diff --git a/.bzrignore b/.bzrignore
index 392572a3f24..ea3f83a4bd5 100644
--- a/.bzrignore
+++ b/.bzrignore
@@ -1444,6 +1444,21 @@ storage/innobase/ib_config.h
 storage/innobase/ib_config.h.in
 storage/innobase/mkinstalldirs
 storage/innobase/stamp-h1
+storage/xtradb/autom4te-2.53.cache/*
+storage/xtradb/autom4te-2.53.cache/output.0
+storage/xtradb/autom4te-2.53.cache/requests
+storage/xtradb/autom4te-2.53.cache/traces.0
+storage/xtradb/autom4te.cache/*
+storage/xtradb/autom4te.cache/output.0
+storage/xtradb/autom4te.cache/requests
+storage/xtradb/autom4te.cache/traces.0
+storage/xtradb/configure.lineno
+storage/xtradb/conftest.s1
+storage/xtradb/conftest.subs
+storage/xtradb/ib_config.h
+storage/xtradb/ib_config.h.in
+storage/xtradb/mkinstalldirs
+storage/xtradb/stamp-h1
 storage/maria/*.MAD
 storage/maria/*.MAI
 storage/maria/ma_rt_test
diff --git a/BUILD/compile-innodb b/BUILD/compile-innodb
new file mode 100755
index 00000000000..82601f03ae9
--- /dev/null
+++ b/BUILD/compile-innodb
@@ -0,0 +1,24 @@
+#! /bin/sh
+#
+# Copyright (c) 2006, 2009, Innobase Oy. All Rights Reserved.
+# 
+# This program is free software; you can redistribute it and/or modify it under
+# the terms of the GNU General Public License as published by the Free Software
+# Foundation; version 2 of the License.
+# 
+# This program is distributed in the hope that it will be useful, but WITHOUT
+# ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+# FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+# 
+# You should have received a copy of the GNU General Public License along with
+# this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+# Place, Suite 330, Boston, MA 02111-1307 USA
+#
+
+path=`dirname $0`
+. "$path/SETUP.sh"
+
+extra_flags="$pentium_cflags $fast_cflags -g"
+extra_configs="$pentium_configs $static_link --with-plugins=innobase"
+
+. "$path/FINISH.sh"
diff --git a/BUILD/compile-innodb-debug b/BUILD/compile-innodb-debug
new file mode 100755
index 00000000000..efb4abf88d5
--- /dev/null
+++ b/BUILD/compile-innodb-debug
@@ -0,0 +1,24 @@
+#! /bin/sh
+#
+# Copyright (c) 2005, 2009, Innobase Oy. All Rights Reserved.
+# 
+# This program is free software; you can redistribute it and/or modify it under
+# the terms of the GNU General Public License as published by the Free Software
+# Foundation; version 2 of the License.
+# 
+# This program is distributed in the hope that it will be useful, but WITHOUT
+# ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+# FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+# 
+# You should have received a copy of the GNU General Public License along with
+# this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+# Place, Suite 330, Boston, MA 02111-1307 USA
+#
+
+path=`dirname $0`
+. "$path/SETUP.sh" $@ --with-debug=full
+
+extra_flags="$pentium_cflags $debug_cflags"
+extra_configs="$pentium_configs $debug_configs --with-plugins=innobase"
+
+. "$path/FINISH.sh"
diff --git a/CMakeLists.txt b/CMakeLists.txt
index b0ce594a425..c6b8e7e2325 100755
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -263,7 +263,7 @@ IF(WITH_FEDERATED_STORAGE_ENGINE)
   ADD_SUBDIRECTORY(storage/federated)
 ENDIF(WITH_FEDERATED_STORAGE_ENGINE)
 IF(WITH_INNOBASE_STORAGE_ENGINE)
-  ADD_SUBDIRECTORY(storage/innobase)
+  ADD_SUBDIRECTORY(storage/xtradb)
 ENDIF(WITH_INNOBASE_STORAGE_ENGINE)
 IF(WITH_MARIA_STORAGE_ENGINE)
   ADD_SUBDIRECTORY(storage/maria)
diff --git a/configure.in b/configure.in
index a46d25b1d12..57c32fbca64 100644
--- a/configure.in
+++ b/configure.in
@@ -1729,6 +1729,30 @@ then
   fi
 fi
 
+AC_CACHE_CHECK([whether the compiler provides atomic builtins],
+	       [mysql_cv_gcc_atomic_builtins], [AC_TRY_RUN([
+  int main()
+  {
+    int foo= -10; int bar= 10;
+    if (!__sync_fetch_and_add(&foo, bar) || foo)
+      return -1;
+    bar= __sync_lock_test_and_set(&foo, bar);
+    if (bar || foo != 10)
+      return -1;
+    bar= __sync_val_compare_and_swap(&bar, foo, 15);
+    if (bar)
+      return -1;
+    return 0;
+  }
+], [mysql_cv_gcc_atomic_builtins=yes],
+   [mysql_cv_gcc_atomic_builtins=no],
+   [mysql_cv_gcc_atomic_builtins=no])])
+
+if test "x$mysql_cv_gcc_atomic_builtins" = xyes; then
+  AC_DEFINE(HAVE_GCC_ATOMIC_BUILTINS, 1,
+	    [Define to 1 if compiler provides atomic builtins.])
+fi
+
 AC_ARG_WITH([atomic-ops],
 	    AC_HELP_STRING([--with-atomic-ops=rwlocks|smp|up],
 	    [Implement atomic operations using pthread rwlocks or atomic CPU
@@ -1742,28 +1766,9 @@ case "$with_atomic_ops" in
                   [Use pthread rwlocks for atomic ops]) ;;
   "smp") ;;
   "")
-    AC_CACHE_CHECK([whether the compiler provides atomic builtins],
-                   [mysql_cv_gcc_atomic_builtins], [AC_TRY_RUN([
-      int main()
-      {
-        int foo= -10; int bar= 10;
-        if (!__sync_fetch_and_add(&foo, bar) || foo)
-          return -1;
-        bar= __sync_lock_test_and_set(&foo, bar);
-        if (bar || foo != 10)
-          return -1;
-        bar= __sync_val_compare_and_swap(&bar, foo, 15);
-        if (bar)
-          return -1;
-        return 0;
-      }
-    ], [mysql_cv_gcc_atomic_builtins=yes_but_disabled],
-       [mysql_cv_gcc_atomic_builtins=no],
-       [mysql_cv_gcc_atomic_builtins=no])])
-
-    if test "x$mysql_cv_gcc_atomic_builtins" = xyes; then
-      AC_DEFINE(HAVE_GCC_ATOMIC_BUILTINS, 1,
-                [Define to 1 if compiler provides atomic builtins.])
+    if test "x$mysql_cv_gcc_atomic_builtins" = xyes_but_disabled; then
+      AC_DEFINE([MY_ATOMIC_MODE_GCC_BUILTINS], [1],
+                [Use GCC atomic builtins for atomic ops])
     fi
    ;;
    *) AC_MSG_ERROR(["$with_atomic_ops" is not a valid value for --with-atomic-ops]) ;;
diff --git a/include/atomic/nolock.h b/include/atomic/nolock.h
index cafd916981d..550b53adcd9 100644
--- a/include/atomic/nolock.h
+++ b/include/atomic/nolock.h
@@ -14,7 +14,7 @@
    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA */
 
 #if defined(__i386__) || defined(_MSC_VER) || \
-    defined(__x86_64__) || defined(HAVE_GCC_ATOMIC_BUILTINS)
+    defined(__x86_64__) || defined(MY_ATOMIC_MODE_GCC_BUILTINS)
 
 #  ifdef MY_ATOMIC_MODE_DUMMY
 #    define LOCK_prefix ""
@@ -22,7 +22,7 @@
 #    define LOCK_prefix "lock"
 #  endif
 
-#  ifdef HAVE_GCC_ATOMIC_BUILTINS
+#  ifdef MY_ATOMIC_MODE_GCC_BUILTINS
 #    include "gcc_builtins.h"
 #  elif __GNUC__
 #    include "x86-gcc.h"
diff --git a/include/my_sys.h b/include/my_sys.h
index fd1c486cead..ad23abafa0a 100644
--- a/include/my_sys.h
+++ b/include/my_sys.h
@@ -791,6 +791,9 @@ extern size_t my_b_gets(IO_CACHE *info, char *to, size_t max_length);
 extern my_off_t my_b_filelength(IO_CACHE *info);
 extern size_t my_b_printf(IO_CACHE *info, const char* fmt, ...);
 extern size_t my_b_vprintf(IO_CACHE *info, const char* fmt, va_list ap);
+extern int init_strvar_from_file(char *var, int max_size, IO_CACHE *f,
+                                 const char *default_val);
+extern int init_intvar_from_file(int* var, IO_CACHE* f, int default_val);
 extern my_bool open_cached_file(IO_CACHE *cache,const char *dir,
 				 const char *prefix, size_t cache_size,
 				 myf cache_myflags);
diff --git a/libmysqld/CMakeLists.txt b/libmysqld/CMakeLists.txt
index 5b22fd07e26..7f311844c6a 100644
--- a/libmysqld/CMakeLists.txt
+++ b/libmysqld/CMakeLists.txt
@@ -132,9 +132,9 @@ IF(WITH_FEDERATED_STORAGE_ENGINE)
 ENDIF(WITH_FEDERATED_STORAGE_ENGINE)
 
 IF(WITH_INNOBASE_STORAGE_ENGINE)
-  INCLUDE(${CMAKE_SOURCE_DIR}/storage/innobase/CMakeLists.txt)
+  INCLUDE(${CMAKE_SOURCE_DIR}/storage/xtradb/CMakeLists.txt)
   FOREACH(rpath ${INNOBASE_SOURCES})
-    SET(LIB_SOURCES ${LIB_SOURCES} ../storage/innobase/${rpath})
+    SET(LIB_SOURCES ${LIB_SOURCES} ../storage/xtradb/${rpath})
   ENDFOREACH(rpath)
 ENDIF(WITH_INNOBASE_STORAGE_ENGINE)
 
diff --git a/mysql-test/include/innodb-index.inc b/mysql-test/include/innodb-index.inc
new file mode 100644
index 00000000000..37de3162abe
--- /dev/null
+++ b/mysql-test/include/innodb-index.inc
@@ -0,0 +1,26 @@
+--eval create table t1(a int not null, b int, c char(10), d varchar(20), primary key (a)) engine = innodb default charset=$charset
+insert into t1 values (1,1,'ab','ab'),(2,2,'ac','ac'),(3,2,'ad','ad'),(4,4,'afe','afe');
+commit;
+--error ER_DUP_ENTRY
+alter table t1 add unique index (b);
+insert into t1 values(8,9,'fff','fff');
+select * from t1;
+show create table t1;
+alter table t1 add index (b);
+insert into t1 values(10,10,'kkk','iii');
+select * from t1;
+select * from t1 force index(b) order by b;
+explain select * from t1 force index(b) order by b;
+show create table t1;
+alter table t1 add unique index (c), add index (d);
+insert into t1 values(11,11,'aaa','mmm');
+select * from t1;
+select * from t1 force index(b) order by b;
+select * from t1 force index(c) order by c;
+select * from t1 force index(d) order by d;
+explain select * from t1 force index(b) order by b;
+explain select * from t1 force index(c) order by c;
+explain select * from t1 force index(d) order by d;
+show create table t1;
+check table t1;
+drop table t1;
diff --git a/mysql-test/r/index_merge_innodb.result b/mysql-test/r/index_merge_innodb.result
index 588de70e6e5..e71f547d36a 100644
--- a/mysql-test/r/index_merge_innodb.result
+++ b/mysql-test/r/index_merge_innodb.result
@@ -111,7 +111,7 @@ count(*)
 explain select count(*) from t1 where
 key1a = 2 and key1b is null and  key2a = 2 and key2b is null;
 id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
-1	SIMPLE	t1	index_merge	i1,i2	i1,i2	10,10	NULL	4	Using intersect(i1,i2); Using where; Using index
+1	SIMPLE	t1	index_merge	i1,i2	i1,i2	10,10	NULL	3	Using intersect(i1,i2); Using where; Using index
 select count(*) from t1 where
 key1a = 2 and key1b is null and key2a = 2 and key2b is null;
 count(*)
@@ -119,7 +119,7 @@ count(*)
 explain select count(*) from t1 where
 key1a = 2 and key1b is null and key3a = 2 and key3b is null;
 id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
-1	SIMPLE	t1	index_merge	i1,i3	i1,i3	10,10	NULL	4	Using intersect(i1,i3); Using where; Using index
+1	SIMPLE	t1	index_merge	i1,i3	i1,i3	10,10	NULL	3	Using intersect(i1,i3); Using where; Using index
 select count(*) from t1 where
 key1a = 2 and key1b is null and key3a = 2 and key3b is null;
 count(*)
diff --git a/mysql-test/r/information_schema.result b/mysql-test/r/information_schema.result
index 9ff2160e909..95ed38697f3 100644
--- a/mysql-test/r/information_schema.result
+++ b/mysql-test/r/information_schema.result
@@ -42,7 +42,7 @@ WHERE table_schema IN ('mysql', 'INFORMATION_SCHEMA', 'test', 'mysqltest') AND
 table_name<>'ndb_binlog_index' AND
 table_name<>'ndb_apply_status' AND
 NOT (table_schema = 'INFORMATION_SCHEMA' AND table_name LIKE 'PBXT_%');
-select * from v1;
+select * from v1 ORDER BY c COLLATE utf8_bin;
 c
 CHARACTER_SETS
 COLLATIONS
@@ -54,6 +54,17 @@ EVENTS
 FILES
 GLOBAL_STATUS
 GLOBAL_VARIABLES
+INNODB_BUFFER_POOL_PAGES
+INNODB_BUFFER_POOL_PAGES_BLOB
+INNODB_BUFFER_POOL_PAGES_INDEX
+INNODB_CMP
+INNODB_CMPMEM
+INNODB_CMPMEM_RESET
+INNODB_CMP_RESET
+INNODB_LOCKS
+INNODB_LOCK_WAITS
+INNODB_RSEG
+INNODB_TRX
 KEY_COLUMN_USAGE
 PARTITIONS
 PLUGINS
@@ -72,6 +83,7 @@ TABLE_PRIVILEGES
 TRIGGERS
 USER_PRIVILEGES
 VIEWS
+XTRADB_ENHANCEMENTS
 columns_priv
 db
 event
@@ -87,6 +99,11 @@ proc
 procs_priv
 servers
 slow_log
+t1
+t2
+t3
+t4
+t5
 tables_priv
 time_zone
 time_zone_leap_second
@@ -94,11 +111,6 @@ time_zone_name
 time_zone_transition
 time_zone_transition_type
 user
-t1
-t4
-t2
-t3
-t5
 v1
 select c,table_name from v1 
 inner join information_schema.TABLES v2 on (v1.c=v2.table_name)
@@ -800,6 +812,8 @@ TABLES	CREATE_TIME	datetime
 TABLES	UPDATE_TIME	datetime
 TABLES	CHECK_TIME	datetime
 TRIGGERS	CREATED	datetime
+INNODB_TRX	trx_started	datetime
+INNODB_TRX	trx_wait_started	datetime
 event	execute_at	datetime
 event	last_executed	datetime
 event	starts	datetime
@@ -848,6 +862,7 @@ TABLES	TABLE_NAME	select
 TABLE_CONSTRAINTS	TABLE_NAME	select
 TABLE_PRIVILEGES	TABLE_NAME	select
 VIEWS	TABLE_NAME	select
+INNODB_BUFFER_POOL_PAGES_INDEX	table_name	select
 delete from mysql.user where user='mysqltest_4';
 delete from mysql.db where user='mysqltest_4';
 flush privileges;
@@ -1223,12 +1238,12 @@ DROP PROCEDURE p1;
 DROP USER mysql_bug20230@localhost;
 SELECT MAX(table_name) FROM information_schema.tables WHERE table_schema IN ('mysql', 'INFORMATION_SCHEMA', 'test');
 MAX(table_name)
-VIEWS
+XTRADB_ENHANCEMENTS
 SELECT table_name from information_schema.tables
 WHERE table_name=(SELECT MAX(table_name)
 FROM information_schema.tables WHERE table_schema IN ('mysql', 'INFORMATION_SCHEMA', 'test'));
 table_name
-VIEWS
+XTRADB_ENHANCEMENTS
 DROP TABLE IF EXISTS bug23037;
 DROP FUNCTION IF EXISTS get_value;
 SELECT COLUMN_NAME, MD5(COLUMN_DEFAULT), LENGTH(COLUMN_DEFAULT) FROM INFORMATION_SCHEMA.COLUMNS WHERE TABLE_NAME='bug23037';
diff --git a/mysql-test/r/information_schema_all_engines.result b/mysql-test/r/information_schema_all_engines.result
index 4ffaa12fd61..ed73f0a42e0 100644
--- a/mysql-test/r/information_schema_all_engines.result
+++ b/mysql-test/r/information_schema_all_engines.result
@@ -29,7 +29,19 @@ TABLE_PRIVILEGES
 TRIGGERS
 USER_PRIVILEGES
 VIEWS
+INNODB_BUFFER_POOL_PAGES
 PBXT_STATISTICS
+INNODB_CMP
+INNODB_RSEG
+XTRADB_ENHANCEMENTS
+INNODB_BUFFER_POOL_PAGES_INDEX
+INNODB_BUFFER_POOL_PAGES_BLOB
+INNODB_TRX
+INNODB_CMP_RESET
+INNODB_LOCK_WAITS
+INNODB_CMPMEM_RESET
+INNODB_LOCKS
+INNODB_CMPMEM
 SELECT t.table_name, c1.column_name
 FROM information_schema.tables t
 INNER JOIN
@@ -73,7 +85,19 @@ TABLE_PRIVILEGES	TABLE_SCHEMA
 TRIGGERS	TRIGGER_SCHEMA
 USER_PRIVILEGES	GRANTEE
 VIEWS	TABLE_SCHEMA
+INNODB_BUFFER_POOL_PAGES	page_type
 PBXT_STATISTICS	ID
+INNODB_CMP	page_size
+INNODB_RSEG	rseg_id
+XTRADB_ENHANCEMENTS	name
+INNODB_BUFFER_POOL_PAGES_INDEX	schema_name
+INNODB_BUFFER_POOL_PAGES_BLOB	space_id
+INNODB_TRX	trx_id
+INNODB_CMP_RESET	page_size
+INNODB_LOCK_WAITS	requesting_trx_id
+INNODB_CMPMEM_RESET	page_size
+INNODB_LOCKS	lock_id
+INNODB_CMPMEM	page_size
 SELECT t.table_name, c1.column_name
 FROM information_schema.tables t
 INNER JOIN
@@ -117,7 +141,19 @@ TABLE_PRIVILEGES	TABLE_SCHEMA
 TRIGGERS	TRIGGER_SCHEMA
 USER_PRIVILEGES	GRANTEE
 VIEWS	TABLE_SCHEMA
+INNODB_BUFFER_POOL_PAGES	page_type
 PBXT_STATISTICS	ID
+INNODB_CMP	page_size
+INNODB_RSEG	rseg_id
+XTRADB_ENHANCEMENTS	name
+INNODB_BUFFER_POOL_PAGES_INDEX	schema_name
+INNODB_BUFFER_POOL_PAGES_BLOB	space_id
+INNODB_TRX	trx_id
+INNODB_CMP_RESET	page_size
+INNODB_LOCK_WAITS	requesting_trx_id
+INNODB_CMPMEM_RESET	page_size
+INNODB_LOCKS	lock_id
+INNODB_CMPMEM	page_size
 select 1 as f1 from information_schema.tables  where "CHARACTER_SETS"=
 (select cast(table_name as char)  from information_schema.tables
 order by table_name limit 1) limit 1;
@@ -149,6 +185,17 @@ EVENTS	information_schema.EVENTS	1
 FILES	information_schema.FILES	1
 GLOBAL_STATUS	information_schema.GLOBAL_STATUS	1
 GLOBAL_VARIABLES	information_schema.GLOBAL_VARIABLES	1
+INNODB_BUFFER_POOL_PAGES	information_schema.INNODB_BUFFER_POOL_PAGES	1
+INNODB_BUFFER_POOL_PAGES_BLOB	information_schema.INNODB_BUFFER_POOL_PAGES_BLOB	1
+INNODB_BUFFER_POOL_PAGES_INDEX	information_schema.INNODB_BUFFER_POOL_PAGES_INDEX	1
+INNODB_CMP	information_schema.INNODB_CMP	1
+INNODB_CMPMEM	information_schema.INNODB_CMPMEM	1
+INNODB_CMPMEM_RESET	information_schema.INNODB_CMPMEM_RESET	1
+INNODB_CMP_RESET	information_schema.INNODB_CMP_RESET	1
+INNODB_LOCKS	information_schema.INNODB_LOCKS	1
+INNODB_LOCK_WAITS	information_schema.INNODB_LOCK_WAITS	1
+INNODB_RSEG	information_schema.INNODB_RSEG	1
+INNODB_TRX	information_schema.INNODB_TRX	1
 KEY_COLUMN_USAGE	information_schema.KEY_COLUMN_USAGE	1
 PARTITIONS	information_schema.PARTITIONS	1
 PBXT_STATISTICS	information_schema.PBXT_STATISTICS	1
@@ -168,6 +215,7 @@ TABLE_PRIVILEGES	information_schema.TABLE_PRIVILEGES	1
 TRIGGERS	information_schema.TRIGGERS	1
 USER_PRIVILEGES	information_schema.USER_PRIVILEGES	1
 VIEWS	information_schema.VIEWS	1
+XTRADB_ENHANCEMENTS	information_schema.XTRADB_ENHANCEMENTS	1
 Database: information_schema
 +---------------------------------------+
 |                Tables                 |
@@ -200,7 +248,19 @@ Database: information_schema
 | TRIGGERS                              |
 | USER_PRIVILEGES                       |
 | VIEWS                                 |
+| INNODB_BUFFER_POOL_PAGES              |
 | PBXT_STATISTICS                       |
+| INNODB_CMP                            |
+| INNODB_RSEG                           |
+| XTRADB_ENHANCEMENTS                   |
+| INNODB_BUFFER_POOL_PAGES_INDEX        |
+| INNODB_BUFFER_POOL_PAGES_BLOB         |
+| INNODB_TRX                            |
+| INNODB_CMP_RESET                      |
+| INNODB_LOCK_WAITS                     |
+| INNODB_CMPMEM_RESET                   |
+| INNODB_LOCKS                          |
+| INNODB_CMPMEM                         |
 +---------------------------------------+
 Database: INFORMATION_SCHEMA
 +---------------------------------------+
@@ -234,7 +294,19 @@ Database: INFORMATION_SCHEMA
 | TRIGGERS                              |
 | USER_PRIVILEGES                       |
 | VIEWS                                 |
+| INNODB_BUFFER_POOL_PAGES              |
 | PBXT_STATISTICS                       |
+| INNODB_CMP                            |
+| INNODB_RSEG                           |
+| XTRADB_ENHANCEMENTS                   |
+| INNODB_BUFFER_POOL_PAGES_INDEX        |
+| INNODB_BUFFER_POOL_PAGES_BLOB         |
+| INNODB_TRX                            |
+| INNODB_CMP_RESET                      |
+| INNODB_LOCK_WAITS                     |
+| INNODB_CMPMEM_RESET                   |
+| INNODB_LOCKS                          |
+| INNODB_CMPMEM                         |
 +---------------------------------------+
 Wildcard: inf_rmation_schema
 +--------------------+
@@ -244,5 +316,5 @@ Wildcard: inf_rmation_schema
 +--------------------+
 SELECT table_schema, count(*) FROM information_schema.TABLES WHERE table_schema IN ('mysql', 'INFORMATION_SCHEMA', 'test', 'mysqltest') AND table_name<>'ndb_binlog_index' AND table_name<>'ndb_apply_status' GROUP BY TABLE_SCHEMA;
 table_schema	count(*)
-information_schema	29
+information_schema	41
 mysql	22
diff --git a/mysql-test/r/innodb-analyze.result b/mysql-test/r/innodb-analyze.result
new file mode 100644
index 00000000000..2aee004a2d6
--- /dev/null
+++ b/mysql-test/r/innodb-analyze.result
@@ -0,0 +1,2 @@
+Variable_name	Value
+innodb_stats_sample_pages	1
diff --git a/mysql-test/r/innodb-autoinc.result b/mysql-test/r/innodb-autoinc.result
index 36d8e26de9e..8174ba52b41 100644
--- a/mysql-test/r/innodb-autoinc.result
+++ b/mysql-test/r/innodb-autoinc.result
@@ -581,11 +581,10 @@ c1
 DROP TABLE t1;
 SET @@SESSION.AUTO_INCREMENT_INCREMENT=1, @@SESSION.AUTO_INCREMENT_OFFSET=1;
 SET @@INSERT_ID=1;
-SHOW VARIABLES LIKE "%auto_inc%";
+SHOW VARIABLES LIKE "auto_inc%";
 Variable_name	Value
 auto_increment_increment	1
 auto_increment_offset	1
-pbxt_auto_increment_mode	0
 CREATE TABLE t1 (c1 DOUBLE NOT NULL AUTO_INCREMENT, c2 INT, PRIMARY KEY (c1)) ENGINE=InnoDB;
 INSERT INTO t1 VALUES(NULL, 1);
 INSERT INTO t1 VALUES(NULL, 2);
diff --git a/mysql-test/r/innodb-index.result b/mysql-test/r/innodb-index.result
new file mode 100644
index 00000000000..e096334dff3
--- /dev/null
+++ b/mysql-test/r/innodb-index.result
@@ -0,0 +1,1137 @@
+SET @save_innodb_file_format_check=@@global.innodb_file_format_check;
+create table t1(a int not null, b int, c char(10) not null, d varchar(20)) engine = innodb;
+insert into t1 values (5,5,'oo','oo'),(4,4,'tr','tr'),(3,4,'ad','ad'),(2,3,'ak','ak');
+commit;
+alter table t1 add index b (b), add index b (b);
+ERROR 42000: Duplicate key name 'b'
+alter table t1 add index (b,b);
+ERROR 42S21: Duplicate column name 'b'
+alter table t1 add index d2 (d);
+show create table t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `a` int(11) NOT NULL,
+  `b` int(11) DEFAULT NULL,
+  `c` char(10) NOT NULL,
+  `d` varchar(20) DEFAULT NULL,
+  KEY `d2` (`d`)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1
+explain select * from t1 force index(d2) order by d;
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	index	NULL	d2	23	NULL	4	
+select * from t1 force index (d2) order by d;
+a	b	c	d
+3	4	ad	ad
+2	3	ak	ak
+5	5	oo	oo
+4	4	tr	tr
+alter table t1 add unique index (b);
+ERROR 23000: Duplicate entry '4' for key 'b'
+show create table t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `a` int(11) NOT NULL,
+  `b` int(11) DEFAULT NULL,
+  `c` char(10) NOT NULL,
+  `d` varchar(20) DEFAULT NULL,
+  KEY `d2` (`d`)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1
+alter table t1 add index (b);
+show create table t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `a` int(11) NOT NULL,
+  `b` int(11) DEFAULT NULL,
+  `c` char(10) NOT NULL,
+  `d` varchar(20) DEFAULT NULL,
+  KEY `d2` (`d`),
+  KEY `b` (`b`)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1
+CREATE TABLE `t1#1`(a INT PRIMARY KEY) ENGINE=InnoDB;
+call mtr.add_suppression(" table `test`\\.`t1#[12]` already exists in InnoDB internal");
+alter table t1 add unique index (c), add index (d);
+ERROR HY000: Table 'test.t1#1' already exists
+rename table `t1#1` to `t1#2`;
+alter table t1 add unique index (c), add index (d);
+ERROR HY000: Table 'test.t1#2' already exists
+drop table `t1#2`;
+alter table t1 add unique index (c), add index (d);
+show create table t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `a` int(11) NOT NULL,
+  `b` int(11) DEFAULT NULL,
+  `c` char(10) NOT NULL,
+  `d` varchar(20) DEFAULT NULL,
+  UNIQUE KEY `c` (`c`),
+  KEY `d2` (`d`),
+  KEY `b` (`b`),
+  KEY `d` (`d`)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1
+explain select * from t1 force index(c) order by c;
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	index	NULL	c	10	NULL	4	
+alter table t1 add primary key (a), drop index c;
+show create table t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `a` int(11) NOT NULL,
+  `b` int(11) DEFAULT NULL,
+  `c` char(10) NOT NULL,
+  `d` varchar(20) DEFAULT NULL,
+  PRIMARY KEY (`a`),
+  KEY `d2` (`d`),
+  KEY `b` (`b`),
+  KEY `d` (`d`)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1
+alter table t1 add primary key (c);
+ERROR 42000: Multiple primary key defined
+alter table t1 drop primary key, add primary key (b);
+ERROR 23000: Duplicate entry '4' for key 'PRIMARY'
+create unique index c on t1 (c);
+show create table t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `a` int(11) NOT NULL,
+  `b` int(11) DEFAULT NULL,
+  `c` char(10) NOT NULL,
+  `d` varchar(20) DEFAULT NULL,
+  PRIMARY KEY (`a`),
+  UNIQUE KEY `c` (`c`),
+  KEY `d2` (`d`),
+  KEY `b` (`b`),
+  KEY `d` (`d`)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1
+explain select * from t1 force index(c) order by c;
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	index	NULL	c	10	NULL	4	
+select * from t1 force index(c) order by c;
+a	b	c	d
+3	4	ad	ad
+2	3	ak	ak
+5	5	oo	oo
+4	4	tr	tr
+alter table t1 drop index b, add index (b);
+show create table t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `a` int(11) NOT NULL,
+  `b` int(11) DEFAULT NULL,
+  `c` char(10) NOT NULL,
+  `d` varchar(20) DEFAULT NULL,
+  PRIMARY KEY (`a`),
+  UNIQUE KEY `c` (`c`),
+  KEY `d2` (`d`),
+  KEY `d` (`d`),
+  KEY `b` (`b`)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1
+insert into t1 values(6,1,'ggg','ggg');
+select * from t1;
+a	b	c	d
+2	3	ak	ak
+3	4	ad	ad
+4	4	tr	tr
+5	5	oo	oo
+6	1	ggg	ggg
+select * from t1 force index(b) order by b;
+a	b	c	d
+6	1	ggg	ggg
+2	3	ak	ak
+3	4	ad	ad
+4	4	tr	tr
+5	5	oo	oo
+select * from t1 force index(c) order by c;
+a	b	c	d
+3	4	ad	ad
+2	3	ak	ak
+6	1	ggg	ggg
+5	5	oo	oo
+4	4	tr	tr
+select * from t1 force index(d) order by d;
+a	b	c	d
+3	4	ad	ad
+2	3	ak	ak
+6	1	ggg	ggg
+5	5	oo	oo
+4	4	tr	tr
+explain select * from t1 force index(b) order by b;
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	index	NULL	b	5	NULL	5	
+explain select * from t1 force index(c) order by c;
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	index	NULL	c	10	NULL	5	
+explain select * from t1 force index(d) order by d;
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	index	NULL	d	23	NULL	5	
+show create table t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `a` int(11) NOT NULL,
+  `b` int(11) DEFAULT NULL,
+  `c` char(10) NOT NULL,
+  `d` varchar(20) DEFAULT NULL,
+  PRIMARY KEY (`a`),
+  UNIQUE KEY `c` (`c`),
+  KEY `d2` (`d`),
+  KEY `d` (`d`),
+  KEY `b` (`b`)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1
+drop table t1;
+create table t1(a int not null, b int, c char(10), d varchar(20), primary key (a)) engine = innodb;
+insert into t1 values (1,1,'ab','ab'),(2,2,'ac','ac'),(3,3,'ad','ad'),(4,4,'afe','afe');
+commit;
+alter table t1 add index (c(2));
+show create table t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `a` int(11) NOT NULL,
+  `b` int(11) DEFAULT NULL,
+  `c` char(10) DEFAULT NULL,
+  `d` varchar(20) DEFAULT NULL,
+  PRIMARY KEY (`a`),
+  KEY `c` (`c`(2))
+) ENGINE=InnoDB DEFAULT CHARSET=latin1
+alter table t1 add unique index (d(10));
+show create table t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `a` int(11) NOT NULL,
+  `b` int(11) DEFAULT NULL,
+  `c` char(10) DEFAULT NULL,
+  `d` varchar(20) DEFAULT NULL,
+  PRIMARY KEY (`a`),
+  UNIQUE KEY `d` (`d`(10)),
+  KEY `c` (`c`(2))
+) ENGINE=InnoDB DEFAULT CHARSET=latin1
+insert into t1 values(5,1,'ggg','ggg');
+select * from t1;
+a	b	c	d
+1	1	ab	ab
+2	2	ac	ac
+3	3	ad	ad
+4	4	afe	afe
+5	1	ggg	ggg
+select * from t1 force index(c) order by c;
+a	b	c	d
+1	1	ab	ab
+2	2	ac	ac
+3	3	ad	ad
+4	4	afe	afe
+5	1	ggg	ggg
+select * from t1 force index(d) order by d;
+a	b	c	d
+1	1	ab	ab
+2	2	ac	ac
+3	3	ad	ad
+4	4	afe	afe
+5	1	ggg	ggg
+explain select * from t1 order by b;
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	ALL	NULL	NULL	NULL	NULL	5	Using filesort
+explain select * from t1 force index(c) order by c;
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	ALL	NULL	NULL	NULL	NULL	5	Using filesort
+explain select * from t1 force index(d) order by d;
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	ALL	NULL	NULL	NULL	NULL	5	Using filesort
+show create table t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `a` int(11) NOT NULL,
+  `b` int(11) DEFAULT NULL,
+  `c` char(10) DEFAULT NULL,
+  `d` varchar(20) DEFAULT NULL,
+  PRIMARY KEY (`a`),
+  UNIQUE KEY `d` (`d`(10)),
+  KEY `c` (`c`(2))
+) ENGINE=InnoDB DEFAULT CHARSET=latin1
+alter table t1 drop index d;
+insert into t1 values(8,9,'fff','fff');
+select * from t1;
+a	b	c	d
+1	1	ab	ab
+2	2	ac	ac
+3	3	ad	ad
+4	4	afe	afe
+5	1	ggg	ggg
+8	9	fff	fff
+select * from t1 force index(c) order by c;
+a	b	c	d
+1	1	ab	ab
+2	2	ac	ac
+3	3	ad	ad
+4	4	afe	afe
+8	9	fff	fff
+5	1	ggg	ggg
+explain select * from t1 order by b;
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	ALL	NULL	NULL	NULL	NULL	6	Using filesort
+explain select * from t1 force index(c) order by c;
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	ALL	NULL	NULL	NULL	NULL	6	Using filesort
+explain select * from t1 order by d;
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	ALL	NULL	NULL	NULL	NULL	6	Using filesort
+show create table t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `a` int(11) NOT NULL,
+  `b` int(11) DEFAULT NULL,
+  `c` char(10) DEFAULT NULL,
+  `d` varchar(20) DEFAULT NULL,
+  PRIMARY KEY (`a`),
+  KEY `c` (`c`(2))
+) ENGINE=InnoDB DEFAULT CHARSET=latin1
+drop table t1;
+create table t1(a int not null, b int, c char(10), d varchar(20), primary key (a)) engine = innodb;
+insert into t1 values (1,1,'ab','ab'),(2,2,'ac','ac'),(3,2,'ad','ad'),(4,4,'afe','afe');
+commit;
+alter table t1 add unique index (b,c);
+insert into t1 values(8,9,'fff','fff');
+select * from t1;
+a	b	c	d
+1	1	ab	ab
+2	2	ac	ac
+3	2	ad	ad
+4	4	afe	afe
+8	9	fff	fff
+select * from t1 force index(b) order by b;
+a	b	c	d
+1	1	ab	ab
+2	2	ac	ac
+3	2	ad	ad
+4	4	afe	afe
+8	9	fff	fff
+explain select * from t1 force index(b) order by b;
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	index	NULL	b	16	NULL	5	
+show create table t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `a` int(11) NOT NULL,
+  `b` int(11) DEFAULT NULL,
+  `c` char(10) DEFAULT NULL,
+  `d` varchar(20) DEFAULT NULL,
+  PRIMARY KEY (`a`),
+  UNIQUE KEY `b` (`b`,`c`)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1
+alter table t1 add index (b,c);
+insert into t1 values(11,11,'kkk','kkk');
+select * from t1;
+a	b	c	d
+1	1	ab	ab
+2	2	ac	ac
+3	2	ad	ad
+4	4	afe	afe
+8	9	fff	fff
+11	11	kkk	kkk
+select * from t1 force index(b) order by b;
+a	b	c	d
+1	1	ab	ab
+2	2	ac	ac
+3	2	ad	ad
+4	4	afe	afe
+8	9	fff	fff
+11	11	kkk	kkk
+explain select * from t1 force index(b) order by b;
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	index	NULL	b	16	NULL	6	
+show create table t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `a` int(11) NOT NULL,
+  `b` int(11) DEFAULT NULL,
+  `c` char(10) DEFAULT NULL,
+  `d` varchar(20) DEFAULT NULL,
+  PRIMARY KEY (`a`),
+  UNIQUE KEY `b` (`b`,`c`),
+  KEY `b_2` (`b`,`c`)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1
+alter table t1 add unique index (c,d);
+insert into t1 values(13,13,'yyy','aaa');
+select * from t1;
+a	b	c	d
+1	1	ab	ab
+2	2	ac	ac
+3	2	ad	ad
+4	4	afe	afe
+8	9	fff	fff
+11	11	kkk	kkk
+13	13	yyy	aaa
+select * from t1 force index(b) order by b;
+a	b	c	d
+1	1	ab	ab
+2	2	ac	ac
+3	2	ad	ad
+4	4	afe	afe
+8	9	fff	fff
+11	11	kkk	kkk
+13	13	yyy	aaa
+select * from t1 force index(c) order by c;
+a	b	c	d
+1	1	ab	ab
+2	2	ac	ac
+3	2	ad	ad
+4	4	afe	afe
+8	9	fff	fff
+11	11	kkk	kkk
+13	13	yyy	aaa
+explain select * from t1 force index(b) order by b;
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	index	NULL	b	16	NULL	7	
+explain select * from t1 force index(c) order by c;
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	index	NULL	c	34	NULL	7	
+show create table t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `a` int(11) NOT NULL,
+  `b` int(11) DEFAULT NULL,
+  `c` char(10) DEFAULT NULL,
+  `d` varchar(20) DEFAULT NULL,
+  PRIMARY KEY (`a`),
+  UNIQUE KEY `b` (`b`,`c`),
+  UNIQUE KEY `c` (`c`,`d`),
+  KEY `b_2` (`b`,`c`)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1
+drop table t1;
+create table t1(a int not null, b int not null, c int, primary key (a), key (b)) engine = innodb;
+create table t3(a int not null, c int not null, d int, primary key (a), key (c)) engine = innodb;
+create table t4(a int not null, d int not null, e int, primary key (a), key (d)) engine = innodb;
+create table t2(a int not null, b int not null, c int not null, d int not null, e int,
+foreign key (b) references t1(b) on delete cascade,
+foreign key (c) references t3(c), foreign key (d) references t4(d))
+engine = innodb;
+alter table t1 drop index b;
+ERROR HY000: Cannot drop index 'b': needed in a foreign key constraint
+alter table t3 drop index c;
+ERROR HY000: Cannot drop index 'c': needed in a foreign key constraint
+alter table t4 drop index d;
+ERROR HY000: Cannot drop index 'd': needed in a foreign key constraint
+alter table t2 drop index b;
+ERROR HY000: Cannot drop index 'b': needed in a foreign key constraint
+alter table t2 drop index b, drop index c, drop index d;
+ERROR HY000: Cannot drop index 'b': needed in a foreign key constraint
+create unique index dc on t2 (d,c);
+create index dc on t1 (b,c);
+alter table t2 add primary key (a);
+insert into t1 values (1,1,1);
+insert into t3 values (1,1,1);
+insert into t4 values (1,1,1);
+insert into t2 values (1,1,1,1,1);
+commit;
+alter table t4 add constraint dc foreign key (a) references t1(a);
+show create table t4;
+Table	Create Table
+t4	CREATE TABLE `t4` (
+  `a` int(11) NOT NULL,
+  `d` int(11) NOT NULL,
+  `e` int(11) DEFAULT NULL,
+  PRIMARY KEY (`a`),
+  KEY `d` (`d`),
+  CONSTRAINT `dc` FOREIGN KEY (`a`) REFERENCES `t1` (`a`)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1
+alter table t3 add constraint dc foreign key (a) references t1(a);
+ERROR HY000: Can't create table '#sql-temporary' (errno: 121)
+show create table t3;
+Table	Create Table
+t3	CREATE TABLE `t3` (
+  `a` int(11) NOT NULL,
+  `c` int(11) NOT NULL,
+  `d` int(11) DEFAULT NULL,
+  PRIMARY KEY (`a`),
+  KEY `c` (`c`)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1
+alter table t2 drop index b, add index (b);
+show create table t2;
+Table	Create Table
+t2	CREATE TABLE `t2` (
+  `a` int(11) NOT NULL,
+  `b` int(11) NOT NULL,
+  `c` int(11) NOT NULL,
+  `d` int(11) NOT NULL,
+  `e` int(11) DEFAULT NULL,
+  PRIMARY KEY (`a`),
+  UNIQUE KEY `dc` (`d`,`c`),
+  KEY `c` (`c`),
+  KEY `b` (`b`),
+  CONSTRAINT `t2_ibfk_1` FOREIGN KEY (`b`) REFERENCES `t1` (`b`) ON DELETE CASCADE,
+  CONSTRAINT `t2_ibfk_2` FOREIGN KEY (`c`) REFERENCES `t3` (`c`),
+  CONSTRAINT `t2_ibfk_3` FOREIGN KEY (`d`) REFERENCES `t4` (`d`)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1
+delete from t1;
+ERROR 23000: Cannot delete or update a parent row: a foreign key constraint fails (`test`.`t4`, CONSTRAINT `dc` FOREIGN KEY (`a`) REFERENCES `t1` (`a`))
+drop index dc on t4;
+ERROR 42000: Can't DROP 'dc'; check that column/key exists
+alter table t3 drop foreign key dc;
+ERROR HY000: Error on rename of './test/t3' to '#sql2-temporary' (errno: 152)
+alter table t4 drop foreign key dc;
+select * from t2;
+a	b	c	d	e
+1	1	1	1	1
+delete from t1;
+select * from t2;
+a	b	c	d	e
+drop table t2,t4,t3,t1;
+create table t1(a int not null, b int, c char(10), d varchar(20), primary key (a)) engine = innodb default charset=utf8;
+insert into t1 values (1,1,'ab','ab'),(2,2,'ac','ac'),(3,2,'ad','ad'),(4,4,'afe','afe');
+commit;
+alter table t1 add unique index (b);
+ERROR 23000: Duplicate entry '2' for key 'b'
+insert into t1 values(8,9,'fff','fff');
+select * from t1;
+a	b	c	d
+1	1	ab	ab
+2	2	ac	ac
+3	2	ad	ad
+4	4	afe	afe
+8	9	fff	fff
+show create table t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `a` int(11) NOT NULL,
+  `b` int(11) DEFAULT NULL,
+  `c` char(10) DEFAULT NULL,
+  `d` varchar(20) DEFAULT NULL,
+  PRIMARY KEY (`a`)
+) ENGINE=InnoDB DEFAULT CHARSET=utf8
+alter table t1 add index (b);
+insert into t1 values(10,10,'kkk','iii');
+select * from t1;
+a	b	c	d
+1	1	ab	ab
+2	2	ac	ac
+3	2	ad	ad
+4	4	afe	afe
+8	9	fff	fff
+10	10	kkk	iii
+select * from t1 force index(b) order by b;
+a	b	c	d
+1	1	ab	ab
+2	2	ac	ac
+3	2	ad	ad
+4	4	afe	afe
+8	9	fff	fff
+10	10	kkk	iii
+explain select * from t1 force index(b) order by b;
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	index	NULL	b	5	NULL	6	
+show create table t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `a` int(11) NOT NULL,
+  `b` int(11) DEFAULT NULL,
+  `c` char(10) DEFAULT NULL,
+  `d` varchar(20) DEFAULT NULL,
+  PRIMARY KEY (`a`),
+  KEY `b` (`b`)
+) ENGINE=InnoDB DEFAULT CHARSET=utf8
+alter table t1 add unique index (c), add index (d);
+insert into t1 values(11,11,'aaa','mmm');
+select * from t1;
+a	b	c	d
+1	1	ab	ab
+2	2	ac	ac
+3	2	ad	ad
+4	4	afe	afe
+8	9	fff	fff
+10	10	kkk	iii
+11	11	aaa	mmm
+select * from t1 force index(b) order by b;
+a	b	c	d
+1	1	ab	ab
+2	2	ac	ac
+3	2	ad	ad
+4	4	afe	afe
+8	9	fff	fff
+10	10	kkk	iii
+11	11	aaa	mmm
+select * from t1 force index(c) order by c;
+a	b	c	d
+11	11	aaa	mmm
+1	1	ab	ab
+2	2	ac	ac
+3	2	ad	ad
+4	4	afe	afe
+8	9	fff	fff
+10	10	kkk	iii
+select * from t1 force index(d) order by d;
+a	b	c	d
+1	1	ab	ab
+2	2	ac	ac
+3	2	ad	ad
+4	4	afe	afe
+8	9	fff	fff
+10	10	kkk	iii
+11	11	aaa	mmm
+explain select * from t1 force index(b) order by b;
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	index	NULL	b	5	NULL	7	
+explain select * from t1 force index(c) order by c;
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	index	NULL	c	31	NULL	7	
+explain select * from t1 force index(d) order by d;
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	index	NULL	d	63	NULL	7	
+show create table t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `a` int(11) NOT NULL,
+  `b` int(11) DEFAULT NULL,
+  `c` char(10) DEFAULT NULL,
+  `d` varchar(20) DEFAULT NULL,
+  PRIMARY KEY (`a`),
+  UNIQUE KEY `c` (`c`),
+  KEY `b` (`b`),
+  KEY `d` (`d`)
+) ENGINE=InnoDB DEFAULT CHARSET=utf8
+check table t1;
+Table	Op	Msg_type	Msg_text
+test.t1	check	status	OK
+drop table t1;
+create table t1(a int not null, b int) engine = innodb;
+insert into t1 values (1,1),(1,1),(1,1),(1,1);
+alter table t1 add unique index (a);
+ERROR 23000: Duplicate entry '1' for key 'a'
+alter table t1 add unique index (b);
+ERROR 23000: Duplicate entry '1' for key 'b'
+alter table t1 add unique index (a), add unique index(b);
+ERROR 23000: Duplicate entry '1' for key 'a'
+show create table t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `a` int(11) NOT NULL,
+  `b` int(11) DEFAULT NULL
+) ENGINE=InnoDB DEFAULT CHARSET=latin1
+drop table t1;
+create table t1(a int not null, c int not null,b int, primary key(a), unique key(c), key(b)) engine = innodb;
+alter table t1 drop index c, drop index b;
+show create table t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `a` int(11) NOT NULL,
+  `c` int(11) NOT NULL,
+  `b` int(11) DEFAULT NULL,
+  PRIMARY KEY (`a`)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1
+drop table t1;
+create table t1(a int not null, b int, primary key(a)) engine = innodb;
+alter table t1 add index (b);
+show create table t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `a` int(11) NOT NULL,
+  `b` int(11) DEFAULT NULL,
+  PRIMARY KEY (`a`),
+  KEY `b` (`b`)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1
+drop table t1;
+create table t1(a int not null, b int, c char(10), d varchar(20), primary key (a)) engine = innodb;
+insert into t1 values (1,1,'ab','ab'),(2,2,'ac','ac'),(3,3,'ac','ac'),(4,4,'afe','afe'),(5,4,'affe','affe');
+alter table t1 add unique index (b), add unique index (c), add unique index (d);
+ERROR 23000: Duplicate entry '4' for key 'b'
+alter table t1 add unique index (c), add unique index (b), add index (d);
+ERROR 23000: Duplicate entry 'ac' for key 'c'
+show create table t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `a` int(11) NOT NULL,
+  `b` int(11) DEFAULT NULL,
+  `c` char(10) DEFAULT NULL,
+  `d` varchar(20) DEFAULT NULL,
+  PRIMARY KEY (`a`)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1
+drop table t1;
+create table t1(a int not null, b int not null, c int, primary key (a), key(c)) engine=innodb;
+insert into t1 values (5,1,5),(4,2,4),(3,3,3),(2,4,2),(1,5,1);
+alter table t1 add unique index (b);
+insert into t1 values (10,20,20),(11,19,19),(12,18,18),(13,17,17);
+show create table t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `a` int(11) NOT NULL,
+  `b` int(11) NOT NULL,
+  `c` int(11) DEFAULT NULL,
+  PRIMARY KEY (`a`),
+  UNIQUE KEY `b` (`b`),
+  KEY `c` (`c`)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1
+check table t1;
+Table	Op	Msg_type	Msg_text
+test.t1	check	status	OK
+explain select * from t1 force index(c) order by c;
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	index	NULL	c	5	NULL	9	
+explain select * from t1 order by a;
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	index	NULL	PRIMARY	4	NULL	9	
+explain select * from t1 force index(b) order by b;
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	index	NULL	b	4	NULL	9	
+select * from t1 order by a;
+a	b	c
+1	5	1
+2	4	2
+3	3	3
+4	2	4
+5	1	5
+10	20	20
+11	19	19
+12	18	18
+13	17	17
+select * from t1 force index(b) order by b;
+a	b	c
+5	1	5
+4	2	4
+3	3	3
+2	4	2
+1	5	1
+13	17	17
+12	18	18
+11	19	19
+10	20	20
+select * from t1 force index(c) order by c;
+a	b	c
+1	5	1
+2	4	2
+3	3	3
+4	2	4
+5	1	5
+13	17	17
+12	18	18
+11	19	19
+10	20	20
+drop table t1;
+create table t1(a int not null, b int not null) engine=innodb;
+insert into t1 values (1,1);
+alter table t1 add primary key(b);
+insert into t1 values (2,2);
+show create table t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `a` int(11) NOT NULL,
+  `b` int(11) NOT NULL,
+  PRIMARY KEY (`b`)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1
+check table t1;
+Table	Op	Msg_type	Msg_text
+test.t1	check	status	OK
+select * from t1;
+a	b
+1	1
+2	2
+explain select * from t1;
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	ALL	NULL	NULL	NULL	NULL	2	
+explain select * from t1 order by a;
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	ALL	NULL	NULL	NULL	NULL	2	Using filesort
+explain select * from t1 order by b;
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	index	NULL	PRIMARY	4	NULL	2	
+checksum table t1;
+Table	Checksum
+test.t1	582702641
+drop table t1;
+create table t1(a int not null) engine=innodb;
+insert into t1 values (1);
+alter table t1 add primary key(a);
+insert into t1 values (2);
+show create table t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `a` int(11) NOT NULL,
+  PRIMARY KEY (`a`)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1
+check table t1;
+Table	Op	Msg_type	Msg_text
+test.t1	check	status	OK
+commit;
+select * from t1;
+a
+1
+2
+explain select * from t1;
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	index	NULL	PRIMARY	4	NULL	2	Using index
+explain select * from t1 order by a;
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	index	NULL	PRIMARY	4	NULL	2	Using index
+drop table t1;
+create table t2(d varchar(17) primary key) engine=innodb default charset=utf8;
+create table t3(a int primary key) engine=innodb;
+insert into t3 values(22),(44),(33),(55),(66);
+insert into t2 values ('jejdkrun87'),('adfd72nh9k'),
+('adfdpplkeock'),('adfdijnmnb78k'),('adfdijn0loKNHJik');
+create table t1(a int, b blob, c text, d text not null)
+engine=innodb default charset = utf8;
+insert into t1
+select a,left(repeat(d,100*a),65535),repeat(d,20*a),d from t2,t3;
+drop table t2, t3;
+select count(*) from t1 where a=44;
+count(*)
+5
+select a,
+length(b),b=left(repeat(d,100*a),65535),length(c),c=repeat(d,20*a),d from t1;
+a	length(b)	b=left(repeat(d,100*a),65535)	length(c)	c=repeat(d,20*a)	d
+22	22000	1	4400	1	adfd72nh9k
+22	35200	1	7040	1	adfdijn0loKNHJik
+22	28600	1	5720	1	adfdijnmnb78k
+22	26400	1	5280	1	adfdpplkeock
+22	22000	1	4400	1	jejdkrun87
+33	33000	1	6600	1	adfd72nh9k
+33	52800	1	10560	1	adfdijn0loKNHJik
+33	42900	1	8580	1	adfdijnmnb78k
+33	39600	1	7920	1	adfdpplkeock
+33	33000	1	6600	1	jejdkrun87
+44	44000	1	8800	1	adfd72nh9k
+44	65535	1	14080	1	adfdijn0loKNHJik
+44	57200	1	11440	1	adfdijnmnb78k
+44	52800	1	10560	1	adfdpplkeock
+44	44000	1	8800	1	jejdkrun87
+55	55000	1	11000	1	adfd72nh9k
+55	65535	1	17600	1	adfdijn0loKNHJik
+55	65535	1	14300	1	adfdijnmnb78k
+55	65535	1	13200	1	adfdpplkeock
+55	55000	1	11000	1	jejdkrun87
+66	65535	1	13200	1	adfd72nh9k
+66	65535	1	21120	1	adfdijn0loKNHJik
+66	65535	1	17160	1	adfdijnmnb78k
+66	65535	1	15840	1	adfdpplkeock
+66	65535	1	13200	1	jejdkrun87
+alter table t1 add primary key (a), add key (b(20));
+ERROR 23000: Duplicate entry '22' for key 'PRIMARY'
+delete from t1 where a%2;
+check table t1;
+Table	Op	Msg_type	Msg_text
+test.t1	check	status	OK
+alter table t1 add primary key (a,b(255),c(255)), add key (b(767));
+select count(*) from t1 where a=44;
+count(*)
+5
+select a,
+length(b),b=left(repeat(d,100*a),65535),length(c),c=repeat(d,20*a),d from t1;
+a	length(b)	b=left(repeat(d,100*a),65535)	length(c)	c=repeat(d,20*a)	d
+22	22000	1	4400	1	adfd72nh9k
+22	35200	1	7040	1	adfdijn0loKNHJik
+22	28600	1	5720	1	adfdijnmnb78k
+22	26400	1	5280	1	adfdpplkeock
+22	22000	1	4400	1	jejdkrun87
+44	44000	1	8800	1	adfd72nh9k
+44	65535	1	14080	1	adfdijn0loKNHJik
+44	57200	1	11440	1	adfdijnmnb78k
+44	52800	1	10560	1	adfdpplkeock
+44	44000	1	8800	1	jejdkrun87
+66	65535	1	13200	1	adfd72nh9k
+66	65535	1	21120	1	adfdijn0loKNHJik
+66	65535	1	17160	1	adfdijnmnb78k
+66	65535	1	15840	1	adfdpplkeock
+66	65535	1	13200	1	jejdkrun87
+show create table t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `a` int(11) NOT NULL DEFAULT '0',
+  `b` blob NOT NULL,
+  `c` text NOT NULL,
+  `d` text NOT NULL,
+  PRIMARY KEY (`a`,`b`(255),`c`(255)),
+  KEY `b` (`b`(767))
+) ENGINE=InnoDB DEFAULT CHARSET=utf8
+check table t1;
+Table	Op	Msg_type	Msg_text
+test.t1	check	status	OK
+explain select * from t1 where b like 'adfd%';
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	ALL	b	NULL	NULL	NULL	15	Using where
+create table t2(a int, b varchar(255), primary key(a,b)) engine=innodb;
+insert into t2 select a,left(b,255) from t1;
+drop table t1;
+rename table t2 to t1;
+set innodb_lock_wait_timeout=1;
+begin;
+select a from t1 limit 1 for update;
+a
+22
+set innodb_lock_wait_timeout=1;
+create index t1ba on t1 (b,a);
+ERROR HY000: Lock wait timeout exceeded; try restarting transaction
+commit;
+begin;
+select a from t1 limit 1 lock in share mode;
+a
+22
+create index t1ba on t1 (b,a);
+drop index t1ba on t1;
+ERROR HY000: Lock wait timeout exceeded; try restarting transaction
+commit;
+explain select a from t1 order by b;
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	index	NULL	t1ba	261	NULL	15	Using index
+select a,sleep(2+a/100) from t1 order by b limit 3;
+select sleep(1);
+sleep(1)
+0
+drop index t1ba on t1;
+a	sleep(2+a/100)
+22	0
+44	0
+66	0
+explain select a from t1 order by b;
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	index	NULL	PRIMARY	261	NULL	15	Using index; Using filesort
+select a from t1 order by b limit 3;
+a
+22
+66
+44
+commit;
+drop table t1;
+set global innodb_file_per_table=on;
+set global innodb_file_format='Barracuda';
+create table t1(a blob,b blob,c blob,d blob,e blob,f blob,g blob,h blob,
+i blob,j blob,k blob,l blob,m blob,n blob,o blob,p blob,
+q blob,r blob,s blob,t blob,u blob)
+engine=innodb row_format=dynamic;
+create index t1a on t1 (a(1));
+create index t1b on t1 (b(1));
+create index t1c on t1 (c(1));
+create index t1d on t1 (d(1));
+create index t1e on t1 (e(1));
+create index t1f on t1 (f(1));
+create index t1g on t1 (g(1));
+create index t1h on t1 (h(1));
+create index t1i on t1 (i(1));
+create index t1j on t1 (j(1));
+create index t1k on t1 (k(1));
+create index t1l on t1 (l(1));
+create index t1m on t1 (m(1));
+create index t1n on t1 (n(1));
+create index t1o on t1 (o(1));
+create index t1p on t1 (p(1));
+create index t1q on t1 (q(1));
+create index t1r on t1 (r(1));
+create index t1s on t1 (s(1));
+create index t1t on t1 (t(1));
+create index t1u on t1 (u(1));
+ERROR HY000: Too big row
+create index t1ut on t1 (u(1), t(1));
+ERROR HY000: Too big row
+create index t1st on t1 (s(1), t(1));
+show create table t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `a` blob,
+  `b` blob,
+  `c` blob,
+  `d` blob,
+  `e` blob,
+  `f` blob,
+  `g` blob,
+  `h` blob,
+  `i` blob,
+  `j` blob,
+  `k` blob,
+  `l` blob,
+  `m` blob,
+  `n` blob,
+  `o` blob,
+  `p` blob,
+  `q` blob,
+  `r` blob,
+  `s` blob,
+  `t` blob,
+  `u` blob,
+  KEY `t1a` (`a`(1)),
+  KEY `t1b` (`b`(1)),
+  KEY `t1c` (`c`(1)),
+  KEY `t1d` (`d`(1)),
+  KEY `t1e` (`e`(1)),
+  KEY `t1f` (`f`(1)),
+  KEY `t1g` (`g`(1)),
+  KEY `t1h` (`h`(1)),
+  KEY `t1i` (`i`(1)),
+  KEY `t1j` (`j`(1)),
+  KEY `t1k` (`k`(1)),
+  KEY `t1l` (`l`(1)),
+  KEY `t1m` (`m`(1)),
+  KEY `t1n` (`n`(1)),
+  KEY `t1o` (`o`(1)),
+  KEY `t1p` (`p`(1)),
+  KEY `t1q` (`q`(1)),
+  KEY `t1r` (`r`(1)),
+  KEY `t1s` (`s`(1)),
+  KEY `t1t` (`t`(1)),
+  KEY `t1st` (`s`(1),`t`(1))
+) ENGINE=InnoDB DEFAULT CHARSET=latin1 ROW_FORMAT=DYNAMIC
+create index t1u on t1 (u(1));
+ERROR HY000: Too big row
+alter table t1 row_format=compact;
+create index t1u on t1 (u(1));
+drop table t1;
+set global innodb_file_per_table=0;
+set global innodb_file_format=Antelope;
+SET @OLD_UNIQUE_CHECKS=@@UNIQUE_CHECKS, UNIQUE_CHECKS=0;
+SET @OLD_FOREIGN_KEY_CHECKS=@@FOREIGN_KEY_CHECKS, FOREIGN_KEY_CHECKS=0;
+CREATE TABLE t1(
+c1	BIGINT(12) NOT NULL,
+PRIMARY KEY (c1)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1;
+CREATE TABLE t2(
+c1	BIGINT(16) NOT NULL,
+c2	BIGINT(12) NOT NULL,
+c3	BIGINT(12) NOT NULL,
+PRIMARY KEY (c1)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1;
+ALTER TABLE t2 ADD CONSTRAINT fk_t2_ca
+FOREIGN KEY (c3) REFERENCES t1(c1);
+SET FOREIGN_KEY_CHECKS=@OLD_FOREIGN_KEY_CHECKS;
+SET UNIQUE_CHECKS=@OLD_UNIQUE_CHECKS;
+SHOW CREATE TABLE t2;
+Table	Create Table
+t2	CREATE TABLE `t2` (
+  `c1` bigint(16) NOT NULL,
+  `c2` bigint(12) NOT NULL,
+  `c3` bigint(12) NOT NULL,
+  PRIMARY KEY (`c1`),
+  KEY `fk_t2_ca` (`c3`),
+  CONSTRAINT `fk_t2_ca` FOREIGN KEY (`c3`) REFERENCES `t1` (`c1`)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1
+CREATE INDEX i_t2_c3_c2 ON t2(c3, c2);
+SHOW CREATE TABLE t2;
+Table	Create Table
+t2	CREATE TABLE `t2` (
+  `c1` bigint(16) NOT NULL,
+  `c2` bigint(12) NOT NULL,
+  `c3` bigint(12) NOT NULL,
+  PRIMARY KEY (`c1`),
+  KEY `i_t2_c3_c2` (`c3`,`c2`),
+  CONSTRAINT `fk_t2_ca` FOREIGN KEY (`c3`) REFERENCES `t1` (`c1`)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1
+SET FOREIGN_KEY_CHECKS=@OLD_FOREIGN_KEY_CHECKS;
+SET UNIQUE_CHECKS=@OLD_UNIQUE_CHECKS;
+INSERT INTO t2 VALUES(0,0,0);
+ERROR 23000: Cannot add or update a child row: a foreign key constraint fails (`test`.`t2`, CONSTRAINT `fk_t2_ca` FOREIGN KEY (`c3`) REFERENCES `t1` (`c1`))
+INSERT INTO t1 VALUES(0);
+INSERT INTO t2 VALUES(0,0,0);
+DROP TABLE t2;
+CREATE TABLE t2(
+c1	BIGINT(16) NOT NULL,
+c2	BIGINT(12) NOT NULL,
+c3	BIGINT(12) NOT NULL,
+PRIMARY KEY (c1,c2,c3)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1;
+ALTER TABLE t2 ADD CONSTRAINT fk_t2_ca
+FOREIGN KEY (c3) REFERENCES t1(c1);
+SHOW CREATE TABLE t2;
+Table	Create Table
+t2	CREATE TABLE `t2` (
+  `c1` bigint(16) NOT NULL,
+  `c2` bigint(12) NOT NULL,
+  `c3` bigint(12) NOT NULL,
+  PRIMARY KEY (`c1`,`c2`,`c3`),
+  KEY `fk_t2_ca` (`c3`),
+  CONSTRAINT `fk_t2_ca` FOREIGN KEY (`c3`) REFERENCES `t1` (`c1`)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1
+CREATE INDEX i_t2_c3_c2 ON t2(c3, c2);
+SHOW CREATE TABLE t2;
+Table	Create Table
+t2	CREATE TABLE `t2` (
+  `c1` bigint(16) NOT NULL,
+  `c2` bigint(12) NOT NULL,
+  `c3` bigint(12) NOT NULL,
+  PRIMARY KEY (`c1`,`c2`,`c3`),
+  KEY `i_t2_c3_c2` (`c3`,`c2`),
+  CONSTRAINT `fk_t2_ca` FOREIGN KEY (`c3`) REFERENCES `t1` (`c1`)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1
+INSERT INTO t2 VALUES(0,0,1);
+ERROR 23000: Cannot add or update a child row: a foreign key constraint fails (`test`.`t2`, CONSTRAINT `fk_t2_ca` FOREIGN KEY (`c3`) REFERENCES `t1` (`c1`))
+INSERT INTO t2 VALUES(0,0,0);
+DELETE FROM t1;
+ERROR 23000: Cannot delete or update a parent row: a foreign key constraint fails (`test`.`t2`, CONSTRAINT `fk_t2_ca` FOREIGN KEY (`c3`) REFERENCES `t1` (`c1`))
+DELETE FROM t2;
+DROP TABLE t2;
+DROP TABLE t1;
+CREATE TABLE t1(
+c1	BIGINT(12) NOT NULL,
+c2	INT(4) NOT NULL,
+PRIMARY KEY (c2,c1)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1;
+CREATE TABLE t2(
+c1	BIGINT(16) NOT NULL,
+c2	BIGINT(12) NOT NULL,
+c3	BIGINT(12) NOT NULL,
+PRIMARY KEY (c1)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1;
+ALTER TABLE t2 ADD CONSTRAINT fk_t2_ca
+FOREIGN KEY (c3,c2) REFERENCES t1(c1,c1);
+ERROR HY000: Can't create table '#sql-temporary' (errno: 150)
+ALTER TABLE t2 ADD CONSTRAINT fk_t2_ca
+FOREIGN KEY (c3,c2) REFERENCES t1(c1,c2);
+ERROR HY000: Can't create table '#sql-temporary' (errno: 150)
+ALTER TABLE t2 ADD CONSTRAINT fk_t2_ca
+FOREIGN KEY (c3,c2) REFERENCES t1(c2,c1);
+ERROR HY000: Can't create table '#sql-temporary' (errno: 150)
+ALTER TABLE t1 MODIFY COLUMN c2 BIGINT(12) NOT NULL;
+ALTER TABLE t2 ADD CONSTRAINT fk_t2_ca
+FOREIGN KEY (c3,c2) REFERENCES t1(c1,c2);
+ERROR HY000: Can't create table '#sql-temporary' (errno: 150)
+ALTER TABLE t2 ADD CONSTRAINT fk_t2_ca
+FOREIGN KEY (c3,c2) REFERENCES t1(c2,c1);
+SHOW CREATE TABLE t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `c1` bigint(12) NOT NULL,
+  `c2` bigint(12) NOT NULL,
+  PRIMARY KEY (`c2`,`c1`)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1
+SHOW CREATE TABLE t2;
+Table	Create Table
+t2	CREATE TABLE `t2` (
+  `c1` bigint(16) NOT NULL,
+  `c2` bigint(12) NOT NULL,
+  `c3` bigint(12) NOT NULL,
+  PRIMARY KEY (`c1`),
+  KEY `fk_t2_ca` (`c3`,`c2`),
+  CONSTRAINT `fk_t2_ca` FOREIGN KEY (`c3`, `c2`) REFERENCES `t1` (`c2`, `c1`)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1
+CREATE INDEX i_t2_c2_c1 ON t2(c2, c1);
+SHOW CREATE TABLE t2;
+Table	Create Table
+t2	CREATE TABLE `t2` (
+  `c1` bigint(16) NOT NULL,
+  `c2` bigint(12) NOT NULL,
+  `c3` bigint(12) NOT NULL,
+  PRIMARY KEY (`c1`),
+  KEY `fk_t2_ca` (`c3`,`c2`),
+  KEY `i_t2_c2_c1` (`c2`,`c1`),
+  CONSTRAINT `fk_t2_ca` FOREIGN KEY (`c3`, `c2`) REFERENCES `t1` (`c2`, `c1`)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1
+CREATE INDEX i_t2_c3_c1_c2 ON t2(c3, c1, c2);
+SHOW CREATE TABLE t2;
+Table	Create Table
+t2	CREATE TABLE `t2` (
+  `c1` bigint(16) NOT NULL,
+  `c2` bigint(12) NOT NULL,
+  `c3` bigint(12) NOT NULL,
+  PRIMARY KEY (`c1`),
+  KEY `fk_t2_ca` (`c3`,`c2`),
+  KEY `i_t2_c2_c1` (`c2`,`c1`),
+  KEY `i_t2_c3_c1_c2` (`c3`,`c1`,`c2`),
+  CONSTRAINT `fk_t2_ca` FOREIGN KEY (`c3`, `c2`) REFERENCES `t1` (`c2`, `c1`)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1
+CREATE INDEX i_t2_c3_c2 ON t2(c3, c2);
+SHOW CREATE TABLE t2;
+Table	Create Table
+t2	CREATE TABLE `t2` (
+  `c1` bigint(16) NOT NULL,
+  `c2` bigint(12) NOT NULL,
+  `c3` bigint(12) NOT NULL,
+  PRIMARY KEY (`c1`),
+  KEY `i_t2_c2_c1` (`c2`,`c1`),
+  KEY `i_t2_c3_c1_c2` (`c3`,`c1`,`c2`),
+  KEY `i_t2_c3_c2` (`c3`,`c2`),
+  CONSTRAINT `fk_t2_ca` FOREIGN KEY (`c3`, `c2`) REFERENCES `t1` (`c2`, `c1`)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1
+DROP TABLE t2;
+DROP TABLE t1;
+SET GLOBAL innodb_file_format_check=@save_innodb_file_format_check;
diff --git a/mysql-test/r/innodb-index_ucs2.result b/mysql-test/r/innodb-index_ucs2.result
new file mode 100644
index 00000000000..c8a1e8c7da1
--- /dev/null
+++ b/mysql-test/r/innodb-index_ucs2.result
@@ -0,0 +1,116 @@
+create table t1(a int not null, b int, c char(10), d varchar(20), primary key (a)) engine = innodb default charset=ucs2;
+insert into t1 values (1,1,'ab','ab'),(2,2,'ac','ac'),(3,2,'ad','ad'),(4,4,'afe','afe');
+commit;
+alter table t1 add unique index (b);
+ERROR 23000: Duplicate entry '2' for key 'b'
+insert into t1 values(8,9,'fff','fff');
+select * from t1;
+a	b	c	d
+1	1	ab	ab
+2	2	ac	ac
+3	2	ad	ad
+4	4	afe	afe
+8	9	fff	fff
+show create table t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `a` int(11) NOT NULL,
+  `b` int(11) DEFAULT NULL,
+  `c` char(10) DEFAULT NULL,
+  `d` varchar(20) DEFAULT NULL,
+  PRIMARY KEY (`a`)
+) ENGINE=InnoDB DEFAULT CHARSET=ucs2
+alter table t1 add index (b);
+insert into t1 values(10,10,'kkk','iii');
+select * from t1;
+a	b	c	d
+1	1	ab	ab
+2	2	ac	ac
+3	2	ad	ad
+4	4	afe	afe
+8	9	fff	fff
+10	10	kkk	iii
+select * from t1 force index(b) order by b;
+a	b	c	d
+1	1	ab	ab
+2	2	ac	ac
+3	2	ad	ad
+4	4	afe	afe
+8	9	fff	fff
+10	10	kkk	iii
+explain select * from t1 force index(b) order by b;
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	index	NULL	b	5	NULL	6	
+show create table t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `a` int(11) NOT NULL,
+  `b` int(11) DEFAULT NULL,
+  `c` char(10) DEFAULT NULL,
+  `d` varchar(20) DEFAULT NULL,
+  PRIMARY KEY (`a`),
+  KEY `b` (`b`)
+) ENGINE=InnoDB DEFAULT CHARSET=ucs2
+alter table t1 add unique index (c), add index (d);
+insert into t1 values(11,11,'aaa','mmm');
+select * from t1;
+a	b	c	d
+1	1	ab	ab
+2	2	ac	ac
+3	2	ad	ad
+4	4	afe	afe
+8	9	fff	fff
+10	10	kkk	iii
+11	11	aaa	mmm
+select * from t1 force index(b) order by b;
+a	b	c	d
+1	1	ab	ab
+2	2	ac	ac
+3	2	ad	ad
+4	4	afe	afe
+8	9	fff	fff
+10	10	kkk	iii
+11	11	aaa	mmm
+select * from t1 force index(c) order by c;
+a	b	c	d
+11	11	aaa	mmm
+1	1	ab	ab
+2	2	ac	ac
+3	2	ad	ad
+4	4	afe	afe
+8	9	fff	fff
+10	10	kkk	iii
+select * from t1 force index(d) order by d;
+a	b	c	d
+1	1	ab	ab
+2	2	ac	ac
+3	2	ad	ad
+4	4	afe	afe
+8	9	fff	fff
+10	10	kkk	iii
+11	11	aaa	mmm
+explain select * from t1 force index(b) order by b;
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	index	NULL	b	5	NULL	7	
+explain select * from t1 force index(c) order by c;
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	index	NULL	c	21	NULL	7	
+explain select * from t1 force index(d) order by d;
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
+1	SIMPLE	t1	index	NULL	d	43	NULL	7	
+show create table t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `a` int(11) NOT NULL,
+  `b` int(11) DEFAULT NULL,
+  `c` char(10) DEFAULT NULL,
+  `d` varchar(20) DEFAULT NULL,
+  PRIMARY KEY (`a`),
+  UNIQUE KEY `c` (`c`),
+  KEY `b` (`b`),
+  KEY `d` (`d`)
+) ENGINE=InnoDB DEFAULT CHARSET=ucs2
+check table t1;
+Table	Op	Msg_type	Msg_text
+test.t1	check	status	OK
+drop table t1;
diff --git a/mysql-test/r/innodb-timeout.result b/mysql-test/r/innodb-timeout.result
new file mode 100644
index 00000000000..be9a688cd72
--- /dev/null
+++ b/mysql-test/r/innodb-timeout.result
@@ -0,0 +1,38 @@
+set global innodb_lock_wait_timeout=42;
+select @@innodb_lock_wait_timeout;
+@@innodb_lock_wait_timeout
+42
+set innodb_lock_wait_timeout=1;
+select @@innodb_lock_wait_timeout;
+@@innodb_lock_wait_timeout
+1
+select @@innodb_lock_wait_timeout;
+@@innodb_lock_wait_timeout
+42
+set global innodb_lock_wait_timeout=347;
+select @@innodb_lock_wait_timeout;
+@@innodb_lock_wait_timeout
+42
+set innodb_lock_wait_timeout=1;
+select @@innodb_lock_wait_timeout;
+@@innodb_lock_wait_timeout
+1
+select @@innodb_lock_wait_timeout;
+@@innodb_lock_wait_timeout
+347
+create table t1(a int primary key)engine=innodb;
+begin;
+insert into t1 values(1),(2),(3);
+select * from t1 for update;
+commit;
+a
+1
+2
+3
+begin;
+insert into t1 values(4);
+select * from t1 for update;
+commit;
+ERROR HY000: Lock wait timeout exceeded; try restarting transaction
+drop table t1;
+set global innodb_lock_wait_timeout=50;
diff --git a/mysql-test/r/innodb-use-sys-malloc.result b/mysql-test/r/innodb-use-sys-malloc.result
new file mode 100644
index 00000000000..2ec4c7c8130
--- /dev/null
+++ b/mysql-test/r/innodb-use-sys-malloc.result
@@ -0,0 +1,48 @@
+SELECT @@GLOBAL.innodb_use_sys_malloc;
+@@GLOBAL.innodb_use_sys_malloc
+1
+1 Expected
+SET @@GLOBAL.innodb_use_sys_malloc=0;
+ERROR HY000: Variable 'innodb_use_sys_malloc' is a read only variable
+Expected error 'Read only variable'
+SELECT @@GLOBAL.innodb_use_sys_malloc;
+@@GLOBAL.innodb_use_sys_malloc
+1
+1 Expected
+drop table if exists t1;
+create table t1(a int not null) engine=innodb DEFAULT CHARSET=latin1;
+insert into t1 values (1),(2),(3),(4),(5),(6),(7);
+select * from t1;
+a
+1
+2
+3
+4
+5
+6
+7
+drop table t1;
+SELECT @@GLOBAL.innodb_use_sys_malloc;
+@@GLOBAL.innodb_use_sys_malloc
+1
+1 Expected
+SET @@GLOBAL.innodb_use_sys_malloc=0;
+ERROR HY000: Variable 'innodb_use_sys_malloc' is a read only variable
+Expected error 'Read only variable'
+SELECT @@GLOBAL.innodb_use_sys_malloc;
+@@GLOBAL.innodb_use_sys_malloc
+1
+1 Expected
+drop table if exists t1;
+create table t1(a int not null) engine=innodb DEFAULT CHARSET=latin1;
+insert into t1 values (1),(2),(3),(4),(5),(6),(7);
+select * from t1;
+a
+1
+2
+3
+4
+5
+6
+7
+drop table t1;
diff --git a/mysql-test/r/innodb-zip.result b/mysql-test/r/innodb-zip.result
new file mode 100644
index 00000000000..b26c4112826
--- /dev/null
+++ b/mysql-test/r/innodb-zip.result
@@ -0,0 +1,421 @@
+set global innodb_file_per_table=off;
+set global innodb_file_format=`0`;
+create table t0(a int primary key) engine=innodb row_format=compressed;
+Warnings:
+Warning	1478	InnoDB: ROW_FORMAT=COMPRESSED requires innodb_file_per_table.
+Warning	1478	InnoDB: assuming ROW_FORMAT=COMPACT.
+create table t00(a int primary key) engine=innodb
+key_block_size=4 row_format=compressed;
+Warnings:
+Warning	1478	InnoDB: KEY_BLOCK_SIZE requires innodb_file_per_table.
+Warning	1478	InnoDB: KEY_BLOCK_SIZE requires innodb_file_format > Antelope.
+Warning	1478	InnoDB: ignoring KEY_BLOCK_SIZE=4.
+Warning	1478	InnoDB: ROW_FORMAT=COMPRESSED requires innodb_file_per_table.
+Warning	1478	InnoDB: assuming ROW_FORMAT=COMPACT.
+create table t1(a int primary key) engine=innodb row_format=dynamic;
+Warnings:
+Warning	1478	InnoDB: ROW_FORMAT=DYNAMIC requires innodb_file_per_table.
+Warning	1478	InnoDB: assuming ROW_FORMAT=COMPACT.
+create table t2(a int primary key) engine=innodb row_format=redundant;
+create table t3(a int primary key) engine=innodb row_format=compact;
+create table t4(a int primary key) engine=innodb key_block_size=9;
+Warnings:
+Warning	1478	InnoDB: KEY_BLOCK_SIZE requires innodb_file_per_table.
+Warning	1478	InnoDB: KEY_BLOCK_SIZE requires innodb_file_format > Antelope.
+Warning	1478	InnoDB: ignoring KEY_BLOCK_SIZE=9.
+create table t5(a int primary key) engine=innodb
+key_block_size=1 row_format=redundant;
+Warnings:
+Warning	1478	InnoDB: KEY_BLOCK_SIZE requires innodb_file_per_table.
+Warning	1478	InnoDB: KEY_BLOCK_SIZE requires innodb_file_format > Antelope.
+Warning	1478	InnoDB: ignoring KEY_BLOCK_SIZE=1.
+set global innodb_file_per_table=on;
+create table t6(a int primary key) engine=innodb
+key_block_size=1 row_format=redundant;
+Warnings:
+Warning	1478	InnoDB: KEY_BLOCK_SIZE requires innodb_file_format > Antelope.
+Warning	1478	InnoDB: ignoring KEY_BLOCK_SIZE=1.
+set global innodb_file_format=`1`;
+create table t7(a int primary key) engine=innodb
+key_block_size=1 row_format=redundant;
+Warnings:
+Warning	1478	InnoDB: ignoring KEY_BLOCK_SIZE=1 unless ROW_FORMAT=COMPRESSED.
+create table t8(a int primary key) engine=innodb
+key_block_size=1 row_format=fixed;
+Warnings:
+Warning	1478	InnoDB: ignoring KEY_BLOCK_SIZE=1 unless ROW_FORMAT=COMPRESSED.
+Warning	1478	InnoDB: assuming ROW_FORMAT=COMPACT.
+create table t9(a int primary key) engine=innodb
+key_block_size=1 row_format=compact;
+Warnings:
+Warning	1478	InnoDB: ignoring KEY_BLOCK_SIZE=1 unless ROW_FORMAT=COMPRESSED.
+create table t10(a int primary key) engine=innodb
+key_block_size=1 row_format=dynamic;
+Warnings:
+Warning	1478	InnoDB: ignoring KEY_BLOCK_SIZE=1 unless ROW_FORMAT=COMPRESSED.
+create table t11(a int primary key) engine=innodb
+key_block_size=1 row_format=compressed;
+create table t12(a int primary key) engine=innodb
+key_block_size=1;
+create table t13(a int primary key) engine=innodb
+row_format=compressed;
+create table t14(a int primary key) engine=innodb key_block_size=9;
+Warnings:
+Warning	1478	InnoDB: ignoring KEY_BLOCK_SIZE=9.
+SELECT table_schema, table_name, row_format
+FROM information_schema.tables WHERE engine='innodb';
+table_schema	table_name	row_format
+test	t0	Compact
+test	t00	Compact
+test	t1	Compact
+test	t10	Dynamic
+test	t11	Compressed
+test	t12	Compressed
+test	t13	Compressed
+test	t14	Compact
+test	t2	Redundant
+test	t3	Compact
+test	t4	Compact
+test	t5	Redundant
+test	t6	Redundant
+test	t7	Redundant
+test	t8	Compact
+test	t9	Compact
+drop table t0,t00,t2,t3,t4,t5,t6,t7,t8,t9,t10,t11,t12,t13,t14;
+alter table t1 key_block_size=0;
+Warnings:
+Warning	1478	InnoDB: ignoring KEY_BLOCK_SIZE=0.
+alter table t1 row_format=dynamic;
+SELECT table_schema, table_name, row_format
+FROM information_schema.tables WHERE engine='innodb';
+table_schema	table_name	row_format
+test	t1	Dynamic
+alter table t1 row_format=compact;
+SELECT table_schema, table_name, row_format
+FROM information_schema.tables WHERE engine='innodb';
+table_schema	table_name	row_format
+test	t1	Compact
+alter table t1 row_format=redundant;
+SELECT table_schema, table_name, row_format
+FROM information_schema.tables WHERE engine='innodb';
+table_schema	table_name	row_format
+test	t1	Redundant
+drop table t1;
+create table t1(a int not null, b text, index(b(10))) engine=innodb
+key_block_size=1;
+create table t2(b text)engine=innodb;
+insert into t2 values(concat('1abcdefghijklmnopqrstuvwxyz', repeat('A',5000)));
+insert into t1 select 1, b from t2;
+commit;
+begin;
+update t1 set b=repeat('B',100);
+select a,left(b,40) from t1 natural join t2;
+a	left(b,40)
+1	1abcdefghijklmnopqrstuvwxyzAAAAAAAAAAAAA
+rollback;
+select a,left(b,40) from t1 natural join t2;
+a	left(b,40)
+1	1abcdefghijklmnopqrstuvwxyzAAAAAAAAAAAAA
+SELECT table_schema, table_name, row_format
+FROM information_schema.tables WHERE engine='innodb';
+table_schema	table_name	row_format
+test	t1	Compressed
+test	t2	Compact
+drop table t1,t2;
+SET SESSION innodb_strict_mode = off;
+CREATE TABLE t1(
+c TEXT NOT NULL, d TEXT NOT NULL,
+PRIMARY KEY (c(767),d(767)))
+ENGINE=InnoDB ROW_FORMAT=COMPRESSED KEY_BLOCK_SIZE=1 CHARSET=ASCII;
+ERROR 42000: Row size too large. The maximum row size for the used table type, not counting BLOBs, is 8126. You have to change some columns to TEXT or BLOBs
+CREATE TABLE t1(
+c TEXT NOT NULL, d TEXT NOT NULL,
+PRIMARY KEY (c(767),d(767)))
+ENGINE=InnoDB ROW_FORMAT=COMPRESSED KEY_BLOCK_SIZE=2 CHARSET=ASCII;
+ERROR 42000: Row size too large. The maximum row size for the used table type, not counting BLOBs, is 8126. You have to change some columns to TEXT or BLOBs
+CREATE TABLE t1(
+c TEXT NOT NULL, d TEXT NOT NULL,
+PRIMARY KEY (c(767),d(767)))
+ENGINE=InnoDB ROW_FORMAT=COMPRESSED KEY_BLOCK_SIZE=4 CHARSET=ASCII;
+drop table t1;
+CREATE TABLE t1(c TEXT, PRIMARY KEY (c(440)))
+ENGINE=InnoDB ROW_FORMAT=COMPRESSED KEY_BLOCK_SIZE=1 CHARSET=ASCII;
+ERROR 42000: Row size too large. The maximum row size for the used table type, not counting BLOBs, is 8126. You have to change some columns to TEXT or BLOBs
+CREATE TABLE t1(c TEXT, PRIMARY KEY (c(438)))
+ENGINE=InnoDB ROW_FORMAT=COMPRESSED KEY_BLOCK_SIZE=1 CHARSET=ASCII;
+INSERT INTO t1 VALUES(REPEAT('A',512)),(REPEAT('B',512));
+DROP TABLE t1;
+create table t1( c1 int not null, c2 blob, c3 blob, c4 blob,
+primary key(c1, c2(22), c3(22)))
+engine = innodb row_format = dynamic;
+begin;
+insert into t1 values(1, repeat('A', 20000), repeat('B', 20000),
+repeat('C', 20000));
+update t1 set c3 = repeat('D', 20000) where c1 = 1;
+commit;
+select count(*) from t1 where c2 = repeat('A', 20000);
+count(*)
+1
+select count(*) from t1 where c3 = repeat('D', 20000);
+count(*)
+1
+select count(*) from t1 where c4 = repeat('C', 20000);
+count(*)
+1
+update t1 set c3 = repeat('E', 20000) where c1 = 1;
+drop table t1;
+set global innodb_file_format=`0`;
+select @@innodb_file_format;
+@@innodb_file_format
+Antelope
+set global innodb_file_format=`1`;
+select @@innodb_file_format;
+@@innodb_file_format
+Barracuda
+set global innodb_file_format=`2`;
+ERROR HY000: Incorrect arguments to SET
+set global innodb_file_format=`-1`;
+ERROR HY000: Incorrect arguments to SET
+set global innodb_file_format=`Antelope`;
+set global innodb_file_format=`Barracuda`;
+set global innodb_file_format=`Cheetah`;
+ERROR HY000: Incorrect arguments to SET
+set global innodb_file_format=`abc`;
+ERROR HY000: Incorrect arguments to SET
+set global innodb_file_format=`1a`;
+ERROR HY000: Incorrect arguments to SET
+set global innodb_file_format=``;
+ERROR HY000: Incorrect arguments to SET
+set global innodb_file_per_table = on;
+set global innodb_file_format = `1`;
+set innodb_strict_mode = off;
+create table t1 (id int primary key) engine = innodb key_block_size = 0;
+Warnings:
+Warning	1478	InnoDB: ignoring KEY_BLOCK_SIZE=0.
+drop table t1;
+set innodb_strict_mode = on;
+create table t1 (id int primary key) engine = innodb key_block_size = 0;
+ERROR HY000: Can't create table 'test.t1' (errno: 1478)
+show errors;
+Level	Code	Message
+Error	1478	InnoDB: invalid KEY_BLOCK_SIZE = 0. Valid values are [1, 2, 4, 8, 16]
+Error	1005	Can't create table 'test.t1' (errno: 1478)
+create table t2 (id int primary key) engine = innodb key_block_size = 9;
+ERROR HY000: Can't create table 'test.t2' (errno: 1478)
+show errors;
+Level	Code	Message
+Error	1478	InnoDB: invalid KEY_BLOCK_SIZE = 9. Valid values are [1, 2, 4, 8, 16]
+Error	1005	Can't create table 'test.t2' (errno: 1478)
+create table t3 (id int primary key) engine = innodb key_block_size = 1;
+create table t4 (id int primary key) engine = innodb key_block_size = 2;
+create table t5 (id int primary key) engine = innodb key_block_size = 4;
+create table t6 (id int primary key) engine = innodb key_block_size = 8;
+create table t7 (id int primary key) engine = innodb key_block_size = 16;
+create table t8 (id int primary key) engine = innodb row_format = compressed;
+create table t9 (id int primary key) engine = innodb row_format = dynamic;
+create table t10(id int primary key) engine = innodb row_format = compact;
+create table t11(id int primary key) engine = innodb row_format = redundant;
+SELECT table_schema, table_name, row_format
+FROM information_schema.tables WHERE engine='innodb';
+table_schema	table_name	row_format
+test	t10	Compact
+test	t11	Redundant
+test	t3	Compressed
+test	t4	Compressed
+test	t5	Compressed
+test	t6	Compressed
+test	t7	Compressed
+test	t8	Compressed
+test	t9	Dynamic
+drop table t3, t4, t5, t6, t7, t8, t9, t10, t11;
+create table t1 (id int primary key) engine = innodb
+key_block_size = 8 row_format = compressed;
+create table t2 (id int primary key) engine = innodb
+key_block_size = 8 row_format = redundant;
+ERROR HY000: Can't create table 'test.t2' (errno: 1478)
+show errors;
+Level	Code	Message
+Error	1478	InnoDB: cannot specify ROW_FORMAT = REDUNDANT with KEY_BLOCK_SIZE.
+Error	1005	Can't create table 'test.t2' (errno: 1478)
+create table t3 (id int primary key) engine = innodb
+key_block_size = 8 row_format = compact;
+ERROR HY000: Can't create table 'test.t3' (errno: 1478)
+show errors;
+Level	Code	Message
+Error	1478	InnoDB: cannot specify ROW_FORMAT = COMPACT with KEY_BLOCK_SIZE.
+Error	1005	Can't create table 'test.t3' (errno: 1478)
+create table t4 (id int primary key) engine = innodb
+key_block_size = 8 row_format = dynamic;
+ERROR HY000: Can't create table 'test.t4' (errno: 1478)
+show errors;
+Level	Code	Message
+Error	1478	InnoDB: cannot specify ROW_FORMAT = DYNAMIC with KEY_BLOCK_SIZE.
+Error	1005	Can't create table 'test.t4' (errno: 1478)
+create table t5 (id int primary key) engine = innodb
+key_block_size = 8 row_format = default;
+ERROR HY000: Can't create table 'test.t5' (errno: 1478)
+show errors;
+Level	Code	Message
+Error	1478	InnoDB: cannot specify ROW_FORMAT = COMPACT with KEY_BLOCK_SIZE.
+Error	1005	Can't create table 'test.t5' (errno: 1478)
+SELECT table_schema, table_name, row_format
+FROM information_schema.tables WHERE engine='innodb';
+table_schema	table_name	row_format
+test	t1	Compressed
+drop table t1;
+create table t1 (id int primary key) engine = innodb
+key_block_size = 9 row_format = redundant;
+ERROR HY000: Can't create table 'test.t1' (errno: 1478)
+show errors;
+Level	Code	Message
+Error	1478	InnoDB: invalid KEY_BLOCK_SIZE = 9. Valid values are [1, 2, 4, 8, 16]
+Error	1478	InnoDB: cannot specify ROW_FORMAT = REDUNDANT with KEY_BLOCK_SIZE.
+Error	1005	Can't create table 'test.t1' (errno: 1478)
+create table t2 (id int primary key) engine = innodb
+key_block_size = 9 row_format = compact;
+ERROR HY000: Can't create table 'test.t2' (errno: 1478)
+show errors;
+Level	Code	Message
+Error	1478	InnoDB: invalid KEY_BLOCK_SIZE = 9. Valid values are [1, 2, 4, 8, 16]
+Error	1478	InnoDB: cannot specify ROW_FORMAT = COMPACT with KEY_BLOCK_SIZE.
+Error	1005	Can't create table 'test.t2' (errno: 1478)
+create table t2 (id int primary key) engine = innodb
+key_block_size = 9 row_format = dynamic;
+ERROR HY000: Can't create table 'test.t2' (errno: 1478)
+show errors;
+Level	Code	Message
+Error	1478	InnoDB: invalid KEY_BLOCK_SIZE = 9. Valid values are [1, 2, 4, 8, 16]
+Error	1478	InnoDB: cannot specify ROW_FORMAT = DYNAMIC with KEY_BLOCK_SIZE.
+Error	1005	Can't create table 'test.t2' (errno: 1478)
+SELECT table_schema, table_name, row_format
+FROM information_schema.tables WHERE engine='innodb';
+table_schema	table_name	row_format
+set global innodb_file_per_table = off;
+create table t1 (id int primary key) engine = innodb key_block_size = 1;
+ERROR HY000: Can't create table 'test.t1' (errno: 1478)
+show errors;
+Level	Code	Message
+Error	1478	InnoDB: KEY_BLOCK_SIZE requires innodb_file_per_table.
+Error	1005	Can't create table 'test.t1' (errno: 1478)
+create table t2 (id int primary key) engine = innodb key_block_size = 2;
+ERROR HY000: Can't create table 'test.t2' (errno: 1478)
+show errors;
+Level	Code	Message
+Error	1478	InnoDB: KEY_BLOCK_SIZE requires innodb_file_per_table.
+Error	1005	Can't create table 'test.t2' (errno: 1478)
+create table t3 (id int primary key) engine = innodb key_block_size = 4;
+ERROR HY000: Can't create table 'test.t3' (errno: 1478)
+show errors;
+Level	Code	Message
+Error	1478	InnoDB: KEY_BLOCK_SIZE requires innodb_file_per_table.
+Error	1005	Can't create table 'test.t3' (errno: 1478)
+create table t4 (id int primary key) engine = innodb key_block_size = 8;
+ERROR HY000: Can't create table 'test.t4' (errno: 1478)
+show errors;
+Level	Code	Message
+Error	1478	InnoDB: KEY_BLOCK_SIZE requires innodb_file_per_table.
+Error	1005	Can't create table 'test.t4' (errno: 1478)
+create table t5 (id int primary key) engine = innodb key_block_size = 16;
+ERROR HY000: Can't create table 'test.t5' (errno: 1478)
+show errors;
+Level	Code	Message
+Error	1478	InnoDB: KEY_BLOCK_SIZE requires innodb_file_per_table.
+Error	1005	Can't create table 'test.t5' (errno: 1478)
+create table t6 (id int primary key) engine = innodb row_format = compressed;
+ERROR HY000: Can't create table 'test.t6' (errno: 1478)
+show errors;
+Level	Code	Message
+Error	1478	InnoDB: ROW_FORMAT=COMPRESSED requires innodb_file_per_table.
+Error	1005	Can't create table 'test.t6' (errno: 1478)
+create table t7 (id int primary key) engine = innodb row_format = dynamic;
+ERROR HY000: Can't create table 'test.t7' (errno: 1478)
+show errors;
+Level	Code	Message
+Error	1478	InnoDB: ROW_FORMAT=DYNAMIC requires innodb_file_per_table.
+Error	1005	Can't create table 'test.t7' (errno: 1478)
+create table t8 (id int primary key) engine = innodb row_format = compact;
+create table t9 (id int primary key) engine = innodb row_format = redundant;
+SELECT table_schema, table_name, row_format
+FROM information_schema.tables WHERE engine='innodb';
+table_schema	table_name	row_format
+test	t8	Compact
+test	t9	Redundant
+drop table t8, t9;
+set global innodb_file_per_table = on;
+set global innodb_file_format = `0`;
+create table t1 (id int primary key) engine = innodb key_block_size = 1;
+ERROR HY000: Can't create table 'test.t1' (errno: 1478)
+show errors;
+Level	Code	Message
+Error	1478	InnoDB: KEY_BLOCK_SIZE requires innodb_file_format > Antelope.
+Error	1005	Can't create table 'test.t1' (errno: 1478)
+create table t2 (id int primary key) engine = innodb key_block_size = 2;
+ERROR HY000: Can't create table 'test.t2' (errno: 1478)
+show errors;
+Level	Code	Message
+Error	1478	InnoDB: KEY_BLOCK_SIZE requires innodb_file_format > Antelope.
+Error	1005	Can't create table 'test.t2' (errno: 1478)
+create table t3 (id int primary key) engine = innodb key_block_size = 4;
+ERROR HY000: Can't create table 'test.t3' (errno: 1478)
+show errors;
+Level	Code	Message
+Error	1478	InnoDB: KEY_BLOCK_SIZE requires innodb_file_format > Antelope.
+Error	1005	Can't create table 'test.t3' (errno: 1478)
+create table t4 (id int primary key) engine = innodb key_block_size = 8;
+ERROR HY000: Can't create table 'test.t4' (errno: 1478)
+show errors;
+Level	Code	Message
+Error	1478	InnoDB: KEY_BLOCK_SIZE requires innodb_file_format > Antelope.
+Error	1005	Can't create table 'test.t4' (errno: 1478)
+create table t5 (id int primary key) engine = innodb key_block_size = 16;
+ERROR HY000: Can't create table 'test.t5' (errno: 1478)
+show errors;
+Level	Code	Message
+Error	1478	InnoDB: KEY_BLOCK_SIZE requires innodb_file_format > Antelope.
+Error	1005	Can't create table 'test.t5' (errno: 1478)
+create table t6 (id int primary key) engine = innodb row_format = compressed;
+ERROR HY000: Can't create table 'test.t6' (errno: 1478)
+show errors;
+Level	Code	Message
+Error	1478	InnoDB: ROW_FORMAT=COMPRESSED requires innodb_file_format > Antelope.
+Error	1005	Can't create table 'test.t6' (errno: 1478)
+create table t7 (id int primary key) engine = innodb row_format = dynamic;
+ERROR HY000: Can't create table 'test.t7' (errno: 1478)
+show errors;
+Level	Code	Message
+Error	1478	InnoDB: ROW_FORMAT=DYNAMIC requires innodb_file_format > Antelope.
+Error	1005	Can't create table 'test.t7' (errno: 1478)
+create table t8 (id int primary key) engine = innodb row_format = compact;
+create table t9 (id int primary key) engine = innodb row_format = redundant;
+SELECT table_schema, table_name, row_format
+FROM information_schema.tables WHERE engine='innodb';
+table_schema	table_name	row_format
+test	t8	Compact
+test	t9	Redundant
+drop table t8, t9;
+set global innodb_file_per_table=0;
+set global innodb_file_format=Antelope;
+set global innodb_file_per_table=on;
+set global innodb_file_format=`Barracuda`;
+set global innodb_file_format_check=`Antelope`;
+create table normal_table (
+c1 int
+) engine = innodb;
+select @@innodb_file_format_check;
+@@innodb_file_format_check
+Antelope
+create table zip_table (
+c1 int
+) engine = innodb key_block_size = 8;
+select @@innodb_file_format_check;
+@@innodb_file_format_check
+Barracuda
+set global innodb_file_format_check=`Antelope`;
+select @@innodb_file_format_check;
+@@innodb_file_format_check
+Antelope
+show table status;
+select @@innodb_file_format_check;
+@@innodb_file_format_check
+Barracuda
+drop table normal_table, zip_table;
diff --git a/mysql-test/r/innodb.result b/mysql-test/r/innodb.result
index 1f46ade27e0..c6379d60ba4 100644
--- a/mysql-test/r/innodb.result
+++ b/mysql-test/r/innodb.result
@@ -1508,7 +1508,7 @@ t2	CREATE TABLE `t2` (
 ) ENGINE=InnoDB DEFAULT CHARSET=latin1
 drop index id2 on t2;
 drop index id on t2;
-Got one of the listed errors
+ERROR HY000: Cannot drop index 'id': needed in a foreign key constraint
 show create table t2;
 Table	Create Table
 t2	CREATE TABLE `t2` (
@@ -1738,7 +1738,7 @@ count(*)
 drop table t1;
 show status like "Innodb_buffer_pool_pages_total";
 Variable_name	Value
-Innodb_buffer_pool_pages_total	512
+Innodb_buffer_pool_pages_total	511
 show status like "Innodb_page_size";
 Variable_name	Value
 Innodb_page_size	16384
@@ -1784,7 +1784,7 @@ innodb_sync_spin_loops	20
 SET @old_innodb_thread_concurrency= @@global.innodb_thread_concurrency;
 show variables like "innodb_thread_concurrency";
 Variable_name	Value
-innodb_thread_concurrency	8
+innodb_thread_concurrency	0
 set global innodb_thread_concurrency=1001;
 Warnings:
 Warning	1292	Truncated incorrect thread_concurrency value: '1001'
@@ -1970,7 +1970,7 @@ explain select count(*) from t1 where v between 'a' and 'a ' and v between 'a  '
 id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
 1	SIMPLE	t1	ref	v	v	13	const	#	Using where; Using index
 alter table t1 add unique(v);
-ERROR 23000: Duplicate entry '{ ' for key 'v_2'
+ERROR 23000: Duplicate entry 'v' for key 'v_2'
 alter table t1 add key(v);
 select concat('*',v,'*',c,'*',t,'*') as qq from t1 where v='a';
 qq
@@ -2377,6 +2377,8 @@ t1	CREATE TABLE `t1` (
 ) ENGINE=InnoDB DEFAULT CHARSET=latin1
 drop table t1;
 create table t1 (v varchar(10), c char(10)) row_format=fixed;
+Warnings:
+Warning	1478	InnoDB: assuming ROW_FORMAT=COMPACT.
 show create table t1;
 Table	Create Table
 t1	CREATE TABLE `t1` (
@@ -3190,6 +3192,7 @@ t1	CREATE TABLE `t1` (
   CONSTRAINT `t1_t2` FOREIGN KEY (`id`) REFERENCES `t2` (`id`)
 ) ENGINE=InnoDB AUTO_INCREMENT=349 DEFAULT CHARSET=latin1
 DROP TABLE t1,t2;
+set innodb_strict_mode=on;
 CREATE TABLE t1 (
 c01 CHAR(255), c02 CHAR(255), c03 CHAR(255), c04 CHAR(255),
 c05 CHAR(255), c06 CHAR(255), c07 CHAR(255), c08 CHAR(255),
diff --git a/mysql-test/r/innodb_bug34300.result b/mysql-test/r/innodb_bug34300.result
index b2732e3047d..ae9fee81ad7 100644
--- a/mysql-test/r/innodb_bug34300.result
+++ b/mysql-test/r/innodb_bug34300.result
@@ -1,8 +1,4 @@
-SELECT f4, f8 FROM bug34300;
 f4	f8
 xxx	zzz
-ALTER TABLE bug34300 ADD COLUMN (f10 INT);
-SELECT f4, f8 FROM bug34300;
 f4	f8
 xxx	zzz
-DROP TABLE bug34300;
diff --git a/mysql-test/r/innodb_bug36169.result b/mysql-test/r/innodb_bug36169.result
new file mode 100644
index 00000000000..0c0cea08955
--- /dev/null
+++ b/mysql-test/r/innodb_bug36169.result
@@ -0,0 +1,5 @@
+SET @save_innodb_file_format=@@global.innodb_file_format;
+SET @save_innodb_file_format_check=@@global.innodb_file_format_check;
+SET @save_innodb_file_per_table=@@global.innodb_file_per_table;
+SET GLOBAL innodb_file_format='Barracuda';
+SET GLOBAL innodb_file_per_table=ON;
diff --git a/mysql-test/r/innodb_bug36172.result b/mysql-test/r/innodb_bug36172.result
new file mode 100644
index 00000000000..195775f74c8
--- /dev/null
+++ b/mysql-test/r/innodb_bug36172.result
@@ -0,0 +1 @@
+SET storage_engine=InnoDB;
diff --git a/mysql-test/r/innodb_bug40360.result b/mysql-test/r/innodb_bug40360.result
new file mode 100644
index 00000000000..ef4cf463903
--- /dev/null
+++ b/mysql-test/r/innodb_bug40360.result
@@ -0,0 +1,4 @@
+SET TX_ISOLATION='READ-COMMITTED';
+CREATE TABLE bug40360 (a INT) engine=innodb;
+INSERT INTO bug40360 VALUES (1);
+DROP TABLE bug40360;
diff --git a/mysql-test/r/innodb_bug41904.result b/mysql-test/r/innodb_bug41904.result
new file mode 100644
index 00000000000..6070d32d181
--- /dev/null
+++ b/mysql-test/r/innodb_bug41904.result
@@ -0,0 +1,4 @@
+CREATE TABLE bug41904 (id INT PRIMARY KEY, uniquecol CHAR(15)) ENGINE=InnoDB;
+INSERT INTO bug41904 VALUES (1,NULL), (2,NULL);
+CREATE UNIQUE INDEX ui ON bug41904 (uniquecol);
+DROP TABLE bug41904;
diff --git a/mysql-test/r/innodb_information_schema.result b/mysql-test/r/innodb_information_schema.result
new file mode 100644
index 00000000000..396cae579ce
--- /dev/null
+++ b/mysql-test/r/innodb_information_schema.result
@@ -0,0 +1,23 @@
+lock_mode	lock_type	lock_table	lock_index	lock_rec	lock_data
+X	RECORD	`test`.```t'\"_str`	`PRIMARY`	2	'1', 'abc', '''abc', 'abc''', 'a''bc', 'a''bc''', '''abc'''''
+X	RECORD	`test`.```t'\"_str`	`PRIMARY`	2	'1', 'abc', '''abc', 'abc''', 'a''bc', 'a''bc''', '''abc'''''
+X	RECORD	`test`.```t'\"_str`	`PRIMARY`	3	'2', 'abc', '"abc', 'abc"', 'a"bc', 'a"bc"', '"abc""'
+X	RECORD	`test`.```t'\"_str`	`PRIMARY`	3	'2', 'abc', '"abc', 'abc"', 'a"bc', 'a"bc"', '"abc""'
+X	RECORD	`test`.```t'\"_str`	`PRIMARY`	4	'3', 'abc', '\\abc', 'abc\\', 'a\\bc', 'a\\bc\\', '\\abc\\\\'
+X	RECORD	`test`.```t'\"_str`	`PRIMARY`	4	'3', 'abc', '\\abc', 'abc\\', 'a\\bc', 'a\\bc\\', '\\abc\\\\'
+X	RECORD	`test`.```t'\"_str`	`PRIMARY`	5	'4', 'abc', '\0abc', 'abc\0', 'a\0bc', 'a\0bc\0', 'a\0bc\0\0'
+X	RECORD	`test`.```t'\"_str`	`PRIMARY`	5	'4', 'abc', '\0abc', 'abc\0', 'a\0bc', 'a\0bc\0', 'a\0bc\0\0'
+X	RECORD	`test`.`t_min`	`PRIMARY`	2	-128, 0, -32768, 0, -8388608, 0, -2147483648, 0, -9223372036854775808, 0
+X	RECORD	`test`.`t_min`	`PRIMARY`	2	-128, 0, -32768, 0, -8388608, 0, -2147483648, 0, -9223372036854775808, 0
+X	RECORD	`test`.`t_max`	`PRIMARY`	2	127, 255, 32767, 65535, 8388607, 16777215, 2147483647, 4294967295, 9223372036854775807, 18446744073709551615
+X	RECORD	`test`.`t_max`	`PRIMARY`	2	127, 255, 32767, 65535, 8388607, 16777215, 2147483647, 4294967295, 9223372036854775807, 18446744073709551615
+X	RECORD	`test`.```t'\"_str`	`PRIMARY`	1	supremum pseudo-record
+X	RECORD	`test`.```t'\"_str`	`PRIMARY`	1	supremum pseudo-record
+lock_table	COUNT(*)
+`test`.`t_max`	2
+`test`.`t_min`	2
+`test`.```t'\"_str`	10
+lock_table	COUNT(*)
+"test"."t_max"	2
+"test"."t_min"	2
+"test"."`t'\""_str"	10
diff --git a/mysql-test/r/innodb_xtradb_bug317074.result b/mysql-test/r/innodb_xtradb_bug317074.result
new file mode 100644
index 00000000000..0c0cea08955
--- /dev/null
+++ b/mysql-test/r/innodb_xtradb_bug317074.result
@@ -0,0 +1,5 @@
+SET @save_innodb_file_format=@@global.innodb_file_format;
+SET @save_innodb_file_format_check=@@global.innodb_file_format_check;
+SET @save_innodb_file_per_table=@@global.innodb_file_per_table;
+SET GLOBAL innodb_file_format='Barracuda';
+SET GLOBAL innodb_file_per_table=ON;
diff --git a/mysql-test/r/mysqlbinlog_row_big.result b/mysql-test/r/mysqlbinlog_row_big.result
index f2ca72f9936..05488be6b08 100644
--- a/mysql-test/r/mysqlbinlog_row_big.result
+++ b/mysql-test/r/mysqlbinlog_row_big.result
@@ -1,4 +1,9 @@
 #
+# We need big packets.
+#
+SET @old_global_max_allowed_packet=@@global.max_allowed_packet;
+SET @@global.max_allowed_packet= 1024*1024*1024;
+#
 # Preparatory cleanup.
 #
 DROP TABLE IF EXISTS t1;
@@ -7,10 +12,6 @@ DROP TABLE IF EXISTS t1;
 #
 SET timestamp=1000000000;
 #
-# We need big packets.
-#
-SET @@global.max_allowed_packet= 1024*1024*1024;
-#
 # Delete all existing binary logs.
 #
 RESET MASTER;
@@ -71,4 +72,5 @@ FLUSH LOGS;
 # Cleanup.
 #
 DROP TABLE t1;
+SET @@global.max_allowed_packet=@old_global_max_allowed_packet;
 remove_file $MYSQLTEST_VARDIR/tmp/mysqlbinlog_big_1.out
diff --git a/mysql-test/r/row-checksum-old.result b/mysql-test/r/row-checksum-old.result
index 3cf5a7104b9..ef523463860 100644
--- a/mysql-test/r/row-checksum-old.result
+++ b/mysql-test/r/row-checksum-old.result
@@ -72,6 +72,8 @@ Table	Checksum
 test.t1	4108368782
 drop table if exists t1;
 create table t1 (a int null, v varchar(100)) engine=innodb checksum=0 row_format=fixed;
+Warnings:
+Warning	1478	InnoDB: assuming ROW_FORMAT=COMPACT.
 insert into t1 values(null, null), (1, "hello");
 checksum table t1;
 Table	Checksum
diff --git a/mysql-test/r/row-checksum.result b/mysql-test/r/row-checksum.result
index 31ae094859b..fb8a1260a1d 100644
--- a/mysql-test/r/row-checksum.result
+++ b/mysql-test/r/row-checksum.result
@@ -72,6 +72,8 @@ Table	Checksum
 test.t1	3885665021
 drop table if exists t1;
 create table t1 (a int null, v varchar(100)) engine=innodb checksum=0 row_format=fixed;
+Warnings:
+Warning	1478	InnoDB: assuming ROW_FORMAT=COMPACT.
 insert into t1 values(null, null), (1, "hello");
 checksum table t1;
 Table	Checksum
diff --git a/mysql-test/r/variables-big.result b/mysql-test/r/variables-big.result
index 35882c7e284..2c178999a2c 100644
--- a/mysql-test/r/variables-big.result
+++ b/mysql-test/r/variables-big.result
@@ -1,20 +1,20 @@
 set session transaction_prealloc_size=1024*1024*1024*1;
-show processlist;
-Id	User	Host	db	Command	Time	State	Info
-#	root	localhost	test	Query	0	NULL	show processlist
+select @pid_temp = (select ID from information_schema.processlist) as 'TRUE';
+TRUE
+1
 set session transaction_prealloc_size=1024*1024*1024*2;
-show processlist;
-Id	User	Host	db	Command	Time	State	Info
-#	root	localhost	test	Query	0	NULL	show processlist
+select @pid_temp = (select ID from information_schema.processlist) as 'TRUE';
+TRUE
+1
 set session transaction_prealloc_size=1024*1024*1024*3;
-show processlist;
-Id	User	Host	db	Command	Time	State	Info
-#	root	localhost	test	Query	0	NULL	show processlist
+select @pid_temp = (select ID from information_schema.processlist) as 'TRUE';
+TRUE
+1
 set session transaction_prealloc_size=1024*1024*1024*4;
-show processlist;
-Id	User	Host	db	Command	Time	State	Info
-#	root	localhost	test	Query	0	NULL	show processlist
+select @pid_temp = (select ID from information_schema.processlist) as 'TRUE';
+TRUE
+1
 set session transaction_prealloc_size=1024*1024*1024*5;
-show processlist;
-Id	User	Host	db	Command	Time	State	Info
-#	root	localhost	test	Query	0	NULL	show processlist
+select @pid_temp = (select ID from information_schema.processlist) as 'TRUE';
+TRUE
+1
diff --git a/mysql-test/t/information_schema.test b/mysql-test/t/information_schema.test
index 5be5d9e3893..d2db589ad91 100644
--- a/mysql-test/t/information_schema.test
+++ b/mysql-test/t/information_schema.test
@@ -47,7 +47,7 @@ create view v1 (c) as
         table_name<>'ndb_binlog_index' AND
         table_name<>'ndb_apply_status' AND
 	NOT (table_schema = 'INFORMATION_SCHEMA' AND table_name LIKE 'PBXT_%');
-select * from v1;
+select * from v1 ORDER BY c COLLATE utf8_bin;
 
 select c,table_name from v1 
 inner join information_schema.TABLES v2 on (v1.c=v2.table_name)
diff --git a/mysql-test/t/innodb-analyze.test b/mysql-test/t/innodb-analyze.test
new file mode 100644
index 00000000000..88eb42991b0
--- /dev/null
+++ b/mysql-test/t/innodb-analyze.test
@@ -0,0 +1,65 @@
+#
+# Test that mysqld does not crash when running ANALYZE TABLE with
+# different values of the parameter innodb_stats_sample_pages.
+#
+
+-- source include/have_innodb.inc
+
+# we care only that the following SQL commands do not produce errors
+# and do not crash the server
+-- disable_query_log
+-- disable_result_log
+-- enable_warnings
+
+SET @save_innodb_stats_sample_pages=@@innodb_stats_sample_pages;
+SET GLOBAL innodb_stats_sample_pages=0;
+
+# check that the value has been adjusted to 1
+-- enable_result_log
+SHOW VARIABLES LIKE 'innodb_stats_sample_pages';
+-- disable_result_log
+
+CREATE TABLE innodb_analyze (
+	a INT,
+	b INT,
+	KEY(a),
+	KEY(b,a)
+) ENGINE=InnoDB;
+
+# test with empty table
+
+ANALYZE TABLE innodb_analyze;
+
+SET GLOBAL innodb_stats_sample_pages=2;
+ANALYZE TABLE innodb_analyze;
+
+SET GLOBAL innodb_stats_sample_pages=4;
+ANALYZE TABLE innodb_analyze;
+
+SET GLOBAL innodb_stats_sample_pages=8;
+ANALYZE TABLE innodb_analyze;
+
+SET GLOBAL innodb_stats_sample_pages=16;
+ANALYZE TABLE innodb_analyze;
+
+INSERT INTO innodb_analyze VALUES
+(1,1), (1,1), (1,2), (1,3), (1,4), (1,5),
+(8,1), (8,8), (8,2), (7,1), (1,4), (3,5);
+
+SET GLOBAL innodb_stats_sample_pages=1;
+ANALYZE TABLE innodb_analyze;
+
+SET GLOBAL innodb_stats_sample_pages=2;
+ANALYZE TABLE innodb_analyze;
+
+SET GLOBAL innodb_stats_sample_pages=4;
+ANALYZE TABLE innodb_analyze;
+
+SET GLOBAL innodb_stats_sample_pages=8;
+ANALYZE TABLE innodb_analyze;
+
+SET GLOBAL innodb_stats_sample_pages=16;
+ANALYZE TABLE innodb_analyze;
+
+SET GLOBAL innodb_stats_sample_pages=@save_innodb_stats_sample_pages;
+DROP TABLE innodb_analyze;
diff --git a/mysql-test/t/innodb-autoinc.test b/mysql-test/t/innodb-autoinc.test
index af23a900ee0..87ad470949d 100644
--- a/mysql-test/t/innodb-autoinc.test
+++ b/mysql-test/t/innodb-autoinc.test
@@ -396,7 +396,7 @@ DROP TABLE t1;
 #
 SET @@SESSION.AUTO_INCREMENT_INCREMENT=1, @@SESSION.AUTO_INCREMENT_OFFSET=1;
 SET @@INSERT_ID=1;
-SHOW VARIABLES LIKE "%auto_inc%";
+SHOW VARIABLES LIKE "auto_inc%";
 CREATE TABLE t1 (c1 DOUBLE NOT NULL AUTO_INCREMENT, c2 INT, PRIMARY KEY (c1)) ENGINE=InnoDB;
 INSERT INTO t1 VALUES(NULL, 1);
 INSERT INTO t1 VALUES(NULL, 2);
diff --git a/mysql-test/t/innodb-index.test b/mysql-test/t/innodb-index.test
new file mode 100644
index 00000000000..f81d70f5dd6
--- /dev/null
+++ b/mysql-test/t/innodb-index.test
@@ -0,0 +1,516 @@
+-- source include/have_innodb.inc
+
+SET @save_innodb_file_format_check=@@global.innodb_file_format_check;
+
+create table t1(a int not null, b int, c char(10) not null, d varchar(20)) engine = innodb;
+insert into t1 values (5,5,'oo','oo'),(4,4,'tr','tr'),(3,4,'ad','ad'),(2,3,'ak','ak');
+commit;
+--error ER_DUP_KEYNAME
+alter table t1 add index b (b), add index b (b);
+--error ER_DUP_FIELDNAME
+alter table t1 add index (b,b);
+alter table t1 add index d2 (d);
+show create table t1;
+explain select * from t1 force index(d2) order by d;
+select * from t1 force index (d2) order by d;
+--error ER_DUP_ENTRY
+alter table t1 add unique index (b);
+show create table t1;
+alter table t1 add index (b);
+show create table t1;
+
+# Check how existing tables interfere with temporary tables.
+CREATE TABLE `t1#1`(a INT PRIMARY KEY) ENGINE=InnoDB;
+
+call mtr.add_suppression(" table `test`\\.`t1#[12]` already exists in InnoDB internal");
+
+--error 156
+alter table t1 add unique index (c), add index (d);
+rename table `t1#1` to `t1#2`;
+--error 156
+alter table t1 add unique index (c), add index (d);
+drop table `t1#2`;
+
+alter table t1 add unique index (c), add index (d);
+show create table t1;
+explain select * from t1 force index(c) order by c;
+alter table t1 add primary key (a), drop index c;
+show create table t1;
+--error ER_MULTIPLE_PRI_KEY
+alter table t1 add primary key (c);
+--error ER_DUP_ENTRY
+alter table t1 drop primary key, add primary key (b);
+create unique index c on t1 (c);
+show create table t1;
+explain select * from t1 force index(c) order by c;
+select * from t1 force index(c) order by c;
+alter table t1 drop index b, add index (b);
+show create table t1;
+insert into t1 values(6,1,'ggg','ggg');
+select * from t1;
+select * from t1 force index(b) order by b;
+select * from t1 force index(c) order by c;
+select * from t1 force index(d) order by d;
+explain select * from t1 force index(b) order by b;
+explain select * from t1 force index(c) order by c;
+explain select * from t1 force index(d) order by d;
+show create table t1;
+drop table t1;
+
+create table t1(a int not null, b int, c char(10), d varchar(20), primary key (a)) engine = innodb;
+insert into t1 values (1,1,'ab','ab'),(2,2,'ac','ac'),(3,3,'ad','ad'),(4,4,'afe','afe');
+commit;
+alter table t1 add index (c(2));
+show create table t1;
+alter table t1 add unique index (d(10));
+show create table t1;
+insert into t1 values(5,1,'ggg','ggg');
+select * from t1;
+select * from t1 force index(c) order by c;
+select * from t1 force index(d) order by d;
+explain select * from t1 order by b;
+explain select * from t1 force index(c) order by c;
+explain select * from t1 force index(d) order by d;
+show create table t1;
+alter table t1 drop index d;
+insert into t1 values(8,9,'fff','fff');
+select * from t1;
+select * from t1 force index(c) order by c;
+explain select * from t1 order by b;
+explain select * from t1 force index(c) order by c;
+explain select * from t1 order by d;
+show create table t1;
+drop table t1;
+
+create table t1(a int not null, b int, c char(10), d varchar(20), primary key (a)) engine = innodb;
+insert into t1 values (1,1,'ab','ab'),(2,2,'ac','ac'),(3,2,'ad','ad'),(4,4,'afe','afe');
+commit;
+alter table t1 add unique index (b,c);
+insert into t1 values(8,9,'fff','fff');
+select * from t1;
+select * from t1 force index(b) order by b;
+explain select * from t1 force index(b) order by b;
+show create table t1;
+alter table t1 add index (b,c);
+insert into t1 values(11,11,'kkk','kkk');
+select * from t1;
+select * from t1 force index(b) order by b;
+explain select * from t1 force index(b) order by b;
+show create table t1;
+alter table t1 add unique index (c,d);
+insert into t1 values(13,13,'yyy','aaa');
+select * from t1;
+select * from t1 force index(b) order by b;
+select * from t1 force index(c) order by c;
+explain select * from t1 force index(b) order by b;
+explain select * from t1 force index(c) order by c;
+show create table t1;
+drop table t1;
+
+create table t1(a int not null, b int not null, c int, primary key (a), key (b)) engine = innodb;
+create table t3(a int not null, c int not null, d int, primary key (a), key (c)) engine = innodb;
+create table t4(a int not null, d int not null, e int, primary key (a), key (d)) engine = innodb;
+create table t2(a int not null, b int not null, c int not null, d int not null, e int,
+foreign key (b) references t1(b) on delete cascade,
+foreign key (c) references t3(c), foreign key (d) references t4(d))
+engine = innodb;
+--error ER_DROP_INDEX_FK
+alter table t1 drop index b;
+--error ER_DROP_INDEX_FK
+alter table t3 drop index c;
+--error ER_DROP_INDEX_FK
+alter table t4 drop index d;
+--error ER_DROP_INDEX_FK
+alter table t2 drop index b;
+--error ER_DROP_INDEX_FK
+alter table t2 drop index b, drop index c, drop index d;
+# Apparently, the following makes mysql_alter_table() drop index d.
+create unique index dc on t2 (d,c);
+create index dc on t1 (b,c);
+# This should preserve the foreign key constraints.
+alter table t2 add primary key (a);
+insert into t1 values (1,1,1);
+insert into t3 values (1,1,1);
+insert into t4 values (1,1,1);
+insert into t2 values (1,1,1,1,1);
+commit;
+alter table t4 add constraint dc foreign key (a) references t1(a);
+show create table t4;
+--replace_regex /'test\.#sql-[0-9a-f_]*'/'#sql-temporary'/
+# a foreign key 'test/dc' already exists
+--error ER_CANT_CREATE_TABLE
+alter table t3 add constraint dc foreign key (a) references t1(a);
+show create table t3;
+alter table t2 drop index b, add index (b);
+show create table t2;
+--error ER_ROW_IS_REFERENCED_2
+delete from t1;
+--error ER_CANT_DROP_FIELD_OR_KEY
+drop index dc on t4;
+# there is no foreign key dc on t3
+--replace_regex /'\.\/test\/#sql2-[0-9a-f-]*'/'#sql2-temporary'/
+--error ER_ERROR_ON_RENAME
+alter table t3 drop foreign key dc;
+alter table t4 drop foreign key dc;
+select * from t2;
+delete from t1;
+select * from t2;
+
+drop table t2,t4,t3,t1;
+
+-- let charset = utf8
+-- source include/innodb-index.inc
+
+create table t1(a int not null, b int) engine = innodb;
+insert into t1 values (1,1),(1,1),(1,1),(1,1);
+--error ER_DUP_ENTRY
+alter table t1 add unique index (a);
+--error ER_DUP_ENTRY
+alter table t1 add unique index (b);
+--error ER_DUP_ENTRY
+alter table t1 add unique index (a), add unique index(b);
+show create table t1;
+drop table t1;
+
+create table t1(a int not null, c int not null,b int, primary key(a), unique key(c), key(b)) engine = innodb;
+alter table t1 drop index c, drop index b;
+show create table t1;
+drop table t1;
+
+create table t1(a int not null, b int, primary key(a)) engine = innodb;
+alter table t1 add index (b);
+show create table t1;
+drop table t1;
+
+create table t1(a int not null, b int, c char(10), d varchar(20), primary key (a)) engine = innodb;
+insert into t1 values (1,1,'ab','ab'),(2,2,'ac','ac'),(3,3,'ac','ac'),(4,4,'afe','afe'),(5,4,'affe','affe');
+--error ER_DUP_ENTRY
+alter table t1 add unique index (b), add unique index (c), add unique index (d);
+--error ER_DUP_ENTRY
+alter table t1 add unique index (c), add unique index (b), add index (d);
+show create table t1;
+drop table t1;
+
+create table t1(a int not null, b int not null, c int, primary key (a), key(c)) engine=innodb;
+insert into t1 values (5,1,5),(4,2,4),(3,3,3),(2,4,2),(1,5,1);
+alter table t1 add unique index (b);
+insert into t1 values (10,20,20),(11,19,19),(12,18,18),(13,17,17);
+show create table t1;
+check table t1;
+explain select * from t1 force index(c) order by c;
+explain select * from t1 order by a;
+explain select * from t1 force index(b) order by b;
+select * from t1 order by a;
+select * from t1 force index(b) order by b;
+select * from t1 force index(c) order by c;
+drop table t1;
+
+create table t1(a int not null, b int not null) engine=innodb;
+insert into t1 values (1,1);
+alter table t1 add primary key(b);
+insert into t1 values (2,2);
+show create table t1;
+check table t1;
+select * from t1;
+explain select * from t1;
+explain select * from t1 order by a;
+explain select * from t1 order by b;
+checksum table t1;
+drop table t1;
+
+create table t1(a int not null) engine=innodb;
+insert into t1 values (1);
+alter table t1 add primary key(a);
+insert into t1 values (2);
+show create table t1;
+check table t1;
+commit;
+select * from t1;
+explain select * from t1;
+explain select * from t1 order by a;
+drop table t1;
+
+create table t2(d varchar(17) primary key) engine=innodb default charset=utf8;
+create table t3(a int primary key) engine=innodb;
+
+insert into t3 values(22),(44),(33),(55),(66);
+
+insert into t2 values ('jejdkrun87'),('adfd72nh9k'),
+('adfdpplkeock'),('adfdijnmnb78k'),('adfdijn0loKNHJik');
+
+create table t1(a int, b blob, c text, d text not null)
+engine=innodb default charset = utf8;
+
+# r2667 The following test is disabled because MySQL behavior changed.
+# r2667 The test was added with this comment:
+# r2667 
+# r2667 ------------------------------------------------------------------------
+# r2667 r1699 | marko | 2007-08-10 19:53:19 +0300 (Fri, 10 Aug 2007) | 5 lines
+# r2667 
+# r2667 branches/zip: Add changes that accidentally omitted from r1698:
+# r2667 
+# r2667 innodb-index.test, innodb-index.result: Add a test for creating
+# r2667 a PRIMARY KEY on a column that contains a NULL value.
+# r2667 ------------------------------------------------------------------------
+# r2667 
+# r2667 but in BZR-r2667:
+# r2667 http://bazaar.launchpad.net/~mysql/mysql-server/mysql-5.1/revision/davi%40mysql.com-20080617141221-8yre8ys9j4uw3xx5?start_revid=joerg%40mysql.com-20080630105418-7qoe5ehomgrcdb89
+# r2667 MySQL changed the behavior to do full table copy when creating PRIMARY INDEX
+# r2667 on a non-NULL column instead of calling ::add_index() which would fail (and
+# r2667 this is what we were testing here). Before r2667 the code execution path was
+# r2667 like this (when adding PRIMARY INDEX on a non-NULL column with ALTER TABLE):
+# r2667 
+# r2667 mysql_alter_table()
+# r2667   compare_tables()  // would return ALTER_TABLE_INDEX_CHANGED
+# r2667   ::add_index()  // would fail with "primary index cannot contain NULL"
+# r2667 
+# r2667 after r2667 the code execution path is the following:
+# r2667 
+# r2667 mysql_alter_table()
+# r2667   compare_tables()  // returns ALTER_TABLE_DATA_CHANGED
+# r2667   full copy is done, without calling ::add_index()
+# r2667 
+# r2667 To enable, remove "# r2667: " below.
+# r2667 
+# r2667: insert into t1 values (null,null,null,'null');
+insert into t1
+select a,left(repeat(d,100*a),65535),repeat(d,20*a),d from t2,t3;
+drop table t2, t3;
+select count(*) from t1 where a=44;
+select a,
+length(b),b=left(repeat(d,100*a),65535),length(c),c=repeat(d,20*a),d from t1;
+# r2667: --error ER_PRIMARY_CANT_HAVE_NULL
+# r2667: alter table t1 add primary key (a), add key (b(20));
+# r2667: delete from t1 where d='null';
+--error ER_DUP_ENTRY
+alter table t1 add primary key (a), add key (b(20));
+delete from t1 where a%2;
+check table t1;
+alter table t1 add primary key (a,b(255),c(255)), add key (b(767));
+select count(*) from t1 where a=44;
+select a,
+length(b),b=left(repeat(d,100*a),65535),length(c),c=repeat(d,20*a),d from t1;
+show create table t1;
+check table t1;
+explain select * from t1 where b like 'adfd%';
+
+#
+# Test locking
+#
+
+create table t2(a int, b varchar(255), primary key(a,b)) engine=innodb;
+insert into t2 select a,left(b,255) from t1;
+drop table t1;
+rename table t2 to t1;
+
+connect (a,localhost,root,,);
+connect (b,localhost,root,,);
+connection a;
+set innodb_lock_wait_timeout=1;
+begin;
+# Obtain an IX lock on the table
+select a from t1 limit 1 for update;
+connection b;
+set innodb_lock_wait_timeout=1;
+# This would require an S lock on the table, conflicting with the IX lock.
+--error ER_LOCK_WAIT_TIMEOUT
+create index t1ba on t1 (b,a);
+connection a;
+commit;
+begin;
+# Obtain an IS lock on the table
+select a from t1 limit 1 lock in share mode;
+connection b;
+# This will require an S lock on the table.  No conflict with the IS lock.
+create index t1ba on t1 (b,a);
+# This would require an X lock on the table, conflicting with the IS lock.
+--error ER_LOCK_WAIT_TIMEOUT
+drop index t1ba on t1;
+connection a;
+commit;
+explain select a from t1 order by b;
+--send
+select a,sleep(2+a/100) from t1 order by b limit 3;
+
+# The following DROP INDEX will succeed, altough the SELECT above has
+# opened a read view.  However, during the execution of the SELECT,
+# MySQL should hold a table lock that should block the execution
+# of the DROP INDEX below.
+
+connection b;
+select sleep(1);
+drop index t1ba on t1;
+
+# After the index was dropped, subsequent SELECTs will use the same
+# read view, but they should not be accessing the dropped index any more.
+
+connection a;
+reap;
+explain select a from t1 order by b;
+select a from t1 order by b limit 3;
+commit;
+
+connection default;
+disconnect a;
+disconnect b;
+
+drop table t1;
+
+let $per_table=`select @@innodb_file_per_table`;
+let $format=`select @@innodb_file_format`;
+set global innodb_file_per_table=on;
+set global innodb_file_format='Barracuda';
+# Test creating a table that could lead to undo log overflow.
+# In the undo log, we write a 768-byte prefix (REC_MAX_INDEX_COL_LEN)
+# of each externally stored column that appears as a column prefix in an index.
+# For this test case, it would suffice to write 1 byte, though.
+create table t1(a blob,b blob,c blob,d blob,e blob,f blob,g blob,h blob,
+		i blob,j blob,k blob,l blob,m blob,n blob,o blob,p blob,
+		q blob,r blob,s blob,t blob,u blob)
+	engine=innodb row_format=dynamic;
+create index t1a on t1 (a(1));
+create index t1b on t1 (b(1));
+create index t1c on t1 (c(1));
+create index t1d on t1 (d(1));
+create index t1e on t1 (e(1));
+create index t1f on t1 (f(1));
+create index t1g on t1 (g(1));
+create index t1h on t1 (h(1));
+create index t1i on t1 (i(1));
+create index t1j on t1 (j(1));
+create index t1k on t1 (k(1));
+create index t1l on t1 (l(1));
+create index t1m on t1 (m(1));
+create index t1n on t1 (n(1));
+create index t1o on t1 (o(1));
+create index t1p on t1 (p(1));
+create index t1q on t1 (q(1));
+create index t1r on t1 (r(1));
+create index t1s on t1 (s(1));
+create index t1t on t1 (t(1));
+--error 139
+create index t1u on t1 (u(1));
+--error 139
+create index t1ut on t1 (u(1), t(1));
+create index t1st on t1 (s(1), t(1));
+show create table t1;
+--error 139
+create index t1u on t1 (u(1));
+alter table t1 row_format=compact;
+create index t1u on t1 (u(1));
+
+drop table t1;
+eval set global innodb_file_per_table=$per_table;
+eval set global innodb_file_format=$format;
+
+#
+# Test to check whether CREATE INDEX handles implicit foreign key
+# constraint modifications (Issue #70, Bug #38786)
+#
+SET @OLD_UNIQUE_CHECKS=@@UNIQUE_CHECKS, UNIQUE_CHECKS=0;
+SET @OLD_FOREIGN_KEY_CHECKS=@@FOREIGN_KEY_CHECKS, FOREIGN_KEY_CHECKS=0;
+
+CREATE TABLE t1(
+  c1	BIGINT(12) NOT NULL,
+  PRIMARY KEY (c1)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1;
+
+CREATE TABLE t2(
+  c1	BIGINT(16) NOT NULL,
+  c2	BIGINT(12) NOT NULL,
+  c3	BIGINT(12) NOT NULL,
+  PRIMARY KEY (c1)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1;
+
+ALTER TABLE t2 ADD CONSTRAINT fk_t2_ca
+ FOREIGN KEY (c3) REFERENCES t1(c1);
+
+SET FOREIGN_KEY_CHECKS=@OLD_FOREIGN_KEY_CHECKS;
+SET UNIQUE_CHECKS=@OLD_UNIQUE_CHECKS;
+
+SHOW CREATE TABLE t2;
+
+CREATE INDEX i_t2_c3_c2 ON t2(c3, c2);
+
+SHOW CREATE TABLE t2;
+
+SET FOREIGN_KEY_CHECKS=@OLD_FOREIGN_KEY_CHECKS;
+SET UNIQUE_CHECKS=@OLD_UNIQUE_CHECKS;
+
+--error ER_NO_REFERENCED_ROW_2
+INSERT INTO t2 VALUES(0,0,0);
+INSERT INTO t1 VALUES(0);
+INSERT INTO t2 VALUES(0,0,0);
+
+DROP TABLE t2;
+
+CREATE TABLE t2(
+  c1	BIGINT(16) NOT NULL,
+  c2	BIGINT(12) NOT NULL,
+  c3	BIGINT(12) NOT NULL,
+  PRIMARY KEY (c1,c2,c3)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1;
+
+ALTER TABLE t2 ADD CONSTRAINT fk_t2_ca
+ FOREIGN KEY (c3) REFERENCES t1(c1);
+
+SHOW CREATE TABLE t2;
+
+CREATE INDEX i_t2_c3_c2 ON t2(c3, c2);
+
+SHOW CREATE TABLE t2;
+--error ER_NO_REFERENCED_ROW_2
+INSERT INTO t2 VALUES(0,0,1);
+INSERT INTO t2 VALUES(0,0,0);
+--error ER_ROW_IS_REFERENCED_2
+DELETE FROM t1;
+DELETE FROM t2;
+
+DROP TABLE t2;
+DROP TABLE t1;
+
+CREATE TABLE t1(
+  c1	BIGINT(12) NOT NULL,
+  c2	INT(4) NOT NULL,
+  PRIMARY KEY (c2,c1)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1;
+
+CREATE TABLE t2(
+  c1	BIGINT(16) NOT NULL,
+  c2	BIGINT(12) NOT NULL,
+  c3	BIGINT(12) NOT NULL,
+  PRIMARY KEY (c1)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1;
+
+--replace_regex /'test\.#sql-[0-9_a-f-]*'/'#sql-temporary'/
+--error ER_CANT_CREATE_TABLE
+ALTER TABLE t2 ADD CONSTRAINT fk_t2_ca
+ FOREIGN KEY (c3,c2) REFERENCES t1(c1,c1);
+--replace_regex /'test\.#sql-[0-9_a-f-]*'/'#sql-temporary'/
+--error ER_CANT_CREATE_TABLE
+ALTER TABLE t2 ADD CONSTRAINT fk_t2_ca
+ FOREIGN KEY (c3,c2) REFERENCES t1(c1,c2);
+--replace_regex /'test\.#sql-[0-9_a-f-]*'/'#sql-temporary'/
+--error ER_CANT_CREATE_TABLE
+ALTER TABLE t2 ADD CONSTRAINT fk_t2_ca
+ FOREIGN KEY (c3,c2) REFERENCES t1(c2,c1);
+ALTER TABLE t1 MODIFY COLUMN c2 BIGINT(12) NOT NULL;
+--replace_regex /'test\.#sql-[0-9_a-f-]*'/'#sql-temporary'/
+--error ER_CANT_CREATE_TABLE
+ALTER TABLE t2 ADD CONSTRAINT fk_t2_ca
+ FOREIGN KEY (c3,c2) REFERENCES t1(c1,c2);
+
+ALTER TABLE t2 ADD CONSTRAINT fk_t2_ca
+ FOREIGN KEY (c3,c2) REFERENCES t1(c2,c1);
+SHOW CREATE TABLE t1;
+SHOW CREATE TABLE t2;
+CREATE INDEX i_t2_c2_c1 ON t2(c2, c1);
+SHOW CREATE TABLE t2;
+CREATE INDEX i_t2_c3_c1_c2 ON t2(c3, c1, c2);
+SHOW CREATE TABLE t2;
+CREATE INDEX i_t2_c3_c2 ON t2(c3, c2);
+SHOW CREATE TABLE t2;
+
+DROP TABLE t2;
+DROP TABLE t1;
+SET GLOBAL innodb_file_format_check=@save_innodb_file_format_check;
diff --git a/mysql-test/t/innodb-index_ucs2.test b/mysql-test/t/innodb-index_ucs2.test
new file mode 100644
index 00000000000..fff9a4da1a8
--- /dev/null
+++ b/mysql-test/t/innodb-index_ucs2.test
@@ -0,0 +1,5 @@
+-- source include/have_innodb.inc
+-- source include/have_ucs2.inc
+
+-- let charset = ucs2
+-- source include/innodb-index.inc
diff --git a/mysql-test/t/innodb-timeout.test b/mysql-test/t/innodb-timeout.test
new file mode 100644
index 00000000000..f23fe3cff2d
--- /dev/null
+++ b/mysql-test/t/innodb-timeout.test
@@ -0,0 +1,64 @@
+-- source include/have_innodb.inc
+
+let $timeout=`select @@innodb_lock_wait_timeout`;
+set global innodb_lock_wait_timeout=42;
+
+connect (a,localhost,root,,);
+connect (b,localhost,root,,);
+
+connection a;
+select @@innodb_lock_wait_timeout;
+set innodb_lock_wait_timeout=1;
+select @@innodb_lock_wait_timeout;
+
+connection b;
+select @@innodb_lock_wait_timeout;
+set global innodb_lock_wait_timeout=347;
+select @@innodb_lock_wait_timeout;
+set innodb_lock_wait_timeout=1;
+select @@innodb_lock_wait_timeout;
+
+connect (c,localhost,root,,);
+connection c;
+select @@innodb_lock_wait_timeout;
+connection default;
+disconnect c;
+
+connection a;
+create table t1(a int primary key)engine=innodb;
+begin;
+insert into t1 values(1),(2),(3);
+
+connection b;
+--send
+select * from t1 for update;
+
+connection a;
+commit;
+
+connection b;
+reap;
+
+connection a;
+begin;
+insert into t1 values(4);
+
+connection b;
+--send
+select * from t1 for update;
+
+connection a;
+sleep 2;
+commit;
+
+connection b;
+--error ER_LOCK_WAIT_TIMEOUT
+reap;
+drop table t1;
+
+connection default;
+
+disconnect a;
+disconnect b;
+
+eval set global innodb_lock_wait_timeout=$timeout;
diff --git a/mysql-test/t/innodb-use-sys-malloc-master.opt b/mysql-test/t/innodb-use-sys-malloc-master.opt
new file mode 100644
index 00000000000..889834add01
--- /dev/null
+++ b/mysql-test/t/innodb-use-sys-malloc-master.opt
@@ -0,0 +1,2 @@
+--innodb-use-sys-malloc=true
+--innodb-use-sys-malloc=true
diff --git a/mysql-test/t/innodb-use-sys-malloc.test b/mysql-test/t/innodb-use-sys-malloc.test
new file mode 100644
index 00000000000..325dd19d086
--- /dev/null
+++ b/mysql-test/t/innodb-use-sys-malloc.test
@@ -0,0 +1,48 @@
+--source include/have_innodb.inc
+
+#display current value of innodb_use_sys_malloc
+SELECT @@GLOBAL.innodb_use_sys_malloc;
+--echo 1 Expected
+
+#try changing it. Should fail.
+--error ER_INCORRECT_GLOBAL_LOCAL_VAR
+SET @@GLOBAL.innodb_use_sys_malloc=0;
+--echo Expected error 'Read only variable'
+
+SELECT @@GLOBAL.innodb_use_sys_malloc;
+--echo 1 Expected
+
+
+#do some stuff to see if it works.
+--disable_warnings
+drop table if exists t1;
+--enable_warnings
+
+create table t1(a int not null) engine=innodb DEFAULT CHARSET=latin1;
+insert into t1 values (1),(2),(3),(4),(5),(6),(7);
+select * from t1;
+drop table t1;
+--source include/have_innodb.inc
+
+#display current value of innodb_use_sys_malloc
+SELECT @@GLOBAL.innodb_use_sys_malloc;
+--echo 1 Expected
+
+#try changing it. Should fail.
+--error ER_INCORRECT_GLOBAL_LOCAL_VAR
+SET @@GLOBAL.innodb_use_sys_malloc=0;
+--echo Expected error 'Read only variable'
+
+SELECT @@GLOBAL.innodb_use_sys_malloc;
+--echo 1 Expected
+
+
+#do some stuff to see if it works.
+--disable_warnings
+drop table if exists t1;
+--enable_warnings
+
+create table t1(a int not null) engine=innodb DEFAULT CHARSET=latin1;
+insert into t1 values (1),(2),(3),(4),(5),(6),(7);
+select * from t1;
+drop table t1;
diff --git a/mysql-test/t/innodb-zip.test b/mysql-test/t/innodb-zip.test
new file mode 100644
index 00000000000..f3f00aea26d
--- /dev/null
+++ b/mysql-test/t/innodb-zip.test
@@ -0,0 +1,347 @@
+-- source include/have_innodb.inc
+
+let $per_table=`select @@innodb_file_per_table`;
+let $format=`select @@innodb_file_format`;
+let $innodb_file_format_check_orig=`select @@innodb_file_format_check`;
+set global innodb_file_per_table=off;
+set global innodb_file_format=`0`;
+
+create table t0(a int primary key) engine=innodb row_format=compressed;
+create table t00(a int primary key) engine=innodb
+key_block_size=4 row_format=compressed;
+create table t1(a int primary key) engine=innodb row_format=dynamic;
+create table t2(a int primary key) engine=innodb row_format=redundant;
+create table t3(a int primary key) engine=innodb row_format=compact;
+create table t4(a int primary key) engine=innodb key_block_size=9;
+create table t5(a int primary key) engine=innodb
+key_block_size=1 row_format=redundant;
+
+set global innodb_file_per_table=on;
+create table t6(a int primary key) engine=innodb
+key_block_size=1 row_format=redundant;
+set global innodb_file_format=`1`;
+create table t7(a int primary key) engine=innodb
+key_block_size=1 row_format=redundant;
+create table t8(a int primary key) engine=innodb
+key_block_size=1 row_format=fixed;
+create table t9(a int primary key) engine=innodb
+key_block_size=1 row_format=compact;
+create table t10(a int primary key) engine=innodb
+key_block_size=1 row_format=dynamic;
+create table t11(a int primary key) engine=innodb
+key_block_size=1 row_format=compressed;
+create table t12(a int primary key) engine=innodb
+key_block_size=1;
+create table t13(a int primary key) engine=innodb
+row_format=compressed;
+create table t14(a int primary key) engine=innodb key_block_size=9;
+
+SELECT table_schema, table_name, row_format
+FROM information_schema.tables WHERE engine='innodb';
+
+drop table t0,t00,t2,t3,t4,t5,t6,t7,t8,t9,t10,t11,t12,t13,t14;
+alter table t1 key_block_size=0;
+alter table t1 row_format=dynamic;
+SELECT table_schema, table_name, row_format
+FROM information_schema.tables WHERE engine='innodb';
+alter table t1 row_format=compact;
+SELECT table_schema, table_name, row_format
+FROM information_schema.tables WHERE engine='innodb';
+alter table t1 row_format=redundant;
+SELECT table_schema, table_name, row_format
+FROM information_schema.tables WHERE engine='innodb';
+drop table t1;
+
+create table t1(a int not null, b text, index(b(10))) engine=innodb
+key_block_size=1;
+
+create table t2(b text)engine=innodb;
+insert into t2 values(concat('1abcdefghijklmnopqrstuvwxyz', repeat('A',5000)));
+
+insert into t1 select 1, b from t2;
+commit;
+
+connect (a,localhost,root,,);
+connect (b,localhost,root,,);
+
+connection a;
+begin;
+update t1 set b=repeat('B',100);
+
+connection b;
+select a,left(b,40) from t1 natural join t2;
+
+connection a;
+rollback;
+
+connection b;
+select a,left(b,40) from t1 natural join t2;
+
+connection default;
+disconnect a;
+disconnect b;
+
+SELECT table_schema, table_name, row_format
+FROM information_schema.tables WHERE engine='innodb';
+drop table t1,t2;
+
+# The following should fail even in non-strict mode.
+SET SESSION innodb_strict_mode = off;
+--error ER_TOO_BIG_ROWSIZE
+CREATE TABLE t1(
+	c TEXT NOT NULL, d TEXT NOT NULL,
+	PRIMARY KEY (c(767),d(767)))
+ENGINE=InnoDB ROW_FORMAT=COMPRESSED KEY_BLOCK_SIZE=1 CHARSET=ASCII;
+--error ER_TOO_BIG_ROWSIZE
+CREATE TABLE t1(
+	c TEXT NOT NULL, d TEXT NOT NULL,
+	PRIMARY KEY (c(767),d(767)))
+ENGINE=InnoDB ROW_FORMAT=COMPRESSED KEY_BLOCK_SIZE=2 CHARSET=ASCII;
+CREATE TABLE t1(
+	c TEXT NOT NULL, d TEXT NOT NULL,
+	PRIMARY KEY (c(767),d(767)))
+ENGINE=InnoDB ROW_FORMAT=COMPRESSED KEY_BLOCK_SIZE=4 CHARSET=ASCII;
+drop table t1;
+--error ER_TOO_BIG_ROWSIZE
+CREATE TABLE t1(c TEXT, PRIMARY KEY (c(440)))
+ENGINE=InnoDB ROW_FORMAT=COMPRESSED KEY_BLOCK_SIZE=1 CHARSET=ASCII;
+# The maximum key size for a compressed row actually depends on the
+# version of libz used, as account must be taken for the maximum
+# compressed size of a key, and this differs between libz
+# versions. Some libz versions allow a size of 439, some only 438.
+CREATE TABLE t1(c TEXT, PRIMARY KEY (c(438)))
+ENGINE=InnoDB ROW_FORMAT=COMPRESSED KEY_BLOCK_SIZE=1 CHARSET=ASCII;
+INSERT INTO t1 VALUES(REPEAT('A',512)),(REPEAT('B',512));
+DROP TABLE t1;
+
+#
+# Test blob column inheritance (mantis issue#36)
+#
+
+create table t1( c1 int not null, c2 blob, c3 blob, c4 blob,
+		primary key(c1, c2(22), c3(22)))
+		engine = innodb row_format = dynamic;
+begin;
+insert into t1 values(1, repeat('A', 20000), repeat('B', 20000),
+			repeat('C', 20000));
+
+update t1 set c3 = repeat('D', 20000) where c1 = 1;
+commit;
+
+# one blob column which is unchanged in update and part of PK
+# one blob column which is changed and part of of PK
+# one blob column which is not part of PK and is unchanged
+select count(*) from t1 where c2 = repeat('A', 20000);
+select count(*) from t1 where c3 = repeat('D', 20000);
+select count(*) from t1 where c4 = repeat('C', 20000);
+
+update t1 set c3 = repeat('E', 20000) where c1 = 1;
+drop table t1;
+
+#
+#
+# Test innodb_file_format
+#
+set global innodb_file_format=`0`;
+select @@innodb_file_format;
+set global innodb_file_format=`1`;
+select @@innodb_file_format;
+-- error ER_WRONG_ARGUMENTS
+set global innodb_file_format=`2`;
+-- error ER_WRONG_ARGUMENTS
+set global innodb_file_format=`-1`;
+set global innodb_file_format=`Antelope`;
+set global innodb_file_format=`Barracuda`;
+-- error ER_WRONG_ARGUMENTS
+set global innodb_file_format=`Cheetah`;
+-- error ER_WRONG_ARGUMENTS
+set global innodb_file_format=`abc`;
+-- error ER_WRONG_ARGUMENTS
+set global innodb_file_format=`1a`;
+-- error ER_WRONG_ARGUMENTS
+set global innodb_file_format=``;
+
+#test strict mode.
+# this does not work anymore, has been removed from mysqltest
+# -- enable_errors
+set global innodb_file_per_table = on;
+set global innodb_file_format = `1`;
+
+set innodb_strict_mode = off;
+create table t1 (id int primary key) engine = innodb key_block_size = 0;
+drop table t1;
+
+#set strict_mode
+set innodb_strict_mode = on;
+
+#Test different values of KEY_BLOCK_SIZE
+
+--error ER_CANT_CREATE_TABLE
+create table t1 (id int primary key) engine = innodb key_block_size = 0;
+show errors;
+
+--error ER_CANT_CREATE_TABLE
+create table t2 (id int primary key) engine = innodb key_block_size = 9;
+show errors;
+
+
+create table t3 (id int primary key) engine = innodb key_block_size = 1;
+create table t4 (id int primary key) engine = innodb key_block_size = 2;
+create table t5 (id int primary key) engine = innodb key_block_size = 4;
+create table t6 (id int primary key) engine = innodb key_block_size = 8;
+create table t7 (id int primary key) engine = innodb key_block_size = 16;
+
+#check various ROW_FORMAT values.
+create table t8 (id int primary key) engine = innodb row_format = compressed;
+create table t9 (id int primary key) engine = innodb row_format = dynamic;
+create table t10(id int primary key) engine = innodb row_format = compact;
+create table t11(id int primary key) engine = innodb row_format = redundant;
+
+SELECT table_schema, table_name, row_format
+FROM information_schema.tables WHERE engine='innodb';
+drop table t3, t4, t5, t6, t7, t8, t9, t10, t11;
+
+#test different values of ROW_FORMAT with KEY_BLOCK_SIZE
+create table t1 (id int primary key) engine = innodb
+key_block_size = 8 row_format = compressed;
+
+--error ER_CANT_CREATE_TABLE
+create table t2 (id int primary key) engine = innodb
+key_block_size = 8 row_format = redundant;
+show errors;
+
+--error ER_CANT_CREATE_TABLE
+create table t3 (id int primary key) engine = innodb
+key_block_size = 8 row_format = compact;
+show errors;
+
+--error ER_CANT_CREATE_TABLE
+create table t4 (id int primary key) engine = innodb
+key_block_size = 8 row_format = dynamic;
+show errors;
+
+--error ER_CANT_CREATE_TABLE
+create table t5 (id int primary key) engine = innodb
+key_block_size = 8 row_format = default;
+show errors;
+
+SELECT table_schema, table_name, row_format
+FROM information_schema.tables WHERE engine='innodb';
+drop table t1;
+
+#test multiple errors
+--error ER_CANT_CREATE_TABLE
+create table t1 (id int primary key) engine = innodb
+key_block_size = 9 row_format = redundant;
+show errors;
+
+--error ER_CANT_CREATE_TABLE
+create table t2 (id int primary key) engine = innodb
+key_block_size = 9 row_format = compact;
+show errors;
+
+--error ER_CANT_CREATE_TABLE
+create table t2 (id int primary key) engine = innodb
+key_block_size = 9 row_format = dynamic;
+show errors;
+
+SELECT table_schema, table_name, row_format
+FROM information_schema.tables WHERE engine='innodb';
+
+#test valid values with innodb_file_per_table unset
+set global innodb_file_per_table = off;
+
+--error ER_CANT_CREATE_TABLE
+create table t1 (id int primary key) engine = innodb key_block_size = 1;
+show errors;
+--error ER_CANT_CREATE_TABLE
+create table t2 (id int primary key) engine = innodb key_block_size = 2;
+show errors;
+--error ER_CANT_CREATE_TABLE
+create table t3 (id int primary key) engine = innodb key_block_size = 4;
+show errors;
+--error ER_CANT_CREATE_TABLE
+create table t4 (id int primary key) engine = innodb key_block_size = 8;
+show errors;
+--error ER_CANT_CREATE_TABLE
+create table t5 (id int primary key) engine = innodb key_block_size = 16;
+show errors;
+--error ER_CANT_CREATE_TABLE
+create table t6 (id int primary key) engine = innodb row_format = compressed;
+show errors;
+--error ER_CANT_CREATE_TABLE
+create table t7 (id int primary key) engine = innodb row_format = dynamic;
+show errors;
+create table t8 (id int primary key) engine = innodb row_format = compact;
+create table t9 (id int primary key) engine = innodb row_format = redundant;
+
+SELECT table_schema, table_name, row_format
+FROM information_schema.tables WHERE engine='innodb';
+drop table t8, t9;
+
+#test valid values with innodb_file_format unset
+set global innodb_file_per_table = on;
+set global innodb_file_format = `0`; 
+
+--error ER_CANT_CREATE_TABLE
+create table t1 (id int primary key) engine = innodb key_block_size = 1;
+show errors;
+--error ER_CANT_CREATE_TABLE
+create table t2 (id int primary key) engine = innodb key_block_size = 2;
+show errors;
+--error ER_CANT_CREATE_TABLE
+create table t3 (id int primary key) engine = innodb key_block_size = 4;
+show errors;
+--error ER_CANT_CREATE_TABLE
+create table t4 (id int primary key) engine = innodb key_block_size = 8;
+show errors;
+--error ER_CANT_CREATE_TABLE
+create table t5 (id int primary key) engine = innodb key_block_size = 16;
+show errors;
+--error ER_CANT_CREATE_TABLE
+create table t6 (id int primary key) engine = innodb row_format = compressed;
+show errors;
+--error ER_CANT_CREATE_TABLE
+create table t7 (id int primary key) engine = innodb row_format = dynamic;
+show errors;
+create table t8 (id int primary key) engine = innodb row_format = compact;
+create table t9 (id int primary key) engine = innodb row_format = redundant;
+
+SELECT table_schema, table_name, row_format
+FROM information_schema.tables WHERE engine='innodb';
+drop table t8, t9;
+
+eval set global innodb_file_per_table=$per_table;
+eval set global innodb_file_format=$format;
+#
+# Testing of tablespace tagging
+#
+-- disable_info
+set global innodb_file_per_table=on;
+set global innodb_file_format=`Barracuda`;
+set global innodb_file_format_check=`Antelope`;
+create table normal_table (
+  c1 int
+) engine = innodb;
+select @@innodb_file_format_check;
+create table zip_table (
+  c1 int
+) engine = innodb key_block_size = 8;
+select @@innodb_file_format_check;
+set global innodb_file_format_check=`Antelope`;
+select @@innodb_file_format_check;
+-- disable_result_log
+show table status;
+-- enable_result_log
+select @@innodb_file_format_check;
+drop table normal_table, zip_table;
+-- disable_result_log
+
+#
+# restore environment to the state it was before this test execution
+#
+
+-- disable_query_log
+eval set global innodb_file_format=$format;
+eval set global innodb_file_per_table=$per_table;
+eval set global innodb_file_format_check=$innodb_file_format_check_orig;
diff --git a/mysql-test/t/innodb.test b/mysql-test/t/innodb.test
index b0353ed5268..b4ba75dd0d1 100644
--- a/mysql-test/t/innodb.test
+++ b/mysql-test/t/innodb.test
@@ -1126,7 +1126,7 @@ show create table t2;
 create index id2 on t2 (id);
 show create table t2;
 drop index id2 on t2;
---error 1025,1025
+--error ER_DROP_INDEX_FK
 drop index id on t2;
 show create table t2;
 drop table t2;
@@ -1294,6 +1294,7 @@ drop table t1;
 
 # Test for testable InnoDB status variables. This test
 # uses previous ones(pages_created, rows_deleted, ...).
+--replace_result 512 511
 show status like "Innodb_buffer_pool_pages_total";
 show status like "Innodb_page_size";
 show status like "Innodb_rows_deleted";
@@ -2357,6 +2358,7 @@ DROP TABLE t1,t2;
 #
 # Bug #21101 (Prints wrong error message if max row size is too large)
 #
+set innodb_strict_mode=on;
 --error 1118
 CREATE TABLE t1 (
 	c01 CHAR(255), c02 CHAR(255), c03 CHAR(255), c04 CHAR(255),
diff --git a/mysql-test/t/innodb_bug34300.test b/mysql-test/t/innodb_bug34300.test
index c4dbce02e05..0068fcd7b60 100644
--- a/mysql-test/t/innodb_bug34300.test
+++ b/mysql-test/t/innodb_bug34300.test
@@ -9,6 +9,7 @@
 -- disable_result_log
 
 # set packet size and reconnect 
+SET @save_max_allowed_packet=@@global.max_allowed_packet;
 SET @@global.max_allowed_packet=16777216;
 --connect (newconn, localhost, root,,)
 
@@ -21,7 +22,6 @@ CREATE TABLE bug34300 (
 
 INSERT INTO bug34300 VALUES ('xxx', repeat('a', 8459264), 'zzz');
 
--- enable_query_log
 -- enable_result_log
 
 SELECT f4, f8 FROM bug34300;
@@ -31,11 +31,6 @@ ALTER TABLE bug34300 ADD COLUMN (f10 INT);
 SELECT f4, f8 FROM bug34300;
 
 DROP TABLE bug34300;
-
 disconnect newconn;
 connection default;
---disable_result_log
---disable_query_log
-SET @@global.max_allowed_packet=default;
---enable_result_log
---enable_query_log
+SET @@global.max_allowed_packet=@save_max_allowed_packet;
diff --git a/mysql-test/t/innodb_bug36169.test b/mysql-test/t/innodb_bug36169.test
new file mode 100644
index 00000000000..6fdf4015417
--- /dev/null
+++ b/mysql-test/t/innodb_bug36169.test
@@ -0,0 +1,1162 @@
+#
+# Bug#36169 create innodb compressed table with too large row size crashed
+# http://bugs.mysql.com/36169
+#
+
+-- source include/have_innodb.inc
+
+SET @save_innodb_file_format=@@global.innodb_file_format;
+SET @save_innodb_file_format_check=@@global.innodb_file_format_check;
+SET @save_innodb_file_per_table=@@global.innodb_file_per_table;
+SET GLOBAL innodb_file_format='Barracuda';
+SET GLOBAL innodb_file_per_table=ON;
+
+#
+# The following is copied from http://bugs.mysql.com/36169
+# (http://bugs.mysql.com/file.php?id=9121)
+# Probably it can be simplified but that is not obvious.
+#
+
+# we care only that the following SQL commands do produce errors
+# as expected and do not crash the server
+-- disable_query_log
+-- disable_result_log
+
+# Generating 10 tables
+# Creating a table with 94 columns and 24 indexes
+DROP TABLE IF EXISTS `table0`;
+--error ER_TOO_BIG_ROWSIZE
+CREATE TABLE IF NOT EXISTS `table0`
+(`col0` BOOL,
+`col1` BOOL,
+`col2` TINYINT,
+`col3` DATE,
+`col4` TIME,
+`col5` SET ('test1','test2','test3'),
+`col6` TIME,
+`col7` TEXT,
+`col8` DECIMAL,
+`col9` SET ('test1','test2','test3'),
+`col10` FLOAT,
+`col11` DOUBLE PRECISION,
+`col12` ENUM ('test1','test2','test3'),
+`col13` TINYBLOB,
+`col14` YEAR,
+`col15` SET ('test1','test2','test3'),
+`col16` NUMERIC,
+`col17` NUMERIC,
+`col18` BLOB,
+`col19` DATETIME,
+`col20` DOUBLE PRECISION,
+`col21` DECIMAL,
+`col22` DATETIME,
+`col23` NUMERIC,
+`col24` NUMERIC,
+`col25` LONGTEXT,
+`col26` TINYBLOB,
+`col27` TIME,
+`col28` TINYBLOB,
+`col29` ENUM ('test1','test2','test3'),
+`col30` SMALLINT,
+`col31` REAL,
+`col32` FLOAT,
+`col33` CHAR (175),
+`col34` TINYTEXT,
+`col35` TINYTEXT,
+`col36` TINYBLOB,
+`col37` TINYBLOB,
+`col38` TINYTEXT,
+`col39` MEDIUMBLOB,
+`col40` TIMESTAMP,
+`col41` DOUBLE,
+`col42` SMALLINT,
+`col43` LONGBLOB,
+`col44` VARCHAR (80),
+`col45` MEDIUMTEXT,
+`col46` NUMERIC,
+`col47` BIGINT,
+`col48` DATE,
+`col49` TINYBLOB,
+`col50` DATE,
+`col51` BOOL,
+`col52` MEDIUMINT,
+`col53` FLOAT,
+`col54` TINYBLOB,
+`col55` LONGTEXT,
+`col56` SMALLINT,
+`col57` ENUM ('test1','test2','test3'),
+`col58` DATETIME,
+`col59` MEDIUMTEXT,
+`col60` VARCHAR (232),
+`col61` NUMERIC,
+`col62` YEAR,
+`col63` SMALLINT,
+`col64` TIMESTAMP,
+`col65` BLOB,
+`col66` LONGBLOB,
+`col67` INT,
+`col68` LONGTEXT,
+`col69` ENUM ('test1','test2','test3'),
+`col70` INT,
+`col71` TIME,
+`col72` TIMESTAMP,
+`col73` TIMESTAMP,
+`col74` VARCHAR (170),
+`col75` SET ('test1','test2','test3'),
+`col76` TINYBLOB,
+`col77` BIGINT,
+`col78` NUMERIC,
+`col79` DATETIME,
+`col80` YEAR,
+`col81` NUMERIC,
+`col82` LONGBLOB,
+`col83` TEXT,
+`col84` CHAR (83),
+`col85` DECIMAL,
+`col86` FLOAT,
+`col87` INT,
+`col88` VARCHAR (145),
+`col89` DATE,
+`col90` DECIMAL,
+`col91` DECIMAL,
+`col92` MEDIUMBLOB,
+`col93` TIME,
+KEY `idx0` (`col69`,`col90`,`col8`),
+KEY `idx1` (`col60`),
+KEY `idx2` (`col60`,`col70`,`col74`),
+KEY `idx3` (`col22`,`col32`,`col72`,`col30`),
+KEY `idx4` (`col29`),
+KEY `idx5` (`col19`,`col45`(143)),
+KEY `idx6` (`col46`,`col48`,`col5`,`col39`(118)),
+KEY `idx7` (`col48`,`col61`),
+KEY `idx8` (`col93`),
+KEY `idx9` (`col31`),
+KEY `idx10` (`col30`,`col21`),
+KEY `idx11` (`col67`),
+KEY `idx12` (`col44`,`col6`,`col8`,`col38`(226)),
+KEY `idx13` (`col71`,`col41`,`col15`,`col49`(88)),
+KEY `idx14` (`col78`),
+KEY `idx15` (`col63`,`col67`,`col64`),
+KEY `idx16` (`col17`,`col86`),
+KEY `idx17` (`col77`,`col56`,`col10`,`col55`(24)),
+KEY `idx18` (`col62`),
+KEY `idx19` (`col31`,`col57`,`col56`,`col53`),
+KEY `idx20` (`col46`),
+KEY `idx21` (`col83`(54)),
+KEY `idx22` (`col51`,`col7`(120)),
+KEY `idx23` (`col7`(163),`col31`,`col71`,`col14`)
+)engine=innodb ROW_FORMAT=COMPRESSED KEY_BLOCK_SIZE=1;
+
+# Creating a table with 10 columns and 32 indexes
+DROP TABLE IF EXISTS `table1`;
+--error ER_TOO_BIG_ROWSIZE
+CREATE TABLE IF NOT EXISTS `table1`
+(`col0` CHAR (113),
+`col1` FLOAT,
+`col2` BIGINT,
+`col3` DECIMAL,
+`col4` BLOB,
+`col5` LONGTEXT,
+`col6` SET ('test1','test2','test3'),
+`col7` BIGINT,
+`col8` BIGINT,
+`col9` TINYBLOB,
+KEY `idx0` (`col5`(101),`col7`,`col8`),
+KEY `idx1` (`col8`),
+KEY `idx2` (`col4`(177),`col9`(126),`col6`,`col3`),
+KEY `idx3` (`col5`(160)),
+KEY `idx4` (`col9`(242)),
+KEY `idx5` (`col4`(139),`col2`,`col3`),
+KEY `idx6` (`col7`),
+KEY `idx7` (`col6`,`col2`,`col0`,`col3`),
+KEY `idx8` (`col9`(66)),
+KEY `idx9` (`col5`(253)),
+KEY `idx10` (`col1`,`col7`,`col2`),
+KEY `idx11` (`col9`(242),`col0`,`col8`,`col5`(163)),
+KEY `idx12` (`col8`),
+KEY `idx13` (`col0`,`col9`(37)),
+KEY `idx14` (`col0`),
+KEY `idx15` (`col5`(111)),
+KEY `idx16` (`col8`,`col0`,`col5`(13)),
+KEY `idx17` (`col4`(139)),
+KEY `idx18` (`col5`(189),`col2`,`col3`,`col9`(136)),
+KEY `idx19` (`col0`,`col3`,`col1`,`col8`),
+KEY `idx20` (`col8`),
+KEY `idx21` (`col0`,`col7`,`col9`(227),`col3`),
+KEY `idx22` (`col0`),
+KEY `idx23` (`col2`),
+KEY `idx24` (`col3`),
+KEY `idx25` (`col2`,`col3`),
+KEY `idx26` (`col0`),
+KEY `idx27` (`col5`(254)),
+KEY `idx28` (`col3`),
+KEY `idx29` (`col3`),
+KEY `idx30` (`col7`,`col3`,`col0`,`col4`(220)),
+KEY `idx31` (`col4`(1),`col0`)
+)engine=innodb ROW_FORMAT=COMPRESSED KEY_BLOCK_SIZE=1;
+
+# Creating a table with 141 columns and 18 indexes
+DROP TABLE IF EXISTS `table2`;
+--error ER_TOO_BIG_ROWSIZE
+CREATE TABLE IF NOT EXISTS `table2`
+(`col0` BOOL,
+`col1` MEDIUMINT,
+`col2` VARCHAR (209),
+`col3` MEDIUMBLOB,
+`col4` CHAR (13),
+`col5` DOUBLE,
+`col6` TINYTEXT,
+`col7` REAL,
+`col8` SMALLINT,
+`col9` BLOB,
+`col10` TINYINT,
+`col11` DECIMAL,
+`col12` BLOB,
+`col13` DECIMAL,
+`col14` LONGBLOB,
+`col15` SMALLINT,
+`col16` LONGBLOB,
+`col17` TINYTEXT,
+`col18` FLOAT,
+`col19` CHAR (78),
+`col20` MEDIUMTEXT,
+`col21` SET ('test1','test2','test3'),
+`col22` MEDIUMINT,
+`col23` INT,
+`col24` MEDIUMBLOB,
+`col25` ENUM ('test1','test2','test3'),
+`col26` TINYBLOB,
+`col27` VARCHAR (116),
+`col28` TIMESTAMP,
+`col29` BLOB,
+`col30` SMALLINT,
+`col31` DOUBLE PRECISION,
+`col32` DECIMAL,
+`col33` DECIMAL,
+`col34` TEXT,
+`col35` MEDIUMINT,
+`col36` MEDIUMINT,
+`col37` BIGINT,
+`col38` VARCHAR (253),
+`col39` TINYBLOB,
+`col40` MEDIUMBLOB,
+`col41` BIGINT,
+`col42` DOUBLE,
+`col43` TEXT,
+`col44` BLOB,
+`col45` TIME,
+`col46` MEDIUMINT,
+`col47` DOUBLE PRECISION,
+`col48` SET ('test1','test2','test3'),
+`col49` DOUBLE PRECISION,
+`col50` VARCHAR (97),
+`col51` TEXT,
+`col52` NUMERIC,
+`col53` ENUM ('test1','test2','test3'),
+`col54` MEDIUMTEXT,
+`col55` MEDIUMINT,
+`col56` DATETIME,
+`col57` DATETIME,
+`col58` MEDIUMTEXT,
+`col59` CHAR (244),
+`col60` LONGBLOB,
+`col61` MEDIUMBLOB,
+`col62` DOUBLE,
+`col63` SMALLINT,
+`col64` BOOL,
+`col65` SMALLINT,
+`col66` VARCHAR (212),
+`col67` TIME,
+`col68` REAL,
+`col69` BOOL,
+`col70` BIGINT,
+`col71` DATE,
+`col72` TINYINT,
+`col73` ENUM ('test1','test2','test3'),
+`col74` DATE,
+`col75` TIME,
+`col76` DATETIME,
+`col77` BOOL,
+`col78` TINYTEXT,
+`col79` MEDIUMINT,
+`col80` NUMERIC,
+`col81` LONGTEXT,
+`col82` SET ('test1','test2','test3'),
+`col83` DOUBLE PRECISION,
+`col84` NUMERIC,
+`col85` VARCHAR (184),
+`col86` DOUBLE PRECISION,
+`col87` MEDIUMTEXT,
+`col88` MEDIUMBLOB,
+`col89` BOOL,
+`col90` SMALLINT,
+`col91` TINYINT,
+`col92` ENUM ('test1','test2','test3'),
+`col93` BOOL,
+`col94` TIMESTAMP,
+`col95` BOOL,
+`col96` MEDIUMTEXT,
+`col97` DECIMAL,
+`col98` BOOL,
+`col99` DECIMAL,
+`col100` MEDIUMINT,
+`col101` DOUBLE PRECISION,
+`col102` TINYINT,
+`col103` BOOL,
+`col104` MEDIUMINT,
+`col105` DECIMAL,
+`col106` NUMERIC,
+`col107` TIMESTAMP,
+`col108` MEDIUMBLOB,
+`col109` TINYBLOB,
+`col110` SET ('test1','test2','test3'),
+`col111` YEAR,
+`col112` TIMESTAMP,
+`col113` CHAR (201),
+`col114` BOOL,
+`col115` TINYINT,
+`col116` DOUBLE,
+`col117` TINYINT,
+`col118` TIMESTAMP,
+`col119` SET ('test1','test2','test3'),
+`col120` SMALLINT,
+`col121` TINYBLOB,
+`col122` TIMESTAMP,
+`col123` BLOB,
+`col124` DATE,
+`col125` SMALLINT,
+`col126` ENUM ('test1','test2','test3'),
+`col127` MEDIUMBLOB,
+`col128` DOUBLE PRECISION,
+`col129` REAL,
+`col130` VARCHAR (159),
+`col131` MEDIUMBLOB,
+`col132` BIGINT,
+`col133` INT,
+`col134` SET ('test1','test2','test3'),
+`col135` CHAR (198),
+`col136` SET ('test1','test2','test3'),
+`col137` MEDIUMTEXT,
+`col138` SMALLINT,
+`col139` BLOB,
+`col140` LONGBLOB,
+KEY `idx0` (`col14`(139),`col24`(208),`col38`,`col35`),
+KEY `idx1` (`col48`,`col118`,`col29`(131),`col100`),
+KEY `idx2` (`col86`,`col67`,`col43`(175)),
+KEY `idx3` (`col19`),
+KEY `idx4` (`col40`(220),`col67`),
+KEY `idx5` (`col99`,`col56`),
+KEY `idx6` (`col68`,`col28`,`col137`(157)),
+KEY `idx7` (`col51`(160),`col99`,`col45`,`col39`(9)),
+KEY `idx8` (`col15`,`col52`,`col90`,`col94`),
+KEY `idx9` (`col24`(3),`col139`(248),`col108`(118),`col41`),
+KEY `idx10` (`col36`,`col92`,`col114`),
+KEY `idx11` (`col115`,`col9`(116)),
+KEY `idx12` (`col130`,`col93`,`col134`),
+KEY `idx13` (`col123`(65)),
+KEY `idx14` (`col44`(90),`col86`,`col119`),
+KEY `idx15` (`col69`),
+KEY `idx16` (`col132`,`col81`(118),`col18`),
+KEY `idx17` (`col24`(250),`col7`,`col92`,`col45`)
+)engine=innodb ROW_FORMAT=COMPRESSED KEY_BLOCK_SIZE=1;
+
+# Creating a table with 199 columns and 1 indexes
+DROP TABLE IF EXISTS `table3`;
+--error ER_TOO_BIG_ROWSIZE
+CREATE TABLE IF NOT EXISTS `table3`
+(`col0` SMALLINT,
+`col1` SET ('test1','test2','test3'),
+`col2` TINYTEXT,
+`col3` DOUBLE,
+`col4` NUMERIC,
+`col5` DATE,
+`col6` BIGINT,
+`col7` DOUBLE,
+`col8` TEXT,
+`col9` INT,
+`col10` REAL,
+`col11` TINYINT,
+`col12` NUMERIC,
+`col13` NUMERIC,
+`col14` TIME,
+`col15` DOUBLE,
+`col16` REAL,
+`col17` MEDIUMBLOB,
+`col18` YEAR,
+`col19` TINYTEXT,
+`col20` YEAR,
+`col21` CHAR (250),
+`col22` TINYINT,
+`col23` TINYINT,
+`col24` SMALLINT,
+`col25` DATETIME,
+`col26` MEDIUMINT,
+`col27` LONGBLOB,
+`col28` VARCHAR (106),
+`col29` FLOAT,
+`col30` MEDIUMTEXT,
+`col31` TINYBLOB,
+`col32` BIGINT,
+`col33` YEAR,
+`col34` REAL,
+`col35` MEDIUMBLOB,
+`col36` LONGTEXT,
+`col37` LONGBLOB,
+`col38` BIGINT,
+`col39` FLOAT,
+`col40` TIME,
+`col41` DATETIME,
+`col42` BOOL,
+`col43` BIGINT,
+`col44` SMALLINT,
+`col45` TIME,
+`col46` DOUBLE PRECISION,
+`col47` TIME,
+`col48` TINYTEXT,
+`col49` DOUBLE PRECISION,
+`col50` BIGINT,
+`col51` NUMERIC,
+`col52` TINYBLOB,
+`col53` DATE,
+`col54` DECIMAL,
+`col55` SMALLINT,
+`col56` TINYTEXT,
+`col57` ENUM ('test1','test2','test3'),
+`col58` YEAR,
+`col59` TIME,
+`col60` TINYINT,
+`col61` DECIMAL,
+`col62` DOUBLE,
+`col63` DATE,
+`col64` LONGTEXT,
+`col65` DOUBLE,
+`col66` VARCHAR (88),
+`col67` MEDIUMTEXT,
+`col68` DATE,
+`col69` MEDIUMINT,
+`col70` DECIMAL,
+`col71` MEDIUMTEXT,
+`col72` LONGTEXT,
+`col73` REAL,
+`col74` DOUBLE,
+`col75` TIME,
+`col76` DATE,
+`col77` DECIMAL,
+`col78` MEDIUMBLOB,
+`col79` NUMERIC,
+`col80` BIGINT,
+`col81` YEAR,
+`col82` SMALLINT,
+`col83` MEDIUMINT,
+`col84` TINYINT,
+`col85` MEDIUMBLOB,
+`col86` TIME,
+`col87` MEDIUMBLOB,
+`col88` LONGTEXT,
+`col89` BOOL,
+`col90` BLOB,
+`col91` LONGBLOB,
+`col92` YEAR,
+`col93` BLOB,
+`col94` INT,
+`col95` TINYTEXT,
+`col96` TINYINT,
+`col97` DECIMAL,
+`col98` ENUM ('test1','test2','test3'),
+`col99` MEDIUMINT,
+`col100` TINYINT,
+`col101` MEDIUMBLOB,
+`col102` TINYINT,
+`col103` SET ('test1','test2','test3'),
+`col104` TIMESTAMP,
+`col105` TEXT,
+`col106` DATETIME,
+`col107` MEDIUMTEXT,
+`col108` CHAR (220),
+`col109` TIME,
+`col110` VARCHAR (131),
+`col111` DECIMAL,
+`col112` FLOAT,
+`col113` SMALLINT,
+`col114` BIGINT,
+`col115` LONGBLOB,
+`col116` SET ('test1','test2','test3'),
+`col117` ENUM ('test1','test2','test3'),
+`col118` BLOB,
+`col119` MEDIUMTEXT,
+`col120` SET ('test1','test2','test3'),
+`col121` DATETIME,
+`col122` FLOAT,
+`col123` VARCHAR (242),
+`col124` YEAR,
+`col125` MEDIUMBLOB,
+`col126` TIME,
+`col127` BOOL,
+`col128` TINYBLOB,
+`col129` DOUBLE,
+`col130` TINYINT,
+`col131` BIGINT,
+`col132` SMALLINT,
+`col133` INT,
+`col134` DOUBLE PRECISION,
+`col135` MEDIUMBLOB,
+`col136` SET ('test1','test2','test3'),
+`col137` TINYTEXT,
+`col138` DOUBLE PRECISION,
+`col139` NUMERIC,
+`col140` BLOB,
+`col141` SET ('test1','test2','test3'),
+`col142` INT,
+`col143` VARCHAR (26),
+`col144` BLOB,
+`col145` REAL,
+`col146` SET ('test1','test2','test3'),
+`col147` LONGBLOB,
+`col148` TEXT,
+`col149` BLOB,
+`col150` CHAR (189),
+`col151` LONGTEXT,
+`col152` INT,
+`col153` FLOAT,
+`col154` LONGTEXT,
+`col155` DATE,
+`col156` LONGBLOB,
+`col157` TINYBLOB,
+`col158` REAL,
+`col159` DATE,
+`col160` TIME,
+`col161` YEAR,
+`col162` DOUBLE,
+`col163` VARCHAR (90),
+`col164` FLOAT,
+`col165` NUMERIC,
+`col166` ENUM ('test1','test2','test3'),
+`col167` DOUBLE PRECISION,
+`col168` DOUBLE PRECISION,
+`col169` TINYBLOB,
+`col170` TIME,
+`col171` SMALLINT,
+`col172` TINYTEXT,
+`col173` SMALLINT,
+`col174` DOUBLE,
+`col175` VARCHAR (14),
+`col176` VARCHAR (90),
+`col177` REAL,
+`col178` MEDIUMINT,
+`col179` TINYBLOB,
+`col180` FLOAT,
+`col181` TIMESTAMP,
+`col182` REAL,
+`col183` DOUBLE PRECISION,
+`col184` BIGINT,
+`col185` INT,
+`col186` MEDIUMTEXT,
+`col187` TIME,
+`col188` FLOAT,
+`col189` TIME,
+`col190` INT,
+`col191` FLOAT,
+`col192` MEDIUMINT,
+`col193` TINYINT,
+`col194` MEDIUMTEXT,
+`col195` DATE,
+`col196` TIME,
+`col197` YEAR,
+`col198` CHAR (206),
+KEY `idx0` (`col39`,`col23`)
+)engine=innodb ROW_FORMAT=COMPRESSED KEY_BLOCK_SIZE=1;
+
+# Creating a table with 133 columns and 16 indexes
+DROP TABLE IF EXISTS `table4`;
+--error ER_TOO_BIG_ROWSIZE
+CREATE TABLE IF NOT EXISTS `table4`
+(`col0` VARCHAR (60),
+`col1` NUMERIC,
+`col2` LONGTEXT,
+`col3` MEDIUMTEXT,
+`col4` LONGTEXT,
+`col5` LONGBLOB,
+`col6` LONGBLOB,
+`col7` DATETIME,
+`col8` TINYTEXT,
+`col9` BLOB,
+`col10` BOOL,
+`col11` BIGINT,
+`col12` TEXT,
+`col13` VARCHAR (213),
+`col14` TINYBLOB,
+`col15` BOOL,
+`col16` MEDIUMTEXT,
+`col17` DOUBLE,
+`col18` TEXT,
+`col19` BLOB,
+`col20` SET ('test1','test2','test3'),
+`col21` TINYINT,
+`col22` DATETIME,
+`col23` TINYINT,
+`col24` ENUM ('test1','test2','test3'),
+`col25` REAL,
+`col26` BOOL,
+`col27` FLOAT,
+`col28` LONGBLOB,
+`col29` DATETIME,
+`col30` FLOAT,
+`col31` SET ('test1','test2','test3'),
+`col32` LONGBLOB,
+`col33` NUMERIC,
+`col34` YEAR,
+`col35` VARCHAR (146),
+`col36` BIGINT,
+`col37` DATETIME,
+`col38` DATE,
+`col39` SET ('test1','test2','test3'),
+`col40` CHAR (112),
+`col41` FLOAT,
+`col42` YEAR,
+`col43` TIME,
+`col44` DOUBLE,
+`col45` NUMERIC,
+`col46` FLOAT,
+`col47` DECIMAL,
+`col48` BIGINT,
+`col49` DECIMAL,
+`col50` YEAR,
+`col51` MEDIUMTEXT,
+`col52` LONGBLOB,
+`col53` SET ('test1','test2','test3'),
+`col54` BLOB,
+`col55` FLOAT,
+`col56` REAL,
+`col57` REAL,
+`col58` TEXT,
+`col59` MEDIUMBLOB,
+`col60` INT,
+`col61` INT,
+`col62` DATE,
+`col63` TEXT,
+`col64` DATE,
+`col65` ENUM ('test1','test2','test3'),
+`col66` DOUBLE PRECISION,
+`col67` TINYTEXT,
+`col68` TINYBLOB,
+`col69` FLOAT,
+`col70` BLOB,
+`col71` DATETIME,
+`col72` DOUBLE,
+`col73` LONGTEXT,
+`col74` TIME,
+`col75` DATETIME,
+`col76` VARCHAR (122),
+`col77` MEDIUMTEXT,
+`col78` MEDIUMTEXT,
+`col79` BOOL,
+`col80` LONGTEXT,
+`col81` TINYTEXT,
+`col82` NUMERIC,
+`col83` DOUBLE PRECISION,
+`col84` DATE,
+`col85` YEAR,
+`col86` BLOB,
+`col87` TINYTEXT,
+`col88` DOUBLE PRECISION,
+`col89` MEDIUMINT,
+`col90` MEDIUMTEXT,
+`col91` NUMERIC,
+`col92` DATETIME,
+`col93` NUMERIC,
+`col94` SET ('test1','test2','test3'),
+`col95` TINYTEXT,
+`col96` SET ('test1','test2','test3'),
+`col97` YEAR,
+`col98` MEDIUMINT,
+`col99` TEXT,
+`col100` TEXT,
+`col101` TIME,
+`col102` VARCHAR (225),
+`col103` TINYTEXT,
+`col104` TEXT,
+`col105` MEDIUMTEXT,
+`col106` TINYINT,
+`col107` TEXT,
+`col108` LONGBLOB,
+`col109` LONGTEXT,
+`col110` TINYTEXT,
+`col111` CHAR (56),
+`col112` YEAR,
+`col113` ENUM ('test1','test2','test3'),
+`col114` TINYBLOB,
+`col115` DATETIME,
+`col116` DATE,
+`col117` TIME,
+`col118` MEDIUMTEXT,
+`col119` DOUBLE PRECISION,
+`col120` FLOAT,
+`col121` TIMESTAMP,
+`col122` MEDIUMINT,
+`col123` YEAR,
+`col124` DATE,
+`col125` TEXT,
+`col126` FLOAT,
+`col127` TINYTEXT,
+`col128` BOOL,
+`col129` NUMERIC,
+`col130` TIMESTAMP,
+`col131` INT,
+`col132` MEDIUMBLOB,
+KEY `idx0` (`col130`),
+KEY `idx1` (`col30`,`col55`,`col19`(31)),
+KEY `idx2` (`col104`(186)),
+KEY `idx3` (`col131`),
+KEY `idx4` (`col64`,`col93`,`col2`(11)),
+KEY `idx5` (`col34`,`col121`,`col22`),
+KEY `idx6` (`col33`,`col55`,`col83`),
+KEY `idx7` (`col17`,`col87`(245),`col99`(17)),
+KEY `idx8` (`col65`,`col120`),
+KEY `idx9` (`col82`),
+KEY `idx10` (`col9`(72)),
+KEY `idx11` (`col88`),
+KEY `idx12` (`col128`,`col9`(200),`col71`,`col66`),
+KEY `idx13` (`col77`(126)),
+KEY `idx14` (`col105`(26),`col13`,`col117`),
+KEY `idx15` (`col4`(246),`col130`,`col115`,`col3`(141))
+)engine=innodb ROW_FORMAT=COMPRESSED KEY_BLOCK_SIZE=1;
+
+# Creating a table with 176 columns and 13 indexes
+DROP TABLE IF EXISTS `table5`;
+--error ER_TOO_BIG_ROWSIZE
+CREATE TABLE IF NOT EXISTS `table5`
+(`col0` MEDIUMTEXT,
+`col1` VARCHAR (90),
+`col2` TINYTEXT,
+`col3` TIME,
+`col4` BOOL,
+`col5` TINYTEXT,
+`col6` BOOL,
+`col7` TIMESTAMP,
+`col8` TINYBLOB,
+`col9` TINYINT,
+`col10` YEAR,
+`col11` SET ('test1','test2','test3'),
+`col12` TEXT,
+`col13` CHAR (248),
+`col14` BIGINT,
+`col15` TEXT,
+`col16` TINYINT,
+`col17` NUMERIC,
+`col18` SET ('test1','test2','test3'),
+`col19` LONGBLOB,
+`col20` FLOAT,
+`col21` INT,
+`col22` TEXT,
+`col23` BOOL,
+`col24` DECIMAL,
+`col25` DOUBLE PRECISION,
+`col26` FLOAT,
+`col27` TINYBLOB,
+`col28` NUMERIC,
+`col29` MEDIUMBLOB,
+`col30` DATE,
+`col31` LONGTEXT,
+`col32` DATE,
+`col33` FLOAT,
+`col34` BIGINT,
+`col35` TINYTEXT,
+`col36` MEDIUMTEXT,
+`col37` TIME,
+`col38` INT,
+`col39` TINYINT,
+`col40` SET ('test1','test2','test3'),
+`col41` CHAR (130),
+`col42` SMALLINT,
+`col43` INT,
+`col44` MEDIUMTEXT,
+`col45` VARCHAR (126),
+`col46` INT,
+`col47` DOUBLE PRECISION,
+`col48` BIGINT,
+`col49` MEDIUMTEXT,
+`col50` TINYBLOB,
+`col51` MEDIUMINT,
+`col52` TEXT,
+`col53` VARCHAR (208),
+`col54` VARCHAR (207),
+`col55` NUMERIC,
+`col56` DATETIME,
+`col57` ENUM ('test1','test2','test3'),
+`col58` NUMERIC,
+`col59` TINYBLOB,
+`col60` VARCHAR (73),
+`col61` MEDIUMTEXT,
+`col62` TINYBLOB,
+`col63` DATETIME,
+`col64` NUMERIC,
+`col65` MEDIUMINT,
+`col66` DATETIME,
+`col67` NUMERIC,
+`col68` TINYINT,
+`col69` VARCHAR (58),
+`col70` DECIMAL,
+`col71` MEDIUMTEXT,
+`col72` DATE,
+`col73` TIME,
+`col74` DOUBLE PRECISION,
+`col75` DECIMAL,
+`col76` MEDIUMBLOB,
+`col77` REAL,
+`col78` YEAR,
+`col79` YEAR,
+`col80` LONGBLOB,
+`col81` BLOB,
+`col82` BIGINT,
+`col83` ENUM ('test1','test2','test3'),
+`col84` NUMERIC,
+`col85` SET ('test1','test2','test3'),
+`col86` MEDIUMTEXT,
+`col87` LONGBLOB,
+`col88` TIME,
+`col89` ENUM ('test1','test2','test3'),
+`col90` DECIMAL,
+`col91` FLOAT,
+`col92` DATETIME,
+`col93` TINYTEXT,
+`col94` TIMESTAMP,
+`col95` TIMESTAMP,
+`col96` TEXT,
+`col97` REAL,
+`col98` VARCHAR (198),
+`col99` TIME,
+`col100` TINYINT,
+`col101` BIGINT,
+`col102` LONGBLOB,
+`col103` LONGBLOB,
+`col104` MEDIUMINT,
+`col105` MEDIUMTEXT,
+`col106` TIMESTAMP,
+`col107` SMALLINT,
+`col108` NUMERIC,
+`col109` DECIMAL,
+`col110` FLOAT,
+`col111` DECIMAL,
+`col112` REAL,
+`col113` TINYTEXT,
+`col114` FLOAT,
+`col115` VARCHAR (7),
+`col116` LONGTEXT,
+`col117` DATE,
+`col118` BIGINT,
+`col119` TEXT,
+`col120` BIGINT,
+`col121` BLOB,
+`col122` CHAR (110),
+`col123` NUMERIC,
+`col124` MEDIUMBLOB,
+`col125` NUMERIC,
+`col126` NUMERIC,
+`col127` BOOL,
+`col128` TIME,
+`col129` TINYBLOB,
+`col130` TINYBLOB,
+`col131` DATE,
+`col132` INT,
+`col133` VARCHAR (123),
+`col134` CHAR (238),
+`col135` VARCHAR (225),
+`col136` LONGTEXT,
+`col137` LONGBLOB,
+`col138` REAL,
+`col139` TINYBLOB,
+`col140` DATETIME,
+`col141` TINYTEXT,
+`col142` LONGBLOB,
+`col143` BIGINT,
+`col144` VARCHAR (236),
+`col145` TEXT,
+`col146` YEAR,
+`col147` DECIMAL,
+`col148` TEXT,
+`col149` MEDIUMBLOB,
+`col150` TINYINT,
+`col151` BOOL,
+`col152` VARCHAR (72),
+`col153` INT,
+`col154` VARCHAR (165),
+`col155` TINYINT,
+`col156` MEDIUMTEXT,
+`col157` DOUBLE PRECISION,
+`col158` TIME,
+`col159` MEDIUMBLOB,
+`col160` LONGBLOB,
+`col161` DATETIME,
+`col162` DOUBLE PRECISION,
+`col163` BLOB,
+`col164` ENUM ('test1','test2','test3'),
+`col165` TIMESTAMP,
+`col166` DATE,
+`col167` TINYBLOB,
+`col168` TINYBLOB,
+`col169` LONGBLOB,
+`col170` DATETIME,
+`col171` BIGINT,
+`col172` VARCHAR (30),
+`col173` LONGTEXT,
+`col174` TIME,
+`col175` FLOAT,
+KEY `idx0` (`col16`,`col156`(139),`col97`,`col120`),
+KEY `idx1` (`col24`,`col0`(108)),
+KEY `idx2` (`col117`,`col173`(34),`col132`,`col82`),
+KEY `idx3` (`col2`(86)),
+KEY `idx4` (`col2`(43)),
+KEY `idx5` (`col83`,`col35`(87),`col111`),
+KEY `idx6` (`col6`,`col134`,`col92`),
+KEY `idx7` (`col56`),
+KEY `idx8` (`col30`,`col53`,`col129`(66)),
+KEY `idx9` (`col53`,`col113`(211),`col32`,`col15`(75)),
+KEY `idx10` (`col34`),
+KEY `idx11` (`col126`),
+KEY `idx12` (`col24`)
+)engine=innodb ROW_FORMAT=COMPRESSED KEY_BLOCK_SIZE=1;
+
+# Creating a table with 179 columns and 46 indexes
+DROP TABLE IF EXISTS `table6`;
+-- error ER_TOO_BIG_ROWSIZE
+--error ER_TOO_BIG_ROWSIZE
+CREATE TABLE IF NOT EXISTS `table6`
+(`col0` ENUM ('test1','test2','test3'),
+`col1` MEDIUMBLOB,
+`col2` MEDIUMBLOB,
+`col3` DATETIME,
+`col4` DATE,
+`col5` YEAR,
+`col6` REAL,
+`col7` NUMERIC,
+`col8` MEDIUMBLOB,
+`col9` TEXT,
+`col10` TIMESTAMP,
+`col11` DOUBLE,
+`col12` DOUBLE,
+`col13` SMALLINT,
+`col14` TIMESTAMP,
+`col15` DECIMAL,
+`col16` DATE,
+`col17` TEXT,
+`col18` LONGBLOB,
+`col19` BIGINT,
+`col20` FLOAT,
+`col21` DATETIME,
+`col22` TINYINT,
+`col23` MEDIUMBLOB,
+`col24` SET ('test1','test2','test3'),
+`col25` TIME,
+`col26` TEXT,
+`col27` LONGTEXT,
+`col28` BIGINT,
+`col29` REAL,
+`col30` YEAR,
+`col31` MEDIUMBLOB,
+`col32` MEDIUMINT,
+`col33` FLOAT,
+`col34` TEXT,
+`col35` DATE,
+`col36` TIMESTAMP,
+`col37` REAL,
+`col38` BLOB,
+`col39` BLOB,
+`col40` BLOB,
+`col41` TINYBLOB,
+`col42` INT,
+`col43` TINYINT,
+`col44` REAL,
+`col45` BIGINT,
+`col46` TIMESTAMP,
+`col47` BLOB,
+`col48` ENUM ('test1','test2','test3'),
+`col49` BOOL,
+`col50` CHAR (109),
+`col51` DOUBLE,
+`col52` DOUBLE PRECISION,
+`col53` ENUM ('test1','test2','test3'),
+`col54` FLOAT,
+`col55` DOUBLE PRECISION,
+`col56` CHAR (166),
+`col57` TEXT,
+`col58` TIME,
+`col59` DECIMAL,
+`col60` TEXT,
+`col61` ENUM ('test1','test2','test3'),
+`col62` LONGTEXT,
+`col63` YEAR,
+`col64` DOUBLE,
+`col65` CHAR (87),
+`col66` DATE,
+`col67` BOOL,
+`col68` MEDIUMBLOB,
+`col69` DATETIME,
+`col70` DECIMAL,
+`col71` TIME,
+`col72` REAL,
+`col73` LONGTEXT,
+`col74` BLOB,
+`col75` REAL,
+`col76` INT,
+`col77` INT,
+`col78` FLOAT,
+`col79` DOUBLE,
+`col80` MEDIUMINT,
+`col81` ENUM ('test1','test2','test3'),
+`col82` VARCHAR (221),
+`col83` BIGINT,
+`col84` TINYINT,
+`col85` BIGINT,
+`col86` FLOAT,
+`col87` MEDIUMBLOB,
+`col88` CHAR (126),
+`col89` MEDIUMBLOB,
+`col90` DATETIME,
+`col91` TINYINT,
+`col92` DOUBLE,
+`col93` NUMERIC,
+`col94` DATE,
+`col95` BLOB,
+`col96` DATETIME,
+`col97` TIME,
+`col98` LONGBLOB,
+`col99` INT,
+`col100` SET ('test1','test2','test3'),
+`col101` TINYBLOB,
+`col102` INT,
+`col103` MEDIUMBLOB,
+`col104` MEDIUMTEXT,
+`col105` FLOAT,
+`col106` TINYBLOB,
+`col107` VARCHAR (26),
+`col108` TINYINT,
+`col109` TIME,
+`col110` TINYBLOB,
+`col111` LONGBLOB,
+`col112` TINYTEXT,
+`col113` FLOAT,
+`col114` TINYINT,
+`col115` NUMERIC,
+`col116` TIME,
+`col117` SET ('test1','test2','test3'),
+`col118` DATE,
+`col119` SMALLINT,
+`col120` BLOB,
+`col121` TINYTEXT,
+`col122` REAL,
+`col123` YEAR,
+`col124` REAL,
+`col125` BOOL,
+`col126` BLOB,
+`col127` REAL,
+`col128` MEDIUMBLOB,
+`col129` TIMESTAMP,
+`col130` LONGBLOB,
+`col131` MEDIUMBLOB,
+`col132` YEAR,
+`col133` YEAR,
+`col134` INT,
+`col135` MEDIUMINT,
+`col136` MEDIUMINT,
+`col137` TINYTEXT,
+`col138` TINYBLOB,
+`col139` BLOB,
+`col140` SET ('test1','test2','test3'),
+`col141` ENUM ('test1','test2','test3'),
+`col142` ENUM ('test1','test2','test3'),
+`col143` TINYTEXT,
+`col144` DATETIME,
+`col145` TEXT,
+`col146` DOUBLE PRECISION,
+`col147` DECIMAL,
+`col148` MEDIUMTEXT,
+`col149` TINYTEXT,
+`col150` SET ('test1','test2','test3'),
+`col151` MEDIUMTEXT,
+`col152` CHAR (126),
+`col153` DOUBLE,
+`col154` CHAR (243),
+`col155` SET ('test1','test2','test3'),
+`col156` SET ('test1','test2','test3'),
+`col157` DATETIME,
+`col158` DOUBLE,
+`col159` NUMERIC,
+`col160` DECIMAL,
+`col161` FLOAT,
+`col162` LONGBLOB,
+`col163` LONGTEXT,
+`col164` INT,
+`col165` TIME,
+`col166` CHAR (27),
+`col167` VARCHAR (63),
+`col168` TEXT,
+`col169` TINYBLOB,
+`col170` TINYBLOB,
+`col171` ENUM ('test1','test2','test3'),
+`col172` INT,
+`col173` TIME,
+`col174` DECIMAL,
+`col175` DOUBLE,
+`col176` MEDIUMBLOB,
+`col177` LONGBLOB,
+`col178` CHAR (43),
+KEY `idx0` (`col131`(219)),
+KEY `idx1` (`col67`,`col122`,`col59`,`col87`(33)),
+KEY `idx2` (`col83`,`col42`,`col57`(152)),
+KEY `idx3` (`col106`(124)),
+KEY `idx4` (`col173`,`col80`,`col165`,`col89`(78)),
+KEY `idx5` (`col174`,`col145`(108),`col23`(228),`col141`),
+KEY `idx6` (`col157`,`col140`),
+KEY `idx7` (`col130`(188),`col15`),
+KEY `idx8` (`col52`),
+KEY `idx9` (`col144`),
+KEY `idx10` (`col155`),
+KEY `idx11` (`col62`(230),`col1`(109)),
+KEY `idx12` (`col151`(24),`col95`(85)),
+KEY `idx13` (`col114`),
+KEY `idx14` (`col42`,`col98`(56),`col146`),
+KEY `idx15` (`col147`,`col39`(254),`col35`),
+KEY `idx16` (`col79`),
+KEY `idx17` (`col65`),
+KEY `idx18` (`col149`(165),`col168`(119),`col32`,`col117`),
+KEY `idx19` (`col64`),
+KEY `idx20` (`col93`),
+KEY `idx21` (`col64`,`col113`,`col104`(182)),
+KEY `idx22` (`col52`,`col111`(189)),
+KEY `idx23` (`col45`),
+KEY `idx24` (`col154`,`col107`,`col110`(159)),
+KEY `idx25` (`col149`(1),`col87`(131)),
+KEY `idx26` (`col58`,`col115`,`col63`),
+KEY `idx27` (`col95`(9),`col0`,`col87`(113)),
+KEY `idx28` (`col92`,`col130`(1)),
+KEY `idx29` (`col151`(129),`col137`(254),`col13`),
+KEY `idx30` (`col49`),
+KEY `idx31` (`col28`),
+KEY `idx32` (`col83`,`col146`),
+KEY `idx33` (`col155`,`col90`,`col17`(245)),
+KEY `idx34` (`col174`,`col169`(44),`col107`),
+KEY `idx35` (`col113`),
+KEY `idx36` (`col52`),
+KEY `idx37` (`col16`,`col120`(190)),
+KEY `idx38` (`col28`),
+KEY `idx39` (`col131`(165)),
+KEY `idx40` (`col135`,`col26`(86)),
+KEY `idx41` (`col69`,`col94`),
+KEY `idx42` (`col105`,`col151`(38),`col97`),
+KEY `idx43` (`col88`),
+KEY `idx44` (`col176`(100),`col42`,`col73`(189),`col94`),
+KEY `idx45` (`col2`(27),`col27`(116))
+)engine=innodb ROW_FORMAT=COMPRESSED KEY_BLOCK_SIZE=1;
+
+SET GLOBAL innodb_file_format=@save_innodb_file_format;
+SET GLOBAL innodb_file_format_check=@save_innodb_file_format_check;
+SET GLOBAL innodb_file_per_table=@save_innodb_file_per_table;
+
+DROP TABLE IF EXISTS table0;
+DROP TABLE IF EXISTS table1;
+DROP TABLE IF EXISTS table2;
+DROP TABLE IF EXISTS table3;
+DROP TABLE IF EXISTS table4;
+DROP TABLE IF EXISTS table5;
+DROP TABLE IF EXISTS table6;
+
diff --git a/mysql-test/t/innodb_bug36172.test b/mysql-test/t/innodb_bug36172.test
new file mode 100644
index 00000000000..ee35d678921
--- /dev/null
+++ b/mysql-test/t/innodb_bug36172.test
@@ -0,0 +1,33 @@
+#
+# Test case for bug 36172
+#
+
+-- source include/not_embedded.inc
+-- source include/have_innodb.inc
+
+SET storage_engine=InnoDB;
+
+# we do not really care about what gets printed, we are only
+# interested in getting success or failure according to our
+# expectations
+
+-- disable_query_log
+-- disable_result_log
+
+SET @save_innodb_file_format=@@global.innodb_file_format;
+SET @save_innodb_file_format_check=@@global.innodb_file_format_check;
+SET @save_innodb_file_per_table=@@global.innodb_file_per_table;
+SET GLOBAL innodb_file_format='Barracuda';
+SET GLOBAL innodb_file_per_table=on;
+
+DROP TABLE IF EXISTS `table0`;
+CREATE TABLE `table0` (   `col0` tinyint(1) DEFAULT NULL,   `col1` tinyint(1) DEFAULT NULL,   `col2` tinyint(4) DEFAULT NULL,   `col3` date DEFAULT NULL,   `col4` time DEFAULT NULL,   `col5` set('test1','test2','test3') DEFAULT NULL,   `col6` time DEFAULT NULL,   `col7` text,   `col8` decimal(10,0) DEFAULT NULL,   `col9` set('test1','test2','test3') DEFAULT NULL,   `col10` float DEFAULT NULL,   `col11` double DEFAULT NULL,   `col12` enum('test1','test2','test3') DEFAULT NULL,   `col13` tinyblob,   `col14` year(4) DEFAULT NULL,   `col15` set('test1','test2','test3') DEFAULT NULL,   `col16` decimal(10,0) DEFAULT NULL,   `col17` decimal(10,0) DEFAULT NULL,   `col18` blob,   `col19` datetime DEFAULT NULL,   `col20` double DEFAULT NULL,   `col21` decimal(10,0) DEFAULT NULL,   `col22` datetime DEFAULT NULL,   `col23` decimal(10,0) DEFAULT NULL,   `col24` decimal(10,0) DEFAULT NULL,   `col25` longtext,   `col26` tinyblob,   `col27` time DEFAULT NULL,   `col28` tinyblob,   `col29` enum('test1','test2','test3') DEFAULT NULL,   `col30` smallint(6) DEFAULT NULL,   `col31` double DEFAULT NULL,   `col32` float DEFAULT NULL,   `col33` char(175) DEFAULT NULL,   `col34` tinytext,   `col35` tinytext,   `col36` tinyblob,   `col37` tinyblob,   `col38` tinytext,   `col39` mediumblob,   `col40` timestamp NOT NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,   `col41` double DEFAULT NULL,   `col42` smallint(6) DEFAULT NULL,   `col43` longblob,   `col44` varchar(80) DEFAULT NULL,   `col45` mediumtext,   `col46` decimal(10,0) DEFAULT NULL,   `col47` bigint(20) DEFAULT NULL,   `col48` date DEFAULT NULL,   `col49` tinyblob,   `col50` date DEFAULT NULL,   `col51` tinyint(1) DEFAULT NULL,   `col52` mediumint(9) DEFAULT NULL,   `col53` float DEFAULT NULL,   `col54` tinyblob,   `col55` longtext,   `col56` smallint(6) DEFAULT NULL,   `col57` enum('test1','test2','test3') DEFAULT NULL,   `col58` datetime DEFAULT NULL,   `col59` mediumtext,   `col60` varchar(232) DEFAULT NULL,   `col61` decimal(10,0) DEFAULT NULL,   `col62` year(4) DEFAULT NULL,   `col63` smallint(6) DEFAULT NULL,   `col64` timestamp NOT NULL DEFAULT '0000-00-00 00:00:00',   `col65` blob,   `col66` longblob,   `col67` int(11) DEFAULT NULL,   `col68` longtext,   `col69` enum('test1','test2','test3') DEFAULT NULL,   `col70` int(11) DEFAULT NULL,   `col71` time DEFAULT NULL,   `col72` timestamp NOT NULL DEFAULT '0000-00-00 00:00:00',   `col73` timestamp NOT NULL DEFAULT '0000-00-00 00:00:00',   `col74` varchar(170) DEFAULT NULL,   `col75` set('test1','test2','test3') DEFAULT NULL,   `col76` tinyblob,   `col77` bigint(20) DEFAULT NULL,   `col78` decimal(10,0) DEFAULT NULL,   `col79` datetime DEFAULT NULL,   `col80` year(4) DEFAULT NULL,   `col81` decimal(10,0) DEFAULT NULL,   `col82` longblob,   `col83` text,   `col84` char(83) DEFAULT NULL,   `col85` decimal(10,0) DEFAULT NULL,   `col86` float DEFAULT NULL,   `col87` int(11) DEFAULT NULL,   `col88` varchar(145) DEFAULT NULL,   `col89` date DEFAULT NULL,   `col90` decimal(10,0) DEFAULT NULL,   `col91` decimal(10,0) DEFAULT NULL,   `col92` mediumblob,   `col93` time DEFAULT NULL,   KEY `idx0` (`col69`,`col90`,`col8`),   KEY `idx1` (`col60`),   KEY `idx2` (`col60`,`col70`,`col74`),   KEY `idx3` (`col22`,`col32`,`col72`,`col30`),   KEY `idx4` (`col29`),   KEY `idx5` (`col19`,`col45`(143)),   KEY `idx6` (`col46`,`col48`,`col5`,`col39`(118)),   KEY `idx7` (`col48`,`col61`),   KEY `idx8` (`col93`),   KEY `idx9` (`col31`),   KEY `idx10` (`col30`,`col21`),   KEY `idx11` (`col67`),   KEY `idx12` (`col44`,`col6`,`col8`,`col38`(226)),   KEY `idx13` (`col71`,`col41`,`col15`,`col49`(88)),   KEY `idx14` (`col78`),   KEY `idx15` (`col63`,`col67`,`col64`),   KEY `idx16` (`col17`,`col86`),   KEY `idx17` (`col77`,`col56`,`col10`,`col55`(24)),   KEY `idx18` (`col62`),   KEY `idx19` (`col31`,`col57`,`col56`,`col53`),   KEY `idx20` (`col46`),   KEY `idx21` (`col83`(54)),   KEY `idx22` (`col51`,`col7`(120)),   KEY `idx23` (`col7`(163),`col31`,`col71`,`col14`) ) ENGINE=InnoDB DEFAULT CHARSET=latin1 ROW_FORMAT=COMPRESSED KEY_BLOCK_SIZE=2;
+insert ignore into `table0` set  `col23` = 7887371.5084383683, `col24` = 4293854615.6906948000, `col25` = 'vitalist', `col26` = 'widespread', `col27` = '3570490', `col28` = 'habitual', `col30` = -5471, `col31` = 4286985783.6771750000, `col32` = 6354540.9826654866, `col33` = 'defoliation', `col34` = 'logarithms', `col35` = 'tegument\'s', `col36` = 'scouting\'s', `col37` = 'intermittency', `col38` = 'elongates', `col39` = 'prophecies', `col40` = '20560103035939', `col41` = 4292809130.0544143000, `col42` = 22057, `col43` = 'Hess\'s', `col44` = 'bandstand', `col45` = 'phenylketonuria', `col46` = 6338767.4018677324, `col47` = 5310247, `col48` = '12592418', `col49` = 'churchman\'s', `col50` = '32226125', `col51` = -58, `col52` = -6207968, `col53` = 1244839.3255104220, `col54` = 'robotized', `col55` = 'monotonous', `col56` = -26909, `col58` = '20720107023550', `col59` = 'suggestiveness\'s', `col60` = 'gemology', `col61` = 4287800670.2229986000, `col62` = '1944', `col63` = -16827, `col64` = '20700107212324', `col65` = 'Nicolais', `col66` = 'apteryx', `col67` = 6935317, `col68` = 'stroganoff', `col70` = 3316430, `col71` = '3277608', `col72` = '19300511045918', `col73` = '20421201003327', `col74` = 'attenuant', `col75` = '15173', `col76` = 'upstroke\'s', `col77` = 8118987, `col78` = 6791516.2735374002, `col79` = '20780701144624', `col80` = '2134', `col81` = 4290682351.3127537000, `col82` = 'unexplainably', `col83` = 'Storm', `col84` = 'Greyso\'s', `col85` = 4289119212.4306774000, `col86` = 7617575.8796655172, `col87` = -6325335, `col88` = 'fondue\'s', `col89` = '40608940', `col90` = 1659421.8093508712, `col91` = 8346904.6584368423, `col92` = 'reloads', `col93` = '5188366';
+CHECK TABLE table0 EXTENDED;
+INSERT IGNORE INTO `table0` SET `col19` = '19940127002709', `col20` = 2383927.9055146948, `col21` = 4293243420.5621204000, `col22` = '20511211123705', `col23` = 4289899778.6573381000, `col24` = 4293449279.0540481000, `col25` = 'emphysemic', `col26` = 'dentally', `col27` = '2347406', `col28` = 'eruct', `col30` = 1222, `col31` = 4294372994.9941406000, `col32` = 4291385574.1173744000, `col33` = 'borrowing\'s', `col34` = 'septics', `col35` = 'ratter\'s', `col36` = 'Kaye', `col37` = 'Florentia', `col38` = 'allium', `col39` = 'barkeep', `col40` = '19510407003441', `col41` = 4293559200.4215522000, `col42` = 22482, `col43` = 'decussate', `col44` = 'Brom\'s', `col45` = 'violated', `col46` = 4925506.4635456400, `col47` = 930549, `col48` = '51296066', `col49` = 'voluminously', `col50` = '29306676', `col51` = -88, `col52` = -2153690, `col53` = 4290250202.1464887000, `col54` = 'expropriation', `col55` = 'Aberdeen\'s', `col56` = 20343, `col58` = '19640415171532', `col59` = 'extern', `col60` = 'Ubana', `col61` = 4290487961.8539081000, `col62` = '2147', `col63` = -24271, `col64` = '20750801194548', `col65` = 'Cunaxa\'s', `col66` = 'pasticcio', `col67` = 2795817, `col68` = 'Indore\'s', `col70` = 6864127, `col71` = '1817832', `col72` = '20540506114211', `col73` = '20040101012300', `col74` = 'rationalized', `col75` = '45522', `col76` = 'indene', `col77` = -6964559, `col78` = 4247535.5266884370, `col79` = '20720416124357', `col80` = '2143', `col81` = 4292060102.4466386000, `col82` = 'striving', `col83` = 'boneblack\'s', `col84` = 'redolent', `col85` = 6489697.9009369183, `col86` = 4287473465.9731131000, `col87` = 7726015, `col88` = 'perplexed', `col89` = '17153791', `col90` = 5478587.1108127078, `col91` = 4287091404.7004304000, `col92` = 'Boulez\'s', `col93` = '2931278';
+CHECK TABLE table0 EXTENDED;
+
+SET GLOBAL innodb_file_format=@save_innodb_file_format;
+SET GLOBAL innodb_file_format_check=@save_innodb_file_format_check;
+SET GLOBAL innodb_file_per_table=@save_innodb_file_per_table;
+DROP TABLE table0;
diff --git a/mysql-test/t/innodb_bug40360.test b/mysql-test/t/innodb_bug40360.test
new file mode 100644
index 00000000000..e88837aab4f
--- /dev/null
+++ b/mysql-test/t/innodb_bug40360.test
@@ -0,0 +1,16 @@
+#
+# Make sure http://bugs.mysql.com/40360 remains fixed.
+#
+
+-- source include/not_embedded.inc
+-- source include/have_innodb.inc
+
+SET TX_ISOLATION='READ-COMMITTED';
+
+# This is the default since MySQL 5.1.29 SET BINLOG_FORMAT='STATEMENT';
+
+CREATE TABLE bug40360 (a INT) engine=innodb;
+
+INSERT INTO bug40360 VALUES (1);
+
+DROP TABLE bug40360;
diff --git a/mysql-test/t/innodb_bug41904.test b/mysql-test/t/innodb_bug41904.test
new file mode 100644
index 00000000000..365c5229adc
--- /dev/null
+++ b/mysql-test/t/innodb_bug41904.test
@@ -0,0 +1,14 @@
+#
+# Make sure http://bugs.mysql.com/41904 remains fixed.
+#
+
+-- source include/not_embedded.inc
+-- source include/have_innodb.inc
+
+CREATE TABLE bug41904 (id INT PRIMARY KEY, uniquecol CHAR(15)) ENGINE=InnoDB;
+
+INSERT INTO bug41904 VALUES (1,NULL), (2,NULL);
+
+CREATE UNIQUE INDEX ui ON bug41904 (uniquecol);
+
+DROP TABLE bug41904;
diff --git a/mysql-test/t/innodb_information_schema.test b/mysql-test/t/innodb_information_schema.test
new file mode 100644
index 00000000000..eaed653854a
--- /dev/null
+++ b/mysql-test/t/innodb_information_schema.test
@@ -0,0 +1,145 @@
+# 
+# Test that user data is correctly "visualized" in
+# INFORMATION_SCHEMA.innodb_locks.lock_data
+#
+
+-- source include/have_innodb.inc
+
+-- disable_query_log
+-- disable_result_log
+
+SET storage_engine=InnoDB;
+
+-- disable_warnings
+DROP TABLE IF EXISTS t_min, t_max;
+-- enable_warnings
+
+let $table_def =
+(
+	c01 TINYINT,
+	c02 TINYINT UNSIGNED,
+	c03 SMALLINT,
+	c04 SMALLINT UNSIGNED,
+	c05 MEDIUMINT,
+	c06 MEDIUMINT UNSIGNED,
+	c07 INT,
+	c08 INT UNSIGNED,
+	c09 BIGINT,
+	c10 BIGINT UNSIGNED,
+	PRIMARY KEY(c01, c02, c03, c04, c05, c06, c07, c08, c09, c10)
+);
+
+-- eval CREATE TABLE t_min $table_def;
+INSERT INTO t_min VALUES
+(-128, 0,
+ -32768, 0,
+ -8388608, 0,
+ -2147483648, 0,
+ -9223372036854775808, 0);
+
+-- eval CREATE TABLE t_max $table_def;
+INSERT INTO t_max VALUES
+(127, 255,
+ 32767, 65535,
+ 8388607, 16777215,
+ 2147483647, 4294967295,
+ 9223372036854775807, 18446744073709551615);
+
+CREATE TABLE ```t'\"_str` (
+	c1 VARCHAR(32),
+	c2 VARCHAR(32),
+	c3 VARCHAR(32),
+	c4 VARCHAR(32),
+	c5 VARCHAR(32),
+	c6 VARCHAR(32),
+	c7 VARCHAR(32),
+	PRIMARY KEY(c1, c2, c3, c4, c5, c6, c7)
+);
+INSERT INTO ```t'\"_str` VALUES
+('1', 'abc', '''abc', 'abc''', 'a''bc', 'a''bc''', '''abc''''');
+INSERT INTO ```t'\"_str` VALUES
+('2', 'abc', '"abc', 'abc"', 'a"bc', 'a"bc"', '"abc""');
+INSERT INTO ```t'\"_str` VALUES
+('3', 'abc', '\\abc', 'abc\\', 'a\\bc', 'a\\bc\\', '\\abc\\\\');
+INSERT INTO ```t'\"_str` VALUES
+('4', 'abc', 0x00616263, 0x61626300, 0x61006263, 0x6100626300, 0x610062630000);
+
+-- connect (con_lock,localhost,root,,)
+-- connect (con_min_trylock,localhost,root,,)
+-- connect (con_max_trylock,localhost,root,,)
+-- connect (con_str_insert_supremum,localhost,root,,)
+-- connect (con_str_lock_row1,localhost,root,,)
+-- connect (con_str_lock_row2,localhost,root,,)
+-- connect (con_str_lock_row3,localhost,root,,)
+-- connect (con_str_lock_row4,localhost,root,,)
+-- connect (con_verify_innodb_locks,localhost,root,,)
+
+-- connection con_lock
+SET autocommit=0;
+SELECT * FROM t_min FOR UPDATE;
+SELECT * FROM t_max FOR UPDATE;
+SELECT * FROM ```t'\"_str` FOR UPDATE;
+
+-- connection con_min_trylock
+-- send
+SELECT * FROM t_min FOR UPDATE;
+
+-- connection con_max_trylock
+-- send
+SELECT * FROM t_max FOR UPDATE;
+
+-- connection con_str_insert_supremum
+-- send
+INSERT INTO ```t'\"_str` VALUES
+('z', 'z', 'z', 'z', 'z', 'z', 'z');
+
+-- connection con_str_lock_row1
+-- send
+SELECT * FROM ```t'\"_str` WHERE c1 = '1' FOR UPDATE;
+
+-- connection con_str_lock_row2
+-- send
+SELECT * FROM ```t'\"_str` WHERE c1 = '2' FOR UPDATE;
+
+-- connection con_str_lock_row3
+-- send
+SELECT * FROM ```t'\"_str` WHERE c1 = '3' FOR UPDATE;
+
+-- connection con_str_lock_row4
+-- send
+SELECT * FROM ```t'\"_str` WHERE c1 = '4' FOR UPDATE;
+
+# Give time to the above 2 queries to execute before continuing.
+# Without this sleep it sometimes happens that the SELECT from innodb_locks
+# executes before some of them, resulting in less than expected number
+# of rows being selected from innodb_locks.
+-- sleep 0.1
+
+-- enable_result_log
+-- connection con_verify_innodb_locks
+SELECT lock_mode, lock_type, lock_table, lock_index, lock_rec, lock_data
+FROM INFORMATION_SCHEMA.INNODB_LOCKS ORDER BY lock_data;
+
+SELECT lock_table,COUNT(*) FROM INFORMATION_SCHEMA.INNODB_LOCKS
+GROUP BY lock_table;
+
+set @save_sql_mode = @@sql_mode;
+SET SQL_MODE='ANSI_QUOTES';
+SELECT lock_table,COUNT(*) FROM INFORMATION_SCHEMA.INNODB_LOCKS
+GROUP BY lock_table;
+SET @@sql_mode=@save_sql_mode;
+-- disable_result_log
+
+-- connection default
+
+-- disconnect con_lock
+-- disconnect con_min_trylock
+-- disconnect con_max_trylock
+-- disconnect con_str_insert_supremum
+-- disconnect con_str_lock_row1
+-- disconnect con_str_lock_row2
+-- disconnect con_str_lock_row3
+-- disconnect con_str_lock_row4
+-- disconnect con_verify_innodb_locks
+
+DROP TABLE t_min, t_max, ```t'\"_str`;
diff --git a/mysql-test/t/innodb_xtradb_bug317074.test b/mysql-test/t/innodb_xtradb_bug317074.test
new file mode 100644
index 00000000000..554acc7196c
--- /dev/null
+++ b/mysql-test/t/innodb_xtradb_bug317074.test
@@ -0,0 +1,45 @@
+-- source include/have_innodb.inc
+
+SET @save_innodb_file_format=@@global.innodb_file_format;
+SET @save_innodb_file_format_check=@@global.innodb_file_format_check;
+SET @save_innodb_file_per_table=@@global.innodb_file_per_table;
+SET GLOBAL innodb_file_format='Barracuda';
+SET GLOBAL innodb_file_per_table=ON;
+
+-- disable_query_log
+-- disable_result_log
+
+DROP TABLE IF EXISTS `test1`;
+CREATE TABLE IF NOT EXISTS `test1` (
+ `a` int primary key auto_increment,
+ `b` int default 0,
+ `c` char(100) default 'testtest'
+) ENGINE=InnoDB ROW_FORMAT=COMPRESSED KEY_BLOCK_SIZE=8;
+
+delimiter |;
+CREATE PROCEDURE insert_many(p1 int)
+BEGIN
+SET @x = 0;
+SET @y = 0;
+REPEAT
+  insert into test1 set b=1;
+  SET @x = @x + 1;
+  SET @y = @y + 1;
+  IF @y >= 100 THEN
+    commit;
+    SET @y = 0;
+  END IF;
+UNTIL @x >= p1 END REPEAT;
+END|
+delimiter ;|
+call insert_many(100000);
+DROP PROCEDURE insert_many;
+
+# The bug is hangup at the following statement
+ALTER TABLE test1 ENGINE=MyISAM;
+
+SET GLOBAL innodb_file_format=@save_innodb_file_format;
+SET GLOBAL innodb_file_format_check=@save_innodb_file_format_check;
+SET GLOBAL innodb_file_per_table=@save_innodb_file_per_table;
+
+DROP TABLE test1;
diff --git a/mysql-test/t/mysqlbinlog_row_big.test b/mysql-test/t/mysqlbinlog_row_big.test
index 14e818e7214..f4e095af021 100644
--- a/mysql-test/t/mysqlbinlog_row_big.test
+++ b/mysql-test/t/mysqlbinlog_row_big.test
@@ -23,6 +23,16 @@
 # This is a big test.
 --source include/big_test.inc
 
+--echo #
+--echo # We need big packets.
+--echo #
+connect (con1, localhost, root,,);
+connection con1;
+SET @old_global_max_allowed_packet=@@global.max_allowed_packet;
+SET @@global.max_allowed_packet= 1024*1024*1024;
+connect (con2, localhost, root,,);
+connection con2;
+
 --echo #
 --echo # Preparatory cleanup.
 --echo #
@@ -35,11 +45,6 @@ DROP TABLE IF EXISTS t1;
 --echo #
 SET timestamp=1000000000;
 
---echo #
---echo # We need big packets.
---echo #
-SET @@global.max_allowed_packet= 1024*1024*1024;
-
 --echo #
 --echo # Delete all existing binary logs.
 --echo #
@@ -122,9 +127,14 @@ let $MYSQLD_DATADIR= `select @@datadir`;
 --echo #
 DROP TABLE t1;
 
+connection con1;
+SET @@global.max_allowed_packet=@old_global_max_allowed_packet;
+
 --echo remove_file \$MYSQLTEST_VARDIR/$mysqlbinlog_output
 #
 # NOTE: If you want to see the *huge* mysqlbinlog output, disable next line:
 #
 --remove_file $MYSQLTEST_VARDIR/$mysqlbinlog_output
 
+disconnect con1
+disconnect con2
diff --git a/mysql-test/t/partition_innodb.test b/mysql-test/t/partition_innodb.test
index 2abbceffbb0..722fad2919e 100644
--- a/mysql-test/t/partition_innodb.test
+++ b/mysql-test/t/partition_innodb.test
@@ -27,14 +27,14 @@ UPDATE t1 SET DATA = data*2 WHERE id = 3;
 
 # grouping/referencing in replace_regex is very slow on long strings,
 # removing all before/after the interesting row before grouping/referencing
---replace_regex /.*---TRANSACTION [0-9]+ [0-9]+, .*, OS thread id [0-9]+// /MySQL thread id [0-9]+, query id [0-9]+ .*// /.*([0-9]+ lock struct\(s\)), heap size [0-9]+, ([0-9]+ row lock\(s\)).*/\1 \2/
+--replace_regex /.*LIST OF TRANSACTIONS FOR EACH SESSION:// /MySQL thread id [0-9]+, query id [0-9]+ .*// /.*([0-9]+ lock struct\(s\)), heap size [0-9]+, ([0-9]+ row lock\(s\)).*/\1 \2/
 SHOW ENGINE InnoDB STATUS;
 
 UPDATE t1 SET data = data*2 WHERE data = 2;
 
 # grouping/referencing in replace_regex is very slow on long strings,
 # removing all before/after the interesting row before grouping/referencing
---replace_regex /.*---TRANSACTION [0-9]+ [0-9]+, .*, OS thread id [0-9]+// /MySQL thread id [0-9]+, query id [0-9]+ .*// /.*([0-9]+ lock struct\(s\)), heap size [0-9]+, ([0-9]+ row lock\(s\)).*/\1 \2/
+--replace_regex /.*LIST OF TRANSACTIONS FOR EACH SESSION:// /MySQL thread id [0-9]+, query id [0-9]+ .*// /.*([0-9]+ lock struct\(s\)), heap size [0-9]+, ([0-9]+ row lock\(s\)).*/\1 \2/
 SHOW ENGINE InnoDB STATUS;
 
 SET @@session.tx_isolation = @old_tx_isolation;
diff --git a/mysql-test/t/type_bit_innodb.test b/mysql-test/t/type_bit_innodb.test
index e7e66da8927..dc5947e25e1 100644
--- a/mysql-test/t/type_bit_innodb.test
+++ b/mysql-test/t/type_bit_innodb.test
@@ -41,7 +41,7 @@ create table t1 (a bit) engine=innodb;
 insert into t1 values (b'0'), (b'1'), (b'000'), (b'100'), (b'001');
 select hex(a) from t1;
 # It is not deterministic which duplicate will be seen first
---replace_regex /(.*Duplicate entry )'.*'( for key.*)/\1''\2/
+--replace_regex /entry '(.*)' for/entry '' for/
 --error ER_DUP_ENTRY
 alter table t1 add unique (a);
 drop table t1;
diff --git a/mysql-test/t/variables-big.test b/mysql-test/t/variables-big.test
index 2dacabfece7..efefa055eab 100644
--- a/mysql-test/t/variables-big.test
+++ b/mysql-test/t/variables-big.test
@@ -8,19 +8,20 @@
 # Bug #27322  	failure to allocate transaction_prealloc_size causes crash
 #
 
+set @pid_temp = (select ID from information_schema.processlist);
 set session transaction_prealloc_size=1024*1024*1024*1;
 --replace_column 1 #
-show processlist;
+select @pid_temp = (select ID from information_schema.processlist) as 'TRUE';
 set session transaction_prealloc_size=1024*1024*1024*2;
 --replace_column 1 #
-show processlist;
+select @pid_temp = (select ID from information_schema.processlist) as 'TRUE';
 --replace_column 1 #
 set session transaction_prealloc_size=1024*1024*1024*3;
 --replace_column 1 #
-show processlist;
+select @pid_temp = (select ID from information_schema.processlist) as 'TRUE';
 set session transaction_prealloc_size=1024*1024*1024*4;
 --replace_column 1 #
-show processlist;
+select @pid_temp = (select ID from information_schema.processlist) as 'TRUE';
 set session transaction_prealloc_size=1024*1024*1024*5;
 --replace_column 1 #
-show processlist;
+select @pid_temp = (select ID from information_schema.processlist) as 'TRUE';
diff --git a/mysys/mf_iocache2.c b/mysys/mf_iocache2.c
index 728501e6c50..87346f26b60 100644
--- a/mysys/mf_iocache2.c
+++ b/mysys/mf_iocache2.c
@@ -464,3 +464,52 @@ process_flags:
 err:
   return (size_t) -1;
 }
+
+int init_strvar_from_file(char *var, int max_size, IO_CACHE *f,
+                          const char *default_val)
+{
+  uint length;
+  DBUG_ENTER("init_strvar_from_file");
+
+  if ((length=my_b_gets(f,var, max_size)))
+  {
+    char* last_p = var + length -1;
+    if (*last_p == '\n')
+      *last_p = 0; /* if we stopped on newline, kill it */
+    else
+    {
+      /*
+        If we truncated a line or stopped on last char, remove all chars
+        up to and including newline.
+      */
+      int c;
+      while (((c=my_b_get(f)) != '\n' && c != my_b_EOF))
+        ;
+    }
+    DBUG_RETURN(0);
+  }
+  else if (default_val)
+  {
+    strmake(var,  default_val, max_size-1);
+    DBUG_RETURN(0);
+  }
+  DBUG_RETURN(1);
+}
+
+int init_intvar_from_file(int* var, IO_CACHE* f, int default_val)
+{
+  char buf[32];
+  DBUG_ENTER("init_intvar_from_file");
+
+  if (my_b_gets(f, buf, sizeof(buf)))
+  {
+    *var = atoi(buf);
+    DBUG_RETURN(0);
+  }
+  else if (default_val)
+  {
+    *var = default_val;
+    DBUG_RETURN(0);
+  }
+  DBUG_RETURN(1);
+}
diff --git a/mysys/thr_mutex.c b/mysys/thr_mutex.c
index 80f21e53473..c46b68761db 100644
--- a/mysys/thr_mutex.c
+++ b/mysys/thr_mutex.c
@@ -149,6 +149,35 @@ static inline void remove_from_active_list(safe_mutex_t *mp)
   mp->prev= mp->next= 0;
 }
 
+/*
+  We initialise the hashes for deadlock detection lazily.
+  This greatly helps with performance when lots of mutexes are initiased but
+  only a few of them are actually used (eg. XtraDB).
+*/
+static int safe_mutex_lazy_init_deadlock_detection(safe_mutex_t *mp)
+{
+  if (!my_multi_malloc(MY_FAE | MY_WME,
+                       &mp->locked_mutex, sizeof(*mp->locked_mutex),
+                       &mp->used_mutex, sizeof(*mp->used_mutex), NullS))
+  {
+    return 1;                                   /* Error */
+  }
+
+  pthread_mutex_lock(&THR_LOCK_mutex);
+  mp->id= ++safe_mutex_id;
+  pthread_mutex_unlock(&THR_LOCK_mutex);
+  hash_init(mp->locked_mutex, &my_charset_bin,
+            1000,
+            offsetof(safe_mutex_deadlock_t, id),
+            sizeof(mp->id),
+            0, 0, HASH_UNIQUE);
+  hash_init(mp->used_mutex, &my_charset_bin,
+            1000,
+            offsetof(safe_mutex_t, id),
+            sizeof(mp->id),
+            0, 0, HASH_UNIQUE);
+  return 0;
+}
 
 int safe_mutex_init(safe_mutex_t *mp,
 		    const pthread_mutexattr_t *attr __attribute__((unused)),
@@ -167,35 +196,8 @@ int safe_mutex_init(safe_mutex_t *mp,
   mp->line= line;
   /* Skip the very common '&' prefix from the autogenerated name */
   mp->name= name[0] == '&' ? name + 1 : name;
+  /* Deadlock detection is initialised only lazily, on first use. */
 
-  if (safe_mutex_deadlock_detector && !( my_flags & MYF_NO_DEADLOCK_DETECTION))
-  {
-    if (!my_multi_malloc(MY_FAE | MY_WME,
-                         &mp->locked_mutex, sizeof(*mp->locked_mutex),
-                         &mp->used_mutex, sizeof(*mp->used_mutex), NullS))
-    {
-      /* Disable deadlock handling for this mutex */
-      my_flags|= MYF_NO_DEADLOCK_DETECTION;
-    }
-    else
-    {
-      pthread_mutex_lock(&THR_LOCK_mutex);
-      mp->id= ++safe_mutex_id;
-      pthread_mutex_unlock(&THR_LOCK_mutex);
-      hash_init(mp->locked_mutex, &my_charset_bin,
-                1000,
-                offsetof(safe_mutex_deadlock_t, id),
-                sizeof(mp->id),
-                0, 0, HASH_UNIQUE);
-      hash_init(mp->used_mutex, &my_charset_bin,
-                1000,
-                offsetof(safe_mutex_t, id),
-                sizeof(mp->id),
-                0, 0, HASH_UNIQUE);
-    }
-  }
-  else
-    my_flags|= MYF_NO_DEADLOCK_DETECTION;
   mp->create_flags= my_flags;
 
 #ifdef SAFE_MUTEX_DETECT_DESTROY
@@ -310,7 +312,8 @@ int safe_mutex_lock(safe_mutex_t *mp, myf my_flags, const char *file,
   /* Deadlock detection */
 
   mp->prev= mp->next= 0;
-  if (!(mp->active_flags & (MYF_TRY_LOCK | MYF_NO_DEADLOCK_DETECTION)))
+  if (!(mp->active_flags & (MYF_TRY_LOCK | MYF_NO_DEADLOCK_DETECTION)) &&
+      (mp->used_mutex != NULL || !safe_mutex_lazy_init_deadlock_detection(mp)))
   {
     safe_mutex_t **mutex_in_use= my_thread_var_mutex_in_use();
 
@@ -643,7 +646,7 @@ int safe_mutex_destroy(safe_mutex_t *mp, const char *file, uint line)
 void safe_mutex_free_deadlock_data(safe_mutex_t *mp)
 {
   /* Free all entries that points to this one */
-  if (!(mp->create_flags & MYF_NO_DEADLOCK_DETECTION))
+  if (!(mp->create_flags & MYF_NO_DEADLOCK_DETECTION) && mp->used_mutex != NULL)
   {
     pthread_mutex_lock(&THR_LOCK_mutex);
     my_hash_iterate(mp->used_mutex,
diff --git a/sql/log_event.cc b/sql/log_event.cc
index 87917b6ac86..485dc2fddc3 100644
--- a/sql/log_event.cc
+++ b/sql/log_event.cc
@@ -9313,3 +9313,29 @@ st_print_event_info::st_print_event_info()
   open_cached_file(&body_cache, NULL, NULL, 0, flags);
 }
 #endif
+
+#if defined(MYSQL_SERVER)
+/*
+  Access to the current replication position.
+
+  There is a dummy replacement for this in the embedded library that returns
+  FALSE; this is used by XtraDB to allow it to access replication stuff while
+  still being able to use the same plugin in both stand-alone and embedded.
+*/
+bool rpl_get_position_info(const char **log_file_name, ulonglong *log_pos,
+                           const char **group_relay_log_name,
+                           ulonglong *relay_log_pos)
+{
+#if defined(EMBEDDED_LIBRARY) || !defined(HAVE_REPLICATION)
+  return FALSE;
+#else
+  const Relay_log_info *rli= &(active_mi->rli);
+  *log_file_name= rli->group_master_log_name;
+  *log_pos= rli->group_master_log_pos +
+    (rli->future_event_relay_log_pos - rli->group_relay_log_pos);
+  *group_relay_log_name= rli->group_relay_log_name;
+  *relay_log_pos= rli->future_event_relay_log_pos;
+  return TRUE;
+#endif
+}
+#endif
diff --git a/sql/log_event.h b/sql/log_event.h
index bda53da8ab0..45dcf297697 100644
--- a/sql/log_event.h
+++ b/sql/log_event.h
@@ -3917,6 +3917,10 @@ static inline bool copy_event_cache_to_file_and_reinit(IO_CACHE *cache,
     reinit_io_cache(cache, WRITE_CACHE, 0, FALSE, TRUE);
 }
 
+bool rpl_get_position_info(const char **log_file_name, ulonglong *log_pos,
+                           const char **group_relay_log_name,
+                           ulonglong *relay_log_pos);
+
 /**
   @} (end of group Replication)
 */
diff --git a/sql/rpl_mi.cc b/sql/rpl_mi.cc
index cb8b0e02ef9..b8af53849f1 100644
--- a/sql/rpl_mi.cc
+++ b/sql/rpl_mi.cc
@@ -22,11 +22,6 @@
 #ifdef HAVE_REPLICATION
 
 
-// Defined in slave.cc
-int init_intvar_from_file(int* var, IO_CACHE* f, int default_val);
-int init_strvar_from_file(char *var, int max_size, IO_CACHE *f,
-			  const char *default_val);
-
 Master_info::Master_info()
   :Slave_reporting_capability("I/O"),
    ssl(0), ssl_verify_server_cert(0), fd(-1), io_thd(0), inited(0),
diff --git a/sql/rpl_rli.cc b/sql/rpl_rli.cc
index 82628143bea..3fbe9b16d23 100644
--- a/sql/rpl_rli.cc
+++ b/sql/rpl_rli.cc
@@ -23,11 +23,6 @@
 
 static int count_relay_log_space(Relay_log_info* rli);
 
-// Defined in slave.cc
-int init_intvar_from_file(int* var, IO_CACHE* f, int default_val);
-int init_strvar_from_file(char *var, int max_size, IO_CACHE *f,
-			  const char *default_val);
-
 
 Relay_log_info::Relay_log_info()
   :Slave_reporting_capability("SQL"),
diff --git a/sql/slave.cc b/sql/slave.cc
index ec86a4c48e0..b6138078e35 100644
--- a/sql/slave.cc
+++ b/sql/slave.cc
@@ -755,57 +755,6 @@ const char *print_slave_db_safe(const char* db)
   DBUG_RETURN((db ? db : ""));
 }
 
-int init_strvar_from_file(char *var, int max_size, IO_CACHE *f,
-                                 const char *default_val)
-{
-  uint length;
-  DBUG_ENTER("init_strvar_from_file");
-
-  if ((length=my_b_gets(f,var, max_size)))
-  {
-    char* last_p = var + length -1;
-    if (*last_p == '\n')
-      *last_p = 0; // if we stopped on newline, kill it
-    else
-    {
-      /*
-        If we truncated a line or stopped on last char, remove all chars
-        up to and including newline.
-      */
-      int c;
-      while (((c=my_b_get(f)) != '\n' && c != my_b_EOF))
-        ;
-    }
-    DBUG_RETURN(0);
-  }
-  else if (default_val)
-  {
-    strmake(var,  default_val, max_size-1);
-    DBUG_RETURN(0);
-  }
-  DBUG_RETURN(1);
-}
-
-
-int init_intvar_from_file(int* var, IO_CACHE* f, int default_val)
-{
-  char buf[32];
-  DBUG_ENTER("init_intvar_from_file");
-
-
-  if (my_b_gets(f, buf, sizeof(buf)))
-  {
-    *var = atoi(buf);
-    DBUG_RETURN(0);
-  }
-  else if (default_val)
-  {
-    *var = default_val;
-    DBUG_RETURN(0);
-  }
-  DBUG_RETURN(1);
-}
-
 /*
   Note that we rely on the master's version (3.23, 4.0.14 etc) instead of
   relying on the binlog's version. This is not perfect: imagine an upgrade
diff --git a/storage/innobase/plug.in b/storage/innobase/plug.in.disabled
similarity index 100%
rename from storage/innobase/plug.in
rename to storage/innobase/plug.in.disabled
diff --git a/storage/xtradb/CMakeLists.txt b/storage/xtradb/CMakeLists.txt
new file mode 100644
index 00000000000..be4206e0076
--- /dev/null
+++ b/storage/xtradb/CMakeLists.txt
@@ -0,0 +1,97 @@
+# Copyright (C) 2006 MySQL AB
+# 
+# This program is free software; you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation; version 2 of the License.
+# 
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+# 
+# You should have received a copy of the GNU General Public License
+# along with this program; if not, write to the Free Software
+# Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
+
+SET(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -DSAFEMALLOC -DSAFE_MUTEX")
+SET(CMAKE_C_FLAGS_DEBUG "${CMAKE_C_FLAGS_DEBUG} -DSAFEMALLOC -DSAFE_MUTEX")
+ADD_DEFINITIONS(-D_WIN32 -D_LIB)
+
+# Bug 19424 - InnoDB: Possibly a memory overrun of the buffer being freed (64-bit Visual C)
+# Removing Win64 compiler optimizations for all innodb/mem/* files.
+IF(CMAKE_GENERATOR MATCHES "Visual Studio" AND CMAKE_SIZEOF_VOID_P MATCHES 8)
+  SET_SOURCE_FILES_PROPERTIES(${CMAKE_SOURCE_DIR}/storage/xtradb/mem/mem0mem.c
+                              ${CMAKE_SOURCE_DIR}/storage/xtradb/mem/mem0pool.c
+                              PROPERTIES COMPILE_FLAGS -Od)
+ENDIF(CMAKE_GENERATOR MATCHES "Visual Studio" AND CMAKE_SIZEOF_VOID_P MATCHES 8)
+
+INCLUDE_DIRECTORIES(${CMAKE_SOURCE_DIR}/include ${CMAKE_SOURCE_DIR}/zlib
+                    ${CMAKE_SOURCE_DIR}/storage/xtradb/include
+                    ${CMAKE_SOURCE_DIR}/storage/xtradb/handler
+                    ${CMAKE_SOURCE_DIR}/sql
+                    ${CMAKE_SOURCE_DIR}/regex
+                    ${CMAKE_SOURCE_DIR}/extra/yassl/include)
+
+SET(INNOBASE_SOURCES  btr/btr0btr.c btr/btr0cur.c btr/btr0pcur.c btr/btr0sea.c 
+					 buf/buf0buddy.c buf/buf0buf.c buf/buf0flu.c buf/buf0lru.c buf/buf0rea.c 
+					 data/data0data.c data/data0type.c 
+					 dict/dict0boot.c dict/dict0crea.c dict/dict0dict.c dict/dict0load.c dict/dict0mem.c 
+					 dyn/dyn0dyn.c 
+					 eval/eval0eval.c eval/eval0proc.c 
+					 fil/fil0fil.c 
+					 fsp/fsp0fsp.c
+					 fut/fut0fut.c fut/fut0lst.c 
+					 ha/ha0ha.c ha/hash0hash.c ha/ha0storage.c 
+					 ibuf/ibuf0ibuf.c 
+					 pars/lexyy.c pars/pars0grm.c pars/pars0opt.c pars/pars0pars.c pars/pars0sym.c 
+					 lock/lock0lock.c lock/lock0iter.c 
+					 log/log0log.c log/log0recv.c 
+					 mach/mach0data.c 
+					 mem/mem0mem.c mem/mem0pool.c 
+					 mtr/mtr0log.c mtr/mtr0mtr.c 
+					 os/os0file.c os/os0proc.c os/os0sync.c os/os0thread.c 
+					 page/page0cur.c page/page0page.c page/page0zip.c 
+					 que/que0que.c 
+					 handler/ha_innodb.cc handler/handler0alter.cc handler/i_s.cc handler/mysql_addons.cc 
+					 read/read0read.c 
+					 rem/rem0cmp.c rem/rem0rec.c
+					 row/row0ext.c row/row0ins.c row/row0merge.c row/row0mysql.c 
+					 row/row0purge.c row/row0row.c row/row0sel.c row/row0uins.c 
+					 row/row0umod.c row/row0undo.c row/row0upd.c row/row0vers.c 
+					 srv/srv0que.c srv/srv0srv.c srv/srv0start.c 
+					 sync/sync0arr.c sync/sync0rw.c sync/sync0sync.c 
+					 thr/thr0loc.c 
+					 trx/trx0i_s.c trx/trx0purge.c trx/trx0rec.c trx/trx0roll.c trx/trx0rseg.c 
+					 trx/trx0sys.c trx/trx0trx.c trx/trx0undo.c 
+					 usr/usr0sess.c 
+					 ut/ut0byte.c ut/ut0dbg.c ut/ut0mem.c ut/ut0rnd.c ut/ut0ut.c ut/ut0vec.c ut/ut0list.c ut/ut0wqueue.c)
+
+IF(NOT SOURCE_SUBLIBS)
+  ADD_LIBRARY(innobase ${INNOBASE_SOURCES})
+  ADD_DEPENDENCIES(innobase GenError)
+  SET_TARGET_PROPERTIES(innobase PROPERTIES COMPILE_FLAGS "-DMYSQL_SERVER")
+
+  IF(INNODB_DYNAMIC_PLUGIN)
+    # The dynamic plugin requires CMake 2.6.0 or later. Otherwise, the /DELAYLOAD property
+    # will not be set
+    CMAKE_MINIMUM_REQUIRED(VERSION 2.6.0 FATAL_ERROR)
+    ADD_LIBRARY(ha_innodb SHARED ${INNOBASE_SOURCES} ha_innodb.def handler/win_delay_loader.cc)
+    ADD_DEPENDENCIES(ha_innodb GenError mysqld)
+    # If build type is not specified as Release, default to Debug
+    # This is a workaround to a problem in CMake 2.6, which does not
+    # set the path of mysqld.lib correctly
+    IF(CMAKE_BUILD_TYPE MATCHES Release)
+       SET(CMAKE_BUILD_TYPE "Release")
+    ELSE(CMAKE_BUILD_TYPE MATCHES Release)
+       SET(CMAKE_BUILD_TYPE "Debug") 
+    ENDIF(CMAKE_BUILD_TYPE MATCHES Release)
+    TARGET_LINK_LIBRARIES(ha_innodb strings zlib)
+    TARGET_LINK_LIBRARIES(ha_innodb ${CMAKE_SOURCE_DIR}/sql/${CMAKE_BUILD_TYPE}/mysqld.lib)
+    SET_TARGET_PROPERTIES(ha_innodb PROPERTIES OUTPUT_NAME ha_innodb)
+    SET_TARGET_PROPERTIES(ha_innodb PROPERTIES LINK_FLAGS "/MAP /MAPINFO:EXPORTS")
+    SET_TARGET_PROPERTIES(ha_innodb PROPERTIES LINK_FLAGS "/ENTRY:\"_DllMainCRTStartup@12\"")
+    SET_TARGET_PROPERTIES(ha_innodb PROPERTIES COMPILE_FLAGS "-DMYSQL_DYNAMIC_PLUGIN")
+    SET_TARGET_PROPERTIES(ha_innodb PROPERTIES LINK_FLAGS "/DELAYLOAD:mysqld.exe")
+  ENDIF(INNODB_DYNAMIC_PLUGIN)
+
+ENDIF(NOT SOURCE_SUBLIBS)
diff --git a/storage/xtradb/COPYING.Google b/storage/xtradb/COPYING.Google
new file mode 100644
index 00000000000..5ade2b0e381
--- /dev/null
+++ b/storage/xtradb/COPYING.Google
@@ -0,0 +1,30 @@
+Portions of this software contain modifications contributed by Google, Inc.
+These contributions are used with the following license:
+
+Copyright (c) 2008, Google Inc. All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions
+are met:
+      * Redistributions of source code must retain the above copyright
+        notice, this list of conditions and the following disclaimer.
+      * Redistributions in binary form must reproduce the above
+        copyright notice, this list of conditions and the following
+        disclaimer in the documentation and/or other materials
+        provided with the distribution.
+      * Neither the name of the Google Inc. nor the names of its
+        contributors may be used to endorse or promote products
+        derived from this software without specific prior written
+        permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
diff --git a/storage/xtradb/ChangeLog b/storage/xtradb/ChangeLog
new file mode 100644
index 00000000000..f662b02eea5
--- /dev/null
+++ b/storage/xtradb/ChangeLog
@@ -0,0 +1,775 @@
+2009-03-05	The InnoDB Team
+
+	* handler/ha_innodb.cc, mysql-test/innodb-autoinc.result,
+	mysql-test/innodb-autoinc.test:
+	Fix Bug#43203 Overflow from auto incrementing causes server segv
+
+2009-02-25	The InnoDB Team
+
+	* handler/ha_innodb.cc, mysql-test/innodb-autoinc.result,
+	mysql-test/innodb-autoinc.test:
+	Fix Bug#42714 AUTO_INCREMENT errors in 5.1.31
+
+2009-02-23	The InnoDB Team
+
+	* btr/btr0cur.c:
+	Fix Bug#43043 Crash on BLOB delete operation
+
+2009-02-20	The InnoDB Team
+
+	* handler/ha_innodb.cc:
+	Make innodb_use_sys_malloc=ON the default.
+
+2009-02-20	The InnoDB Team
+
+	* handler/ha_innodb.cc, mysql-test/innodb-autoinc.result,
+	mysql-test/innodb-autoinc.test:
+	Fix Bug#42400 InnoDB autoinc code can't handle floating-point columns
+
+2009-02-18	The InnoDB Team
+
+	* include/ut0mem.h, os/os0proc.c, ut/ut0mem.c:
+	Protect ut_total_allocated_memory with ut_list_mutex in
+	os_mem_alloc_large() and os_mem_free_large(). The lack of this mutex
+	protection could cause an assertion failure during fast index
+	creation. Also, add UNIV_MEM_ALLOC and UNIV_MEM_FREE instrumentation
+	to os_mem_alloc_large() and os_mem_free_large(), so that Valgrind can
+	detect more errors.
+
+2009-02-11	The InnoDB Team
+
+	* handler/ha_innodb.cc:
+	Make innodb_thread_concurrency=0 the default. The old default value
+	was 8. A non-zero setting may be useful when InnoDB is showing severe
+	scalability problems under multiple concurrent connections.
+
+2009-02-10	The InnoDB Team
+
+	* handler/ha_innodb.cc, handler/ha_innodb.h:
+	Fix Bug#41676 Table names are case insensitive in locking
+
+2009-02-10	The InnoDB Team
+
+	* mem/mem0dbg.c, mem/mem0mem.c, mem/mem0pool.c:
+	When innodb_use_sys_malloc is set, ignore
+	innodb_additional_mem_pool_size, because nothing will be allocated
+	from mem_comm_pool.
+
+2009-02-10	The InnoDB Team
+
+	* ut/ut0mem.c:
+	Map ut_malloc_low(), ut_realloc(), and ut_free() directly to malloc(),
+	realloc(), and free() when innodb_use_sys_malloc is set. As a side
+	effect, ut_total_allocated_memory ("Total memory allocated" in the
+	"BUFFER POOL AND MEMORY" section of SHOW ENGINE INNODB STATUS) will
+	exclude any memory allocated by these functions when
+	innodb_use_sys_malloc is set.
+
+2009-02-10	The InnoDB Team
+
+	* btr/btr0cur.c, btr/btr0sea.c, buf/buf0buf.c, handler/ha_innodb.cc,
+	include/buf0buf.ic, include/os0sync.h, include/srv0srv.h,
+	include/sync0rw.h, include/sync0rw.ic, include/sync0sync.h,
+	include/sync0sync.ic, include/univ.i, row/row0sel.c, srv/srv0srv.c,
+	srv/srv0start.c, sync/sync0arr.c, sync/sync0rw.c, sync/sync0sync.c:
+	On those platforms that support it, implement the synchronization
+	primitives of InnoDB mutexes and read/write locks with GCC atomic
+	builtins instead of Pthreads mutexes and InnoDB mutexes. These changes
+	are based on a patch supplied by Mark Callaghan of Google under a BSD
+	license.
+
+2009-01-30	The InnoDB Team
+
+	* btr/btr0cur.c, btr/btr0sea.c, buf/buf0buf.c, handler/ha_innodb.cc,
+	include/btr0sea.h, include/buf0buf.h, include/sync0sync.h,
+	sync/sync0sync.c:
+	Make the configuration parameter innodb_adaptive_hash_index dynamic,
+	so that it can be changed at runtime.
+
+2009-01-29	The InnoDB Team
+
+	* handler/ha_innodb.cc, ibuf/ibuf0ibuf.c, include/ibuf0ibuf.h,
+	include/ibuf0ibuf.ic:
+	Implement the settable global variable innodb_change_buffering,
+	with the allowed values 'none' and 'inserts'. The default value
+	'inserts' enables the buffering of inserts to non-unique secondary
+	index trees when the B-tree leaf page is not in the buffer pool.
+
+2009-01-27	The InnoDB Team
+
+	* buf/buf0lru.c:
+	Fix a race condition in buf_LRU_invalidate_tablespace(): The
+	compressed page size (zip_size) was read while the block descriptor
+	was no longer protected by a mutex. This could lead to corruption
+	when a table is dropped on a busy system that contains compressed
+	tables.
+
+2009-01-26	The InnoDB Team
+
+	* btr/btr0sea.c, buf/buf0buf.c, include/buf0buf.h, include/buf0buf.ic,
+	include/mtr0log.ic, include/row0upd.ic, mtr/mtr0mtr.c:
+	Implement buf_block_align() with pointer arithmetics, as it is in the
+	built-in InnoDB distributed with MySQL. Do not acquire the buffer pool
+	mutex before buf_block_align(). This removes a scalability bottleneck
+	in the adaptive hash index lookup. In CHECK TABLE, check that
+	buf_pool->page_hash is consistent with buf_block_align().
+
+2009-01-23	The InnoDB Team
+
+	* btr/btr0sea.c:
+	Fix Bug#42279 Race condition in btr_search_drop_page_hash_when_freed()
+
+2009-01-23	The InnoDB Team
+
+	* buf/buf0buf.c, include/buf0buf.h:
+	Remove the unused mode BUF_GET_NOWAIT of buf_page_get_gen()
+
+2009-01-20	The InnoDB Team
+
+	* include/rem0rec.h, include/rem0rec.ic:
+	Fix Bug#41571 MySQL segfaults after innodb recovery
+
+2009-01-20	The InnoDB Team
+
+	* lock/lock0lock.c:
+	Fix Bug#42152 Race condition in lock_is_table_exclusive()
+
+2009-01-14	The InnoDB Team
+
+	* include/trx0roll.h, trx/trx0roll.c, trx/trx0trx.c:
+	Fix Bug#38187 Error 153 when creating savepoints
+
+2009-01-14	The InnoDB Team
+
+	* dict/dict0load.c:
+	Fix Bug#42075 dict_load_indexes failure in dict_load_table will
+	corrupt the dictionary cache
+
+2009-01-13	The InnoDB Team
+
+	* buf/buf0buddy.c, dict/dict0dict.c, dict/dict0mem.c, fil/fil0fil.c,
+	ha/ha0storage.c, handler/ha_innodb.cc, handler/win_delay_loader.cc,
+	include/buf0buf.ic, include/dict0dict.ic, include/hash0hash.h,
+	thr/thr0loc.c, trx/trx0i_s.c:
+	Add the parameter ASSERTION to HASH_SEARCH() macro, and use it for
+	light validation of the traversed items in hash table lookups when
+	UNIV_DEBUG is enabled.
+
+2009-01-09	The InnoDB Team
+
+	* buf/buf0flu.c, include/buf0flu.h, include/buf0flu.ic:
+	Remove unused code from the functions
+	buf_flush_insert_into_flush_list() and
+	buf_flush_insert_sorted_into_flush_list().
+
+2009-01-09	The InnoDB Team
+
+	* buf/buf0flu.c:
+	Simplify the functions buf_flush_try_page() and buf_flush_batch(). Add
+	debug assertions and an explanation to buf_flush_write_block_low().
+
+2009-01-07	The InnoDB Team
+
+	* row/row0merge.c:
+	Fix a bug in recovery when dropping temporary indexes.
+
+2009-01-07	The InnoDB Team
+
+	* handler/ha_innodb.cc, handler/ha_innodb.h, handler/handler0alter.cc:
+	Fix Bug#41680 calls to trx_allocate_for_mysql are not consistent
+
+2009-01-07	The InnoDB Team
+
+	* mysql-test/innodb_bug41904.result, mysql-test/innodb_bug41904.test,
+	row/row0merge.c:
+	Fix Bug#41904 create unique index problem
+
+2009-01-02	The InnoDB Team
+
+	* handler/ha_innodb.cc, include/srv0srv.h, mem/mem0pool.c,
+	mysql-test/innodb-use-sys-malloc-master.opt,
+	mysql-test/innodb-use-sys-malloc.result,
+	mysql-test/innodb-use-sys-malloc.test, srv/srv0srv.c, srv/srv0start.c:
+	Implement the configuration parameter innodb_use_sys_malloc (false by
+	default), for disabling InnoDB's internal memory allocator and using
+	system malloc/free instead. The "BUFFER POOL AND MEMORY" section of
+	SHOW ENGINE INNODB STATUS will report "in additional pool allocated
+	allocated 0" when innodb_use_sys_malloc is set.
+
+2008-12-30	The InnoDB Team
+
+	* btr/btr0btr.c:
+	When setting the PAGE_LEVEL of a compressed B-tree page from or to 0,
+	compress the page at the same time. This is necessary, because the
+	column information stored on the compressed page will differ between
+	leaf and non-leaf pages. Leaf pages are identified by PAGE_LEVEL=0.
+	This bug can make InnoDB crash when all rows of a compressed table are
+	deleted.
+
+2008-12-17	The InnoDB Team
+
+	* include/row0sel.h, include/row0upd.h, pars/pars0pars.c,
+	row/row0mysql.c, row/row0sel.c, row/row0upd.c:
+	Remove update-in-place select from the internal SQL interpreter. It
+	was only used for updating the InnoDB internal data dictionary when
+	renaming or dropping tables. It could have caused deadlocks when
+	acquiring latches on insert buffer bitmap pages.
+
+2008-12-17	The InnoDB Team
+
+	* btr/btr0sea.c, buf/buf0buf.c, buf/buf0lru.c, ha/ha0ha.c,
+	ha/hash0hash.c, include/buf0buf.h, include/ha0ha.h, include/ha0ha.ic,
+	include/hash0hash.h, include/univ.i:
+	Introduce the preprocessor symbol UNIV_AHI_DEBUG for enabling adaptive
+	hash index debugging independently of UNIV_DEBUG.
+
+2008-12-16	The InnoDB Team
+
+	* btr/btr0cur.c:
+	Do not update the free bits in the insert buffer bitmap when inserting
+	or deleting from the insert buffer B-tree. Assert that records in the
+	insert buffer B-tree are never updated.
+
+2008-12-12	The InnoDB Team
+
+	* buf/buf0buf.c, fil/fil0fil.c, fsp/fsp0fsp.c, ibuf/ibuf0ibuf.c,
+	include/fil0fil.h, include/ibuf0ibuf.h, include/ibuf0ibuf.ic,
+	include/ibuf0types.h:
+	Clean up the insert buffer subsystem so that only one insert
+	buffer B-tree exists.
+	Originally, there were provisions in InnoDB for multiple insert
+	buffer B-trees, apparently one for each tablespace.
+	When Heikki Tuuri implemented multiple InnoDB tablespaces in
+	MySQL/InnoDB 4.1, he made the insert buffer live only in the
+	system tablespace (space 0) but left the provisions in the code.
+
+2008-12-11	The InnoDB Team
+
+	* include/srv0srv.h, os/os0proc.c, srv/srv0srv.c:
+	Fix the issue that the InnoDB plugin fails if innodb_buffer_pool_size
+	is defined bigger than 4096M on 64-bit Windows. This bug should not
+	have affected other 64-bit systems.
+
+2008-12-09	The InnoDB Team
+
+	* handler/ha_innodb.cc:
+	Fix Bug#40386 Not flushing query cache after truncate.
+
+2008-12-09	The InnoDB Team
+
+	* handler/ha_innodb.cc, srv/srv0srv.c, trx/trx0trx.c:
+	Fix Bug#40760 "set global innodb_thread_concurrency = 0;" is not safe
+
+2008-12-04	The InnoDB Team
+
+	* handler/ha_innodb.cc, handler/mysql_addons.cc,
+	include/mysql_addons.h, trx/trx0i_s.c, win-plugin/win-plugin.diff:
+	Remove dependencies to MySQL internals (defining MYSQL_SERVER).
+
+2008-12-02	The InnoDB Team
+
+	* page/page0cur.c:
+	When allocating space for a record from the free list of previously
+	purged records, zero out the DB_TRX_ID and DB_ROLL_PTR of the purged
+	record if the new record would not overwrite these fields. This fixes
+	a harmless content mismatch reported by page_zip_validate().
+
+2008-12-02	The InnoDB Team
+
+	* row/row0merge.c:
+	Replace the WHILE 1 with WHILE 1=1 in the SQL procedure, so that the
+	loop will actually be entered and temporary indexes be dropped during
+	crash recovery.
+
+2008-12-01	The InnoDB Team
+
+	InnoDB Plugin 1.0.2 released
+
+2008-10-31	The InnoDB Team
+
+	* dict/dict0mem.c, include/dict0mem.h, include/lock0lock.h,
+	include/row0mysql.h, include/trx0trx.h, include/univ.i,
+	include/ut0vec.h, include/ut0vec.ic, lock/lock0lock.c,
+	row/row0mysql.c, trx/trx0trx.c:
+	Fix Bug#26316 Triggers create duplicate entries on auto-increment
+	columns
+
+2008-10-30	The InnoDB Team
+
+	* handler/ha_innodb.cc, handler/handler0vars.h,
+	handler/win_delay_loader.cc, mysql-test/innodb_bug40360.result,
+	mysql-test/innodb_bug40360.test:
+	Fix Bug#40360 Binlog related errors with binlog off
+
+2008-10-29	The InnoDB Team
+
+	* include/data0type.ic:
+	Fix Bug#40369 dtype_get_sql_null_size() returns 0 or 1, not the size
+
+2008-10-29	The InnoDB Team
+
+	* handler/ha_innodb.cc, include/srv0srv.h, srv/srv0srv.c:
+	Fix Bug#38189 innodb_stats_on_metadata missing
+
+2008-10-28	The InnoDB Team
+
+	* CMakeLists.txt, ha_innodb.def, handler/ha_innodb.cc,
+	handler/handler0alter.cc, handler/handler0vars.h, handler/i_s.cc,
+	handler/win_delay_loader.cc, win-plugin/*:
+	Implemented the delayloading of externals for the plugin on Windows.
+	This makes it possible to build a dynamic plugin (ha_innodb.dll) on
+	Windows.
+
+2008-10-27	The InnoDB Team
+
+	* CMakeLists.txt:
+	Fix Bug#19424 InnoDB: Possibly a memory overrun of the buffer being
+	freed (64-bit Visual C)
+
+2008-10-23	The InnoDB Team
+
+	* ibuf/ibuf0ibuf.c:
+	ibuf_delete_rec(): When the cursor to the insert buffer record
+	cannot be restored, do not complain if the tablespace does not
+	exist, because the insert buffer record may have been discarded by
+	some other thread. This bug has existed in MySQL/InnoDB since
+	version 4.1, when innodb_file_per_table was implemented.
+	This may fix Bug#27276 InnoDB Error: ibuf cursor restoration fails.
+
+2008-10-22	The InnoDB Team
+
+	* dict/dict0dict.c, dict/dict0mem.c, handler/ha_innodb.cc,
+	handler/ha_innodb.h, include/dict0dict.h, include/dict0mem.h,
+	row/row0mysql.c:
+	Fix Bug#39830 Table autoinc value not updated on first insert
+	Fix Bug#35498 Cannot get table test/table1 auto-inccounter value in
+	::info
+	Fix Bug#36411 "Failed to read auto-increment value from storage
+	engine" in 5.1.24 auto-inc
+
+2008-10-22	The InnoDB Team
+
+	* handler/ha_innodb.cc, include/row0mysql.h, row/row0mysql.c:
+	Fix Bug#40224 New AUTOINC changes mask reporting of deadlock/timeout
+	errors
+
+2008-10-16	The InnoDB Team
+
+	* dict/dict0dict.c, mysql-test/innodb-index.result,
+	mysql-test/innodb-index.test:
+	Skip the undo log size check when creating REDUNDANT and COMPACT
+	tables. In ROW_FORMAT=DYNAMIC and ROW_FORMAT=COMPRESSED, column
+	prefix indexes require that prefixes of externally stored columns
+	be written to the undo log. This may make the undo log record
+	bigger than the record on the B-tree page. The maximum size of an
+	undo log record is the page size. That must be checked for, in
+	dict_index_add_to_cache(). However, this restriction must not
+	be enforced on REDUNDANT or COMPACT tables.
+
+2008-10-15	The InnoDB Team
+
+	* btr/btr0cur.c, include/btr0cur.h, row/row0ext.c, row/row0sel.c,
+	row/row0upd.c:
+	When the server crashes while freeing an externally stored column
+	of a compressed table, the BTR_EXTERN_LEN field in the BLOB
+	pointer will be written as 0. Tolerate this in the functions that
+	deal with externally stored columns. This fixes problems after
+	crash recovery, in the rollback of incomplete transactions, and in
+	the purge of delete-marked records.
+
+2008-10-15	The InnoDB Team
+
+	* btr/btr0btr.c, include/page0zip.h, page/page0zip.c, include/univ.i:
+	When a B-tree node of a compressed table is split or merged, the
+	compression may fail. In this case, the entire compressed page
+	will be copied and the excess records will be deleted. However,
+	page_zip_copy(), now renamed to page_zip_copy_recs(), copied too
+	many fields in the page header, overwriting PAGE_BTR_SEG_LEAF and
+	PAGE_BTR_SEG_TOP when splitting the B-tree root. This caused
+	corruption of compressed tables. Furthermore, the lock table and
+	the adaptive hash index would be corrupted, because we forgot to
+	update them when invoking page_zip_copy_recs().
+
+	Introduce the symbol UNIV_ZIP_DEBUG for triggering the copying of
+	compressed pages more often, for debugging purposes.
+
+2008-10-10	The InnoDB Team
+
+	* handler/handler0alter.cc, include/row0merge.h, row/row0merge.c,
+	row/row0mysql.c:
+	Fix some locking issues, mainly in fast index creation. The
+	InnoDB data dictionary cache should be latched whenever a
+	transaction is holding locks on any data dictionary tables.
+	Otherwise, lock waits or deadlocks could occur. Furthermore, the
+	data dictionary transaction must be committed (and the locks
+	released) before the data dictionary latch is released.
+
+	ha_innobase::add_index(): Lock the data dictionary before renaming
+	or dropping the created indexes, because neither operation will
+	commit the data dictionary transaction.
+
+	ha_innobase::final_drop_index(): Commit the transactions before
+	unlocking the data dictionary.
+
+2008-10-09	The InnoDB Team
+
+	* buf/buf0lru.c:
+	Fix Bug#39939 DROP TABLE/DISCARD TABLESPACE takes long time in
+	buf_LRU_invalidate_tablespace()
+
+2008-10-08	The InnoDB Team
+
+	* dict/dict0crea.c, trx/trx0roll.c, include/row0mysql.h,
+	row/row0merge.c, row/row0mysql.c:
+	When dropping a table, hold the data dictionary latch until the
+	transaction has been committed. The data dictionary latch is
+	supposed to prevent lock waits and deadlocks in the data
+	dictionary tables. Due to this bug, DROP TABLE could cause a
+	deadlock or hang. Note that because of Bug#33650 and Bug#39833,
+	MySQL may also drop a (temporary) table when executing CREATE INDEX
+	or ALTER TABLE ... ADD INDEX.
+
+2008-10-04	The InnoDB Team
+
+	* handler/ha_innodb.cc, mysql-test/innodb_bug39438-master.opt,
+	mysql-test/innodb_bug39438.result, mysql-test/innodb_bug39438.test:
+	Fix Bug#39438 Testcase for Bug#39436 crashes on 5.1 in
+	fil_space_get_latch
+
+2008-10-04	The InnoDB Team
+
+	* include/lock0lock.h, lock/lock0lock.c,
+	mysql-test/innodb_bug38231.result, mysql-test/innodb_bug38231.test,
+	row/row0mysql.c:
+	Fix Bug#38231 Innodb crash in lock_reset_all_on_table() on TRUNCATE +
+	LOCK / UNLOCK
+
+2008-10-04	The InnoDB Team
+
+	* handler/ha_innodb.cc:
+	Fix Bug#35498 Cannot get table test/table1 auto-inccounter value in
+	::info
+
+2008-10-04	The InnoDB Team
+
+	* handler/ha_innodb.cc, handler/ha_innodb.h:
+	Fix Bug#37788 InnoDB Plugin: AUTO_INCREMENT wrong for compressed
+	tables
+
+2008-10-04	The InnoDB Team
+
+	* dict/dict0dict.c, handler/ha_innodb.cc, handler/ha_innodb.h,
+	include/dict0dict.h, include/dict0mem.h, row/row0mysql.c:
+	Fix Bug#39830 Table autoinc value not updated on first insert
+
+2008-10-03	The InnoDB Team
+
+	* mysql-test/innodb-index.test, mysql-test/innodb-index.result,
+	mysql-test/innodb-timeout.test, mysql-test/innodb-timeout.result,
+	srv/srv0srv.c, include/srv0srv.h, handler/ha_innodb.cc,
+	include/ha_prototypes.h:
+	Fix Bug#36285 innodb_lock_wait_timeout is not dynamic, not per session
+
+2008-09-19	The InnoDB Team
+
+	* os/os0proc.c:
+	Fix a memory leak on Windows. The memory leak was due to wrong
+	parameters passed into VirtualFree() call. As the result, the
+	call fails with Windows error 87.
+
+2008-09-17	The InnoDB Team
+
+	* mysql-test/innodb.result, mysql-test/innodb-zip.result,
+	mysql-test/innodb-zip.test, mysql-test/innodb.test, ibuf/ibuf0ibuf.c,
+	dict/dict0crea.c, dict/dict0load.c, dict/dict0boot.c,
+	include/dict0dict.h, include/trx0trx.h, dict/dict0dict.c,
+	trx/trx0trx.c, include/ha_prototypes.h, handler/ha_innodb.cc:
+	When creating an index in innodb_strict_mode, check that the
+	maximum record size will never exceed the B-tree page size limit.
+	For uncompressed tables, there should always be enough space for
+	two records in an empty B-tree page. For compressed tables, there
+	should be enough space for storing two node pointer records or one
+	data record in an empty page in uncompressed format.
+	The purpose of this check is to guarantee that INSERT or UPDATE
+	will never fail due to too big record size.
+
+2008-09-17	The InnoDB Team
+
+	* btr/btr0cur.c, data/data0data.c, include/page0zip.h,
+	include/page0zip.ic, page/page0zip.c, mysql-test/innodb_bug36172.test:
+	Prevent infinite B-tree page splits in compressed tables by
+	ensuring that there will always be enough space for two node
+	pointer records in an empty B-tree page. Also, require that at
+	least one data record will fit in an empty compressed page. This
+	will reduce the maximum size of records in compressed tables.
+
+2008-09-09	The InnoDB Team
+
+	* mysql-test/innodb.result:
+	Fix the failing innodb test by merging changes that MySQL made to
+	that file (r2646.12.1 in MySQL BZR repository)
+
+2008-09-09	The InnoDB Team
+
+	* handler/ha_innodb.cc, mysql-test/innodb-autoinc.result,
+	mysql-test/innodb-autoinc.test:
+	Fix Bug#38839 auto increment does not work properly with InnoDB after
+	update
+
+2008-09-09	The InnoDB Team
+
+	* dict/dict0dict.c, handler/handler0alter.cc, include/dict0dict.h,
+	mysql-test/innodb-index.result, mysql-test/innodb-index.test:
+	Fix Bug#38786 InnoDB plugin crashes on drop table/create table with FK
+
+2008-08-21	The InnoDB Team
+
+	* handler/ha_innodb.cc, include/ha_prototypes.h, row/row0sel.c:
+	Fix Bug#37885 row_search_for_mysql may gap lock unnecessarily with SQL
+	comments in query
+
+2008-08-21	The InnoDB Team
+
+	* handler/ha_innodb.cc:
+	Fix Bug#38185 ha_innobase::info can hold locks even when called with
+	HA_STATUS_NO_LOCK
+
+2008-08-18	The InnoDB Team
+
+	* buf/buf0buf.c, buf/buf0lru.c, include/buf0buf.ic, include/univ.i:
+	Introduce UNIV_LRU_DEBUG for debugging the LRU buffer pool cache
+
+2008-08-08	The InnoDB Team
+
+	* buf/buf0lru.c, include/buf0buf.h:
+	Fix two recovery bugs that could lead to a crash in debug builds with
+	small buffer size
+
+2008-08-07	The InnoDB Team
+
+	* btr/btr0cur.c, handler/ha_innodb.cc, include/srv0srv.h,
+	srv/srv0srv.c:
+	Add a parameter innodb_stats_sample_pages to allow users to control
+	the number of index dives when InnoDB estimates the cardinality of
+	an index (ANALYZE TABLE, SHOW TABLE STATUS etc)
+
+2008-08-07	The InnoDB Team
+
+	* trx/trx0i_s.c:
+	Fix a bug that would lead to a crash if a SELECT was issued from the
+	INFORMATION_SCHEMA tables and there are rolling back transactions at
+	the same time
+
+2008-08-06	The InnoDB Team
+
+	* btr/btr0btr.c, btr/btr0cur.c, ibuf/ibuf0ibuf.c, include/btr0cur.h,
+	include/trx0roll.h, include/trx0types.h, row/row0purge.c,
+	row/row0uins.c, row/row0umod.c, trx/trx0roll.c:
+	In the rollback of incomplete transactions after crash recovery,
+	tolerate clustered index records whose externally stored columns
+	have not been written.
+
+2008-07-30	The InnoDB Team
+
+	* trx/trx0trx.c:
+	Fixes a race in recovery where the recovery thread recovering a
+	PREPARED trx and the background rollback thread can both try
+	to free the trx after its status is set to COMMITTED_IN_MEMORY.
+
+2008-07-29	The InnoDB Team
+
+	* include/trx0rec.h, row/row0purge.c, row/row0vers.c, trx/trx0rec.c:
+	Fix a BLOB corruption bug
+
+2008-07-15	The InnoDB Team
+
+	* btr/btr0sea.c, dict/dict0dict.c, include/btr0sea.h:
+	Fixed a timing hole where a thread dropping an index can free the
+	in-memory index struct while another thread is still using that
+	structure to remove entries from adaptive hash index belonging
+	to one of the pages that belongs to the index being dropped.
+
+2008-07-04	The InnoDB Team
+
+	* mysql-test/innodb-index.result:
+	Fix the failing innodb-index test by adjusting the result to a new
+	MySQL behavior (the change occured in BZR-r2667)
+
+2008-07-03	The InnoDB Team
+
+	* mysql-test/innodb-zip.result, mysql-test/innodb-zip.test:
+	Remove the negative test cases that produce warnings
+
+2008-07-02	The InnoDB Team
+
+	* mysql-test/innodb-replace.result, mysql-test/innodb-index.test:
+	Disable part of innodb-index test because MySQL changed its behavior
+	and is not calling ::add_index() anymore when adding primary index on
+	non-NULL column
+
+2008-07-01	The InnoDB Team
+
+	* mysql-test/innodb-replace.result, mysql-test/innodb-replace.test:
+	Fix the failing innodb-replace test by merging changes that MySQL
+	made to that file (r2659 in MySQL BZR repository)
+
+2008-07-01	The InnoDB Team
+
+	* lock/lock0lock.c:
+	Fix Bug#36942 Performance problem in lock_get_n_rec_locks (SHOW INNODB
+	STATUS)
+
+2008-07-01	The InnoDB Team
+
+	* ha/ha0ha.c:
+	Fix Bug#36941 Performance problem in ha_print_info (SHOW INNODB
+	STATUS)
+
+2008-07-01	The InnoDB Team
+
+	* handler/ha_innodb.cc, mysql-test/innodb-autoinc.result,
+	mysql-test/innodb-autoinc.test:
+	Fix Bug#37531 After truncate, auto_increment behaves incorrectly for
+	InnoDB
+
+2008-06-19	The InnoDB Team
+
+	* handler/ha_innodb.cc:
+	Rewrite the function innodb_plugin_init() to support parameters in
+	different order (in static and dynamic InnoDB) and to support more
+	parameters in the static InnoDB
+
+2008-06-19	The InnoDB Team
+
+	* handler/handler0alter.cc:
+	Fix a bug in ::add_index() which set the transaction state to "active"
+	but never restored it to the original value. This bug caused warnings
+	to be printed by the rpl.rpl_ddl mysql-test.
+
+2008-06-19	The InnoDB Team
+
+	* mysql-test/patches:
+	Add a directory which contains patches, which need to be applied to
+	MySQL source in order to get some mysql-tests to succeed. The patches
+	cannot be committed in MySQL repository because they are specific to
+	the InnoDB plugin.
+
+2008-06-19	The InnoDB Team
+
+	* mysql-test/innodb-zip.result, mysql-test/innodb-zip.test,
+	row/row0row.c:
+	Fix an anomaly when updating a record with BLOB prefix
+
+2008-06-18	The InnoDB Team
+
+	* include/trx0sys.h, srv/srv0start.c, trx/trx0sys.c:
+	Fix a bug in recovery which was a side effect of the file_format_check
+	changes
+
+2008-06-09	The InnoDB Team
+
+	* mysql-test/innodb.result:
+	Fix the failing innodb test by merging changes that MySQL made to that
+	file
+
+2008-06-06	The InnoDB Team
+
+	* buf/buf0buf.c, handler/ha_innodb.cc, include/buf0buf.h,
+	include/srv0srv.h, srv/srv0srv.c:
+	Fix Bug#36600 SHOW STATUS takes a lot of CPU in
+	buf_get_latched_pages_number
+
+	* handler/ha_innodb.cc, os/os0file.c:
+	Fix Bug#11894 innodb_file_per_table crashes w/ Windows .sym symbolic
+	link hack
+
+	* include/ut0ut.h, srv/srv0srv.c, ut/ut0ut.c:
+	Fix Bug#36819 ut_usectime does not handle errors from gettimeofday
+
+	* handler/ha_innodb.cc:
+	Fix Bug#35602 Failed to read auto-increment value from storage engine
+
+	* srv/srv0start.c:
+	Fix Bug#36149 Read buffer overflow in srv0start.c found during "make
+	test"
+
+2008-05-08	The InnoDB Team
+
+	* btr/btr0btr.c, mysql-test/innodb_bug36172.result,
+	mysql-test/innodb_bug36172.test:
+	Fix Bug#36172 insert into compressed innodb table crashes
+
+2008-05-08	The InnoDB Team
+
+	InnoDB Plugin 1.0.1 released
+
+2008-05-06	The InnoDB Team
+
+	* handler/ha_innodb.cc, include/srv0srv.h, include/sync0sync.h,
+	include/trx0sys.h, mysql-test/innodb-zip.result,
+	mysql-test/innodb-zip.test, srv/srv0srv.c, srv/srv0start.c,
+	sync/sync0sync.c, trx/trx0sys.c:
+	Implement the system tablespace tagging
+
+	* handler/ha_innodb.cc, handler/i_s.cc, include/univ.i,
+	srv/srv0start.c:
+	Add InnoDB version in INFORMATION_SCHEMA.PLUGINS.PLUGIN_VERSION,
+	in the startup message and in a server variable innodb_version.
+
+	* sync/sync0sync.c:
+	Fix a bug in the sync debug code where a lock with level
+	SYNC_LEVEL_VARYING would cause an assertion failure when a thread
+	tried to release it.
+
+2008-04-30	The InnoDB Team
+
+	* Makefile.am:
+	Fix Bug#36434 ha_innodb.so is installed in the wrong directory
+
+	* handler/ha_innodb.cc:
+	Merge change from MySQL (Fix Bug#35406 5.1-opt crashes on select from
+	I_S.REFERENTIAL_CONSTRAINTS):
+	ChangeSet@1.2563, 2008-03-18 19:42:04+04:00, gluh@mysql.com +1 -0
+
+	* scripts/install_innodb_plugins.sql:
+	Added
+
+	* mysql-test/innodb.result:
+	Merge change from MySQL (this fixes the failing innodb test):
+	ChangeSet@1.1810.3601.4, 2008-02-07 02:33:21+04:00
+
+	* row/row0sel.c:
+	Fix Bug#35226 RBR event crashes slave
+
+	* handler/ha_innodb.cc:
+	Change the fix for Bug#32440 to show bytes instead of kilobytes in
+	INFORMATION_SCHEMA.TABLES.DATA_FREE
+
+	* handler/ha_innodb.cc, mysql-test/innodb.result,
+	mysql-test/innodb.test:
+	Fix Bug#29507 TRUNCATE shows to many rows effected
+
+	* handler/ha_innodb.cc, mysql-test/innodb.result,
+	mysql-test/innodb.test:
+	Fix Bug#35537 Innodb doesn't increment handler_update and
+	handler_delete
+
+2008-04-29	The InnoDB Team
+
+	* handler/i_s.cc, include/srv0start.h, srv/srv0start.c:
+	Fix Bug#36310 InnoDB plugin crash
+
+2008-04-23	The InnoDB Team
+
+	* mysql-test/innodb_bug36169.result, mysql-test/innodb_bug36169.test,
+	row/row0mysql.c:
+	Fix Bug#36169 create innodb compressed table with too large row size
+	crashed
+
+	* (outside the source tree):
+	Fix Bug#36222 New InnoDB plugin 1.0 has wrong MKDIR_P defined in
+	Makefile.in
+
+2008-04-15	The InnoDB Team
+
+	InnoDB Plugin 1.0.0 released
diff --git a/storage/xtradb/Makefile.am b/storage/xtradb/Makefile.am
new file mode 100644
index 00000000000..8f64aedb9b0
--- /dev/null
+++ b/storage/xtradb/Makefile.am
@@ -0,0 +1,195 @@
+# Copyright (C) 2001, 2004, 2006 MySQL AB & Innobase Oy
+#
+# This program is free software; you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation; version 2 of the License.
+# 
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+# 
+# You should have received a copy of the GNU General Public License
+# along with this program; if not, write to the Free Software
+# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+
+# Process this file with automake to create Makefile.in
+
+MYSQLDATAdir=		$(localstatedir)
+MYSQLSHAREdir=		$(pkgdatadir)
+MYSQLBASEdir=		$(prefix)
+MYSQLLIBdir=		$(pkglibdir)
+pkgplugindir=		$(pkglibdir)/plugin
+INCLUDES=		-I$(top_srcdir)/include -I$(top_builddir)/include \
+			-I$(top_srcdir)/regex \
+			-I$(top_srcdir)/storage/xtradb/include \
+			-I$(top_srcdir)/sql \
+			-I$(srcdir)
+
+DEFS=			@DEFS@
+
+
+noinst_HEADERS=		include/btr0btr.h include/btr0btr.ic		\
+			include/btr0cur.h include/btr0cur.ic		\
+			include/btr0pcur.h include/btr0pcur.ic		\
+			include/btr0sea.h include/btr0sea.ic		\
+			include/btr0types.h include/buf0buddy.h		\
+			include/buf0buddy.ic include/buf0buf.h		\
+			include/buf0buf.ic include/buf0flu.h		\
+			include/buf0flu.ic include/buf0lru.h		\
+			include/buf0lru.ic include/buf0rea.h		\
+			include/buf0types.h include/data0data.h		\
+			include/data0data.ic include/data0type.h	\
+			include/data0type.ic include/data0types.h	\
+			include/db0err.h include/dict0boot.h		\
+			include/dict0boot.ic include/dict0crea.h	\
+			include/dict0crea.ic include/dict0dict.h	\
+			include/dict0dict.ic include/dict0load.h	\
+			include/dict0load.ic include/dict0mem.h		\
+			include/dict0mem.ic include/dict0types.h	\
+			include/dyn0dyn.h include/dyn0dyn.ic		\
+			include/eval0eval.h include/eval0eval.ic	\
+			include/eval0proc.h include/eval0proc.ic	\
+			include/fil0fil.h include/fsp0fsp.h		\
+			include/fsp0fsp.ic include/fut0fut.h		\
+			include/fut0fut.ic include/fut0lst.h		\
+			include/fut0lst.ic include/ha0ha.h		\
+			include/ha0ha.ic				\
+			include/ha0storage.h				\
+			include/ha0storage.ic				\
+			include/hash0hash.h				\
+			include/hash0hash.ic include/ibuf0ibuf.h	\
+			include/ibuf0ibuf.ic include/ibuf0types.h	\
+			include/lock0iter.h				\
+			include/lock0lock.h include/lock0lock.ic	\
+			include/lock0priv.h include/lock0priv.ic	\
+			include/lock0types.h include/log0log.h		\
+			include/log0log.ic include/log0recv.h		\
+			include/log0recv.ic include/mach0data.h		\
+			include/mach0data.ic include/mem0dbg.h		\
+			include/mem0dbg.ic mem/mem0dbg.c		\
+			include/mem0mem.h include/mem0mem.ic		\
+			include/mem0pool.h include/mem0pool.ic		\
+			include/mtr0log.h include/mtr0log.ic		\
+			include/mtr0mtr.h include/mtr0mtr.ic		\
+			include/mtr0types.h				\
+			include/mysql_addons.h				\
+			include/os0file.h				\
+			include/os0proc.h include/os0proc.ic		\
+			include/os0sync.h include/os0sync.ic		\
+			include/os0thread.h include/os0thread.ic	\
+			include/page0cur.h include/page0cur.ic		\
+			include/page0page.h include/page0page.ic	\
+			include/page0zip.h include/page0zip.ic		\
+			include/page0types.h include/pars0grm.h		\
+			include/pars0opt.h include/pars0opt.ic		\
+			include/pars0pars.h include/pars0pars.ic	\
+			include/pars0sym.h include/pars0sym.ic		\
+			include/pars0types.h include/que0que.h		\
+			include/que0que.ic include/que0types.h		\
+			include/read0read.h include/read0read.ic	\
+			include/read0types.h include/rem0cmp.h		\
+			include/rem0cmp.ic include/rem0rec.h		\
+			include/rem0rec.ic include/rem0types.h		\
+			include/row0ext.h include/row0ext.ic		\
+			include/row0ins.h include/row0ins.ic		\
+			include/row0merge.h				\
+			include/row0mysql.h include/row0mysql.ic	\
+			include/row0purge.h include/row0purge.ic	\
+			include/row0row.h include/row0row.ic		\
+			include/row0sel.h include/row0sel.ic		\
+			include/row0types.h include/row0uins.h		\
+			include/row0uins.ic include/row0umod.h		\
+			include/row0umod.ic include/row0undo.h		\
+			include/row0undo.ic include/row0upd.h		\
+			include/row0upd.ic include/row0vers.h		\
+			include/row0vers.ic include/srv0que.h		\
+			include/srv0srv.h include/srv0srv.ic		\
+			include/srv0start.h include/sync0arr.h		\
+			include/sync0arr.ic include/sync0rw.h		\
+			include/sync0rw.ic include/sync0sync.h		\
+			include/sync0sync.ic include/sync0types.h	\
+			include/thr0loc.h include/thr0loc.ic		\
+			include/trx0i_s.h				\
+			include/trx0purge.h include/trx0purge.ic	\
+			include/trx0rec.h include/trx0rec.ic		\
+			include/trx0roll.h include/trx0roll.ic		\
+			include/trx0rseg.h include/trx0rseg.ic		\
+			include/trx0sys.h include/trx0sys.ic		\
+			include/trx0trx.h include/trx0trx.ic		\
+			include/trx0types.h include/trx0undo.h		\
+			include/trx0undo.ic include/trx0xa.h		\
+			include/univ.i include/usr0sess.h		\
+			include/usr0sess.ic include/usr0types.h		\
+			include/ut0byte.h include/ut0byte.ic		\
+			include/ut0dbg.h include/ut0lst.h		\
+			include/ut0mem.h include/ut0mem.ic		\
+			include/ut0rnd.h include/ut0rnd.ic		\
+			include/ut0sort.h include/ut0ut.h		\
+			include/ut0ut.ic include/ut0vec.h		\
+			include/ut0vec.ic include/ut0list.h		\
+			include/ut0list.ic include/ut0wqueue.h		\
+			include/ha_prototypes.h handler/ha_innodb.h	\
+			include/handler0alter.h				\
+			handler/i_s.h handler/innodb_patch_info.h
+
+EXTRA_LIBRARIES=	libinnobase.a
+noinst_LIBRARIES=	@plugin_innobase_static_target@
+libinnobase_a_SOURCES=	btr/btr0btr.c btr/btr0cur.c btr/btr0pcur.c	\
+			btr/btr0sea.c buf/buf0buddy.c			\
+			buf/buf0buf.c buf/buf0flu.c			\
+			buf/buf0lru.c buf/buf0rea.c data/data0data.c	\
+			data/data0type.c dict/dict0boot.c		\
+			dict/dict0crea.c dict/dict0dict.c		\
+			dict/dict0load.c dict/dict0mem.c dyn/dyn0dyn.c	\
+			eval/eval0eval.c eval/eval0proc.c		\
+			fil/fil0fil.c fsp/fsp0fsp.c fut/fut0fut.c	\
+			fut/fut0lst.c ha/ha0ha.c			\
+			ha/ha0storage.c					\
+			ha/hash0hash.c					\
+			ibuf/ibuf0ibuf.c lock/lock0iter.c		\
+			lock/lock0lock.c				\
+			log/log0log.c log/log0recv.c mach/mach0data.c	\
+			mem/mem0mem.c mem/mem0pool.c mtr/mtr0log.c	\
+			mtr/mtr0mtr.c os/os0file.c os/os0proc.c		\
+			os/os0sync.c os/os0thread.c page/page0cur.c	\
+			page/page0page.c page/page0zip.c		\
+			pars/lexyy.c pars/pars0grm.c			\
+			pars/pars0opt.c pars/pars0pars.c		\
+			pars/pars0sym.c que/que0que.c read/read0read.c	\
+			rem/rem0cmp.c rem/rem0rec.c row/row0ext.c	\
+			row/row0ins.c row/row0merge.c			\
+			row/row0mysql.c row/row0purge.c row/row0row.c	\
+			row/row0sel.c row/row0uins.c row/row0umod.c	\
+			row/row0undo.c row/row0upd.c row/row0vers.c	\
+			srv/srv0que.c srv/srv0srv.c srv/srv0start.c	\
+			sync/sync0arr.c sync/sync0rw.c			\
+			sync/sync0sync.c thr/thr0loc.c			\
+			trx/trx0i_s.c					\
+			trx/trx0purge.c					\
+			trx/trx0rec.c trx/trx0roll.c trx/trx0rseg.c	\
+			trx/trx0sys.c trx/trx0trx.c trx/trx0undo.c	\
+			usr/usr0sess.c ut/ut0byte.c ut/ut0dbg.c		\
+			ut/ut0list.c ut/ut0mem.c ut/ut0rnd.c		\
+			ut/ut0ut.c ut/ut0vec.c ut/ut0wqueue.c		\
+			handler/ha_innodb.cc handler/handler0alter.cc	\
+			handler/i_s.cc					\
+			handler/mysql_addons.cc
+
+libinnobase_a_CXXFLAGS=	$(AM_CFLAGS)
+libinnobase_a_CFLAGS=	$(AM_CFLAGS)
+
+EXTRA_LTLIBRARIES=	ha_innodb.la
+pkgplugin_LTLIBRARIES=	@plugin_innobase_shared_target@
+
+ha_innodb_la_LDFLAGS=	-module -rpath $(pkgplugindir)
+ha_innodb_la_CXXFLAGS=	$(AM_CFLAGS) $(INNODB_DYNAMIC_CFLAGS)
+ha_innodb_la_CFLAGS=	$(AM_CFLAGS) $(INNODB_DYNAMIC_CFLAGS)
+ha_innodb_la_SOURCES=	$(libinnobase_a_SOURCES)
+
+EXTRA_DIST=		CMakeLists.txt plug.in \
+			pars/make_bison.sh pars/make_flex.sh \
+			pars/pars0grm.y pars/pars0lex.l
+
+# Don't update the files from bitkeeper
+%::SCCS/s.%
diff --git a/storage/xtradb/btr/btr0btr.c b/storage/xtradb/btr/btr0btr.c
new file mode 100644
index 00000000000..2029a95cc19
--- /dev/null
+++ b/storage/xtradb/btr/btr0btr.c
@@ -0,0 +1,3656 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The B-tree
+
+Created 6/2/1994 Heikki Tuuri
+*******************************************************/
+
+#include "btr0btr.h"
+
+#ifdef UNIV_NONINL
+#include "btr0btr.ic"
+#endif
+
+#include "fsp0fsp.h"
+#include "page0page.h"
+#include "page0zip.h"
+#include "btr0cur.h"
+#include "btr0sea.h"
+#include "btr0pcur.h"
+#include "rem0cmp.h"
+#include "lock0lock.h"
+#include "ibuf0ibuf.h"
+#include "trx0trx.h"
+
+/*
+Latching strategy of the InnoDB B-tree
+--------------------------------------
+A tree latch protects all non-leaf nodes of the tree. Each node of a tree
+also has a latch of its own.
+
+A B-tree operation normally first acquires an S-latch on the tree. It
+searches down the tree and releases the tree latch when it has the
+leaf node latch. To save CPU time we do not acquire any latch on
+non-leaf nodes of the tree during a search, those pages are only bufferfixed.
+
+If an operation needs to restructure the tree, it acquires an X-latch on
+the tree before searching to a leaf node. If it needs, for example, to
+split a leaf,
+(1) InnoDB decides the split point in the leaf,
+(2) allocates a new page,
+(3) inserts the appropriate node pointer to the first non-leaf level,
+(4) releases the tree X-latch,
+(5) and then moves records from the leaf to the new allocated page.
+
+Node pointers
+-------------
+Leaf pages of a B-tree contain the index records stored in the
+tree. On levels n > 0 we store 'node pointers' to pages on level
+n - 1. For each page there is exactly one node pointer stored:
+thus the our tree is an ordinary B-tree, not a B-link tree.
+
+A node pointer contains a prefix P of an index record. The prefix
+is long enough so that it determines an index record uniquely.
+The file page number of the child page is added as the last
+field. To the child page we can store node pointers or index records
+which are >= P in the alphabetical order, but < P1 if there is
+a next node pointer on the level, and P1 is its prefix.
+
+If a node pointer with a prefix P points to a non-leaf child,
+then the leftmost record in the child must have the same
+prefix P. If it points to a leaf node, the child is not required
+to contain any record with a prefix equal to P. The leaf case
+is decided this way to allow arbitrary deletions in a leaf node
+without touching upper levels of the tree.
+
+We have predefined a special minimum record which we
+define as the smallest record in any alphabetical order.
+A minimum record is denoted by setting a bit in the record
+header. A minimum record acts as the prefix of a node pointer
+which points to a leftmost node on any level of the tree.
+
+File page allocation
+--------------------
+In the root node of a B-tree there are two file segment headers.
+The leaf pages of a tree are allocated from one file segment, to
+make them consecutive on disk if possible. From the other file segment
+we allocate pages for the non-leaf levels of the tree.
+*/
+
+#ifdef UNIV_BTR_DEBUG
+/******************************************************************
+Checks a file segment header within a B-tree root page. */
+static
+ibool
+btr_root_fseg_validate(
+/*===================*/
+						/* out: TRUE if valid */
+	const fseg_header_t*	seg_header,	/* in: segment header */
+	ulint			space)		/* in: tablespace identifier */
+{
+	ulint	offset = mach_read_from_2(seg_header + FSEG_HDR_OFFSET);
+
+	ut_a(mach_read_from_4(seg_header + FSEG_HDR_SPACE) == space);
+	ut_a(offset >= FIL_PAGE_DATA);
+	ut_a(offset <= UNIV_PAGE_SIZE - FIL_PAGE_DATA_END);
+	return(TRUE);
+}
+#endif /* UNIV_BTR_DEBUG */
+
+/******************************************************************
+Gets the root node of a tree and x-latches it. */
+static
+buf_block_t*
+btr_root_block_get(
+/*===============*/
+				/* out: root page, x-latched */
+	dict_index_t*	index,	/* in: index tree */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	ulint		space;
+	ulint		zip_size;
+	ulint		root_page_no;
+	buf_block_t*	block;
+
+	space = dict_index_get_space(index);
+	zip_size = dict_table_zip_size(index->table);
+	root_page_no = dict_index_get_page(index);
+
+	block = btr_block_get(space, zip_size, root_page_no, RW_X_LATCH, mtr);
+	ut_a((ibool)!!page_is_comp(buf_block_get_frame(block))
+	     == dict_table_is_comp(index->table));
+#ifdef UNIV_BTR_DEBUG
+	if (!dict_index_is_ibuf(index)) {
+		const page_t*	root = buf_block_get_frame(block);
+
+		ut_a(btr_root_fseg_validate(FIL_PAGE_DATA + PAGE_BTR_SEG_LEAF
+					    + root, space));
+		ut_a(btr_root_fseg_validate(FIL_PAGE_DATA + PAGE_BTR_SEG_TOP
+					    + root, space));
+	}
+#endif /* UNIV_BTR_DEBUG */
+
+	return(block);
+}
+
+/******************************************************************
+Gets the root node of a tree and x-latches it. */
+UNIV_INTERN
+page_t*
+btr_root_get(
+/*=========*/
+				/* out: root page, x-latched */
+	dict_index_t*	index,	/* in: index tree */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	return(buf_block_get_frame(btr_root_block_get(index, mtr)));
+}
+
+/*****************************************************************
+Gets pointer to the previous user record in the tree. It is assumed that
+the caller has appropriate latches on the page and its neighbor. */
+UNIV_INTERN
+rec_t*
+btr_get_prev_user_rec(
+/*==================*/
+			/* out: previous user record, NULL if there is none */
+	rec_t*	rec,	/* in: record on leaf level */
+	mtr_t*	mtr)	/* in: mtr holding a latch on the page, and if
+			needed, also to the previous page */
+{
+	page_t*	page;
+	page_t*	prev_page;
+	ulint	prev_page_no;
+
+	if (!page_rec_is_infimum(rec)) {
+
+		rec_t*	prev_rec = page_rec_get_prev(rec);
+
+		if (!page_rec_is_infimum(prev_rec)) {
+
+			return(prev_rec);
+		}
+	}
+
+	page = page_align(rec);
+	prev_page_no = btr_page_get_prev(page, mtr);
+
+	if (prev_page_no != FIL_NULL) {
+
+		ulint		space;
+		ulint		zip_size;
+		buf_block_t*	prev_block;
+
+		space = page_get_space_id(page);
+		zip_size = fil_space_get_zip_size(space);
+
+		prev_block = buf_page_get_with_no_latch(space, zip_size,
+							prev_page_no, mtr);
+		prev_page = buf_block_get_frame(prev_block);
+		/* The caller must already have a latch to the brother */
+		ut_ad(mtr_memo_contains(mtr, prev_block,
+					MTR_MEMO_PAGE_S_FIX)
+		      || mtr_memo_contains(mtr, prev_block,
+					   MTR_MEMO_PAGE_X_FIX));
+#ifdef UNIV_BTR_DEBUG
+		ut_a(page_is_comp(prev_page) == page_is_comp(page));
+		ut_a(btr_page_get_next(prev_page, mtr)
+		     == page_get_page_no(page));
+#endif /* UNIV_BTR_DEBUG */
+
+		return(page_rec_get_prev(page_get_supremum_rec(prev_page)));
+	}
+
+	return(NULL);
+}
+
+/*****************************************************************
+Gets pointer to the next user record in the tree. It is assumed that the
+caller has appropriate latches on the page and its neighbor. */
+UNIV_INTERN
+rec_t*
+btr_get_next_user_rec(
+/*==================*/
+			/* out: next user record, NULL if there is none */
+	rec_t*	rec,	/* in: record on leaf level */
+	mtr_t*	mtr)	/* in: mtr holding a latch on the page, and if
+			needed, also to the next page */
+{
+	page_t*	page;
+	page_t*	next_page;
+	ulint	next_page_no;
+
+	if (!page_rec_is_supremum(rec)) {
+
+		rec_t*	next_rec = page_rec_get_next(rec);
+
+		if (!page_rec_is_supremum(next_rec)) {
+
+			return(next_rec);
+		}
+	}
+
+	page = page_align(rec);
+	next_page_no = btr_page_get_next(page, mtr);
+
+	if (next_page_no != FIL_NULL) {
+		ulint		space;
+		ulint		zip_size;
+		buf_block_t*	next_block;
+
+		space = page_get_space_id(page);
+		zip_size = fil_space_get_zip_size(space);
+
+		next_block = buf_page_get_with_no_latch(space, zip_size,
+							next_page_no, mtr);
+		next_page = buf_block_get_frame(next_block);
+		/* The caller must already have a latch to the brother */
+		ut_ad(mtr_memo_contains(mtr, next_block, MTR_MEMO_PAGE_S_FIX)
+		      || mtr_memo_contains(mtr, next_block,
+					   MTR_MEMO_PAGE_X_FIX));
+#ifdef UNIV_BTR_DEBUG
+		ut_a(page_is_comp(next_page) == page_is_comp(page));
+		ut_a(btr_page_get_prev(next_page, mtr)
+		     == page_get_page_no(page));
+#endif /* UNIV_BTR_DEBUG */
+
+		return(page_rec_get_next(page_get_infimum_rec(next_page)));
+	}
+
+	return(NULL);
+}
+
+/******************************************************************
+Creates a new index page (not the root, and also not
+used in page reorganization).  @see btr_page_empty(). */
+static
+void
+btr_page_create(
+/*============*/
+	buf_block_t*	block,	/* in/out: page to be created */
+	page_zip_des_t*	page_zip,/* in/out: compressed page, or NULL */
+	dict_index_t*	index,	/* in: index */
+	ulint		level,	/* in: the B-tree level of the page */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	page_t*		page = buf_block_get_frame(block);
+
+	ut_ad(mtr_memo_contains(mtr, block, MTR_MEMO_PAGE_X_FIX));
+
+	if (UNIV_LIKELY_NULL(page_zip)) {
+		page_create_zip(block, index, level, mtr);
+	} else {
+		page_create(block, mtr, dict_table_is_comp(index->table));
+		/* Set the level of the new index page */
+		btr_page_set_level(page, NULL, level, mtr);
+	}
+
+	block->check_index_page_at_flush = TRUE;
+
+	btr_page_set_index_id(page, page_zip, index->id, mtr);
+}
+
+/******************************************************************
+Allocates a new file page to be used in an ibuf tree. Takes the page from
+the free list of the tree, which must contain pages! */
+static
+buf_block_t*
+btr_page_alloc_for_ibuf(
+/*====================*/
+				/* out: new allocated block, x-latched */
+	dict_index_t*	index,	/* in: index tree */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	fil_addr_t	node_addr;
+	page_t*		root;
+	page_t*		new_page;
+	buf_block_t*	new_block;
+
+	root = btr_root_get(index, mtr);
+
+	node_addr = flst_get_first(root + PAGE_HEADER
+				   + PAGE_BTR_IBUF_FREE_LIST, mtr);
+	ut_a(node_addr.page != FIL_NULL);
+
+	new_block = buf_page_get(dict_index_get_space(index),
+				 dict_table_zip_size(index->table),
+				 node_addr.page, RW_X_LATCH, mtr);
+	new_page = buf_block_get_frame(new_block);
+	buf_block_dbg_add_level(new_block, SYNC_TREE_NODE_NEW);
+
+	flst_remove(root + PAGE_HEADER + PAGE_BTR_IBUF_FREE_LIST,
+		    new_page + PAGE_HEADER + PAGE_BTR_IBUF_FREE_LIST_NODE,
+		    mtr);
+	ut_ad(flst_validate(root + PAGE_HEADER + PAGE_BTR_IBUF_FREE_LIST,
+			    mtr));
+
+	return(new_block);
+}
+
+/******************************************************************
+Allocates a new file page to be used in an index tree. NOTE: we assume
+that the caller has made the reservation for free extents! */
+UNIV_INTERN
+buf_block_t*
+btr_page_alloc(
+/*===========*/
+					/* out: new allocated block, x-latched;
+					NULL if out of space */
+	dict_index_t*	index,		/* in: index */
+	ulint		hint_page_no,	/* in: hint of a good page */
+	byte		file_direction,	/* in: direction where a possible
+					page split is made */
+	ulint		level,		/* in: level where the page is placed
+					in the tree */
+	mtr_t*		mtr)		/* in: mtr */
+{
+	fseg_header_t*	seg_header;
+	page_t*		root;
+	buf_block_t*	new_block;
+	ulint		new_page_no;
+
+	if (dict_index_is_ibuf(index)) {
+
+		return(btr_page_alloc_for_ibuf(index, mtr));
+	}
+
+	root = btr_root_get(index, mtr);
+
+	if (level == 0) {
+		seg_header = root + PAGE_HEADER + PAGE_BTR_SEG_LEAF;
+	} else {
+		seg_header = root + PAGE_HEADER + PAGE_BTR_SEG_TOP;
+	}
+
+	/* Parameter TRUE below states that the caller has made the
+	reservation for free extents, and thus we know that a page can
+	be allocated: */
+
+	new_page_no = fseg_alloc_free_page_general(seg_header, hint_page_no,
+						   file_direction, TRUE, mtr);
+	if (new_page_no == FIL_NULL) {
+
+		return(NULL);
+	}
+
+	new_block = buf_page_get(dict_index_get_space(index),
+				 dict_table_zip_size(index->table),
+				 new_page_no, RW_X_LATCH, mtr);
+	buf_block_dbg_add_level(new_block, SYNC_TREE_NODE_NEW);
+
+	return(new_block);
+}
+
+/******************************************************************
+Gets the number of pages in a B-tree. */
+UNIV_INTERN
+ulint
+btr_get_size(
+/*=========*/
+				/* out: number of pages */
+	dict_index_t*	index,	/* in: index */
+	ulint		flag)	/* in: BTR_N_LEAF_PAGES or BTR_TOTAL_SIZE */
+{
+	fseg_header_t*	seg_header;
+	page_t*		root;
+	ulint		n;
+	ulint		dummy;
+	mtr_t		mtr;
+
+	mtr_start(&mtr);
+
+	mtr_s_lock(dict_index_get_lock(index), &mtr);
+
+	root = btr_root_get(index, &mtr);
+
+	if (flag == BTR_N_LEAF_PAGES) {
+		seg_header = root + PAGE_HEADER + PAGE_BTR_SEG_LEAF;
+
+		fseg_n_reserved_pages(seg_header, &n, &mtr);
+
+	} else if (flag == BTR_TOTAL_SIZE) {
+		seg_header = root + PAGE_HEADER + PAGE_BTR_SEG_TOP;
+
+		n = fseg_n_reserved_pages(seg_header, &dummy, &mtr);
+
+		seg_header = root + PAGE_HEADER + PAGE_BTR_SEG_LEAF;
+
+		n += fseg_n_reserved_pages(seg_header, &dummy, &mtr);
+	} else {
+		ut_error;
+	}
+
+	mtr_commit(&mtr);
+
+	return(n);
+}
+
+/******************************************************************
+Frees a page used in an ibuf tree. Puts the page to the free list of the
+ibuf tree. */
+static
+void
+btr_page_free_for_ibuf(
+/*===================*/
+	dict_index_t*	index,	/* in: index tree */
+	buf_block_t*	block,	/* in: block to be freed, x-latched */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	page_t*		root;
+
+	ut_ad(mtr_memo_contains(mtr, block, MTR_MEMO_PAGE_X_FIX));
+	root = btr_root_get(index, mtr);
+
+	flst_add_first(root + PAGE_HEADER + PAGE_BTR_IBUF_FREE_LIST,
+		       buf_block_get_frame(block)
+		       + PAGE_HEADER + PAGE_BTR_IBUF_FREE_LIST_NODE, mtr);
+
+	ut_ad(flst_validate(root + PAGE_HEADER + PAGE_BTR_IBUF_FREE_LIST,
+			    mtr));
+}
+
+/******************************************************************
+Frees a file page used in an index tree. Can be used also to (BLOB)
+external storage pages, because the page level 0 can be given as an
+argument. */
+UNIV_INTERN
+void
+btr_page_free_low(
+/*==============*/
+	dict_index_t*	index,	/* in: index tree */
+	buf_block_t*	block,	/* in: block to be freed, x-latched */
+	ulint		level,	/* in: page level */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	fseg_header_t*	seg_header;
+	page_t*		root;
+
+	ut_ad(mtr_memo_contains(mtr, block, MTR_MEMO_PAGE_X_FIX));
+	/* The page gets invalid for optimistic searches: increment the frame
+	modify clock */
+
+	buf_block_modify_clock_inc(block);
+
+	if (dict_index_is_ibuf(index)) {
+
+		btr_page_free_for_ibuf(index, block, mtr);
+
+		return;
+	}
+
+	root = btr_root_get(index, mtr);
+
+	if (level == 0) {
+		seg_header = root + PAGE_HEADER + PAGE_BTR_SEG_LEAF;
+	} else {
+		seg_header = root + PAGE_HEADER + PAGE_BTR_SEG_TOP;
+	}
+
+	fseg_free_page(seg_header,
+		       buf_block_get_space(block),
+		       buf_block_get_page_no(block), mtr);
+}
+
+/******************************************************************
+Frees a file page used in an index tree. NOTE: cannot free field external
+storage pages because the page must contain info on its level. */
+UNIV_INTERN
+void
+btr_page_free(
+/*==========*/
+	dict_index_t*	index,	/* in: index tree */
+	buf_block_t*	block,	/* in: block to be freed, x-latched */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	ulint		level;
+
+	level = btr_page_get_level(buf_block_get_frame(block), mtr);
+
+	btr_page_free_low(index, block, level, mtr);
+}
+
+/******************************************************************
+Sets the child node file address in a node pointer. */
+UNIV_INLINE
+void
+btr_node_ptr_set_child_page_no(
+/*===========================*/
+	rec_t*		rec,	/* in: node pointer record */
+	page_zip_des_t*	page_zip,/* in/out: compressed page whose uncompressed
+				part will be updated, or NULL */
+	const ulint*	offsets,/* in: array returned by rec_get_offsets() */
+	ulint		page_no,/* in: child node address */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	byte*	field;
+	ulint	len;
+
+	ut_ad(rec_offs_validate(rec, NULL, offsets));
+	ut_ad(!page_is_leaf(page_align(rec)));
+	ut_ad(!rec_offs_comp(offsets) || rec_get_node_ptr_flag(rec));
+
+	/* The child address is in the last field */
+	field = rec_get_nth_field(rec, offsets,
+				  rec_offs_n_fields(offsets) - 1, &len);
+
+	ut_ad(len == REC_NODE_PTR_SIZE);
+
+	if (UNIV_LIKELY_NULL(page_zip)) {
+		page_zip_write_node_ptr(page_zip, rec,
+					rec_offs_data_size(offsets),
+					page_no, mtr);
+	} else {
+		mlog_write_ulint(field, page_no, MLOG_4BYTES, mtr);
+	}
+}
+
+/****************************************************************
+Returns the child page of a node pointer and x-latches it. */
+static
+buf_block_t*
+btr_node_ptr_get_child(
+/*===================*/
+				/* out: child page, x-latched */
+	const rec_t*	node_ptr,/* in: node pointer */
+	dict_index_t*	index,	/* in: index */
+	const ulint*	offsets,/* in: array returned by rec_get_offsets() */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	ulint	page_no;
+	ulint	space;
+
+	ut_ad(rec_offs_validate(node_ptr, index, offsets));
+	space = page_get_space_id(page_align(node_ptr));
+	page_no = btr_node_ptr_get_child_page_no(node_ptr, offsets);
+
+	return(btr_block_get(space, dict_table_zip_size(index->table),
+			     page_no, RW_X_LATCH, mtr));
+}
+
+/****************************************************************
+Returns the upper level node pointer to a page. It is assumed that mtr holds
+an x-latch on the tree. */
+static
+ulint*
+btr_page_get_father_node_ptr(
+/*=========================*/
+				/* out: rec_get_offsets() of the
+				node pointer record */
+	ulint*		offsets,/* in: work area for the return value */
+	mem_heap_t*	heap,	/* in: memory heap to use */
+	btr_cur_t*	cursor,	/* in: cursor pointing to user record,
+				out: cursor on node pointer record,
+				its page x-latched */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	dtuple_t*	tuple;
+	rec_t*		user_rec;
+	rec_t*		node_ptr;
+	ulint		level;
+	ulint		page_no;
+	dict_index_t*	index;
+
+	page_no = buf_block_get_page_no(btr_cur_get_block(cursor));
+	index = btr_cur_get_index(cursor);
+
+	ut_ad(mtr_memo_contains(mtr, dict_index_get_lock(index),
+				MTR_MEMO_X_LOCK));
+
+	ut_ad(dict_index_get_page(index) != page_no);
+
+	level = btr_page_get_level(btr_cur_get_page(cursor), mtr);
+	user_rec = btr_cur_get_rec(cursor);
+	ut_a(page_rec_is_user_rec(user_rec));
+	tuple = dict_index_build_node_ptr(index, user_rec, 0, heap, level);
+
+	btr_cur_search_to_nth_level(index, level + 1, tuple, PAGE_CUR_LE,
+				    BTR_CONT_MODIFY_TREE, cursor, 0, mtr);
+
+	node_ptr = btr_cur_get_rec(cursor);
+	ut_ad(!page_rec_is_comp(node_ptr)
+	      || rec_get_status(node_ptr) == REC_STATUS_NODE_PTR);
+	offsets = rec_get_offsets(node_ptr, index, offsets,
+				  ULINT_UNDEFINED, &heap);
+
+	if (UNIV_UNLIKELY(btr_node_ptr_get_child_page_no(node_ptr, offsets)
+			  != page_no)) {
+		rec_t*	print_rec;
+		fputs("InnoDB: Dump of the child page:\n", stderr);
+		buf_page_print(page_align(user_rec), 0);
+		fputs("InnoDB: Dump of the parent page:\n", stderr);
+		buf_page_print(page_align(node_ptr), 0);
+
+		fputs("InnoDB: Corruption of an index tree: table ", stderr);
+		ut_print_name(stderr, NULL, TRUE, index->table_name);
+		fputs(", index ", stderr);
+		ut_print_name(stderr, NULL, FALSE, index->name);
+		fprintf(stderr, ",\n"
+			"InnoDB: father ptr page no %lu, child page no %lu\n",
+			(ulong)
+			btr_node_ptr_get_child_page_no(node_ptr, offsets),
+			(ulong) page_no);
+		print_rec = page_rec_get_next(
+			page_get_infimum_rec(page_align(user_rec)));
+		offsets = rec_get_offsets(print_rec, index,
+					  offsets, ULINT_UNDEFINED, &heap);
+		page_rec_print(print_rec, offsets);
+		offsets = rec_get_offsets(node_ptr, index, offsets,
+					  ULINT_UNDEFINED, &heap);
+		page_rec_print(node_ptr, offsets);
+
+		fputs("InnoDB: You should dump + drop + reimport the table"
+		      " to fix the\n"
+		      "InnoDB: corruption. If the crash happens at "
+		      "the database startup, see\n"
+		      "InnoDB: http://dev.mysql.com/doc/refman/5.1/en/"
+		      "forcing-recovery.html about\n"
+		      "InnoDB: forcing recovery. "
+		      "Then dump + drop + reimport.\n", stderr);
+
+		ut_error;
+	}
+
+	return(offsets);
+}
+
+/****************************************************************
+Returns the upper level node pointer to a page. It is assumed that mtr holds
+an x-latch on the tree. */
+static
+ulint*
+btr_page_get_father_block(
+/*======================*/
+				/* out: rec_get_offsets() of the
+				node pointer record */
+	ulint*		offsets,/* in: work area for the return value */
+	mem_heap_t*	heap,	/* in: memory heap to use */
+	dict_index_t*	index,	/* in: b-tree index */
+	buf_block_t*	block,	/* in: child page in the index */
+	mtr_t*		mtr,	/* in: mtr */
+	btr_cur_t*	cursor)	/* out: cursor on node pointer record,
+				its page x-latched */
+{
+	rec_t*	rec
+		= page_rec_get_next(page_get_infimum_rec(buf_block_get_frame(
+								 block)));
+	btr_cur_position(index, rec, block, cursor);
+	return(btr_page_get_father_node_ptr(offsets, heap, cursor, mtr));
+}
+
+/****************************************************************
+Seeks to the upper level node pointer to a page.
+It is assumed that mtr holds an x-latch on the tree. */
+static
+void
+btr_page_get_father(
+/*================*/
+	dict_index_t*	index,	/* in: b-tree index */
+	buf_block_t*	block,	/* in: child page in the index */
+	mtr_t*		mtr,	/* in: mtr */
+	btr_cur_t*	cursor)	/* out: cursor on node pointer record,
+				its page x-latched */
+{
+	mem_heap_t*	heap;
+	rec_t*		rec
+		= page_rec_get_next(page_get_infimum_rec(buf_block_get_frame(
+								 block)));
+	btr_cur_position(index, rec, block, cursor);
+
+	heap = mem_heap_create(100);
+	btr_page_get_father_node_ptr(NULL, heap, cursor, mtr);
+	mem_heap_free(heap);
+}
+
+/****************************************************************
+Creates the root node for a new index tree. */
+UNIV_INTERN
+ulint
+btr_create(
+/*=======*/
+				/* out: page number of the created root,
+				FIL_NULL if did not succeed */
+	ulint		type,	/* in: type of the index */
+	ulint		space,	/* in: space where created */
+	ulint		zip_size,/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	dulint		index_id,/* in: index id */
+	dict_index_t*	index,	/* in: index */
+	mtr_t*		mtr)	/* in: mini-transaction handle */
+{
+	ulint		page_no;
+	buf_block_t*	block;
+	buf_frame_t*	frame;
+	page_t*		page;
+	page_zip_des_t*	page_zip;
+
+	/* Create the two new segments (one, in the case of an ibuf tree) for
+	the index tree; the segment headers are put on the allocated root page
+	(for an ibuf tree, not in the root, but on a separate ibuf header
+	page) */
+
+	if (type & DICT_IBUF) {
+		/* Allocate first the ibuf header page */
+		buf_block_t*	ibuf_hdr_block = fseg_create(
+			space, 0,
+			IBUF_HEADER + IBUF_TREE_SEG_HEADER, mtr);
+
+		buf_block_dbg_add_level(ibuf_hdr_block, SYNC_TREE_NODE_NEW);
+
+		ut_ad(buf_block_get_page_no(ibuf_hdr_block)
+		      == IBUF_HEADER_PAGE_NO);
+		/* Allocate then the next page to the segment: it will be the
+		tree root page */
+
+		page_no = fseg_alloc_free_page(buf_block_get_frame(
+						       ibuf_hdr_block)
+					       + IBUF_HEADER
+					       + IBUF_TREE_SEG_HEADER,
+					       IBUF_TREE_ROOT_PAGE_NO,
+					       FSP_UP, mtr);
+		ut_ad(page_no == IBUF_TREE_ROOT_PAGE_NO);
+
+		block = buf_page_get(space, zip_size, page_no,
+				     RW_X_LATCH, mtr);
+	} else {
+		block = fseg_create(space, 0,
+				    PAGE_HEADER + PAGE_BTR_SEG_TOP, mtr);
+	}
+
+	if (block == NULL) {
+
+		return(FIL_NULL);
+	}
+
+	page_no = buf_block_get_page_no(block);
+	frame = buf_block_get_frame(block);
+
+	buf_block_dbg_add_level(block, SYNC_TREE_NODE_NEW);
+
+	if (type & DICT_IBUF) {
+		/* It is an insert buffer tree: initialize the free list */
+
+		ut_ad(page_no == IBUF_TREE_ROOT_PAGE_NO);
+
+		flst_init(frame + PAGE_HEADER + PAGE_BTR_IBUF_FREE_LIST, mtr);
+	} else {
+		/* It is a non-ibuf tree: create a file segment for leaf
+		pages */
+		fseg_create(space, page_no,
+			    PAGE_HEADER + PAGE_BTR_SEG_LEAF, mtr);
+		/* The fseg create acquires a second latch on the page,
+		therefore we must declare it: */
+		buf_block_dbg_add_level(block, SYNC_TREE_NODE_NEW);
+	}
+
+	/* Create a new index page on the the allocated segment page */
+	page_zip = buf_block_get_page_zip(block);
+
+	if (UNIV_LIKELY_NULL(page_zip)) {
+		page = page_create_zip(block, index, 0, mtr);
+	} else {
+		page = page_create(block, mtr,
+				   dict_table_is_comp(index->table));
+		/* Set the level of the new index page */
+		btr_page_set_level(page, NULL, 0, mtr);
+	}
+
+	block->check_index_page_at_flush = TRUE;
+
+	/* Set the index id of the page */
+	btr_page_set_index_id(page, page_zip, index_id, mtr);
+
+	/* Set the next node and previous node fields */
+	btr_page_set_next(page, page_zip, FIL_NULL, mtr);
+	btr_page_set_prev(page, page_zip, FIL_NULL, mtr);
+
+	/* We reset the free bits for the page to allow creation of several
+	trees in the same mtr, otherwise the latch on a bitmap page would
+	prevent it because of the latching order */
+
+	if (!(type & DICT_CLUSTERED)) {
+		ibuf_reset_free_bits(block);
+	}
+
+	/* In the following assertion we test that two records of maximum
+	allowed size fit on the root page: this fact is needed to ensure
+	correctness of split algorithms */
+
+	ut_ad(page_get_max_insert_size(page, 2) > 2 * BTR_PAGE_MAX_REC_SIZE);
+
+	return(page_no);
+}
+
+/****************************************************************
+Frees a B-tree except the root page, which MUST be freed after this
+by calling btr_free_root. */
+UNIV_INTERN
+void
+btr_free_but_not_root(
+/*==================*/
+	ulint	space,		/* in: space where created */
+	ulint	zip_size,	/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	ulint	root_page_no)	/* in: root page number */
+{
+	ibool	finished;
+	page_t*	root;
+	mtr_t	mtr;
+
+leaf_loop:
+	mtr_start(&mtr);
+
+	root = btr_page_get(space, zip_size, root_page_no, RW_X_LATCH, &mtr);
+#ifdef UNIV_BTR_DEBUG
+	ut_a(btr_root_fseg_validate(FIL_PAGE_DATA + PAGE_BTR_SEG_LEAF
+				    + root, space));
+	ut_a(btr_root_fseg_validate(FIL_PAGE_DATA + PAGE_BTR_SEG_TOP
+				    + root, space));
+#endif /* UNIV_BTR_DEBUG */
+
+	/* NOTE: page hash indexes are dropped when a page is freed inside
+	fsp0fsp. */
+
+	finished = fseg_free_step(root + PAGE_HEADER + PAGE_BTR_SEG_LEAF,
+				  &mtr);
+	mtr_commit(&mtr);
+
+	if (!finished) {
+
+		goto leaf_loop;
+	}
+top_loop:
+	mtr_start(&mtr);
+
+	root = btr_page_get(space, zip_size, root_page_no, RW_X_LATCH, &mtr);
+#ifdef UNIV_BTR_DEBUG
+	ut_a(btr_root_fseg_validate(FIL_PAGE_DATA + PAGE_BTR_SEG_TOP
+				    + root, space));
+#endif /* UNIV_BTR_DEBUG */
+
+	finished = fseg_free_step_not_header(
+		root + PAGE_HEADER + PAGE_BTR_SEG_TOP, &mtr);
+	mtr_commit(&mtr);
+
+	if (!finished) {
+
+		goto top_loop;
+	}
+}
+
+/****************************************************************
+Frees the B-tree root page. Other tree MUST already have been freed. */
+UNIV_INTERN
+void
+btr_free_root(
+/*==========*/
+	ulint	space,		/* in: space where created */
+	ulint	zip_size,	/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	ulint	root_page_no,	/* in: root page number */
+	mtr_t*	mtr)		/* in: a mini-transaction which has already
+				been started */
+{
+	buf_block_t*	block;
+	fseg_header_t*	header;
+
+	block = btr_block_get(space, zip_size, root_page_no, RW_X_LATCH, mtr);
+
+	btr_search_drop_page_hash_index(block);
+
+	header = buf_block_get_frame(block) + PAGE_HEADER + PAGE_BTR_SEG_TOP;
+#ifdef UNIV_BTR_DEBUG
+	ut_a(btr_root_fseg_validate(header, space));
+#endif /* UNIV_BTR_DEBUG */
+
+	while (!fseg_free_step(header, mtr));
+}
+
+/*****************************************************************
+Reorganizes an index page. */
+static
+ibool
+btr_page_reorganize_low(
+/*====================*/
+	ibool		recovery,/* in: TRUE if called in recovery:
+				locks should not be updated, i.e.,
+				there cannot exist locks on the
+				page, and a hash index should not be
+				dropped: it cannot exist */
+	buf_block_t*	block,	/* in: page to be reorganized */
+	dict_index_t*	index,	/* in: record descriptor */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	page_t*		page		= buf_block_get_frame(block);
+	page_zip_des_t*	page_zip	= buf_block_get_page_zip(block);
+	buf_block_t*	temp_block;
+	page_t*		temp_page;
+	ulint		log_mode;
+	ulint		data_size1;
+	ulint		data_size2;
+	ulint		max_ins_size1;
+	ulint		max_ins_size2;
+	ibool		success		= FALSE;
+
+	ut_ad(mtr_memo_contains(mtr, block, MTR_MEMO_PAGE_X_FIX));
+	ut_ad(!!page_is_comp(page) == dict_table_is_comp(index->table));
+#ifdef UNIV_ZIP_DEBUG
+	ut_a(!page_zip || page_zip_validate(page_zip, page));
+#endif /* UNIV_ZIP_DEBUG */
+	data_size1 = page_get_data_size(page);
+	max_ins_size1 = page_get_max_insert_size_after_reorganize(page, 1);
+
+	/* Write the log record */
+	mlog_open_and_write_index(mtr, page, index, page_is_comp(page)
+				  ? MLOG_COMP_PAGE_REORGANIZE
+				  : MLOG_PAGE_REORGANIZE, 0);
+
+	/* Turn logging off */
+	log_mode = mtr_set_log_mode(mtr, MTR_LOG_NONE);
+
+	temp_block = buf_block_alloc(0);
+	temp_page = temp_block->frame;
+
+	/* Copy the old page to temporary space */
+	buf_frame_copy(temp_page, page);
+
+	if (UNIV_LIKELY(!recovery)) {
+		btr_search_drop_page_hash_index(block);
+	}
+
+	/* Recreate the page: note that global data on page (possible
+	segment headers, next page-field, etc.) is preserved intact */
+
+	page_create(block, mtr, dict_table_is_comp(index->table));
+	block->check_index_page_at_flush = TRUE;
+
+	/* Copy the records from the temporary space to the recreated page;
+	do not copy the lock bits yet */
+
+	page_copy_rec_list_end_no_locks(block, temp_block,
+					page_get_infimum_rec(temp_page),
+					index, mtr);
+	/* Copy max trx id to recreated page */
+	page_set_max_trx_id(block, NULL, page_get_max_trx_id(temp_page));
+
+	if (UNIV_LIKELY_NULL(page_zip)
+	    && UNIV_UNLIKELY
+	    (!page_zip_compress(page_zip, page, index, NULL))) {
+
+		/* Restore the old page and exit. */
+		buf_frame_copy(page, temp_page);
+
+		goto func_exit;
+	}
+
+	if (UNIV_LIKELY(!recovery)) {
+		/* Update the record lock bitmaps */
+		lock_move_reorganize_page(block, temp_block);
+	}
+
+	data_size2 = page_get_data_size(page);
+	max_ins_size2 = page_get_max_insert_size_after_reorganize(page, 1);
+
+	if (UNIV_UNLIKELY(data_size1 != data_size2)
+	    || UNIV_UNLIKELY(max_ins_size1 != max_ins_size2)) {
+		buf_page_print(page, 0);
+		buf_page_print(temp_page, 0);
+		fprintf(stderr,
+			"InnoDB: Error: page old data size %lu"
+			" new data size %lu\n"
+			"InnoDB: Error: page old max ins size %lu"
+			" new max ins size %lu\n"
+			"InnoDB: Submit a detailed bug report"
+			" to http://bugs.mysql.com\n",
+			(unsigned long) data_size1, (unsigned long) data_size2,
+			(unsigned long) max_ins_size1,
+			(unsigned long) max_ins_size2);
+	} else {
+		success = TRUE;
+	}
+
+func_exit:
+#ifdef UNIV_ZIP_DEBUG
+	ut_a(!page_zip || page_zip_validate(page_zip, page));
+#endif /* UNIV_ZIP_DEBUG */
+	buf_block_free(temp_block);
+
+	/* Restore logging mode */
+	mtr_set_log_mode(mtr, log_mode);
+
+	return(success);
+}
+
+/*****************************************************************
+Reorganizes an index page.
+IMPORTANT: if btr_page_reorganize() is invoked on a compressed leaf
+page of a non-clustered index, the caller must update the insert
+buffer free bits in the same mini-transaction in such a way that the
+modification will be redo-logged. */
+UNIV_INTERN
+ibool
+btr_page_reorganize(
+/*================*/
+				/* out: TRUE on success, FALSE on failure */
+	buf_block_t*	block,	/* in: page to be reorganized */
+	dict_index_t*	index,	/* in: record descriptor */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	return(btr_page_reorganize_low(FALSE, block, index, mtr));
+}
+
+/***************************************************************
+Parses a redo log record of reorganizing a page. */
+UNIV_INTERN
+byte*
+btr_parse_page_reorganize(
+/*======================*/
+				/* out: end of log record or NULL */
+	byte*		ptr,	/* in: buffer */
+	byte*		end_ptr __attribute__((unused)),
+				/* in: buffer end */
+	dict_index_t*	index,	/* in: record descriptor */
+	buf_block_t*	block,	/* in: page to be reorganized, or NULL */
+	mtr_t*		mtr)	/* in: mtr or NULL */
+{
+	ut_ad(ptr && end_ptr);
+
+	/* The record is empty, except for the record initial part */
+
+	if (UNIV_LIKELY(block != NULL)) {
+		btr_page_reorganize_low(TRUE, block, index, mtr);
+	}
+
+	return(ptr);
+}
+
+/*****************************************************************
+Empties an index page.  @see btr_page_create().*/
+static
+void
+btr_page_empty(
+/*===========*/
+	buf_block_t*	block,	/* in: page to be emptied */
+	page_zip_des_t*	page_zip,/* out: compressed page, or NULL */
+	dict_index_t*	index,	/* in: index of the page */
+	ulint		level,	/* in: the B-tree level of the page */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	page_t*	page = buf_block_get_frame(block);
+
+	ut_ad(mtr_memo_contains(mtr, block, MTR_MEMO_PAGE_X_FIX));
+	ut_ad(page_zip == buf_block_get_page_zip(block));
+#ifdef UNIV_ZIP_DEBUG
+	ut_a(!page_zip || page_zip_validate(page_zip, page));
+#endif /* UNIV_ZIP_DEBUG */
+
+	btr_search_drop_page_hash_index(block);
+
+	/* Recreate the page: note that global data on page (possible
+	segment headers, next page-field, etc.) is preserved intact */
+
+	if (UNIV_LIKELY_NULL(page_zip)) {
+		page_create_zip(block, index, level, mtr);
+	} else {
+		page_create(block, mtr, dict_table_is_comp(index->table));
+		btr_page_set_level(page, NULL, level, mtr);
+	}
+
+	block->check_index_page_at_flush = TRUE;
+}
+
+/*****************************************************************
+Makes tree one level higher by splitting the root, and inserts
+the tuple. It is assumed that mtr contains an x-latch on the tree.
+NOTE that the operation of this function must always succeed,
+we cannot reverse it: therefore enough free disk space must be
+guaranteed to be available before this function is called. */
+UNIV_INTERN
+rec_t*
+btr_root_raise_and_insert(
+/*======================*/
+				/* out: inserted record */
+	btr_cur_t*	cursor,	/* in: cursor at which to insert: must be
+				on the root page; when the function returns,
+				the cursor is positioned on the predecessor
+				of the inserted record */
+	const dtuple_t*	tuple,	/* in: tuple to insert */
+	ulint		n_ext,	/* in: number of externally stored columns */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	dict_index_t*	index;
+	page_t*		root;
+	page_t*		new_page;
+	ulint		new_page_no;
+	rec_t*		rec;
+	mem_heap_t*	heap;
+	dtuple_t*	node_ptr;
+	ulint		level;
+	rec_t*		node_ptr_rec;
+	page_cur_t*	page_cursor;
+	page_zip_des_t*	root_page_zip;
+	page_zip_des_t*	new_page_zip;
+	buf_block_t*	root_block;
+	buf_block_t*	new_block;
+
+	root = btr_cur_get_page(cursor);
+	root_block = btr_cur_get_block(cursor);
+	root_page_zip = buf_block_get_page_zip(root_block);
+#ifdef UNIV_ZIP_DEBUG
+	ut_a(!root_page_zip || page_zip_validate(root_page_zip, root));
+#endif /* UNIV_ZIP_DEBUG */
+	index = btr_cur_get_index(cursor);
+#ifdef UNIV_BTR_DEBUG
+	if (!dict_index_is_ibuf(index)) {
+		ulint	space = dict_index_get_space(index);
+
+		ut_a(btr_root_fseg_validate(FIL_PAGE_DATA + PAGE_BTR_SEG_LEAF
+					    + root, space));
+		ut_a(btr_root_fseg_validate(FIL_PAGE_DATA + PAGE_BTR_SEG_TOP
+					    + root, space));
+	}
+
+	ut_a(dict_index_get_page(index) == page_get_page_no(root));
+#endif /* UNIV_BTR_DEBUG */
+	ut_ad(mtr_memo_contains(mtr, dict_index_get_lock(index),
+				MTR_MEMO_X_LOCK));
+	ut_ad(mtr_memo_contains(mtr, root_block, MTR_MEMO_PAGE_X_FIX));
+
+	/* Allocate a new page to the tree. Root splitting is done by first
+	moving the root records to the new page, emptying the root, putting
+	a node pointer to the new page, and then splitting the new page. */
+
+	level = btr_page_get_level(root, mtr);
+
+	new_block = btr_page_alloc(index, 0, FSP_NO_DIR, level, mtr);
+	new_page = buf_block_get_frame(new_block);
+	new_page_zip = buf_block_get_page_zip(new_block);
+	ut_a(!new_page_zip == !root_page_zip);
+	ut_a(!new_page_zip
+	     || page_zip_get_size(new_page_zip)
+	     == page_zip_get_size(root_page_zip));
+
+	btr_page_create(new_block, new_page_zip, index, level, mtr);
+
+	/* Set the next node and previous node fields of new page */
+	btr_page_set_next(new_page, new_page_zip, FIL_NULL, mtr);
+	btr_page_set_prev(new_page, new_page_zip, FIL_NULL, mtr);
+
+	/* Copy the records from root to the new page one by one. */
+
+	if (0
+#ifdef UNIV_ZIP_COPY
+	    || new_page_zip
+#endif /* UNIV_ZIP_COPY */
+	    || UNIV_UNLIKELY
+	    (!page_copy_rec_list_end(new_block, root_block,
+				     page_get_infimum_rec(root),
+				     index, mtr))) {
+		ut_a(new_page_zip);
+
+		/* Copy the page byte for byte. */
+		page_zip_copy_recs(new_page_zip, new_page,
+				   root_page_zip, root, index, mtr);
+
+		/* Update the lock table and possible hash index. */
+
+		lock_move_rec_list_end(new_block, root_block,
+				       page_get_infimum_rec(root));
+
+		btr_search_move_or_delete_hash_entries(new_block, root_block,
+						       index);
+	}
+
+	/* If this is a pessimistic insert which is actually done to
+	perform a pessimistic update then we have stored the lock
+	information of the record to be inserted on the infimum of the
+	root page: we cannot discard the lock structs on the root page */
+
+	lock_update_root_raise(new_block, root_block);
+
+	/* Create a memory heap where the node pointer is stored */
+	heap = mem_heap_create(100);
+
+	rec = page_rec_get_next(page_get_infimum_rec(new_page));
+	new_page_no = buf_block_get_page_no(new_block);
+
+	/* Build the node pointer (= node key and page address) for the
+	child */
+
+	node_ptr = dict_index_build_node_ptr(index, rec, new_page_no, heap,
+					     level);
+	/* The node pointer must be marked as the predefined minimum record,
+	as there is no lower alphabetical limit to records in the leftmost
+	node of a level: */
+	dtuple_set_info_bits(node_ptr,
+			     dtuple_get_info_bits(node_ptr)
+			     | REC_INFO_MIN_REC_FLAG);
+
+	/* Rebuild the root page to get free space */
+	btr_page_empty(root_block, root_page_zip, index, level + 1, mtr);
+
+	/* Set the next node and previous node fields, although
+	they should already have been set.  The previous node field
+	must be FIL_NULL if root_page_zip != NULL, because the
+	REC_INFO_MIN_REC_FLAG (of the first user record) will be
+	set if and only if btr_page_get_prev() == FIL_NULL. */
+	btr_page_set_next(root, root_page_zip, FIL_NULL, mtr);
+	btr_page_set_prev(root, root_page_zip, FIL_NULL, mtr);
+
+	page_cursor = btr_cur_get_page_cur(cursor);
+
+	/* Insert node pointer to the root */
+
+	page_cur_set_before_first(root_block, page_cursor);
+
+	node_ptr_rec = page_cur_tuple_insert(page_cursor, node_ptr,
+					     index, 0, mtr);
+
+	/* The root page should only contain the node pointer
+	to new_page at this point.  Thus, the data should fit. */
+	ut_a(node_ptr_rec);
+
+	/* Free the memory heap */
+	mem_heap_free(heap);
+
+	/* We play safe and reset the free bits for the new page */
+
+#if 0
+	fprintf(stderr, "Root raise new page no %lu\n", new_page_no);
+#endif
+
+	if (!dict_index_is_clust(index)) {
+		ibuf_reset_free_bits(new_block);
+	}
+
+	/* Reposition the cursor to the child node */
+	page_cur_search(new_block, index, tuple,
+			PAGE_CUR_LE, page_cursor);
+
+	/* Split the child and insert tuple */
+	return(btr_page_split_and_insert(cursor, tuple, n_ext, mtr));
+}
+
+/*****************************************************************
+Decides if the page should be split at the convergence point of inserts
+converging to the left. */
+UNIV_INTERN
+ibool
+btr_page_get_split_rec_to_left(
+/*===========================*/
+				/* out: TRUE if split recommended */
+	btr_cur_t*	cursor,	/* in: cursor at which to insert */
+	rec_t**		split_rec) /* out: if split recommended,
+				the first record on upper half page,
+				or NULL if tuple to be inserted should
+				be first */
+{
+	page_t*	page;
+	rec_t*	insert_point;
+	rec_t*	infimum;
+
+	page = btr_cur_get_page(cursor);
+	insert_point = btr_cur_get_rec(cursor);
+
+	if (page_header_get_ptr(page, PAGE_LAST_INSERT)
+	    == page_rec_get_next(insert_point)) {
+
+		infimum = page_get_infimum_rec(page);
+
+		/* If the convergence is in the middle of a page, include also
+		the record immediately before the new insert to the upper
+		page. Otherwise, we could repeatedly move from page to page
+		lots of records smaller than the convergence point. */
+
+		if (infimum != insert_point
+		    && page_rec_get_next(infimum) != insert_point) {
+
+			*split_rec = insert_point;
+		} else {
+			*split_rec = page_rec_get_next(insert_point);
+		}
+
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
+
+/*****************************************************************
+Decides if the page should be split at the convergence point of inserts
+converging to the right. */
+UNIV_INTERN
+ibool
+btr_page_get_split_rec_to_right(
+/*============================*/
+				/* out: TRUE if split recommended */
+	btr_cur_t*	cursor,	/* in: cursor at which to insert */
+	rec_t**		split_rec) /* out: if split recommended,
+				the first record on upper half page,
+				or NULL if tuple to be inserted should
+				be first */
+{
+	page_t*	page;
+	rec_t*	insert_point;
+
+	page = btr_cur_get_page(cursor);
+	insert_point = btr_cur_get_rec(cursor);
+
+	/* We use eager heuristics: if the new insert would be right after
+	the previous insert on the same page, we assume that there is a
+	pattern of sequential inserts here. */
+
+	if (UNIV_LIKELY(page_header_get_ptr(page, PAGE_LAST_INSERT)
+			== insert_point)) {
+
+		rec_t*	next_rec;
+
+		next_rec = page_rec_get_next(insert_point);
+
+		if (page_rec_is_supremum(next_rec)) {
+split_at_new:
+			/* Split at the new record to insert */
+			*split_rec = NULL;
+		} else {
+			rec_t*	next_next_rec = page_rec_get_next(next_rec);
+			if (page_rec_is_supremum(next_next_rec)) {
+
+				goto split_at_new;
+			}
+
+			/* If there are >= 2 user records up from the insert
+			point, split all but 1 off. We want to keep one because
+			then sequential inserts can use the adaptive hash
+			index, as they can do the necessary checks of the right
+			search position just by looking at the records on this
+			page. */
+
+			*split_rec = next_next_rec;
+		}
+
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
+
+/*****************************************************************
+Calculates a split record such that the tuple will certainly fit on
+its half-page when the split is performed. We assume in this function
+only that the cursor page has at least one user record. */
+static
+rec_t*
+btr_page_get_sure_split_rec(
+/*========================*/
+				/* out: split record, or NULL if tuple
+				will be the first record on upper half-page */
+	btr_cur_t*	cursor,	/* in: cursor at which insert should be made */
+	const dtuple_t*	tuple,	/* in: tuple to insert */
+	ulint		n_ext)	/* in: number of externally stored columns */
+{
+	page_t*		page;
+	page_zip_des_t*	page_zip;
+	ulint		insert_size;
+	ulint		free_space;
+	ulint		total_data;
+	ulint		total_n_recs;
+	ulint		total_space;
+	ulint		incl_data;
+	rec_t*		ins_rec;
+	rec_t*		rec;
+	rec_t*		next_rec;
+	ulint		n;
+	mem_heap_t*	heap;
+	ulint*		offsets;
+
+	page = btr_cur_get_page(cursor);
+
+	insert_size = rec_get_converted_size(cursor->index, tuple, n_ext);
+	free_space  = page_get_free_space_of_empty(page_is_comp(page));
+
+	page_zip = btr_cur_get_page_zip(cursor);
+	if (UNIV_LIKELY_NULL(page_zip)) {
+		/* Estimate the free space of an empty compressed page. */
+		ulint	free_space_zip = page_zip_empty_size(
+			cursor->index->n_fields,
+			page_zip_get_size(page_zip));
+
+		if (UNIV_LIKELY(free_space > (ulint) free_space_zip)) {
+			free_space = (ulint) free_space_zip;
+		}
+	}
+
+	/* free_space is now the free space of a created new page */
+
+	total_data   = page_get_data_size(page) + insert_size;
+	total_n_recs = page_get_n_recs(page) + 1;
+	ut_ad(total_n_recs >= 2);
+	total_space  = total_data + page_dir_calc_reserved_space(total_n_recs);
+
+	n = 0;
+	incl_data = 0;
+	ins_rec = btr_cur_get_rec(cursor);
+	rec = page_get_infimum_rec(page);
+
+	heap = NULL;
+	offsets = NULL;
+
+	/* We start to include records to the left half, and when the
+	space reserved by them exceeds half of total_space, then if
+	the included records fit on the left page, they will be put there
+	if something was left over also for the right page,
+	otherwise the last included record will be the first on the right
+	half page */
+
+	do {
+		/* Decide the next record to include */
+		if (rec == ins_rec) {
+			rec = NULL;	/* NULL denotes that tuple is
+					now included */
+		} else if (rec == NULL) {
+			rec = page_rec_get_next(ins_rec);
+		} else {
+			rec = page_rec_get_next(rec);
+		}
+
+		if (rec == NULL) {
+			/* Include tuple */
+			incl_data += insert_size;
+		} else {
+			offsets = rec_get_offsets(rec, cursor->index,
+						  offsets, ULINT_UNDEFINED,
+						  &heap);
+			incl_data += rec_offs_size(offsets);
+		}
+
+		n++;
+	} while (incl_data + page_dir_calc_reserved_space(n)
+		 < total_space / 2);
+
+	if (incl_data + page_dir_calc_reserved_space(n) <= free_space) {
+		/* The next record will be the first on
+		the right half page if it is not the
+		supremum record of page */
+
+		if (rec == ins_rec) {
+			rec = NULL;
+
+			goto func_exit;
+		} else if (rec == NULL) {
+			next_rec = page_rec_get_next(ins_rec);
+		} else {
+			next_rec = page_rec_get_next(rec);
+		}
+		ut_ad(next_rec);
+		if (!page_rec_is_supremum(next_rec)) {
+			rec = next_rec;
+		}
+	}
+
+func_exit:
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+	return(rec);
+}
+
+/*****************************************************************
+Returns TRUE if the insert fits on the appropriate half-page with the
+chosen split_rec. */
+static
+ibool
+btr_page_insert_fits(
+/*=================*/
+				/* out: TRUE if fits */
+	btr_cur_t*	cursor,	/* in: cursor at which insert
+				should be made */
+	const rec_t*	split_rec,/* in: suggestion for first record
+				on upper half-page, or NULL if
+				tuple to be inserted should be first */
+	const ulint*	offsets,/* in: rec_get_offsets(
+				split_rec, cursor->index) */
+	const dtuple_t*	tuple,	/* in: tuple to insert */
+	ulint		n_ext,	/* in: number of externally stored columns */
+	mem_heap_t*	heap)	/* in: temporary memory heap */
+{
+	page_t*		page;
+	ulint		insert_size;
+	ulint		free_space;
+	ulint		total_data;
+	ulint		total_n_recs;
+	const rec_t*	rec;
+	const rec_t*	end_rec;
+	ulint*		offs;
+
+	page = btr_cur_get_page(cursor);
+
+	ut_ad(!split_rec == !offsets);
+	ut_ad(!offsets
+	      || !page_is_comp(page) == !rec_offs_comp(offsets));
+	ut_ad(!offsets
+	      || rec_offs_validate(split_rec, cursor->index, offsets));
+
+	insert_size = rec_get_converted_size(cursor->index, tuple, n_ext);
+	free_space  = page_get_free_space_of_empty(page_is_comp(page));
+
+	/* free_space is now the free space of a created new page */
+
+	total_data   = page_get_data_size(page) + insert_size;
+	total_n_recs = page_get_n_recs(page) + 1;
+
+	/* We determine which records (from rec to end_rec, not including
+	end_rec) will end up on the other half page from tuple when it is
+	inserted. */
+
+	if (split_rec == NULL) {
+		rec = page_rec_get_next(page_get_infimum_rec(page));
+		end_rec = page_rec_get_next(btr_cur_get_rec(cursor));
+
+	} else if (cmp_dtuple_rec(tuple, split_rec, offsets) >= 0) {
+
+		rec = page_rec_get_next(page_get_infimum_rec(page));
+		end_rec = split_rec;
+	} else {
+		rec = split_rec;
+		end_rec = page_get_supremum_rec(page);
+	}
+
+	if (total_data + page_dir_calc_reserved_space(total_n_recs)
+	    <= free_space) {
+
+		/* Ok, there will be enough available space on the
+		half page where the tuple is inserted */
+
+		return(TRUE);
+	}
+
+	offs = NULL;
+
+	while (rec != end_rec) {
+		/* In this loop we calculate the amount of reserved
+		space after rec is removed from page. */
+
+		offs = rec_get_offsets(rec, cursor->index, offs,
+				       ULINT_UNDEFINED, &heap);
+
+		total_data -= rec_offs_size(offs);
+		total_n_recs--;
+
+		if (total_data + page_dir_calc_reserved_space(total_n_recs)
+		    <= free_space) {
+
+			/* Ok, there will be enough available space on the
+			half page where the tuple is inserted */
+
+			return(TRUE);
+		}
+
+		rec = page_rec_get_next_const(rec);
+	}
+
+	return(FALSE);
+}
+
+/***********************************************************
+Inserts a data tuple to a tree on a non-leaf level. It is assumed
+that mtr holds an x-latch on the tree. */
+UNIV_INTERN
+void
+btr_insert_on_non_leaf_level(
+/*=========================*/
+	dict_index_t*	index,	/* in: index */
+	ulint		level,	/* in: level, must be > 0 */
+	dtuple_t*	tuple,	/* in: the record to be inserted */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	big_rec_t*	dummy_big_rec;
+	btr_cur_t	cursor;
+	ulint		err;
+	rec_t*		rec;
+
+	ut_ad(level > 0);
+
+	btr_cur_search_to_nth_level(index, level, tuple, PAGE_CUR_LE,
+				    BTR_CONT_MODIFY_TREE,
+				    &cursor, 0, mtr);
+
+	err = btr_cur_pessimistic_insert(BTR_NO_LOCKING_FLAG
+					 | BTR_KEEP_SYS_FLAG
+					 | BTR_NO_UNDO_LOG_FLAG,
+					 &cursor, tuple, &rec,
+					 &dummy_big_rec, 0, NULL, mtr);
+	ut_a(err == DB_SUCCESS);
+}
+
+/******************************************************************
+Attaches the halves of an index page on the appropriate level in an
+index tree. */
+static
+void
+btr_attach_half_pages(
+/*==================*/
+	dict_index_t*	index,		/* in: the index tree */
+	buf_block_t*	block,		/* in/out: page to be split */
+	rec_t*		split_rec,	/* in: first record on upper
+					half page */
+	buf_block_t*	new_block,	/* in/out: the new half page */
+	ulint		direction,	/* in: FSP_UP or FSP_DOWN */
+	mtr_t*		mtr)		/* in: mtr */
+{
+	ulint		space;
+	ulint		zip_size;
+	ulint		prev_page_no;
+	ulint		next_page_no;
+	ulint		level;
+	page_t*		page		= buf_block_get_frame(block);
+	page_t*		lower_page;
+	page_t*		upper_page;
+	ulint		lower_page_no;
+	ulint		upper_page_no;
+	page_zip_des_t*	lower_page_zip;
+	page_zip_des_t*	upper_page_zip;
+	dtuple_t*	node_ptr_upper;
+	mem_heap_t*	heap;
+
+	ut_ad(mtr_memo_contains(mtr, block, MTR_MEMO_PAGE_X_FIX));
+	ut_ad(mtr_memo_contains(mtr, new_block, MTR_MEMO_PAGE_X_FIX));
+
+	/* Create a memory heap where the data tuple is stored */
+	heap = mem_heap_create(1024);
+
+	/* Based on split direction, decide upper and lower pages */
+	if (direction == FSP_DOWN) {
+
+		btr_cur_t	cursor;
+		ulint*		offsets;
+
+		lower_page = buf_block_get_frame(new_block);
+		lower_page_no = buf_block_get_page_no(new_block);
+		lower_page_zip = buf_block_get_page_zip(new_block);
+		upper_page = buf_block_get_frame(block);
+		upper_page_no = buf_block_get_page_no(block);
+		upper_page_zip = buf_block_get_page_zip(block);
+
+		/* Look up the index for the node pointer to page */
+		offsets = btr_page_get_father_block(NULL, heap, index,
+						    block, mtr, &cursor);
+
+		/* Replace the address of the old child node (= page) with the
+		address of the new lower half */
+
+		btr_node_ptr_set_child_page_no(
+			btr_cur_get_rec(&cursor),
+			btr_cur_get_page_zip(&cursor),
+			offsets, lower_page_no, mtr);
+		mem_heap_empty(heap);
+	} else {
+		lower_page = buf_block_get_frame(block);
+		lower_page_no = buf_block_get_page_no(block);
+		lower_page_zip = buf_block_get_page_zip(block);
+		upper_page = buf_block_get_frame(new_block);
+		upper_page_no = buf_block_get_page_no(new_block);
+		upper_page_zip = buf_block_get_page_zip(new_block);
+	}
+
+	/* Get the level of the split pages */
+	level = btr_page_get_level(buf_block_get_frame(block), mtr);
+	ut_ad(level
+	      == btr_page_get_level(buf_block_get_frame(new_block), mtr));
+
+	/* Build the node pointer (= node key and page address) for the upper
+	half */
+
+	node_ptr_upper = dict_index_build_node_ptr(index, split_rec,
+						   upper_page_no, heap, level);
+
+	/* Insert it next to the pointer to the lower half. Note that this
+	may generate recursion leading to a split on the higher level. */
+
+	btr_insert_on_non_leaf_level(index, level + 1, node_ptr_upper, mtr);
+
+	/* Free the memory heap */
+	mem_heap_free(heap);
+
+	/* Get the previous and next pages of page */
+
+	prev_page_no = btr_page_get_prev(page, mtr);
+	next_page_no = btr_page_get_next(page, mtr);
+	space = buf_block_get_space(block);
+	zip_size = buf_block_get_zip_size(block);
+
+	/* Update page links of the level */
+
+	if (prev_page_no != FIL_NULL) {
+		buf_block_t*	prev_block = btr_block_get(space, zip_size,
+							   prev_page_no,
+							   RW_X_LATCH, mtr);
+#ifdef UNIV_BTR_DEBUG
+		ut_a(page_is_comp(prev_block->frame) == page_is_comp(page));
+		ut_a(btr_page_get_next(prev_block->frame, mtr)
+		     == buf_block_get_page_no(block));
+#endif /* UNIV_BTR_DEBUG */
+
+		btr_page_set_next(buf_block_get_frame(prev_block),
+				  buf_block_get_page_zip(prev_block),
+				  lower_page_no, mtr);
+	}
+
+	if (next_page_no != FIL_NULL) {
+		buf_block_t*	next_block = btr_block_get(space, zip_size,
+							   next_page_no,
+							   RW_X_LATCH, mtr);
+#ifdef UNIV_BTR_DEBUG
+		ut_a(page_is_comp(next_block->frame) == page_is_comp(page));
+		ut_a(btr_page_get_prev(next_block->frame, mtr)
+		     == page_get_page_no(page));
+#endif /* UNIV_BTR_DEBUG */
+
+		btr_page_set_prev(buf_block_get_frame(next_block),
+				  buf_block_get_page_zip(next_block),
+				  upper_page_no, mtr);
+	}
+
+	btr_page_set_prev(lower_page, lower_page_zip, prev_page_no, mtr);
+	btr_page_set_next(lower_page, lower_page_zip, upper_page_no, mtr);
+
+	btr_page_set_prev(upper_page, upper_page_zip, lower_page_no, mtr);
+	btr_page_set_next(upper_page, upper_page_zip, next_page_no, mtr);
+}
+
+/*****************************************************************
+Splits an index page to halves and inserts the tuple. It is assumed
+that mtr holds an x-latch to the index tree. NOTE: the tree x-latch
+is released within this function! NOTE that the operation of this
+function must always succeed, we cannot reverse it: therefore
+enough free disk space must be guaranteed to be available before
+this function is called. */
+UNIV_INTERN
+rec_t*
+btr_page_split_and_insert(
+/*======================*/
+				/* out: inserted record; NOTE: the tree
+				x-latch is released! NOTE: 2 free disk
+				pages must be available! */
+	btr_cur_t*	cursor,	/* in: cursor at which to insert; when the
+				function returns, the cursor is positioned
+				on the predecessor of the inserted record */
+	const dtuple_t*	tuple,	/* in: tuple to insert */
+	ulint		n_ext,	/* in: number of externally stored columns */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	buf_block_t*	block;
+	page_t*		page;
+	page_zip_des_t*	page_zip;
+	ulint		page_no;
+	byte		direction;
+	ulint		hint_page_no;
+	buf_block_t*	new_block;
+	page_t*		new_page;
+	page_zip_des_t*	new_page_zip;
+	rec_t*		split_rec;
+	buf_block_t*	left_block;
+	buf_block_t*	right_block;
+	buf_block_t*	insert_block;
+	page_t*		insert_page;
+	page_cur_t*	page_cursor;
+	rec_t*		first_rec;
+	byte*		buf = 0; /* remove warning */
+	rec_t*		move_limit;
+	ibool		insert_will_fit;
+	ibool		insert_left;
+	ulint		n_iterations = 0;
+	rec_t*		rec;
+	mem_heap_t*	heap;
+	ulint		n_uniq;
+	ulint*		offsets;
+
+	heap = mem_heap_create(1024);
+	n_uniq = dict_index_get_n_unique_in_tree(cursor->index);
+func_start:
+	mem_heap_empty(heap);
+	offsets = NULL;
+
+	ut_ad(mtr_memo_contains(mtr, dict_index_get_lock(cursor->index),
+				MTR_MEMO_X_LOCK));
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(rw_lock_own(dict_index_get_lock(cursor->index), RW_LOCK_EX));
+#endif /* UNIV_SYNC_DEBUG */
+
+	block = btr_cur_get_block(cursor);
+	page = buf_block_get_frame(block);
+	page_zip = buf_block_get_page_zip(block);
+
+	ut_ad(mtr_memo_contains(mtr, block, MTR_MEMO_PAGE_X_FIX));
+	ut_ad(page_get_n_recs(page) >= 1);
+
+	page_no = buf_block_get_page_no(block);
+
+	/* 1. Decide the split record; split_rec == NULL means that the
+	tuple to be inserted should be the first record on the upper
+	half-page */
+
+	if (n_iterations > 0) {
+		direction = FSP_UP;
+		hint_page_no = page_no + 1;
+		split_rec = btr_page_get_sure_split_rec(cursor, tuple, n_ext);
+
+	} else if (btr_page_get_split_rec_to_right(cursor, &split_rec)) {
+		direction = FSP_UP;
+		hint_page_no = page_no + 1;
+
+	} else if (btr_page_get_split_rec_to_left(cursor, &split_rec)) {
+		direction = FSP_DOWN;
+		hint_page_no = page_no - 1;
+	} else {
+		direction = FSP_UP;
+		hint_page_no = page_no + 1;
+
+		if (page_get_n_recs(page) == 1) {
+			page_cur_t	pcur;
+
+			/* There is only one record in the index page
+			therefore we can't split the node in the middle
+			by default. We need to determine whether the
+			new record will be inserted to the left or right. */
+
+			/* Read the first (and only) record in the page. */
+			page_cur_set_before_first(block, &pcur);
+			page_cur_move_to_next(&pcur);
+			first_rec = page_cur_get_rec(&pcur);
+
+			offsets = rec_get_offsets(
+				first_rec, cursor->index, offsets,
+				n_uniq, &heap);
+
+			/* If the new record is less than the existing record
+			the the split in the middle will copy the existing
+			record to the new node. */
+			if (cmp_dtuple_rec(tuple, first_rec, offsets) < 0) {
+				split_rec = page_get_middle_rec(page);
+			} else {
+				split_rec = NULL;
+			}
+		} else {
+			split_rec = page_get_middle_rec(page);
+		}
+	}
+
+	/* 2. Allocate a new page to the index */
+	new_block = btr_page_alloc(cursor->index, hint_page_no, direction,
+				   btr_page_get_level(page, mtr), mtr);
+	new_page = buf_block_get_frame(new_block);
+	new_page_zip = buf_block_get_page_zip(new_block);
+	btr_page_create(new_block, new_page_zip, cursor->index,
+			btr_page_get_level(page, mtr), mtr);
+
+	/* 3. Calculate the first record on the upper half-page, and the
+	first record (move_limit) on original page which ends up on the
+	upper half */
+
+	if (split_rec) {
+		first_rec = move_limit = split_rec;
+
+		offsets = rec_get_offsets(split_rec, cursor->index, offsets,
+					  n_uniq, &heap);
+
+		insert_left = cmp_dtuple_rec(tuple, split_rec, offsets) < 0;
+
+		if (UNIV_UNLIKELY(!insert_left && new_page_zip
+				  && n_iterations > 0)) {
+			/* If a compressed page has already been split,
+			avoid further splits by inserting the record
+			to an empty page. */
+			split_rec = NULL;
+			goto insert_right;
+		}
+	} else {
+insert_right:
+		insert_left = FALSE;
+		buf = mem_alloc(rec_get_converted_size(cursor->index,
+						       tuple, n_ext));
+
+		first_rec = rec_convert_dtuple_to_rec(buf, cursor->index,
+						      tuple, n_ext);
+		move_limit = page_rec_get_next(btr_cur_get_rec(cursor));
+	}
+
+	/* 4. Do first the modifications in the tree structure */
+
+	btr_attach_half_pages(cursor->index, block,
+			      first_rec, new_block, direction, mtr);
+
+	/* If the split is made on the leaf level and the insert will fit
+	on the appropriate half-page, we may release the tree x-latch.
+	We can then move the records after releasing the tree latch,
+	thus reducing the tree latch contention. */
+
+	if (split_rec) {
+		insert_will_fit = !new_page_zip
+			&& btr_page_insert_fits(cursor, split_rec,
+						offsets, tuple, n_ext, heap);
+	} else {
+		mem_free(buf);
+		insert_will_fit = !new_page_zip
+			&& btr_page_insert_fits(cursor, NULL,
+						NULL, tuple, n_ext, heap);
+	}
+
+	if (insert_will_fit && page_is_leaf(page)) {
+
+		mtr_memo_release(mtr, dict_index_get_lock(cursor->index),
+				 MTR_MEMO_X_LOCK);
+	}
+
+	/* 5. Move then the records to the new page */
+	if (direction == FSP_DOWN) {
+		/*		fputs("Split left\n", stderr); */
+
+		if (0
+#ifdef UNIV_ZIP_COPY
+		    || page_zip
+#endif /* UNIV_ZIP_COPY */
+		    || UNIV_UNLIKELY
+		    (!page_move_rec_list_start(new_block, block, move_limit,
+					       cursor->index, mtr))) {
+			/* For some reason, compressing new_page failed,
+			even though it should contain fewer records than
+			the original page.  Copy the page byte for byte
+			and then delete the records from both pages
+			as appropriate.  Deleting will always succeed. */
+			ut_a(new_page_zip);
+
+			page_zip_copy_recs(new_page_zip, new_page,
+					   page_zip, page, cursor->index, mtr);
+			page_delete_rec_list_end(move_limit - page + new_page,
+						 new_block, cursor->index,
+						 ULINT_UNDEFINED,
+						 ULINT_UNDEFINED, mtr);
+
+			/* Update the lock table and possible hash index. */
+
+			lock_move_rec_list_start(
+				new_block, block, move_limit,
+				new_page + PAGE_NEW_INFIMUM);
+
+			btr_search_move_or_delete_hash_entries(
+				new_block, block, cursor->index);
+
+			/* Delete the records from the source page. */
+
+			page_delete_rec_list_start(move_limit, block,
+						   cursor->index, mtr);
+		}
+
+		left_block = new_block;
+		right_block = block;
+
+		lock_update_split_left(right_block, left_block);
+	} else {
+		/*		fputs("Split right\n", stderr); */
+
+		if (0
+#ifdef UNIV_ZIP_COPY
+		    || page_zip
+#endif /* UNIV_ZIP_COPY */
+		    || UNIV_UNLIKELY
+		    (!page_move_rec_list_end(new_block, block, move_limit,
+					     cursor->index, mtr))) {
+			/* For some reason, compressing new_page failed,
+			even though it should contain fewer records than
+			the original page.  Copy the page byte for byte
+			and then delete the records from both pages
+			as appropriate.  Deleting will always succeed. */
+			ut_a(new_page_zip);
+
+			page_zip_copy_recs(new_page_zip, new_page,
+					   page_zip, page, cursor->index, mtr);
+			page_delete_rec_list_start(move_limit - page
+						   + new_page, new_block,
+						   cursor->index, mtr);
+
+			/* Update the lock table and possible hash index. */
+
+			lock_move_rec_list_end(new_block, block, move_limit);
+
+			btr_search_move_or_delete_hash_entries(
+				new_block, block, cursor->index);
+
+			/* Delete the records from the source page. */
+
+			page_delete_rec_list_end(move_limit, block,
+						 cursor->index,
+						 ULINT_UNDEFINED,
+						 ULINT_UNDEFINED, mtr);
+		}
+
+		left_block = block;
+		right_block = new_block;
+
+		lock_update_split_right(right_block, left_block);
+	}
+
+#ifdef UNIV_ZIP_DEBUG
+	if (UNIV_LIKELY_NULL(page_zip)) {
+		ut_a(page_zip_validate(page_zip, page));
+		ut_a(page_zip_validate(new_page_zip, new_page));
+	}
+#endif /* UNIV_ZIP_DEBUG */
+
+	/* At this point, split_rec, move_limit and first_rec may point
+	to garbage on the old page. */
+
+	/* 6. The split and the tree modification is now completed. Decide the
+	page where the tuple should be inserted */
+
+	if (insert_left) {
+		insert_block = left_block;
+	} else {
+		insert_block = right_block;
+	}
+
+	insert_page = buf_block_get_frame(insert_block);
+
+	/* 7. Reposition the cursor for insert and try insertion */
+	page_cursor = btr_cur_get_page_cur(cursor);
+
+	page_cur_search(insert_block, cursor->index, tuple,
+			PAGE_CUR_LE, page_cursor);
+
+	rec = page_cur_tuple_insert(page_cursor, tuple,
+				    cursor->index, n_ext, mtr);
+
+#ifdef UNIV_ZIP_DEBUG
+	{
+		page_zip_des_t*	insert_page_zip
+			= buf_block_get_page_zip(insert_block);
+		ut_a(!insert_page_zip
+		     || page_zip_validate(insert_page_zip, insert_page));
+	}
+#endif /* UNIV_ZIP_DEBUG */
+
+	if (UNIV_LIKELY(rec != NULL)) {
+
+		goto func_exit;
+	}
+
+	/* 8. If insert did not fit, try page reorganization */
+
+	if (UNIV_UNLIKELY
+	    (!btr_page_reorganize(insert_block, cursor->index, mtr))) {
+
+		goto insert_failed;
+	}
+
+	page_cur_search(insert_block, cursor->index, tuple,
+			PAGE_CUR_LE, page_cursor);
+	rec = page_cur_tuple_insert(page_cursor, tuple, cursor->index,
+				    n_ext, mtr);
+
+	if (UNIV_UNLIKELY(rec == NULL)) {
+		/* The insert did not fit on the page: loop back to the
+		start of the function for a new split */
+insert_failed:
+		/* We play safe and reset the free bits for new_page */
+		if (!dict_index_is_clust(cursor->index)) {
+			ibuf_reset_free_bits(new_block);
+		}
+
+		/* fprintf(stderr, "Split second round %lu\n",
+		page_get_page_no(page)); */
+		n_iterations++;
+		ut_ad(n_iterations < 2
+		      || buf_block_get_page_zip(insert_block));
+		ut_ad(!insert_will_fit);
+
+		goto func_start;
+	}
+
+func_exit:
+	/* Insert fit on the page: update the free bits for the
+	left and right pages in the same mtr */
+
+	if (!dict_index_is_clust(cursor->index) && page_is_leaf(page)) {
+		ibuf_update_free_bits_for_two_pages_low(
+			buf_block_get_zip_size(left_block),
+			left_block, right_block, mtr);
+	}
+
+#if 0
+	fprintf(stderr, "Split and insert done %lu %lu\n",
+		buf_block_get_page_no(left_block),
+		buf_block_get_page_no(right_block));
+#endif
+
+	ut_ad(page_validate(buf_block_get_frame(left_block), cursor->index));
+	ut_ad(page_validate(buf_block_get_frame(right_block), cursor->index));
+
+	mem_heap_free(heap);
+	return(rec);
+}
+
+/*****************************************************************
+Removes a page from the level list of pages. */
+static
+void
+btr_level_list_remove(
+/*==================*/
+	ulint		space,	/* in: space where removed */
+	ulint		zip_size,/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	page_t*		page,	/* in: page to remove */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	ulint	prev_page_no;
+	ulint	next_page_no;
+
+	ut_ad(page && mtr);
+	ut_ad(mtr_memo_contains_page(mtr, page, MTR_MEMO_PAGE_X_FIX));
+	ut_ad(space == page_get_space_id(page));
+	/* Get the previous and next page numbers of page */
+
+	prev_page_no = btr_page_get_prev(page, mtr);
+	next_page_no = btr_page_get_next(page, mtr);
+
+	/* Update page links of the level */
+
+	if (prev_page_no != FIL_NULL) {
+		buf_block_t*	prev_block
+			= btr_block_get(space, zip_size, prev_page_no,
+					RW_X_LATCH, mtr);
+		page_t*		prev_page
+			= buf_block_get_frame(prev_block);
+#ifdef UNIV_BTR_DEBUG
+		ut_a(page_is_comp(prev_page) == page_is_comp(page));
+		ut_a(btr_page_get_next(prev_page, mtr)
+		     == page_get_page_no(page));
+#endif /* UNIV_BTR_DEBUG */
+
+		btr_page_set_next(prev_page,
+				  buf_block_get_page_zip(prev_block),
+				  next_page_no, mtr);
+	}
+
+	if (next_page_no != FIL_NULL) {
+		buf_block_t*	next_block
+			= btr_block_get(space, zip_size, next_page_no,
+					RW_X_LATCH, mtr);
+		page_t*		next_page
+			= buf_block_get_frame(next_block);
+#ifdef UNIV_BTR_DEBUG
+		ut_a(page_is_comp(next_page) == page_is_comp(page));
+		ut_a(btr_page_get_prev(next_page, mtr)
+		     == page_get_page_no(page));
+#endif /* UNIV_BTR_DEBUG */
+
+		btr_page_set_prev(next_page,
+				  buf_block_get_page_zip(next_block),
+				  prev_page_no, mtr);
+	}
+}
+
+/********************************************************************
+Writes the redo log record for setting an index record as the predefined
+minimum record. */
+UNIV_INLINE
+void
+btr_set_min_rec_mark_log(
+/*=====================*/
+	rec_t*	rec,	/* in: record */
+	byte	type,	/* in: MLOG_COMP_REC_MIN_MARK or MLOG_REC_MIN_MARK */
+	mtr_t*	mtr)	/* in: mtr */
+{
+	mlog_write_initial_log_record(rec, type, mtr);
+
+	/* Write rec offset as a 2-byte ulint */
+	mlog_catenate_ulint(mtr, page_offset(rec), MLOG_2BYTES);
+}
+
+/********************************************************************
+Parses the redo log record for setting an index record as the predefined
+minimum record. */
+UNIV_INTERN
+byte*
+btr_parse_set_min_rec_mark(
+/*=======================*/
+			/* out: end of log record or NULL */
+	byte*	ptr,	/* in: buffer */
+	byte*	end_ptr,/* in: buffer end */
+	ulint	comp,	/* in: nonzero=compact page format */
+	page_t*	page,	/* in: page or NULL */
+	mtr_t*	mtr)	/* in: mtr or NULL */
+{
+	rec_t*	rec;
+
+	if (end_ptr < ptr + 2) {
+
+		return(NULL);
+	}
+
+	if (page) {
+		ut_a(!page_is_comp(page) == !comp);
+
+		rec = page + mach_read_from_2(ptr);
+
+		btr_set_min_rec_mark(rec, mtr);
+	}
+
+	return(ptr + 2);
+}
+
+/********************************************************************
+Sets a record as the predefined minimum record. */
+UNIV_INTERN
+void
+btr_set_min_rec_mark(
+/*=================*/
+	rec_t*	rec,	/* in: record */
+	mtr_t*	mtr)	/* in: mtr */
+{
+	ulint	info_bits;
+
+	if (UNIV_LIKELY(page_rec_is_comp(rec))) {
+		info_bits = rec_get_info_bits(rec, TRUE);
+
+		rec_set_info_bits_new(rec, info_bits | REC_INFO_MIN_REC_FLAG);
+
+		btr_set_min_rec_mark_log(rec, MLOG_COMP_REC_MIN_MARK, mtr);
+	} else {
+		info_bits = rec_get_info_bits(rec, FALSE);
+
+		rec_set_info_bits_old(rec, info_bits | REC_INFO_MIN_REC_FLAG);
+
+		btr_set_min_rec_mark_log(rec, MLOG_REC_MIN_MARK, mtr);
+	}
+}
+
+/*****************************************************************
+Deletes on the upper level the node pointer to a page. */
+UNIV_INTERN
+void
+btr_node_ptr_delete(
+/*================*/
+	dict_index_t*	index,	/* in: index tree */
+	buf_block_t*	block,	/* in: page whose node pointer is deleted */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	btr_cur_t	cursor;
+	ibool		compressed;
+	ulint		err;
+
+	ut_ad(mtr_memo_contains(mtr, block, MTR_MEMO_PAGE_X_FIX));
+
+	/* Delete node pointer on father page */
+	btr_page_get_father(index, block, mtr, &cursor);
+
+	compressed = btr_cur_pessimistic_delete(&err, TRUE, &cursor, RB_NONE,
+						mtr);
+	ut_a(err == DB_SUCCESS);
+
+	if (!compressed) {
+		btr_cur_compress_if_useful(&cursor, mtr);
+	}
+}
+
+/*****************************************************************
+If page is the only on its level, this function moves its records to the
+father page, thus reducing the tree height. */
+static
+void
+btr_lift_page_up(
+/*=============*/
+	dict_index_t*	index,	/* in: index tree */
+	buf_block_t*	block,	/* in: page which is the only on its level;
+				must not be empty: use
+				btr_discard_only_page_on_level if the last
+				record from the page should be removed */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	buf_block_t*	father_block;
+	page_t*		father_page;
+	ulint		page_level;
+	page_zip_des_t*	father_page_zip;
+	page_t*		page		= buf_block_get_frame(block);
+	ulint		root_page_no;
+	buf_block_t*	blocks[BTR_MAX_LEVELS];
+	ulint		n_blocks;	/* last used index in blocks[] */
+	ulint		i;
+
+	ut_ad(btr_page_get_prev(page, mtr) == FIL_NULL);
+	ut_ad(btr_page_get_next(page, mtr) == FIL_NULL);
+	ut_ad(mtr_memo_contains(mtr, block, MTR_MEMO_PAGE_X_FIX));
+
+	page_level = btr_page_get_level(page, mtr);
+	root_page_no = dict_index_get_page(index);
+
+	{
+		btr_cur_t	cursor;
+		mem_heap_t*	heap	= mem_heap_create(100);
+		ulint*		offsets;
+		buf_block_t*	b;
+
+		offsets = btr_page_get_father_block(NULL, heap, index,
+						    block, mtr, &cursor);
+		father_block = btr_cur_get_block(&cursor);
+		father_page_zip = buf_block_get_page_zip(father_block);
+		father_page = buf_block_get_frame(father_block);
+
+		n_blocks = 0;
+
+		/* Store all ancestor pages so we can reset their
+		levels later on.  We have to do all the searches on
+		the tree now because later on, after we've replaced
+		the first level, the tree is in an inconsistent state
+		and can not be searched. */
+		for (b = father_block;
+		     buf_block_get_page_no(b) != root_page_no; ) {
+			ut_a(n_blocks < BTR_MAX_LEVELS);
+
+			offsets = btr_page_get_father_block(offsets, heap,
+							    index, b,
+							    mtr, &cursor);
+
+			blocks[n_blocks++] = b = btr_cur_get_block(&cursor);
+		}
+
+		mem_heap_free(heap);
+	}
+
+	btr_search_drop_page_hash_index(block);
+
+	/* Make the father empty */
+	btr_page_empty(father_block, father_page_zip, index, page_level, mtr);
+
+	/* Copy the records to the father page one by one. */
+	if (0
+#ifdef UNIV_ZIP_COPY
+	    || father_page_zip
+#endif /* UNIV_ZIP_COPY */
+	    || UNIV_UNLIKELY
+	    (!page_copy_rec_list_end(father_block, block,
+				     page_get_infimum_rec(page),
+				     index, mtr))) {
+		const page_zip_des_t*	page_zip
+			= buf_block_get_page_zip(block);
+		ut_a(father_page_zip);
+		ut_a(page_zip);
+
+		/* Copy the page byte for byte. */
+		page_zip_copy_recs(father_page_zip, father_page,
+				   page_zip, page, index, mtr);
+
+		/* Update the lock table and possible hash index. */
+
+		lock_move_rec_list_end(father_block, block,
+				       page_get_infimum_rec(page));
+
+		btr_search_move_or_delete_hash_entries(father_block, block,
+						       index);
+	}
+
+	lock_update_copy_and_discard(father_block, block);
+
+	/* Go upward to root page, decrementing levels by one. */
+	for (i = 0; i < n_blocks; i++, page_level++) {
+		page_t*		page	= buf_block_get_frame(blocks[i]);
+		page_zip_des_t*	page_zip= buf_block_get_page_zip(blocks[i]);
+
+		ut_ad(btr_page_get_level(page, mtr) == page_level + 1);
+
+		btr_page_set_level(page, page_zip, page_level, mtr);
+#ifdef UNIV_ZIP_DEBUG
+		ut_a(!page_zip || page_zip_validate(page_zip, page));
+#endif /* UNIV_ZIP_DEBUG */
+	}
+
+	/* Free the file page */
+	btr_page_free(index, block, mtr);
+
+	/* We play it safe and reset the free bits for the father */
+	if (!dict_index_is_clust(index)) {
+		ibuf_reset_free_bits(father_block);
+	}
+	ut_ad(page_validate(father_page, index));
+	ut_ad(btr_check_node_ptr(index, father_block, mtr));
+}
+
+/*****************************************************************
+Tries to merge the page first to the left immediate brother if such a
+brother exists, and the node pointers to the current page and to the brother
+reside on the same page. If the left brother does not satisfy these
+conditions, looks at the right brother. If the page is the only one on that
+level lifts the records of the page to the father page, thus reducing the
+tree height. It is assumed that mtr holds an x-latch on the tree and on the
+page. If cursor is on the leaf level, mtr must also hold x-latches to the
+brothers, if they exist. */
+UNIV_INTERN
+ibool
+btr_compress(
+/*=========*/
+				/* out: TRUE on success */
+	btr_cur_t*	cursor,	/* in: cursor on the page to merge or lift;
+				the page must not be empty: in record delete
+				use btr_discard_page if the page would become
+				empty */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	dict_index_t*	index;
+	ulint		space;
+	ulint		zip_size;
+	ulint		left_page_no;
+	ulint		right_page_no;
+	buf_block_t*	merge_block;
+	page_t*		merge_page;
+	page_zip_des_t*	merge_page_zip;
+	ibool		is_left;
+	buf_block_t*	block;
+	page_t*		page;
+	btr_cur_t	father_cursor;
+	mem_heap_t*	heap;
+	ulint*		offsets;
+	ulint		data_size;
+	ulint		n_recs;
+	ulint		max_ins_size;
+	ulint		max_ins_size_reorg;
+	ulint		level;
+
+	block = btr_cur_get_block(cursor);
+	page = btr_cur_get_page(cursor);
+	index = btr_cur_get_index(cursor);
+	ut_a((ibool) !!page_is_comp(page) == dict_table_is_comp(index->table));
+
+	ut_ad(mtr_memo_contains(mtr, dict_index_get_lock(index),
+				MTR_MEMO_X_LOCK));
+	ut_ad(mtr_memo_contains(mtr, block, MTR_MEMO_PAGE_X_FIX));
+	level = btr_page_get_level(page, mtr);
+	space = dict_index_get_space(index);
+	zip_size = dict_table_zip_size(index->table);
+
+	left_page_no = btr_page_get_prev(page, mtr);
+	right_page_no = btr_page_get_next(page, mtr);
+
+#if 0
+	fprintf(stderr, "Merge left page %lu right %lu \n",
+		left_page_no, right_page_no);
+#endif
+
+	heap = mem_heap_create(100);
+	offsets = btr_page_get_father_block(NULL, heap, index, block, mtr,
+					    &father_cursor);
+
+	/* Decide the page to which we try to merge and which will inherit
+	the locks */
+
+	is_left = left_page_no != FIL_NULL;
+
+	if (is_left) {
+
+		merge_block = btr_block_get(space, zip_size, left_page_no,
+					    RW_X_LATCH, mtr);
+		merge_page = buf_block_get_frame(merge_block);
+#ifdef UNIV_BTR_DEBUG
+		ut_a(btr_page_get_next(merge_page, mtr)
+		     == buf_block_get_page_no(block));
+#endif /* UNIV_BTR_DEBUG */
+	} else if (right_page_no != FIL_NULL) {
+
+		merge_block = btr_block_get(space, zip_size, right_page_no,
+					    RW_X_LATCH, mtr);
+		merge_page = buf_block_get_frame(merge_block);
+#ifdef UNIV_BTR_DEBUG
+		ut_a(btr_page_get_prev(merge_page, mtr)
+		     == buf_block_get_page_no(block));
+#endif /* UNIV_BTR_DEBUG */
+	} else {
+		/* The page is the only one on the level, lift the records
+		to the father */
+		btr_lift_page_up(index, block, mtr);
+		mem_heap_free(heap);
+		return(TRUE);
+	}
+
+	n_recs = page_get_n_recs(page);
+	data_size = page_get_data_size(page);
+#ifdef UNIV_BTR_DEBUG
+	ut_a(page_is_comp(merge_page) == page_is_comp(page));
+#endif /* UNIV_BTR_DEBUG */
+
+	max_ins_size_reorg = page_get_max_insert_size_after_reorganize(
+		merge_page, n_recs);
+	if (data_size > max_ins_size_reorg) {
+
+		/* No space for merge */
+err_exit:
+		/* We play it safe and reset the free bits. */
+		if (zip_size
+		    && page_is_leaf(merge_page)
+		    && !dict_index_is_clust(index)) {
+			ibuf_reset_free_bits(merge_block);
+		}
+
+		mem_heap_free(heap);
+		return(FALSE);
+	}
+
+	ut_ad(page_validate(merge_page, index));
+
+	max_ins_size = page_get_max_insert_size(merge_page, n_recs);
+
+	if (UNIV_UNLIKELY(data_size > max_ins_size)) {
+
+		/* We have to reorganize merge_page */
+
+		if (UNIV_UNLIKELY(!btr_page_reorganize(merge_block,
+						       index, mtr))) {
+
+			goto err_exit;
+		}
+
+		max_ins_size = page_get_max_insert_size(merge_page, n_recs);
+
+		ut_ad(page_validate(merge_page, index));
+		ut_ad(max_ins_size == max_ins_size_reorg);
+
+		if (UNIV_UNLIKELY(data_size > max_ins_size)) {
+
+			/* Add fault tolerance, though this should
+			never happen */
+
+			goto err_exit;
+		}
+	}
+
+	merge_page_zip = buf_block_get_page_zip(merge_block);
+#ifdef UNIV_ZIP_DEBUG
+	if (UNIV_LIKELY_NULL(merge_page_zip)) {
+		const page_zip_des_t*	page_zip
+			= buf_block_get_page_zip(block);
+		ut_a(page_zip);
+		ut_a(page_zip_validate(merge_page_zip, merge_page));
+		ut_a(page_zip_validate(page_zip, page));
+	}
+#endif /* UNIV_ZIP_DEBUG */
+
+	/* Move records to the merge page */
+	if (is_left) {
+		rec_t*	orig_pred = page_copy_rec_list_start(
+			merge_block, block, page_get_supremum_rec(page),
+			index, mtr);
+
+		if (UNIV_UNLIKELY(!orig_pred)) {
+			goto err_exit;
+		}
+
+		btr_search_drop_page_hash_index(block);
+
+		/* Remove the page from the level list */
+		btr_level_list_remove(space, zip_size, page, mtr);
+
+		btr_node_ptr_delete(index, block, mtr);
+		lock_update_merge_left(merge_block, orig_pred, block);
+	} else {
+		rec_t*		orig_succ;
+#ifdef UNIV_BTR_DEBUG
+		byte		fil_page_prev[4];
+#endif /* UNIV_BTR_DEBUG */
+
+		if (UNIV_LIKELY_NULL(merge_page_zip)) {
+			/* The function page_zip_compress(), which will be
+			invoked by page_copy_rec_list_end() below,
+			requires that FIL_PAGE_PREV be FIL_NULL.
+			Clear the field, but prepare to restore it. */
+#ifdef UNIV_BTR_DEBUG
+			memcpy(fil_page_prev, merge_page + FIL_PAGE_PREV, 4);
+#endif /* UNIV_BTR_DEBUG */
+#if FIL_NULL != 0xffffffff
+# error "FIL_NULL != 0xffffffff"
+#endif
+			memset(merge_page + FIL_PAGE_PREV, 0xff, 4);
+		}
+
+		orig_succ = page_copy_rec_list_end(merge_block, block,
+						   page_get_infimum_rec(page),
+						   cursor->index, mtr);
+
+		if (UNIV_UNLIKELY(!orig_succ)) {
+			ut_a(merge_page_zip);
+#ifdef UNIV_BTR_DEBUG
+			/* FIL_PAGE_PREV was restored from merge_page_zip. */
+			ut_a(!memcmp(fil_page_prev,
+				     merge_page + FIL_PAGE_PREV, 4));
+#endif /* UNIV_BTR_DEBUG */
+			goto err_exit;
+		}
+
+		btr_search_drop_page_hash_index(block);
+
+#ifdef UNIV_BTR_DEBUG
+		if (UNIV_LIKELY_NULL(merge_page_zip)) {
+			/* Restore FIL_PAGE_PREV in order to avoid an assertion
+			failure in btr_level_list_remove(), which will set
+			the field again to FIL_NULL.  Even though this makes
+			merge_page and merge_page_zip inconsistent for a
+			split second, it is harmless, because the pages
+			are X-latched. */
+			memcpy(merge_page + FIL_PAGE_PREV, fil_page_prev, 4);
+		}
+#endif /* UNIV_BTR_DEBUG */
+
+		/* Remove the page from the level list */
+		btr_level_list_remove(space, zip_size, page, mtr);
+
+		/* Replace the address of the old child node (= page) with the
+		address of the merge page to the right */
+
+		btr_node_ptr_set_child_page_no(
+			btr_cur_get_rec(&father_cursor),
+			btr_cur_get_page_zip(&father_cursor),
+			offsets, right_page_no, mtr);
+		btr_node_ptr_delete(index, merge_block, mtr);
+
+		lock_update_merge_right(merge_block, orig_succ, block);
+	}
+
+	mem_heap_free(heap);
+
+	if (!dict_index_is_clust(index) && page_is_leaf(merge_page)) {
+		/* Update the free bits of the B-tree page in the
+		insert buffer bitmap.  This has to be done in a
+		separate mini-transaction that is committed before the
+		main mini-transaction.  We cannot update the insert
+		buffer bitmap in this mini-transaction, because
+		btr_compress() can be invoked recursively without
+		committing the mini-transaction in between.  Since
+		insert buffer bitmap pages have a lower rank than
+		B-tree pages, we must not access other pages in the
+		same mini-transaction after accessing an insert buffer
+		bitmap page. */
+
+		/* The free bits in the insert buffer bitmap must
+		never exceed the free space on a page.  It is safe to
+		decrement or reset the bits in the bitmap in a
+		mini-transaction that is committed before the
+		mini-transaction that affects the free space. */
+
+		/* It is unsafe to increment the bits in a separately
+		committed mini-transaction, because in crash recovery,
+		the free bits could momentarily be set too high. */
+
+		if (zip_size) {
+			/* Because the free bits may be incremented
+			and we cannot update the insert buffer bitmap
+			in the same mini-transaction, the only safe
+			thing we can do here is the pessimistic
+			approach: reset the free bits. */
+			ibuf_reset_free_bits(merge_block);
+		} else {
+			/* On uncompressed pages, the free bits will
+			never increase here.  Thus, it is safe to
+			write the bits accurately in a separate
+			mini-transaction. */
+			ibuf_update_free_bits_if_full(merge_block,
+						      UNIV_PAGE_SIZE,
+						      ULINT_UNDEFINED);
+		}
+	}
+
+	ut_ad(page_validate(merge_page, index));
+#ifdef UNIV_ZIP_DEBUG
+	ut_a(!merge_page_zip || page_zip_validate(merge_page_zip, merge_page));
+#endif /* UNIV_ZIP_DEBUG */
+
+	/* Free the file page */
+	btr_page_free(index, block, mtr);
+
+	ut_ad(btr_check_node_ptr(index, merge_block, mtr));
+	return(TRUE);
+}
+
+/*****************************************************************
+Discards a page that is the only page on its level.  This will empty
+the whole B-tree, leaving just an empty root page.  This function
+should never be reached, because btr_compress(), which is invoked in
+delete operations, calls btr_lift_page_up() to flatten the B-tree. */
+static
+void
+btr_discard_only_page_on_level(
+/*===========================*/
+	dict_index_t*	index,	/* in: index tree */
+	buf_block_t*	block,	/* in: page which is the only on its level */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	ulint	page_level = 0;
+
+	while (buf_block_get_page_no(block) != dict_index_get_page(index)) {
+		btr_cur_t	cursor;
+		buf_block_t*	father;
+		const page_t*	page	= buf_block_get_frame(block);
+
+		ut_a(page_get_n_recs(page) == 1);
+		ut_a(page_level == btr_page_get_level(page, mtr));
+		ut_a(btr_page_get_prev(page, mtr) == FIL_NULL);
+		ut_a(btr_page_get_next(page, mtr) == FIL_NULL);
+
+		ut_ad(mtr_memo_contains(mtr, block, MTR_MEMO_PAGE_X_FIX));
+		btr_search_drop_page_hash_index(block);
+
+		btr_page_get_father(index, block, mtr, &cursor);
+		father = btr_cur_get_block(&cursor);
+
+		lock_update_discard(father, PAGE_HEAP_NO_SUPREMUM, block);
+
+		/* Free the file page */
+		btr_page_free(index, block, mtr);
+
+		block = father;
+		page_level++;
+	}
+
+	/* block is the root page, which must be empty, except
+	for the node pointer to the (now discarded) block(s). */
+
+#ifdef UNIV_BTR_DEBUG
+	if (!dict_index_is_ibuf(index)) {
+		const page_t*	root	= buf_block_get_frame(block);
+		const ulint	space	= dict_index_get_space(index);
+		ut_a(btr_root_fseg_validate(FIL_PAGE_DATA + PAGE_BTR_SEG_LEAF
+					    + root, space));
+		ut_a(btr_root_fseg_validate(FIL_PAGE_DATA + PAGE_BTR_SEG_TOP
+					    + root, space));
+	}
+#endif /* UNIV_BTR_DEBUG */
+
+	btr_page_empty(block, buf_block_get_page_zip(block), index, 0, mtr);
+
+	/* We play it safe and reset the free bits for the root */
+	if (!dict_index_is_clust(index)) {
+		ibuf_reset_free_bits(block);
+	}
+}
+
+/*****************************************************************
+Discards a page from a B-tree. This is used to remove the last record from
+a B-tree page: the whole page must be removed at the same time. This cannot
+be used for the root page, which is allowed to be empty. */
+UNIV_INTERN
+void
+btr_discard_page(
+/*=============*/
+	btr_cur_t*	cursor,	/* in: cursor on the page to discard: not on
+				the root page */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	dict_index_t*	index;
+	ulint		space;
+	ulint		zip_size;
+	ulint		left_page_no;
+	ulint		right_page_no;
+	buf_block_t*	merge_block;
+	page_t*		merge_page;
+	buf_block_t*	block;
+	page_t*		page;
+	rec_t*		node_ptr;
+
+	block = btr_cur_get_block(cursor);
+	index = btr_cur_get_index(cursor);
+
+	ut_ad(dict_index_get_page(index) != buf_block_get_page_no(block));
+	ut_ad(mtr_memo_contains(mtr, dict_index_get_lock(index),
+				MTR_MEMO_X_LOCK));
+	ut_ad(mtr_memo_contains(mtr, block, MTR_MEMO_PAGE_X_FIX));
+	space = dict_index_get_space(index);
+	zip_size = dict_table_zip_size(index->table);
+
+	/* Decide the page which will inherit the locks */
+
+	left_page_no = btr_page_get_prev(buf_block_get_frame(block), mtr);
+	right_page_no = btr_page_get_next(buf_block_get_frame(block), mtr);
+
+	if (left_page_no != FIL_NULL) {
+		merge_block = btr_block_get(space, zip_size, left_page_no,
+					    RW_X_LATCH, mtr);
+		merge_page = buf_block_get_frame(merge_block);
+#ifdef UNIV_BTR_DEBUG
+		ut_a(btr_page_get_next(merge_page, mtr)
+		     == buf_block_get_page_no(block));
+#endif /* UNIV_BTR_DEBUG */
+	} else if (right_page_no != FIL_NULL) {
+		merge_block = btr_block_get(space, zip_size, right_page_no,
+					    RW_X_LATCH, mtr);
+		merge_page = buf_block_get_frame(merge_block);
+#ifdef UNIV_BTR_DEBUG
+		ut_a(btr_page_get_prev(merge_page, mtr)
+		     == buf_block_get_page_no(block));
+#endif /* UNIV_BTR_DEBUG */
+	} else {
+		btr_discard_only_page_on_level(index, block, mtr);
+
+		return;
+	}
+
+	page = buf_block_get_frame(block);
+	ut_a(page_is_comp(merge_page) == page_is_comp(page));
+	btr_search_drop_page_hash_index(block);
+
+	if (left_page_no == FIL_NULL && !page_is_leaf(page)) {
+
+		/* We have to mark the leftmost node pointer on the right
+		side page as the predefined minimum record */
+		node_ptr = page_rec_get_next(page_get_infimum_rec(merge_page));
+
+		ut_ad(page_rec_is_user_rec(node_ptr));
+
+		/* This will make page_zip_validate() fail on merge_page
+		until btr_level_list_remove() completes.  This is harmless,
+		because everything will take place within a single
+		mini-transaction and because writing to the redo log
+		is an atomic operation (performed by mtr_commit()). */
+		btr_set_min_rec_mark(node_ptr, mtr);
+	}
+
+	btr_node_ptr_delete(index, block, mtr);
+
+	/* Remove the page from the level list */
+	btr_level_list_remove(space, zip_size, page, mtr);
+#ifdef UNIV_ZIP_DEBUG
+	{
+		page_zip_des_t*	merge_page_zip
+			= buf_block_get_page_zip(merge_block);
+		ut_a(!merge_page_zip
+		     || page_zip_validate(merge_page_zip, merge_page));
+	}
+#endif /* UNIV_ZIP_DEBUG */
+
+	if (left_page_no != FIL_NULL) {
+		lock_update_discard(merge_block, PAGE_HEAP_NO_SUPREMUM,
+				    block);
+	} else {
+		lock_update_discard(merge_block,
+				    lock_get_min_heap_no(merge_block),
+				    block);
+	}
+
+	/* Free the file page */
+	btr_page_free(index, block, mtr);
+
+	ut_ad(btr_check_node_ptr(index, merge_block, mtr));
+}
+
+#ifdef UNIV_BTR_PRINT
+/*****************************************************************
+Prints size info of a B-tree. */
+UNIV_INTERN
+void
+btr_print_size(
+/*===========*/
+	dict_index_t*	index)	/* in: index tree */
+{
+	page_t*		root;
+	fseg_header_t*	seg;
+	mtr_t		mtr;
+
+	if (dict_index_is_ibuf(index)) {
+		fputs("Sorry, cannot print info of an ibuf tree:"
+		      " use ibuf functions\n", stderr);
+
+		return;
+	}
+
+	mtr_start(&mtr);
+
+	root = btr_root_get(index, &mtr);
+
+	seg = root + PAGE_HEADER + PAGE_BTR_SEG_TOP;
+
+	fputs("INFO OF THE NON-LEAF PAGE SEGMENT\n", stderr);
+	fseg_print(seg, &mtr);
+
+	if (!(index->type & DICT_UNIVERSAL)) {
+
+		seg = root + PAGE_HEADER + PAGE_BTR_SEG_LEAF;
+
+		fputs("INFO OF THE LEAF PAGE SEGMENT\n", stderr);
+		fseg_print(seg, &mtr);
+	}
+
+	mtr_commit(&mtr);
+}
+
+/****************************************************************
+Prints recursively index tree pages. */
+static
+void
+btr_print_recursive(
+/*================*/
+	dict_index_t*	index,	/* in: index tree */
+	buf_block_t*	block,	/* in: index page */
+	ulint		width,	/* in: print this many entries from start
+				and end */
+	mem_heap_t**	heap,	/* in/out: heap for rec_get_offsets() */
+	ulint**		offsets,/* in/out: buffer for rec_get_offsets() */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	const page_t*	page	= buf_block_get_frame(block);
+	page_cur_t	cursor;
+	ulint		n_recs;
+	ulint		i	= 0;
+	mtr_t		mtr2;
+
+	ut_ad(mtr_memo_contains(mtr, block, MTR_MEMO_PAGE_X_FIX));
+	fprintf(stderr, "NODE ON LEVEL %lu page number %lu\n",
+		(ulong) btr_page_get_level(page, mtr),
+		(ulong) buf_block_get_page_no(block));
+
+	page_print(block, index, width, width);
+
+	n_recs = page_get_n_recs(page);
+
+	page_cur_set_before_first(block, &cursor);
+	page_cur_move_to_next(&cursor);
+
+	while (!page_cur_is_after_last(&cursor)) {
+
+		if (page_is_leaf(page)) {
+
+			/* If this is the leaf level, do nothing */
+
+		} else if ((i <= width) || (i >= n_recs - width)) {
+
+			const rec_t*	node_ptr;
+
+			mtr_start(&mtr2);
+
+			node_ptr = page_cur_get_rec(&cursor);
+
+			*offsets = rec_get_offsets(node_ptr, index, *offsets,
+						   ULINT_UNDEFINED, heap);
+			btr_print_recursive(index,
+					    btr_node_ptr_get_child(node_ptr,
+								   index,
+								   *offsets,
+								   &mtr2),
+					    width, heap, offsets, &mtr2);
+			mtr_commit(&mtr2);
+		}
+
+		page_cur_move_to_next(&cursor);
+		i++;
+	}
+}
+
+/******************************************************************
+Prints directories and other info of all nodes in the tree. */
+UNIV_INTERN
+void
+btr_print_index(
+/*============*/
+	dict_index_t*	index,	/* in: index */
+	ulint		width)	/* in: print this many entries from start
+				and end */
+{
+	mtr_t		mtr;
+	buf_block_t*	root;
+	mem_heap_t*	heap	= NULL;
+	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint*		offsets	= offsets_;
+	rec_offs_init(offsets_);
+
+	fputs("--------------------------\n"
+	      "INDEX TREE PRINT\n", stderr);
+
+	mtr_start(&mtr);
+
+	root = btr_root_block_get(index, &mtr);
+
+	btr_print_recursive(index, root, width, &heap, &offsets, &mtr);
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+
+	mtr_commit(&mtr);
+
+	btr_validate_index(index, NULL);
+}
+#endif /* UNIV_BTR_PRINT */
+
+#ifdef UNIV_DEBUG
+/****************************************************************
+Checks that the node pointer to a page is appropriate. */
+UNIV_INTERN
+ibool
+btr_check_node_ptr(
+/*===============*/
+				/* out: TRUE */
+	dict_index_t*	index,	/* in: index tree */
+	buf_block_t*	block,	/* in: index page */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	mem_heap_t*	heap;
+	dtuple_t*	tuple;
+	ulint*		offsets;
+	btr_cur_t	cursor;
+	page_t*		page = buf_block_get_frame(block);
+
+	ut_ad(mtr_memo_contains(mtr, block, MTR_MEMO_PAGE_X_FIX));
+	if (dict_index_get_page(index) == buf_block_get_page_no(block)) {
+
+		return(TRUE);
+	}
+
+	heap = mem_heap_create(256);
+	offsets = btr_page_get_father_block(NULL, heap, index, block, mtr,
+					    &cursor);
+
+	if (page_is_leaf(page)) {
+
+		goto func_exit;
+	}
+
+	tuple = dict_index_build_node_ptr(
+		index, page_rec_get_next(page_get_infimum_rec(page)), 0, heap,
+		btr_page_get_level(page, mtr));
+
+	ut_a(!cmp_dtuple_rec(tuple, btr_cur_get_rec(&cursor), offsets));
+func_exit:
+	mem_heap_free(heap);
+
+	return(TRUE);
+}
+#endif /* UNIV_DEBUG */
+
+/****************************************************************
+Display identification information for a record. */
+static
+void
+btr_index_rec_validate_report(
+/*==========================*/
+	const page_t*		page,	/* in: index page */
+	const rec_t*		rec,	/* in: index record */
+	const dict_index_t*	index)	/* in: index */
+{
+	fputs("InnoDB: Record in ", stderr);
+	dict_index_name_print(stderr, NULL, index);
+	fprintf(stderr, ", page %lu, at offset %lu\n",
+		page_get_page_no(page), (ulint) page_offset(rec));
+}
+
+/****************************************************************
+Checks the size and number of fields in a record based on the definition of
+the index. */
+UNIV_INTERN
+ibool
+btr_index_rec_validate(
+/*===================*/
+						/* out: TRUE if ok */
+	const rec_t*		rec,		/* in: index record */
+	const dict_index_t*	index,		/* in: index */
+	ibool			dump_on_error)	/* in: TRUE if the function
+						should print hex dump of record
+						and page on error */
+{
+	ulint		len;
+	ulint		n;
+	ulint		i;
+	const page_t*	page;
+	mem_heap_t*	heap	= NULL;
+	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint*		offsets	= offsets_;
+	rec_offs_init(offsets_);
+
+	page = page_align(rec);
+
+	if (UNIV_UNLIKELY(index->type & DICT_UNIVERSAL)) {
+		/* The insert buffer index tree can contain records from any
+		other index: we cannot check the number of fields or
+		their length */
+
+		return(TRUE);
+	}
+
+	if (UNIV_UNLIKELY((ibool)!!page_is_comp(page)
+			  != dict_table_is_comp(index->table))) {
+		btr_index_rec_validate_report(page, rec, index);
+		fprintf(stderr, "InnoDB: compact flag=%lu, should be %lu\n",
+			(ulong) !!page_is_comp(page),
+			(ulong) dict_table_is_comp(index->table));
+
+		return(FALSE);
+	}
+
+	n = dict_index_get_n_fields(index);
+
+	if (!page_is_comp(page)
+	    && UNIV_UNLIKELY(rec_get_n_fields_old(rec) != n)) {
+		btr_index_rec_validate_report(page, rec, index);
+		fprintf(stderr, "InnoDB: has %lu fields, should have %lu\n",
+			(ulong) rec_get_n_fields_old(rec), (ulong) n);
+
+		if (dump_on_error) {
+			buf_page_print(page, 0);
+
+			fputs("InnoDB: corrupt record ", stderr);
+			rec_print_old(stderr, rec);
+			putc('\n', stderr);
+		}
+		return(FALSE);
+	}
+
+	offsets = rec_get_offsets(rec, index, offsets, ULINT_UNDEFINED, &heap);
+
+	for (i = 0; i < n; i++) {
+		ulint	fixed_size = dict_col_get_fixed_size(
+			dict_index_get_nth_col(index, i));
+
+		rec_get_nth_field_offs(offsets, i, &len);
+
+		/* Note that if fixed_size != 0, it equals the
+		length of a fixed-size column in the clustered index.
+		A prefix index of the column is of fixed, but different
+		length.  When fixed_size == 0, prefix_len is the maximum
+		length of the prefix index column. */
+
+		if ((dict_index_get_nth_field(index, i)->prefix_len == 0
+		     && len != UNIV_SQL_NULL && fixed_size
+		     && len != fixed_size)
+		    || (dict_index_get_nth_field(index, i)->prefix_len > 0
+			&& len != UNIV_SQL_NULL
+			&& len
+			> dict_index_get_nth_field(index, i)->prefix_len)) {
+
+			btr_index_rec_validate_report(page, rec, index);
+			fprintf(stderr,
+				"InnoDB: field %lu len is %lu,"
+				" should be %lu\n",
+				(ulong) i, (ulong) len, (ulong) fixed_size);
+
+			if (dump_on_error) {
+				buf_page_print(page, 0);
+
+				fputs("InnoDB: corrupt record ", stderr);
+				rec_print_new(stderr, rec, offsets);
+				putc('\n', stderr);
+			}
+			if (UNIV_LIKELY_NULL(heap)) {
+				mem_heap_free(heap);
+			}
+			return(FALSE);
+		}
+	}
+
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+	return(TRUE);
+}
+
+/****************************************************************
+Checks the size and number of fields in records based on the definition of
+the index. */
+static
+ibool
+btr_index_page_validate(
+/*====================*/
+				/* out: TRUE if ok */
+	buf_block_t*	block,	/* in: index page */
+	dict_index_t*	index)	/* in: index */
+{
+	page_cur_t	cur;
+	ibool		ret	= TRUE;
+
+	page_cur_set_before_first(block, &cur);
+	page_cur_move_to_next(&cur);
+
+	for (;;) {
+		if (page_cur_is_after_last(&cur)) {
+
+			break;
+		}
+
+		if (!btr_index_rec_validate(cur.rec, index, TRUE)) {
+
+			return(FALSE);
+		}
+
+		page_cur_move_to_next(&cur);
+	}
+
+	return(ret);
+}
+
+/****************************************************************
+Report an error on one page of an index tree. */
+static
+void
+btr_validate_report1(
+/*=================*/
+					/* out: TRUE if ok */
+	dict_index_t*		index,	/* in: index */
+	ulint			level,	/* in: B-tree level */
+	const buf_block_t*	block)	/* in: index page */
+{
+	fprintf(stderr, "InnoDB: Error in page %lu of ",
+		buf_block_get_page_no(block));
+	dict_index_name_print(stderr, NULL, index);
+	if (level) {
+		fprintf(stderr, ", index tree level %lu", level);
+	}
+	putc('\n', stderr);
+}
+
+/****************************************************************
+Report an error on two pages of an index tree. */
+static
+void
+btr_validate_report2(
+/*=================*/
+					/* out: TRUE if ok */
+	const dict_index_t*	index,	/* in: index */
+	ulint			level,	/* in: B-tree level */
+	const buf_block_t*	block1,	/* in: first index page */
+	const buf_block_t*	block2)	/* in: second index page */
+{
+	fprintf(stderr, "InnoDB: Error in pages %lu and %lu of ",
+		buf_block_get_page_no(block1),
+		buf_block_get_page_no(block2));
+	dict_index_name_print(stderr, NULL, index);
+	if (level) {
+		fprintf(stderr, ", index tree level %lu", level);
+	}
+	putc('\n', stderr);
+}
+
+/****************************************************************
+Validates index tree level. */
+static
+ibool
+btr_validate_level(
+/*===============*/
+				/* out: TRUE if ok */
+	dict_index_t*	index,	/* in: index tree */
+	trx_t*		trx,	/* in: transaction or NULL */
+	ulint		level)	/* in: level number */
+{
+	ulint		space;
+	ulint		zip_size;
+	buf_block_t*	block;
+	page_t*		page;
+	buf_block_t*	right_block = 0; /* remove warning */
+	page_t*		right_page = 0; /* remove warning */
+	page_t*		father_page;
+	btr_cur_t	node_cur;
+	btr_cur_t	right_node_cur;
+	rec_t*		rec;
+	ulint		right_page_no;
+	ulint		left_page_no;
+	page_cur_t	cursor;
+	dtuple_t*	node_ptr_tuple;
+	ibool		ret	= TRUE;
+	mtr_t		mtr;
+	mem_heap_t*	heap	= mem_heap_create(256);
+	ulint*		offsets	= NULL;
+	ulint*		offsets2= NULL;
+#ifdef UNIV_ZIP_DEBUG
+	page_zip_des_t*	page_zip;
+#endif /* UNIV_ZIP_DEBUG */
+
+	mtr_start(&mtr);
+
+	mtr_x_lock(dict_index_get_lock(index), &mtr);
+
+	block = btr_root_block_get(index, &mtr);
+	page = buf_block_get_frame(block);
+
+	space = dict_index_get_space(index);
+	zip_size = dict_table_zip_size(index->table);
+
+	while (level != btr_page_get_level(page, &mtr)) {
+		const rec_t*	node_ptr;
+
+		ut_a(space == buf_block_get_space(block));
+		ut_a(space == page_get_space_id(page));
+#ifdef UNIV_ZIP_DEBUG
+		page_zip = buf_block_get_page_zip(block);
+		ut_a(!page_zip || page_zip_validate(page_zip, page));
+#endif /* UNIV_ZIP_DEBUG */
+		ut_a(!page_is_leaf(page));
+
+		page_cur_set_before_first(block, &cursor);
+		page_cur_move_to_next(&cursor);
+
+		node_ptr = page_cur_get_rec(&cursor);
+		offsets = rec_get_offsets(node_ptr, index, offsets,
+					  ULINT_UNDEFINED, &heap);
+		block = btr_node_ptr_get_child(node_ptr, index, offsets, &mtr);
+		page = buf_block_get_frame(block);
+	}
+
+	/* Now we are on the desired level. Loop through the pages on that
+	level. */
+loop:
+	if (trx_is_interrupted(trx)) {
+		mtr_commit(&mtr);
+		mem_heap_free(heap);
+		return(ret);
+	}
+	mem_heap_empty(heap);
+	offsets = offsets2 = NULL;
+	mtr_x_lock(dict_index_get_lock(index), &mtr);
+
+#ifdef UNIV_ZIP_DEBUG
+	page_zip = buf_block_get_page_zip(block);
+	ut_a(!page_zip || page_zip_validate(page_zip, page));
+#endif /* UNIV_ZIP_DEBUG */
+
+	/* Check ordering etc. of records */
+
+	if (!page_validate(page, index)) {
+		btr_validate_report1(index, level, block);
+
+		ret = FALSE;
+	} else if (level == 0) {
+		/* We are on level 0. Check that the records have the right
+		number of fields, and field lengths are right. */
+
+		if (!btr_index_page_validate(block, index)) {
+
+			ret = FALSE;
+		}
+	}
+
+	ut_a(btr_page_get_level(page, &mtr) == level);
+
+	right_page_no = btr_page_get_next(page, &mtr);
+	left_page_no = btr_page_get_prev(page, &mtr);
+
+	ut_a(page_get_n_recs(page) > 0 || (level == 0
+					   && page_get_page_no(page)
+					   == dict_index_get_page(index)));
+
+	if (right_page_no != FIL_NULL) {
+		const rec_t*	right_rec;
+		right_block = btr_block_get(space, zip_size, right_page_no,
+					    RW_X_LATCH, &mtr);
+		right_page = buf_block_get_frame(right_block);
+		if (UNIV_UNLIKELY(btr_page_get_prev(right_page, &mtr)
+				  != page_get_page_no(page))) {
+			btr_validate_report2(index, level, block, right_block);
+			fputs("InnoDB: broken FIL_PAGE_NEXT"
+			      " or FIL_PAGE_PREV links\n", stderr);
+			buf_page_print(page, 0);
+			buf_page_print(right_page, 0);
+
+			ret = FALSE;
+		}
+
+		if (UNIV_UNLIKELY(page_is_comp(right_page)
+				  != page_is_comp(page))) {
+			btr_validate_report2(index, level, block, right_block);
+			fputs("InnoDB: 'compact' flag mismatch\n", stderr);
+			buf_page_print(page, 0);
+			buf_page_print(right_page, 0);
+
+			ret = FALSE;
+
+			goto node_ptr_fails;
+		}
+
+		rec = page_rec_get_prev(page_get_supremum_rec(page));
+		right_rec = page_rec_get_next(page_get_infimum_rec(
+						      right_page));
+		offsets = rec_get_offsets(rec, index,
+					  offsets, ULINT_UNDEFINED, &heap);
+		offsets2 = rec_get_offsets(right_rec, index,
+					   offsets2, ULINT_UNDEFINED, &heap);
+		if (UNIV_UNLIKELY(cmp_rec_rec(rec, right_rec,
+					      offsets, offsets2,
+					      index) >= 0)) {
+
+			btr_validate_report2(index, level, block, right_block);
+
+			fputs("InnoDB: records in wrong order"
+			      " on adjacent pages\n", stderr);
+
+			buf_page_print(page, 0);
+			buf_page_print(right_page, 0);
+
+			fputs("InnoDB: record ", stderr);
+			rec = page_rec_get_prev(page_get_supremum_rec(page));
+			rec_print(stderr, rec, index);
+			putc('\n', stderr);
+			fputs("InnoDB: record ", stderr);
+			rec = page_rec_get_next(
+				page_get_infimum_rec(right_page));
+			rec_print(stderr, rec, index);
+			putc('\n', stderr);
+
+			ret = FALSE;
+		}
+	}
+
+	if (level > 0 && left_page_no == FIL_NULL) {
+		ut_a(REC_INFO_MIN_REC_FLAG & rec_get_info_bits(
+			     page_rec_get_next(page_get_infimum_rec(page)),
+			     page_is_comp(page)));
+	}
+
+	if (buf_block_get_page_no(block) != dict_index_get_page(index)) {
+
+		/* Check father node pointers */
+
+		rec_t*	node_ptr;
+
+		offsets = btr_page_get_father_block(offsets, heap, index,
+						    block, &mtr, &node_cur);
+		father_page = btr_cur_get_page(&node_cur);
+		node_ptr = btr_cur_get_rec(&node_cur);
+
+		btr_cur_position(
+			index, page_rec_get_prev(page_get_supremum_rec(page)),
+			block, &node_cur);
+		offsets = btr_page_get_father_node_ptr(offsets, heap,
+						       &node_cur, &mtr);
+
+		if (UNIV_UNLIKELY(node_ptr != btr_cur_get_rec(&node_cur))
+		    || UNIV_UNLIKELY(btr_node_ptr_get_child_page_no(node_ptr,
+								    offsets)
+				     != buf_block_get_page_no(block))) {
+
+			btr_validate_report1(index, level, block);
+
+			fputs("InnoDB: node pointer to the page is wrong\n",
+			      stderr);
+
+			buf_page_print(father_page, 0);
+			buf_page_print(page, 0);
+
+			fputs("InnoDB: node ptr ", stderr);
+			rec_print(stderr, node_ptr, index);
+
+			rec = btr_cur_get_rec(&node_cur);
+			fprintf(stderr, "\n"
+				"InnoDB: node ptr child page n:o %lu\n",
+				(ulong) btr_node_ptr_get_child_page_no(
+					rec, offsets));
+
+			fputs("InnoDB: record on page ", stderr);
+			rec_print_new(stderr, rec, offsets);
+			putc('\n', stderr);
+			ret = FALSE;
+
+			goto node_ptr_fails;
+		}
+
+		if (!page_is_leaf(page)) {
+			node_ptr_tuple = dict_index_build_node_ptr(
+				index,
+				page_rec_get_next(page_get_infimum_rec(page)),
+				0, heap, btr_page_get_level(page, &mtr));
+
+			if (cmp_dtuple_rec(node_ptr_tuple, node_ptr,
+					   offsets)) {
+				const rec_t* first_rec = page_rec_get_next(
+					page_get_infimum_rec(page));
+
+				btr_validate_report1(index, level, block);
+
+				buf_page_print(father_page, 0);
+				buf_page_print(page, 0);
+
+				fputs("InnoDB: Error: node ptrs differ"
+				      " on levels > 0\n"
+				      "InnoDB: node ptr ", stderr);
+				rec_print_new(stderr, node_ptr, offsets);
+				fputs("InnoDB: first rec ", stderr);
+				rec_print(stderr, first_rec, index);
+				putc('\n', stderr);
+				ret = FALSE;
+
+				goto node_ptr_fails;
+			}
+		}
+
+		if (left_page_no == FIL_NULL) {
+			ut_a(node_ptr == page_rec_get_next(
+				     page_get_infimum_rec(father_page)));
+			ut_a(btr_page_get_prev(father_page, &mtr) == FIL_NULL);
+		}
+
+		if (right_page_no == FIL_NULL) {
+			ut_a(node_ptr == page_rec_get_prev(
+				     page_get_supremum_rec(father_page)));
+			ut_a(btr_page_get_next(father_page, &mtr) == FIL_NULL);
+		} else {
+			const rec_t*	right_node_ptr
+				= page_rec_get_next(node_ptr);
+
+			offsets = btr_page_get_father_block(
+				offsets, heap, index, right_block,
+				&mtr, &right_node_cur);
+			if (right_node_ptr
+			    != page_get_supremum_rec(father_page)) {
+
+				if (btr_cur_get_rec(&right_node_cur)
+				    != right_node_ptr) {
+					ret = FALSE;
+					fputs("InnoDB: node pointer to"
+					      " the right page is wrong\n",
+					      stderr);
+
+					btr_validate_report1(index, level,
+							     block);
+
+					buf_page_print(father_page, 0);
+					buf_page_print(page, 0);
+					buf_page_print(right_page, 0);
+				}
+			} else {
+				page_t*	right_father_page
+					= btr_cur_get_page(&right_node_cur);
+
+				if (btr_cur_get_rec(&right_node_cur)
+				    != page_rec_get_next(
+					    page_get_infimum_rec(
+						    right_father_page))) {
+					ret = FALSE;
+					fputs("InnoDB: node pointer 2 to"
+					      " the right page is wrong\n",
+					      stderr);
+
+					btr_validate_report1(index, level,
+							     block);
+
+					buf_page_print(father_page, 0);
+					buf_page_print(right_father_page, 0);
+					buf_page_print(page, 0);
+					buf_page_print(right_page, 0);
+				}
+
+				if (page_get_page_no(right_father_page)
+				    != btr_page_get_next(father_page, &mtr)) {
+
+					ret = FALSE;
+					fputs("InnoDB: node pointer 3 to"
+					      " the right page is wrong\n",
+					      stderr);
+
+					btr_validate_report1(index, level,
+							     block);
+
+					buf_page_print(father_page, 0);
+					buf_page_print(right_father_page, 0);
+					buf_page_print(page, 0);
+					buf_page_print(right_page, 0);
+				}
+			}
+		}
+	}
+
+node_ptr_fails:
+	/* Commit the mini-transaction to release the latch on 'page'.
+	Re-acquire the latch on right_page, which will become 'page'
+	on the next loop.  The page has already been checked. */
+	mtr_commit(&mtr);
+
+	if (right_page_no != FIL_NULL) {
+		mtr_start(&mtr);
+
+		block = btr_block_get(space, zip_size, right_page_no,
+				      RW_X_LATCH, &mtr);
+		page = buf_block_get_frame(block);
+
+		goto loop;
+	}
+
+	mem_heap_free(heap);
+	return(ret);
+}
+
+/******************************************************************
+Checks the consistency of an index tree. */
+UNIV_INTERN
+ibool
+btr_validate_index(
+/*===============*/
+				/* out: TRUE if ok */
+	dict_index_t*	index,	/* in: index */
+	trx_t*		trx)	/* in: transaction or NULL */
+{
+	mtr_t	mtr;
+	page_t*	root;
+	ulint	i;
+	ulint	n;
+
+	mtr_start(&mtr);
+	mtr_x_lock(dict_index_get_lock(index), &mtr);
+
+	root = btr_root_get(index, &mtr);
+	n = btr_page_get_level(root, &mtr);
+
+	for (i = 0; i <= n && !trx_is_interrupted(trx); i++) {
+		if (!btr_validate_level(index, trx, n - i)) {
+
+			mtr_commit(&mtr);
+
+			return(FALSE);
+		}
+	}
+
+	mtr_commit(&mtr);
+
+	return(TRUE);
+}
diff --git a/storage/xtradb/btr/btr0cur.c b/storage/xtradb/btr/btr0cur.c
new file mode 100644
index 00000000000..70cf06342aa
--- /dev/null
+++ b/storage/xtradb/btr/btr0cur.c
@@ -0,0 +1,4809 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+Copyright (c) 2008, Google Inc.
+
+Portions of this file contain modifications contributed and copyrighted by
+Google, Inc. Those modifications are gratefully acknowledged and are described
+briefly in the InnoDB documentation. The contributions by Google are
+incorporated with their permission, and subject to the conditions contained in
+the file COPYING.Google.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The index tree cursor
+
+All changes that row operations make to a B-tree or the records
+there must go through this module! Undo log records are written here
+of every modify or insert of a clustered index record.
+
+			NOTE!!!
+To make sure we do not run out of disk space during a pessimistic
+insert or update, we have to reserve 2 x the height of the index tree
+many pages in the tablespace before we start the operation, because
+if leaf splitting has been started, it is difficult to undo, except
+by crashing the database and doing a roll-forward.
+
+Created 10/16/1994 Heikki Tuuri
+*******************************************************/
+
+#include "btr0cur.h"
+
+#ifdef UNIV_NONINL
+#include "btr0cur.ic"
+#endif
+
+#include "page0page.h"
+#include "page0zip.h"
+#include "rem0rec.h"
+#include "rem0cmp.h"
+#include "buf0lru.h"
+#include "btr0btr.h"
+#include "btr0sea.h"
+#include "row0upd.h"
+#include "trx0rec.h"
+#include "trx0roll.h" /* trx_is_recv() */
+#include "que0que.h"
+#include "row0row.h"
+#include "srv0srv.h"
+#include "ibuf0ibuf.h"
+#include "lock0lock.h"
+#include "zlib.h"
+
+#ifdef UNIV_DEBUG
+/* If the following is set to TRUE, this module prints a lot of
+trace information of individual record operations */
+UNIV_INTERN ibool	btr_cur_print_record_ops = FALSE;
+#endif /* UNIV_DEBUG */
+
+UNIV_INTERN ulint	btr_cur_n_non_sea	= 0;
+UNIV_INTERN ulint	btr_cur_n_sea		= 0;
+UNIV_INTERN ulint	btr_cur_n_non_sea_old	= 0;
+UNIV_INTERN ulint	btr_cur_n_sea_old	= 0;
+
+/* In the optimistic insert, if the insert does not fit, but this much space
+can be released by page reorganize, then it is reorganized */
+
+#define BTR_CUR_PAGE_REORGANIZE_LIMIT	(UNIV_PAGE_SIZE / 32)
+
+/* The structure of a BLOB part header */
+/*--------------------------------------*/
+#define BTR_BLOB_HDR_PART_LEN		0	/* BLOB part len on this
+						page */
+#define BTR_BLOB_HDR_NEXT_PAGE_NO	4	/* next BLOB part page no,
+						FIL_NULL if none */
+/*--------------------------------------*/
+#define BTR_BLOB_HDR_SIZE		8
+
+/* A BLOB field reference full of zero, for use in assertions and tests.
+Initially, BLOB field references are set to zero, in
+dtuple_convert_big_rec(). */
+UNIV_INTERN const byte field_ref_zero[BTR_EXTERN_FIELD_REF_SIZE];
+
+/***********************************************************************
+Marks all extern fields in a record as owned by the record. This function
+should be called if the delete mark of a record is removed: a not delete
+marked record always owns all its extern fields. */
+static
+void
+btr_cur_unmark_extern_fields(
+/*=========================*/
+	page_zip_des_t*	page_zip,/* in/out: compressed page whose uncompressed
+				part will be updated, or NULL */
+	rec_t*		rec,	/* in/out: record in a clustered index */
+	dict_index_t*	index,	/* in: index of the page */
+	const ulint*	offsets,/* in: array returned by rec_get_offsets() */
+	mtr_t*		mtr);	/* in: mtr, or NULL if not logged */
+/***********************************************************************
+Adds path information to the cursor for the current page, for which
+the binary search has been performed. */
+static
+void
+btr_cur_add_path_info(
+/*==================*/
+	btr_cur_t*	cursor,		/* in: cursor positioned on a page */
+	ulint		height,		/* in: height of the page in tree;
+					0 means leaf node */
+	ulint		root_height);	/* in: root node height in tree */
+/***************************************************************
+Frees the externally stored fields for a record, if the field is mentioned
+in the update vector. */
+static
+void
+btr_rec_free_updated_extern_fields(
+/*===============================*/
+	dict_index_t*	index,	/* in: index of rec; the index tree MUST be
+				X-latched */
+	rec_t*		rec,	/* in: record */
+	page_zip_des_t*	page_zip,/* in: compressed page whose uncompressed
+				part will be updated, or NULL */
+	const ulint*	offsets,/* in: rec_get_offsets(rec, index) */
+	const upd_t*	update,	/* in: update vector */
+	enum trx_rb_ctx	rb_ctx,	/* in: rollback context */
+	mtr_t*		mtr);	/* in: mini-transaction handle which contains
+				an X-latch to record page and to the tree */
+/***************************************************************
+Frees the externally stored fields for a record. */
+static
+void
+btr_rec_free_externally_stored_fields(
+/*==================================*/
+	dict_index_t*	index,	/* in: index of the data, the index
+				tree MUST be X-latched */
+	rec_t*		rec,	/* in: record */
+	const ulint*	offsets,/* in: rec_get_offsets(rec, index) */
+	page_zip_des_t*	page_zip,/* in: compressed page whose uncompressed
+				part will be updated, or NULL */
+	enum trx_rb_ctx	rb_ctx,	/* in: rollback context */
+	mtr_t*		mtr);	/* in: mini-transaction handle which contains
+				an X-latch to record page and to the index
+				tree */
+/***************************************************************
+Gets the externally stored size of a record, in units of a database page. */
+static
+ulint
+btr_rec_get_externally_stored_len(
+/*==============================*/
+				/* out: externally stored part,
+				in units of a database page */
+	rec_t*		rec,	/* in: record */
+	const ulint*	offsets);/* in: array returned by rec_get_offsets() */
+
+/**********************************************************
+The following function is used to set the deleted bit of a record. */
+UNIV_INLINE
+void
+btr_rec_set_deleted_flag(
+/*=====================*/
+				/* out: TRUE on success;
+				FALSE on page_zip overflow */
+	rec_t*		rec,	/* in/out: physical record */
+	page_zip_des_t*	page_zip,/* in/out: compressed page (or NULL) */
+	ulint		flag)	/* in: nonzero if delete marked */
+{
+	if (page_rec_is_comp(rec)) {
+		rec_set_deleted_flag_new(rec, page_zip, flag);
+	} else {
+		ut_ad(!page_zip);
+		rec_set_deleted_flag_old(rec, flag);
+	}
+}
+
+/*==================== B-TREE SEARCH =========================*/
+
+/************************************************************************
+Latches the leaf page or pages requested. */
+static
+void
+btr_cur_latch_leaves(
+/*=================*/
+	page_t*		page,		/* in: leaf page where the search
+					converged */
+	ulint		space,		/* in: space id */
+	ulint		zip_size,	/* in: compressed page size in bytes
+					or 0 for uncompressed pages */
+	ulint		page_no,	/* in: page number of the leaf */
+	ulint		latch_mode,	/* in: BTR_SEARCH_LEAF, ... */
+	btr_cur_t*	cursor,		/* in: cursor */
+	mtr_t*		mtr)		/* in: mtr */
+{
+	ulint		mode;
+	ulint		left_page_no;
+	ulint		right_page_no;
+	buf_block_t*	get_block;
+
+	ut_ad(page && mtr);
+
+	switch (latch_mode) {
+	case BTR_SEARCH_LEAF:
+	case BTR_MODIFY_LEAF:
+		mode = latch_mode == BTR_SEARCH_LEAF ? RW_S_LATCH : RW_X_LATCH;
+		get_block = btr_block_get(space, zip_size, page_no, mode, mtr);
+#ifdef UNIV_BTR_DEBUG
+		ut_a(page_is_comp(get_block->frame) == page_is_comp(page));
+#endif /* UNIV_BTR_DEBUG */
+		get_block->check_index_page_at_flush = TRUE;
+		return;
+	case BTR_MODIFY_TREE:
+		/* x-latch also brothers from left to right */
+		left_page_no = btr_page_get_prev(page, mtr);
+
+		if (left_page_no != FIL_NULL) {
+			get_block = btr_block_get(space, zip_size,
+						  left_page_no,
+						  RW_X_LATCH, mtr);
+#ifdef UNIV_BTR_DEBUG
+			ut_a(page_is_comp(get_block->frame)
+			     == page_is_comp(page));
+			ut_a(btr_page_get_next(get_block->frame, mtr)
+			     == page_get_page_no(page));
+#endif /* UNIV_BTR_DEBUG */
+			get_block->check_index_page_at_flush = TRUE;
+		}
+
+		get_block = btr_block_get(space, zip_size, page_no,
+					  RW_X_LATCH, mtr);
+#ifdef UNIV_BTR_DEBUG
+		ut_a(page_is_comp(get_block->frame) == page_is_comp(page));
+#endif /* UNIV_BTR_DEBUG */
+		get_block->check_index_page_at_flush = TRUE;
+
+		right_page_no = btr_page_get_next(page, mtr);
+
+		if (right_page_no != FIL_NULL) {
+			get_block = btr_block_get(space, zip_size,
+						  right_page_no,
+						  RW_X_LATCH, mtr);
+#ifdef UNIV_BTR_DEBUG
+			ut_a(page_is_comp(get_block->frame)
+			     == page_is_comp(page));
+			ut_a(btr_page_get_prev(get_block->frame, mtr)
+			     == page_get_page_no(page));
+#endif /* UNIV_BTR_DEBUG */
+			get_block->check_index_page_at_flush = TRUE;
+		}
+
+		return;
+
+	case BTR_SEARCH_PREV:
+	case BTR_MODIFY_PREV:
+		mode = latch_mode == BTR_SEARCH_PREV ? RW_S_LATCH : RW_X_LATCH;
+		/* latch also left brother */
+		left_page_no = btr_page_get_prev(page, mtr);
+
+		if (left_page_no != FIL_NULL) {
+			get_block = btr_block_get(space, zip_size,
+						  left_page_no, mode, mtr);
+			cursor->left_block = get_block;
+#ifdef UNIV_BTR_DEBUG
+			ut_a(page_is_comp(get_block->frame)
+			     == page_is_comp(page));
+			ut_a(btr_page_get_next(get_block->frame, mtr)
+			     == page_get_page_no(page));
+#endif /* UNIV_BTR_DEBUG */
+			get_block->check_index_page_at_flush = TRUE;
+		}
+
+		get_block = btr_block_get(space, zip_size, page_no, mode, mtr);
+#ifdef UNIV_BTR_DEBUG
+		ut_a(page_is_comp(get_block->frame) == page_is_comp(page));
+#endif /* UNIV_BTR_DEBUG */
+		get_block->check_index_page_at_flush = TRUE;
+		return;
+	}
+
+	ut_error;
+}
+
+/************************************************************************
+Searches an index tree and positions a tree cursor on a given level.
+NOTE: n_fields_cmp in tuple must be set so that it cannot be compared
+to node pointer page number fields on the upper levels of the tree!
+Note that if mode is PAGE_CUR_LE, which is used in inserts, then
+cursor->up_match and cursor->low_match both will have sensible values.
+If mode is PAGE_CUR_GE, then up_match will a have a sensible value.
+
+If mode is PAGE_CUR_LE , cursor is left at the place where an insert of the
+search tuple should be performed in the B-tree. InnoDB does an insert
+immediately after the cursor. Thus, the cursor may end up on a user record,
+or on a page infimum record. */
+UNIV_INTERN
+void
+btr_cur_search_to_nth_level(
+/*========================*/
+	dict_index_t*	index,	/* in: index */
+	ulint		level,	/* in: the tree level of search */
+	const dtuple_t*	tuple,	/* in: data tuple; NOTE: n_fields_cmp in
+				tuple must be set so that it cannot get
+				compared to the node ptr page number field! */
+	ulint		mode,	/* in: PAGE_CUR_L, ...;
+				Inserts should always be made using
+				PAGE_CUR_LE to search the position! */
+	ulint		latch_mode, /* in: BTR_SEARCH_LEAF, ..., ORed with
+				BTR_INSERT and BTR_ESTIMATE;
+				cursor->left_block is used to store a pointer
+				to the left neighbor page, in the cases
+				BTR_SEARCH_PREV and BTR_MODIFY_PREV;
+				NOTE that if has_search_latch
+				is != 0, we maybe do not have a latch set
+				on the cursor page, we assume
+				the caller uses his search latch
+				to protect the record! */
+	btr_cur_t*	cursor, /* in/out: tree cursor; the cursor page is
+				s- or x-latched, but see also above! */
+	ulint		has_search_latch,/* in: info on the latch mode the
+				caller currently has on btr_search_latch:
+				RW_S_LATCH, or 0 */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	page_cur_t*	page_cursor;
+	page_t*		page;
+	buf_block_t*	guess;
+	rec_t*		node_ptr;
+	ulint		page_no;
+	ulint		space;
+	ulint		up_match;
+	ulint		up_bytes;
+	ulint		low_match;
+	ulint		low_bytes;
+	ulint		height;
+	ulint		savepoint;
+	ulint		page_mode;
+	ulint		insert_planned;
+	ulint		estimate;
+	ulint		ignore_sec_unique;
+	ulint		root_height = 0; /* remove warning */
+#ifdef BTR_CUR_ADAPT
+	btr_search_t*	info;
+#endif
+	mem_heap_t*	heap		= NULL;
+	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint*		offsets		= offsets_;
+	rec_offs_init(offsets_);
+	/* Currently, PAGE_CUR_LE is the only search mode used for searches
+	ending to upper levels */
+
+	ut_ad(level == 0 || mode == PAGE_CUR_LE);
+	ut_ad(dict_index_check_search_tuple(index, tuple));
+	ut_ad(!dict_index_is_ibuf(index) || ibuf_inside());
+	ut_ad(dtuple_check_typed(tuple));
+
+#ifdef UNIV_DEBUG
+	cursor->up_match = ULINT_UNDEFINED;
+	cursor->low_match = ULINT_UNDEFINED;
+#endif
+	insert_planned = latch_mode & BTR_INSERT;
+	estimate = latch_mode & BTR_ESTIMATE;
+	ignore_sec_unique = latch_mode & BTR_IGNORE_SEC_UNIQUE;
+	latch_mode = latch_mode & ~(BTR_INSERT | BTR_ESTIMATE
+				    | BTR_IGNORE_SEC_UNIQUE);
+
+	ut_ad(!insert_planned || (mode == PAGE_CUR_LE));
+
+	cursor->flag = BTR_CUR_BINARY;
+	cursor->index = index;
+
+#ifndef BTR_CUR_ADAPT
+	guess = NULL;
+#else
+	info = btr_search_get_info(index);
+
+	guess = info->root_guess;
+
+#ifdef BTR_CUR_HASH_ADAPT
+
+#ifdef UNIV_SEARCH_PERF_STAT
+	info->n_searches++;
+#endif
+	if (rw_lock_get_writer(&btr_search_latch) == RW_LOCK_NOT_LOCKED
+	    && latch_mode <= BTR_MODIFY_LEAF && info->last_hash_succ
+	    && !estimate
+#ifdef PAGE_CUR_LE_OR_EXTENDS
+	    && mode != PAGE_CUR_LE_OR_EXTENDS
+#endif /* PAGE_CUR_LE_OR_EXTENDS */
+	    /* If !has_search_latch, we do a dirty read of
+	    btr_search_enabled below, and btr_search_guess_on_hash()
+	    will have to check it again. */
+	    && UNIV_LIKELY(btr_search_enabled)
+	    && btr_search_guess_on_hash(index, info, tuple, mode,
+					latch_mode, cursor,
+					has_search_latch, mtr)) {
+
+		/* Search using the hash index succeeded */
+
+		ut_ad(cursor->up_match != ULINT_UNDEFINED
+		      || mode != PAGE_CUR_GE);
+		ut_ad(cursor->up_match != ULINT_UNDEFINED
+		      || mode != PAGE_CUR_LE);
+		ut_ad(cursor->low_match != ULINT_UNDEFINED
+		      || mode != PAGE_CUR_LE);
+		btr_cur_n_sea++;
+
+		return;
+	}
+#endif /* BTR_CUR_HASH_ADAPT */
+#endif /* BTR_CUR_ADAPT */
+	btr_cur_n_non_sea++;
+
+	/* If the hash search did not succeed, do binary search down the
+	tree */
+
+	if (has_search_latch) {
+		/* Release possible search latch to obey latching order */
+		rw_lock_s_unlock(&btr_search_latch);
+	}
+
+	/* Store the position of the tree latch we push to mtr so that we
+	know how to release it when we have latched leaf node(s) */
+
+	savepoint = mtr_set_savepoint(mtr);
+
+	if (latch_mode == BTR_MODIFY_TREE) {
+		mtr_x_lock(dict_index_get_lock(index), mtr);
+
+	} else if (latch_mode == BTR_CONT_MODIFY_TREE) {
+		/* Do nothing */
+		ut_ad(mtr_memo_contains(mtr, dict_index_get_lock(index),
+					MTR_MEMO_X_LOCK));
+	} else {
+		mtr_s_lock(dict_index_get_lock(index), mtr);
+	}
+
+	page_cursor = btr_cur_get_page_cur(cursor);
+
+	space = dict_index_get_space(index);
+	page_no = dict_index_get_page(index);
+
+	up_match = 0;
+	up_bytes = 0;
+	low_match = 0;
+	low_bytes = 0;
+
+	height = ULINT_UNDEFINED;
+
+	/* We use these modified search modes on non-leaf levels of the
+	B-tree. These let us end up in the right B-tree leaf. In that leaf
+	we use the original search mode. */
+
+	switch (mode) {
+	case PAGE_CUR_GE:
+		page_mode = PAGE_CUR_L;
+		break;
+	case PAGE_CUR_G:
+		page_mode = PAGE_CUR_LE;
+		break;
+	default:
+#ifdef PAGE_CUR_LE_OR_EXTENDS
+		ut_ad(mode == PAGE_CUR_L || mode == PAGE_CUR_LE
+		      || mode == PAGE_CUR_LE_OR_EXTENDS);
+#else /* PAGE_CUR_LE_OR_EXTENDS */
+		ut_ad(mode == PAGE_CUR_L || mode == PAGE_CUR_LE);
+#endif /* PAGE_CUR_LE_OR_EXTENDS */
+		page_mode = mode;
+		break;
+	}
+
+	/* Loop and search until we arrive at the desired level */
+
+	for (;;) {
+		ulint		zip_size;
+		buf_block_t*	block;
+		ulint		rw_latch;
+		ulint		buf_mode;
+
+		zip_size = dict_table_zip_size(index->table);
+		rw_latch = RW_NO_LATCH;
+		buf_mode = BUF_GET;
+
+		if (height == 0 && latch_mode <= BTR_MODIFY_LEAF) {
+
+			rw_latch = latch_mode;
+
+			if (insert_planned
+			    && ibuf_should_try(index, ignore_sec_unique)) {
+
+				/* Try insert to the insert buffer if the
+				page is not in the buffer pool */
+
+				buf_mode = BUF_GET_IF_IN_POOL;
+			}
+		}
+
+retry_page_get:
+		block = buf_page_get_gen(space, zip_size, page_no,
+					 rw_latch, guess, buf_mode,
+					 __FILE__, __LINE__, mtr);
+		if (block == NULL) {
+			/* This must be a search to perform an insert;
+			try insert to the insert buffer */
+
+			ut_ad(buf_mode == BUF_GET_IF_IN_POOL);
+			ut_ad(insert_planned);
+			ut_ad(cursor->thr);
+
+			if (ibuf_insert(tuple, index, space, zip_size,
+					page_no, cursor->thr)) {
+				/* Insertion to the insert buffer succeeded */
+				cursor->flag = BTR_CUR_INSERT_TO_IBUF;
+				if (UNIV_LIKELY_NULL(heap)) {
+					mem_heap_free(heap);
+				}
+				goto func_exit;
+			}
+
+			/* Insert to the insert buffer did not succeed:
+			retry page get */
+
+			buf_mode = BUF_GET;
+
+			goto retry_page_get;
+		}
+
+		page = buf_block_get_frame(block);
+
+		block->check_index_page_at_flush = TRUE;
+
+		if (rw_latch != RW_NO_LATCH) {
+#ifdef UNIV_ZIP_DEBUG
+			const page_zip_des_t*	page_zip
+				= buf_block_get_page_zip(block);
+			ut_a(!page_zip || page_zip_validate(page_zip, page));
+#endif /* UNIV_ZIP_DEBUG */
+
+			buf_block_dbg_add_level(block, SYNC_TREE_NODE);
+		}
+
+		ut_ad(0 == ut_dulint_cmp(index->id,
+					 btr_page_get_index_id(page)));
+
+		if (UNIV_UNLIKELY(height == ULINT_UNDEFINED)) {
+			/* We are in the root node */
+
+			height = btr_page_get_level(page, mtr);
+			root_height = height;
+			cursor->tree_height = root_height + 1;
+#ifdef BTR_CUR_ADAPT
+			if (block != guess) {
+				info->root_guess = block;
+			}
+#endif
+		}
+
+		if (height == 0) {
+			if (rw_latch == RW_NO_LATCH) {
+
+				btr_cur_latch_leaves(page, space, zip_size,
+						     page_no, latch_mode,
+						     cursor, mtr);
+			}
+
+			if ((latch_mode != BTR_MODIFY_TREE)
+			    && (latch_mode != BTR_CONT_MODIFY_TREE)) {
+
+				/* Release the tree s-latch */
+
+				mtr_release_s_latch_at_savepoint(
+					mtr, savepoint,
+					dict_index_get_lock(index));
+			}
+
+			page_mode = mode;
+		}
+
+		page_cur_search_with_match(block, index, tuple, page_mode,
+					   &up_match, &up_bytes,
+					   &low_match, &low_bytes,
+					   page_cursor);
+
+		if (estimate) {
+			btr_cur_add_path_info(cursor, height, root_height);
+		}
+
+		/* If this is the desired level, leave the loop */
+
+		ut_ad(height == btr_page_get_level(
+			      page_cur_get_page(page_cursor), mtr));
+
+		if (level == height) {
+
+			if (level > 0) {
+				/* x-latch the page */
+				page = btr_page_get(space, zip_size,
+						    page_no, RW_X_LATCH, mtr);
+				ut_a((ibool)!!page_is_comp(page)
+				     == dict_table_is_comp(index->table));
+			}
+
+			break;
+		}
+
+		ut_ad(height > 0);
+
+		height--;
+
+		guess = NULL;
+
+		node_ptr = page_cur_get_rec(page_cursor);
+		offsets = rec_get_offsets(node_ptr, cursor->index, offsets,
+					  ULINT_UNDEFINED, &heap);
+		/* Go to the child node */
+		page_no = btr_node_ptr_get_child_page_no(node_ptr, offsets);
+	}
+
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+
+	if (level == 0) {
+		cursor->low_match = low_match;
+		cursor->low_bytes = low_bytes;
+		cursor->up_match = up_match;
+		cursor->up_bytes = up_bytes;
+
+#ifdef BTR_CUR_ADAPT
+		/* We do a dirty read of btr_search_enabled here.  We
+		will properly check btr_search_enabled again in
+		btr_search_build_page_hash_index() before building a
+		page hash index, while holding btr_search_latch. */
+		if (UNIV_LIKELY(btr_search_enabled)) {
+
+			btr_search_info_update(index, cursor);
+		}
+#endif
+		ut_ad(cursor->up_match != ULINT_UNDEFINED
+		      || mode != PAGE_CUR_GE);
+		ut_ad(cursor->up_match != ULINT_UNDEFINED
+		      || mode != PAGE_CUR_LE);
+		ut_ad(cursor->low_match != ULINT_UNDEFINED
+		      || mode != PAGE_CUR_LE);
+	}
+
+func_exit:
+	if (has_search_latch) {
+
+		rw_lock_s_lock(&btr_search_latch);
+	}
+}
+
+/*********************************************************************
+Opens a cursor at either end of an index. */
+UNIV_INTERN
+void
+btr_cur_open_at_index_side(
+/*=======================*/
+	ibool		from_left,	/* in: TRUE if open to the low end,
+					FALSE if to the high end */
+	dict_index_t*	index,		/* in: index */
+	ulint		latch_mode,	/* in: latch mode */
+	btr_cur_t*	cursor,		/* in: cursor */
+	mtr_t*		mtr)		/* in: mtr */
+{
+	page_cur_t*	page_cursor;
+	ulint		page_no;
+	ulint		space;
+	ulint		zip_size;
+	ulint		height;
+	ulint		root_height = 0; /* remove warning */
+	rec_t*		node_ptr;
+	ulint		estimate;
+	ulint		savepoint;
+	mem_heap_t*	heap		= NULL;
+	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint*		offsets		= offsets_;
+	rec_offs_init(offsets_);
+
+	estimate = latch_mode & BTR_ESTIMATE;
+	latch_mode = latch_mode & ~BTR_ESTIMATE;
+
+	/* Store the position of the tree latch we push to mtr so that we
+	know how to release it when we have latched the leaf node */
+
+	savepoint = mtr_set_savepoint(mtr);
+
+	if (latch_mode == BTR_MODIFY_TREE) {
+		mtr_x_lock(dict_index_get_lock(index), mtr);
+	} else {
+		mtr_s_lock(dict_index_get_lock(index), mtr);
+	}
+
+	page_cursor = btr_cur_get_page_cur(cursor);
+	cursor->index = index;
+
+	space = dict_index_get_space(index);
+	zip_size = dict_table_zip_size(index->table);
+	page_no = dict_index_get_page(index);
+
+	height = ULINT_UNDEFINED;
+
+	for (;;) {
+		buf_block_t*	block;
+		page_t*		page;
+		block = buf_page_get_gen(space, zip_size, page_no,
+					 RW_NO_LATCH, NULL, BUF_GET,
+					 __FILE__, __LINE__, mtr);
+		page = buf_block_get_frame(block);
+		ut_ad(0 == ut_dulint_cmp(index->id,
+					 btr_page_get_index_id(page)));
+
+		block->check_index_page_at_flush = TRUE;
+
+		if (height == ULINT_UNDEFINED) {
+			/* We are in the root node */
+
+			height = btr_page_get_level(page, mtr);
+			root_height = height;
+		}
+
+		if (height == 0) {
+			btr_cur_latch_leaves(page, space, zip_size, page_no,
+					     latch_mode, cursor, mtr);
+
+			/* In versions <= 3.23.52 we had forgotten to
+			release the tree latch here. If in an index scan
+			we had to scan far to find a record visible to the
+			current transaction, that could starve others
+			waiting for the tree latch. */
+
+			if ((latch_mode != BTR_MODIFY_TREE)
+			    && (latch_mode != BTR_CONT_MODIFY_TREE)) {
+
+				/* Release the tree s-latch */
+
+				mtr_release_s_latch_at_savepoint(
+					mtr, savepoint,
+					dict_index_get_lock(index));
+			}
+		}
+
+		if (from_left) {
+			page_cur_set_before_first(block, page_cursor);
+		} else {
+			page_cur_set_after_last(block, page_cursor);
+		}
+
+		if (height == 0) {
+			if (estimate) {
+				btr_cur_add_path_info(cursor, height,
+						      root_height);
+			}
+
+			break;
+		}
+
+		ut_ad(height > 0);
+
+		if (from_left) {
+			page_cur_move_to_next(page_cursor);
+		} else {
+			page_cur_move_to_prev(page_cursor);
+		}
+
+		if (estimate) {
+			btr_cur_add_path_info(cursor, height, root_height);
+		}
+
+		height--;
+
+		node_ptr = page_cur_get_rec(page_cursor);
+		offsets = rec_get_offsets(node_ptr, cursor->index, offsets,
+					  ULINT_UNDEFINED, &heap);
+		/* Go to the child node */
+		page_no = btr_node_ptr_get_child_page_no(node_ptr, offsets);
+	}
+
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+}
+
+/**************************************************************************
+Positions a cursor at a randomly chosen position within a B-tree. */
+UNIV_INTERN
+void
+btr_cur_open_at_rnd_pos(
+/*====================*/
+	dict_index_t*	index,		/* in: index */
+	ulint		latch_mode,	/* in: BTR_SEARCH_LEAF, ... */
+	btr_cur_t*	cursor,		/* in/out: B-tree cursor */
+	mtr_t*		mtr)		/* in: mtr */
+{
+	page_cur_t*	page_cursor;
+	ulint		page_no;
+	ulint		space;
+	ulint		zip_size;
+	ulint		height;
+	rec_t*		node_ptr;
+	mem_heap_t*	heap		= NULL;
+	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint*		offsets		= offsets_;
+	rec_offs_init(offsets_);
+
+	if (latch_mode == BTR_MODIFY_TREE) {
+		mtr_x_lock(dict_index_get_lock(index), mtr);
+	} else {
+		mtr_s_lock(dict_index_get_lock(index), mtr);
+	}
+
+	page_cursor = btr_cur_get_page_cur(cursor);
+	cursor->index = index;
+
+	space = dict_index_get_space(index);
+	zip_size = dict_table_zip_size(index->table);
+	page_no = dict_index_get_page(index);
+
+	height = ULINT_UNDEFINED;
+
+	for (;;) {
+		buf_block_t*	block;
+		page_t*		page;
+
+		block = buf_page_get_gen(space, zip_size, page_no,
+					 RW_NO_LATCH, NULL, BUF_GET,
+					 __FILE__, __LINE__, mtr);
+		page = buf_block_get_frame(block);
+		ut_ad(0 == ut_dulint_cmp(index->id,
+					 btr_page_get_index_id(page)));
+
+		if (height == ULINT_UNDEFINED) {
+			/* We are in the root node */
+
+			height = btr_page_get_level(page, mtr);
+		}
+
+		if (height == 0) {
+			btr_cur_latch_leaves(page, space, zip_size, page_no,
+					     latch_mode, cursor, mtr);
+		}
+
+		page_cur_open_on_rnd_user_rec(block, page_cursor);
+
+		if (height == 0) {
+
+			break;
+		}
+
+		ut_ad(height > 0);
+
+		height--;
+
+		node_ptr = page_cur_get_rec(page_cursor);
+		offsets = rec_get_offsets(node_ptr, cursor->index, offsets,
+					  ULINT_UNDEFINED, &heap);
+		/* Go to the child node */
+		page_no = btr_node_ptr_get_child_page_no(node_ptr, offsets);
+	}
+
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+}
+
+/*==================== B-TREE INSERT =========================*/
+
+/*****************************************************************
+Inserts a record if there is enough space, or if enough space can
+be freed by reorganizing. Differs from btr_cur_optimistic_insert because
+no heuristics is applied to whether it pays to use CPU time for
+reorganizing the page or not. */
+static
+rec_t*
+btr_cur_insert_if_possible(
+/*=======================*/
+				/* out: pointer to inserted record if succeed,
+				else NULL */
+	btr_cur_t*	cursor,	/* in: cursor on page after which to insert;
+				cursor stays valid */
+	const dtuple_t*	tuple,	/* in: tuple to insert; the size info need not
+				have been stored to tuple */
+	ulint		n_ext,	/* in: number of externally stored columns */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	page_cur_t*	page_cursor;
+	buf_block_t*	block;
+	rec_t*		rec;
+
+	ut_ad(dtuple_check_typed(tuple));
+
+	block = btr_cur_get_block(cursor);
+
+	ut_ad(mtr_memo_contains(mtr, block, MTR_MEMO_PAGE_X_FIX));
+	page_cursor = btr_cur_get_page_cur(cursor);
+
+	/* Now, try the insert */
+	rec = page_cur_tuple_insert(page_cursor, tuple,
+				    cursor->index, n_ext, mtr);
+
+	if (UNIV_UNLIKELY(!rec)) {
+		/* If record did not fit, reorganize */
+
+		if (btr_page_reorganize(block, cursor->index, mtr)) {
+
+			page_cur_search(block, cursor->index, tuple,
+					PAGE_CUR_LE, page_cursor);
+
+			rec = page_cur_tuple_insert(page_cursor, tuple,
+						    cursor->index, n_ext, mtr);
+		}
+	}
+
+	return(rec);
+}
+
+/*****************************************************************
+For an insert, checks the locks and does the undo logging if desired. */
+UNIV_INLINE
+ulint
+btr_cur_ins_lock_and_undo(
+/*======================*/
+				/* out: DB_SUCCESS, DB_WAIT_LOCK,
+				DB_FAIL, or error number */
+	ulint		flags,	/* in: undo logging and locking flags: if
+				not zero, the parameters index and thr
+				should be specified */
+	btr_cur_t*	cursor,	/* in: cursor on page after which to insert */
+	const dtuple_t*	entry,	/* in: entry to insert */
+	que_thr_t*	thr,	/* in: query thread or NULL */
+	ibool*		inherit)/* out: TRUE if the inserted new record maybe
+				should inherit LOCK_GAP type locks from the
+				successor record */
+{
+	dict_index_t*	index;
+	ulint		err;
+	rec_t*		rec;
+	dulint		roll_ptr;
+
+	/* Check if we have to wait for a lock: enqueue an explicit lock
+	request if yes */
+
+	rec = btr_cur_get_rec(cursor);
+	index = cursor->index;
+
+	err = lock_rec_insert_check_and_lock(flags, rec,
+					     btr_cur_get_block(cursor),
+					     index, thr, inherit);
+
+	if (err != DB_SUCCESS) {
+
+		return(err);
+	}
+
+	if (dict_index_is_clust(index) && !dict_index_is_ibuf(index)) {
+
+		err = trx_undo_report_row_operation(flags, TRX_UNDO_INSERT_OP,
+						    thr, index, entry,
+						    NULL, 0, NULL,
+						    &roll_ptr);
+		if (err != DB_SUCCESS) {
+
+			return(err);
+		}
+
+		/* Now we can fill in the roll ptr field in entry */
+
+		if (!(flags & BTR_KEEP_SYS_FLAG)) {
+
+			row_upd_index_entry_sys_field(entry, index,
+						      DATA_ROLL_PTR, roll_ptr);
+		}
+	}
+
+	return(DB_SUCCESS);
+}
+
+#ifdef UNIV_DEBUG
+/*****************************************************************
+Report information about a transaction. */
+static
+void
+btr_cur_trx_report(
+/*===============*/
+	trx_t*			trx,	/* in: transaction */
+	const dict_index_t*	index,	/* in: index */
+	const char*		op)	/* in: operation */
+{
+	fprintf(stderr, "Trx with id " TRX_ID_FMT " going to ",
+		TRX_ID_PREP_PRINTF(trx->id));
+	fputs(op, stderr);
+	dict_index_name_print(stderr, trx, index);
+	putc('\n', stderr);
+}
+#endif /* UNIV_DEBUG */
+
+/*****************************************************************
+Tries to perform an insert to a page in an index tree, next to cursor.
+It is assumed that mtr holds an x-latch on the page. The operation does
+not succeed if there is too little space on the page. If there is just
+one record on the page, the insert will always succeed; this is to
+prevent trying to split a page with just one record. */
+UNIV_INTERN
+ulint
+btr_cur_optimistic_insert(
+/*======================*/
+				/* out: DB_SUCCESS, DB_WAIT_LOCK,
+				DB_FAIL, or error number */
+	ulint		flags,	/* in: undo logging and locking flags: if not
+				zero, the parameters index and thr should be
+				specified */
+	btr_cur_t*	cursor,	/* in: cursor on page after which to insert;
+				cursor stays valid */
+	dtuple_t*	entry,	/* in/out: entry to insert */
+	rec_t**		rec,	/* out: pointer to inserted record if
+				succeed */
+	big_rec_t**	big_rec,/* out: big rec vector whose fields have to
+				be stored externally by the caller, or
+				NULL */
+	ulint		n_ext,	/* in: number of externally stored columns */
+	que_thr_t*	thr,	/* in: query thread or NULL */
+	mtr_t*		mtr)	/* in: mtr; if this function returns
+				DB_SUCCESS on a leaf page of a secondary
+				index in a compressed tablespace, the
+				mtr must be committed before latching
+				any further pages */
+{
+	big_rec_t*	big_rec_vec	= NULL;
+	dict_index_t*	index;
+	page_cur_t*	page_cursor;
+	buf_block_t*	block;
+	page_t*		page;
+	ulint		max_size;
+	rec_t*		dummy_rec;
+	ibool		leaf;
+	ibool		reorg;
+	ibool		inherit;
+	ulint		zip_size;
+	ulint		rec_size;
+	mem_heap_t*	heap		= NULL;
+	ulint		err;
+
+	*big_rec = NULL;
+
+	block = btr_cur_get_block(cursor);
+	page = buf_block_get_frame(block);
+	index = cursor->index;
+	zip_size = buf_block_get_zip_size(block);
+#ifdef UNIV_DEBUG_VALGRIND
+	if (zip_size) {
+		UNIV_MEM_ASSERT_RW(page, UNIV_PAGE_SIZE);
+		UNIV_MEM_ASSERT_RW(block->page.zip.data, zip_size);
+	}
+#endif /* UNIV_DEBUG_VALGRIND */
+
+	if (!dtuple_check_typed_no_assert(entry)) {
+		fputs("InnoDB: Error in a tuple to insert into ", stderr);
+		dict_index_name_print(stderr, thr_get_trx(thr), index);
+	}
+#ifdef UNIV_DEBUG
+	if (btr_cur_print_record_ops && thr) {
+		btr_cur_trx_report(thr_get_trx(thr), index, "insert into ");
+		dtuple_print(stderr, entry);
+	}
+#endif /* UNIV_DEBUG */
+
+	ut_ad(mtr_memo_contains(mtr, block, MTR_MEMO_PAGE_X_FIX));
+	max_size = page_get_max_insert_size_after_reorganize(page, 1);
+	leaf = page_is_leaf(page);
+
+	/* Calculate the record size when entry is converted to a record */
+	rec_size = rec_get_converted_size(index, entry, n_ext);
+
+	if (page_zip_rec_needs_ext(rec_size, page_is_comp(page),
+				   dtuple_get_n_fields(entry), zip_size)) {
+
+		/* The record is so big that we have to store some fields
+		externally on separate database pages */
+		big_rec_vec = dtuple_convert_big_rec(index, entry, &n_ext);
+
+		if (UNIV_UNLIKELY(big_rec_vec == NULL)) {
+
+			return(DB_TOO_BIG_RECORD);
+		}
+
+		rec_size = rec_get_converted_size(index, entry, n_ext);
+	}
+
+	if (UNIV_UNLIKELY(zip_size)) {
+		/* Estimate the free space of an empty compressed page.
+		Subtract one byte for the encoded heap_no in the
+		modification log. */
+		ulint	free_space_zip = page_zip_empty_size(
+			cursor->index->n_fields, zip_size) - 1;
+		ulint	n_uniq = dict_index_get_n_unique_in_tree(index);
+
+		ut_ad(dict_table_is_comp(index->table));
+
+		/* There should be enough room for two node pointer
+		records on an empty non-leaf page.  This prevents
+		infinite page splits. */
+
+		if (UNIV_LIKELY(entry->n_fields >= n_uniq)
+		    && UNIV_UNLIKELY(REC_NODE_PTR_SIZE
+				     + rec_get_converted_size_comp_prefix(
+					     index, entry->fields, n_uniq,
+					     NULL)
+				     /* On a compressed page, there is
+				     a two-byte entry in the dense
+				     page directory for every record.
+				     But there is no record header. */
+				     - (REC_N_NEW_EXTRA_BYTES - 2)
+				     > free_space_zip / 2)) {
+
+			if (big_rec_vec) {
+				dtuple_convert_back_big_rec(
+					index, entry, big_rec_vec);
+			}
+
+			if (heap) {
+				mem_heap_free(heap);
+			}
+
+			return(DB_TOO_BIG_RECORD);
+		}
+	}
+
+	/* If there have been many consecutive inserts, and we are on the leaf
+	level, check if we have to split the page to reserve enough free space
+	for future updates of records. */
+
+	if (dict_index_is_clust(index)
+	    && (page_get_n_recs(page) >= 2)
+	    && UNIV_LIKELY(leaf)
+	    && (dict_index_get_space_reserve() + rec_size > max_size)
+	    && (btr_page_get_split_rec_to_right(cursor, &dummy_rec)
+		|| btr_page_get_split_rec_to_left(cursor, &dummy_rec))) {
+fail:
+		err = DB_FAIL;
+fail_err:
+
+		if (big_rec_vec) {
+			dtuple_convert_back_big_rec(index, entry, big_rec_vec);
+		}
+
+		if (UNIV_LIKELY_NULL(heap)) {
+			mem_heap_free(heap);
+		}
+
+		return(err);
+	}
+
+	if (UNIV_UNLIKELY(max_size < BTR_CUR_PAGE_REORGANIZE_LIMIT
+	     || max_size < rec_size)
+	    && UNIV_LIKELY(page_get_n_recs(page) > 1)
+	    && page_get_max_insert_size(page, 1) < rec_size) {
+
+		goto fail;
+	}
+
+	/* Check locks and write to the undo log, if specified */
+	err = btr_cur_ins_lock_and_undo(flags, cursor, entry, thr, &inherit);
+
+	if (UNIV_UNLIKELY(err != DB_SUCCESS)) {
+
+		goto fail_err;
+	}
+
+	page_cursor = btr_cur_get_page_cur(cursor);
+
+	/* Now, try the insert */
+
+	{
+		const rec_t* page_cursor_rec = page_cur_get_rec(page_cursor);
+		*rec = page_cur_tuple_insert(page_cursor, entry, index,
+					     n_ext, mtr);
+		reorg = page_cursor_rec != page_cur_get_rec(page_cursor);
+
+		if (UNIV_UNLIKELY(reorg)) {
+			ut_a(zip_size);
+			ut_a(*rec);
+		}
+	}
+
+	if (UNIV_UNLIKELY(!*rec) && UNIV_LIKELY(!reorg)) {
+		/* If the record did not fit, reorganize */
+		if (UNIV_UNLIKELY(!btr_page_reorganize(block, index, mtr))) {
+			ut_a(zip_size);
+
+			goto fail;
+		}
+
+		ut_ad(zip_size
+		      || page_get_max_insert_size(page, 1) == max_size);
+
+		reorg = TRUE;
+
+		page_cur_search(block, index, entry, PAGE_CUR_LE, page_cursor);
+
+		*rec = page_cur_tuple_insert(page_cursor, entry, index,
+					     n_ext, mtr);
+
+		if (UNIV_UNLIKELY(!*rec)) {
+			if (UNIV_LIKELY(zip_size != 0)) {
+
+				goto fail;
+			}
+
+			fputs("InnoDB: Error: cannot insert tuple ", stderr);
+			dtuple_print(stderr, entry);
+			fputs(" into ", stderr);
+			dict_index_name_print(stderr, thr_get_trx(thr), index);
+			fprintf(stderr, "\nInnoDB: max insert size %lu\n",
+				(ulong) max_size);
+			ut_error;
+		}
+	}
+
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+
+#ifdef BTR_CUR_HASH_ADAPT
+	if (!reorg && leaf && (cursor->flag == BTR_CUR_HASH)) {
+		btr_search_update_hash_node_on_insert(cursor);
+	} else {
+		btr_search_update_hash_on_insert(cursor);
+	}
+#endif
+
+	if (!(flags & BTR_NO_LOCKING_FLAG) && inherit) {
+
+		lock_update_insert(block, *rec);
+	}
+
+#if 0
+	fprintf(stderr, "Insert into page %lu, max ins size %lu,"
+		" rec %lu ind type %lu\n",
+		buf_block_get_page_no(block), max_size,
+		rec_size + PAGE_DIR_SLOT_SIZE, index->type);
+#endif
+	if (leaf
+	    && !dict_index_is_clust(index)
+	    && !dict_index_is_ibuf(index)) {
+		/* Update the free bits of the B-tree page in the
+		insert buffer bitmap. */
+
+		/* The free bits in the insert buffer bitmap must
+		never exceed the free space on a page.  It is safe to
+		decrement or reset the bits in the bitmap in a
+		mini-transaction that is committed before the
+		mini-transaction that affects the free space. */
+
+		/* It is unsafe to increment the bits in a separately
+		committed mini-transaction, because in crash recovery,
+		the free bits could momentarily be set too high. */
+
+		if (zip_size) {
+			/* Update the bits in the same mini-transaction. */
+			ibuf_update_free_bits_zip(block, mtr);
+		} else {
+			/* Decrement the bits in a separate
+			mini-transaction. */
+			ibuf_update_free_bits_if_full(
+				block, max_size,
+				rec_size + PAGE_DIR_SLOT_SIZE);
+		}
+	}
+
+	*big_rec = big_rec_vec;
+
+	return(DB_SUCCESS);
+}
+
+/*****************************************************************
+Performs an insert on a page of an index tree. It is assumed that mtr
+holds an x-latch on the tree and on the cursor page. If the insert is
+made on the leaf level, to avoid deadlocks, mtr must also own x-latches
+to brothers of page, if those brothers exist. */
+UNIV_INTERN
+ulint
+btr_cur_pessimistic_insert(
+/*=======================*/
+				/* out: DB_SUCCESS or error number */
+	ulint		flags,	/* in: undo logging and locking flags: if not
+				zero, the parameter thr should be
+				specified; if no undo logging is specified,
+				then the caller must have reserved enough
+				free extents in the file space so that the
+				insertion will certainly succeed */
+	btr_cur_t*	cursor,	/* in: cursor after which to insert;
+				cursor stays valid */
+	dtuple_t*	entry,	/* in/out: entry to insert */
+	rec_t**		rec,	/* out: pointer to inserted record if
+				succeed */
+	big_rec_t**	big_rec,/* out: big rec vector whose fields have to
+				be stored externally by the caller, or
+				NULL */
+	ulint		n_ext,	/* in: number of externally stored columns */
+	que_thr_t*	thr,	/* in: query thread or NULL */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	dict_index_t*	index		= cursor->index;
+	ulint		zip_size	= dict_table_zip_size(index->table);
+	big_rec_t*	big_rec_vec	= NULL;
+	mem_heap_t*	heap		= NULL;
+	ulint		err;
+	ibool		dummy_inh;
+	ibool		success;
+	ulint		n_extents	= 0;
+	ulint		n_reserved;
+
+	ut_ad(dtuple_check_typed(entry));
+
+	*big_rec = NULL;
+
+	ut_ad(mtr_memo_contains(mtr,
+				dict_index_get_lock(btr_cur_get_index(cursor)),
+				MTR_MEMO_X_LOCK));
+	ut_ad(mtr_memo_contains(mtr, btr_cur_get_block(cursor),
+				MTR_MEMO_PAGE_X_FIX));
+
+	/* Try first an optimistic insert; reset the cursor flag: we do not
+	assume anything of how it was positioned */
+
+	cursor->flag = BTR_CUR_BINARY;
+
+	err = btr_cur_optimistic_insert(flags, cursor, entry, rec,
+					big_rec, n_ext, thr, mtr);
+	if (err != DB_FAIL) {
+
+		return(err);
+	}
+
+	/* Retry with a pessimistic insert. Check locks and write to undo log,
+	if specified */
+
+	err = btr_cur_ins_lock_and_undo(flags, cursor, entry, thr, &dummy_inh);
+
+	if (err != DB_SUCCESS) {
+
+		return(err);
+	}
+
+	if (!(flags & BTR_NO_UNDO_LOG_FLAG)) {
+		/* First reserve enough free space for the file segments
+		of the index tree, so that the insert will not fail because
+		of lack of space */
+
+		n_extents = cursor->tree_height / 16 + 3;
+
+		success = fsp_reserve_free_extents(&n_reserved, index->space,
+						   n_extents, FSP_NORMAL, mtr);
+		if (!success) {
+			return(DB_OUT_OF_FILE_SPACE);
+		}
+	}
+
+	if (page_zip_rec_needs_ext(rec_get_converted_size(index, entry, n_ext),
+				   dict_table_is_comp(index->table),
+				   dict_index_get_n_fields(index),
+				   zip_size)) {
+		/* The record is so big that we have to store some fields
+		externally on separate database pages */
+
+		if (UNIV_LIKELY_NULL(big_rec_vec)) {
+			/* This should never happen, but we handle
+			the situation in a robust manner. */
+			ut_ad(0);
+			dtuple_convert_back_big_rec(index, entry, big_rec_vec);
+		}
+
+		big_rec_vec = dtuple_convert_big_rec(index, entry, &n_ext);
+
+		if (big_rec_vec == NULL) {
+
+			if (n_extents > 0) {
+				fil_space_release_free_extents(index->space,
+							       n_reserved);
+			}
+			return(DB_TOO_BIG_RECORD);
+		}
+	}
+
+	if (dict_index_get_page(index)
+	    == buf_block_get_page_no(btr_cur_get_block(cursor))) {
+
+		/* The page is the root page */
+		*rec = btr_root_raise_and_insert(cursor, entry, n_ext, mtr);
+	} else {
+		*rec = btr_page_split_and_insert(cursor, entry, n_ext, mtr);
+	}
+
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+
+	ut_ad(page_rec_get_next(btr_cur_get_rec(cursor)) == *rec);
+
+#ifdef BTR_CUR_ADAPT
+	btr_search_update_hash_on_insert(cursor);
+#endif
+	if (!(flags & BTR_NO_LOCKING_FLAG)) {
+
+		lock_update_insert(btr_cur_get_block(cursor), *rec);
+	}
+
+	if (n_extents > 0) {
+		fil_space_release_free_extents(index->space, n_reserved);
+	}
+
+	*big_rec = big_rec_vec;
+
+	return(DB_SUCCESS);
+}
+
+/*==================== B-TREE UPDATE =========================*/
+
+/*****************************************************************
+For an update, checks the locks and does the undo logging. */
+UNIV_INLINE
+ulint
+btr_cur_upd_lock_and_undo(
+/*======================*/
+				/* out: DB_SUCCESS, DB_WAIT_LOCK, or error
+				number */
+	ulint		flags,	/* in: undo logging and locking flags */
+	btr_cur_t*	cursor,	/* in: cursor on record to update */
+	const upd_t*	update,	/* in: update vector */
+	ulint		cmpl_info,/* in: compiler info on secondary index
+				updates */
+	que_thr_t*	thr,	/* in: query thread */
+	dulint*		roll_ptr)/* out: roll pointer */
+{
+	dict_index_t*	index;
+	rec_t*		rec;
+	ulint		err;
+
+	ut_ad(cursor && update && thr && roll_ptr);
+
+	rec = btr_cur_get_rec(cursor);
+	index = cursor->index;
+
+	if (!dict_index_is_clust(index)) {
+		/* We do undo logging only when we update a clustered index
+		record */
+		return(lock_sec_rec_modify_check_and_lock(
+			       flags, btr_cur_get_block(cursor), rec,
+			       index, thr));
+	}
+
+	/* Check if we have to wait for a lock: enqueue an explicit lock
+	request if yes */
+
+	err = DB_SUCCESS;
+
+	if (!(flags & BTR_NO_LOCKING_FLAG)) {
+		mem_heap_t*	heap		= NULL;
+		ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+		rec_offs_init(offsets_);
+
+		err = lock_clust_rec_modify_check_and_lock(
+			flags, btr_cur_get_block(cursor), rec, index,
+			rec_get_offsets(rec, index, offsets_,
+					ULINT_UNDEFINED, &heap), thr);
+		if (UNIV_LIKELY_NULL(heap)) {
+			mem_heap_free(heap);
+		}
+		if (err != DB_SUCCESS) {
+
+			return(err);
+		}
+	}
+
+	/* Append the info about the update in the undo log */
+
+	err = trx_undo_report_row_operation(flags, TRX_UNDO_MODIFY_OP, thr,
+					    index, NULL, update,
+					    cmpl_info, rec, roll_ptr);
+	return(err);
+}
+
+/***************************************************************
+Writes a redo log record of updating a record in-place. */
+UNIV_INLINE
+void
+btr_cur_update_in_place_log(
+/*========================*/
+	ulint		flags,		/* in: flags */
+	rec_t*		rec,		/* in: record */
+	dict_index_t*	index,		/* in: index where cursor positioned */
+	const upd_t*	update,		/* in: update vector */
+	trx_t*		trx,		/* in: transaction */
+	dulint		roll_ptr,	/* in: roll ptr */
+	mtr_t*		mtr)		/* in: mtr */
+{
+	byte*	log_ptr;
+	page_t*	page	= page_align(rec);
+	ut_ad(flags < 256);
+	ut_ad(!!page_is_comp(page) == dict_table_is_comp(index->table));
+
+	log_ptr = mlog_open_and_write_index(mtr, rec, index, page_is_comp(page)
+					    ? MLOG_COMP_REC_UPDATE_IN_PLACE
+					    : MLOG_REC_UPDATE_IN_PLACE,
+					    1 + DATA_ROLL_PTR_LEN + 14 + 2
+					    + MLOG_BUF_MARGIN);
+
+	if (!log_ptr) {
+		/* Logging in mtr is switched off during crash recovery */
+		return;
+	}
+
+	/* The code below assumes index is a clustered index: change index to
+	the clustered index if we are updating a secondary index record (or we
+	could as well skip writing the sys col values to the log in this case
+	because they are not needed for a secondary index record update) */
+
+	index = dict_table_get_first_index(index->table);
+
+	mach_write_to_1(log_ptr, flags);
+	log_ptr++;
+
+	log_ptr = row_upd_write_sys_vals_to_log(index, trx, roll_ptr, log_ptr,
+						mtr);
+	mach_write_to_2(log_ptr, page_offset(rec));
+	log_ptr += 2;
+
+	row_upd_index_write_log(update, log_ptr, mtr);
+}
+
+/***************************************************************
+Parses a redo log record of updating a record in-place. */
+UNIV_INTERN
+byte*
+btr_cur_parse_update_in_place(
+/*==========================*/
+				/* out: end of log record or NULL */
+	byte*		ptr,	/* in: buffer */
+	byte*		end_ptr,/* in: buffer end */
+	page_t*		page,	/* in/out: page or NULL */
+	page_zip_des_t*	page_zip,/* in/out: compressed page, or NULL */
+	dict_index_t*	index)	/* in: index corresponding to page */
+{
+	ulint	flags;
+	rec_t*	rec;
+	upd_t*	update;
+	ulint	pos;
+	dulint	trx_id;
+	dulint	roll_ptr;
+	ulint	rec_offset;
+	mem_heap_t* heap;
+	ulint*	offsets;
+
+	if (end_ptr < ptr + 1) {
+
+		return(NULL);
+	}
+
+	flags = mach_read_from_1(ptr);
+	ptr++;
+
+	ptr = row_upd_parse_sys_vals(ptr, end_ptr, &pos, &trx_id, &roll_ptr);
+
+	if (ptr == NULL) {
+
+		return(NULL);
+	}
+
+	if (end_ptr < ptr + 2) {
+
+		return(NULL);
+	}
+
+	rec_offset = mach_read_from_2(ptr);
+	ptr += 2;
+
+	ut_a(rec_offset <= UNIV_PAGE_SIZE);
+
+	heap = mem_heap_create(256);
+
+	ptr = row_upd_index_parse(ptr, end_ptr, heap, &update);
+
+	if (!ptr || !page) {
+
+		goto func_exit;
+	}
+
+	ut_a((ibool)!!page_is_comp(page) == dict_table_is_comp(index->table));
+	rec = page + rec_offset;
+
+	/* We do not need to reserve btr_search_latch, as the page is only
+	being recovered, and there cannot be a hash index to it. */
+
+	offsets = rec_get_offsets(rec, index, NULL, ULINT_UNDEFINED, &heap);
+
+	if (!(flags & BTR_KEEP_SYS_FLAG)) {
+		row_upd_rec_sys_fields_in_recovery(rec, page_zip, offsets,
+						   pos, trx_id, roll_ptr);
+	}
+
+	row_upd_rec_in_place(rec, index, offsets, update, page_zip);
+
+func_exit:
+	mem_heap_free(heap);
+
+	return(ptr);
+}
+
+/*****************************************************************
+See if there is enough place in the page modification log to log
+an update-in-place. */
+static
+ibool
+btr_cur_update_alloc_zip(
+/*=====================*/
+				/* out: TRUE if enough place */
+	page_zip_des_t*	page_zip,/* in/out: compressed page */
+	buf_block_t*	block,	/* in/out: buffer page */
+	dict_index_t*	index,	/* in: the index corresponding to the block */
+	ulint		length,	/* in: size needed */
+	mtr_t*		mtr)	/* in: mini-transaction */
+{
+	ut_a(page_zip == buf_block_get_page_zip(block));
+	ut_ad(page_zip);
+	ut_ad(!dict_index_is_ibuf(index));
+
+	if (page_zip_available(page_zip, dict_index_is_clust(index),
+			       length, 0)) {
+		return(TRUE);
+	}
+
+	if (!page_zip->m_nonempty) {
+		/* The page has been freshly compressed, so
+		recompressing it will not help. */
+		return(FALSE);
+	}
+
+	if (!page_zip_compress(page_zip, buf_block_get_frame(block),
+			       index, mtr)) {
+		/* Unable to compress the page */
+		return(FALSE);
+	}
+
+	/* After recompressing a page, we must make sure that the free
+	bits in the insert buffer bitmap will not exceed the free
+	space on the page.  Because this function will not attempt
+	recompression unless page_zip_available() fails above, it is
+	safe to reset the free bits if page_zip_available() fails
+	again, below.  The free bits can safely be reset in a separate
+	mini-transaction.  If page_zip_available() succeeds below, we
+	can be sure that the page_zip_compress() above did not reduce
+	the free space available on the page. */
+
+	if (!page_zip_available(page_zip, dict_index_is_clust(index),
+				length, 0)) {
+		/* Out of space: reset the free bits. */
+		if (!dict_index_is_clust(index)
+		    && page_is_leaf(buf_block_get_frame(block))) {
+			ibuf_reset_free_bits(block);
+		}
+		return(FALSE);
+	}
+
+	return(TRUE);
+}
+
+/*****************************************************************
+Updates a record when the update causes no size changes in its fields.
+We assume here that the ordering fields of the record do not change. */
+UNIV_INTERN
+ulint
+btr_cur_update_in_place(
+/*====================*/
+				/* out: DB_SUCCESS or error number */
+	ulint		flags,	/* in: undo logging and locking flags */
+	btr_cur_t*	cursor,	/* in: cursor on the record to update;
+				cursor stays valid and positioned on the
+				same record */
+	const upd_t*	update,	/* in: update vector */
+	ulint		cmpl_info,/* in: compiler info on secondary index
+				updates */
+	que_thr_t*	thr,	/* in: query thread */
+	mtr_t*		mtr)	/* in: mtr; must be committed before
+				latching any further pages */
+{
+	dict_index_t*	index;
+	buf_block_t*	block;
+	page_zip_des_t*	page_zip;
+	ulint		err;
+	rec_t*		rec;
+	dulint		roll_ptr	= ut_dulint_zero;
+	trx_t*		trx;
+	ulint		was_delete_marked;
+	mem_heap_t*	heap		= NULL;
+	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint*		offsets		= offsets_;
+	rec_offs_init(offsets_);
+
+	rec = btr_cur_get_rec(cursor);
+	index = cursor->index;
+	ut_ad(!!page_rec_is_comp(rec) == dict_table_is_comp(index->table));
+	/* The insert buffer tree should never be updated in place. */
+	ut_ad(!dict_index_is_ibuf(index));
+
+	trx = thr_get_trx(thr);
+	offsets = rec_get_offsets(rec, index, offsets, ULINT_UNDEFINED, &heap);
+#ifdef UNIV_DEBUG
+	if (btr_cur_print_record_ops && thr) {
+		btr_cur_trx_report(trx, index, "update ");
+		rec_print_new(stderr, rec, offsets);
+	}
+#endif /* UNIV_DEBUG */
+
+	block = btr_cur_get_block(cursor);
+	page_zip = buf_block_get_page_zip(block);
+
+	/* Check that enough space is available on the compressed page. */
+	if (UNIV_LIKELY_NULL(page_zip)
+	    && !btr_cur_update_alloc_zip(page_zip, block, index,
+					 rec_offs_size(offsets), mtr)) {
+		return(DB_ZIP_OVERFLOW);
+	}
+
+	/* Do lock checking and undo logging */
+	err = btr_cur_upd_lock_and_undo(flags, cursor, update, cmpl_info,
+					thr, &roll_ptr);
+	if (UNIV_UNLIKELY(err != DB_SUCCESS)) {
+
+		if (UNIV_LIKELY_NULL(heap)) {
+			mem_heap_free(heap);
+		}
+		return(err);
+	}
+
+	if (block->is_hashed) {
+		/* The function row_upd_changes_ord_field_binary works only
+		if the update vector was built for a clustered index, we must
+		NOT call it if index is secondary */
+
+		if (!dict_index_is_clust(index)
+		    || row_upd_changes_ord_field_binary(NULL, index, update)) {
+
+			/* Remove possible hash index pointer to this record */
+			btr_search_update_hash_on_delete(cursor);
+		}
+
+		rw_lock_x_lock(&btr_search_latch);
+	}
+
+	if (!(flags & BTR_KEEP_SYS_FLAG)) {
+		row_upd_rec_sys_fields(rec, NULL,
+				       index, offsets, trx, roll_ptr);
+	}
+
+	was_delete_marked = rec_get_deleted_flag(
+		rec, page_is_comp(buf_block_get_frame(block)));
+
+	row_upd_rec_in_place(rec, index, offsets, update, page_zip);
+
+	if (block->is_hashed) {
+		rw_lock_x_unlock(&btr_search_latch);
+	}
+
+	if (page_zip && !dict_index_is_clust(index)
+	    && page_is_leaf(buf_block_get_frame(block))) {
+		/* Update the free bits in the insert buffer. */
+		ibuf_update_free_bits_zip(block, mtr);
+	}
+
+	btr_cur_update_in_place_log(flags, rec, index, update,
+				    trx, roll_ptr, mtr);
+
+	if (was_delete_marked
+	    && !rec_get_deleted_flag(rec, page_is_comp(
+					     buf_block_get_frame(block)))) {
+		/* The new updated record owns its possible externally
+		stored fields */
+
+		btr_cur_unmark_extern_fields(page_zip,
+					     rec, index, offsets, mtr);
+	}
+
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+	return(DB_SUCCESS);
+}
+
+/*****************************************************************
+Tries to update a record on a page in an index tree. It is assumed that mtr
+holds an x-latch on the page. The operation does not succeed if there is too
+little space on the page or if the update would result in too empty a page,
+so that tree compression is recommended. We assume here that the ordering
+fields of the record do not change. */
+UNIV_INTERN
+ulint
+btr_cur_optimistic_update(
+/*======================*/
+				/* out: DB_SUCCESS, or DB_OVERFLOW if the
+				updated record does not fit, DB_UNDERFLOW
+				if the page would become too empty, or
+				DB_ZIP_OVERFLOW if there is not enough
+				space left on the compressed page */
+	ulint		flags,	/* in: undo logging and locking flags */
+	btr_cur_t*	cursor,	/* in: cursor on the record to update;
+				cursor stays valid and positioned on the
+				same record */
+	const upd_t*	update,	/* in: update vector; this must also
+				contain trx id and roll ptr fields */
+	ulint		cmpl_info,/* in: compiler info on secondary index
+				updates */
+	que_thr_t*	thr,	/* in: query thread */
+	mtr_t*		mtr)	/* in: mtr; must be committed before
+				latching any further pages */
+{
+	dict_index_t*	index;
+	page_cur_t*	page_cursor;
+	ulint		err;
+	buf_block_t*	block;
+	page_t*		page;
+	page_zip_des_t*	page_zip;
+	rec_t*		rec;
+	rec_t*		orig_rec;
+	ulint		max_size;
+	ulint		new_rec_size;
+	ulint		old_rec_size;
+	dtuple_t*	new_entry;
+	dulint		roll_ptr;
+	trx_t*		trx;
+	mem_heap_t*	heap;
+	ulint		i;
+	ulint		n_ext;
+	ulint*		offsets;
+
+	block = btr_cur_get_block(cursor);
+	page = buf_block_get_frame(block);
+	orig_rec = rec = btr_cur_get_rec(cursor);
+	index = cursor->index;
+	ut_ad(!!page_rec_is_comp(rec) == dict_table_is_comp(index->table));
+	ut_ad(mtr_memo_contains(mtr, block, MTR_MEMO_PAGE_X_FIX));
+	/* The insert buffer tree should never be updated in place. */
+	ut_ad(!dict_index_is_ibuf(index));
+
+	heap = mem_heap_create(1024);
+	offsets = rec_get_offsets(rec, index, NULL, ULINT_UNDEFINED, &heap);
+
+#ifdef UNIV_DEBUG
+	if (btr_cur_print_record_ops && thr) {
+		btr_cur_trx_report(thr_get_trx(thr), index, "update ");
+		rec_print_new(stderr, rec, offsets);
+	}
+#endif /* UNIV_DEBUG */
+
+	if (!row_upd_changes_field_size_or_external(index, offsets, update)) {
+
+		/* The simplest and the most common case: the update does not
+		change the size of any field and none of the updated fields is
+		externally stored in rec or update, and there is enough space
+		on the compressed page to log the update. */
+
+		mem_heap_free(heap);
+		return(btr_cur_update_in_place(flags, cursor, update,
+					       cmpl_info, thr, mtr));
+	}
+
+	if (rec_offs_any_extern(offsets)) {
+any_extern:
+		/* Externally stored fields are treated in pessimistic
+		update */
+
+		mem_heap_free(heap);
+		return(DB_OVERFLOW);
+	}
+
+	for (i = 0; i < upd_get_n_fields(update); i++) {
+		if (dfield_is_ext(&upd_get_nth_field(update, i)->new_val)) {
+
+			goto any_extern;
+		}
+	}
+
+	page_cursor = btr_cur_get_page_cur(cursor);
+
+	new_entry = row_rec_to_index_entry(ROW_COPY_DATA, rec, index, offsets,
+					   &n_ext, heap);
+	/* We checked above that there are no externally stored fields. */
+	ut_a(!n_ext);
+
+	/* The page containing the clustered index record
+	corresponding to new_entry is latched in mtr.
+	Thus the following call is safe. */
+	row_upd_index_replace_new_col_vals_index_pos(new_entry, index, update,
+						     FALSE, heap);
+	old_rec_size = rec_offs_size(offsets);
+	new_rec_size = rec_get_converted_size(index, new_entry, 0);
+
+	page_zip = buf_block_get_page_zip(block);
+#ifdef UNIV_ZIP_DEBUG
+	ut_a(!page_zip || page_zip_validate(page_zip, page));
+#endif /* UNIV_ZIP_DEBUG */
+
+	if (UNIV_LIKELY_NULL(page_zip)
+	    && !btr_cur_update_alloc_zip(page_zip, block, index,
+					 new_rec_size, mtr)) {
+		err = DB_ZIP_OVERFLOW;
+		goto err_exit;
+	}
+
+	if (UNIV_UNLIKELY(new_rec_size
+			  >= (page_get_free_space_of_empty(page_is_comp(page))
+			      / 2))) {
+
+		err = DB_OVERFLOW;
+		goto err_exit;
+	}
+
+	if (UNIV_UNLIKELY(page_get_data_size(page)
+			  - old_rec_size + new_rec_size
+			  < BTR_CUR_PAGE_COMPRESS_LIMIT)) {
+
+		/* The page would become too empty */
+
+		err = DB_UNDERFLOW;
+		goto err_exit;
+	}
+
+	max_size = old_rec_size
+		+ page_get_max_insert_size_after_reorganize(page, 1);
+
+	if (!(((max_size >= BTR_CUR_PAGE_REORGANIZE_LIMIT)
+	       && (max_size >= new_rec_size))
+	      || (page_get_n_recs(page) <= 1))) {
+
+		/* There was not enough space, or it did not pay to
+		reorganize: for simplicity, we decide what to do assuming a
+		reorganization is needed, though it might not be necessary */
+
+		err = DB_OVERFLOW;
+		goto err_exit;
+	}
+
+	/* Do lock checking and undo logging */
+	err = btr_cur_upd_lock_and_undo(flags, cursor, update, cmpl_info, thr,
+					&roll_ptr);
+	if (err != DB_SUCCESS) {
+err_exit:
+		mem_heap_free(heap);
+		return(err);
+	}
+
+	/* Ok, we may do the replacement. Store on the page infimum the
+	explicit locks on rec, before deleting rec (see the comment in
+	btr_cur_pessimistic_update). */
+
+	lock_rec_store_on_page_infimum(block, rec);
+
+	btr_search_update_hash_on_delete(cursor);
+
+	/* The call to row_rec_to_index_entry(ROW_COPY_DATA, ...) above
+	invokes rec_offs_make_valid() to point to the copied record that
+	the fields of new_entry point to.  We have to undo it here. */
+	ut_ad(rec_offs_validate(NULL, index, offsets));
+	rec_offs_make_valid(page_cur_get_rec(page_cursor), index, offsets);
+
+	page_cur_delete_rec(page_cursor, index, offsets, mtr);
+
+	page_cur_move_to_prev(page_cursor);
+
+	trx = thr_get_trx(thr);
+
+	if (!(flags & BTR_KEEP_SYS_FLAG)) {
+		row_upd_index_entry_sys_field(new_entry, index, DATA_ROLL_PTR,
+					      roll_ptr);
+		row_upd_index_entry_sys_field(new_entry, index, DATA_TRX_ID,
+					      trx->id);
+	}
+
+	/* There are no externally stored columns in new_entry */
+	rec = btr_cur_insert_if_possible(cursor, new_entry, 0/*n_ext*/, mtr);
+	ut_a(rec); /* <- We calculated above the insert would fit */
+
+	if (page_zip && !dict_index_is_clust(index)
+	    && page_is_leaf(page)) {
+		/* Update the free bits in the insert buffer. */
+		ibuf_update_free_bits_zip(block, mtr);
+	}
+
+	/* Restore the old explicit lock state on the record */
+
+	lock_rec_restore_from_page_infimum(block, rec, block);
+
+	page_cur_move_to_next(page_cursor);
+
+	mem_heap_free(heap);
+
+	return(DB_SUCCESS);
+}
+
+/*****************************************************************
+If, in a split, a new supremum record was created as the predecessor of the
+updated record, the supremum record must inherit exactly the locks on the
+updated record. In the split it may have inherited locks from the successor
+of the updated record, which is not correct. This function restores the
+right locks for the new supremum. */
+static
+void
+btr_cur_pess_upd_restore_supremum(
+/*==============================*/
+	buf_block_t*	block,	/* in: buffer block of rec */
+	const rec_t*	rec,	/* in: updated record */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	page_t*		page;
+	buf_block_t*	prev_block;
+	ulint		space;
+	ulint		zip_size;
+	ulint		prev_page_no;
+
+	page = buf_block_get_frame(block);
+
+	if (page_rec_get_next(page_get_infimum_rec(page)) != rec) {
+		/* Updated record is not the first user record on its page */
+
+		return;
+	}
+
+	space = buf_block_get_space(block);
+	zip_size = buf_block_get_zip_size(block);
+	prev_page_no = btr_page_get_prev(page, mtr);
+
+	ut_ad(prev_page_no != FIL_NULL);
+	prev_block = buf_page_get_with_no_latch(space, zip_size,
+						prev_page_no, mtr);
+#ifdef UNIV_BTR_DEBUG
+	ut_a(btr_page_get_next(prev_block->frame, mtr)
+	     == page_get_page_no(page));
+#endif /* UNIV_BTR_DEBUG */
+
+	/* We must already have an x-latch on prev_block! */
+	ut_ad(mtr_memo_contains(mtr, prev_block, MTR_MEMO_PAGE_X_FIX));
+
+	lock_rec_reset_and_inherit_gap_locks(prev_block, block,
+					     PAGE_HEAP_NO_SUPREMUM,
+					     page_rec_get_heap_no(rec));
+}
+
+/*****************************************************************
+Performs an update of a record on a page of a tree. It is assumed
+that mtr holds an x-latch on the tree and on the cursor page. If the
+update is made on the leaf level, to avoid deadlocks, mtr must also
+own x-latches to brothers of page, if those brothers exist. We assume
+here that the ordering fields of the record do not change. */
+UNIV_INTERN
+ulint
+btr_cur_pessimistic_update(
+/*=======================*/
+				/* out: DB_SUCCESS or error code */
+	ulint		flags,	/* in: undo logging, locking, and rollback
+				flags */
+	btr_cur_t*	cursor,	/* in: cursor on the record to update */
+	mem_heap_t**	heap,	/* in/out: pointer to memory heap, or NULL */
+	big_rec_t**	big_rec,/* out: big rec vector whose fields have to
+				be stored externally by the caller, or NULL */
+	const upd_t*	update,	/* in: update vector; this is allowed also
+				contain trx id and roll ptr fields, but
+				the values in update vector have no effect */
+	ulint		cmpl_info,/* in: compiler info on secondary index
+				updates */
+	que_thr_t*	thr,	/* in: query thread */
+	mtr_t*		mtr)	/* in: mtr; must be committed before
+				latching any further pages */
+{
+	big_rec_t*	big_rec_vec	= NULL;
+	big_rec_t*	dummy_big_rec;
+	dict_index_t*	index;
+	buf_block_t*	block;
+	page_t*		page;
+	page_zip_des_t*	page_zip;
+	rec_t*		rec;
+	page_cur_t*	page_cursor;
+	dtuple_t*	new_entry;
+	ulint		err;
+	ulint		optim_err;
+	dulint		roll_ptr;
+	trx_t*		trx;
+	ibool		was_first;
+	ulint		n_extents	= 0;
+	ulint		n_reserved;
+	ulint		n_ext;
+	ulint*		offsets		= NULL;
+
+	*big_rec = NULL;
+
+	block = btr_cur_get_block(cursor);
+	page = buf_block_get_frame(block);
+	page_zip = buf_block_get_page_zip(block);
+	rec = btr_cur_get_rec(cursor);
+	index = cursor->index;
+
+	ut_ad(mtr_memo_contains(mtr, dict_index_get_lock(index),
+				MTR_MEMO_X_LOCK));
+	ut_ad(mtr_memo_contains(mtr, block, MTR_MEMO_PAGE_X_FIX));
+#ifdef UNIV_ZIP_DEBUG
+	ut_a(!page_zip || page_zip_validate(page_zip, page));
+#endif /* UNIV_ZIP_DEBUG */
+	/* The insert buffer tree should never be updated in place. */
+	ut_ad(!dict_index_is_ibuf(index));
+
+	optim_err = btr_cur_optimistic_update(flags, cursor, update,
+					      cmpl_info, thr, mtr);
+
+	switch (optim_err) {
+	case DB_UNDERFLOW:
+	case DB_OVERFLOW:
+	case DB_ZIP_OVERFLOW:
+		break;
+	default:
+		return(optim_err);
+	}
+
+	/* Do lock checking and undo logging */
+	err = btr_cur_upd_lock_and_undo(flags, cursor, update, cmpl_info,
+					thr, &roll_ptr);
+	if (err != DB_SUCCESS) {
+
+		return(err);
+	}
+
+	if (optim_err == DB_OVERFLOW) {
+		ulint	reserve_flag;
+
+		/* First reserve enough free space for the file segments
+		of the index tree, so that the update will not fail because
+		of lack of space */
+
+		n_extents = cursor->tree_height / 16 + 3;
+
+		if (flags & BTR_NO_UNDO_LOG_FLAG) {
+			reserve_flag = FSP_CLEANING;
+		} else {
+			reserve_flag = FSP_NORMAL;
+		}
+
+		if (!fsp_reserve_free_extents(&n_reserved, index->space,
+					      n_extents, reserve_flag, mtr)) {
+			return(DB_OUT_OF_FILE_SPACE);
+		}
+	}
+
+	if (!*heap) {
+		*heap = mem_heap_create(1024);
+	}
+	offsets = rec_get_offsets(rec, index, NULL, ULINT_UNDEFINED, heap);
+
+	trx = thr_get_trx(thr);
+
+	new_entry = row_rec_to_index_entry(ROW_COPY_DATA, rec, index, offsets,
+					   &n_ext, *heap);
+	/* The call to row_rec_to_index_entry(ROW_COPY_DATA, ...) above
+	invokes rec_offs_make_valid() to point to the copied record that
+	the fields of new_entry point to.  We have to undo it here. */
+	ut_ad(rec_offs_validate(NULL, index, offsets));
+	rec_offs_make_valid(rec, index, offsets);
+
+	/* The page containing the clustered index record
+	corresponding to new_entry is latched in mtr.  If the
+	clustered index record is delete-marked, then its externally
+	stored fields cannot have been purged yet, because then the
+	purge would also have removed the clustered index record
+	itself.  Thus the following call is safe. */
+	row_upd_index_replace_new_col_vals_index_pos(new_entry, index, update,
+						     FALSE, *heap);
+	if (!(flags & BTR_KEEP_SYS_FLAG)) {
+		row_upd_index_entry_sys_field(new_entry, index, DATA_ROLL_PTR,
+					      roll_ptr);
+		row_upd_index_entry_sys_field(new_entry, index, DATA_TRX_ID,
+					      trx->id);
+	}
+
+	if ((flags & BTR_NO_UNDO_LOG_FLAG) && rec_offs_any_extern(offsets)) {
+		/* We are in a transaction rollback undoing a row
+		update: we must free possible externally stored fields
+		which got new values in the update, if they are not
+		inherited values. They can be inherited if we have
+		updated the primary key to another value, and then
+		update it back again. */
+
+		ut_ad(big_rec_vec == NULL);
+
+		btr_rec_free_updated_extern_fields(
+			index, rec, page_zip, offsets, update,
+			trx_is_recv(trx) ? RB_RECOVERY : RB_NORMAL, mtr);
+	}
+
+	/* We have to set appropriate extern storage bits in the new
+	record to be inserted: we have to remember which fields were such */
+
+	ut_ad(!page_is_comp(page) || !rec_get_node_ptr_flag(rec));
+	offsets = rec_get_offsets(rec, index, offsets, ULINT_UNDEFINED, heap);
+	n_ext += btr_push_update_extern_fields(new_entry, update, *heap);
+
+	if (UNIV_LIKELY_NULL(page_zip)) {
+		ut_ad(page_is_comp(page));
+		if (page_zip_rec_needs_ext(
+			    rec_get_converted_size(index, new_entry, n_ext),
+			    TRUE,
+			    dict_index_get_n_fields(index),
+			    page_zip_get_size(page_zip))) {
+
+			goto make_external;
+		}
+	} else if (page_zip_rec_needs_ext(
+			   rec_get_converted_size(index, new_entry, n_ext),
+			   page_is_comp(page), 0, 0)) {
+make_external:
+		big_rec_vec = dtuple_convert_big_rec(index, new_entry, &n_ext);
+		if (UNIV_UNLIKELY(big_rec_vec == NULL)) {
+
+			err = DB_TOO_BIG_RECORD;
+			goto return_after_reservations;
+		}
+	}
+
+	/* Store state of explicit locks on rec on the page infimum record,
+	before deleting rec. The page infimum acts as a dummy carrier of the
+	locks, taking care also of lock releases, before we can move the locks
+	back on the actual record. There is a special case: if we are
+	inserting on the root page and the insert causes a call of
+	btr_root_raise_and_insert. Therefore we cannot in the lock system
+	delete the lock structs set on the root page even if the root
+	page carries just node pointers. */
+
+	lock_rec_store_on_page_infimum(block, rec);
+
+	btr_search_update_hash_on_delete(cursor);
+
+#ifdef UNIV_ZIP_DEBUG
+	ut_a(!page_zip || page_zip_validate(page_zip, page));
+#endif /* UNIV_ZIP_DEBUG */
+	page_cursor = btr_cur_get_page_cur(cursor);
+
+	page_cur_delete_rec(page_cursor, index, offsets, mtr);
+
+	page_cur_move_to_prev(page_cursor);
+
+	rec = btr_cur_insert_if_possible(cursor, new_entry, n_ext, mtr);
+
+	if (rec) {
+		lock_rec_restore_from_page_infimum(btr_cur_get_block(cursor),
+						   rec, block);
+
+		offsets = rec_get_offsets(rec, index, offsets,
+					  ULINT_UNDEFINED, heap);
+
+		if (!rec_get_deleted_flag(rec, rec_offs_comp(offsets))) {
+			/* The new inserted record owns its possible externally
+			stored fields */
+			btr_cur_unmark_extern_fields(page_zip,
+						     rec, index, offsets, mtr);
+		}
+
+		btr_cur_compress_if_useful(cursor, mtr);
+
+		if (page_zip && !dict_index_is_clust(index)
+		    && page_is_leaf(page)) {
+			/* Update the free bits in the insert buffer. */
+			ibuf_update_free_bits_zip(block, mtr);
+		}
+
+		err = DB_SUCCESS;
+		goto return_after_reservations;
+	} else {
+		ut_a(optim_err != DB_UNDERFLOW);
+
+		/* Out of space: reset the free bits. */
+		if (!dict_index_is_clust(index)
+		    && page_is_leaf(page)) {
+			ibuf_reset_free_bits(block);
+		}
+	}
+
+	/* Was the record to be updated positioned as the first user
+	record on its page? */
+	was_first = page_cur_is_before_first(page_cursor);
+
+	/* The first parameter means that no lock checking and undo logging
+	is made in the insert */
+
+	err = btr_cur_pessimistic_insert(BTR_NO_UNDO_LOG_FLAG
+					 | BTR_NO_LOCKING_FLAG
+					 | BTR_KEEP_SYS_FLAG,
+					 cursor, new_entry, &rec,
+					 &dummy_big_rec, n_ext, NULL, mtr);
+	ut_a(rec);
+	ut_a(err == DB_SUCCESS);
+	ut_a(dummy_big_rec == NULL);
+
+	if (!rec_get_deleted_flag(rec, rec_offs_comp(offsets))) {
+		/* The new inserted record owns its possible externally
+		stored fields */
+		buf_block_t*	rec_block = btr_cur_get_block(cursor);
+
+#ifdef UNIV_ZIP_DEBUG
+		ut_a(!page_zip || page_zip_validate(page_zip, page));
+		page = buf_block_get_frame(rec_block);
+#endif /* UNIV_ZIP_DEBUG */
+		page_zip = buf_block_get_page_zip(rec_block);
+
+		offsets = rec_get_offsets(rec, index, offsets,
+					  ULINT_UNDEFINED, heap);
+		btr_cur_unmark_extern_fields(page_zip,
+					     rec, index, offsets, mtr);
+	}
+
+	lock_rec_restore_from_page_infimum(btr_cur_get_block(cursor),
+					   rec, block);
+
+	/* If necessary, restore also the correct lock state for a new,
+	preceding supremum record created in a page split. While the old
+	record was nonexistent, the supremum might have inherited its locks
+	from a wrong record. */
+
+	if (!was_first) {
+		btr_cur_pess_upd_restore_supremum(btr_cur_get_block(cursor),
+						  rec, mtr);
+	}
+
+return_after_reservations:
+#ifdef UNIV_ZIP_DEBUG
+	ut_a(!page_zip || page_zip_validate(page_zip, page));
+#endif /* UNIV_ZIP_DEBUG */
+
+	if (n_extents > 0) {
+		fil_space_release_free_extents(index->space, n_reserved);
+	}
+
+	*big_rec = big_rec_vec;
+
+	return(err);
+}
+
+/*==================== B-TREE DELETE MARK AND UNMARK ===============*/
+
+/********************************************************************
+Writes the redo log record for delete marking or unmarking of an index
+record. */
+UNIV_INLINE
+void
+btr_cur_del_mark_set_clust_rec_log(
+/*===============================*/
+	ulint		flags,	/* in: flags */
+	rec_t*		rec,	/* in: record */
+	dict_index_t*	index,	/* in: index of the record */
+	ibool		val,	/* in: value to set */
+	trx_t*		trx,	/* in: deleting transaction */
+	dulint		roll_ptr,/* in: roll ptr to the undo log record */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	byte*	log_ptr;
+	ut_ad(flags < 256);
+	ut_ad(val <= 1);
+
+	ut_ad(!!page_rec_is_comp(rec) == dict_table_is_comp(index->table));
+
+	log_ptr = mlog_open_and_write_index(mtr, rec, index,
+					    page_rec_is_comp(rec)
+					    ? MLOG_COMP_REC_CLUST_DELETE_MARK
+					    : MLOG_REC_CLUST_DELETE_MARK,
+					    1 + 1 + DATA_ROLL_PTR_LEN
+					    + 14 + 2);
+
+	if (!log_ptr) {
+		/* Logging in mtr is switched off during crash recovery */
+		return;
+	}
+
+	mach_write_to_1(log_ptr, flags);
+	log_ptr++;
+	mach_write_to_1(log_ptr, val);
+	log_ptr++;
+
+	log_ptr = row_upd_write_sys_vals_to_log(index, trx, roll_ptr, log_ptr,
+						mtr);
+	mach_write_to_2(log_ptr, page_offset(rec));
+	log_ptr += 2;
+
+	mlog_close(mtr, log_ptr);
+}
+
+/********************************************************************
+Parses the redo log record for delete marking or unmarking of a clustered
+index record. */
+UNIV_INTERN
+byte*
+btr_cur_parse_del_mark_set_clust_rec(
+/*=================================*/
+				/* out: end of log record or NULL */
+	byte*		ptr,	/* in: buffer */
+	byte*		end_ptr,/* in: buffer end */
+	page_t*		page,	/* in/out: page or NULL */
+	page_zip_des_t*	page_zip,/* in/out: compressed page, or NULL */
+	dict_index_t*	index)	/* in: index corresponding to page */
+{
+	ulint	flags;
+	ulint	val;
+	ulint	pos;
+	dulint	trx_id;
+	dulint	roll_ptr;
+	ulint	offset;
+	rec_t*	rec;
+
+	ut_ad(!page
+	      || !!page_is_comp(page) == dict_table_is_comp(index->table));
+
+	if (end_ptr < ptr + 2) {
+
+		return(NULL);
+	}
+
+	flags = mach_read_from_1(ptr);
+	ptr++;
+	val = mach_read_from_1(ptr);
+	ptr++;
+
+	ptr = row_upd_parse_sys_vals(ptr, end_ptr, &pos, &trx_id, &roll_ptr);
+
+	if (ptr == NULL) {
+
+		return(NULL);
+	}
+
+	if (end_ptr < ptr + 2) {
+
+		return(NULL);
+	}
+
+	offset = mach_read_from_2(ptr);
+	ptr += 2;
+
+	ut_a(offset <= UNIV_PAGE_SIZE);
+
+	if (page) {
+		rec = page + offset;
+
+		/* We do not need to reserve btr_search_latch, as the page
+		is only being recovered, and there cannot be a hash index to
+		it. */
+
+		btr_rec_set_deleted_flag(rec, page_zip, val);
+
+		if (!(flags & BTR_KEEP_SYS_FLAG)) {
+			mem_heap_t*	heap		= NULL;
+			ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+			rec_offs_init(offsets_);
+
+			row_upd_rec_sys_fields_in_recovery(
+				rec, page_zip,
+				rec_get_offsets(rec, index, offsets_,
+						ULINT_UNDEFINED, &heap),
+				pos, trx_id, roll_ptr);
+			if (UNIV_LIKELY_NULL(heap)) {
+				mem_heap_free(heap);
+			}
+		}
+	}
+
+	return(ptr);
+}
+
+/***************************************************************
+Marks a clustered index record deleted. Writes an undo log record to
+undo log on this delete marking. Writes in the trx id field the id
+of the deleting transaction, and in the roll ptr field pointer to the
+undo log record created. */
+UNIV_INTERN
+ulint
+btr_cur_del_mark_set_clust_rec(
+/*===========================*/
+				/* out: DB_SUCCESS, DB_LOCK_WAIT, or error
+				number */
+	ulint		flags,	/* in: undo logging and locking flags */
+	btr_cur_t*	cursor,	/* in: cursor */
+	ibool		val,	/* in: value to set */
+	que_thr_t*	thr,	/* in: query thread */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	dict_index_t*	index;
+	buf_block_t*	block;
+	dulint		roll_ptr;
+	ulint		err;
+	rec_t*		rec;
+	page_zip_des_t*	page_zip;
+	trx_t*		trx;
+	mem_heap_t*	heap		= NULL;
+	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint*		offsets		= offsets_;
+	rec_offs_init(offsets_);
+
+	rec = btr_cur_get_rec(cursor);
+	index = cursor->index;
+	ut_ad(!!page_rec_is_comp(rec) == dict_table_is_comp(index->table));
+	offsets = rec_get_offsets(rec, index, offsets, ULINT_UNDEFINED, &heap);
+
+#ifdef UNIV_DEBUG
+	if (btr_cur_print_record_ops && thr) {
+		btr_cur_trx_report(thr_get_trx(thr), index, "del mark ");
+		rec_print_new(stderr, rec, offsets);
+	}
+#endif /* UNIV_DEBUG */
+
+	ut_ad(dict_index_is_clust(index));
+	ut_ad(!rec_get_deleted_flag(rec, rec_offs_comp(offsets)));
+
+	err = lock_clust_rec_modify_check_and_lock(flags,
+						   btr_cur_get_block(cursor),
+						   rec, index, offsets, thr);
+
+	if (err != DB_SUCCESS) {
+
+		goto func_exit;
+	}
+
+	err = trx_undo_report_row_operation(flags, TRX_UNDO_MODIFY_OP, thr,
+					    index, NULL, NULL, 0, rec,
+					    &roll_ptr);
+	if (err != DB_SUCCESS) {
+
+		goto func_exit;
+	}
+
+	block = btr_cur_get_block(cursor);
+
+	if (block->is_hashed) {
+		rw_lock_x_lock(&btr_search_latch);
+	}
+
+	page_zip = buf_block_get_page_zip(block);
+
+	btr_rec_set_deleted_flag(rec, page_zip, val);
+
+	trx = thr_get_trx(thr);
+
+	if (!(flags & BTR_KEEP_SYS_FLAG)) {
+		row_upd_rec_sys_fields(rec, page_zip,
+				       index, offsets, trx, roll_ptr);
+	}
+
+	if (block->is_hashed) {
+		rw_lock_x_unlock(&btr_search_latch);
+	}
+
+	btr_cur_del_mark_set_clust_rec_log(flags, rec, index, val, trx,
+					   roll_ptr, mtr);
+
+func_exit:
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+	return(err);
+}
+
+/********************************************************************
+Writes the redo log record for a delete mark setting of a secondary
+index record. */
+UNIV_INLINE
+void
+btr_cur_del_mark_set_sec_rec_log(
+/*=============================*/
+	rec_t*		rec,	/* in: record */
+	ibool		val,	/* in: value to set */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	byte*	log_ptr;
+	ut_ad(val <= 1);
+
+	log_ptr = mlog_open(mtr, 11 + 1 + 2);
+
+	if (!log_ptr) {
+		/* Logging in mtr is switched off during crash recovery:
+		in that case mlog_open returns NULL */
+		return;
+	}
+
+	log_ptr = mlog_write_initial_log_record_fast(
+		rec, MLOG_REC_SEC_DELETE_MARK, log_ptr, mtr);
+	mach_write_to_1(log_ptr, val);
+	log_ptr++;
+
+	mach_write_to_2(log_ptr, page_offset(rec));
+	log_ptr += 2;
+
+	mlog_close(mtr, log_ptr);
+}
+
+/********************************************************************
+Parses the redo log record for delete marking or unmarking of a secondary
+index record. */
+UNIV_INTERN
+byte*
+btr_cur_parse_del_mark_set_sec_rec(
+/*===============================*/
+				/* out: end of log record or NULL */
+	byte*		ptr,	/* in: buffer */
+	byte*		end_ptr,/* in: buffer end */
+	page_t*		page,	/* in/out: page or NULL */
+	page_zip_des_t*	page_zip)/* in/out: compressed page, or NULL */
+{
+	ulint	val;
+	ulint	offset;
+	rec_t*	rec;
+
+	if (end_ptr < ptr + 3) {
+
+		return(NULL);
+	}
+
+	val = mach_read_from_1(ptr);
+	ptr++;
+
+	offset = mach_read_from_2(ptr);
+	ptr += 2;
+
+	ut_a(offset <= UNIV_PAGE_SIZE);
+
+	if (page) {
+		rec = page + offset;
+
+		/* We do not need to reserve btr_search_latch, as the page
+		is only being recovered, and there cannot be a hash index to
+		it. */
+
+		btr_rec_set_deleted_flag(rec, page_zip, val);
+	}
+
+	return(ptr);
+}
+
+/***************************************************************
+Sets a secondary index record delete mark to TRUE or FALSE. */
+UNIV_INTERN
+ulint
+btr_cur_del_mark_set_sec_rec(
+/*=========================*/
+				/* out: DB_SUCCESS, DB_LOCK_WAIT, or error
+				number */
+	ulint		flags,	/* in: locking flag */
+	btr_cur_t*	cursor,	/* in: cursor */
+	ibool		val,	/* in: value to set */
+	que_thr_t*	thr,	/* in: query thread */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	buf_block_t*	block;
+	rec_t*		rec;
+	ulint		err;
+
+	block = btr_cur_get_block(cursor);
+	rec = btr_cur_get_rec(cursor);
+
+#ifdef UNIV_DEBUG
+	if (btr_cur_print_record_ops && thr) {
+		btr_cur_trx_report(thr_get_trx(thr), cursor->index,
+				   "del mark ");
+		rec_print(stderr, rec, cursor->index);
+	}
+#endif /* UNIV_DEBUG */
+
+	err = lock_sec_rec_modify_check_and_lock(flags,
+						 btr_cur_get_block(cursor),
+						 rec, cursor->index, thr);
+	if (err != DB_SUCCESS) {
+
+		return(err);
+	}
+
+	ut_ad(!!page_rec_is_comp(rec)
+	      == dict_table_is_comp(cursor->index->table));
+
+	if (block->is_hashed) {
+		rw_lock_x_lock(&btr_search_latch);
+	}
+
+	btr_rec_set_deleted_flag(rec, buf_block_get_page_zip(block), val);
+
+	if (block->is_hashed) {
+		rw_lock_x_unlock(&btr_search_latch);
+	}
+
+	btr_cur_del_mark_set_sec_rec_log(rec, val, mtr);
+
+	return(DB_SUCCESS);
+}
+
+/***************************************************************
+Clear a secondary index record's delete mark.  This function is only
+used by the insert buffer insert merge mechanism. */
+UNIV_INTERN
+void
+btr_cur_del_unmark_for_ibuf(
+/*========================*/
+	rec_t*		rec,		/* in/out: record to delete unmark */
+	page_zip_des_t*	page_zip,	/* in/out: compressed page
+					corresponding to rec, or NULL
+					when the tablespace is
+					uncompressed */
+	mtr_t*		mtr)		/* in: mtr */
+{
+	/* We do not need to reserve btr_search_latch, as the page has just
+	been read to the buffer pool and there cannot be a hash index to it. */
+
+	btr_rec_set_deleted_flag(rec, page_zip, FALSE);
+
+	btr_cur_del_mark_set_sec_rec_log(rec, FALSE, mtr);
+}
+
+/*==================== B-TREE RECORD REMOVE =========================*/
+
+/*****************************************************************
+Tries to compress a page of the tree if it seems useful. It is assumed
+that mtr holds an x-latch on the tree and on the cursor page. To avoid
+deadlocks, mtr must also own x-latches to brothers of page, if those
+brothers exist. NOTE: it is assumed that the caller has reserved enough
+free extents so that the compression will always succeed if done! */
+UNIV_INTERN
+ibool
+btr_cur_compress_if_useful(
+/*=======================*/
+				/* out: TRUE if compression occurred */
+	btr_cur_t*	cursor,	/* in: cursor on the page to compress;
+				cursor does not stay valid if compression
+				occurs */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	ut_ad(mtr_memo_contains(mtr,
+				dict_index_get_lock(btr_cur_get_index(cursor)),
+				MTR_MEMO_X_LOCK));
+	ut_ad(mtr_memo_contains(mtr, btr_cur_get_block(cursor),
+				MTR_MEMO_PAGE_X_FIX));
+
+	return(btr_cur_compress_recommendation(cursor, mtr)
+	       && btr_compress(cursor, mtr));
+}
+
+/***********************************************************
+Removes the record on which the tree cursor is positioned on a leaf page.
+It is assumed that the mtr has an x-latch on the page where the cursor is
+positioned, but no latch on the whole tree. */
+UNIV_INTERN
+ibool
+btr_cur_optimistic_delete(
+/*======================*/
+				/* out: TRUE if success, i.e., the page
+				did not become too empty */
+	btr_cur_t*	cursor,	/* in: cursor on leaf page, on the record to
+				delete; cursor stays valid: if deletion
+				succeeds, on function exit it points to the
+				successor of the deleted record */
+	mtr_t*		mtr)	/* in: mtr; if this function returns
+				TRUE on a leaf page of a secondary
+				index, the mtr must be committed
+				before latching any further pages */
+{
+	buf_block_t*	block;
+	rec_t*		rec;
+	mem_heap_t*	heap		= NULL;
+	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint*		offsets		= offsets_;
+	ibool		no_compress_needed;
+	rec_offs_init(offsets_);
+
+	ut_ad(mtr_memo_contains(mtr, btr_cur_get_block(cursor),
+				MTR_MEMO_PAGE_X_FIX));
+	/* This is intended only for leaf page deletions */
+
+	block = btr_cur_get_block(cursor);
+
+	ut_ad(page_is_leaf(buf_block_get_frame(block)));
+
+	rec = btr_cur_get_rec(cursor);
+	offsets = rec_get_offsets(rec, cursor->index, offsets,
+				  ULINT_UNDEFINED, &heap);
+
+	no_compress_needed = !rec_offs_any_extern(offsets)
+		&& btr_cur_can_delete_without_compress(
+			cursor, rec_offs_size(offsets), mtr);
+
+	if (no_compress_needed) {
+
+		page_t*		page	= buf_block_get_frame(block);
+		page_zip_des_t*	page_zip= buf_block_get_page_zip(block);
+		ulint		max_ins	= 0;
+
+		lock_update_delete(block, rec);
+
+		btr_search_update_hash_on_delete(cursor);
+
+		if (!page_zip) {
+			max_ins = page_get_max_insert_size_after_reorganize(
+				page, 1);
+		}
+#ifdef UNIV_ZIP_DEBUG
+		ut_a(!page_zip || page_zip_validate(page_zip, page));
+#endif /* UNIV_ZIP_DEBUG */
+		page_cur_delete_rec(btr_cur_get_page_cur(cursor),
+				    cursor->index, offsets, mtr);
+#ifdef UNIV_ZIP_DEBUG
+		ut_a(!page_zip || page_zip_validate(page_zip, page));
+#endif /* UNIV_ZIP_DEBUG */
+
+		if (dict_index_is_clust(cursor->index)
+		    || dict_index_is_ibuf(cursor->index)
+		    || !page_is_leaf(page)) {
+			/* The insert buffer does not handle
+			inserts to clustered indexes, to
+			non-leaf pages of secondary index B-trees,
+			or to the insert buffer. */
+		} else if (page_zip) {
+			ibuf_update_free_bits_zip(block, mtr);
+		} else {
+			ibuf_update_free_bits_low(block, max_ins, mtr);
+		}
+	}
+
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+
+	return(no_compress_needed);
+}
+
+/*****************************************************************
+Removes the record on which the tree cursor is positioned. Tries
+to compress the page if its fillfactor drops below a threshold
+or if it is the only page on the level. It is assumed that mtr holds
+an x-latch on the tree and on the cursor page. To avoid deadlocks,
+mtr must also own x-latches to brothers of page, if those brothers
+exist. */
+UNIV_INTERN
+ibool
+btr_cur_pessimistic_delete(
+/*=======================*/
+				/* out: TRUE if compression occurred */
+	ulint*		err,	/* out: DB_SUCCESS or DB_OUT_OF_FILE_SPACE;
+				the latter may occur because we may have
+				to update node pointers on upper levels,
+				and in the case of variable length keys
+				these may actually grow in size */
+	ibool		has_reserved_extents, /* in: TRUE if the
+				caller has already reserved enough free
+				extents so that he knows that the operation
+				will succeed */
+	btr_cur_t*	cursor,	/* in: cursor on the record to delete;
+				if compression does not occur, the cursor
+				stays valid: it points to successor of
+				deleted record on function exit */
+	enum trx_rb_ctx	rb_ctx,	/* in: rollback context */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	buf_block_t*	block;
+	page_t*		page;
+	page_zip_des_t*	page_zip;
+	dict_index_t*	index;
+	rec_t*		rec;
+	dtuple_t*	node_ptr;
+	ulint		n_extents	= 0;
+	ulint		n_reserved;
+	ibool		success;
+	ibool		ret		= FALSE;
+	ulint		level;
+	mem_heap_t*	heap;
+	ulint*		offsets;
+
+	block = btr_cur_get_block(cursor);
+	page = buf_block_get_frame(block);
+	index = btr_cur_get_index(cursor);
+
+	ut_ad(mtr_memo_contains(mtr, dict_index_get_lock(index),
+				MTR_MEMO_X_LOCK));
+	ut_ad(mtr_memo_contains(mtr, block, MTR_MEMO_PAGE_X_FIX));
+	if (!has_reserved_extents) {
+		/* First reserve enough free space for the file segments
+		of the index tree, so that the node pointer updates will
+		not fail because of lack of space */
+
+		n_extents = cursor->tree_height / 32 + 1;
+
+		success = fsp_reserve_free_extents(&n_reserved,
+						   index->space,
+						   n_extents,
+						   FSP_CLEANING, mtr);
+		if (!success) {
+			*err = DB_OUT_OF_FILE_SPACE;
+
+			return(FALSE);
+		}
+	}
+
+	heap = mem_heap_create(1024);
+	rec = btr_cur_get_rec(cursor);
+	page_zip = buf_block_get_page_zip(block);
+#ifdef UNIV_ZIP_DEBUG
+	ut_a(!page_zip || page_zip_validate(page_zip, page));
+#endif /* UNIV_ZIP_DEBUG */
+
+	offsets = rec_get_offsets(rec, index, NULL, ULINT_UNDEFINED, &heap);
+
+	if (rec_offs_any_extern(offsets)) {
+		btr_rec_free_externally_stored_fields(index,
+						      rec, offsets, page_zip,
+						      rb_ctx, mtr);
+#ifdef UNIV_ZIP_DEBUG
+		ut_a(!page_zip || page_zip_validate(page_zip, page));
+#endif /* UNIV_ZIP_DEBUG */
+	}
+
+	if (UNIV_UNLIKELY(page_get_n_recs(page) < 2)
+	    && UNIV_UNLIKELY(dict_index_get_page(index)
+			     != buf_block_get_page_no(block))) {
+
+		/* If there is only one record, drop the whole page in
+		btr_discard_page, if this is not the root page */
+
+		btr_discard_page(cursor, mtr);
+
+		*err = DB_SUCCESS;
+		ret = TRUE;
+
+		goto return_after_reservations;
+	}
+
+	lock_update_delete(block, rec);
+	level = btr_page_get_level(page, mtr);
+
+	if (level > 0
+	    && UNIV_UNLIKELY(rec == page_rec_get_next(
+				     page_get_infimum_rec(page)))) {
+
+		rec_t*	next_rec = page_rec_get_next(rec);
+
+		if (btr_page_get_prev(page, mtr) == FIL_NULL) {
+
+			/* If we delete the leftmost node pointer on a
+			non-leaf level, we must mark the new leftmost node
+			pointer as the predefined minimum record */
+
+			/* This will make page_zip_validate() fail until
+			page_cur_delete_rec() completes.  This is harmless,
+			because everything will take place within a single
+			mini-transaction and because writing to the redo log
+			is an atomic operation (performed by mtr_commit()). */
+			btr_set_min_rec_mark(next_rec, mtr);
+		} else {
+			/* Otherwise, if we delete the leftmost node pointer
+			on a page, we have to change the father node pointer
+			so that it is equal to the new leftmost node pointer
+			on the page */
+
+			btr_node_ptr_delete(index, block, mtr);
+
+			node_ptr = dict_index_build_node_ptr(
+				index, next_rec, buf_block_get_page_no(block),
+				heap, level);
+
+			btr_insert_on_non_leaf_level(index,
+						     level + 1, node_ptr, mtr);
+		}
+	}
+
+	btr_search_update_hash_on_delete(cursor);
+
+	page_cur_delete_rec(btr_cur_get_page_cur(cursor), index, offsets, mtr);
+#ifdef UNIV_ZIP_DEBUG
+	ut_a(!page_zip || page_zip_validate(page_zip, page));
+#endif /* UNIV_ZIP_DEBUG */
+
+	ut_ad(btr_check_node_ptr(index, block, mtr));
+
+	*err = DB_SUCCESS;
+
+return_after_reservations:
+	mem_heap_free(heap);
+
+	if (ret == FALSE) {
+		ret = btr_cur_compress_if_useful(cursor, mtr);
+	}
+
+	if (n_extents > 0) {
+		fil_space_release_free_extents(index->space, n_reserved);
+	}
+
+	return(ret);
+}
+
+/***********************************************************************
+Adds path information to the cursor for the current page, for which
+the binary search has been performed. */
+static
+void
+btr_cur_add_path_info(
+/*==================*/
+	btr_cur_t*	cursor,		/* in: cursor positioned on a page */
+	ulint		height,		/* in: height of the page in tree;
+					0 means leaf node */
+	ulint		root_height)	/* in: root node height in tree */
+{
+	btr_path_t*	slot;
+	rec_t*		rec;
+
+	ut_a(cursor->path_arr);
+
+	if (root_height >= BTR_PATH_ARRAY_N_SLOTS - 1) {
+		/* Do nothing; return empty path */
+
+		slot = cursor->path_arr;
+		slot->nth_rec = ULINT_UNDEFINED;
+
+		return;
+	}
+
+	if (height == 0) {
+		/* Mark end of slots for path */
+		slot = cursor->path_arr + root_height + 1;
+		slot->nth_rec = ULINT_UNDEFINED;
+	}
+
+	rec = btr_cur_get_rec(cursor);
+
+	slot = cursor->path_arr + (root_height - height);
+
+	slot->nth_rec = page_rec_get_n_recs_before(rec);
+	slot->n_recs = page_get_n_recs(page_align(rec));
+}
+
+/***********************************************************************
+Estimates the number of rows in a given index range. */
+UNIV_INTERN
+ib_int64_t
+btr_estimate_n_rows_in_range(
+/*=========================*/
+				/* out: estimated number of rows */
+	dict_index_t*	index,	/* in: index */
+	const dtuple_t*	tuple1,	/* in: range start, may also be empty tuple */
+	ulint		mode1,	/* in: search mode for range start */
+	const dtuple_t*	tuple2,	/* in: range end, may also be empty tuple */
+	ulint		mode2)	/* in: search mode for range end */
+{
+	btr_path_t	path1[BTR_PATH_ARRAY_N_SLOTS];
+	btr_path_t	path2[BTR_PATH_ARRAY_N_SLOTS];
+	btr_cur_t	cursor;
+	btr_path_t*	slot1;
+	btr_path_t*	slot2;
+	ibool		diverged;
+	ibool		diverged_lot;
+	ulint		divergence_level;
+	ib_int64_t	n_rows;
+	ulint		i;
+	mtr_t		mtr;
+
+	mtr_start(&mtr);
+
+	cursor.path_arr = path1;
+
+	if (dtuple_get_n_fields(tuple1) > 0) {
+
+		btr_cur_search_to_nth_level(index, 0, tuple1, mode1,
+					    BTR_SEARCH_LEAF | BTR_ESTIMATE,
+					    &cursor, 0, &mtr);
+	} else {
+		btr_cur_open_at_index_side(TRUE, index,
+					   BTR_SEARCH_LEAF | BTR_ESTIMATE,
+					   &cursor, &mtr);
+	}
+
+	mtr_commit(&mtr);
+
+	mtr_start(&mtr);
+
+	cursor.path_arr = path2;
+
+	if (dtuple_get_n_fields(tuple2) > 0) {
+
+		btr_cur_search_to_nth_level(index, 0, tuple2, mode2,
+					    BTR_SEARCH_LEAF | BTR_ESTIMATE,
+					    &cursor, 0, &mtr);
+	} else {
+		btr_cur_open_at_index_side(FALSE, index,
+					   BTR_SEARCH_LEAF | BTR_ESTIMATE,
+					   &cursor, &mtr);
+	}
+
+	mtr_commit(&mtr);
+
+	/* We have the path information for the range in path1 and path2 */
+
+	n_rows = 1;
+	diverged = FALSE;	    /* This becomes true when the path is not
+				    the same any more */
+	diverged_lot = FALSE;	    /* This becomes true when the paths are
+				    not the same or adjacent any more */
+	divergence_level = 1000000; /* This is the level where paths diverged
+				    a lot */
+	for (i = 0; ; i++) {
+		ut_ad(i < BTR_PATH_ARRAY_N_SLOTS);
+
+		slot1 = path1 + i;
+		slot2 = path2 + i;
+
+		if (slot1->nth_rec == ULINT_UNDEFINED
+		    || slot2->nth_rec == ULINT_UNDEFINED) {
+
+			if (i > divergence_level + 1) {
+				/* In trees whose height is > 1 our algorithm
+				tends to underestimate: multiply the estimate
+				by 2: */
+
+				n_rows = n_rows * 2;
+			}
+
+			/* Do not estimate the number of rows in the range
+			to over 1 / 2 of the estimated rows in the whole
+			table */
+
+			if (n_rows > index->table->stat_n_rows / 2) {
+				n_rows = index->table->stat_n_rows / 2;
+
+				/* If there are just 0 or 1 rows in the table,
+				then we estimate all rows are in the range */
+
+				if (n_rows == 0) {
+					n_rows = index->table->stat_n_rows;
+				}
+			}
+
+			return(n_rows);
+		}
+
+		if (!diverged && slot1->nth_rec != slot2->nth_rec) {
+
+			diverged = TRUE;
+
+			if (slot1->nth_rec < slot2->nth_rec) {
+				n_rows = slot2->nth_rec - slot1->nth_rec;
+
+				if (n_rows > 1) {
+					diverged_lot = TRUE;
+					divergence_level = i;
+				}
+			} else {
+				/* Maybe the tree has changed between
+				searches */
+
+				return(10);
+			}
+
+		} else if (diverged && !diverged_lot) {
+
+			if (slot1->nth_rec < slot1->n_recs
+			    || slot2->nth_rec > 1) {
+
+				diverged_lot = TRUE;
+				divergence_level = i;
+
+				n_rows = 0;
+
+				if (slot1->nth_rec < slot1->n_recs) {
+					n_rows += slot1->n_recs
+						- slot1->nth_rec;
+				}
+
+				if (slot2->nth_rec > 1) {
+					n_rows += slot2->nth_rec - 1;
+				}
+			}
+		} else if (diverged_lot) {
+
+			n_rows = (n_rows * (slot1->n_recs + slot2->n_recs))
+				/ 2;
+		}
+	}
+}
+
+/***********************************************************************
+Estimates the number of different key values in a given index, for
+each n-column prefix of the index where n <= dict_index_get_n_unique(index).
+The estimates are stored in the array index->stat_n_diff_key_vals. */
+UNIV_INTERN
+void
+btr_estimate_number_of_different_key_vals(
+/*======================================*/
+	dict_index_t*	index)	/* in: index */
+{
+	btr_cur_t	cursor;
+	page_t*		page;
+	rec_t*		rec;
+	ulint		n_cols;
+	ulint		matched_fields;
+	ulint		matched_bytes;
+	ib_int64_t*	n_diff;
+	ullint		n_sample_pages; /* number of pages to sample */
+	ulint		not_empty_flag	= 0;
+	ulint		total_external_size = 0;
+	ulint		i;
+	ulint		j;
+	ullint		add_on;
+	mtr_t		mtr;
+	mem_heap_t*	heap		= NULL;
+	ulint		offsets_rec_[REC_OFFS_NORMAL_SIZE];
+	ulint		offsets_next_rec_[REC_OFFS_NORMAL_SIZE];
+	ulint*		offsets_rec	= offsets_rec_;
+	ulint*		offsets_next_rec= offsets_next_rec_;
+	rec_offs_init(offsets_rec_);
+	rec_offs_init(offsets_next_rec_);
+
+	n_cols = dict_index_get_n_unique(index);
+
+	n_diff = mem_zalloc((n_cols + 1) * sizeof(ib_int64_t));
+
+	/* It makes no sense to test more pages than are contained
+	in the index, thus we lower the number if it is too high */
+	if (srv_stats_sample_pages > index->stat_index_size) {
+		if (index->stat_index_size > 0) {
+			n_sample_pages = index->stat_index_size;
+		} else {
+			n_sample_pages = 1;
+		}
+	} else {
+		n_sample_pages = srv_stats_sample_pages;
+	}
+
+	/* We sample some pages in the index to get an estimate */
+
+	for (i = 0; i < n_sample_pages; i++) {
+		rec_t*	supremum;
+		mtr_start(&mtr);
+
+		btr_cur_open_at_rnd_pos(index, BTR_SEARCH_LEAF, &cursor, &mtr);
+
+		/* Count the number of different key values for each prefix of
+		the key on this index page. If the prefix does not determine
+		the index record uniquely in the B-tree, then we subtract one
+		because otherwise our algorithm would give a wrong estimate
+		for an index where there is just one key value. */
+
+		page = btr_cur_get_page(&cursor);
+
+		supremum = page_get_supremum_rec(page);
+		rec = page_rec_get_next(page_get_infimum_rec(page));
+
+		if (rec != supremum) {
+			not_empty_flag = 1;
+			offsets_rec = rec_get_offsets(rec, index, offsets_rec,
+						      ULINT_UNDEFINED, &heap);
+		}
+
+		while (rec != supremum) {
+			rec_t*	next_rec = page_rec_get_next(rec);
+			if (next_rec == supremum) {
+				break;
+			}
+
+			matched_fields = 0;
+			matched_bytes = 0;
+			offsets_next_rec = rec_get_offsets(next_rec, index,
+							   offsets_next_rec,
+							   n_cols, &heap);
+
+			cmp_rec_rec_with_match(rec, next_rec,
+					       offsets_rec, offsets_next_rec,
+					       index, &matched_fields,
+					       &matched_bytes);
+
+			for (j = matched_fields + 1; j <= n_cols; j++) {
+				/* We add one if this index record has
+				a different prefix from the previous */
+
+				n_diff[j]++;
+			}
+
+			total_external_size
+				+= btr_rec_get_externally_stored_len(
+					rec, offsets_rec);
+
+			rec = next_rec;
+			/* Initialize offsets_rec for the next round
+			and assign the old offsets_rec buffer to
+			offsets_next_rec. */
+			{
+				ulint*	offsets_tmp = offsets_rec;
+				offsets_rec = offsets_next_rec;
+				offsets_next_rec = offsets_tmp;
+			}
+		}
+
+
+		if (n_cols == dict_index_get_n_unique_in_tree(index)) {
+
+			/* If there is more than one leaf page in the tree,
+			we add one because we know that the first record
+			on the page certainly had a different prefix than the
+			last record on the previous index page in the
+			alphabetical order. Before this fix, if there was
+			just one big record on each clustered index page, the
+			algorithm grossly underestimated the number of rows
+			in the table. */
+
+			if (btr_page_get_prev(page, &mtr) != FIL_NULL
+			    || btr_page_get_next(page, &mtr) != FIL_NULL) {
+
+				n_diff[n_cols]++;
+			}
+		}
+
+		offsets_rec = rec_get_offsets(rec, index, offsets_rec,
+					      ULINT_UNDEFINED, &heap);
+		total_external_size += btr_rec_get_externally_stored_len(
+			rec, offsets_rec);
+		mtr_commit(&mtr);
+	}
+
+	/* If we saw k borders between different key values on
+	n_sample_pages leaf pages, we can estimate how many
+	there will be in index->stat_n_leaf_pages */
+
+	/* We must take into account that our sample actually represents
+	also the pages used for external storage of fields (those pages are
+	included in index->stat_n_leaf_pages) */
+
+	for (j = 0; j <= n_cols; j++) {
+		index->stat_n_diff_key_vals[j]
+			= ((n_diff[j]
+			    * (ib_int64_t)index->stat_n_leaf_pages
+			    + n_sample_pages - 1
+			    + total_external_size
+			    + not_empty_flag)
+			   / (n_sample_pages
+			      + total_external_size));
+
+		/* If the tree is small, smaller than
+		10 * n_sample_pages + total_external_size, then
+		the above estimate is ok. For bigger trees it is common that we
+		do not see any borders between key values in the few pages
+		we pick. But still there may be n_sample_pages
+		different key values, or even more. Let us try to approximate
+		that: */
+
+		add_on = index->stat_n_leaf_pages
+			/ (10 * (n_sample_pages
+				 + total_external_size));
+
+		if (add_on > n_sample_pages) {
+			add_on = n_sample_pages;
+		}
+
+		index->stat_n_diff_key_vals[j] += add_on;
+	}
+
+	mem_free(n_diff);
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+}
+
+/*================== EXTERNAL STORAGE OF BIG FIELDS ===================*/
+
+/***************************************************************
+Gets the externally stored size of a record, in units of a database page. */
+static
+ulint
+btr_rec_get_externally_stored_len(
+/*==============================*/
+				/* out: externally stored part,
+				in units of a database page */
+	rec_t*		rec,	/* in: record */
+	const ulint*	offsets)/* in: array returned by rec_get_offsets() */
+{
+	ulint	n_fields;
+	byte*	data;
+	ulint	local_len;
+	ulint	extern_len;
+	ulint	total_extern_len = 0;
+	ulint	i;
+
+	ut_ad(!rec_offs_comp(offsets) || !rec_get_node_ptr_flag(rec));
+	n_fields = rec_offs_n_fields(offsets);
+
+	for (i = 0; i < n_fields; i++) {
+		if (rec_offs_nth_extern(offsets, i)) {
+
+			data = rec_get_nth_field(rec, offsets, i, &local_len);
+
+			local_len -= BTR_EXTERN_FIELD_REF_SIZE;
+
+			extern_len = mach_read_from_4(data + local_len
+						      + BTR_EXTERN_LEN + 4);
+
+			total_extern_len += ut_calc_align(extern_len,
+							  UNIV_PAGE_SIZE);
+		}
+	}
+
+	return(total_extern_len / UNIV_PAGE_SIZE);
+}
+
+/***********************************************************************
+Sets the ownership bit of an externally stored field in a record. */
+static
+void
+btr_cur_set_ownership_of_extern_field(
+/*==================================*/
+	page_zip_des_t*	page_zip,/* in/out: compressed page whose uncompressed
+				part will be updated, or NULL */
+	rec_t*		rec,	/* in/out: clustered index record */
+	dict_index_t*	index,	/* in: index of the page */
+	const ulint*	offsets,/* in: array returned by rec_get_offsets() */
+	ulint		i,	/* in: field number */
+	ibool		val,	/* in: value to set */
+	mtr_t*		mtr)	/* in: mtr, or NULL if not logged */
+{
+	byte*	data;
+	ulint	local_len;
+	ulint	byte_val;
+
+	data = rec_get_nth_field(rec, offsets, i, &local_len);
+
+	ut_a(local_len >= BTR_EXTERN_FIELD_REF_SIZE);
+
+	local_len -= BTR_EXTERN_FIELD_REF_SIZE;
+
+	byte_val = mach_read_from_1(data + local_len + BTR_EXTERN_LEN);
+
+	if (val) {
+		byte_val = byte_val & (~BTR_EXTERN_OWNER_FLAG);
+	} else {
+		byte_val = byte_val | BTR_EXTERN_OWNER_FLAG;
+	}
+
+	if (UNIV_LIKELY_NULL(page_zip)) {
+		mach_write_to_1(data + local_len + BTR_EXTERN_LEN, byte_val);
+		page_zip_write_blob_ptr(page_zip, rec, index, offsets, i, mtr);
+	} else if (UNIV_LIKELY(mtr != NULL)) {
+
+		mlog_write_ulint(data + local_len + BTR_EXTERN_LEN, byte_val,
+				 MLOG_1BYTE, mtr);
+	} else {
+		mach_write_to_1(data + local_len + BTR_EXTERN_LEN, byte_val);
+	}
+}
+
+/***********************************************************************
+Marks not updated extern fields as not-owned by this record. The ownership
+is transferred to the updated record which is inserted elsewhere in the
+index tree. In purge only the owner of externally stored field is allowed
+to free the field. */
+UNIV_INTERN
+void
+btr_cur_mark_extern_inherited_fields(
+/*=================================*/
+	page_zip_des_t*	page_zip,/* in/out: compressed page whose uncompressed
+				part will be updated, or NULL */
+	rec_t*		rec,	/* in/out: record in a clustered index */
+	dict_index_t*	index,	/* in: index of the page */
+	const ulint*	offsets,/* in: array returned by rec_get_offsets() */
+	const upd_t*	update,	/* in: update vector */
+	mtr_t*		mtr)	/* in: mtr, or NULL if not logged */
+{
+	ulint	n;
+	ulint	j;
+	ulint	i;
+
+	ut_ad(rec_offs_validate(rec, NULL, offsets));
+	ut_ad(!rec_offs_comp(offsets) || !rec_get_node_ptr_flag(rec));
+
+	if (!rec_offs_any_extern(offsets)) {
+
+		return;
+	}
+
+	n = rec_offs_n_fields(offsets);
+
+	for (i = 0; i < n; i++) {
+		if (rec_offs_nth_extern(offsets, i)) {
+
+			/* Check it is not in updated fields */
+
+			if (update) {
+				for (j = 0; j < upd_get_n_fields(update);
+				     j++) {
+					if (upd_get_nth_field(update, j)
+					    ->field_no == i) {
+
+						goto updated;
+					}
+				}
+			}
+
+			btr_cur_set_ownership_of_extern_field(
+				page_zip, rec, index, offsets, i, FALSE, mtr);
+updated:
+			;
+		}
+	}
+}
+
+/***********************************************************************
+The complement of the previous function: in an update entry may inherit
+some externally stored fields from a record. We must mark them as inherited
+in entry, so that they are not freed in a rollback. */
+UNIV_INTERN
+void
+btr_cur_mark_dtuple_inherited_extern(
+/*=================================*/
+	dtuple_t*	entry,		/* in/out: updated entry to be
+					inserted to clustered index */
+	const upd_t*	update)		/* in: update vector */
+{
+	ulint		i;
+
+	for (i = 0; i < dtuple_get_n_fields(entry); i++) {
+
+		dfield_t*	dfield = dtuple_get_nth_field(entry, i);
+		byte*		data;
+		ulint		len;
+		ulint		j;
+
+		if (!dfield_is_ext(dfield)) {
+			continue;
+		}
+
+		/* Check if it is in updated fields */
+
+		for (j = 0; j < upd_get_n_fields(update); j++) {
+			if (upd_get_nth_field(update, j)->field_no == i) {
+
+				goto is_updated;
+			}
+		}
+
+		data = dfield_get_data(dfield);
+		len = dfield_get_len(dfield);
+		data[len - BTR_EXTERN_FIELD_REF_SIZE + BTR_EXTERN_LEN]
+			|= BTR_EXTERN_INHERITED_FLAG;
+
+is_updated:
+		;
+	}
+}
+
+/***********************************************************************
+Marks all extern fields in a record as owned by the record. This function
+should be called if the delete mark of a record is removed: a not delete
+marked record always owns all its extern fields. */
+static
+void
+btr_cur_unmark_extern_fields(
+/*=========================*/
+	page_zip_des_t*	page_zip,/* in/out: compressed page whose uncompressed
+				part will be updated, or NULL */
+	rec_t*		rec,	/* in/out: record in a clustered index */
+	dict_index_t*	index,	/* in: index of the page */
+	const ulint*	offsets,/* in: array returned by rec_get_offsets() */
+	mtr_t*		mtr)	/* in: mtr, or NULL if not logged */
+{
+	ulint	n;
+	ulint	i;
+
+	ut_ad(!rec_offs_comp(offsets) || !rec_get_node_ptr_flag(rec));
+	n = rec_offs_n_fields(offsets);
+
+	if (!rec_offs_any_extern(offsets)) {
+
+		return;
+	}
+
+	for (i = 0; i < n; i++) {
+		if (rec_offs_nth_extern(offsets, i)) {
+
+			btr_cur_set_ownership_of_extern_field(
+				page_zip, rec, index, offsets, i, TRUE, mtr);
+		}
+	}
+}
+
+/***********************************************************************
+Marks all extern fields in a dtuple as owned by the record. */
+UNIV_INTERN
+void
+btr_cur_unmark_dtuple_extern_fields(
+/*================================*/
+	dtuple_t*	entry)		/* in/out: clustered index entry */
+{
+	ulint	i;
+
+	for (i = 0; i < dtuple_get_n_fields(entry); i++) {
+		dfield_t* dfield = dtuple_get_nth_field(entry, i);
+
+		if (dfield_is_ext(dfield)) {
+			byte*	data = dfield_get_data(dfield);
+			ulint	len = dfield_get_len(dfield);
+
+			data[len - BTR_EXTERN_FIELD_REF_SIZE + BTR_EXTERN_LEN]
+				&= ~BTR_EXTERN_OWNER_FLAG;
+		}
+	}
+}
+
+/***********************************************************************
+Flags the data tuple fields that are marked as extern storage in the
+update vector.  We use this function to remember which fields we must
+mark as extern storage in a record inserted for an update. */
+UNIV_INTERN
+ulint
+btr_push_update_extern_fields(
+/*==========================*/
+				/* out: number of flagged external columns */
+	dtuple_t*	tuple,	/* in/out: data tuple */
+	const upd_t*	update,	/* in: update vector */
+	mem_heap_t*	heap)	/* in: memory heap */
+{
+	ulint			n_pushed	= 0;
+	ulint			n;
+	const upd_field_t*	uf;
+
+	ut_ad(tuple);
+	ut_ad(update);
+
+	uf = update->fields;
+	n = upd_get_n_fields(update);
+
+	for (; n--; uf++) {
+		if (dfield_is_ext(&uf->new_val)) {
+			dfield_t*	field
+				= dtuple_get_nth_field(tuple, uf->field_no);
+
+			if (!dfield_is_ext(field)) {
+				dfield_set_ext(field);
+				n_pushed++;
+			}
+
+			switch (uf->orig_len) {
+				byte*	data;
+				ulint	len;
+				byte*	buf;
+			case 0:
+				break;
+			case BTR_EXTERN_FIELD_REF_SIZE:
+				/* Restore the original locally stored
+				part of the column.  In the undo log,
+				InnoDB writes a longer prefix of externally
+				stored columns, so that column prefixes
+				in secondary indexes can be reconstructed. */
+				dfield_set_data(field, (byte*) dfield_get_data(field)
+						+ dfield_get_len(field)
+						- BTR_EXTERN_FIELD_REF_SIZE,
+						BTR_EXTERN_FIELD_REF_SIZE);
+				dfield_set_ext(field);
+				break;
+			default:
+				/* Reconstruct the original locally
+				stored part of the column.  The data
+				will have to be copied. */
+				ut_a(uf->orig_len > BTR_EXTERN_FIELD_REF_SIZE);
+
+				data = dfield_get_data(field);
+				len = dfield_get_len(field);
+
+				buf = mem_heap_alloc(heap, uf->orig_len);
+				/* Copy the locally stored prefix. */
+				memcpy(buf, data,
+				       uf->orig_len
+				       - BTR_EXTERN_FIELD_REF_SIZE);
+				/* Copy the BLOB pointer. */
+				memcpy(buf + uf->orig_len
+				       - BTR_EXTERN_FIELD_REF_SIZE,
+				       data + len - BTR_EXTERN_FIELD_REF_SIZE,
+				       BTR_EXTERN_FIELD_REF_SIZE);
+
+				dfield_set_data(field, buf, uf->orig_len);
+				dfield_set_ext(field);
+			}
+		}
+	}
+
+	return(n_pushed);
+}
+
+/***********************************************************************
+Returns the length of a BLOB part stored on the header page. */
+static
+ulint
+btr_blob_get_part_len(
+/*==================*/
+					/* out: part length */
+	const byte*	blob_header)	/* in: blob header */
+{
+	return(mach_read_from_4(blob_header + BTR_BLOB_HDR_PART_LEN));
+}
+
+/***********************************************************************
+Returns the page number where the next BLOB part is stored. */
+static
+ulint
+btr_blob_get_next_page_no(
+/*======================*/
+					/* out: page number or FIL_NULL if
+					no more pages */
+	const byte*	blob_header)	/* in: blob header */
+{
+	return(mach_read_from_4(blob_header + BTR_BLOB_HDR_NEXT_PAGE_NO));
+}
+
+/***********************************************************************
+Deallocate a buffer block that was reserved for a BLOB part. */
+static
+void
+btr_blob_free(
+/*==========*/
+	buf_block_t*	block,	/* in: buffer block */
+	ibool		all,	/* in: TRUE=remove also the compressed page
+				if there is one */
+	mtr_t*		mtr)	/* in: mini-transaction to commit */
+{
+	ulint	space	= buf_block_get_space(block);
+	ulint	page_no	= buf_block_get_page_no(block);
+
+	ut_ad(mtr_memo_contains(mtr, block, MTR_MEMO_PAGE_X_FIX));
+
+	mtr_commit(mtr);
+
+	buf_pool_mutex_enter();
+	mutex_enter(&block->mutex);
+
+	/* Only free the block if it is still allocated to
+	the same file page. */
+
+	if (buf_block_get_state(block)
+	    == BUF_BLOCK_FILE_PAGE
+	    && buf_block_get_space(block) == space
+	    && buf_block_get_page_no(block) == page_no) {
+
+		if (buf_LRU_free_block(&block->page, all, NULL)
+		    != BUF_LRU_FREED
+		    && all && block->page.zip.data) {
+			/* Attempt to deallocate the uncompressed page
+			if the whole block cannot be deallocted. */
+
+			buf_LRU_free_block(&block->page, FALSE, NULL);
+		}
+	}
+
+	buf_pool_mutex_exit();
+	mutex_exit(&block->mutex);
+}
+
+/***********************************************************************
+Stores the fields in big_rec_vec to the tablespace and puts pointers to
+them in rec.  The extern flags in rec will have to be set beforehand.
+The fields are stored on pages allocated from leaf node
+file segment of the index tree. */
+UNIV_INTERN
+ulint
+btr_store_big_rec_extern_fields(
+/*============================*/
+					/* out: DB_SUCCESS or error */
+	dict_index_t*	index,		/* in: index of rec; the index tree
+					MUST be X-latched */
+	buf_block_t*	rec_block,	/* in/out: block containing rec */
+	rec_t*		rec,		/* in/out: record */
+	const ulint*	offsets,	/* in: rec_get_offsets(rec, index);
+					the "external storage" flags in offsets
+					will not correspond to rec when
+					this function returns */
+	big_rec_t*	big_rec_vec,	/* in: vector containing fields
+					to be stored externally */
+	mtr_t*		local_mtr __attribute__((unused))) /* in: mtr
+					containing the latch to rec and to the
+					tree */
+{
+	ulint	rec_page_no;
+	byte*	field_ref;
+	ulint	extern_len;
+	ulint	store_len;
+	ulint	page_no;
+	ulint	space_id;
+	ulint	zip_size;
+	ulint	prev_page_no;
+	ulint	hint_page_no;
+	ulint	i;
+	mtr_t	mtr;
+	mem_heap_t* heap = NULL;
+	page_zip_des_t*	page_zip;
+	z_stream c_stream;
+
+	ut_ad(rec_offs_validate(rec, index, offsets));
+	ut_ad(mtr_memo_contains(local_mtr, dict_index_get_lock(index),
+				MTR_MEMO_X_LOCK));
+	ut_ad(mtr_memo_contains(local_mtr, rec_block, MTR_MEMO_PAGE_X_FIX));
+	ut_ad(buf_block_get_frame(rec_block) == page_align(rec));
+	ut_a(dict_index_is_clust(index));
+
+	page_zip = buf_block_get_page_zip(rec_block);
+	ut_a(dict_table_zip_size(index->table)
+	     == buf_block_get_zip_size(rec_block));
+
+	space_id = buf_block_get_space(rec_block);
+	zip_size = buf_block_get_zip_size(rec_block);
+	rec_page_no = buf_block_get_page_no(rec_block);
+	ut_a(fil_page_get_type(page_align(rec)) == FIL_PAGE_INDEX);
+
+	if (UNIV_LIKELY_NULL(page_zip)) {
+		int	err;
+
+		/* Zlib deflate needs 128 kilobytes for the default
+		window size, plus 512 << memLevel, plus a few
+		kilobytes for small objects.  We use reduced memLevel
+		to limit the memory consumption, and preallocate the
+		heap, hoping to avoid memory fragmentation. */
+		heap = mem_heap_create(250000);
+		page_zip_set_alloc(&c_stream, heap);
+
+		err = deflateInit2(&c_stream, Z_DEFAULT_COMPRESSION,
+				   Z_DEFLATED, 15, 7, Z_DEFAULT_STRATEGY);
+		ut_a(err == Z_OK);
+	}
+
+	/* We have to create a file segment to the tablespace
+	for each field and put the pointer to the field in rec */
+
+	for (i = 0; i < big_rec_vec->n_fields; i++) {
+		ut_ad(rec_offs_nth_extern(offsets,
+					  big_rec_vec->fields[i].field_no));
+		{
+			ulint	local_len;
+			field_ref = rec_get_nth_field(
+				rec, offsets, big_rec_vec->fields[i].field_no,
+				&local_len);
+			ut_a(local_len >= BTR_EXTERN_FIELD_REF_SIZE);
+			local_len -= BTR_EXTERN_FIELD_REF_SIZE;
+			field_ref += local_len;
+		}
+		extern_len = big_rec_vec->fields[i].len;
+
+		ut_a(extern_len > 0);
+
+		prev_page_no = FIL_NULL;
+
+		if (UNIV_LIKELY_NULL(page_zip)) {
+			int	err = deflateReset(&c_stream);
+			ut_a(err == Z_OK);
+
+			c_stream.next_in = (void*) big_rec_vec->fields[i].data;
+			c_stream.avail_in = extern_len;
+		}
+
+		for (;;) {
+			buf_block_t*	block;
+			page_t*		page;
+
+			mtr_start(&mtr);
+
+			if (prev_page_no == FIL_NULL) {
+				hint_page_no = 1 + rec_page_no;
+			} else {
+				hint_page_no = prev_page_no + 1;
+			}
+
+			block = btr_page_alloc(index, hint_page_no,
+					       FSP_NO_DIR, 0, &mtr);
+			if (UNIV_UNLIKELY(block == NULL)) {
+
+				mtr_commit(&mtr);
+
+				if (UNIV_LIKELY_NULL(page_zip)) {
+					deflateEnd(&c_stream);
+					mem_heap_free(heap);
+				}
+
+				return(DB_OUT_OF_FILE_SPACE);
+			}
+
+			page_no = buf_block_get_page_no(block);
+			page = buf_block_get_frame(block);
+
+			if (prev_page_no != FIL_NULL) {
+				buf_block_t*	prev_block;
+				page_t*		prev_page;
+
+				prev_block = buf_page_get(space_id, zip_size,
+							  prev_page_no,
+							  RW_X_LATCH, &mtr);
+				buf_block_dbg_add_level(prev_block,
+							SYNC_EXTERN_STORAGE);
+				prev_page = buf_block_get_frame(prev_block);
+
+				if (UNIV_LIKELY_NULL(page_zip)) {
+					mlog_write_ulint(
+						prev_page + FIL_PAGE_NEXT,
+						page_no, MLOG_4BYTES, &mtr);
+					memcpy(buf_block_get_page_zip(
+						       prev_block)
+					       ->data + FIL_PAGE_NEXT,
+					       prev_page + FIL_PAGE_NEXT, 4);
+				} else {
+					mlog_write_ulint(
+						prev_page + FIL_PAGE_DATA
+						+ BTR_BLOB_HDR_NEXT_PAGE_NO,
+						page_no, MLOG_4BYTES, &mtr);
+				}
+
+			}
+
+			if (UNIV_LIKELY_NULL(page_zip)) {
+				int		err;
+				page_zip_des_t*	blob_page_zip;
+
+				mach_write_to_2(page + FIL_PAGE_TYPE,
+						prev_page_no == FIL_NULL
+						? FIL_PAGE_TYPE_ZBLOB
+						: FIL_PAGE_TYPE_ZBLOB2);
+
+				c_stream.next_out = page
+					+ FIL_PAGE_DATA;
+				c_stream.avail_out
+					= page_zip_get_size(page_zip)
+					- FIL_PAGE_DATA;
+
+				err = deflate(&c_stream, Z_FINISH);
+				ut_a(err == Z_OK || err == Z_STREAM_END);
+				ut_a(err == Z_STREAM_END
+				     || c_stream.avail_out == 0);
+
+				/* Write the "next BLOB page" pointer */
+				mlog_write_ulint(page + FIL_PAGE_NEXT,
+						 FIL_NULL, MLOG_4BYTES, &mtr);
+				/* Initialize the unused "prev page" pointer */
+				mlog_write_ulint(page + FIL_PAGE_PREV,
+						 FIL_NULL, MLOG_4BYTES, &mtr);
+				/* Write a back pointer to the record
+				into the otherwise unused area.  This
+				information could be useful in
+				debugging.  Later, we might want to
+				implement the possibility to relocate
+				BLOB pages.  Then, we would need to be
+				able to adjust the BLOB pointer in the
+				record.  We do not store the heap
+				number of the record, because it can
+				change in page_zip_reorganize() or
+				btr_page_reorganize().  However, also
+				the page number of the record may
+				change when B-tree nodes are split or
+				merged. */
+				mlog_write_ulint(page
+						 + FIL_PAGE_FILE_FLUSH_LSN,
+						 space_id,
+						 MLOG_4BYTES, &mtr);
+				mlog_write_ulint(page
+						 + FIL_PAGE_FILE_FLUSH_LSN + 4,
+						 rec_page_no,
+						 MLOG_4BYTES, &mtr);
+
+				/* Zero out the unused part of the page. */
+				memset(page + page_zip_get_size(page_zip)
+				       - c_stream.avail_out,
+				       0, c_stream.avail_out);
+				mlog_log_string(page + FIL_PAGE_TYPE,
+						page_zip_get_size(page_zip)
+						- FIL_PAGE_TYPE,
+						&mtr);
+				/* Copy the page to compressed storage,
+				because it will be flushed to disk
+				from there. */
+				blob_page_zip = buf_block_get_page_zip(block);
+				ut_ad(blob_page_zip);
+				ut_ad(page_zip_get_size(blob_page_zip)
+				      == page_zip_get_size(page_zip));
+				memcpy(blob_page_zip->data, page,
+				       page_zip_get_size(page_zip));
+
+				if (err == Z_OK && prev_page_no != FIL_NULL) {
+
+					goto next_zip_page;
+				}
+
+				rec_block = buf_page_get(space_id, zip_size,
+							 rec_page_no,
+							 RW_X_LATCH, &mtr);
+				buf_block_dbg_add_level(rec_block,
+							SYNC_NO_ORDER_CHECK);
+
+				if (err == Z_STREAM_END) {
+					mach_write_to_4(field_ref
+							+ BTR_EXTERN_LEN, 0);
+					mach_write_to_4(field_ref
+							+ BTR_EXTERN_LEN + 4,
+							c_stream.total_in);
+				} else {
+					memset(field_ref + BTR_EXTERN_LEN,
+					       0, 8);
+				}
+
+				if (prev_page_no == FIL_NULL) {
+					mach_write_to_4(field_ref
+							+ BTR_EXTERN_SPACE_ID,
+							space_id);
+
+					mach_write_to_4(field_ref
+							+ BTR_EXTERN_PAGE_NO,
+							page_no);
+
+					mach_write_to_4(field_ref
+							+ BTR_EXTERN_OFFSET,
+							FIL_PAGE_NEXT);
+				}
+
+				page_zip_write_blob_ptr(
+					page_zip, rec, index, offsets,
+					big_rec_vec->fields[i].field_no, &mtr);
+
+next_zip_page:
+				prev_page_no = page_no;
+
+				/* Commit mtr and release the
+				uncompressed page frame to save memory. */
+				btr_blob_free(block, FALSE, &mtr);
+
+				if (err == Z_STREAM_END) {
+					break;
+				}
+			} else {
+				mlog_write_ulint(page + FIL_PAGE_TYPE,
+						 FIL_PAGE_TYPE_BLOB,
+						 MLOG_2BYTES, &mtr);
+
+				if (extern_len > (UNIV_PAGE_SIZE
+						  - FIL_PAGE_DATA
+						  - BTR_BLOB_HDR_SIZE
+						  - FIL_PAGE_DATA_END)) {
+					store_len = UNIV_PAGE_SIZE
+						- FIL_PAGE_DATA
+						- BTR_BLOB_HDR_SIZE
+						- FIL_PAGE_DATA_END;
+				} else {
+					store_len = extern_len;
+				}
+
+				mlog_write_string(page + FIL_PAGE_DATA
+						  + BTR_BLOB_HDR_SIZE,
+						  (const byte*)
+						  big_rec_vec->fields[i].data
+						  + big_rec_vec->fields[i].len
+						  - extern_len,
+						  store_len, &mtr);
+				mlog_write_ulint(page + FIL_PAGE_DATA
+						 + BTR_BLOB_HDR_PART_LEN,
+						 store_len, MLOG_4BYTES, &mtr);
+				mlog_write_ulint(page + FIL_PAGE_DATA
+						 + BTR_BLOB_HDR_NEXT_PAGE_NO,
+						 FIL_NULL, MLOG_4BYTES, &mtr);
+
+				extern_len -= store_len;
+
+				rec_block = buf_page_get(space_id, zip_size,
+							 rec_page_no,
+							 RW_X_LATCH, &mtr);
+				buf_block_dbg_add_level(rec_block,
+							SYNC_NO_ORDER_CHECK);
+
+				mlog_write_ulint(field_ref + BTR_EXTERN_LEN, 0,
+						 MLOG_4BYTES, &mtr);
+				mlog_write_ulint(field_ref
+						 + BTR_EXTERN_LEN + 4,
+						 big_rec_vec->fields[i].len
+						 - extern_len,
+						 MLOG_4BYTES, &mtr);
+
+				if (prev_page_no == FIL_NULL) {
+					mlog_write_ulint(field_ref
+							 + BTR_EXTERN_SPACE_ID,
+							 space_id,
+							 MLOG_4BYTES, &mtr);
+
+					mlog_write_ulint(field_ref
+							 + BTR_EXTERN_PAGE_NO,
+							 page_no,
+							 MLOG_4BYTES, &mtr);
+
+					mlog_write_ulint(field_ref
+							 + BTR_EXTERN_OFFSET,
+							 FIL_PAGE_DATA,
+							 MLOG_4BYTES, &mtr);
+				}
+
+				prev_page_no = page_no;
+
+				mtr_commit(&mtr);
+
+				if (extern_len == 0) {
+					break;
+				}
+			}
+		}
+	}
+
+	if (UNIV_LIKELY_NULL(page_zip)) {
+		deflateEnd(&c_stream);
+		mem_heap_free(heap);
+	}
+
+	return(DB_SUCCESS);
+}
+
+/***********************************************************************
+Check the FIL_PAGE_TYPE on an uncompressed BLOB page. */
+static
+void
+btr_check_blob_fil_page_type(
+/*=========================*/
+	ulint		space_id,	/* in: space id */
+	ulint		page_no,	/* in: page number */
+	const page_t*	page,		/* in: page */
+	ibool		read)		/* in: TRUE=read, FALSE=purge */
+{
+	ulint	type = fil_page_get_type(page);
+
+	ut_a(space_id == page_get_space_id(page));
+	ut_a(page_no == page_get_page_no(page));
+
+	if (UNIV_UNLIKELY(type != FIL_PAGE_TYPE_BLOB)) {
+		ulint	flags = fil_space_get_flags(space_id);
+
+		if (UNIV_LIKELY
+		    ((flags & DICT_TF_FORMAT_MASK) == DICT_TF_FORMAT_51)) {
+			/* Old versions of InnoDB did not initialize
+			FIL_PAGE_TYPE on BLOB pages.  Do not print
+			anything about the type mismatch when reading
+			a BLOB page that is in Antelope format.*/
+			return;
+		}
+
+		ut_print_timestamp(stderr);
+		fprintf(stderr,
+			"  InnoDB: FIL_PAGE_TYPE=%lu"
+			" on BLOB %s space %lu page %lu flags %lx\n",
+			(ulong) type, read ? "read" : "purge",
+			(ulong) space_id, (ulong) page_no, (ulong) flags);
+		ut_error;
+	}
+}
+
+/***********************************************************************
+Frees the space in an externally stored field to the file space
+management if the field in data is owned by the externally stored field,
+in a rollback we may have the additional condition that the field must
+not be inherited. */
+UNIV_INTERN
+void
+btr_free_externally_stored_field(
+/*=============================*/
+	dict_index_t*	index,		/* in: index of the data, the index
+					tree MUST be X-latched; if the tree
+					height is 1, then also the root page
+					must be X-latched! (this is relevant
+					in the case this function is called
+					from purge where 'data' is located on
+					an undo log page, not an index
+					page) */
+	byte*		field_ref,	/* in/out: field reference */
+	const rec_t*	rec,		/* in: record containing field_ref, for
+					page_zip_write_blob_ptr(), or NULL */
+	const ulint*	offsets,	/* in: rec_get_offsets(rec, index),
+					or NULL */
+	page_zip_des_t*	page_zip,	/* in: compressed page corresponding
+					to rec, or NULL if rec == NULL */
+	ulint		i,		/* in: field number of field_ref;
+					ignored if rec == NULL */
+	enum trx_rb_ctx	rb_ctx,		/* in: rollback context */
+	mtr_t*		local_mtr __attribute__((unused))) /* in: mtr
+					containing the latch to data an an
+					X-latch to the index tree */
+{
+	page_t*		page;
+	ulint		space_id;
+	ulint		rec_zip_size = dict_table_zip_size(index->table);
+	ulint		ext_zip_size;
+	ulint		page_no;
+	ulint		next_page_no;
+	mtr_t		mtr;
+#ifdef UNIV_DEBUG
+	ut_ad(mtr_memo_contains(local_mtr, dict_index_get_lock(index),
+				MTR_MEMO_X_LOCK));
+	ut_ad(mtr_memo_contains_page(local_mtr, field_ref,
+				     MTR_MEMO_PAGE_X_FIX));
+	ut_ad(!rec || rec_offs_validate(rec, index, offsets));
+
+	if (rec) {
+		ulint	local_len;
+		const byte*	f = rec_get_nth_field(rec, offsets,
+						      i, &local_len);
+		ut_a(local_len >= BTR_EXTERN_FIELD_REF_SIZE);
+		local_len -= BTR_EXTERN_FIELD_REF_SIZE;
+		f += local_len;
+		ut_ad(f == field_ref);
+	}
+#endif /* UNIV_DEBUG */
+
+	if (UNIV_UNLIKELY(!memcmp(field_ref, field_ref_zero,
+				  BTR_EXTERN_FIELD_REF_SIZE))) {
+		/* In the rollback of uncommitted transactions, we may
+		encounter a clustered index record whose BLOBs have
+		not been written.  There is nothing to free then. */
+		ut_a(rb_ctx == RB_RECOVERY);
+		return;
+	}
+
+	space_id = mach_read_from_4(field_ref + BTR_EXTERN_SPACE_ID);
+
+	if (UNIV_UNLIKELY(space_id != dict_index_get_space(index))) {
+		ext_zip_size = fil_space_get_zip_size(space_id);
+		/* This must be an undo log record in the system tablespace,
+		that is, in row_purge_upd_exist_or_extern().
+		Currently, externally stored records are stored in the
+		same tablespace as the referring records. */
+		ut_ad(!page_get_space_id(page_align(field_ref)));
+		ut_ad(!rec);
+		ut_ad(!page_zip);
+	} else {
+		ext_zip_size = rec_zip_size;
+	}
+
+	if (!rec) {
+		/* This is a call from row_purge_upd_exist_or_extern(). */
+		ut_ad(!page_zip);
+		rec_zip_size = 0;
+	}
+
+	for (;;) {
+		buf_block_t*	rec_block;
+		buf_block_t*	ext_block;
+
+		mtr_start(&mtr);
+
+		rec_block = buf_page_get(page_get_space_id(
+						 page_align(field_ref)),
+					 rec_zip_size,
+					 page_get_page_no(
+						 page_align(field_ref)),
+					 RW_X_LATCH, &mtr);
+		buf_block_dbg_add_level(rec_block, SYNC_NO_ORDER_CHECK);
+		page_no = mach_read_from_4(field_ref + BTR_EXTERN_PAGE_NO);
+
+		if (/* There is no external storage data */
+		    page_no == FIL_NULL
+		    /* This field does not own the externally stored field */
+		    || (mach_read_from_1(field_ref + BTR_EXTERN_LEN)
+			& BTR_EXTERN_OWNER_FLAG)
+		    /* Rollback and inherited field */
+		    || (rb_ctx != RB_NONE
+			&& (mach_read_from_1(field_ref + BTR_EXTERN_LEN)
+			    & BTR_EXTERN_INHERITED_FLAG))) {
+
+			/* Do not free */
+			mtr_commit(&mtr);
+
+			return;
+		}
+
+		ext_block = buf_page_get(space_id, ext_zip_size, page_no,
+					 RW_X_LATCH, &mtr);
+		buf_block_dbg_add_level(ext_block, SYNC_EXTERN_STORAGE);
+		page = buf_block_get_frame(ext_block);
+
+		if (ext_zip_size) {
+			/* Note that page_zip will be NULL
+			in row_purge_upd_exist_or_extern(). */
+			switch (fil_page_get_type(page)) {
+			case FIL_PAGE_TYPE_ZBLOB:
+			case FIL_PAGE_TYPE_ZBLOB2:
+				break;
+			default:
+				ut_error;
+			}
+			next_page_no = mach_read_from_4(page + FIL_PAGE_NEXT);
+
+			btr_page_free_low(index, ext_block, 0, &mtr);
+
+			if (UNIV_LIKELY(page_zip != NULL)) {
+				mach_write_to_4(field_ref + BTR_EXTERN_PAGE_NO,
+						next_page_no);
+				mach_write_to_4(field_ref + BTR_EXTERN_LEN + 4,
+						0);
+				page_zip_write_blob_ptr(page_zip, rec, index,
+							offsets, i, &mtr);
+			} else {
+				mlog_write_ulint(field_ref
+						 + BTR_EXTERN_PAGE_NO,
+						 next_page_no,
+						 MLOG_4BYTES, &mtr);
+				mlog_write_ulint(field_ref
+						 + BTR_EXTERN_LEN + 4, 0,
+						 MLOG_4BYTES, &mtr);
+			}
+		} else {
+			ut_a(!page_zip);
+			btr_check_blob_fil_page_type(space_id, page_no, page,
+						     FALSE);
+
+			next_page_no = mach_read_from_4(
+				page + FIL_PAGE_DATA
+				+ BTR_BLOB_HDR_NEXT_PAGE_NO);
+
+			/* We must supply the page level (= 0) as an argument
+			because we did not store it on the page (we save the
+			space overhead from an index page header. */
+
+			btr_page_free_low(index, ext_block, 0, &mtr);
+
+			mlog_write_ulint(field_ref + BTR_EXTERN_PAGE_NO,
+					 next_page_no,
+					 MLOG_4BYTES, &mtr);
+			/* Zero out the BLOB length.  If the server
+			crashes during the execution of this function,
+			trx_rollback_or_clean_all_recovered() could
+			dereference the half-deleted BLOB, fetching a
+			wrong prefix for the BLOB. */
+			mlog_write_ulint(field_ref + BTR_EXTERN_LEN + 4,
+					 0,
+					 MLOG_4BYTES, &mtr);
+		}
+
+		/* Commit mtr and release the BLOB block to save memory. */
+		btr_blob_free(ext_block, TRUE, &mtr);
+	}
+}
+
+/***************************************************************
+Frees the externally stored fields for a record. */
+static
+void
+btr_rec_free_externally_stored_fields(
+/*==================================*/
+	dict_index_t*	index,	/* in: index of the data, the index
+				tree MUST be X-latched */
+	rec_t*		rec,	/* in/out: record */
+	const ulint*	offsets,/* in: rec_get_offsets(rec, index) */
+	page_zip_des_t*	page_zip,/* in: compressed page whose uncompressed
+				part will be updated, or NULL */
+	enum trx_rb_ctx	rb_ctx,	/* in: rollback context */
+	mtr_t*		mtr)	/* in: mini-transaction handle which contains
+				an X-latch to record page and to the index
+				tree */
+{
+	ulint	n_fields;
+	ulint	i;
+
+	ut_ad(rec_offs_validate(rec, index, offsets));
+	ut_ad(mtr_memo_contains_page(mtr, rec, MTR_MEMO_PAGE_X_FIX));
+	/* Free possible externally stored fields in the record */
+
+	ut_ad(dict_table_is_comp(index->table) == !!rec_offs_comp(offsets));
+	n_fields = rec_offs_n_fields(offsets);
+
+	for (i = 0; i < n_fields; i++) {
+		if (rec_offs_nth_extern(offsets, i)) {
+			ulint	len;
+			byte*	data
+				= rec_get_nth_field(rec, offsets, i, &len);
+			ut_a(len >= BTR_EXTERN_FIELD_REF_SIZE);
+
+			btr_free_externally_stored_field(
+				index, data + len - BTR_EXTERN_FIELD_REF_SIZE,
+				rec, offsets, page_zip, i, rb_ctx, mtr);
+		}
+	}
+}
+
+/***************************************************************
+Frees the externally stored fields for a record, if the field is mentioned
+in the update vector. */
+static
+void
+btr_rec_free_updated_extern_fields(
+/*===============================*/
+	dict_index_t*	index,	/* in: index of rec; the index tree MUST be
+				X-latched */
+	rec_t*		rec,	/* in/out: record */
+	page_zip_des_t*	page_zip,/* in: compressed page whose uncompressed
+				part will be updated, or NULL */
+	const ulint*	offsets,/* in: rec_get_offsets(rec, index) */
+	const upd_t*	update,	/* in: update vector */
+	enum trx_rb_ctx	rb_ctx,	/* in: rollback context */
+	mtr_t*		mtr)	/* in: mini-transaction handle which contains
+				an X-latch to record page and to the tree */
+{
+	ulint	n_fields;
+	ulint	i;
+
+	ut_ad(rec_offs_validate(rec, index, offsets));
+	ut_ad(mtr_memo_contains_page(mtr, rec, MTR_MEMO_PAGE_X_FIX));
+
+	/* Free possible externally stored fields in the record */
+
+	n_fields = upd_get_n_fields(update);
+
+	for (i = 0; i < n_fields; i++) {
+		const upd_field_t* ufield = upd_get_nth_field(update, i);
+
+		if (rec_offs_nth_extern(offsets, ufield->field_no)) {
+			ulint	len;
+			byte*	data = rec_get_nth_field(
+				rec, offsets, ufield->field_no, &len);
+			ut_a(len >= BTR_EXTERN_FIELD_REF_SIZE);
+
+			btr_free_externally_stored_field(
+				index, data + len - BTR_EXTERN_FIELD_REF_SIZE,
+				rec, offsets, page_zip,
+				ufield->field_no, rb_ctx, mtr);
+		}
+	}
+}
+
+/***********************************************************************
+Copies the prefix of an uncompressed BLOB.  The clustered index record
+that points to this BLOB must be protected by a lock or a page latch. */
+static
+ulint
+btr_copy_blob_prefix(
+/*=================*/
+				/* out: number of bytes written to buf */
+	byte*		buf,	/* out: the externally stored part of
+				the field, or a prefix of it */
+	ulint		len,	/* in: length of buf, in bytes */
+	ulint		space_id,/* in: space id of the BLOB pages */
+	ulint		page_no,/* in: page number of the first BLOB page */
+	ulint		offset)	/* in: offset on the first BLOB page */
+{
+	ulint	copied_len	= 0;
+
+	for (;;) {
+		mtr_t		mtr;
+		buf_block_t*	block;
+		const page_t*	page;
+		const byte*	blob_header;
+		ulint		part_len;
+		ulint		copy_len;
+
+		mtr_start(&mtr);
+
+		block = buf_page_get(space_id, 0, page_no, RW_S_LATCH, &mtr);
+		buf_block_dbg_add_level(block, SYNC_EXTERN_STORAGE);
+		page = buf_block_get_frame(block);
+
+		btr_check_blob_fil_page_type(space_id, page_no, page, TRUE);
+
+		blob_header = page + offset;
+		part_len = btr_blob_get_part_len(blob_header);
+		copy_len = ut_min(part_len, len - copied_len);
+
+		memcpy(buf + copied_len,
+		       blob_header + BTR_BLOB_HDR_SIZE, copy_len);
+		copied_len += copy_len;
+
+		page_no = btr_blob_get_next_page_no(blob_header);
+
+		mtr_commit(&mtr);
+
+		if (page_no == FIL_NULL || copy_len != part_len) {
+			return(copied_len);
+		}
+
+		/* On other BLOB pages except the first the BLOB header
+		always is at the page data start: */
+
+		offset = FIL_PAGE_DATA;
+
+		ut_ad(copied_len <= len);
+	}
+}
+
+/***********************************************************************
+Copies the prefix of a compressed BLOB.  The clustered index record
+that points to this BLOB must be protected by a lock or a page latch. */
+static
+void
+btr_copy_zblob_prefix(
+/*==================*/
+	z_stream*	d_stream,/* in/out: the decompressing stream */
+	ulint		zip_size,/* in: compressed BLOB page size */
+	ulint		space_id,/* in: space id of the BLOB pages */
+	ulint		page_no,/* in: page number of the first BLOB page */
+	ulint		offset)	/* in: offset on the first BLOB page */
+{
+	ulint	page_type = FIL_PAGE_TYPE_ZBLOB;
+
+	ut_ad(ut_is_2pow(zip_size));
+	ut_ad(zip_size >= PAGE_ZIP_MIN_SIZE);
+	ut_ad(zip_size <= UNIV_PAGE_SIZE);
+	ut_ad(space_id);
+
+	for (;;) {
+		buf_page_t*	bpage;
+		int		err;
+		ulint		next_page_no;
+
+		/* There is no latch on bpage directly.  Instead,
+		bpage is protected by the B-tree page latch that
+		is being held on the clustered index record, or,
+		in row_merge_copy_blobs(), by an exclusive table lock. */
+		bpage = buf_page_get_zip(space_id, zip_size, page_no);
+
+		if (UNIV_UNLIKELY(!bpage)) {
+			ut_print_timestamp(stderr);
+			fprintf(stderr,
+				"  InnoDB: Cannot load"
+				" compressed BLOB"
+				" page %lu space %lu\n",
+				(ulong) page_no, (ulong) space_id);
+			return;
+		}
+
+		if (UNIV_UNLIKELY
+		    (fil_page_get_type(bpage->zip.data) != page_type)) {
+			ut_print_timestamp(stderr);
+			fprintf(stderr,
+				"  InnoDB: Unexpected type %lu of"
+				" compressed BLOB"
+				" page %lu space %lu\n",
+				(ulong) fil_page_get_type(bpage->zip.data),
+				(ulong) page_no, (ulong) space_id);
+			goto end_of_blob;
+		}
+
+		next_page_no = mach_read_from_4(bpage->zip.data + offset);
+
+		if (UNIV_LIKELY(offset == FIL_PAGE_NEXT)) {
+			/* When the BLOB begins at page header,
+			the compressed data payload does not
+			immediately follow the next page pointer. */
+			offset = FIL_PAGE_DATA;
+		} else {
+			offset += 4;
+		}
+
+		d_stream->next_in = bpage->zip.data + offset;
+		d_stream->avail_in = zip_size - offset;
+
+		err = inflate(d_stream, Z_NO_FLUSH);
+		switch (err) {
+		case Z_OK:
+			if (!d_stream->avail_out) {
+				goto end_of_blob;
+			}
+			break;
+		case Z_STREAM_END:
+			if (next_page_no == FIL_NULL) {
+				goto end_of_blob;
+			}
+			/* fall through */
+		default:
+inflate_error:
+			ut_print_timestamp(stderr);
+			fprintf(stderr,
+				"  InnoDB: inflate() of"
+				" compressed BLOB"
+				" page %lu space %lu returned %d (%s)\n",
+				(ulong) page_no, (ulong) space_id,
+				err, d_stream->msg);
+		case Z_BUF_ERROR:
+			goto end_of_blob;
+		}
+
+		if (next_page_no == FIL_NULL) {
+			if (!d_stream->avail_in) {
+				ut_print_timestamp(stderr);
+				fprintf(stderr,
+					"  InnoDB: unexpected end of"
+					" compressed BLOB"
+					" page %lu space %lu\n",
+					(ulong) page_no,
+					(ulong) space_id);
+			} else {
+				err = inflate(d_stream, Z_FINISH);
+				switch (err) {
+				case Z_STREAM_END:
+				case Z_BUF_ERROR:
+					break;
+				default:
+					goto inflate_error;
+				}
+			}
+
+end_of_blob:
+			buf_page_release_zip(bpage);
+			return;
+		}
+
+		buf_page_release_zip(bpage);
+
+		/* On other BLOB pages except the first
+		the BLOB header always is at the page header: */
+
+		page_no = next_page_no;
+		offset = FIL_PAGE_NEXT;
+		page_type = FIL_PAGE_TYPE_ZBLOB2;
+	}
+}
+
+/***********************************************************************
+Copies the prefix of an externally stored field of a record.  The
+clustered index record that points to this BLOB must be protected by a
+lock or a page latch. */
+static
+ulint
+btr_copy_externally_stored_field_prefix_low(
+/*========================================*/
+				/* out: number of bytes written to buf */
+	byte*		buf,	/* out: the externally stored part of
+				the field, or a prefix of it */
+	ulint		len,	/* in: length of buf, in bytes */
+	ulint		zip_size,/* in: nonzero=compressed BLOB page size,
+				zero for uncompressed BLOBs */
+	ulint		space_id,/* in: space id of the first BLOB page */
+	ulint		page_no,/* in: page number of the first BLOB page */
+	ulint		offset)	/* in: offset on the first BLOB page */
+{
+	if (UNIV_UNLIKELY(len == 0)) {
+		return(0);
+	}
+
+	if (UNIV_UNLIKELY(zip_size)) {
+		int		err;
+		z_stream	d_stream;
+		mem_heap_t*	heap;
+
+		/* Zlib inflate needs 32 kilobytes for the default
+		window size, plus a few kilobytes for small objects. */
+		heap = mem_heap_create(40000);
+		page_zip_set_alloc(&d_stream, heap);
+
+		err = inflateInit(&d_stream);
+		ut_a(err == Z_OK);
+
+		d_stream.next_out = buf;
+		d_stream.avail_out = len;
+		d_stream.avail_in = 0;
+
+		btr_copy_zblob_prefix(&d_stream, zip_size,
+				      space_id, page_no, offset);
+		inflateEnd(&d_stream);
+		mem_heap_free(heap);
+		return(d_stream.total_out);
+	} else {
+		return(btr_copy_blob_prefix(buf, len, space_id,
+					    page_no, offset));
+	}
+}
+
+/***********************************************************************
+Copies the prefix of an externally stored field of a record.  The
+clustered index record must be protected by a lock or a page latch. */
+UNIV_INTERN
+ulint
+btr_copy_externally_stored_field_prefix(
+/*====================================*/
+				/* out: the length of the copied field,
+				or 0 if the column was being or has been
+				deleted */
+	byte*		buf,	/* out: the field, or a prefix of it */
+	ulint		len,	/* in: length of buf, in bytes */
+	ulint		zip_size,/* in: nonzero=compressed BLOB page size,
+				zero for uncompressed BLOBs */
+	const byte*	data,	/* in: 'internally' stored part of the
+				field containing also the reference to
+				the external part; must be protected by
+				a lock or a page latch */
+	ulint		local_len)/* in: length of data, in bytes */
+{
+	ulint	space_id;
+	ulint	page_no;
+	ulint	offset;
+
+	ut_a(local_len >= BTR_EXTERN_FIELD_REF_SIZE);
+
+	local_len -= BTR_EXTERN_FIELD_REF_SIZE;
+
+	if (UNIV_UNLIKELY(local_len >= len)) {
+		memcpy(buf, data, len);
+		return(len);
+	}
+
+	memcpy(buf, data, local_len);
+	data += local_len;
+
+	ut_a(memcmp(data, field_ref_zero, BTR_EXTERN_FIELD_REF_SIZE));
+
+	if (!mach_read_from_4(data + BTR_EXTERN_LEN + 4)) {
+		/* The externally stored part of the column has been
+		(partially) deleted.  Signal the half-deleted BLOB
+		to the caller. */
+
+		return(0);
+	}
+
+	space_id = mach_read_from_4(data + BTR_EXTERN_SPACE_ID);
+
+	page_no = mach_read_from_4(data + BTR_EXTERN_PAGE_NO);
+
+	offset = mach_read_from_4(data + BTR_EXTERN_OFFSET);
+
+	return(local_len
+	       + btr_copy_externally_stored_field_prefix_low(buf + local_len,
+							     len - local_len,
+							     zip_size,
+							     space_id, page_no,
+							     offset));
+}
+
+/***********************************************************************
+Copies an externally stored field of a record to mem heap.  The
+clustered index record must be protected by a lock or a page latch. */
+static
+byte*
+btr_copy_externally_stored_field(
+/*=============================*/
+				/* out: the whole field copied to heap */
+	ulint*		len,	/* out: length of the whole field */
+	const byte*	data,	/* in: 'internally' stored part of the
+				field containing also the reference to
+				the external part; must be protected by
+				a lock or a page latch */
+	ulint		zip_size,/* in: nonzero=compressed BLOB page size,
+				zero for uncompressed BLOBs */
+	ulint		local_len,/* in: length of data */
+	mem_heap_t*	heap)	/* in: mem heap */
+{
+	ulint	space_id;
+	ulint	page_no;
+	ulint	offset;
+	ulint	extern_len;
+	byte*	buf;
+
+	ut_a(local_len >= BTR_EXTERN_FIELD_REF_SIZE);
+
+	local_len -= BTR_EXTERN_FIELD_REF_SIZE;
+
+	space_id = mach_read_from_4(data + local_len + BTR_EXTERN_SPACE_ID);
+
+	page_no = mach_read_from_4(data + local_len + BTR_EXTERN_PAGE_NO);
+
+	offset = mach_read_from_4(data + local_len + BTR_EXTERN_OFFSET);
+
+	/* Currently a BLOB cannot be bigger than 4 GB; we
+	leave the 4 upper bytes in the length field unused */
+
+	extern_len = mach_read_from_4(data + local_len + BTR_EXTERN_LEN + 4);
+
+	buf = mem_heap_alloc(heap, local_len + extern_len);
+
+	memcpy(buf, data, local_len);
+	*len = local_len
+		+ btr_copy_externally_stored_field_prefix_low(buf + local_len,
+							      extern_len,
+							      zip_size,
+							      space_id,
+							      page_no, offset);
+
+	return(buf);
+}
+
+/***********************************************************************
+Copies an externally stored field of a record to mem heap. */
+UNIV_INTERN
+byte*
+btr_rec_copy_externally_stored_field(
+/*=================================*/
+				/* out: the field copied to heap */
+	const rec_t*	rec,	/* in: record in a clustered index;
+				must be protected by a lock or a page latch */
+	const ulint*	offsets,/* in: array returned by rec_get_offsets() */
+	ulint		zip_size,/* in: nonzero=compressed BLOB page size,
+				zero for uncompressed BLOBs */
+	ulint		no,	/* in: field number */
+	ulint*		len,	/* out: length of the field */
+	mem_heap_t*	heap)	/* in: mem heap */
+{
+	ulint		local_len;
+	const byte*	data;
+
+	ut_a(rec_offs_nth_extern(offsets, no));
+
+	/* An externally stored field can contain some initial
+	data from the field, and in the last 20 bytes it has the
+	space id, page number, and offset where the rest of the
+	field data is stored, and the data length in addition to
+	the data stored locally. We may need to store some data
+	locally to get the local record length above the 128 byte
+	limit so that field offsets are stored in two bytes, and
+	the extern bit is available in those two bytes. */
+
+	data = rec_get_nth_field(rec, offsets, no, &local_len);
+
+	return(btr_copy_externally_stored_field(len, data,
+						zip_size, local_len, heap));
+}
diff --git a/storage/xtradb/btr/btr0pcur.c b/storage/xtradb/btr/btr0pcur.c
new file mode 100644
index 00000000000..b14efefe13f
--- /dev/null
+++ b/storage/xtradb/btr/btr0pcur.c
@@ -0,0 +1,584 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The index tree persistent cursor
+
+Created 2/23/1996 Heikki Tuuri
+*******************************************************/
+
+#include "btr0pcur.h"
+
+#ifdef UNIV_NONINL
+#include "btr0pcur.ic"
+#endif
+
+#include "ut0byte.h"
+#include "rem0cmp.h"
+#include "trx0trx.h"
+
+/******************************************************************
+Allocates memory for a persistent cursor object and initializes the cursor. */
+UNIV_INTERN
+btr_pcur_t*
+btr_pcur_create_for_mysql(void)
+/*============================*/
+				/* out, own: persistent cursor */
+{
+	btr_pcur_t*	pcur;
+
+	pcur = mem_alloc(sizeof(btr_pcur_t));
+
+	pcur->btr_cur.index = NULL;
+	btr_pcur_init(pcur);
+
+	return(pcur);
+}
+
+/******************************************************************
+Frees the memory for a persistent cursor object. */
+UNIV_INTERN
+void
+btr_pcur_free_for_mysql(
+/*====================*/
+	btr_pcur_t*	cursor)	/* in, own: persistent cursor */
+{
+	if (cursor->old_rec_buf != NULL) {
+
+		mem_free(cursor->old_rec_buf);
+
+		cursor->old_rec_buf = NULL;
+	}
+
+	cursor->btr_cur.page_cur.rec = NULL;
+	cursor->old_rec = NULL;
+	cursor->old_n_fields = 0;
+	cursor->old_stored = BTR_PCUR_OLD_NOT_STORED;
+
+	cursor->latch_mode = BTR_NO_LATCHES;
+	cursor->pos_state = BTR_PCUR_NOT_POSITIONED;
+
+	mem_free(cursor);
+}
+
+/******************************************************************
+The position of the cursor is stored by taking an initial segment of the
+record the cursor is positioned on, before, or after, and copying it to the
+cursor data structure, or just setting a flag if the cursor id before the
+first in an EMPTY tree, or after the last in an EMPTY tree. NOTE that the
+page where the cursor is positioned must not be empty if the index tree is
+not totally empty! */
+UNIV_INTERN
+void
+btr_pcur_store_position(
+/*====================*/
+	btr_pcur_t*	cursor, /* in: persistent cursor */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	page_cur_t*	page_cursor;
+	buf_block_t*	block;
+	rec_t*		rec;
+	dict_index_t*	index;
+	page_t*		page;
+	ulint		offs;
+
+	ut_a(cursor->pos_state == BTR_PCUR_IS_POSITIONED);
+	ut_ad(cursor->latch_mode != BTR_NO_LATCHES);
+
+	block = btr_pcur_get_block(cursor);
+	index = btr_cur_get_index(btr_pcur_get_btr_cur(cursor));
+
+	page_cursor = btr_pcur_get_page_cur(cursor);
+
+	rec = page_cur_get_rec(page_cursor);
+	page = page_align(rec);
+	offs = page_offset(rec);
+
+	ut_ad(mtr_memo_contains(mtr, block, MTR_MEMO_PAGE_S_FIX)
+	      || mtr_memo_contains(mtr, block, MTR_MEMO_PAGE_X_FIX));
+	ut_a(cursor->latch_mode != BTR_NO_LATCHES);
+
+	if (UNIV_UNLIKELY(page_get_n_recs(page) == 0)) {
+		/* It must be an empty index tree; NOTE that in this case
+		we do not store the modify_clock, but always do a search
+		if we restore the cursor position */
+
+		ut_a(btr_page_get_next(page, mtr) == FIL_NULL);
+		ut_a(btr_page_get_prev(page, mtr) == FIL_NULL);
+
+		cursor->old_stored = BTR_PCUR_OLD_STORED;
+
+		if (page_rec_is_supremum_low(offs)) {
+
+			cursor->rel_pos = BTR_PCUR_AFTER_LAST_IN_TREE;
+		} else {
+			cursor->rel_pos = BTR_PCUR_BEFORE_FIRST_IN_TREE;
+		}
+
+		return;
+	}
+
+	if (page_rec_is_supremum_low(offs)) {
+
+		rec = page_rec_get_prev(rec);
+
+		cursor->rel_pos = BTR_PCUR_AFTER;
+
+	} else if (page_rec_is_infimum_low(offs)) {
+
+		rec = page_rec_get_next(rec);
+
+		cursor->rel_pos = BTR_PCUR_BEFORE;
+	} else {
+		cursor->rel_pos = BTR_PCUR_ON;
+	}
+
+	cursor->old_stored = BTR_PCUR_OLD_STORED;
+	cursor->old_rec = dict_index_copy_rec_order_prefix(
+		index, rec, &cursor->old_n_fields,
+		&cursor->old_rec_buf, &cursor->buf_size);
+
+	cursor->block_when_stored = block;
+	cursor->modify_clock = buf_block_get_modify_clock(block);
+}
+
+/******************************************************************
+Copies the stored position of a pcur to another pcur. */
+UNIV_INTERN
+void
+btr_pcur_copy_stored_position(
+/*==========================*/
+	btr_pcur_t*	pcur_receive,	/* in: pcur which will receive the
+					position info */
+	btr_pcur_t*	pcur_donate)	/* in: pcur from which the info is
+					copied */
+{
+	if (pcur_receive->old_rec_buf) {
+		mem_free(pcur_receive->old_rec_buf);
+	}
+
+	ut_memcpy(pcur_receive, pcur_donate, sizeof(btr_pcur_t));
+
+	if (pcur_donate->old_rec_buf) {
+
+		pcur_receive->old_rec_buf = mem_alloc(pcur_donate->buf_size);
+
+		ut_memcpy(pcur_receive->old_rec_buf, pcur_donate->old_rec_buf,
+			  pcur_donate->buf_size);
+		pcur_receive->old_rec = pcur_receive->old_rec_buf
+			+ (pcur_donate->old_rec - pcur_donate->old_rec_buf);
+	}
+
+	pcur_receive->old_n_fields = pcur_donate->old_n_fields;
+}
+
+/******************************************************************
+Restores the stored position of a persistent cursor bufferfixing the page and
+obtaining the specified latches. If the cursor position was saved when the
+(1) cursor was positioned on a user record: this function restores the position
+to the last record LESS OR EQUAL to the stored record;
+(2) cursor was positioned on a page infimum record: restores the position to
+the last record LESS than the user record which was the successor of the page
+infimum;
+(3) cursor was positioned on the page supremum: restores to the first record
+GREATER than the user record which was the predecessor of the supremum.
+(4) cursor was positioned before the first or after the last in an empty tree:
+restores to before first or after the last in the tree. */
+UNIV_INTERN
+ibool
+btr_pcur_restore_position(
+/*======================*/
+					/* out: TRUE if the cursor position
+					was stored when it was on a user record
+					and it can be restored on a user record
+					whose ordering fields are identical to
+					the ones of the original user record */
+	ulint		latch_mode,	/* in: BTR_SEARCH_LEAF, ... */
+	btr_pcur_t*	cursor,		/* in: detached persistent cursor */
+	mtr_t*		mtr)		/* in: mtr */
+{
+	dict_index_t*	index;
+	dtuple_t*	tuple;
+	ulint		mode;
+	ulint		old_mode;
+	mem_heap_t*	heap;
+
+	index = btr_cur_get_index(btr_pcur_get_btr_cur(cursor));
+
+	if (UNIV_UNLIKELY(cursor->old_stored != BTR_PCUR_OLD_STORED)
+	    || UNIV_UNLIKELY(cursor->pos_state != BTR_PCUR_WAS_POSITIONED
+			     && cursor->pos_state != BTR_PCUR_IS_POSITIONED)) {
+		ut_print_buf(stderr, cursor, sizeof(btr_pcur_t));
+		putc('\n', stderr);
+		if (cursor->trx_if_known) {
+			trx_print(stderr, cursor->trx_if_known, 0);
+		}
+
+		ut_error;
+	}
+
+	if (UNIV_UNLIKELY
+	    (cursor->rel_pos == BTR_PCUR_AFTER_LAST_IN_TREE
+	     || cursor->rel_pos == BTR_PCUR_BEFORE_FIRST_IN_TREE)) {
+
+		/* In these cases we do not try an optimistic restoration,
+		but always do a search */
+
+		btr_cur_open_at_index_side(
+			cursor->rel_pos == BTR_PCUR_BEFORE_FIRST_IN_TREE,
+			index, latch_mode, btr_pcur_get_btr_cur(cursor), mtr);
+
+		cursor->block_when_stored = btr_pcur_get_block(cursor);
+
+		return(FALSE);
+	}
+
+	ut_a(cursor->old_rec);
+	ut_a(cursor->old_n_fields);
+
+	if (UNIV_LIKELY(latch_mode == BTR_SEARCH_LEAF)
+	    || UNIV_LIKELY(latch_mode == BTR_MODIFY_LEAF)) {
+		/* Try optimistic restoration */
+
+		if (UNIV_LIKELY(buf_page_optimistic_get(
+					latch_mode,
+					cursor->block_when_stored,
+					cursor->modify_clock, mtr))) {
+			cursor->pos_state = BTR_PCUR_IS_POSITIONED;
+
+			buf_block_dbg_add_level(btr_pcur_get_block(cursor),
+						SYNC_TREE_NODE);
+
+			if (cursor->rel_pos == BTR_PCUR_ON) {
+#ifdef UNIV_DEBUG
+				const rec_t*	rec;
+				const ulint*	offsets1;
+				const ulint*	offsets2;
+#endif /* UNIV_DEBUG */
+				cursor->latch_mode = latch_mode;
+#ifdef UNIV_DEBUG
+				rec = btr_pcur_get_rec(cursor);
+
+				heap = mem_heap_create(256);
+				offsets1 = rec_get_offsets(
+					cursor->old_rec, index, NULL,
+					cursor->old_n_fields, &heap);
+				offsets2 = rec_get_offsets(
+					rec, index, NULL,
+					cursor->old_n_fields, &heap);
+
+				ut_ad(!cmp_rec_rec(cursor->old_rec,
+						   rec, offsets1, offsets2,
+						   index));
+				mem_heap_free(heap);
+#endif /* UNIV_DEBUG */
+				return(TRUE);
+			}
+
+			return(FALSE);
+		}
+	}
+
+	/* If optimistic restoration did not succeed, open the cursor anew */
+
+	heap = mem_heap_create(256);
+
+	tuple = dict_index_build_data_tuple(index, cursor->old_rec,
+					    cursor->old_n_fields, heap);
+
+	/* Save the old search mode of the cursor */
+	old_mode = cursor->search_mode;
+
+	if (UNIV_LIKELY(cursor->rel_pos == BTR_PCUR_ON)) {
+		mode = PAGE_CUR_LE;
+	} else if (cursor->rel_pos == BTR_PCUR_AFTER) {
+		mode = PAGE_CUR_G;
+	} else {
+		ut_ad(cursor->rel_pos == BTR_PCUR_BEFORE);
+		mode = PAGE_CUR_L;
+	}
+
+	btr_pcur_open_with_no_init(index, tuple, mode, latch_mode,
+				   cursor, 0, mtr);
+
+	/* Restore the old search mode */
+	cursor->search_mode = old_mode;
+
+	if (cursor->rel_pos == BTR_PCUR_ON
+	    && btr_pcur_is_on_user_rec(cursor)
+	    && 0 == cmp_dtuple_rec(tuple, btr_pcur_get_rec(cursor),
+				   rec_get_offsets(
+					   btr_pcur_get_rec(cursor), index,
+					   NULL, ULINT_UNDEFINED, &heap))) {
+
+		/* We have to store the NEW value for the modify clock, since
+		the cursor can now be on a different page! But we can retain
+		the value of old_rec */
+
+		cursor->block_when_stored = btr_pcur_get_block(cursor);
+		cursor->modify_clock = buf_block_get_modify_clock(
+			cursor->block_when_stored);
+		cursor->old_stored = BTR_PCUR_OLD_STORED;
+
+		mem_heap_free(heap);
+
+		return(TRUE);
+	}
+
+	mem_heap_free(heap);
+
+	/* We have to store new position information, modify_clock etc.,
+	to the cursor because it can now be on a different page, the record
+	under it may have been removed, etc. */
+
+	btr_pcur_store_position(cursor, mtr);
+
+	return(FALSE);
+}
+
+/******************************************************************
+If the latch mode of the cursor is BTR_LEAF_SEARCH or BTR_LEAF_MODIFY,
+releases the page latch and bufferfix reserved by the cursor.
+NOTE! In the case of BTR_LEAF_MODIFY, there should not exist changes
+made by the current mini-transaction to the data protected by the
+cursor latch, as then the latch must not be released until mtr_commit. */
+UNIV_INTERN
+void
+btr_pcur_release_leaf(
+/*==================*/
+	btr_pcur_t*	cursor, /* in: persistent cursor */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	buf_block_t*	block;
+
+	ut_a(cursor->pos_state == BTR_PCUR_IS_POSITIONED);
+	ut_ad(cursor->latch_mode != BTR_NO_LATCHES);
+
+	block = btr_pcur_get_block(cursor);
+
+	btr_leaf_page_release(block, cursor->latch_mode, mtr);
+
+	cursor->latch_mode = BTR_NO_LATCHES;
+
+	cursor->pos_state = BTR_PCUR_WAS_POSITIONED;
+}
+
+/*************************************************************
+Moves the persistent cursor to the first record on the next page. Releases the
+latch on the current page, and bufferunfixes it. Note that there must not be
+modifications on the current page, as then the x-latch can be released only in
+mtr_commit. */
+UNIV_INTERN
+void
+btr_pcur_move_to_next_page(
+/*=======================*/
+	btr_pcur_t*	cursor,	/* in: persistent cursor; must be on the
+				last record of the current page */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	ulint		next_page_no;
+	ulint		space;
+	ulint		zip_size;
+	page_t*		page;
+	buf_block_t*	next_block;
+	page_t*		next_page;
+
+	ut_a(cursor->pos_state == BTR_PCUR_IS_POSITIONED);
+	ut_ad(cursor->latch_mode != BTR_NO_LATCHES);
+	ut_ad(btr_pcur_is_after_last_on_page(cursor));
+
+	cursor->old_stored = BTR_PCUR_OLD_NOT_STORED;
+
+	page = btr_pcur_get_page(cursor);
+	next_page_no = btr_page_get_next(page, mtr);
+	space = buf_block_get_space(btr_pcur_get_block(cursor));
+	zip_size = buf_block_get_zip_size(btr_pcur_get_block(cursor));
+
+	ut_ad(next_page_no != FIL_NULL);
+
+	next_block = btr_block_get(space, zip_size, next_page_no,
+				   cursor->latch_mode, mtr);
+	next_page = buf_block_get_frame(next_block);
+#ifdef UNIV_BTR_DEBUG
+	ut_a(page_is_comp(next_page) == page_is_comp(page));
+	ut_a(btr_page_get_prev(next_page, mtr)
+	     == buf_block_get_page_no(btr_pcur_get_block(cursor)));
+#endif /* UNIV_BTR_DEBUG */
+	next_block->check_index_page_at_flush = TRUE;
+
+	btr_leaf_page_release(btr_pcur_get_block(cursor),
+			      cursor->latch_mode, mtr);
+
+	page_cur_set_before_first(next_block, btr_pcur_get_page_cur(cursor));
+
+	page_check_dir(next_page);
+}
+
+/*************************************************************
+Moves the persistent cursor backward if it is on the first record of the page.
+Commits mtr. Note that to prevent a possible deadlock, the operation
+first stores the position of the cursor, commits mtr, acquires the necessary
+latches and restores the cursor position again before returning. The
+alphabetical position of the cursor is guaranteed to be sensible on
+return, but it may happen that the cursor is not positioned on the last
+record of any page, because the structure of the tree may have changed
+during the time when the cursor had no latches. */
+UNIV_INTERN
+void
+btr_pcur_move_backward_from_page(
+/*=============================*/
+	btr_pcur_t*	cursor,	/* in: persistent cursor, must be on the first
+				record of the current page */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	ulint		prev_page_no;
+	ulint		space;
+	page_t*		page;
+	buf_block_t*	prev_block;
+	ulint		latch_mode;
+	ulint		latch_mode2;
+
+	ut_a(cursor->pos_state == BTR_PCUR_IS_POSITIONED);
+	ut_ad(cursor->latch_mode != BTR_NO_LATCHES);
+	ut_ad(btr_pcur_is_before_first_on_page(cursor));
+	ut_ad(!btr_pcur_is_before_first_in_tree(cursor, mtr));
+
+	latch_mode = cursor->latch_mode;
+
+	if (latch_mode == BTR_SEARCH_LEAF) {
+
+		latch_mode2 = BTR_SEARCH_PREV;
+
+	} else if (latch_mode == BTR_MODIFY_LEAF) {
+
+		latch_mode2 = BTR_MODIFY_PREV;
+	} else {
+		latch_mode2 = 0; /* To eliminate compiler warning */
+		ut_error;
+	}
+
+	btr_pcur_store_position(cursor, mtr);
+
+	mtr_commit(mtr);
+
+	mtr_start(mtr);
+
+	btr_pcur_restore_position(latch_mode2, cursor, mtr);
+
+	page = btr_pcur_get_page(cursor);
+
+	prev_page_no = btr_page_get_prev(page, mtr);
+	space = buf_block_get_space(btr_pcur_get_block(cursor));
+
+	if (prev_page_no == FIL_NULL) {
+	} else if (btr_pcur_is_before_first_on_page(cursor)) {
+
+		prev_block = btr_pcur_get_btr_cur(cursor)->left_block;
+
+		btr_leaf_page_release(btr_pcur_get_block(cursor),
+				      latch_mode, mtr);
+
+		page_cur_set_after_last(prev_block,
+					btr_pcur_get_page_cur(cursor));
+	} else {
+
+		/* The repositioned cursor did not end on an infimum record on
+		a page. Cursor repositioning acquired a latch also on the
+		previous page, but we do not need the latch: release it. */
+
+		prev_block = btr_pcur_get_btr_cur(cursor)->left_block;
+
+		btr_leaf_page_release(prev_block, latch_mode, mtr);
+	}
+
+	cursor->latch_mode = latch_mode;
+
+	cursor->old_stored = BTR_PCUR_OLD_NOT_STORED;
+}
+
+/*************************************************************
+Moves the persistent cursor to the previous record in the tree. If no records
+are left, the cursor stays 'before first in tree'. */
+UNIV_INTERN
+ibool
+btr_pcur_move_to_prev(
+/*==================*/
+				/* out: TRUE if the cursor was not before first
+				in tree */
+	btr_pcur_t*	cursor,	/* in: persistent cursor; NOTE that the
+				function may release the page latch */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	ut_ad(cursor->pos_state == BTR_PCUR_IS_POSITIONED);
+	ut_ad(cursor->latch_mode != BTR_NO_LATCHES);
+
+	cursor->old_stored = BTR_PCUR_OLD_NOT_STORED;
+
+	if (btr_pcur_is_before_first_on_page(cursor)) {
+
+		if (btr_pcur_is_before_first_in_tree(cursor, mtr)) {
+
+			return(FALSE);
+		}
+
+		btr_pcur_move_backward_from_page(cursor, mtr);
+
+		return(TRUE);
+	}
+
+	btr_pcur_move_to_prev_on_page(cursor);
+
+	return(TRUE);
+}
+
+/******************************************************************
+If mode is PAGE_CUR_G or PAGE_CUR_GE, opens a persistent cursor on the first
+user record satisfying the search condition, in the case PAGE_CUR_L or
+PAGE_CUR_LE, on the last user record. If no such user record exists, then
+in the first case sets the cursor after last in tree, and in the latter case
+before first in tree. The latching mode must be BTR_SEARCH_LEAF or
+BTR_MODIFY_LEAF. */
+UNIV_INTERN
+void
+btr_pcur_open_on_user_rec(
+/*======================*/
+	dict_index_t*	index,		/* in: index */
+	const dtuple_t*	tuple,		/* in: tuple on which search done */
+	ulint		mode,		/* in: PAGE_CUR_L, ... */
+	ulint		latch_mode,	/* in: BTR_SEARCH_LEAF or
+					BTR_MODIFY_LEAF */
+	btr_pcur_t*	cursor,		/* in: memory buffer for persistent
+					cursor */
+	mtr_t*		mtr)		/* in: mtr */
+{
+	btr_pcur_open(index, tuple, mode, latch_mode, cursor, mtr);
+
+	if ((mode == PAGE_CUR_GE) || (mode == PAGE_CUR_G)) {
+
+		if (btr_pcur_is_after_last_on_page(cursor)) {
+
+			btr_pcur_move_to_next_user_rec(cursor, mtr);
+		}
+	} else {
+		ut_ad((mode == PAGE_CUR_LE) || (mode == PAGE_CUR_L));
+
+		/* Not implemented yet */
+
+		ut_error;
+	}
+}
diff --git a/storage/xtradb/btr/btr0sea.c b/storage/xtradb/btr/btr0sea.c
new file mode 100644
index 00000000000..8aafd738542
--- /dev/null
+++ b/storage/xtradb/btr/btr0sea.c
@@ -0,0 +1,1872 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+Copyright (c) 2008, Google Inc.
+
+Portions of this file contain modifications contributed and copyrighted by
+Google, Inc. Those modifications are gratefully acknowledged and are described
+briefly in the InnoDB documentation. The contributions by Google are
+incorporated with their permission, and subject to the conditions contained in
+the file COPYING.Google.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/************************************************************************
+The index tree adaptive search
+
+Created 2/17/1996 Heikki Tuuri
+*************************************************************************/
+
+#include "btr0sea.h"
+#ifdef UNIV_NONINL
+#include "btr0sea.ic"
+#endif
+
+#include "buf0buf.h"
+#include "page0page.h"
+#include "page0cur.h"
+#include "btr0cur.h"
+#include "btr0pcur.h"
+#include "btr0btr.h"
+#include "ha0ha.h"
+
+/* Flag: has the search system been enabled?
+Protected by btr_search_latch and btr_search_enabled_mutex. */
+UNIV_INTERN char		btr_search_enabled	= TRUE;
+
+static mutex_t			btr_search_enabled_mutex;
+
+/* A dummy variable to fool the compiler */
+UNIV_INTERN ulint		btr_search_this_is_zero = 0;
+
+#ifdef UNIV_SEARCH_PERF_STAT
+UNIV_INTERN ulint		btr_search_n_succ	= 0;
+UNIV_INTERN ulint		btr_search_n_hash_fail	= 0;
+#endif /* UNIV_SEARCH_PERF_STAT */
+
+/* padding to prevent other memory update
+hotspots from residing on the same memory
+cache line as btr_search_latch */
+UNIV_INTERN byte		btr_sea_pad1[64];
+
+/* The latch protecting the adaptive search system: this latch protects the
+(1) positions of records on those pages where a hash index has been built.
+NOTE: It does not protect values of non-ordering fields within a record from
+being updated in-place! We can use fact (1) to perform unique searches to
+indexes. */
+
+/* We will allocate the latch from dynamic memory to get it to the
+same DRAM page as other hotspot semaphores */
+UNIV_INTERN rw_lock_t*		btr_search_latch_temp;
+
+/* padding to prevent other memory update hotspots from residing on
+the same memory cache line */
+UNIV_INTERN byte		btr_sea_pad2[64];
+
+UNIV_INTERN btr_search_sys_t*	btr_search_sys;
+
+/* If the number of records on the page divided by this parameter
+would have been successfully accessed using a hash index, the index
+is then built on the page, assuming the global limit has been reached */
+
+#define BTR_SEARCH_PAGE_BUILD_LIMIT	16
+
+/* The global limit for consecutive potentially successful hash searches,
+before hash index building is started */
+
+#define BTR_SEARCH_BUILD_LIMIT		100
+
+/************************************************************************
+Builds a hash index on a page with the given parameters. If the page already
+has a hash index with different parameters, the old hash index is removed.
+If index is non-NULL, this function checks if n_fields and n_bytes are
+sensible values, and does not build a hash index if not. */
+static
+void
+btr_search_build_page_hash_index(
+/*=============================*/
+	dict_index_t*	index,	/* in: index for which to build, or NULL if
+				not known */
+	buf_block_t*	block,	/* in: index page, s- or x-latched */
+	ulint		n_fields,/* in: hash this many full fields */
+	ulint		n_bytes,/* in: hash this many bytes from the next
+				field */
+	ibool		left_side);/* in: hash for searches from left side? */
+
+/*********************************************************************
+This function should be called before reserving any btr search mutex, if
+the intended operation might add nodes to the search system hash table.
+Because of the latching order, once we have reserved the btr search system
+latch, we cannot allocate a free frame from the buffer pool. Checks that
+there is a free buffer frame allocated for hash table heap in the btr search
+system. If not, allocates a free frames for the heap. This check makes it
+probable that, when have reserved the btr search system latch and we need to
+allocate a new node to the hash table, it will succeed. However, the check
+will not guarantee success. */
+static
+void
+btr_search_check_free_space_in_heap(void)
+/*=====================================*/
+{
+	hash_table_t*	table;
+	mem_heap_t*	heap;
+
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(!rw_lock_own(&btr_search_latch, RW_LOCK_SHARED));
+	ut_ad(!rw_lock_own(&btr_search_latch, RW_LOCK_EX));
+#endif /* UNIV_SYNC_DEBUG */
+
+	table = btr_search_sys->hash_index;
+
+	heap = table->heap;
+
+	/* Note that we peek the value of heap->free_block without reserving
+	the latch: this is ok, because we will not guarantee that there will
+	be enough free space in the hash table. */
+
+	if (heap->free_block == NULL) {
+		buf_block_t*	block = buf_block_alloc(0);
+
+		rw_lock_x_lock(&btr_search_latch);
+
+		if (heap->free_block == NULL) {
+			heap->free_block = block;
+		} else {
+			buf_block_free(block);
+		}
+
+		rw_lock_x_unlock(&btr_search_latch);
+	}
+}
+
+/*********************************************************************
+Creates and initializes the adaptive search system at a database start. */
+UNIV_INTERN
+void
+btr_search_sys_create(
+/*==================*/
+	ulint	hash_size)	/* in: hash index hash table size */
+{
+	/* We allocate the search latch from dynamic memory:
+	see above at the global variable definition */
+
+	btr_search_latch_temp = mem_alloc(sizeof(rw_lock_t));
+
+	rw_lock_create(&btr_search_latch, SYNC_SEARCH_SYS);
+	mutex_create(&btr_search_enabled_mutex, SYNC_SEARCH_SYS_CONF);
+
+	btr_search_sys = mem_alloc(sizeof(btr_search_sys_t));
+
+	btr_search_sys->hash_index = ha_create(hash_size, 0, 0);
+}
+
+/************************************************************************
+Disable the adaptive hash search system and empty the index. */
+UNIV_INTERN
+void
+btr_search_disable(void)
+/*====================*/
+{
+	mutex_enter(&btr_search_enabled_mutex);
+	rw_lock_x_lock(&btr_search_latch);
+
+	btr_search_enabled = FALSE;
+
+	/* Clear all block->is_hashed flags and remove all entries
+	from btr_search_sys->hash_index. */
+	buf_pool_drop_hash_index();
+
+	/* btr_search_enabled_mutex should guarantee this. */
+	ut_ad(!btr_search_enabled);
+
+	rw_lock_x_unlock(&btr_search_latch);
+	mutex_exit(&btr_search_enabled_mutex);
+}
+
+/************************************************************************
+Enable the adaptive hash search system. */
+UNIV_INTERN
+void
+btr_search_enable(void)
+/*====================*/
+{
+	mutex_enter(&btr_search_enabled_mutex);
+	rw_lock_x_lock(&btr_search_latch);
+
+	btr_search_enabled = TRUE;
+
+	rw_lock_x_unlock(&btr_search_latch);
+	mutex_exit(&btr_search_enabled_mutex);
+}
+
+/*********************************************************************
+Creates and initializes a search info struct. */
+UNIV_INTERN
+btr_search_t*
+btr_search_info_create(
+/*===================*/
+				/* out, own: search info struct */
+	mem_heap_t*	heap)	/* in: heap where created */
+{
+	btr_search_t*	info;
+
+	info = mem_heap_alloc(heap, sizeof(btr_search_t));
+
+#ifdef UNIV_DEBUG
+	info->magic_n = BTR_SEARCH_MAGIC_N;
+#endif /* UNIV_DEBUG */
+
+	info->ref_count = 0;
+	info->root_guess = NULL;
+
+	info->hash_analysis = 0;
+	info->n_hash_potential = 0;
+
+	info->last_hash_succ = FALSE;
+
+#ifdef UNIV_SEARCH_PERF_STAT
+	info->n_hash_succ = 0;
+	info->n_hash_fail = 0;
+	info->n_patt_succ = 0;
+	info->n_searches = 0;
+#endif /* UNIV_SEARCH_PERF_STAT */
+
+	/* Set some sensible values */
+	info->n_fields = 1;
+	info->n_bytes = 0;
+
+	info->left_side = TRUE;
+
+	return(info);
+}
+
+/*********************************************************************
+Returns the value of ref_count. The value is protected by
+btr_search_latch. */
+UNIV_INTERN
+ulint
+btr_search_info_get_ref_count(
+/*==========================*/
+				/* out: ref_count value. */
+	btr_search_t*   info)	/* in: search info. */
+{
+	ulint ret;
+
+	ut_ad(info);
+
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(!rw_lock_own(&btr_search_latch, RW_LOCK_SHARED));
+	ut_ad(!rw_lock_own(&btr_search_latch, RW_LOCK_EX));
+#endif /* UNIV_SYNC_DEBUG */
+
+	rw_lock_s_lock(&btr_search_latch);
+	ret = info->ref_count;
+	rw_lock_s_unlock(&btr_search_latch);
+
+	return(ret);
+}
+
+/*************************************************************************
+Updates the search info of an index about hash successes. NOTE that info
+is NOT protected by any semaphore, to save CPU time! Do not assume its fields
+are consistent. */
+static
+void
+btr_search_info_update_hash(
+/*========================*/
+	btr_search_t*	info,	/* in/out: search info */
+	btr_cur_t*	cursor)	/* in: cursor which was just positioned */
+{
+	dict_index_t*	index;
+	ulint		n_unique;
+	int		cmp;
+
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(!rw_lock_own(&btr_search_latch, RW_LOCK_SHARED));
+	ut_ad(!rw_lock_own(&btr_search_latch, RW_LOCK_EX));
+#endif /* UNIV_SYNC_DEBUG */
+
+	index = cursor->index;
+
+	if (dict_index_is_ibuf(index)) {
+		/* So many deletes are performed on an insert buffer tree
+		that we do not consider a hash index useful on it: */
+
+		return;
+	}
+
+	n_unique = dict_index_get_n_unique_in_tree(index);
+
+	if (info->n_hash_potential == 0) {
+
+		goto set_new_recomm;
+	}
+
+	/* Test if the search would have succeeded using the recommended
+	hash prefix */
+
+	if (info->n_fields >= n_unique && cursor->up_match >= n_unique) {
+increment_potential:
+		info->n_hash_potential++;
+
+		return;
+	}
+
+	cmp = ut_pair_cmp(info->n_fields, info->n_bytes,
+			  cursor->low_match, cursor->low_bytes);
+
+	if (info->left_side ? cmp <= 0 : cmp > 0) {
+
+		goto set_new_recomm;
+	}
+
+	cmp = ut_pair_cmp(info->n_fields, info->n_bytes,
+			  cursor->up_match, cursor->up_bytes);
+
+	if (info->left_side ? cmp <= 0 : cmp > 0) {
+
+		goto increment_potential;
+	}
+
+set_new_recomm:
+	/* We have to set a new recommendation; skip the hash analysis
+	for a while to avoid unnecessary CPU time usage when there is no
+	chance for success */
+
+	info->hash_analysis = 0;
+
+	cmp = ut_pair_cmp(cursor->up_match, cursor->up_bytes,
+			  cursor->low_match, cursor->low_bytes);
+	if (cmp == 0) {
+		info->n_hash_potential = 0;
+
+		/* For extra safety, we set some sensible values here */
+
+		info->n_fields = 1;
+		info->n_bytes = 0;
+
+		info->left_side = TRUE;
+
+	} else if (cmp > 0) {
+		info->n_hash_potential = 1;
+
+		if (cursor->up_match >= n_unique) {
+
+			info->n_fields = n_unique;
+			info->n_bytes = 0;
+
+		} else if (cursor->low_match < cursor->up_match) {
+
+			info->n_fields = cursor->low_match + 1;
+			info->n_bytes = 0;
+		} else {
+			info->n_fields = cursor->low_match;
+			info->n_bytes = cursor->low_bytes + 1;
+		}
+
+		info->left_side = TRUE;
+	} else {
+		info->n_hash_potential = 1;
+
+		if (cursor->low_match >= n_unique) {
+
+			info->n_fields = n_unique;
+			info->n_bytes = 0;
+
+		} else if (cursor->low_match > cursor->up_match) {
+
+			info->n_fields = cursor->up_match + 1;
+			info->n_bytes = 0;
+		} else {
+			info->n_fields = cursor->up_match;
+			info->n_bytes = cursor->up_bytes + 1;
+		}
+
+		info->left_side = FALSE;
+	}
+}
+
+/*************************************************************************
+Updates the block search info on hash successes. NOTE that info and
+block->n_hash_helps, n_fields, n_bytes, side are NOT protected by any
+semaphore, to save CPU time! Do not assume the fields are consistent. */
+static
+ibool
+btr_search_update_block_hash_info(
+/*==============================*/
+				/* out: TRUE if building a (new) hash index on
+				the block is recommended */
+	btr_search_t*	info,	/* in: search info */
+	buf_block_t*	block,	/* in: buffer block */
+	btr_cur_t*	cursor __attribute__((unused)))
+				/* in: cursor */
+{
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(!rw_lock_own(&btr_search_latch, RW_LOCK_SHARED));
+	ut_ad(!rw_lock_own(&btr_search_latch, RW_LOCK_EX));
+	ut_ad(rw_lock_own(&block->lock, RW_LOCK_SHARED)
+	      || rw_lock_own(&block->lock, RW_LOCK_EX));
+#endif /* UNIV_SYNC_DEBUG */
+	ut_ad(cursor);
+
+	info->last_hash_succ = FALSE;
+
+	ut_a(buf_block_state_valid(block));
+	ut_ad(info->magic_n == BTR_SEARCH_MAGIC_N);
+
+	if ((block->n_hash_helps > 0)
+	    && (info->n_hash_potential > 0)
+	    && (block->n_fields == info->n_fields)
+	    && (block->n_bytes == info->n_bytes)
+	    && (block->left_side == info->left_side)) {
+
+		if ((block->is_hashed)
+		    && (block->curr_n_fields == info->n_fields)
+		    && (block->curr_n_bytes == info->n_bytes)
+		    && (block->curr_left_side == info->left_side)) {
+
+			/* The search would presumably have succeeded using
+			the hash index */
+
+			info->last_hash_succ = TRUE;
+		}
+
+		block->n_hash_helps++;
+	} else {
+		block->n_hash_helps = 1;
+		block->n_fields = info->n_fields;
+		block->n_bytes = info->n_bytes;
+		block->left_side = info->left_side;
+	}
+
+#ifdef UNIV_DEBUG
+	if (cursor->index->table->does_not_fit_in_memory) {
+		block->n_hash_helps = 0;
+	}
+#endif /* UNIV_DEBUG */
+
+	if ((block->n_hash_helps > page_get_n_recs(block->frame)
+	     / BTR_SEARCH_PAGE_BUILD_LIMIT)
+	    && (info->n_hash_potential >= BTR_SEARCH_BUILD_LIMIT)) {
+
+		if ((!block->is_hashed)
+		    || (block->n_hash_helps
+			> 2 * page_get_n_recs(block->frame))
+		    || (block->n_fields != block->curr_n_fields)
+		    || (block->n_bytes != block->curr_n_bytes)
+		    || (block->left_side != block->curr_left_side)) {
+
+			/* Build a new hash index on the page */
+
+			return(TRUE);
+		}
+	}
+
+	return(FALSE);
+}
+
+/*************************************************************************
+Updates a hash node reference when it has been unsuccessfully used in a
+search which could have succeeded with the used hash parameters. This can
+happen because when building a hash index for a page, we do not check
+what happens at page boundaries, and therefore there can be misleading
+hash nodes. Also, collisions in the fold value can lead to misleading
+references. This function lazily fixes these imperfections in the hash
+index. */
+static
+void
+btr_search_update_hash_ref(
+/*=======================*/
+	btr_search_t*	info,	/* in: search info */
+	buf_block_t*	block,	/* in: buffer block where cursor positioned */
+	btr_cur_t*	cursor)	/* in: cursor */
+{
+	ulint	fold;
+	rec_t*	rec;
+	dulint	index_id;
+
+	ut_ad(cursor->flag == BTR_CUR_HASH_FAIL);
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(rw_lock_own(&btr_search_latch, RW_LOCK_EX));
+	ut_ad(rw_lock_own(&(block->lock), RW_LOCK_SHARED)
+	      || rw_lock_own(&(block->lock), RW_LOCK_EX));
+#endif /* UNIV_SYNC_DEBUG */
+	ut_ad(page_align(btr_cur_get_rec(cursor))
+	      == buf_block_get_frame(block));
+
+	if (!block->is_hashed) {
+
+		return;
+	}
+
+	ut_a(block->index == cursor->index);
+	ut_a(!dict_index_is_ibuf(cursor->index));
+
+	if ((info->n_hash_potential > 0)
+	    && (block->curr_n_fields == info->n_fields)
+	    && (block->curr_n_bytes == info->n_bytes)
+	    && (block->curr_left_side == info->left_side)) {
+		mem_heap_t*	heap		= NULL;
+		ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+		rec_offs_init(offsets_);
+
+		rec = btr_cur_get_rec(cursor);
+
+		if (!page_rec_is_user_rec(rec)) {
+
+			return;
+		}
+
+		index_id = cursor->index->id;
+		fold = rec_fold(rec,
+				rec_get_offsets(rec, cursor->index, offsets_,
+						ULINT_UNDEFINED, &heap),
+				block->curr_n_fields,
+				block->curr_n_bytes, index_id);
+		if (UNIV_LIKELY_NULL(heap)) {
+			mem_heap_free(heap);
+		}
+#ifdef UNIV_SYNC_DEBUG
+		ut_ad(rw_lock_own(&btr_search_latch, RW_LOCK_EX));
+#endif /* UNIV_SYNC_DEBUG */
+
+		ha_insert_for_fold(btr_search_sys->hash_index, fold,
+				   block, rec);
+	}
+}
+
+/*************************************************************************
+Updates the search info. */
+UNIV_INTERN
+void
+btr_search_info_update_slow(
+/*========================*/
+	btr_search_t*	info,	/* in/out: search info */
+	btr_cur_t*	cursor)	/* in: cursor which was just positioned */
+{
+	buf_block_t*	block;
+	ibool		build_index;
+	ulint*		params;
+	ulint*		params2;
+
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(!rw_lock_own(&btr_search_latch, RW_LOCK_SHARED));
+	ut_ad(!rw_lock_own(&btr_search_latch, RW_LOCK_EX));
+#endif /* UNIV_SYNC_DEBUG */
+
+	block = btr_cur_get_block(cursor);
+
+	/* NOTE that the following two function calls do NOT protect
+	info or block->n_fields etc. with any semaphore, to save CPU time!
+	We cannot assume the fields are consistent when we return from
+	those functions! */
+
+	btr_search_info_update_hash(info, cursor);
+
+	build_index = btr_search_update_block_hash_info(info, block, cursor);
+
+	if (build_index || (cursor->flag == BTR_CUR_HASH_FAIL)) {
+
+		btr_search_check_free_space_in_heap();
+	}
+
+	if (cursor->flag == BTR_CUR_HASH_FAIL) {
+		/* Update the hash node reference, if appropriate */
+
+#ifdef UNIV_SEARCH_PERF_STAT
+		btr_search_n_hash_fail++;
+#endif /* UNIV_SEARCH_PERF_STAT */
+
+		rw_lock_x_lock(&btr_search_latch);
+
+		btr_search_update_hash_ref(info, block, cursor);
+
+		rw_lock_x_unlock(&btr_search_latch);
+	}
+
+	if (build_index) {
+		/* Note that since we did not protect block->n_fields etc.
+		with any semaphore, the values can be inconsistent. We have
+		to check inside the function call that they make sense. We
+		also malloc an array and store the values there to make sure
+		the compiler does not let the function call parameters change
+		inside the called function. It might be that the compiler
+		would optimize the call just to pass pointers to block. */
+
+		params = mem_alloc(3 * sizeof(ulint));
+		params[0] = block->n_fields;
+		params[1] = block->n_bytes;
+		params[2] = block->left_side;
+
+		/* Make sure the compiler cannot deduce the values and do
+		optimizations */
+
+		params2 = params + btr_search_this_is_zero;
+
+		btr_search_build_page_hash_index(cursor->index,
+						 block,
+						 params2[0],
+						 params2[1],
+						 params2[2]);
+		mem_free(params);
+	}
+}
+
+/**********************************************************************
+Checks if a guessed position for a tree cursor is right. Note that if
+mode is PAGE_CUR_LE, which is used in inserts, and the function returns
+TRUE, then cursor->up_match and cursor->low_match both have sensible values. */
+static
+ibool
+btr_search_check_guess(
+/*===================*/
+				/* out: TRUE if success */
+	btr_cur_t*	cursor,	/* in: guessed cursor position */
+	ibool		can_only_compare_to_cursor_rec,
+				/* in: if we do not have a latch on the page
+				of cursor, but only a latch on
+				btr_search_latch, then ONLY the columns
+				of the record UNDER the cursor are
+				protected, not the next or previous record
+				in the chain: we cannot look at the next or
+				previous record to check our guess! */
+	const dtuple_t*	tuple,	/* in: data tuple */
+	ulint		mode,	/* in: PAGE_CUR_L, PAGE_CUR_LE, PAGE_CUR_G,
+				or PAGE_CUR_GE */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	rec_t*		rec;
+	ulint		n_unique;
+	ulint		match;
+	ulint		bytes;
+	int		cmp;
+	mem_heap_t*	heap		= NULL;
+	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint*		offsets		= offsets_;
+	ibool		success		= FALSE;
+	rec_offs_init(offsets_);
+
+	n_unique = dict_index_get_n_unique_in_tree(cursor->index);
+
+	rec = btr_cur_get_rec(cursor);
+
+	ut_ad(page_rec_is_user_rec(rec));
+
+	match = 0;
+	bytes = 0;
+
+	offsets = rec_get_offsets(rec, cursor->index, offsets,
+				  n_unique, &heap);
+	cmp = page_cmp_dtuple_rec_with_match(tuple, rec,
+					     offsets, &match, &bytes);
+
+	if (mode == PAGE_CUR_GE) {
+		if (cmp == 1) {
+			goto exit_func;
+		}
+
+		cursor->up_match = match;
+
+		if (match >= n_unique) {
+			success = TRUE;
+			goto exit_func;
+		}
+	} else if (mode == PAGE_CUR_LE) {
+		if (cmp == -1) {
+			goto exit_func;
+		}
+
+		cursor->low_match = match;
+
+	} else if (mode == PAGE_CUR_G) {
+		if (cmp != -1) {
+			goto exit_func;
+		}
+	} else if (mode == PAGE_CUR_L) {
+		if (cmp != 1) {
+			goto exit_func;
+		}
+	}
+
+	if (can_only_compare_to_cursor_rec) {
+		/* Since we could not determine if our guess is right just by
+		looking at the record under the cursor, return FALSE */
+		goto exit_func;
+	}
+
+	match = 0;
+	bytes = 0;
+
+	if ((mode == PAGE_CUR_G) || (mode == PAGE_CUR_GE)) {
+		rec_t*	prev_rec;
+
+		ut_ad(!page_rec_is_infimum(rec));
+
+		prev_rec = page_rec_get_prev(rec);
+
+		if (page_rec_is_infimum(prev_rec)) {
+			success = btr_page_get_prev(page_align(prev_rec), mtr)
+				== FIL_NULL;
+
+			goto exit_func;
+		}
+
+		offsets = rec_get_offsets(prev_rec, cursor->index, offsets,
+					  n_unique, &heap);
+		cmp = page_cmp_dtuple_rec_with_match(tuple, prev_rec,
+						     offsets, &match, &bytes);
+		if (mode == PAGE_CUR_GE) {
+			success = cmp == 1;
+		} else {
+			success = cmp != -1;
+		}
+
+		goto exit_func;
+	} else {
+		rec_t*	next_rec;
+
+		ut_ad(!page_rec_is_supremum(rec));
+
+		next_rec = page_rec_get_next(rec);
+
+		if (page_rec_is_supremum(next_rec)) {
+			if (btr_page_get_next(page_align(next_rec), mtr)
+			    == FIL_NULL) {
+
+				cursor->up_match = 0;
+				success = TRUE;
+			}
+
+			goto exit_func;
+		}
+
+		offsets = rec_get_offsets(next_rec, cursor->index, offsets,
+					  n_unique, &heap);
+		cmp = page_cmp_dtuple_rec_with_match(tuple, next_rec,
+						     offsets, &match, &bytes);
+		if (mode == PAGE_CUR_LE) {
+			success = cmp == -1;
+			cursor->up_match = match;
+		} else {
+			success = cmp != 1;
+		}
+	}
+exit_func:
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+	return(success);
+}
+
+/**********************************************************************
+Tries to guess the right search position based on the hash search info
+of the index. Note that if mode is PAGE_CUR_LE, which is used in inserts,
+and the function returns TRUE, then cursor->up_match and cursor->low_match
+both have sensible values. */
+UNIV_INTERN
+ibool
+btr_search_guess_on_hash(
+/*=====================*/
+					/* out: TRUE if succeeded */
+	dict_index_t*	index,		/* in: index */
+	btr_search_t*	info,		/* in: index search info */
+	const dtuple_t*	tuple,		/* in: logical record */
+	ulint		mode,		/* in: PAGE_CUR_L, ... */
+	ulint		latch_mode,	/* in: BTR_SEARCH_LEAF, ...;
+					NOTE that only if has_search_latch
+					is 0, we will have a latch set on
+					the cursor page, otherwise we assume
+					the caller uses his search latch
+					to protect the record! */
+	btr_cur_t*	cursor,		/* out: tree cursor */
+	ulint		has_search_latch,/* in: latch mode the caller
+					currently has on btr_search_latch:
+					RW_S_LATCH, RW_X_LATCH, or 0 */
+	mtr_t*		mtr)		/* in: mtr */
+{
+	buf_block_t*	block;
+	rec_t*		rec;
+	ulint		fold;
+	dulint		index_id;
+#ifdef notdefined
+	btr_cur_t	cursor2;
+	btr_pcur_t	pcur;
+#endif
+	ut_ad(index && info && tuple && cursor && mtr);
+	ut_ad((latch_mode == BTR_SEARCH_LEAF)
+	      || (latch_mode == BTR_MODIFY_LEAF));
+
+	/* Note that, for efficiency, the struct info may not be protected by
+	any latch here! */
+
+	if (UNIV_UNLIKELY(info->n_hash_potential == 0)) {
+
+		return(FALSE);
+	}
+
+	cursor->n_fields = info->n_fields;
+	cursor->n_bytes = info->n_bytes;
+
+	if (UNIV_UNLIKELY(dtuple_get_n_fields(tuple)
+			  < cursor->n_fields + (cursor->n_bytes > 0))) {
+
+		return(FALSE);
+	}
+
+	index_id = index->id;
+
+#ifdef UNIV_SEARCH_PERF_STAT
+	info->n_hash_succ++;
+#endif
+	fold = dtuple_fold(tuple, cursor->n_fields, cursor->n_bytes, index_id);
+
+	cursor->fold = fold;
+	cursor->flag = BTR_CUR_HASH;
+
+	if (UNIV_LIKELY(!has_search_latch)) {
+		rw_lock_s_lock(&btr_search_latch);
+
+		if (UNIV_UNLIKELY(!btr_search_enabled)) {
+			goto failure_unlock;
+		}
+	}
+
+	ut_ad(rw_lock_get_writer(&btr_search_latch) != RW_LOCK_EX);
+	ut_ad(rw_lock_get_reader_count(&btr_search_latch) > 0);
+
+	rec = ha_search_and_get_data(btr_search_sys->hash_index, fold);
+
+	if (UNIV_UNLIKELY(!rec)) {
+		goto failure_unlock;
+	}
+
+	block = buf_block_align(rec);
+
+	if (UNIV_LIKELY(!has_search_latch)) {
+
+		if (UNIV_UNLIKELY(
+			    !buf_page_get_known_nowait(latch_mode, block,
+						       BUF_MAKE_YOUNG,
+						       __FILE__, __LINE__,
+						       mtr))) {
+			goto failure_unlock;
+		}
+
+		rw_lock_s_unlock(&btr_search_latch);
+
+		buf_block_dbg_add_level(block, SYNC_TREE_NODE_FROM_HASH);
+	}
+
+	if (UNIV_UNLIKELY(buf_block_get_state(block) != BUF_BLOCK_FILE_PAGE)) {
+		ut_ad(buf_block_get_state(block) == BUF_BLOCK_REMOVE_HASH);
+
+		if (UNIV_LIKELY(!has_search_latch)) {
+
+			btr_leaf_page_release(block, latch_mode, mtr);
+		}
+
+		goto failure;
+	}
+
+	ut_ad(page_rec_is_user_rec(rec));
+
+	btr_cur_position(index, rec, block, cursor);
+
+	/* Check the validity of the guess within the page */
+
+	/* If we only have the latch on btr_search_latch, not on the
+	page, it only protects the columns of the record the cursor
+	is positioned on. We cannot look at the next of the previous
+	record to determine if our guess for the cursor position is
+	right. */
+	if (UNIV_EXPECT
+	    (ut_dulint_cmp(index_id, btr_page_get_index_id(block->frame)), 0)
+	    || !btr_search_check_guess(cursor,
+				       has_search_latch,
+				       tuple, mode, mtr)) {
+		if (UNIV_LIKELY(!has_search_latch)) {
+			btr_leaf_page_release(block, latch_mode, mtr);
+		}
+
+		goto failure;
+	}
+
+	if (UNIV_LIKELY(info->n_hash_potential < BTR_SEARCH_BUILD_LIMIT + 5)) {
+
+		info->n_hash_potential++;
+	}
+
+#ifdef notdefined
+	/* These lines of code can be used in a debug version to check
+	the correctness of the searched cursor position: */
+
+	info->last_hash_succ = FALSE;
+
+	/* Currently, does not work if the following fails: */
+	ut_ad(!has_search_latch);
+
+	btr_leaf_page_release(block, latch_mode, mtr);
+
+	btr_cur_search_to_nth_level(index, 0, tuple, mode, latch_mode,
+				    &cursor2, 0, mtr);
+	if (mode == PAGE_CUR_GE
+	    && page_rec_is_supremum(btr_cur_get_rec(&cursor2))) {
+
+		/* If mode is PAGE_CUR_GE, then the binary search
+		in the index tree may actually take us to the supremum
+		of the previous page */
+
+		info->last_hash_succ = FALSE;
+
+		btr_pcur_open_on_user_rec(index, tuple, mode, latch_mode,
+					  &pcur, mtr);
+		ut_ad(btr_pcur_get_rec(&pcur) == btr_cur_get_rec(cursor));
+	} else {
+		ut_ad(btr_cur_get_rec(&cursor2) == btr_cur_get_rec(cursor));
+	}
+
+	/* NOTE that it is theoretically possible that the above assertions
+	fail if the page of the cursor gets removed from the buffer pool
+	meanwhile! Thus it might not be a bug. */
+#endif
+	info->last_hash_succ = TRUE;
+
+#ifdef UNIV_SEARCH_PERF_STAT
+	btr_search_n_succ++;
+#endif
+	if (UNIV_LIKELY(!has_search_latch)
+	    && buf_page_peek_if_too_old(&block->page)) {
+
+		buf_page_make_young(&block->page);
+	}
+
+	/* Increment the page get statistics though we did not really
+	fix the page: for user info only */
+
+	buf_pool->n_page_gets++;
+
+	return(TRUE);
+
+	/*-------------------------------------------*/
+failure_unlock:
+	if (UNIV_LIKELY(!has_search_latch)) {
+		rw_lock_s_unlock(&btr_search_latch);
+	}
+failure:
+	cursor->flag = BTR_CUR_HASH_FAIL;
+
+#ifdef UNIV_SEARCH_PERF_STAT
+	info->n_hash_fail++;
+
+	if (info->n_hash_succ > 0) {
+		info->n_hash_succ--;
+	}
+#endif
+	info->last_hash_succ = FALSE;
+
+	return(FALSE);
+}
+
+/************************************************************************
+Drops a page hash index. */
+UNIV_INTERN
+void
+btr_search_drop_page_hash_index(
+/*============================*/
+	buf_block_t*	block)	/* in: block containing index page,
+				s- or x-latched, or an index page
+				for which we know that
+				block->buf_fix_count == 0 */
+{
+	hash_table_t*		table;
+	ulint			n_fields;
+	ulint			n_bytes;
+	const page_t*		page;
+	const rec_t*		rec;
+	ulint			fold;
+	ulint			prev_fold;
+	dulint			index_id;
+	ulint			n_cached;
+	ulint			n_recs;
+	ulint*			folds;
+	ulint			i;
+	mem_heap_t*		heap;
+	const dict_index_t*	index;
+	ulint*			offsets;
+
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(!rw_lock_own(&btr_search_latch, RW_LOCK_SHARED));
+	ut_ad(!rw_lock_own(&btr_search_latch, RW_LOCK_EX));
+#endif /* UNIV_SYNC_DEBUG */
+
+retry:
+	rw_lock_s_lock(&btr_search_latch);
+	page = block->frame;
+
+	if (UNIV_LIKELY(!block->is_hashed)) {
+
+		rw_lock_s_unlock(&btr_search_latch);
+
+		return;
+	}
+
+	table = btr_search_sys->hash_index;
+
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(rw_lock_own(&(block->lock), RW_LOCK_SHARED)
+	      || rw_lock_own(&(block->lock), RW_LOCK_EX)
+	      || (block->page.buf_fix_count == 0));
+#endif /* UNIV_SYNC_DEBUG */
+
+	n_fields = block->curr_n_fields;
+	n_bytes = block->curr_n_bytes;
+	index = block->index;
+	ut_a(!dict_index_is_ibuf(index));
+
+	/* NOTE: The fields of block must not be accessed after
+	releasing btr_search_latch, as the index page might only
+	be s-latched! */
+
+	rw_lock_s_unlock(&btr_search_latch);
+
+	ut_a(n_fields + n_bytes > 0);
+
+	n_recs = page_get_n_recs(page);
+
+	/* Calculate and cache fold values into an array for fast deletion
+	from the hash index */
+
+	folds = mem_alloc(n_recs * sizeof(ulint));
+
+	n_cached = 0;
+
+	rec = page_get_infimum_rec(page);
+	rec = page_rec_get_next_low(rec, page_is_comp(page));
+
+	index_id = btr_page_get_index_id(page);
+
+	ut_a(0 == ut_dulint_cmp(index_id, index->id));
+
+	prev_fold = 0;
+
+	heap = NULL;
+	offsets = NULL;
+
+	while (!page_rec_is_supremum(rec)) {
+		offsets = rec_get_offsets(rec, index, offsets,
+					  n_fields + (n_bytes > 0), &heap);
+		ut_a(rec_offs_n_fields(offsets) == n_fields + (n_bytes > 0));
+		fold = rec_fold(rec, offsets, n_fields, n_bytes, index_id);
+
+		if (fold == prev_fold && prev_fold != 0) {
+
+			goto next_rec;
+		}
+
+		/* Remove all hash nodes pointing to this page from the
+		hash chain */
+
+		folds[n_cached] = fold;
+		n_cached++;
+next_rec:
+		rec = page_rec_get_next_low(rec, page_rec_is_comp(rec));
+		prev_fold = fold;
+	}
+
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+
+	rw_lock_x_lock(&btr_search_latch);
+
+	if (UNIV_UNLIKELY(!block->is_hashed)) {
+		/* Someone else has meanwhile dropped the hash index */
+
+		goto cleanup;
+	}
+
+	ut_a(block->index == index);
+
+	if (UNIV_UNLIKELY(block->curr_n_fields != n_fields)
+	    || UNIV_UNLIKELY(block->curr_n_bytes != n_bytes)) {
+
+		/* Someone else has meanwhile built a new hash index on the
+		page, with different parameters */
+
+		rw_lock_x_unlock(&btr_search_latch);
+
+		mem_free(folds);
+		goto retry;
+	}
+
+	for (i = 0; i < n_cached; i++) {
+
+		ha_remove_all_nodes_to_page(table, folds[i], page);
+	}
+
+	ut_a(index->search_info->ref_count > 0);
+	index->search_info->ref_count--;
+
+	block->is_hashed = FALSE;
+	block->index = NULL;
+	
+cleanup:
+#if defined UNIV_AHI_DEBUG || defined UNIV_DEBUG
+	if (UNIV_UNLIKELY(block->n_pointers)) {
+		/* Corruption */
+		ut_print_timestamp(stderr);
+		fprintf(stderr,
+			"  InnoDB: Corruption of adaptive hash index."
+			" After dropping\n"
+			"InnoDB: the hash index to a page of %s,"
+			" still %lu hash nodes remain.\n",
+			index->name, (ulong) block->n_pointers);
+		rw_lock_x_unlock(&btr_search_latch);
+
+		btr_search_validate();
+	} else {
+		rw_lock_x_unlock(&btr_search_latch);
+	}
+#else /* UNIV_AHI_DEBUG || UNIV_DEBUG */
+	rw_lock_x_unlock(&btr_search_latch);
+#endif /* UNIV_AHI_DEBUG || UNIV_DEBUG */
+
+	mem_free(folds);
+}
+
+/************************************************************************
+Drops a page hash index when a page is freed from a fseg to the file system.
+Drops possible hash index if the page happens to be in the buffer pool. */
+UNIV_INTERN
+void
+btr_search_drop_page_hash_when_freed(
+/*=================================*/
+	ulint	space,		/* in: space id */
+	ulint	zip_size,	/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	ulint	page_no)	/* in: page number */
+{
+	buf_block_t*	block;
+	mtr_t		mtr;
+
+	if (!buf_page_peek_if_search_hashed(space, page_no)) {
+
+		return;
+	}
+
+	mtr_start(&mtr);
+
+	/* We assume that if the caller has a latch on the page, then the
+	caller has already dropped the hash index for the page, and we never
+	get here. Therefore we can acquire the s-latch to the page without
+	having to fear a deadlock. */
+
+	block = buf_page_get_gen(space, zip_size, page_no, RW_S_LATCH, NULL,
+				BUF_GET_IF_IN_POOL, __FILE__, __LINE__,
+				&mtr);
+	/* Because the buffer pool mutex was released by
+	buf_page_peek_if_search_hashed(), it is possible that the
+	block was removed from the buffer pool by another thread
+	before buf_page_get_gen() got a chance to acquire the buffer
+	pool mutex again.  Thus, we must check for a NULL return. */
+
+	if (UNIV_LIKELY(block != NULL)) {
+
+		buf_block_dbg_add_level(block, SYNC_TREE_NODE_FROM_HASH);
+
+		btr_search_drop_page_hash_index(block);
+	}
+
+	mtr_commit(&mtr);
+}
+
+/************************************************************************
+Builds a hash index on a page with the given parameters. If the page already
+has a hash index with different parameters, the old hash index is removed.
+If index is non-NULL, this function checks if n_fields and n_bytes are
+sensible values, and does not build a hash index if not. */
+static
+void
+btr_search_build_page_hash_index(
+/*=============================*/
+	dict_index_t*	index,	/* in: index for which to build */
+	buf_block_t*	block,	/* in: index page, s- or x-latched */
+	ulint		n_fields,/* in: hash this many full fields */
+	ulint		n_bytes,/* in: hash this many bytes from the next
+				field */
+	ibool		left_side)/* in: hash for searches from left side? */
+{
+	hash_table_t*	table;
+	page_t*		page;
+	rec_t*		rec;
+	rec_t*		next_rec;
+	ulint		fold;
+	ulint		next_fold;
+	dulint		index_id;
+	ulint		n_cached;
+	ulint		n_recs;
+	ulint*		folds;
+	rec_t**		recs;
+	ulint		i;
+	mem_heap_t*	heap		= NULL;
+	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint*		offsets		= offsets_;
+	rec_offs_init(offsets_);
+
+	ut_ad(index);
+	ut_a(!dict_index_is_ibuf(index));
+
+	table = btr_search_sys->hash_index;
+	page = buf_block_get_frame(block);
+
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(!rw_lock_own(&btr_search_latch, RW_LOCK_EX));
+	ut_ad(rw_lock_own(&(block->lock), RW_LOCK_SHARED)
+	      || rw_lock_own(&(block->lock), RW_LOCK_EX));
+#endif /* UNIV_SYNC_DEBUG */
+
+	rw_lock_s_lock(&btr_search_latch);
+
+	if (block->is_hashed && ((block->curr_n_fields != n_fields)
+				 || (block->curr_n_bytes != n_bytes)
+				 || (block->curr_left_side != left_side))) {
+
+		rw_lock_s_unlock(&btr_search_latch);
+
+		btr_search_drop_page_hash_index(block);
+	} else {
+		rw_lock_s_unlock(&btr_search_latch);
+	}
+
+	n_recs = page_get_n_recs(page);
+
+	if (n_recs == 0) {
+
+		return;
+	}
+
+	/* Check that the values for hash index build are sensible */
+
+	if (n_fields + n_bytes == 0) {
+
+		return;
+	}
+
+	if (dict_index_get_n_unique_in_tree(index) < n_fields
+	    || (dict_index_get_n_unique_in_tree(index) == n_fields
+		&& n_bytes > 0)) {
+		return;
+	}
+
+	/* Calculate and cache fold values and corresponding records into
+	an array for fast insertion to the hash index */
+
+	folds = mem_alloc(n_recs * sizeof(ulint));
+	recs = mem_alloc(n_recs * sizeof(rec_t*));
+
+	n_cached = 0;
+
+	index_id = btr_page_get_index_id(page);
+
+	rec = page_rec_get_next(page_get_infimum_rec(page));
+
+	offsets = rec_get_offsets(rec, index, offsets,
+				  n_fields + (n_bytes > 0), &heap);
+
+	if (!page_rec_is_supremum(rec)) {
+		ut_a(n_fields <= rec_offs_n_fields(offsets));
+
+		if (n_bytes > 0) {
+			ut_a(n_fields < rec_offs_n_fields(offsets));
+		}
+	}
+
+	fold = rec_fold(rec, offsets, n_fields, n_bytes, index_id);
+
+	if (left_side) {
+
+		folds[n_cached] = fold;
+		recs[n_cached] = rec;
+		n_cached++;
+	}
+
+	for (;;) {
+		next_rec = page_rec_get_next(rec);
+
+		if (page_rec_is_supremum(next_rec)) {
+
+			if (!left_side) {
+
+				folds[n_cached] = fold;
+				recs[n_cached] = rec;
+				n_cached++;
+			}
+
+			break;
+		}
+
+		offsets = rec_get_offsets(next_rec, index, offsets,
+					  n_fields + (n_bytes > 0), &heap);
+		next_fold = rec_fold(next_rec, offsets, n_fields,
+				     n_bytes, index_id);
+
+		if (fold != next_fold) {
+			/* Insert an entry into the hash index */
+
+			if (left_side) {
+
+				folds[n_cached] = next_fold;
+				recs[n_cached] = next_rec;
+				n_cached++;
+			} else {
+				folds[n_cached] = fold;
+				recs[n_cached] = rec;
+				n_cached++;
+			}
+		}
+
+		rec = next_rec;
+		fold = next_fold;
+	}
+
+	btr_search_check_free_space_in_heap();
+
+	rw_lock_x_lock(&btr_search_latch);
+
+	if (UNIV_UNLIKELY(!btr_search_enabled)) {
+		goto exit_func;
+	}
+
+	if (block->is_hashed && ((block->curr_n_fields != n_fields)
+				 || (block->curr_n_bytes != n_bytes)
+				 || (block->curr_left_side != left_side))) {
+		goto exit_func;
+	}
+
+	/* This counter is decremented every time we drop page
+	hash index entries and is incremented here. Since we can
+	rebuild hash index for a page that is already hashed, we
+	have to take care not to increment the counter in that
+	case. */
+	if (!block->is_hashed) {
+		index->search_info->ref_count++;
+	}
+
+	block->is_hashed = TRUE;
+	block->n_hash_helps = 0;
+
+	block->curr_n_fields = n_fields;
+	block->curr_n_bytes = n_bytes;
+	block->curr_left_side = left_side;
+	block->index = index;
+
+	for (i = 0; i < n_cached; i++) {
+
+		ha_insert_for_fold(table, folds[i], block, recs[i]);
+	}
+
+exit_func:
+	rw_lock_x_unlock(&btr_search_latch);
+
+	mem_free(folds);
+	mem_free(recs);
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+}
+
+/************************************************************************
+Moves or deletes hash entries for moved records. If new_page is already hashed,
+then the hash index for page, if any, is dropped. If new_page is not hashed,
+and page is hashed, then a new hash index is built to new_page with the same
+parameters as page (this often happens when a page is split). */
+UNIV_INTERN
+void
+btr_search_move_or_delete_hash_entries(
+/*===================================*/
+	buf_block_t*	new_block,	/* in: records are copied
+					to this page */
+	buf_block_t*	block,		/* in: index page from which
+					records were copied, and the
+					copied records will be deleted
+					from this page */
+	dict_index_t*	index)		/* in: record descriptor */
+{
+	ulint	n_fields;
+	ulint	n_bytes;
+	ibool	left_side;
+
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(rw_lock_own(&(block->lock), RW_LOCK_EX));
+	ut_ad(rw_lock_own(&(new_block->lock), RW_LOCK_EX));
+#endif /* UNIV_SYNC_DEBUG */
+	ut_a(!new_block->is_hashed || new_block->index == index);
+	ut_a(!block->is_hashed || block->index == index);
+	ut_a(!(new_block->is_hashed || block->is_hashed)
+	     || !dict_index_is_ibuf(index));
+
+	rw_lock_s_lock(&btr_search_latch);
+
+	if (new_block->is_hashed) {
+
+		rw_lock_s_unlock(&btr_search_latch);
+
+		btr_search_drop_page_hash_index(block);
+
+		return;
+	}
+
+	if (block->is_hashed) {
+
+		n_fields = block->curr_n_fields;
+		n_bytes = block->curr_n_bytes;
+		left_side = block->curr_left_side;
+
+		new_block->n_fields = block->curr_n_fields;
+		new_block->n_bytes = block->curr_n_bytes;
+		new_block->left_side = left_side;
+
+		rw_lock_s_unlock(&btr_search_latch);
+
+		ut_a(n_fields + n_bytes > 0);
+
+		btr_search_build_page_hash_index(index, new_block, n_fields,
+						 n_bytes, left_side);
+		ut_ad(n_fields == block->curr_n_fields);
+		ut_ad(n_bytes == block->curr_n_bytes);
+		ut_ad(left_side == block->curr_left_side);
+		return;
+	}
+
+	rw_lock_s_unlock(&btr_search_latch);
+}
+
+/************************************************************************
+Updates the page hash index when a single record is deleted from a page. */
+UNIV_INTERN
+void
+btr_search_update_hash_on_delete(
+/*=============================*/
+	btr_cur_t*	cursor)	/* in: cursor which was positioned on the
+				record to delete using btr_cur_search_...,
+				the record is not yet deleted */
+{
+	hash_table_t*	table;
+	buf_block_t*	block;
+	rec_t*		rec;
+	ulint		fold;
+	dulint		index_id;
+	ibool		found;
+	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+	mem_heap_t*	heap		= NULL;
+	rec_offs_init(offsets_);
+
+	rec = btr_cur_get_rec(cursor);
+
+	block = btr_cur_get_block(cursor);
+
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(rw_lock_own(&(block->lock), RW_LOCK_EX));
+#endif /* UNIV_SYNC_DEBUG */
+
+	if (!block->is_hashed) {
+
+		return;
+	}
+
+	ut_a(block->index == cursor->index);
+	ut_a(block->curr_n_fields + block->curr_n_bytes > 0);
+	ut_a(!dict_index_is_ibuf(cursor->index));
+
+	table = btr_search_sys->hash_index;
+
+	index_id = cursor->index->id;
+	fold = rec_fold(rec, rec_get_offsets(rec, cursor->index, offsets_,
+					     ULINT_UNDEFINED, &heap),
+			block->curr_n_fields, block->curr_n_bytes, index_id);
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+	rw_lock_x_lock(&btr_search_latch);
+
+	found = ha_search_and_delete_if_found(table, fold, rec);
+
+	rw_lock_x_unlock(&btr_search_latch);
+}
+
+/************************************************************************
+Updates the page hash index when a single record is inserted on a page. */
+UNIV_INTERN
+void
+btr_search_update_hash_node_on_insert(
+/*==================================*/
+	btr_cur_t*	cursor)	/* in: cursor which was positioned to the
+				place to insert using btr_cur_search_...,
+				and the new record has been inserted next
+				to the cursor */
+{
+	hash_table_t*	table;
+	buf_block_t*	block;
+	rec_t*		rec;
+
+	rec = btr_cur_get_rec(cursor);
+
+	block = btr_cur_get_block(cursor);
+
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(rw_lock_own(&(block->lock), RW_LOCK_EX));
+#endif /* UNIV_SYNC_DEBUG */
+
+	if (!block->is_hashed) {
+
+		return;
+	}
+
+	ut_a(block->index == cursor->index);
+	ut_a(!dict_index_is_ibuf(cursor->index));
+
+	rw_lock_x_lock(&btr_search_latch);
+
+	if ((cursor->flag == BTR_CUR_HASH)
+	    && (cursor->n_fields == block->curr_n_fields)
+	    && (cursor->n_bytes == block->curr_n_bytes)
+	    && !block->curr_left_side) {
+
+		table = btr_search_sys->hash_index;
+
+		ha_search_and_update_if_found(table, cursor->fold, rec,
+					      block, page_rec_get_next(rec));
+
+		rw_lock_x_unlock(&btr_search_latch);
+	} else {
+		rw_lock_x_unlock(&btr_search_latch);
+
+		btr_search_update_hash_on_insert(cursor);
+	}
+}
+
+/************************************************************************
+Updates the page hash index when a single record is inserted on a page. */
+UNIV_INTERN
+void
+btr_search_update_hash_on_insert(
+/*=============================*/
+	btr_cur_t*	cursor)	/* in: cursor which was positioned to the
+				place to insert using btr_cur_search_...,
+				and the new record has been inserted next
+				to the cursor */
+{
+	hash_table_t*	table;
+	buf_block_t*	block;
+	rec_t*		rec;
+	rec_t*		ins_rec;
+	rec_t*		next_rec;
+	dulint		index_id;
+	ulint		fold;
+	ulint		ins_fold;
+	ulint		next_fold = 0; /* remove warning (??? bug ???) */
+	ulint		n_fields;
+	ulint		n_bytes;
+	ibool		left_side;
+	ibool		locked		= FALSE;
+	mem_heap_t*	heap		= NULL;
+	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint*		offsets		= offsets_;
+	rec_offs_init(offsets_);
+
+	table = btr_search_sys->hash_index;
+
+	btr_search_check_free_space_in_heap();
+
+	rec = btr_cur_get_rec(cursor);
+
+	block = btr_cur_get_block(cursor);
+
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(rw_lock_own(&(block->lock), RW_LOCK_EX));
+#endif /* UNIV_SYNC_DEBUG */
+
+	if (!block->is_hashed) {
+
+		return;
+	}
+
+	ut_a(block->index == cursor->index);
+	ut_a(!dict_index_is_ibuf(cursor->index));
+
+	index_id = cursor->index->id;
+
+	n_fields = block->curr_n_fields;
+	n_bytes = block->curr_n_bytes;
+	left_side = block->curr_left_side;
+
+	ins_rec = page_rec_get_next(rec);
+	next_rec = page_rec_get_next(ins_rec);
+
+	offsets = rec_get_offsets(ins_rec, cursor->index, offsets,
+				  ULINT_UNDEFINED, &heap);
+	ins_fold = rec_fold(ins_rec, offsets, n_fields, n_bytes, index_id);
+
+	if (!page_rec_is_supremum(next_rec)) {
+		offsets = rec_get_offsets(next_rec, cursor->index, offsets,
+					  n_fields + (n_bytes > 0), &heap);
+		next_fold = rec_fold(next_rec, offsets, n_fields,
+				     n_bytes, index_id);
+	}
+
+	if (!page_rec_is_infimum(rec)) {
+		offsets = rec_get_offsets(rec, cursor->index, offsets,
+					  n_fields + (n_bytes > 0), &heap);
+		fold = rec_fold(rec, offsets, n_fields, n_bytes, index_id);
+	} else {
+		if (left_side) {
+
+			rw_lock_x_lock(&btr_search_latch);
+
+			locked = TRUE;
+
+			ha_insert_for_fold(table, ins_fold, block, ins_rec);
+		}
+
+		goto check_next_rec;
+	}
+
+	if (fold != ins_fold) {
+
+		if (!locked) {
+
+			rw_lock_x_lock(&btr_search_latch);
+
+			locked = TRUE;
+		}
+
+		if (!left_side) {
+			ha_insert_for_fold(table, fold, block, rec);
+		} else {
+			ha_insert_for_fold(table, ins_fold, block, ins_rec);
+		}
+	}
+
+check_next_rec:
+	if (page_rec_is_supremum(next_rec)) {
+
+		if (!left_side) {
+
+			if (!locked) {
+				rw_lock_x_lock(&btr_search_latch);
+
+				locked = TRUE;
+			}
+
+			ha_insert_for_fold(table, ins_fold, block, ins_rec);
+		}
+
+		goto function_exit;
+	}
+
+	if (ins_fold != next_fold) {
+
+		if (!locked) {
+
+			rw_lock_x_lock(&btr_search_latch);
+
+			locked = TRUE;
+		}
+
+		if (!left_side) {
+
+			ha_insert_for_fold(table, ins_fold, block, ins_rec);
+			/*
+			fputs("Hash insert for ", stderr);
+			dict_index_name_print(stderr, cursor->index);
+			fprintf(stderr, " fold %lu\n", ins_fold);
+			*/
+		} else {
+			ha_insert_for_fold(table, next_fold, block, next_rec);
+		}
+	}
+
+function_exit:
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+	if (locked) {
+		rw_lock_x_unlock(&btr_search_latch);
+	}
+}
+
+/************************************************************************
+Validates the search system. */
+UNIV_INTERN
+ibool
+btr_search_validate(void)
+/*=====================*/
+				/* out: TRUE if ok */
+{
+	ha_node_t*	node;
+	ulint		n_page_dumps	= 0;
+	ibool		ok		= TRUE;
+	ulint		i;
+	ulint		cell_count;
+	mem_heap_t*	heap		= NULL;
+	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint*		offsets		= offsets_;
+
+	/* How many cells to check before temporarily releasing
+	btr_search_latch. */
+	ulint		chunk_size = 10000;
+
+	rec_offs_init(offsets_);
+
+	rw_lock_x_lock(&btr_search_latch);
+	buf_pool_mutex_enter();
+
+	cell_count = hash_get_n_cells(btr_search_sys->hash_index);
+
+	for (i = 0; i < cell_count; i++) {
+		/* We release btr_search_latch every once in a while to
+		give other queries a chance to run. */
+		if ((i != 0) && ((i % chunk_size) == 0)) {
+			buf_pool_mutex_exit();
+			rw_lock_x_unlock(&btr_search_latch);
+			os_thread_yield();
+			rw_lock_x_lock(&btr_search_latch);
+			buf_pool_mutex_enter();
+		}
+
+		node = hash_get_nth_cell(btr_search_sys->hash_index, i)->node;
+
+		for (; node != NULL; node = node->next) {
+			const buf_block_t*	block
+				= buf_block_align(node->data);
+			const buf_block_t*	hash_block;
+
+			if (UNIV_LIKELY(buf_block_get_state(block)
+					== BUF_BLOCK_FILE_PAGE)) {
+
+				/* The space and offset are only valid
+				for file blocks.  It is possible that
+				the block is being freed
+				(BUF_BLOCK_REMOVE_HASH, see the
+				assertion and the comment below) */
+				hash_block = buf_block_hash_get(
+					buf_block_get_space(block),
+					buf_block_get_page_no(block));
+			} else {
+				hash_block = NULL;
+			}
+
+			if (hash_block) {
+				ut_a(hash_block == block);
+			} else {
+				/* When a block is being freed,
+				buf_LRU_search_and_free_block() first
+				removes the block from
+				buf_pool->page_hash by calling
+				buf_LRU_block_remove_hashed_page().
+				After that, it invokes
+				btr_search_drop_page_hash_index() to
+				remove the block from
+				btr_search_sys->hash_index. */
+
+				ut_a(buf_block_get_state(block)
+				     == BUF_BLOCK_REMOVE_HASH);
+			}
+
+			ut_a(!dict_index_is_ibuf(block->index));
+
+			offsets = rec_get_offsets((const rec_t*) node->data,
+						  block->index, offsets,
+						  block->curr_n_fields
+						  + (block->curr_n_bytes > 0),
+						  &heap);
+
+			if (!block->is_hashed || node->fold
+			    != rec_fold((rec_t*)(node->data),
+					offsets,
+					block->curr_n_fields,
+					block->curr_n_bytes,
+					btr_page_get_index_id(block->frame))) {
+				const page_t*	page = block->frame;
+
+				ok = FALSE;
+				ut_print_timestamp(stderr);
+
+				fprintf(stderr,
+					"  InnoDB: Error in an adaptive hash"
+					" index pointer to page %lu\n"
+					"InnoDB: ptr mem address %p"
+					" index id %lu %lu,"
+					" node fold %lu, rec fold %lu\n",
+					(ulong) page_get_page_no(page),
+					node->data,
+					(ulong) ut_dulint_get_high(
+						btr_page_get_index_id(page)),
+					(ulong) ut_dulint_get_low(
+						btr_page_get_index_id(page)),
+					(ulong) node->fold,
+					(ulong) rec_fold((rec_t*)(node->data),
+							 offsets,
+							 block->curr_n_fields,
+							 block->curr_n_bytes,
+							 btr_page_get_index_id(
+								 page)));
+
+				fputs("InnoDB: Record ", stderr);
+				rec_print_new(stderr, (rec_t*)node->data,
+					      offsets);
+				fprintf(stderr, "\nInnoDB: on that page."
+					" Page mem address %p, is hashed %lu,"
+					" n fields %lu, n bytes %lu\n"
+					"InnoDB: side %lu\n",
+					(void*) page, (ulong) block->is_hashed,
+					(ulong) block->curr_n_fields,
+					(ulong) block->curr_n_bytes,
+					(ulong) block->curr_left_side);
+
+				if (n_page_dumps < 20) {
+					buf_page_print(page, 0);
+					n_page_dumps++;
+				}
+			}
+		}
+	}
+
+	for (i = 0; i < cell_count; i += chunk_size) {
+		ulint end_index = ut_min(i + chunk_size - 1, cell_count - 1);
+
+		/* We release btr_search_latch every once in a while to
+		give other queries a chance to run. */
+		if (i != 0) {
+			buf_pool_mutex_exit();
+			rw_lock_x_unlock(&btr_search_latch);
+			os_thread_yield();
+			rw_lock_x_lock(&btr_search_latch);
+			buf_pool_mutex_enter();
+		}
+
+		if (!ha_validate(btr_search_sys->hash_index, i, end_index)) {
+			ok = FALSE;
+		}
+	}
+
+	buf_pool_mutex_exit();
+	rw_lock_x_unlock(&btr_search_latch);
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+
+	return(ok);
+}
diff --git a/storage/xtradb/buf/buf0buddy.c b/storage/xtradb/buf/buf0buddy.c
new file mode 100644
index 00000000000..99ee7554132
--- /dev/null
+++ b/storage/xtradb/buf/buf0buddy.c
@@ -0,0 +1,682 @@
+/*****************************************************************************
+
+Copyright (c) 2006, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Binary buddy allocator for compressed pages
+
+Created December 2006 by Marko Makela
+*******************************************************/
+
+#define THIS_MODULE
+#include "buf0buddy.h"
+#ifdef UNIV_NONINL
+# include "buf0buddy.ic"
+#endif
+#undef THIS_MODULE
+#include "buf0buf.h"
+#include "buf0lru.h"
+#include "buf0flu.h"
+#include "page0zip.h"
+
+/* Statistic counters */
+
+#ifdef UNIV_DEBUG
+/** Number of frames allocated from the buffer pool to the buddy system.
+Protected by buf_pool_mutex. */
+static ulint buf_buddy_n_frames;
+#endif /* UNIV_DEBUG */
+/** Statistics of the buddy system, indexed by block size.
+Protected by buf_pool_mutex. */
+UNIV_INTERN buf_buddy_stat_t buf_buddy_stat[BUF_BUDDY_SIZES + 1];
+
+/**************************************************************************
+Get the offset of the buddy of a compressed page frame. */
+UNIV_INLINE
+byte*
+buf_buddy_get(
+/*==========*/
+			/* out: the buddy relative of page */
+	byte*	page,	/* in: compressed page */
+	ulint	size)	/* in: page size in bytes */
+{
+	ut_ad(ut_is_2pow(size));
+	ut_ad(size >= BUF_BUDDY_LOW);
+	ut_ad(size < BUF_BUDDY_HIGH);
+	ut_ad(!ut_align_offset(page, size));
+
+	if (((ulint) page) & size) {
+		return(page - size);
+	} else {
+		return(page + size);
+	}
+}
+
+/**************************************************************************
+Add a block to the head of the appropriate buddy free list. */
+UNIV_INLINE
+void
+buf_buddy_add_to_free(
+/*==================*/
+	buf_page_t*	bpage,	/* in,own: block to be freed */
+	ulint		i)	/* in: index of buf_pool->zip_free[] */
+{
+#ifdef UNIV_DEBUG_VALGRIND
+	buf_page_t*	b  = UT_LIST_GET_FIRST(buf_pool->zip_free[i]);
+
+	if (b) UNIV_MEM_VALID(b, BUF_BUDDY_LOW << i);
+#endif /* UNIV_DEBUG_VALGRIND */
+
+	ut_ad(buf_pool->zip_free[i].start != bpage);
+	UT_LIST_ADD_FIRST(list, buf_pool->zip_free[i], bpage);
+
+#ifdef UNIV_DEBUG_VALGRIND
+	if (b) UNIV_MEM_FREE(b, BUF_BUDDY_LOW << i);
+	UNIV_MEM_ASSERT_AND_FREE(bpage, BUF_BUDDY_LOW << i);
+#endif /* UNIV_DEBUG_VALGRIND */
+}
+
+/**************************************************************************
+Remove a block from the appropriate buddy free list. */
+UNIV_INLINE
+void
+buf_buddy_remove_from_free(
+/*=======================*/
+	buf_page_t*	bpage,	/* in: block to be removed */
+	ulint		i)	/* in: index of buf_pool->zip_free[] */
+{
+#ifdef UNIV_DEBUG_VALGRIND
+	buf_page_t*	prev = UT_LIST_GET_PREV(list, bpage);
+	buf_page_t*	next = UT_LIST_GET_NEXT(list, bpage);
+
+	if (prev) UNIV_MEM_VALID(prev, BUF_BUDDY_LOW << i);
+	if (next) UNIV_MEM_VALID(next, BUF_BUDDY_LOW << i);
+
+	ut_ad(!prev || buf_page_get_state(prev) == BUF_BLOCK_ZIP_FREE);
+	ut_ad(!next || buf_page_get_state(next) == BUF_BLOCK_ZIP_FREE);
+#endif /* UNIV_DEBUG_VALGRIND */
+
+	ut_ad(buf_page_get_state(bpage) == BUF_BLOCK_ZIP_FREE);
+	UT_LIST_REMOVE(list, buf_pool->zip_free[i], bpage);
+
+#ifdef UNIV_DEBUG_VALGRIND
+	if (prev) UNIV_MEM_FREE(prev, BUF_BUDDY_LOW << i);
+	if (next) UNIV_MEM_FREE(next, BUF_BUDDY_LOW << i);
+#endif /* UNIV_DEBUG_VALGRIND */
+}
+
+/**************************************************************************
+Try to allocate a block from buf_pool->zip_free[]. */
+static
+void*
+buf_buddy_alloc_zip(
+/*================*/
+			/* out: allocated block, or NULL
+			if buf_pool->zip_free[] was empty */
+	ulint	i)	/* in: index of buf_pool->zip_free[] */
+{
+	buf_page_t*	bpage;
+
+	ut_ad(buf_pool_mutex_own());
+	ut_a(i < BUF_BUDDY_SIZES);
+
+#if defined UNIV_DEBUG && !defined UNIV_DEBUG_VALGRIND
+	/* Valgrind would complain about accessing free memory. */
+	UT_LIST_VALIDATE(list, buf_page_t, buf_pool->zip_free[i]);
+#endif /* UNIV_DEBUG && !UNIV_DEBUG_VALGRIND */
+	bpage = UT_LIST_GET_LAST(buf_pool->zip_free[i]);
+
+	if (bpage) {
+		UNIV_MEM_VALID(bpage, BUF_BUDDY_LOW << i);
+		ut_a(buf_page_get_state(bpage) == BUF_BLOCK_ZIP_FREE);
+
+		buf_buddy_remove_from_free(bpage, i);
+	} else if (i + 1 < BUF_BUDDY_SIZES) {
+		/* Attempt to split. */
+		bpage = buf_buddy_alloc_zip(i + 1);
+
+		if (bpage) {
+			buf_page_t*	buddy = (buf_page_t*)
+				(((char*) bpage) + (BUF_BUDDY_LOW << i));
+
+			ut_ad(!buf_pool_contains_zip(buddy));
+			ut_d(memset(buddy, i, BUF_BUDDY_LOW << i));
+			buddy->state = BUF_BLOCK_ZIP_FREE;
+			buf_buddy_add_to_free(buddy, i);
+		}
+	}
+
+#ifdef UNIV_DEBUG
+	if (bpage) {
+		memset(bpage, ~i, BUF_BUDDY_LOW << i);
+	}
+#endif /* UNIV_DEBUG */
+
+	UNIV_MEM_ALLOC(bpage, BUF_BUDDY_SIZES << i);
+
+	return(bpage);
+}
+
+/**************************************************************************
+Deallocate a buffer frame of UNIV_PAGE_SIZE. */
+static
+void
+buf_buddy_block_free(
+/*=================*/
+	void*	buf)	/* in: buffer frame to deallocate */
+{
+	const ulint	fold	= BUF_POOL_ZIP_FOLD_PTR(buf);
+	buf_page_t*	bpage;
+	buf_block_t*	block;
+
+	ut_ad(buf_pool_mutex_own());
+	ut_ad(!mutex_own(&buf_pool_zip_mutex));
+	ut_a(!ut_align_offset(buf, UNIV_PAGE_SIZE));
+
+	HASH_SEARCH(hash, buf_pool->zip_hash, fold, buf_page_t*, bpage,
+		    ut_ad(buf_page_get_state(bpage) == BUF_BLOCK_MEMORY
+			  && bpage->in_zip_hash && !bpage->in_page_hash),
+		    ((buf_block_t*) bpage)->frame == buf);
+	ut_a(bpage);
+	ut_a(buf_page_get_state(bpage) == BUF_BLOCK_MEMORY);
+	ut_ad(!bpage->in_page_hash);
+	ut_ad(bpage->in_zip_hash);
+	ut_d(bpage->in_zip_hash = FALSE);
+	HASH_DELETE(buf_page_t, hash, buf_pool->zip_hash, fold, bpage);
+
+	ut_d(memset(buf, 0, UNIV_PAGE_SIZE));
+	UNIV_MEM_INVALID(buf, UNIV_PAGE_SIZE);
+
+	block = (buf_block_t*) bpage;
+	mutex_enter(&block->mutex);
+	buf_LRU_block_free_non_file_page(block);
+	mutex_exit(&block->mutex);
+
+	ut_ad(buf_buddy_n_frames > 0);
+	ut_d(buf_buddy_n_frames--);
+}
+
+/**************************************************************************
+Allocate a buffer block to the buddy allocator. */
+static
+void
+buf_buddy_block_register(
+/*=====================*/
+	buf_block_t*	block)	/* in: buffer frame to allocate */
+{
+	const ulint	fold = BUF_POOL_ZIP_FOLD(block);
+	ut_ad(buf_pool_mutex_own());
+	ut_ad(!mutex_own(&buf_pool_zip_mutex));
+
+	buf_block_set_state(block, BUF_BLOCK_MEMORY);
+
+	ut_a(block->frame);
+	ut_a(!ut_align_offset(block->frame, UNIV_PAGE_SIZE));
+
+	ut_ad(!block->page.in_page_hash);
+	ut_ad(!block->page.in_zip_hash);
+	ut_d(block->page.in_zip_hash = TRUE);
+	HASH_INSERT(buf_page_t, hash, buf_pool->zip_hash, fold, &block->page);
+
+	ut_d(buf_buddy_n_frames++);
+}
+
+/**************************************************************************
+Allocate a block from a bigger object. */
+static
+void*
+buf_buddy_alloc_from(
+/*=================*/
+				/* out: allocated block */
+	void*		buf,	/* in: a block that is free to use */
+	ulint		i,	/* in: index of buf_pool->zip_free[] */
+	ulint		j)	/* in: size of buf as an index
+				of buf_pool->zip_free[] */
+{
+	ulint	offs	= BUF_BUDDY_LOW << j;
+	ut_ad(j <= BUF_BUDDY_SIZES);
+	ut_ad(j >= i);
+	ut_ad(!ut_align_offset(buf, offs));
+
+	/* Add the unused parts of the block to the free lists. */
+	while (j > i) {
+		buf_page_t*	bpage;
+
+		offs >>= 1;
+		j--;
+
+		bpage = (buf_page_t*) ((byte*) buf + offs);
+		ut_d(memset(bpage, j, BUF_BUDDY_LOW << j));
+		bpage->state = BUF_BLOCK_ZIP_FREE;
+#if defined UNIV_DEBUG && !defined UNIV_DEBUG_VALGRIND
+		/* Valgrind would complain about accessing free memory. */
+		UT_LIST_VALIDATE(list, buf_page_t, buf_pool->zip_free[j]);
+#endif /* UNIV_DEBUG && !UNIV_DEBUG_VALGRIND */
+		buf_buddy_add_to_free(bpage, j);
+	}
+
+	return(buf);
+}
+
+/**************************************************************************
+Allocate a block.  The thread calling this function must hold
+buf_pool_mutex and must not hold buf_pool_zip_mutex or any block->mutex.
+The buf_pool_mutex may only be released and reacquired if lru != NULL. */
+UNIV_INTERN
+void*
+buf_buddy_alloc_low(
+/*================*/
+			/* out: allocated block,
+			possibly NULL if lru==NULL */
+	ulint	i,	/* in: index of buf_pool->zip_free[],
+			or BUF_BUDDY_SIZES */
+	ibool*	lru)	/* in: pointer to a variable that will be assigned
+			TRUE if storage was allocated from the LRU list
+			and buf_pool_mutex was temporarily released,
+			or NULL if the LRU list should not be used */
+{
+	buf_block_t*	block;
+
+	ut_ad(buf_pool_mutex_own());
+	ut_ad(!mutex_own(&buf_pool_zip_mutex));
+
+	if (i < BUF_BUDDY_SIZES) {
+		/* Try to allocate from the buddy system. */
+		block = buf_buddy_alloc_zip(i);
+
+		if (block) {
+
+			goto func_exit;
+		}
+	}
+
+	/* Try allocating from the buf_pool->free list. */
+	block = buf_LRU_get_free_only();
+
+	if (block) {
+
+		goto alloc_big;
+	}
+
+	if (!lru) {
+
+		return(NULL);
+	}
+
+	/* Try replacing an uncompressed page in the buffer pool. */
+	buf_pool_mutex_exit();
+	block = buf_LRU_get_free_block(0);
+	*lru = TRUE;
+	buf_pool_mutex_enter();
+
+alloc_big:
+	buf_buddy_block_register(block);
+
+	block = buf_buddy_alloc_from(block->frame, i, BUF_BUDDY_SIZES);
+
+func_exit:
+	buf_buddy_stat[i].used++;
+	return(block);
+}
+
+/**************************************************************************
+Try to relocate the control block of a compressed page. */
+static
+ibool
+buf_buddy_relocate_block(
+/*=====================*/
+				/* out: TRUE if relocated */
+	buf_page_t*	bpage,	/* in: block to relocate */
+	buf_page_t*	dpage)	/* in: free block to relocate to */
+{
+	buf_page_t*	b;
+
+	ut_ad(buf_pool_mutex_own());
+
+	switch (buf_page_get_state(bpage)) {
+	case BUF_BLOCK_ZIP_FREE:
+	case BUF_BLOCK_NOT_USED:
+	case BUF_BLOCK_READY_FOR_USE:
+	case BUF_BLOCK_FILE_PAGE:
+	case BUF_BLOCK_MEMORY:
+	case BUF_BLOCK_REMOVE_HASH:
+		ut_error;
+	case BUF_BLOCK_ZIP_DIRTY:
+		/* Cannot relocate dirty pages. */
+		return(FALSE);
+
+	case BUF_BLOCK_ZIP_PAGE:
+		break;
+	}
+
+	mutex_enter(&buf_pool_zip_mutex);
+
+	if (!buf_page_can_relocate(bpage)) {
+		mutex_exit(&buf_pool_zip_mutex);
+		return(FALSE);
+	}
+
+	buf_relocate(bpage, dpage);
+	ut_d(bpage->state = BUF_BLOCK_ZIP_FREE);
+
+	/* relocate buf_pool->zip_clean */
+	b = UT_LIST_GET_PREV(list, dpage);
+	UT_LIST_REMOVE(list, buf_pool->zip_clean, dpage);
+
+	if (b) {
+		UT_LIST_INSERT_AFTER(list, buf_pool->zip_clean, b, dpage);
+	} else {
+		UT_LIST_ADD_FIRST(list, buf_pool->zip_clean, dpage);
+	}
+
+	mutex_exit(&buf_pool_zip_mutex);
+	return(TRUE);
+}
+
+/**************************************************************************
+Try to relocate a block. */
+static
+ibool
+buf_buddy_relocate(
+/*===============*/
+			/* out: TRUE if relocated */
+	void*	src,	/* in: block to relocate */
+	void*	dst,	/* in: free block to relocate to */
+	ulint	i)	/* in: index of buf_pool->zip_free[] */
+{
+	buf_page_t*	bpage;
+	const ulint	size	= BUF_BUDDY_LOW << i;
+	ullint		usec	= ut_time_us(NULL);
+
+	ut_ad(buf_pool_mutex_own());
+	ut_ad(!mutex_own(&buf_pool_zip_mutex));
+	ut_ad(!ut_align_offset(src, size));
+	ut_ad(!ut_align_offset(dst, size));
+	UNIV_MEM_ASSERT_W(dst, size);
+
+	/* We assume that all memory from buf_buddy_alloc()
+	is used for either compressed pages or buf_page_t
+	objects covering compressed pages. */
+
+	/* We look inside the allocated objects returned by
+	buf_buddy_alloc() and assume that anything of
+	PAGE_ZIP_MIN_SIZE or larger is a compressed page that contains
+	a valid space_id and page_no in the page header.  Should the
+	fields be invalid, we will be unable to relocate the block.
+	We also assume that anything that fits sizeof(buf_page_t)
+	actually is a properly initialized buf_page_t object. */
+
+	if (size >= PAGE_ZIP_MIN_SIZE) {
+		/* This is a compressed page. */
+		mutex_t*	mutex;
+
+		/* The src block may be split into smaller blocks,
+		some of which may be free.  Thus, the
+		mach_read_from_4() calls below may attempt to read
+		from free memory.  The memory is "owned" by the buddy
+		allocator (and it has been allocated from the buffer
+		pool), so there is nothing wrong about this.  The
+		mach_read_from_4() calls here will only trigger bogus
+		Valgrind memcheck warnings in UNIV_DEBUG_VALGRIND builds. */
+		bpage = buf_page_hash_get(
+			mach_read_from_4((const byte*) src
+					 + FIL_PAGE_ARCH_LOG_NO_OR_SPACE_ID),
+			mach_read_from_4((const byte*) src
+					 + FIL_PAGE_OFFSET));
+
+		if (!bpage || bpage->zip.data != src) {
+			/* The block has probably been freshly
+			allocated by buf_LRU_get_free_block() but not
+			added to buf_pool->page_hash yet.  Obviously,
+			it cannot be relocated. */
+
+			return(FALSE);
+		}
+
+		if (page_zip_get_size(&bpage->zip) != size) {
+			/* The block is of different size.  We would
+			have to relocate all blocks covered by src.
+			For the sake of simplicity, give up. */
+			ut_ad(page_zip_get_size(&bpage->zip) < size);
+
+			return(FALSE);
+		}
+
+		/* The block must have been allocated, but it may
+		contain uninitialized data. */
+		UNIV_MEM_ASSERT_W(src, size);
+
+		mutex = buf_page_get_mutex(bpage);
+
+		mutex_enter(mutex);
+
+		if (buf_page_can_relocate(bpage)) {
+			/* Relocate the compressed page. */
+			ut_a(bpage->zip.data == src);
+			memcpy(dst, src, size);
+			bpage->zip.data = dst;
+			mutex_exit(mutex);
+success:
+			UNIV_MEM_INVALID(src, size);
+			{
+				buf_buddy_stat_t*	buddy_stat
+					= &buf_buddy_stat[i];
+				buddy_stat->relocated++;
+				buddy_stat->relocated_usec
+					+= ut_time_us(NULL) - usec;
+			}
+			return(TRUE);
+		}
+
+		mutex_exit(mutex);
+	} else if (i == buf_buddy_get_slot(sizeof(buf_page_t))) {
+		/* This must be a buf_page_t object. */
+		UNIV_MEM_ASSERT_RW(src, size);
+		if (buf_buddy_relocate_block(src, dst)) {
+
+			goto success;
+		}
+	}
+
+	return(FALSE);
+}
+
+/**************************************************************************
+Deallocate a block. */
+UNIV_INTERN
+void
+buf_buddy_free_low(
+/*===============*/
+	void*	buf,	/* in: block to be freed, must not be
+			pointed to by the buffer pool */
+	ulint	i)	/* in: index of buf_pool->zip_free[] */
+{
+	buf_page_t*	bpage;
+	buf_page_t*	buddy;
+
+	ut_ad(buf_pool_mutex_own());
+	ut_ad(!mutex_own(&buf_pool_zip_mutex));
+	ut_ad(i <= BUF_BUDDY_SIZES);
+	ut_ad(buf_buddy_stat[i].used > 0);
+
+	buf_buddy_stat[i].used--;
+recombine:
+	UNIV_MEM_ASSERT_AND_ALLOC(buf, BUF_BUDDY_LOW << i);
+	ut_d(((buf_page_t*) buf)->state = BUF_BLOCK_ZIP_FREE);
+
+	if (i == BUF_BUDDY_SIZES) {
+		buf_buddy_block_free(buf);
+		return;
+	}
+
+	ut_ad(i < BUF_BUDDY_SIZES);
+	ut_ad(buf == ut_align_down(buf, BUF_BUDDY_LOW << i));
+	ut_ad(!buf_pool_contains_zip(buf));
+
+	/* Try to combine adjacent blocks. */
+
+	buddy = (buf_page_t*) buf_buddy_get(((byte*) buf), BUF_BUDDY_LOW << i);
+
+#ifndef UNIV_DEBUG_VALGRIND
+	/* Valgrind would complain about accessing free memory. */
+
+	if (buddy->state != BUF_BLOCK_ZIP_FREE) {
+
+		goto buddy_nonfree;
+	}
+
+	/* The field buddy->state can only be trusted for free blocks.
+	If buddy->state == BUF_BLOCK_ZIP_FREE, the block is free if
+	it is in the free list. */
+#endif /* !UNIV_DEBUG_VALGRIND */
+
+	for (bpage = UT_LIST_GET_FIRST(buf_pool->zip_free[i]); bpage; ) {
+		UNIV_MEM_VALID(bpage, BUF_BUDDY_LOW << i);
+		ut_ad(buf_page_get_state(bpage) == BUF_BLOCK_ZIP_FREE);
+
+		if (bpage == buddy) {
+buddy_free:
+			/* The buddy is free: recombine */
+			buf_buddy_remove_from_free(bpage, i);
+buddy_free2:
+			ut_ad(buf_page_get_state(buddy) == BUF_BLOCK_ZIP_FREE);
+			ut_ad(!buf_pool_contains_zip(buddy));
+			i++;
+			buf = ut_align_down(buf, BUF_BUDDY_LOW << i);
+
+			goto recombine;
+		}
+
+		ut_a(bpage != buf);
+
+		{
+			buf_page_t*	next = UT_LIST_GET_NEXT(list, bpage);
+			UNIV_MEM_ASSERT_AND_FREE(bpage, BUF_BUDDY_LOW << i);
+			bpage = next;
+		}
+	}
+
+#ifndef UNIV_DEBUG_VALGRIND
+buddy_nonfree:
+	/* Valgrind would complain about accessing free memory. */
+	ut_d(UT_LIST_VALIDATE(list, buf_page_t, buf_pool->zip_free[i]));
+#endif /* UNIV_DEBUG_VALGRIND */
+
+	/* The buddy is not free. Is there a free block of this size? */
+	bpage = UT_LIST_GET_FIRST(buf_pool->zip_free[i]);
+
+	if (bpage) {
+		/* Remove the block from the free list, because a successful
+		buf_buddy_relocate() will overwrite bpage->list. */
+
+		UNIV_MEM_VALID(bpage, BUF_BUDDY_LOW << i);
+		buf_buddy_remove_from_free(bpage, i);
+
+		/* Try to relocate the buddy of buf to the free block. */
+		if (buf_buddy_relocate(buddy, bpage, i)) {
+
+			ut_d(buddy->state = BUF_BLOCK_ZIP_FREE);
+			goto buddy_free2;
+		}
+
+		buf_buddy_add_to_free(bpage, i);
+
+		/* Try to relocate the buddy of the free block to buf. */
+		buddy = (buf_page_t*) buf_buddy_get(((byte*) bpage),
+						    BUF_BUDDY_LOW << i);
+
+#if defined UNIV_DEBUG && !defined UNIV_DEBUG_VALGRIND
+		{
+			const buf_page_t* b;
+
+			/* The buddy must not be (completely) free, because
+			we always recombine adjacent free blocks.
+			(Parts of the buddy can be free in
+			buf_pool->zip_free[j] with j < i.)*/
+			for (b = UT_LIST_GET_FIRST(buf_pool->zip_free[i]);
+			     b; b = UT_LIST_GET_NEXT(list, b)) {
+
+				ut_a(b != buddy);
+			}
+		}
+#endif /* UNIV_DEBUG && !UNIV_DEBUG_VALGRIND */
+
+		if (buf_buddy_relocate(buddy, buf, i)) {
+
+			buf = bpage;
+			UNIV_MEM_VALID(bpage, BUF_BUDDY_LOW << i);
+			ut_d(buddy->state = BUF_BLOCK_ZIP_FREE);
+			goto buddy_free;
+		}
+	}
+
+	/* Free the block to the buddy list. */
+	bpage = buf;
+#ifdef UNIV_DEBUG
+	if (i < buf_buddy_get_slot(PAGE_ZIP_MIN_SIZE)) {
+		/* This area has most likely been allocated for at
+		least one compressed-only block descriptor.  Check
+		that there are no live objects in the area.  This is
+		not a complete check: it may yield false positives as
+		well as false negatives.  Also, due to buddy blocks
+		being recombined, it is possible (although unlikely)
+		that this branch is never reached. */
+
+		char* c;
+
+# ifndef UNIV_DEBUG_VALGRIND
+		/* Valgrind would complain about accessing
+		uninitialized memory.  Besides, Valgrind performs a
+		more exhaustive check, at every memory access. */
+		const buf_page_t* b = buf;
+		const buf_page_t* const b_end = (buf_page_t*)
+			((char*) b + (BUF_BUDDY_LOW << i));
+
+		for (; b < b_end; b++) {
+			/* Avoid false positives (and cause false
+			negatives) by checking for b->space < 1000. */
+
+			if ((b->state == BUF_BLOCK_ZIP_PAGE
+			     || b->state == BUF_BLOCK_ZIP_DIRTY)
+			    && b->space > 0 && b->space < 1000) {
+				fprintf(stderr,
+					"buddy dirty %p %u (%u,%u) %p,%lu\n",
+					(void*) b,
+					b->state, b->space, b->offset,
+					buf, i);
+			}
+		}
+# endif /* !UNIV_DEBUG_VALGRIND */
+
+		/* Scramble the block.  This should make any pointers
+		invalid and trigger a segmentation violation.  Because
+		the scrambling can be reversed, it may be possible to
+		track down the object pointing to the freed data by
+		dereferencing the unscrambled bpage->LRU or
+		bpage->list pointers. */
+		for (c = (char*) buf + (BUF_BUDDY_LOW << i);
+		     c-- > (char*) buf; ) {
+			*c = ~*c ^ i;
+		}
+	} else {
+		/* Fill large blocks with a constant pattern. */
+		memset(bpage, i, BUF_BUDDY_LOW << i);
+	}
+#endif /* UNIV_DEBUG */
+	bpage->state = BUF_BLOCK_ZIP_FREE;
+	buf_buddy_add_to_free(bpage, i);
+}
diff --git a/storage/xtradb/buf/buf0buf.c b/storage/xtradb/buf/buf0buf.c
new file mode 100644
index 00000000000..67013780ac6
--- /dev/null
+++ b/storage/xtradb/buf/buf0buf.c
@@ -0,0 +1,3920 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+Copyright (c) 2008, Google Inc.
+
+Portions of this file contain modifications contributed and copyrighted by
+Google, Inc. Those modifications are gratefully acknowledged and are described
+briefly in the InnoDB documentation. The contributions by Google are
+incorporated with their permission, and subject to the conditions contained in
+the file COPYING.Google.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The database buffer buf_pool
+
+Created 11/5/1995 Heikki Tuuri
+*******************************************************/
+
+#include "buf0buf.h"
+
+#ifdef UNIV_NONINL
+#include "buf0buf.ic"
+#endif
+
+#include "buf0buddy.h"
+#include "mem0mem.h"
+#include "btr0btr.h"
+#include "fil0fil.h"
+#include "lock0lock.h"
+#include "btr0sea.h"
+#include "ibuf0ibuf.h"
+#include "dict0dict.h"
+#include "log0recv.h"
+#include "trx0undo.h"
+#include "srv0srv.h"
+#include "page0zip.h"
+
+/*
+		IMPLEMENTATION OF THE BUFFER POOL
+		=================================
+
+Performance improvement:
+------------------------
+Thread scheduling in NT may be so slow that the OS wait mechanism should
+not be used even in waiting for disk reads to complete.
+Rather, we should put waiting query threads to the queue of
+waiting jobs, and let the OS thread do something useful while the i/o
+is processed. In this way we could remove most OS thread switches in
+an i/o-intensive benchmark like TPC-C.
+
+A possibility is to put a user space thread library between the database
+and NT. User space thread libraries might be very fast.
+
+SQL Server 7.0 can be configured to use 'fibers' which are lightweight
+threads in NT. These should be studied.
+
+		Buffer frames and blocks
+		------------------------
+Following the terminology of Gray and Reuter, we call the memory
+blocks where file pages are loaded buffer frames. For each buffer
+frame there is a control block, or shortly, a block, in the buffer
+control array. The control info which does not need to be stored
+in the file along with the file page, resides in the control block.
+
+		Buffer pool struct
+		------------------
+The buffer buf_pool contains a single mutex which protects all the
+control data structures of the buf_pool. The content of a buffer frame is
+protected by a separate read-write lock in its control block, though.
+These locks can be locked and unlocked without owning the buf_pool mutex.
+The OS events in the buf_pool struct can be waited for without owning the
+buf_pool mutex.
+
+The buf_pool mutex is a hot-spot in main memory, causing a lot of
+memory bus traffic on multiprocessor systems when processors
+alternately access the mutex. On our Pentium, the mutex is accessed
+maybe every 10 microseconds. We gave up the solution to have mutexes
+for each control block, for instance, because it seemed to be
+complicated.
+
+A solution to reduce mutex contention of the buf_pool mutex is to
+create a separate mutex for the page hash table. On Pentium,
+accessing the hash table takes 2 microseconds, about half
+of the total buf_pool mutex hold time.
+
+		Control blocks
+		--------------
+
+The control block contains, for instance, the bufferfix count
+which is incremented when a thread wants a file page to be fixed
+in a buffer frame. The bufferfix operation does not lock the
+contents of the frame, however. For this purpose, the control
+block contains a read-write lock.
+
+The buffer frames have to be aligned so that the start memory
+address of a frame is divisible by the universal page size, which
+is a power of two.
+
+We intend to make the buffer buf_pool size on-line reconfigurable,
+that is, the buf_pool size can be changed without closing the database.
+Then the database administarator may adjust it to be bigger
+at night, for example. The control block array must
+contain enough control blocks for the maximum buffer buf_pool size
+which is used in the particular database.
+If the buf_pool size is cut, we exploit the virtual memory mechanism of
+the OS, and just refrain from using frames at high addresses. Then the OS
+can swap them to disk.
+
+The control blocks containing file pages are put to a hash table
+according to the file address of the page.
+We could speed up the access to an individual page by using
+"pointer swizzling": we could replace the page references on
+non-leaf index pages by direct pointers to the page, if it exists
+in the buf_pool. We could make a separate hash table where we could
+chain all the page references in non-leaf pages residing in the buf_pool,
+using the page reference as the hash key,
+and at the time of reading of a page update the pointers accordingly.
+Drawbacks of this solution are added complexity and,
+possibly, extra space required on non-leaf pages for memory pointers.
+A simpler solution is just to speed up the hash table mechanism
+in the database, using tables whose size is a power of 2.
+
+		Lists of blocks
+		---------------
+
+There are several lists of control blocks.
+
+The free list (buf_pool->free) contains blocks which are currently not
+used.
+
+The common LRU list contains all the blocks holding a file page
+except those for which the bufferfix count is non-zero.
+The pages are in the LRU list roughly in the order of the last
+access to the page, so that the oldest pages are at the end of the
+list. We also keep a pointer to near the end of the LRU list,
+which we can use when we want to artificially age a page in the
+buf_pool. This is used if we know that some page is not needed
+again for some time: we insert the block right after the pointer,
+causing it to be replaced sooner than would noramlly be the case.
+Currently this aging mechanism is used for read-ahead mechanism
+of pages, and it can also be used when there is a scan of a full
+table which cannot fit in the memory. Putting the pages near the
+of the LRU list, we make sure that most of the buf_pool stays in the
+main memory, undisturbed.
+
+The unzip_LRU list contains a subset of the common LRU list.  The
+blocks on the unzip_LRU list hold a compressed file page and the
+corresponding uncompressed page frame.  A block is in unzip_LRU if and
+only if the predicate buf_page_belongs_to_unzip_LRU(&block->page)
+holds.  The blocks in unzip_LRU will be in same order as they are in
+the common LRU list.  That is, each manipulation of the common LRU
+list will result in the same manipulation of the unzip_LRU list.
+
+The chain of modified blocks (buf_pool->flush_list) contains the blocks
+holding file pages that have been modified in the memory
+but not written to disk yet. The block with the oldest modification
+which has not yet been written to disk is at the end of the chain.
+
+The chain of unmodified compressed blocks (buf_pool->zip_clean)
+contains the control blocks (buf_page_t) of those compressed pages
+that are not in buf_pool->flush_list and for which no uncompressed
+page has been allocated in the buffer pool.  The control blocks for
+uncompressed pages are accessible via buf_block_t objects that are
+reachable via buf_pool->chunks[].
+
+The chains of free memory blocks (buf_pool->zip_free[]) are used by
+the buddy allocator (buf0buddy.c) to keep track of currently unused
+memory blocks of size sizeof(buf_page_t)..UNIV_PAGE_SIZE / 2.  These
+blocks are inside the UNIV_PAGE_SIZE-sized memory blocks of type
+BUF_BLOCK_MEMORY that the buddy allocator requests from the buffer
+pool.  The buddy allocator is solely used for allocating control
+blocks for compressed pages (buf_page_t) and compressed page frames.
+
+		Loading a file page
+		-------------------
+
+First, a victim block for replacement has to be found in the
+buf_pool. It is taken from the free list or searched for from the
+end of the LRU-list. An exclusive lock is reserved for the frame,
+the io_fix field is set in the block fixing the block in buf_pool,
+and the io-operation for loading the page is queued. The io-handler thread
+releases the X-lock on the frame and resets the io_fix field
+when the io operation completes.
+
+A thread may request the above operation using the function
+buf_page_get(). It may then continue to request a lock on the frame.
+The lock is granted when the io-handler releases the x-lock.
+
+		Read-ahead
+		----------
+
+The read-ahead mechanism is intended to be intelligent and
+isolated from the semantically higher levels of the database
+index management. From the higher level we only need the
+information if a file page has a natural successor or
+predecessor page. On the leaf level of a B-tree index,
+these are the next and previous pages in the natural
+order of the pages.
+
+Let us first explain the read-ahead mechanism when the leafs
+of a B-tree are scanned in an ascending or descending order.
+When a read page is the first time referenced in the buf_pool,
+the buffer manager checks if it is at the border of a so-called
+linear read-ahead area. The tablespace is divided into these
+areas of size 64 blocks, for example. So if the page is at the
+border of such an area, the read-ahead mechanism checks if
+all the other blocks in the area have been accessed in an
+ascending or descending order. If this is the case, the system
+looks at the natural successor or predecessor of the page,
+checks if that is at the border of another area, and in this case
+issues read-requests for all the pages in that area. Maybe
+we could relax the condition that all the pages in the area
+have to be accessed: if data is deleted from a table, there may
+appear holes of unused pages in the area.
+
+A different read-ahead mechanism is used when there appears
+to be a random access pattern to a file.
+If a new page is referenced in the buf_pool, and several pages
+of its random access area (for instance, 32 consecutive pages
+in a tablespace) have recently been referenced, we may predict
+that the whole area may be needed in the near future, and issue
+the read requests for the whole area.
+*/
+
+/* Value in microseconds */
+static const int WAIT_FOR_READ	= 5000;
+
+/* The buffer buf_pool of the database */
+UNIV_INTERN buf_pool_t*	buf_pool = NULL;
+
+/* mutex protecting the buffer pool struct and control blocks, except the
+read-write lock in them */
+UNIV_INTERN mutex_t		buf_pool_mutex;
+/* mutex protecting the control blocks of compressed-only pages
+(of type buf_page_t, not buf_block_t) */
+UNIV_INTERN mutex_t		buf_pool_zip_mutex;
+
+#if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
+static ulint	buf_dbg_counter	= 0; /* This is used to insert validation
+					operations in excution in the
+					debug version */
+/** Flag to forbid the release of the buffer pool mutex.
+Protected by buf_pool_mutex. */
+UNIV_INTERN ulint		buf_pool_mutex_exit_forbidden = 0;
+#endif /* UNIV_DEBUG || UNIV_BUF_DEBUG */
+#ifdef UNIV_DEBUG
+/* If this is set TRUE, the program prints info whenever
+read-ahead or flush occurs */
+UNIV_INTERN ibool		buf_debug_prints = FALSE;
+#endif /* UNIV_DEBUG */
+
+/* A chunk of buffers.  The buffer pool is allocated in chunks. */
+struct buf_chunk_struct{
+	ulint		mem_size;	/* allocated size of the chunk */
+	ulint		size;		/* size of frames[] and blocks[] */
+	void*		mem;		/* pointer to the memory area which
+					was allocated for the frames */
+	buf_block_t*	blocks;		/* array of buffer control blocks */
+};
+
+/************************************************************************
+Calculates a page checksum which is stored to the page when it is written
+to a file. Note that we must be careful to calculate the same value on
+32-bit and 64-bit architectures. */
+UNIV_INTERN
+ulint
+buf_calc_page_new_checksum(
+/*=======================*/
+				/* out: checksum */
+	const byte*	page)	/* in: buffer page */
+{
+	ulint checksum;
+
+	/* Since the field FIL_PAGE_FILE_FLUSH_LSN, and in versions <= 4.1.x
+	..._ARCH_LOG_NO, are written outside the buffer pool to the first
+	pages of data files, we have to skip them in the page checksum
+	calculation.
+	We must also skip the field FIL_PAGE_SPACE_OR_CHKSUM where the
+	checksum is stored, and also the last 8 bytes of page because
+	there we store the old formula checksum. */
+
+	checksum = ut_fold_binary(page + FIL_PAGE_OFFSET,
+				  FIL_PAGE_FILE_FLUSH_LSN - FIL_PAGE_OFFSET)
+		+ ut_fold_binary(page + FIL_PAGE_DATA,
+				 UNIV_PAGE_SIZE - FIL_PAGE_DATA
+				 - FIL_PAGE_END_LSN_OLD_CHKSUM);
+	checksum = checksum & 0xFFFFFFFFUL;
+
+	return(checksum);
+}
+
+/************************************************************************
+In versions < 4.0.14 and < 4.1.1 there was a bug that the checksum only
+looked at the first few bytes of the page. This calculates that old
+checksum.
+NOTE: we must first store the new formula checksum to
+FIL_PAGE_SPACE_OR_CHKSUM before calculating and storing this old checksum
+because this takes that field as an input! */
+UNIV_INTERN
+ulint
+buf_calc_page_old_checksum(
+/*=======================*/
+				/* out: checksum */
+	const byte*	page)	/* in: buffer page */
+{
+	ulint checksum;
+
+	checksum = ut_fold_binary(page, FIL_PAGE_FILE_FLUSH_LSN);
+
+	checksum = checksum & 0xFFFFFFFFUL;
+
+	return(checksum);
+}
+
+/************************************************************************
+Checks if a page is corrupt. */
+UNIV_INTERN
+ibool
+buf_page_is_corrupted(
+/*==================*/
+					/* out: TRUE if corrupted */
+	const byte*	read_buf,	/* in: a database page */
+	ulint		zip_size)	/* in: size of compressed page;
+					0 for uncompressed pages */
+{
+	ulint		checksum_field;
+	ulint		old_checksum_field;
+#ifndef UNIV_HOTBACKUP
+	ib_uint64_t	current_lsn;
+#endif
+	if (UNIV_LIKELY(!zip_size)
+	    && memcmp(read_buf + FIL_PAGE_LSN + 4,
+		      read_buf + UNIV_PAGE_SIZE
+		      - FIL_PAGE_END_LSN_OLD_CHKSUM + 4, 4)) {
+
+		/* Stored log sequence numbers at the start and the end
+		of page do not match */
+
+		return(TRUE);
+	}
+
+#ifndef UNIV_HOTBACKUP
+	if (recv_lsn_checks_on && log_peek_lsn(&current_lsn)) {
+		if (current_lsn < mach_read_ull(read_buf + FIL_PAGE_LSN)) {
+			ut_print_timestamp(stderr);
+
+			fprintf(stderr,
+				"  InnoDB: Error: page %lu log sequence number"
+				" %llu\n"
+				"InnoDB: is in the future! Current system "
+				"log sequence number %llu.\n"
+				"InnoDB: Your database may be corrupt or "
+				"you may have copied the InnoDB\n"
+				"InnoDB: tablespace but not the InnoDB "
+				"log files. See\n"
+				"InnoDB: http://dev.mysql.com/doc/refman/"
+				"5.1/en/forcing-recovery.html\n"
+				"InnoDB: for more information.\n",
+				(ulong) mach_read_from_4(read_buf
+							 + FIL_PAGE_OFFSET),
+				mach_read_ull(read_buf + FIL_PAGE_LSN),
+				current_lsn);
+		}
+	}
+#endif
+
+	/* If we use checksums validation, make additional check before
+	returning TRUE to ensure that the checksum is not equal to
+	BUF_NO_CHECKSUM_MAGIC which might be stored by InnoDB with checksums
+	disabled. Otherwise, skip checksum calculation and return FALSE */
+
+	if (UNIV_LIKELY(srv_use_checksums)) {
+		checksum_field = mach_read_from_4(read_buf
+						  + FIL_PAGE_SPACE_OR_CHKSUM);
+
+		if (UNIV_UNLIKELY(zip_size)) {
+			return(checksum_field != BUF_NO_CHECKSUM_MAGIC
+			       && checksum_field
+			       != page_zip_calc_checksum(read_buf, zip_size));
+		}
+
+		old_checksum_field = mach_read_from_4(
+			read_buf + UNIV_PAGE_SIZE
+			- FIL_PAGE_END_LSN_OLD_CHKSUM);
+
+		/* There are 2 valid formulas for old_checksum_field:
+
+		1. Very old versions of InnoDB only stored 8 byte lsn to the
+		start and the end of the page.
+
+		2. Newer InnoDB versions store the old formula checksum
+		there. */
+
+		if (old_checksum_field != mach_read_from_4(read_buf
+							   + FIL_PAGE_LSN)
+		    && old_checksum_field != BUF_NO_CHECKSUM_MAGIC
+		    && old_checksum_field
+		    != buf_calc_page_old_checksum(read_buf)) {
+
+			return(TRUE);
+		}
+
+		/* InnoDB versions < 4.0.14 and < 4.1.1 stored the space id
+		(always equal to 0), to FIL_PAGE_SPACE_OR_CHKSUM */
+
+		if (checksum_field != 0
+		    && checksum_field != BUF_NO_CHECKSUM_MAGIC
+		    && checksum_field
+		    != buf_calc_page_new_checksum(read_buf)) {
+
+			return(TRUE);
+		}
+	}
+
+	return(FALSE);
+}
+
+/************************************************************************
+Prints a page to stderr. */
+UNIV_INTERN
+void
+buf_page_print(
+/*===========*/
+	const byte*	read_buf,	/* in: a database page */
+	ulint		zip_size)	/* in: compressed page size, or
+				0 for uncompressed pages */
+{
+	dict_index_t*	index;
+	ulint		checksum;
+	ulint		old_checksum;
+	ulint		size	= zip_size;
+
+	if (!size) {
+		size = UNIV_PAGE_SIZE;
+	}
+
+	ut_print_timestamp(stderr);
+	fprintf(stderr, "  InnoDB: Page dump in ascii and hex (%lu bytes):\n",
+		(ulong) size);
+	ut_print_buf(stderr, read_buf, size);
+	fputs("\nInnoDB: End of page dump\n", stderr);
+
+	if (zip_size) {
+		/* Print compressed page. */
+
+		switch (fil_page_get_type(read_buf)) {
+		case FIL_PAGE_TYPE_ZBLOB:
+		case FIL_PAGE_TYPE_ZBLOB2:
+			checksum = srv_use_checksums
+				? page_zip_calc_checksum(read_buf, zip_size)
+				: BUF_NO_CHECKSUM_MAGIC;
+			ut_print_timestamp(stderr);
+			fprintf(stderr,
+				"  InnoDB: Compressed BLOB page"
+				" checksum %lu, stored %lu\n"
+				"InnoDB: Page lsn %lu %lu\n"
+				"InnoDB: Page number (if stored"
+				" to page already) %lu,\n"
+				"InnoDB: space id (if stored"
+				" to page already) %lu\n",
+				(ulong) checksum,
+				(ulong) mach_read_from_4(
+					read_buf + FIL_PAGE_SPACE_OR_CHKSUM),
+				(ulong) mach_read_from_4(
+					read_buf + FIL_PAGE_LSN),
+				(ulong) mach_read_from_4(
+					read_buf + (FIL_PAGE_LSN + 4)),
+				(ulong) mach_read_from_4(
+					read_buf + FIL_PAGE_OFFSET),
+				(ulong) mach_read_from_4(
+					read_buf
+					+ FIL_PAGE_ARCH_LOG_NO_OR_SPACE_ID));
+			return;
+		default:
+			ut_print_timestamp(stderr);
+			fprintf(stderr,
+				"  InnoDB: unknown page type %lu,"
+				" assuming FIL_PAGE_INDEX\n",
+				fil_page_get_type(read_buf));
+			/* fall through */
+		case FIL_PAGE_INDEX:
+			checksum = srv_use_checksums
+				? page_zip_calc_checksum(read_buf, zip_size)
+				: BUF_NO_CHECKSUM_MAGIC;
+
+			ut_print_timestamp(stderr);
+			fprintf(stderr,
+				"  InnoDB: Compressed page checksum %lu,"
+				" stored %lu\n"
+				"InnoDB: Page lsn %lu %lu\n"
+				"InnoDB: Page number (if stored"
+				" to page already) %lu,\n"
+				"InnoDB: space id (if stored"
+				" to page already) %lu\n",
+				(ulong) checksum,
+				(ulong) mach_read_from_4(
+					read_buf + FIL_PAGE_SPACE_OR_CHKSUM),
+				(ulong) mach_read_from_4(
+					read_buf + FIL_PAGE_LSN),
+				(ulong) mach_read_from_4(
+					read_buf + (FIL_PAGE_LSN + 4)),
+				(ulong) mach_read_from_4(
+					read_buf + FIL_PAGE_OFFSET),
+				(ulong) mach_read_from_4(
+					read_buf
+					+ FIL_PAGE_ARCH_LOG_NO_OR_SPACE_ID));
+			return;
+		case FIL_PAGE_TYPE_XDES:
+			/* This is an uncompressed page. */
+			break;
+		}
+	}
+
+	checksum = srv_use_checksums
+		? buf_calc_page_new_checksum(read_buf) : BUF_NO_CHECKSUM_MAGIC;
+	old_checksum = srv_use_checksums
+		? buf_calc_page_old_checksum(read_buf) : BUF_NO_CHECKSUM_MAGIC;
+
+	ut_print_timestamp(stderr);
+	fprintf(stderr,
+		"  InnoDB: Page checksum %lu, prior-to-4.0.14-form"
+		" checksum %lu\n"
+		"InnoDB: stored checksum %lu, prior-to-4.0.14-form"
+		" stored checksum %lu\n"
+		"InnoDB: Page lsn %lu %lu, low 4 bytes of lsn"
+		" at page end %lu\n"
+		"InnoDB: Page number (if stored to page already) %lu,\n"
+		"InnoDB: space id (if created with >= MySQL-4.1.1"
+		" and stored already) %lu\n",
+		(ulong) checksum, (ulong) old_checksum,
+		(ulong) mach_read_from_4(read_buf + FIL_PAGE_SPACE_OR_CHKSUM),
+		(ulong) mach_read_from_4(read_buf + UNIV_PAGE_SIZE
+					 - FIL_PAGE_END_LSN_OLD_CHKSUM),
+		(ulong) mach_read_from_4(read_buf + FIL_PAGE_LSN),
+		(ulong) mach_read_from_4(read_buf + FIL_PAGE_LSN + 4),
+		(ulong) mach_read_from_4(read_buf + UNIV_PAGE_SIZE
+					 - FIL_PAGE_END_LSN_OLD_CHKSUM + 4),
+		(ulong) mach_read_from_4(read_buf + FIL_PAGE_OFFSET),
+		(ulong) mach_read_from_4(read_buf
+					 + FIL_PAGE_ARCH_LOG_NO_OR_SPACE_ID));
+
+	if (mach_read_from_2(read_buf + TRX_UNDO_PAGE_HDR + TRX_UNDO_PAGE_TYPE)
+	    == TRX_UNDO_INSERT) {
+		fprintf(stderr,
+			"InnoDB: Page may be an insert undo log page\n");
+	} else if (mach_read_from_2(read_buf + TRX_UNDO_PAGE_HDR
+				    + TRX_UNDO_PAGE_TYPE)
+		   == TRX_UNDO_UPDATE) {
+		fprintf(stderr,
+			"InnoDB: Page may be an update undo log page\n");
+	}
+
+	switch (fil_page_get_type(read_buf)) {
+	case FIL_PAGE_INDEX:
+		fprintf(stderr,
+			"InnoDB: Page may be an index page where"
+			" index id is %lu %lu\n",
+			(ulong) ut_dulint_get_high(
+				btr_page_get_index_id(read_buf)),
+			(ulong) ut_dulint_get_low(
+				btr_page_get_index_id(read_buf)));
+
+#ifdef UNIV_HOTBACKUP
+		/* If the code is in ibbackup, dict_sys may be uninitialized,
+		i.e., NULL */
+
+		if (dict_sys == NULL) {
+			break;
+		}
+#endif /* UNIV_HOTBACKUP */
+
+		index = dict_index_find_on_id_low(
+			btr_page_get_index_id(read_buf));
+		if (index) {
+			fputs("InnoDB: (", stderr);
+			dict_index_name_print(stderr, NULL, index);
+			fputs(")\n", stderr);
+		}
+		break;
+	case FIL_PAGE_INODE:
+		fputs("InnoDB: Page may be an 'inode' page\n", stderr);
+		break;
+	case FIL_PAGE_IBUF_FREE_LIST:
+		fputs("InnoDB: Page may be an insert buffer free list page\n",
+		      stderr);
+		break;
+	case FIL_PAGE_TYPE_ALLOCATED:
+		fputs("InnoDB: Page may be a freshly allocated page\n",
+		      stderr);
+		break;
+	case FIL_PAGE_IBUF_BITMAP:
+		fputs("InnoDB: Page may be an insert buffer bitmap page\n",
+		      stderr);
+		break;
+	case FIL_PAGE_TYPE_SYS:
+		fputs("InnoDB: Page may be a system page\n",
+		      stderr);
+		break;
+	case FIL_PAGE_TYPE_TRX_SYS:
+		fputs("InnoDB: Page may be a transaction system page\n",
+		      stderr);
+		break;
+	case FIL_PAGE_TYPE_FSP_HDR:
+		fputs("InnoDB: Page may be a file space header page\n",
+		      stderr);
+		break;
+	case FIL_PAGE_TYPE_XDES:
+		fputs("InnoDB: Page may be an extent descriptor page\n",
+		      stderr);
+		break;
+	case FIL_PAGE_TYPE_BLOB:
+		fputs("InnoDB: Page may be a BLOB page\n",
+		      stderr);
+		break;
+	case FIL_PAGE_TYPE_ZBLOB:
+	case FIL_PAGE_TYPE_ZBLOB2:
+		fputs("InnoDB: Page may be a compressed BLOB page\n",
+		      stderr);
+		break;
+	}
+}
+
+/************************************************************************
+Initializes a buffer control block when the buf_pool is created. */
+static
+void
+buf_block_init(
+/*===========*/
+	buf_block_t*	block,	/* in: pointer to control block */
+	byte*		frame)	/* in: pointer to buffer frame */
+{
+	UNIV_MEM_DESC(frame, UNIV_PAGE_SIZE, block);
+
+	block->frame = frame;
+
+	block->page.state = BUF_BLOCK_NOT_USED;
+	block->page.buf_fix_count = 0;
+	block->page.io_fix = BUF_IO_NONE;
+
+	block->modify_clock = 0;
+
+#ifdef UNIV_DEBUG_FILE_ACCESSES
+	block->page.file_page_was_freed = FALSE;
+#endif /* UNIV_DEBUG_FILE_ACCESSES */
+
+	block->check_index_page_at_flush = FALSE;
+	block->index = NULL;
+
+#ifdef UNIV_DEBUG
+	block->page.in_page_hash = FALSE;
+	block->page.in_zip_hash = FALSE;
+	block->page.in_flush_list = FALSE;
+	block->page.in_free_list = FALSE;
+	block->in_unzip_LRU_list = FALSE;
+#endif /* UNIV_DEBUG */
+	block->page.in_LRU_list = FALSE;
+#if defined UNIV_AHI_DEBUG || defined UNIV_DEBUG
+	block->n_pointers = 0;
+#endif /* UNIV_AHI_DEBUG || UNIV_DEBUG */
+	page_zip_des_init(&block->page.zip);
+
+	mutex_create(&block->mutex, SYNC_BUF_BLOCK);
+
+	rw_lock_create(&block->lock, SYNC_LEVEL_VARYING);
+	ut_ad(rw_lock_validate(&(block->lock)));
+
+#ifdef UNIV_SYNC_DEBUG
+	rw_lock_create(&block->debug_latch, SYNC_NO_ORDER_CHECK);
+#endif /* UNIV_SYNC_DEBUG */
+}
+
+/************************************************************************
+Allocates a chunk of buffer frames. */
+static
+buf_chunk_t*
+buf_chunk_init(
+/*===========*/
+					/* out: chunk, or NULL on failure */
+	buf_chunk_t*	chunk,		/* out: chunk of buffers */
+	ulint		mem_size)	/* in: requested size in bytes */
+{
+	buf_block_t*	block;
+	byte*		frame;
+	ulint		i;
+
+	/* Round down to a multiple of page size,
+	although it already should be. */
+	mem_size = ut_2pow_round(mem_size, UNIV_PAGE_SIZE);
+	/* Reserve space for the block descriptors. */
+	mem_size += ut_2pow_round((mem_size / UNIV_PAGE_SIZE) * (sizeof *block)
+				  + (UNIV_PAGE_SIZE - 1), UNIV_PAGE_SIZE);
+
+	chunk->mem_size = mem_size;
+	chunk->mem = os_mem_alloc_large(&chunk->mem_size);
+
+	if (UNIV_UNLIKELY(chunk->mem == NULL)) {
+
+		return(NULL);
+	}
+
+	/* Allocate the block descriptors from
+	the start of the memory block. */
+	chunk->blocks = chunk->mem;
+
+	/* Align a pointer to the first frame.  Note that when
+	os_large_page_size is smaller than UNIV_PAGE_SIZE,
+	we may allocate one fewer block than requested.  When
+	it is bigger, we may allocate more blocks than requested. */
+
+	frame = ut_align(chunk->mem, UNIV_PAGE_SIZE);
+	chunk->size = chunk->mem_size / UNIV_PAGE_SIZE
+		- (frame != chunk->mem);
+
+	/* Subtract the space needed for block descriptors. */
+	{
+		ulint	size = chunk->size;
+
+		while (frame < (byte*) (chunk->blocks + size)) {
+			frame += UNIV_PAGE_SIZE;
+			size--;
+		}
+
+		chunk->size = size;
+	}
+
+	/* Init block structs and assign frames for them. Then we
+	assign the frames to the first blocks (we already mapped the
+	memory above). */
+
+	block = chunk->blocks;
+
+	for (i = chunk->size; i--; ) {
+
+		buf_block_init(block, frame);
+
+#ifdef HAVE_purify
+		/* Wipe contents of frame to eliminate a Purify warning */
+		memset(block->frame, '\0', UNIV_PAGE_SIZE);
+#endif
+		/* Add the block to the free list */
+		UT_LIST_ADD_LAST(list, buf_pool->free, (&block->page));
+		ut_d(block->page.in_free_list = TRUE);
+
+		block++;
+		frame += UNIV_PAGE_SIZE;
+	}
+
+	return(chunk);
+}
+
+#ifdef UNIV_DEBUG
+/*************************************************************************
+Finds a block in the given buffer chunk that points to a
+given compressed page. */
+static
+buf_block_t*
+buf_chunk_contains_zip(
+/*===================*/
+				/* out: buffer block pointing to
+				the compressed page, or NULL */
+	buf_chunk_t*	chunk,	/* in: chunk being checked */
+	const void*	data)	/* in: pointer to compressed page */
+{
+	buf_block_t*	block;
+	ulint		i;
+
+	ut_ad(buf_pool);
+	ut_ad(buf_pool_mutex_own());
+
+	block = chunk->blocks;
+
+	for (i = chunk->size; i--; block++) {
+		if (block->page.zip.data == data) {
+
+			return(block);
+		}
+	}
+
+	return(NULL);
+}
+
+/*************************************************************************
+Finds a block in the buffer pool that points to a
+given compressed page. */
+UNIV_INTERN
+buf_block_t*
+buf_pool_contains_zip(
+/*==================*/
+				/* out: buffer block pointing to
+				the compressed page, or NULL */
+	const void*	data)	/* in: pointer to compressed page */
+{
+	ulint		n;
+	buf_chunk_t*	chunk = buf_pool->chunks;
+
+	for (n = buf_pool->n_chunks; n--; chunk++) {
+		buf_block_t* block = buf_chunk_contains_zip(chunk, data);
+
+		if (block) {
+			return(block);
+		}
+	}
+
+	return(NULL);
+}
+#endif /* UNIV_DEBUG */
+
+/*************************************************************************
+Checks that all file pages in the buffer chunk are in a replaceable state. */
+static
+const buf_block_t*
+buf_chunk_not_freed(
+/*================*/
+				/* out: address of a non-free block,
+				or NULL if all freed */
+	buf_chunk_t*	chunk)	/* in: chunk being checked */
+{
+	buf_block_t*	block;
+	ulint		i;
+
+	ut_ad(buf_pool);
+	ut_ad(buf_pool_mutex_own());
+
+	block = chunk->blocks;
+
+	for (i = chunk->size; i--; block++) {
+		mutex_enter(&block->mutex);
+
+		if (buf_block_get_state(block) == BUF_BLOCK_FILE_PAGE
+		    && !buf_flush_ready_for_replace(&block->page)) {
+
+			mutex_exit(&block->mutex);
+			return(block);
+		}
+
+		mutex_exit(&block->mutex);
+	}
+
+	return(NULL);
+}
+
+/*************************************************************************
+Checks that all blocks in the buffer chunk are in BUF_BLOCK_NOT_USED state. */
+static
+ibool
+buf_chunk_all_free(
+/*===============*/
+					/* out: TRUE if all freed */
+	const buf_chunk_t*	chunk)	/* in: chunk being checked */
+{
+	const buf_block_t*	block;
+	ulint			i;
+
+	ut_ad(buf_pool);
+	ut_ad(buf_pool_mutex_own());
+
+	block = chunk->blocks;
+
+	for (i = chunk->size; i--; block++) {
+
+		if (buf_block_get_state(block) != BUF_BLOCK_NOT_USED) {
+
+			return(FALSE);
+		}
+	}
+
+	return(TRUE);
+}
+
+/************************************************************************
+Frees a chunk of buffer frames. */
+static
+void
+buf_chunk_free(
+/*===========*/
+	buf_chunk_t*	chunk)		/* out: chunk of buffers */
+{
+	buf_block_t*		block;
+	const buf_block_t*	block_end;
+
+	ut_ad(buf_pool_mutex_own());
+
+	block_end = chunk->blocks + chunk->size;
+
+	for (block = chunk->blocks; block < block_end; block++) {
+		ut_a(buf_block_get_state(block) == BUF_BLOCK_NOT_USED);
+		ut_a(!block->page.zip.data);
+
+		ut_ad(!block->page.in_LRU_list);
+		ut_ad(!block->in_unzip_LRU_list);
+		ut_ad(!block->page.in_flush_list);
+		/* Remove the block from the free list. */
+		ut_ad(block->page.in_free_list);
+		UT_LIST_REMOVE(list, buf_pool->free, (&block->page));
+
+		/* Free the latches. */
+		mutex_free(&block->mutex);
+		rw_lock_free(&block->lock);
+#ifdef UNIV_SYNC_DEBUG
+		rw_lock_free(&block->debug_latch);
+#endif /* UNIV_SYNC_DEBUG */
+		UNIV_MEM_UNDESC(block);
+	}
+
+	os_mem_free_large(chunk->mem, chunk->mem_size);
+}
+
+/************************************************************************
+Creates the buffer pool. */
+UNIV_INTERN
+buf_pool_t*
+buf_pool_init(void)
+/*===============*/
+				/* out, own: buf_pool object, NULL if not
+				enough memory or error */
+{
+	buf_chunk_t*	chunk;
+	ulint		i;
+
+	buf_pool = mem_zalloc(sizeof(buf_pool_t));
+
+	/* 1. Initialize general fields
+	------------------------------- */
+	mutex_create(&buf_pool_mutex, SYNC_BUF_POOL);
+	mutex_create(&buf_pool_zip_mutex, SYNC_BUF_BLOCK);
+
+	buf_pool_mutex_enter();
+
+	buf_pool->n_chunks = 1;
+	buf_pool->chunks = chunk = mem_alloc(sizeof *chunk);
+
+	UT_LIST_INIT(buf_pool->free);
+
+	if (!buf_chunk_init(chunk, srv_buf_pool_size)) {
+		mem_free(chunk);
+		mem_free(buf_pool);
+		buf_pool = NULL;
+		return(NULL);
+	}
+
+	srv_buf_pool_old_size = srv_buf_pool_size;
+	buf_pool->curr_size = chunk->size;
+	srv_buf_pool_curr_size = buf_pool->curr_size * UNIV_PAGE_SIZE;
+
+	buf_pool->page_hash = hash_create(2 * buf_pool->curr_size);
+	buf_pool->zip_hash = hash_create(2 * buf_pool->curr_size);
+
+	buf_pool->last_printout_time = time(NULL);
+
+	/* 2. Initialize flushing fields
+	-------------------------------- */
+
+	for (i = BUF_FLUSH_LRU; i < BUF_FLUSH_N_TYPES; i++) {
+		buf_pool->no_flush[i] = os_event_create(NULL);
+	}
+
+	buf_pool->ulint_clock = 1;
+
+	/* 3. Initialize LRU fields
+	--------------------------- */
+	/* All fields are initialized by mem_zalloc(). */
+
+	buf_pool_mutex_exit();
+
+	btr_search_sys_create(buf_pool->curr_size
+			      * UNIV_PAGE_SIZE / sizeof(void*) / 64);
+
+	/* 4. Initialize the buddy allocator fields */
+	/* All fields are initialized by mem_zalloc(). */
+
+	return(buf_pool);
+}
+
+/************************************************************************
+Frees the buffer pool at shutdown.  This must not be invoked before
+freeing all mutexes. */
+UNIV_INTERN
+void
+buf_pool_free(void)
+/*===============*/
+{
+	buf_chunk_t*	chunk;
+	buf_chunk_t*	chunks;
+
+	chunks = buf_pool->chunks;
+	chunk = chunks + buf_pool->n_chunks;
+
+	while (--chunk >= chunks) {
+		/* Bypass the checks of buf_chunk_free(), since they
+		would fail at shutdown. */
+		os_mem_free_large(chunk->mem, chunk->mem_size);
+	}
+
+	buf_pool->n_chunks = 0;
+}
+
+
+/************************************************************************
+Drops the adaptive hash index.  To prevent a livelock, this function
+is only to be called while holding btr_search_latch and while
+btr_search_enabled == FALSE. */
+UNIV_INTERN
+void
+buf_pool_drop_hash_index(void)
+/*==========================*/
+{
+	ibool		released_search_latch;
+
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(rw_lock_own(&btr_search_latch, RW_LOCK_EX));
+#endif /* UNIV_SYNC_DEBUG */
+	ut_ad(!btr_search_enabled);
+
+	do {
+		buf_chunk_t*	chunks	= buf_pool->chunks;
+		buf_chunk_t*	chunk	= chunks + buf_pool->n_chunks;
+
+		released_search_latch = FALSE;
+
+		while (--chunk >= chunks) {
+			buf_block_t*	block	= chunk->blocks;
+			ulint		i	= chunk->size;
+
+			for (; i--; block++) {
+				/* block->is_hashed cannot be modified
+				when we have an x-latch on btr_search_latch;
+				see the comment in buf0buf.h */
+
+				if (!block->is_hashed) {
+					continue;
+				}
+
+				/* To follow the latching order, we
+				have to release btr_search_latch
+				before acquiring block->latch. */
+				rw_lock_x_unlock(&btr_search_latch);
+				/* When we release the search latch,
+				we must rescan all blocks, because
+				some may become hashed again. */
+				released_search_latch = TRUE;
+
+				rw_lock_x_lock(&block->lock);
+
+				/* This should be guaranteed by the
+				callers, which will be holding
+				btr_search_enabled_mutex. */
+				ut_ad(!btr_search_enabled);
+
+				/* Because we did not buffer-fix the
+				block by calling buf_block_get_gen(),
+				it is possible that the block has been
+				allocated for some other use after
+				btr_search_latch was released above.
+				We do not care which file page the
+				block is mapped to.  All we want to do
+				is to drop any hash entries referring
+				to the page. */
+
+				/* It is possible that
+				block->page.state != BUF_FILE_PAGE.
+				Even that does not matter, because
+				btr_search_drop_page_hash_index() will
+				check block->is_hashed before doing
+				anything.  block->is_hashed can only
+				be set on uncompressed file pages. */
+
+				btr_search_drop_page_hash_index(block);
+
+				rw_lock_x_unlock(&block->lock);
+
+				rw_lock_x_lock(&btr_search_latch);
+
+				ut_ad(!btr_search_enabled);
+			}
+		}
+	} while (released_search_latch);
+}
+
+/************************************************************************
+Relocate a buffer control block.  Relocates the block on the LRU list
+and in buf_pool->page_hash.  Does not relocate bpage->list.
+The caller must take care of relocating bpage->list. */
+UNIV_INTERN
+void
+buf_relocate(
+/*=========*/
+	buf_page_t*	bpage,	/* in/out: control block being relocated;
+				buf_page_get_state(bpage) must be
+				BUF_BLOCK_ZIP_DIRTY or BUF_BLOCK_ZIP_PAGE */
+	buf_page_t*	dpage)	/* in/out: destination control block */
+{
+	buf_page_t*	b;
+	ulint		fold;
+
+	ut_ad(buf_pool_mutex_own());
+	ut_ad(mutex_own(buf_page_get_mutex(bpage)));
+	ut_a(buf_page_get_io_fix(bpage) == BUF_IO_NONE);
+	ut_a(bpage->buf_fix_count == 0);
+	ut_ad(bpage->in_LRU_list);
+	ut_ad(!bpage->in_zip_hash);
+	ut_ad(bpage->in_page_hash);
+	ut_ad(bpage == buf_page_hash_get(bpage->space, bpage->offset));
+#ifdef UNIV_DEBUG
+	switch (buf_page_get_state(bpage)) {
+	case BUF_BLOCK_ZIP_FREE:
+	case BUF_BLOCK_NOT_USED:
+	case BUF_BLOCK_READY_FOR_USE:
+	case BUF_BLOCK_FILE_PAGE:
+	case BUF_BLOCK_MEMORY:
+	case BUF_BLOCK_REMOVE_HASH:
+		ut_error;
+	case BUF_BLOCK_ZIP_DIRTY:
+	case BUF_BLOCK_ZIP_PAGE:
+		break;
+	}
+#endif /* UNIV_DEBUG */
+
+	memcpy(dpage, bpage, sizeof *dpage);
+
+	bpage->in_LRU_list = FALSE;
+	ut_d(bpage->in_page_hash = FALSE);
+
+	/* relocate buf_pool->LRU */
+	b = UT_LIST_GET_PREV(LRU, bpage);
+	UT_LIST_REMOVE(LRU, buf_pool->LRU, bpage);
+
+	if (b) {
+		UT_LIST_INSERT_AFTER(LRU, buf_pool->LRU, b, dpage);
+	} else {
+		UT_LIST_ADD_FIRST(LRU, buf_pool->LRU, dpage);
+	}
+
+	if (UNIV_UNLIKELY(buf_pool->LRU_old == bpage)) {
+		buf_pool->LRU_old = dpage;
+#ifdef UNIV_LRU_DEBUG
+		/* buf_pool->LRU_old must be the first item in the LRU list
+		whose "old" flag is set. */
+		ut_a(!UT_LIST_GET_PREV(LRU, buf_pool->LRU_old)
+		     || !UT_LIST_GET_PREV(LRU, buf_pool->LRU_old)->old);
+		ut_a(!UT_LIST_GET_NEXT(LRU, buf_pool->LRU_old)
+		     || UT_LIST_GET_NEXT(LRU, buf_pool->LRU_old)->old);
+#endif /* UNIV_LRU_DEBUG */
+	}
+
+	ut_d(UT_LIST_VALIDATE(LRU, buf_page_t, buf_pool->LRU));
+
+	/* relocate buf_pool->page_hash */
+	fold = buf_page_address_fold(bpage->space, bpage->offset);
+
+	HASH_DELETE(buf_page_t, hash, buf_pool->page_hash, fold, bpage);
+	HASH_INSERT(buf_page_t, hash, buf_pool->page_hash, fold, dpage);
+
+	UNIV_MEM_INVALID(bpage, sizeof *bpage);
+}
+
+/************************************************************************
+Shrinks the buffer pool. */
+static
+void
+buf_pool_shrink(
+/*============*/
+				/* out: TRUE if shrunk */
+	ulint	chunk_size)	/* in: number of pages to remove */
+{
+	buf_chunk_t*	chunks;
+	buf_chunk_t*	chunk;
+	ulint		max_size;
+	ulint		max_free_size;
+	buf_chunk_t*	max_chunk;
+	buf_chunk_t*	max_free_chunk;
+
+	ut_ad(!buf_pool_mutex_own());
+
+try_again:
+	btr_search_disable(); /* Empty the adaptive hash index again */
+	buf_pool_mutex_enter();
+
+shrink_again:
+	if (buf_pool->n_chunks <= 1) {
+
+		/* Cannot shrink if there is only one chunk */
+		goto func_done;
+	}
+
+	/* Search for the largest free chunk
+	not larger than the size difference */
+	chunks = buf_pool->chunks;
+	chunk = chunks + buf_pool->n_chunks;
+	max_size = max_free_size = 0;
+	max_chunk = max_free_chunk = NULL;
+
+	while (--chunk >= chunks) {
+		if (chunk->size <= chunk_size
+		    && chunk->size > max_free_size) {
+			if (chunk->size > max_size) {
+				max_size = chunk->size;
+				max_chunk = chunk;
+			}
+
+			if (buf_chunk_all_free(chunk)) {
+				max_free_size = chunk->size;
+				max_free_chunk = chunk;
+			}
+		}
+	}
+
+	if (!max_free_size) {
+
+		ulint		dirty	= 0;
+		ulint		nonfree	= 0;
+		buf_block_t*	block;
+		buf_block_t*	bend;
+
+		/* Cannot shrink: try again later
+		(do not assign srv_buf_pool_old_size) */
+		if (!max_chunk) {
+
+			goto func_exit;
+		}
+
+		block = max_chunk->blocks;
+		bend = block + max_chunk->size;
+
+		/* Move the blocks of chunk to the end of the
+		LRU list and try to flush them. */
+		for (; block < bend; block++) {
+			switch (buf_block_get_state(block)) {
+			case BUF_BLOCK_NOT_USED:
+				continue;
+			case BUF_BLOCK_FILE_PAGE:
+				break;
+			default:
+				nonfree++;
+				continue;
+			}
+
+			mutex_enter(&block->mutex);
+			/* The following calls will temporarily
+			release block->mutex and buf_pool_mutex.
+			Therefore, we have to always retry,
+			even if !dirty && !nonfree. */
+
+			if (!buf_flush_ready_for_replace(&block->page)) {
+
+				buf_LRU_make_block_old(&block->page);
+				dirty++;
+			} else if (buf_LRU_free_block(&block->page, TRUE, NULL)
+				   != BUF_LRU_FREED) {
+				nonfree++;
+			}
+
+			mutex_exit(&block->mutex);
+		}
+
+		buf_pool_mutex_exit();
+
+		/* Request for a flush of the chunk if it helps.
+		Do not flush if there are non-free blocks, since
+		flushing will not make the chunk freeable. */
+		if (nonfree) {
+			/* Avoid busy-waiting. */
+			os_thread_sleep(100000);
+		} else if (dirty
+			   && buf_flush_batch(BUF_FLUSH_LRU, dirty, 0)
+			   == ULINT_UNDEFINED) {
+
+			buf_flush_wait_batch_end(BUF_FLUSH_LRU);
+		}
+
+		goto try_again;
+	}
+
+	max_size = max_free_size;
+	max_chunk = max_free_chunk;
+
+	srv_buf_pool_old_size = srv_buf_pool_size;
+
+	/* Rewrite buf_pool->chunks.  Copy everything but max_chunk. */
+	chunks = mem_alloc((buf_pool->n_chunks - 1) * sizeof *chunks);
+	memcpy(chunks, buf_pool->chunks,
+	       (max_chunk - buf_pool->chunks) * sizeof *chunks);
+	memcpy(chunks + (max_chunk - buf_pool->chunks),
+	       max_chunk + 1,
+	       buf_pool->chunks + buf_pool->n_chunks
+	       - (max_chunk + 1));
+	ut_a(buf_pool->curr_size > max_chunk->size);
+	buf_pool->curr_size -= max_chunk->size;
+	srv_buf_pool_curr_size = buf_pool->curr_size * UNIV_PAGE_SIZE;
+	chunk_size -= max_chunk->size;
+	buf_chunk_free(max_chunk);
+	mem_free(buf_pool->chunks);
+	buf_pool->chunks = chunks;
+	buf_pool->n_chunks--;
+
+	/* Allow a slack of one megabyte. */
+	if (chunk_size > 1048576 / UNIV_PAGE_SIZE) {
+
+		goto shrink_again;
+	}
+
+func_done:
+	srv_buf_pool_old_size = srv_buf_pool_size;
+func_exit:
+	buf_pool_mutex_exit();
+	btr_search_enable();
+}
+
+/************************************************************************
+Rebuild buf_pool->page_hash. */
+static
+void
+buf_pool_page_hash_rebuild(void)
+/*============================*/
+{
+	ulint		i;
+	ulint		n_chunks;
+	buf_chunk_t*	chunk;
+	hash_table_t*	page_hash;
+	hash_table_t*	zip_hash;
+	buf_page_t*	b;
+
+	buf_pool_mutex_enter();
+
+	/* Free, create, and populate the hash table. */
+	hash_table_free(buf_pool->page_hash);
+	buf_pool->page_hash = page_hash = hash_create(2 * buf_pool->curr_size);
+	zip_hash = hash_create(2 * buf_pool->curr_size);
+
+	HASH_MIGRATE(buf_pool->zip_hash, zip_hash, buf_page_t, hash,
+		     BUF_POOL_ZIP_FOLD_BPAGE);
+
+	hash_table_free(buf_pool->zip_hash);
+	buf_pool->zip_hash = zip_hash;
+
+	/* Insert the uncompressed file pages to buf_pool->page_hash. */
+
+	chunk = buf_pool->chunks;
+	n_chunks = buf_pool->n_chunks;
+
+	for (i = 0; i < n_chunks; i++, chunk++) {
+		ulint		j;
+		buf_block_t*	block = chunk->blocks;
+
+		for (j = 0; j < chunk->size; j++, block++) {
+			if (buf_block_get_state(block)
+			    == BUF_BLOCK_FILE_PAGE) {
+				ut_ad(!block->page.in_zip_hash);
+				ut_ad(block->page.in_page_hash);
+
+				HASH_INSERT(buf_page_t, hash, page_hash,
+					    buf_page_address_fold(
+						    block->page.space,
+						    block->page.offset),
+					    &block->page);
+			}
+		}
+	}
+
+	/* Insert the compressed-only pages to buf_pool->page_hash.
+	All such blocks are either in buf_pool->zip_clean or
+	in buf_pool->flush_list. */
+
+	for (b = UT_LIST_GET_FIRST(buf_pool->zip_clean); b;
+	     b = UT_LIST_GET_NEXT(list, b)) {
+		ut_a(buf_page_get_state(b) == BUF_BLOCK_ZIP_PAGE);
+		ut_ad(!b->in_flush_list);
+		ut_ad(b->in_LRU_list);
+		ut_ad(b->in_page_hash);
+		ut_ad(!b->in_zip_hash);
+
+		HASH_INSERT(buf_page_t, hash, page_hash,
+			    buf_page_address_fold(b->space, b->offset), b);
+	}
+
+	for (b = UT_LIST_GET_FIRST(buf_pool->flush_list); b;
+	     b = UT_LIST_GET_NEXT(list, b)) {
+		ut_ad(b->in_flush_list);
+		ut_ad(b->in_LRU_list);
+		ut_ad(b->in_page_hash);
+		ut_ad(!b->in_zip_hash);
+
+		switch (buf_page_get_state(b)) {
+		case BUF_BLOCK_ZIP_DIRTY:
+			HASH_INSERT(buf_page_t, hash, page_hash,
+				    buf_page_address_fold(b->space,
+							  b->offset), b);
+			break;
+		case BUF_BLOCK_FILE_PAGE:
+			/* uncompressed page */
+			break;
+		case BUF_BLOCK_ZIP_FREE:
+		case BUF_BLOCK_ZIP_PAGE:
+		case BUF_BLOCK_NOT_USED:
+		case BUF_BLOCK_READY_FOR_USE:
+		case BUF_BLOCK_MEMORY:
+		case BUF_BLOCK_REMOVE_HASH:
+			ut_error;
+			break;
+		}
+	}
+
+	buf_pool_mutex_exit();
+}
+
+/************************************************************************
+Resizes the buffer pool. */
+UNIV_INTERN
+void
+buf_pool_resize(void)
+/*=================*/
+{
+	buf_pool_mutex_enter();
+
+	if (srv_buf_pool_old_size == srv_buf_pool_size) {
+
+		buf_pool_mutex_exit();
+		return;
+	}
+
+	if (srv_buf_pool_curr_size + 1048576 > srv_buf_pool_size) {
+
+		buf_pool_mutex_exit();
+
+		/* Disable adaptive hash indexes and empty the index
+		in order to free up memory in the buffer pool chunks. */
+		buf_pool_shrink((srv_buf_pool_curr_size - srv_buf_pool_size)
+				/ UNIV_PAGE_SIZE);
+	} else if (srv_buf_pool_curr_size + 1048576 < srv_buf_pool_size) {
+
+		/* Enlarge the buffer pool by at least one megabyte */
+
+		ulint		mem_size
+			= srv_buf_pool_size - srv_buf_pool_curr_size;
+		buf_chunk_t*	chunks;
+		buf_chunk_t*	chunk;
+
+		chunks = mem_alloc((buf_pool->n_chunks + 1) * sizeof *chunks);
+
+		memcpy(chunks, buf_pool->chunks, buf_pool->n_chunks
+		       * sizeof *chunks);
+
+		chunk = &chunks[buf_pool->n_chunks];
+
+		if (!buf_chunk_init(chunk, mem_size)) {
+			mem_free(chunks);
+		} else {
+			buf_pool->curr_size += chunk->size;
+			srv_buf_pool_curr_size = buf_pool->curr_size
+				* UNIV_PAGE_SIZE;
+			mem_free(buf_pool->chunks);
+			buf_pool->chunks = chunks;
+			buf_pool->n_chunks++;
+		}
+
+		srv_buf_pool_old_size = srv_buf_pool_size;
+		buf_pool_mutex_exit();
+	}
+
+	buf_pool_page_hash_rebuild();
+}
+
+/************************************************************************
+Moves to the block to the start of the LRU list if there is a danger
+that the block would drift out of the buffer pool. */
+UNIV_INLINE
+void
+buf_block_make_young(
+/*=================*/
+	buf_page_t*	bpage)	/* in: block to make younger */
+{
+	ut_ad(!buf_pool_mutex_own());
+
+	/* Note that we read freed_page_clock's without holding any mutex:
+	this is allowed since the result is used only in heuristics */
+
+	if (buf_page_peek_if_too_old(bpage)) {
+
+		buf_pool_mutex_enter();
+		/* There has been freeing activity in the LRU list:
+		best to move to the head of the LRU list */
+
+		buf_LRU_make_block_young(bpage);
+		buf_pool_mutex_exit();
+	}
+}
+
+/************************************************************************
+Moves a page to the start of the buffer pool LRU list. This high-level
+function can be used to prevent an important page from from slipping out of
+the buffer pool. */
+UNIV_INTERN
+void
+buf_page_make_young(
+/*================*/
+	buf_page_t*	bpage)	/* in: buffer block of a file page */
+{
+	buf_pool_mutex_enter();
+
+	ut_a(buf_page_in_file(bpage));
+
+	buf_LRU_make_block_young(bpage);
+
+	buf_pool_mutex_exit();
+}
+
+/************************************************************************
+Resets the check_index_page_at_flush field of a page if found in the buffer
+pool. */
+UNIV_INTERN
+void
+buf_reset_check_index_page_at_flush(
+/*================================*/
+	ulint	space,	/* in: space id */
+	ulint	offset)	/* in: page number */
+{
+	buf_block_t*	block;
+
+	buf_pool_mutex_enter();
+
+	block = (buf_block_t*) buf_page_hash_get(space, offset);
+
+	if (block && buf_block_get_state(block) == BUF_BLOCK_FILE_PAGE) {
+		block->check_index_page_at_flush = FALSE;
+	}
+
+	buf_pool_mutex_exit();
+}
+
+/************************************************************************
+Returns the current state of is_hashed of a page. FALSE if the page is
+not in the pool. NOTE that this operation does not fix the page in the
+pool if it is found there. */
+UNIV_INTERN
+ibool
+buf_page_peek_if_search_hashed(
+/*===========================*/
+			/* out: TRUE if page hash index is built in search
+			system */
+	ulint	space,	/* in: space id */
+	ulint	offset)	/* in: page number */
+{
+	buf_block_t*	block;
+	ibool		is_hashed;
+
+	buf_pool_mutex_enter();
+
+	block = (buf_block_t*) buf_page_hash_get(space, offset);
+
+	if (!block || buf_block_get_state(block) != BUF_BLOCK_FILE_PAGE) {
+		is_hashed = FALSE;
+	} else {
+		is_hashed = block->is_hashed;
+	}
+
+	buf_pool_mutex_exit();
+
+	return(is_hashed);
+}
+
+#ifdef UNIV_DEBUG_FILE_ACCESSES
+/************************************************************************
+Sets file_page_was_freed TRUE if the page is found in the buffer pool.
+This function should be called when we free a file page and want the
+debug version to check that it is not accessed any more unless
+reallocated. */
+UNIV_INTERN
+buf_page_t*
+buf_page_set_file_page_was_freed(
+/*=============================*/
+			/* out: control block if found in page hash table,
+			otherwise NULL */
+	ulint	space,	/* in: space id */
+	ulint	offset)	/* in: page number */
+{
+	buf_page_t*	bpage;
+
+	buf_pool_mutex_enter();
+
+	bpage = buf_page_hash_get(space, offset);
+
+	if (bpage) {
+		bpage->file_page_was_freed = TRUE;
+	}
+
+	buf_pool_mutex_exit();
+
+	return(bpage);
+}
+
+/************************************************************************
+Sets file_page_was_freed FALSE if the page is found in the buffer pool.
+This function should be called when we free a file page and want the
+debug version to check that it is not accessed any more unless
+reallocated. */
+UNIV_INTERN
+buf_page_t*
+buf_page_reset_file_page_was_freed(
+/*===============================*/
+			/* out: control block if found in page hash table,
+			otherwise NULL */
+	ulint	space,	/* in: space id */
+	ulint	offset)	/* in: page number */
+{
+	buf_page_t*	bpage;
+
+	buf_pool_mutex_enter();
+
+	bpage = buf_page_hash_get(space, offset);
+
+	if (bpage) {
+		bpage->file_page_was_freed = FALSE;
+	}
+
+	buf_pool_mutex_exit();
+
+	return(bpage);
+}
+#endif /* UNIV_DEBUG_FILE_ACCESSES */
+
+/************************************************************************
+Get read access to a compressed page (usually of type
+FIL_PAGE_TYPE_ZBLOB or FIL_PAGE_TYPE_ZBLOB2).
+The page must be released with buf_page_release_zip().
+NOTE: the page is not protected by any latch.  Mutual exclusion has to
+be implemented at a higher level.  In other words, all possible
+accesses to a given page through this function must be protected by
+the same set of mutexes or latches. */
+UNIV_INTERN
+buf_page_t*
+buf_page_get_zip(
+/*=============*/
+				/* out: pointer to the block */
+	ulint		space,	/* in: space id */
+	ulint		zip_size,/* in: compressed page size */
+	ulint		offset)	/* in: page number */
+{
+	buf_page_t*	bpage;
+	mutex_t*	block_mutex;
+	ibool		must_read;
+
+#ifndef UNIV_LOG_DEBUG
+	ut_ad(!ibuf_inside());
+#endif
+	buf_pool->n_page_gets++;
+
+	for (;;) {
+		buf_pool_mutex_enter();
+lookup:
+		bpage = buf_page_hash_get(space, offset);
+		if (bpage) {
+			break;
+		}
+
+		/* Page not in buf_pool: needs to be read from file */
+
+		buf_pool_mutex_exit();
+
+		buf_read_page(space, zip_size, offset);
+
+#if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
+		ut_a(++buf_dbg_counter % 37 || buf_validate());
+#endif /* UNIV_DEBUG || UNIV_BUF_DEBUG */
+	}
+
+	if (UNIV_UNLIKELY(!bpage->zip.data)) {
+		/* There is no compressed page. */
+		buf_pool_mutex_exit();
+		return(NULL);
+	}
+
+	block_mutex = buf_page_get_mutex(bpage);
+	mutex_enter(block_mutex);
+
+	switch (buf_page_get_state(bpage)) {
+	case BUF_BLOCK_NOT_USED:
+	case BUF_BLOCK_READY_FOR_USE:
+	case BUF_BLOCK_MEMORY:
+	case BUF_BLOCK_REMOVE_HASH:
+	case BUF_BLOCK_ZIP_FREE:
+		ut_error;
+		break;
+	case BUF_BLOCK_ZIP_PAGE:
+	case BUF_BLOCK_ZIP_DIRTY:
+		bpage->buf_fix_count++;
+		break;
+	case BUF_BLOCK_FILE_PAGE:
+		/* Discard the uncompressed page frame if possible. */
+		if (buf_LRU_free_block(bpage, FALSE, NULL)
+		    == BUF_LRU_FREED) {
+
+			mutex_exit(block_mutex);
+			goto lookup;
+		}
+
+		buf_block_buf_fix_inc((buf_block_t*) bpage,
+				      __FILE__, __LINE__);
+		break;
+	}
+
+	must_read = buf_page_get_io_fix(bpage) == BUF_IO_READ;
+
+	buf_pool_mutex_exit();
+
+	buf_page_set_accessed(bpage, TRUE);
+
+	mutex_exit(block_mutex);
+
+	buf_block_make_young(bpage);
+
+#ifdef UNIV_DEBUG_FILE_ACCESSES
+	ut_a(!bpage->file_page_was_freed);
+#endif
+
+#if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
+	ut_a(++buf_dbg_counter % 5771 || buf_validate());
+	ut_a(bpage->buf_fix_count > 0);
+	ut_a(buf_page_in_file(bpage));
+#endif /* UNIV_DEBUG || UNIV_BUF_DEBUG */
+
+	if (must_read) {
+		/* Let us wait until the read operation
+		completes */
+
+		for (;;) {
+			enum buf_io_fix	io_fix;
+
+			mutex_enter(block_mutex);
+			io_fix = buf_page_get_io_fix(bpage);
+			mutex_exit(block_mutex);
+
+			if (io_fix == BUF_IO_READ) {
+
+				os_thread_sleep(WAIT_FOR_READ);
+			} else {
+				break;
+			}
+		}
+	}
+
+#ifdef UNIV_IBUF_COUNT_DEBUG
+	ut_a(ibuf_count_get(buf_page_get_space(bpage),
+			    buf_page_get_page_no(bpage)) == 0);
+#endif
+	return(bpage);
+}
+
+/************************************************************************
+Initialize some fields of a control block. */
+UNIV_INLINE
+void
+buf_block_init_low(
+/*===============*/
+	buf_block_t*	block)	/* in: block to init */
+{
+	block->check_index_page_at_flush = FALSE;
+	block->index		= NULL;
+
+	block->n_hash_helps	= 0;
+	block->is_hashed	= FALSE;
+	block->n_fields		= 1;
+	block->n_bytes		= 0;
+	block->left_side	= TRUE;
+}
+
+/************************************************************************
+Decompress a block. */
+static
+ibool
+buf_zip_decompress(
+/*===============*/
+				/* out: TRUE if successful */
+	buf_block_t*	block,	/* in/out: block */
+	ibool		check)	/* in: TRUE=verify the page checksum */
+{
+	const byte* frame = block->page.zip.data;
+
+	ut_ad(buf_block_get_zip_size(block));
+	ut_a(buf_block_get_space(block) != 0);
+
+	if (UNIV_LIKELY(check)) {
+		ulint	stamp_checksum	= mach_read_from_4(
+			frame + FIL_PAGE_SPACE_OR_CHKSUM);
+		ulint	calc_checksum	= page_zip_calc_checksum(
+			frame, page_zip_get_size(&block->page.zip));
+
+		if (UNIV_UNLIKELY(stamp_checksum != calc_checksum)) {
+			ut_print_timestamp(stderr);
+			fprintf(stderr,
+				"  InnoDB: compressed page checksum mismatch"
+				" (space %u page %u): %lu != %lu\n",
+				block->page.space, block->page.offset,
+				stamp_checksum, calc_checksum);
+			return(FALSE);
+		}
+	}
+
+	switch (fil_page_get_type(frame)) {
+	case FIL_PAGE_INDEX:
+		if (page_zip_decompress(&block->page.zip,
+					block->frame)) {
+			return(TRUE);
+		}
+
+		fprintf(stderr,
+			"InnoDB: unable to decompress space %lu page %lu\n",
+			(ulong) block->page.space,
+			(ulong) block->page.offset);
+		return(FALSE);
+
+	case FIL_PAGE_TYPE_ALLOCATED:
+	case FIL_PAGE_INODE:
+	case FIL_PAGE_IBUF_BITMAP:
+	case FIL_PAGE_TYPE_FSP_HDR:
+	case FIL_PAGE_TYPE_XDES:
+	case FIL_PAGE_TYPE_ZBLOB:
+	case FIL_PAGE_TYPE_ZBLOB2:
+		/* Copy to uncompressed storage. */
+		memcpy(block->frame, frame,
+		       buf_block_get_zip_size(block));
+		return(TRUE);
+	}
+
+	ut_print_timestamp(stderr);
+	fprintf(stderr,
+		"  InnoDB: unknown compressed page"
+		" type %lu\n",
+		fil_page_get_type(frame));
+	return(FALSE);
+}
+
+/***********************************************************************
+Gets the block to whose frame the pointer is pointing to. */
+UNIV_INTERN
+buf_block_t*
+buf_block_align(
+/*============*/
+				/* out: pointer to block, never NULL */
+	const byte*	ptr)	/* in: pointer to a frame */
+{
+	buf_chunk_t*	chunk;
+	ulint		i;
+
+	/* TODO: protect buf_pool->chunks with a mutex (it will
+	currently remain constant after buf_pool_init()) */
+	for (chunk = buf_pool->chunks, i = buf_pool->n_chunks; i--; chunk++) {
+		lint	offs = ptr - chunk->blocks->frame;
+
+		if (UNIV_UNLIKELY(offs < 0)) {
+
+			continue;
+		}
+
+		offs >>= UNIV_PAGE_SIZE_SHIFT;
+
+		if (UNIV_LIKELY((ulint) offs < chunk->size)) {
+			buf_block_t*	block = &chunk->blocks[offs];
+
+			/* The function buf_chunk_init() invokes
+			buf_block_init() so that block[n].frame ==
+			block->frame + n * UNIV_PAGE_SIZE.  Check it. */
+			ut_ad(block->frame == page_align(ptr));
+#ifdef UNIV_DEBUG
+			/* A thread that updates these fields must
+			hold buf_pool_mutex and block->mutex.  Acquire
+			only the latter. */
+			mutex_enter(&block->mutex);
+
+			switch (buf_block_get_state(block)) {
+			case BUF_BLOCK_ZIP_FREE:
+			case BUF_BLOCK_ZIP_PAGE:
+			case BUF_BLOCK_ZIP_DIRTY:
+				/* These types should only be used in
+				the compressed buffer pool, whose
+				memory is allocated from
+				buf_pool->chunks, in UNIV_PAGE_SIZE
+				blocks flagged as BUF_BLOCK_MEMORY. */
+				ut_error;
+				break;
+			case BUF_BLOCK_NOT_USED:
+			case BUF_BLOCK_READY_FOR_USE:
+			case BUF_BLOCK_MEMORY:
+				/* Some data structures contain
+				"guess" pointers to file pages.  The
+				file pages may have been freed and
+				reused.  Do not complain. */
+				break;
+			case BUF_BLOCK_REMOVE_HASH:
+				/* buf_LRU_block_remove_hashed_page()
+				will overwrite the FIL_PAGE_OFFSET and
+				FIL_PAGE_ARCH_LOG_NO_OR_SPACE_ID with
+				0xff and set the state to
+				BUF_BLOCK_REMOVE_HASH. */
+				ut_ad(page_get_space_id(page_align(ptr))
+				      == 0xffffffff);
+				ut_ad(page_get_page_no(page_align(ptr))
+				      == 0xffffffff);
+				break;
+			case BUF_BLOCK_FILE_PAGE:
+				ut_ad(block->page.space
+				      == page_get_space_id(page_align(ptr)));
+				ut_ad(block->page.offset
+				      == page_get_page_no(page_align(ptr)));
+				break;
+			}
+
+			mutex_exit(&block->mutex);
+#endif /* UNIV_DEBUG */
+
+			return(block);
+		}
+	}
+
+	/* The block should always be found. */
+	ut_error;
+	return(NULL);
+}
+
+/************************************************************************
+Find out if a buffer block was created by buf_chunk_init(). */
+static
+ibool
+buf_block_is_uncompressed(
+/*======================*/
+					/* out: TRUE if "block" has
+					been added to buf_pool->free
+					by buf_chunk_init() */
+	const buf_block_t*	block)	/* in: pointer to block,
+					not dereferenced */
+{
+	const buf_chunk_t*		chunk	= buf_pool->chunks;
+	const buf_chunk_t* const	echunk	= chunk + buf_pool->n_chunks;
+
+	ut_ad(buf_pool_mutex_own());
+
+	if (UNIV_UNLIKELY((((ulint) block) % sizeof *block) != 0)) {
+		/* The pointer should be aligned. */
+		return(FALSE);
+	}
+
+	while (chunk < echunk) {
+		if (block >= chunk->blocks
+		    && block < chunk->blocks + chunk->size) {
+
+			return(TRUE);
+		}
+
+		chunk++;
+	}
+
+	return(FALSE);
+}
+
+/************************************************************************
+This is the general function used to get access to a database page. */
+UNIV_INTERN
+buf_block_t*
+buf_page_get_gen(
+/*=============*/
+				/* out: pointer to the block or NULL */
+	ulint		space,	/* in: space id */
+	ulint		zip_size,/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	ulint		offset,	/* in: page number */
+	ulint		rw_latch,/* in: RW_S_LATCH, RW_X_LATCH, RW_NO_LATCH */
+	buf_block_t*	guess,	/* in: guessed block or NULL */
+	ulint		mode,	/* in: BUF_GET, BUF_GET_IF_IN_POOL,
+				BUF_GET_NO_LATCH */
+	const char*	file,	/* in: file name */
+	ulint		line,	/* in: line where called */
+	mtr_t*		mtr)	/* in: mini-transaction */
+{
+	buf_block_t*	block;
+	ibool		accessed;
+	ulint		fix_type;
+	ibool		must_read;
+
+	ut_ad(mtr);
+	ut_ad((rw_latch == RW_S_LATCH)
+	      || (rw_latch == RW_X_LATCH)
+	      || (rw_latch == RW_NO_LATCH));
+	ut_ad((mode != BUF_GET_NO_LATCH) || (rw_latch == RW_NO_LATCH));
+	ut_ad((mode == BUF_GET) || (mode == BUF_GET_IF_IN_POOL)
+	      || (mode == BUF_GET_NO_LATCH));
+	ut_ad(zip_size == fil_space_get_zip_size(space));
+#ifndef UNIV_LOG_DEBUG
+	ut_ad(!ibuf_inside() || ibuf_page(space, zip_size, offset, NULL));
+#endif
+	buf_pool->n_page_gets++;
+loop:
+	block = guess;
+	buf_pool_mutex_enter();
+
+	if (block) {
+		/* If the guess is a compressed page descriptor that
+		has been allocated by buf_buddy_alloc(), it may have
+		been invalidated by buf_buddy_relocate().  In that
+		case, block could point to something that happens to
+		contain the expected bits in block->page.  Similarly,
+		the guess may be pointing to a buffer pool chunk that
+		has been released when resizing the buffer pool. */
+
+		if (!buf_block_is_uncompressed(block)
+		    || offset != block->page.offset
+		    || space != block->page.space
+		    || buf_block_get_state(block) != BUF_BLOCK_FILE_PAGE) {
+
+			block = guess = NULL;
+		} else {
+			ut_ad(!block->page.in_zip_hash);
+			ut_ad(block->page.in_page_hash);
+		}
+	}
+
+	if (block == NULL) {
+		block = (buf_block_t*) buf_page_hash_get(space, offset);
+	}
+
+loop2:
+	if (block == NULL) {
+		/* Page not in buf_pool: needs to be read from file */
+
+		buf_pool_mutex_exit();
+
+		if (mode == BUF_GET_IF_IN_POOL) {
+
+			return(NULL);
+		}
+
+		buf_read_page(space, zip_size, offset);
+
+#if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
+		ut_a(++buf_dbg_counter % 37 || buf_validate());
+#endif /* UNIV_DEBUG || UNIV_BUF_DEBUG */
+		goto loop;
+	}
+
+	ut_ad(page_zip_get_size(&block->page.zip) == zip_size);
+
+	must_read = buf_block_get_io_fix(block) == BUF_IO_READ;
+
+	if (must_read && mode == BUF_GET_IF_IN_POOL) {
+		/* The page is only being read to buffer */
+		buf_pool_mutex_exit();
+
+		return(NULL);
+	}
+
+	switch (buf_block_get_state(block)) {
+		buf_page_t*	bpage;
+		ibool		success;
+
+	case BUF_BLOCK_FILE_PAGE:
+		break;
+
+	case BUF_BLOCK_ZIP_PAGE:
+	case BUF_BLOCK_ZIP_DIRTY:
+		bpage = &block->page;
+
+		if (bpage->buf_fix_count
+		    || buf_page_get_io_fix(bpage) != BUF_IO_NONE) {
+			/* This condition often occurs when the buffer
+			is not buffer-fixed, but I/O-fixed by
+			buf_page_init_for_read(). */
+wait_until_unfixed:
+			/* The block is buffer-fixed or I/O-fixed.
+			Try again later. */
+			buf_pool_mutex_exit();
+			os_thread_sleep(WAIT_FOR_READ);
+
+			goto loop;
+		}
+
+		/* Allocate an uncompressed page. */
+		buf_pool_mutex_exit();
+
+		block = buf_LRU_get_free_block(0);
+		ut_a(block);
+
+		buf_pool_mutex_enter();
+		mutex_enter(&block->mutex);
+
+		{
+			buf_page_t*	hash_bpage
+				= buf_page_hash_get(space, offset);
+
+			if (UNIV_UNLIKELY(bpage != hash_bpage)) {
+				/* The buf_pool->page_hash was modified
+				while buf_pool_mutex was released.
+				Free the block that was allocated. */
+
+				buf_LRU_block_free_non_file_page(block);
+				mutex_exit(&block->mutex);
+
+				block = (buf_block_t*) hash_bpage;
+				goto loop2;
+			}
+		}
+
+		if (UNIV_UNLIKELY
+		    (bpage->buf_fix_count
+		     || buf_page_get_io_fix(bpage) != BUF_IO_NONE)) {
+
+			/* The block was buffer-fixed or I/O-fixed
+			while buf_pool_mutex was not held by this thread.
+			Free the block that was allocated and try again.
+			This should be extremely unlikely. */
+
+			buf_LRU_block_free_non_file_page(block);
+			mutex_exit(&block->mutex);
+
+			goto wait_until_unfixed;
+		}
+
+		/* Move the compressed page from bpage to block,
+		and uncompress it. */
+
+		mutex_enter(&buf_pool_zip_mutex);
+
+		buf_relocate(bpage, &block->page);
+		buf_block_init_low(block);
+		block->lock_hash_val = lock_rec_hash(space, offset);
+
+		UNIV_MEM_DESC(&block->page.zip.data,
+			      page_zip_get_size(&block->page.zip), block);
+
+		if (buf_page_get_state(&block->page)
+		    == BUF_BLOCK_ZIP_PAGE) {
+			UT_LIST_REMOVE(list, buf_pool->zip_clean,
+				       &block->page);
+			ut_ad(!block->page.in_flush_list);
+		} else {
+			/* Relocate buf_pool->flush_list. */
+			buf_page_t*	b;
+
+			b = UT_LIST_GET_PREV(list, &block->page);
+			ut_ad(block->page.in_flush_list);
+			UT_LIST_REMOVE(list, buf_pool->flush_list,
+				       &block->page);
+
+			if (b) {
+				UT_LIST_INSERT_AFTER(
+					list, buf_pool->flush_list, b,
+					&block->page);
+			} else {
+				UT_LIST_ADD_FIRST(
+					list, buf_pool->flush_list,
+					&block->page);
+			}
+		}
+
+		/* Buffer-fix, I/O-fix, and X-latch the block
+		for the duration of the decompression.
+		Also add the block to the unzip_LRU list. */
+		block->page.state = BUF_BLOCK_FILE_PAGE;
+
+		/* Insert at the front of unzip_LRU list */
+		buf_unzip_LRU_add_block(block, FALSE);
+
+		block->page.buf_fix_count = 1;
+		buf_block_set_io_fix(block, BUF_IO_READ);
+		buf_pool->n_pend_unzip++;
+		rw_lock_x_lock(&block->lock);
+		mutex_exit(&block->mutex);
+		mutex_exit(&buf_pool_zip_mutex);
+
+		buf_buddy_free(bpage, sizeof *bpage);
+
+		buf_pool_mutex_exit();
+
+		/* Decompress the page and apply buffered operations
+		while not holding buf_pool_mutex or block->mutex. */
+		success = buf_zip_decompress(block, srv_use_checksums);
+
+		if (UNIV_LIKELY(success)) {
+			ibuf_merge_or_delete_for_page(block, space, offset,
+						      zip_size, TRUE);
+		}
+
+		/* Unfix and unlatch the block. */
+		buf_pool_mutex_enter();
+		mutex_enter(&block->mutex);
+		buf_pool->n_pend_unzip--;
+		block->page.buf_fix_count--;
+		buf_block_set_io_fix(block, BUF_IO_NONE);
+		mutex_exit(&block->mutex);
+		rw_lock_x_unlock(&block->lock);
+
+		if (UNIV_UNLIKELY(!success)) {
+
+			buf_pool_mutex_exit();
+			return(NULL);
+		}
+
+		break;
+
+	case BUF_BLOCK_ZIP_FREE:
+	case BUF_BLOCK_NOT_USED:
+	case BUF_BLOCK_READY_FOR_USE:
+	case BUF_BLOCK_MEMORY:
+	case BUF_BLOCK_REMOVE_HASH:
+		ut_error;
+		break;
+	}
+
+	ut_ad(buf_block_get_state(block) == BUF_BLOCK_FILE_PAGE);
+
+	mutex_enter(&block->mutex);
+	UNIV_MEM_ASSERT_RW(&block->page, sizeof block->page);
+
+	buf_block_buf_fix_inc(block, file, line);
+	buf_pool_mutex_exit();
+
+	/* Check if this is the first access to the page */
+
+	accessed = buf_page_is_accessed(&block->page);
+
+	buf_page_set_accessed(&block->page, TRUE);
+
+	mutex_exit(&block->mutex);
+
+	buf_block_make_young(&block->page);
+
+#ifdef UNIV_DEBUG_FILE_ACCESSES
+	ut_a(!block->page.file_page_was_freed);
+#endif
+
+#if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
+	ut_a(++buf_dbg_counter % 5771 || buf_validate());
+	ut_a(block->page.buf_fix_count > 0);
+	ut_a(buf_block_get_state(block) == BUF_BLOCK_FILE_PAGE);
+#endif /* UNIV_DEBUG || UNIV_BUF_DEBUG */
+
+	switch (rw_latch) {
+	case RW_NO_LATCH:
+		if (must_read) {
+			/* Let us wait until the read operation
+			completes */
+
+			for (;;) {
+				enum buf_io_fix	io_fix;
+
+				mutex_enter(&block->mutex);
+				io_fix = buf_block_get_io_fix(block);
+				mutex_exit(&block->mutex);
+
+				if (io_fix == BUF_IO_READ) {
+
+					os_thread_sleep(WAIT_FOR_READ);
+				} else {
+					break;
+				}
+			}
+		}
+
+		fix_type = MTR_MEMO_BUF_FIX;
+		break;
+
+	case RW_S_LATCH:
+		rw_lock_s_lock_func(&(block->lock), 0, file, line);
+
+		fix_type = MTR_MEMO_PAGE_S_FIX;
+		break;
+
+	default:
+		ut_ad(rw_latch == RW_X_LATCH);
+		rw_lock_x_lock_func(&(block->lock), 0, file, line);
+
+		fix_type = MTR_MEMO_PAGE_X_FIX;
+		break;
+	}
+
+	mtr_memo_push(mtr, block, fix_type);
+
+	if (!accessed) {
+		/* In the case of a first access, try to apply linear
+		read-ahead */
+
+		buf_read_ahead_linear(space, zip_size, offset);
+	}
+
+#ifdef UNIV_IBUF_COUNT_DEBUG
+	ut_a(ibuf_count_get(buf_block_get_space(block),
+			    buf_block_get_page_no(block)) == 0);
+#endif
+	return(block);
+}
+
+/************************************************************************
+This is the general function used to get optimistic access to a database
+page. */
+UNIV_INTERN
+ibool
+buf_page_optimistic_get_func(
+/*=========================*/
+				/* out: TRUE if success */
+	ulint		rw_latch,/* in: RW_S_LATCH, RW_X_LATCH */
+	buf_block_t*	block,	/* in: guessed buffer block */
+	ib_uint64_t	modify_clock,/* in: modify clock value if mode is
+				..._GUESS_ON_CLOCK */
+	const char*	file,	/* in: file name */
+	ulint		line,	/* in: line where called */
+	mtr_t*		mtr)	/* in: mini-transaction */
+{
+	ibool		accessed;
+	ibool		success;
+	ulint		fix_type;
+
+	ut_ad(mtr && block);
+	ut_ad((rw_latch == RW_S_LATCH) || (rw_latch == RW_X_LATCH));
+
+	mutex_enter(&block->mutex);
+
+	if (UNIV_UNLIKELY(buf_block_get_state(block) != BUF_BLOCK_FILE_PAGE)) {
+
+		mutex_exit(&block->mutex);
+
+		return(FALSE);
+	}
+
+	buf_block_buf_fix_inc(block, file, line);
+	accessed = buf_page_is_accessed(&block->page);
+	buf_page_set_accessed(&block->page, TRUE);
+
+	mutex_exit(&block->mutex);
+
+	buf_block_make_young(&block->page);
+
+	/* Check if this is the first access to the page */
+
+	ut_ad(!ibuf_inside()
+	      || ibuf_page(buf_block_get_space(block),
+			   buf_block_get_zip_size(block),
+			   buf_block_get_page_no(block), NULL));
+
+	if (rw_latch == RW_S_LATCH) {
+		success = rw_lock_s_lock_nowait(&(block->lock),
+						file, line);
+		fix_type = MTR_MEMO_PAGE_S_FIX;
+	} else {
+		success = rw_lock_x_lock_func_nowait(&(block->lock),
+						     file, line);
+		fix_type = MTR_MEMO_PAGE_X_FIX;
+	}
+
+	if (UNIV_UNLIKELY(!success)) {
+		mutex_enter(&block->mutex);
+		buf_block_buf_fix_dec(block);
+		mutex_exit(&block->mutex);
+
+		return(FALSE);
+	}
+
+	if (UNIV_UNLIKELY(modify_clock != block->modify_clock)) {
+		buf_block_dbg_add_level(block, SYNC_NO_ORDER_CHECK);
+
+		if (rw_latch == RW_S_LATCH) {
+			rw_lock_s_unlock(&(block->lock));
+		} else {
+			rw_lock_x_unlock(&(block->lock));
+		}
+
+		mutex_enter(&block->mutex);
+		buf_block_buf_fix_dec(block);
+		mutex_exit(&block->mutex);
+
+		return(FALSE);
+	}
+
+	mtr_memo_push(mtr, block, fix_type);
+
+#if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
+	ut_a(++buf_dbg_counter % 5771 || buf_validate());
+	ut_a(block->page.buf_fix_count > 0);
+	ut_a(buf_block_get_state(block) == BUF_BLOCK_FILE_PAGE);
+#endif /* UNIV_DEBUG || UNIV_BUF_DEBUG */
+
+#ifdef UNIV_DEBUG_FILE_ACCESSES
+	ut_a(block->page.file_page_was_freed == FALSE);
+#endif
+	if (UNIV_UNLIKELY(!accessed)) {
+		/* In the case of a first access, try to apply linear
+		read-ahead */
+
+		buf_read_ahead_linear(buf_block_get_space(block),
+				      buf_block_get_zip_size(block),
+				      buf_block_get_page_no(block));
+	}
+
+#ifdef UNIV_IBUF_COUNT_DEBUG
+	ut_a(ibuf_count_get(buf_block_get_space(block),
+			    buf_block_get_page_no(block)) == 0);
+#endif
+	buf_pool->n_page_gets++;
+
+	return(TRUE);
+}
+
+/************************************************************************
+This is used to get access to a known database page, when no waiting can be
+done. For example, if a search in an adaptive hash index leads us to this
+frame. */
+UNIV_INTERN
+ibool
+buf_page_get_known_nowait(
+/*======================*/
+				/* out: TRUE if success */
+	ulint		rw_latch,/* in: RW_S_LATCH, RW_X_LATCH */
+	buf_block_t*	block,	/* in: the known page */
+	ulint		mode,	/* in: BUF_MAKE_YOUNG or BUF_KEEP_OLD */
+	const char*	file,	/* in: file name */
+	ulint		line,	/* in: line where called */
+	mtr_t*		mtr)	/* in: mini-transaction */
+{
+	ibool		success;
+	ulint		fix_type;
+
+	ut_ad(mtr);
+	ut_ad((rw_latch == RW_S_LATCH) || (rw_latch == RW_X_LATCH));
+
+	mutex_enter(&block->mutex);
+
+	if (buf_block_get_state(block) == BUF_BLOCK_REMOVE_HASH) {
+		/* Another thread is just freeing the block from the LRU list
+		of the buffer pool: do not try to access this page; this
+		attempt to access the page can only come through the hash
+		index because when the buffer block state is ..._REMOVE_HASH,
+		we have already removed it from the page address hash table
+		of the buffer pool. */
+
+		mutex_exit(&block->mutex);
+
+		return(FALSE);
+	}
+
+	ut_a(buf_block_get_state(block) == BUF_BLOCK_FILE_PAGE);
+
+	buf_block_buf_fix_inc(block, file, line);
+
+	mutex_exit(&block->mutex);
+
+	if (mode == BUF_MAKE_YOUNG) {
+		buf_block_make_young(&block->page);
+	}
+
+	ut_ad(!ibuf_inside() || (mode == BUF_KEEP_OLD));
+
+	if (rw_latch == RW_S_LATCH) {
+		success = rw_lock_s_lock_nowait(&(block->lock),
+						file, line);
+		fix_type = MTR_MEMO_PAGE_S_FIX;
+	} else {
+		success = rw_lock_x_lock_func_nowait(&(block->lock),
+						     file, line);
+		fix_type = MTR_MEMO_PAGE_X_FIX;
+	}
+
+	if (!success) {
+		mutex_enter(&block->mutex);
+		buf_block_buf_fix_dec(block);
+		mutex_exit(&block->mutex);
+
+		return(FALSE);
+	}
+
+	mtr_memo_push(mtr, block, fix_type);
+
+#if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
+	ut_a(++buf_dbg_counter % 5771 || buf_validate());
+	ut_a(block->page.buf_fix_count > 0);
+	ut_a(buf_block_get_state(block) == BUF_BLOCK_FILE_PAGE);
+#endif /* UNIV_DEBUG || UNIV_BUF_DEBUG */
+#ifdef UNIV_DEBUG_FILE_ACCESSES
+	ut_a(block->page.file_page_was_freed == FALSE);
+#endif
+
+#ifdef UNIV_IBUF_COUNT_DEBUG
+	ut_a((mode == BUF_KEEP_OLD)
+	     || (ibuf_count_get(buf_block_get_space(block),
+				buf_block_get_page_no(block)) == 0));
+#endif
+	buf_pool->n_page_gets++;
+
+	return(TRUE);
+}
+
+/***********************************************************************
+Given a tablespace id and page number tries to get that page. If the
+page is not in the buffer pool it is not loaded and NULL is returned.
+Suitable for using when holding the kernel mutex. */
+UNIV_INTERN
+const buf_block_t*
+buf_page_try_get_func(
+/*==================*/
+				/* out: pointer to a page or NULL */
+	ulint		space_id,/* in: tablespace id */
+	ulint		page_no,/* in: page number */
+	const char*	file,	/* in: file name */
+	ulint		line,	/* in: line where called */
+	mtr_t*		mtr)	/* in: mini-transaction */
+{
+	buf_block_t*	block;
+	ibool		success;
+	ulint		fix_type;
+
+	buf_pool_mutex_enter();
+	block = buf_block_hash_get(space_id, page_no);
+
+	if (!block) {
+		buf_pool_mutex_exit();
+		return(NULL);
+	}
+
+	mutex_enter(&block->mutex);
+	buf_pool_mutex_exit();
+
+#if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
+	ut_a(buf_block_get_state(block) == BUF_BLOCK_FILE_PAGE);
+	ut_a(buf_block_get_space(block) == space_id);
+	ut_a(buf_block_get_page_no(block) == page_no);
+#endif /* UNIV_DEBUG || UNIV_BUF_DEBUG */
+
+	buf_block_buf_fix_inc(block, file, line);
+	mutex_exit(&block->mutex);
+
+	fix_type = MTR_MEMO_PAGE_S_FIX;
+	success = rw_lock_s_lock_nowait(&block->lock, file, line);
+
+	if (!success) {
+		/* Let us try to get an X-latch. If the current thread
+		is holding an X-latch on the page, we cannot get an
+		S-latch. */
+
+		fix_type = MTR_MEMO_PAGE_X_FIX;
+		success = rw_lock_x_lock_func_nowait(&block->lock,
+						     file, line);
+	}
+
+	if (!success) {
+		mutex_enter(&block->mutex);
+		buf_block_buf_fix_dec(block);
+		mutex_exit(&block->mutex);
+
+		return(NULL);
+	}
+
+	mtr_memo_push(mtr, block, fix_type);
+#if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
+	ut_a(++buf_dbg_counter % 5771 || buf_validate());
+	ut_a(block->page.buf_fix_count > 0);
+	ut_a(buf_block_get_state(block) == BUF_BLOCK_FILE_PAGE);
+#endif /* UNIV_DEBUG || UNIV_BUF_DEBUG */
+#ifdef UNIV_DEBUG_FILE_ACCESSES
+	ut_a(block->page.file_page_was_freed == FALSE);
+#endif /* UNIV_DEBUG_FILE_ACCESSES */
+	buf_block_dbg_add_level(block, SYNC_NO_ORDER_CHECK);
+
+	buf_pool->n_page_gets++;
+
+#ifdef UNIV_IBUF_COUNT_DEBUG
+	ut_a(ibuf_count_get(buf_block_get_space(block),
+			    buf_block_get_page_no(block)) == 0);
+#endif
+
+	return(block);
+}
+
+/************************************************************************
+Initialize some fields of a control block. */
+UNIV_INLINE
+void
+buf_page_init_low(
+/*==============*/
+	buf_page_t*	bpage)	/* in: block to init */
+{
+	bpage->flush_type = BUF_FLUSH_LRU;
+	bpage->accessed = FALSE;
+	bpage->io_fix = BUF_IO_NONE;
+	bpage->buf_fix_count = 0;
+	bpage->freed_page_clock = 0;
+	bpage->newest_modification = 0;
+	bpage->oldest_modification = 0;
+	HASH_INVALIDATE(bpage, hash);
+#ifdef UNIV_DEBUG_FILE_ACCESSES
+	bpage->file_page_was_freed = FALSE;
+#endif /* UNIV_DEBUG_FILE_ACCESSES */
+}
+
+#ifdef UNIV_HOTBACKUP
+/************************************************************************
+Inits a page to the buffer buf_pool, for use in ibbackup --restore. */
+UNIV_INTERN
+void
+buf_page_init_for_backup_restore(
+/*=============================*/
+	ulint		space,	/* in: space id */
+	ulint		offset,	/* in: offset of the page within space
+				in units of a page */
+	ulint		zip_size,/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	buf_block_t*	block)	/* in: block to init */
+{
+	buf_block_init_low(block);
+
+	block->lock_hash_val	= 0;
+
+	buf_page_init_low(&block->page);
+	block->page.state	= BUF_BLOCK_FILE_PAGE;
+	block->page.space	= space;
+	block->page.offset	= offset;
+
+	page_zip_des_init(&block->page.zip);
+
+	/* We assume that block->page.data has been allocated
+	with zip_size == UNIV_PAGE_SIZE. */
+	ut_ad(zip_size <= UNIV_PAGE_SIZE);
+	ut_ad(ut_is_2pow(zip_size));
+	page_zip_set_size(&block->page.zip, zip_size);
+}
+#endif /* UNIV_HOTBACKUP */
+
+/************************************************************************
+Inits a page to the buffer buf_pool. */
+static
+void
+buf_page_init(
+/*==========*/
+	ulint		space,	/* in: space id */
+	ulint		offset,	/* in: offset of the page within space
+				in units of a page */
+	buf_block_t*	block)	/* in: block to init */
+{
+	buf_page_t*	hash_page;
+
+	ut_ad(buf_pool_mutex_own());
+	ut_ad(mutex_own(&(block->mutex)));
+	ut_a(buf_block_get_state(block) != BUF_BLOCK_FILE_PAGE);
+
+	/* Set the state of the block */
+	buf_block_set_file_page(block, space, offset);
+
+#ifdef UNIV_DEBUG_VALGRIND
+	if (!space) {
+		/* Silence valid Valgrind warnings about uninitialized
+		data being written to data files.  There are some unused
+		bytes on some pages that InnoDB does not initialize. */
+		UNIV_MEM_VALID(block->frame, UNIV_PAGE_SIZE);
+	}
+#endif /* UNIV_DEBUG_VALGRIND */
+
+	buf_block_init_low(block);
+
+	block->lock_hash_val	= lock_rec_hash(space, offset);
+
+	/* Insert into the hash table of file pages */
+
+	hash_page = buf_page_hash_get(space, offset);
+
+	if (UNIV_LIKELY_NULL(hash_page)) {
+		fprintf(stderr,
+			"InnoDB: Error: page %lu %lu already found"
+			" in the hash table: %p, %p\n",
+			(ulong) space,
+			(ulong) offset,
+			(const void*) hash_page, (const void*) block);
+#if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
+		mutex_exit(&block->mutex);
+		buf_pool_mutex_exit();
+		buf_print();
+		buf_LRU_print();
+		buf_validate();
+		buf_LRU_validate();
+#endif /* UNIV_DEBUG || UNIV_BUF_DEBUG */
+		ut_error;
+	}
+
+	buf_page_init_low(&block->page);
+
+	ut_ad(!block->page.in_zip_hash);
+	ut_ad(!block->page.in_page_hash);
+	ut_d(block->page.in_page_hash = TRUE);
+	HASH_INSERT(buf_page_t, hash, buf_pool->page_hash,
+		    buf_page_address_fold(space, offset), &block->page);
+}
+
+/************************************************************************
+Function which inits a page for read to the buffer buf_pool. If the page is
+(1) already in buf_pool, or
+(2) if we specify to read only ibuf pages and the page is not an ibuf page, or
+(3) if the space is deleted or being deleted,
+then this function does nothing.
+Sets the io_fix flag to BUF_IO_READ and sets a non-recursive exclusive lock
+on the buffer frame. The io-handler must take care that the flag is cleared
+and the lock released later. */
+UNIV_INTERN
+buf_page_t*
+buf_page_init_for_read(
+/*===================*/
+				/* out: pointer to the block or NULL */
+	ulint*		err,	/* out: DB_SUCCESS or DB_TABLESPACE_DELETED */
+	ulint		mode,	/* in: BUF_READ_IBUF_PAGES_ONLY, ... */
+	ulint		space,	/* in: space id */
+	ulint		zip_size,/* in: compressed page size, or 0 */
+	ibool		unzip,	/* in: TRUE=request uncompressed page */
+	ib_int64_t	tablespace_version,/* in: prevents reading from a wrong
+				version of the tablespace in case we have done
+				DISCARD + IMPORT */
+	ulint		offset)	/* in: page number */
+{
+	buf_block_t*	block;
+	buf_page_t*	bpage;
+	mtr_t		mtr;
+	ibool		lru	= FALSE;
+	void*		data;
+
+	ut_ad(buf_pool);
+
+	*err = DB_SUCCESS;
+
+	if (mode == BUF_READ_IBUF_PAGES_ONLY) {
+		/* It is a read-ahead within an ibuf routine */
+
+		ut_ad(!ibuf_bitmap_page(zip_size, offset));
+		ut_ad(ibuf_inside());
+
+		mtr_start(&mtr);
+
+		if (!recv_no_ibuf_operations
+		    && !ibuf_page(space, zip_size, offset, &mtr)) {
+
+			mtr_commit(&mtr);
+
+			return(NULL);
+		}
+	} else {
+		ut_ad(mode == BUF_READ_ANY_PAGE);
+	}
+
+	if (zip_size && UNIV_LIKELY(!unzip)
+	    && UNIV_LIKELY(!recv_recovery_is_on())) {
+		block = NULL;
+	} else {
+		block = buf_LRU_get_free_block(0);
+		ut_ad(block);
+	}
+
+	buf_pool_mutex_enter();
+
+	if (buf_page_hash_get(space, offset)) {
+		/* The page is already in the buffer pool. */
+err_exit:
+		if (block) {
+			mutex_enter(&block->mutex);
+			buf_LRU_block_free_non_file_page(block);
+			mutex_exit(&block->mutex);
+		}
+
+		bpage = NULL;
+		goto func_exit;
+	}
+
+	if (fil_tablespace_deleted_or_being_deleted_in_mem(
+		    space, tablespace_version)) {
+		/* The page belongs to a space which has been
+		deleted or is being deleted. */
+		*err = DB_TABLESPACE_DELETED;
+
+		goto err_exit;
+	}
+
+	if (block) {
+		bpage = &block->page;
+		mutex_enter(&block->mutex);
+		buf_page_init(space, offset, block);
+
+		/* The block must be put to the LRU list, to the old blocks */
+		buf_LRU_add_block(bpage, TRUE/* to old blocks */);
+
+		/* We set a pass-type x-lock on the frame because then
+		the same thread which called for the read operation
+		(and is running now at this point of code) can wait
+		for the read to complete by waiting for the x-lock on
+		the frame; if the x-lock were recursive, the same
+		thread would illegally get the x-lock before the page
+		read is completed.  The x-lock is cleared by the
+		io-handler thread. */
+
+		rw_lock_x_lock_gen(&block->lock, BUF_IO_READ);
+		buf_page_set_io_fix(bpage, BUF_IO_READ);
+
+		if (UNIV_UNLIKELY(zip_size)) {
+			page_zip_set_size(&block->page.zip, zip_size);
+
+			/* buf_pool_mutex may be released and
+			reacquired by buf_buddy_alloc().  Thus, we
+			must release block->mutex in order not to
+			break the latching order in the reacquisition
+			of buf_pool_mutex.  We also must defer this
+			operation until after the block descriptor has
+			been added to buf_pool->LRU and
+			buf_pool->page_hash. */
+			mutex_exit(&block->mutex);
+			data = buf_buddy_alloc(zip_size, &lru);
+			mutex_enter(&block->mutex);
+			block->page.zip.data = data;
+
+			/* To maintain the invariant
+			block->in_unzip_LRU_list
+			== buf_page_belongs_to_unzip_LRU(&block->page)
+			we have to add this block to unzip_LRU
+			after block->page.zip.data is set. */
+			ut_ad(buf_page_belongs_to_unzip_LRU(&block->page));
+			buf_unzip_LRU_add_block(block, TRUE);
+		}
+
+		mutex_exit(&block->mutex);
+	} else {
+		/* Defer buf_buddy_alloc() until after the block has
+		been found not to exist.  The buf_buddy_alloc() and
+		buf_buddy_free() calls may be expensive because of
+		buf_buddy_relocate(). */
+
+		/* The compressed page must be allocated before the
+		control block (bpage), in order to avoid the
+		invocation of buf_buddy_relocate_block() on
+		uninitialized data. */
+		data = buf_buddy_alloc(zip_size, &lru);
+		bpage = buf_buddy_alloc(sizeof *bpage, &lru);
+
+		/* If buf_buddy_alloc() allocated storage from the LRU list,
+		it released and reacquired buf_pool_mutex.  Thus, we must
+		check the page_hash again, as it may have been modified. */
+		if (UNIV_UNLIKELY(lru)
+		    && UNIV_LIKELY_NULL(buf_page_hash_get(space, offset))) {
+
+			/* The block was added by some other thread. */
+			buf_buddy_free(bpage, sizeof *bpage);
+			buf_buddy_free(data, zip_size);
+
+			bpage = NULL;
+			goto func_exit;
+		}
+
+		page_zip_des_init(&bpage->zip);
+		page_zip_set_size(&bpage->zip, zip_size);
+		bpage->zip.data = data;
+
+		mutex_enter(&buf_pool_zip_mutex);
+		UNIV_MEM_DESC(bpage->zip.data,
+			      page_zip_get_size(&bpage->zip), bpage);
+		buf_page_init_low(bpage);
+		bpage->state	= BUF_BLOCK_ZIP_PAGE;
+		bpage->space	= space;
+		bpage->offset	= offset;
+
+#ifdef UNIV_DEBUG
+		bpage->in_page_hash = FALSE;
+		bpage->in_zip_hash = FALSE;
+		bpage->in_flush_list = FALSE;
+		bpage->in_free_list = FALSE;
+#endif /* UNIV_DEBUG */
+		bpage->in_LRU_list = FALSE;
+
+		ut_d(bpage->in_page_hash = TRUE);
+		HASH_INSERT(buf_page_t, hash, buf_pool->page_hash,
+			    buf_page_address_fold(space, offset), bpage);
+
+		/* The block must be put to the LRU list, to the old blocks */
+		buf_LRU_add_block(bpage, TRUE/* to old blocks */);
+		buf_LRU_insert_zip_clean(bpage);
+
+		buf_page_set_io_fix(bpage, BUF_IO_READ);
+
+		mutex_exit(&buf_pool_zip_mutex);
+	}
+
+	buf_pool->n_pend_reads++;
+func_exit:
+	buf_pool_mutex_exit();
+
+	if (mode == BUF_READ_IBUF_PAGES_ONLY) {
+
+		mtr_commit(&mtr);
+	}
+
+	ut_ad(!bpage || buf_page_in_file(bpage));
+	return(bpage);
+}
+
+/************************************************************************
+Initializes a page to the buffer buf_pool. The page is usually not read
+from a file even if it cannot be found in the buffer buf_pool. This is one
+of the functions which perform to a block a state transition NOT_USED =>
+FILE_PAGE (the other is buf_page_get_gen). */
+UNIV_INTERN
+buf_block_t*
+buf_page_create(
+/*============*/
+			/* out: pointer to the block, page bufferfixed */
+	ulint	space,	/* in: space id */
+	ulint	offset,	/* in: offset of the page within space in units of
+			a page */
+	ulint	zip_size,/* in: compressed page size, or 0 */
+	mtr_t*	mtr)	/* in: mini-transaction handle */
+{
+	buf_frame_t*	frame;
+	buf_block_t*	block;
+	buf_block_t*	free_block	= NULL;
+
+	ut_ad(mtr);
+	ut_ad(space || !zip_size);
+
+	free_block = buf_LRU_get_free_block(0);
+
+	buf_pool_mutex_enter();
+
+	block = (buf_block_t*) buf_page_hash_get(space, offset);
+
+	if (block && buf_page_in_file(&block->page)) {
+#ifdef UNIV_IBUF_COUNT_DEBUG
+		ut_a(ibuf_count_get(space, offset) == 0);
+#endif
+#ifdef UNIV_DEBUG_FILE_ACCESSES
+		block->page.file_page_was_freed = FALSE;
+#endif /* UNIV_DEBUG_FILE_ACCESSES */
+
+		/* Page can be found in buf_pool */
+		buf_pool_mutex_exit();
+
+		buf_block_free(free_block);
+
+		return(buf_page_get_with_no_latch(space, zip_size,
+						  offset, mtr));
+	}
+
+	/* If we get here, the page was not in buf_pool: init it there */
+
+#ifdef UNIV_DEBUG
+	if (buf_debug_prints) {
+		fprintf(stderr, "Creating space %lu page %lu to buffer\n",
+			(ulong) space, (ulong) offset);
+	}
+#endif /* UNIV_DEBUG */
+
+	block = free_block;
+
+	mutex_enter(&block->mutex);
+
+	buf_page_init(space, offset, block);
+
+	/* The block must be put to the LRU list */
+	buf_LRU_add_block(&block->page, FALSE);
+
+	buf_block_buf_fix_inc(block, __FILE__, __LINE__);
+	buf_pool->n_pages_created++;
+
+	if (zip_size) {
+		void*	data;
+		ibool	lru;
+
+		/* Prevent race conditions during buf_buddy_alloc(),
+		which may release and reacquire buf_pool_mutex,
+		by IO-fixing and X-latching the block. */
+
+		buf_page_set_io_fix(&block->page, BUF_IO_READ);
+		rw_lock_x_lock(&block->lock);
+
+		page_zip_set_size(&block->page.zip, zip_size);
+		mutex_exit(&block->mutex);
+		/* buf_pool_mutex may be released and reacquired by
+		buf_buddy_alloc().  Thus, we must release block->mutex
+		in order not to break the latching order in
+		the reacquisition of buf_pool_mutex.  We also must
+		defer this operation until after the block descriptor
+		has been added to buf_pool->LRU and buf_pool->page_hash. */
+		data = buf_buddy_alloc(zip_size, &lru);
+		mutex_enter(&block->mutex);
+		block->page.zip.data = data;
+
+		/* To maintain the invariant
+		block->in_unzip_LRU_list
+		== buf_page_belongs_to_unzip_LRU(&block->page)
+		we have to add this block to unzip_LRU after
+		block->page.zip.data is set. */
+		ut_ad(buf_page_belongs_to_unzip_LRU(&block->page));
+		buf_unzip_LRU_add_block(block, FALSE);
+
+		buf_page_set_io_fix(&block->page, BUF_IO_NONE);
+		rw_lock_x_unlock(&block->lock);
+	}
+
+	buf_pool_mutex_exit();
+
+	mtr_memo_push(mtr, block, MTR_MEMO_BUF_FIX);
+
+	buf_page_set_accessed(&block->page, TRUE);
+
+	mutex_exit(&block->mutex);
+
+	/* Delete possible entries for the page from the insert buffer:
+	such can exist if the page belonged to an index which was dropped */
+
+	ibuf_merge_or_delete_for_page(NULL, space, offset, zip_size, TRUE);
+
+	/* Flush pages from the end of the LRU list if necessary */
+	buf_flush_free_margin(FALSE);
+
+	frame = block->frame;
+
+	memset(frame + FIL_PAGE_PREV, 0xff, 4);
+	memset(frame + FIL_PAGE_NEXT, 0xff, 4);
+	mach_write_to_2(frame + FIL_PAGE_TYPE, FIL_PAGE_TYPE_ALLOCATED);
+
+	/* Reset to zero the file flush lsn field in the page; if the first
+	page of an ibdata file is 'created' in this function into the buffer
+	pool then we lose the original contents of the file flush lsn stamp.
+	Then InnoDB could in a crash recovery print a big, false, corruption
+	warning if the stamp contains an lsn bigger than the ib_logfile lsn. */
+
+	memset(frame + FIL_PAGE_FILE_FLUSH_LSN, 0, 8);
+
+#if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
+	ut_a(++buf_dbg_counter % 357 || buf_validate());
+#endif /* UNIV_DEBUG || UNIV_BUF_DEBUG */
+#ifdef UNIV_IBUF_COUNT_DEBUG
+	ut_a(ibuf_count_get(buf_block_get_space(block),
+			    buf_block_get_page_no(block)) == 0);
+#endif
+	return(block);
+}
+
+/************************************************************************
+Completes an asynchronous read or write request of a file page to or from
+the buffer pool. */
+UNIV_INTERN
+void
+buf_page_io_complete(
+/*=================*/
+	buf_page_t*	bpage)	/* in: pointer to the block in question */
+{
+	enum buf_io_fix	io_type;
+	const ibool	uncompressed = (buf_page_get_state(bpage)
+					== BUF_BLOCK_FILE_PAGE);
+
+	ut_a(buf_page_in_file(bpage));
+
+	/* We do not need protect io_fix here by mutex to read
+	it because this is the only function where we can change the value
+	from BUF_IO_READ or BUF_IO_WRITE to some other value, and our code
+	ensures that this is the only thread that handles the i/o for this
+	block. */
+
+	io_type = buf_page_get_io_fix(bpage);
+	ut_ad(io_type == BUF_IO_READ || io_type == BUF_IO_WRITE);
+
+	if (io_type == BUF_IO_READ) {
+		ulint	read_page_no;
+		ulint	read_space_id;
+		byte*	frame;
+
+		if (buf_page_get_zip_size(bpage)) {
+			frame = bpage->zip.data;
+			buf_pool->n_pend_unzip++;
+			if (uncompressed
+			    && !buf_zip_decompress((buf_block_t*) bpage,
+						   FALSE)) {
+
+				buf_pool->n_pend_unzip--;
+				goto corrupt;
+			}
+			buf_pool->n_pend_unzip--;
+		} else {
+			ut_a(uncompressed);
+			frame = ((buf_block_t*) bpage)->frame;
+		}
+
+		/* If this page is not uninitialized and not in the
+		doublewrite buffer, then the page number and space id
+		should be the same as in block. */
+		read_page_no = mach_read_from_4(frame + FIL_PAGE_OFFSET);
+		read_space_id = mach_read_from_4(
+			frame + FIL_PAGE_ARCH_LOG_NO_OR_SPACE_ID);
+
+		if (bpage->space == TRX_SYS_SPACE
+		    && trx_doublewrite_page_inside(bpage->offset)) {
+
+			ut_print_timestamp(stderr);
+			fprintf(stderr,
+				"  InnoDB: Error: reading page %lu\n"
+				"InnoDB: which is in the"
+				" doublewrite buffer!\n",
+				(ulong) bpage->offset);
+		} else if (!read_space_id && !read_page_no) {
+			/* This is likely an uninitialized page. */
+		} else if ((bpage->space
+			    && bpage->space != read_space_id)
+			   || bpage->offset != read_page_no) {
+			/* We did not compare space_id to read_space_id
+			if bpage->space == 0, because the field on the
+			page may contain garbage in MySQL < 4.1.1,
+			which only supported bpage->space == 0. */
+
+			ut_print_timestamp(stderr);
+			fprintf(stderr,
+				"  InnoDB: Error: space id and page n:o"
+				" stored in the page\n"
+				"InnoDB: read in are %lu:%lu,"
+				" should be %lu:%lu!\n",
+				(ulong) read_space_id, (ulong) read_page_no,
+				(ulong) bpage->space,
+				(ulong) bpage->offset);
+		}
+
+		/* From version 3.23.38 up we store the page checksum
+		to the 4 first bytes of the page end lsn field */
+
+		if (buf_page_is_corrupted(frame,
+					  buf_page_get_zip_size(bpage))) {
+corrupt:
+			fprintf(stderr,
+				"InnoDB: Database page corruption on disk"
+				" or a failed\n"
+				"InnoDB: file read of page %lu.\n"
+				"InnoDB: You may have to recover"
+				" from a backup.\n",
+				(ulong) bpage->offset);
+			buf_page_print(frame, buf_page_get_zip_size(bpage));
+			fprintf(stderr,
+				"InnoDB: Database page corruption on disk"
+				" or a failed\n"
+				"InnoDB: file read of page %lu.\n"
+				"InnoDB: You may have to recover"
+				" from a backup.\n",
+				(ulong) bpage->offset);
+			fputs("InnoDB: It is also possible that"
+			      " your operating\n"
+			      "InnoDB: system has corrupted its"
+			      " own file cache\n"
+			      "InnoDB: and rebooting your computer"
+			      " removes the\n"
+			      "InnoDB: error.\n"
+			      "InnoDB: If the corrupt page is an index page\n"
+			      "InnoDB: you can also try to"
+			      " fix the corruption\n"
+			      "InnoDB: by dumping, dropping,"
+			      " and reimporting\n"
+			      "InnoDB: the corrupt table."
+			      " You can use CHECK\n"
+			      "InnoDB: TABLE to scan your"
+			      " table for corruption.\n"
+			      "InnoDB: See also"
+			      " http://dev.mysql.com/doc/refman/5.1/en/"
+			      "forcing-recovery.html\n"
+			      "InnoDB: about forcing recovery.\n", stderr);
+
+			if (srv_force_recovery < SRV_FORCE_IGNORE_CORRUPT) {
+				fputs("InnoDB: Ending processing because of"
+				      " a corrupt database page.\n",
+				      stderr);
+				exit(1);
+			}
+		}
+
+		if (recv_recovery_is_on()) {
+			/* Pages must be uncompressed for crash recovery. */
+			ut_a(uncompressed);
+			recv_recover_page(FALSE, TRUE, (buf_block_t*) bpage);
+		}
+
+		if (uncompressed && !recv_no_ibuf_operations) {
+			ibuf_merge_or_delete_for_page(
+				(buf_block_t*) bpage, bpage->space,
+				bpage->offset, buf_page_get_zip_size(bpage),
+				TRUE);
+		}
+	}
+
+	buf_pool_mutex_enter();
+	mutex_enter(buf_page_get_mutex(bpage));
+
+#ifdef UNIV_IBUF_COUNT_DEBUG
+	if (io_type == BUF_IO_WRITE || uncompressed) {
+		/* For BUF_IO_READ of compressed-only blocks, the
+		buffered operations will be merged by buf_page_get_gen()
+		after the block has been uncompressed. */
+		ut_a(ibuf_count_get(bpage->space, bpage->offset) == 0);
+	}
+#endif
+	/* Because this thread which does the unlocking is not the same that
+	did the locking, we use a pass value != 0 in unlock, which simply
+	removes the newest lock debug record, without checking the thread
+	id. */
+
+	buf_page_set_io_fix(bpage, BUF_IO_NONE);
+
+	switch (io_type) {
+	case BUF_IO_READ:
+		/* NOTE that the call to ibuf may have moved the ownership of
+		the x-latch to this OS thread: do not let this confuse you in
+		debugging! */
+
+		ut_ad(buf_pool->n_pend_reads > 0);
+		buf_pool->n_pend_reads--;
+		buf_pool->n_pages_read++;
+
+		if (uncompressed) {
+			rw_lock_x_unlock_gen(&((buf_block_t*) bpage)->lock,
+					     BUF_IO_READ);
+		}
+
+		break;
+
+	case BUF_IO_WRITE:
+		/* Write means a flush operation: call the completion
+		routine in the flush system */
+
+		buf_flush_write_complete(bpage);
+
+		if (uncompressed) {
+			rw_lock_s_unlock_gen(&((buf_block_t*) bpage)->lock,
+					     BUF_IO_WRITE);
+		}
+
+		buf_pool->n_pages_written++;
+
+		break;
+
+	default:
+		ut_error;
+	}
+
+#ifdef UNIV_DEBUG
+	if (buf_debug_prints) {
+		fprintf(stderr, "Has %s page space %lu page no %lu\n",
+			io_type == BUF_IO_READ ? "read" : "written",
+			(ulong) buf_page_get_space(bpage),
+			(ulong) buf_page_get_page_no(bpage));
+	}
+#endif /* UNIV_DEBUG */
+
+	mutex_exit(buf_page_get_mutex(bpage));
+	buf_pool_mutex_exit();
+}
+
+/*************************************************************************
+Invalidates the file pages in the buffer pool when an archive recovery is
+completed. All the file pages buffered must be in a replaceable state when
+this function is called: not latched and not modified. */
+UNIV_INTERN
+void
+buf_pool_invalidate(void)
+/*=====================*/
+{
+	ibool	freed;
+
+	ut_ad(buf_all_freed());
+
+	freed = TRUE;
+
+	while (freed) {
+		freed = buf_LRU_search_and_free_block(100);
+	}
+
+	buf_pool_mutex_enter();
+
+	ut_ad(UT_LIST_GET_LEN(buf_pool->LRU) == 0);
+	ut_ad(UT_LIST_GET_LEN(buf_pool->unzip_LRU) == 0);
+
+	buf_pool_mutex_exit();
+}
+
+#if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
+/*************************************************************************
+Validates the buffer buf_pool data structure. */
+UNIV_INTERN
+ibool
+buf_validate(void)
+/*==============*/
+{
+	buf_page_t*	b;
+	buf_chunk_t*	chunk;
+	ulint		i;
+	ulint		n_single_flush	= 0;
+	ulint		n_lru_flush	= 0;
+	ulint		n_list_flush	= 0;
+	ulint		n_lru		= 0;
+	ulint		n_flush		= 0;
+	ulint		n_free		= 0;
+	ulint		n_zip		= 0;
+
+	ut_ad(buf_pool);
+
+	buf_pool_mutex_enter();
+
+	chunk = buf_pool->chunks;
+
+	/* Check the uncompressed blocks. */
+
+	for (i = buf_pool->n_chunks; i--; chunk++) {
+
+		ulint		j;
+		buf_block_t*	block = chunk->blocks;
+
+		for (j = chunk->size; j--; block++) {
+
+			mutex_enter(&block->mutex);
+
+			switch (buf_block_get_state(block)) {
+			case BUF_BLOCK_ZIP_FREE:
+			case BUF_BLOCK_ZIP_PAGE:
+			case BUF_BLOCK_ZIP_DIRTY:
+				/* These should only occur on
+				zip_clean, zip_free[], or flush_list. */
+				ut_error;
+				break;
+
+			case BUF_BLOCK_FILE_PAGE:
+				ut_a(buf_page_hash_get(buf_block_get_space(
+							       block),
+						       buf_block_get_page_no(
+							       block))
+				     == &block->page);
+
+#ifdef UNIV_IBUF_COUNT_DEBUG
+				ut_a(buf_page_get_io_fix(&block->page)
+				     == BUF_IO_READ
+				     || !ibuf_count_get(buf_block_get_space(
+								block),
+							buf_block_get_page_no(
+								block)));
+#endif
+				switch (buf_page_get_io_fix(&block->page)) {
+				case BUF_IO_NONE:
+					break;
+
+				case BUF_IO_WRITE:
+					switch (buf_page_get_flush_type(
+							&block->page)) {
+					case BUF_FLUSH_LRU:
+						n_lru_flush++;
+						ut_a(rw_lock_is_locked(
+							     &block->lock,
+							     RW_LOCK_SHARED));
+						break;
+					case BUF_FLUSH_LIST:
+						n_list_flush++;
+						break;
+					case BUF_FLUSH_SINGLE_PAGE:
+						n_single_flush++;
+						break;
+					default:
+						ut_error;
+					}
+
+					break;
+
+				case BUF_IO_READ:
+
+					ut_a(rw_lock_is_locked(&block->lock,
+							       RW_LOCK_EX));
+					break;
+				}
+
+				n_lru++;
+
+				if (block->page.oldest_modification > 0) {
+					n_flush++;
+				}
+
+				break;
+
+			case BUF_BLOCK_NOT_USED:
+				n_free++;
+				break;
+
+			case BUF_BLOCK_READY_FOR_USE:
+			case BUF_BLOCK_MEMORY:
+			case BUF_BLOCK_REMOVE_HASH:
+				/* do nothing */
+				break;
+			}
+
+			mutex_exit(&block->mutex);
+		}
+	}
+
+	mutex_enter(&buf_pool_zip_mutex);
+
+	/* Check clean compressed-only blocks. */
+
+	for (b = UT_LIST_GET_FIRST(buf_pool->zip_clean); b;
+	     b = UT_LIST_GET_NEXT(list, b)) {
+		ut_a(buf_page_get_state(b) == BUF_BLOCK_ZIP_PAGE);
+		switch (buf_page_get_io_fix(b)) {
+		case BUF_IO_NONE:
+			/* All clean blocks should be I/O-unfixed. */
+			break;
+		case BUF_IO_READ:
+			/* In buf_LRU_free_block(), we temporarily set
+			b->io_fix = BUF_IO_READ for a newly allocated
+			control block in order to prevent
+			buf_page_get_gen() from decompressing the block. */
+			break;
+		default:
+			ut_error;
+			break;
+		}
+		ut_a(!b->oldest_modification);
+		ut_a(buf_page_hash_get(b->space, b->offset) == b);
+
+		n_lru++;
+		n_zip++;
+	}
+
+	/* Check dirty compressed-only blocks. */
+
+	for (b = UT_LIST_GET_FIRST(buf_pool->flush_list); b;
+	     b = UT_LIST_GET_NEXT(list, b)) {
+		ut_ad(b->in_flush_list);
+
+		switch (buf_page_get_state(b)) {
+		case BUF_BLOCK_ZIP_DIRTY:
+			ut_a(b->oldest_modification);
+			n_lru++;
+			n_flush++;
+			n_zip++;
+			switch (buf_page_get_io_fix(b)) {
+			case BUF_IO_NONE:
+			case BUF_IO_READ:
+				break;
+
+			case BUF_IO_WRITE:
+				switch (buf_page_get_flush_type(b)) {
+				case BUF_FLUSH_LRU:
+					n_lru_flush++;
+					break;
+				case BUF_FLUSH_LIST:
+					n_list_flush++;
+					break;
+				case BUF_FLUSH_SINGLE_PAGE:
+					n_single_flush++;
+					break;
+				default:
+					ut_error;
+				}
+				break;
+			}
+			break;
+		case BUF_BLOCK_FILE_PAGE:
+			/* uncompressed page */
+			break;
+		case BUF_BLOCK_ZIP_FREE:
+		case BUF_BLOCK_ZIP_PAGE:
+		case BUF_BLOCK_NOT_USED:
+		case BUF_BLOCK_READY_FOR_USE:
+		case BUF_BLOCK_MEMORY:
+		case BUF_BLOCK_REMOVE_HASH:
+			ut_error;
+			break;
+		}
+		ut_a(buf_page_hash_get(b->space, b->offset) == b);
+	}
+
+	mutex_exit(&buf_pool_zip_mutex);
+
+	if (n_lru + n_free > buf_pool->curr_size + n_zip) {
+		fprintf(stderr, "n LRU %lu, n free %lu, pool %lu zip %lu\n",
+			(ulong) n_lru, (ulong) n_free,
+			(ulong) buf_pool->curr_size, (ulong) n_zip);
+		ut_error;
+	}
+
+	ut_a(UT_LIST_GET_LEN(buf_pool->LRU) == n_lru);
+	if (UT_LIST_GET_LEN(buf_pool->free) != n_free) {
+		fprintf(stderr, "Free list len %lu, free blocks %lu\n",
+			(ulong) UT_LIST_GET_LEN(buf_pool->free),
+			(ulong) n_free);
+		ut_error;
+	}
+	ut_a(UT_LIST_GET_LEN(buf_pool->flush_list) == n_flush);
+
+	ut_a(buf_pool->n_flush[BUF_FLUSH_SINGLE_PAGE] == n_single_flush);
+	ut_a(buf_pool->n_flush[BUF_FLUSH_LIST] == n_list_flush);
+	ut_a(buf_pool->n_flush[BUF_FLUSH_LRU] == n_lru_flush);
+
+	buf_pool_mutex_exit();
+
+	ut_a(buf_LRU_validate());
+	ut_a(buf_flush_validate());
+
+	return(TRUE);
+}
+#endif /* UNIV_DEBUG || UNIV_BUF_DEBUG */
+
+#if defined UNIV_DEBUG_PRINT || defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
+/*************************************************************************
+Prints info of the buffer buf_pool data structure. */
+UNIV_INTERN
+void
+buf_print(void)
+/*===========*/
+{
+	dulint*		index_ids;
+	ulint*		counts;
+	ulint		size;
+	ulint		i;
+	ulint		j;
+	dulint		id;
+	ulint		n_found;
+	buf_chunk_t*	chunk;
+	dict_index_t*	index;
+
+	ut_ad(buf_pool);
+
+	size = buf_pool->curr_size;
+
+	index_ids = mem_alloc(sizeof(dulint) * size);
+	counts = mem_alloc(sizeof(ulint) * size);
+
+	buf_pool_mutex_enter();
+
+	fprintf(stderr,
+		"buf_pool size %lu\n"
+		"database pages %lu\n"
+		"free pages %lu\n"
+		"modified database pages %lu\n"
+		"n pending decompressions %lu\n"
+		"n pending reads %lu\n"
+		"n pending flush LRU %lu list %lu single page %lu\n"
+		"pages read %lu, created %lu, written %lu\n",
+		(ulong) size,
+		(ulong) UT_LIST_GET_LEN(buf_pool->LRU),
+		(ulong) UT_LIST_GET_LEN(buf_pool->free),
+		(ulong) UT_LIST_GET_LEN(buf_pool->flush_list),
+		(ulong) buf_pool->n_pend_unzip,
+		(ulong) buf_pool->n_pend_reads,
+		(ulong) buf_pool->n_flush[BUF_FLUSH_LRU],
+		(ulong) buf_pool->n_flush[BUF_FLUSH_LIST],
+		(ulong) buf_pool->n_flush[BUF_FLUSH_SINGLE_PAGE],
+		(ulong) buf_pool->n_pages_read, buf_pool->n_pages_created,
+		(ulong) buf_pool->n_pages_written);
+
+	/* Count the number of blocks belonging to each index in the buffer */
+
+	n_found = 0;
+
+	chunk = buf_pool->chunks;
+
+	for (i = buf_pool->n_chunks; i--; chunk++) {
+		buf_block_t*	block		= chunk->blocks;
+		ulint		n_blocks	= chunk->size;
+
+		for (; n_blocks--; block++) {
+			const buf_frame_t* frame = block->frame;
+
+			if (fil_page_get_type(frame) == FIL_PAGE_INDEX) {
+
+				id = btr_page_get_index_id(frame);
+
+				/* Look for the id in the index_ids array */
+				j = 0;
+
+				while (j < n_found) {
+
+					if (ut_dulint_cmp(index_ids[j],
+							  id) == 0) {
+						counts[j]++;
+
+						break;
+					}
+					j++;
+				}
+
+				if (j == n_found) {
+					n_found++;
+					index_ids[j] = id;
+					counts[j] = 1;
+				}
+			}
+		}
+	}
+
+	buf_pool_mutex_exit();
+
+	for (i = 0; i < n_found; i++) {
+		index = dict_index_get_if_in_cache(index_ids[i]);
+
+		fprintf(stderr,
+			"Block count for index %lu in buffer is about %lu",
+			(ulong) ut_dulint_get_low(index_ids[i]),
+			(ulong) counts[i]);
+
+		if (index) {
+			putc(' ', stderr);
+			dict_index_name_print(stderr, NULL, index);
+		}
+
+		putc('\n', stderr);
+	}
+
+	mem_free(index_ids);
+	mem_free(counts);
+
+	ut_a(buf_validate());
+}
+#endif /* UNIV_DEBUG_PRINT || UNIV_DEBUG || UNIV_BUF_DEBUG */
+
+#ifdef UNIV_DEBUG
+/*************************************************************************
+Returns the number of latched pages in the buffer pool. */
+UNIV_INTERN
+ulint
+buf_get_latched_pages_number(void)
+/*==============================*/
+{
+	buf_chunk_t*	chunk;
+	buf_page_t*	b;
+	ulint		i;
+	ulint		fixed_pages_number = 0;
+
+	buf_pool_mutex_enter();
+
+	chunk = buf_pool->chunks;
+
+	for (i = buf_pool->n_chunks; i--; chunk++) {
+		buf_block_t*	block;
+		ulint		j;
+
+		block = chunk->blocks;
+
+		for (j = chunk->size; j--; block++) {
+			if (buf_block_get_state(block)
+			    != BUF_BLOCK_FILE_PAGE) {
+
+				continue;
+			}
+
+			mutex_enter(&block->mutex);
+
+			if (block->page.buf_fix_count != 0
+			    || buf_page_get_io_fix(&block->page)
+			    != BUF_IO_NONE) {
+				fixed_pages_number++;
+			}
+
+			mutex_exit(&block->mutex);
+		}
+	}
+
+	mutex_enter(&buf_pool_zip_mutex);
+
+	/* Traverse the lists of clean and dirty compressed-only blocks. */
+
+	for (b = UT_LIST_GET_FIRST(buf_pool->zip_clean); b;
+	     b = UT_LIST_GET_NEXT(list, b)) {
+		ut_a(buf_page_get_state(b) == BUF_BLOCK_ZIP_PAGE);
+		ut_a(buf_page_get_io_fix(b) != BUF_IO_WRITE);
+
+		if (b->buf_fix_count != 0
+		    || buf_page_get_io_fix(b) != BUF_IO_NONE) {
+			fixed_pages_number++;
+		}
+	}
+
+	for (b = UT_LIST_GET_FIRST(buf_pool->flush_list); b;
+	     b = UT_LIST_GET_NEXT(list, b)) {
+		ut_ad(b->in_flush_list);
+
+		switch (buf_page_get_state(b)) {
+		case BUF_BLOCK_ZIP_DIRTY:
+			if (b->buf_fix_count != 0
+			    || buf_page_get_io_fix(b) != BUF_IO_NONE) {
+				fixed_pages_number++;
+			}
+			break;
+		case BUF_BLOCK_FILE_PAGE:
+			/* uncompressed page */
+			break;
+		case BUF_BLOCK_ZIP_FREE:
+		case BUF_BLOCK_ZIP_PAGE:
+		case BUF_BLOCK_NOT_USED:
+		case BUF_BLOCK_READY_FOR_USE:
+		case BUF_BLOCK_MEMORY:
+		case BUF_BLOCK_REMOVE_HASH:
+			ut_error;
+			break;
+		}
+	}
+
+	mutex_exit(&buf_pool_zip_mutex);
+	buf_pool_mutex_exit();
+
+	return(fixed_pages_number);
+}
+#endif /* UNIV_DEBUG */
+
+/*************************************************************************
+Returns the number of pending buf pool ios. */
+UNIV_INTERN
+ulint
+buf_get_n_pending_ios(void)
+/*=======================*/
+{
+	return(buf_pool->n_pend_reads
+	       + buf_pool->n_flush[BUF_FLUSH_LRU]
+	       + buf_pool->n_flush[BUF_FLUSH_LIST]
+	       + buf_pool->n_flush[BUF_FLUSH_SINGLE_PAGE]);
+}
+
+/*************************************************************************
+Returns the ratio in percents of modified pages in the buffer pool /
+database pages in the buffer pool. */
+UNIV_INTERN
+ulint
+buf_get_modified_ratio_pct(void)
+/*============================*/
+{
+	ulint	ratio;
+
+	//buf_pool_mutex_enter(); /* optimistic */
+
+	ratio = (100 * UT_LIST_GET_LEN(buf_pool->flush_list))
+		/ (1 + UT_LIST_GET_LEN(buf_pool->LRU)
+		   + UT_LIST_GET_LEN(buf_pool->free));
+
+	/* 1 + is there to avoid division by zero */
+
+	//buf_pool_mutex_exit(); /* optimistic */
+
+	return(ratio);
+}
+
+/*************************************************************************
+Prints info of the buffer i/o. */
+UNIV_INTERN
+void
+buf_print_io(
+/*=========*/
+	FILE*	file)	/* in/out: buffer where to print */
+{
+	time_t	current_time;
+	double	time_elapsed;
+	ulint	size;
+
+	ut_ad(buf_pool);
+	size = buf_pool->curr_size;
+
+	buf_pool_mutex_enter();
+
+	fprintf(file,
+		"Buffer pool size        %lu\n"
+		"Buffer pool size, bytes %lu\n"
+		"Free buffers            %lu\n"
+		"Database pages          %lu\n"
+		"Modified db pages       %lu\n"
+		"Pending reads %lu\n"
+		"Pending writes: LRU %lu, flush list %lu, single page %lu\n",
+		(ulong) size,
+		(ulong) size * UNIV_PAGE_SIZE,
+		(ulong) UT_LIST_GET_LEN(buf_pool->free),
+		(ulong) UT_LIST_GET_LEN(buf_pool->LRU),
+		(ulong) UT_LIST_GET_LEN(buf_pool->flush_list),
+		(ulong) buf_pool->n_pend_reads,
+		(ulong) buf_pool->n_flush[BUF_FLUSH_LRU]
+		+ buf_pool->init_flush[BUF_FLUSH_LRU],
+		(ulong) buf_pool->n_flush[BUF_FLUSH_LIST]
+		+ buf_pool->init_flush[BUF_FLUSH_LIST],
+		(ulong) buf_pool->n_flush[BUF_FLUSH_SINGLE_PAGE]);
+
+	current_time = time(NULL);
+	time_elapsed = 0.001 + difftime(current_time,
+					buf_pool->last_printout_time);
+	buf_pool->last_printout_time = current_time;
+
+	fprintf(file,
+		"Pages read %lu, created %lu, written %lu\n"
+		"%.2f reads/s, %.2f creates/s, %.2f writes/s\n",
+		(ulong) buf_pool->n_pages_read,
+		(ulong) buf_pool->n_pages_created,
+		(ulong) buf_pool->n_pages_written,
+		(buf_pool->n_pages_read - buf_pool->n_pages_read_old)
+		/ time_elapsed,
+		(buf_pool->n_pages_created - buf_pool->n_pages_created_old)
+		/ time_elapsed,
+		(buf_pool->n_pages_written - buf_pool->n_pages_written_old)
+		/ time_elapsed);
+
+	if (buf_pool->n_page_gets > buf_pool->n_page_gets_old) {
+		fprintf(file, "Buffer pool hit rate %lu / 1000\n",
+			(ulong)
+			(1000 - ((1000 * (buf_pool->n_pages_read
+					  - buf_pool->n_pages_read_old))
+				 / (buf_pool->n_page_gets
+				    - buf_pool->n_page_gets_old))));
+	} else {
+		fputs("No buffer pool page gets since the last printout\n",
+		      file);
+	}
+
+	buf_pool->n_page_gets_old = buf_pool->n_page_gets;
+	buf_pool->n_pages_read_old = buf_pool->n_pages_read;
+	buf_pool->n_pages_created_old = buf_pool->n_pages_created;
+	buf_pool->n_pages_written_old = buf_pool->n_pages_written;
+
+	/* Print some values to help us with visualizing what is
+	happening with LRU eviction. */
+	fprintf(file,
+		"LRU len: %lu, unzip_LRU len: %lu\n"
+		"I/O sum[%lu]:cur[%lu], unzip sum[%lu]:cur[%lu]\n",
+		UT_LIST_GET_LEN(buf_pool->LRU),
+		UT_LIST_GET_LEN(buf_pool->unzip_LRU),
+		buf_LRU_stat_sum.io, buf_LRU_stat_cur.io,
+		buf_LRU_stat_sum.unzip, buf_LRU_stat_cur.unzip);
+
+	buf_pool_mutex_exit();
+}
+
+/**************************************************************************
+Refreshes the statistics used to print per-second averages. */
+UNIV_INTERN
+void
+buf_refresh_io_stats(void)
+/*======================*/
+{
+	buf_pool->last_printout_time = time(NULL);
+	buf_pool->n_page_gets_old = buf_pool->n_page_gets;
+	buf_pool->n_pages_read_old = buf_pool->n_pages_read;
+	buf_pool->n_pages_created_old = buf_pool->n_pages_created;
+	buf_pool->n_pages_written_old = buf_pool->n_pages_written;
+}
+
+/*************************************************************************
+Checks that all file pages in the buffer are in a replaceable state. */
+UNIV_INTERN
+ibool
+buf_all_freed(void)
+/*===============*/
+{
+	buf_chunk_t*	chunk;
+	ulint		i;
+
+	ut_ad(buf_pool);
+
+	buf_pool_mutex_enter();
+
+	chunk = buf_pool->chunks;
+
+	for (i = buf_pool->n_chunks; i--; chunk++) {
+
+		const buf_block_t* block = buf_chunk_not_freed(chunk);
+
+		if (UNIV_LIKELY_NULL(block)) {
+			fprintf(stderr,
+				"Page %lu %lu still fixed or dirty\n",
+				(ulong) block->page.space,
+				(ulong) block->page.offset);
+			ut_error;
+		}
+	}
+
+	buf_pool_mutex_exit();
+
+	return(TRUE);
+}
+
+/*************************************************************************
+Checks that there currently are no pending i/o-operations for the buffer
+pool. */
+UNIV_INTERN
+ibool
+buf_pool_check_no_pending_io(void)
+/*==============================*/
+				/* out: TRUE if there is no pending i/o */
+{
+	ibool	ret;
+
+	buf_pool_mutex_enter();
+
+	if (buf_pool->n_pend_reads + buf_pool->n_flush[BUF_FLUSH_LRU]
+	    + buf_pool->n_flush[BUF_FLUSH_LIST]
+	    + buf_pool->n_flush[BUF_FLUSH_SINGLE_PAGE]) {
+		ret = FALSE;
+	} else {
+		ret = TRUE;
+	}
+
+	buf_pool_mutex_exit();
+
+	return(ret);
+}
+
+/*************************************************************************
+Gets the current length of the free list of buffer blocks. */
+UNIV_INTERN
+ulint
+buf_get_free_list_len(void)
+/*=======================*/
+{
+	ulint	len;
+
+	buf_pool_mutex_enter();
+
+	len = UT_LIST_GET_LEN(buf_pool->free);
+
+	buf_pool_mutex_exit();
+
+	return(len);
+}
diff --git a/storage/xtradb/buf/buf0flu.c b/storage/xtradb/buf/buf0flu.c
new file mode 100644
index 00000000000..7cb5bb91994
--- /dev/null
+++ b/storage/xtradb/buf/buf0flu.c
@@ -0,0 +1,1277 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The database buffer buf_pool flush algorithm
+
+Created 11/11/1995 Heikki Tuuri
+*******************************************************/
+
+#include "buf0flu.h"
+
+#ifdef UNIV_NONINL
+#include "buf0flu.ic"
+#include "trx0sys.h"
+#endif
+
+#include "ut0byte.h"
+#include "ut0lst.h"
+#include "page0page.h"
+#include "page0zip.h"
+#include "fil0fil.h"
+#include "buf0buf.h"
+#include "buf0lru.h"
+#include "buf0rea.h"
+#include "ibuf0ibuf.h"
+#include "log0log.h"
+#include "os0file.h"
+#include "trx0sys.h"
+#include "srv0srv.h"
+
+#if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
+/**********************************************************************
+Validates the flush list. */
+static
+ibool
+buf_flush_validate_low(void);
+/*========================*/
+		/* out: TRUE if ok */
+#endif /* UNIV_DEBUG || UNIV_BUF_DEBUG */
+
+/************************************************************************
+Inserts a modified block into the flush list. */
+UNIV_INTERN
+void
+buf_flush_insert_into_flush_list(
+/*=============================*/
+	buf_block_t*	block)	/* in/out: block which is modified */
+{
+	ut_ad(buf_pool_mutex_own());
+	ut_ad((UT_LIST_GET_FIRST(buf_pool->flush_list) == NULL)
+	      || (UT_LIST_GET_FIRST(buf_pool->flush_list)->oldest_modification
+		  <= block->page.oldest_modification));
+
+	ut_ad(buf_block_get_state(block) == BUF_BLOCK_FILE_PAGE);
+	ut_ad(block->page.in_LRU_list);
+	ut_ad(block->page.in_page_hash);
+	ut_ad(!block->page.in_zip_hash);
+	ut_ad(!block->page.in_flush_list);
+	ut_d(block->page.in_flush_list = TRUE);
+	UT_LIST_ADD_FIRST(list, buf_pool->flush_list, &block->page);
+
+#if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
+	ut_a(buf_flush_validate_low());
+#endif /* UNIV_DEBUG || UNIV_BUF_DEBUG */
+}
+
+/************************************************************************
+Inserts a modified block into the flush list in the right sorted position.
+This function is used by recovery, because there the modifications do not
+necessarily come in the order of lsn's. */
+UNIV_INTERN
+void
+buf_flush_insert_sorted_into_flush_list(
+/*====================================*/
+	buf_block_t*	block)	/* in/out: block which is modified */
+{
+	buf_page_t*	prev_b;
+	buf_page_t*	b;
+
+	ut_ad(buf_pool_mutex_own());
+	ut_ad(buf_block_get_state(block) == BUF_BLOCK_FILE_PAGE);
+
+	ut_ad(block->page.in_LRU_list);
+	ut_ad(block->page.in_page_hash);
+	ut_ad(!block->page.in_zip_hash);
+	ut_ad(!block->page.in_flush_list);
+	ut_d(block->page.in_flush_list = TRUE);
+
+	prev_b = NULL;
+	b = UT_LIST_GET_FIRST(buf_pool->flush_list);
+
+	while (b && b->oldest_modification > block->page.oldest_modification) {
+		ut_ad(b->in_flush_list);
+		prev_b = b;
+		b = UT_LIST_GET_NEXT(list, b);
+	}
+
+	if (prev_b == NULL) {
+		UT_LIST_ADD_FIRST(list, buf_pool->flush_list, &block->page);
+	} else {
+		UT_LIST_INSERT_AFTER(list, buf_pool->flush_list,
+				     prev_b, &block->page);
+	}
+
+#if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
+	ut_a(buf_flush_validate_low());
+#endif /* UNIV_DEBUG || UNIV_BUF_DEBUG */
+}
+
+/************************************************************************
+Returns TRUE if the file page block is immediately suitable for replacement,
+i.e., the transition FILE_PAGE => NOT_USED allowed. */
+UNIV_INTERN
+ibool
+buf_flush_ready_for_replace(
+/*========================*/
+				/* out: TRUE if can replace immediately */
+	buf_page_t*	bpage)	/* in: buffer control block, must be
+				buf_page_in_file(bpage) and in the LRU list */
+{
+	//ut_ad(buf_pool_mutex_own());
+	//ut_ad(mutex_own(buf_page_get_mutex(bpage)));
+	//ut_ad(bpage->in_LRU_list); /* optimistic use */
+
+	if (UNIV_LIKELY(bpage->in_LRU_list && buf_page_in_file(bpage))) {
+
+		return(bpage->oldest_modification == 0
+		       && buf_page_get_io_fix(bpage) == BUF_IO_NONE
+		       && bpage->buf_fix_count == 0);
+	}
+
+	/* permited not to own LRU_mutex..  */
+/*
+	ut_print_timestamp(stderr);
+	fprintf(stderr,
+		"  InnoDB: Error: buffer block state %lu"
+		" in the LRU list!\n",
+		(ulong) buf_page_get_state(bpage));
+	ut_print_buf(stderr, bpage, sizeof(buf_page_t));
+	putc('\n', stderr);
+*/
+
+	return(FALSE);
+}
+
+/************************************************************************
+Returns TRUE if the block is modified and ready for flushing. */
+UNIV_INLINE
+ibool
+buf_flush_ready_for_flush(
+/*======================*/
+				/* out: TRUE if can flush immediately */
+	buf_page_t*	bpage,	/* in: buffer control block, must be
+				buf_page_in_file(bpage) */
+	enum buf_flush	flush_type)/* in: BUF_FLUSH_LRU or BUF_FLUSH_LIST */
+{
+	ut_a(buf_page_in_file(bpage));
+	ut_ad(buf_pool_mutex_own());
+	ut_ad(mutex_own(buf_page_get_mutex(bpage)));
+	ut_ad(flush_type == BUF_FLUSH_LRU || BUF_FLUSH_LIST);
+
+	if (bpage->oldest_modification != 0
+	    && buf_page_get_io_fix(bpage) == BUF_IO_NONE) {
+		ut_ad(bpage->in_flush_list);
+
+		if (flush_type != BUF_FLUSH_LRU) {
+
+			return(TRUE);
+
+		} else if (bpage->buf_fix_count == 0) {
+
+			/* If we are flushing the LRU list, to avoid deadlocks
+			we require the block not to be bufferfixed, and hence
+			not latched. */
+
+			return(TRUE);
+		}
+	}
+
+	return(FALSE);
+}
+
+/************************************************************************
+Remove a block from the flush list of modified blocks. */
+UNIV_INTERN
+void
+buf_flush_remove(
+/*=============*/
+	buf_page_t*	bpage)	/* in: pointer to the block in question */
+{
+	ut_ad(buf_pool_mutex_own());
+	ut_ad(mutex_own(buf_page_get_mutex(bpage)));
+	ut_ad(bpage->in_flush_list);
+	ut_d(bpage->in_flush_list = FALSE);
+
+	switch (buf_page_get_state(bpage)) {
+	case BUF_BLOCK_ZIP_PAGE:
+		/* clean compressed pages should not be on the flush list */
+	case BUF_BLOCK_ZIP_FREE:
+	case BUF_BLOCK_NOT_USED:
+	case BUF_BLOCK_READY_FOR_USE:
+	case BUF_BLOCK_MEMORY:
+	case BUF_BLOCK_REMOVE_HASH:
+		ut_error;
+		return;
+	case BUF_BLOCK_ZIP_DIRTY:
+		buf_page_set_state(bpage, BUF_BLOCK_ZIP_PAGE);
+		UT_LIST_REMOVE(list, buf_pool->flush_list, bpage);
+		buf_LRU_insert_zip_clean(bpage);
+		break;
+	case BUF_BLOCK_FILE_PAGE:
+		UT_LIST_REMOVE(list, buf_pool->flush_list, bpage);
+		break;
+	}
+
+	bpage->oldest_modification = 0;
+
+	ut_d(UT_LIST_VALIDATE(list, buf_page_t, buf_pool->flush_list));
+}
+
+/************************************************************************
+Updates the flush system data structures when a write is completed. */
+UNIV_INTERN
+void
+buf_flush_write_complete(
+/*=====================*/
+	buf_page_t*	bpage)	/* in: pointer to the block in question */
+{
+	enum buf_flush	flush_type;
+
+	ut_ad(bpage);
+
+	buf_flush_remove(bpage);
+
+	flush_type = buf_page_get_flush_type(bpage);
+	buf_pool->n_flush[flush_type]--;
+
+	if (flush_type == BUF_FLUSH_LRU) {
+		/* Put the block to the end of the LRU list to wait to be
+		moved to the free list */
+
+		buf_LRU_make_block_old(bpage);
+
+		buf_pool->LRU_flush_ended++;
+	}
+
+	/* fprintf(stderr, "n pending flush %lu\n",
+	buf_pool->n_flush[flush_type]); */
+
+	if ((buf_pool->n_flush[flush_type] == 0)
+	    && (buf_pool->init_flush[flush_type] == FALSE)) {
+
+		/* The running flush batch has ended */
+
+		os_event_set(buf_pool->no_flush[flush_type]);
+	}
+}
+
+/************************************************************************
+Flushes possible buffered writes from the doublewrite memory buffer to disk,
+and also wakes up the aio thread if simulated aio is used. It is very
+important to call this function after a batch of writes has been posted,
+and also when we may have to wait for a page latch! Otherwise a deadlock
+of threads can occur. */
+static
+void
+buf_flush_buffered_writes(void)
+/*===========================*/
+{
+	byte*		write_buf;
+	ulint		len;
+	ulint		len2;
+	ulint		i;
+
+	if (!srv_use_doublewrite_buf || trx_doublewrite == NULL) {
+		os_aio_simulated_wake_handler_threads();
+
+		return;
+	}
+
+	mutex_enter(&(trx_doublewrite->mutex));
+
+	/* Write first to doublewrite buffer blocks. We use synchronous
+	aio and thus know that file write has been completed when the
+	control returns. */
+
+	if (trx_doublewrite->first_free == 0) {
+
+		mutex_exit(&(trx_doublewrite->mutex));
+
+		return;
+	}
+
+	for (i = 0; i < trx_doublewrite->first_free; i++) {
+
+		const buf_block_t*	block;
+
+		block = (buf_block_t*) trx_doublewrite->buf_block_arr[i];
+
+		if (buf_block_get_state(block) != BUF_BLOCK_FILE_PAGE
+		    || block->page.zip.data) {
+			/* No simple validate for compressed pages exists. */
+			continue;
+		}
+
+		if (UNIV_UNLIKELY
+		    (memcmp(block->frame + (FIL_PAGE_LSN + 4),
+			    block->frame + (UNIV_PAGE_SIZE
+					    - FIL_PAGE_END_LSN_OLD_CHKSUM + 4),
+			    4))) {
+			ut_print_timestamp(stderr);
+			fprintf(stderr,
+				"  InnoDB: ERROR: The page to be written"
+				" seems corrupt!\n"
+				"InnoDB: The lsn fields do not match!"
+				" Noticed in the buffer pool\n"
+				"InnoDB: before posting to the"
+				" doublewrite buffer.\n");
+		}
+
+		if (!block->check_index_page_at_flush) {
+		} else if (page_is_comp(block->frame)) {
+			if (UNIV_UNLIKELY
+			    (!page_simple_validate_new(block->frame))) {
+corrupted_page:
+				buf_page_print(block->frame, 0);
+
+				ut_print_timestamp(stderr);
+				fprintf(stderr,
+					"  InnoDB: Apparent corruption of an"
+					" index page n:o %lu in space %lu\n"
+					"InnoDB: to be written to data file."
+					" We intentionally crash server\n"
+					"InnoDB: to prevent corrupt data"
+					" from ending up in data\n"
+					"InnoDB: files.\n",
+					(ulong) buf_block_get_page_no(block),
+					(ulong) buf_block_get_space(block));
+
+				ut_error;
+			}
+		} else if (UNIV_UNLIKELY
+			   (!page_simple_validate_old(block->frame))) {
+
+			goto corrupted_page;
+		}
+	}
+
+	/* increment the doublewrite flushed pages counter */
+	srv_dblwr_pages_written+= trx_doublewrite->first_free;
+	srv_dblwr_writes++;
+
+	len = ut_min(TRX_SYS_DOUBLEWRITE_BLOCK_SIZE,
+		     trx_doublewrite->first_free) * UNIV_PAGE_SIZE;
+
+	write_buf = trx_doublewrite->write_buf;
+	i = 0;
+
+	fil_io(OS_FILE_WRITE, TRUE, TRX_SYS_SPACE, 0,
+	       trx_doublewrite->block1, 0, len,
+	       (void*) write_buf, NULL);
+
+	for (len2 = 0; len2 + UNIV_PAGE_SIZE <= len;
+	     len2 += UNIV_PAGE_SIZE, i++) {
+		const buf_block_t* block = (buf_block_t*)
+			trx_doublewrite->buf_block_arr[i];
+
+		if (UNIV_LIKELY(!block->page.zip.data)
+		    && UNIV_LIKELY(buf_block_get_state(block)
+				   == BUF_BLOCK_FILE_PAGE)
+		    && UNIV_UNLIKELY
+		    (memcmp(write_buf + len2 + (FIL_PAGE_LSN + 4),
+			    write_buf + len2
+			    + (UNIV_PAGE_SIZE
+			       - FIL_PAGE_END_LSN_OLD_CHKSUM + 4), 4))) {
+			ut_print_timestamp(stderr);
+			fprintf(stderr,
+				"  InnoDB: ERROR: The page to be written"
+				" seems corrupt!\n"
+				"InnoDB: The lsn fields do not match!"
+				" Noticed in the doublewrite block1.\n");
+		}
+	}
+
+	if (trx_doublewrite->first_free <= TRX_SYS_DOUBLEWRITE_BLOCK_SIZE) {
+		goto flush;
+	}
+
+	len = (trx_doublewrite->first_free - TRX_SYS_DOUBLEWRITE_BLOCK_SIZE)
+		* UNIV_PAGE_SIZE;
+
+	write_buf = trx_doublewrite->write_buf
+		+ TRX_SYS_DOUBLEWRITE_BLOCK_SIZE * UNIV_PAGE_SIZE;
+	ut_ad(i == TRX_SYS_DOUBLEWRITE_BLOCK_SIZE);
+
+	fil_io(OS_FILE_WRITE, TRUE, TRX_SYS_SPACE, 0,
+	       trx_doublewrite->block2, 0, len,
+	       (void*) write_buf, NULL);
+
+	for (len2 = 0; len2 + UNIV_PAGE_SIZE <= len;
+	     len2 += UNIV_PAGE_SIZE, i++) {
+		const buf_block_t* block = (buf_block_t*)
+			trx_doublewrite->buf_block_arr[i];
+
+		if (UNIV_LIKELY(!block->page.zip.data)
+		    && UNIV_LIKELY(buf_block_get_state(block)
+				   == BUF_BLOCK_FILE_PAGE)
+		    && UNIV_UNLIKELY
+		    (memcmp(write_buf + len2 + (FIL_PAGE_LSN + 4),
+			    write_buf + len2
+			    + (UNIV_PAGE_SIZE
+			       - FIL_PAGE_END_LSN_OLD_CHKSUM + 4), 4))) {
+			ut_print_timestamp(stderr);
+			fprintf(stderr,
+				"  InnoDB: ERROR: The page to be"
+				" written seems corrupt!\n"
+				"InnoDB: The lsn fields do not match!"
+				" Noticed in"
+				" the doublewrite block2.\n");
+		}
+	}
+
+flush:
+	/* Now flush the doublewrite buffer data to disk */
+
+	fil_flush(TRX_SYS_SPACE);
+
+	/* We know that the writes have been flushed to disk now
+	and in recovery we will find them in the doublewrite buffer
+	blocks. Next do the writes to the intended positions. */
+
+	for (i = 0; i < trx_doublewrite->first_free; i++) {
+		const buf_block_t* block = (buf_block_t*)
+			trx_doublewrite->buf_block_arr[i];
+
+		ut_a(buf_page_in_file(&block->page));
+		if (UNIV_LIKELY_NULL(block->page.zip.data)) {
+			fil_io(OS_FILE_WRITE | OS_AIO_SIMULATED_WAKE_LATER,
+			       FALSE, buf_page_get_space(&block->page),
+			       buf_page_get_zip_size(&block->page),
+			       buf_page_get_page_no(&block->page), 0,
+			       buf_page_get_zip_size(&block->page),
+			       (void*)block->page.zip.data,
+			       (void*)block);
+
+			/* Increment the counter of I/O operations used
+			for selecting LRU policy. */
+			buf_LRU_stat_inc_io();
+
+			continue;
+		}
+
+		ut_a(buf_block_get_state(block) == BUF_BLOCK_FILE_PAGE);
+
+		if (UNIV_UNLIKELY(memcmp(block->frame + (FIL_PAGE_LSN + 4),
+					 block->frame
+					 + (UNIV_PAGE_SIZE
+					    - FIL_PAGE_END_LSN_OLD_CHKSUM + 4),
+					 4))) {
+			ut_print_timestamp(stderr);
+			fprintf(stderr,
+				"  InnoDB: ERROR: The page to be written"
+				" seems corrupt!\n"
+				"InnoDB: The lsn fields do not match!"
+				" Noticed in the buffer pool\n"
+				"InnoDB: after posting and flushing"
+				" the doublewrite buffer.\n"
+				"InnoDB: Page buf fix count %lu,"
+				" io fix %lu, state %lu\n",
+				(ulong)block->page.buf_fix_count,
+				(ulong)buf_block_get_io_fix(block),
+				(ulong)buf_block_get_state(block));
+		}
+
+		fil_io(OS_FILE_WRITE | OS_AIO_SIMULATED_WAKE_LATER,
+		       FALSE, buf_block_get_space(block), 0,
+		       buf_block_get_page_no(block), 0, UNIV_PAGE_SIZE,
+		       (void*)block->frame, (void*)block);
+
+		/* Increment the counter of I/O operations used
+		for selecting LRU policy. */
+		buf_LRU_stat_inc_io();
+	}
+
+	/* Wake possible simulated aio thread to actually post the
+	writes to the operating system */
+
+	os_aio_simulated_wake_handler_threads();
+
+	/* Wait that all async writes to tablespaces have been posted to
+	the OS */
+
+	os_aio_wait_until_no_pending_writes();
+
+	/* Now we flush the data to disk (for example, with fsync) */
+
+	fil_flush_file_spaces(FIL_TABLESPACE);
+
+	/* We can now reuse the doublewrite memory buffer: */
+
+	trx_doublewrite->first_free = 0;
+
+	mutex_exit(&(trx_doublewrite->mutex));
+}
+
+/************************************************************************
+Posts a buffer page for writing. If the doublewrite memory buffer is
+full, calls buf_flush_buffered_writes and waits for for free space to
+appear. */
+static
+void
+buf_flush_post_to_doublewrite_buf(
+/*==============================*/
+	buf_page_t*	bpage)	/* in: buffer block to write */
+{
+	ulint	zip_size;
+try_again:
+	mutex_enter(&(trx_doublewrite->mutex));
+
+	ut_a(buf_page_in_file(bpage));
+
+	if (trx_doublewrite->first_free
+	    >= 2 * TRX_SYS_DOUBLEWRITE_BLOCK_SIZE) {
+		mutex_exit(&(trx_doublewrite->mutex));
+
+		buf_flush_buffered_writes();
+
+		goto try_again;
+	}
+
+	zip_size = buf_page_get_zip_size(bpage);
+
+	if (UNIV_UNLIKELY(zip_size)) {
+		/* Copy the compressed page and clear the rest. */
+		memcpy(trx_doublewrite->write_buf
+		       + UNIV_PAGE_SIZE * trx_doublewrite->first_free,
+		       bpage->zip.data, zip_size);
+		memset(trx_doublewrite->write_buf
+		       + UNIV_PAGE_SIZE * trx_doublewrite->first_free
+		       + zip_size, 0, UNIV_PAGE_SIZE - zip_size);
+	} else {
+		ut_a(buf_page_get_state(bpage) == BUF_BLOCK_FILE_PAGE);
+
+		memcpy(trx_doublewrite->write_buf
+		       + UNIV_PAGE_SIZE * trx_doublewrite->first_free,
+		       ((buf_block_t*) bpage)->frame, UNIV_PAGE_SIZE);
+	}
+
+	trx_doublewrite->buf_block_arr[trx_doublewrite->first_free] = bpage;
+
+	trx_doublewrite->first_free++;
+
+	if (trx_doublewrite->first_free
+	    >= 2 * TRX_SYS_DOUBLEWRITE_BLOCK_SIZE) {
+		mutex_exit(&(trx_doublewrite->mutex));
+
+		buf_flush_buffered_writes();
+
+		return;
+	}
+
+	mutex_exit(&(trx_doublewrite->mutex));
+}
+
+/************************************************************************
+Initializes a page for writing to the tablespace. */
+UNIV_INTERN
+void
+buf_flush_init_for_writing(
+/*=======================*/
+	byte*		page,		/* in/out: page */
+	void*		page_zip_,	/* in/out: compressed page, or NULL */
+	ib_uint64_t	newest_lsn)	/* in: newest modification lsn
+					to the page */
+{
+	ut_ad(page);
+
+	if (page_zip_) {
+		page_zip_des_t*	page_zip = page_zip_;
+		ulint		zip_size = page_zip_get_size(page_zip);
+		ut_ad(zip_size);
+		ut_ad(ut_is_2pow(zip_size));
+		ut_ad(zip_size <= UNIV_PAGE_SIZE);
+
+		switch (UNIV_EXPECT(fil_page_get_type(page), FIL_PAGE_INDEX)) {
+		case FIL_PAGE_TYPE_ALLOCATED:
+		case FIL_PAGE_INODE:
+		case FIL_PAGE_IBUF_BITMAP:
+		case FIL_PAGE_TYPE_FSP_HDR:
+		case FIL_PAGE_TYPE_XDES:
+			/* These are essentially uncompressed pages. */
+			memcpy(page_zip->data, page, zip_size);
+			/* fall through */
+		case FIL_PAGE_TYPE_ZBLOB:
+		case FIL_PAGE_TYPE_ZBLOB2:
+		case FIL_PAGE_INDEX:
+			mach_write_ull(page_zip->data
+				       + FIL_PAGE_LSN, newest_lsn);
+			memset(page_zip->data + FIL_PAGE_FILE_FLUSH_LSN, 0, 8);
+			mach_write_to_4(page_zip->data
+					+ FIL_PAGE_SPACE_OR_CHKSUM,
+					srv_use_checksums
+					? page_zip_calc_checksum(
+						page_zip->data, zip_size)
+					: BUF_NO_CHECKSUM_MAGIC);
+			return;
+		}
+
+		ut_print_timestamp(stderr);
+		fputs("  InnoDB: ERROR: The compressed page to be written"
+		      " seems corrupt:", stderr);
+		ut_print_buf(stderr, page, zip_size);
+		fputs("\nInnoDB: Possibly older version of the page:", stderr);
+		ut_print_buf(stderr, page_zip->data, zip_size);
+		putc('\n', stderr);
+		ut_error;
+	}
+
+	/* Write the newest modification lsn to the page header and trailer */
+	mach_write_ull(page + FIL_PAGE_LSN, newest_lsn);
+
+	mach_write_ull(page + UNIV_PAGE_SIZE - FIL_PAGE_END_LSN_OLD_CHKSUM,
+		       newest_lsn);
+
+	/* Store the new formula checksum */
+
+	mach_write_to_4(page + FIL_PAGE_SPACE_OR_CHKSUM,
+			srv_use_checksums
+			? buf_calc_page_new_checksum(page)
+			: BUF_NO_CHECKSUM_MAGIC);
+
+	/* We overwrite the first 4 bytes of the end lsn field to store
+	the old formula checksum. Since it depends also on the field
+	FIL_PAGE_SPACE_OR_CHKSUM, it has to be calculated after storing the
+	new formula checksum. */
+
+	mach_write_to_4(page + UNIV_PAGE_SIZE - FIL_PAGE_END_LSN_OLD_CHKSUM,
+			srv_use_checksums
+			? buf_calc_page_old_checksum(page)
+			: BUF_NO_CHECKSUM_MAGIC);
+}
+
+/************************************************************************
+Does an asynchronous write of a buffer page. NOTE: in simulated aio and
+also when the doublewrite buffer is used, we must call
+buf_flush_buffered_writes after we have posted a batch of writes! */
+static
+void
+buf_flush_write_block_low(
+/*======================*/
+	buf_page_t*	bpage)	/* in: buffer block to write */
+{
+	ulint	zip_size	= buf_page_get_zip_size(bpage);
+	page_t*	frame		= NULL;
+#ifdef UNIV_LOG_DEBUG
+	static ibool univ_log_debug_warned;
+#endif /* UNIV_LOG_DEBUG */
+
+	ut_ad(buf_page_in_file(bpage));
+
+	/* We are not holding buf_pool_mutex or block_mutex here.
+	Nevertheless, it is safe to access bpage, because it is
+	io_fixed and oldest_modification != 0.  Thus, it cannot be
+	relocated in the buffer pool or removed from flush_list or
+	LRU_list. */
+	ut_ad(!buf_pool_mutex_own());
+	ut_ad(!mutex_own(buf_page_get_mutex(bpage)));
+	ut_ad(buf_page_get_io_fix(bpage) == BUF_IO_WRITE);
+	ut_ad(bpage->oldest_modification != 0);
+
+#ifdef UNIV_IBUF_COUNT_DEBUG
+	ut_a(ibuf_count_get(bpage->space, bpage->offset) == 0);
+#endif
+	ut_ad(bpage->newest_modification != 0);
+
+#ifdef UNIV_LOG_DEBUG
+	if (!univ_log_debug_warned) {
+		univ_log_debug_warned = TRUE;
+		fputs("Warning: cannot force log to disk if"
+		      " UNIV_LOG_DEBUG is defined!\n"
+		      "Crash recovery will not work!\n",
+		      stderr);
+	}
+#else
+	/* Force the log to the disk before writing the modified block */
+	log_write_up_to(bpage->newest_modification, LOG_WAIT_ALL_GROUPS, TRUE);
+#endif
+	switch (buf_page_get_state(bpage)) {
+	case BUF_BLOCK_ZIP_FREE:
+	case BUF_BLOCK_ZIP_PAGE: /* The page should be dirty. */
+	case BUF_BLOCK_NOT_USED:
+	case BUF_BLOCK_READY_FOR_USE:
+	case BUF_BLOCK_MEMORY:
+	case BUF_BLOCK_REMOVE_HASH:
+		ut_error;
+		break;
+	case BUF_BLOCK_ZIP_DIRTY:
+		frame = bpage->zip.data;
+		if (UNIV_LIKELY(srv_use_checksums)) {
+			ut_a(mach_read_from_4(frame + FIL_PAGE_SPACE_OR_CHKSUM)
+			     == page_zip_calc_checksum(frame, zip_size));
+		}
+		mach_write_ull(frame + FIL_PAGE_LSN,
+			       bpage->newest_modification);
+		memset(frame + FIL_PAGE_FILE_FLUSH_LSN, 0, 8);
+		break;
+	case BUF_BLOCK_FILE_PAGE:
+		frame = bpage->zip.data;
+		if (!frame) {
+			frame = ((buf_block_t*) bpage)->frame;
+		}
+
+		buf_flush_init_for_writing(((buf_block_t*) bpage)->frame,
+					   bpage->zip.data
+					   ? &bpage->zip : NULL,
+					   bpage->newest_modification);
+		break;
+	}
+
+	if (!srv_use_doublewrite_buf || !trx_doublewrite) {
+		fil_io(OS_FILE_WRITE | OS_AIO_SIMULATED_WAKE_LATER,
+		       FALSE, buf_page_get_space(bpage), zip_size,
+		       buf_page_get_page_no(bpage), 0,
+		       zip_size ? zip_size : UNIV_PAGE_SIZE,
+		       frame, bpage);
+	} else {
+		buf_flush_post_to_doublewrite_buf(bpage);
+	}
+}
+
+/************************************************************************
+Writes a flushable page asynchronously from the buffer pool to a file.
+NOTE: in simulated aio we must call
+os_aio_simulated_wake_handler_threads after we have posted a batch of
+writes! NOTE: buf_pool_mutex and buf_page_get_mutex(bpage) must be
+held upon entering this function, and they will be released by this
+function. */
+static
+void
+buf_flush_page(
+/*===========*/
+	buf_page_t*	bpage,		/* in: buffer control block */
+	enum buf_flush	flush_type)	/* in: BUF_FLUSH_LRU
+					or BUF_FLUSH_LIST */
+{
+	mutex_t*	block_mutex;
+	ibool		is_uncompressed;
+
+	ut_ad(flush_type == BUF_FLUSH_LRU || flush_type == BUF_FLUSH_LIST);
+	ut_ad(buf_pool_mutex_own());
+	ut_ad(buf_page_in_file(bpage));
+
+	block_mutex = buf_page_get_mutex(bpage);
+	ut_ad(mutex_own(block_mutex));
+
+	ut_ad(buf_flush_ready_for_flush(bpage, flush_type));
+
+	buf_page_set_io_fix(bpage, BUF_IO_WRITE);
+
+	buf_page_set_flush_type(bpage, flush_type);
+
+	if (buf_pool->n_flush[flush_type] == 0) {
+
+		os_event_reset(buf_pool->no_flush[flush_type]);
+	}
+
+	buf_pool->n_flush[flush_type]++;
+
+	is_uncompressed = (buf_page_get_state(bpage) == BUF_BLOCK_FILE_PAGE);
+	ut_ad(is_uncompressed == (block_mutex != &buf_pool_zip_mutex));
+
+	switch (flush_type) {
+		ibool	is_s_latched;
+	case BUF_FLUSH_LIST:
+		/* If the simulated aio thread is not running, we must
+		not wait for any latch, as we may end up in a deadlock:
+		if buf_fix_count == 0, then we know we need not wait */
+
+		is_s_latched = (bpage->buf_fix_count == 0);
+		if (is_s_latched && is_uncompressed) {
+			rw_lock_s_lock_gen(&((buf_block_t*) bpage)->lock,
+					   BUF_IO_WRITE);
+		}
+
+		mutex_exit(block_mutex);
+		buf_pool_mutex_exit();
+
+		/* Even though bpage is not protected by any mutex at
+		this point, it is safe to access bpage, because it is
+		io_fixed and oldest_modification != 0.  Thus, it
+		cannot be relocated in the buffer pool or removed from
+		flush_list or LRU_list. */
+
+		if (!is_s_latched) {
+			buf_flush_buffered_writes();
+
+			if (is_uncompressed) {
+				rw_lock_s_lock_gen(&((buf_block_t*) bpage)
+						   ->lock, BUF_IO_WRITE);
+			}
+		}
+
+		break;
+
+	case BUF_FLUSH_LRU:
+		/* VERY IMPORTANT:
+		Because any thread may call the LRU flush, even when owning
+		locks on pages, to avoid deadlocks, we must make sure that the
+		s-lock is acquired on the page without waiting: this is
+		accomplished because buf_flush_ready_for_flush() must hold,
+		and that requires the page not to be bufferfixed. */
+
+		if (is_uncompressed) {
+			rw_lock_s_lock_gen(&((buf_block_t*) bpage)->lock,
+					   BUF_IO_WRITE);
+		}
+
+		/* Note that the s-latch is acquired before releasing the
+		buf_pool mutex: this ensures that the latch is acquired
+		immediately. */
+
+		mutex_exit(block_mutex);
+		buf_pool_mutex_exit();
+		break;
+
+	default:
+		ut_error;
+	}
+
+	/* Even though bpage is not protected by any mutex at this
+	point, it is safe to access bpage, because it is io_fixed and
+	oldest_modification != 0.  Thus, it cannot be relocated in the
+	buffer pool or removed from flush_list or LRU_list. */
+
+#ifdef UNIV_DEBUG
+	if (buf_debug_prints) {
+		fprintf(stderr,
+			"Flushing %u space %u page %u\n",
+			flush_type, bpage->space, bpage->offset);
+	}
+#endif /* UNIV_DEBUG */
+	buf_flush_write_block_low(bpage);
+}
+
+/***************************************************************
+Flushes to disk all flushable pages within the flush area. */
+static
+ulint
+buf_flush_try_neighbors(
+/*====================*/
+					/* out: number of pages flushed */
+	ulint		space,		/* in: space id */
+	ulint		offset,		/* in: page offset */
+	enum buf_flush	flush_type)	/* in: BUF_FLUSH_LRU or
+					BUF_FLUSH_LIST */
+{
+	buf_page_t*	bpage;
+	ulint		low, high;
+	ulint		count		= 0;
+	ulint		i;
+
+	ut_ad(flush_type == BUF_FLUSH_LRU || flush_type == BUF_FLUSH_LIST);
+
+	if (UT_LIST_GET_LEN(buf_pool->LRU) < BUF_LRU_OLD_MIN_LEN) {
+		/* If there is little space, it is better not to flush any
+		block except from the end of the LRU list */
+
+		low = offset;
+		high = offset + 1;
+	} else {
+		/* When flushed, dirty blocks are searched in neighborhoods of
+		this size, and flushed along with the original page. */
+
+		ulint	buf_flush_area	= ut_min(BUF_READ_AHEAD_AREA,
+						 buf_pool->curr_size / 16);
+
+		low = (offset / buf_flush_area) * buf_flush_area;
+		high = (offset / buf_flush_area + 1) * buf_flush_area;
+	}
+
+	/* fprintf(stderr, "Flush area: low %lu high %lu\n", low, high); */
+
+	if (high > fil_space_get_size(space)) {
+		high = fil_space_get_size(space);
+	}
+
+	buf_pool_mutex_enter();
+
+	for (i = low; i < high; i++) {
+
+		bpage = buf_page_hash_get(space, i);
+
+		if (!bpage) {
+
+			continue;
+		}
+
+		ut_a(buf_page_in_file(bpage));
+
+		/* We avoid flushing 'non-old' blocks in an LRU flush,
+		because the flushed blocks are soon freed */
+
+		if (flush_type != BUF_FLUSH_LRU
+		    || i == offset
+		    || buf_page_is_old(bpage)) {
+			mutex_t* block_mutex = buf_page_get_mutex(bpage);
+
+			mutex_enter(block_mutex);
+
+			if (buf_flush_ready_for_flush(bpage, flush_type)
+			    && (i == offset || !bpage->buf_fix_count)) {
+				/* We only try to flush those
+				neighbors != offset where the buf fix count is
+				zero, as we then know that we probably can
+				latch the page without a semaphore wait.
+				Semaphore waits are expensive because we must
+				flush the doublewrite buffer before we start
+				waiting. */
+
+				buf_flush_page(bpage, flush_type);
+				ut_ad(!mutex_own(block_mutex));
+				count++;
+
+				buf_pool_mutex_enter();
+			} else {
+				mutex_exit(block_mutex);
+			}
+		}
+	}
+
+	buf_pool_mutex_exit();
+
+	return(count);
+}
+
+/***********************************************************************
+This utility flushes dirty blocks from the end of the LRU list or flush_list.
+NOTE 1: in the case of an LRU flush the calling thread may own latches to
+pages: to avoid deadlocks, this function must be written so that it cannot
+end up waiting for these latches! NOTE 2: in the case of a flush list flush,
+the calling thread is not allowed to own any latches on pages! */
+UNIV_INTERN
+ulint
+buf_flush_batch(
+/*============*/
+					/* out: number of blocks for which the
+					write request was queued;
+					ULINT_UNDEFINED if there was a flush
+					of the same type already running */
+	enum buf_flush	flush_type,	/* in: BUF_FLUSH_LRU or
+					BUF_FLUSH_LIST; if BUF_FLUSH_LIST,
+					then the caller must not own any
+					latches on pages */
+	ulint		min_n,		/* in: wished minimum mumber of blocks
+					flushed (it is not guaranteed that the
+					actual number is that big, though) */
+	ib_uint64_t	lsn_limit)	/* in the case BUF_FLUSH_LIST all
+					blocks whose oldest_modification is
+					smaller than this should be flushed
+					(if their number does not exceed
+					min_n), otherwise ignored */
+{
+	buf_page_t*	bpage;
+	ulint		page_count	= 0;
+	ulint		old_page_count;
+	ulint		space;
+	ulint		offset;
+
+	ut_ad((flush_type == BUF_FLUSH_LRU)
+	      || (flush_type == BUF_FLUSH_LIST));
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad((flush_type != BUF_FLUSH_LIST)
+	      || sync_thread_levels_empty_gen(TRUE));
+#endif /* UNIV_SYNC_DEBUG */
+	buf_pool_mutex_enter();
+
+	if ((buf_pool->n_flush[flush_type] > 0)
+	    || (buf_pool->init_flush[flush_type] == TRUE)) {
+
+		/* There is already a flush batch of the same type running */
+
+		buf_pool_mutex_exit();
+
+		return(ULINT_UNDEFINED);
+	}
+
+	buf_pool->init_flush[flush_type] = TRUE;
+
+	for (;;) {
+flush_next:
+		/* If we have flushed enough, leave the loop */
+		if (page_count >= min_n) {
+
+			break;
+		}
+
+		/* Start from the end of the list looking for a suitable
+		block to be flushed. */
+
+		if (flush_type == BUF_FLUSH_LRU) {
+			bpage = UT_LIST_GET_LAST(buf_pool->LRU);
+		} else {
+			ut_ad(flush_type == BUF_FLUSH_LIST);
+
+			bpage = UT_LIST_GET_LAST(buf_pool->flush_list);
+			if (!bpage
+			    || bpage->oldest_modification >= lsn_limit) {
+				/* We have flushed enough */
+
+				break;
+			}
+			ut_ad(bpage->in_flush_list);
+		}
+
+		/* Note that after finding a single flushable page, we try to
+		flush also all its neighbors, and after that start from the
+		END of the LRU list or flush list again: the list may change
+		during the flushing and we cannot safely preserve within this
+		function a pointer to a block in the list! */
+
+		do {
+			mutex_t*block_mutex = buf_page_get_mutex(bpage);
+			ibool	ready;
+
+			ut_a(buf_page_in_file(bpage));
+
+			mutex_enter(block_mutex);
+			ready = buf_flush_ready_for_flush(bpage, flush_type);
+			mutex_exit(block_mutex);
+
+			if (ready) {
+				space = buf_page_get_space(bpage);
+				offset = buf_page_get_page_no(bpage);
+
+				buf_pool_mutex_exit();
+
+				old_page_count = page_count;
+
+				if (srv_flush_neighbor_pages) {
+				/* Try to flush also all the neighbors */
+				page_count += buf_flush_try_neighbors(
+					space, offset, flush_type);
+				} else {
+					/* Try to flush the page only */
+					buf_pool_mutex_enter();
+
+					mutex_t* block_mutex = buf_page_get_mutex(bpage);
+					mutex_enter(block_mutex);
+
+					buf_page_t* bpage_tmp = buf_page_hash_get(space, offset);
+					if (bpage_tmp) {
+						buf_flush_page(bpage_tmp, flush_type);
+						page_count++;
+					}
+				}
+				/* fprintf(stderr,
+				"Flush type %lu, page no %lu, neighb %lu\n",
+				flush_type, offset,
+				page_count - old_page_count); */
+
+				buf_pool_mutex_enter();
+				goto flush_next;
+
+			} else if (flush_type == BUF_FLUSH_LRU) {
+				bpage = UT_LIST_GET_PREV(LRU, bpage);
+			} else {
+				ut_ad(flush_type == BUF_FLUSH_LIST);
+
+				bpage = UT_LIST_GET_PREV(list, bpage);
+				ut_ad(!bpage || bpage->in_flush_list);
+			}
+		} while (bpage != NULL);
+
+		/* If we could not find anything to flush, leave the loop */
+
+		break;
+	}
+
+	buf_pool->init_flush[flush_type] = FALSE;
+
+	if (buf_pool->n_flush[flush_type] == 0) {
+
+		/* The running flush batch has ended */
+
+		os_event_set(buf_pool->no_flush[flush_type]);
+	}
+
+	buf_pool_mutex_exit();
+
+	buf_flush_buffered_writes();
+
+#ifdef UNIV_DEBUG
+	if (buf_debug_prints && page_count > 0) {
+		ut_a(flush_type == BUF_FLUSH_LRU
+		     || flush_type == BUF_FLUSH_LIST);
+		fprintf(stderr, flush_type == BUF_FLUSH_LRU
+			? "Flushed %lu pages in LRU flush\n"
+			: "Flushed %lu pages in flush list flush\n",
+			(ulong) page_count);
+	}
+#endif /* UNIV_DEBUG */
+
+	srv_buf_pool_flushed += page_count;
+
+	return(page_count);
+}
+
+/**********************************************************************
+Waits until a flush batch of the given type ends */
+UNIV_INTERN
+void
+buf_flush_wait_batch_end(
+/*=====================*/
+	enum buf_flush	type)	/* in: BUF_FLUSH_LRU or BUF_FLUSH_LIST */
+{
+	ut_ad((type == BUF_FLUSH_LRU) || (type == BUF_FLUSH_LIST));
+
+	os_event_wait(buf_pool->no_flush[type]);
+}
+
+/**********************************************************************
+Gives a recommendation of how many blocks should be flushed to establish
+a big enough margin of replaceable blocks near the end of the LRU list
+and in the free list. */
+static
+ulint
+buf_flush_LRU_recommendation(void)
+/*==============================*/
+			/* out: number of blocks which should be flushed
+			from the end of the LRU list */
+{
+	buf_page_t*	bpage;
+	ulint		n_replaceable;
+	ulint		distance	= 0;
+	ibool		have_LRU_mutex = FALSE;
+
+	if(UT_LIST_GET_LEN(buf_pool->unzip_LRU))
+		have_LRU_mutex = TRUE;
+
+	//buf_pool_mutex_enter();
+	if (have_LRU_mutex)
+		buf_pool_mutex_enter();
+
+	n_replaceable = UT_LIST_GET_LEN(buf_pool->free);
+
+	bpage = UT_LIST_GET_LAST(buf_pool->LRU);
+
+	while ((bpage != NULL)
+	       && (n_replaceable < BUF_FLUSH_FREE_BLOCK_MARGIN
+		   + BUF_FLUSH_EXTRA_MARGIN)
+	       && (distance < BUF_LRU_FREE_SEARCH_LEN)) {
+
+		if (!bpage->in_LRU_list) {
+			/* reatart. but it is very optimistic */
+			bpage = UT_LIST_GET_LAST(buf_pool->LRU);
+			continue;
+		}
+
+		mutex_t* block_mutex = buf_page_get_mutex(bpage);
+
+		mutex_enter(block_mutex);
+
+		if (buf_flush_ready_for_replace(bpage)) {
+			n_replaceable++;
+		}
+
+		mutex_exit(block_mutex);
+
+		distance++;
+
+		bpage = UT_LIST_GET_PREV(LRU, bpage);
+	}
+
+	//buf_pool_mutex_exit();
+	if (have_LRU_mutex)
+		buf_pool_mutex_exit();
+
+	if (n_replaceable >= BUF_FLUSH_FREE_BLOCK_MARGIN) {
+
+		return(0);
+	}
+
+	return(BUF_FLUSH_FREE_BLOCK_MARGIN + BUF_FLUSH_EXTRA_MARGIN
+	       - n_replaceable);
+}
+
+/*************************************************************************
+Flushes pages from the end of the LRU list if there is too small a margin
+of replaceable pages there or in the free list. VERY IMPORTANT: this function
+is called also by threads which have locks on pages. To avoid deadlocks, we
+flush only pages such that the s-lock required for flushing can be acquired
+immediately, without waiting. */
+UNIV_INTERN
+void
+buf_flush_free_margin(
+/*=======================*/
+	ibool	wait)
+{
+	ulint	n_to_flush;
+	ulint	n_flushed;
+
+	n_to_flush = buf_flush_LRU_recommendation();
+
+	if (n_to_flush > 0) {
+		n_flushed = buf_flush_batch(BUF_FLUSH_LRU, n_to_flush, 0);
+		if (wait && n_flushed == ULINT_UNDEFINED) {
+			/* There was an LRU type flush batch already running;
+			let us wait for it to end */
+
+			buf_flush_wait_batch_end(BUF_FLUSH_LRU);
+		}
+	}
+}
+
+#if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
+/**********************************************************************
+Validates the flush list. */
+static
+ibool
+buf_flush_validate_low(void)
+/*========================*/
+		/* out: TRUE if ok */
+{
+	buf_page_t*	bpage;
+
+	UT_LIST_VALIDATE(list, buf_page_t, buf_pool->flush_list);
+
+	bpage = UT_LIST_GET_FIRST(buf_pool->flush_list);
+
+	while (bpage != NULL) {
+		const ib_uint64_t om = bpage->oldest_modification;
+		ut_ad(bpage->in_flush_list);
+		ut_a(buf_page_in_file(bpage));
+		ut_a(om > 0);
+
+		bpage = UT_LIST_GET_NEXT(list, bpage);
+
+		ut_a(!bpage || om >= bpage->oldest_modification);
+	}
+
+	return(TRUE);
+}
+
+/**********************************************************************
+Validates the flush list. */
+UNIV_INTERN
+ibool
+buf_flush_validate(void)
+/*====================*/
+		/* out: TRUE if ok */
+{
+	ibool	ret;
+
+	buf_pool_mutex_enter();
+
+	ret = buf_flush_validate_low();
+
+	buf_pool_mutex_exit();
+
+	return(ret);
+}
+#endif /* UNIV_DEBUG || UNIV_BUF_DEBUG */
diff --git a/storage/xtradb/buf/buf0lru.c b/storage/xtradb/buf/buf0lru.c
new file mode 100644
index 00000000000..dcf4d70c3eb
--- /dev/null
+++ b/storage/xtradb/buf/buf0lru.c
@@ -0,0 +1,2060 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The database buffer replacement algorithm
+
+Created 11/5/1995 Heikki Tuuri
+*******************************************************/
+
+#include "buf0lru.h"
+
+#ifdef UNIV_NONINL
+#include "buf0lru.ic"
+#endif
+
+#include "ut0byte.h"
+#include "ut0lst.h"
+#include "ut0rnd.h"
+#include "sync0sync.h"
+#include "sync0rw.h"
+#include "hash0hash.h"
+#include "os0sync.h"
+#include "fil0fil.h"
+#include "btr0btr.h"
+#include "buf0buddy.h"
+#include "buf0buf.h"
+#include "buf0flu.h"
+#include "buf0rea.h"
+#include "btr0sea.h"
+#include "ibuf0ibuf.h"
+#include "os0file.h"
+#include "page0zip.h"
+#include "log0recv.h"
+#include "srv0srv.h"
+
+/* The number of blocks from the LRU_old pointer onward, including the block
+pointed to, must be 3/8 of the whole LRU list length, except that the
+tolerance defined below is allowed. Note that the tolerance must be small
+enough such that for even the BUF_LRU_OLD_MIN_LEN long LRU list, the
+LRU_old pointer is not allowed to point to either end of the LRU list. */
+
+#define BUF_LRU_OLD_TOLERANCE	20
+
+/* The whole LRU list length is divided by this number to determine an
+initial segment in buf_LRU_get_recent_limit */
+
+#define BUF_LRU_INITIAL_RATIO	8
+
+/* When dropping the search hash index entries before deleting an ibd
+file, we build a local array of pages belonging to that tablespace
+in the buffer pool. Following is the size of that array. */
+#define BUF_LRU_DROP_SEARCH_HASH_SIZE	1024
+
+/* If we switch on the InnoDB monitor because there are too few available
+frames in the buffer pool, we set this to TRUE */
+UNIV_INTERN ibool	buf_lru_switched_on_innodb_mon	= FALSE;
+
+/**********************************************************************
+These statistics are not 'of' LRU but 'for' LRU.  We keep count of I/O
+and page_zip_decompress() operations.  Based on the statistics,
+buf_LRU_evict_from_unzip_LRU() decides if we want to evict from
+unzip_LRU or the regular LRU.  From unzip_LRU, we will only evict the
+uncompressed frame (meaning we can evict dirty blocks as well).  From
+the regular LRU, we will evict the entire block (i.e.: both the
+uncompressed and compressed data), which must be clean. */
+
+/* Number of intervals for which we keep the history of these stats.
+Each interval is 1 second, defined by the rate at which
+srv_error_monitor_thread() calls buf_LRU_stat_update(). */
+#define BUF_LRU_STAT_N_INTERVAL 50
+
+/* Co-efficient with which we multiply I/O operations to equate them
+with page_zip_decompress() operations. */
+#define BUF_LRU_IO_TO_UNZIP_FACTOR 50
+
+/* Sampled values buf_LRU_stat_cur.
+Protected by buf_pool_mutex.  Updated by buf_LRU_stat_update(). */
+static buf_LRU_stat_t		buf_LRU_stat_arr[BUF_LRU_STAT_N_INTERVAL];
+/* Cursor to buf_LRU_stat_arr[] that is updated in a round-robin fashion. */
+static ulint			buf_LRU_stat_arr_ind;
+
+/* Current operation counters.  Not protected by any mutex.  Cleared
+by buf_LRU_stat_update(). */
+UNIV_INTERN buf_LRU_stat_t	buf_LRU_stat_cur;
+
+/* Running sum of past values of buf_LRU_stat_cur.
+Updated by buf_LRU_stat_update().  Protected by buf_pool_mutex. */
+UNIV_INTERN buf_LRU_stat_t	buf_LRU_stat_sum;
+
+/**********************************************************************
+Takes a block out of the LRU list and page hash table.
+If the block is compressed-only (BUF_BLOCK_ZIP_PAGE),
+the object will be freed and buf_pool_zip_mutex will be released.
+
+If a compressed page or a compressed-only block descriptor is freed,
+other compressed pages or compressed-only block descriptors may be
+relocated. */
+static
+enum buf_page_state
+buf_LRU_block_remove_hashed_page(
+/*=============================*/
+				/* out: the new state of the block
+				(BUF_BLOCK_ZIP_FREE if the state was
+				BUF_BLOCK_ZIP_PAGE, or BUF_BLOCK_REMOVE_HASH
+				otherwise) */
+	buf_page_t*	bpage,	/* in: block, must contain a file page and
+				be in a state where it can be freed; there
+				may or may not be a hash index to the page */
+	ibool		zip);	/* in: TRUE if should remove also the
+				compressed page of an uncompressed page */
+/**********************************************************************
+Puts a file page whose has no hash index to the free list. */
+static
+void
+buf_LRU_block_free_hashed_page(
+/*===========================*/
+	buf_block_t*	block);	/* in: block, must contain a file page and
+				be in a state where it can be freed */
+
+/**********************************************************************
+Determines if the unzip_LRU list should be used for evicting a victim
+instead of the general LRU list. */
+UNIV_INLINE
+ibool
+buf_LRU_evict_from_unzip_LRU(void)
+/*==============================*/
+				/* out: TRUE if should use unzip_LRU */
+{
+	ulint	io_avg;
+	ulint	unzip_avg;
+
+	ut_ad(buf_pool_mutex_own());
+
+	/* If the unzip_LRU list is empty, we can only use the LRU. */
+	if (UT_LIST_GET_LEN(buf_pool->unzip_LRU) == 0) {
+		return(FALSE);
+	}
+
+	/* If unzip_LRU is at most 10% of the size of the LRU list,
+	then use the LRU.  This slack allows us to keep hot
+	decompressed pages in the buffer pool. */
+	if (UT_LIST_GET_LEN(buf_pool->unzip_LRU)
+	    <= UT_LIST_GET_LEN(buf_pool->LRU) / 10) {
+		return(FALSE);
+	}
+
+	/* If eviction hasn't started yet, we assume by default
+	that a workload is disk bound. */
+	if (buf_pool->freed_page_clock == 0) {
+		return(TRUE);
+	}
+
+	/* Calculate the average over past intervals, and add the values
+	of the current interval. */
+	io_avg = buf_LRU_stat_sum.io / BUF_LRU_STAT_N_INTERVAL
+		+ buf_LRU_stat_cur.io;
+	unzip_avg = buf_LRU_stat_sum.unzip / BUF_LRU_STAT_N_INTERVAL
+		+ buf_LRU_stat_cur.unzip;
+
+	/* Decide based on our formula.  If the load is I/O bound
+	(unzip_avg is smaller than the weighted io_avg), evict an
+	uncompressed frame from unzip_LRU.  Otherwise we assume that
+	the load is CPU bound and evict from the regular LRU. */
+	return(unzip_avg <= io_avg * BUF_LRU_IO_TO_UNZIP_FACTOR);
+}
+
+/**********************************************************************
+Attempts to drop page hash index on a batch of pages belonging to a
+particular space id. */
+static
+void
+buf_LRU_drop_page_hash_batch(
+/*=========================*/
+	ulint		space_id,	/* in: space id */
+	ulint		zip_size,	/* in: compressed page size in bytes
+					or 0 for uncompressed pages */
+	const ulint*	arr,		/* in: array of page_no */
+	ulint		count)		/* in: number of entries in array */
+{
+	ulint	i;
+
+	ut_ad(arr != NULL);
+	ut_ad(count <= BUF_LRU_DROP_SEARCH_HASH_SIZE);
+
+	for (i = 0; i < count; ++i) {
+		btr_search_drop_page_hash_when_freed(space_id, zip_size,
+						     arr[i]);
+	}
+}
+
+/**********************************************************************
+When doing a DROP TABLE/DISCARD TABLESPACE we have to drop all page
+hash index entries belonging to that table. This function tries to
+do that in batch. Note that this is a 'best effort' attempt and does
+not guarantee that ALL hash entries will be removed. */
+static
+void
+buf_LRU_drop_page_hash_for_tablespace(
+/*==================================*/
+	ulint	id)	/* in: space id */
+{
+	buf_page_t*	bpage;
+	ulint*		page_arr;
+	ulint		num_entries;
+	ulint		zip_size;
+
+	zip_size = fil_space_get_zip_size(id);
+
+	if (UNIV_UNLIKELY(zip_size == ULINT_UNDEFINED)) {
+		/* Somehow, the tablespace does not exist.  Nothing to drop. */
+		ut_ad(0);
+		return;
+	}
+
+	page_arr = ut_malloc(sizeof(ulint)
+			     * BUF_LRU_DROP_SEARCH_HASH_SIZE);
+	buf_pool_mutex_enter();
+
+scan_again:
+	num_entries = 0;
+	bpage = UT_LIST_GET_LAST(buf_pool->LRU);
+
+	while (bpage != NULL) {
+		mutex_t*	block_mutex = buf_page_get_mutex(bpage);
+		buf_page_t*	prev_bpage;
+
+		mutex_enter(block_mutex);
+		prev_bpage = UT_LIST_GET_PREV(LRU, bpage);
+
+		ut_a(buf_page_in_file(bpage));
+
+		if (buf_page_get_state(bpage) != BUF_BLOCK_FILE_PAGE
+		    || bpage->space != id
+		    || bpage->buf_fix_count > 0
+		    || bpage->io_fix != BUF_IO_NONE) {
+			/* We leave the fixed pages as is in this scan.
+			To be dealt with later in the final scan. */
+			mutex_exit(block_mutex);
+			goto next_page;
+		}
+
+		if (((buf_block_t*) bpage)->is_hashed) {
+
+			/* Store the offset(i.e.: page_no) in the array
+			so that we can drop hash index in a batch
+			later. */
+			page_arr[num_entries] = bpage->offset;
+			mutex_exit(block_mutex);
+			ut_a(num_entries < BUF_LRU_DROP_SEARCH_HASH_SIZE);
+			++num_entries;
+
+			if (num_entries < BUF_LRU_DROP_SEARCH_HASH_SIZE) {
+				goto next_page;
+			}
+			/* Array full. We release the buf_pool_mutex to
+			obey the latching order. */
+			buf_pool_mutex_exit();
+
+			buf_LRU_drop_page_hash_batch(id, zip_size, page_arr,
+						     num_entries);
+			num_entries = 0;
+			buf_pool_mutex_enter();
+		} else {
+			mutex_exit(block_mutex);
+		}
+
+next_page:
+		/* Note that we may have released the buf_pool mutex
+		above after reading the prev_bpage during processing
+		of a page_hash_batch (i.e.: when the array was full).
+		This means that prev_bpage can change in LRU list.
+		This is OK because this function is a 'best effort'
+		to drop as many search hash entries as possible and
+		it does not guarantee that ALL such entries will be
+		dropped. */
+		bpage = prev_bpage;
+
+		/* If, however, bpage has been removed from LRU list
+		to the free list then we should restart the scan.
+		bpage->state is protected by buf_pool mutex. */
+		if (bpage && !buf_page_in_file(bpage)) {
+			ut_a(num_entries == 0);
+			goto scan_again;
+		}
+	}
+
+	buf_pool_mutex_exit();
+
+	/* Drop any remaining batch of search hashed pages. */
+	buf_LRU_drop_page_hash_batch(id, zip_size, page_arr, num_entries);
+	ut_free(page_arr);
+}
+
+/**********************************************************************
+Invalidates all pages belonging to a given tablespace when we are deleting
+the data file(s) of that tablespace. */
+UNIV_INTERN
+void
+buf_LRU_invalidate_tablespace(
+/*==========================*/
+	ulint	id)	/* in: space id */
+{
+	buf_page_t*	bpage;
+	ibool		all_freed;
+
+	/* Before we attempt to drop pages one by one we first
+	attempt to drop page hash index entries in batches to make
+	it more efficient. The batching attempt is a best effort
+	attempt and does not guarantee that all pages hash entries
+	will be dropped. We get rid of remaining page hash entries
+	one by one below. */
+	buf_LRU_drop_page_hash_for_tablespace(id);
+
+scan_again:
+	buf_pool_mutex_enter();
+
+	all_freed = TRUE;
+
+	bpage = UT_LIST_GET_LAST(buf_pool->LRU);
+
+	while (bpage != NULL) {
+		mutex_t*	block_mutex = buf_page_get_mutex(bpage);
+		buf_page_t*	prev_bpage;
+
+		ut_a(buf_page_in_file(bpage));
+
+		mutex_enter(block_mutex);
+		prev_bpage = UT_LIST_GET_PREV(LRU, bpage);
+
+		if (buf_page_get_space(bpage) == id) {
+			if (bpage->buf_fix_count > 0
+			    || buf_page_get_io_fix(bpage) != BUF_IO_NONE) {
+
+				/* We cannot remove this page during
+				this scan yet; maybe the system is
+				currently reading it in, or flushing
+				the modifications to the file */
+
+				all_freed = FALSE;
+
+				goto next_page;
+			}
+
+#ifdef UNIV_DEBUG
+			if (buf_debug_prints) {
+				fprintf(stderr,
+					"Dropping space %lu page %lu\n",
+					(ulong) buf_page_get_space(bpage),
+					(ulong) buf_page_get_page_no(bpage));
+			}
+#endif
+			if (buf_page_get_state(bpage) == BUF_BLOCK_FILE_PAGE
+			    && ((buf_block_t*) bpage)->is_hashed) {
+				ulint	page_no;
+				ulint	zip_size;
+
+				buf_pool_mutex_exit();
+
+				zip_size = buf_page_get_zip_size(bpage);
+				page_no = buf_page_get_page_no(bpage);
+
+				mutex_exit(block_mutex);
+
+				/* Note that the following call will acquire
+				an S-latch on the page */
+
+				btr_search_drop_page_hash_when_freed(
+					id, zip_size, page_no);
+				goto scan_again;
+			}
+
+			if (bpage->oldest_modification != 0) {
+
+				buf_flush_remove(bpage);
+			}
+
+			/* Remove from the LRU list */
+			if (buf_LRU_block_remove_hashed_page(bpage, TRUE)
+			    != BUF_BLOCK_ZIP_FREE) {
+				buf_LRU_block_free_hashed_page((buf_block_t*)
+							       bpage);
+			} else {
+				/* The block_mutex should have been
+				released by buf_LRU_block_remove_hashed_page()
+				when it returns BUF_BLOCK_ZIP_FREE. */
+				ut_ad(block_mutex == &buf_pool_zip_mutex);
+				ut_ad(!mutex_own(block_mutex));
+
+				/* The compressed block descriptor
+				(bpage) has been deallocated and
+				block_mutex released.  Also,
+				buf_buddy_free() may have relocated
+				prev_bpage.  Rescan the LRU list. */
+
+				bpage = UT_LIST_GET_LAST(buf_pool->LRU);
+				continue;
+			}
+		}
+next_page:
+		mutex_exit(block_mutex);
+		bpage = prev_bpage;
+	}
+
+	buf_pool_mutex_exit();
+
+	if (!all_freed) {
+		os_thread_sleep(20000);
+
+		goto scan_again;
+	}
+}
+
+/**********************************************************************
+Gets the minimum LRU_position field for the blocks in an initial segment
+(determined by BUF_LRU_INITIAL_RATIO) of the LRU list. The limit is not
+guaranteed to be precise, because the ulint_clock may wrap around. */
+UNIV_INTERN
+ulint
+buf_LRU_get_recent_limit(void)
+/*==========================*/
+			/* out: the limit; zero if could not determine it */
+{
+	const buf_page_t*	bpage;
+	ulint			len;
+	ulint			limit;
+
+	buf_pool_mutex_enter();
+
+	len = UT_LIST_GET_LEN(buf_pool->LRU);
+
+	if (len < BUF_LRU_OLD_MIN_LEN) {
+		/* The LRU list is too short to do read-ahead */
+
+		buf_pool_mutex_exit();
+
+		return(0);
+	}
+
+	bpage = UT_LIST_GET_FIRST(buf_pool->LRU);
+
+	limit = buf_page_get_LRU_position(bpage) - len / BUF_LRU_INITIAL_RATIO;
+
+	buf_pool_mutex_exit();
+
+	return(limit);
+}
+
+/************************************************************************
+Insert a compressed block into buf_pool->zip_clean in the LRU order. */
+UNIV_INTERN
+void
+buf_LRU_insert_zip_clean(
+/*=====================*/
+	buf_page_t*	bpage)	/* in: pointer to the block in question */
+{
+	buf_page_t*	b;
+
+	ut_ad(buf_pool_mutex_own());
+	ut_ad(buf_page_get_state(bpage) == BUF_BLOCK_ZIP_PAGE);
+
+	/* Find the first successor of bpage in the LRU list
+	that is in the zip_clean list. */
+	b = bpage;
+	do {
+		b = UT_LIST_GET_NEXT(LRU, b);
+	} while (b && buf_page_get_state(b) != BUF_BLOCK_ZIP_PAGE);
+
+	/* Insert bpage before b, i.e., after the predecessor of b. */
+	if (b) {
+		b = UT_LIST_GET_PREV(list, b);
+	}
+
+	if (b) {
+		UT_LIST_INSERT_AFTER(list, buf_pool->zip_clean, b, bpage);
+	} else {
+		UT_LIST_ADD_FIRST(list, buf_pool->zip_clean, bpage);
+	}
+}
+
+/**********************************************************************
+Try to free an uncompressed page of a compressed block from the unzip
+LRU list.  The compressed page is preserved, and it need not be clean. */
+UNIV_INLINE
+ibool
+buf_LRU_free_from_unzip_LRU_list(
+/*=============================*/
+				/* out: TRUE if freed */
+	ulint	n_iterations)	/* in: how many times this has been called
+				repeatedly without result: a high value means
+				that we should search farther; we will search
+				n_iterations / 5 of the unzip_LRU list,
+				or nothing if n_iterations >= 5 */
+{
+	buf_block_t*	block;
+	ulint		distance;
+
+	ut_ad(buf_pool_mutex_own());
+
+	/* Theoratically it should be much easier to find a victim
+	from unzip_LRU as we can choose even a dirty block (as we'll
+	be evicting only the uncompressed frame).  In a very unlikely
+	eventuality that we are unable to find a victim from
+	unzip_LRU, we fall back to the regular LRU list.  We do this
+	if we have done five iterations so far. */
+
+	if (UNIV_UNLIKELY(n_iterations >= 5)
+	    || !buf_LRU_evict_from_unzip_LRU()) {
+
+		return(FALSE);
+	}
+
+	distance = 100 + (n_iterations
+			  * UT_LIST_GET_LEN(buf_pool->unzip_LRU)) / 5;
+
+	for (block = UT_LIST_GET_LAST(buf_pool->unzip_LRU);
+	     UNIV_LIKELY(block != NULL) && UNIV_LIKELY(distance > 0);
+	     block = UT_LIST_GET_PREV(unzip_LRU, block), distance--) {
+
+		enum buf_lru_free_block_status	freed;
+
+		ut_ad(buf_block_get_state(block) == BUF_BLOCK_FILE_PAGE);
+		ut_ad(block->in_unzip_LRU_list);
+		ut_ad(block->page.in_LRU_list);
+
+		mutex_enter(&block->mutex);
+		freed = buf_LRU_free_block(&block->page, FALSE, NULL);
+		mutex_exit(&block->mutex);
+
+		switch (freed) {
+		case BUF_LRU_FREED:
+			return(TRUE);
+
+		case BUF_LRU_CANNOT_RELOCATE:
+			/* If we failed to relocate, try
+			regular LRU eviction. */
+			return(FALSE);
+
+		case BUF_LRU_NOT_FREED:
+			/* The block was buffer-fixed or I/O-fixed.
+			Keep looking. */
+			continue;
+		}
+
+		/* inappropriate return value from
+		buf_LRU_free_block() */
+		ut_error;
+	}
+
+	return(FALSE);
+}
+
+/**********************************************************************
+Try to free a clean page from the common LRU list. */
+UNIV_INLINE
+ibool
+buf_LRU_free_from_common_LRU_list(
+/*==============================*/
+				/* out: TRUE if freed */
+	ulint	n_iterations)	/* in: how many times this has been called
+				repeatedly without result: a high value means
+				that we should search farther; if
+				n_iterations < 10, then we search
+				n_iterations / 10 * buf_pool->curr_size
+				pages from the end of the LRU list */
+{
+	buf_page_t*	bpage;
+	ulint		distance;
+
+	ut_ad(buf_pool_mutex_own());
+
+	distance = 100 + (n_iterations * buf_pool->curr_size) / 10;
+
+	for (bpage = UT_LIST_GET_LAST(buf_pool->LRU);
+	     UNIV_LIKELY(bpage != NULL) && UNIV_LIKELY(distance > 0);
+	     bpage = UT_LIST_GET_PREV(LRU, bpage), distance--) {
+
+		enum buf_lru_free_block_status	freed;
+		mutex_t*			block_mutex
+			= buf_page_get_mutex(bpage);
+
+		ut_ad(buf_page_in_file(bpage));
+		ut_ad(bpage->in_LRU_list);
+
+		mutex_enter(block_mutex);
+		freed = buf_LRU_free_block(bpage, TRUE, NULL);
+		mutex_exit(block_mutex);
+
+		switch (freed) {
+		case BUF_LRU_FREED:
+			return(TRUE);
+
+		case BUF_LRU_NOT_FREED:
+			/* The block was dirty, buffer-fixed, or I/O-fixed.
+			Keep looking. */
+			continue;
+
+		case BUF_LRU_CANNOT_RELOCATE:
+			/* This should never occur, because we
+			want to discard the compressed page too. */
+			break;
+		}
+
+		/* inappropriate return value from
+		buf_LRU_free_block() */
+		ut_error;
+	}
+
+	return(FALSE);
+}
+
+/**********************************************************************
+Try to free a replaceable block. */
+UNIV_INTERN
+ibool
+buf_LRU_search_and_free_block(
+/*==========================*/
+				/* out: TRUE if found and freed */
+	ulint	n_iterations)	/* in: how many times this has been called
+				repeatedly without result: a high value means
+				that we should search farther; if
+				n_iterations < 10, then we search
+				n_iterations / 10 * buf_pool->curr_size
+				pages from the end of the LRU list; if
+				n_iterations < 5, then we will also search
+				n_iterations / 5 of the unzip_LRU list. */
+{
+	ibool	freed = FALSE;
+
+	buf_pool_mutex_enter();
+
+	freed = buf_LRU_free_from_unzip_LRU_list(n_iterations);
+
+	if (!freed) {
+		freed = buf_LRU_free_from_common_LRU_list(n_iterations);
+	}
+
+	if (!freed) {
+		buf_pool->LRU_flush_ended = 0;
+	} else if (buf_pool->LRU_flush_ended > 0) {
+		buf_pool->LRU_flush_ended--;
+	}
+
+	buf_pool_mutex_exit();
+
+	return(freed);
+}
+
+/**********************************************************************
+Tries to remove LRU flushed blocks from the end of the LRU list and put them
+to the free list. This is beneficial for the efficiency of the insert buffer
+operation, as flushed pages from non-unique non-clustered indexes are here
+taken out of the buffer pool, and their inserts redirected to the insert
+buffer. Otherwise, the flushed blocks could get modified again before read
+operations need new buffer blocks, and the i/o work done in flushing would be
+wasted. */
+UNIV_INTERN
+void
+buf_LRU_try_free_flushed_blocks(void)
+/*=================================*/
+{
+	buf_pool_mutex_enter();
+
+	while (buf_pool->LRU_flush_ended > 0) {
+
+		buf_pool_mutex_exit();
+
+		buf_LRU_search_and_free_block(1);
+
+		buf_pool_mutex_enter();
+	}
+
+	buf_pool_mutex_exit();
+}
+
+/**********************************************************************
+Returns TRUE if less than 25 % of the buffer pool is available. This can be
+used in heuristics to prevent huge transactions eating up the whole buffer
+pool for their locks. */
+UNIV_INTERN
+ibool
+buf_LRU_buf_pool_running_out(void)
+/*==============================*/
+				/* out: TRUE if less than 25 % of buffer pool
+				left */
+{
+	ibool	ret	= FALSE;
+
+	buf_pool_mutex_enter();
+
+	if (!recv_recovery_on && UT_LIST_GET_LEN(buf_pool->free)
+	    + UT_LIST_GET_LEN(buf_pool->LRU) < buf_pool->curr_size / 4) {
+
+		ret = TRUE;
+	}
+
+	buf_pool_mutex_exit();
+
+	return(ret);
+}
+
+/**********************************************************************
+Returns a free block from the buf_pool.  The block is taken off the
+free list.  If it is empty, returns NULL. */
+UNIV_INTERN
+buf_block_t*
+buf_LRU_get_free_only(void)
+/*=======================*/
+				/* out: a free control block, or NULL
+				if the buf_block->free list is empty */
+{
+	buf_block_t*	block;
+
+	ut_ad(buf_pool_mutex_own());
+
+	block = (buf_block_t*) UT_LIST_GET_FIRST(buf_pool->free);
+
+	if (block) {
+		ut_ad(block->page.in_free_list);
+		ut_d(block->page.in_free_list = FALSE);
+		ut_ad(!block->page.in_flush_list);
+		ut_ad(!block->page.in_LRU_list);
+		ut_a(!buf_page_in_file(&block->page));
+		UT_LIST_REMOVE(list, buf_pool->free, (&block->page));
+
+		mutex_enter(&block->mutex);
+
+		buf_block_set_state(block, BUF_BLOCK_READY_FOR_USE);
+		UNIV_MEM_ALLOC(block->frame, UNIV_PAGE_SIZE);
+
+		mutex_exit(&block->mutex);
+	}
+
+	return(block);
+}
+
+/**********************************************************************
+Returns a free block from the buf_pool. The block is taken off the
+free list. If it is empty, blocks are moved from the end of the
+LRU list to the free list. */
+UNIV_INTERN
+buf_block_t*
+buf_LRU_get_free_block(
+/*===================*/
+				/* out: the free control block,
+				in state BUF_BLOCK_READY_FOR_USE */
+	ulint	zip_size)	/* in: compressed page size in bytes,
+				or 0 if uncompressed tablespace */
+{
+	buf_block_t*	block		= NULL;
+	ibool		freed;
+	ulint		n_iterations	= 1;
+	ibool		mon_value_was	= FALSE;
+	ibool		started_monitor	= FALSE;
+loop:
+	buf_pool_mutex_enter();
+
+	if (!recv_recovery_on && UT_LIST_GET_LEN(buf_pool->free)
+	    + UT_LIST_GET_LEN(buf_pool->LRU) < buf_pool->curr_size / 20) {
+		ut_print_timestamp(stderr);
+
+		fprintf(stderr,
+			"  InnoDB: ERROR: over 95 percent of the buffer pool"
+			" is occupied by\n"
+			"InnoDB: lock heaps or the adaptive hash index!"
+			" Check that your\n"
+			"InnoDB: transactions do not set too many row locks.\n"
+			"InnoDB: Your buffer pool size is %lu MB."
+			" Maybe you should make\n"
+			"InnoDB: the buffer pool bigger?\n"
+			"InnoDB: We intentionally generate a seg fault"
+			" to print a stack trace\n"
+			"InnoDB: on Linux!\n",
+			(ulong) (buf_pool->curr_size
+				 / (1024 * 1024 / UNIV_PAGE_SIZE)));
+
+		ut_error;
+
+	} else if (!recv_recovery_on
+		   && (UT_LIST_GET_LEN(buf_pool->free)
+		       + UT_LIST_GET_LEN(buf_pool->LRU))
+		   < buf_pool->curr_size / 3) {
+
+		if (!buf_lru_switched_on_innodb_mon) {
+
+			/* Over 67 % of the buffer pool is occupied by lock
+			heaps or the adaptive hash index. This may be a memory
+			leak! */
+
+			ut_print_timestamp(stderr);
+			fprintf(stderr,
+				"  InnoDB: WARNING: over 67 percent of"
+				" the buffer pool is occupied by\n"
+				"InnoDB: lock heaps or the adaptive"
+				" hash index! Check that your\n"
+				"InnoDB: transactions do not set too many"
+				" row locks.\n"
+				"InnoDB: Your buffer pool size is %lu MB."
+				" Maybe you should make\n"
+				"InnoDB: the buffer pool bigger?\n"
+				"InnoDB: Starting the InnoDB Monitor to print"
+				" diagnostics, including\n"
+				"InnoDB: lock heap and hash index sizes.\n",
+				(ulong) (buf_pool->curr_size
+					 / (1024 * 1024 / UNIV_PAGE_SIZE)));
+
+			buf_lru_switched_on_innodb_mon = TRUE;
+			srv_print_innodb_monitor = TRUE;
+			os_event_set(srv_lock_timeout_thread_event);
+		}
+	} else if (buf_lru_switched_on_innodb_mon) {
+
+		/* Switch off the InnoDB Monitor; this is a simple way
+		to stop the monitor if the situation becomes less urgent,
+		but may also surprise users if the user also switched on the
+		monitor! */
+
+		buf_lru_switched_on_innodb_mon = FALSE;
+		srv_print_innodb_monitor = FALSE;
+	}
+
+	/* If there is a block in the free list, take it */
+	block = buf_LRU_get_free_only();
+	if (block) {
+
+#ifdef UNIV_DEBUG
+		block->page.zip.m_start =
+#endif /* UNIV_DEBUG */
+			block->page.zip.m_end =
+			block->page.zip.m_nonempty =
+			block->page.zip.n_blobs = 0;
+
+		if (UNIV_UNLIKELY(zip_size)) {
+			ibool	lru;
+			page_zip_set_size(&block->page.zip, zip_size);
+			block->page.zip.data = buf_buddy_alloc(zip_size, &lru);
+			UNIV_MEM_DESC(block->page.zip.data, zip_size, block);
+		} else {
+			page_zip_set_size(&block->page.zip, 0);
+			block->page.zip.data = NULL;
+		}
+
+		buf_pool_mutex_exit();
+
+		if (started_monitor) {
+			srv_print_innodb_monitor = mon_value_was;
+		}
+
+		return(block);
+	}
+
+	/* If no block was in the free list, search from the end of the LRU
+	list and try to free a block there */
+
+	buf_pool_mutex_exit();
+
+	freed = buf_LRU_search_and_free_block(n_iterations);
+
+	if (freed > 0) {
+		goto loop;
+	}
+
+	if (n_iterations > 30) {
+		ut_print_timestamp(stderr);
+		fprintf(stderr,
+			"  InnoDB: Warning: difficult to find free blocks in\n"
+			"InnoDB: the buffer pool (%lu search iterations)!"
+			" Consider\n"
+			"InnoDB: increasing the buffer pool size.\n"
+			"InnoDB: It is also possible that"
+			" in your Unix version\n"
+			"InnoDB: fsync is very slow, or"
+			" completely frozen inside\n"
+			"InnoDB: the OS kernel. Then upgrading to"
+			" a newer version\n"
+			"InnoDB: of your operating system may help."
+			" Look at the\n"
+			"InnoDB: number of fsyncs in diagnostic info below.\n"
+			"InnoDB: Pending flushes (fsync) log: %lu;"
+			" buffer pool: %lu\n"
+			"InnoDB: %lu OS file reads, %lu OS file writes,"
+			" %lu OS fsyncs\n"
+			"InnoDB: Starting InnoDB Monitor to print further\n"
+			"InnoDB: diagnostics to the standard output.\n",
+			(ulong) n_iterations,
+			(ulong) fil_n_pending_log_flushes,
+			(ulong) fil_n_pending_tablespace_flushes,
+			(ulong) os_n_file_reads, (ulong) os_n_file_writes,
+			(ulong) os_n_fsyncs);
+
+		mon_value_was = srv_print_innodb_monitor;
+		started_monitor = TRUE;
+		srv_print_innodb_monitor = TRUE;
+		os_event_set(srv_lock_timeout_thread_event);
+	}
+
+	/* No free block was found: try to flush the LRU list */
+
+	buf_flush_free_margin(TRUE);
+	++srv_buf_pool_wait_free;
+
+	os_aio_simulated_wake_handler_threads();
+
+	buf_pool_mutex_enter();
+
+	if (buf_pool->LRU_flush_ended > 0) {
+		/* We have written pages in an LRU flush. To make the insert
+		buffer more efficient, we try to move these pages to the free
+		list. */
+
+		buf_pool_mutex_exit();
+
+		buf_LRU_try_free_flushed_blocks();
+	} else {
+		buf_pool_mutex_exit();
+	}
+
+	if (n_iterations > 10) {
+
+		os_thread_sleep(500000);
+	}
+
+	n_iterations++;
+
+	goto loop;
+}
+
+/***********************************************************************
+Moves the LRU_old pointer so that the length of the old blocks list
+is inside the allowed limits. */
+UNIV_INLINE
+void
+buf_LRU_old_adjust_len(void)
+/*========================*/
+{
+	ulint	old_len;
+	ulint	new_len;
+
+	ut_a(buf_pool->LRU_old);
+	ut_ad(buf_pool_mutex_own());
+#if 3 * (BUF_LRU_OLD_MIN_LEN / 8) <= BUF_LRU_OLD_TOLERANCE + 5
+# error "3 * (BUF_LRU_OLD_MIN_LEN / 8) <= BUF_LRU_OLD_TOLERANCE + 5"
+#endif
+#ifdef UNIV_LRU_DEBUG
+	/* buf_pool->LRU_old must be the first item in the LRU list
+	whose "old" flag is set. */
+	ut_a(buf_pool->LRU_old->old);
+	ut_a(!UT_LIST_GET_PREV(LRU, buf_pool->LRU_old)
+	     || !UT_LIST_GET_PREV(LRU, buf_pool->LRU_old)->old);
+	ut_a(!UT_LIST_GET_NEXT(LRU, buf_pool->LRU_old)
+	     || UT_LIST_GET_NEXT(LRU, buf_pool->LRU_old)->old);
+#endif /* UNIV_LRU_DEBUG */
+
+	for (;;) {
+		old_len = buf_pool->LRU_old_len;
+		new_len = 3 * (UT_LIST_GET_LEN(buf_pool->LRU) / 8);
+
+		ut_ad(buf_pool->LRU_old->in_LRU_list);
+		ut_a(buf_pool->LRU_old);
+#ifdef UNIV_LRU_DEBUG
+		ut_a(buf_pool->LRU_old->old);
+#endif /* UNIV_LRU_DEBUG */
+
+		/* Update the LRU_old pointer if necessary */
+
+		if (old_len < new_len - BUF_LRU_OLD_TOLERANCE) {
+
+			buf_pool->LRU_old = UT_LIST_GET_PREV(
+				LRU, buf_pool->LRU_old);
+#ifdef UNIV_LRU_DEBUG
+			ut_a(!buf_pool->LRU_old->old);
+#endif /* UNIV_LRU_DEBUG */
+			buf_page_set_old(buf_pool->LRU_old, TRUE);
+			buf_pool->LRU_old_len++;
+
+		} else if (old_len > new_len + BUF_LRU_OLD_TOLERANCE) {
+
+			buf_page_set_old(buf_pool->LRU_old, FALSE);
+			buf_pool->LRU_old = UT_LIST_GET_NEXT(
+				LRU, buf_pool->LRU_old);
+			buf_pool->LRU_old_len--;
+		} else {
+			return;
+		}
+	}
+}
+
+/***********************************************************************
+Initializes the old blocks pointer in the LRU list. This function should be
+called when the LRU list grows to BUF_LRU_OLD_MIN_LEN length. */
+static
+void
+buf_LRU_old_init(void)
+/*==================*/
+{
+	buf_page_t*	bpage;
+
+	ut_ad(buf_pool_mutex_own());
+	ut_a(UT_LIST_GET_LEN(buf_pool->LRU) == BUF_LRU_OLD_MIN_LEN);
+
+	/* We first initialize all blocks in the LRU list as old and then use
+	the adjust function to move the LRU_old pointer to the right
+	position */
+
+	bpage = UT_LIST_GET_FIRST(buf_pool->LRU);
+
+	while (bpage != NULL) {
+		ut_ad(bpage->in_LRU_list);
+		buf_page_set_old(bpage, TRUE);
+		bpage = UT_LIST_GET_NEXT(LRU, bpage);
+	}
+
+	buf_pool->LRU_old = UT_LIST_GET_FIRST(buf_pool->LRU);
+	buf_pool->LRU_old_len = UT_LIST_GET_LEN(buf_pool->LRU);
+
+	buf_LRU_old_adjust_len();
+}
+
+/**********************************************************************
+Remove a block from the unzip_LRU list if it belonged to the list. */
+static
+void
+buf_unzip_LRU_remove_block_if_needed(
+/*=================================*/
+	buf_page_t*	bpage)	/* in/out: control block */
+{
+	ut_ad(buf_pool);
+	ut_ad(bpage);
+	ut_ad(buf_page_in_file(bpage));
+	ut_ad(buf_pool_mutex_own());
+
+	if (buf_page_belongs_to_unzip_LRU(bpage)) {
+		buf_block_t*	block = (buf_block_t*) bpage;
+
+		ut_ad(block->in_unzip_LRU_list);
+		ut_d(block->in_unzip_LRU_list = FALSE);
+
+		UT_LIST_REMOVE(unzip_LRU, buf_pool->unzip_LRU, block);
+	}
+}
+
+/**********************************************************************
+Removes a block from the LRU list. */
+UNIV_INLINE
+void
+buf_LRU_remove_block(
+/*=================*/
+	buf_page_t*	bpage)	/* in: control block */
+{
+	ut_ad(buf_pool);
+	ut_ad(bpage);
+	ut_ad(buf_pool_mutex_own());
+
+	ut_a(buf_page_in_file(bpage));
+
+	ut_ad(bpage->in_LRU_list);
+
+	/* If the LRU_old pointer is defined and points to just this block,
+	move it backward one step */
+
+	if (UNIV_UNLIKELY(bpage == buf_pool->LRU_old)) {
+
+		/* Below: the previous block is guaranteed to exist, because
+		the LRU_old pointer is only allowed to differ by the
+		tolerance value from strict 3/8 of the LRU list length. */
+
+		buf_pool->LRU_old = UT_LIST_GET_PREV(LRU, bpage);
+		ut_a(buf_pool->LRU_old);
+#ifdef UNIV_LRU_DEBUG
+		ut_a(!buf_pool->LRU_old->old);
+#endif /* UNIV_LRU_DEBUG */
+		buf_page_set_old(buf_pool->LRU_old, TRUE);
+
+		buf_pool->LRU_old_len++;
+	}
+
+	/* Remove the block from the LRU list */
+	UT_LIST_REMOVE(LRU, buf_pool->LRU, bpage);
+	bpage->in_LRU_list = FALSE;
+
+	buf_unzip_LRU_remove_block_if_needed(bpage);
+
+	/* If the LRU list is so short that LRU_old not defined, return */
+	if (UT_LIST_GET_LEN(buf_pool->LRU) < BUF_LRU_OLD_MIN_LEN) {
+
+		buf_pool->LRU_old = NULL;
+
+		return;
+	}
+
+	ut_ad(buf_pool->LRU_old);
+
+	/* Update the LRU_old_len field if necessary */
+	if (buf_page_is_old(bpage)) {
+
+		buf_pool->LRU_old_len--;
+	}
+
+	/* Adjust the length of the old block list if necessary */
+	buf_LRU_old_adjust_len();
+}
+
+/**********************************************************************
+Adds a block to the LRU list of decompressed zip pages. */
+UNIV_INTERN
+void
+buf_unzip_LRU_add_block(
+/*====================*/
+	buf_block_t*	block,	/* in: control block */
+	ibool		old)	/* in: TRUE if should be put to the end
+				of the list, else put to the start */
+{
+	ut_ad(buf_pool);
+	ut_ad(block);
+	ut_ad(buf_pool_mutex_own());
+
+	ut_a(buf_page_belongs_to_unzip_LRU(&block->page));
+
+	ut_ad(!block->in_unzip_LRU_list);
+	ut_d(block->in_unzip_LRU_list = TRUE);
+
+	if (old) {
+		UT_LIST_ADD_LAST(unzip_LRU, buf_pool->unzip_LRU, block);
+	} else {
+		UT_LIST_ADD_FIRST(unzip_LRU, buf_pool->unzip_LRU, block);
+	}
+}
+
+/**********************************************************************
+Adds a block to the LRU list end. */
+UNIV_INLINE
+void
+buf_LRU_add_block_to_end_low(
+/*=========================*/
+	buf_page_t*	bpage)	/* in: control block */
+{
+	buf_page_t*	last_bpage;
+
+	ut_ad(buf_pool);
+	ut_ad(bpage);
+	ut_ad(buf_pool_mutex_own());
+
+	ut_a(buf_page_in_file(bpage));
+
+	last_bpage = UT_LIST_GET_LAST(buf_pool->LRU);
+
+	if (last_bpage) {
+		bpage->LRU_position = last_bpage->LRU_position;
+	} else {
+		bpage->LRU_position = buf_pool_clock_tic();
+	}
+
+	ut_ad(!bpage->in_LRU_list);
+	UT_LIST_ADD_LAST(LRU, buf_pool->LRU, bpage);
+	bpage->in_LRU_list = TRUE;
+
+	buf_page_set_old(bpage, TRUE);
+
+	if (UT_LIST_GET_LEN(buf_pool->LRU) >= BUF_LRU_OLD_MIN_LEN) {
+
+		buf_pool->LRU_old_len++;
+	}
+
+	if (UT_LIST_GET_LEN(buf_pool->LRU) > BUF_LRU_OLD_MIN_LEN) {
+
+		ut_ad(buf_pool->LRU_old);
+
+		/* Adjust the length of the old block list if necessary */
+
+		buf_LRU_old_adjust_len();
+
+	} else if (UT_LIST_GET_LEN(buf_pool->LRU) == BUF_LRU_OLD_MIN_LEN) {
+
+		/* The LRU list is now long enough for LRU_old to become
+		defined: init it */
+
+		buf_LRU_old_init();
+	}
+
+	/* If this is a zipped block with decompressed frame as well
+	then put it on the unzip_LRU list */
+	if (buf_page_belongs_to_unzip_LRU(bpage)) {
+		buf_unzip_LRU_add_block((buf_block_t*) bpage, TRUE);
+	}
+}
+
+/**********************************************************************
+Adds a block to the LRU list. */
+UNIV_INLINE
+void
+buf_LRU_add_block_low(
+/*==================*/
+	buf_page_t*	bpage,	/* in: control block */
+	ibool		old)	/* in: TRUE if should be put to the old blocks
+				in the LRU list, else put to the start; if the
+				LRU list is very short, the block is added to
+				the start, regardless of this parameter */
+{
+	ut_ad(buf_pool);
+	ut_ad(bpage);
+	ut_ad(buf_pool_mutex_own());
+
+	ut_a(buf_page_in_file(bpage));
+	ut_ad(!bpage->in_LRU_list);
+
+	if (!old || (UT_LIST_GET_LEN(buf_pool->LRU) < BUF_LRU_OLD_MIN_LEN)) {
+
+		UT_LIST_ADD_FIRST(LRU, buf_pool->LRU, bpage);
+
+		bpage->LRU_position = buf_pool_clock_tic();
+		bpage->freed_page_clock = buf_pool->freed_page_clock;
+	} else {
+#ifdef UNIV_LRU_DEBUG
+		/* buf_pool->LRU_old must be the first item in the LRU list
+		whose "old" flag is set. */
+		ut_a(buf_pool->LRU_old->old);
+		ut_a(!UT_LIST_GET_PREV(LRU, buf_pool->LRU_old)
+		     || !UT_LIST_GET_PREV(LRU, buf_pool->LRU_old)->old);
+		ut_a(!UT_LIST_GET_NEXT(LRU, buf_pool->LRU_old)
+		     || UT_LIST_GET_NEXT(LRU, buf_pool->LRU_old)->old);
+#endif /* UNIV_LRU_DEBUG */
+		UT_LIST_INSERT_AFTER(LRU, buf_pool->LRU, buf_pool->LRU_old,
+				     bpage);
+		buf_pool->LRU_old_len++;
+
+		/* We copy the LRU position field of the previous block
+		to the new block */
+
+		bpage->LRU_position = (buf_pool->LRU_old)->LRU_position;
+	}
+
+	bpage->in_LRU_list = TRUE;
+
+	buf_page_set_old(bpage, old);
+
+	if (UT_LIST_GET_LEN(buf_pool->LRU) > BUF_LRU_OLD_MIN_LEN) {
+
+		ut_ad(buf_pool->LRU_old);
+
+		/* Adjust the length of the old block list if necessary */
+
+		buf_LRU_old_adjust_len();
+
+	} else if (UT_LIST_GET_LEN(buf_pool->LRU) == BUF_LRU_OLD_MIN_LEN) {
+
+		/* The LRU list is now long enough for LRU_old to become
+		defined: init it */
+
+		buf_LRU_old_init();
+	}
+
+	/* If this is a zipped block with decompressed frame as well
+	then put it on the unzip_LRU list */
+	if (buf_page_belongs_to_unzip_LRU(bpage)) {
+		buf_unzip_LRU_add_block((buf_block_t*) bpage, old);
+	}
+}
+
+/**********************************************************************
+Adds a block to the LRU list. */
+UNIV_INTERN
+void
+buf_LRU_add_block(
+/*==============*/
+	buf_page_t*	bpage,	/* in: control block */
+	ibool		old)	/* in: TRUE if should be put to the old
+				blocks in the LRU list, else put to the start;
+				if the LRU list is very short, the block is
+				added to the start, regardless of this
+				parameter */
+{
+	buf_LRU_add_block_low(bpage, old);
+}
+
+/**********************************************************************
+Moves a block to the start of the LRU list. */
+UNIV_INTERN
+void
+buf_LRU_make_block_young(
+/*=====================*/
+	buf_page_t*	bpage)	/* in: control block */
+{
+	buf_LRU_remove_block(bpage);
+	buf_LRU_add_block_low(bpage, FALSE);
+}
+
+/**********************************************************************
+Moves a block to the end of the LRU list. */
+UNIV_INTERN
+void
+buf_LRU_make_block_old(
+/*===================*/
+	buf_page_t*	bpage)	/* in: control block */
+{
+	buf_LRU_remove_block(bpage);
+	buf_LRU_add_block_to_end_low(bpage);
+}
+
+/**********************************************************************
+Try to free a block.  If bpage is a descriptor of a compressed-only
+page, the descriptor object will be freed as well.
+
+NOTE: If this function returns BUF_LRU_FREED, it will not temporarily
+release buf_pool_mutex.  Furthermore, the page frame will no longer be
+accessible via bpage.
+
+The caller must hold buf_pool_mutex and buf_page_get_mutex(bpage) and
+release these two mutexes after the call.  No other
+buf_page_get_mutex() may be held when calling this function. */
+UNIV_INTERN
+enum buf_lru_free_block_status
+buf_LRU_free_block(
+/*===============*/
+				/* out: BUF_LRU_FREED if freed,
+				BUF_LRU_CANNOT_RELOCATE or
+				BUF_LRU_NOT_FREED otherwise. */
+	buf_page_t*	bpage,	/* in: block to be freed */
+	ibool		zip,	/* in: TRUE if should remove also the
+				compressed page of an uncompressed page */
+	ibool*		buf_pool_mutex_released)
+				/* in: pointer to a variable that will
+				be assigned TRUE if buf_pool_mutex
+				was temporarily released, or NULL */
+{
+	buf_page_t*	b = NULL;
+	mutex_t*	block_mutex = buf_page_get_mutex(bpage);
+
+	ut_ad(buf_pool_mutex_own());
+	ut_ad(mutex_own(block_mutex));
+	ut_ad(buf_page_in_file(bpage));
+	ut_ad(bpage->in_LRU_list);
+	ut_ad(!bpage->in_flush_list == !bpage->oldest_modification);
+	UNIV_MEM_ASSERT_RW(bpage, sizeof *bpage);
+
+	if (!buf_page_can_relocate(bpage)) {
+
+		/* Do not free buffer-fixed or I/O-fixed blocks. */
+		return(BUF_LRU_NOT_FREED);
+	}
+
+#ifdef UNIV_IBUF_COUNT_DEBUG
+	ut_a(ibuf_count_get(bpage->space, bpage->offset) == 0);
+#endif /* UNIV_IBUF_COUNT_DEBUG */
+
+	if (zip || !bpage->zip.data) {
+		/* This would completely free the block. */
+		/* Do not completely free dirty blocks. */
+
+		if (bpage->oldest_modification) {
+			return(BUF_LRU_NOT_FREED);
+		}
+	} else if (bpage->oldest_modification) {
+		/* Do not completely free dirty blocks. */
+
+		if (buf_page_get_state(bpage) != BUF_BLOCK_FILE_PAGE) {
+			ut_ad(buf_page_get_state(bpage)
+			      == BUF_BLOCK_ZIP_DIRTY);
+			return(BUF_LRU_NOT_FREED);
+		}
+
+		goto alloc;
+	} else if (buf_page_get_state(bpage) == BUF_BLOCK_FILE_PAGE) {
+		/* Allocate the control block for the compressed page.
+		If it cannot be allocated (without freeing a block
+		from the LRU list), refuse to free bpage. */
+alloc:
+		buf_pool_mutex_exit_forbid();
+		b = buf_buddy_alloc(sizeof *b, NULL);
+		buf_pool_mutex_exit_allow();
+
+		if (UNIV_UNLIKELY(!b)) {
+			return(BUF_LRU_CANNOT_RELOCATE);
+		}
+
+		memcpy(b, bpage, sizeof *b);
+	}
+
+#ifdef UNIV_DEBUG
+	if (buf_debug_prints) {
+		fprintf(stderr, "Putting space %lu page %lu to free list\n",
+			(ulong) buf_page_get_space(bpage),
+			(ulong) buf_page_get_page_no(bpage));
+	}
+#endif /* UNIV_DEBUG */
+
+	if (buf_LRU_block_remove_hashed_page(bpage, zip)
+	    != BUF_BLOCK_ZIP_FREE) {
+		ut_a(bpage->buf_fix_count == 0);
+
+		if (b) {
+			buf_page_t*	prev_b	= UT_LIST_GET_PREV(LRU, b);
+			const ulint	fold	= buf_page_address_fold(
+				bpage->space, bpage->offset);
+
+			ut_a(!buf_page_hash_get(bpage->space, bpage->offset));
+
+			b->state = b->oldest_modification
+				? BUF_BLOCK_ZIP_DIRTY
+				: BUF_BLOCK_ZIP_PAGE;
+			UNIV_MEM_DESC(b->zip.data,
+				      page_zip_get_size(&b->zip), b);
+
+			/* The fields in_page_hash and in_LRU_list of
+			the to-be-freed block descriptor should have
+			been cleared in
+			buf_LRU_block_remove_hashed_page(), which
+			invokes buf_LRU_remove_block(). */
+			ut_ad(!bpage->in_page_hash);
+			ut_ad(!bpage->in_LRU_list);
+			/* bpage->state was BUF_BLOCK_FILE_PAGE because
+			b != NULL. The type cast below is thus valid. */
+			ut_ad(!((buf_block_t*) bpage)->in_unzip_LRU_list);
+
+			/* The fields of bpage were copied to b before
+			buf_LRU_block_remove_hashed_page() was invoked. */
+			ut_ad(!b->in_zip_hash);
+			ut_ad(b->in_page_hash);
+			ut_ad(b->in_LRU_list);
+
+			HASH_INSERT(buf_page_t, hash,
+				    buf_pool->page_hash, fold, b);
+
+			/* Insert b where bpage was in the LRU list. */
+			if (UNIV_LIKELY(prev_b != NULL)) {
+				ulint	lru_len;
+
+				ut_ad(prev_b->in_LRU_list);
+				ut_ad(buf_page_in_file(prev_b));
+				UNIV_MEM_ASSERT_RW(prev_b, sizeof *prev_b);
+
+				UT_LIST_INSERT_AFTER(LRU, buf_pool->LRU,
+						     prev_b, b);
+
+				if (buf_page_is_old(b)) {
+					buf_pool->LRU_old_len++;
+					if (UNIV_UNLIKELY
+					    (buf_pool->LRU_old
+					     == UT_LIST_GET_NEXT(LRU, b))) {
+
+						buf_pool->LRU_old = b;
+					}
+#ifdef UNIV_LRU_DEBUG
+					ut_a(prev_b->old
+					     || !UT_LIST_GET_NEXT(LRU, b)
+					     || UT_LIST_GET_NEXT(LRU, b)->old);
+				} else {
+					ut_a(!prev_b->old
+					     || !UT_LIST_GET_NEXT(LRU, b)
+					     || !UT_LIST_GET_NEXT(LRU, b)->old);
+#endif /* UNIV_LRU_DEBUG */
+				}
+
+				lru_len = UT_LIST_GET_LEN(buf_pool->LRU);
+
+				if (lru_len > BUF_LRU_OLD_MIN_LEN) {
+					ut_ad(buf_pool->LRU_old);
+					/* Adjust the length of the
+					old block list if necessary */
+					buf_LRU_old_adjust_len();
+				} else if (lru_len == BUF_LRU_OLD_MIN_LEN) {
+					/* The LRU list is now long
+					enough for LRU_old to become
+					defined: init it */
+					buf_LRU_old_init();
+				}
+			} else {
+				b->in_LRU_list = FALSE;
+				buf_LRU_add_block_low(b, buf_page_is_old(b));
+			}
+
+			if (b->state == BUF_BLOCK_ZIP_PAGE) {
+				buf_LRU_insert_zip_clean(b);
+			} else {
+				buf_page_t* prev;
+
+				ut_ad(b->in_flush_list);
+				ut_d(bpage->in_flush_list = FALSE);
+
+				prev = UT_LIST_GET_PREV(list, b);
+				UT_LIST_REMOVE(list, buf_pool->flush_list, b);
+
+				if (prev) {
+					ut_ad(prev->in_flush_list);
+					UT_LIST_INSERT_AFTER(
+						list,
+						buf_pool->flush_list,
+						prev, b);
+				} else {
+					UT_LIST_ADD_FIRST(
+						list,
+						buf_pool->flush_list,
+						b);
+				}
+			}
+
+			bpage->zip.data = NULL;
+			page_zip_set_size(&bpage->zip, 0);
+
+			/* Prevent buf_page_get_gen() from
+			decompressing the block while we release
+			buf_pool_mutex and block_mutex. */
+			b->buf_fix_count++;
+			b->io_fix = BUF_IO_READ;
+		}
+
+		if (buf_pool_mutex_released) {
+			*buf_pool_mutex_released = TRUE;
+		}
+
+		buf_pool_mutex_exit();
+		mutex_exit(block_mutex);
+
+		/* Remove possible adaptive hash index on the page.
+		The page was declared uninitialized by
+		buf_LRU_block_remove_hashed_page().  We need to flag
+		the contents of the page valid (which it still is) in
+		order to avoid bogus Valgrind warnings.*/
+
+		UNIV_MEM_VALID(((buf_block_t*) bpage)->frame,
+			       UNIV_PAGE_SIZE);
+		btr_search_drop_page_hash_index((buf_block_t*) bpage);
+		UNIV_MEM_INVALID(((buf_block_t*) bpage)->frame,
+				 UNIV_PAGE_SIZE);
+
+		if (b) {
+			/* Compute and stamp the compressed page
+			checksum while not holding any mutex.  The
+			block is already half-freed
+			(BUF_BLOCK_REMOVE_HASH) and removed from
+			buf_pool->page_hash, thus inaccessible by any
+			other thread. */
+
+			mach_write_to_4(
+				b->zip.data + FIL_PAGE_SPACE_OR_CHKSUM,
+				UNIV_LIKELY(srv_use_checksums)
+				? page_zip_calc_checksum(
+					b->zip.data,
+					page_zip_get_size(&b->zip))
+				: BUF_NO_CHECKSUM_MAGIC);
+		}
+
+		buf_pool_mutex_enter();
+		mutex_enter(block_mutex);
+
+		if (b) {
+			mutex_enter(&buf_pool_zip_mutex);
+			b->buf_fix_count--;
+			buf_page_set_io_fix(b, BUF_IO_NONE);
+			mutex_exit(&buf_pool_zip_mutex);
+		}
+
+		buf_LRU_block_free_hashed_page((buf_block_t*) bpage);
+	} else {
+		/* The block_mutex should have been released by
+		buf_LRU_block_remove_hashed_page() when it returns
+		BUF_BLOCK_ZIP_FREE. */
+		ut_ad(block_mutex == &buf_pool_zip_mutex);
+		mutex_enter(block_mutex);
+	}
+
+	return(BUF_LRU_FREED);
+}
+
+/**********************************************************************
+Puts a block back to the free list. */
+UNIV_INTERN
+void
+buf_LRU_block_free_non_file_page(
+/*=============================*/
+	buf_block_t*	block)	/* in: block, must not contain a file page */
+{
+	void*	data;
+
+	ut_ad(block);
+	ut_ad(buf_pool_mutex_own());
+	ut_ad(mutex_own(&block->mutex));
+
+	switch (buf_block_get_state(block)) {
+	case BUF_BLOCK_MEMORY:
+	case BUF_BLOCK_READY_FOR_USE:
+		break;
+	default:
+		ut_error;
+	}
+
+#if defined UNIV_AHI_DEBUG || defined UNIV_DEBUG
+	ut_a(block->n_pointers == 0);
+#endif /* UNIV_AHI_DEBUG || UNIV_DEBUG */
+	ut_ad(!block->page.in_free_list);
+	ut_ad(!block->page.in_flush_list);
+	ut_ad(!block->page.in_LRU_list);
+
+	buf_block_set_state(block, BUF_BLOCK_NOT_USED);
+
+	UNIV_MEM_ALLOC(block->frame, UNIV_PAGE_SIZE);
+#ifdef UNIV_DEBUG
+	/* Wipe contents of page to reveal possible stale pointers to it */
+	memset(block->frame, '\0', UNIV_PAGE_SIZE);
+#else
+	/* Wipe page_no and space_id */
+	memset(block->frame + FIL_PAGE_OFFSET, 0xfe, 4);
+	memset(block->frame + FIL_PAGE_ARCH_LOG_NO_OR_SPACE_ID, 0xfe, 4);
+#endif
+	data = block->page.zip.data;
+
+	if (data) {
+		block->page.zip.data = NULL;
+		mutex_exit(&block->mutex);
+		buf_pool_mutex_exit_forbid();
+		buf_buddy_free(data, page_zip_get_size(&block->page.zip));
+		buf_pool_mutex_exit_allow();
+		mutex_enter(&block->mutex);
+		page_zip_set_size(&block->page.zip, 0);
+	}
+
+	UT_LIST_ADD_FIRST(list, buf_pool->free, (&block->page));
+	ut_d(block->page.in_free_list = TRUE);
+
+	UNIV_MEM_ASSERT_AND_FREE(block->frame, UNIV_PAGE_SIZE);
+}
+
+/**********************************************************************
+Takes a block out of the LRU list and page hash table.
+If the block is compressed-only (BUF_BLOCK_ZIP_PAGE),
+the object will be freed and buf_pool_zip_mutex will be released.
+
+If a compressed page or a compressed-only block descriptor is freed,
+other compressed pages or compressed-only block descriptors may be
+relocated. */
+static
+enum buf_page_state
+buf_LRU_block_remove_hashed_page(
+/*=============================*/
+				/* out: the new state of the block
+				(BUF_BLOCK_ZIP_FREE if the state was
+				BUF_BLOCK_ZIP_PAGE, or BUF_BLOCK_REMOVE_HASH
+				otherwise) */
+	buf_page_t*	bpage,	/* in: block, must contain a file page and
+				be in a state where it can be freed; there
+				may or may not be a hash index to the page */
+	ibool		zip)	/* in: TRUE if should remove also the
+				compressed page of an uncompressed page */
+{
+	const buf_page_t*	hashed_bpage;
+	ut_ad(bpage);
+	ut_ad(buf_pool_mutex_own());
+	ut_ad(mutex_own(buf_page_get_mutex(bpage)));
+
+	ut_a(buf_page_get_io_fix(bpage) == BUF_IO_NONE);
+	ut_a(bpage->buf_fix_count == 0);
+
+	UNIV_MEM_ASSERT_RW(bpage, sizeof *bpage);
+
+	buf_LRU_remove_block(bpage);
+
+	buf_pool->freed_page_clock += 1;
+
+	switch (buf_page_get_state(bpage)) {
+	case BUF_BLOCK_FILE_PAGE:
+		UNIV_MEM_ASSERT_W(bpage, sizeof(buf_block_t));
+		UNIV_MEM_ASSERT_W(((buf_block_t*) bpage)->frame,
+				  UNIV_PAGE_SIZE);
+		buf_block_modify_clock_inc((buf_block_t*) bpage);
+		if (bpage->zip.data) {
+			const page_t*	page = ((buf_block_t*) bpage)->frame;
+			const ulint	zip_size
+				= page_zip_get_size(&bpage->zip);
+
+			ut_a(!zip || bpage->oldest_modification == 0);
+
+			switch (UNIV_EXPECT(fil_page_get_type(page),
+					    FIL_PAGE_INDEX)) {
+			case FIL_PAGE_TYPE_ALLOCATED:
+			case FIL_PAGE_INODE:
+			case FIL_PAGE_IBUF_BITMAP:
+			case FIL_PAGE_TYPE_FSP_HDR:
+			case FIL_PAGE_TYPE_XDES:
+				/* These are essentially uncompressed pages. */
+				if (!zip) {
+					/* InnoDB writes the data to the
+					uncompressed page frame.  Copy it
+					to the compressed page, which will
+					be preserved. */
+					memcpy(bpage->zip.data, page,
+					       zip_size);
+				}
+				break;
+			case FIL_PAGE_TYPE_ZBLOB:
+			case FIL_PAGE_TYPE_ZBLOB2:
+				break;
+			case FIL_PAGE_INDEX:
+#ifdef UNIV_ZIP_DEBUG
+				ut_a(page_zip_validate(&bpage->zip, page));
+#endif /* UNIV_ZIP_DEBUG */
+				break;
+			default:
+				ut_print_timestamp(stderr);
+				fputs("  InnoDB: ERROR: The compressed page"
+				      " to be evicted seems corrupt:", stderr);
+				ut_print_buf(stderr, page, zip_size);
+				fputs("\nInnoDB: Possibly older version"
+				      " of the page:", stderr);
+				ut_print_buf(stderr, bpage->zip.data,
+					     zip_size);
+				putc('\n', stderr);
+				ut_error;
+			}
+
+			break;
+		}
+		/* fall through */
+	case BUF_BLOCK_ZIP_PAGE:
+		ut_a(bpage->oldest_modification == 0);
+		UNIV_MEM_ASSERT_W(bpage->zip.data,
+				  page_zip_get_size(&bpage->zip));
+		break;
+	case BUF_BLOCK_ZIP_FREE:
+	case BUF_BLOCK_ZIP_DIRTY:
+	case BUF_BLOCK_NOT_USED:
+	case BUF_BLOCK_READY_FOR_USE:
+	case BUF_BLOCK_MEMORY:
+	case BUF_BLOCK_REMOVE_HASH:
+		ut_error;
+		break;
+	}
+
+	hashed_bpage = buf_page_hash_get(bpage->space, bpage->offset);
+
+	if (UNIV_UNLIKELY(bpage != hashed_bpage)) {
+		fprintf(stderr,
+			"InnoDB: Error: page %lu %lu not found"
+			" in the hash table\n",
+			(ulong) bpage->space,
+			(ulong) bpage->offset);
+		if (hashed_bpage) {
+			fprintf(stderr,
+				"InnoDB: In hash table we find block"
+				" %p of %lu %lu which is not %p\n",
+				(const void*) hashed_bpage,
+				(ulong) hashed_bpage->space,
+				(ulong) hashed_bpage->offset,
+				(const void*) bpage);
+		}
+
+#if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
+		mutex_exit(buf_page_get_mutex(bpage));
+		buf_pool_mutex_exit();
+		buf_print();
+		buf_LRU_print();
+		buf_validate();
+		buf_LRU_validate();
+#endif /* UNIV_DEBUG || UNIV_BUF_DEBUG */
+		ut_error;
+	}
+
+	ut_ad(!bpage->in_zip_hash);
+	ut_ad(bpage->in_page_hash);
+	ut_d(bpage->in_page_hash = FALSE);
+	HASH_DELETE(buf_page_t, hash, buf_pool->page_hash,
+		    buf_page_address_fold(bpage->space, bpage->offset),
+		    bpage);
+	switch (buf_page_get_state(bpage)) {
+	case BUF_BLOCK_ZIP_PAGE:
+		ut_ad(!bpage->in_free_list);
+		ut_ad(!bpage->in_flush_list);
+		ut_ad(!bpage->in_LRU_list);
+		ut_a(bpage->zip.data);
+		ut_a(buf_page_get_zip_size(bpage));
+
+		UT_LIST_REMOVE(list, buf_pool->zip_clean, bpage);
+
+		mutex_exit(&buf_pool_zip_mutex);
+		buf_pool_mutex_exit_forbid();
+		buf_buddy_free(bpage->zip.data,
+			       page_zip_get_size(&bpage->zip));
+		buf_buddy_free(bpage, sizeof(*bpage));
+		buf_pool_mutex_exit_allow();
+		UNIV_MEM_UNDESC(bpage);
+		return(BUF_BLOCK_ZIP_FREE);
+
+	case BUF_BLOCK_FILE_PAGE:
+		memset(((buf_block_t*) bpage)->frame
+		       + FIL_PAGE_OFFSET, 0xff, 4);
+		memset(((buf_block_t*) bpage)->frame
+		       + FIL_PAGE_ARCH_LOG_NO_OR_SPACE_ID, 0xff, 4);
+		UNIV_MEM_INVALID(((buf_block_t*) bpage)->frame,
+				 UNIV_PAGE_SIZE);
+		buf_page_set_state(bpage, BUF_BLOCK_REMOVE_HASH);
+
+		if (zip && bpage->zip.data) {
+			/* Free the compressed page. */
+			void*	data = bpage->zip.data;
+			bpage->zip.data = NULL;
+
+			mutex_exit(&((buf_block_t*) bpage)->mutex);
+			buf_pool_mutex_exit_forbid();
+			buf_buddy_free(data, page_zip_get_size(&bpage->zip));
+			buf_pool_mutex_exit_allow();
+			mutex_enter(&((buf_block_t*) bpage)->mutex);
+			page_zip_set_size(&bpage->zip, 0);
+		}
+
+		return(BUF_BLOCK_REMOVE_HASH);
+
+	case BUF_BLOCK_ZIP_FREE:
+	case BUF_BLOCK_ZIP_DIRTY:
+	case BUF_BLOCK_NOT_USED:
+	case BUF_BLOCK_READY_FOR_USE:
+	case BUF_BLOCK_MEMORY:
+	case BUF_BLOCK_REMOVE_HASH:
+		break;
+	}
+
+	ut_error;
+	return(BUF_BLOCK_ZIP_FREE);
+}
+
+/**********************************************************************
+Puts a file page whose has no hash index to the free list. */
+static
+void
+buf_LRU_block_free_hashed_page(
+/*===========================*/
+	buf_block_t*	block)	/* in: block, must contain a file page and
+				be in a state where it can be freed */
+{
+	ut_ad(buf_pool_mutex_own());
+	ut_ad(mutex_own(&block->mutex));
+
+	buf_block_set_state(block, BUF_BLOCK_MEMORY);
+
+	buf_LRU_block_free_non_file_page(block);
+}
+
+/************************************************************************
+Update the historical stats that we are collecting for LRU eviction
+policy at the end of each interval. */
+UNIV_INTERN
+void
+buf_LRU_stat_update(void)
+/*=====================*/
+{
+	buf_LRU_stat_t*	item;
+
+	/* If we haven't started eviction yet then don't update stats. */
+	if (buf_pool->freed_page_clock == 0) {
+		goto func_exit;
+	}
+
+	buf_pool_mutex_enter();
+
+	/* Update the index. */
+	item = &buf_LRU_stat_arr[buf_LRU_stat_arr_ind];
+	buf_LRU_stat_arr_ind++;
+	buf_LRU_stat_arr_ind %= BUF_LRU_STAT_N_INTERVAL;
+
+	/* Add the current value and subtract the obsolete entry. */
+	buf_LRU_stat_sum.io += buf_LRU_stat_cur.io - item->io;
+	buf_LRU_stat_sum.unzip += buf_LRU_stat_cur.unzip - item->unzip;
+
+	/* Put current entry in the array. */
+	memcpy(item, &buf_LRU_stat_cur, sizeof *item);
+
+	buf_pool_mutex_exit();
+
+func_exit:
+	/* Clear the current entry. */
+	memset(&buf_LRU_stat_cur, 0, sizeof buf_LRU_stat_cur);
+}
+
+#if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
+/**************************************************************************
+Validates the LRU list. */
+UNIV_INTERN
+ibool
+buf_LRU_validate(void)
+/*==================*/
+{
+	buf_page_t*	bpage;
+	buf_block_t*	block;
+	ulint		old_len;
+	ulint		new_len;
+	ulint		LRU_pos;
+
+	ut_ad(buf_pool);
+	buf_pool_mutex_enter();
+
+	if (UT_LIST_GET_LEN(buf_pool->LRU) >= BUF_LRU_OLD_MIN_LEN) {
+
+		ut_a(buf_pool->LRU_old);
+		old_len = buf_pool->LRU_old_len;
+		new_len = 3 * (UT_LIST_GET_LEN(buf_pool->LRU) / 8);
+		ut_a(old_len >= new_len - BUF_LRU_OLD_TOLERANCE);
+		ut_a(old_len <= new_len + BUF_LRU_OLD_TOLERANCE);
+	}
+
+	UT_LIST_VALIDATE(LRU, buf_page_t, buf_pool->LRU);
+
+	bpage = UT_LIST_GET_FIRST(buf_pool->LRU);
+
+	old_len = 0;
+
+	while (bpage != NULL) {
+
+		switch (buf_page_get_state(bpage)) {
+		case BUF_BLOCK_ZIP_FREE:
+		case BUF_BLOCK_NOT_USED:
+		case BUF_BLOCK_READY_FOR_USE:
+		case BUF_BLOCK_MEMORY:
+		case BUF_BLOCK_REMOVE_HASH:
+			ut_error;
+			break;
+		case BUF_BLOCK_FILE_PAGE:
+			ut_ad(((buf_block_t*) bpage)->in_unzip_LRU_list
+			      == buf_page_belongs_to_unzip_LRU(bpage));
+		case BUF_BLOCK_ZIP_PAGE:
+		case BUF_BLOCK_ZIP_DIRTY:
+			break;
+		}
+
+		if (buf_page_is_old(bpage)) {
+			old_len++;
+		}
+
+		if (buf_pool->LRU_old && (old_len == 1)) {
+			ut_a(buf_pool->LRU_old == bpage);
+		}
+
+		LRU_pos	= buf_page_get_LRU_position(bpage);
+
+		bpage = UT_LIST_GET_NEXT(LRU, bpage);
+
+		if (bpage) {
+			/* If the following assert fails, it may
+			not be an error: just the buf_pool clock
+			has wrapped around */
+			ut_a(LRU_pos >= buf_page_get_LRU_position(bpage));
+		}
+	}
+
+	if (buf_pool->LRU_old) {
+		ut_a(buf_pool->LRU_old_len == old_len);
+	}
+
+	UT_LIST_VALIDATE(list, buf_page_t, buf_pool->free);
+
+	for (bpage = UT_LIST_GET_FIRST(buf_pool->free);
+	     bpage != NULL;
+	     bpage = UT_LIST_GET_NEXT(list, bpage)) {
+
+		ut_a(buf_page_get_state(bpage) == BUF_BLOCK_NOT_USED);
+	}
+
+	UT_LIST_VALIDATE(unzip_LRU, buf_block_t, buf_pool->unzip_LRU);
+
+	for (block = UT_LIST_GET_FIRST(buf_pool->unzip_LRU);
+	     block;
+	     block = UT_LIST_GET_NEXT(unzip_LRU, block)) {
+
+		ut_ad(block->in_unzip_LRU_list);
+		ut_ad(block->page.in_LRU_list);
+		ut_a(buf_page_belongs_to_unzip_LRU(&block->page));
+	}
+
+	buf_pool_mutex_exit();
+	return(TRUE);
+}
+#endif /* UNIV_DEBUG || UNIV_BUF_DEBUG */
+
+#if defined UNIV_DEBUG_PRINT || defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
+/**************************************************************************
+Prints the LRU list. */
+UNIV_INTERN
+void
+buf_LRU_print(void)
+/*===============*/
+{
+	const buf_page_t*	bpage;
+
+	ut_ad(buf_pool);
+	buf_pool_mutex_enter();
+
+	fprintf(stderr, "Pool ulint clock %lu\n",
+		(ulong) buf_pool->ulint_clock);
+
+	bpage = UT_LIST_GET_FIRST(buf_pool->LRU);
+
+	while (bpage != NULL) {
+
+		fprintf(stderr, "BLOCK space %lu page %lu ",
+			(ulong) buf_page_get_space(bpage),
+			(ulong) buf_page_get_page_no(bpage));
+
+		if (buf_page_is_old(bpage)) {
+			fputs("old ", stderr);
+		}
+
+		if (bpage->buf_fix_count) {
+			fprintf(stderr, "buffix count %lu ",
+				(ulong) bpage->buf_fix_count);
+		}
+
+		if (buf_page_get_io_fix(bpage)) {
+			fprintf(stderr, "io_fix %lu ",
+				(ulong) buf_page_get_io_fix(bpage));
+		}
+
+		if (bpage->oldest_modification) {
+			fputs("modif. ", stderr);
+		}
+
+		switch (buf_page_get_state(bpage)) {
+			const byte*	frame;
+		case BUF_BLOCK_FILE_PAGE:
+			frame = buf_block_get_frame((buf_block_t*) bpage);
+			fprintf(stderr, "\nLRU pos %lu type %lu"
+				" index id %lu\n",
+				(ulong) buf_page_get_LRU_position(bpage),
+				(ulong) fil_page_get_type(frame),
+				(ulong) ut_dulint_get_low(
+					btr_page_get_index_id(frame)));
+			break;
+		case BUF_BLOCK_ZIP_PAGE:
+			frame = bpage->zip.data;
+			fprintf(stderr, "\nLRU pos %lu type %lu size %lu"
+				" index id %lu\n",
+				(ulong) buf_page_get_LRU_position(bpage),
+				(ulong) fil_page_get_type(frame),
+				(ulong) buf_page_get_zip_size(bpage),
+				(ulong) ut_dulint_get_low(
+					btr_page_get_index_id(frame)));
+			break;
+
+		default:
+			fprintf(stderr, "\nLRU pos %lu !state %lu!\n",
+				(ulong) buf_page_get_LRU_position(bpage),
+				(ulong) buf_page_get_state(bpage));
+			break;
+		}
+
+		bpage = UT_LIST_GET_NEXT(LRU, bpage);
+	}
+
+	buf_pool_mutex_exit();
+}
+#endif /* UNIV_DEBUG_PRINT || UNIV_DEBUG || UNIV_BUF_DEBUG */
diff --git a/storage/xtradb/buf/buf0rea.c b/storage/xtradb/buf/buf0rea.c
new file mode 100644
index 00000000000..149be7f3bdd
--- /dev/null
+++ b/storage/xtradb/buf/buf0rea.c
@@ -0,0 +1,818 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The database buffer read
+
+Created 11/5/1995 Heikki Tuuri
+*******************************************************/
+
+#include "buf0rea.h"
+
+#include "fil0fil.h"
+#include "mtr0mtr.h"
+
+#include "buf0buf.h"
+#include "buf0flu.h"
+#include "buf0lru.h"
+#include "ibuf0ibuf.h"
+#include "log0recv.h"
+#include "trx0sys.h"
+#include "os0file.h"
+#include "srv0start.h"
+
+extern ulint srv_read_ahead;
+extern ulint srv_read_ahead_rnd;
+extern ulint srv_read_ahead_seq;
+extern ulint srv_buf_pool_reads;
+
+/* The size in blocks of the area where the random read-ahead algorithm counts
+the accessed pages when deciding whether to read-ahead */
+#define	BUF_READ_AHEAD_RANDOM_AREA	BUF_READ_AHEAD_AREA
+
+/* There must be at least this many pages in buf_pool in the area to start
+a random read-ahead */
+#define BUF_READ_AHEAD_RANDOM_THRESHOLD	(5 + buf_read_ahead_random_area / 8)
+
+/* The linear read-ahead area size */
+#define	BUF_READ_AHEAD_LINEAR_AREA	BUF_READ_AHEAD_AREA
+
+/* The linear read-ahead threshold */
+#define LINEAR_AREA_THRESHOLD_COEF	5 / 8
+
+/* If there are buf_pool->curr_size per the number below pending reads, then
+read-ahead is not done: this is to prevent flooding the buffer pool with
+i/o-fixed buffer blocks */
+#define BUF_READ_AHEAD_PEND_LIMIT	2
+
+/************************************************************************
+Low-level function which reads a page asynchronously from a file to the
+buffer buf_pool if it is not already there, in which case does nothing.
+Sets the io_fix flag and sets an exclusive lock on the buffer frame. The
+flag is cleared and the x-lock released by an i/o-handler thread. */
+static
+ulint
+buf_read_page_low(
+/*==============*/
+			/* out: 1 if a read request was queued, 0 if the page
+			already resided in buf_pool, or if the page is in
+			the doublewrite buffer blocks in which case it is never
+			read into the pool, or if the tablespace does not
+			exist or is being dropped */
+	ulint*	err,	/* out: DB_SUCCESS or DB_TABLESPACE_DELETED if we are
+			trying to read from a non-existent tablespace, or a
+			tablespace which is just now being dropped */
+	ibool	sync,	/* in: TRUE if synchronous aio is desired */
+	ulint	mode,	/* in: BUF_READ_IBUF_PAGES_ONLY, ...,
+			ORed to OS_AIO_SIMULATED_WAKE_LATER (see below
+			at read-ahead functions) */
+	ulint	space,	/* in: space id */
+	ulint	zip_size,/* in: compressed page size, or 0 */
+	ibool	unzip,	/* in: TRUE=request uncompressed page */
+	ib_int64_t tablespace_version, /* in: if the space memory object has
+			this timestamp different from what we are giving here,
+			treat the tablespace as dropped; this is a timestamp we
+			use to stop dangling page reads from a tablespace
+			which we have DISCARDed + IMPORTed back */
+	ulint	offset)	/* in: page number */
+{
+	buf_page_t*	bpage;
+	ulint		wake_later;
+
+	*err = DB_SUCCESS;
+
+	wake_later = mode & OS_AIO_SIMULATED_WAKE_LATER;
+	mode = mode & ~OS_AIO_SIMULATED_WAKE_LATER;
+
+	if (trx_doublewrite && space == TRX_SYS_SPACE
+	    && (   (offset >= trx_doublewrite->block1
+		    && offset < trx_doublewrite->block1
+		    + TRX_SYS_DOUBLEWRITE_BLOCK_SIZE)
+		   || (offset >= trx_doublewrite->block2
+		       && offset < trx_doublewrite->block2
+		       + TRX_SYS_DOUBLEWRITE_BLOCK_SIZE))) {
+		ut_print_timestamp(stderr);
+		fprintf(stderr,
+			"  InnoDB: Warning: trying to read"
+			" doublewrite buffer page %lu\n",
+			(ulong) offset);
+
+		return(0);
+	}
+
+	if (ibuf_bitmap_page(zip_size, offset)
+	    || trx_sys_hdr_page(space, offset)) {
+
+		/* Trx sys header is so low in the latching order that we play
+		safe and do not leave the i/o-completion to an asynchronous
+		i/o-thread. Ibuf bitmap pages must always be read with
+		syncronous i/o, to make sure they do not get involved in
+		thread deadlocks. */
+
+		sync = TRUE;
+	}
+
+	/* The following call will also check if the tablespace does not exist
+	or is being dropped; if we succeed in initing the page in the buffer
+	pool for read, then DISCARD cannot proceed until the read has
+	completed */
+	bpage = buf_page_init_for_read(err, mode, space, zip_size, unzip,
+				       tablespace_version, offset);
+	if (bpage == NULL) {
+
+		return(0);
+	}
+
+#ifdef UNIV_DEBUG
+	if (buf_debug_prints) {
+		fprintf(stderr,
+			"Posting read request for page %lu, sync %lu\n",
+			(ulong) offset,
+			(ulong) sync);
+	}
+#endif
+
+	ut_ad(buf_page_in_file(bpage));
+
+	if (zip_size) {
+		*err = fil_io(OS_FILE_READ | wake_later,
+			      sync, space, zip_size, offset, 0, zip_size,
+			      bpage->zip.data, bpage);
+	} else {
+		ut_a(buf_page_get_state(bpage) == BUF_BLOCK_FILE_PAGE);
+
+		*err = fil_io(OS_FILE_READ | wake_later,
+			      sync, space, 0, offset, 0, UNIV_PAGE_SIZE,
+			      ((buf_block_t*) bpage)->frame, bpage);
+	}
+	ut_a(*err == DB_SUCCESS);
+
+	if (sync) {
+		/* The i/o is already completed when we arrive from
+		fil_read */
+		buf_page_io_complete(bpage);
+	}
+
+	return(1);
+}
+
+/************************************************************************
+Applies a random read-ahead in buf_pool if there are at least a threshold
+value of accessed pages from the random read-ahead area. Does not read any
+page, not even the one at the position (space, offset), if the read-ahead
+mechanism is not activated. NOTE 1: the calling thread may own latches on
+pages: to avoid deadlocks this function must be written such that it cannot
+end up waiting for these latches! NOTE 2: the calling thread must want
+access to the page given: this rule is set to prevent unintended read-aheads
+performed by ibuf routines, a situation which could result in a deadlock if
+the OS does not support asynchronous i/o. */
+static
+ulint
+buf_read_ahead_random(
+/*==================*/
+			/* out: number of page read requests issued; NOTE
+			that if we read ibuf pages, it may happen that
+			the page at the given page number does not get
+			read even if we return a value > 0! */
+	ulint	space,	/* in: space id */
+	ulint	zip_size,/* in: compressed page size in bytes, or 0 */
+	ulint	offset)	/* in: page number of a page which the current thread
+			wants to access */
+{
+	ib_int64_t	tablespace_version;
+	ulint		recent_blocks	= 0;
+	ulint		count;
+	ulint		LRU_recent_limit;
+	ulint		ibuf_mode;
+	ulint		low, high;
+	ulint		err;
+	ulint		i;
+	ulint		buf_read_ahead_random_area;
+
+	if (!(srv_read_ahead & 1)) {
+		return(0);
+	}
+
+	if (srv_startup_is_before_trx_rollback_phase) {
+		/* No read-ahead to avoid thread deadlocks */
+		return(0);
+	}
+
+	if (ibuf_bitmap_page(zip_size, offset)
+	    || trx_sys_hdr_page(space, offset)) {
+
+		/* If it is an ibuf bitmap page or trx sys hdr, we do
+		no read-ahead, as that could break the ibuf page access
+		order */
+
+		return(0);
+	}
+
+	/* Remember the tablespace version before we ask te tablespace size
+	below: if DISCARD + IMPORT changes the actual .ibd file meanwhile, we
+	do not try to read outside the bounds of the tablespace! */
+
+	tablespace_version = fil_space_get_version(space);
+
+	buf_read_ahead_random_area = BUF_READ_AHEAD_RANDOM_AREA;
+
+	low  = (offset / buf_read_ahead_random_area)
+		* buf_read_ahead_random_area;
+	high = (offset / buf_read_ahead_random_area + 1)
+		* buf_read_ahead_random_area;
+	if (high > fil_space_get_size(space)) {
+
+		high = fil_space_get_size(space);
+	}
+
+	/* Get the minimum LRU_position field value for an initial segment
+	of the LRU list, to determine which blocks have recently been added
+	to the start of the list. */
+
+	LRU_recent_limit = buf_LRU_get_recent_limit();
+
+	buf_pool_mutex_enter();
+
+	if (buf_pool->n_pend_reads
+	    > buf_pool->curr_size / BUF_READ_AHEAD_PEND_LIMIT) {
+		buf_pool_mutex_exit();
+
+		return(0);
+	}
+
+	/* Count how many blocks in the area have been recently accessed,
+	that is, reside near the start of the LRU list. */
+
+	for (i = low; i < high; i++) {
+		const buf_page_t*	bpage = buf_page_hash_get(space, i);
+
+		if (bpage
+		    && buf_page_is_accessed(bpage)
+		    && (buf_page_get_LRU_position(bpage) > LRU_recent_limit)) {
+
+			recent_blocks++;
+
+			if (recent_blocks >= BUF_READ_AHEAD_RANDOM_THRESHOLD) {
+
+				buf_pool_mutex_exit();
+				goto read_ahead;
+			}
+		}
+	}
+
+	buf_pool_mutex_exit();
+	/* Do nothing */
+	return(0);
+
+read_ahead:
+	/* Read all the suitable blocks within the area */
+
+	if (ibuf_inside()) {
+		ibuf_mode = BUF_READ_IBUF_PAGES_ONLY;
+	} else {
+		ibuf_mode = BUF_READ_ANY_PAGE;
+	}
+
+	count = 0;
+
+	for (i = low; i < high; i++) {
+		/* It is only sensible to do read-ahead in the non-sync aio
+		mode: hence FALSE as the first parameter */
+
+		if (!ibuf_bitmap_page(zip_size, i)) {
+			count += buf_read_page_low(
+				&err, FALSE,
+				ibuf_mode | OS_AIO_SIMULATED_WAKE_LATER,
+				space, zip_size, FALSE,
+				tablespace_version, i);
+			if (err == DB_TABLESPACE_DELETED) {
+				ut_print_timestamp(stderr);
+				fprintf(stderr,
+					"  InnoDB: Warning: in random"
+					" readahead trying to access\n"
+					"InnoDB: tablespace %lu page %lu,\n"
+					"InnoDB: but the tablespace does not"
+					" exist or is just being dropped.\n",
+					(ulong) space, (ulong) i);
+			}
+		}
+	}
+
+	/* In simulated aio we wake the aio handler threads only after
+	queuing all aio requests, in native aio the following call does
+	nothing: */
+
+	os_aio_simulated_wake_handler_threads();
+
+#ifdef UNIV_DEBUG
+	if (buf_debug_prints && (count > 0)) {
+		fprintf(stderr,
+			"Random read-ahead space %lu offset %lu pages %lu\n",
+			(ulong) space, (ulong) offset,
+			(ulong) count);
+	}
+#endif /* UNIV_DEBUG */
+
+	++srv_read_ahead_rnd;
+	return(count);
+}
+
+/************************************************************************
+High-level function which reads a page asynchronously from a file to the
+buffer buf_pool if it is not already there. Sets the io_fix flag and sets
+an exclusive lock on the buffer frame. The flag is cleared and the x-lock
+released by the i/o-handler thread. Does a random read-ahead if it seems
+sensible. */
+UNIV_INTERN
+ulint
+buf_read_page(
+/*==========*/
+			/* out: number of page read requests issued: this can
+			be > 1 if read-ahead occurred */
+	ulint	space,	/* in: space id */
+	ulint	zip_size,/* in: compressed page size in bytes, or 0 */
+	ulint	offset)	/* in: page number */
+{
+	ib_int64_t	tablespace_version;
+	ulint		count;
+	ulint		count2;
+	ulint		err;
+
+	tablespace_version = fil_space_get_version(space);
+
+	count = buf_read_ahead_random(space, zip_size, offset);
+
+	/* We do the i/o in the synchronous aio mode to save thread
+	switches: hence TRUE */
+
+	count2 = buf_read_page_low(&err, TRUE, BUF_READ_ANY_PAGE, space,
+				   zip_size, FALSE,
+				   tablespace_version, offset);
+	srv_buf_pool_reads+= count2;
+	if (err == DB_TABLESPACE_DELETED) {
+		ut_print_timestamp(stderr);
+		fprintf(stderr,
+			"  InnoDB: Error: trying to access"
+			" tablespace %lu page no. %lu,\n"
+			"InnoDB: but the tablespace does not exist"
+			" or is just being dropped.\n",
+			(ulong) space, (ulong) offset);
+	}
+
+	/* Flush pages from the end of the LRU list if necessary */
+	buf_flush_free_margin(FALSE);
+
+	/* Increment number of I/O operations used for LRU policy. */
+	buf_LRU_stat_inc_io();
+
+	return(count + count2);
+}
+
+/************************************************************************
+Applies linear read-ahead if in the buf_pool the page is a border page of
+a linear read-ahead area and all the pages in the area have been accessed.
+Does not read any page if the read-ahead mechanism is not activated. Note
+that the the algorithm looks at the 'natural' adjacent successor and
+predecessor of the page, which on the leaf level of a B-tree are the next
+and previous page in the chain of leaves. To know these, the page specified
+in (space, offset) must already be present in the buf_pool. Thus, the
+natural way to use this function is to call it when a page in the buf_pool
+is accessed the first time, calling this function just after it has been
+bufferfixed.
+NOTE 1: as this function looks at the natural predecessor and successor
+fields on the page, what happens, if these are not initialized to any
+sensible value? No problem, before applying read-ahead we check that the
+area to read is within the span of the space, if not, read-ahead is not
+applied. An uninitialized value may result in a useless read operation, but
+only very improbably.
+NOTE 2: the calling thread may own latches on pages: to avoid deadlocks this
+function must be written such that it cannot end up waiting for these
+latches!
+NOTE 3: the calling thread must want access to the page given: this rule is
+set to prevent unintended read-aheads performed by ibuf routines, a situation
+which could result in a deadlock if the OS does not support asynchronous io. */
+UNIV_INTERN
+ulint
+buf_read_ahead_linear(
+/*==================*/
+			/* out: number of page read requests issued */
+	ulint	space,	/* in: space id */
+	ulint	zip_size,/* in: compressed page size in bytes, or 0 */
+	ulint	offset)	/* in: page number of a page; NOTE: the current thread
+			must want access to this page (see NOTE 3 above) */
+{
+	ib_int64_t	tablespace_version;
+	buf_page_t*	bpage;
+	buf_frame_t*	frame;
+	buf_page_t*	pred_bpage	= NULL;
+	ulint		pred_offset;
+	ulint		succ_offset;
+	ulint		count;
+	int		asc_or_desc;
+	ulint		new_offset;
+	ulint		fail_count;
+	ulint		ibuf_mode;
+	ulint		low, high;
+	ulint		err;
+	ulint		i;
+	const ulint	buf_read_ahead_linear_area
+		= BUF_READ_AHEAD_LINEAR_AREA;
+
+ 	if (!(srv_read_ahead & 2)) {
+ 		return(0);
+ 	}
+ 
+	if (UNIV_UNLIKELY(srv_startup_is_before_trx_rollback_phase)) {
+		/* No read-ahead to avoid thread deadlocks */
+		return(0);
+	}
+
+	low  = (offset / buf_read_ahead_linear_area)
+		* buf_read_ahead_linear_area;
+	high = (offset / buf_read_ahead_linear_area + 1)
+		* buf_read_ahead_linear_area;
+
+	if ((offset != low) && (offset != high - 1)) {
+		/* This is not a border page of the area: return */
+
+		return(0);
+	}
+
+	if (ibuf_bitmap_page(zip_size, offset)
+	    || trx_sys_hdr_page(space, offset)) {
+
+		/* If it is an ibuf bitmap page or trx sys hdr, we do
+		no read-ahead, as that could break the ibuf page access
+		order */
+
+		return(0);
+	}
+
+	/* Remember the tablespace version before we ask te tablespace size
+	below: if DISCARD + IMPORT changes the actual .ibd file meanwhile, we
+	do not try to read outside the bounds of the tablespace! */
+
+	tablespace_version = fil_space_get_version(space);
+
+	buf_pool_mutex_enter();
+
+	if (high > fil_space_get_size(space)) {
+		buf_pool_mutex_exit();
+		/* The area is not whole, return */
+
+		return(0);
+	}
+
+	if (buf_pool->n_pend_reads
+	    > buf_pool->curr_size / BUF_READ_AHEAD_PEND_LIMIT) {
+		buf_pool_mutex_exit();
+
+		return(0);
+	}
+
+	/* Check that almost all pages in the area have been accessed; if
+	offset == low, the accesses must be in a descending order, otherwise,
+	in an ascending order. */
+
+	asc_or_desc = 1;
+
+	if (offset == low) {
+		asc_or_desc = -1;
+	}
+
+	fail_count = 0;
+
+	for (i = low; i < high; i++) {
+		bpage = buf_page_hash_get(space, i);
+
+		if ((bpage == NULL) || !buf_page_is_accessed(bpage)) {
+			/* Not accessed */
+			fail_count++;
+
+		} else if (pred_bpage
+			   && (ut_ulint_cmp(
+				       buf_page_get_LRU_position(bpage),
+				       buf_page_get_LRU_position(pred_bpage))
+			       != asc_or_desc)) {
+			/* Accesses not in the right order */
+
+			fail_count++;
+			pred_bpage = bpage;
+		}
+	}
+
+	if (fail_count > buf_read_ahead_linear_area
+	    * LINEAR_AREA_THRESHOLD_COEF) {
+		/* Too many failures: return */
+
+		buf_pool_mutex_exit();
+
+		return(0);
+	}
+
+	/* If we got this far, we know that enough pages in the area have
+	been accessed in the right order: linear read-ahead can be sensible */
+
+	bpage = buf_page_hash_get(space, offset);
+
+	if (bpage == NULL) {
+		buf_pool_mutex_exit();
+
+		return(0);
+	}
+
+	switch (buf_page_get_state(bpage)) {
+	case BUF_BLOCK_ZIP_PAGE:
+		frame = bpage->zip.data;
+		break;
+	case BUF_BLOCK_FILE_PAGE:
+		frame = ((buf_block_t*) bpage)->frame;
+		break;
+	default:
+		ut_error;
+		break;
+	}
+
+	/* Read the natural predecessor and successor page addresses from
+	the page; NOTE that because the calling thread may have an x-latch
+	on the page, we do not acquire an s-latch on the page, this is to
+	prevent deadlocks. Even if we read values which are nonsense, the
+	algorithm will work. */
+
+	pred_offset = fil_page_get_prev(frame);
+	succ_offset = fil_page_get_next(frame);
+
+	buf_pool_mutex_exit();
+
+	if ((offset == low) && (succ_offset == offset + 1)) {
+
+		/* This is ok, we can continue */
+		new_offset = pred_offset;
+
+	} else if ((offset == high - 1) && (pred_offset == offset - 1)) {
+
+		/* This is ok, we can continue */
+		new_offset = succ_offset;
+	} else {
+		/* Successor or predecessor not in the right order */
+
+		return(0);
+	}
+
+	low  = (new_offset / buf_read_ahead_linear_area)
+		* buf_read_ahead_linear_area;
+	high = (new_offset / buf_read_ahead_linear_area + 1)
+		* buf_read_ahead_linear_area;
+
+	if ((new_offset != low) && (new_offset != high - 1)) {
+		/* This is not a border page of the area: return */
+
+		return(0);
+	}
+
+	if (high > fil_space_get_size(space)) {
+		/* The area is not whole, return */
+
+		return(0);
+	}
+
+	/* If we got this far, read-ahead can be sensible: do it */
+
+	if (ibuf_inside()) {
+		ibuf_mode = BUF_READ_IBUF_PAGES_ONLY;
+	} else {
+		ibuf_mode = BUF_READ_ANY_PAGE;
+	}
+
+	count = 0;
+
+	/* Since Windows XP seems to schedule the i/o handler thread
+	very eagerly, and consequently it does not wait for the
+	full read batch to be posted, we use special heuristics here */
+
+	os_aio_simulated_put_read_threads_to_sleep();
+
+	for (i = low; i < high; i++) {
+		/* It is only sensible to do read-ahead in the non-sync
+		aio mode: hence FALSE as the first parameter */
+
+		if (!ibuf_bitmap_page(zip_size, i)) {
+			count += buf_read_page_low(
+				&err, FALSE,
+				ibuf_mode | OS_AIO_SIMULATED_WAKE_LATER,
+				space, zip_size, FALSE, tablespace_version, i);
+			if (err == DB_TABLESPACE_DELETED) {
+				ut_print_timestamp(stderr);
+				fprintf(stderr,
+					"  InnoDB: Warning: in"
+					" linear readahead trying to access\n"
+					"InnoDB: tablespace %lu page %lu,\n"
+					"InnoDB: but the tablespace does not"
+					" exist or is just being dropped.\n",
+					(ulong) space, (ulong) i);
+			}
+		}
+	}
+
+	/* In simulated aio we wake the aio handler threads only after
+	queuing all aio requests, in native aio the following call does
+	nothing: */
+
+	os_aio_simulated_wake_handler_threads();
+
+	/* Flush pages from the end of the LRU list if necessary */
+	buf_flush_free_margin(FALSE);
+
+#ifdef UNIV_DEBUG
+	if (buf_debug_prints && (count > 0)) {
+		fprintf(stderr,
+			"LINEAR read-ahead space %lu offset %lu pages %lu\n",
+			(ulong) space, (ulong) offset, (ulong) count);
+	}
+#endif /* UNIV_DEBUG */
+
+	/* Read ahead is considered one I/O operation for the purpose of
+	LRU policy decision. */
+	buf_LRU_stat_inc_io();
+
+	++srv_read_ahead_seq;
+	return(count);
+}
+
+/************************************************************************
+Issues read requests for pages which the ibuf module wants to read in, in
+order to contract the insert buffer tree. Technically, this function is like
+a read-ahead function. */
+UNIV_INTERN
+void
+buf_read_ibuf_merge_pages(
+/*======================*/
+	ibool		sync,		/* in: TRUE if the caller
+					wants this function to wait
+					for the highest address page
+					to get read in, before this
+					function returns */
+	const ulint*	space_ids,	/* in: array of space ids */
+	const ib_int64_t* space_versions,/* in: the spaces must have
+					this version number
+					(timestamp), otherwise we
+					discard the read; we use this
+					to cancel reads if DISCARD +
+					IMPORT may have changed the
+					tablespace size */
+	const ulint*	page_nos,	/* in: array of page numbers
+					to read, with the highest page
+					number the last in the
+					array */
+	ulint		n_stored)	/* in: number of elements
+					in the arrays */
+{
+	ulint	i;
+
+	ut_ad(!ibuf_inside());
+#ifdef UNIV_IBUF_DEBUG
+	ut_a(n_stored < UNIV_PAGE_SIZE);
+#endif
+	while (buf_pool->n_pend_reads
+	       > buf_pool->curr_size / BUF_READ_AHEAD_PEND_LIMIT) {
+		os_thread_sleep(500000);
+	}
+
+	for (i = 0; i < n_stored; i++) {
+		ulint	zip_size = fil_space_get_zip_size(space_ids[i]);
+		ulint	err;
+
+		if (UNIV_UNLIKELY(zip_size == ULINT_UNDEFINED)) {
+
+			goto tablespace_deleted;
+		}
+
+		buf_read_page_low(&err, sync && (i + 1 == n_stored),
+				  BUF_READ_ANY_PAGE, space_ids[i],
+				  zip_size, TRUE, space_versions[i],
+				  page_nos[i]);
+
+		if (UNIV_UNLIKELY(err == DB_TABLESPACE_DELETED)) {
+tablespace_deleted:
+			/* We have deleted or are deleting the single-table
+			tablespace: remove the entries for that page */
+
+			ibuf_merge_or_delete_for_page(NULL, space_ids[i],
+						      page_nos[i],
+						      zip_size, FALSE);
+		}
+	}
+
+	os_aio_simulated_wake_handler_threads();
+
+	/* Flush pages from the end of the LRU list if necessary */
+	buf_flush_free_margin(FALSE);
+
+#ifdef UNIV_DEBUG
+	if (buf_debug_prints) {
+		fprintf(stderr,
+			"Ibuf merge read-ahead space %lu pages %lu\n",
+			(ulong) space_ids[0], (ulong) n_stored);
+	}
+#endif /* UNIV_DEBUG */
+}
+
+/************************************************************************
+Issues read requests for pages which recovery wants to read in. */
+UNIV_INTERN
+void
+buf_read_recv_pages(
+/*================*/
+	ibool		sync,		/* in: TRUE if the caller
+					wants this function to wait
+					for the highest address page
+					to get read in, before this
+					function returns */
+	ulint		space,		/* in: space id */
+	ulint		zip_size,	/* in: compressed page size in
+					bytes, or 0 */
+	const ulint*	page_nos,	/* in: array of page numbers
+					to read, with the highest page
+					number the last in the
+					array */
+	ulint		n_stored)	/* in: number of page numbers
+					in the array */
+{
+	ib_int64_t	tablespace_version;
+	ulint		count;
+	ulint		err;
+	ulint		i;
+
+	zip_size = fil_space_get_zip_size(space);
+	tablespace_version = fil_space_get_version(space);
+
+	for (i = 0; i < n_stored; i++) {
+
+		count = 0;
+
+		os_aio_print_debug = FALSE;
+
+		while (buf_pool->n_pend_reads >= recv_n_pool_free_frames / 2) {
+
+			os_aio_simulated_wake_handler_threads();
+			os_thread_sleep(500000);
+
+			count++;
+
+			if (count > 100) {
+				fprintf(stderr,
+					"InnoDB: Error: InnoDB has waited for"
+					" 50 seconds for pending\n"
+					"InnoDB: reads to the buffer pool to"
+					" be finished.\n"
+					"InnoDB: Number of pending reads %lu,"
+					" pending pread calls %lu\n",
+					(ulong) buf_pool->n_pend_reads,
+					(ulong)os_file_n_pending_preads);
+
+				os_aio_print_debug = TRUE;
+			}
+		}
+
+		os_aio_print_debug = FALSE;
+
+		if ((i + 1 == n_stored) && sync) {
+			buf_read_page_low(&err, TRUE, BUF_READ_ANY_PAGE, space,
+					  zip_size, TRUE, tablespace_version,
+					  page_nos[i]);
+		} else {
+			buf_read_page_low(&err, FALSE, BUF_READ_ANY_PAGE
+					  | OS_AIO_SIMULATED_WAKE_LATER,
+					  space, zip_size, TRUE,
+					  tablespace_version, page_nos[i]);
+		}
+	}
+
+	os_aio_simulated_wake_handler_threads();
+
+	/* Flush pages from the end of the LRU list if necessary */
+	buf_flush_free_margin(FALSE);
+
+#ifdef UNIV_DEBUG
+	if (buf_debug_prints) {
+		fprintf(stderr,
+			"Recovery applies read-ahead pages %lu\n",
+			(ulong) n_stored);
+	}
+#endif /* UNIV_DEBUG */
+}
diff --git a/storage/xtradb/data/data0data.c b/storage/xtradb/data/data0data.c
new file mode 100644
index 00000000000..1cb3803b187
--- /dev/null
+++ b/storage/xtradb/data/data0data.c
@@ -0,0 +1,758 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/************************************************************************
+SQL data field and tuple
+
+Created 5/30/1994 Heikki Tuuri
+*************************************************************************/
+
+#include "data0data.h"
+
+#ifdef UNIV_NONINL
+#include "data0data.ic"
+#endif
+
+#include "rem0rec.h"
+#include "rem0cmp.h"
+#include "page0page.h"
+#include "page0zip.h"
+#include "dict0dict.h"
+#include "btr0cur.h"
+
+#include <ctype.h>
+
+#ifdef UNIV_DEBUG
+/* data pointers of tuple fields are initialized to point here
+for error checking */
+UNIV_INTERN byte	data_error;
+
+# ifndef UNIV_DEBUG_VALGRIND
+/* this is used to fool the compiler in dtuple_validate */
+UNIV_INTERN ulint	data_dummy;
+# endif /* !UNIV_DEBUG_VALGRIND */
+#endif /* UNIV_DEBUG */
+
+/*************************************************************************
+Tests if dfield data length and content is equal to the given. */
+UNIV_INTERN
+ibool
+dfield_data_is_binary_equal(
+/*========================*/
+				/* out: TRUE if equal */
+	const dfield_t*	field,	/* in: field */
+	ulint		len,	/* in: data length or UNIV_SQL_NULL */
+	const byte*	data)	/* in: data */
+{
+	if (len != dfield_get_len(field)) {
+
+		return(FALSE);
+	}
+
+	if (len == UNIV_SQL_NULL) {
+
+		return(TRUE);
+	}
+
+	if (0 != memcmp(dfield_get_data(field), data, len)) {
+
+		return(FALSE);
+	}
+
+	return(TRUE);
+}
+
+/****************************************************************
+Compare two data tuples, respecting the collation of character fields. */
+UNIV_INTERN
+int
+dtuple_coll_cmp(
+/*============*/
+				/* out: 1, 0 , -1 if tuple1 is greater, equal,
+				less, respectively, than tuple2 */
+	const dtuple_t*	tuple1,	/* in: tuple 1 */
+	const dtuple_t*	tuple2)	/* in: tuple 2 */
+{
+	ulint	n_fields;
+	ulint	i;
+
+	ut_ad(tuple1 && tuple2);
+	ut_ad(tuple1->magic_n == DATA_TUPLE_MAGIC_N);
+	ut_ad(tuple2->magic_n == DATA_TUPLE_MAGIC_N);
+	ut_ad(dtuple_check_typed(tuple1));
+	ut_ad(dtuple_check_typed(tuple2));
+
+	n_fields = dtuple_get_n_fields(tuple1);
+
+	if (n_fields != dtuple_get_n_fields(tuple2)) {
+
+		return(n_fields < dtuple_get_n_fields(tuple2) ? -1 : 1);
+	}
+
+	for (i = 0; i < n_fields; i++) {
+		int		cmp;
+		const dfield_t*	field1	= dtuple_get_nth_field(tuple1, i);
+		const dfield_t*	field2	= dtuple_get_nth_field(tuple2, i);
+
+		cmp = cmp_dfield_dfield(field1, field2);
+
+		if (cmp) {
+			return(cmp);
+		}
+	}
+
+	return(0);
+}
+
+/*************************************************************************
+Sets number of fields used in a tuple. Normally this is set in
+dtuple_create, but if you want later to set it smaller, you can use this. */
+UNIV_INTERN
+void
+dtuple_set_n_fields(
+/*================*/
+	dtuple_t*	tuple,		/* in: tuple */
+	ulint		n_fields)	/* in: number of fields */
+{
+	ut_ad(tuple);
+
+	tuple->n_fields = n_fields;
+	tuple->n_fields_cmp = n_fields;
+}
+
+/**************************************************************
+Checks that a data field is typed. */
+static
+ibool
+dfield_check_typed_no_assert(
+/*=========================*/
+				/* out: TRUE if ok */
+	const dfield_t*	field)	/* in: data field */
+{
+	if (dfield_get_type(field)->mtype > DATA_MYSQL
+	    || dfield_get_type(field)->mtype < DATA_VARCHAR) {
+
+		fprintf(stderr,
+			"InnoDB: Error: data field type %lu, len %lu\n",
+			(ulong) dfield_get_type(field)->mtype,
+			(ulong) dfield_get_len(field));
+		return(FALSE);
+	}
+
+	return(TRUE);
+}
+
+/**************************************************************
+Checks that a data tuple is typed. */
+UNIV_INTERN
+ibool
+dtuple_check_typed_no_assert(
+/*=========================*/
+				/* out: TRUE if ok */
+	const dtuple_t*	tuple)	/* in: tuple */
+{
+	const dfield_t*	field;
+	ulint		i;
+
+	if (dtuple_get_n_fields(tuple) > REC_MAX_N_FIELDS) {
+		fprintf(stderr,
+			"InnoDB: Error: index entry has %lu fields\n",
+			(ulong) dtuple_get_n_fields(tuple));
+dump:
+		fputs("InnoDB: Tuple contents: ", stderr);
+		dtuple_print(stderr, tuple);
+		putc('\n', stderr);
+
+		return(FALSE);
+	}
+
+	for (i = 0; i < dtuple_get_n_fields(tuple); i++) {
+
+		field = dtuple_get_nth_field(tuple, i);
+
+		if (!dfield_check_typed_no_assert(field)) {
+			goto dump;
+		}
+	}
+
+	return(TRUE);
+}
+
+/**************************************************************
+Checks that a data field is typed. Asserts an error if not. */
+UNIV_INTERN
+ibool
+dfield_check_typed(
+/*===============*/
+				/* out: TRUE if ok */
+	const dfield_t*	field)	/* in: data field */
+{
+	if (dfield_get_type(field)->mtype > DATA_MYSQL
+	    || dfield_get_type(field)->mtype < DATA_VARCHAR) {
+
+		fprintf(stderr,
+			"InnoDB: Error: data field type %lu, len %lu\n",
+			(ulong) dfield_get_type(field)->mtype,
+			(ulong) dfield_get_len(field));
+
+		ut_error;
+	}
+
+	return(TRUE);
+}
+
+/**************************************************************
+Checks that a data tuple is typed. Asserts an error if not. */
+UNIV_INTERN
+ibool
+dtuple_check_typed(
+/*===============*/
+				/* out: TRUE if ok */
+	const dtuple_t*	tuple)	/* in: tuple */
+{
+	const dfield_t*	field;
+	ulint		i;
+
+	for (i = 0; i < dtuple_get_n_fields(tuple); i++) {
+
+		field = dtuple_get_nth_field(tuple, i);
+
+		ut_a(dfield_check_typed(field));
+	}
+
+	return(TRUE);
+}
+
+#ifdef UNIV_DEBUG
+/**************************************************************
+Validates the consistency of a tuple which must be complete, i.e,
+all fields must have been set. */
+UNIV_INTERN
+ibool
+dtuple_validate(
+/*============*/
+				/* out: TRUE if ok */
+	const dtuple_t*	tuple)	/* in: tuple */
+{
+	const dfield_t*	field;
+	ulint		n_fields;
+	ulint		len;
+	ulint		i;
+
+	ut_ad(tuple->magic_n == DATA_TUPLE_MAGIC_N);
+
+	n_fields = dtuple_get_n_fields(tuple);
+
+	/* We dereference all the data of each field to test
+	for memory traps */
+
+	for (i = 0; i < n_fields; i++) {
+
+		field = dtuple_get_nth_field(tuple, i);
+		len = dfield_get_len(field);
+
+		if (!dfield_is_null(field)) {
+
+			const byte*	data = dfield_get_data(field);
+#ifndef UNIV_DEBUG_VALGRIND
+			ulint		j;
+
+			for (j = 0; j < len; j++) {
+
+				data_dummy  += *data; /* fool the compiler not
+						      to optimize out this
+						      code */
+				data++;
+			}
+#endif /* !UNIV_DEBUG_VALGRIND */
+
+			UNIV_MEM_ASSERT_RW(data, len);
+		}
+	}
+
+	ut_a(dtuple_check_typed(tuple));
+
+	return(TRUE);
+}
+#endif /* UNIV_DEBUG */
+
+/*****************************************************************
+Pretty prints a dfield value according to its data type. */
+UNIV_INTERN
+void
+dfield_print(
+/*=========*/
+	const dfield_t*	dfield)	/* in: dfield */
+{
+	const byte*	data;
+	ulint		len;
+	ulint		i;
+
+	len = dfield_get_len(dfield);
+	data = dfield_get_data(dfield);
+
+	if (dfield_is_null(dfield)) {
+		fputs("NULL", stderr);
+
+		return;
+	}
+
+	switch (dtype_get_mtype(dfield_get_type(dfield))) {
+	case DATA_CHAR:
+	case DATA_VARCHAR:
+		for (i = 0; i < len; i++) {
+			int	c = *data++;
+			putc(isprint(c) ? c : ' ', stderr);
+		}
+
+		if (dfield_is_ext(dfield)) {
+			fputs("(external)", stderr);
+		}
+		break;
+	case DATA_INT:
+		ut_a(len == 4); /* only works for 32-bit integers */
+		fprintf(stderr, "%d", (int)mach_read_from_4(data));
+		break;
+	default:
+		ut_error;
+	}
+}
+
+/*****************************************************************
+Pretty prints a dfield value according to its data type. Also the hex string
+is printed if a string contains non-printable characters. */
+UNIV_INTERN
+void
+dfield_print_also_hex(
+/*==================*/
+	const dfield_t*	dfield)	/* in: dfield */
+{
+	const byte*	data;
+	ulint		len;
+	ulint		prtype;
+	ulint		i;
+	ibool		print_also_hex;
+
+	len = dfield_get_len(dfield);
+	data = dfield_get_data(dfield);
+
+	if (dfield_is_null(dfield)) {
+		fputs("NULL", stderr);
+
+		return;
+	}
+
+	prtype = dtype_get_prtype(dfield_get_type(dfield));
+
+	switch (dtype_get_mtype(dfield_get_type(dfield))) {
+		dulint	id;
+	case DATA_INT:
+		switch (len) {
+			ulint	val;
+		case 1:
+			val = mach_read_from_1(data);
+
+			if (!(prtype & DATA_UNSIGNED)) {
+				val &= ~0x80;
+				fprintf(stderr, "%ld", (long) val);
+			} else {
+				fprintf(stderr, "%lu", (ulong) val);
+			}
+			break;
+
+		case 2:
+			val = mach_read_from_2(data);
+
+			if (!(prtype & DATA_UNSIGNED)) {
+				val &= ~0x8000;
+				fprintf(stderr, "%ld", (long) val);
+			} else {
+				fprintf(stderr, "%lu", (ulong) val);
+			}
+			break;
+
+		case 3:
+			val = mach_read_from_3(data);
+
+			if (!(prtype & DATA_UNSIGNED)) {
+				val &= ~0x800000;
+				fprintf(stderr, "%ld", (long) val);
+			} else {
+				fprintf(stderr, "%lu", (ulong) val);
+			}
+			break;
+
+		case 4:
+			val = mach_read_from_4(data);
+
+			if (!(prtype & DATA_UNSIGNED)) {
+				val &= ~0x80000000;
+				fprintf(stderr, "%ld", (long) val);
+			} else {
+				fprintf(stderr, "%lu", (ulong) val);
+			}
+			break;
+
+		case 6:
+			id = mach_read_from_6(data);
+			fprintf(stderr, "{%lu %lu}",
+				ut_dulint_get_high(id),
+				ut_dulint_get_low(id));
+			break;
+
+		case 7:
+			id = mach_read_from_7(data);
+			fprintf(stderr, "{%lu %lu}",
+				ut_dulint_get_high(id),
+				ut_dulint_get_low(id));
+			break;
+		case 8:
+			id = mach_read_from_8(data);
+			fprintf(stderr, "{%lu %lu}",
+				ut_dulint_get_high(id),
+				ut_dulint_get_low(id));
+			break;
+		default:
+			goto print_hex;
+		}
+		break;
+
+	case DATA_SYS:
+		switch (prtype & DATA_SYS_PRTYPE_MASK) {
+		case DATA_TRX_ID:
+			id = mach_read_from_6(data);
+
+			fprintf(stderr, "trx_id " TRX_ID_FMT,
+				TRX_ID_PREP_PRINTF(id));
+			break;
+
+		case DATA_ROLL_PTR:
+			id = mach_read_from_7(data);
+
+			fprintf(stderr, "roll_ptr {%lu %lu}",
+				ut_dulint_get_high(id), ut_dulint_get_low(id));
+			break;
+
+		case DATA_ROW_ID:
+			id = mach_read_from_6(data);
+
+			fprintf(stderr, "row_id {%lu %lu}",
+				ut_dulint_get_high(id), ut_dulint_get_low(id));
+			break;
+
+		default:
+			id = mach_dulint_read_compressed(data);
+
+			fprintf(stderr, "mix_id {%lu %lu}",
+				ut_dulint_get_high(id), ut_dulint_get_low(id));
+		}
+		break;
+
+	case DATA_CHAR:
+	case DATA_VARCHAR:
+		print_also_hex = FALSE;
+
+		for (i = 0; i < len; i++) {
+			int c = *data++;
+
+			if (!isprint(c)) {
+				print_also_hex = TRUE;
+
+				fprintf(stderr, "\\x%02x", (unsigned char) c);
+			} else {
+				putc(c, stderr);
+			}
+		}
+
+		if (dfield_is_ext(dfield)) {
+			fputs("(external)", stderr);
+		}
+
+		if (!print_also_hex) {
+			break;
+		}
+
+		data = dfield_get_data(dfield);
+		/* fall through */
+
+	case DATA_BINARY:
+	default:
+print_hex:
+		fputs(" Hex: ",stderr);
+
+		for (i = 0; i < len; i++) {
+			fprintf(stderr, "%02lx", (ulint) *data++);
+		}
+
+		if (dfield_is_ext(dfield)) {
+			fputs("(external)", stderr);
+		}
+	}
+}
+
+/*****************************************************************
+Print a dfield value using ut_print_buf. */
+static
+void
+dfield_print_raw(
+/*=============*/
+	FILE*		f,		/* in: output stream */
+	const dfield_t*	dfield)		/* in: dfield */
+{
+	ulint	len	= dfield_get_len(dfield);
+	if (!dfield_is_null(dfield)) {
+		ulint	print_len = ut_min(len, 1000);
+		ut_print_buf(f, dfield_get_data(dfield), print_len);
+		if (len != print_len) {
+			fprintf(f, "(total %lu bytes%s)",
+				(ulong) len,
+				dfield_is_ext(dfield) ? ", external" : "");
+		}
+	} else {
+		fputs(" SQL NULL", f);
+	}
+}
+
+/**************************************************************
+The following function prints the contents of a tuple. */
+UNIV_INTERN
+void
+dtuple_print(
+/*=========*/
+	FILE*		f,	/* in: output stream */
+	const dtuple_t*	tuple)	/* in: tuple */
+{
+	ulint		n_fields;
+	ulint		i;
+
+	n_fields = dtuple_get_n_fields(tuple);
+
+	fprintf(f, "DATA TUPLE: %lu fields;\n", (ulong) n_fields);
+
+	for (i = 0; i < n_fields; i++) {
+		fprintf(f, " %lu:", (ulong) i);
+
+		dfield_print_raw(f, dtuple_get_nth_field(tuple, i));
+
+		putc(';', f);
+		putc('\n', f);
+	}
+
+	ut_ad(dtuple_validate(tuple));
+}
+
+/******************************************************************
+Moves parts of long fields in entry to the big record vector so that
+the size of tuple drops below the maximum record size allowed in the
+database. Moves data only from those fields which are not necessary
+to determine uniquely the insertion place of the tuple in the index. */
+UNIV_INTERN
+big_rec_t*
+dtuple_convert_big_rec(
+/*===================*/
+				/* out, own: created big record vector,
+				NULL if we are not able to shorten
+				the entry enough, i.e., if there are
+				too many fixed-length or short fields
+				in entry or the index is clustered */
+	dict_index_t*	index,	/* in: index */
+	dtuple_t*	entry,	/* in/out: index entry */
+	ulint*		n_ext)	/* in/out: number of
+				externally stored columns */
+{
+	mem_heap_t*	heap;
+	big_rec_t*	vector;
+	dfield_t*	dfield;
+	dict_field_t*	ifield;
+	ulint		size;
+	ulint		n_fields;
+	ulint		local_len;
+	ulint		local_prefix_len;
+
+	if (UNIV_UNLIKELY(!dict_index_is_clust(index))) {
+		return(NULL);
+	}
+
+	if (dict_table_get_format(index->table) < DICT_TF_FORMAT_ZIP) {
+		/* up to MySQL 5.1: store a 768-byte prefix locally */
+		local_len = BTR_EXTERN_FIELD_REF_SIZE + DICT_MAX_INDEX_COL_LEN;
+	} else {
+		/* new-format table: do not store any BLOB prefix locally */
+		local_len = BTR_EXTERN_FIELD_REF_SIZE;
+	}
+
+	ut_a(dtuple_check_typed_no_assert(entry));
+
+	size = rec_get_converted_size(index, entry, *n_ext);
+
+	if (UNIV_UNLIKELY(size > 1000000000)) {
+		fprintf(stderr,
+			"InnoDB: Warning: tuple size very big: %lu\n",
+			(ulong) size);
+		fputs("InnoDB: Tuple contents: ", stderr);
+		dtuple_print(stderr, entry);
+		putc('\n', stderr);
+	}
+
+	heap = mem_heap_create(size + dtuple_get_n_fields(entry)
+			       * sizeof(big_rec_field_t) + 1000);
+
+	vector = mem_heap_alloc(heap, sizeof(big_rec_t));
+
+	vector->heap = heap;
+	vector->fields = mem_heap_alloc(heap, dtuple_get_n_fields(entry)
+					* sizeof(big_rec_field_t));
+
+	/* Decide which fields to shorten: the algorithm is to look for
+	a variable-length field that yields the biggest savings when
+	stored externally */
+
+	n_fields = 0;
+
+	while (page_zip_rec_needs_ext(rec_get_converted_size(index, entry,
+							     *n_ext),
+				      dict_table_is_comp(index->table),
+				      dict_index_get_n_fields(index),
+				      dict_table_zip_size(index->table))) {
+		ulint			i;
+		ulint			longest		= 0;
+		ulint			longest_i	= ULINT_MAX;
+		byte*			data;
+		big_rec_field_t*	b;
+
+		for (i = dict_index_get_n_unique_in_tree(index);
+		     i < dtuple_get_n_fields(entry); i++) {
+			ulint	savings;
+
+			dfield = dtuple_get_nth_field(entry, i);
+			ifield = dict_index_get_nth_field(index, i);
+
+			/* Skip fixed-length, NULL, externally stored,
+			or short columns */
+
+			if (ifield->fixed_len
+			    || dfield_is_null(dfield)
+			    || dfield_is_ext(dfield)
+			    || dfield_get_len(dfield) <= local_len
+			    || dfield_get_len(dfield)
+			    <= BTR_EXTERN_FIELD_REF_SIZE * 2) {
+				goto skip_field;
+			}
+
+			savings = dfield_get_len(dfield) - local_len;
+
+			/* Check that there would be savings */
+			if (longest >= savings) {
+				goto skip_field;
+			}
+
+			longest_i = i;
+			longest = savings;
+
+skip_field:
+			continue;
+		}
+
+		if (!longest) {
+			/* Cannot shorten more */
+
+			mem_heap_free(heap);
+
+			return(NULL);
+		}
+
+		/* Move data from field longest_i to big rec vector.
+
+		We store the first bytes locally to the record. Then
+		we can calculate all ordering fields in all indexes
+		from locally stored data. */
+
+		dfield = dtuple_get_nth_field(entry, longest_i);
+		ifield = dict_index_get_nth_field(index, longest_i);
+		local_prefix_len = local_len - BTR_EXTERN_FIELD_REF_SIZE;
+
+		b = &vector->fields[n_fields];
+		b->field_no = longest_i;
+		b->len = dfield_get_len(dfield) - local_prefix_len;
+		b->data = (char*) dfield_get_data(dfield) + local_prefix_len;
+
+		/* Allocate the locally stored part of the column. */
+		data = mem_heap_alloc(heap, local_len);
+
+		/* Copy the local prefix. */
+		memcpy(data, dfield_get_data(dfield), local_prefix_len);
+		/* Clear the extern field reference (BLOB pointer). */
+		memset(data + local_prefix_len, 0, BTR_EXTERN_FIELD_REF_SIZE);
+#if 0
+		/* The following would fail the Valgrind checks in
+		page_cur_insert_rec_low() and page_cur_insert_rec_zip().
+		The BLOB pointers in the record will be initialized after
+		the record and the BLOBs have been written. */
+		UNIV_MEM_ALLOC(data + local_prefix_len,
+			       BTR_EXTERN_FIELD_REF_SIZE);
+#endif
+
+		dfield_set_data(dfield, data, local_len);
+		dfield_set_ext(dfield);
+
+		n_fields++;
+		(*n_ext)++;
+		ut_ad(n_fields < dtuple_get_n_fields(entry));
+	}
+
+	vector->n_fields = n_fields;
+	return(vector);
+}
+
+/******************************************************************
+Puts back to entry the data stored in vector. Note that to ensure the
+fields in entry can accommodate the data, vector must have been created
+from entry with dtuple_convert_big_rec. */
+UNIV_INTERN
+void
+dtuple_convert_back_big_rec(
+/*========================*/
+	dict_index_t*	index __attribute__((unused)),	/* in: index */
+	dtuple_t*	entry,	/* in: entry whose data was put to vector */
+	big_rec_t*	vector)	/* in, own: big rec vector; it is
+				freed in this function */
+{
+	big_rec_field_t*		b	= vector->fields;
+	const big_rec_field_t* const	end	= b + vector->n_fields;
+
+	for (; b < end; b++) {
+		dfield_t*	dfield;
+		ulint		local_len;
+
+		dfield = dtuple_get_nth_field(entry, b->field_no);
+		local_len = dfield_get_len(dfield);
+
+		ut_ad(dfield_is_ext(dfield));
+		ut_ad(local_len >= BTR_EXTERN_FIELD_REF_SIZE);
+
+		local_len -= BTR_EXTERN_FIELD_REF_SIZE;
+
+		ut_ad(local_len <= DICT_MAX_INDEX_COL_LEN);
+
+		dfield_set_data(dfield,
+				(char*) b->data - local_len,
+				b->len + local_len);
+	}
+
+	mem_heap_free(vector->heap);
+}
diff --git a/storage/xtradb/data/data0type.c b/storage/xtradb/data/data0type.c
new file mode 100644
index 00000000000..5df933ef9fd
--- /dev/null
+++ b/storage/xtradb/data/data0type.c
@@ -0,0 +1,300 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Data types
+
+Created 1/16/1996 Heikki Tuuri
+*******************************************************/
+
+#include "data0type.h"
+
+#ifdef UNIV_NONINL
+#include "data0type.ic"
+#endif
+
+/**********************************************************************
+This function is used to find the storage length in bytes of the first n
+characters for prefix indexes using a multibyte character set. The function
+finds charset information and returns length of prefix_len characters in the
+index field in bytes.
+
+NOTE: the prototype of this function is copied from ha_innodb.cc! If you change
+this function, you MUST change also the prototype here! */
+UNIV_INTERN
+ulint
+innobase_get_at_most_n_mbchars(
+/*===========================*/
+				/* out: number of bytes occupied by the first
+				n characters */
+	ulint charset_id,	/* in: character set id */
+	ulint prefix_len,	/* in: prefix length in bytes of the index
+				(this has to be divided by mbmaxlen to get the
+				number of CHARACTERS n in the prefix) */
+	ulint data_len,		/* in: length of the string in bytes */
+	const char* str);	/* in: character string */
+
+/* At the database startup we store the default-charset collation number of
+this MySQL installation to this global variable. If we have < 4.1.2 format
+column definitions, or records in the insert buffer, we use this
+charset-collation code for them. */
+
+UNIV_INTERN ulint	data_mysql_default_charset_coll;
+
+/*************************************************************************
+Determine how many bytes the first n characters of the given string occupy.
+If the string is shorter than n characters, returns the number of bytes
+the characters in the string occupy. */
+UNIV_INTERN
+ulint
+dtype_get_at_most_n_mbchars(
+/*========================*/
+					/* out: length of the prefix,
+					in bytes */
+	ulint		prtype,		/* in: precise type */
+	ulint		mbminlen,	/* in: minimum length of a
+					multi-byte character */
+	ulint		mbmaxlen,	/* in: maximum length of a
+					multi-byte character */
+	ulint		prefix_len,	/* in: length of the requested
+					prefix, in characters, multiplied by
+					dtype_get_mbmaxlen(dtype) */
+	ulint		data_len,	/* in: length of str (in bytes) */
+	const char*	str)		/* in: the string whose prefix
+					length is being determined */
+{
+#ifndef UNIV_HOTBACKUP
+	ut_a(data_len != UNIV_SQL_NULL);
+	ut_ad(!mbmaxlen || !(prefix_len % mbmaxlen));
+
+	if (mbminlen != mbmaxlen) {
+		ut_a(!(prefix_len % mbmaxlen));
+		return(innobase_get_at_most_n_mbchars(
+			dtype_get_charset_coll(prtype),
+			prefix_len, data_len, str));
+	}
+
+	if (prefix_len < data_len) {
+
+		return(prefix_len);
+
+	}
+
+	return(data_len);
+#else /* UNIV_HOTBACKUP */
+	/* This function depends on MySQL code that is not included in
+	InnoDB Hot Backup builds.  Besides, this function should never
+	be called in InnoDB Hot Backup. */
+	ut_error;
+#endif /* UNIV_HOTBACKUP */
+}
+
+/*************************************************************************
+Checks if a data main type is a string type. Also a BLOB is considered a
+string type. */
+UNIV_INTERN
+ibool
+dtype_is_string_type(
+/*=================*/
+			/* out: TRUE if string type */
+	ulint	mtype)	/* in: InnoDB main data type code: DATA_CHAR, ... */
+{
+	if (mtype <= DATA_BLOB
+	    || mtype == DATA_MYSQL
+	    || mtype == DATA_VARMYSQL) {
+
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
+
+/*************************************************************************
+Checks if a type is a binary string type. Note that for tables created with
+< 4.0.14, we do not know if a DATA_BLOB column is a BLOB or a TEXT column. For
+those DATA_BLOB columns this function currently returns FALSE. */
+UNIV_INTERN
+ibool
+dtype_is_binary_string_type(
+/*========================*/
+			/* out: TRUE if binary string type */
+	ulint	mtype,	/* in: main data type */
+	ulint	prtype)	/* in: precise type */
+{
+	if ((mtype == DATA_FIXBINARY)
+	    || (mtype == DATA_BINARY)
+	    || (mtype == DATA_BLOB && (prtype & DATA_BINARY_TYPE))) {
+
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
+
+/*************************************************************************
+Checks if a type is a non-binary string type. That is, dtype_is_string_type is
+TRUE and dtype_is_binary_string_type is FALSE. Note that for tables created
+with < 4.0.14, we do not know if a DATA_BLOB column is a BLOB or a TEXT column.
+For those DATA_BLOB columns this function currently returns TRUE. */
+UNIV_INTERN
+ibool
+dtype_is_non_binary_string_type(
+/*============================*/
+			/* out: TRUE if non-binary string type */
+	ulint	mtype,	/* in: main data type */
+	ulint	prtype)	/* in: precise type */
+{
+	if (dtype_is_string_type(mtype) == TRUE
+	    && dtype_is_binary_string_type(mtype, prtype) == FALSE) {
+
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
+
+/*************************************************************************
+Forms a precise type from the < 4.1.2 format precise type plus the
+charset-collation code. */
+UNIV_INTERN
+ulint
+dtype_form_prtype(
+/*==============*/
+	ulint	old_prtype,	/* in: the MySQL type code and the flags
+				DATA_BINARY_TYPE etc. */
+	ulint	charset_coll)	/* in: MySQL charset-collation code */
+{
+	ut_a(old_prtype < 256 * 256);
+	ut_a(charset_coll < 256);
+
+	return(old_prtype + (charset_coll << 16));
+}
+
+/*************************************************************************
+Validates a data type structure. */
+UNIV_INTERN
+ibool
+dtype_validate(
+/*===========*/
+				/* out: TRUE if ok */
+	const dtype_t*	type)	/* in: type struct to validate */
+{
+	ut_a(type);
+	ut_a(type->mtype >= DATA_VARCHAR);
+	ut_a(type->mtype <= DATA_MYSQL);
+
+	if (type->mtype == DATA_SYS) {
+		ut_a((type->prtype & DATA_MYSQL_TYPE_MASK) < DATA_N_SYS_COLS);
+	}
+
+	ut_a(type->mbminlen <= type->mbmaxlen);
+
+	return(TRUE);
+}
+
+/*************************************************************************
+Prints a data type structure. */
+UNIV_INTERN
+void
+dtype_print(
+/*========*/
+	const dtype_t*	type)	/* in: type */
+{
+	ulint	mtype;
+	ulint	prtype;
+	ulint	len;
+
+	ut_a(type);
+
+	mtype = type->mtype;
+	prtype = type->prtype;
+
+	switch (mtype) {
+	case DATA_VARCHAR:
+		fputs("DATA_VARCHAR", stderr);
+		break;
+
+	case DATA_CHAR:
+		fputs("DATA_CHAR", stderr);
+		break;
+
+	case DATA_BINARY:
+		fputs("DATA_BINARY", stderr);
+		break;
+
+	case DATA_FIXBINARY:
+		fputs("DATA_FIXBINARY", stderr);
+		break;
+
+	case DATA_BLOB:
+		fputs("DATA_BLOB", stderr);
+		break;
+
+	case DATA_INT:
+		fputs("DATA_INT", stderr);
+		break;
+
+	case DATA_MYSQL:
+		fputs("DATA_MYSQL", stderr);
+		break;
+
+	case DATA_SYS:
+		fputs("DATA_SYS", stderr);
+		break;
+
+	default:
+		fprintf(stderr, "type %lu", (ulong) mtype);
+		break;
+	}
+
+	len = type->len;
+
+	if ((type->mtype == DATA_SYS)
+	    || (type->mtype == DATA_VARCHAR)
+	    || (type->mtype == DATA_CHAR)) {
+		putc(' ', stderr);
+		if (prtype == DATA_ROW_ID) {
+			fputs("DATA_ROW_ID", stderr);
+			len = DATA_ROW_ID_LEN;
+		} else if (prtype == DATA_ROLL_PTR) {
+			fputs("DATA_ROLL_PTR", stderr);
+			len = DATA_ROLL_PTR_LEN;
+		} else if (prtype == DATA_TRX_ID) {
+			fputs("DATA_TRX_ID", stderr);
+			len = DATA_TRX_ID_LEN;
+		} else if (prtype == DATA_ENGLISH) {
+			fputs("DATA_ENGLISH", stderr);
+		} else {
+			fprintf(stderr, "prtype %lu", (ulong) prtype);
+		}
+	} else {
+		if (prtype & DATA_UNSIGNED) {
+			fputs(" DATA_UNSIGNED", stderr);
+		}
+
+		if (prtype & DATA_BINARY_TYPE) {
+			fputs(" DATA_BINARY_TYPE", stderr);
+		}
+
+		if (prtype & DATA_NOT_NULL) {
+			fputs(" DATA_NOT_NULL", stderr);
+		}
+	}
+
+	fprintf(stderr, " len %lu", (ulong) len);
+}
diff --git a/storage/xtradb/dict/dict0boot.c b/storage/xtradb/dict/dict0boot.c
new file mode 100644
index 00000000000..505ba53ec52
--- /dev/null
+++ b/storage/xtradb/dict/dict0boot.c
@@ -0,0 +1,462 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Data dictionary creation and booting
+
+Created 4/18/1996 Heikki Tuuri
+*******************************************************/
+
+#include "dict0boot.h"
+
+#ifdef UNIV_NONINL
+#include "dict0boot.ic"
+#endif
+
+#include "dict0crea.h"
+#include "btr0btr.h"
+#include "dict0load.h"
+#include "dict0load.h"
+#include "trx0trx.h"
+#include "srv0srv.h"
+#include "ibuf0ibuf.h"
+#include "buf0flu.h"
+#include "log0recv.h"
+#include "os0file.h"
+
+/**************************************************************************
+Gets a pointer to the dictionary header and x-latches its page. */
+UNIV_INTERN
+dict_hdr_t*
+dict_hdr_get(
+/*=========*/
+			/* out: pointer to the dictionary header,
+			page x-latched */
+	mtr_t*	mtr)	/* in: mtr */
+{
+	buf_block_t*	block;
+	dict_hdr_t*	header;
+
+	block = buf_page_get(DICT_HDR_SPACE, 0, DICT_HDR_PAGE_NO,
+			     RW_X_LATCH, mtr);
+	header = DICT_HDR + buf_block_get_frame(block);
+
+	buf_block_dbg_add_level(block, SYNC_DICT_HEADER);
+
+	return(header);
+}
+
+/**************************************************************************
+Returns a new table, index, or tree id. */
+UNIV_INTERN
+dulint
+dict_hdr_get_new_id(
+/*================*/
+			/* out: the new id */
+	ulint	type)	/* in: DICT_HDR_ROW_ID, ... */
+{
+	dict_hdr_t*	dict_hdr;
+	dulint		id;
+	mtr_t		mtr;
+
+	ut_ad((type == DICT_HDR_TABLE_ID) || (type == DICT_HDR_INDEX_ID));
+
+	mtr_start(&mtr);
+
+	dict_hdr = dict_hdr_get(&mtr);
+
+	id = mtr_read_dulint(dict_hdr + type, &mtr);
+	id = ut_dulint_add(id, 1);
+
+	mlog_write_dulint(dict_hdr + type, id, &mtr);
+
+	mtr_commit(&mtr);
+
+	return(id);
+}
+
+/**************************************************************************
+Writes the current value of the row id counter to the dictionary header file
+page. */
+UNIV_INTERN
+void
+dict_hdr_flush_row_id(void)
+/*=======================*/
+{
+	dict_hdr_t*	dict_hdr;
+	dulint		id;
+	mtr_t		mtr;
+
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+
+	id = dict_sys->row_id;
+
+	mtr_start(&mtr);
+
+	dict_hdr = dict_hdr_get(&mtr);
+
+	mlog_write_dulint(dict_hdr + DICT_HDR_ROW_ID, id, &mtr);
+
+	mtr_commit(&mtr);
+}
+
+/*********************************************************************
+Creates the file page for the dictionary header. This function is
+called only at the database creation. */
+static
+ibool
+dict_hdr_create(
+/*============*/
+			/* out: TRUE if succeed */
+	mtr_t*	mtr)	/* in: mtr */
+{
+	buf_block_t*	block;
+	dict_hdr_t*	dict_header;
+	ulint		root_page_no;
+
+	ut_ad(mtr);
+
+	/* Create the dictionary header file block in a new, allocated file
+	segment in the system tablespace */
+	block = fseg_create(DICT_HDR_SPACE, 0,
+			    DICT_HDR + DICT_HDR_FSEG_HEADER, mtr);
+
+	ut_a(DICT_HDR_PAGE_NO == buf_block_get_page_no(block));
+
+	dict_header = dict_hdr_get(mtr);
+
+	/* Start counting row, table, index, and tree ids from
+	DICT_HDR_FIRST_ID */
+	mlog_write_dulint(dict_header + DICT_HDR_ROW_ID,
+			  ut_dulint_create(0, DICT_HDR_FIRST_ID), mtr);
+
+	mlog_write_dulint(dict_header + DICT_HDR_TABLE_ID,
+			  ut_dulint_create(0, DICT_HDR_FIRST_ID), mtr);
+
+	mlog_write_dulint(dict_header + DICT_HDR_INDEX_ID,
+			  ut_dulint_create(0, DICT_HDR_FIRST_ID), mtr);
+
+	/* Obsolete, but we must initialize it to 0 anyway. */
+	mlog_write_dulint(dict_header + DICT_HDR_MIX_ID,
+			  ut_dulint_create(0, DICT_HDR_FIRST_ID), mtr);
+
+	/* Create the B-tree roots for the clustered indexes of the basic
+	system tables */
+
+	/*--------------------------*/
+	root_page_no = btr_create(DICT_CLUSTERED | DICT_UNIQUE,
+				  DICT_HDR_SPACE, 0, DICT_TABLES_ID,
+				  srv_sys->dummy_ind1, mtr);
+	if (root_page_no == FIL_NULL) {
+
+		return(FALSE);
+	}
+
+	mlog_write_ulint(dict_header + DICT_HDR_TABLES, root_page_no,
+			 MLOG_4BYTES, mtr);
+	/*--------------------------*/
+	root_page_no = btr_create(DICT_UNIQUE, DICT_HDR_SPACE, 0,
+				  DICT_TABLE_IDS_ID,
+				  srv_sys->dummy_ind1, mtr);
+	if (root_page_no == FIL_NULL) {
+
+		return(FALSE);
+	}
+
+	mlog_write_ulint(dict_header + DICT_HDR_TABLE_IDS, root_page_no,
+			 MLOG_4BYTES, mtr);
+	/*--------------------------*/
+	root_page_no = btr_create(DICT_CLUSTERED | DICT_UNIQUE,
+				  DICT_HDR_SPACE, 0, DICT_COLUMNS_ID,
+				  srv_sys->dummy_ind1, mtr);
+	if (root_page_no == FIL_NULL) {
+
+		return(FALSE);
+	}
+
+	mlog_write_ulint(dict_header + DICT_HDR_COLUMNS, root_page_no,
+			 MLOG_4BYTES, mtr);
+	/*--------------------------*/
+	root_page_no = btr_create(DICT_CLUSTERED | DICT_UNIQUE,
+				  DICT_HDR_SPACE, 0, DICT_INDEXES_ID,
+				  srv_sys->dummy_ind1, mtr);
+	if (root_page_no == FIL_NULL) {
+
+		return(FALSE);
+	}
+
+	mlog_write_ulint(dict_header + DICT_HDR_INDEXES, root_page_no,
+			 MLOG_4BYTES, mtr);
+	/*--------------------------*/
+	root_page_no = btr_create(DICT_CLUSTERED | DICT_UNIQUE,
+				  DICT_HDR_SPACE, 0, DICT_FIELDS_ID,
+				  srv_sys->dummy_ind1, mtr);
+	if (root_page_no == FIL_NULL) {
+
+		return(FALSE);
+	}
+
+	mlog_write_ulint(dict_header + DICT_HDR_FIELDS, root_page_no,
+			 MLOG_4BYTES, mtr);
+	/*--------------------------*/
+
+	return(TRUE);
+}
+
+/*********************************************************************
+Initializes the data dictionary memory structures when the database is
+started. This function is also called when the data dictionary is created. */
+UNIV_INTERN
+void
+dict_boot(void)
+/*===========*/
+{
+	dict_table_t*	table;
+	dict_index_t*	index;
+	dict_hdr_t*	dict_hdr;
+	mem_heap_t*	heap;
+	mtr_t		mtr;
+	ulint		error;
+
+	mtr_start(&mtr);
+
+	/* Create the hash tables etc. */
+	dict_init();
+
+	heap = mem_heap_create(450);
+
+	mutex_enter(&(dict_sys->mutex));
+
+	/* Get the dictionary header */
+	dict_hdr = dict_hdr_get(&mtr);
+
+	/* Because we only write new row ids to disk-based data structure
+	(dictionary header) when it is divisible by
+	DICT_HDR_ROW_ID_WRITE_MARGIN, in recovery we will not recover
+	the latest value of the row id counter. Therefore we advance
+	the counter at the database startup to avoid overlapping values.
+	Note that when a user after database startup first time asks for
+	a new row id, then because the counter is now divisible by
+	..._MARGIN, it will immediately be updated to the disk-based
+	header. */
+
+	dict_sys->row_id = ut_dulint_add(
+		ut_dulint_align_up(mtr_read_dulint(dict_hdr + DICT_HDR_ROW_ID,
+						   &mtr),
+				   DICT_HDR_ROW_ID_WRITE_MARGIN),
+		DICT_HDR_ROW_ID_WRITE_MARGIN);
+
+	/* Insert into the dictionary cache the descriptions of the basic
+	system tables */
+	/*-------------------------*/
+	table = dict_mem_table_create("SYS_TABLES", DICT_HDR_SPACE, 8, 0);
+
+	dict_mem_table_add_col(table, heap, "NAME", DATA_BINARY, 0, 0);
+	dict_mem_table_add_col(table, heap, "ID", DATA_BINARY, 0, 0);
+	/* ROW_FORMAT = (N_COLS >> 31) ? COMPACT : REDUNDANT */
+	dict_mem_table_add_col(table, heap, "N_COLS", DATA_INT, 0, 4);
+	/* TYPE is either DICT_TABLE_ORDINARY, or (TYPE & DICT_TF_COMPACT)
+	and (TYPE & DICT_TF_FORMAT_MASK) are nonzero and TYPE = table->flags */
+	dict_mem_table_add_col(table, heap, "TYPE", DATA_INT, 0, 4);
+	dict_mem_table_add_col(table, heap, "MIX_ID", DATA_BINARY, 0, 0);
+	dict_mem_table_add_col(table, heap, "MIX_LEN", DATA_INT, 0, 4);
+	dict_mem_table_add_col(table, heap, "CLUSTER_NAME", DATA_BINARY, 0, 0);
+	dict_mem_table_add_col(table, heap, "SPACE", DATA_INT, 0, 4);
+
+	table->id = DICT_TABLES_ID;
+
+	dict_table_add_to_cache(table, heap);
+	dict_sys->sys_tables = table;
+	mem_heap_empty(heap);
+
+	index = dict_mem_index_create("SYS_TABLES", "CLUST_IND",
+				      DICT_HDR_SPACE,
+				      DICT_UNIQUE | DICT_CLUSTERED, 1);
+
+	dict_mem_index_add_field(index, "NAME", 0);
+
+	index->id = DICT_TABLES_ID;
+
+	error = dict_index_add_to_cache(table, index,
+					mtr_read_ulint(dict_hdr
+						       + DICT_HDR_TABLES,
+						       MLOG_4BYTES, &mtr),
+					FALSE);
+	ut_a(error == DB_SUCCESS);
+
+	/*-------------------------*/
+	index = dict_mem_index_create("SYS_TABLES", "ID_IND",
+				      DICT_HDR_SPACE, DICT_UNIQUE, 1);
+	dict_mem_index_add_field(index, "ID", 0);
+
+	index->id = DICT_TABLE_IDS_ID;
+	error = dict_index_add_to_cache(table, index,
+					mtr_read_ulint(dict_hdr
+						       + DICT_HDR_TABLE_IDS,
+						       MLOG_4BYTES, &mtr),
+					FALSE);
+	ut_a(error == DB_SUCCESS);
+
+	/*-------------------------*/
+	table = dict_mem_table_create("SYS_COLUMNS", DICT_HDR_SPACE, 7, 0);
+
+	dict_mem_table_add_col(table, heap, "TABLE_ID", DATA_BINARY, 0, 0);
+	dict_mem_table_add_col(table, heap, "POS", DATA_INT, 0, 4);
+	dict_mem_table_add_col(table, heap, "NAME", DATA_BINARY, 0, 0);
+	dict_mem_table_add_col(table, heap, "MTYPE", DATA_INT, 0, 4);
+	dict_mem_table_add_col(table, heap, "PRTYPE", DATA_INT, 0, 4);
+	dict_mem_table_add_col(table, heap, "LEN", DATA_INT, 0, 4);
+	dict_mem_table_add_col(table, heap, "PREC", DATA_INT, 0, 4);
+
+	table->id = DICT_COLUMNS_ID;
+
+	dict_table_add_to_cache(table, heap);
+	dict_sys->sys_columns = table;
+	mem_heap_empty(heap);
+
+	index = dict_mem_index_create("SYS_COLUMNS", "CLUST_IND",
+				      DICT_HDR_SPACE,
+				      DICT_UNIQUE | DICT_CLUSTERED, 2);
+
+	dict_mem_index_add_field(index, "TABLE_ID", 0);
+	dict_mem_index_add_field(index, "POS", 0);
+
+	index->id = DICT_COLUMNS_ID;
+	error = dict_index_add_to_cache(table, index,
+					mtr_read_ulint(dict_hdr
+						       + DICT_HDR_COLUMNS,
+						       MLOG_4BYTES, &mtr),
+					FALSE);
+	ut_a(error == DB_SUCCESS);
+
+	/*-------------------------*/
+	table = dict_mem_table_create("SYS_INDEXES", DICT_HDR_SPACE, 7, 0);
+
+	dict_mem_table_add_col(table, heap, "TABLE_ID", DATA_BINARY, 0, 0);
+	dict_mem_table_add_col(table, heap, "ID", DATA_BINARY, 0, 0);
+	dict_mem_table_add_col(table, heap, "NAME", DATA_BINARY, 0, 0);
+	dict_mem_table_add_col(table, heap, "N_FIELDS", DATA_INT, 0, 4);
+	dict_mem_table_add_col(table, heap, "TYPE", DATA_INT, 0, 4);
+	dict_mem_table_add_col(table, heap, "SPACE", DATA_INT, 0, 4);
+	dict_mem_table_add_col(table, heap, "PAGE_NO", DATA_INT, 0, 4);
+
+	/* The '+ 2' below comes from the 2 system fields */
+#if DICT_SYS_INDEXES_PAGE_NO_FIELD != 6 + 2
+#error "DICT_SYS_INDEXES_PAGE_NO_FIELD != 6 + 2"
+#endif
+#if DICT_SYS_INDEXES_SPACE_NO_FIELD != 5 + 2
+#error "DICT_SYS_INDEXES_SPACE_NO_FIELD != 5 + 2"
+#endif
+#if DICT_SYS_INDEXES_TYPE_FIELD != 4 + 2
+#error "DICT_SYS_INDEXES_TYPE_FIELD != 4 + 2"
+#endif
+
+	table->id = DICT_INDEXES_ID;
+	dict_table_add_to_cache(table, heap);
+	dict_sys->sys_indexes = table;
+	mem_heap_empty(heap);
+
+	index = dict_mem_index_create("SYS_INDEXES", "CLUST_IND",
+				      DICT_HDR_SPACE,
+				      DICT_UNIQUE | DICT_CLUSTERED, 2);
+
+	dict_mem_index_add_field(index, "TABLE_ID", 0);
+	dict_mem_index_add_field(index, "ID", 0);
+
+	index->id = DICT_INDEXES_ID;
+	error = dict_index_add_to_cache(table, index,
+					mtr_read_ulint(dict_hdr
+						       + DICT_HDR_INDEXES,
+						       MLOG_4BYTES, &mtr),
+					FALSE);
+	ut_a(error == DB_SUCCESS);
+
+	/*-------------------------*/
+	table = dict_mem_table_create("SYS_FIELDS", DICT_HDR_SPACE, 3, 0);
+
+	dict_mem_table_add_col(table, heap, "INDEX_ID", DATA_BINARY, 0, 0);
+	dict_mem_table_add_col(table, heap, "POS", DATA_INT, 0, 4);
+	dict_mem_table_add_col(table, heap, "COL_NAME", DATA_BINARY, 0, 0);
+
+	table->id = DICT_FIELDS_ID;
+	dict_table_add_to_cache(table, heap);
+	dict_sys->sys_fields = table;
+	mem_heap_free(heap);
+
+	index = dict_mem_index_create("SYS_FIELDS", "CLUST_IND",
+				      DICT_HDR_SPACE,
+				      DICT_UNIQUE | DICT_CLUSTERED, 2);
+
+	dict_mem_index_add_field(index, "INDEX_ID", 0);
+	dict_mem_index_add_field(index, "POS", 0);
+
+	index->id = DICT_FIELDS_ID;
+	error = dict_index_add_to_cache(table, index,
+					mtr_read_ulint(dict_hdr
+						       + DICT_HDR_FIELDS,
+						       MLOG_4BYTES, &mtr),
+					FALSE);
+	ut_a(error == DB_SUCCESS);
+
+	mtr_commit(&mtr);
+	/*-------------------------*/
+
+	/* Initialize the insert buffer table and index for each tablespace */
+
+	ibuf_init_at_db_start();
+
+	/* Load definitions of other indexes on system tables */
+
+	dict_load_sys_table(dict_sys->sys_tables);
+	dict_load_sys_table(dict_sys->sys_columns);
+	dict_load_sys_table(dict_sys->sys_indexes);
+	dict_load_sys_table(dict_sys->sys_fields);
+
+	mutex_exit(&(dict_sys->mutex));
+}
+
+/*********************************************************************
+Inserts the basic system table data into themselves in the database
+creation. */
+static
+void
+dict_insert_initial_data(void)
+/*==========================*/
+{
+	/* Does nothing yet */
+}
+
+/*********************************************************************
+Creates and initializes the data dictionary at the database creation. */
+UNIV_INTERN
+void
+dict_create(void)
+/*=============*/
+{
+	mtr_t	mtr;
+
+	mtr_start(&mtr);
+
+	dict_hdr_create(&mtr);
+
+	mtr_commit(&mtr);
+
+	dict_boot();
+
+	dict_insert_initial_data();
+}
diff --git a/storage/xtradb/dict/dict0crea.c b/storage/xtradb/dict/dict0crea.c
new file mode 100644
index 00000000000..b9662c9a44c
--- /dev/null
+++ b/storage/xtradb/dict/dict0crea.c
@@ -0,0 +1,1503 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Database object creation
+
+Created 1/8/1996 Heikki Tuuri
+*******************************************************/
+
+#include "dict0crea.h"
+
+#ifdef UNIV_NONINL
+#include "dict0crea.ic"
+#endif
+
+#include "btr0pcur.h"
+#include "btr0btr.h"
+#include "page0page.h"
+#include "mach0data.h"
+#include "dict0boot.h"
+#include "dict0dict.h"
+#include "que0que.h"
+#include "row0ins.h"
+#include "row0mysql.h"
+#include "pars0pars.h"
+#include "trx0roll.h"
+#include "usr0sess.h"
+#include "ut0vec.h"
+
+/*********************************************************************
+Based on a table object, this function builds the entry to be inserted
+in the SYS_TABLES system table. */
+static
+dtuple_t*
+dict_create_sys_tables_tuple(
+/*=========================*/
+				/* out: the tuple which should be inserted */
+	dict_table_t*	table,	/* in: table */
+	mem_heap_t*	heap)	/* in: memory heap from which the memory for
+				the built tuple is allocated */
+{
+	dict_table_t*	sys_tables;
+	dtuple_t*	entry;
+	dfield_t*	dfield;
+	byte*		ptr;
+
+	ut_ad(table && heap);
+
+	sys_tables = dict_sys->sys_tables;
+
+	entry = dtuple_create(heap, 8 + DATA_N_SYS_COLS);
+
+	dict_table_copy_types(entry, sys_tables);
+
+	/* 0: NAME -----------------------------*/
+	dfield = dtuple_get_nth_field(entry, 0);
+
+	dfield_set_data(dfield, table->name, ut_strlen(table->name));
+	/* 3: ID -------------------------------*/
+	dfield = dtuple_get_nth_field(entry, 1);
+
+	ptr = mem_heap_alloc(heap, 8);
+	mach_write_to_8(ptr, table->id);
+
+	dfield_set_data(dfield, ptr, 8);
+	/* 4: N_COLS ---------------------------*/
+	dfield = dtuple_get_nth_field(entry, 2);
+
+#if DICT_TF_COMPACT != 1
+#error
+#endif
+
+	ptr = mem_heap_alloc(heap, 4);
+	mach_write_to_4(ptr, table->n_def
+			| ((table->flags & DICT_TF_COMPACT) << 31));
+	dfield_set_data(dfield, ptr, 4);
+	/* 5: TYPE -----------------------------*/
+	dfield = dtuple_get_nth_field(entry, 3);
+
+	ptr = mem_heap_alloc(heap, 4);
+	if (table->flags & ~DICT_TF_COMPACT) {
+		ut_a(table->flags & DICT_TF_COMPACT);
+		ut_a(dict_table_get_format(table) >= DICT_TF_FORMAT_ZIP);
+		ut_a((table->flags & DICT_TF_ZSSIZE_MASK)
+		     <= (DICT_TF_ZSSIZE_MAX << DICT_TF_ZSSIZE_SHIFT));
+		ut_a(!(table->flags & (~0 << DICT_TF_BITS)));
+		mach_write_to_4(ptr, table->flags);
+	} else {
+		mach_write_to_4(ptr, DICT_TABLE_ORDINARY);
+	}
+
+	dfield_set_data(dfield, ptr, 4);
+	/* 6: MIX_ID (obsolete) ---------------------------*/
+	dfield = dtuple_get_nth_field(entry, 4);
+
+	ptr = mem_heap_zalloc(heap, 8);
+
+	dfield_set_data(dfield, ptr, 8);
+	/* 7: MIX_LEN (obsolete) --------------------------*/
+
+	dfield = dtuple_get_nth_field(entry, 5);
+
+	ptr = mem_heap_zalloc(heap, 4);
+
+	dfield_set_data(dfield, ptr, 4);
+	/* 8: CLUSTER_NAME ---------------------*/
+	dfield = dtuple_get_nth_field(entry, 6);
+	dfield_set_null(dfield); /* not supported */
+
+	/* 9: SPACE ----------------------------*/
+	dfield = dtuple_get_nth_field(entry, 7);
+
+	ptr = mem_heap_alloc(heap, 4);
+	mach_write_to_4(ptr, table->space);
+
+	dfield_set_data(dfield, ptr, 4);
+	/*----------------------------------*/
+
+	return(entry);
+}
+
+/*********************************************************************
+Based on a table object, this function builds the entry to be inserted
+in the SYS_COLUMNS system table. */
+static
+dtuple_t*
+dict_create_sys_columns_tuple(
+/*==========================*/
+				/* out: the tuple which should be inserted */
+	dict_table_t*	table,	/* in: table */
+	ulint		i,	/* in: column number */
+	mem_heap_t*	heap)	/* in: memory heap from which the memory for
+				the built tuple is allocated */
+{
+	dict_table_t*		sys_columns;
+	dtuple_t*		entry;
+	const dict_col_t*	column;
+	dfield_t*		dfield;
+	byte*			ptr;
+	const char*	col_name;
+
+	ut_ad(table && heap);
+
+	column = dict_table_get_nth_col(table, i);
+
+	sys_columns = dict_sys->sys_columns;
+
+	entry = dtuple_create(heap, 7 + DATA_N_SYS_COLS);
+
+	dict_table_copy_types(entry, sys_columns);
+
+	/* 0: TABLE_ID -----------------------*/
+	dfield = dtuple_get_nth_field(entry, 0);
+
+	ptr = mem_heap_alloc(heap, 8);
+	mach_write_to_8(ptr, table->id);
+
+	dfield_set_data(dfield, ptr, 8);
+	/* 1: POS ----------------------------*/
+	dfield = dtuple_get_nth_field(entry, 1);
+
+	ptr = mem_heap_alloc(heap, 4);
+	mach_write_to_4(ptr, i);
+
+	dfield_set_data(dfield, ptr, 4);
+	/* 4: NAME ---------------------------*/
+	dfield = dtuple_get_nth_field(entry, 2);
+
+	col_name = dict_table_get_col_name(table, i);
+	dfield_set_data(dfield, col_name, ut_strlen(col_name));
+	/* 5: MTYPE --------------------------*/
+	dfield = dtuple_get_nth_field(entry, 3);
+
+	ptr = mem_heap_alloc(heap, 4);
+	mach_write_to_4(ptr, column->mtype);
+
+	dfield_set_data(dfield, ptr, 4);
+	/* 6: PRTYPE -------------------------*/
+	dfield = dtuple_get_nth_field(entry, 4);
+
+	ptr = mem_heap_alloc(heap, 4);
+	mach_write_to_4(ptr, column->prtype);
+
+	dfield_set_data(dfield, ptr, 4);
+	/* 7: LEN ----------------------------*/
+	dfield = dtuple_get_nth_field(entry, 5);
+
+	ptr = mem_heap_alloc(heap, 4);
+	mach_write_to_4(ptr, column->len);
+
+	dfield_set_data(dfield, ptr, 4);
+	/* 8: PREC ---------------------------*/
+	dfield = dtuple_get_nth_field(entry, 6);
+
+	ptr = mem_heap_alloc(heap, 4);
+	mach_write_to_4(ptr, 0/* unused */);
+
+	dfield_set_data(dfield, ptr, 4);
+	/*---------------------------------*/
+
+	return(entry);
+}
+
+/*******************************************************************
+Builds a table definition to insert. */
+static
+ulint
+dict_build_table_def_step(
+/*======================*/
+				/* out: DB_SUCCESS or error code */
+	que_thr_t*	thr,	/* in: query thread */
+	tab_node_t*	node)	/* in: table create node */
+{
+	dict_table_t*	table;
+	dtuple_t*	row;
+	ulint		error;
+	const char*	path_or_name;
+	ibool		is_path;
+	mtr_t		mtr;
+
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+
+	table = node->table;
+
+	table->id = dict_hdr_get_new_id(DICT_HDR_TABLE_ID);
+
+	thr_get_trx(thr)->table_id = table->id;
+
+	if (srv_file_per_table) {
+		/* We create a new single-table tablespace for the table.
+		We initially let it be 4 pages:
+		- page 0 is the fsp header and an extent descriptor page,
+		- page 1 is an ibuf bitmap page,
+		- page 2 is the first inode page,
+		- page 3 will contain the root of the clustered index of the
+		table we create here. */
+
+		ulint	space = 0;	/* reset to zero for the call below */
+
+		if (table->dir_path_of_temp_table) {
+			/* We place tables created with CREATE TEMPORARY
+			TABLE in the tmp dir of mysqld server */
+
+			path_or_name = table->dir_path_of_temp_table;
+			is_path = TRUE;
+		} else {
+			path_or_name = table->name;
+			is_path = FALSE;
+		}
+
+		ut_ad(dict_table_get_format(table) <= DICT_TF_FORMAT_MAX);
+		ut_ad(!dict_table_zip_size(table)
+		      || dict_table_get_format(table) >= DICT_TF_FORMAT_ZIP);
+
+		error = fil_create_new_single_table_tablespace(
+			&space, path_or_name, is_path,
+			table->flags == DICT_TF_COMPACT ? 0 : table->flags,
+			FIL_IBD_FILE_INITIAL_SIZE);
+		table->space = (unsigned int) space;
+
+		if (error != DB_SUCCESS) {
+
+			return(error);
+		}
+
+		mtr_start(&mtr);
+
+		fsp_header_init(table->space, FIL_IBD_FILE_INITIAL_SIZE, &mtr);
+
+		mtr_commit(&mtr);
+	} else {
+		/* Create in the system tablespace: disallow new features */
+		table->flags &= DICT_TF_COMPACT;
+	}
+
+	row = dict_create_sys_tables_tuple(table, node->heap);
+
+	ins_node_set_new_row(node->tab_def, row);
+
+	return(DB_SUCCESS);
+}
+
+/*******************************************************************
+Builds a column definition to insert. */
+static
+ulint
+dict_build_col_def_step(
+/*====================*/
+				/* out: DB_SUCCESS */
+	tab_node_t*	node)	/* in: table create node */
+{
+	dtuple_t*	row;
+
+	row = dict_create_sys_columns_tuple(node->table, node->col_no,
+					    node->heap);
+	ins_node_set_new_row(node->col_def, row);
+
+	return(DB_SUCCESS);
+}
+
+/*********************************************************************
+Based on an index object, this function builds the entry to be inserted
+in the SYS_INDEXES system table. */
+static
+dtuple_t*
+dict_create_sys_indexes_tuple(
+/*==========================*/
+				/* out: the tuple which should be inserted */
+	dict_index_t*	index,	/* in: index */
+	mem_heap_t*	heap)	/* in: memory heap from which the memory for
+				the built tuple is allocated */
+{
+	dict_table_t*	sys_indexes;
+	dict_table_t*	table;
+	dtuple_t*	entry;
+	dfield_t*	dfield;
+	byte*		ptr;
+
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+	ut_ad(index && heap);
+
+	sys_indexes = dict_sys->sys_indexes;
+
+	table = dict_table_get_low(index->table_name);
+
+	entry = dtuple_create(heap, 7 + DATA_N_SYS_COLS);
+
+	dict_table_copy_types(entry, sys_indexes);
+
+	/* 0: TABLE_ID -----------------------*/
+	dfield = dtuple_get_nth_field(entry, 0);
+
+	ptr = mem_heap_alloc(heap, 8);
+	mach_write_to_8(ptr, table->id);
+
+	dfield_set_data(dfield, ptr, 8);
+	/* 1: ID ----------------------------*/
+	dfield = dtuple_get_nth_field(entry, 1);
+
+	ptr = mem_heap_alloc(heap, 8);
+	mach_write_to_8(ptr, index->id);
+
+	dfield_set_data(dfield, ptr, 8);
+	/* 4: NAME --------------------------*/
+	dfield = dtuple_get_nth_field(entry, 2);
+
+	dfield_set_data(dfield, index->name, ut_strlen(index->name));
+	/* 5: N_FIELDS ----------------------*/
+	dfield = dtuple_get_nth_field(entry, 3);
+
+	ptr = mem_heap_alloc(heap, 4);
+	mach_write_to_4(ptr, index->n_fields);
+
+	dfield_set_data(dfield, ptr, 4);
+	/* 6: TYPE --------------------------*/
+	dfield = dtuple_get_nth_field(entry, 4);
+
+	ptr = mem_heap_alloc(heap, 4);
+	mach_write_to_4(ptr, index->type);
+
+	dfield_set_data(dfield, ptr, 4);
+	/* 7: SPACE --------------------------*/
+
+#if DICT_SYS_INDEXES_SPACE_NO_FIELD != 7
+#error "DICT_SYS_INDEXES_SPACE_NO_FIELD != 7"
+#endif
+
+	dfield = dtuple_get_nth_field(entry, 5);
+
+	ptr = mem_heap_alloc(heap, 4);
+	mach_write_to_4(ptr, index->space);
+
+	dfield_set_data(dfield, ptr, 4);
+	/* 8: PAGE_NO --------------------------*/
+
+#if DICT_SYS_INDEXES_PAGE_NO_FIELD != 8
+#error "DICT_SYS_INDEXES_PAGE_NO_FIELD != 8"
+#endif
+
+	dfield = dtuple_get_nth_field(entry, 6);
+
+	ptr = mem_heap_alloc(heap, 4);
+	mach_write_to_4(ptr, FIL_NULL);
+
+	dfield_set_data(dfield, ptr, 4);
+	/*--------------------------------*/
+
+	return(entry);
+}
+
+/*********************************************************************
+Based on an index object, this function builds the entry to be inserted
+in the SYS_FIELDS system table. */
+static
+dtuple_t*
+dict_create_sys_fields_tuple(
+/*=========================*/
+				/* out: the tuple which should be inserted */
+	dict_index_t*	index,	/* in: index */
+	ulint		i,	/* in: field number */
+	mem_heap_t*	heap)	/* in: memory heap from which the memory for
+				the built tuple is allocated */
+{
+	dict_table_t*	sys_fields;
+	dtuple_t*	entry;
+	dict_field_t*	field;
+	dfield_t*	dfield;
+	byte*		ptr;
+	ibool		index_contains_column_prefix_field	= FALSE;
+	ulint		j;
+
+	ut_ad(index && heap);
+
+	for (j = 0; j < index->n_fields; j++) {
+		if (dict_index_get_nth_field(index, j)->prefix_len > 0) {
+			index_contains_column_prefix_field = TRUE;
+			break;
+		}
+	}
+
+	field = dict_index_get_nth_field(index, i);
+
+	sys_fields = dict_sys->sys_fields;
+
+	entry = dtuple_create(heap, 3 + DATA_N_SYS_COLS);
+
+	dict_table_copy_types(entry, sys_fields);
+
+	/* 0: INDEX_ID -----------------------*/
+	dfield = dtuple_get_nth_field(entry, 0);
+
+	ptr = mem_heap_alloc(heap, 8);
+	mach_write_to_8(ptr, index->id);
+
+	dfield_set_data(dfield, ptr, 8);
+	/* 1: POS + PREFIX LENGTH ----------------------------*/
+
+	dfield = dtuple_get_nth_field(entry, 1);
+
+	ptr = mem_heap_alloc(heap, 4);
+
+	if (index_contains_column_prefix_field) {
+		/* If there are column prefix fields in the index, then
+		we store the number of the field to the 2 HIGH bytes
+		and the prefix length to the 2 low bytes, */
+
+		mach_write_to_4(ptr, (i << 16) + field->prefix_len);
+	} else {
+		/* Else we store the number of the field to the 2 LOW bytes.
+		This is to keep the storage format compatible with
+		InnoDB versions < 4.0.14. */
+
+		mach_write_to_4(ptr, i);
+	}
+
+	dfield_set_data(dfield, ptr, 4);
+	/* 4: COL_NAME -------------------------*/
+	dfield = dtuple_get_nth_field(entry, 2);
+
+	dfield_set_data(dfield, field->name,
+			ut_strlen(field->name));
+	/*---------------------------------*/
+
+	return(entry);
+}
+
+/*********************************************************************
+Creates the tuple with which the index entry is searched for writing the index
+tree root page number, if such a tree is created. */
+static
+dtuple_t*
+dict_create_search_tuple(
+/*=====================*/
+				/* out: the tuple for search */
+	const dtuple_t*	tuple,	/* in: the tuple inserted in the SYS_INDEXES
+				table */
+	mem_heap_t*	heap)	/* in: memory heap from which the memory for
+				the built tuple is allocated */
+{
+	dtuple_t*	search_tuple;
+	const dfield_t*	field1;
+	dfield_t*	field2;
+
+	ut_ad(tuple && heap);
+
+	search_tuple = dtuple_create(heap, 2);
+
+	field1 = dtuple_get_nth_field(tuple, 0);
+	field2 = dtuple_get_nth_field(search_tuple, 0);
+
+	dfield_copy(field2, field1);
+
+	field1 = dtuple_get_nth_field(tuple, 1);
+	field2 = dtuple_get_nth_field(search_tuple, 1);
+
+	dfield_copy(field2, field1);
+
+	ut_ad(dtuple_validate(search_tuple));
+
+	return(search_tuple);
+}
+
+/*******************************************************************
+Builds an index definition row to insert. */
+static
+ulint
+dict_build_index_def_step(
+/*======================*/
+				/* out: DB_SUCCESS or error code */
+	que_thr_t*	thr,	/* in: query thread */
+	ind_node_t*	node)	/* in: index create node */
+{
+	dict_table_t*	table;
+	dict_index_t*	index;
+	dtuple_t*	row;
+	trx_t*		trx;
+
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+
+	trx = thr_get_trx(thr);
+
+	index = node->index;
+
+	table = dict_table_get_low(index->table_name);
+
+	if (table == NULL) {
+		return(DB_TABLE_NOT_FOUND);
+	}
+
+	trx->table_id = table->id;
+
+	node->table = table;
+
+	ut_ad((UT_LIST_GET_LEN(table->indexes) > 0)
+	      || dict_index_is_clust(index));
+
+	index->id = dict_hdr_get_new_id(DICT_HDR_INDEX_ID);
+
+	/* Inherit the space id from the table; we store all indexes of a
+	table in the same tablespace */
+
+	index->space = table->space;
+	node->page_no = FIL_NULL;
+	row = dict_create_sys_indexes_tuple(index, node->heap);
+	node->ind_row = row;
+
+	ins_node_set_new_row(node->ind_def, row);
+
+#ifdef ROW_MERGE_IS_INDEX_USABLE
+	/* Note that the index was created by this transaction. */
+	index->trx_id = trx->id;
+#endif /* ROW_MERGE_IS_INDEX_USABLE */
+
+	return(DB_SUCCESS);
+}
+
+/*******************************************************************
+Builds a field definition row to insert. */
+static
+ulint
+dict_build_field_def_step(
+/*======================*/
+				/* out: DB_SUCCESS */
+	ind_node_t*	node)	/* in: index create node */
+{
+	dict_index_t*	index;
+	dtuple_t*	row;
+
+	index = node->index;
+
+	row = dict_create_sys_fields_tuple(index, node->field_no, node->heap);
+
+	ins_node_set_new_row(node->field_def, row);
+
+	return(DB_SUCCESS);
+}
+
+/*******************************************************************
+Creates an index tree for the index if it is not a member of a cluster. */
+static
+ulint
+dict_create_index_tree_step(
+/*========================*/
+				/* out: DB_SUCCESS or DB_OUT_OF_FILE_SPACE */
+	ind_node_t*	node)	/* in: index create node */
+{
+	dict_index_t*	index;
+	dict_table_t*	sys_indexes;
+	dict_table_t*	table;
+	dtuple_t*	search_tuple;
+	btr_pcur_t	pcur;
+	mtr_t		mtr;
+
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+
+	index = node->index;
+	table = node->table;
+
+	sys_indexes = dict_sys->sys_indexes;
+
+	/* Run a mini-transaction in which the index tree is allocated for
+	the index and its root address is written to the index entry in
+	sys_indexes */
+
+	mtr_start(&mtr);
+
+	search_tuple = dict_create_search_tuple(node->ind_row, node->heap);
+
+	btr_pcur_open(UT_LIST_GET_FIRST(sys_indexes->indexes),
+		      search_tuple, PAGE_CUR_L, BTR_MODIFY_LEAF,
+		      &pcur, &mtr);
+
+	btr_pcur_move_to_next_user_rec(&pcur, &mtr);
+
+	node->page_no = btr_create(index->type, index->space,
+				   dict_table_zip_size(index->table),
+				   index->id, index, &mtr);
+	/* printf("Created a new index tree in space %lu root page %lu\n",
+	index->space, index->page_no); */
+
+	page_rec_write_index_page_no(btr_pcur_get_rec(&pcur),
+				     DICT_SYS_INDEXES_PAGE_NO_FIELD,
+				     node->page_no, &mtr);
+	btr_pcur_close(&pcur);
+	mtr_commit(&mtr);
+
+	if (node->page_no == FIL_NULL) {
+
+		return(DB_OUT_OF_FILE_SPACE);
+	}
+
+	return(DB_SUCCESS);
+}
+
+/***********************************************************************
+Drops the index tree associated with a row in SYS_INDEXES table. */
+UNIV_INTERN
+void
+dict_drop_index_tree(
+/*=================*/
+	rec_t*	rec,	/* in/out: record in the clustered index
+			of SYS_INDEXES table */
+	mtr_t*	mtr)	/* in: mtr having the latch on the record page */
+{
+	ulint		root_page_no;
+	ulint		space;
+	ulint		zip_size;
+	const byte*	ptr;
+	ulint		len;
+
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+	ut_a(!dict_table_is_comp(dict_sys->sys_indexes));
+	ptr = rec_get_nth_field_old(rec, DICT_SYS_INDEXES_PAGE_NO_FIELD, &len);
+
+	ut_ad(len == 4);
+
+	root_page_no = mtr_read_ulint(ptr, MLOG_4BYTES, mtr);
+
+	if (root_page_no == FIL_NULL) {
+		/* The tree has already been freed */
+
+		return;
+	}
+
+	ptr = rec_get_nth_field_old(rec,
+				    DICT_SYS_INDEXES_SPACE_NO_FIELD, &len);
+
+	ut_ad(len == 4);
+
+	space = mtr_read_ulint(ptr, MLOG_4BYTES, mtr);
+	zip_size = fil_space_get_zip_size(space);
+
+	if (UNIV_UNLIKELY(zip_size == ULINT_UNDEFINED)) {
+		/* It is a single table tablespace and the .ibd file is
+		missing: do nothing */
+
+		return;
+	}
+
+	/* We free all the pages but the root page first; this operation
+	may span several mini-transactions */
+
+	btr_free_but_not_root(space, zip_size, root_page_no);
+
+	/* Then we free the root page in the same mini-transaction where
+	we write FIL_NULL to the appropriate field in the SYS_INDEXES
+	record: this mini-transaction marks the B-tree totally freed */
+
+	/* printf("Dropping index tree in space %lu root page %lu\n", space,
+	root_page_no); */
+	btr_free_root(space, zip_size, root_page_no, mtr);
+
+	page_rec_write_index_page_no(rec,
+				     DICT_SYS_INDEXES_PAGE_NO_FIELD,
+				     FIL_NULL, mtr);
+}
+
+/***********************************************************************
+Truncates the index tree associated with a row in SYS_INDEXES table. */
+UNIV_INTERN
+ulint
+dict_truncate_index_tree(
+/*=====================*/
+				/* out: new root page number, or
+				FIL_NULL on failure */
+	dict_table_t*	table,	/* in: the table the index belongs to */
+	ulint		space,	/* in: 0=truncate,
+				nonzero=create the index tree in the
+				given tablespace */
+	btr_pcur_t*	pcur,	/* in/out: persistent cursor pointing to
+				record in the clustered index of
+				SYS_INDEXES table. The cursor may be
+				repositioned in this call. */
+	mtr_t*		mtr)	/* in: mtr having the latch
+				on the record page. The mtr may be
+				committed and restarted in this call. */
+{
+	ulint		root_page_no;
+	ibool		drop = !space;
+	ulint		zip_size;
+	ulint		type;
+	dulint		index_id;
+	rec_t*		rec;
+	const byte*	ptr;
+	ulint		len;
+	dict_index_t*	index;
+
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+	ut_a(!dict_table_is_comp(dict_sys->sys_indexes));
+	rec = btr_pcur_get_rec(pcur);
+	ptr = rec_get_nth_field_old(rec, DICT_SYS_INDEXES_PAGE_NO_FIELD, &len);
+
+	ut_ad(len == 4);
+
+	root_page_no = mtr_read_ulint(ptr, MLOG_4BYTES, mtr);
+
+	if (drop && root_page_no == FIL_NULL) {
+		/* The tree has been freed. */
+
+		ut_print_timestamp(stderr);
+		fprintf(stderr, "  InnoDB: Trying to TRUNCATE"
+			" a missing index of table %s!\n", table->name);
+		drop = FALSE;
+	}
+
+	ptr = rec_get_nth_field_old(rec,
+				    DICT_SYS_INDEXES_SPACE_NO_FIELD, &len);
+
+	ut_ad(len == 4);
+
+	if (drop) {
+		space = mtr_read_ulint(ptr, MLOG_4BYTES, mtr);
+	}
+
+	zip_size = fil_space_get_zip_size(space);
+
+	if (UNIV_UNLIKELY(zip_size == ULINT_UNDEFINED)) {
+		/* It is a single table tablespace and the .ibd file is
+		missing: do nothing */
+
+		ut_print_timestamp(stderr);
+		fprintf(stderr, "  InnoDB: Trying to TRUNCATE"
+			" a missing .ibd file of table %s!\n", table->name);
+		return(FIL_NULL);
+	}
+
+	ptr = rec_get_nth_field_old(rec,
+				    DICT_SYS_INDEXES_TYPE_FIELD, &len);
+	ut_ad(len == 4);
+	type = mach_read_from_4(ptr);
+
+	ptr = rec_get_nth_field_old(rec, 1, &len);
+	ut_ad(len == 8);
+	index_id = mach_read_from_8(ptr);
+
+	if (!drop) {
+
+		goto create;
+	}
+
+	/* We free all the pages but the root page first; this operation
+	may span several mini-transactions */
+
+	btr_free_but_not_root(space, zip_size, root_page_no);
+
+	/* Then we free the root page in the same mini-transaction where
+	we create the b-tree and write its new root page number to the
+	appropriate field in the SYS_INDEXES record: this mini-transaction
+	marks the B-tree totally truncated */
+
+	btr_page_get(space, zip_size, root_page_no, RW_X_LATCH, mtr);
+
+	btr_free_root(space, zip_size, root_page_no, mtr);
+create:
+	/* We will temporarily write FIL_NULL to the PAGE_NO field
+	in SYS_INDEXES, so that the database will not get into an
+	inconsistent state in case it crashes between the mtr_commit()
+	below and the following mtr_commit() call. */
+	page_rec_write_index_page_no(rec, DICT_SYS_INDEXES_PAGE_NO_FIELD,
+				     FIL_NULL, mtr);
+
+	/* We will need to commit the mini-transaction in order to avoid
+	deadlocks in the btr_create() call, because otherwise we would
+	be freeing and allocating pages in the same mini-transaction. */
+	btr_pcur_store_position(pcur, mtr);
+	mtr_commit(mtr);
+
+	mtr_start(mtr);
+	btr_pcur_restore_position(BTR_MODIFY_LEAF, pcur, mtr);
+
+	/* Find the index corresponding to this SYS_INDEXES record. */
+	for (index = UT_LIST_GET_FIRST(table->indexes);
+	     index;
+	     index = UT_LIST_GET_NEXT(indexes, index)) {
+		if (!ut_dulint_cmp(index->id, index_id)) {
+			root_page_no = btr_create(type, space, zip_size,
+						  index_id, index, mtr);
+			index->page = (unsigned int) root_page_no;
+			return(root_page_no);
+		}
+	}
+
+	ut_print_timestamp(stderr);
+	fprintf(stderr,
+		"  InnoDB: Index %lu %lu of table %s is missing\n"
+		"InnoDB: from the data dictionary during TRUNCATE!\n",
+		ut_dulint_get_high(index_id),
+		ut_dulint_get_low(index_id),
+		table->name);
+
+	return(FIL_NULL);
+}
+
+/*************************************************************************
+Creates a table create graph. */
+UNIV_INTERN
+tab_node_t*
+tab_create_graph_create(
+/*====================*/
+				/* out, own: table create node */
+	dict_table_t*	table,	/* in: table to create, built as a memory data
+				structure */
+	mem_heap_t*	heap)	/* in: heap where created */
+{
+	tab_node_t*	node;
+
+	node = mem_heap_alloc(heap, sizeof(tab_node_t));
+
+	node->common.type = QUE_NODE_CREATE_TABLE;
+
+	node->table = table;
+
+	node->state = TABLE_BUILD_TABLE_DEF;
+	node->heap = mem_heap_create(256);
+
+	node->tab_def = ins_node_create(INS_DIRECT, dict_sys->sys_tables,
+					heap);
+	node->tab_def->common.parent = node;
+
+	node->col_def = ins_node_create(INS_DIRECT, dict_sys->sys_columns,
+					heap);
+	node->col_def->common.parent = node;
+
+	node->commit_node = commit_node_create(heap);
+	node->commit_node->common.parent = node;
+
+	return(node);
+}
+
+/*************************************************************************
+Creates an index create graph. */
+UNIV_INTERN
+ind_node_t*
+ind_create_graph_create(
+/*====================*/
+				/* out, own: index create node */
+	dict_index_t*	index,	/* in: index to create, built as a memory data
+				structure */
+	mem_heap_t*	heap)	/* in: heap where created */
+{
+	ind_node_t*	node;
+
+	node = mem_heap_alloc(heap, sizeof(ind_node_t));
+
+	node->common.type = QUE_NODE_CREATE_INDEX;
+
+	node->index = index;
+
+	node->state = INDEX_BUILD_INDEX_DEF;
+	node->page_no = FIL_NULL;
+	node->heap = mem_heap_create(256);
+
+	node->ind_def = ins_node_create(INS_DIRECT,
+					dict_sys->sys_indexes, heap);
+	node->ind_def->common.parent = node;
+
+	node->field_def = ins_node_create(INS_DIRECT,
+					  dict_sys->sys_fields, heap);
+	node->field_def->common.parent = node;
+
+	node->commit_node = commit_node_create(heap);
+	node->commit_node->common.parent = node;
+
+	return(node);
+}
+
+/***************************************************************
+Creates a table. This is a high-level function used in SQL execution graphs. */
+UNIV_INTERN
+que_thr_t*
+dict_create_table_step(
+/*===================*/
+				/* out: query thread to run next or NULL */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	tab_node_t*	node;
+	ulint		err	= DB_ERROR;
+	trx_t*		trx;
+
+	ut_ad(thr);
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+
+	trx = thr_get_trx(thr);
+
+	node = thr->run_node;
+
+	ut_ad(que_node_get_type(node) == QUE_NODE_CREATE_TABLE);
+
+	if (thr->prev_node == que_node_get_parent(node)) {
+		node->state = TABLE_BUILD_TABLE_DEF;
+	}
+
+	if (node->state == TABLE_BUILD_TABLE_DEF) {
+
+		/* DO THE CHECKS OF THE CONSISTENCY CONSTRAINTS HERE */
+
+		err = dict_build_table_def_step(thr, node);
+
+		if (err != DB_SUCCESS) {
+
+			goto function_exit;
+		}
+
+		node->state = TABLE_BUILD_COL_DEF;
+		node->col_no = 0;
+
+		thr->run_node = node->tab_def;
+
+		return(thr);
+	}
+
+	if (node->state == TABLE_BUILD_COL_DEF) {
+
+		if (node->col_no < (node->table)->n_def) {
+
+			err = dict_build_col_def_step(node);
+
+			if (err != DB_SUCCESS) {
+
+				goto function_exit;
+			}
+
+			node->col_no++;
+
+			thr->run_node = node->col_def;
+
+			return(thr);
+		} else {
+			node->state = TABLE_COMMIT_WORK;
+		}
+	}
+
+	if (node->state == TABLE_COMMIT_WORK) {
+
+		/* Table was correctly defined: do NOT commit the transaction
+		(CREATE TABLE does NOT do an implicit commit of the current
+		transaction) */
+
+		node->state = TABLE_ADD_TO_CACHE;
+
+		/* thr->run_node = node->commit_node;
+
+		return(thr); */
+	}
+
+	if (node->state == TABLE_ADD_TO_CACHE) {
+
+		dict_table_add_to_cache(node->table, node->heap);
+
+		err = DB_SUCCESS;
+	}
+
+function_exit:
+	trx->error_state = err;
+
+	if (err == DB_SUCCESS) {
+		/* Ok: do nothing */
+
+	} else if (err == DB_LOCK_WAIT) {
+
+		return(NULL);
+	} else {
+		/* SQL error detected */
+
+		return(NULL);
+	}
+
+	thr->run_node = que_node_get_parent(node);
+
+	return(thr);
+}
+
+/***************************************************************
+Creates an index. This is a high-level function used in SQL execution
+graphs. */
+UNIV_INTERN
+que_thr_t*
+dict_create_index_step(
+/*===================*/
+				/* out: query thread to run next or NULL */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	ind_node_t*	node;
+	ulint		err	= DB_ERROR;
+	trx_t*		trx;
+
+	ut_ad(thr);
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+
+	trx = thr_get_trx(thr);
+
+	node = thr->run_node;
+
+	ut_ad(que_node_get_type(node) == QUE_NODE_CREATE_INDEX);
+
+	if (thr->prev_node == que_node_get_parent(node)) {
+		node->state = INDEX_BUILD_INDEX_DEF;
+	}
+
+	if (node->state == INDEX_BUILD_INDEX_DEF) {
+		/* DO THE CHECKS OF THE CONSISTENCY CONSTRAINTS HERE */
+		err = dict_build_index_def_step(thr, node);
+
+		if (err != DB_SUCCESS) {
+
+			goto function_exit;
+		}
+
+		node->state = INDEX_BUILD_FIELD_DEF;
+		node->field_no = 0;
+
+		thr->run_node = node->ind_def;
+
+		return(thr);
+	}
+
+	if (node->state == INDEX_BUILD_FIELD_DEF) {
+
+		if (node->field_no < (node->index)->n_fields) {
+
+			err = dict_build_field_def_step(node);
+
+			if (err != DB_SUCCESS) {
+
+				goto function_exit;
+			}
+
+			node->field_no++;
+
+			thr->run_node = node->field_def;
+
+			return(thr);
+		} else {
+			node->state = INDEX_ADD_TO_CACHE;
+		}
+	}
+
+	if (node->state == INDEX_ADD_TO_CACHE) {
+
+		dulint	index_id = node->index->id;
+
+		err = dict_index_add_to_cache(node->table, node->index,
+					      FIL_NULL, TRUE);
+
+		node->index = dict_index_get_if_in_cache_low(index_id);
+		ut_a(!node->index == (err != DB_SUCCESS));
+
+		if (err != DB_SUCCESS) {
+
+			goto function_exit;
+		}
+
+		node->state = INDEX_CREATE_INDEX_TREE;
+	}
+
+	if (node->state == INDEX_CREATE_INDEX_TREE) {
+
+		err = dict_create_index_tree_step(node);
+
+		if (err != DB_SUCCESS) {
+			dict_index_remove_from_cache(node->table, node->index);
+			node->index = NULL;
+
+			goto function_exit;
+		}
+
+		node->index->page = node->page_no;
+		node->state = INDEX_COMMIT_WORK;
+	}
+
+	if (node->state == INDEX_COMMIT_WORK) {
+
+		/* Index was correctly defined: do NOT commit the transaction
+		(CREATE INDEX does NOT currently do an implicit commit of
+		the current transaction) */
+
+		node->state = INDEX_CREATE_INDEX_TREE;
+
+		/* thr->run_node = node->commit_node;
+
+		return(thr); */
+	}
+
+function_exit:
+	trx->error_state = err;
+
+	if (err == DB_SUCCESS) {
+		/* Ok: do nothing */
+
+	} else if (err == DB_LOCK_WAIT) {
+
+		return(NULL);
+	} else {
+		/* SQL error detected */
+
+		return(NULL);
+	}
+
+	thr->run_node = que_node_get_parent(node);
+
+	return(thr);
+}
+
+#ifndef UNIV_HOTBACKUP
+/********************************************************************
+Creates the foreign key constraints system tables inside InnoDB
+at database creation or database start if they are not found or are
+not of the right form. */
+UNIV_INTERN
+ulint
+dict_create_or_check_foreign_constraint_tables(void)
+/*================================================*/
+				/* out: DB_SUCCESS or error code */
+{
+	dict_table_t*	table1;
+	dict_table_t*	table2;
+	ulint		error;
+	trx_t*		trx;
+
+	mutex_enter(&(dict_sys->mutex));
+
+	table1 = dict_table_get_low("SYS_FOREIGN");
+	table2 = dict_table_get_low("SYS_FOREIGN_COLS");
+
+	if (table1 && table2
+	    && UT_LIST_GET_LEN(table1->indexes) == 3
+	    && UT_LIST_GET_LEN(table2->indexes) == 1) {
+
+		/* Foreign constraint system tables have already been
+		created, and they are ok */
+
+		mutex_exit(&(dict_sys->mutex));
+
+		return(DB_SUCCESS);
+	}
+
+	mutex_exit(&(dict_sys->mutex));
+
+	trx = trx_allocate_for_mysql();
+
+	trx->op_info = "creating foreign key sys tables";
+
+	row_mysql_lock_data_dictionary(trx);
+
+	if (table1) {
+		fprintf(stderr,
+			"InnoDB: dropping incompletely created"
+			" SYS_FOREIGN table\n");
+		row_drop_table_for_mysql("SYS_FOREIGN", trx, TRUE);
+	}
+
+	if (table2) {
+		fprintf(stderr,
+			"InnoDB: dropping incompletely created"
+			" SYS_FOREIGN_COLS table\n");
+		row_drop_table_for_mysql("SYS_FOREIGN_COLS", trx, TRUE);
+	}
+
+	fprintf(stderr,
+		"InnoDB: Creating foreign key constraint system tables\n");
+
+	/* NOTE: in dict_load_foreigns we use the fact that
+	there are 2 secondary indexes on SYS_FOREIGN, and they
+	are defined just like below */
+
+	/* NOTE: when designing InnoDB's foreign key support in 2001, we made
+	an error and made the table names and the foreign key id of type
+	'CHAR' (internally, really a VARCHAR). We should have made the type
+	VARBINARY, like in other InnoDB system tables, to get a clean
+	design. */
+
+	error = que_eval_sql(NULL,
+			     "PROCEDURE CREATE_FOREIGN_SYS_TABLES_PROC () IS\n"
+			     "BEGIN\n"
+			     "CREATE TABLE\n"
+			     "SYS_FOREIGN(ID CHAR, FOR_NAME CHAR,"
+			     " REF_NAME CHAR, N_COLS INT);\n"
+			     "CREATE UNIQUE CLUSTERED INDEX ID_IND"
+			     " ON SYS_FOREIGN (ID);\n"
+			     "CREATE INDEX FOR_IND"
+			     " ON SYS_FOREIGN (FOR_NAME);\n"
+			     "CREATE INDEX REF_IND"
+			     " ON SYS_FOREIGN (REF_NAME);\n"
+			     "CREATE TABLE\n"
+			     "SYS_FOREIGN_COLS(ID CHAR, POS INT,"
+			     " FOR_COL_NAME CHAR, REF_COL_NAME CHAR);\n"
+			     "CREATE UNIQUE CLUSTERED INDEX ID_IND"
+			     " ON SYS_FOREIGN_COLS (ID, POS);\n"
+			     "END;\n"
+			     , FALSE, trx);
+
+	if (error != DB_SUCCESS) {
+		fprintf(stderr, "InnoDB: error %lu in creation\n",
+			(ulong) error);
+
+		ut_a(error == DB_OUT_OF_FILE_SPACE
+		     || error == DB_TOO_MANY_CONCURRENT_TRXS);
+
+		fprintf(stderr,
+			"InnoDB: creation failed\n"
+			"InnoDB: tablespace is full\n"
+			"InnoDB: dropping incompletely created"
+			" SYS_FOREIGN tables\n");
+
+		row_drop_table_for_mysql("SYS_FOREIGN", trx, TRUE);
+		row_drop_table_for_mysql("SYS_FOREIGN_COLS", trx, TRUE);
+
+		error = DB_MUST_GET_MORE_FILE_SPACE;
+	}
+
+	trx_commit_for_mysql(trx);
+
+	row_mysql_unlock_data_dictionary(trx);
+
+	trx_free_for_mysql(trx);
+
+	if (error == DB_SUCCESS) {
+		fprintf(stderr,
+			"InnoDB: Foreign key constraint system tables"
+			" created\n");
+	}
+
+	return(error);
+}
+
+/********************************************************************
+Evaluate the given foreign key SQL statement. */
+static
+ulint
+dict_foreign_eval_sql(
+/*==================*/
+				/* out: error code or DB_SUCCESS */
+	pars_info_t*	info,	/* in: info struct, or NULL */
+	const char*	sql,	/* in: SQL string to evaluate */
+	dict_table_t*	table,	/* in: table */
+	dict_foreign_t*	foreign,/* in: foreign */
+	trx_t*		trx)	/* in: transaction */
+{
+	ulint		error;
+	FILE*		ef	= dict_foreign_err_file;
+
+	error = que_eval_sql(info, sql, FALSE, trx);
+
+	if (error == DB_DUPLICATE_KEY) {
+		mutex_enter(&dict_foreign_err_mutex);
+		rewind(ef);
+		ut_print_timestamp(ef);
+		fputs(" Error in foreign key constraint creation for table ",
+		      ef);
+		ut_print_name(ef, trx, TRUE, table->name);
+		fputs(".\nA foreign key constraint of name ", ef);
+		ut_print_name(ef, trx, TRUE, foreign->id);
+		fputs("\nalready exists."
+		      " (Note that internally InnoDB adds 'databasename'\n"
+		      "in front of the user-defined constraint name.)\n"
+		      "Note that InnoDB's FOREIGN KEY system tables store\n"
+		      "constraint names as case-insensitive, with the\n"
+		      "MySQL standard latin1_swedish_ci collation. If you\n"
+		      "create tables or databases whose names differ only in\n"
+		      "the character case, then collisions in constraint\n"
+		      "names can occur. Workaround: name your constraints\n"
+		      "explicitly with unique names.\n",
+		      ef);
+
+		mutex_exit(&dict_foreign_err_mutex);
+
+		return(error);
+	}
+
+	if (error != DB_SUCCESS) {
+		fprintf(stderr,
+			"InnoDB: Foreign key constraint creation failed:\n"
+			"InnoDB: internal error number %lu\n", (ulong) error);
+
+		mutex_enter(&dict_foreign_err_mutex);
+		ut_print_timestamp(ef);
+		fputs(" Internal error in foreign key constraint creation"
+		      " for table ", ef);
+		ut_print_name(ef, trx, TRUE, table->name);
+		fputs(".\n"
+		      "See the MySQL .err log in the datadir"
+		      " for more information.\n", ef);
+		mutex_exit(&dict_foreign_err_mutex);
+
+		return(error);
+	}
+
+	return(DB_SUCCESS);
+}
+
+/************************************************************************
+Add a single foreign key field definition to the data dictionary tables in
+the database.  */
+static
+ulint
+dict_create_add_foreign_field_to_dictionary(
+/*========================================*/
+					/* out: error code or DB_SUCCESS */
+	ulint		field_nr,	/* in: foreign field number */
+	dict_table_t*	table,		/* in: table */
+	dict_foreign_t*	foreign,	/* in: foreign */
+	trx_t*		trx)		/* in: transaction */
+{
+	pars_info_t*	info = pars_info_create();
+
+	pars_info_add_str_literal(info, "id", foreign->id);
+
+	pars_info_add_int4_literal(info, "pos", field_nr);
+
+	pars_info_add_str_literal(info, "for_col_name",
+				  foreign->foreign_col_names[field_nr]);
+
+	pars_info_add_str_literal(info, "ref_col_name",
+				  foreign->referenced_col_names[field_nr]);
+
+	return(dict_foreign_eval_sql(
+		       info,
+		       "PROCEDURE P () IS\n"
+		       "BEGIN\n"
+		       "INSERT INTO SYS_FOREIGN_COLS VALUES"
+		       "(:id, :pos, :for_col_name, :ref_col_name);\n"
+		       "END;\n",
+		       table, foreign, trx));
+}
+
+/************************************************************************
+Add a single foreign key definition to the data dictionary tables in the
+database. We also generate names to constraints that were not named by the
+user. A generated constraint has a name of the format
+databasename/tablename_ibfk_<number>, where the numbers start from 1, and
+are given locally for this table, that is, the number is not global, as in
+the old format constraints < 4.0.18 it used to be. */
+static
+ulint
+dict_create_add_foreign_to_dictionary(
+/*==================================*/
+				/* out: error code or DB_SUCCESS */
+	ulint*		id_nr,	/* in/out: number to use in id generation;
+				incremented if used */
+	dict_table_t*	table,	/* in: table */
+	dict_foreign_t*	foreign,/* in: foreign */
+	trx_t*		trx)	/* in: transaction */
+{
+	ulint		error;
+	ulint		i;
+
+	pars_info_t*	info = pars_info_create();
+
+	if (foreign->id == NULL) {
+		/* Generate a new constraint id */
+		ulint	namelen	= strlen(table->name);
+		char*	id	= mem_heap_alloc(foreign->heap, namelen + 20);
+		/* no overflow if number < 1e13 */
+		sprintf(id, "%s_ibfk_%lu", table->name, (ulong) (*id_nr)++);
+		foreign->id = id;
+	}
+
+	pars_info_add_str_literal(info, "id", foreign->id);
+
+	pars_info_add_str_literal(info, "for_name", table->name);
+
+	pars_info_add_str_literal(info, "ref_name",
+				  foreign->referenced_table_name);
+
+	pars_info_add_int4_literal(info, "n_cols",
+				   foreign->n_fields + (foreign->type << 24));
+
+	error = dict_foreign_eval_sql(info,
+				      "PROCEDURE P () IS\n"
+				      "BEGIN\n"
+				      "INSERT INTO SYS_FOREIGN VALUES"
+				      "(:id, :for_name, :ref_name, :n_cols);\n"
+				      "END;\n"
+				      , table, foreign, trx);
+
+	if (error != DB_SUCCESS) {
+
+		return(error);
+	}
+
+	for (i = 0; i < foreign->n_fields; i++) {
+		error = dict_create_add_foreign_field_to_dictionary(
+			i, table, foreign, trx);
+
+		if (error != DB_SUCCESS) {
+
+			return(error);
+		}
+	}
+
+	error = dict_foreign_eval_sql(NULL,
+				      "PROCEDURE P () IS\n"
+				      "BEGIN\n"
+				      "COMMIT WORK;\n"
+				      "END;\n"
+				      , table, foreign, trx);
+
+	return(error);
+}
+
+/************************************************************************
+Adds foreign key definitions to data dictionary tables in the database. */
+UNIV_INTERN
+ulint
+dict_create_add_foreigns_to_dictionary(
+/*===================================*/
+				/* out: error code or DB_SUCCESS */
+	ulint		start_id,/* in: if we are actually doing ALTER TABLE
+				ADD CONSTRAINT, we want to generate constraint
+				numbers which are bigger than in the table so
+				far; we number the constraints from
+				start_id + 1 up; start_id should be set to 0 if
+				we are creating a new table, or if the table
+				so far has no constraints for which the name
+				was generated here */
+	dict_table_t*	table,	/* in: table */
+	trx_t*		trx)	/* in: transaction */
+{
+	dict_foreign_t*	foreign;
+	ulint		number	= start_id + 1;
+	ulint		error;
+
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+
+	if (NULL == dict_table_get_low("SYS_FOREIGN")) {
+		fprintf(stderr,
+			"InnoDB: table SYS_FOREIGN not found"
+			" in internal data dictionary\n");
+
+		return(DB_ERROR);
+	}
+
+	for (foreign = UT_LIST_GET_FIRST(table->foreign_list);
+	     foreign;
+	     foreign = UT_LIST_GET_NEXT(foreign_list, foreign)) {
+
+		error = dict_create_add_foreign_to_dictionary(&number, table,
+							      foreign, trx);
+
+		if (error != DB_SUCCESS) {
+
+			return(error);
+		}
+	}
+
+	return(DB_SUCCESS);
+}
+#endif /* !UNIV_HOTBACKUP */
diff --git a/storage/xtradb/dict/dict0dict.c b/storage/xtradb/dict/dict0dict.c
new file mode 100644
index 00000000000..134be1dadba
--- /dev/null
+++ b/storage/xtradb/dict/dict0dict.c
@@ -0,0 +1,4747 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/**********************************************************************
+Data dictionary system
+
+Created 1/8/1996 Heikki Tuuri
+***********************************************************************/
+
+#include "dict0dict.h"
+
+#ifdef UNIV_NONINL
+#include "dict0dict.ic"
+#endif
+
+#include "buf0buf.h"
+#include "data0type.h"
+#include "mach0data.h"
+#include "dict0boot.h"
+#include "dict0mem.h"
+#include "dict0crea.h"
+#include "trx0undo.h"
+#include "btr0btr.h"
+#include "btr0cur.h"
+#include "btr0sea.h"
+#include "page0zip.h"
+#include "page0page.h"
+#include "pars0pars.h"
+#include "pars0sym.h"
+#include "que0que.h"
+#include "rem0cmp.h"
+#include "row0merge.h"
+#ifndef UNIV_HOTBACKUP
+# include "m_ctype.h" /* my_isspace() */
+# include "ha_prototypes.h" /* innobase_strcasecmp() */
+#endif /* !UNIV_HOTBACKUP */
+
+#include <ctype.h>
+
+/* the dictionary system */
+UNIV_INTERN dict_sys_t*	dict_sys	= NULL;
+
+/* table create, drop, etc. reserve this in X-mode; implicit or
+backround operations purge, rollback, foreign key checks reserve this
+in S-mode; we cannot trust that MySQL protects implicit or background
+operations a table drop since MySQL does not know of them; therefore
+we need this; NOTE: a transaction which reserves this must keep book
+on the mode in trx->dict_operation_lock_mode */
+UNIV_INTERN rw_lock_t	dict_operation_lock;
+
+#define	DICT_HEAP_SIZE		100	/* initial memory heap size when
+					creating a table or index object */
+#define DICT_POOL_PER_TABLE_HASH 512	/* buffer pool max size per table
+					hash table fixed size in bytes */
+#define DICT_POOL_PER_VARYING	4	/* buffer pool max size per data
+					dictionary varying size in bytes */
+
+/* Identifies generated InnoDB foreign key names */
+static char	dict_ibfk[] = "_ibfk_";
+
+/***********************************************************************
+Tries to find column names for the index and sets the col field of the
+index. */
+static
+void
+dict_index_find_cols(
+/*=================*/
+	dict_table_t*	table,	/* in: table */
+	dict_index_t*	index);	/* in: index */
+/***********************************************************************
+Builds the internal dictionary cache representation for a clustered
+index, containing also system fields not defined by the user. */
+static
+dict_index_t*
+dict_index_build_internal_clust(
+/*============================*/
+					/* out, own: the internal
+					representation of the clustered
+					index */
+	const dict_table_t*	table,	/* in: table */
+	dict_index_t*		index);	/* in: user representation of
+					a clustered index */
+/***********************************************************************
+Builds the internal dictionary cache representation for a non-clustered
+index, containing also system fields not defined by the user. */
+static
+dict_index_t*
+dict_index_build_internal_non_clust(
+/*================================*/
+					/* out, own: the internal
+					representation of the non-clustered
+					index */
+	const dict_table_t*	table,	/* in: table */
+	dict_index_t*		index);	/* in: user representation of
+					a non-clustered index */
+/**************************************************************************
+Removes a foreign constraint struct from the dictionary cache. */
+static
+void
+dict_foreign_remove_from_cache(
+/*===========================*/
+	dict_foreign_t*	foreign);	/* in, own: foreign constraint */
+/**************************************************************************
+Prints a column data. */
+static
+void
+dict_col_print_low(
+/*===============*/
+	const dict_table_t*	table,	/* in: table */
+	const dict_col_t*	col);	/* in: column */
+/**************************************************************************
+Prints an index data. */
+static
+void
+dict_index_print_low(
+/*=================*/
+	dict_index_t*	index);	/* in: index */
+/**************************************************************************
+Prints a field data. */
+static
+void
+dict_field_print_low(
+/*=================*/
+	dict_field_t*	field);	/* in: field */
+/*************************************************************************
+Frees a foreign key struct. */
+static
+void
+dict_foreign_free(
+/*==============*/
+	dict_foreign_t*	foreign);	/* in, own: foreign key struct */
+
+/* Stream for storing detailed information about the latest foreign key
+and unique key errors */
+UNIV_INTERN FILE*	dict_foreign_err_file		= NULL;
+/* mutex protecting the foreign and unique error buffers */
+UNIV_INTERN mutex_t	dict_foreign_err_mutex;
+
+#ifndef UNIV_HOTBACKUP
+/**********************************************************************
+Makes all characters in a NUL-terminated UTF-8 string lower case. */
+UNIV_INTERN
+void
+dict_casedn_str(
+/*============*/
+	char*	a)	/* in/out: string to put in lower case */
+{
+	innobase_casedn_str(a);
+}
+#endif /* !UNIV_HOTBACKUP */
+
+/************************************************************************
+Checks if the database name in two table names is the same. */
+UNIV_INTERN
+ibool
+dict_tables_have_same_db(
+/*=====================*/
+				/* out: TRUE if same db name */
+	const char*	name1,	/* in: table name in the form
+				dbname '/' tablename */
+	const char*	name2)	/* in: table name in the form
+				dbname '/' tablename */
+{
+	for (; *name1 == *name2; name1++, name2++) {
+		if (*name1 == '/') {
+			return(TRUE);
+		}
+		ut_a(*name1); /* the names must contain '/' */
+	}
+	return(FALSE);
+}
+
+/************************************************************************
+Return the end of table name where we have removed dbname and '/'. */
+UNIV_INTERN
+const char*
+dict_remove_db_name(
+/*================*/
+				/* out: table name */
+	const char*	name)	/* in: table name in the form
+				dbname '/' tablename */
+{
+	const char*	s = strchr(name, '/');
+	ut_a(s);
+
+	return(s + 1);
+}
+
+/************************************************************************
+Get the database name length in a table name. */
+UNIV_INTERN
+ulint
+dict_get_db_name_len(
+/*=================*/
+				/* out: database name length */
+	const char*	name)	/* in: table name in the form
+				dbname '/' tablename */
+{
+	const char*	s;
+	s = strchr(name, '/');
+	ut_a(s);
+	return(s - name);
+}
+
+/************************************************************************
+Reserves the dictionary system mutex for MySQL. */
+UNIV_INTERN
+void
+dict_mutex_enter_for_mysql(void)
+/*============================*/
+{
+	mutex_enter(&(dict_sys->mutex));
+}
+
+/************************************************************************
+Releases the dictionary system mutex for MySQL. */
+UNIV_INTERN
+void
+dict_mutex_exit_for_mysql(void)
+/*===========================*/
+{
+	mutex_exit(&(dict_sys->mutex));
+}
+
+/************************************************************************
+Decrements the count of open MySQL handles to a table. */
+UNIV_INTERN
+void
+dict_table_decrement_handle_count(
+/*==============================*/
+	dict_table_t*	table,		/* in/out: table */
+	ibool		dict_locked)	/* in: TRUE=data dictionary locked */
+{
+	if (!dict_locked) {
+		mutex_enter(&dict_sys->mutex);
+	}
+
+	ut_ad(mutex_own(&dict_sys->mutex));
+	ut_a(table->n_mysql_handles_opened > 0);
+
+	table->n_mysql_handles_opened--;
+
+	if (!dict_locked) {
+		mutex_exit(&dict_sys->mutex);
+	}
+}
+
+/**************************************************************************
+Returns a column's name. */
+UNIV_INTERN
+const char*
+dict_table_get_col_name(
+/*====================*/
+					/* out: column name. NOTE: not
+					guaranteed to stay valid if table is
+					modified in any way (columns added,
+					etc.). */
+	const dict_table_t*	table,	/* in: table */
+	ulint			col_nr)	/* in: column number */
+{
+	ulint		i;
+	const char*	s;
+
+	ut_ad(table);
+	ut_ad(col_nr < table->n_def);
+	ut_ad(table->magic_n == DICT_TABLE_MAGIC_N);
+
+	s = table->col_names;
+	if (s) {
+		for (i = 0; i < col_nr; i++) {
+			s += strlen(s) + 1;
+		}
+	}
+
+	return(s);
+}
+
+
+/************************************************************************
+Acquire the autoinc lock.*/
+UNIV_INTERN
+void
+dict_table_autoinc_lock(
+/*====================*/
+	dict_table_t*	table)	/* in/out: table */
+{
+	mutex_enter(&table->autoinc_mutex);
+}
+
+/************************************************************************
+Unconditionally set the autoinc counter. */
+UNIV_INTERN
+void
+dict_table_autoinc_initialize(
+/*==========================*/
+	dict_table_t*	table,	/* in/out: table */
+	ib_uint64_t	value)	/* in: next value to assign to a row */
+{
+	ut_ad(mutex_own(&table->autoinc_mutex));
+
+	table->autoinc = value;
+}
+
+/************************************************************************
+Reads the next autoinc value (== autoinc counter value), 0 if not yet
+initialized. */
+UNIV_INTERN
+ib_uint64_t
+dict_table_autoinc_read(
+/*====================*/
+					/* out: value for a new row, or 0 */
+	const dict_table_t*	table)	/* in: table */
+{
+	ut_ad(mutex_own(&table->autoinc_mutex));
+
+	return(table->autoinc);
+}
+
+/************************************************************************
+Updates the autoinc counter if the value supplied is greater than the
+current value. */
+UNIV_INTERN
+void
+dict_table_autoinc_update_if_greater(
+/*=================================*/
+
+	dict_table_t*	table,	/* in/out: table */
+	ib_uint64_t	value)	/* in: value which was assigned to a row */
+{
+	ut_ad(mutex_own(&table->autoinc_mutex));
+
+	if (value > table->autoinc) {
+
+		table->autoinc = value;
+	}
+}
+
+/************************************************************************
+Release the autoinc lock.*/
+UNIV_INTERN
+void
+dict_table_autoinc_unlock(
+/*======================*/
+	dict_table_t*	table)	/* in/out: table */
+{
+	mutex_exit(&table->autoinc_mutex);
+}
+
+/**************************************************************************
+Looks for an index with the given table and index id.
+NOTE that we do not reserve the dictionary mutex. */
+UNIV_INTERN
+dict_index_t*
+dict_index_get_on_id_low(
+/*=====================*/
+				/* out: index or NULL if not found
+				from cache */
+	dict_table_t*	table,	/* in: table */
+	dulint		id)	/* in: index id */
+{
+	dict_index_t*	index;
+
+	index = dict_table_get_first_index(table);
+
+	while (index) {
+		if (0 == ut_dulint_cmp(id, index->id)) {
+			/* Found */
+
+			return(index);
+		}
+
+		index = dict_table_get_next_index(index);
+	}
+
+	return(NULL);
+}
+
+/************************************************************************
+Looks for column n in an index. */
+UNIV_INTERN
+ulint
+dict_index_get_nth_col_pos(
+/*=======================*/
+					/* out: position in internal
+					representation of the index;
+					if not contained, returns
+					ULINT_UNDEFINED */
+	const dict_index_t*	index,	/* in: index */
+	ulint			n)	/* in: column number */
+{
+	const dict_field_t*	field;
+	const dict_col_t*	col;
+	ulint			pos;
+	ulint			n_fields;
+
+	ut_ad(index);
+	ut_ad(index->magic_n == DICT_INDEX_MAGIC_N);
+
+	col = dict_table_get_nth_col(index->table, n);
+
+	if (dict_index_is_clust(index)) {
+
+		return(dict_col_get_clust_pos(col, index));
+	}
+
+	n_fields = dict_index_get_n_fields(index);
+
+	for (pos = 0; pos < n_fields; pos++) {
+		field = dict_index_get_nth_field(index, pos);
+
+		if (col == field->col && field->prefix_len == 0) {
+
+			return(pos);
+		}
+	}
+
+	return(ULINT_UNDEFINED);
+}
+
+/************************************************************************
+Returns TRUE if the index contains a column or a prefix of that column. */
+UNIV_INTERN
+ibool
+dict_index_contains_col_or_prefix(
+/*==============================*/
+					/* out: TRUE if contains the column
+					or its prefix */
+	const dict_index_t*	index,	/* in: index */
+	ulint			n)	/* in: column number */
+{
+	const dict_field_t*	field;
+	const dict_col_t*	col;
+	ulint			pos;
+	ulint			n_fields;
+
+	ut_ad(index);
+	ut_ad(index->magic_n == DICT_INDEX_MAGIC_N);
+
+	if (dict_index_is_clust(index)) {
+
+		return(TRUE);
+	}
+
+	col = dict_table_get_nth_col(index->table, n);
+
+	n_fields = dict_index_get_n_fields(index);
+
+	for (pos = 0; pos < n_fields; pos++) {
+		field = dict_index_get_nth_field(index, pos);
+
+		if (col == field->col) {
+
+			return(TRUE);
+		}
+	}
+
+	return(FALSE);
+}
+
+/************************************************************************
+Looks for a matching field in an index. The column has to be the same. The
+column in index must be complete, or must contain a prefix longer than the
+column in index2. That is, we must be able to construct the prefix in index2
+from the prefix in index. */
+UNIV_INTERN
+ulint
+dict_index_get_nth_field_pos(
+/*=========================*/
+					/* out: position in internal
+					representation of the index;
+					if not contained, returns
+					ULINT_UNDEFINED */
+	const dict_index_t*	index,	/* in: index from which to search */
+	const dict_index_t*	index2,	/* in: index */
+	ulint			n)	/* in: field number in index2 */
+{
+	const dict_field_t*	field;
+	const dict_field_t*	field2;
+	ulint			n_fields;
+	ulint			pos;
+
+	ut_ad(index);
+	ut_ad(index->magic_n == DICT_INDEX_MAGIC_N);
+
+	field2 = dict_index_get_nth_field(index2, n);
+
+	n_fields = dict_index_get_n_fields(index);
+
+	for (pos = 0; pos < n_fields; pos++) {
+		field = dict_index_get_nth_field(index, pos);
+
+		if (field->col == field2->col
+		    && (field->prefix_len == 0
+			|| (field->prefix_len >= field2->prefix_len
+			    && field2->prefix_len != 0))) {
+
+			return(pos);
+		}
+	}
+
+	return(ULINT_UNDEFINED);
+}
+
+/**************************************************************************
+Returns a table object based on table id. */
+UNIV_INTERN
+dict_table_t*
+dict_table_get_on_id(
+/*=================*/
+				/* out: table, NULL if does not exist */
+	dulint	table_id,	/* in: table id */
+	trx_t*	trx)		/* in: transaction handle */
+{
+	dict_table_t*	table;
+
+	if (ut_dulint_cmp(table_id, DICT_FIELDS_ID) <= 0
+	    || trx->dict_operation_lock_mode == RW_X_LATCH) {
+		/* It is a system table which will always exist in the table
+		cache: we avoid acquiring the dictionary mutex, because
+		if we are doing a rollback to handle an error in TABLE
+		CREATE, for example, we already have the mutex! */
+
+		ut_ad(mutex_own(&(dict_sys->mutex))
+		      || trx->dict_operation_lock_mode == RW_X_LATCH);
+
+		return(dict_table_get_on_id_low(table_id));
+	}
+
+	mutex_enter(&(dict_sys->mutex));
+
+	table = dict_table_get_on_id_low(table_id);
+
+	mutex_exit(&(dict_sys->mutex));
+
+	return(table);
+}
+
+/************************************************************************
+Looks for column n position in the clustered index. */
+UNIV_INTERN
+ulint
+dict_table_get_nth_col_pos(
+/*=======================*/
+					/* out: position in internal
+					representation of
+					the clustered index */
+	const dict_table_t*	table,	/* in: table */
+	ulint			n)	/* in: column number */
+{
+	return(dict_index_get_nth_col_pos(dict_table_get_first_index(table),
+					  n));
+}
+
+/************************************************************************
+Checks if a column is in the ordering columns of the clustered index of a
+table. Column prefixes are treated like whole columns. */
+UNIV_INTERN
+ibool
+dict_table_col_in_clustered_key(
+/*============================*/
+					/* out: TRUE if the column, or its
+					prefix, is in the clustered key */
+	const dict_table_t*	table,	/* in: table */
+	ulint			n)	/* in: column number */
+{
+	const dict_index_t*	index;
+	const dict_field_t*	field;
+	const dict_col_t*	col;
+	ulint			pos;
+	ulint			n_fields;
+
+	ut_ad(table);
+
+	col = dict_table_get_nth_col(table, n);
+
+	index = dict_table_get_first_index(table);
+
+	n_fields = dict_index_get_n_unique(index);
+
+	for (pos = 0; pos < n_fields; pos++) {
+		field = dict_index_get_nth_field(index, pos);
+
+		if (col == field->col) {
+
+			return(TRUE);
+		}
+	}
+
+	return(FALSE);
+}
+
+/**************************************************************************
+Inits the data dictionary module. */
+UNIV_INTERN
+void
+dict_init(void)
+/*===========*/
+{
+	dict_sys = mem_alloc(sizeof(dict_sys_t));
+
+	mutex_create(&dict_sys->mutex, SYNC_DICT);
+
+	dict_sys->table_hash = hash_create(buf_pool_get_curr_size()
+					   / (DICT_POOL_PER_TABLE_HASH
+					      * UNIV_WORD_SIZE));
+	dict_sys->table_id_hash = hash_create(buf_pool_get_curr_size()
+					      / (DICT_POOL_PER_TABLE_HASH
+						 * UNIV_WORD_SIZE));
+	dict_sys->size = 0;
+
+	UT_LIST_INIT(dict_sys->table_LRU);
+
+	rw_lock_create(&dict_operation_lock, SYNC_DICT_OPERATION);
+
+	dict_foreign_err_file = os_file_create_tmpfile();
+	ut_a(dict_foreign_err_file);
+
+	mutex_create(&dict_foreign_err_mutex, SYNC_ANY_LATCH);
+}
+
+/**************************************************************************
+Returns a table object and optionally increment its MySQL open handle count.
+NOTE! This is a high-level function to be used mainly from outside the
+'dict' directory. Inside this directory dict_table_get_low is usually the
+appropriate function. */
+UNIV_INTERN
+dict_table_t*
+dict_table_get(
+/*===========*/
+					/* out: table, NULL if
+					does not exist */
+	const char*	table_name,	/* in: table name */
+	ibool		inc_mysql_count)
+					/* in: whether to increment the open
+					handle count on the table */
+{
+	dict_table_t*	table;
+
+	mutex_enter(&(dict_sys->mutex));
+
+	table = dict_table_get_low(table_name);
+
+	if (inc_mysql_count && table) {
+		table->n_mysql_handles_opened++;
+	}
+
+	mutex_exit(&(dict_sys->mutex));
+
+	if (table != NULL) {
+		if (!table->stat_initialized) {
+			/* If table->ibd_file_missing == TRUE, this will
+			print an error message and return without doing
+			anything. */
+			dict_update_statistics(table);
+		}
+	}
+
+	return(table);
+}
+
+/**************************************************************************
+Adds system columns to a table object. */
+UNIV_INTERN
+void
+dict_table_add_system_columns(
+/*==========================*/
+	dict_table_t*	table,	/* in/out: table */
+	mem_heap_t*	heap)	/* in: temporary heap */
+{
+	ut_ad(table);
+	ut_ad(table->n_def == table->n_cols - DATA_N_SYS_COLS);
+	ut_ad(table->magic_n == DICT_TABLE_MAGIC_N);
+	ut_ad(!table->cached);
+
+	/* NOTE: the system columns MUST be added in the following order
+	(so that they can be indexed by the numerical value of DATA_ROW_ID,
+	etc.) and as the last columns of the table memory object.
+	The clustered index will not always physically contain all
+	system columns. */
+
+	dict_mem_table_add_col(table, heap, "DB_ROW_ID", DATA_SYS,
+			       DATA_ROW_ID | DATA_NOT_NULL,
+			       DATA_ROW_ID_LEN);
+#if DATA_ROW_ID != 0
+#error "DATA_ROW_ID != 0"
+#endif
+	dict_mem_table_add_col(table, heap, "DB_TRX_ID", DATA_SYS,
+			       DATA_TRX_ID | DATA_NOT_NULL,
+			       DATA_TRX_ID_LEN);
+#if DATA_TRX_ID != 1
+#error "DATA_TRX_ID != 1"
+#endif
+	dict_mem_table_add_col(table, heap, "DB_ROLL_PTR", DATA_SYS,
+			       DATA_ROLL_PTR | DATA_NOT_NULL,
+			       DATA_ROLL_PTR_LEN);
+#if DATA_ROLL_PTR != 2
+#error "DATA_ROLL_PTR != 2"
+#endif
+
+	/* This check reminds that if a new system column is added to
+	the program, it should be dealt with here */
+#if DATA_N_SYS_COLS != 3
+#error "DATA_N_SYS_COLS != 3"
+#endif
+}
+
+/**************************************************************************
+Adds a table object to the dictionary cache. */
+UNIV_INTERN
+void
+dict_table_add_to_cache(
+/*====================*/
+	dict_table_t*	table,	/* in: table */
+	mem_heap_t*	heap)	/* in: temporary heap */
+{
+	ulint	fold;
+	ulint	id_fold;
+	ulint	i;
+	ulint	row_len;
+
+	/* The lower limit for what we consider a "big" row */
+#define BIG_ROW_SIZE 1024
+
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+
+	dict_table_add_system_columns(table, heap);
+
+	table->cached = TRUE;
+
+	fold = ut_fold_string(table->name);
+	id_fold = ut_fold_dulint(table->id);
+
+	row_len = 0;
+	for (i = 0; i < table->n_def; i++) {
+		ulint	col_len = dict_col_get_max_size(
+			dict_table_get_nth_col(table, i));
+
+		row_len += col_len;
+
+		/* If we have a single unbounded field, or several gigantic
+		fields, mark the maximum row size as BIG_ROW_SIZE. */
+		if (row_len >= BIG_ROW_SIZE || col_len >= BIG_ROW_SIZE) {
+			row_len = BIG_ROW_SIZE;
+
+			break;
+		}
+	}
+
+	table->big_rows = row_len >= BIG_ROW_SIZE;
+
+	/* Look for a table with the same name: error if such exists */
+	{
+		dict_table_t*	table2;
+		HASH_SEARCH(name_hash, dict_sys->table_hash, fold,
+			    dict_table_t*, table2, ut_ad(table2->cached),
+			    ut_strcmp(table2->name, table->name) == 0);
+		ut_a(table2 == NULL);
+
+#ifdef UNIV_DEBUG
+		/* Look for the same table pointer with a different name */
+		HASH_SEARCH_ALL(name_hash, dict_sys->table_hash,
+				dict_table_t*, table2, ut_ad(table2->cached),
+				table2 == table);
+		ut_ad(table2 == NULL);
+#endif /* UNIV_DEBUG */
+	}
+
+	/* Look for a table with the same id: error if such exists */
+	{
+		dict_table_t*	table2;
+		HASH_SEARCH(id_hash, dict_sys->table_id_hash, id_fold,
+			    dict_table_t*, table2, ut_ad(table2->cached),
+			    ut_dulint_cmp(table2->id, table->id) == 0);
+		ut_a(table2 == NULL);
+
+#ifdef UNIV_DEBUG
+		/* Look for the same table pointer with a different id */
+		HASH_SEARCH_ALL(id_hash, dict_sys->table_id_hash,
+				dict_table_t*, table2, ut_ad(table2->cached),
+				table2 == table);
+		ut_ad(table2 == NULL);
+#endif /* UNIV_DEBUG */
+	}
+
+	/* Add table to hash table of tables */
+	HASH_INSERT(dict_table_t, name_hash, dict_sys->table_hash, fold,
+		    table);
+
+	/* Add table to hash table of tables based on table id */
+	HASH_INSERT(dict_table_t, id_hash, dict_sys->table_id_hash, id_fold,
+		    table);
+	/* Add table to LRU list of tables */
+	UT_LIST_ADD_FIRST(table_LRU, dict_sys->table_LRU, table);
+
+	dict_sys->size += mem_heap_get_size(table->heap);
+}
+
+/**************************************************************************
+Looks for an index with the given id. NOTE that we do not reserve
+the dictionary mutex: this function is for emergency purposes like
+printing info of a corrupt database page! */
+UNIV_INTERN
+dict_index_t*
+dict_index_find_on_id_low(
+/*======================*/
+			/* out: index or NULL if not found from cache */
+	dulint	id)	/* in: index id */
+{
+	dict_table_t*	table;
+	dict_index_t*	index;
+
+	table = UT_LIST_GET_FIRST(dict_sys->table_LRU);
+
+	while (table) {
+		index = dict_table_get_first_index(table);
+
+		while (index) {
+			if (0 == ut_dulint_cmp(id, index->id)) {
+				/* Found */
+
+				return(index);
+			}
+
+			index = dict_table_get_next_index(index);
+		}
+
+		table = UT_LIST_GET_NEXT(table_LRU, table);
+	}
+
+	return(NULL);
+}
+
+/**************************************************************************
+Renames a table object. */
+UNIV_INTERN
+ibool
+dict_table_rename_in_cache(
+/*=======================*/
+					/* out: TRUE if success */
+	dict_table_t*	table,		/* in/out: table */
+	const char*	new_name,	/* in: new name */
+	ibool		rename_also_foreigns)/* in: in ALTER TABLE we want
+					to preserve the original table name
+					in constraints which reference it */
+{
+	dict_foreign_t*	foreign;
+	dict_index_t*	index;
+	ulint		fold;
+	ulint		old_size;
+	const char*	old_name;
+
+	ut_ad(table);
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+
+	old_size = mem_heap_get_size(table->heap);
+	old_name = table->name;
+
+	fold = ut_fold_string(new_name);
+
+	/* Look for a table with the same name: error if such exists */
+	{
+		dict_table_t*	table2;
+		HASH_SEARCH(name_hash, dict_sys->table_hash, fold,
+			    dict_table_t*, table2, ut_ad(table2->cached),
+			    (ut_strcmp(table2->name, new_name) == 0));
+		if (UNIV_LIKELY_NULL(table2)) {
+			ut_print_timestamp(stderr);
+			fputs("  InnoDB: Error: dictionary cache"
+			      " already contains a table ", stderr);
+			ut_print_name(stderr, NULL, TRUE, new_name);
+			fputs("\n"
+			      "InnoDB: cannot rename table ", stderr);
+			ut_print_name(stderr, NULL, TRUE, old_name);
+			putc('\n', stderr);
+			return(FALSE);
+		}
+	}
+
+	/* If the table is stored in a single-table tablespace, rename the
+	.ibd file */
+
+	if (table->space != 0) {
+		if (table->dir_path_of_temp_table != NULL) {
+			ut_print_timestamp(stderr);
+			fputs("  InnoDB: Error: trying to rename a"
+			      " TEMPORARY TABLE ", stderr);
+			ut_print_name(stderr, NULL, TRUE, old_name);
+			fputs(" (", stderr);
+			ut_print_filename(stderr,
+					  table->dir_path_of_temp_table);
+			fputs(" )\n", stderr);
+			return(FALSE);
+		} else if (!fil_rename_tablespace(old_name, table->space,
+						  new_name)) {
+			return(FALSE);
+		}
+	}
+
+	/* Remove table from the hash tables of tables */
+	HASH_DELETE(dict_table_t, name_hash, dict_sys->table_hash,
+		    ut_fold_string(old_name), table);
+	table->name = mem_heap_strdup(table->heap, new_name);
+
+	/* Add table to hash table of tables */
+	HASH_INSERT(dict_table_t, name_hash, dict_sys->table_hash, fold,
+		    table);
+	dict_sys->size += (mem_heap_get_size(table->heap) - old_size);
+
+	/* Update the table_name field in indexes */
+	index = dict_table_get_first_index(table);
+
+	while (index != NULL) {
+		index->table_name = table->name;
+
+		index = dict_table_get_next_index(index);
+	}
+
+	if (!rename_also_foreigns) {
+		/* In ALTER TABLE we think of the rename table operation
+		in the direction table -> temporary table (#sql...)
+		as dropping the table with the old name and creating
+		a new with the new name. Thus we kind of drop the
+		constraints from the dictionary cache here. The foreign key
+		constraints will be inherited to the new table from the
+		system tables through a call of dict_load_foreigns. */
+
+		/* Remove the foreign constraints from the cache */
+		foreign = UT_LIST_GET_LAST(table->foreign_list);
+
+		while (foreign != NULL) {
+			dict_foreign_remove_from_cache(foreign);
+			foreign = UT_LIST_GET_LAST(table->foreign_list);
+		}
+
+		/* Reset table field in referencing constraints */
+
+		foreign = UT_LIST_GET_FIRST(table->referenced_list);
+
+		while (foreign != NULL) {
+			foreign->referenced_table = NULL;
+			foreign->referenced_index = NULL;
+
+			foreign = UT_LIST_GET_NEXT(referenced_list, foreign);
+		}
+
+		/* Make the list of referencing constraints empty */
+
+		UT_LIST_INIT(table->referenced_list);
+
+		return(TRUE);
+	}
+
+	/* Update the table name fields in foreign constraints, and update also
+	the constraint id of new format >= 4.0.18 constraints. Note that at
+	this point we have already changed table->name to the new name. */
+
+	foreign = UT_LIST_GET_FIRST(table->foreign_list);
+
+	while (foreign != NULL) {
+		if (ut_strlen(foreign->foreign_table_name)
+		    < ut_strlen(table->name)) {
+			/* Allocate a longer name buffer;
+			TODO: store buf len to save memory */
+
+			foreign->foreign_table_name
+				= mem_heap_alloc(foreign->heap,
+						 ut_strlen(table->name) + 1);
+		}
+
+		strcpy(foreign->foreign_table_name, table->name);
+
+		if (strchr(foreign->id, '/')) {
+			ulint	db_len;
+			char*	old_id;
+
+			/* This is a >= 4.0.18 format id */
+
+			old_id = mem_strdup(foreign->id);
+
+			if (ut_strlen(foreign->id) > ut_strlen(old_name)
+			    + ((sizeof dict_ibfk) - 1)
+			    && !memcmp(foreign->id, old_name,
+				       ut_strlen(old_name))
+			    && !memcmp(foreign->id + ut_strlen(old_name),
+				       dict_ibfk, (sizeof dict_ibfk) - 1)) {
+
+				/* This is a generated >= 4.0.18 format id */
+
+				if (strlen(table->name) > strlen(old_name)) {
+					foreign->id = mem_heap_alloc(
+						foreign->heap,
+						strlen(table->name)
+						+ strlen(old_id) + 1);
+				}
+
+				/* Replace the prefix 'databasename/tablename'
+				with the new names */
+				strcpy(foreign->id, table->name);
+				strcat(foreign->id,
+				       old_id + ut_strlen(old_name));
+			} else {
+				/* This is a >= 4.0.18 format id where the user
+				gave the id name */
+				db_len = dict_get_db_name_len(table->name) + 1;
+
+				if (dict_get_db_name_len(table->name)
+				    > dict_get_db_name_len(foreign->id)) {
+
+					foreign->id = mem_heap_alloc(
+						foreign->heap,
+						db_len + strlen(old_id) + 1);
+				}
+
+				/* Replace the database prefix in id with the
+				one from table->name */
+
+				ut_memcpy(foreign->id, table->name, db_len);
+
+				strcpy(foreign->id + db_len,
+				       dict_remove_db_name(old_id));
+			}
+
+			mem_free(old_id);
+		}
+
+		foreign = UT_LIST_GET_NEXT(foreign_list, foreign);
+	}
+
+	foreign = UT_LIST_GET_FIRST(table->referenced_list);
+
+	while (foreign != NULL) {
+		if (ut_strlen(foreign->referenced_table_name)
+		    < ut_strlen(table->name)) {
+			/* Allocate a longer name buffer;
+			TODO: store buf len to save memory */
+
+			foreign->referenced_table_name = mem_heap_alloc(
+				foreign->heap, strlen(table->name) + 1);
+		}
+
+		strcpy(foreign->referenced_table_name, table->name);
+
+		foreign = UT_LIST_GET_NEXT(referenced_list, foreign);
+	}
+
+	return(TRUE);
+}
+
+/**************************************************************************
+Change the id of a table object in the dictionary cache. This is used in
+DISCARD TABLESPACE. */
+UNIV_INTERN
+void
+dict_table_change_id_in_cache(
+/*==========================*/
+	dict_table_t*	table,	/* in/out: table object already in cache */
+	dulint		new_id)	/* in: new id to set */
+{
+	ut_ad(table);
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+	ut_ad(table->magic_n == DICT_TABLE_MAGIC_N);
+
+	/* Remove the table from the hash table of id's */
+
+	HASH_DELETE(dict_table_t, id_hash, dict_sys->table_id_hash,
+		    ut_fold_dulint(table->id), table);
+	table->id = new_id;
+
+	/* Add the table back to the hash table */
+	HASH_INSERT(dict_table_t, id_hash, dict_sys->table_id_hash,
+		    ut_fold_dulint(table->id), table);
+}
+
+/**************************************************************************
+Removes a table object from the dictionary cache. */
+UNIV_INTERN
+void
+dict_table_remove_from_cache(
+/*=========================*/
+	dict_table_t*	table)	/* in, own: table */
+{
+	dict_foreign_t*	foreign;
+	dict_index_t*	index;
+	ulint		size;
+
+	ut_ad(table);
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+	ut_ad(table->magic_n == DICT_TABLE_MAGIC_N);
+
+#if 0
+	fputs("Removing table ", stderr);
+	ut_print_name(stderr, table->name, ULINT_UNDEFINED);
+	fputs(" from dictionary cache\n", stderr);
+#endif
+
+	/* Remove the foreign constraints from the cache */
+	foreign = UT_LIST_GET_LAST(table->foreign_list);
+
+	while (foreign != NULL) {
+		dict_foreign_remove_from_cache(foreign);
+		foreign = UT_LIST_GET_LAST(table->foreign_list);
+	}
+
+	/* Reset table field in referencing constraints */
+
+	foreign = UT_LIST_GET_FIRST(table->referenced_list);
+
+	while (foreign != NULL) {
+		foreign->referenced_table = NULL;
+		foreign->referenced_index = NULL;
+
+		foreign = UT_LIST_GET_NEXT(referenced_list, foreign);
+	}
+
+	/* Remove the indexes from the cache */
+	index = UT_LIST_GET_LAST(table->indexes);
+
+	while (index != NULL) {
+		dict_index_remove_from_cache(table, index);
+		index = UT_LIST_GET_LAST(table->indexes);
+	}
+
+	/* Remove table from the hash tables of tables */
+	HASH_DELETE(dict_table_t, name_hash, dict_sys->table_hash,
+		    ut_fold_string(table->name), table);
+	HASH_DELETE(dict_table_t, id_hash, dict_sys->table_id_hash,
+		    ut_fold_dulint(table->id), table);
+
+	/* Remove table from LRU list of tables */
+	UT_LIST_REMOVE(table_LRU, dict_sys->table_LRU, table);
+
+	size = mem_heap_get_size(table->heap);
+
+	ut_ad(dict_sys->size >= size);
+
+	dict_sys->size -= size;
+
+	dict_mem_table_free(table);
+}
+
+/********************************************************************
+If the given column name is reserved for InnoDB system columns, return
+TRUE. */
+UNIV_INTERN
+ibool
+dict_col_name_is_reserved(
+/*======================*/
+				/* out: TRUE if name is reserved */
+	const char*	name)	/* in: column name */
+{
+	/* This check reminds that if a new system column is added to
+	the program, it should be dealt with here. */
+#if DATA_N_SYS_COLS != 3
+#error "DATA_N_SYS_COLS != 3"
+#endif
+
+	static const char*	reserved_names[] = {
+		"DB_ROW_ID", "DB_TRX_ID", "DB_ROLL_PTR"
+	};
+
+	ulint			i;
+
+	for (i = 0; i < UT_ARR_SIZE(reserved_names); i++) {
+		if (strcmp(name, reserved_names[i]) == 0) {
+
+			return(TRUE);
+		}
+	}
+
+	return(FALSE);
+}
+
+/********************************************************************
+If an undo log record for this table might not fit on a single page,
+return TRUE. */
+static
+ibool
+dict_index_too_big_for_undo(
+/*========================*/
+						/* out: TRUE if the undo log
+						record could become too big */
+	const dict_table_t*	table,		/* in: table */
+	const dict_index_t*	new_index)	/* in: index */
+{
+	/* Make sure that all column prefixes will fit in the undo log record
+	in trx_undo_page_report_modify() right after trx_undo_page_init(). */
+
+	ulint			i;
+	const dict_index_t*	clust_index
+		= dict_table_get_first_index(table);
+	ulint			undo_page_len
+		= TRX_UNDO_PAGE_HDR - TRX_UNDO_PAGE_HDR_SIZE
+		+ 2 /* next record pointer */
+		+ 1 /* type_cmpl */
+		+ 11 /* trx->undo_no */ - 11 /* table->id */
+		+ 1 /* rec_get_info_bits() */
+		+ 11 /* DB_TRX_ID */
+		+ 11 /* DB_ROLL_PTR */
+		+ 10 + FIL_PAGE_DATA_END /* trx_undo_left() */
+		+ 2/* pointer to previous undo log record */;
+
+	if (UNIV_UNLIKELY(!clust_index)) {
+		ut_a(dict_index_is_clust(new_index));
+		clust_index = new_index;
+	}
+
+	/* Add the size of the ordering columns in the
+	clustered index. */
+	for (i = 0; i < clust_index->n_uniq; i++) {
+		const dict_col_t*	col
+			= dict_index_get_nth_col(clust_index, i);
+
+		/* Use the maximum output size of
+		mach_write_compressed(), although the encoded
+		length should always fit in 2 bytes. */
+		undo_page_len += 5 + dict_col_get_max_size(col);
+	}
+
+	/* Add the old values of the columns to be updated.
+	First, the amount and the numbers of the columns.
+	These are written by mach_write_compressed() whose
+	maximum output length is 5 bytes.  However, given that
+	the quantities are below REC_MAX_N_FIELDS (10 bits),
+	the maximum length is 2 bytes per item. */
+	undo_page_len += 2 * (dict_table_get_n_cols(table) + 1);
+
+	for (i = 0; i < clust_index->n_def; i++) {
+		const dict_col_t*	col
+			= dict_index_get_nth_col(clust_index, i);
+		ulint			max_size
+			= dict_col_get_max_size(col);
+		ulint			fixed_size
+			= dict_col_get_fixed_size(col);
+
+		if (fixed_size) {
+			/* Fixed-size columns are stored locally. */
+			max_size = fixed_size;
+		} else if (max_size <= BTR_EXTERN_FIELD_REF_SIZE * 2) {
+			/* Short columns are stored locally. */
+		} else if (!col->ord_part) {
+			/* See if col->ord_part would be set
+			because of new_index. */
+			ulint	j;
+
+			for (j = 0; j < new_index->n_uniq; j++) {
+				if (dict_index_get_nth_col(
+					    new_index, j) == col) {
+
+					goto is_ord_part;
+				}
+			}
+
+			/* This is not an ordering column in any index.
+			Thus, it can be stored completely externally. */
+			max_size = BTR_EXTERN_FIELD_REF_SIZE;
+		} else {
+is_ord_part:
+			/* This is an ordering column in some index.
+			A long enough prefix must be written to the
+			undo log.  See trx_undo_page_fetch_ext(). */
+
+			if (max_size > REC_MAX_INDEX_COL_LEN) {
+				max_size = REC_MAX_INDEX_COL_LEN;
+			}
+
+			max_size += BTR_EXTERN_FIELD_REF_SIZE;
+		}
+
+		undo_page_len += 5 + max_size;
+	}
+
+	return(undo_page_len >= UNIV_PAGE_SIZE);
+}
+
+/********************************************************************
+If a record of this index might not fit on a single B-tree page,
+return TRUE. */
+static
+ibool
+dict_index_too_big_for_tree(
+/*========================*/
+						/* out: TRUE if the index
+						record could become too big */
+	const dict_table_t*	table,		/* in: table */
+	const dict_index_t*	new_index)	/* in: index */
+{
+	ulint	zip_size;
+	ulint	comp;
+	ulint	i;
+	/* maximum possible storage size of a record */
+	ulint	rec_max_size;
+	/* maximum allowed size of a record on a leaf page */
+	ulint	page_rec_max;
+	/* maximum allowed size of a node pointer record */
+	ulint	page_ptr_max;
+
+	comp = dict_table_is_comp(table);
+	zip_size = dict_table_zip_size(table);
+
+	if (zip_size && zip_size < UNIV_PAGE_SIZE) {
+		/* On a compressed page, two records must fit in the
+		uncompressed page modification log.  On compressed
+		pages with zip_size == UNIV_PAGE_SIZE, this limit will
+		never be reached. */
+		ut_ad(comp);
+		/* The maximum allowed record size is the size of
+		an empty page, minus a byte for recoding the heap
+		number in the page modification log.  The maximum
+		allowed node pointer size is half that. */
+		page_rec_max = page_zip_empty_size(new_index->n_fields,
+						   zip_size) - 1;
+		page_ptr_max = page_rec_max / 2;
+		/* On a compressed page, there is a two-byte entry in
+		the dense page directory for every record.  But there
+		is no record header. */
+		rec_max_size = 2;
+	} else {
+		/* The maximum allowed record size is half a B-tree
+		page.  No additional sparse page directory entry will
+		be generated for the first few user records. */
+		page_rec_max = page_get_free_space_of_empty(comp) / 2;
+		page_ptr_max = page_rec_max;
+		/* Each record has a header. */
+		rec_max_size = comp
+			? REC_N_NEW_EXTRA_BYTES
+			: REC_N_OLD_EXTRA_BYTES;
+	}
+
+	if (comp) {
+		/* Include the "null" flags in the
+		maximum possible record size. */
+		rec_max_size += UT_BITS_IN_BYTES(new_index->n_nullable);
+	} else {
+		/* For each column, include a 2-byte offset and a
+		"null" flag.  The 1-byte format is only used in short
+		records that do not contain externally stored columns.
+		Such records could never exceed the page limit, even
+		when using the 2-byte format. */
+		rec_max_size += 2 * new_index->n_fields;
+	}
+
+	/* Compute the maximum possible record size. */
+	for (i = 0; i < new_index->n_fields; i++) {
+		const dict_field_t*	field
+			= dict_index_get_nth_field(new_index, i);
+		const dict_col_t*	col
+			= dict_field_get_col(field);
+		ulint			field_max_size;
+		ulint			field_ext_max_size;
+
+		/* In dtuple_convert_big_rec(), variable-length columns
+		that are longer than BTR_EXTERN_FIELD_REF_SIZE * 2
+		may be chosen for external storage.
+
+		Fixed-length columns, and all columns of secondary
+		index records are always stored inline. */
+
+		/* Determine the maximum length of the index field.
+		The field_ext_max_size should be computed as the worst
+		case in rec_get_converted_size_comp() for
+		REC_STATUS_ORDINARY records. */
+
+		field_max_size = dict_col_get_fixed_size(col);
+		if (field_max_size) {
+			/* dict_index_add_col() should guarantee this */
+			ut_ad(!field->prefix_len
+			      || field->fixed_len == field->prefix_len);
+			/* Fixed lengths are not encoded
+			in ROW_FORMAT=COMPACT. */
+			field_ext_max_size = 0;
+			goto add_field_size;
+		}
+
+		field_max_size = dict_col_get_max_size(col);
+		field_ext_max_size = field_max_size < 256 ? 1 : 2;
+
+		if (field->prefix_len) {
+			if (field->prefix_len < field_max_size) {
+				field_max_size = field->prefix_len;
+			}
+		} else if (field_max_size > BTR_EXTERN_FIELD_REF_SIZE * 2
+			   && dict_index_is_clust(new_index)) {
+
+			/* In the worst case, we have a locally stored
+			column of BTR_EXTERN_FIELD_REF_SIZE * 2 bytes.
+			The length can be stored in one byte.  If the
+			column were stored externally, the lengths in
+			the clustered index page would be
+			BTR_EXTERN_FIELD_REF_SIZE and 2. */
+			field_max_size = BTR_EXTERN_FIELD_REF_SIZE * 2;
+			field_ext_max_size = 1;
+		}
+
+		if (comp) {
+			/* Add the extra size for ROW_FORMAT=COMPACT.
+			For ROW_FORMAT=REDUNDANT, these bytes were
+			added to rec_max_size before this loop. */
+			rec_max_size += field_ext_max_size;
+		}
+add_field_size:
+		rec_max_size += field_max_size;
+
+		/* Check the size limit on leaf pages. */
+		if (UNIV_UNLIKELY(rec_max_size >= page_rec_max)) {
+
+			return(TRUE);
+		}
+
+		/* Check the size limit on non-leaf pages.  Records
+		stored in non-leaf B-tree pages consist of the unique
+		columns of the record (the key columns of the B-tree)
+		and a node pointer field.  When we have processed the
+		unique columns, rec_max_size equals the size of the
+		node pointer record minus the node pointer column. */
+		if (i + 1 == dict_index_get_n_unique_in_tree(new_index)
+		    && rec_max_size + REC_NODE_PTR_SIZE >= page_ptr_max) {
+
+			return(TRUE);
+		}
+	}
+
+	return(FALSE);
+}
+
+/**************************************************************************
+Adds an index to the dictionary cache. */
+UNIV_INTERN
+ulint
+dict_index_add_to_cache(
+/*====================*/
+				/* out: DB_SUCCESS or DB_TOO_BIG_RECORD */
+	dict_table_t*	table,	/* in: table on which the index is */
+	dict_index_t*	index,	/* in, own: index; NOTE! The index memory
+				object is freed in this function! */
+	ulint		page_no,/* in: root page number of the index */
+	ibool		strict)	/* in: TRUE=refuse to create the index
+				if records could be too big to fit in
+				an B-tree page */
+{
+	dict_index_t*	new_index;
+	ulint		n_ord;
+	ulint		i;
+
+	ut_ad(index);
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+	ut_ad(index->n_def == index->n_fields);
+	ut_ad(index->magic_n == DICT_INDEX_MAGIC_N);
+
+	ut_ad(mem_heap_validate(index->heap));
+	ut_a(!dict_index_is_clust(index)
+	     || UT_LIST_GET_LEN(table->indexes) == 0);
+
+	dict_index_find_cols(table, index);
+
+	/* Build the cache internal representation of the index,
+	containing also the added system fields */
+
+	if (dict_index_is_clust(index)) {
+		new_index = dict_index_build_internal_clust(table, index);
+	} else {
+		new_index = dict_index_build_internal_non_clust(table, index);
+	}
+
+	/* Set the n_fields value in new_index to the actual defined
+	number of fields in the cache internal representation */
+
+	new_index->n_fields = new_index->n_def;
+
+	if (strict && dict_index_too_big_for_tree(table, new_index)) {
+too_big:
+		dict_mem_index_free(new_index);
+		dict_mem_index_free(index);
+		return(DB_TOO_BIG_RECORD);
+	}
+
+	if (UNIV_UNLIKELY(index->type & DICT_UNIVERSAL)) {
+		n_ord = new_index->n_fields;
+	} else {
+		n_ord = new_index->n_uniq;
+	}
+
+	switch (dict_table_get_format(table)) {
+	case DICT_TF_FORMAT_51:
+		/* ROW_FORMAT=REDUNDANT and ROW_FORMAT=COMPACT store
+		prefixes of externally stored columns locally within
+		the record.  There are no special considerations for
+		the undo log record size. */
+		goto undo_size_ok;
+
+	case DICT_TF_FORMAT_ZIP:
+		/* In ROW_FORMAT=DYNAMIC and ROW_FORMAT=COMPRESSED,
+		column prefix indexes require that prefixes of
+		externally stored columns are written to the undo log.
+		This may make the undo log record bigger than the
+		record on the B-tree page.  The maximum size of an
+		undo log record is the page size.  That must be
+		checked for below. */
+		break;
+
+#if DICT_TF_FORMAT_ZIP != DICT_TF_FORMAT_MAX
+# error "DICT_TF_FORMAT_ZIP != DICT_TF_FORMAT_MAX"
+#endif
+	}
+
+	for (i = 0; i < n_ord; i++) {
+		const dict_field_t*	field
+			= dict_index_get_nth_field(new_index, i);
+		const dict_col_t*	col
+			= dict_field_get_col(field);
+
+		/* In dtuple_convert_big_rec(), variable-length columns
+		that are longer than BTR_EXTERN_FIELD_REF_SIZE * 2
+		may be chosen for external storage.  If the column appears
+		in an ordering column of an index, a longer prefix of
+		REC_MAX_INDEX_COL_LEN will be copied to the undo log
+		by trx_undo_page_report_modify() and
+		trx_undo_page_fetch_ext().  It suffices to check the
+		capacity of the undo log whenever new_index includes
+		a column prefix on a column that may be stored externally. */
+
+		if (field->prefix_len /* prefix index */
+		    && !col->ord_part /* not yet ordering column */
+		    && !dict_col_get_fixed_size(col) /* variable-length */
+		    && dict_col_get_max_size(col)
+		    > BTR_EXTERN_FIELD_REF_SIZE * 2 /* long enough */) {
+
+			if (dict_index_too_big_for_undo(table, new_index)) {
+				/* An undo log record might not fit in
+				a single page.  Refuse to create this index. */
+
+				goto too_big;
+			}
+
+			break;
+		}
+	}
+
+undo_size_ok:
+	/* Flag the ordering columns */
+
+	for (i = 0; i < n_ord; i++) {
+
+		dict_index_get_nth_field(new_index, i)->col->ord_part = 1;
+	}
+
+	/* Add the new index as the last index for the table */
+
+	UT_LIST_ADD_LAST(indexes, table->indexes, new_index);
+	new_index->table = table;
+	new_index->table_name = table->name;
+
+	new_index->search_info = btr_search_info_create(new_index->heap);
+
+	new_index->stat_index_size = 1;
+	new_index->stat_n_leaf_pages = 1;
+
+	new_index->page = page_no;
+	rw_lock_create(&new_index->lock, SYNC_INDEX_TREE);
+
+	if (!UNIV_UNLIKELY(new_index->type & DICT_UNIVERSAL)) {
+
+		new_index->stat_n_diff_key_vals = mem_heap_alloc(
+			new_index->heap,
+			(1 + dict_index_get_n_unique(new_index))
+			* sizeof(ib_int64_t));
+		/* Give some sensible values to stat_n_... in case we do
+		not calculate statistics quickly enough */
+
+		for (i = 0; i <= dict_index_get_n_unique(new_index); i++) {
+
+			new_index->stat_n_diff_key_vals[i] = 100;
+		}
+	}
+
+	dict_sys->size += mem_heap_get_size(new_index->heap);
+
+	dict_mem_index_free(index);
+
+	return(DB_SUCCESS);
+}
+
+/**************************************************************************
+Removes an index from the dictionary cache. */
+UNIV_INTERN
+void
+dict_index_remove_from_cache(
+/*=========================*/
+	dict_table_t*	table,	/* in/out: table */
+	dict_index_t*	index)	/* in, own: index */
+{
+	ulint		size;
+	ulint		retries = 0;
+	btr_search_t*	info;
+
+	ut_ad(table && index);
+	ut_ad(table->magic_n == DICT_TABLE_MAGIC_N);
+	ut_ad(index->magic_n == DICT_INDEX_MAGIC_N);
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+
+	/* We always create search info whether or not adaptive
+	hash index is enabled or not. */
+	info = index->search_info;
+	ut_ad(info);
+
+	/* We are not allowed to free the in-memory index struct
+ 	dict_index_t until all entries in the adaptive hash index
+	that point to any of the page belonging to his b-tree index
+	are dropped. This is so because dropping of these entries
+	require access to dict_index_t struct. To avoid such scenario
+	We keep a count of number of such pages in the search_info and
+	only free the dict_index_t struct when this count drops to
+	zero. */
+
+	for (;;) {
+		ulint ref_count = btr_search_info_get_ref_count(info);
+		if (ref_count == 0) {
+			break;
+		}
+
+		/* Sleep for 10ms before trying again. */
+		os_thread_sleep(10000);
+		++retries;
+
+		if (retries % 500 == 0) {
+			/* No luck after 5 seconds of wait. */
+			fprintf(stderr, "InnoDB: Error: Waited for"
+					" %lu secs for hash index"
+					" ref_count (%lu) to drop"
+					" to 0.\n"
+					"index: \"%s\""
+					" table: \"%s\"\n",
+					retries/100,
+					ref_count,
+					index->name,
+					table->name);
+		}
+
+		/* To avoid a hang here we commit suicide if the
+		ref_count doesn't drop to zero in 600 seconds. */
+		if (retries >= 60000) {
+			ut_error;
+		}
+	}
+
+	rw_lock_free(&index->lock);
+
+	/* Remove the index from the list of indexes of the table */
+	UT_LIST_REMOVE(indexes, table->indexes, index);
+
+	size = mem_heap_get_size(index->heap);
+
+	ut_ad(dict_sys->size >= size);
+
+	dict_sys->size -= size;
+
+	dict_mem_index_free(index);
+}
+
+/***********************************************************************
+Tries to find column names for the index and sets the col field of the
+index. */
+static
+void
+dict_index_find_cols(
+/*=================*/
+	dict_table_t*	table,	/* in: table */
+	dict_index_t*	index)	/* in: index */
+{
+	ulint		i;
+
+	ut_ad(table && index);
+	ut_ad(table->magic_n == DICT_TABLE_MAGIC_N);
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+
+	for (i = 0; i < index->n_fields; i++) {
+		ulint		j;
+		dict_field_t*	field = dict_index_get_nth_field(index, i);
+
+		for (j = 0; j < table->n_cols; j++) {
+			if (!strcmp(dict_table_get_col_name(table, j),
+				    field->name)) {
+				field->col = dict_table_get_nth_col(table, j);
+
+				goto found;
+			}
+		}
+
+		/* It is an error not to find a matching column. */
+		ut_error;
+
+found:
+		;
+	}
+}
+
+/***********************************************************************
+Adds a column to index. */
+UNIV_INTERN
+void
+dict_index_add_col(
+/*===============*/
+	dict_index_t*		index,		/* in/out: index */
+	const dict_table_t*	table,		/* in: table */
+	dict_col_t*		col,		/* in: column */
+	ulint			prefix_len)	/* in: column prefix length */
+{
+	dict_field_t*	field;
+	const char*	col_name;
+
+	col_name = dict_table_get_col_name(table, dict_col_get_no(col));
+
+	dict_mem_index_add_field(index, col_name, prefix_len);
+
+	field = dict_index_get_nth_field(index, index->n_def - 1);
+
+	field->col = col;
+	field->fixed_len = (unsigned int) dict_col_get_fixed_size(col);
+
+	if (prefix_len && field->fixed_len > prefix_len) {
+		field->fixed_len = (unsigned int) prefix_len;
+	}
+
+	/* Long fixed-length fields that need external storage are treated as
+	variable-length fields, so that the extern flag can be embedded in
+	the length word. */
+
+	if (field->fixed_len > DICT_MAX_INDEX_COL_LEN) {
+		field->fixed_len = 0;
+	}
+#if DICT_MAX_INDEX_COL_LEN != 768
+	/* The comparison limit above must be constant.  If it were
+	changed, the disk format of some fixed-length columns would
+	change, which would be a disaster. */
+# error "DICT_MAX_INDEX_COL_LEN != 768"
+#endif
+
+	if (!(col->prtype & DATA_NOT_NULL)) {
+		index->n_nullable++;
+	}
+}
+
+/***********************************************************************
+Copies fields contained in index2 to index1. */
+static
+void
+dict_index_copy(
+/*============*/
+	dict_index_t*		index1,	/* in: index to copy to */
+	dict_index_t*		index2,	/* in: index to copy from */
+	const dict_table_t*	table,	/* in: table */
+	ulint			start,	/* in: first position to copy */
+	ulint			end)	/* in: last position to copy */
+{
+	dict_field_t*	field;
+	ulint		i;
+
+	/* Copy fields contained in index2 */
+
+	for (i = start; i < end; i++) {
+
+		field = dict_index_get_nth_field(index2, i);
+		dict_index_add_col(index1, table, field->col,
+				   field->prefix_len);
+	}
+}
+
+/***********************************************************************
+Copies types of fields contained in index to tuple. */
+UNIV_INTERN
+void
+dict_index_copy_types(
+/*==================*/
+	dtuple_t*		tuple,		/* in/out: data tuple */
+	const dict_index_t*	index,		/* in: index */
+	ulint			n_fields)	/* in: number of
+						field types to copy */
+{
+	ulint		i;
+
+	if (UNIV_UNLIKELY(index->type & DICT_UNIVERSAL)) {
+		dtuple_set_types_binary(tuple, n_fields);
+
+		return;
+	}
+
+	for (i = 0; i < n_fields; i++) {
+		const dict_field_t*	ifield;
+		dtype_t*		dfield_type;
+
+		ifield = dict_index_get_nth_field(index, i);
+		dfield_type = dfield_get_type(dtuple_get_nth_field(tuple, i));
+		dict_col_copy_type(dict_field_get_col(ifield), dfield_type);
+	}
+}
+
+/***********************************************************************
+Copies types of columns contained in table to tuple and sets all
+fields of the tuple to the SQL NULL value.  This function should
+be called right after dtuple_create(). */
+UNIV_INTERN
+void
+dict_table_copy_types(
+/*==================*/
+	dtuple_t*		tuple,	/* in/out: data tuple */
+	const dict_table_t*	table)	/* in: table */
+{
+	ulint		i;
+
+	for (i = 0; i < dtuple_get_n_fields(tuple); i++) {
+
+		dfield_t*	dfield	= dtuple_get_nth_field(tuple, i);
+		dtype_t*	dtype	= dfield_get_type(dfield);
+
+		dfield_set_null(dfield);
+		dict_col_copy_type(dict_table_get_nth_col(table, i), dtype);
+	}
+}
+
+/***********************************************************************
+Builds the internal dictionary cache representation for a clustered
+index, containing also system fields not defined by the user. */
+static
+dict_index_t*
+dict_index_build_internal_clust(
+/*============================*/
+					/* out, own: the internal
+					representation of the clustered
+					index */
+	const dict_table_t*	table,	/* in: table */
+	dict_index_t*		index)	/* in: user representation of
+					a clustered index */
+{
+	dict_index_t*	new_index;
+	dict_field_t*	field;
+	ulint		fixed_size;
+	ulint		trx_id_pos;
+	ulint		i;
+	ibool*		indexed;
+
+	ut_ad(table && index);
+	ut_ad(dict_index_is_clust(index));
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+	ut_ad(table->magic_n == DICT_TABLE_MAGIC_N);
+
+	/* Create a new index object with certainly enough fields */
+	new_index = dict_mem_index_create(table->name,
+					  index->name, table->space,
+					  index->type,
+					  index->n_fields + table->n_cols);
+
+	/* Copy other relevant data from the old index struct to the new
+	struct: it inherits the values */
+
+	new_index->n_user_defined_cols = index->n_fields;
+
+	new_index->id = index->id;
+
+	/* Copy the fields of index */
+	dict_index_copy(new_index, index, table, 0, index->n_fields);
+
+	if (UNIV_UNLIKELY(index->type & DICT_UNIVERSAL)) {
+		/* No fixed number of fields determines an entry uniquely */
+
+		new_index->n_uniq = REC_MAX_N_FIELDS;
+
+	} else if (dict_index_is_unique(index)) {
+		/* Only the fields defined so far are needed to identify
+		the index entry uniquely */
+
+		new_index->n_uniq = new_index->n_def;
+	} else {
+		/* Also the row id is needed to identify the entry */
+		new_index->n_uniq = 1 + new_index->n_def;
+	}
+
+	new_index->trx_id_offset = 0;
+
+	if (!dict_index_is_ibuf(index)) {
+		/* Add system columns, trx id first */
+
+		trx_id_pos = new_index->n_def;
+
+#if DATA_ROW_ID != 0
+# error "DATA_ROW_ID != 0"
+#endif
+#if DATA_TRX_ID != 1
+# error "DATA_TRX_ID != 1"
+#endif
+#if DATA_ROLL_PTR != 2
+# error "DATA_ROLL_PTR != 2"
+#endif
+
+		if (!dict_index_is_unique(index)) {
+			dict_index_add_col(new_index, table,
+					   dict_table_get_sys_col(
+						   table, DATA_ROW_ID),
+					   0);
+			trx_id_pos++;
+		}
+
+		dict_index_add_col(new_index, table,
+				   dict_table_get_sys_col(table, DATA_TRX_ID),
+				   0);
+
+		dict_index_add_col(new_index, table,
+				   dict_table_get_sys_col(table,
+							  DATA_ROLL_PTR),
+				   0);
+
+		for (i = 0; i < trx_id_pos; i++) {
+
+			fixed_size = dict_col_get_fixed_size(
+				dict_index_get_nth_col(new_index, i));
+
+			if (fixed_size == 0) {
+				new_index->trx_id_offset = 0;
+
+				break;
+			}
+
+			if (dict_index_get_nth_field(new_index, i)->prefix_len
+			    > 0) {
+				new_index->trx_id_offset = 0;
+
+				break;
+			}
+
+			new_index->trx_id_offset += (unsigned int) fixed_size;
+		}
+
+	}
+
+	/* Remember the table columns already contained in new_index */
+	indexed = mem_zalloc(table->n_cols * sizeof *indexed);
+
+	/* Mark the table columns already contained in new_index */
+	for (i = 0; i < new_index->n_def; i++) {
+
+		field = dict_index_get_nth_field(new_index, i);
+
+		/* If there is only a prefix of the column in the index
+		field, do not mark the column as contained in the index */
+
+		if (field->prefix_len == 0) {
+
+			indexed[field->col->ind] = TRUE;
+		}
+	}
+
+	/* Add to new_index non-system columns of table not yet included
+	there */
+	for (i = 0; i + DATA_N_SYS_COLS < (ulint) table->n_cols; i++) {
+
+		dict_col_t*	col = dict_table_get_nth_col(table, i);
+		ut_ad(col->mtype != DATA_SYS);
+
+		if (!indexed[col->ind]) {
+			dict_index_add_col(new_index, table, col, 0);
+		}
+	}
+
+	mem_free(indexed);
+
+	ut_ad(dict_index_is_ibuf(index)
+	      || (UT_LIST_GET_LEN(table->indexes) == 0));
+
+	new_index->cached = TRUE;
+
+	return(new_index);
+}
+
+/***********************************************************************
+Builds the internal dictionary cache representation for a non-clustered
+index, containing also system fields not defined by the user. */
+static
+dict_index_t*
+dict_index_build_internal_non_clust(
+/*================================*/
+					/* out, own: the internal
+					representation of the non-clustered
+					index */
+	const dict_table_t*	table,	/* in: table */
+	dict_index_t*		index)	/* in: user representation of
+					a non-clustered index */
+{
+	dict_field_t*	field;
+	dict_index_t*	new_index;
+	dict_index_t*	clust_index;
+	ulint		i;
+	ibool*		indexed;
+
+	ut_ad(table && index);
+	ut_ad(!dict_index_is_clust(index));
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+	ut_ad(table->magic_n == DICT_TABLE_MAGIC_N);
+
+	/* The clustered index should be the first in the list of indexes */
+	clust_index = UT_LIST_GET_FIRST(table->indexes);
+
+	ut_ad(clust_index);
+	ut_ad(dict_index_is_clust(clust_index));
+	ut_ad(!(clust_index->type & DICT_UNIVERSAL));
+
+	/* Create a new index */
+	new_index = dict_mem_index_create(
+		table->name, index->name, index->space, index->type,
+		index->n_fields + 1 + clust_index->n_uniq);
+
+	/* Copy other relevant data from the old index
+	struct to the new struct: it inherits the values */
+
+	new_index->n_user_defined_cols = index->n_fields;
+
+	new_index->id = index->id;
+
+	/* Copy fields from index to new_index */
+	dict_index_copy(new_index, index, table, 0, index->n_fields);
+
+	/* Remember the table columns already contained in new_index */
+	indexed = mem_zalloc(table->n_cols * sizeof *indexed);
+
+	/* Mark the table columns already contained in new_index */
+	for (i = 0; i < new_index->n_def; i++) {
+
+		field = dict_index_get_nth_field(new_index, i);
+
+		/* If there is only a prefix of the column in the index
+		field, do not mark the column as contained in the index */
+
+		if (field->prefix_len == 0) {
+
+			indexed[field->col->ind] = TRUE;
+		}
+	}
+
+	/* Add to new_index the columns necessary to determine the clustered
+	index entry uniquely */
+
+	for (i = 0; i < clust_index->n_uniq; i++) {
+
+		field = dict_index_get_nth_field(clust_index, i);
+
+		if (!indexed[field->col->ind]) {
+			dict_index_add_col(new_index, table, field->col,
+					   field->prefix_len);
+		}
+	}
+
+	mem_free(indexed);
+
+	if (dict_index_is_unique(index)) {
+		new_index->n_uniq = index->n_fields;
+	} else {
+		new_index->n_uniq = new_index->n_def;
+	}
+
+	/* Set the n_fields value in new_index to the actual defined
+	number of fields */
+
+	new_index->n_fields = new_index->n_def;
+
+	new_index->cached = TRUE;
+
+	return(new_index);
+}
+
+/*====================== FOREIGN KEY PROCESSING ========================*/
+
+/*************************************************************************
+Checks if a table is referenced by foreign keys. */
+UNIV_INTERN
+ibool
+dict_table_is_referenced_by_foreign_key(
+/*====================================*/
+					/* out: TRUE if table is referenced
+					by a foreign key */
+	const dict_table_t*	table)	/* in: InnoDB table */
+{
+	return(UT_LIST_GET_LEN(table->referenced_list) > 0);
+}
+
+/*************************************************************************
+Check if the index is referenced by a foreign key, if TRUE return foreign
+else return NULL */
+UNIV_INTERN
+dict_foreign_t*
+dict_table_get_referenced_constraint(
+/*=================================*/
+				/* out: pointer to foreign key struct if index
+				is defined for foreign key, otherwise NULL */
+	dict_table_t*	table,	/* in: InnoDB table */
+	dict_index_t*	index)	/* in: InnoDB index */
+{
+	dict_foreign_t*	foreign;
+
+	ut_ad(index != NULL);
+	ut_ad(table != NULL);
+
+	for (foreign = UT_LIST_GET_FIRST(table->referenced_list);
+	     foreign;
+	     foreign = UT_LIST_GET_NEXT(referenced_list, foreign)) {
+
+		if (foreign->referenced_index == index) {
+
+			return(foreign);
+		}
+	}
+
+	return(NULL);
+}
+
+/*************************************************************************
+Checks if a index is defined for a foreign key constraint. Index is a part
+of a foreign key constraint if the index is referenced by foreign key
+or index is a foreign key index. */
+UNIV_INTERN
+dict_foreign_t*
+dict_table_get_foreign_constraint(
+/*==============================*/
+				/* out: pointer to foreign key struct if index
+				is defined for foreign key, otherwise NULL */
+	dict_table_t*	table,	/* in: InnoDB table */
+	dict_index_t*	index)	/* in: InnoDB index */
+{
+	dict_foreign_t*	foreign;
+
+	ut_ad(index != NULL);
+	ut_ad(table != NULL);
+
+	for (foreign = UT_LIST_GET_FIRST(table->foreign_list);
+	     foreign;
+	     foreign = UT_LIST_GET_NEXT(foreign_list, foreign)) {
+
+		if (foreign->foreign_index == index
+		    || foreign->referenced_index == index) {
+
+			return(foreign);
+		}
+	}
+
+	return(NULL);
+}
+
+/*************************************************************************
+Frees a foreign key struct. */
+static
+void
+dict_foreign_free(
+/*==============*/
+	dict_foreign_t*	foreign)	/* in, own: foreign key struct */
+{
+	mem_heap_free(foreign->heap);
+}
+
+/**************************************************************************
+Removes a foreign constraint struct from the dictionary cache. */
+static
+void
+dict_foreign_remove_from_cache(
+/*===========================*/
+	dict_foreign_t*	foreign)	/* in, own: foreign constraint */
+{
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+	ut_a(foreign);
+
+	if (foreign->referenced_table) {
+		UT_LIST_REMOVE(referenced_list,
+			       foreign->referenced_table->referenced_list,
+			       foreign);
+	}
+
+	if (foreign->foreign_table) {
+		UT_LIST_REMOVE(foreign_list,
+			       foreign->foreign_table->foreign_list,
+			       foreign);
+	}
+
+	dict_foreign_free(foreign);
+}
+
+#ifndef UNIV_HOTBACKUP
+/**************************************************************************
+Looks for the foreign constraint from the foreign and referenced lists
+of a table. */
+static
+dict_foreign_t*
+dict_foreign_find(
+/*==============*/
+				/* out: foreign constraint */
+	dict_table_t*	table,	/* in: table object */
+	const char*	id)	/* in: foreign constraint id */
+{
+	dict_foreign_t*	foreign;
+
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+
+	foreign = UT_LIST_GET_FIRST(table->foreign_list);
+
+	while (foreign) {
+		if (ut_strcmp(id, foreign->id) == 0) {
+
+			return(foreign);
+		}
+
+		foreign = UT_LIST_GET_NEXT(foreign_list, foreign);
+	}
+
+	foreign = UT_LIST_GET_FIRST(table->referenced_list);
+
+	while (foreign) {
+		if (ut_strcmp(id, foreign->id) == 0) {
+
+			return(foreign);
+		}
+
+		foreign = UT_LIST_GET_NEXT(referenced_list, foreign);
+	}
+
+	return(NULL);
+}
+
+/*************************************************************************
+Tries to find an index whose first fields are the columns in the array,
+in the same order and is not marked for deletion and is not the same
+as types_idx. */
+static
+dict_index_t*
+dict_foreign_find_index(
+/*====================*/
+				/* out: matching index, NULL if not found */
+	dict_table_t*	table,	/* in: table */
+	const char**	columns,/* in: array of column names */
+	ulint		n_cols,	/* in: number of columns */
+	dict_index_t*	types_idx, /* in: NULL or an index to whose types the
+				   column types must match */
+	ibool		check_charsets,
+				/* in: whether to check charsets.
+				only has an effect if types_idx != NULL */
+	ulint		check_null)
+				/* in: nonzero if none of the columns must
+				be declared NOT NULL */
+{
+	dict_index_t*	index;
+
+	index = dict_table_get_first_index(table);
+
+	while (index != NULL) {
+		/* Ignore matches that refer to the same instance
+		or the index is to be dropped */
+		if (index->to_be_dropped || types_idx == index) {
+
+			goto next_rec;
+
+		} else if (dict_index_get_n_fields(index) >= n_cols) {
+			ulint		i;
+
+			for (i = 0; i < n_cols; i++) {
+				dict_field_t*	field;
+				const char*	col_name;
+
+				field = dict_index_get_nth_field(index, i);
+
+				col_name = dict_table_get_col_name(
+					table, dict_col_get_no(field->col));
+
+				if (field->prefix_len != 0) {
+					/* We do not accept column prefix
+					indexes here */
+
+					break;
+				}
+
+				if (0 != innobase_strcasecmp(columns[i],
+							     col_name)) {
+					break;
+				}
+
+				if (check_null
+				    && (field->col->prtype & DATA_NOT_NULL)) {
+
+					return(NULL);
+				}
+
+				if (types_idx && !cmp_cols_are_equal(
+					    dict_index_get_nth_col(index, i),
+					    dict_index_get_nth_col(types_idx,
+								   i),
+					    check_charsets)) {
+
+					break;
+				}
+			}
+
+			if (i == n_cols) {
+				/* We found a matching index */
+
+				return(index);
+			}
+		}
+
+next_rec:
+		index = dict_table_get_next_index(index);
+	}
+
+	return(NULL);
+}
+
+/**************************************************************************
+Find an index that is equivalent to the one passed in and is not marked
+for deletion. */
+UNIV_INTERN
+dict_index_t*
+dict_foreign_find_equiv_index(
+/*==========================*/
+				/* out: index equivalent to
+				foreign->foreign_index, or NULL */
+	dict_foreign_t*	foreign)/* in: foreign key */
+{
+	ut_a(foreign != NULL);
+
+	/* Try to find an index which contains the columns as the
+	first fields and in the right order, and the types are the
+	same as in foreign->foreign_index */
+
+	return(dict_foreign_find_index(
+		       foreign->foreign_table,
+		       foreign->foreign_col_names, foreign->n_fields,
+		       foreign->foreign_index, TRUE, /* check types */
+		       FALSE/* allow columns to be NULL */));
+}
+
+/**************************************************************************
+Returns an index object by matching on the name and column names and
+if more than one index matches return the index with the max id */
+UNIV_INTERN
+dict_index_t*
+dict_table_get_index_by_max_id(
+/*===========================*/
+				/* out: matching index, NULL if not found */
+	dict_table_t*	table,	/* in: table */
+	const char*	name,	/* in: the index name to find */
+	const char**	columns,/* in: array of column names */
+	ulint		n_cols)	/* in: number of columns */
+{
+	dict_index_t*	index;
+	dict_index_t*	found;
+
+	found = NULL;
+	index = dict_table_get_first_index(table);
+
+	while (index != NULL) {
+		if (ut_strcmp(index->name, name) == 0
+		    && dict_index_get_n_ordering_defined_by_user(index)
+		    == n_cols) {
+
+			ulint		i;
+
+			for (i = 0; i < n_cols; i++) {
+				dict_field_t*	field;
+				const char*	col_name;
+
+				field = dict_index_get_nth_field(index, i);
+
+				col_name = dict_table_get_col_name(
+					table, dict_col_get_no(field->col));
+
+				if (0 != innobase_strcasecmp(
+					    columns[i], col_name)) {
+
+					break;
+				}
+			}
+
+			if (i == n_cols) {
+				/* We found a matching index, select
+				the index with the higher id*/
+
+				if (!found
+				    || ut_dulint_cmp(index->id, found->id) > 0) {
+
+					found = index;
+				}
+			}
+		}
+
+		index = dict_table_get_next_index(index);
+	}
+
+	return(found);
+}
+
+/**************************************************************************
+Report an error in a foreign key definition. */
+static
+void
+dict_foreign_error_report_low(
+/*==========================*/
+	FILE*		file,	/* in: output stream */
+	const char*	name)	/* in: table name */
+{
+	rewind(file);
+	ut_print_timestamp(file);
+	fprintf(file, " Error in foreign key constraint of table %s:\n",
+		name);
+}
+
+/**************************************************************************
+Report an error in a foreign key definition. */
+static
+void
+dict_foreign_error_report(
+/*======================*/
+	FILE*		file,	/* in: output stream */
+	dict_foreign_t*	fk,	/* in: foreign key constraint */
+	const char*	msg)	/* in: the error message */
+{
+	mutex_enter(&dict_foreign_err_mutex);
+	dict_foreign_error_report_low(file, fk->foreign_table_name);
+	fputs(msg, file);
+	fputs(" Constraint:\n", file);
+	dict_print_info_on_foreign_key_in_create_format(file, NULL, fk, TRUE);
+	putc('\n', file);
+	if (fk->foreign_index) {
+		fputs("The index in the foreign key in table is ", file);
+		ut_print_name(file, NULL, FALSE, fk->foreign_index->name);
+		fputs("\n"
+		      "See http://dev.mysql.com/doc/refman/5.1/en/"
+		      "innodb-foreign-key-constraints.html\n"
+		      "for correct foreign key definition.\n",
+		      file);
+	}
+	mutex_exit(&dict_foreign_err_mutex);
+}
+
+/**************************************************************************
+Adds a foreign key constraint object to the dictionary cache. May free
+the object if there already is an object with the same identifier in.
+At least one of the foreign table and the referenced table must already
+be in the dictionary cache! */
+UNIV_INTERN
+ulint
+dict_foreign_add_to_cache(
+/*======================*/
+					/* out: DB_SUCCESS or error code */
+	dict_foreign_t*	foreign,	/* in, own: foreign key constraint */
+	ibool		check_charsets)	/* in: TRUE=check charset
+					compatibility */
+{
+	dict_table_t*	for_table;
+	dict_table_t*	ref_table;
+	dict_foreign_t*	for_in_cache		= NULL;
+	dict_index_t*	index;
+	ibool		added_to_referenced_list= FALSE;
+	FILE*		ef			= dict_foreign_err_file;
+
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+
+	for_table = dict_table_check_if_in_cache_low(
+		foreign->foreign_table_name);
+
+	ref_table = dict_table_check_if_in_cache_low(
+		foreign->referenced_table_name);
+	ut_a(for_table || ref_table);
+
+	if (for_table) {
+		for_in_cache = dict_foreign_find(for_table, foreign->id);
+	}
+
+	if (!for_in_cache && ref_table) {
+		for_in_cache = dict_foreign_find(ref_table, foreign->id);
+	}
+
+	if (for_in_cache) {
+		/* Free the foreign object */
+		mem_heap_free(foreign->heap);
+	} else {
+		for_in_cache = foreign;
+	}
+
+	if (for_in_cache->referenced_table == NULL && ref_table) {
+		index = dict_foreign_find_index(
+			ref_table,
+			for_in_cache->referenced_col_names,
+			for_in_cache->n_fields, for_in_cache->foreign_index,
+			check_charsets, FALSE);
+
+		if (index == NULL) {
+			dict_foreign_error_report(
+				ef, for_in_cache,
+				"there is no index in referenced table"
+				" which would contain\n"
+				"the columns as the first columns,"
+				" or the data types in the\n"
+				"referenced table do not match"
+				" the ones in table.");
+
+			if (for_in_cache == foreign) {
+				mem_heap_free(foreign->heap);
+			}
+
+			return(DB_CANNOT_ADD_CONSTRAINT);
+		}
+
+		for_in_cache->referenced_table = ref_table;
+		for_in_cache->referenced_index = index;
+		UT_LIST_ADD_LAST(referenced_list,
+				 ref_table->referenced_list,
+				 for_in_cache);
+		added_to_referenced_list = TRUE;
+	}
+
+	if (for_in_cache->foreign_table == NULL && for_table) {
+		index = dict_foreign_find_index(
+			for_table,
+			for_in_cache->foreign_col_names,
+			for_in_cache->n_fields,
+			for_in_cache->referenced_index, check_charsets,
+			for_in_cache->type
+			& (DICT_FOREIGN_ON_DELETE_SET_NULL
+			   | DICT_FOREIGN_ON_UPDATE_SET_NULL));
+
+		if (index == NULL) {
+			dict_foreign_error_report(
+				ef, for_in_cache,
+				"there is no index in the table"
+				" which would contain\n"
+				"the columns as the first columns,"
+				" or the data types in the\n"
+				"table do not match"
+				" the ones in the referenced table\n"
+				"or one of the ON ... SET NULL columns"
+				" is declared NOT NULL.");
+
+			if (for_in_cache == foreign) {
+				if (added_to_referenced_list) {
+					UT_LIST_REMOVE(
+						referenced_list,
+						ref_table->referenced_list,
+						for_in_cache);
+				}
+
+				mem_heap_free(foreign->heap);
+			}
+
+			return(DB_CANNOT_ADD_CONSTRAINT);
+		}
+
+		for_in_cache->foreign_table = for_table;
+		for_in_cache->foreign_index = index;
+		UT_LIST_ADD_LAST(foreign_list,
+				 for_table->foreign_list,
+				 for_in_cache);
+	}
+
+	return(DB_SUCCESS);
+}
+
+/*************************************************************************
+Scans from pointer onwards. Stops if is at the start of a copy of
+'string' where characters are compared without case sensitivity, and
+only outside `` or "" quotes. Stops also at '\0'. */
+static
+const char*
+dict_scan_to(
+/*=========*/
+				/* out: scanned up to this */
+	const char*	ptr,	/* in: scan from */
+	const char*	string)	/* in: look for this */
+{
+	char	quote	= '\0';
+
+	for (; *ptr; ptr++) {
+		if (*ptr == quote) {
+			/* Closing quote character: do not look for
+			starting quote or the keyword. */
+			quote = '\0';
+		} else if (quote) {
+			/* Within quotes: do nothing. */
+		} else if (*ptr == '`' || *ptr == '"') {
+			/* Starting quote: remember the quote character. */
+			quote = *ptr;
+		} else {
+			/* Outside quotes: look for the keyword. */
+			ulint	i;
+			for (i = 0; string[i]; i++) {
+				if (toupper((int)(unsigned char)(ptr[i]))
+				    != toupper((int)(unsigned char)
+					       (string[i]))) {
+					goto nomatch;
+				}
+			}
+			break;
+nomatch:
+			;
+		}
+	}
+
+	return(ptr);
+}
+
+/*************************************************************************
+Accepts a specified string. Comparisons are case-insensitive. */
+static
+const char*
+dict_accept(
+/*========*/
+				/* out: if string was accepted, the pointer
+				is moved after that, else ptr is returned */
+	struct charset_info_st*	cs,/* in: the character set of ptr */
+	const char*	ptr,	/* in: scan from this */
+	const char*	string,	/* in: accept only this string as the next
+				non-whitespace string */
+	ibool*		success)/* out: TRUE if accepted */
+{
+	const char*	old_ptr = ptr;
+	const char*	old_ptr2;
+
+	*success = FALSE;
+
+	while (my_isspace(cs, *ptr)) {
+		ptr++;
+	}
+
+	old_ptr2 = ptr;
+
+	ptr = dict_scan_to(ptr, string);
+
+	if (*ptr == '\0' || old_ptr2 != ptr) {
+		return(old_ptr);
+	}
+
+	*success = TRUE;
+
+	return(ptr + ut_strlen(string));
+}
+
+/*************************************************************************
+Scans an id. For the lexical definition of an 'id', see the code below.
+Strips backquotes or double quotes from around the id. */
+static
+const char*
+dict_scan_id(
+/*=========*/
+				/* out: scanned to */
+	struct charset_info_st*	cs,/* in: the character set of ptr */
+	const char*	ptr,	/* in: scanned to */
+	mem_heap_t*	heap,	/* in: heap where to allocate the id
+				(NULL=id will not be allocated, but it
+				will point to string near ptr) */
+	const char**	id,	/* out,own: the id; NULL if no id was
+				scannable */
+	ibool		table_id,/* in: TRUE=convert the allocated id
+				as a table name; FALSE=convert to UTF-8 */
+	ibool		accept_also_dot)
+				/* in: TRUE if also a dot can appear in a
+				non-quoted id; in a quoted id it can appear
+				always */
+{
+	char		quote	= '\0';
+	ulint		len	= 0;
+	const char*	s;
+	char*		str;
+	char*		dst;
+
+	*id = NULL;
+
+	while (my_isspace(cs, *ptr)) {
+		ptr++;
+	}
+
+	if (*ptr == '\0') {
+
+		return(ptr);
+	}
+
+	if (*ptr == '`' || *ptr == '"') {
+		quote = *ptr++;
+	}
+
+	s = ptr;
+
+	if (quote) {
+		for (;;) {
+			if (!*ptr) {
+				/* Syntax error */
+				return(ptr);
+			}
+			if (*ptr == quote) {
+				ptr++;
+				if (*ptr != quote) {
+					break;
+				}
+			}
+			ptr++;
+			len++;
+		}
+	} else {
+		while (!my_isspace(cs, *ptr) && *ptr != '(' && *ptr != ')'
+		       && (accept_also_dot || *ptr != '.')
+		       && *ptr != ',' && *ptr != '\0') {
+
+			ptr++;
+		}
+
+		len = ptr - s;
+	}
+
+	if (UNIV_UNLIKELY(!heap)) {
+		/* no heap given: id will point to source string */
+		*id = s;
+		return(ptr);
+	}
+
+	if (quote) {
+		char*	d;
+		str = d = mem_heap_alloc(heap, len + 1);
+		while (len--) {
+			if ((*d++ = *s++) == quote) {
+				s++;
+			}
+		}
+		*d++ = 0;
+		len = d - str;
+		ut_ad(*s == quote);
+		ut_ad(s + 1 == ptr);
+	} else {
+		str = mem_heap_strdupl(heap, s, len);
+	}
+
+	if (!table_id) {
+convert_id:
+		/* Convert the identifier from connection character set
+		to UTF-8. */
+		len = 3 * len + 1;
+		*id = dst = mem_heap_alloc(heap, len);
+
+		innobase_convert_from_id(cs, dst, str, len);
+	} else if (!strncmp(str, srv_mysql50_table_name_prefix,
+			    sizeof srv_mysql50_table_name_prefix)) {
+		/* This is a pre-5.1 table name
+		containing chars other than [A-Za-z0-9].
+		Discard the prefix and use raw UTF-8 encoding. */
+		str += sizeof srv_mysql50_table_name_prefix;
+		len -= sizeof srv_mysql50_table_name_prefix;
+		goto convert_id;
+	} else {
+		/* Encode using filename-safe characters. */
+		len = 5 * len + 1;
+		*id = dst = mem_heap_alloc(heap, len);
+
+		innobase_convert_from_table_id(cs, dst, str, len);
+	}
+
+	return(ptr);
+}
+
+/*************************************************************************
+Tries to scan a column name. */
+static
+const char*
+dict_scan_col(
+/*==========*/
+					/* out: scanned to */
+	struct charset_info_st*	cs,	/* in: the character set of ptr */
+	const char*		ptr,	/* in: scanned to */
+	ibool*			success,/* out: TRUE if success */
+	dict_table_t*		table,	/* in: table in which the column is */
+	const dict_col_t**	column,	/* out: pointer to column if success */
+	mem_heap_t*		heap,	/* in: heap where to allocate */
+	const char**		name)	/* out,own: the column name;
+					NULL if no name was scannable */
+{
+	ulint		i;
+
+	*success = FALSE;
+
+	ptr = dict_scan_id(cs, ptr, heap, name, FALSE, TRUE);
+
+	if (*name == NULL) {
+
+		return(ptr);	/* Syntax error */
+	}
+
+	if (table == NULL) {
+		*success = TRUE;
+		*column = NULL;
+	} else {
+		for (i = 0; i < dict_table_get_n_cols(table); i++) {
+
+			const char*	col_name = dict_table_get_col_name(
+				table, i);
+
+			if (0 == innobase_strcasecmp(col_name, *name)) {
+				/* Found */
+
+				*success = TRUE;
+				*column = dict_table_get_nth_col(table, i);
+				strcpy((char*) *name, col_name);
+
+				break;
+			}
+		}
+	}
+
+	return(ptr);
+}
+
+/*************************************************************************
+Scans a table name from an SQL string. */
+static
+const char*
+dict_scan_table_name(
+/*=================*/
+				/* out: scanned to */
+	struct charset_info_st*	cs,/* in: the character set of ptr */
+	const char*	ptr,	/* in: scanned to */
+	dict_table_t**	table,	/* out: table object or NULL */
+	const char*	name,	/* in: foreign key table name */
+	ibool*		success,/* out: TRUE if ok name found */
+	mem_heap_t*	heap,	/* in: heap where to allocate the id */
+	const char**	ref_name)/* out,own: the table name;
+				NULL if no name was scannable */
+{
+	const char*	database_name	= NULL;
+	ulint		database_name_len = 0;
+	const char*	table_name	= NULL;
+	ulint		table_name_len;
+	const char*	scan_name;
+	char*		ref;
+
+	*success = FALSE;
+	*table = NULL;
+
+	ptr = dict_scan_id(cs, ptr, heap, &scan_name, TRUE, FALSE);
+
+	if (scan_name == NULL) {
+
+		return(ptr);	/* Syntax error */
+	}
+
+	if (*ptr == '.') {
+		/* We scanned the database name; scan also the table name */
+
+		ptr++;
+
+		database_name = scan_name;
+		database_name_len = strlen(database_name);
+
+		ptr = dict_scan_id(cs, ptr, heap, &table_name, TRUE, FALSE);
+
+		if (table_name == NULL) {
+
+			return(ptr);	/* Syntax error */
+		}
+	} else {
+		/* To be able to read table dumps made with InnoDB-4.0.17 or
+		earlier, we must allow the dot separator between the database
+		name and the table name also to appear within a quoted
+		identifier! InnoDB used to print a constraint as:
+		... REFERENCES `databasename.tablename` ...
+		starting from 4.0.18 it is
+		... REFERENCES `databasename`.`tablename` ... */
+		const char* s;
+
+		for (s = scan_name; *s; s++) {
+			if (*s == '.') {
+				database_name = scan_name;
+				database_name_len = s - scan_name;
+				scan_name = ++s;
+				break;/* to do: multiple dots? */
+			}
+		}
+
+		table_name = scan_name;
+	}
+
+	if (database_name == NULL) {
+		/* Use the database name of the foreign key table */
+
+		database_name = name;
+		database_name_len = dict_get_db_name_len(name);
+	}
+
+	table_name_len = strlen(table_name);
+
+	/* Copy database_name, '/', table_name, '\0' */
+	ref = mem_heap_alloc(heap, database_name_len + table_name_len + 2);
+	memcpy(ref, database_name, database_name_len);
+	ref[database_name_len] = '/';
+	memcpy(ref + database_name_len + 1, table_name, table_name_len + 1);
+#ifndef __WIN__
+	if (srv_lower_case_table_names) {
+#endif /* !__WIN__ */
+		/* The table name is always put to lower case on Windows. */
+		innobase_casedn_str(ref);
+#ifndef __WIN__
+	}
+#endif /* !__WIN__ */
+
+	*success = TRUE;
+	*ref_name = ref;
+	*table = dict_table_get_low(ref);
+
+	return(ptr);
+}
+
+/*************************************************************************
+Skips one id. The id is allowed to contain also '.'. */
+static
+const char*
+dict_skip_word(
+/*===========*/
+				/* out: scanned to */
+	struct charset_info_st*	cs,/* in: the character set of ptr */
+	const char*	ptr,	/* in: scanned to */
+	ibool*		success)/* out: TRUE if success, FALSE if just spaces
+				left in string or a syntax error */
+{
+	const char*	start;
+
+	*success = FALSE;
+
+	ptr = dict_scan_id(cs, ptr, NULL, &start, FALSE, TRUE);
+
+	if (start) {
+		*success = TRUE;
+	}
+
+	return(ptr);
+}
+
+/*************************************************************************
+Removes MySQL comments from an SQL string. A comment is either
+(a) '#' to the end of the line,
+(b) '--<space>' to the end of the line, or
+(c) '<slash><asterisk>' till the next '<asterisk><slash>' (like the familiar
+C comment syntax). */
+static
+char*
+dict_strip_comments(
+/*================*/
+					/* out, own: SQL string stripped from
+					comments; the caller must free this
+					with mem_free()! */
+	const char*	sql_string)	/* in: SQL string */
+{
+	char*		str;
+	const char*	sptr;
+	char*		ptr;
+	/* unclosed quote character (0 if none) */
+	char		quote	= 0;
+
+	str = mem_alloc(strlen(sql_string) + 1);
+
+	sptr = sql_string;
+	ptr = str;
+
+	for (;;) {
+scan_more:
+		if (*sptr == '\0') {
+			*ptr = '\0';
+
+			ut_a(ptr <= str + strlen(sql_string));
+
+			return(str);
+		}
+
+		if (*sptr == quote) {
+			/* Closing quote character: do not look for
+			starting quote or comments. */
+			quote = 0;
+		} else if (quote) {
+			/* Within quotes: do not look for
+			starting quotes or comments. */
+		} else if (*sptr == '"' || *sptr == '`') {
+			/* Starting quote: remember the quote character. */
+			quote = *sptr;
+		} else if (*sptr == '#'
+			   || (sptr[0] == '-' && sptr[1] == '-'
+			       && sptr[2] == ' ')) {
+			for (;;) {
+				/* In Unix a newline is 0x0A while in Windows
+				it is 0x0D followed by 0x0A */
+
+				if (*sptr == (char)0x0A
+				    || *sptr == (char)0x0D
+				    || *sptr == '\0') {
+
+					goto scan_more;
+				}
+
+				sptr++;
+			}
+		} else if (!quote && *sptr == '/' && *(sptr + 1) == '*') {
+			for (;;) {
+				if (*sptr == '*' && *(sptr + 1) == '/') {
+
+					sptr += 2;
+
+					goto scan_more;
+				}
+
+				if (*sptr == '\0') {
+
+					goto scan_more;
+				}
+
+				sptr++;
+			}
+		}
+
+		*ptr = *sptr;
+
+		ptr++;
+		sptr++;
+	}
+}
+
+/*************************************************************************
+Finds the highest <number> for foreign key constraints of the table. Looks
+only at the >= 4.0.18-format id's, which are of the form
+databasename/tablename_ibfk_<number>. */
+static
+ulint
+dict_table_get_highest_foreign_id(
+/*==============================*/
+				/* out: highest number, 0 if table has no new
+				format foreign key constraints */
+	dict_table_t*	table)	/* in: table in the dictionary memory cache */
+{
+	dict_foreign_t*	foreign;
+	char*		endp;
+	ulint		biggest_id	= 0;
+	ulint		id;
+	ulint		len;
+
+	ut_a(table);
+
+	len = ut_strlen(table->name);
+	foreign = UT_LIST_GET_FIRST(table->foreign_list);
+
+	while (foreign) {
+		if (ut_strlen(foreign->id) > ((sizeof dict_ibfk) - 1) + len
+		    && 0 == ut_memcmp(foreign->id, table->name, len)
+		    && 0 == ut_memcmp(foreign->id + len,
+				      dict_ibfk, (sizeof dict_ibfk) - 1)
+		    && foreign->id[len + ((sizeof dict_ibfk) - 1)] != '0') {
+			/* It is of the >= 4.0.18 format */
+
+			id = strtoul(foreign->id + len
+				     + ((sizeof dict_ibfk) - 1),
+				     &endp, 10);
+			if (*endp == '\0') {
+				ut_a(id != biggest_id);
+
+				if (id > biggest_id) {
+					biggest_id = id;
+				}
+			}
+		}
+
+		foreign = UT_LIST_GET_NEXT(foreign_list, foreign);
+	}
+
+	return(biggest_id);
+}
+
+/*************************************************************************
+Reports a simple foreign key create clause syntax error. */
+static
+void
+dict_foreign_report_syntax_err(
+/*===========================*/
+	const char*	name,		/* in: table name */
+	const char*	start_of_latest_foreign,
+					/* in: start of the foreign key clause
+					in the SQL string */
+	const char*	ptr)		/* in: place of the syntax error */
+{
+	FILE*	ef = dict_foreign_err_file;
+
+	mutex_enter(&dict_foreign_err_mutex);
+	dict_foreign_error_report_low(ef, name);
+	fprintf(ef, "%s:\nSyntax error close to:\n%s\n",
+		start_of_latest_foreign, ptr);
+	mutex_exit(&dict_foreign_err_mutex);
+}
+
+/*************************************************************************
+Scans a table create SQL string and adds to the data dictionary the foreign
+key constraints declared in the string. This function should be called after
+the indexes for a table have been created. Each foreign key constraint must
+be accompanied with indexes in both participating tables. The indexes are
+allowed to contain more fields than mentioned in the constraint. */
+static
+ulint
+dict_create_foreign_constraints_low(
+/*================================*/
+				/* out: error code or DB_SUCCESS */
+	trx_t*		trx,	/* in: transaction */
+	mem_heap_t*	heap,	/* in: memory heap */
+	struct charset_info_st*	cs,/* in: the character set of sql_string */
+	const char*	sql_string,
+				/* in: CREATE TABLE or ALTER TABLE statement
+				where foreign keys are declared like:
+				FOREIGN KEY (a, b) REFERENCES table2(c, d),
+				table2 can be written also with the database
+				name before it: test.table2; the default
+				database is the database of parameter name */
+	const char*	name,	/* in: table full name in the normalized form
+				database_name/table_name */
+	ibool		reject_fks)
+				/* in: if TRUE, fail with error code
+				DB_CANNOT_ADD_CONSTRAINT if any foreign
+				keys are found. */
+{
+	dict_table_t*	table;
+	dict_table_t*	referenced_table;
+	dict_table_t*	table_to_alter;
+	ulint		highest_id_so_far	= 0;
+	dict_index_t*	index;
+	dict_foreign_t*	foreign;
+	const char*	ptr			= sql_string;
+	const char*	start_of_latest_foreign	= sql_string;
+	FILE*		ef			= dict_foreign_err_file;
+	const char*	constraint_name;
+	ibool		success;
+	ulint		error;
+	const char*	ptr1;
+	const char*	ptr2;
+	ulint		i;
+	ulint		j;
+	ibool		is_on_delete;
+	ulint		n_on_deletes;
+	ulint		n_on_updates;
+	const dict_col_t*columns[500];
+	const char*	column_names[500];
+	const char*	referenced_table_name;
+
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+
+	table = dict_table_get_low(name);
+
+	if (table == NULL) {
+		mutex_enter(&dict_foreign_err_mutex);
+		dict_foreign_error_report_low(ef, name);
+		fprintf(ef,
+			"Cannot find the table in the internal"
+			" data dictionary of InnoDB.\n"
+			"Create table statement:\n%s\n", sql_string);
+		mutex_exit(&dict_foreign_err_mutex);
+
+		return(DB_ERROR);
+	}
+
+	/* First check if we are actually doing an ALTER TABLE, and in that
+	case look for the table being altered */
+
+	ptr = dict_accept(cs, ptr, "ALTER", &success);
+
+	if (!success) {
+
+		goto loop;
+	}
+
+	ptr = dict_accept(cs, ptr, "TABLE", &success);
+
+	if (!success) {
+
+		goto loop;
+	}
+
+	/* We are doing an ALTER TABLE: scan the table name we are altering */
+
+	ptr = dict_scan_table_name(cs, ptr, &table_to_alter, name,
+				   &success, heap, &referenced_table_name);
+	if (!success) {
+		fprintf(stderr,
+			"InnoDB: Error: could not find"
+			" the table being ALTERED in:\n%s\n",
+			sql_string);
+
+		return(DB_ERROR);
+	}
+
+	/* Starting from 4.0.18 and 4.1.2, we generate foreign key id's in the
+	format databasename/tablename_ibfk_<number>, where <number> is local
+	to the table; look for the highest <number> for table_to_alter, so
+	that we can assign to new constraints higher numbers. */
+
+	/* If we are altering a temporary table, the table name after ALTER
+	TABLE does not correspond to the internal table name, and
+	table_to_alter is NULL. TODO: should we fix this somehow? */
+
+	if (table_to_alter == NULL) {
+		highest_id_so_far = 0;
+	} else {
+		highest_id_so_far = dict_table_get_highest_foreign_id(
+			table_to_alter);
+	}
+
+	/* Scan for foreign key declarations in a loop */
+loop:
+	/* Scan either to "CONSTRAINT" or "FOREIGN", whichever is closer */
+
+	ptr1 = dict_scan_to(ptr, "CONSTRAINT");
+	ptr2 = dict_scan_to(ptr, "FOREIGN");
+
+	constraint_name = NULL;
+
+	if (ptr1 < ptr2) {
+		/* The user may have specified a constraint name. Pick it so
+		that we can store 'databasename/constraintname' as the id of
+		of the constraint to system tables. */
+		ptr = ptr1;
+
+		ptr = dict_accept(cs, ptr, "CONSTRAINT", &success);
+
+		ut_a(success);
+
+		if (!my_isspace(cs, *ptr) && *ptr != '"' && *ptr != '`') {
+			goto loop;
+		}
+
+		while (my_isspace(cs, *ptr)) {
+			ptr++;
+		}
+
+		/* read constraint name unless got "CONSTRAINT FOREIGN" */
+		if (ptr != ptr2) {
+			ptr = dict_scan_id(cs, ptr, heap,
+					   &constraint_name, FALSE, FALSE);
+		}
+	} else {
+		ptr = ptr2;
+	}
+
+	if (*ptr == '\0') {
+		/* The proper way to reject foreign keys for temporary
+		tables would be to split the lexing and syntactical
+		analysis of foreign key clauses from the actual adding
+		of them, so that ha_innodb.cc could first parse the SQL
+		command, determine if there are any foreign keys, and
+		if so, immediately reject the command if the table is a
+		temporary one. For now, this kludge will work. */
+		if (reject_fks && (UT_LIST_GET_LEN(table->foreign_list) > 0)) {
+
+			return(DB_CANNOT_ADD_CONSTRAINT);
+		}
+
+		/**********************************************************/
+		/* The following call adds the foreign key constraints
+		to the data dictionary system tables on disk */
+
+		error = dict_create_add_foreigns_to_dictionary(
+			highest_id_so_far, table, trx);
+		return(error);
+	}
+
+	start_of_latest_foreign = ptr;
+
+	ptr = dict_accept(cs, ptr, "FOREIGN", &success);
+
+	if (!success) {
+		goto loop;
+	}
+
+	if (!my_isspace(cs, *ptr)) {
+		goto loop;
+	}
+
+	ptr = dict_accept(cs, ptr, "KEY", &success);
+
+	if (!success) {
+		goto loop;
+	}
+
+	ptr = dict_accept(cs, ptr, "(", &success);
+
+	if (!success) {
+		/* MySQL allows also an index id before the '('; we
+		skip it */
+		ptr = dict_skip_word(cs, ptr, &success);
+
+		if (!success) {
+			dict_foreign_report_syntax_err(
+				name, start_of_latest_foreign, ptr);
+
+			return(DB_CANNOT_ADD_CONSTRAINT);
+		}
+
+		ptr = dict_accept(cs, ptr, "(", &success);
+
+		if (!success) {
+			/* We do not flag a syntax error here because in an
+			ALTER TABLE we may also have DROP FOREIGN KEY abc */
+
+			goto loop;
+		}
+	}
+
+	i = 0;
+
+	/* Scan the columns in the first list */
+col_loop1:
+	ut_a(i < (sizeof column_names) / sizeof *column_names);
+	ptr = dict_scan_col(cs, ptr, &success, table, columns + i,
+			    heap, column_names + i);
+	if (!success) {
+		mutex_enter(&dict_foreign_err_mutex);
+		dict_foreign_error_report_low(ef, name);
+		fprintf(ef, "%s:\nCannot resolve column name close to:\n%s\n",
+			start_of_latest_foreign, ptr);
+		mutex_exit(&dict_foreign_err_mutex);
+
+		return(DB_CANNOT_ADD_CONSTRAINT);
+	}
+
+	i++;
+
+	ptr = dict_accept(cs, ptr, ",", &success);
+
+	if (success) {
+		goto col_loop1;
+	}
+
+	ptr = dict_accept(cs, ptr, ")", &success);
+
+	if (!success) {
+		dict_foreign_report_syntax_err(
+			name, start_of_latest_foreign, ptr);
+		return(DB_CANNOT_ADD_CONSTRAINT);
+	}
+
+	/* Try to find an index which contains the columns
+	as the first fields and in the right order */
+
+	index = dict_foreign_find_index(table, column_names, i,
+					NULL, TRUE, FALSE);
+
+	if (!index) {
+		mutex_enter(&dict_foreign_err_mutex);
+		dict_foreign_error_report_low(ef, name);
+		fputs("There is no index in table ", ef);
+		ut_print_name(ef, NULL, TRUE, name);
+		fprintf(ef, " where the columns appear\n"
+			"as the first columns. Constraint:\n%s\n"
+			"See http://dev.mysql.com/doc/refman/5.1/en/"
+			"innodb-foreign-key-constraints.html\n"
+			"for correct foreign key definition.\n",
+			start_of_latest_foreign);
+		mutex_exit(&dict_foreign_err_mutex);
+
+		return(DB_CANNOT_ADD_CONSTRAINT);
+	}
+	ptr = dict_accept(cs, ptr, "REFERENCES", &success);
+
+	if (!success || !my_isspace(cs, *ptr)) {
+		dict_foreign_report_syntax_err(
+			name, start_of_latest_foreign, ptr);
+		return(DB_CANNOT_ADD_CONSTRAINT);
+	}
+
+	/* Let us create a constraint struct */
+
+	foreign = dict_mem_foreign_create();
+
+	if (constraint_name) {
+		ulint	db_len;
+
+		/* Catenate 'databasename/' to the constraint name specified
+		by the user: we conceive the constraint as belonging to the
+		same MySQL 'database' as the table itself. We store the name
+		to foreign->id. */
+
+		db_len = dict_get_db_name_len(table->name);
+
+		foreign->id = mem_heap_alloc(
+			foreign->heap, db_len + strlen(constraint_name) + 2);
+
+		ut_memcpy(foreign->id, table->name, db_len);
+		foreign->id[db_len] = '/';
+		strcpy(foreign->id + db_len + 1, constraint_name);
+	}
+
+	foreign->foreign_table = table;
+	foreign->foreign_table_name = mem_heap_strdup(foreign->heap,
+						      table->name);
+	foreign->foreign_index = index;
+	foreign->n_fields = (unsigned int) i;
+	foreign->foreign_col_names = mem_heap_alloc(foreign->heap,
+						    i * sizeof(void*));
+	for (i = 0; i < foreign->n_fields; i++) {
+		foreign->foreign_col_names[i] = mem_heap_strdup(
+			foreign->heap,
+			dict_table_get_col_name(table,
+						dict_col_get_no(columns[i])));
+	}
+
+	ptr = dict_scan_table_name(cs, ptr, &referenced_table, name,
+				   &success, heap, &referenced_table_name);
+
+	/* Note that referenced_table can be NULL if the user has suppressed
+	checking of foreign key constraints! */
+
+	if (!success || (!referenced_table && trx->check_foreigns)) {
+		dict_foreign_free(foreign);
+
+		mutex_enter(&dict_foreign_err_mutex);
+		dict_foreign_error_report_low(ef, name);
+		fprintf(ef, "%s:\nCannot resolve table name close to:\n"
+			"%s\n",
+			start_of_latest_foreign, ptr);
+		mutex_exit(&dict_foreign_err_mutex);
+
+		return(DB_CANNOT_ADD_CONSTRAINT);
+	}
+
+	ptr = dict_accept(cs, ptr, "(", &success);
+
+	if (!success) {
+		dict_foreign_free(foreign);
+		dict_foreign_report_syntax_err(name, start_of_latest_foreign,
+					       ptr);
+		return(DB_CANNOT_ADD_CONSTRAINT);
+	}
+
+	/* Scan the columns in the second list */
+	i = 0;
+
+col_loop2:
+	ptr = dict_scan_col(cs, ptr, &success, referenced_table, columns + i,
+			    heap, column_names + i);
+	i++;
+
+	if (!success) {
+		dict_foreign_free(foreign);
+
+		mutex_enter(&dict_foreign_err_mutex);
+		dict_foreign_error_report_low(ef, name);
+		fprintf(ef, "%s:\nCannot resolve column name close to:\n"
+			"%s\n",
+			start_of_latest_foreign, ptr);
+		mutex_exit(&dict_foreign_err_mutex);
+
+		return(DB_CANNOT_ADD_CONSTRAINT);
+	}
+
+	ptr = dict_accept(cs, ptr, ",", &success);
+
+	if (success) {
+		goto col_loop2;
+	}
+
+	ptr = dict_accept(cs, ptr, ")", &success);
+
+	if (!success || foreign->n_fields != i) {
+		dict_foreign_free(foreign);
+
+		dict_foreign_report_syntax_err(name, start_of_latest_foreign,
+					       ptr);
+		return(DB_CANNOT_ADD_CONSTRAINT);
+	}
+
+	n_on_deletes = 0;
+	n_on_updates = 0;
+
+scan_on_conditions:
+	/* Loop here as long as we can find ON ... conditions */
+
+	ptr = dict_accept(cs, ptr, "ON", &success);
+
+	if (!success) {
+
+		goto try_find_index;
+	}
+
+	ptr = dict_accept(cs, ptr, "DELETE", &success);
+
+	if (!success) {
+		ptr = dict_accept(cs, ptr, "UPDATE", &success);
+
+		if (!success) {
+			dict_foreign_free(foreign);
+
+			dict_foreign_report_syntax_err(
+				name, start_of_latest_foreign, ptr);
+			return(DB_CANNOT_ADD_CONSTRAINT);
+		}
+
+		is_on_delete = FALSE;
+		n_on_updates++;
+	} else {
+		is_on_delete = TRUE;
+		n_on_deletes++;
+	}
+
+	ptr = dict_accept(cs, ptr, "RESTRICT", &success);
+
+	if (success) {
+		goto scan_on_conditions;
+	}
+
+	ptr = dict_accept(cs, ptr, "CASCADE", &success);
+
+	if (success) {
+		if (is_on_delete) {
+			foreign->type |= DICT_FOREIGN_ON_DELETE_CASCADE;
+		} else {
+			foreign->type |= DICT_FOREIGN_ON_UPDATE_CASCADE;
+		}
+
+		goto scan_on_conditions;
+	}
+
+	ptr = dict_accept(cs, ptr, "NO", &success);
+
+	if (success) {
+		ptr = dict_accept(cs, ptr, "ACTION", &success);
+
+		if (!success) {
+			dict_foreign_free(foreign);
+			dict_foreign_report_syntax_err(
+				name, start_of_latest_foreign, ptr);
+
+			return(DB_CANNOT_ADD_CONSTRAINT);
+		}
+
+		if (is_on_delete) {
+			foreign->type |= DICT_FOREIGN_ON_DELETE_NO_ACTION;
+		} else {
+			foreign->type |= DICT_FOREIGN_ON_UPDATE_NO_ACTION;
+		}
+
+		goto scan_on_conditions;
+	}
+
+	ptr = dict_accept(cs, ptr, "SET", &success);
+
+	if (!success) {
+		dict_foreign_free(foreign);
+		dict_foreign_report_syntax_err(name, start_of_latest_foreign,
+					       ptr);
+		return(DB_CANNOT_ADD_CONSTRAINT);
+	}
+
+	ptr = dict_accept(cs, ptr, "NULL", &success);
+
+	if (!success) {
+		dict_foreign_free(foreign);
+		dict_foreign_report_syntax_err(name, start_of_latest_foreign,
+					       ptr);
+		return(DB_CANNOT_ADD_CONSTRAINT);
+	}
+
+	for (j = 0; j < foreign->n_fields; j++) {
+		if ((dict_index_get_nth_col(foreign->foreign_index, j)->prtype)
+		    & DATA_NOT_NULL) {
+
+			/* It is not sensible to define SET NULL
+			if the column is not allowed to be NULL! */
+
+			dict_foreign_free(foreign);
+
+			mutex_enter(&dict_foreign_err_mutex);
+			dict_foreign_error_report_low(ef, name);
+			fprintf(ef, "%s:\n"
+				"You have defined a SET NULL condition"
+				" though some of the\n"
+				"columns are defined as NOT NULL.\n",
+				start_of_latest_foreign);
+			mutex_exit(&dict_foreign_err_mutex);
+
+			return(DB_CANNOT_ADD_CONSTRAINT);
+		}
+	}
+
+	if (is_on_delete) {
+		foreign->type |= DICT_FOREIGN_ON_DELETE_SET_NULL;
+	} else {
+		foreign->type |= DICT_FOREIGN_ON_UPDATE_SET_NULL;
+	}
+
+	goto scan_on_conditions;
+
+try_find_index:
+	if (n_on_deletes > 1 || n_on_updates > 1) {
+		/* It is an error to define more than 1 action */
+
+		dict_foreign_free(foreign);
+
+		mutex_enter(&dict_foreign_err_mutex);
+		dict_foreign_error_report_low(ef, name);
+		fprintf(ef, "%s:\n"
+			"You have twice an ON DELETE clause"
+			" or twice an ON UPDATE clause.\n",
+			start_of_latest_foreign);
+		mutex_exit(&dict_foreign_err_mutex);
+
+		return(DB_CANNOT_ADD_CONSTRAINT);
+	}
+
+	/* Try to find an index which contains the columns as the first fields
+	and in the right order, and the types are the same as in
+	foreign->foreign_index */
+
+	if (referenced_table) {
+		index = dict_foreign_find_index(referenced_table,
+						column_names, i,
+						foreign->foreign_index,
+						TRUE, FALSE);
+		if (!index) {
+			dict_foreign_free(foreign);
+			mutex_enter(&dict_foreign_err_mutex);
+			dict_foreign_error_report_low(ef, name);
+			fprintf(ef, "%s:\n"
+				"Cannot find an index in the"
+				" referenced table where the\n"
+				"referenced columns appear as the"
+				" first columns, or column types\n"
+				"in the table and the referenced table"
+				" do not match for constraint.\n"
+				"Note that the internal storage type of"
+				" ENUM and SET changed in\n"
+				"tables created with >= InnoDB-4.1.12,"
+				" and such columns in old tables\n"
+				"cannot be referenced by such columns"
+				" in new tables.\n"
+				"See http://dev.mysql.com/doc/refman/5.1/en/"
+				"innodb-foreign-key-constraints.html\n"
+				"for correct foreign key definition.\n",
+				start_of_latest_foreign);
+			mutex_exit(&dict_foreign_err_mutex);
+
+			return(DB_CANNOT_ADD_CONSTRAINT);
+		}
+	} else {
+		ut_a(trx->check_foreigns == FALSE);
+		index = NULL;
+	}
+
+	foreign->referenced_index = index;
+	foreign->referenced_table = referenced_table;
+
+	foreign->referenced_table_name
+		= mem_heap_strdup(foreign->heap, referenced_table_name);
+
+	foreign->referenced_col_names = mem_heap_alloc(foreign->heap,
+						       i * sizeof(void*));
+	for (i = 0; i < foreign->n_fields; i++) {
+		foreign->referenced_col_names[i]
+			= mem_heap_strdup(foreign->heap, column_names[i]);
+	}
+
+	/* We found an ok constraint definition: add to the lists */
+
+	UT_LIST_ADD_LAST(foreign_list, table->foreign_list, foreign);
+
+	if (referenced_table) {
+		UT_LIST_ADD_LAST(referenced_list,
+				 referenced_table->referenced_list,
+				 foreign);
+	}
+
+	goto loop;
+}
+
+/*************************************************************************
+Scans a table create SQL string and adds to the data dictionary the foreign
+key constraints declared in the string. This function should be called after
+the indexes for a table have been created. Each foreign key constraint must
+be accompanied with indexes in both participating tables. The indexes are
+allowed to contain more fields than mentioned in the constraint. */
+UNIV_INTERN
+ulint
+dict_create_foreign_constraints(
+/*============================*/
+					/* out: error code or DB_SUCCESS */
+	trx_t*		trx,		/* in: transaction */
+	const char*	sql_string,	/* in: table create statement where
+					foreign keys are declared like:
+					FOREIGN KEY (a, b) REFERENCES
+					table2(c, d), table2 can be written
+					also with the database
+					name before it: test.table2; the
+					default database id the database of
+					parameter name */
+	const char*	name,		/* in: table full name in the
+					normalized form
+					database_name/table_name */
+	ibool		reject_fks)	/* in: if TRUE, fail with error
+					code DB_CANNOT_ADD_CONSTRAINT if
+					any foreign keys are found. */
+{
+	char*			str;
+	ulint			err;
+	mem_heap_t*		heap;
+
+	ut_a(trx);
+	ut_a(trx->mysql_thd);
+
+	str = dict_strip_comments(sql_string);
+	heap = mem_heap_create(10000);
+
+	err = dict_create_foreign_constraints_low(
+		trx, heap, innobase_get_charset(trx->mysql_thd), str, name,
+		reject_fks);
+
+	mem_heap_free(heap);
+	mem_free(str);
+
+	return(err);
+}
+
+/**************************************************************************
+Parses the CONSTRAINT id's to be dropped in an ALTER TABLE statement. */
+UNIV_INTERN
+ulint
+dict_foreign_parse_drop_constraints(
+/*================================*/
+						/* out: DB_SUCCESS or
+						DB_CANNOT_DROP_CONSTRAINT if
+						syntax error or the constraint
+						id does not match */
+	mem_heap_t*	heap,			/* in: heap from which we can
+						allocate memory */
+	trx_t*		trx,			/* in: transaction */
+	dict_table_t*	table,			/* in: table */
+	ulint*		n,			/* out: number of constraints
+						to drop */
+	const char***	constraints_to_drop)	/* out: id's of the
+						constraints to drop */
+{
+	dict_foreign_t*		foreign;
+	ibool			success;
+	char*			str;
+	const char*		ptr;
+	const char*		id;
+	FILE*			ef	= dict_foreign_err_file;
+	struct charset_info_st*	cs;
+
+	ut_a(trx);
+	ut_a(trx->mysql_thd);
+
+	cs = innobase_get_charset(trx->mysql_thd);
+
+	*n = 0;
+
+	*constraints_to_drop = mem_heap_alloc(heap, 1000 * sizeof(char*));
+
+	str = dict_strip_comments(*(trx->mysql_query_str));
+	ptr = str;
+
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+loop:
+	ptr = dict_scan_to(ptr, "DROP");
+
+	if (*ptr == '\0') {
+		mem_free(str);
+
+		return(DB_SUCCESS);
+	}
+
+	ptr = dict_accept(cs, ptr, "DROP", &success);
+
+	if (!my_isspace(cs, *ptr)) {
+
+		goto loop;
+	}
+
+	ptr = dict_accept(cs, ptr, "FOREIGN", &success);
+
+	if (!success || !my_isspace(cs, *ptr)) {
+
+		goto loop;
+	}
+
+	ptr = dict_accept(cs, ptr, "KEY", &success);
+
+	if (!success) {
+
+		goto syntax_error;
+	}
+
+	ptr = dict_scan_id(cs, ptr, heap, &id, FALSE, TRUE);
+
+	if (id == NULL) {
+
+		goto syntax_error;
+	}
+
+	ut_a(*n < 1000);
+	(*constraints_to_drop)[*n] = id;
+	(*n)++;
+
+	/* Look for the given constraint id */
+
+	foreign = UT_LIST_GET_FIRST(table->foreign_list);
+
+	while (foreign != NULL) {
+		if (0 == strcmp(foreign->id, id)
+		    || (strchr(foreign->id, '/')
+			&& 0 == strcmp(id,
+				       dict_remove_db_name(foreign->id)))) {
+			/* Found */
+			break;
+		}
+
+		foreign = UT_LIST_GET_NEXT(foreign_list, foreign);
+	}
+
+	if (foreign == NULL) {
+		mutex_enter(&dict_foreign_err_mutex);
+		rewind(ef);
+		ut_print_timestamp(ef);
+		fputs(" Error in dropping of a foreign key constraint"
+		      " of table ", ef);
+		ut_print_name(ef, NULL, TRUE, table->name);
+		fputs(",\n"
+		      "in SQL command\n", ef);
+		fputs(str, ef);
+		fputs("\nCannot find a constraint with the given id ", ef);
+		ut_print_name(ef, NULL, FALSE, id);
+		fputs(".\n", ef);
+		mutex_exit(&dict_foreign_err_mutex);
+
+		mem_free(str);
+
+		return(DB_CANNOT_DROP_CONSTRAINT);
+	}
+
+	goto loop;
+
+syntax_error:
+	mutex_enter(&dict_foreign_err_mutex);
+	rewind(ef);
+	ut_print_timestamp(ef);
+	fputs(" Syntax error in dropping of a"
+	      " foreign key constraint of table ", ef);
+	ut_print_name(ef, NULL, TRUE, table->name);
+	fprintf(ef, ",\n"
+		"close to:\n%s\n in SQL command\n%s\n", ptr, str);
+	mutex_exit(&dict_foreign_err_mutex);
+
+	mem_free(str);
+
+	return(DB_CANNOT_DROP_CONSTRAINT);
+}
+#endif /* UNIV_HOTBACKUP */
+
+/*==================== END OF FOREIGN KEY PROCESSING ====================*/
+
+/**************************************************************************
+Returns an index object if it is found in the dictionary cache.
+Assumes that dict_sys->mutex is already being held. */
+UNIV_INTERN
+dict_index_t*
+dict_index_get_if_in_cache_low(
+/*===========================*/
+				/* out: index, NULL if not found */
+	dulint	index_id)	/* in: index id */
+{
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+
+	return(dict_index_find_on_id_low(index_id));
+}
+
+#if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
+/**************************************************************************
+Returns an index object if it is found in the dictionary cache. */
+UNIV_INTERN
+dict_index_t*
+dict_index_get_if_in_cache(
+/*=======================*/
+				/* out: index, NULL if not found */
+	dulint	index_id)	/* in: index id */
+{
+	dict_index_t*	index;
+
+	if (dict_sys == NULL) {
+		return(NULL);
+	}
+
+	mutex_enter(&(dict_sys->mutex));
+
+	index = dict_index_get_if_in_cache_low(index_id);
+
+	mutex_exit(&(dict_sys->mutex));
+
+	return(index);
+}
+#endif /* UNIV_DEBUG || UNIV_BUF_DEBUG */
+
+#ifdef UNIV_DEBUG
+/**************************************************************************
+Checks that a tuple has n_fields_cmp value in a sensible range, so that
+no comparison can occur with the page number field in a node pointer. */
+UNIV_INTERN
+ibool
+dict_index_check_search_tuple(
+/*==========================*/
+					/* out: TRUE if ok */
+	const dict_index_t*	index,	/* in: index tree */
+	const dtuple_t*		tuple)	/* in: tuple used in a search */
+{
+	ut_a(index);
+	ut_a(dtuple_get_n_fields_cmp(tuple)
+	     <= dict_index_get_n_unique_in_tree(index));
+	return(TRUE);
+}
+#endif /* UNIV_DEBUG */
+
+/**************************************************************************
+Builds a node pointer out of a physical record and a page number. */
+UNIV_INTERN
+dtuple_t*
+dict_index_build_node_ptr(
+/*======================*/
+					/* out, own: node pointer */
+	const dict_index_t*	index,	/* in: index */
+	const rec_t*		rec,	/* in: record for which to build node
+					pointer */
+	ulint			page_no,/* in: page number to put in node
+					pointer */
+	mem_heap_t*		heap,	/* in: memory heap where pointer
+					created */
+	ulint			level)	/* in: level of rec in tree:
+					0 means leaf level */
+{
+	dtuple_t*	tuple;
+	dfield_t*	field;
+	byte*		buf;
+	ulint		n_unique;
+
+	if (UNIV_UNLIKELY(index->type & DICT_UNIVERSAL)) {
+		/* In a universal index tree, we take the whole record as
+		the node pointer if the record is on the leaf level,
+		on non-leaf levels we remove the last field, which
+		contains the page number of the child page */
+
+		ut_a(!dict_table_is_comp(index->table));
+		n_unique = rec_get_n_fields_old(rec);
+
+		if (level > 0) {
+			ut_a(n_unique > 1);
+			n_unique--;
+		}
+	} else {
+		n_unique = dict_index_get_n_unique_in_tree(index);
+	}
+
+	tuple = dtuple_create(heap, n_unique + 1);
+
+	/* When searching in the tree for the node pointer, we must not do
+	comparison on the last field, the page number field, as on upper
+	levels in the tree there may be identical node pointers with a
+	different page number; therefore, we set the n_fields_cmp to one
+	less: */
+
+	dtuple_set_n_fields_cmp(tuple, n_unique);
+
+	dict_index_copy_types(tuple, index, n_unique);
+
+	buf = mem_heap_alloc(heap, 4);
+
+	mach_write_to_4(buf, page_no);
+
+	field = dtuple_get_nth_field(tuple, n_unique);
+	dfield_set_data(field, buf, 4);
+
+	dtype_set(dfield_get_type(field), DATA_SYS_CHILD, DATA_NOT_NULL, 4);
+
+	rec_copy_prefix_to_dtuple(tuple, rec, index, n_unique, heap);
+	dtuple_set_info_bits(tuple, dtuple_get_info_bits(tuple)
+			     | REC_STATUS_NODE_PTR);
+
+	ut_ad(dtuple_check_typed(tuple));
+
+	return(tuple);
+}
+
+/**************************************************************************
+Copies an initial segment of a physical record, long enough to specify an
+index entry uniquely. */
+UNIV_INTERN
+rec_t*
+dict_index_copy_rec_order_prefix(
+/*=============================*/
+					/* out: pointer to the prefix record */
+	const dict_index_t*	index,	/* in: index */
+	const rec_t*		rec,	/* in: record for which to
+					copy prefix */
+	ulint*			n_fields,/* out: number of fields copied */
+	byte**			buf,	/* in/out: memory buffer for the
+					copied prefix, or NULL */
+	ulint*			buf_size)/* in/out: buffer size */
+{
+	ulint		n;
+
+	UNIV_PREFETCH_R(rec);
+
+	if (UNIV_UNLIKELY(index->type & DICT_UNIVERSAL)) {
+		ut_a(!dict_table_is_comp(index->table));
+		n = rec_get_n_fields_old(rec);
+	} else {
+		n = dict_index_get_n_unique_in_tree(index);
+	}
+
+	*n_fields = n;
+	return(rec_copy_prefix_to_buf(rec, index, n, buf, buf_size));
+}
+
+/**************************************************************************
+Builds a typed data tuple out of a physical record. */
+UNIV_INTERN
+dtuple_t*
+dict_index_build_data_tuple(
+/*========================*/
+				/* out, own: data tuple */
+	dict_index_t*	index,	/* in: index tree */
+	rec_t*		rec,	/* in: record for which to build data tuple */
+	ulint		n_fields,/* in: number of data fields */
+	mem_heap_t*	heap)	/* in: memory heap where tuple created */
+{
+	dtuple_t*	tuple;
+
+	ut_ad(dict_table_is_comp(index->table)
+	      || n_fields <= rec_get_n_fields_old(rec));
+
+	tuple = dtuple_create(heap, n_fields);
+
+	dict_index_copy_types(tuple, index, n_fields);
+
+	rec_copy_prefix_to_dtuple(tuple, rec, index, n_fields, heap);
+
+	ut_ad(dtuple_check_typed(tuple));
+
+	return(tuple);
+}
+
+/*************************************************************************
+Calculates the minimum record length in an index. */
+UNIV_INTERN
+ulint
+dict_index_calc_min_rec_len(
+/*========================*/
+	const dict_index_t*	index)	/* in: index */
+{
+	ulint	sum	= 0;
+	ulint	i;
+
+	if (dict_table_is_comp(index->table)) {
+		ulint nullable = 0;
+		sum = REC_N_NEW_EXTRA_BYTES;
+		for (i = 0; i < dict_index_get_n_fields(index); i++) {
+			const dict_col_t*	col
+				= dict_index_get_nth_col(index, i);
+			ulint	size = dict_col_get_fixed_size(col);
+			sum += size;
+			if (!size) {
+				size = col->len;
+				sum += size < 128 ? 1 : 2;
+			}
+			if (!(col->prtype & DATA_NOT_NULL)) {
+				nullable++;
+			}
+		}
+
+		/* round the NULL flags up to full bytes */
+		sum += UT_BITS_IN_BYTES(nullable);
+
+		return(sum);
+	}
+
+	for (i = 0; i < dict_index_get_n_fields(index); i++) {
+		sum += dict_col_get_fixed_size(
+			dict_index_get_nth_col(index, i));
+	}
+
+	if (sum > 127) {
+		sum += 2 * dict_index_get_n_fields(index);
+	} else {
+		sum += dict_index_get_n_fields(index);
+	}
+
+	sum += REC_N_OLD_EXTRA_BYTES;
+
+	return(sum);
+}
+
+/*************************************************************************
+Calculates new estimates for table and index statistics. The statistics
+are used in query optimization. */
+UNIV_INTERN
+void
+dict_update_statistics_low(
+/*=======================*/
+	dict_table_t*	table,		/* in/out: table */
+	ibool		has_dict_mutex __attribute__((unused)))
+					/* in: TRUE if the caller has the
+					dictionary mutex */
+{
+	dict_index_t*	index;
+	ulint		size;
+	ulint		sum_of_index_sizes	= 0;
+
+	if (table->ibd_file_missing) {
+		ut_print_timestamp(stderr);
+		fprintf(stderr,
+			"  InnoDB: cannot calculate statistics for table %s\n"
+			"InnoDB: because the .ibd file is missing.  For help,"
+			" please refer to\n"
+			"InnoDB: http://dev.mysql.com/doc/refman/5.1/en/"
+			"innodb-troubleshooting.html\n",
+			table->name);
+
+		return;
+	}
+
+	/* If we have set a high innodb_force_recovery level, do not calculate
+	statistics, as a badly corrupted index can cause a crash in it. */
+
+	if (srv_force_recovery >= SRV_FORCE_NO_IBUF_MERGE) {
+
+		return;
+	}
+
+	/* Find out the sizes of the indexes and how many different values
+	for the key they approximately have */
+
+	index = dict_table_get_first_index(table);
+
+	if (index == NULL) {
+		/* Table definition is corrupt */
+
+		return;
+	}
+
+	while (index) {
+		size = btr_get_size(index, BTR_TOTAL_SIZE);
+
+		index->stat_index_size = size;
+
+		sum_of_index_sizes += size;
+
+		size = btr_get_size(index, BTR_N_LEAF_PAGES);
+
+		if (size == 0) {
+			/* The root node of the tree is a leaf */
+			size = 1;
+		}
+
+		index->stat_n_leaf_pages = size;
+
+		btr_estimate_number_of_different_key_vals(index);
+
+		index = dict_table_get_next_index(index);
+	}
+
+	index = dict_table_get_first_index(table);
+
+	table->stat_n_rows = index->stat_n_diff_key_vals[
+		dict_index_get_n_unique(index)];
+
+	table->stat_clustered_index_size = index->stat_index_size;
+
+	table->stat_sum_of_other_index_sizes = sum_of_index_sizes
+		- index->stat_index_size;
+
+	table->stat_initialized = TRUE;
+
+	table->stat_modified_counter = 0;
+}
+
+/*************************************************************************
+Calculates new estimates for table and index statistics. The statistics
+are used in query optimization. */
+UNIV_INTERN
+void
+dict_update_statistics(
+/*===================*/
+	dict_table_t*	table)	/* in/out: table */
+{
+	dict_update_statistics_low(table, FALSE);
+}
+
+/**************************************************************************
+Prints info of a foreign key constraint. */
+static
+void
+dict_foreign_print_low(
+/*===================*/
+	dict_foreign_t*	foreign)	/* in: foreign key constraint */
+{
+	ulint	i;
+
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+
+	fprintf(stderr, "  FOREIGN KEY CONSTRAINT %s: %s (",
+		foreign->id, foreign->foreign_table_name);
+
+	for (i = 0; i < foreign->n_fields; i++) {
+		fprintf(stderr, " %s", foreign->foreign_col_names[i]);
+	}
+
+	fprintf(stderr, " )\n"
+		"             REFERENCES %s (",
+		foreign->referenced_table_name);
+
+	for (i = 0; i < foreign->n_fields; i++) {
+		fprintf(stderr, " %s", foreign->referenced_col_names[i]);
+	}
+
+	fputs(" )\n", stderr);
+}
+
+/**************************************************************************
+Prints a table data. */
+UNIV_INTERN
+void
+dict_table_print(
+/*=============*/
+	dict_table_t*	table)	/* in: table */
+{
+	mutex_enter(&(dict_sys->mutex));
+	dict_table_print_low(table);
+	mutex_exit(&(dict_sys->mutex));
+}
+
+/**************************************************************************
+Prints a table data when we know the table name. */
+UNIV_INTERN
+void
+dict_table_print_by_name(
+/*=====================*/
+	const char*	name)
+{
+	dict_table_t*	table;
+
+	mutex_enter(&(dict_sys->mutex));
+
+	table = dict_table_get_low(name);
+
+	ut_a(table);
+
+	dict_table_print_low(table);
+	mutex_exit(&(dict_sys->mutex));
+}
+
+/**************************************************************************
+Prints a table data. */
+UNIV_INTERN
+void
+dict_table_print_low(
+/*=================*/
+	dict_table_t*	table)	/* in: table */
+{
+	dict_index_t*	index;
+	dict_foreign_t*	foreign;
+	ulint		i;
+
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+
+	dict_update_statistics_low(table, TRUE);
+
+	fprintf(stderr,
+		"--------------------------------------\n"
+		"TABLE: name %s, id %lu %lu, flags %lx, columns %lu,"
+		" indexes %lu, appr.rows %lu\n"
+		"  COLUMNS: ",
+		table->name,
+		(ulong) ut_dulint_get_high(table->id),
+		(ulong) ut_dulint_get_low(table->id),
+		(ulong) table->flags,
+		(ulong) table->n_cols,
+		(ulong) UT_LIST_GET_LEN(table->indexes),
+		(ulong) table->stat_n_rows);
+
+	for (i = 0; i + 1 < (ulint) table->n_cols; i++) {
+		dict_col_print_low(table, dict_table_get_nth_col(table, i));
+		fputs("; ", stderr);
+	}
+
+	putc('\n', stderr);
+
+	index = UT_LIST_GET_FIRST(table->indexes);
+
+	while (index != NULL) {
+		dict_index_print_low(index);
+		index = UT_LIST_GET_NEXT(indexes, index);
+	}
+
+	foreign = UT_LIST_GET_FIRST(table->foreign_list);
+
+	while (foreign != NULL) {
+		dict_foreign_print_low(foreign);
+		foreign = UT_LIST_GET_NEXT(foreign_list, foreign);
+	}
+
+	foreign = UT_LIST_GET_FIRST(table->referenced_list);
+
+	while (foreign != NULL) {
+		dict_foreign_print_low(foreign);
+		foreign = UT_LIST_GET_NEXT(referenced_list, foreign);
+	}
+}
+
+/**************************************************************************
+Prints a column data. */
+static
+void
+dict_col_print_low(
+/*===============*/
+	const dict_table_t*	table,	/* in: table */
+	const dict_col_t*	col)	/* in: column */
+{
+	dtype_t	type;
+
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+
+	dict_col_copy_type(col, &type);
+	fprintf(stderr, "%s: ", dict_table_get_col_name(table,
+							dict_col_get_no(col)));
+
+	dtype_print(&type);
+}
+
+/**************************************************************************
+Prints an index data. */
+static
+void
+dict_index_print_low(
+/*=================*/
+	dict_index_t*	index)	/* in: index */
+{
+	ib_int64_t	n_vals;
+	ulint		i;
+	const char*	type_string;
+
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+
+	if (index->n_user_defined_cols > 0) {
+		n_vals = index->stat_n_diff_key_vals[
+			index->n_user_defined_cols];
+	} else {
+		n_vals = index->stat_n_diff_key_vals[1];
+	}
+
+	if (dict_index_is_clust(index)) {
+		type_string = "clustered index";
+	} else if (dict_index_is_unique(index)) {
+		type_string = "unique index";
+	} else {
+		type_string = "secondary index";
+	}
+
+	fprintf(stderr,
+		"  INDEX: name %s, id %lu %lu, fields %lu/%lu,"
+		" uniq %lu, type %lu\n"
+		"   root page %lu, appr.key vals %lu,"
+		" leaf pages %lu, size pages %lu\n"
+		"   FIELDS: ",
+		index->name,
+		(ulong) ut_dulint_get_high(index->id),
+		(ulong) ut_dulint_get_low(index->id),
+		(ulong) index->n_user_defined_cols,
+		(ulong) index->n_fields,
+		(ulong) index->n_uniq,
+		(ulong) index->type,
+		(ulong) index->page,
+		(ulong) n_vals,
+		(ulong) index->stat_n_leaf_pages,
+		(ulong) index->stat_index_size);
+
+	for (i = 0; i < index->n_fields; i++) {
+		dict_field_print_low(dict_index_get_nth_field(index, i));
+	}
+
+	putc('\n', stderr);
+
+#ifdef UNIV_BTR_PRINT
+	btr_print_size(index);
+
+	btr_print_index(index, 7);
+#endif /* UNIV_BTR_PRINT */
+}
+
+/**************************************************************************
+Prints a field data. */
+static
+void
+dict_field_print_low(
+/*=================*/
+	dict_field_t*	field)	/* in: field */
+{
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+
+	fprintf(stderr, " %s", field->name);
+
+	if (field->prefix_len != 0) {
+		fprintf(stderr, "(%lu)", (ulong) field->prefix_len);
+	}
+}
+
+/**************************************************************************
+Outputs info on a foreign key of a table in a format suitable for
+CREATE TABLE. */
+UNIV_INTERN
+void
+dict_print_info_on_foreign_key_in_create_format(
+/*============================================*/
+	FILE*		file,		/* in: file where to print */
+	trx_t*		trx,		/* in: transaction */
+	dict_foreign_t*	foreign,	/* in: foreign key constraint */
+	ibool		add_newline)	/* in: whether to add a newline */
+{
+	const char*	stripped_id;
+	ulint	i;
+
+	if (strchr(foreign->id, '/')) {
+		/* Strip the preceding database name from the constraint id */
+		stripped_id = foreign->id + 1
+			+ dict_get_db_name_len(foreign->id);
+	} else {
+		stripped_id = foreign->id;
+	}
+
+	putc(',', file);
+
+	if (add_newline) {
+		/* SHOW CREATE TABLE wants constraints each printed nicely
+		on its own line, while error messages want no newlines
+		inserted. */
+		fputs("\n ", file);
+	}
+
+	fputs(" CONSTRAINT ", file);
+	ut_print_name(file, trx, FALSE, stripped_id);
+	fputs(" FOREIGN KEY (", file);
+
+	for (i = 0;;) {
+		ut_print_name(file, trx, FALSE, foreign->foreign_col_names[i]);
+		if (++i < foreign->n_fields) {
+			fputs(", ", file);
+		} else {
+			break;
+		}
+	}
+
+	fputs(") REFERENCES ", file);
+
+	if (dict_tables_have_same_db(foreign->foreign_table_name,
+				     foreign->referenced_table_name)) {
+		/* Do not print the database name of the referenced table */
+		ut_print_name(file, trx, TRUE,
+			      dict_remove_db_name(
+				      foreign->referenced_table_name));
+	} else {
+		ut_print_name(file, trx, TRUE,
+			      foreign->referenced_table_name);
+	}
+
+	putc(' ', file);
+	putc('(', file);
+
+	for (i = 0;;) {
+		ut_print_name(file, trx, FALSE,
+			      foreign->referenced_col_names[i]);
+		if (++i < foreign->n_fields) {
+			fputs(", ", file);
+		} else {
+			break;
+		}
+	}
+
+	putc(')', file);
+
+	if (foreign->type & DICT_FOREIGN_ON_DELETE_CASCADE) {
+		fputs(" ON DELETE CASCADE", file);
+	}
+
+	if (foreign->type & DICT_FOREIGN_ON_DELETE_SET_NULL) {
+		fputs(" ON DELETE SET NULL", file);
+	}
+
+	if (foreign->type & DICT_FOREIGN_ON_DELETE_NO_ACTION) {
+		fputs(" ON DELETE NO ACTION", file);
+	}
+
+	if (foreign->type & DICT_FOREIGN_ON_UPDATE_CASCADE) {
+		fputs(" ON UPDATE CASCADE", file);
+	}
+
+	if (foreign->type & DICT_FOREIGN_ON_UPDATE_SET_NULL) {
+		fputs(" ON UPDATE SET NULL", file);
+	}
+
+	if (foreign->type & DICT_FOREIGN_ON_UPDATE_NO_ACTION) {
+		fputs(" ON UPDATE NO ACTION", file);
+	}
+}
+
+/**************************************************************************
+Outputs info on foreign keys of a table. */
+UNIV_INTERN
+void
+dict_print_info_on_foreign_keys(
+/*============================*/
+	ibool		create_table_format, /* in: if TRUE then print in
+				a format suitable to be inserted into
+				a CREATE TABLE, otherwise in the format
+				of SHOW TABLE STATUS */
+	FILE*		file,	/* in: file where to print */
+	trx_t*		trx,	/* in: transaction */
+	dict_table_t*	table)	/* in: table */
+{
+	dict_foreign_t*	foreign;
+
+	mutex_enter(&(dict_sys->mutex));
+
+	foreign = UT_LIST_GET_FIRST(table->foreign_list);
+
+	if (foreign == NULL) {
+		mutex_exit(&(dict_sys->mutex));
+
+		return;
+	}
+
+	while (foreign != NULL) {
+		if (create_table_format) {
+			dict_print_info_on_foreign_key_in_create_format(
+				file, trx, foreign, TRUE);
+		} else {
+			ulint	i;
+			fputs("; (", file);
+
+			for (i = 0; i < foreign->n_fields; i++) {
+				if (i) {
+					putc(' ', file);
+				}
+
+				ut_print_name(file, trx, FALSE,
+					      foreign->foreign_col_names[i]);
+			}
+
+			fputs(") REFER ", file);
+			ut_print_name(file, trx, TRUE,
+				      foreign->referenced_table_name);
+			putc('(', file);
+
+			for (i = 0; i < foreign->n_fields; i++) {
+				if (i) {
+					putc(' ', file);
+				}
+				ut_print_name(
+					file, trx, FALSE,
+					foreign->referenced_col_names[i]);
+			}
+
+			putc(')', file);
+
+			if (foreign->type == DICT_FOREIGN_ON_DELETE_CASCADE) {
+				fputs(" ON DELETE CASCADE", file);
+			}
+
+			if (foreign->type == DICT_FOREIGN_ON_DELETE_SET_NULL) {
+				fputs(" ON DELETE SET NULL", file);
+			}
+
+			if (foreign->type & DICT_FOREIGN_ON_DELETE_NO_ACTION) {
+				fputs(" ON DELETE NO ACTION", file);
+			}
+
+			if (foreign->type & DICT_FOREIGN_ON_UPDATE_CASCADE) {
+				fputs(" ON UPDATE CASCADE", file);
+			}
+
+			if (foreign->type & DICT_FOREIGN_ON_UPDATE_SET_NULL) {
+				fputs(" ON UPDATE SET NULL", file);
+			}
+
+			if (foreign->type & DICT_FOREIGN_ON_UPDATE_NO_ACTION) {
+				fputs(" ON UPDATE NO ACTION", file);
+			}
+		}
+
+		foreign = UT_LIST_GET_NEXT(foreign_list, foreign);
+	}
+
+	mutex_exit(&(dict_sys->mutex));
+}
+
+/************************************************************************
+Displays the names of the index and the table. */
+UNIV_INTERN
+void
+dict_index_name_print(
+/*==================*/
+	FILE*			file,	/* in: output stream */
+	trx_t*			trx,	/* in: transaction */
+	const dict_index_t*	index)	/* in: index to print */
+{
+	fputs("index ", file);
+	ut_print_name(file, trx, FALSE, index->name);
+	fputs(" of table ", file);
+	ut_print_name(file, trx, TRUE, index->table_name);
+}
+
+/**************************************************************************
+Get index by name */
+UNIV_INTERN
+dict_index_t*
+dict_table_get_index_on_name(
+/*=========================*/
+				/* out: index, NULL if does not exist */
+	dict_table_t*	table,	/* in: table */
+	const char*	name)	/* in: name of the index to find */
+{
+	dict_index_t*	index;
+
+	index = dict_table_get_first_index(table);
+
+	while (index != NULL) {
+		if (ut_strcmp(index->name, name) == 0) {
+
+			return(index);
+		}
+
+		index = dict_table_get_next_index(index);
+	}
+
+	return(NULL);
+
+}
+
+/**************************************************************************
+Replace the index passed in with another equivalent index in the tables
+foreign key list. */
+UNIV_INTERN
+void
+dict_table_replace_index_in_foreign_list(
+/*=====================================*/
+	dict_table_t*	table,  /* in/out: table */
+	dict_index_t*	index)	/* in: index to be replaced */
+{
+	dict_foreign_t*	foreign;
+
+	for (foreign = UT_LIST_GET_FIRST(table->foreign_list);
+	     foreign;
+	     foreign = UT_LIST_GET_NEXT(foreign_list, foreign)) {
+
+		if (foreign->foreign_index == index) {
+			dict_index_t*	new_index
+				= dict_foreign_find_equiv_index(foreign);
+			ut_a(new_index);
+
+			foreign->foreign_index = new_index;
+		}
+	}
+}
+
+/**************************************************************************
+In case there is more than one index with the same name return the index
+with the min(id). */
+UNIV_INTERN
+dict_index_t*
+dict_table_get_index_on_name_and_min_id(
+/*=====================================*/
+				/* out: index, NULL if does not exist */
+	dict_table_t*	table,	/* in: table */
+	const char*	name)	/* in: name of the index to find */
+{
+	dict_index_t*	index;
+	dict_index_t*	min_index; /* Index with matching name and min(id) */
+
+	min_index = NULL;
+	index = dict_table_get_first_index(table);
+
+	while (index != NULL) {
+		if (ut_strcmp(index->name, name) == 0) {
+			if (!min_index
+			    || ut_dulint_cmp(index->id, min_index->id) < 0) {
+
+				min_index = index;
+			}
+		}
+
+		index = dict_table_get_next_index(index);
+	}
+
+	return(min_index);
+
+}
+
+#ifdef UNIV_DEBUG
+/**************************************************************************
+Check for duplicate index entries in a table [using the index name] */
+UNIV_INTERN
+void
+dict_table_check_for_dup_indexes(
+/*=============================*/
+	const dict_table_t*	table)	/* in: Check for dup indexes
+					in this table */
+{
+	/* Check for duplicates, ignoring indexes that are marked
+	as to be dropped */
+
+	const dict_index_t*	index1;
+	const dict_index_t*	index2;
+
+	/* The primary index _must_ exist */
+	ut_a(UT_LIST_GET_LEN(table->indexes) > 0);
+
+	index1 = UT_LIST_GET_FIRST(table->indexes);
+	index2 = UT_LIST_GET_NEXT(indexes, index1);
+
+	while (index1 && index2) {
+
+		while (index2) {
+
+			if (!index2->to_be_dropped) {
+				ut_ad(ut_strcmp(index1->name, index2->name));
+			}
+
+			index2 = UT_LIST_GET_NEXT(indexes, index2);
+		}
+
+		index1 = UT_LIST_GET_NEXT(indexes, index1);
+		index2 = UT_LIST_GET_NEXT(indexes, index1);
+	}
+}
+#endif /* UNIV_DEBUG */
diff --git a/storage/xtradb/dict/dict0load.c b/storage/xtradb/dict/dict0load.c
new file mode 100644
index 00000000000..44590a261a6
--- /dev/null
+++ b/storage/xtradb/dict/dict0load.c
@@ -0,0 +1,1460 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Loads to the memory cache database object definitions
+from dictionary tables
+
+Created 4/24/1996 Heikki Tuuri
+*******************************************************/
+
+#include "dict0load.h"
+#ifndef UNIV_HOTBACKUP
+#include "mysql_version.h"
+#endif /* !UNIV_HOTBACKUP */
+
+#ifdef UNIV_NONINL
+#include "dict0load.ic"
+#endif
+
+#include "btr0pcur.h"
+#include "btr0btr.h"
+#include "page0page.h"
+#include "mach0data.h"
+#include "dict0dict.h"
+#include "dict0boot.h"
+#include "rem0cmp.h"
+#include "srv0start.h"
+#include "srv0srv.h"
+
+/********************************************************************
+Returns TRUE if index's i'th column's name is 'name' .*/
+static
+ibool
+name_of_col_is(
+/*===========*/
+				/* out: */
+	dict_table_t*	table,	/* in: table */
+	dict_index_t*	index,	/* in: index */
+	ulint		i,	/* in:  */
+	const char*	name)	/* in: name to compare to */
+{
+	ulint	tmp = dict_col_get_no(dict_field_get_col(
+					      dict_index_get_nth_field(
+						      index, i)));
+
+	return(strcmp(name, dict_table_get_col_name(table, tmp)) == 0);
+}
+
+/************************************************************************
+Finds the first table name in the given database. */
+UNIV_INTERN
+char*
+dict_get_first_table_name_in_db(
+/*============================*/
+				/* out, own: table name, NULL if
+				does not exist; the caller must
+				free the memory in the string! */
+	const char*	name)	/* in: database name which ends in '/' */
+{
+	dict_table_t*	sys_tables;
+	btr_pcur_t	pcur;
+	dict_index_t*	sys_index;
+	dtuple_t*	tuple;
+	mem_heap_t*	heap;
+	dfield_t*	dfield;
+	const rec_t*	rec;
+	const byte*	field;
+	ulint		len;
+	mtr_t		mtr;
+
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+
+	heap = mem_heap_create(1000);
+
+	mtr_start(&mtr);
+
+	sys_tables = dict_table_get_low("SYS_TABLES");
+	sys_index = UT_LIST_GET_FIRST(sys_tables->indexes);
+	ut_a(!dict_table_is_comp(sys_tables));
+
+	tuple = dtuple_create(heap, 1);
+	dfield = dtuple_get_nth_field(tuple, 0);
+
+	dfield_set_data(dfield, name, ut_strlen(name));
+	dict_index_copy_types(tuple, sys_index, 1);
+
+	btr_pcur_open_on_user_rec(sys_index, tuple, PAGE_CUR_GE,
+				  BTR_SEARCH_LEAF, &pcur, &mtr);
+loop:
+	rec = btr_pcur_get_rec(&pcur);
+
+	if (!btr_pcur_is_on_user_rec(&pcur)) {
+		/* Not found */
+
+		btr_pcur_close(&pcur);
+		mtr_commit(&mtr);
+		mem_heap_free(heap);
+
+		return(NULL);
+	}
+
+	field = rec_get_nth_field_old(rec, 0, &len);
+
+	if (len < strlen(name)
+	    || ut_memcmp(name, field, strlen(name)) != 0) {
+		/* Not found */
+
+		btr_pcur_close(&pcur);
+		mtr_commit(&mtr);
+		mem_heap_free(heap);
+
+		return(NULL);
+	}
+
+	if (!rec_get_deleted_flag(rec, 0)) {
+
+		/* We found one */
+
+		char*	table_name = mem_strdupl((char*) field, len);
+
+		btr_pcur_close(&pcur);
+		mtr_commit(&mtr);
+		mem_heap_free(heap);
+
+		return(table_name);
+	}
+
+	btr_pcur_move_to_next_user_rec(&pcur, &mtr);
+
+	goto loop;
+}
+
+/************************************************************************
+Prints to the standard output information on all tables found in the data
+dictionary system table. */
+UNIV_INTERN
+void
+dict_print(void)
+/*============*/
+{
+	dict_table_t*	sys_tables;
+	dict_index_t*	sys_index;
+	dict_table_t*	table;
+	btr_pcur_t	pcur;
+	const rec_t*	rec;
+	const byte*	field;
+	ulint		len;
+	mtr_t		mtr;
+
+	/* Enlarge the fatal semaphore wait timeout during the InnoDB table
+	monitor printout */
+
+	mutex_enter(&kernel_mutex);
+	srv_fatal_semaphore_wait_threshold += 7200; /* 2 hours */
+	mutex_exit(&kernel_mutex);
+
+	mutex_enter(&(dict_sys->mutex));
+
+	mtr_start(&mtr);
+
+	sys_tables = dict_table_get_low("SYS_TABLES");
+	sys_index = UT_LIST_GET_FIRST(sys_tables->indexes);
+
+	btr_pcur_open_at_index_side(TRUE, sys_index, BTR_SEARCH_LEAF, &pcur,
+				    TRUE, &mtr);
+loop:
+	btr_pcur_move_to_next_user_rec(&pcur, &mtr);
+
+	rec = btr_pcur_get_rec(&pcur);
+
+	if (!btr_pcur_is_on_user_rec(&pcur)) {
+		/* end of index */
+
+		btr_pcur_close(&pcur);
+		mtr_commit(&mtr);
+
+		mutex_exit(&(dict_sys->mutex));
+
+		/* Restore the fatal semaphore wait timeout */
+
+		mutex_enter(&kernel_mutex);
+		srv_fatal_semaphore_wait_threshold -= 7200; /* 2 hours */
+		mutex_exit(&kernel_mutex);
+
+		return;
+	}
+
+	field = rec_get_nth_field_old(rec, 0, &len);
+
+	if (!rec_get_deleted_flag(rec, 0)) {
+
+		/* We found one */
+
+		char*	table_name = mem_strdupl((char*) field, len);
+
+		btr_pcur_store_position(&pcur, &mtr);
+
+		mtr_commit(&mtr);
+
+		table = dict_table_get_low(table_name);
+		mem_free(table_name);
+
+		if (table == NULL) {
+			fputs("InnoDB: Failed to load table ", stderr);
+			ut_print_namel(stderr, NULL, TRUE, (char*) field, len);
+			putc('\n', stderr);
+		} else {
+			/* The table definition was corrupt if there
+			is no index */
+
+			if (dict_table_get_first_index(table)) {
+				dict_update_statistics_low(table, TRUE);
+			}
+
+			dict_table_print_low(table);
+		}
+
+		mtr_start(&mtr);
+
+		btr_pcur_restore_position(BTR_SEARCH_LEAF, &pcur, &mtr);
+	}
+
+	goto loop;
+}
+
+/************************************************************************
+Determine the flags of a table described in SYS_TABLES. */
+static
+ulint
+dict_sys_tables_get_flags(
+/*======================*/
+				/* out: compressed page size in kilobytes;
+				or 0 if the tablespace is uncompressed,
+				ULINT_UNDEFINED on error */
+	const rec_t*	rec)	/* in: a record of SYS_TABLES */
+{
+	const byte*	field;
+	ulint		len;
+	ulint		n_cols;
+	ulint		flags;
+
+	field = rec_get_nth_field_old(rec, 5, &len);
+	ut_a(len == 4);
+
+	flags = mach_read_from_4(field);
+
+	if (UNIV_LIKELY(flags == DICT_TABLE_ORDINARY)) {
+		return(0);
+	}
+
+	field = rec_get_nth_field_old(rec, 4, &len);
+	n_cols = mach_read_from_4(field);
+
+	if (UNIV_UNLIKELY(!(n_cols & 0x80000000UL))) {
+		/* New file formats require ROW_FORMAT=COMPACT. */
+		return(ULINT_UNDEFINED);
+	}
+
+	switch (flags & (DICT_TF_FORMAT_MASK | DICT_TF_COMPACT)) {
+	default:
+	case DICT_TF_FORMAT_51 << DICT_TF_FORMAT_SHIFT:
+	case DICT_TF_FORMAT_51 << DICT_TF_FORMAT_SHIFT | DICT_TF_COMPACT:
+		/* flags should be DICT_TABLE_ORDINARY,
+		or DICT_TF_FORMAT_MASK should be nonzero. */
+		return(ULINT_UNDEFINED);
+
+	case DICT_TF_FORMAT_ZIP << DICT_TF_FORMAT_SHIFT | DICT_TF_COMPACT:
+#if DICT_TF_FORMAT_MAX > DICT_TF_FORMAT_ZIP
+# error "missing case labels for DICT_TF_FORMAT_ZIP .. DICT_TF_FORMAT_MAX"
+#endif
+		/* We support this format. */
+		break;
+	}
+
+	if (UNIV_UNLIKELY((flags & DICT_TF_ZSSIZE_MASK)
+			  > (DICT_TF_ZSSIZE_MAX << DICT_TF_ZSSIZE_SHIFT))) {
+		/* Unsupported compressed page size. */
+		return(ULINT_UNDEFINED);
+	}
+
+	if (UNIV_UNLIKELY(flags & (~0 << DICT_TF_BITS))) {
+		/* Some unused bits are set. */
+		return(ULINT_UNDEFINED);
+	}
+
+	return(flags);
+}
+
+/************************************************************************
+In a crash recovery we already have all the tablespace objects created.
+This function compares the space id information in the InnoDB data dictionary
+to what we already read with fil_load_single_table_tablespaces().
+
+In a normal startup, we create the tablespace objects for every table in
+InnoDB's data dictionary, if the corresponding .ibd file exists.
+We also scan the biggest space id, and store it to fil_system. */
+UNIV_INTERN
+void
+dict_check_tablespaces_and_store_max_id(
+/*====================================*/
+	ibool	in_crash_recovery)	/* in: are we doing a crash recovery */
+{
+	dict_table_t*	sys_tables;
+	dict_index_t*	sys_index;
+	btr_pcur_t	pcur;
+	const rec_t*	rec;
+	ulint		max_space_id	= 0;
+	mtr_t		mtr;
+
+	mutex_enter(&(dict_sys->mutex));
+
+	mtr_start(&mtr);
+
+	sys_tables = dict_table_get_low("SYS_TABLES");
+	sys_index = UT_LIST_GET_FIRST(sys_tables->indexes);
+	ut_a(!dict_table_is_comp(sys_tables));
+
+	btr_pcur_open_at_index_side(TRUE, sys_index, BTR_SEARCH_LEAF, &pcur,
+				    TRUE, &mtr);
+loop:
+	btr_pcur_move_to_next_user_rec(&pcur, &mtr);
+
+	rec = btr_pcur_get_rec(&pcur);
+
+	if (!btr_pcur_is_on_user_rec(&pcur)) {
+		/* end of index */
+
+		btr_pcur_close(&pcur);
+		mtr_commit(&mtr);
+
+		/* We must make the tablespace cache aware of the biggest
+		known space id */
+
+		/* printf("Biggest space id in data dictionary %lu\n",
+		max_space_id); */
+		fil_set_max_space_id_if_bigger(max_space_id);
+
+		mutex_exit(&(dict_sys->mutex));
+
+		return;
+	}
+
+	if (!rec_get_deleted_flag(rec, 0)) {
+
+		/* We found one */
+		const byte*	field;
+		ulint		len;
+		ulint		space_id;
+		ulint		flags;
+		char*		name;
+
+		field = rec_get_nth_field_old(rec, 0, &len);
+		name = mem_strdupl((char*) field, len);
+
+		flags = dict_sys_tables_get_flags(rec);
+		if (UNIV_UNLIKELY(flags == ULINT_UNDEFINED)) {
+
+			field = rec_get_nth_field_old(rec, 5, &len);
+			flags = mach_read_from_4(field);
+
+			ut_print_timestamp(stderr);
+			fputs("  InnoDB: Error: table ", stderr);
+			ut_print_filename(stderr, name);
+			fprintf(stderr, "\n"
+				"InnoDB: in InnoDB data dictionary"
+				" has unknown type %lx.\n",
+				(ulong) flags);
+
+			goto loop;
+		}
+
+		field = rec_get_nth_field_old(rec, 9, &len);
+		ut_a(len == 4);
+
+		space_id = mach_read_from_4(field);
+
+		btr_pcur_store_position(&pcur, &mtr);
+
+		mtr_commit(&mtr);
+
+		if (space_id != 0 && in_crash_recovery) {
+			/* Check that the tablespace (the .ibd file) really
+			exists; print a warning to the .err log if not */
+
+			fil_space_for_table_exists_in_mem(space_id, name,
+							  FALSE, TRUE, TRUE);
+		}
+
+		if (space_id != 0 && !in_crash_recovery) {
+			/* It is a normal database startup: create the space
+			object and check that the .ibd file exists. */
+
+			fil_open_single_table_tablespace(FALSE, space_id,
+							 flags, name);
+		}
+
+		mem_free(name);
+
+		if (space_id > max_space_id) {
+			max_space_id = space_id;
+		}
+
+		mtr_start(&mtr);
+
+		btr_pcur_restore_position(BTR_SEARCH_LEAF, &pcur, &mtr);
+	}
+
+	goto loop;
+}
+
+/************************************************************************
+Loads definitions for table columns. */
+static
+void
+dict_load_columns(
+/*==============*/
+	dict_table_t*	table,	/* in: table */
+	mem_heap_t*	heap)	/* in: memory heap for temporary storage */
+{
+	dict_table_t*	sys_columns;
+	dict_index_t*	sys_index;
+	btr_pcur_t	pcur;
+	dtuple_t*	tuple;
+	dfield_t*	dfield;
+	const rec_t*	rec;
+	const byte*	field;
+	ulint		len;
+	byte*		buf;
+	char*		name;
+	ulint		mtype;
+	ulint		prtype;
+	ulint		col_len;
+	ulint		i;
+	mtr_t		mtr;
+
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+
+	mtr_start(&mtr);
+
+	sys_columns = dict_table_get_low("SYS_COLUMNS");
+	sys_index = UT_LIST_GET_FIRST(sys_columns->indexes);
+	ut_a(!dict_table_is_comp(sys_columns));
+
+	tuple = dtuple_create(heap, 1);
+	dfield = dtuple_get_nth_field(tuple, 0);
+
+	buf = mem_heap_alloc(heap, 8);
+	mach_write_to_8(buf, table->id);
+
+	dfield_set_data(dfield, buf, 8);
+	dict_index_copy_types(tuple, sys_index, 1);
+
+	btr_pcur_open_on_user_rec(sys_index, tuple, PAGE_CUR_GE,
+				  BTR_SEARCH_LEAF, &pcur, &mtr);
+	for (i = 0; i + DATA_N_SYS_COLS < (ulint) table->n_cols; i++) {
+
+		rec = btr_pcur_get_rec(&pcur);
+
+		ut_a(btr_pcur_is_on_user_rec(&pcur));
+
+		ut_a(!rec_get_deleted_flag(rec, 0));
+
+		field = rec_get_nth_field_old(rec, 0, &len);
+		ut_ad(len == 8);
+		ut_a(ut_dulint_cmp(table->id, mach_read_from_8(field)) == 0);
+
+		field = rec_get_nth_field_old(rec, 1, &len);
+		ut_ad(len == 4);
+		ut_a(i == mach_read_from_4(field));
+
+		ut_a(name_of_col_is(sys_columns, sys_index, 4, "NAME"));
+
+		field = rec_get_nth_field_old(rec, 4, &len);
+		name = mem_heap_strdupl(heap, (char*) field, len);
+
+		field = rec_get_nth_field_old(rec, 5, &len);
+		mtype = mach_read_from_4(field);
+
+		field = rec_get_nth_field_old(rec, 6, &len);
+		prtype = mach_read_from_4(field);
+
+		if (dtype_get_charset_coll(prtype) == 0
+		    && dtype_is_string_type(mtype)) {
+			/* The table was created with < 4.1.2. */
+
+			if (dtype_is_binary_string_type(mtype, prtype)) {
+				/* Use the binary collation for
+				string columns of binary type. */
+
+				prtype = dtype_form_prtype(
+					prtype,
+					DATA_MYSQL_BINARY_CHARSET_COLL);
+			} else {
+				/* Use the default charset for
+				other than binary columns. */
+
+				prtype = dtype_form_prtype(
+					prtype,
+					data_mysql_default_charset_coll);
+			}
+		}
+
+		field = rec_get_nth_field_old(rec, 7, &len);
+		col_len = mach_read_from_4(field);
+
+		ut_a(name_of_col_is(sys_columns, sys_index, 8, "PREC"));
+
+		dict_mem_table_add_col(table, heap, name,
+				       mtype, prtype, col_len);
+		btr_pcur_move_to_next_user_rec(&pcur, &mtr);
+	}
+
+	btr_pcur_close(&pcur);
+	mtr_commit(&mtr);
+}
+
+/************************************************************************
+Loads definitions for index fields. */
+static
+void
+dict_load_fields(
+/*=============*/
+	dict_index_t*	index,	/* in: index whose fields to load */
+	mem_heap_t*	heap)	/* in: memory heap for temporary storage */
+{
+	dict_table_t*	sys_fields;
+	dict_index_t*	sys_index;
+	btr_pcur_t	pcur;
+	dtuple_t*	tuple;
+	dfield_t*	dfield;
+	ulint		pos_and_prefix_len;
+	ulint		prefix_len;
+	const rec_t*	rec;
+	const byte*	field;
+	ulint		len;
+	byte*		buf;
+	ulint		i;
+	mtr_t		mtr;
+
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+
+	mtr_start(&mtr);
+
+	sys_fields = dict_table_get_low("SYS_FIELDS");
+	sys_index = UT_LIST_GET_FIRST(sys_fields->indexes);
+	ut_a(!dict_table_is_comp(sys_fields));
+
+	tuple = dtuple_create(heap, 1);
+	dfield = dtuple_get_nth_field(tuple, 0);
+
+	buf = mem_heap_alloc(heap, 8);
+	mach_write_to_8(buf, index->id);
+
+	dfield_set_data(dfield, buf, 8);
+	dict_index_copy_types(tuple, sys_index, 1);
+
+	btr_pcur_open_on_user_rec(sys_index, tuple, PAGE_CUR_GE,
+				  BTR_SEARCH_LEAF, &pcur, &mtr);
+	for (i = 0; i < index->n_fields; i++) {
+
+		rec = btr_pcur_get_rec(&pcur);
+
+		ut_a(btr_pcur_is_on_user_rec(&pcur));
+
+		/* There could be delete marked records in SYS_FIELDS
+		because SYS_FIELDS.INDEX_ID can be updated
+		by ALTER TABLE ADD INDEX. */
+
+		if (rec_get_deleted_flag(rec, 0)) {
+
+			goto next_rec;
+		}
+
+		field = rec_get_nth_field_old(rec, 0, &len);
+		ut_ad(len == 8);
+
+		field = rec_get_nth_field_old(rec, 1, &len);
+		ut_a(len == 4);
+
+		/* The next field stores the field position in the index
+		and a possible column prefix length if the index field
+		does not contain the whole column. The storage format is
+		like this: if there is at least one prefix field in the index,
+		then the HIGH 2 bytes contain the field number (== i) and the
+		low 2 bytes the prefix length for the field. Otherwise the
+		field number (== i) is contained in the 2 LOW bytes. */
+
+		pos_and_prefix_len = mach_read_from_4(field);
+
+		ut_a((pos_and_prefix_len & 0xFFFFUL) == i
+		     || (pos_and_prefix_len & 0xFFFF0000UL) == (i << 16));
+
+		if ((i == 0 && pos_and_prefix_len > 0)
+		    || (pos_and_prefix_len & 0xFFFF0000UL) > 0) {
+
+			prefix_len = pos_and_prefix_len & 0xFFFFUL;
+		} else {
+			prefix_len = 0;
+		}
+
+		ut_a(name_of_col_is(sys_fields, sys_index, 4, "COL_NAME"));
+
+		field = rec_get_nth_field_old(rec, 4, &len);
+
+		dict_mem_index_add_field(index,
+					 mem_heap_strdupl(heap,
+							  (char*) field, len),
+					 prefix_len);
+
+next_rec:
+		btr_pcur_move_to_next_user_rec(&pcur, &mtr);
+	}
+
+	btr_pcur_close(&pcur);
+	mtr_commit(&mtr);
+}
+
+/************************************************************************
+Loads definitions for table indexes. Adds them to the data dictionary
+cache. */
+static
+ulint
+dict_load_indexes(
+/*==============*/
+				/* out: DB_SUCCESS if ok, DB_CORRUPTION
+				if corruption of dictionary table or
+				DB_UNSUPPORTED if table has unknown index
+				type */
+	dict_table_t*	table,	/* in: table */
+	mem_heap_t*	heap)	/* in: memory heap for temporary storage */
+{
+	dict_table_t*	sys_indexes;
+	dict_index_t*	sys_index;
+	dict_index_t*	index;
+	btr_pcur_t	pcur;
+	dtuple_t*	tuple;
+	dfield_t*	dfield;
+	const rec_t*	rec;
+	const byte*	field;
+	ulint		len;
+	ulint		name_len;
+	char*		name_buf;
+	ulint		type;
+	ulint		space;
+	ulint		page_no;
+	ulint		n_fields;
+	byte*		buf;
+	ibool		is_sys_table;
+	dulint		id;
+	mtr_t		mtr;
+	ulint		error = DB_SUCCESS;
+
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+
+	if ((ut_dulint_get_high(table->id) == 0)
+	    && (ut_dulint_get_low(table->id) < DICT_HDR_FIRST_ID)) {
+		is_sys_table = TRUE;
+	} else {
+		is_sys_table = FALSE;
+	}
+
+	mtr_start(&mtr);
+
+	sys_indexes = dict_table_get_low("SYS_INDEXES");
+	sys_index = UT_LIST_GET_FIRST(sys_indexes->indexes);
+	ut_a(!dict_table_is_comp(sys_indexes));
+
+	tuple = dtuple_create(heap, 1);
+	dfield = dtuple_get_nth_field(tuple, 0);
+
+	buf = mem_heap_alloc(heap, 8);
+	mach_write_to_8(buf, table->id);
+
+	dfield_set_data(dfield, buf, 8);
+	dict_index_copy_types(tuple, sys_index, 1);
+
+	btr_pcur_open_on_user_rec(sys_index, tuple, PAGE_CUR_GE,
+				  BTR_SEARCH_LEAF, &pcur, &mtr);
+	for (;;) {
+		if (!btr_pcur_is_on_user_rec(&pcur)) {
+
+			break;
+		}
+
+		rec = btr_pcur_get_rec(&pcur);
+
+		field = rec_get_nth_field_old(rec, 0, &len);
+		ut_ad(len == 8);
+
+		if (ut_memcmp(buf, field, len) != 0) {
+			break;
+		} else if (rec_get_deleted_flag(rec, 0)) {
+			/* Skip delete marked records */
+			goto next_rec;
+		}
+
+		field = rec_get_nth_field_old(rec, 1, &len);
+		ut_ad(len == 8);
+		id = mach_read_from_8(field);
+
+		ut_a(name_of_col_is(sys_indexes, sys_index, 4, "NAME"));
+
+		field = rec_get_nth_field_old(rec, 4, &name_len);
+		name_buf = mem_heap_strdupl(heap, (char*) field, name_len);
+
+		field = rec_get_nth_field_old(rec, 5, &len);
+		n_fields = mach_read_from_4(field);
+
+		field = rec_get_nth_field_old(rec, 6, &len);
+		type = mach_read_from_4(field);
+
+		field = rec_get_nth_field_old(rec, 7, &len);
+		space = mach_read_from_4(field);
+
+		ut_a(name_of_col_is(sys_indexes, sys_index, 8, "PAGE_NO"));
+
+		field = rec_get_nth_field_old(rec, 8, &len);
+		page_no = mach_read_from_4(field);
+
+		/* We check for unsupported types first, so that the
+		subsequent checks are relevant for the supported types. */
+		if (type & ~(DICT_CLUSTERED | DICT_UNIQUE)) {
+
+			fprintf(stderr,
+				"InnoDB: Error: unknown type %lu"
+				" of index %s of table %s\n",
+				(ulong) type, name_buf, table->name);
+
+			error = DB_UNSUPPORTED;
+			goto func_exit;
+		} else if (page_no == FIL_NULL) {
+
+			fprintf(stderr,
+				"InnoDB: Error: trying to load index %s"
+				" for table %s\n"
+				"InnoDB: but the index tree has been freed!\n",
+				name_buf, table->name);
+
+			error = DB_CORRUPTION;
+			goto func_exit;
+		} else if ((type & DICT_CLUSTERED) == 0
+			    && NULL == dict_table_get_first_index(table)) {
+
+			fputs("InnoDB: Error: trying to load index ",
+			      stderr);
+			ut_print_name(stderr, NULL, FALSE, name_buf);
+			fputs(" for table ", stderr);
+			ut_print_name(stderr, NULL, TRUE, table->name);
+			fputs("\nInnoDB: but the first index"
+			      " is not clustered!\n", stderr);
+
+			error = DB_CORRUPTION;
+			goto func_exit;
+		} else if (is_sys_table
+			   && ((type & DICT_CLUSTERED)
+			       || ((table == dict_sys->sys_tables)
+				   && (name_len == (sizeof "ID_IND") - 1)
+				   && (0 == ut_memcmp(name_buf,
+						      "ID_IND", name_len))))) {
+
+			/* The index was created in memory already at booting
+			of the database server */
+		} else {
+			index = dict_mem_index_create(table->name, name_buf,
+						      space, type, n_fields);
+			index->id = id;
+
+			dict_load_fields(index, heap);
+			error = dict_index_add_to_cache(table, index, page_no,
+							FALSE);
+			/* The data dictionary tables should never contain
+			invalid index definitions.  If we ignored this error
+			and simply did not load this index definition, the
+			.frm file would disagree with the index definitions
+			inside InnoDB. */
+			if (UNIV_UNLIKELY(error != DB_SUCCESS)) {
+
+				goto func_exit;
+			}
+		}
+
+next_rec:
+		btr_pcur_move_to_next_user_rec(&pcur, &mtr);
+	}
+
+func_exit:
+	btr_pcur_close(&pcur);
+	mtr_commit(&mtr);
+
+	return(error);
+}
+
+/************************************************************************
+Loads a table definition and also all its index definitions, and also
+the cluster definition if the table is a member in a cluster. Also loads
+all foreign key constraints where the foreign key is in the table or where
+a foreign key references columns in this table. Adds all these to the data
+dictionary cache. */
+UNIV_INTERN
+dict_table_t*
+dict_load_table(
+/*============*/
+				/* out: table, NULL if does not exist;
+				if the table is stored in an .ibd file,
+				but the file does not exist,
+				then we set the ibd_file_missing flag TRUE
+				in the table object we return */
+	const char*	name)	/* in: table name in the
+				databasename/tablename format */
+{
+	ibool		ibd_file_missing	= FALSE;
+	dict_table_t*	table;
+	dict_table_t*	sys_tables;
+	btr_pcur_t	pcur;
+	dict_index_t*	sys_index;
+	dtuple_t*	tuple;
+	mem_heap_t*	heap;
+	dfield_t*	dfield;
+	const rec_t*	rec;
+	const byte*	field;
+	ulint		len;
+	ulint		space;
+	ulint		n_cols;
+	ulint		flags;
+	ulint		err;
+	mtr_t		mtr;
+
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+
+	heap = mem_heap_create(32000);
+
+	mtr_start(&mtr);
+
+	sys_tables = dict_table_get_low("SYS_TABLES");
+	sys_index = UT_LIST_GET_FIRST(sys_tables->indexes);
+	ut_a(!dict_table_is_comp(sys_tables));
+
+	tuple = dtuple_create(heap, 1);
+	dfield = dtuple_get_nth_field(tuple, 0);
+
+	dfield_set_data(dfield, name, ut_strlen(name));
+	dict_index_copy_types(tuple, sys_index, 1);
+
+	btr_pcur_open_on_user_rec(sys_index, tuple, PAGE_CUR_GE,
+				  BTR_SEARCH_LEAF, &pcur, &mtr);
+	rec = btr_pcur_get_rec(&pcur);
+
+	if (!btr_pcur_is_on_user_rec(&pcur)
+	    || rec_get_deleted_flag(rec, 0)) {
+		/* Not found */
+err_exit:
+		btr_pcur_close(&pcur);
+		mtr_commit(&mtr);
+		mem_heap_free(heap);
+
+		return(NULL);
+	}
+
+	field = rec_get_nth_field_old(rec, 0, &len);
+
+	/* Check if the table name in record is the searched one */
+	if (len != ut_strlen(name) || ut_memcmp(name, field, len) != 0) {
+
+		goto err_exit;
+	}
+
+	ut_a(name_of_col_is(sys_tables, sys_index, 9, "SPACE"));
+
+	field = rec_get_nth_field_old(rec, 9, &len);
+	space = mach_read_from_4(field);
+
+	/* Check if the tablespace exists and has the right name */
+	if (space != 0) {
+		flags = dict_sys_tables_get_flags(rec);
+
+		if (UNIV_UNLIKELY(flags == ULINT_UNDEFINED)) {
+			field = rec_get_nth_field_old(rec, 5, &len);
+			flags = mach_read_from_4(field);
+
+			ut_print_timestamp(stderr);
+			fputs("  InnoDB: Error: table ", stderr);
+			ut_print_filename(stderr, name);
+			fprintf(stderr, "\n"
+				"InnoDB: in InnoDB data dictionary"
+				" has unknown type %lx.\n",
+				(ulong) flags);
+			goto err_exit;
+		}
+
+		if (fil_space_for_table_exists_in_mem(space, name, FALSE,
+						      FALSE, FALSE)) {
+			/* Ok; (if we did a crash recovery then the tablespace
+			can already be in the memory cache) */
+		} else {
+			/* In >= 4.1.9, InnoDB scans the data dictionary also
+			at a normal mysqld startup. It is an error if the
+			space object does not exist in memory. */
+
+			ut_print_timestamp(stderr);
+			fprintf(stderr,
+				"  InnoDB: error: space object of table %s,\n"
+				"InnoDB: space id %lu did not exist in memory."
+				" Retrying an open.\n",
+				name, (ulong)space);
+			/* Try to open the tablespace */
+			if (!fil_open_single_table_tablespace(
+				    TRUE, space, flags, name)) {
+				/* We failed to find a sensible tablespace
+				file */
+
+				ibd_file_missing = TRUE;
+			}
+		}
+	} else {
+		flags = 0;
+	}
+
+	ut_a(name_of_col_is(sys_tables, sys_index, 4, "N_COLS"));
+
+	field = rec_get_nth_field_old(rec, 4, &len);
+	n_cols = mach_read_from_4(field);
+
+	/* The high-order bit of N_COLS is the "compact format" flag. */
+	if (n_cols & 0x80000000UL) {
+		flags |= DICT_TF_COMPACT;
+	}
+
+	table = dict_mem_table_create(name, space, n_cols & ~0x80000000UL,
+				      flags);
+
+	table->ibd_file_missing = (unsigned int) ibd_file_missing;
+
+	ut_a(name_of_col_is(sys_tables, sys_index, 3, "ID"));
+
+	field = rec_get_nth_field_old(rec, 3, &len);
+	table->id = mach_read_from_8(field);
+
+	btr_pcur_close(&pcur);
+	mtr_commit(&mtr);
+
+	dict_load_columns(table, heap);
+
+	dict_table_add_to_cache(table, heap);
+
+	mem_heap_empty(heap);
+
+	err = dict_load_indexes(table, heap);
+#ifndef UNIV_HOTBACKUP
+	/* If the force recovery flag is set, we open the table irrespective
+	of the error condition, since the user may want to dump data from the
+	clustered index. However we load the foreign key information only if
+	all indexes were loaded. */
+	if (err == DB_SUCCESS) {
+		err = dict_load_foreigns(table->name, TRUE);
+	} else if (!srv_force_recovery) {
+		dict_table_remove_from_cache(table);
+		table = NULL;
+	}
+# if 0
+	if (err != DB_SUCCESS && table != NULL) {
+
+		mutex_enter(&dict_foreign_err_mutex);
+
+		ut_print_timestamp(stderr);
+
+		fprintf(stderr,
+			"  InnoDB: Error: could not make a foreign key"
+			" definition to match\n"
+			"InnoDB: the foreign key table"
+			" or the referenced table!\n"
+			"InnoDB: The data dictionary of InnoDB is corrupt."
+			" You may need to drop\n"
+			"InnoDB: and recreate the foreign key table"
+			" or the referenced table.\n"
+			"InnoDB: Submit a detailed bug report"
+			" to http://bugs.mysql.com\n"
+			"InnoDB: Latest foreign key error printout:\n%s\n",
+			dict_foreign_err_buf);
+
+		mutex_exit(&dict_foreign_err_mutex);
+	}
+# endif /* 0 */
+#endif /* !UNIV_HOTBACKUP */
+	mem_heap_free(heap);
+
+	return(table);
+}
+
+/***************************************************************************
+Loads a table object based on the table id. */
+UNIV_INTERN
+dict_table_t*
+dict_load_table_on_id(
+/*==================*/
+				/* out: table; NULL if table does not exist */
+	dulint	table_id)	/* in: table id */
+{
+	byte		id_buf[8];
+	btr_pcur_t	pcur;
+	mem_heap_t*	heap;
+	dtuple_t*	tuple;
+	dfield_t*	dfield;
+	dict_index_t*	sys_table_ids;
+	dict_table_t*	sys_tables;
+	const rec_t*	rec;
+	const byte*	field;
+	ulint		len;
+	dict_table_t*	table;
+	mtr_t		mtr;
+
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+
+	/* NOTE that the operation of this function is protected by
+	the dictionary mutex, and therefore no deadlocks can occur
+	with other dictionary operations. */
+
+	mtr_start(&mtr);
+	/*---------------------------------------------------*/
+	/* Get the secondary index based on ID for table SYS_TABLES */
+	sys_tables = dict_sys->sys_tables;
+	sys_table_ids = dict_table_get_next_index(
+		dict_table_get_first_index(sys_tables));
+	ut_a(!dict_table_is_comp(sys_tables));
+	heap = mem_heap_create(256);
+
+	tuple  = dtuple_create(heap, 1);
+	dfield = dtuple_get_nth_field(tuple, 0);
+
+	/* Write the table id in byte format to id_buf */
+	mach_write_to_8(id_buf, table_id);
+
+	dfield_set_data(dfield, id_buf, 8);
+	dict_index_copy_types(tuple, sys_table_ids, 1);
+
+	btr_pcur_open_on_user_rec(sys_table_ids, tuple, PAGE_CUR_GE,
+				  BTR_SEARCH_LEAF, &pcur, &mtr);
+	rec = btr_pcur_get_rec(&pcur);
+
+	if (!btr_pcur_is_on_user_rec(&pcur)
+	    || rec_get_deleted_flag(rec, 0)) {
+		/* Not found */
+
+		btr_pcur_close(&pcur);
+		mtr_commit(&mtr);
+		mem_heap_free(heap);
+
+		return(NULL);
+	}
+
+	/*---------------------------------------------------*/
+	/* Now we have the record in the secondary index containing the
+	table ID and NAME */
+
+	rec = btr_pcur_get_rec(&pcur);
+	field = rec_get_nth_field_old(rec, 0, &len);
+	ut_ad(len == 8);
+
+	/* Check if the table id in record is the one searched for */
+	if (ut_dulint_cmp(table_id, mach_read_from_8(field)) != 0) {
+
+		btr_pcur_close(&pcur);
+		mtr_commit(&mtr);
+		mem_heap_free(heap);
+
+		return(NULL);
+	}
+
+	/* Now we get the table name from the record */
+	field = rec_get_nth_field_old(rec, 1, &len);
+	/* Load the table definition to memory */
+	table = dict_load_table(mem_heap_strdupl(heap, (char*) field, len));
+
+	btr_pcur_close(&pcur);
+	mtr_commit(&mtr);
+	mem_heap_free(heap);
+
+	return(table);
+}
+
+/************************************************************************
+This function is called when the database is booted. Loads system table
+index definitions except for the clustered index which is added to the
+dictionary cache at booting before calling this function. */
+UNIV_INTERN
+void
+dict_load_sys_table(
+/*================*/
+	dict_table_t*	table)	/* in: system table */
+{
+	mem_heap_t*	heap;
+
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+
+	heap = mem_heap_create(1000);
+
+	dict_load_indexes(table, heap);
+
+	mem_heap_free(heap);
+}
+
+#ifndef UNIV_HOTBACKUP
+/************************************************************************
+Loads foreign key constraint col names (also for the referenced table). */
+static
+void
+dict_load_foreign_cols(
+/*===================*/
+	const char*	id,	/* in: foreign constraint id as a
+				null-terminated string */
+	dict_foreign_t*	foreign)/* in: foreign constraint object */
+{
+	dict_table_t*	sys_foreign_cols;
+	dict_index_t*	sys_index;
+	btr_pcur_t	pcur;
+	dtuple_t*	tuple;
+	dfield_t*	dfield;
+	const rec_t*	rec;
+	const byte*	field;
+	ulint		len;
+	ulint		i;
+	mtr_t		mtr;
+
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+
+	foreign->foreign_col_names = mem_heap_alloc(
+		foreign->heap, foreign->n_fields * sizeof(void*));
+
+	foreign->referenced_col_names = mem_heap_alloc(
+		foreign->heap, foreign->n_fields * sizeof(void*));
+	mtr_start(&mtr);
+
+	sys_foreign_cols = dict_table_get_low("SYS_FOREIGN_COLS");
+	sys_index = UT_LIST_GET_FIRST(sys_foreign_cols->indexes);
+	ut_a(!dict_table_is_comp(sys_foreign_cols));
+
+	tuple = dtuple_create(foreign->heap, 1);
+	dfield = dtuple_get_nth_field(tuple, 0);
+
+	dfield_set_data(dfield, id, ut_strlen(id));
+	dict_index_copy_types(tuple, sys_index, 1);
+
+	btr_pcur_open_on_user_rec(sys_index, tuple, PAGE_CUR_GE,
+				  BTR_SEARCH_LEAF, &pcur, &mtr);
+	for (i = 0; i < foreign->n_fields; i++) {
+
+		rec = btr_pcur_get_rec(&pcur);
+
+		ut_a(btr_pcur_is_on_user_rec(&pcur));
+		ut_a(!rec_get_deleted_flag(rec, 0));
+
+		field = rec_get_nth_field_old(rec, 0, &len);
+		ut_a(len == ut_strlen(id));
+		ut_a(ut_memcmp(id, field, len) == 0);
+
+		field = rec_get_nth_field_old(rec, 1, &len);
+		ut_a(len == 4);
+		ut_a(i == mach_read_from_4(field));
+
+		field = rec_get_nth_field_old(rec, 4, &len);
+		foreign->foreign_col_names[i] = mem_heap_strdupl(
+			foreign->heap, (char*) field, len);
+
+		field = rec_get_nth_field_old(rec, 5, &len);
+		foreign->referenced_col_names[i] = mem_heap_strdupl(
+			foreign->heap, (char*) field, len);
+
+		btr_pcur_move_to_next_user_rec(&pcur, &mtr);
+	}
+
+	btr_pcur_close(&pcur);
+	mtr_commit(&mtr);
+}
+
+/***************************************************************************
+Loads a foreign key constraint to the dictionary cache. */
+static
+ulint
+dict_load_foreign(
+/*==============*/
+				/* out: DB_SUCCESS or error code */
+	const char*	id,	/* in: foreign constraint id as a
+				null-terminated string */
+	ibool		check_charsets)
+				/* in: TRUE=check charset compatibility */
+{
+	dict_foreign_t*	foreign;
+	dict_table_t*	sys_foreign;
+	btr_pcur_t	pcur;
+	dict_index_t*	sys_index;
+	dtuple_t*	tuple;
+	mem_heap_t*	heap2;
+	dfield_t*	dfield;
+	const rec_t*	rec;
+	const byte*	field;
+	ulint		len;
+	ulint		n_fields_and_type;
+	mtr_t		mtr;
+
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+
+	heap2 = mem_heap_create(1000);
+
+	mtr_start(&mtr);
+
+	sys_foreign = dict_table_get_low("SYS_FOREIGN");
+	sys_index = UT_LIST_GET_FIRST(sys_foreign->indexes);
+	ut_a(!dict_table_is_comp(sys_foreign));
+
+	tuple = dtuple_create(heap2, 1);
+	dfield = dtuple_get_nth_field(tuple, 0);
+
+	dfield_set_data(dfield, id, ut_strlen(id));
+	dict_index_copy_types(tuple, sys_index, 1);
+
+	btr_pcur_open_on_user_rec(sys_index, tuple, PAGE_CUR_GE,
+				  BTR_SEARCH_LEAF, &pcur, &mtr);
+	rec = btr_pcur_get_rec(&pcur);
+
+	if (!btr_pcur_is_on_user_rec(&pcur)
+	    || rec_get_deleted_flag(rec, 0)) {
+		/* Not found */
+
+		fprintf(stderr,
+			"InnoDB: Error A: cannot load foreign constraint %s\n",
+			id);
+
+		btr_pcur_close(&pcur);
+		mtr_commit(&mtr);
+		mem_heap_free(heap2);
+
+		return(DB_ERROR);
+	}
+
+	field = rec_get_nth_field_old(rec, 0, &len);
+
+	/* Check if the id in record is the searched one */
+	if (len != ut_strlen(id) || ut_memcmp(id, field, len) != 0) {
+
+		fprintf(stderr,
+			"InnoDB: Error B: cannot load foreign constraint %s\n",
+			id);
+
+		btr_pcur_close(&pcur);
+		mtr_commit(&mtr);
+		mem_heap_free(heap2);
+
+		return(DB_ERROR);
+	}
+
+	/* Read the table names and the number of columns associated
+	with the constraint */
+
+	mem_heap_free(heap2);
+
+	foreign = dict_mem_foreign_create();
+
+	n_fields_and_type = mach_read_from_4(
+		rec_get_nth_field_old(rec, 5, &len));
+
+	ut_a(len == 4);
+
+	/* We store the type in the bits 24..29 of n_fields_and_type. */
+
+	foreign->type = (unsigned int) (n_fields_and_type >> 24);
+	foreign->n_fields = (unsigned int) (n_fields_and_type & 0x3FFUL);
+
+	foreign->id = mem_heap_strdup(foreign->heap, id);
+
+	field = rec_get_nth_field_old(rec, 3, &len);
+	foreign->foreign_table_name = mem_heap_strdupl(
+		foreign->heap, (char*) field, len);
+
+	field = rec_get_nth_field_old(rec, 4, &len);
+	foreign->referenced_table_name = mem_heap_strdupl(
+		foreign->heap, (char*) field, len);
+
+	btr_pcur_close(&pcur);
+	mtr_commit(&mtr);
+
+	dict_load_foreign_cols(id, foreign);
+
+	/* If the foreign table is not yet in the dictionary cache, we
+	have to load it so that we are able to make type comparisons
+	in the next function call. */
+
+	dict_table_get_low(foreign->foreign_table_name);
+
+	/* Note that there may already be a foreign constraint object in
+	the dictionary cache for this constraint: then the following
+	call only sets the pointers in it to point to the appropriate table
+	and index objects and frees the newly created object foreign.
+	Adding to the cache should always succeed since we are not creating
+	a new foreign key constraint but loading one from the data
+	dictionary. */
+
+	return(dict_foreign_add_to_cache(foreign, check_charsets));
+}
+
+/***************************************************************************
+Loads foreign key constraints where the table is either the foreign key
+holder or where the table is referenced by a foreign key. Adds these
+constraints to the data dictionary. Note that we know that the dictionary
+cache already contains all constraints where the other relevant table is
+already in the dictionary cache. */
+UNIV_INTERN
+ulint
+dict_load_foreigns(
+/*===============*/
+					/* out: DB_SUCCESS or error code */
+	const char*	table_name,	/* in: table name */
+	ibool		check_charsets)	/* in: TRUE=check charset
+					compatibility */
+{
+	btr_pcur_t	pcur;
+	mem_heap_t*	heap;
+	dtuple_t*	tuple;
+	dfield_t*	dfield;
+	dict_index_t*	sec_index;
+	dict_table_t*	sys_foreign;
+	const rec_t*	rec;
+	const byte*	field;
+	ulint		len;
+	char*		id ;
+	ulint		err;
+	mtr_t		mtr;
+
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+
+	sys_foreign = dict_table_get_low("SYS_FOREIGN");
+
+	if (sys_foreign == NULL) {
+		/* No foreign keys defined yet in this database */
+
+		fprintf(stderr,
+			"InnoDB: Error: no foreign key system tables"
+			" in the database\n");
+
+		return(DB_ERROR);
+	}
+
+	ut_a(!dict_table_is_comp(sys_foreign));
+	mtr_start(&mtr);
+
+	/* Get the secondary index based on FOR_NAME from table
+	SYS_FOREIGN */
+
+	sec_index = dict_table_get_next_index(
+		dict_table_get_first_index(sys_foreign));
+start_load:
+	heap = mem_heap_create(256);
+
+	tuple  = dtuple_create(heap, 1);
+	dfield = dtuple_get_nth_field(tuple, 0);
+
+	dfield_set_data(dfield, table_name, ut_strlen(table_name));
+	dict_index_copy_types(tuple, sec_index, 1);
+
+	btr_pcur_open_on_user_rec(sec_index, tuple, PAGE_CUR_GE,
+				  BTR_SEARCH_LEAF, &pcur, &mtr);
+loop:
+	rec = btr_pcur_get_rec(&pcur);
+
+	if (!btr_pcur_is_on_user_rec(&pcur)) {
+		/* End of index */
+
+		goto load_next_index;
+	}
+
+	/* Now we have the record in the secondary index containing a table
+	name and a foreign constraint ID */
+
+	rec = btr_pcur_get_rec(&pcur);
+	field = rec_get_nth_field_old(rec, 0, &len);
+
+	/* Check if the table name in the record is the one searched for; the
+	following call does the comparison in the latin1_swedish_ci
+	charset-collation, in a case-insensitive way. */
+
+	if (0 != cmp_data_data(dfield_get_type(dfield)->mtype,
+			       dfield_get_type(dfield)->prtype,
+			       dfield_get_data(dfield), dfield_get_len(dfield),
+			       field, len)) {
+
+		goto load_next_index;
+	}
+
+	/* Since table names in SYS_FOREIGN are stored in a case-insensitive
+	order, we have to check that the table name matches also in a binary
+	string comparison. On Unix, MySQL allows table names that only differ
+	in character case. */
+
+	if (0 != ut_memcmp(field, table_name, len)) {
+
+		goto next_rec;
+	}
+
+	if (rec_get_deleted_flag(rec, 0)) {
+
+		goto next_rec;
+	}
+
+	/* Now we get a foreign key constraint id */
+	field = rec_get_nth_field_old(rec, 1, &len);
+	id = mem_heap_strdupl(heap, (char*) field, len);
+
+	btr_pcur_store_position(&pcur, &mtr);
+
+	mtr_commit(&mtr);
+
+	/* Load the foreign constraint definition to the dictionary cache */
+
+	err = dict_load_foreign(id, check_charsets);
+
+	if (err != DB_SUCCESS) {
+		btr_pcur_close(&pcur);
+		mem_heap_free(heap);
+
+		return(err);
+	}
+
+	mtr_start(&mtr);
+
+	btr_pcur_restore_position(BTR_SEARCH_LEAF, &pcur, &mtr);
+next_rec:
+	btr_pcur_move_to_next_user_rec(&pcur, &mtr);
+
+	goto loop;
+
+load_next_index:
+	btr_pcur_close(&pcur);
+	mtr_commit(&mtr);
+	mem_heap_free(heap);
+
+	sec_index = dict_table_get_next_index(sec_index);
+
+	if (sec_index != NULL) {
+
+		mtr_start(&mtr);
+
+		goto start_load;
+	}
+
+	return(DB_SUCCESS);
+}
+#endif /* !UNIV_HOTBACKUP */
diff --git a/storage/xtradb/dict/dict0mem.c b/storage/xtradb/dict/dict0mem.c
new file mode 100644
index 00000000000..31c0c23f450
--- /dev/null
+++ b/storage/xtradb/dict/dict0mem.c
@@ -0,0 +1,308 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/**********************************************************************
+Data dictionary memory object creation
+
+Created 1/8/1996 Heikki Tuuri
+***********************************************************************/
+
+#include "dict0mem.h"
+
+#ifdef UNIV_NONINL
+#include "dict0mem.ic"
+#endif
+
+#include "rem0rec.h"
+#include "data0type.h"
+#include "mach0data.h"
+#include "dict0dict.h"
+#include "lock0lock.h"
+
+#define	DICT_HEAP_SIZE		100	/* initial memory heap size when
+					creating a table or index object */
+
+/**************************************************************************
+Creates a table memory object. */
+UNIV_INTERN
+dict_table_t*
+dict_mem_table_create(
+/*==================*/
+				/* out, own: table object */
+	const char*	name,	/* in: table name */
+	ulint		space,	/* in: space where the clustered index of
+				the table is placed; this parameter is
+				ignored if the table is made a member of
+				a cluster */
+	ulint		n_cols,	/* in: number of columns */
+	ulint		flags)	/* in: table flags */
+{
+	dict_table_t*	table;
+	mem_heap_t*	heap;
+
+	ut_ad(name);
+	ut_a(!(flags & (~0 << DICT_TF_BITS)));
+
+	heap = mem_heap_create(DICT_HEAP_SIZE);
+
+	table = mem_heap_zalloc(heap, sizeof(dict_table_t));
+
+	table->heap = heap;
+
+	table->flags = (unsigned int) flags;
+	table->name = mem_heap_strdup(heap, name);
+	table->space = (unsigned int) space;
+	table->n_cols = (unsigned int) (n_cols + DATA_N_SYS_COLS);
+
+	table->cols = mem_heap_alloc(heap, (n_cols + DATA_N_SYS_COLS)
+				     * sizeof(dict_col_t));
+
+	table->autoinc_lock = mem_heap_alloc(heap, lock_get_size());
+
+	mutex_create(&table->autoinc_mutex, SYNC_DICT_AUTOINC_MUTEX);
+
+	table->autoinc = 0;
+
+	/* The number of transactions that are either waiting on the
+	AUTOINC lock or have been granted the lock. */
+	table->n_waiting_or_granted_auto_inc_locks = 0;
+
+#ifdef UNIV_DEBUG
+	table->magic_n = DICT_TABLE_MAGIC_N;
+#endif /* UNIV_DEBUG */
+	return(table);
+}
+
+/********************************************************************
+Free a table memory object. */
+UNIV_INTERN
+void
+dict_mem_table_free(
+/*================*/
+	dict_table_t*	table)		/* in: table */
+{
+	ut_ad(table);
+	ut_ad(table->magic_n == DICT_TABLE_MAGIC_N);
+	ut_d(table->cached = FALSE);
+
+	mutex_free(&(table->autoinc_mutex));
+	mem_heap_free(table->heap);
+}
+
+/********************************************************************
+Append 'name' to 'col_names' (@see dict_table_t::col_names). */
+static
+const char*
+dict_add_col_name(
+/*==============*/
+					/* out: new column names array */
+	const char*	col_names,	/* in: existing column names, or
+					NULL */
+	ulint		cols,		/* in: number of existing columns */
+	const char*	name,		/* in: new column name */
+	mem_heap_t*	heap)		/* in: heap */
+{
+	ulint	old_len;
+	ulint	new_len;
+	ulint	total_len;
+	char*	res;
+
+	ut_ad(!cols == !col_names);
+
+	/* Find out length of existing array. */
+	if (col_names) {
+		const char*	s = col_names;
+		ulint		i;
+
+		for (i = 0; i < cols; i++) {
+			s += strlen(s) + 1;
+		}
+
+		old_len = s - col_names;
+	} else {
+		old_len = 0;
+	}
+
+	new_len = strlen(name) + 1;
+	total_len = old_len + new_len;
+
+	res = mem_heap_alloc(heap, total_len);
+
+	if (old_len > 0) {
+		memcpy(res, col_names, old_len);
+	}
+
+	memcpy(res + old_len, name, new_len);
+
+	return(res);
+}
+
+/**************************************************************************
+Adds a column definition to a table. */
+UNIV_INTERN
+void
+dict_mem_table_add_col(
+/*===================*/
+	dict_table_t*	table,	/* in: table */
+	mem_heap_t*	heap,	/* in: temporary memory heap, or NULL */
+	const char*	name,	/* in: column name, or NULL */
+	ulint		mtype,	/* in: main datatype */
+	ulint		prtype,	/* in: precise type */
+	ulint		len)	/* in: precision */
+{
+	dict_col_t*	col;
+	ulint		mbminlen;
+	ulint		mbmaxlen;
+	ulint		i;
+
+	ut_ad(table);
+	ut_ad(table->magic_n == DICT_TABLE_MAGIC_N);
+	ut_ad(!heap == !name);
+
+	i = table->n_def++;
+
+	if (name) {
+		if (UNIV_UNLIKELY(table->n_def == table->n_cols)) {
+			heap = table->heap;
+		}
+		if (UNIV_LIKELY(i) && UNIV_UNLIKELY(!table->col_names)) {
+			/* All preceding column names are empty. */
+			char* s = mem_heap_zalloc(heap, table->n_def);
+			table->col_names = s;
+		}
+
+		table->col_names = dict_add_col_name(table->col_names,
+						     i, name, heap);
+	}
+
+	col = dict_table_get_nth_col(table, i);
+
+	col->ind = (unsigned int) i;
+	col->ord_part = 0;
+
+	col->mtype = (unsigned int) mtype;
+	col->prtype = (unsigned int) prtype;
+	col->len = (unsigned int) len;
+
+	dtype_get_mblen(mtype, prtype, &mbminlen, &mbmaxlen);
+
+	col->mbminlen = (unsigned int) mbminlen;
+	col->mbmaxlen = (unsigned int) mbmaxlen;
+}
+
+/**************************************************************************
+Creates an index memory object. */
+UNIV_INTERN
+dict_index_t*
+dict_mem_index_create(
+/*==================*/
+					/* out, own: index object */
+	const char*	table_name,	/* in: table name */
+	const char*	index_name,	/* in: index name */
+	ulint		space,		/* in: space where the index tree is
+					placed, ignored if the index is of
+					the clustered type */
+	ulint		type,		/* in: DICT_UNIQUE,
+					DICT_CLUSTERED, ... ORed */
+	ulint		n_fields)	/* in: number of fields */
+{
+	dict_index_t*	index;
+	mem_heap_t*	heap;
+
+	ut_ad(table_name && index_name);
+
+	heap = mem_heap_create(DICT_HEAP_SIZE);
+	index = mem_heap_zalloc(heap, sizeof(dict_index_t));
+
+	index->heap = heap;
+
+	index->type = type;
+	index->space = (unsigned int) space;
+	index->name = mem_heap_strdup(heap, index_name);
+	index->table_name = table_name;
+	index->n_fields = (unsigned int) n_fields;
+	index->fields = mem_heap_alloc(heap, 1 + n_fields
+				       * sizeof(dict_field_t));
+	/* The '1 +' above prevents allocation
+	of an empty mem block */
+#ifdef UNIV_DEBUG
+	index->magic_n = DICT_INDEX_MAGIC_N;
+#endif /* UNIV_DEBUG */
+	return(index);
+}
+
+/**************************************************************************
+Creates and initializes a foreign constraint memory object. */
+UNIV_INTERN
+dict_foreign_t*
+dict_mem_foreign_create(void)
+/*=========================*/
+				/* out, own: foreign constraint struct */
+{
+	dict_foreign_t*	foreign;
+	mem_heap_t*	heap;
+
+	heap = mem_heap_create(100);
+
+	foreign = mem_heap_zalloc(heap, sizeof(dict_foreign_t));
+
+	foreign->heap = heap;
+
+	return(foreign);
+}
+
+/**************************************************************************
+Adds a field definition to an index. NOTE: does not take a copy
+of the column name if the field is a column. The memory occupied
+by the column name may be released only after publishing the index. */
+UNIV_INTERN
+void
+dict_mem_index_add_field(
+/*=====================*/
+	dict_index_t*	index,		/* in: index */
+	const char*	name,		/* in: column name */
+	ulint		prefix_len)	/* in: 0 or the column prefix length
+					in a MySQL index like
+					INDEX (textcol(25)) */
+{
+	dict_field_t*	field;
+
+	ut_ad(index);
+	ut_ad(index->magic_n == DICT_INDEX_MAGIC_N);
+
+	index->n_def++;
+
+	field = dict_index_get_nth_field(index, index->n_def - 1);
+
+	field->name = name;
+	field->prefix_len = (unsigned int) prefix_len;
+}
+
+/**************************************************************************
+Frees an index memory object. */
+UNIV_INTERN
+void
+dict_mem_index_free(
+/*================*/
+	dict_index_t*	index)	/* in: index */
+{
+	ut_ad(index);
+	ut_ad(index->magic_n == DICT_INDEX_MAGIC_N);
+
+	mem_heap_free(index->heap);
+}
diff --git a/storage/xtradb/dyn/dyn0dyn.c b/storage/xtradb/dyn/dyn0dyn.c
new file mode 100644
index 00000000000..16e82eaed66
--- /dev/null
+++ b/storage/xtradb/dyn/dyn0dyn.c
@@ -0,0 +1,64 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The dynamically allocated array
+
+Created 2/5/1996 Heikki Tuuri
+*******************************************************/
+
+#include "dyn0dyn.h"
+#ifdef UNIV_NONINL
+#include "dyn0dyn.ic"
+#endif
+
+/****************************************************************
+Adds a new block to a dyn array. */
+UNIV_INTERN
+dyn_block_t*
+dyn_array_add_block(
+/*================*/
+				/* out: created block */
+	dyn_array_t*	arr)	/* in: dyn array */
+{
+	mem_heap_t*	heap;
+	dyn_block_t*	block;
+
+	ut_ad(arr);
+	ut_ad(arr->magic_n == DYN_BLOCK_MAGIC_N);
+
+	if (arr->heap == NULL) {
+		UT_LIST_INIT(arr->base);
+		UT_LIST_ADD_FIRST(list, arr->base, arr);
+
+		arr->heap = mem_heap_create(sizeof(dyn_block_t));
+	}
+
+	block = dyn_array_get_last_block(arr);
+	block->used = block->used | DYN_BLOCK_FULL_FLAG;
+
+	heap = arr->heap;
+
+	block = mem_heap_alloc(heap, sizeof(dyn_block_t));
+
+	block->used = 0;
+
+	UT_LIST_ADD_LAST(list, arr->base, block);
+
+	return(block);
+}
diff --git a/storage/xtradb/eval/eval0eval.c b/storage/xtradb/eval/eval0eval.c
new file mode 100644
index 00000000000..a2590c63c38
--- /dev/null
+++ b/storage/xtradb/eval/eval0eval.c
@@ -0,0 +1,851 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+SQL evaluator: evaluates simple data structures, like expressions, in
+a query graph
+
+Created 12/29/1997 Heikki Tuuri
+*******************************************************/
+
+#include "eval0eval.h"
+
+#ifdef UNIV_NONINL
+#include "eval0eval.ic"
+#endif
+
+#include "data0data.h"
+#include "row0sel.h"
+
+/* The RND function seed */
+static ulint	eval_rnd	= 128367121;
+
+/* Dummy adress used when we should allocate a buffer of size 0 in
+the function below */
+
+static byte	eval_dummy;
+
+/*********************************************************************
+Allocate a buffer from global dynamic memory for a value of a que_node.
+NOTE that this memory must be explicitly freed when the query graph is
+freed. If the node already has an allocated buffer, that buffer is freed
+here. NOTE that this is the only function where dynamic memory should be
+allocated for a query node val field. */
+UNIV_INTERN
+byte*
+eval_node_alloc_val_buf(
+/*====================*/
+				/* out: pointer to allocated buffer */
+	que_node_t*	node,	/* in: query graph node; sets the val field
+				data field to point to the new buffer, and
+				len field equal to size */
+	ulint		size)	/* in: buffer size */
+{
+	dfield_t*	dfield;
+	byte*		data;
+
+	ut_ad(que_node_get_type(node) == QUE_NODE_SYMBOL
+	      || que_node_get_type(node) == QUE_NODE_FUNC);
+
+	dfield = que_node_get_val(node);
+
+	data = dfield_get_data(dfield);
+
+	if (data && data != &eval_dummy) {
+		mem_free(data);
+	}
+
+	if (size == 0) {
+		data = &eval_dummy;
+	} else {
+		data = mem_alloc(size);
+	}
+
+	que_node_set_val_buf_size(node, size);
+
+	dfield_set_data(dfield, data, size);
+
+	return(data);
+}
+
+/*********************************************************************
+Free the buffer from global dynamic memory for a value of a que_node,
+if it has been allocated in the above function. The freeing for pushed
+column values is done in sel_col_prefetch_buf_free. */
+UNIV_INTERN
+void
+eval_node_free_val_buf(
+/*===================*/
+	que_node_t*	node)	/* in: query graph node */
+{
+	dfield_t*	dfield;
+	byte*		data;
+
+	ut_ad(que_node_get_type(node) == QUE_NODE_SYMBOL
+	      || que_node_get_type(node) == QUE_NODE_FUNC);
+
+	dfield = que_node_get_val(node);
+
+	data = dfield_get_data(dfield);
+
+	if (que_node_get_val_buf_size(node) > 0) {
+		ut_a(data);
+
+		mem_free(data);
+	}
+}
+
+/*********************************************************************
+Evaluates a comparison node. */
+UNIV_INTERN
+ibool
+eval_cmp(
+/*=====*/
+					/* out: the result of the comparison */
+	func_node_t*	cmp_node)	/* in: comparison node */
+{
+	que_node_t*	arg1;
+	que_node_t*	arg2;
+	int		res;
+	ibool		val;
+	int		func;
+
+	ut_ad(que_node_get_type(cmp_node) == QUE_NODE_FUNC);
+
+	arg1 = cmp_node->args;
+	arg2 = que_node_get_next(arg1);
+
+	res = cmp_dfield_dfield(que_node_get_val(arg1),
+				que_node_get_val(arg2));
+	val = TRUE;
+
+	func = cmp_node->func;
+
+	if (func == '=') {
+		if (res != 0) {
+			val = FALSE;
+		}
+	} else if (func == '<') {
+		if (res != -1) {
+			val = FALSE;
+		}
+	} else if (func == PARS_LE_TOKEN) {
+		if (res == 1) {
+			val = FALSE;
+		}
+	} else if (func == PARS_NE_TOKEN) {
+		if (res == 0) {
+			val = FALSE;
+		}
+	} else if (func == PARS_GE_TOKEN) {
+		if (res == -1) {
+			val = FALSE;
+		}
+	} else {
+		ut_ad(func == '>');
+
+		if (res != 1) {
+			val = FALSE;
+		}
+	}
+
+	eval_node_set_ibool_val(cmp_node, val);
+
+	return(val);
+}
+
+/*********************************************************************
+Evaluates a logical operation node. */
+UNIV_INLINE
+void
+eval_logical(
+/*=========*/
+	func_node_t*	logical_node)	/* in: logical operation node */
+{
+	que_node_t*	arg1;
+	que_node_t*	arg2;
+	ibool		val1;
+	ibool		val2 = 0; /* remove warning */
+	ibool		val = 0;  /* remove warning */
+	int		func;
+
+	ut_ad(que_node_get_type(logical_node) == QUE_NODE_FUNC);
+
+	arg1 = logical_node->args;
+	arg2 = que_node_get_next(arg1); /* arg2 is NULL if func is 'NOT' */
+
+	val1 = eval_node_get_ibool_val(arg1);
+
+	if (arg2) {
+		val2 = eval_node_get_ibool_val(arg2);
+	}
+
+	func = logical_node->func;
+
+	if (func == PARS_AND_TOKEN) {
+		val = val1 & val2;
+	} else if (func == PARS_OR_TOKEN) {
+		val = val1 | val2;
+	} else if (func == PARS_NOT_TOKEN) {
+		val = TRUE - val1;
+	} else {
+		ut_error;
+	}
+
+	eval_node_set_ibool_val(logical_node, val);
+}
+
+/*********************************************************************
+Evaluates an arithmetic operation node. */
+UNIV_INLINE
+void
+eval_arith(
+/*=======*/
+	func_node_t*	arith_node)	/* in: arithmetic operation node */
+{
+	que_node_t*	arg1;
+	que_node_t*	arg2;
+	lint		val1;
+	lint		val2 = 0; /* remove warning */
+	lint		val;
+	int		func;
+
+	ut_ad(que_node_get_type(arith_node) == QUE_NODE_FUNC);
+
+	arg1 = arith_node->args;
+	arg2 = que_node_get_next(arg1); /* arg2 is NULL if func is unary '-' */
+
+	val1 = eval_node_get_int_val(arg1);
+
+	if (arg2) {
+		val2 = eval_node_get_int_val(arg2);
+	}
+
+	func = arith_node->func;
+
+	if (func == '+') {
+		val = val1 + val2;
+	} else if ((func == '-') && arg2) {
+		val = val1 - val2;
+	} else if (func == '-') {
+		val = -val1;
+	} else if (func == '*') {
+		val = val1 * val2;
+	} else {
+		ut_ad(func == '/');
+		val = val1 / val2;
+	}
+
+	eval_node_set_int_val(arith_node, val);
+}
+
+/*********************************************************************
+Evaluates an aggregate operation node. */
+UNIV_INLINE
+void
+eval_aggregate(
+/*===========*/
+	func_node_t*	node)	/* in: aggregate operation node */
+{
+	que_node_t*	arg;
+	lint		val;
+	lint		arg_val;
+	int		func;
+
+	ut_ad(que_node_get_type(node) == QUE_NODE_FUNC);
+
+	val = eval_node_get_int_val(node);
+
+	func = node->func;
+
+	if (func == PARS_COUNT_TOKEN) {
+
+		val = val + 1;
+	} else {
+		ut_ad(func == PARS_SUM_TOKEN);
+
+		arg = node->args;
+		arg_val = eval_node_get_int_val(arg);
+
+		val = val + arg_val;
+	}
+
+	eval_node_set_int_val(node, val);
+}
+
+/*********************************************************************
+Evaluates a predefined function node where the function is not relevant
+in benchmarks. */
+static
+void
+eval_predefined_2(
+/*==============*/
+	func_node_t*	func_node)	/* in: predefined function node */
+{
+	que_node_t*	arg;
+	que_node_t*	arg1;
+	que_node_t*	arg2 = 0; /* remove warning (??? bug ???) */
+	lint		int_val;
+	byte*		data;
+	ulint		len1;
+	ulint		len2;
+	int		func;
+	ulint		i;
+
+	ut_ad(que_node_get_type(func_node) == QUE_NODE_FUNC);
+
+	arg1 = func_node->args;
+
+	if (arg1) {
+		arg2 = que_node_get_next(arg1);
+	}
+
+	func = func_node->func;
+
+	if (func == PARS_PRINTF_TOKEN) {
+
+		arg = arg1;
+
+		while (arg) {
+			dfield_print(que_node_get_val(arg));
+
+			arg = que_node_get_next(arg);
+		}
+
+		putc('\n', stderr);
+
+	} else if (func == PARS_ASSERT_TOKEN) {
+
+		if (!eval_node_get_ibool_val(arg1)) {
+			fputs("SQL assertion fails in a stored procedure!\n",
+			      stderr);
+		}
+
+		ut_a(eval_node_get_ibool_val(arg1));
+
+		/* This function, or more precisely, a debug procedure,
+		returns no value */
+
+	} else if (func == PARS_RND_TOKEN) {
+
+		len1 = (ulint)eval_node_get_int_val(arg1);
+		len2 = (ulint)eval_node_get_int_val(arg2);
+
+		ut_ad(len2 >= len1);
+
+		if (len2 > len1) {
+			int_val = (lint) (len1
+					  + (eval_rnd % (len2 - len1 + 1)));
+		} else {
+			int_val = (lint) len1;
+		}
+
+		eval_rnd = ut_rnd_gen_next_ulint(eval_rnd);
+
+		eval_node_set_int_val(func_node, int_val);
+
+	} else if (func == PARS_RND_STR_TOKEN) {
+
+		len1 = (ulint)eval_node_get_int_val(arg1);
+
+		data = eval_node_ensure_val_buf(func_node, len1);
+
+		for (i = 0; i < len1; i++) {
+			data[i] = (byte)(97 + (eval_rnd % 3));
+
+			eval_rnd = ut_rnd_gen_next_ulint(eval_rnd);
+		}
+	} else {
+		ut_error;
+	}
+}
+
+/*********************************************************************
+Evaluates a notfound-function node. */
+UNIV_INLINE
+void
+eval_notfound(
+/*==========*/
+	func_node_t*	func_node)	/* in: function node */
+{
+	que_node_t*	arg1;
+	que_node_t*	arg2;
+	sym_node_t*	cursor;
+	sel_node_t*	sel_node;
+	ibool		ibool_val;
+
+	arg1 = func_node->args;
+	arg2 = que_node_get_next(arg1);
+
+	ut_ad(func_node->func == PARS_NOTFOUND_TOKEN);
+
+	cursor = arg1;
+
+	ut_ad(que_node_get_type(cursor) == QUE_NODE_SYMBOL);
+
+	if (cursor->token_type == SYM_LIT) {
+
+		ut_ad(ut_memcmp(dfield_get_data(que_node_get_val(cursor)),
+				"SQL", 3) == 0);
+
+		sel_node = cursor->sym_table->query_graph->last_sel_node;
+	} else {
+		sel_node = cursor->alias->cursor_def;
+	}
+
+	if (sel_node->state == SEL_NODE_NO_MORE_ROWS) {
+		ibool_val = TRUE;
+	} else {
+		ibool_val = FALSE;
+	}
+
+	eval_node_set_ibool_val(func_node, ibool_val);
+}
+
+/*********************************************************************
+Evaluates a substr-function node. */
+UNIV_INLINE
+void
+eval_substr(
+/*========*/
+	func_node_t*	func_node)	/* in: function node */
+{
+	que_node_t*	arg1;
+	que_node_t*	arg2;
+	que_node_t*	arg3;
+	dfield_t*	dfield;
+	byte*		str1;
+	ulint		len1;
+	ulint		len2;
+
+	arg1 = func_node->args;
+	arg2 = que_node_get_next(arg1);
+
+	ut_ad(func_node->func == PARS_SUBSTR_TOKEN);
+
+	arg3 = que_node_get_next(arg2);
+
+	str1 = dfield_get_data(que_node_get_val(arg1));
+
+	len1 = (ulint)eval_node_get_int_val(arg2);
+	len2 = (ulint)eval_node_get_int_val(arg3);
+
+	dfield = que_node_get_val(func_node);
+
+	dfield_set_data(dfield, str1 + len1, len2);
+}
+
+/*********************************************************************
+Evaluates a replstr-procedure node. */
+static
+void
+eval_replstr(
+/*=========*/
+	func_node_t*	func_node)	/* in: function node */
+{
+	que_node_t*	arg1;
+	que_node_t*	arg2;
+	que_node_t*	arg3;
+	que_node_t*	arg4;
+	byte*		str1;
+	byte*		str2;
+	ulint		len1;
+	ulint		len2;
+
+	arg1 = func_node->args;
+	arg2 = que_node_get_next(arg1);
+
+	ut_ad(que_node_get_type(arg1) == QUE_NODE_SYMBOL);
+
+	arg3 = que_node_get_next(arg2);
+	arg4 = que_node_get_next(arg3);
+
+	str1 = dfield_get_data(que_node_get_val(arg1));
+	str2 = dfield_get_data(que_node_get_val(arg2));
+
+	len1 = (ulint)eval_node_get_int_val(arg3);
+	len2 = (ulint)eval_node_get_int_val(arg4);
+
+	if ((dfield_get_len(que_node_get_val(arg1)) < len1 + len2)
+	    || (dfield_get_len(que_node_get_val(arg2)) < len2)) {
+
+		ut_error;
+	}
+
+	ut_memcpy(str1 + len1, str2, len2);
+}
+
+/*********************************************************************
+Evaluates an instr-function node. */
+static
+void
+eval_instr(
+/*=======*/
+	func_node_t*	func_node)	/* in: function node */
+{
+	que_node_t*	arg1;
+	que_node_t*	arg2;
+	dfield_t*	dfield1;
+	dfield_t*	dfield2;
+	lint		int_val;
+	byte*		str1;
+	byte*		str2;
+	byte		match_char;
+	ulint		len1;
+	ulint		len2;
+	ulint		i;
+	ulint		j;
+
+	arg1 = func_node->args;
+	arg2 = que_node_get_next(arg1);
+
+	dfield1 = que_node_get_val(arg1);
+	dfield2 = que_node_get_val(arg2);
+
+	str1 = dfield_get_data(dfield1);
+	str2 = dfield_get_data(dfield2);
+
+	len1 = dfield_get_len(dfield1);
+	len2 = dfield_get_len(dfield2);
+
+	if (len2 == 0) {
+		ut_error;
+	}
+
+	match_char = str2[0];
+
+	for (i = 0; i < len1; i++) {
+		/* In this outer loop, the number of matched characters is 0 */
+
+		if (str1[i] == match_char) {
+
+			if (i + len2 > len1) {
+
+				break;
+			}
+
+			for (j = 1;; j++) {
+				/* We have already matched j characters */
+
+				if (j == len2) {
+					int_val = i + 1;
+
+					goto match_found;
+				}
+
+				if (str1[i + j] != str2[j]) {
+
+					break;
+				}
+			}
+		}
+	}
+
+	int_val = 0;
+
+match_found:
+	eval_node_set_int_val(func_node, int_val);
+}
+
+/*********************************************************************
+Evaluates a predefined function node. */
+UNIV_INLINE
+void
+eval_binary_to_number(
+/*==================*/
+	func_node_t*	func_node)	/* in: function node */
+{
+	que_node_t*	arg1;
+	dfield_t*	dfield;
+	byte*		str1;
+	byte*		str2;
+	ulint		len1;
+	ulint		int_val;
+
+	arg1 = func_node->args;
+
+	dfield = que_node_get_val(arg1);
+
+	str1 = dfield_get_data(dfield);
+	len1 = dfield_get_len(dfield);
+
+	if (len1 > 4) {
+		ut_error;
+	}
+
+	if (len1 == 4) {
+		str2 = str1;
+	} else {
+		int_val = 0;
+		str2 = (byte*)&int_val;
+
+		ut_memcpy(str2 + (4 - len1), str1, len1);
+	}
+
+	eval_node_copy_and_alloc_val(func_node, str2, 4);
+}
+
+/*********************************************************************
+Evaluates a predefined function node. */
+static
+void
+eval_concat(
+/*========*/
+	func_node_t*	func_node)	/* in: function node */
+{
+	que_node_t*	arg;
+	dfield_t*	dfield;
+	byte*		data;
+	ulint		len;
+	ulint		len1;
+
+	arg = func_node->args;
+	len = 0;
+
+	while (arg) {
+		len1 = dfield_get_len(que_node_get_val(arg));
+
+		len += len1;
+
+		arg = que_node_get_next(arg);
+	}
+
+	data = eval_node_ensure_val_buf(func_node, len);
+
+	arg = func_node->args;
+	len = 0;
+
+	while (arg) {
+		dfield = que_node_get_val(arg);
+		len1 = dfield_get_len(dfield);
+
+		ut_memcpy(data + len, dfield_get_data(dfield), len1);
+
+		len += len1;
+
+		arg = que_node_get_next(arg);
+	}
+}
+
+/*********************************************************************
+Evaluates a predefined function node. If the first argument is an integer,
+this function looks at the second argument which is the integer length in
+bytes, and converts the integer to a VARCHAR.
+If the first argument is of some other type, this function converts it to
+BINARY. */
+UNIV_INLINE
+void
+eval_to_binary(
+/*===========*/
+	func_node_t*	func_node)	/* in: function node */
+{
+	que_node_t*	arg1;
+	que_node_t*	arg2;
+	dfield_t*	dfield;
+	byte*		str1;
+	ulint		len;
+	ulint		len1;
+
+	arg1 = func_node->args;
+
+	str1 = dfield_get_data(que_node_get_val(arg1));
+
+	if (dtype_get_mtype(que_node_get_data_type(arg1)) != DATA_INT) {
+
+		len = dfield_get_len(que_node_get_val(arg1));
+
+		dfield = que_node_get_val(func_node);
+
+		dfield_set_data(dfield, str1, len);
+
+		return;
+	}
+
+	arg2 = que_node_get_next(arg1);
+
+	len1 = (ulint)eval_node_get_int_val(arg2);
+
+	if (len1 > 4) {
+
+		ut_error;
+	}
+
+	dfield = que_node_get_val(func_node);
+
+	dfield_set_data(dfield, str1 + (4 - len1), len1);
+}
+
+/*********************************************************************
+Evaluates a predefined function node. */
+UNIV_INLINE
+void
+eval_predefined(
+/*============*/
+	func_node_t*	func_node)	/* in: function node */
+{
+	que_node_t*	arg1;
+	lint		int_val;
+	byte*		data;
+	int		func;
+
+	func = func_node->func;
+
+	arg1 = func_node->args;
+
+	if (func == PARS_LENGTH_TOKEN) {
+
+		int_val = (lint)dfield_get_len(que_node_get_val(arg1));
+
+	} else if (func == PARS_TO_CHAR_TOKEN) {
+
+		/* Convert number to character string as a
+		signed decimal integer. */
+
+		ulint	uint_val;
+		int	int_len;
+
+		int_val = eval_node_get_int_val(arg1);
+
+		/* Determine the length of the string. */
+
+		if (int_val == 0) {
+			int_len = 1; /* the number 0 occupies 1 byte */
+		} else {
+			int_len = 0;
+			if (int_val < 0) {
+				uint_val = ((ulint) -int_val - 1) + 1;
+				int_len++; /* reserve space for minus sign */
+			} else {
+				uint_val = (ulint) int_val;
+			}
+			for (; uint_val > 0; int_len++) {
+				uint_val /= 10;
+			}
+		}
+
+		/* allocate the string */
+		data = eval_node_ensure_val_buf(func_node, int_len + 1);
+
+		/* add terminating NUL character */
+		data[int_len] = 0;
+
+		/* convert the number */
+
+		if (int_val == 0) {
+			data[0] = '0';
+		} else {
+			int tmp;
+			if (int_val < 0) {
+				data[0] = '-'; /* preceding minus sign */
+				uint_val = ((ulint) -int_val - 1) + 1;
+			} else {
+				uint_val = (ulint) int_val;
+			}
+			for (tmp = int_len; uint_val > 0; uint_val /= 10) {
+				data[--tmp] = (byte)
+					('0' + (byte)(uint_val % 10));
+			}
+		}
+
+		dfield_set_len(que_node_get_val(func_node), int_len);
+
+		return;
+
+	} else if (func == PARS_TO_NUMBER_TOKEN) {
+
+		int_val = atoi((char*)
+			       dfield_get_data(que_node_get_val(arg1)));
+
+	} else if (func == PARS_SYSDATE_TOKEN) {
+		int_val = (lint)ut_time();
+	} else {
+		eval_predefined_2(func_node);
+
+		return;
+	}
+
+	eval_node_set_int_val(func_node, int_val);
+}
+
+/*********************************************************************
+Evaluates a function node. */
+UNIV_INTERN
+void
+eval_func(
+/*======*/
+	func_node_t*	func_node)	/* in: function node */
+{
+	que_node_t*	arg;
+	ulint		class;
+	ulint		func;
+
+	ut_ad(que_node_get_type(func_node) == QUE_NODE_FUNC);
+
+	class = func_node->class;
+	func = func_node->func;
+
+	arg = func_node->args;
+
+	/* Evaluate first the argument list */
+	while (arg) {
+		eval_exp(arg);
+
+		/* The functions are not defined for SQL null argument
+		values, except for eval_cmp and notfound */
+
+		if (dfield_is_null(que_node_get_val(arg))
+		    && (class != PARS_FUNC_CMP)
+		    && (func != PARS_NOTFOUND_TOKEN)
+		    && (func != PARS_PRINTF_TOKEN)) {
+			ut_error;
+		}
+
+		arg = que_node_get_next(arg);
+	}
+
+	if (class == PARS_FUNC_CMP) {
+		eval_cmp(func_node);
+	} else if (class == PARS_FUNC_ARITH) {
+		eval_arith(func_node);
+	} else if (class == PARS_FUNC_AGGREGATE) {
+		eval_aggregate(func_node);
+	} else if (class == PARS_FUNC_PREDEFINED) {
+
+		if (func == PARS_NOTFOUND_TOKEN) {
+			eval_notfound(func_node);
+		} else if (func == PARS_SUBSTR_TOKEN) {
+			eval_substr(func_node);
+		} else if (func == PARS_REPLSTR_TOKEN) {
+			eval_replstr(func_node);
+		} else if (func == PARS_INSTR_TOKEN) {
+			eval_instr(func_node);
+		} else if (func == PARS_BINARY_TO_NUMBER_TOKEN) {
+			eval_binary_to_number(func_node);
+		} else if (func == PARS_CONCAT_TOKEN) {
+			eval_concat(func_node);
+		} else if (func == PARS_TO_BINARY_TOKEN) {
+			eval_to_binary(func_node);
+		} else {
+			eval_predefined(func_node);
+		}
+	} else {
+		ut_ad(class == PARS_FUNC_LOGICAL);
+
+		eval_logical(func_node);
+	}
+}
diff --git a/storage/xtradb/eval/eval0proc.c b/storage/xtradb/eval/eval0proc.c
new file mode 100644
index 00000000000..9c7563e8c7d
--- /dev/null
+++ b/storage/xtradb/eval/eval0proc.c
@@ -0,0 +1,294 @@
+/*****************************************************************************
+
+Copyright (c) 1998, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Executes SQL stored procedures and their control structures
+
+Created 1/20/1998 Heikki Tuuri
+*******************************************************/
+
+#include "eval0proc.h"
+
+#ifdef UNIV_NONINL
+#include "eval0proc.ic"
+#endif
+
+/**************************************************************************
+Performs an execution step of an if-statement node. */
+UNIV_INTERN
+que_thr_t*
+if_step(
+/*====*/
+				/* out: query thread to run next or NULL */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	if_node_t*	node;
+	elsif_node_t*	elsif_node;
+
+	ut_ad(thr);
+
+	node = thr->run_node;
+	ut_ad(que_node_get_type(node) == QUE_NODE_IF);
+
+	if (thr->prev_node == que_node_get_parent(node)) {
+
+		/* Evaluate the condition */
+
+		eval_exp(node->cond);
+
+		if (eval_node_get_ibool_val(node->cond)) {
+
+			/* The condition evaluated to TRUE: start execution
+			from the first statement in the statement list */
+
+			thr->run_node = node->stat_list;
+
+		} else if (node->else_part) {
+			thr->run_node = node->else_part;
+
+		} else if (node->elsif_list) {
+			elsif_node = node->elsif_list;
+
+			for (;;) {
+				eval_exp(elsif_node->cond);
+
+				if (eval_node_get_ibool_val(
+					    elsif_node->cond)) {
+
+					/* The condition evaluated to TRUE:
+					start execution from the first
+					statement in the statement list */
+
+					thr->run_node = elsif_node->stat_list;
+
+					break;
+				}
+
+				elsif_node = que_node_get_next(elsif_node);
+
+				if (elsif_node == NULL) {
+					thr->run_node = NULL;
+
+					break;
+				}
+			}
+		} else {
+			thr->run_node = NULL;
+		}
+	} else {
+		/* Move to the next statement */
+		ut_ad(que_node_get_next(thr->prev_node) == NULL);
+
+		thr->run_node = NULL;
+	}
+
+	if (thr->run_node == NULL) {
+		thr->run_node = que_node_get_parent(node);
+	}
+
+	return(thr);
+}
+
+/**************************************************************************
+Performs an execution step of a while-statement node. */
+UNIV_INTERN
+que_thr_t*
+while_step(
+/*=======*/
+				/* out: query thread to run next or NULL */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	while_node_t*	node;
+
+	ut_ad(thr);
+
+	node = thr->run_node;
+	ut_ad(que_node_get_type(node) == QUE_NODE_WHILE);
+
+	ut_ad((thr->prev_node == que_node_get_parent(node))
+	      || (que_node_get_next(thr->prev_node) == NULL));
+
+	/* Evaluate the condition */
+
+	eval_exp(node->cond);
+
+	if (eval_node_get_ibool_val(node->cond)) {
+
+		/* The condition evaluated to TRUE: start execution
+		from the first statement in the statement list */
+
+		thr->run_node = node->stat_list;
+	} else {
+		thr->run_node = que_node_get_parent(node);
+	}
+
+	return(thr);
+}
+
+/**************************************************************************
+Performs an execution step of an assignment statement node. */
+UNIV_INTERN
+que_thr_t*
+assign_step(
+/*========*/
+				/* out: query thread to run next or NULL */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	assign_node_t*	node;
+
+	ut_ad(thr);
+
+	node = thr->run_node;
+	ut_ad(que_node_get_type(node) == QUE_NODE_ASSIGNMENT);
+
+	/* Evaluate the value to assign */
+
+	eval_exp(node->val);
+
+	eval_node_copy_val(node->var->alias, node->val);
+
+	thr->run_node = que_node_get_parent(node);
+
+	return(thr);
+}
+
+/**************************************************************************
+Performs an execution step of a for-loop node. */
+UNIV_INTERN
+que_thr_t*
+for_step(
+/*=====*/
+				/* out: query thread to run next or NULL */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	for_node_t*	node;
+	que_node_t*	parent;
+	lint		loop_var_value;
+
+	ut_ad(thr);
+
+	node = thr->run_node;
+
+	ut_ad(que_node_get_type(node) == QUE_NODE_FOR);
+
+	parent = que_node_get_parent(node);
+
+	if (thr->prev_node != parent) {
+
+		/* Move to the next statement */
+		thr->run_node = que_node_get_next(thr->prev_node);
+
+		if (thr->run_node != NULL) {
+
+			return(thr);
+		}
+
+		/* Increment the value of loop_var */
+
+		loop_var_value = 1 + eval_node_get_int_val(node->loop_var);
+	} else {
+		/* Initialize the loop */
+
+		eval_exp(node->loop_start_limit);
+		eval_exp(node->loop_end_limit);
+
+		loop_var_value = eval_node_get_int_val(node->loop_start_limit);
+
+		node->loop_end_value
+                  = (int) eval_node_get_int_val(node->loop_end_limit);
+	}
+
+	/* Check if we should do another loop */
+
+	if (loop_var_value > node->loop_end_value) {
+
+		/* Enough loops done */
+
+		thr->run_node = parent;
+	} else {
+		eval_node_set_int_val(node->loop_var, loop_var_value);
+
+		thr->run_node = node->stat_list;
+	}
+
+	return(thr);
+}
+
+/**************************************************************************
+Performs an execution step of an exit statement node. */
+UNIV_INTERN
+que_thr_t*
+exit_step(
+/*======*/
+				/* out: query thread to run next or NULL */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	exit_node_t*	node;
+	que_node_t*	loop_node;
+
+	ut_ad(thr);
+
+	node = thr->run_node;
+
+	ut_ad(que_node_get_type(node) == QUE_NODE_EXIT);
+
+	/* Loops exit by setting thr->run_node as the loop node's parent, so
+	find our containing loop node and get its parent. */
+
+	loop_node = que_node_get_containing_loop_node(node);
+
+	/* If someone uses an EXIT statement outside of a loop, this will
+	trigger. */
+	ut_a(loop_node);
+
+	thr->run_node = que_node_get_parent(loop_node);
+
+	return(thr);
+}
+
+/**************************************************************************
+Performs an execution step of a return-statement node. */
+UNIV_INTERN
+que_thr_t*
+return_step(
+/*========*/
+				/* out: query thread to run next or NULL */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	return_node_t*	node;
+	que_node_t*	parent;
+
+	ut_ad(thr);
+
+	node = thr->run_node;
+
+	ut_ad(que_node_get_type(node) == QUE_NODE_RETURN);
+
+	parent = node;
+
+	while (que_node_get_type(parent) != QUE_NODE_PROC) {
+
+		parent = que_node_get_parent(parent);
+	}
+
+	ut_a(parent);
+
+	thr->run_node = que_node_get_parent(parent);
+
+	return(thr);
+}
diff --git a/storage/xtradb/fil/fil0fil.c b/storage/xtradb/fil/fil0fil.c
new file mode 100644
index 00000000000..dad045fe067
--- /dev/null
+++ b/storage/xtradb/fil/fil0fil.c
@@ -0,0 +1,4817 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The tablespace memory cache
+
+Created 10/25/1995 Heikki Tuuri
+*******************************************************/
+
+#include "fil0fil.h"
+
+#include "mem0mem.h"
+#include "sync0sync.h"
+#include "hash0hash.h"
+#include "os0file.h"
+#include "os0sync.h"
+#include "mach0data.h"
+#include "ibuf0ibuf.h"
+#include "buf0buf.h"
+#include "buf0flu.h"
+#include "buf0lru.h"
+#include "log0recv.h"
+#include "fsp0fsp.h"
+#include "srv0srv.h"
+#include "srv0start.h"
+#include "mtr0mtr.h"
+#include "mtr0log.h"
+#include "dict0dict.h"
+#include "page0zip.h"
+
+
+/*
+		IMPLEMENTATION OF THE TABLESPACE MEMORY CACHE
+		=============================================
+
+The tablespace cache is responsible for providing fast read/write access to
+tablespaces and logs of the database. File creation and deletion is done
+in other modules which know more of the logic of the operation, however.
+
+A tablespace consists of a chain of files. The size of the files does not
+have to be divisible by the database block size, because we may just leave
+the last incomplete block unused. When a new file is appended to the
+tablespace, the maximum size of the file is also specified. At the moment,
+we think that it is best to extend the file to its maximum size already at
+the creation of the file, because then we can avoid dynamically extending
+the file when more space is needed for the tablespace.
+
+A block's position in the tablespace is specified with a 32-bit unsigned
+integer. The files in the chain are thought to be catenated, and the block
+corresponding to an address n is the nth block in the catenated file (where
+the first block is named the 0th block, and the incomplete block fragments
+at the end of files are not taken into account). A tablespace can be extended
+by appending a new file at the end of the chain.
+
+Our tablespace concept is similar to the one of Oracle.
+
+To acquire more speed in disk transfers, a technique called disk striping is
+sometimes used. This means that logical block addresses are divided in a
+round-robin fashion across several disks. Windows NT supports disk striping,
+so there we do not need to support it in the database. Disk striping is
+implemented in hardware in RAID disks. We conclude that it is not necessary
+to implement it in the database. Oracle 7 does not support disk striping,
+either.
+
+Another trick used at some database sites is replacing tablespace files by
+raw disks, that is, the whole physical disk drive, or a partition of it, is
+opened as a single file, and it is accessed through byte offsets calculated
+from the start of the disk or the partition. This is recommended in some
+books on database tuning to achieve more speed in i/o. Using raw disk
+certainly prevents the OS from fragmenting disk space, but it is not clear
+if it really adds speed. We measured on the Pentium 100 MHz + NT + NTFS file
+system + EIDE Conner disk only a negligible difference in speed when reading
+from a file, versus reading from a raw disk.
+
+To have fast access to a tablespace or a log file, we put the data structures
+to a hash table. Each tablespace and log file is given an unique 32-bit
+identifier.
+
+Some operating systems do not support many open files at the same time,
+though NT seems to tolerate at least 900 open files. Therefore, we put the
+open files in an LRU-list. If we need to open another file, we may close the
+file at the end of the LRU-list. When an i/o-operation is pending on a file,
+the file cannot be closed. We take the file nodes with pending i/o-operations
+out of the LRU-list and keep a count of pending operations. When an operation
+completes, we decrement the count and return the file node to the LRU-list if
+the count drops to zero. */
+
+/* When mysqld is run, the default directory "." is the mysqld datadir,
+but in the MySQL Embedded Server Library and ibbackup it is not the default
+directory, and we must set the base file path explicitly */
+UNIV_INTERN const char*	fil_path_to_mysql_datadir	= ".";
+
+/* The number of fsyncs done to the log */
+UNIV_INTERN ulint	fil_n_log_flushes			= 0;
+
+UNIV_INTERN ulint	fil_n_pending_log_flushes		= 0;
+UNIV_INTERN ulint	fil_n_pending_tablespace_flushes	= 0;
+
+/* Null file address */
+UNIV_INTERN fil_addr_t	fil_addr_null = {FIL_NULL, 0};
+
+/* File node of a tablespace or the log data space */
+struct fil_node_struct {
+	fil_space_t*	space;	/* backpointer to the space where this node
+				belongs */
+	char*		name;	/* path to the file */
+	ibool		open;	/* TRUE if file open */
+	os_file_t	handle;	/* OS handle to the file, if file open */
+	ibool		is_raw_disk;/* TRUE if the 'file' is actually a raw
+				device or a raw disk partition */
+	ulint		size;	/* size of the file in database pages, 0 if
+				not known yet; the possible last incomplete
+				megabyte may be ignored if space == 0 */
+	ulint		n_pending;
+				/* count of pending i/o's on this file;
+				closing of the file is not allowed if
+				this is > 0 */
+	ulint		n_pending_flushes;
+				/* count of pending flushes on this file;
+				closing of the file is not allowed if
+				this is > 0 */
+	ib_int64_t	modification_counter;/* when we write to the file we
+				increment this by one */
+	ib_int64_t	flush_counter;/* up to what modification_counter value
+				we have flushed the modifications to disk */
+	UT_LIST_NODE_T(fil_node_t) chain;
+				/* link field for the file chain */
+	UT_LIST_NODE_T(fil_node_t) LRU;
+				/* link field for the LRU list */
+	ulint		magic_n;
+};
+
+#define	FIL_NODE_MAGIC_N	89389
+
+/* Tablespace or log data space: let us call them by a common name space */
+struct fil_space_struct {
+	char*		name;	/* space name = the path to the first file in
+				it */
+	ulint		id;	/* space id */
+	ib_int64_t	tablespace_version;
+				/* in DISCARD/IMPORT this timestamp is used to
+				check if we should ignore an insert buffer
+				merge request for a page because it actually
+				was for the previous incarnation of the
+				space */
+	ibool		mark;	/* this is set to TRUE at database startup if
+				the space corresponds to a table in the InnoDB
+				data dictionary; so we can print a warning of
+				orphaned tablespaces */
+	ibool		stop_ios;/* TRUE if we want to rename the .ibd file of
+				tablespace and want to stop temporarily
+				posting of new i/o requests on the file */
+	ibool		stop_ibuf_merges;
+				/* we set this TRUE when we start deleting a
+				single-table tablespace */
+	ibool		is_being_deleted;
+				/* this is set to TRUE when we start
+				deleting a single-table tablespace and its
+				file; when this flag is set no further i/o
+				or flush requests can be placed on this space,
+				though there may be such requests still being
+				processed on this space */
+	ulint		purpose;/* FIL_TABLESPACE, FIL_LOG, or FIL_ARCH_LOG */
+	UT_LIST_BASE_NODE_T(fil_node_t) chain;
+				/* base node for the file chain */
+	ulint		size;	/* space size in pages; 0 if a single-table
+				tablespace whose size we do not know yet;
+				last incomplete megabytes in data files may be
+				ignored if space == 0 */
+	ulint		flags;	/* in: compressed page size
+				and file format, or 0 */
+	ulint		n_reserved_extents;
+				/* number of reserved free extents for
+				ongoing operations like B-tree page split */
+	ulint		n_pending_flushes; /* this is > 0 when flushing
+				the tablespace to disk; dropping of the
+				tablespace is forbidden if this is > 0 */
+	ulint		n_pending_ibuf_merges;/* this is > 0 when merging
+				insert buffer entries to a page so that we
+				may need to access the ibuf bitmap page in the
+				tablespade: dropping of the tablespace is
+				forbidden if this is > 0 */
+	hash_node_t	hash;	/* hash chain node */
+	hash_node_t	name_hash;/* hash chain the name_hash table */
+	rw_lock_t	latch;	/* latch protecting the file space storage
+				allocation */
+	UT_LIST_NODE_T(fil_space_t) unflushed_spaces;
+				/* list of spaces with at least one unflushed
+				file we have written to */
+	ibool		is_in_unflushed_spaces; /* TRUE if this space is
+				currently in the list above */
+	UT_LIST_NODE_T(fil_space_t) space_list;
+				/* list of all spaces */
+	ulint		magic_n;
+};
+
+#define	FIL_SPACE_MAGIC_N	89472
+
+/* The tablespace memory cache; also the totality of logs = the log data space,
+is stored here; below we talk about tablespaces, but also the ib_logfiles
+form a 'space' and it is handled here */
+
+typedef	struct fil_system_struct	fil_system_t;
+struct fil_system_struct {
+	mutex_t		mutex;		/* The mutex protecting the cache */
+	hash_table_t*	spaces;		/* The hash table of spaces in the
+					system; they are hashed on the space
+					id */
+	hash_table_t*	name_hash;	/* hash table based on the space
+					name */
+	UT_LIST_BASE_NODE_T(fil_node_t) LRU;
+					/* base node for the LRU list of the
+					most recently used open files with no
+					pending i/o's; if we start an i/o on
+					the file, we first remove it from this
+					list, and return it to the start of
+					the list when the i/o ends;
+					log files and the system tablespace are
+					not put to this list: they are opened
+					after the startup, and kept open until
+					shutdown */
+	UT_LIST_BASE_NODE_T(fil_space_t) unflushed_spaces;
+					/* base node for the list of those
+					tablespaces whose files contain
+					unflushed writes; those spaces have
+					at least one file node where
+					modification_counter > flush_counter */
+	ulint		n_open;		/* number of files currently open */
+	ulint		max_n_open;	/* n_open is not allowed to exceed
+					this */
+	ib_int64_t	modification_counter;/* when we write to a file we
+					increment this by one */
+	ulint		max_assigned_id;/* maximum space id in the existing
+					tables, or assigned during the time
+					mysqld has been up; at an InnoDB
+					startup we scan the data dictionary
+					and set here the maximum of the
+					space id's of the tables there */
+	ib_int64_t	tablespace_version;
+					/* a counter which is incremented for
+					every space object memory creation;
+					every space mem object gets a
+					'timestamp' from this; in DISCARD/
+					IMPORT this is used to check if we
+					should ignore an insert buffer merge
+					request */
+	UT_LIST_BASE_NODE_T(fil_space_t) space_list;
+					/* list of all file spaces */
+};
+
+/* The tablespace memory cache. This variable is NULL before the module is
+initialized. */
+UNIV_INTERN fil_system_t*	fil_system	= NULL;
+
+
+/************************************************************************
+NOTE: you must call fil_mutex_enter_and_prepare_for_io() first!
+
+Prepares a file node for i/o. Opens the file if it is closed. Updates the
+pending i/o's field in the node and the system appropriately. Takes the node
+off the LRU list if it is in the LRU list. The caller must hold the fil_sys
+mutex. */
+static
+void
+fil_node_prepare_for_io(
+/*====================*/
+	fil_node_t*	node,	/* in: file node */
+	fil_system_t*	system,	/* in: tablespace memory cache */
+	fil_space_t*	space);	/* in: space */
+/************************************************************************
+Updates the data structures when an i/o operation finishes. Updates the
+pending i/o's field in the node appropriately. */
+static
+void
+fil_node_complete_io(
+/*=================*/
+	fil_node_t*	node,	/* in: file node */
+	fil_system_t*	system,	/* in: tablespace memory cache */
+	ulint		type);	/* in: OS_FILE_WRITE or OS_FILE_READ; marks
+				the node as modified if
+				type == OS_FILE_WRITE */
+/***********************************************************************
+Checks if a single-table tablespace for a given table name exists in the
+tablespace memory cache. */
+static
+ulint
+fil_get_space_id_for_table(
+/*=======================*/
+				/* out: space id, ULINT_UNDEFINED if not
+				found */
+	const char*	name);	/* in: table name in the standard
+				'databasename/tablename' format */
+/************************************************************************
+Reads data from a space to a buffer. Remember that the possible incomplete
+blocks at the end of file are ignored: they are not taken into account when
+calculating the byte offset within a space. */
+UNIV_INLINE
+ulint
+fil_read(
+/*=====*/
+				/* out: DB_SUCCESS, or DB_TABLESPACE_DELETED
+				if we are trying to do i/o on a tablespace
+				which does not exist */
+	ibool	sync,		/* in: TRUE if synchronous aio is desired */
+	ulint	space_id,	/* in: space id */
+	ulint	zip_size,	/* in: compressed page size in bytes;
+				0 for uncompressed pages */
+	ulint	block_offset,	/* in: offset in number of blocks */
+	ulint	byte_offset,	/* in: remainder of offset in bytes; in aio
+				this must be divisible by the OS block size */
+	ulint	len,		/* in: how many bytes to read; this must not
+				cross a file boundary; in aio this must be a
+				block size multiple */
+	void*	buf,		/* in/out: buffer where to store data read;
+				in aio this must be appropriately aligned */
+	void*	message)	/* in: message for aio handler if non-sync
+				aio used, else ignored */
+{
+	return(fil_io(OS_FILE_READ, sync, space_id, zip_size, block_offset,
+					  byte_offset, len, buf, message));
+}
+
+/************************************************************************
+Writes data to a space from a buffer. Remember that the possible incomplete
+blocks at the end of file are ignored: they are not taken into account when
+calculating the byte offset within a space. */
+UNIV_INLINE
+ulint
+fil_write(
+/*======*/
+				/* out: DB_SUCCESS, or DB_TABLESPACE_DELETED
+				if we are trying to do i/o on a tablespace
+				which does not exist */
+	ibool	sync,		/* in: TRUE if synchronous aio is desired */
+	ulint	space_id,	/* in: space id */
+	ulint	zip_size,	/* in: compressed page size in bytes;
+				0 for uncompressed pages */
+	ulint	block_offset,	/* in: offset in number of blocks */
+	ulint	byte_offset,	/* in: remainder of offset in bytes; in aio
+				this must be divisible by the OS block size */
+	ulint	len,		/* in: how many bytes to write; this must
+				not cross a file boundary; in aio this must
+				be a block size multiple */
+	void*	buf,		/* in: buffer from which to write; in aio
+				this must be appropriately aligned */
+	void*	message)	/* in: message for aio handler if non-sync
+				aio used, else ignored */
+{
+	return(fil_io(OS_FILE_WRITE, sync, space_id, zip_size, block_offset,
+					   byte_offset, len, buf, message));
+}
+
+/***********************************************************************
+Returns the table space by a given id, NULL if not found. */
+UNIV_INLINE
+fil_space_t*
+fil_space_get_by_id(
+/*================*/
+	ulint	id)	/* in: space id */
+{
+	fil_space_t*	space;
+
+	ut_ad(mutex_own(&fil_system->mutex));
+
+	HASH_SEARCH(hash, fil_system->spaces, id,
+		    fil_space_t*, space,
+		    ut_ad(space->magic_n == FIL_SPACE_MAGIC_N),
+		    space->id == id);
+
+	return(space);
+}
+
+/***********************************************************************
+Returns the table space by a given name, NULL if not found. */
+UNIV_INLINE
+fil_space_t*
+fil_space_get_by_name(
+/*==================*/
+	const char*	name)	/* in: space name */
+{
+	fil_space_t*	space;
+	ulint		fold;
+
+	ut_ad(mutex_own(&fil_system->mutex));
+
+	fold = ut_fold_string(name);
+
+	HASH_SEARCH(name_hash, fil_system->name_hash, fold,
+		    fil_space_t*, space,
+		    ut_ad(space->magic_n == FIL_SPACE_MAGIC_N),
+		    !strcmp(name, space->name));
+
+	return(space);
+}
+
+/***********************************************************************
+Returns the version number of a tablespace, -1 if not found. */
+UNIV_INTERN
+ib_int64_t
+fil_space_get_version(
+/*==================*/
+			/* out: version number, -1 if the tablespace does not
+			exist in the memory cache */
+	ulint	id)	/* in: space id */
+{
+	fil_system_t*	system		= fil_system;
+	fil_space_t*	space;
+	ib_int64_t	version		= -1;
+
+	ut_ad(system);
+
+	mutex_enter(&(system->mutex));
+
+	space = fil_space_get_by_id(id);
+
+	if (space) {
+		version = space->tablespace_version;
+	}
+
+	mutex_exit(&(system->mutex));
+
+	return(version);
+}
+
+/***********************************************************************
+Returns the latch of a file space. */
+UNIV_INTERN
+rw_lock_t*
+fil_space_get_latch(
+/*================*/
+			/* out: latch protecting storage allocation */
+	ulint	id,	/* in: space id */
+	ulint*	flags)	/* out: tablespace flags */
+{
+	fil_system_t*	system		= fil_system;
+	fil_space_t*	space;
+
+	ut_ad(system);
+
+	mutex_enter(&(system->mutex));
+
+	space = fil_space_get_by_id(id);
+
+	ut_a(space);
+
+	if (flags) {
+		*flags = space->flags;
+	}
+
+	mutex_exit(&(system->mutex));
+
+	return(&(space->latch));
+}
+
+/***********************************************************************
+Returns the type of a file space. */
+UNIV_INTERN
+ulint
+fil_space_get_type(
+/*===============*/
+			/* out: FIL_TABLESPACE or FIL_LOG */
+	ulint	id)	/* in: space id */
+{
+	fil_system_t*	system		= fil_system;
+	fil_space_t*	space;
+
+	ut_ad(system);
+
+	mutex_enter(&(system->mutex));
+
+	space = fil_space_get_by_id(id);
+
+	ut_a(space);
+
+	mutex_exit(&(system->mutex));
+
+	return(space->purpose);
+}
+
+/**************************************************************************
+Checks if all the file nodes in a space are flushed. The caller must hold
+the fil_system mutex. */
+static
+ibool
+fil_space_is_flushed(
+/*=================*/
+				/* out: TRUE if all are flushed */
+	fil_space_t*	space)	/* in: space */
+{
+	fil_node_t*	node;
+
+	ut_ad(mutex_own(&(fil_system->mutex)));
+
+	node = UT_LIST_GET_FIRST(space->chain);
+
+	while (node) {
+		if (node->modification_counter > node->flush_counter) {
+
+			return(FALSE);
+		}
+
+		node = UT_LIST_GET_NEXT(chain, node);
+	}
+
+	return(TRUE);
+}
+
+/***********************************************************************
+Appends a new file to the chain of files of a space. File must be closed. */
+UNIV_INTERN
+void
+fil_node_create(
+/*============*/
+	const char*	name,	/* in: file name (file must be closed) */
+	ulint		size,	/* in: file size in database blocks, rounded
+				downwards to an integer */
+	ulint		id,	/* in: space id where to append */
+	ibool		is_raw)	/* in: TRUE if a raw device or
+				a raw disk partition */
+{
+	fil_system_t*	system	= fil_system;
+	fil_node_t*	node;
+	fil_space_t*	space;
+
+	ut_a(system);
+	ut_a(name);
+
+	mutex_enter(&(system->mutex));
+
+	node = mem_alloc(sizeof(fil_node_t));
+
+	node->name = mem_strdup(name);
+	node->open = FALSE;
+
+	ut_a(!is_raw || srv_start_raw_disk_in_use);
+
+	node->is_raw_disk = is_raw;
+	node->size = size;
+	node->magic_n = FIL_NODE_MAGIC_N;
+	node->n_pending = 0;
+	node->n_pending_flushes = 0;
+
+	node->modification_counter = 0;
+	node->flush_counter = 0;
+
+	space = fil_space_get_by_id(id);
+
+	if (!space) {
+		ut_print_timestamp(stderr);
+		fprintf(stderr,
+			"  InnoDB: Error: Could not find tablespace %lu for\n"
+			"InnoDB: file ", (ulong) id);
+		ut_print_filename(stderr, name);
+		fputs(" in the tablespace memory cache.\n", stderr);
+		mem_free(node->name);
+
+		mem_free(node);
+
+		mutex_exit(&(system->mutex));
+
+		return;
+	}
+
+	space->size += size;
+
+	node->space = space;
+
+	UT_LIST_ADD_LAST(chain, space->chain, node);
+
+	mutex_exit(&(system->mutex));
+}
+
+/************************************************************************
+Opens a the file of a node of a tablespace. The caller must own the fil_system
+mutex. */
+static
+void
+fil_node_open_file(
+/*===============*/
+	fil_node_t*	node,	/* in: file node */
+	fil_system_t*	system,	/* in: tablespace memory cache */
+	fil_space_t*	space)	/* in: space */
+{
+	ib_int64_t	size_bytes;
+	ulint		size_low;
+	ulint		size_high;
+	ibool		ret;
+	ibool		success;
+#ifndef UNIV_HOTBACKUP
+	byte*		buf2;
+	byte*		page;
+	ulint		space_id;
+	ulint		flags;
+#endif /* !UNIV_HOTBACKUP */
+
+	ut_ad(mutex_own(&(system->mutex)));
+	ut_a(node->n_pending == 0);
+	ut_a(node->open == FALSE);
+
+	if (node->size == 0) {
+		/* It must be a single-table tablespace and we do not know the
+		size of the file yet. First we open the file in the normal
+		mode, no async I/O here, for simplicity. Then do some checks,
+		and close the file again.
+		NOTE that we could not use the simple file read function
+		os_file_read() in Windows to read from a file opened for
+		async I/O! */
+
+		node->handle = os_file_create_simple_no_error_handling(
+			node->name, OS_FILE_OPEN, OS_FILE_READ_ONLY, &success);
+		if (!success) {
+			/* The following call prints an error message */
+			os_file_get_last_error(TRUE);
+
+			ut_print_timestamp(stderr);
+
+			fprintf(stderr,
+				"  InnoDB: Fatal error: cannot open %s\n."
+				"InnoDB: Have you deleted .ibd files"
+				" under a running mysqld server?\n",
+				node->name);
+			ut_a(0);
+		}
+
+		os_file_get_size(node->handle, &size_low, &size_high);
+
+		size_bytes = (((ib_int64_t)size_high) << 32)
+			+ (ib_int64_t)size_low;
+#ifdef UNIV_HOTBACKUP
+		node->size = (ulint) (size_bytes / UNIV_PAGE_SIZE);
+		/* TODO: adjust to zip_size, like below? */
+#else
+		ut_a(space->purpose != FIL_LOG);
+		ut_a(space->id != 0);
+
+		if (size_bytes < FIL_IBD_FILE_INITIAL_SIZE * UNIV_PAGE_SIZE) {
+			fprintf(stderr,
+				"InnoDB: Error: the size of single-table"
+				" tablespace file %s\n"
+				"InnoDB: is only %lu %lu,"
+				" should be at least %lu!\n",
+				node->name,
+				(ulong) size_high,
+				(ulong) size_low,
+				(ulong) (FIL_IBD_FILE_INITIAL_SIZE
+					 * UNIV_PAGE_SIZE));
+
+			ut_a(0);
+		}
+
+		/* Read the first page of the tablespace */
+
+		buf2 = ut_malloc(2 * UNIV_PAGE_SIZE);
+		/* Align the memory for file i/o if we might have O_DIRECT
+		set */
+		page = ut_align(buf2, UNIV_PAGE_SIZE);
+
+		success = os_file_read(node->handle, page, 0, 0,
+				       UNIV_PAGE_SIZE);
+		space_id = fsp_header_get_space_id(page);
+		flags = fsp_header_get_flags(page);
+
+		ut_free(buf2);
+
+		/* Close the file now that we have read the space id from it */
+
+		os_file_close(node->handle);
+
+		if (UNIV_UNLIKELY(space_id != space->id)) {
+			fprintf(stderr,
+				"InnoDB: Error: tablespace id is %lu"
+				" in the data dictionary\n"
+				"InnoDB: but in file %s it is %lu!\n",
+				space->id, node->name, space_id);
+
+			ut_error;
+		}
+
+		if (UNIV_UNLIKELY(space_id == ULINT_UNDEFINED
+				  || space_id == 0)) {
+			fprintf(stderr,
+				"InnoDB: Error: tablespace id %lu"
+				" in file %s is not sensible\n",
+				(ulong) space_id, node->name);
+
+			ut_error;
+		}
+
+		if (UNIV_UNLIKELY(space->flags != flags)) {
+			fprintf(stderr,
+				"InnoDB: Error: table flags are %lx"
+				" in the data dictionary\n"
+				"InnoDB: but the flags in file %s are %lx!\n",
+				space->flags, node->name, flags);
+
+			ut_error;
+		}
+
+		if (size_bytes >= 1024 * 1024) {
+			/* Truncate the size to whole megabytes. */
+			size_bytes = ut_2pow_round(size_bytes, 1024 * 1024);
+		}
+
+		if (!(flags & DICT_TF_ZSSIZE_MASK)) {
+			node->size = (ulint) (size_bytes / UNIV_PAGE_SIZE);
+		} else {
+			node->size = (ulint)
+				(size_bytes
+				 / dict_table_flags_to_zip_size(flags));
+		}
+#endif
+		space->size += node->size;
+	}
+
+	/* printf("Opening file %s\n", node->name); */
+
+	/* Open the file for reading and writing, in Windows normally in the
+	unbuffered async I/O mode, though global variables may make
+	os_file_create() to fall back to the normal file I/O mode. */
+
+	if (space->purpose == FIL_LOG) {
+		node->handle = os_file_create(node->name, OS_FILE_OPEN,
+					      OS_FILE_AIO, OS_LOG_FILE, &ret);
+	} else if (node->is_raw_disk) {
+		node->handle = os_file_create(node->name,
+					      OS_FILE_OPEN_RAW,
+					      OS_FILE_AIO, OS_DATA_FILE, &ret);
+	} else {
+		node->handle = os_file_create(node->name, OS_FILE_OPEN,
+					      OS_FILE_AIO, OS_DATA_FILE, &ret);
+	}
+
+	ut_a(ret);
+
+	node->open = TRUE;
+
+	system->n_open++;
+
+	if (space->purpose == FIL_TABLESPACE && space->id != 0) {
+		/* Put the node to the LRU list */
+		UT_LIST_ADD_FIRST(LRU, system->LRU, node);
+	}
+}
+
+/**************************************************************************
+Closes a file. */
+static
+void
+fil_node_close_file(
+/*================*/
+	fil_node_t*	node,	/* in: file node */
+	fil_system_t*	system)	/* in: tablespace memory cache */
+{
+	ibool	ret;
+
+	ut_ad(node && system);
+	ut_ad(mutex_own(&(system->mutex)));
+	ut_a(node->open);
+	ut_a(node->n_pending == 0);
+	ut_a(node->n_pending_flushes == 0);
+	ut_a(node->modification_counter == node->flush_counter);
+
+	ret = os_file_close(node->handle);
+	ut_a(ret);
+
+	/* printf("Closing file %s\n", node->name); */
+
+	node->open = FALSE;
+	ut_a(system->n_open > 0);
+	system->n_open--;
+
+	if (node->space->purpose == FIL_TABLESPACE && node->space->id != 0) {
+		ut_a(UT_LIST_GET_LEN(system->LRU) > 0);
+
+		/* The node is in the LRU list, remove it */
+		UT_LIST_REMOVE(LRU, system->LRU, node);
+	}
+}
+
+/************************************************************************
+Tries to close a file in the LRU list. The caller must hold the fil_sys
+mutex. */
+static
+ibool
+fil_try_to_close_file_in_LRU(
+/*=========================*/
+				/* out: TRUE if success, FALSE if should retry
+				later; since i/o's generally complete in <
+				100 ms, and as InnoDB writes at most 128 pages
+				from the buffer pool in a batch, and then
+				immediately flushes the files, there is a good
+				chance that the next time we find a suitable
+				node from the LRU list */
+	ibool	print_info)	/* in: if TRUE, prints information why it
+				cannot close a file */
+{
+	fil_system_t*	system		= fil_system;
+	fil_node_t*	node;
+
+	ut_ad(mutex_own(&(system->mutex)));
+
+	node = UT_LIST_GET_LAST(system->LRU);
+
+	if (print_info) {
+		fprintf(stderr,
+			"InnoDB: fil_sys open file LRU len %lu\n",
+			(ulong) UT_LIST_GET_LEN(system->LRU));
+	}
+
+	while (node != NULL) {
+		if (node->modification_counter == node->flush_counter
+		    && node->n_pending_flushes == 0) {
+
+			fil_node_close_file(node, system);
+
+			return(TRUE);
+		}
+
+		if (print_info && node->n_pending_flushes > 0) {
+			fputs("InnoDB: cannot close file ", stderr);
+			ut_print_filename(stderr, node->name);
+			fprintf(stderr, ", because n_pending_flushes %lu\n",
+				(ulong) node->n_pending_flushes);
+		}
+
+		if (print_info
+		    && node->modification_counter != node->flush_counter) {
+			fputs("InnoDB: cannot close file ", stderr);
+			ut_print_filename(stderr, node->name);
+			fprintf(stderr,
+				", because mod_count %ld != fl_count %ld\n",
+				(long) node->modification_counter,
+				(long) node->flush_counter);
+		}
+
+		node = UT_LIST_GET_PREV(LRU, node);
+	}
+
+	return(FALSE);
+}
+
+/***********************************************************************
+Reserves the fil_system mutex and tries to make sure we can open at least one
+file while holding it. This should be called before calling
+fil_node_prepare_for_io(), because that function may need to open a file. */
+static
+void
+fil_mutex_enter_and_prepare_for_io(
+/*===============================*/
+	ulint	space_id)	/* in: space id */
+{
+	fil_system_t*	system		= fil_system;
+	fil_space_t*	space;
+	ibool		success;
+	ibool		print_info	= FALSE;
+	ulint		count		= 0;
+	ulint		count2		= 0;
+
+	ut_ad(!mutex_own(&(system->mutex)));
+retry:
+	mutex_enter(&(system->mutex));
+
+	if (space_id == 0 || space_id >= SRV_LOG_SPACE_FIRST_ID) {
+		/* We keep log files and system tablespace files always open;
+		this is important in preventing deadlocks in this module, as
+		a page read completion often performs another read from the
+		insert buffer. The insert buffer is in tablespace 0, and we
+		cannot end up waiting in this function. */
+
+		return;
+	}
+
+	if (system->n_open < system->max_n_open) {
+
+		return;
+	}
+
+	space = fil_space_get_by_id(space_id);
+
+	if (space != NULL && space->stop_ios) {
+		/* We are going to do a rename file and want to stop new i/o's
+		for a while */
+
+		if (count2 > 20000) {
+			fputs("InnoDB: Warning: tablespace ", stderr);
+			ut_print_filename(stderr, space->name);
+			fprintf(stderr,
+				" has i/o ops stopped for a long time %lu\n",
+				(ulong) count2);
+		}
+
+		mutex_exit(&(system->mutex));
+
+		os_thread_sleep(20000);
+
+		count2++;
+
+		goto retry;
+	}
+
+	/* If the file is already open, no need to do anything; if the space
+	does not exist, we handle the situation in the function which called
+	this function */
+
+	if (!space || UT_LIST_GET_FIRST(space->chain)->open) {
+
+		return;
+	}
+
+	if (count > 1) {
+		print_info = TRUE;
+	}
+
+	/* Too many files are open, try to close some */
+close_more:
+	success = fil_try_to_close_file_in_LRU(print_info);
+
+	if (success && system->n_open >= system->max_n_open) {
+
+		goto close_more;
+	}
+
+	if (system->n_open < system->max_n_open) {
+		/* Ok */
+
+		return;
+	}
+
+	if (count >= 2) {
+		ut_print_timestamp(stderr);
+		fprintf(stderr,
+			"  InnoDB: Warning: too many (%lu) files stay open"
+			" while the maximum\n"
+			"InnoDB: allowed value would be %lu.\n"
+			"InnoDB: You may need to raise the value of"
+			" innodb_max_files_open in\n"
+			"InnoDB: my.cnf.\n",
+			(ulong) system->n_open, (ulong) system->max_n_open);
+
+		return;
+	}
+
+	mutex_exit(&(system->mutex));
+
+#ifndef UNIV_HOTBACKUP
+	/* Wake the i/o-handler threads to make sure pending i/o's are
+	performed */
+	os_aio_simulated_wake_handler_threads();
+
+	os_thread_sleep(20000);
+#endif
+	/* Flush tablespaces so that we can close modified files in the LRU
+	list */
+
+	fil_flush_file_spaces(FIL_TABLESPACE);
+
+	count++;
+
+	goto retry;
+}
+
+/***********************************************************************
+Frees a file node object from a tablespace memory cache. */
+static
+void
+fil_node_free(
+/*==========*/
+	fil_node_t*	node,	/* in, own: file node */
+	fil_system_t*	system,	/* in: tablespace memory cache */
+	fil_space_t*	space)	/* in: space where the file node is chained */
+{
+	ut_ad(node && system && space);
+	ut_ad(mutex_own(&(system->mutex)));
+	ut_a(node->magic_n == FIL_NODE_MAGIC_N);
+	ut_a(node->n_pending == 0);
+
+	if (node->open) {
+		/* We fool the assertion in fil_node_close_file() to think
+		there are no unflushed modifications in the file */
+
+		node->modification_counter = node->flush_counter;
+
+		if (space->is_in_unflushed_spaces
+		    && fil_space_is_flushed(space)) {
+
+			space->is_in_unflushed_spaces = FALSE;
+
+			UT_LIST_REMOVE(unflushed_spaces,
+				       system->unflushed_spaces,
+				       space);
+		}
+
+		fil_node_close_file(node, system);
+	}
+
+	space->size -= node->size;
+
+	UT_LIST_REMOVE(chain, space->chain, node);
+
+	mem_free(node->name);
+	mem_free(node);
+}
+
+#ifdef UNIV_LOG_ARCHIVE
+/********************************************************************
+Drops files from the start of a file space, so that its size is cut by
+the amount given. */
+UNIV_INTERN
+void
+fil_space_truncate_start(
+/*=====================*/
+	ulint	id,		/* in: space id */
+	ulint	trunc_len)	/* in: truncate by this much; it is an error
+				if this does not equal to the combined size of
+				some initial files in the space */
+{
+	fil_system_t*	system		= fil_system;
+	fil_node_t*	node;
+	fil_space_t*	space;
+
+	mutex_enter(&(system->mutex));
+
+	space = fil_space_get_by_id(id);
+
+	ut_a(space);
+
+	while (trunc_len > 0) {
+		node = UT_LIST_GET_FIRST(space->chain);
+
+		ut_a(node->size * UNIV_PAGE_SIZE <= trunc_len);
+
+		trunc_len -= node->size * UNIV_PAGE_SIZE;
+
+		fil_node_free(node, system, space);
+	}
+
+	mutex_exit(&(system->mutex));
+}
+#endif /* UNIV_LOG_ARCHIVE */
+
+/***********************************************************************
+Creates a space memory object and puts it to the tablespace memory cache. If
+there is an error, prints an error message to the .err log. */
+UNIV_INTERN
+ibool
+fil_space_create(
+/*=============*/
+				/* out: TRUE if success */
+	const char*	name,	/* in: space name */
+	ulint		id,	/* in: space id */
+	ulint		flags,	/* in: compressed page size
+				and file format, or 0 */
+	ulint		purpose)/* in: FIL_TABLESPACE, or FIL_LOG if log */
+{
+	fil_system_t*	system		= fil_system;
+	fil_space_t*	space;
+
+	/* The tablespace flags (FSP_SPACE_FLAGS) should be 0 for
+	ROW_FORMAT=COMPACT (table->flags == DICT_TF_COMPACT) and
+	ROW_FORMAT=REDUNDANT (table->flags == 0).  For any other
+	format, the tablespace flags should equal table->flags. */
+	ut_a(flags != DICT_TF_COMPACT);
+
+try_again:
+	/*printf(
+	"InnoDB: Adding tablespace %lu of name %s, purpose %lu\n", id, name,
+	purpose);*/
+
+	ut_a(system);
+	ut_a(name);
+
+	mutex_enter(&(system->mutex));
+
+	space = fil_space_get_by_name(name);
+
+	if (UNIV_LIKELY_NULL(space)) {
+		ulint	namesake_id;
+
+		ut_print_timestamp(stderr);
+		fprintf(stderr,
+			"  InnoDB: Warning: trying to init to the"
+			" tablespace memory cache\n"
+			"InnoDB: a tablespace %lu of name ", (ulong) id);
+		ut_print_filename(stderr, name);
+		fprintf(stderr, ",\n"
+			"InnoDB: but a tablespace %lu of the same name\n"
+			"InnoDB: already exists in the"
+			" tablespace memory cache!\n",
+			(ulong) space->id);
+
+		if (id == 0 || purpose != FIL_TABLESPACE) {
+
+			mutex_exit(&(system->mutex));
+
+			return(FALSE);
+		}
+
+		fprintf(stderr,
+			"InnoDB: We assume that InnoDB did a crash recovery,"
+			" and you had\n"
+			"InnoDB: an .ibd file for which the table"
+			" did not exist in the\n"
+			"InnoDB: InnoDB internal data dictionary in the"
+			" ibdata files.\n"
+			"InnoDB: We assume that you later removed the"
+			" .ibd and .frm files,\n"
+			"InnoDB: and are now trying to recreate the table."
+			" We now remove the\n"
+			"InnoDB: conflicting tablespace object"
+			" from the memory cache and try\n"
+			"InnoDB: the init again.\n");
+
+		namesake_id = space->id;
+
+		mutex_exit(&(system->mutex));
+
+		fil_space_free(namesake_id);
+
+		goto try_again;
+	}
+
+	space = fil_space_get_by_id(id);
+
+	if (UNIV_LIKELY_NULL(space)) {
+		fprintf(stderr,
+			"InnoDB: Error: trying to add tablespace %lu"
+			" of name ", (ulong) id);
+		ut_print_filename(stderr, name);
+		fprintf(stderr, "\n"
+			"InnoDB: to the tablespace memory cache,"
+			" but tablespace\n"
+			"InnoDB: %lu of name ", (ulong) space->id);
+		ut_print_filename(stderr, space->name);
+		fputs(" already exists in the tablespace\n"
+		      "InnoDB: memory cache!\n", stderr);
+
+		mutex_exit(&(system->mutex));
+
+		return(FALSE);
+	}
+
+	space = mem_alloc(sizeof(fil_space_t));
+
+	space->name = mem_strdup(name);
+	space->id = id;
+
+	system->tablespace_version++;
+	space->tablespace_version = system->tablespace_version;
+	space->mark = FALSE;
+
+	if (purpose == FIL_TABLESPACE && id > system->max_assigned_id) {
+		system->max_assigned_id = id;
+	}
+
+	space->stop_ios = FALSE;
+	space->stop_ibuf_merges = FALSE;
+	space->is_being_deleted = FALSE;
+	space->purpose = purpose;
+	space->size = 0;
+	space->flags = flags;
+
+	space->n_reserved_extents = 0;
+
+	space->n_pending_flushes = 0;
+	space->n_pending_ibuf_merges = 0;
+
+	UT_LIST_INIT(space->chain);
+	space->magic_n = FIL_SPACE_MAGIC_N;
+
+	rw_lock_create(&space->latch, SYNC_FSP);
+
+	HASH_INSERT(fil_space_t, hash, system->spaces, id, space);
+
+	HASH_INSERT(fil_space_t, name_hash, system->name_hash,
+		    ut_fold_string(name), space);
+	space->is_in_unflushed_spaces = FALSE;
+
+	UT_LIST_ADD_LAST(space_list, system->space_list, space);
+
+	mutex_exit(&(system->mutex));
+
+	return(TRUE);
+}
+
+/***********************************************************************
+Assigns a new space id for a new single-table tablespace. This works simply by
+incrementing the global counter. If 4 billion id's is not enough, we may need
+to recycle id's. */
+static
+ulint
+fil_assign_new_space_id(void)
+/*=========================*/
+			/* out: new tablespace id; ULINT_UNDEFINED if could
+			not assign an id */
+{
+	fil_system_t*	system = fil_system;
+	ulint		id;
+
+	mutex_enter(&(system->mutex));
+
+	system->max_assigned_id++;
+
+	id = system->max_assigned_id;
+
+	if (id > (SRV_LOG_SPACE_FIRST_ID / 2) && (id % 1000000UL == 0)) {
+		ut_print_timestamp(stderr);
+		fprintf(stderr,
+			"InnoDB: Warning: you are running out of new"
+			" single-table tablespace id's.\n"
+			"InnoDB: Current counter is %lu and it"
+			" must not exceed %lu!\n"
+			"InnoDB: To reset the counter to zero"
+			" you have to dump all your tables and\n"
+			"InnoDB: recreate the whole InnoDB installation.\n",
+			(ulong) id,
+			(ulong) SRV_LOG_SPACE_FIRST_ID);
+	}
+
+	if (id >= SRV_LOG_SPACE_FIRST_ID) {
+		ut_print_timestamp(stderr);
+		fprintf(stderr,
+			"InnoDB: You have run out of single-table"
+			" tablespace id's!\n"
+			"InnoDB: Current counter is %lu.\n"
+			"InnoDB: To reset the counter to zero you"
+			" have to dump all your tables and\n"
+			"InnoDB: recreate the whole InnoDB installation.\n",
+			(ulong) id);
+		system->max_assigned_id--;
+
+		id = ULINT_UNDEFINED;
+	}
+
+	mutex_exit(&(system->mutex));
+
+	return(id);
+}
+
+/***********************************************************************
+Frees a space object from the tablespace memory cache. Closes the files in
+the chain but does not delete them. There must not be any pending i/o's or
+flushes on the files. */
+UNIV_INTERN
+ibool
+fil_space_free(
+/*===========*/
+			/* out: TRUE if success */
+	ulint	id)	/* in: space id */
+{
+	fil_system_t*	system = fil_system;
+	fil_space_t*	space;
+	fil_space_t*	namespace;
+	fil_node_t*	fil_node;
+
+	mutex_enter(&(system->mutex));
+
+	space = fil_space_get_by_id(id);
+
+	if (!space) {
+		ut_print_timestamp(stderr);
+		fprintf(stderr,
+			"  InnoDB: Error: trying to remove tablespace %lu"
+			" from the cache but\n"
+			"InnoDB: it is not there.\n", (ulong) id);
+
+		mutex_exit(&(system->mutex));
+
+		return(FALSE);
+	}
+
+	HASH_DELETE(fil_space_t, hash, system->spaces, id, space);
+
+	namespace = fil_space_get_by_name(space->name);
+	ut_a(namespace);
+	ut_a(space == namespace);
+
+	HASH_DELETE(fil_space_t, name_hash, system->name_hash,
+		    ut_fold_string(space->name), space);
+
+	if (space->is_in_unflushed_spaces) {
+		space->is_in_unflushed_spaces = FALSE;
+
+		UT_LIST_REMOVE(unflushed_spaces, system->unflushed_spaces,
+			       space);
+	}
+
+	UT_LIST_REMOVE(space_list, system->space_list, space);
+
+	ut_a(space->magic_n == FIL_SPACE_MAGIC_N);
+	ut_a(0 == space->n_pending_flushes);
+
+	fil_node = UT_LIST_GET_FIRST(space->chain);
+
+	while (fil_node != NULL) {
+		fil_node_free(fil_node, system, space);
+
+		fil_node = UT_LIST_GET_FIRST(space->chain);
+	}
+
+	ut_a(0 == UT_LIST_GET_LEN(space->chain));
+
+	mutex_exit(&(system->mutex));
+
+	rw_lock_free(&(space->latch));
+
+	mem_free(space->name);
+	mem_free(space);
+
+	return(TRUE);
+}
+
+#ifdef UNIV_HOTBACKUP
+/***********************************************************************
+Returns the tablespace object for a given id, or NULL if not found from the
+tablespace memory cache. */
+static
+fil_space_t*
+fil_get_space_for_id_low(
+/*=====================*/
+			/* out: tablespace object or NULL; NOTE that you must
+			own &(fil_system->mutex) to call this function! */
+	ulint	id)	/* in: space id */
+{
+	fil_system_t*	system		= fil_system;
+	fil_space_t*	space;
+
+	ut_ad(system);
+
+	space = fil_space_get_by_id(id);
+
+	return(space);
+}
+#endif
+
+/***********************************************************************
+Returns the size of the space in pages. The tablespace must be cached in the
+memory cache. */
+UNIV_INTERN
+ulint
+fil_space_get_size(
+/*===============*/
+			/* out: space size, 0 if space not found */
+	ulint	id)	/* in: space id */
+{
+	fil_system_t*	system		= fil_system;
+	fil_node_t*	node;
+	fil_space_t*	space;
+	ulint		size;
+
+	ut_ad(system);
+
+	fil_mutex_enter_and_prepare_for_io(id);
+
+	space = fil_space_get_by_id(id);
+
+	if (space == NULL) {
+		mutex_exit(&(system->mutex));
+
+		return(0);
+	}
+
+	if (space->size == 0 && space->purpose == FIL_TABLESPACE) {
+		ut_a(id != 0);
+
+		ut_a(1 == UT_LIST_GET_LEN(space->chain));
+
+		node = UT_LIST_GET_FIRST(space->chain);
+
+		/* It must be a single-table tablespace and we have not opened
+		the file yet; the following calls will open it and update the
+		size fields */
+
+		fil_node_prepare_for_io(node, system, space);
+		fil_node_complete_io(node, system, OS_FILE_READ);
+	}
+
+	size = space->size;
+
+	mutex_exit(&(system->mutex));
+
+	return(size);
+}
+
+/***********************************************************************
+Returns the flags of the space. The tablespace must be cached
+in the memory cache. */
+UNIV_INTERN
+ulint
+fil_space_get_flags(
+/*================*/
+			/* out: flags, ULINT_UNDEFINED if space not found */
+	ulint	id)	/* in: space id */
+{
+	fil_system_t*	system		= fil_system;
+	fil_node_t*	node;
+	fil_space_t*	space;
+	ulint		flags;
+
+	ut_ad(system);
+
+	if (UNIV_UNLIKELY(!id)) {
+		return(0);
+	}
+
+	fil_mutex_enter_and_prepare_for_io(id);
+
+	space = fil_space_get_by_id(id);
+
+	if (space == NULL) {
+		mutex_exit(&(system->mutex));
+
+		return(ULINT_UNDEFINED);
+	}
+
+	if (space->size == 0 && space->purpose == FIL_TABLESPACE) {
+		ut_a(id != 0);
+
+		ut_a(1 == UT_LIST_GET_LEN(space->chain));
+
+		node = UT_LIST_GET_FIRST(space->chain);
+
+		/* It must be a single-table tablespace and we have not opened
+		the file yet; the following calls will open it and update the
+		size fields */
+
+		fil_node_prepare_for_io(node, system, space);
+		fil_node_complete_io(node, system, OS_FILE_READ);
+	}
+
+	flags = space->flags;
+
+	mutex_exit(&(system->mutex));
+
+	return(flags);
+}
+
+/***********************************************************************
+Returns the compressed page size of the space, or 0 if the space
+is not compressed. The tablespace must be cached in the memory cache. */
+UNIV_INTERN
+ulint
+fil_space_get_zip_size(
+/*===================*/
+			/* out: compressed page size, ULINT_UNDEFINED
+			if space not found */
+	ulint	id)	/* in: space id */
+{
+	ulint	flags;
+
+	flags = fil_space_get_flags(id);
+
+	if (flags && flags != ULINT_UNDEFINED) {
+
+		return(dict_table_flags_to_zip_size(flags));
+	}
+
+	return(flags);
+}
+
+/***********************************************************************
+Checks if the pair space, page_no refers to an existing page in a tablespace
+file space. The tablespace must be cached in the memory cache. */
+UNIV_INTERN
+ibool
+fil_check_adress_in_tablespace(
+/*===========================*/
+			/* out: TRUE if the address is meaningful */
+	ulint	id,	/* in: space id */
+	ulint	page_no)/* in: page number */
+{
+	if (fil_space_get_size(id) > page_no) {
+
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
+
+/********************************************************************
+Creates a the tablespace memory cache. */
+static
+fil_system_t*
+fil_system_create(
+/*==============*/
+				/* out, own: tablespace memory cache */
+	ulint	hash_size,	/* in: hash table size */
+	ulint	max_n_open)	/* in: maximum number of open files; must be
+				> 10 */
+{
+	fil_system_t*	system;
+
+	ut_a(hash_size > 0);
+	ut_a(max_n_open > 0);
+
+	system = mem_alloc(sizeof(fil_system_t));
+
+	mutex_create(&system->mutex, SYNC_ANY_LATCH);
+
+	system->spaces = hash_create(hash_size);
+	system->name_hash = hash_create(hash_size);
+
+	UT_LIST_INIT(system->LRU);
+
+	system->n_open = 0;
+	system->max_n_open = max_n_open;
+
+	system->modification_counter = 0;
+	system->max_assigned_id = 0;
+
+	system->tablespace_version = 0;
+
+	UT_LIST_INIT(system->unflushed_spaces);
+	UT_LIST_INIT(system->space_list);
+
+	return(system);
+}
+
+/********************************************************************
+Initializes the tablespace memory cache. */
+UNIV_INTERN
+void
+fil_init(
+/*=====*/
+	ulint	max_n_open)	/* in: max number of open files */
+{
+	ulint	hash_size;
+
+	ut_a(fil_system == NULL);
+
+	if (srv_file_per_table) {
+		hash_size = 50000;
+	} else {
+		hash_size = 5000;
+	}
+
+	fil_system = fil_system_create(hash_size, max_n_open);
+}
+
+/***********************************************************************
+Opens all log files and system tablespace data files. They stay open until the
+database server shutdown. This should be called at a server startup after the
+space objects for the log and the system tablespace have been created. The
+purpose of this operation is to make sure we never run out of file descriptors
+if we need to read from the insert buffer or to write to the log. */
+UNIV_INTERN
+void
+fil_open_log_and_system_tablespace_files(void)
+/*==========================================*/
+{
+	fil_system_t*	system = fil_system;
+	fil_space_t*	space;
+	fil_node_t*	node;
+
+	mutex_enter(&(system->mutex));
+
+	space = UT_LIST_GET_FIRST(system->space_list);
+
+	while (space != NULL) {
+		if (space->purpose != FIL_TABLESPACE || space->id == 0) {
+			node = UT_LIST_GET_FIRST(space->chain);
+
+			while (node != NULL) {
+				if (!node->open) {
+					fil_node_open_file(node, system,
+							   space);
+				}
+				if (system->max_n_open < 10 + system->n_open) {
+					fprintf(stderr,
+						"InnoDB: Warning: you must"
+						" raise the value of"
+						" innodb_max_open_files in\n"
+						"InnoDB: my.cnf! Remember that"
+						" InnoDB keeps all log files"
+						" and all system\n"
+						"InnoDB: tablespace files open"
+						" for the whole time mysqld is"
+						" running, and\n"
+						"InnoDB: needs to open also"
+						" some .ibd files if the"
+						" file-per-table storage\n"
+						"InnoDB: model is used."
+						" Current open files %lu,"
+						" max allowed"
+						" open files %lu.\n",
+						(ulong) system->n_open,
+						(ulong) system->max_n_open);
+				}
+				node = UT_LIST_GET_NEXT(chain, node);
+			}
+		}
+		space = UT_LIST_GET_NEXT(space_list, space);
+	}
+
+	mutex_exit(&(system->mutex));
+}
+
+/***********************************************************************
+Closes all open files. There must not be any pending i/o's or not flushed
+modifications in the files. */
+UNIV_INTERN
+void
+fil_close_all_files(void)
+/*=====================*/
+{
+	fil_system_t*	system = fil_system;
+	fil_space_t*	space;
+	fil_node_t*	node;
+
+	mutex_enter(&(system->mutex));
+
+	space = UT_LIST_GET_FIRST(system->space_list);
+
+	while (space != NULL) {
+		node = UT_LIST_GET_FIRST(space->chain);
+
+		while (node != NULL) {
+			if (node->open) {
+				fil_node_close_file(node, system);
+			}
+			node = UT_LIST_GET_NEXT(chain, node);
+		}
+		space = UT_LIST_GET_NEXT(space_list, space);
+	}
+
+	mutex_exit(&(system->mutex));
+}
+
+/***********************************************************************
+Sets the max tablespace id counter if the given number is bigger than the
+previous value. */
+UNIV_INTERN
+void
+fil_set_max_space_id_if_bigger(
+/*===========================*/
+	ulint	max_id)	/* in: maximum known id */
+{
+	fil_system_t*	system = fil_system;
+
+	if (max_id >= SRV_LOG_SPACE_FIRST_ID) {
+		fprintf(stderr,
+			"InnoDB: Fatal error: max tablespace id"
+			" is too high, %lu\n", (ulong) max_id);
+		ut_a(0);
+	}
+
+	mutex_enter(&(system->mutex));
+
+	if (system->max_assigned_id < max_id) {
+
+		system->max_assigned_id = max_id;
+	}
+
+	mutex_exit(&(system->mutex));
+}
+
+/********************************************************************
+Writes the flushed lsn and the latest archived log number to the page header
+of the first page of a data file of the system tablespace (space 0),
+which is uncompressed. */
+static
+ulint
+fil_write_lsn_and_arch_no_to_file(
+/*==============================*/
+	ulint		sum_of_sizes,	/* in: combined size of previous files
+					in space, in database pages */
+	ib_uint64_t	lsn,		/* in: lsn to write */
+	ulint		arch_log_no	/* in: archived log number to write */
+	__attribute__((unused)))
+{
+	byte*	buf1;
+	byte*	buf;
+
+	buf1 = mem_alloc(2 * UNIV_PAGE_SIZE);
+	buf = ut_align(buf1, UNIV_PAGE_SIZE);
+
+	fil_read(TRUE, 0, 0, sum_of_sizes, 0, UNIV_PAGE_SIZE, buf, NULL);
+
+	mach_write_ull(buf + FIL_PAGE_FILE_FLUSH_LSN, lsn);
+
+	fil_write(TRUE, 0, 0, sum_of_sizes, 0, UNIV_PAGE_SIZE, buf, NULL);
+
+	mem_free(buf1);
+
+	return(DB_SUCCESS);
+}
+
+/********************************************************************
+Writes the flushed lsn and the latest archived log number to the page
+header of the first page of each data file in the system tablespace. */
+UNIV_INTERN
+ulint
+fil_write_flushed_lsn_to_data_files(
+/*================================*/
+					/* out: DB_SUCCESS or error number */
+	ib_uint64_t	lsn,		/* in: lsn to write */
+	ulint		arch_log_no)	/* in: latest archived log
+					file number */
+{
+	fil_space_t*	space;
+	fil_node_t*	node;
+	ulint		sum_of_sizes;
+	ulint		err;
+
+	mutex_enter(&(fil_system->mutex));
+
+	space = UT_LIST_GET_FIRST(fil_system->space_list);
+
+	while (space) {
+		/* We only write the lsn to all existing data files which have
+		been open during the lifetime of the mysqld process; they are
+		represented by the space objects in the tablespace memory
+		cache. Note that all data files in the system tablespace 0 are
+		always open. */
+
+		if (space->purpose == FIL_TABLESPACE
+		    && space->id == 0) {
+			sum_of_sizes = 0;
+
+			node = UT_LIST_GET_FIRST(space->chain);
+			while (node) {
+				mutex_exit(&(fil_system->mutex));
+
+				err = fil_write_lsn_and_arch_no_to_file(
+					sum_of_sizes, lsn, arch_log_no);
+				if (err != DB_SUCCESS) {
+
+					return(err);
+				}
+
+				mutex_enter(&(fil_system->mutex));
+
+				sum_of_sizes += node->size;
+				node = UT_LIST_GET_NEXT(chain, node);
+			}
+		}
+		space = UT_LIST_GET_NEXT(space_list, space);
+	}
+
+	mutex_exit(&(fil_system->mutex));
+
+	return(DB_SUCCESS);
+}
+
+/***********************************************************************
+Reads the flushed lsn and arch no fields from a data file at database
+startup. */
+UNIV_INTERN
+void
+fil_read_flushed_lsn_and_arch_log_no(
+/*=================================*/
+	os_file_t	data_file,		/* in: open data file */
+	ibool		one_read_already,	/* in: TRUE if min and max
+						parameters below already
+						contain sensible data */
+#ifdef UNIV_LOG_ARCHIVE
+	ulint*		min_arch_log_no,	/* in/out: */
+	ulint*		max_arch_log_no,	/* in/out: */
+#endif /* UNIV_LOG_ARCHIVE */
+	ib_uint64_t*	min_flushed_lsn,	/* in/out: */
+	ib_uint64_t*	max_flushed_lsn)	/* in/out: */
+{
+	byte*		buf;
+	byte*		buf2;
+	ib_uint64_t	flushed_lsn;
+
+	buf2 = ut_malloc(2 * UNIV_PAGE_SIZE);
+	/* Align the memory for a possible read from a raw device */
+	buf = ut_align(buf2, UNIV_PAGE_SIZE);
+
+	os_file_read(data_file, buf, 0, 0, UNIV_PAGE_SIZE);
+
+	flushed_lsn = mach_read_ull(buf + FIL_PAGE_FILE_FLUSH_LSN);
+
+	ut_free(buf2);
+
+	if (!one_read_already) {
+		*min_flushed_lsn = flushed_lsn;
+		*max_flushed_lsn = flushed_lsn;
+#ifdef UNIV_LOG_ARCHIVE
+		*min_arch_log_no = arch_log_no;
+		*max_arch_log_no = arch_log_no;
+#endif /* UNIV_LOG_ARCHIVE */
+		return;
+	}
+
+	if (*min_flushed_lsn > flushed_lsn) {
+		*min_flushed_lsn = flushed_lsn;
+	}
+	if (*max_flushed_lsn < flushed_lsn) {
+		*max_flushed_lsn = flushed_lsn;
+	}
+#ifdef UNIV_LOG_ARCHIVE
+	if (*min_arch_log_no > arch_log_no) {
+		*min_arch_log_no = arch_log_no;
+	}
+	if (*max_arch_log_no < arch_log_no) {
+		*max_arch_log_no = arch_log_no;
+	}
+#endif /* UNIV_LOG_ARCHIVE */
+}
+
+/*================ SINGLE-TABLE TABLESPACES ==========================*/
+
+/***********************************************************************
+Increments the count of pending insert buffer page merges, if space is not
+being deleted. */
+UNIV_INTERN
+ibool
+fil_inc_pending_ibuf_merges(
+/*========================*/
+			/* out: TRUE if being deleted, and ibuf merges should
+			be skipped */
+	ulint	id)	/* in: space id */
+{
+	fil_system_t*	system		= fil_system;
+	fil_space_t*	space;
+
+	mutex_enter(&(system->mutex));
+
+	space = fil_space_get_by_id(id);
+
+	if (space == NULL) {
+		fprintf(stderr,
+			"InnoDB: Error: trying to do ibuf merge to a"
+			" dropped tablespace %lu\n",
+			(ulong) id);
+	}
+
+	if (space == NULL || space->stop_ibuf_merges) {
+		mutex_exit(&(system->mutex));
+
+		return(TRUE);
+	}
+
+	space->n_pending_ibuf_merges++;
+
+	mutex_exit(&(system->mutex));
+
+	return(FALSE);
+}
+
+/***********************************************************************
+Decrements the count of pending insert buffer page merges. */
+UNIV_INTERN
+void
+fil_decr_pending_ibuf_merges(
+/*=========================*/
+	ulint	id)	/* in: space id */
+{
+	fil_system_t*	system		= fil_system;
+	fil_space_t*	space;
+
+	mutex_enter(&(system->mutex));
+
+	space = fil_space_get_by_id(id);
+
+	if (space == NULL) {
+		fprintf(stderr,
+			"InnoDB: Error: decrementing ibuf merge of a"
+			" dropped tablespace %lu\n",
+			(ulong) id);
+	}
+
+	if (space != NULL) {
+		space->n_pending_ibuf_merges--;
+	}
+
+	mutex_exit(&(system->mutex));
+}
+
+/************************************************************
+Creates the database directory for a table if it does not exist yet. */
+static
+void
+fil_create_directory_for_tablename(
+/*===============================*/
+	const char*	name)	/* in: name in the standard
+				'databasename/tablename' format */
+{
+	const char*	namend;
+	char*		path;
+	ulint		len;
+
+	len = strlen(fil_path_to_mysql_datadir);
+	namend = strchr(name, '/');
+	ut_a(namend);
+	path = mem_alloc(len + (namend - name) + 2);
+
+	memcpy(path, fil_path_to_mysql_datadir, len);
+	path[len] = '/';
+	memcpy(path + len + 1, name, namend - name);
+	path[len + (namend - name) + 1] = 0;
+
+	srv_normalize_path_for_win(path);
+
+	ut_a(os_file_create_directory(path, FALSE));
+	mem_free(path);
+}
+
+#ifndef UNIV_HOTBACKUP
+/************************************************************
+Writes a log record about an .ibd file create/rename/delete. */
+static
+void
+fil_op_write_log(
+/*=============*/
+	ulint		type,		/* in: MLOG_FILE_CREATE,
+					MLOG_FILE_CREATE2,
+					MLOG_FILE_DELETE, or
+					MLOG_FILE_RENAME */
+	ulint		space_id,	/* in: space id */
+	ulint		flags,		/* in: compressed page size
+					and file format
+					if type==MLOG_FILE_CREATE2, or 0 */
+	const char*	name,		/* in: table name in the familiar
+					'databasename/tablename' format, or
+					the file path in the case of
+					MLOG_FILE_DELETE */
+	const char*	new_name,	/* in: if type is MLOG_FILE_RENAME,
+					the new table name in the
+					'databasename/tablename' format */
+	mtr_t*		mtr)		/* in: mini-transaction handle */
+{
+	byte*	log_ptr;
+	ulint	len;
+
+	log_ptr = mlog_open(mtr, 11 + 2 + 1);
+
+	if (!log_ptr) {
+		/* Logging in mtr is switched off during crash recovery:
+		in that case mlog_open returns NULL */
+		return;
+	}
+
+	log_ptr = mlog_write_initial_log_record_for_file_op(type, space_id, 0,
+							    log_ptr, mtr);
+	if (type == MLOG_FILE_CREATE2) {
+		mach_write_to_4(log_ptr, flags);
+		log_ptr += 4;
+	}
+	/* Let us store the strings as null-terminated for easier readability
+	and handling */
+
+	len = strlen(name) + 1;
+
+	mach_write_to_2(log_ptr, len);
+	log_ptr += 2;
+	mlog_close(mtr, log_ptr);
+
+	mlog_catenate_string(mtr, (byte*) name, len);
+
+	if (type == MLOG_FILE_RENAME) {
+		len = strlen(new_name) + 1;
+		log_ptr = mlog_open(mtr, 2 + len);
+		ut_a(log_ptr);
+		mach_write_to_2(log_ptr, len);
+		log_ptr += 2;
+		mlog_close(mtr, log_ptr);
+
+		mlog_catenate_string(mtr, (byte*) new_name, len);
+	}
+}
+#endif
+
+/***********************************************************************
+Parses the body of a log record written about an .ibd file operation. That is,
+the log record part after the standard (type, space id, page no) header of the
+log record.
+
+If desired, also replays the delete or rename operation if the .ibd file
+exists and the space id in it matches. Replays the create operation if a file
+at that path does not exist yet. If the database directory for the file to be
+created does not exist, then we create the directory, too.
+
+Note that ibbackup --apply-log sets fil_path_to_mysql_datadir to point to the
+datadir that we should use in replaying the file operations. */
+UNIV_INTERN
+byte*
+fil_op_log_parse_or_replay(
+/*=======================*/
+				/* out: end of log record, or NULL if the
+				record was not completely contained between
+				ptr and end_ptr */
+	byte*	ptr,		/* in: buffer containing the log record body,
+				or an initial segment of it, if the record does
+				not fir completely between ptr and end_ptr */
+	byte*	end_ptr,	/* in: buffer end */
+	ulint	type,		/* in: the type of this log record */
+	ulint	space_id)	/* in: the space id of the tablespace in
+				question, or 0 if the log record should
+				only be parsed but not replayed */
+{
+	ulint		name_len;
+	ulint		new_name_len;
+	const char*	name;
+	const char*	new_name	= NULL;
+	ulint		flags		= 0;
+
+	if (type == MLOG_FILE_CREATE2) {
+		if (end_ptr < ptr + 4) {
+
+			return(NULL);
+		}
+
+		flags = mach_read_from_4(ptr);
+		ptr += 4;
+	}
+
+	if (end_ptr < ptr + 2) {
+
+		return(NULL);
+	}
+
+	name_len = mach_read_from_2(ptr);
+
+	ptr += 2;
+
+	if (end_ptr < ptr + name_len) {
+
+		return(NULL);
+	}
+
+	name = (const char*) ptr;
+
+	ptr += name_len;
+
+	if (type == MLOG_FILE_RENAME) {
+		if (end_ptr < ptr + 2) {
+
+			return(NULL);
+		}
+
+		new_name_len = mach_read_from_2(ptr);
+
+		ptr += 2;
+
+		if (end_ptr < ptr + new_name_len) {
+
+			return(NULL);
+		}
+
+		new_name = (const char*) ptr;
+
+		ptr += new_name_len;
+	}
+
+	/* We managed to parse a full log record body */
+	/*
+	printf("Parsed log rec of type %lu space %lu\n"
+	"name %s\n", type, space_id, name);
+
+	if (type == MLOG_FILE_RENAME) {
+	printf("new name %s\n", new_name);
+	}
+	*/
+	if (!space_id) {
+
+		return(ptr);
+	}
+
+	/* Let us try to perform the file operation, if sensible. Note that
+	ibbackup has at this stage already read in all space id info to the
+	fil0fil.c data structures.
+
+	NOTE that our algorithm is not guaranteed to work correctly if there
+	were renames of tables during the backup. See ibbackup code for more
+	on the problem. */
+
+	switch (type) {
+	case MLOG_FILE_DELETE:
+		if (fil_tablespace_exists_in_mem(space_id)) {
+			ut_a(fil_delete_tablespace(space_id));
+		}
+
+		break;
+
+	case MLOG_FILE_RENAME:
+		/* We do the rename based on space id, not old file name;
+		this should guarantee that after the log replay each .ibd file
+		has the correct name for the latest log sequence number; the
+		proof is left as an exercise :) */
+
+		if (fil_tablespace_exists_in_mem(space_id)) {
+			/* Create the database directory for the new name, if
+			it does not exist yet */
+			fil_create_directory_for_tablename(new_name);
+
+			/* Rename the table if there is not yet a tablespace
+			with the same name */
+
+			if (fil_get_space_id_for_table(new_name)
+			    == ULINT_UNDEFINED) {
+				/* We do not care of the old name, that is
+				why we pass NULL as the first argument */
+				if (!fil_rename_tablespace(NULL, space_id,
+							   new_name)) {
+					ut_error;
+				}
+			}
+		}
+
+		break;
+
+	case MLOG_FILE_CREATE:
+	case MLOG_FILE_CREATE2:
+		if (fil_tablespace_exists_in_mem(space_id)) {
+			/* Do nothing */
+		} else if (fil_get_space_id_for_table(name)
+			   != ULINT_UNDEFINED) {
+			/* Do nothing */
+		} else {
+			/* Create the database directory for name, if it does
+			not exist yet */
+			fil_create_directory_for_tablename(name);
+
+			if (fil_create_new_single_table_tablespace(
+				    &space_id, name, FALSE, flags,
+				    FIL_IBD_FILE_INITIAL_SIZE) != DB_SUCCESS) {
+				ut_error;
+			}
+		}
+
+		break;
+
+	default:
+		ut_error;
+	}
+
+	return(ptr);
+}
+
+/***********************************************************************
+Deletes a single-table tablespace. The tablespace must be cached in the
+memory cache. */
+UNIV_INTERN
+ibool
+fil_delete_tablespace(
+/*==================*/
+			/* out: TRUE if success */
+	ulint	id)	/* in: space id */
+{
+	fil_system_t*	system		= fil_system;
+	ibool		success;
+	fil_space_t*	space;
+	fil_node_t*	node;
+	ulint		count		= 0;
+	char*		path;
+
+	ut_a(id != 0);
+stop_ibuf_merges:
+	mutex_enter(&(system->mutex));
+
+	space = fil_space_get_by_id(id);
+
+	if (space != NULL) {
+		space->stop_ibuf_merges = TRUE;
+
+		if (space->n_pending_ibuf_merges == 0) {
+			mutex_exit(&(system->mutex));
+
+			count = 0;
+
+			goto try_again;
+		} else {
+			if (count > 5000) {
+				ut_print_timestamp(stderr);
+				fputs("  InnoDB: Warning: trying to"
+				      " delete tablespace ", stderr);
+				ut_print_filename(stderr, space->name);
+				fprintf(stderr, ",\n"
+					"InnoDB: but there are %lu pending"
+					" ibuf merges on it.\n"
+					"InnoDB: Loop %lu.\n",
+					(ulong) space->n_pending_ibuf_merges,
+					(ulong) count);
+			}
+
+			mutex_exit(&(system->mutex));
+
+			os_thread_sleep(20000);
+			count++;
+
+			goto stop_ibuf_merges;
+		}
+	}
+
+	mutex_exit(&(system->mutex));
+	count = 0;
+
+try_again:
+	mutex_enter(&(system->mutex));
+
+	space = fil_space_get_by_id(id);
+
+	if (space == NULL) {
+		ut_print_timestamp(stderr);
+		fprintf(stderr,
+			"  InnoDB: Error: cannot delete tablespace %lu\n"
+			"InnoDB: because it is not found in the"
+			" tablespace memory cache.\n",
+			(ulong) id);
+
+		mutex_exit(&(system->mutex));
+
+		return(FALSE);
+	}
+
+	ut_a(space);
+	ut_a(space->n_pending_ibuf_merges == 0);
+
+	space->is_being_deleted = TRUE;
+
+	ut_a(UT_LIST_GET_LEN(space->chain) == 1);
+	node = UT_LIST_GET_FIRST(space->chain);
+
+	if (space->n_pending_flushes > 0 || node->n_pending > 0) {
+		if (count > 1000) {
+			ut_print_timestamp(stderr);
+			fputs("  InnoDB: Warning: trying to"
+			      " delete tablespace ", stderr);
+			ut_print_filename(stderr, space->name);
+			fprintf(stderr, ",\n"
+				"InnoDB: but there are %lu flushes"
+				" and %lu pending i/o's on it\n"
+				"InnoDB: Loop %lu.\n",
+				(ulong) space->n_pending_flushes,
+				(ulong) node->n_pending,
+				(ulong) count);
+		}
+		mutex_exit(&(system->mutex));
+		os_thread_sleep(20000);
+
+		count++;
+
+		goto try_again;
+	}
+
+	path = mem_strdup(space->name);
+
+	mutex_exit(&(system->mutex));
+#ifndef UNIV_HOTBACKUP
+	/* Invalidate in the buffer pool all pages belonging to the
+	tablespace. Since we have set space->is_being_deleted = TRUE, readahead
+	or ibuf merge can no longer read more pages of this tablespace to the
+	buffer pool. Thus we can clean the tablespace out of the buffer pool
+	completely and permanently. The flag is_being_deleted also prevents
+	fil_flush() from being applied to this tablespace. */
+
+	buf_LRU_invalidate_tablespace(id);
+#endif
+	/* printf("Deleting tablespace %s id %lu\n", space->name, id); */
+
+	success = fil_space_free(id);
+
+	if (success) {
+		success = os_file_delete(path);
+
+		if (!success) {
+			success = os_file_delete_if_exists(path);
+		}
+	}
+
+	if (success) {
+#ifndef UNIV_HOTBACKUP
+		/* Write a log record about the deletion of the .ibd
+		file, so that ibbackup can replay it in the
+		--apply-log phase. We use a dummy mtr and the familiar
+		log write mechanism. */
+		mtr_t		mtr;
+
+		/* When replaying the operation in ibbackup, do not try
+		to write any log record */
+		mtr_start(&mtr);
+
+		fil_op_write_log(MLOG_FILE_DELETE, id, 0, path, NULL, &mtr);
+		mtr_commit(&mtr);
+#endif
+		mem_free(path);
+
+		return(TRUE);
+	}
+
+	mem_free(path);
+
+	return(FALSE);
+}
+
+/***********************************************************************
+Discards a single-table tablespace. The tablespace must be cached in the
+memory cache. Discarding is like deleting a tablespace, but
+1) we do not drop the table from the data dictionary;
+2) we remove all insert buffer entries for the tablespace immediately; in DROP
+TABLE they are only removed gradually in the background;
+3) when the user does IMPORT TABLESPACE, the tablespace will have the same id
+as it originally had. */
+UNIV_INTERN
+ibool
+fil_discard_tablespace(
+/*===================*/
+			/* out: TRUE if success */
+	ulint	id)	/* in: space id */
+{
+	ibool	success;
+
+	success = fil_delete_tablespace(id);
+
+	if (!success) {
+		fprintf(stderr,
+			"InnoDB: Warning: cannot delete tablespace %lu"
+			" in DISCARD TABLESPACE.\n"
+			"InnoDB: But let us remove the"
+			" insert buffer entries for this tablespace.\n",
+			(ulong) id);
+	}
+
+	/* Remove all insert buffer entries for the tablespace */
+
+	ibuf_delete_for_discarded_space(id);
+
+	return(success);
+}
+
+/***********************************************************************
+Renames the memory cache structures of a single-table tablespace. */
+static
+ibool
+fil_rename_tablespace_in_mem(
+/*=========================*/
+				/* out: TRUE if success */
+	fil_space_t*	space,	/* in: tablespace memory object */
+	fil_node_t*	node,	/* in: file node of that tablespace */
+	const char*	path)	/* in: new name */
+{
+	fil_system_t*	system		= fil_system;
+	fil_space_t*	space2;
+	const char*	old_name	= space->name;
+
+	space2 = fil_space_get_by_name(old_name);
+	if (space != space2) {
+		fputs("InnoDB: Error: cannot find ", stderr);
+		ut_print_filename(stderr, old_name);
+		fputs(" in tablespace memory cache\n", stderr);
+
+		return(FALSE);
+	}
+
+	space2 = fil_space_get_by_name(path);
+	if (space2 != NULL) {
+		fputs("InnoDB: Error: ", stderr);
+		ut_print_filename(stderr, path);
+		fputs(" is already in tablespace memory cache\n", stderr);
+
+		return(FALSE);
+	}
+
+	HASH_DELETE(fil_space_t, name_hash, system->name_hash,
+		    ut_fold_string(space->name), space);
+	mem_free(space->name);
+	mem_free(node->name);
+
+	space->name = mem_strdup(path);
+	node->name = mem_strdup(path);
+
+	HASH_INSERT(fil_space_t, name_hash, system->name_hash,
+		    ut_fold_string(path), space);
+	return(TRUE);
+}
+
+/***********************************************************************
+Allocates a file name for a single-table tablespace. The string must be freed
+by caller with mem_free(). */
+static
+char*
+fil_make_ibd_name(
+/*==============*/
+					/* out, own: file name */
+	const char*	name,		/* in: table name or a dir path of a
+					TEMPORARY table */
+	ibool		is_temp)	/* in: TRUE if it is a dir path */
+{
+	ulint	namelen		= strlen(name);
+	ulint	dirlen		= strlen(fil_path_to_mysql_datadir);
+	char*	filename	= mem_alloc(namelen + dirlen + sizeof "/.ibd");
+
+	if (is_temp) {
+		memcpy(filename, name, namelen);
+		memcpy(filename + namelen, ".ibd", sizeof ".ibd");
+	} else {
+		memcpy(filename, fil_path_to_mysql_datadir, dirlen);
+		filename[dirlen] = '/';
+
+		memcpy(filename + dirlen + 1, name, namelen);
+		memcpy(filename + dirlen + namelen + 1, ".ibd", sizeof ".ibd");
+	}
+
+	srv_normalize_path_for_win(filename);
+
+	return(filename);
+}
+
+/***********************************************************************
+Renames a single-table tablespace. The tablespace must be cached in the
+tablespace memory cache. */
+UNIV_INTERN
+ibool
+fil_rename_tablespace(
+/*==================*/
+					/* out: TRUE if success */
+	const char*	old_name,	/* in: old table name in the standard
+					databasename/tablename format of
+					InnoDB, or NULL if we do the rename
+					based on the space id only */
+	ulint		id,		/* in: space id */
+	const char*	new_name)	/* in: new table name in the standard
+					databasename/tablename format
+					of InnoDB */
+{
+	fil_system_t*	system		= fil_system;
+	ibool		success;
+	fil_space_t*	space;
+	fil_node_t*	node;
+	ulint		count		= 0;
+	char*		path;
+	ibool		old_name_was_specified		= TRUE;
+	char*		old_path;
+
+	ut_a(id != 0);
+
+	if (old_name == NULL) {
+		old_name = "(name not specified)";
+		old_name_was_specified = FALSE;
+	}
+retry:
+	count++;
+
+	if (count > 1000) {
+		ut_print_timestamp(stderr);
+		fputs("  InnoDB: Warning: problems renaming ", stderr);
+		ut_print_filename(stderr, old_name);
+		fputs(" to ", stderr);
+		ut_print_filename(stderr, new_name);
+		fprintf(stderr, ", %lu iterations\n", (ulong) count);
+	}
+
+	mutex_enter(&(system->mutex));
+
+	space = fil_space_get_by_id(id);
+
+	if (space == NULL) {
+		fprintf(stderr,
+			"InnoDB: Error: cannot find space id %lu"
+			" in the tablespace memory cache\n"
+			"InnoDB: though the table ", (ulong) id);
+		ut_print_filename(stderr, old_name);
+		fputs(" in a rename operation should have that id\n", stderr);
+		mutex_exit(&(system->mutex));
+
+		return(FALSE);
+	}
+
+	if (count > 25000) {
+		space->stop_ios = FALSE;
+		mutex_exit(&(system->mutex));
+
+		return(FALSE);
+	}
+
+	/* We temporarily close the .ibd file because we do not trust that
+	operating systems can rename an open file. For the closing we have to
+	wait until there are no pending i/o's or flushes on the file. */
+
+	space->stop_ios = TRUE;
+
+	ut_a(UT_LIST_GET_LEN(space->chain) == 1);
+	node = UT_LIST_GET_FIRST(space->chain);
+
+	if (node->n_pending > 0 || node->n_pending_flushes > 0) {
+		/* There are pending i/o's or flushes, sleep for a while and
+		retry */
+
+		mutex_exit(&(system->mutex));
+
+		os_thread_sleep(20000);
+
+		goto retry;
+
+	} else if (node->modification_counter > node->flush_counter) {
+		/* Flush the space */
+
+		mutex_exit(&(system->mutex));
+
+		os_thread_sleep(20000);
+
+		fil_flush(id);
+
+		goto retry;
+
+	} else if (node->open) {
+		/* Close the file */
+
+		fil_node_close_file(node, system);
+	}
+
+	/* Check that the old name in the space is right */
+
+	if (old_name_was_specified) {
+		old_path = fil_make_ibd_name(old_name, FALSE);
+
+		ut_a(strcmp(space->name, old_path) == 0);
+		ut_a(strcmp(node->name, old_path) == 0);
+	} else {
+		old_path = mem_strdup(space->name);
+	}
+
+	/* Rename the tablespace and the node in the memory cache */
+	path = fil_make_ibd_name(new_name, FALSE);
+	success = fil_rename_tablespace_in_mem(space, node, path);
+
+	if (success) {
+		success = os_file_rename(old_path, path);
+
+		if (!success) {
+			/* We have to revert the changes we made
+			to the tablespace memory cache */
+
+			ut_a(fil_rename_tablespace_in_mem(space, node,
+							  old_path));
+		}
+	}
+
+	mem_free(path);
+	mem_free(old_path);
+
+	space->stop_ios = FALSE;
+
+	mutex_exit(&(system->mutex));
+
+#ifndef UNIV_HOTBACKUP
+	if (success) {
+		mtr_t		mtr;
+
+		mtr_start(&mtr);
+
+		fil_op_write_log(MLOG_FILE_RENAME, id, 0, old_name, new_name,
+				 &mtr);
+		mtr_commit(&mtr);
+	}
+#endif
+	return(success);
+}
+
+/***********************************************************************
+Creates a new single-table tablespace to a database directory of MySQL.
+Database directories are under the 'datadir' of MySQL. The datadir is the
+directory of a running mysqld program. We can refer to it by simply the
+path '.'. Tables created with CREATE TEMPORARY TABLE we place in the temp
+dir of the mysqld server. */
+UNIV_INTERN
+ulint
+fil_create_new_single_table_tablespace(
+/*===================================*/
+					/* out: DB_SUCCESS or error code */
+	ulint*		space_id,	/* in/out: space id; if this is != 0,
+					then this is an input parameter,
+					otherwise output */
+	const char*	tablename,	/* in: the table name in the usual
+					databasename/tablename format
+					of InnoDB, or a dir path to a temp
+					table */
+	ibool		is_temp,	/* in: TRUE if a table created with
+					CREATE TEMPORARY TABLE */
+	ulint		flags,		/* in: tablespace flags */
+	ulint		size)		/* in: the initial size of the
+					tablespace file in pages,
+					must be >= FIL_IBD_FILE_INITIAL_SIZE */
+{
+	os_file_t	file;
+	ibool		ret;
+	ulint		err;
+	byte*		buf2;
+	byte*		page;
+	ibool		success;
+	char*		path;
+
+	ut_a(size >= FIL_IBD_FILE_INITIAL_SIZE);
+	/* The tablespace flags (FSP_SPACE_FLAGS) should be 0 for
+	ROW_FORMAT=COMPACT (table->flags == DICT_TF_COMPACT) and
+	ROW_FORMAT=REDUNDANT (table->flags == 0).  For any other
+	format, the tablespace flags should equal table->flags. */
+	ut_a(flags != DICT_TF_COMPACT);
+
+	path = fil_make_ibd_name(tablename, is_temp);
+
+	file = os_file_create(path, OS_FILE_CREATE, OS_FILE_NORMAL,
+			      OS_DATA_FILE, &ret);
+	if (ret == FALSE) {
+		ut_print_timestamp(stderr);
+		fputs("  InnoDB: Error creating file ", stderr);
+		ut_print_filename(stderr, path);
+		fputs(".\n", stderr);
+
+		/* The following call will print an error message */
+
+		err = os_file_get_last_error(TRUE);
+
+		if (err == OS_FILE_ALREADY_EXISTS) {
+			fputs("InnoDB: The file already exists though"
+			      " the corresponding table did not\n"
+			      "InnoDB: exist in the InnoDB data dictionary."
+			      " Have you moved InnoDB\n"
+			      "InnoDB: .ibd files around without using the"
+			      " SQL commands\n"
+			      "InnoDB: DISCARD TABLESPACE and"
+			      " IMPORT TABLESPACE, or did\n"
+			      "InnoDB: mysqld crash in the middle of"
+			      " CREATE TABLE? You can\n"
+			      "InnoDB: resolve the problem by"
+			      " removing the file ", stderr);
+			ut_print_filename(stderr, path);
+			fputs("\n"
+			      "InnoDB: under the 'datadir' of MySQL.\n",
+			      stderr);
+
+			mem_free(path);
+			return(DB_TABLESPACE_ALREADY_EXISTS);
+		}
+
+		if (err == OS_FILE_DISK_FULL) {
+
+			mem_free(path);
+			return(DB_OUT_OF_FILE_SPACE);
+		}
+
+		mem_free(path);
+		return(DB_ERROR);
+	}
+
+	buf2 = ut_malloc(3 * UNIV_PAGE_SIZE);
+	/* Align the memory for file i/o if we might have O_DIRECT set */
+	page = ut_align(buf2, UNIV_PAGE_SIZE);
+
+	ret = os_file_set_size(path, file, size * UNIV_PAGE_SIZE, 0);
+
+	if (!ret) {
+		ut_free(buf2);
+		os_file_close(file);
+		os_file_delete(path);
+
+		mem_free(path);
+		return(DB_OUT_OF_FILE_SPACE);
+	}
+
+	if (*space_id == 0) {
+		*space_id = fil_assign_new_space_id();
+	}
+
+	/* printf("Creating tablespace %s id %lu\n", path, *space_id); */
+
+	if (*space_id == ULINT_UNDEFINED) {
+		ut_free(buf2);
+error_exit:
+		os_file_close(file);
+error_exit2:
+		os_file_delete(path);
+
+		mem_free(path);
+		return(DB_ERROR);
+	}
+
+	/* We have to write the space id to the file immediately and flush the
+	file to disk. This is because in crash recovery we must be aware what
+	tablespaces exist and what are their space id's, so that we can apply
+	the log records to the right file. It may take quite a while until
+	buffer pool flush algorithms write anything to the file and flush it to
+	disk. If we would not write here anything, the file would be filled
+	with zeros from the call of os_file_set_size(), until a buffer pool
+	flush would write to it. */
+
+	memset(page, '\0', UNIV_PAGE_SIZE);
+
+	fsp_header_init_fields(page, *space_id, flags);
+	mach_write_to_4(page + FIL_PAGE_ARCH_LOG_NO_OR_SPACE_ID, *space_id);
+
+	if (!(flags & DICT_TF_ZSSIZE_MASK)) {
+		buf_flush_init_for_writing(page, NULL, 0);
+		ret = os_file_write(path, file, page, 0, 0, UNIV_PAGE_SIZE);
+	} else {
+		page_zip_des_t	page_zip;
+		ulint		zip_size;
+
+		zip_size = ((PAGE_ZIP_MIN_SIZE >> 1)
+			    << ((flags & DICT_TF_ZSSIZE_MASK)
+				>> DICT_TF_ZSSIZE_SHIFT));
+
+		page_zip_set_size(&page_zip, zip_size);
+		page_zip.data = page + UNIV_PAGE_SIZE;
+#ifdef UNIV_DEBUG
+		page_zip.m_start =
+#endif /* UNIV_DEBUG */
+			page_zip.m_end = page_zip.m_nonempty =
+			page_zip.n_blobs = 0;
+		buf_flush_init_for_writing(page, &page_zip, 0);
+		ret = os_file_write(path, file, page_zip.data, 0, 0, zip_size);
+	}
+
+	ut_free(buf2);
+
+	if (!ret) {
+		fputs("InnoDB: Error: could not write the first page"
+		      " to tablespace ", stderr);
+		ut_print_filename(stderr, path);
+		putc('\n', stderr);
+		goto error_exit;
+	}
+
+	ret = os_file_flush(file);
+
+	if (!ret) {
+		fputs("InnoDB: Error: file flush of tablespace ", stderr);
+		ut_print_filename(stderr, path);
+		fputs(" failed\n", stderr);
+		goto error_exit;
+	}
+
+	os_file_close(file);
+
+	if (*space_id == ULINT_UNDEFINED) {
+		goto error_exit2;
+	}
+
+	success = fil_space_create(path, *space_id, flags, FIL_TABLESPACE);
+
+	if (!success) {
+		goto error_exit2;
+	}
+
+	fil_node_create(path, size, *space_id, FALSE);
+
+#ifndef UNIV_HOTBACKUP
+	{
+		mtr_t		mtr;
+
+		mtr_start(&mtr);
+
+		fil_op_write_log(flags
+				 ? MLOG_FILE_CREATE2
+				 : MLOG_FILE_CREATE,
+				 *space_id, flags,
+				 tablename, NULL, &mtr);
+
+		mtr_commit(&mtr);
+	}
+#endif
+	mem_free(path);
+	return(DB_SUCCESS);
+}
+
+/************************************************************************
+It is possible, though very improbable, that the lsn's in the tablespace to be
+imported have risen above the current system lsn, if a lengthy purge, ibuf
+merge, or rollback was performed on a backup taken with ibbackup. If that is
+the case, reset page lsn's in the file. We assume that mysqld was shut down
+after it performed these cleanup operations on the .ibd file, so that it at
+the shutdown stamped the latest lsn to the FIL_PAGE_FILE_FLUSH_LSN in the
+first page of the .ibd file, and we can determine whether we need to reset the
+lsn's just by looking at that flush lsn. */
+UNIV_INTERN
+ibool
+fil_reset_too_high_lsns(
+/*====================*/
+					/* out: TRUE if success */
+	const char*	name,		/* in: table name in the
+					databasename/tablename format */
+	ib_uint64_t	current_lsn)	/* in: reset lsn's if the lsn stamped
+					to FIL_PAGE_FILE_FLUSH_LSN in the
+					first page is too high */
+{
+	os_file_t	file;
+	char*		filepath;
+	byte*		page;
+	byte*		buf2;
+	ib_uint64_t	flush_lsn;
+	ulint		space_id;
+	ib_int64_t	file_size;
+	ib_int64_t	offset;
+	ulint		zip_size;
+	ibool		success;
+
+	filepath = fil_make_ibd_name(name, FALSE);
+
+	file = os_file_create_simple_no_error_handling(
+		filepath, OS_FILE_OPEN, OS_FILE_READ_WRITE, &success);
+	if (!success) {
+		/* The following call prints an error message */
+		os_file_get_last_error(TRUE);
+
+		ut_print_timestamp(stderr);
+
+		fputs("  InnoDB: Error: trying to open a table,"
+		      " but could not\n"
+		      "InnoDB: open the tablespace file ", stderr);
+		ut_print_filename(stderr, filepath);
+		fputs("!\n", stderr);
+		mem_free(filepath);
+
+		return(FALSE);
+	}
+
+	/* Read the first page of the tablespace */
+
+	buf2 = ut_malloc(3 * UNIV_PAGE_SIZE);
+	/* Align the memory for file i/o if we might have O_DIRECT set */
+	page = ut_align(buf2, UNIV_PAGE_SIZE);
+
+	success = os_file_read(file, page, 0, 0, UNIV_PAGE_SIZE);
+	if (!success) {
+
+		goto func_exit;
+	}
+
+	/* We have to read the file flush lsn from the header of the file */
+
+	flush_lsn = mach_read_ull(page + FIL_PAGE_FILE_FLUSH_LSN);
+
+	if (current_lsn >= flush_lsn) {
+		/* Ok */
+		success = TRUE;
+
+		goto func_exit;
+	}
+
+	space_id = fsp_header_get_space_id(page);
+	zip_size = fsp_header_get_zip_size(page);
+
+	ut_print_timestamp(stderr);
+	fprintf(stderr,
+		"  InnoDB: Flush lsn in the tablespace file %lu"
+		" to be imported\n"
+		"InnoDB: is %llu, which exceeds current"
+		" system lsn %llu.\n"
+		"InnoDB: We reset the lsn's in the file ",
+		(ulong) space_id,
+		flush_lsn, current_lsn);
+	ut_print_filename(stderr, filepath);
+	fputs(".\n", stderr);
+
+	ut_a(ut_is_2pow(zip_size));
+	ut_a(zip_size <= UNIV_PAGE_SIZE);
+
+	/* Loop through all the pages in the tablespace and reset the lsn and
+	the page checksum if necessary */
+
+	file_size = os_file_get_size_as_iblonglong(file);
+
+	for (offset = 0; offset < file_size;
+	     offset += zip_size ? zip_size : UNIV_PAGE_SIZE) {
+		success = os_file_read(file, page,
+				       (ulint)(offset & 0xFFFFFFFFUL),
+				       (ulint)(offset >> 32),
+				       zip_size ? zip_size : UNIV_PAGE_SIZE);
+		if (!success) {
+
+			goto func_exit;
+		}
+		if (mach_read_ull(page + FIL_PAGE_LSN) > current_lsn) {
+			/* We have to reset the lsn */
+
+			if (zip_size) {
+				memcpy(page + UNIV_PAGE_SIZE, page, zip_size);
+				buf_flush_init_for_writing(
+					page, page + UNIV_PAGE_SIZE,
+					current_lsn);
+			} else {
+				buf_flush_init_for_writing(
+					page, NULL, current_lsn);
+			}
+			success = os_file_write(filepath, file, page,
+						(ulint)(offset & 0xFFFFFFFFUL),
+						(ulint)(offset >> 32),
+						zip_size
+						? zip_size
+						: UNIV_PAGE_SIZE);
+			if (!success) {
+
+				goto func_exit;
+			}
+		}
+	}
+
+	success = os_file_flush(file);
+	if (!success) {
+
+		goto func_exit;
+	}
+
+	/* We now update the flush_lsn stamp at the start of the file */
+	success = os_file_read(file, page, 0, 0,
+			       zip_size ? zip_size : UNIV_PAGE_SIZE);
+	if (!success) {
+
+		goto func_exit;
+	}
+
+	mach_write_ull(page + FIL_PAGE_FILE_FLUSH_LSN, current_lsn);
+
+	success = os_file_write(filepath, file, page, 0, 0,
+				zip_size ? zip_size : UNIV_PAGE_SIZE);
+	if (!success) {
+
+		goto func_exit;
+	}
+	success = os_file_flush(file);
+func_exit:
+	os_file_close(file);
+	ut_free(buf2);
+	mem_free(filepath);
+
+	return(success);
+}
+
+/************************************************************************
+Tries to open a single-table tablespace and optionally checks the space id is
+right in it. If does not succeed, prints an error message to the .err log. This
+function is used to open a tablespace when we start up mysqld, and also in
+IMPORT TABLESPACE.
+NOTE that we assume this operation is used either at the database startup
+or under the protection of the dictionary mutex, so that two users cannot
+race here. This operation does not leave the file associated with the
+tablespace open, but closes it after we have looked at the space id in it. */
+UNIV_INTERN
+ibool
+fil_open_single_table_tablespace(
+/*=============================*/
+					/* out: TRUE if success */
+	ibool		check_space_id,	/* in: should we check that the space
+					id in the file is right; we assume
+					that this function runs much faster
+					if no check is made, since accessing
+					the file inode probably is much
+					faster (the OS caches them) than
+					accessing the first page of the file */
+	ulint		id,		/* in: space id */
+	ulint		flags,		/* in: tablespace flags */
+	const char*	name)		/* in: table name in the
+					databasename/tablename format */
+{
+	os_file_t	file;
+	char*		filepath;
+	ibool		success;
+	byte*		buf2;
+	byte*		page;
+	ulint		space_id;
+	ulint		space_flags;
+	ibool		ret		= TRUE;
+
+	filepath = fil_make_ibd_name(name, FALSE);
+
+	/* The tablespace flags (FSP_SPACE_FLAGS) should be 0 for
+	ROW_FORMAT=COMPACT (table->flags == DICT_TF_COMPACT) and
+	ROW_FORMAT=REDUNDANT (table->flags == 0).  For any other
+	format, the tablespace flags should equal table->flags. */
+	ut_a(flags != DICT_TF_COMPACT);
+
+	file = os_file_create_simple_no_error_handling(
+		filepath, OS_FILE_OPEN, OS_FILE_READ_ONLY, &success);
+	if (!success) {
+		/* The following call prints an error message */
+		os_file_get_last_error(TRUE);
+
+		ut_print_timestamp(stderr);
+
+		fputs("  InnoDB: Error: trying to open a table,"
+		      " but could not\n"
+		      "InnoDB: open the tablespace file ", stderr);
+		ut_print_filename(stderr, filepath);
+		fputs("!\n"
+		      "InnoDB: Have you moved InnoDB .ibd files around"
+		      " without using the\n"
+		      "InnoDB: commands DISCARD TABLESPACE and"
+		      " IMPORT TABLESPACE?\n"
+		      "InnoDB: It is also possible that this is"
+		      " a temporary table #sql...,\n"
+		      "InnoDB: and MySQL removed the .ibd file for this.\n"
+		      "InnoDB: Please refer to\n"
+		      "InnoDB: http://dev.mysql.com/doc/refman/5.1/en/"
+		      "innodb-troubleshooting.html\n"
+		      "InnoDB: for how to resolve the issue.\n", stderr);
+
+		mem_free(filepath);
+
+		return(FALSE);
+	}
+
+	if (!check_space_id) {
+		space_id = id;
+
+		goto skip_check;
+	}
+
+	/* Read the first page of the tablespace */
+
+	buf2 = ut_malloc(2 * UNIV_PAGE_SIZE);
+	/* Align the memory for file i/o if we might have O_DIRECT set */
+	page = ut_align(buf2, UNIV_PAGE_SIZE);
+
+	success = os_file_read(file, page, 0, 0, UNIV_PAGE_SIZE);
+
+	/* We have to read the tablespace id and flags from the file. */
+
+	space_id = fsp_header_get_space_id(page);
+	space_flags = fsp_header_get_flags(page);
+
+	ut_free(buf2);
+
+	if (UNIV_UNLIKELY(space_id != id || space_flags != flags)) {
+		ut_print_timestamp(stderr);
+
+		fputs("  InnoDB: Error: tablespace id and flags in file ",
+		      stderr);
+		ut_print_filename(stderr, filepath);
+		fprintf(stderr, " are %lu and %lu, but in the InnoDB\n"
+			"InnoDB: data dictionary they are %lu and %lu.\n"
+			"InnoDB: Have you moved InnoDB .ibd files"
+			" around without using the\n"
+			"InnoDB: commands DISCARD TABLESPACE and"
+			" IMPORT TABLESPACE?\n"
+			"InnoDB: Please refer to\n"
+			"InnoDB: http://dev.mysql.com/doc/refman/5.1/en/"
+			"innodb-troubleshooting.html\n"
+			"InnoDB: for how to resolve the issue.\n",
+			(ulong) space_id, (ulong) space_flags,
+			(ulong) id, (ulong) flags);
+
+		ret = FALSE;
+
+		goto func_exit;
+	}
+
+skip_check:
+	success = fil_space_create(filepath, space_id, flags, FIL_TABLESPACE);
+
+	if (!success) {
+		goto func_exit;
+	}
+
+	/* We do not measure the size of the file, that is why we pass the 0
+	below */
+
+	fil_node_create(filepath, 0, space_id, FALSE);
+func_exit:
+	os_file_close(file);
+	mem_free(filepath);
+
+	return(ret);
+}
+
+#ifdef UNIV_HOTBACKUP
+/***********************************************************************
+Allocates a file name for an old version of a single-table tablespace.
+The string must be freed by caller with mem_free()! */
+static
+char*
+fil_make_ibbackup_old_name(
+/*=======================*/
+					/* out, own: file name */
+	const char*	name)		/* in: original file name */
+{
+	static const char suffix[] = "_ibbackup_old_vers_";
+	ulint	len	= strlen(name);
+	char*	path	= mem_alloc(len + (15 + sizeof suffix));
+
+	memcpy(path, name, len);
+	memcpy(path + len, suffix, (sizeof suffix) - 1);
+	ut_sprintf_timestamp_without_extra_chars(path + len + sizeof suffix);
+	return(path);
+}
+#endif /* UNIV_HOTBACKUP */
+
+/************************************************************************
+Opens an .ibd file and adds the associated single-table tablespace to the
+InnoDB fil0fil.c data structures. */
+static
+void
+fil_load_single_table_tablespace(
+/*=============================*/
+	const char*	dbname,		/* in: database name */
+	const char*	filename)	/* in: file name (not a path),
+					including the .ibd extension */
+{
+	os_file_t	file;
+	char*		filepath;
+	ibool		success;
+	byte*		buf2;
+	byte*		page;
+	ulint		space_id;
+	ulint		flags;
+	ulint		size_low;
+	ulint		size_high;
+	ib_int64_t	size;
+#ifdef UNIV_HOTBACKUP
+	fil_space_t*	space;
+#endif
+	filepath = mem_alloc(strlen(dbname) + strlen(filename)
+			     + strlen(fil_path_to_mysql_datadir) + 3);
+
+	sprintf(filepath, "%s/%s/%s", fil_path_to_mysql_datadir, dbname,
+		filename);
+	srv_normalize_path_for_win(filepath);
+#ifdef __WIN__
+# ifndef UNIV_HOTBACKUP
+	/* If lower_case_table_names is 0 or 2, then MySQL allows database
+	directory names with upper case letters. On Windows, all table and
+	database names in InnoDB are internally always in lower case. Put the
+	file path to lower case, so that we are consistent with InnoDB's
+	internal data dictionary. */
+
+	dict_casedn_str(filepath);
+# endif /* !UNIV_HOTBACKUP */
+#endif
+	file = os_file_create_simple_no_error_handling(
+		filepath, OS_FILE_OPEN, OS_FILE_READ_ONLY, &success);
+	if (!success) {
+		/* The following call prints an error message */
+		os_file_get_last_error(TRUE);
+
+		fprintf(stderr,
+			"InnoDB: Error: could not open single-table tablespace"
+			" file\n"
+			"InnoDB: %s!\n"
+			"InnoDB: We do not continue the crash recovery,"
+			" because the table may become\n"
+			"InnoDB: corrupt if we cannot apply the log records"
+			" in the InnoDB log to it.\n"
+			"InnoDB: To fix the problem and start mysqld:\n"
+			"InnoDB: 1) If there is a permission problem"
+			" in the file and mysqld cannot\n"
+			"InnoDB: open the file, you should"
+			" modify the permissions.\n"
+			"InnoDB: 2) If the table is not needed, or you can"
+			" restore it from a backup,\n"
+			"InnoDB: then you can remove the .ibd file,"
+			" and InnoDB will do a normal\n"
+			"InnoDB: crash recovery and ignore that table.\n"
+			"InnoDB: 3) If the file system or the"
+			" disk is broken, and you cannot remove\n"
+			"InnoDB: the .ibd file, you can set"
+			" innodb_force_recovery > 0 in my.cnf\n"
+			"InnoDB: and force InnoDB to continue crash"
+			" recovery here.\n", filepath);
+
+		mem_free(filepath);
+
+		if (srv_force_recovery > 0) {
+			fprintf(stderr,
+				"InnoDB: innodb_force_recovery"
+				" was set to %lu. Continuing crash recovery\n"
+				"InnoDB: even though we cannot access"
+				" the .ibd file of this table.\n",
+				srv_force_recovery);
+			return;
+		}
+
+		exit(1);
+	}
+
+	success = os_file_get_size(file, &size_low, &size_high);
+
+	if (!success) {
+		/* The following call prints an error message */
+		os_file_get_last_error(TRUE);
+
+		fprintf(stderr,
+			"InnoDB: Error: could not measure the size"
+			" of single-table tablespace file\n"
+			"InnoDB: %s!\n"
+			"InnoDB: We do not continue crash recovery,"
+			" because the table will become\n"
+			"InnoDB: corrupt if we cannot apply the log records"
+			" in the InnoDB log to it.\n"
+			"InnoDB: To fix the problem and start mysqld:\n"
+			"InnoDB: 1) If there is a permission problem"
+			" in the file and mysqld cannot\n"
+			"InnoDB: access the file, you should"
+			" modify the permissions.\n"
+			"InnoDB: 2) If the table is not needed,"
+			" or you can restore it from a backup,\n"
+			"InnoDB: then you can remove the .ibd file,"
+			" and InnoDB will do a normal\n"
+			"InnoDB: crash recovery and ignore that table.\n"
+			"InnoDB: 3) If the file system or the disk is broken,"
+			" and you cannot remove\n"
+			"InnoDB: the .ibd file, you can set"
+			" innodb_force_recovery > 0 in my.cnf\n"
+			"InnoDB: and force InnoDB to continue"
+			" crash recovery here.\n", filepath);
+
+		os_file_close(file);
+		mem_free(filepath);
+
+		if (srv_force_recovery > 0) {
+			fprintf(stderr,
+				"InnoDB: innodb_force_recovery"
+				" was set to %lu. Continuing crash recovery\n"
+				"InnoDB: even though we cannot access"
+				" the .ibd file of this table.\n",
+				srv_force_recovery);
+			return;
+		}
+
+		exit(1);
+	}
+
+	/* TODO: What to do in other cases where we cannot access an .ibd
+	file during a crash recovery? */
+
+	/* Every .ibd file is created >= 4 pages in size. Smaller files
+	cannot be ok. */
+
+	size = (((ib_int64_t)size_high) << 32) + (ib_int64_t)size_low;
+#ifndef UNIV_HOTBACKUP
+	if (size < FIL_IBD_FILE_INITIAL_SIZE * UNIV_PAGE_SIZE) {
+		fprintf(stderr,
+			"InnoDB: Error: the size of single-table tablespace"
+			" file %s\n"
+			"InnoDB: is only %lu %lu, should be at least %lu!",
+			filepath,
+			(ulong) size_high,
+			(ulong) size_low, (ulong) (4 * UNIV_PAGE_SIZE));
+		os_file_close(file);
+		mem_free(filepath);
+
+		return;
+	}
+#endif
+	/* Read the first page of the tablespace if the size big enough */
+
+	buf2 = ut_malloc(2 * UNIV_PAGE_SIZE);
+	/* Align the memory for file i/o if we might have O_DIRECT set */
+	page = ut_align(buf2, UNIV_PAGE_SIZE);
+
+	if (size >= FIL_IBD_FILE_INITIAL_SIZE * UNIV_PAGE_SIZE) {
+		success = os_file_read(file, page, 0, 0, UNIV_PAGE_SIZE);
+
+		/* We have to read the tablespace id from the file */
+
+		space_id = fsp_header_get_space_id(page);
+		flags = fsp_header_get_flags(page);
+	} else {
+		space_id = ULINT_UNDEFINED;
+		flags = 0;
+	}
+
+#ifndef UNIV_HOTBACKUP
+	if (space_id == ULINT_UNDEFINED || space_id == 0) {
+		fprintf(stderr,
+			"InnoDB: Error: tablespace id %lu in file %s"
+			" is not sensible\n",
+			(ulong) space_id,
+			filepath);
+		goto func_exit;
+	}
+#else
+	if (space_id == ULINT_UNDEFINED || space_id == 0) {
+		char*	new_path;
+
+		fprintf(stderr,
+			"InnoDB: Renaming tablespace %s of id %lu,\n"
+			"InnoDB: to %s_ibbackup_old_vers_<timestamp>\n"
+			"InnoDB: because its size %lld is too small"
+			" (< 4 pages 16 kB each),\n"
+			"InnoDB: or the space id in the file header"
+			" is not sensible.\n"
+			"InnoDB: This can happen in an ibbackup run,"
+			" and is not dangerous.\n",
+			filepath, space_id, filepath, size);
+		os_file_close(file);
+
+		new_path = fil_make_ibbackup_old_name(filepath);
+		ut_a(os_file_rename(filepath, new_path));
+
+		ut_free(buf2);
+		mem_free(filepath);
+		mem_free(new_path);
+
+		return;
+	}
+
+	/* A backup may contain the same space several times, if the space got
+	renamed at a sensitive time. Since it is enough to have one version of
+	the space, we rename the file if a space with the same space id
+	already exists in the tablespace memory cache. We rather rename the
+	file than delete it, because if there is a bug, we do not want to
+	destroy valuable data. */
+
+	mutex_enter(&(fil_system->mutex));
+
+	space = fil_get_space_for_id_low(space_id);
+
+	if (space) {
+		char*	new_path;
+
+		fprintf(stderr,
+			"InnoDB: Renaming tablespace %s of id %lu,\n"
+			"InnoDB: to %s_ibbackup_old_vers_<timestamp>\n"
+			"InnoDB: because space %s with the same id\n"
+			"InnoDB: was scanned earlier. This can happen"
+			" if you have renamed tables\n"
+			"InnoDB: during an ibbackup run.\n",
+			filepath, space_id, filepath,
+			space->name);
+		os_file_close(file);
+
+		new_path = fil_make_ibbackup_old_name(filepath);
+
+		mutex_exit(&(fil_system->mutex));
+
+		ut_a(os_file_rename(filepath, new_path));
+
+		ut_free(buf2);
+		mem_free(filepath);
+		mem_free(new_path);
+
+		return;
+	}
+	mutex_exit(&(fil_system->mutex));
+#endif
+	success = fil_space_create(filepath, space_id, flags, FIL_TABLESPACE);
+
+	if (!success) {
+
+		goto func_exit;
+	}
+
+	/* We do not use the size information we have about the file, because
+	the rounding formula for extents and pages is somewhat complex; we
+	let fil_node_open() do that task. */
+
+	fil_node_create(filepath, 0, space_id, FALSE);
+func_exit:
+	os_file_close(file);
+	ut_free(buf2);
+	mem_free(filepath);
+}
+
+/***************************************************************************
+A fault-tolerant function that tries to read the next file name in the
+directory. We retry 100 times if os_file_readdir_next_file() returns -1. The
+idea is to read as much good data as we can and jump over bad data. */
+static
+int
+fil_file_readdir_next_file(
+/*=======================*/
+				/* out: 0 if ok, -1 if error even after the
+				retries, 1 if at the end of the directory */
+	ulint*		err,	/* out: this is set to DB_ERROR if an error
+				was encountered, otherwise not changed */
+	const char*	dirname,/* in: directory name or path */
+	os_file_dir_t	dir,	/* in: directory stream */
+	os_file_stat_t*	info)	/* in/out: buffer where the info is returned */
+{
+	ulint	i;
+	int	ret;
+
+	for (i = 0; i < 100; i++) {
+		ret = os_file_readdir_next_file(dirname, dir, info);
+
+		if (ret != -1) {
+
+			return(ret);
+		}
+
+		fprintf(stderr,
+			"InnoDB: Error: os_file_readdir_next_file()"
+			" returned -1 in\n"
+			"InnoDB: directory %s\n"
+			"InnoDB: Crash recovery may have failed"
+			" for some .ibd files!\n", dirname);
+
+		*err = DB_ERROR;
+	}
+
+	return(-1);
+}
+
+/************************************************************************
+At the server startup, if we need crash recovery, scans the database
+directories under the MySQL datadir, looking for .ibd files. Those files are
+single-table tablespaces. We need to know the space id in each of them so that
+we know into which file we should look to check the contents of a page stored
+in the doublewrite buffer, also to know where to apply log records where the
+space id is != 0. */
+UNIV_INTERN
+ulint
+fil_load_single_table_tablespaces(void)
+/*===================================*/
+			/* out: DB_SUCCESS or error number */
+{
+	int		ret;
+	char*		dbpath		= NULL;
+	ulint		dbpath_len	= 100;
+	os_file_dir_t	dir;
+	os_file_dir_t	dbdir;
+	os_file_stat_t	dbinfo;
+	os_file_stat_t	fileinfo;
+	ulint		err		= DB_SUCCESS;
+
+	/* The datadir of MySQL is always the default directory of mysqld */
+
+	dir = os_file_opendir(fil_path_to_mysql_datadir, TRUE);
+
+	if (dir == NULL) {
+
+		return(DB_ERROR);
+	}
+
+	dbpath = mem_alloc(dbpath_len);
+
+	/* Scan all directories under the datadir. They are the database
+	directories of MySQL. */
+
+	ret = fil_file_readdir_next_file(&err, fil_path_to_mysql_datadir, dir,
+					 &dbinfo);
+	while (ret == 0) {
+		ulint len;
+		/* printf("Looking at %s in datadir\n", dbinfo.name); */
+
+		if (dbinfo.type == OS_FILE_TYPE_FILE
+		    || dbinfo.type == OS_FILE_TYPE_UNKNOWN) {
+
+			goto next_datadir_item;
+		}
+
+		/* We found a symlink or a directory; try opening it to see
+		if a symlink is a directory */
+
+		len = strlen(fil_path_to_mysql_datadir)
+			+ strlen (dbinfo.name) + 2;
+		if (len > dbpath_len) {
+			dbpath_len = len;
+
+			if (dbpath) {
+				mem_free(dbpath);
+			}
+
+			dbpath = mem_alloc(dbpath_len);
+		}
+		sprintf(dbpath, "%s/%s", fil_path_to_mysql_datadir,
+			dbinfo.name);
+		srv_normalize_path_for_win(dbpath);
+
+		dbdir = os_file_opendir(dbpath, FALSE);
+
+		if (dbdir != NULL) {
+			/* printf("Opened dir %s\n", dbinfo.name); */
+
+			/* We found a database directory; loop through it,
+			looking for possible .ibd files in it */
+
+			ret = fil_file_readdir_next_file(&err, dbpath, dbdir,
+							 &fileinfo);
+			while (ret == 0) {
+				/* printf(
+				"     Looking at file %s\n", fileinfo.name); */
+
+				if (fileinfo.type == OS_FILE_TYPE_DIR) {
+
+					goto next_file_item;
+				}
+
+				/* We found a symlink or a file */
+				if (strlen(fileinfo.name) > 4
+				    && 0 == strcmp(fileinfo.name
+						   + strlen(fileinfo.name) - 4,
+						   ".ibd")) {
+					/* The name ends in .ibd; try opening
+					the file */
+					fil_load_single_table_tablespace(
+						dbinfo.name, fileinfo.name);
+				}
+next_file_item:
+				ret = fil_file_readdir_next_file(&err,
+								 dbpath, dbdir,
+								 &fileinfo);
+			}
+
+			if (0 != os_file_closedir(dbdir)) {
+				fputs("InnoDB: Warning: could not"
+				      " close database directory ", stderr);
+				ut_print_filename(stderr, dbpath);
+				putc('\n', stderr);
+
+				err = DB_ERROR;
+			}
+		}
+
+next_datadir_item:
+		ret = fil_file_readdir_next_file(&err,
+						 fil_path_to_mysql_datadir,
+						 dir, &dbinfo);
+	}
+
+	mem_free(dbpath);
+
+	if (0 != os_file_closedir(dir)) {
+		fprintf(stderr,
+			"InnoDB: Error: could not close MySQL datadir\n");
+
+		return(DB_ERROR);
+	}
+
+	return(err);
+}
+
+/************************************************************************
+If we need crash recovery, and we have called
+fil_load_single_table_tablespaces() and dict_load_single_table_tablespaces(),
+we can call this function to print an error message of orphaned .ibd files
+for which there is not a data dictionary entry with a matching table name
+and space id. */
+UNIV_INTERN
+void
+fil_print_orphaned_tablespaces(void)
+/*================================*/
+{
+	fil_system_t*	system		= fil_system;
+	fil_space_t*	space;
+
+	mutex_enter(&(system->mutex));
+
+	space = UT_LIST_GET_FIRST(system->space_list);
+
+	while (space) {
+		if (space->purpose == FIL_TABLESPACE && space->id != 0
+		    && !space->mark) {
+			fputs("InnoDB: Warning: tablespace ", stderr);
+			ut_print_filename(stderr, space->name);
+			fprintf(stderr, " of id %lu has no matching table in\n"
+				"InnoDB: the InnoDB data dictionary.\n",
+				(ulong) space->id);
+		}
+
+		space = UT_LIST_GET_NEXT(space_list, space);
+	}
+
+	mutex_exit(&(system->mutex));
+}
+
+/***********************************************************************
+Returns TRUE if a single-table tablespace does not exist in the memory cache,
+or is being deleted there. */
+UNIV_INTERN
+ibool
+fil_tablespace_deleted_or_being_deleted_in_mem(
+/*===========================================*/
+				/* out: TRUE if does not exist or is being\
+				deleted */
+	ulint		id,	/* in: space id */
+	ib_int64_t	version)/* in: tablespace_version should be this; if
+				you pass -1 as the value of this, then this
+				parameter is ignored */
+{
+	fil_system_t*	system	= fil_system;
+	fil_space_t*	space;
+
+	ut_ad(system);
+
+	mutex_enter(&(system->mutex));
+
+	space = fil_space_get_by_id(id);
+
+	if (space == NULL || space->is_being_deleted) {
+		mutex_exit(&(system->mutex));
+
+		return(TRUE);
+	}
+
+	if (version != ((ib_int64_t)-1)
+	    && space->tablespace_version != version) {
+		mutex_exit(&(system->mutex));
+
+		return(TRUE);
+	}
+
+	mutex_exit(&(system->mutex));
+
+	return(FALSE);
+}
+
+/***********************************************************************
+Returns TRUE if a single-table tablespace exists in the memory cache. */
+UNIV_INTERN
+ibool
+fil_tablespace_exists_in_mem(
+/*=========================*/
+			/* out: TRUE if exists */
+	ulint	id)	/* in: space id */
+{
+	fil_system_t*	system		= fil_system;
+	fil_space_t*	space;
+
+	ut_ad(system);
+
+	mutex_enter(&(system->mutex));
+
+	space = fil_space_get_by_id(id);
+
+	if (space == NULL) {
+		mutex_exit(&(system->mutex));
+
+		return(FALSE);
+	}
+
+	mutex_exit(&(system->mutex));
+
+	return(TRUE);
+}
+
+/***********************************************************************
+Returns TRUE if a matching tablespace exists in the InnoDB tablespace memory
+cache. Note that if we have not done a crash recovery at the database startup,
+there may be many tablespaces which are not yet in the memory cache. */
+UNIV_INTERN
+ibool
+fil_space_for_table_exists_in_mem(
+/*==============================*/
+					/* out: TRUE if a matching tablespace
+					exists in the memory cache */
+	ulint		id,		/* in: space id */
+	const char*	name,		/* in: table name in the standard
+					'databasename/tablename' format or
+					the dir path to a temp table */
+	ibool		is_temp,	/* in: TRUE if created with CREATE
+					TEMPORARY TABLE */
+	ibool		mark_space,	/* in: in crash recovery, at database
+					startup we mark all spaces which have
+					an associated table in the InnoDB
+					data dictionary, so that
+					we can print a warning about orphaned
+					tablespaces */
+	ibool		print_error_if_does_not_exist)
+					/* in: print detailed error
+					information to the .err log if a
+					matching tablespace is not found from
+					memory */
+{
+	fil_system_t*	system		= fil_system;
+	fil_space_t*	namespace;
+	fil_space_t*	space;
+	char*		path;
+
+	ut_ad(system);
+
+	mutex_enter(&(system->mutex));
+
+	path = fil_make_ibd_name(name, is_temp);
+
+	/* Look if there is a space with the same id */
+
+	space = fil_space_get_by_id(id);
+
+	/* Look if there is a space with the same name; the name is the
+	directory path from the datadir to the file */
+
+	namespace = fil_space_get_by_name(path);
+	if (space && space == namespace) {
+		/* Found */
+
+		if (mark_space) {
+			space->mark = TRUE;
+		}
+
+		mem_free(path);
+		mutex_exit(&(system->mutex));
+
+		return(TRUE);
+	}
+
+	if (!print_error_if_does_not_exist) {
+
+		mem_free(path);
+		mutex_exit(&(system->mutex));
+
+		return(FALSE);
+	}
+
+	if (space == NULL) {
+		if (namespace == NULL) {
+			ut_print_timestamp(stderr);
+			fputs("  InnoDB: Error: table ", stderr);
+			ut_print_filename(stderr, name);
+			fprintf(stderr, "\n"
+				"InnoDB: in InnoDB data dictionary"
+				" has tablespace id %lu,\n"
+				"InnoDB: but tablespace with that id"
+				" or name does not exist. Have\n"
+				"InnoDB: you deleted or moved .ibd files?\n"
+				"InnoDB: This may also be a table created with"
+				" CREATE TEMPORARY TABLE\n"
+				"InnoDB: whose .ibd and .frm files"
+				" MySQL automatically removed, but the\n"
+				"InnoDB: table still exists in the"
+				" InnoDB internal data dictionary.\n",
+				(ulong) id);
+		} else {
+			ut_print_timestamp(stderr);
+			fputs("  InnoDB: Error: table ", stderr);
+			ut_print_filename(stderr, name);
+			fprintf(stderr, "\n"
+				"InnoDB: in InnoDB data dictionary has"
+				" tablespace id %lu,\n"
+				"InnoDB: but a tablespace with that id"
+				" does not exist. There is\n"
+				"InnoDB: a tablespace of name %s and id %lu,"
+				" though. Have\n"
+				"InnoDB: you deleted or moved .ibd files?\n",
+				(ulong) id, namespace->name,
+				(ulong) namespace->id);
+		}
+error_exit:
+		fputs("InnoDB: Please refer to\n"
+		      "InnoDB: http://dev.mysql.com/doc/refman/5.1/en/"
+		      "innodb-troubleshooting.html\n"
+		      "InnoDB: for how to resolve the issue.\n", stderr);
+
+		mem_free(path);
+		mutex_exit(&(system->mutex));
+
+		return(FALSE);
+	}
+
+	if (0 != strcmp(space->name, path)) {
+		ut_print_timestamp(stderr);
+		fputs("  InnoDB: Error: table ", stderr);
+		ut_print_filename(stderr, name);
+		fprintf(stderr, "\n"
+			"InnoDB: in InnoDB data dictionary has"
+			" tablespace id %lu,\n"
+			"InnoDB: but the tablespace with that id"
+			" has name %s.\n"
+			"InnoDB: Have you deleted or moved .ibd files?\n",
+			(ulong) id, space->name);
+
+		if (namespace != NULL) {
+			fputs("InnoDB: There is a tablespace"
+			      " with the right name\n"
+			      "InnoDB: ", stderr);
+			ut_print_filename(stderr, namespace->name);
+			fprintf(stderr, ", but its id is %lu.\n",
+				(ulong) namespace->id);
+		}
+
+		goto error_exit;
+	}
+
+	mem_free(path);
+	mutex_exit(&(system->mutex));
+
+	return(FALSE);
+}
+
+/***********************************************************************
+Checks if a single-table tablespace for a given table name exists in the
+tablespace memory cache. */
+static
+ulint
+fil_get_space_id_for_table(
+/*=======================*/
+				/* out: space id, ULINT_UNDEFINED if not
+				found */
+	const char*	name)	/* in: table name in the standard
+				'databasename/tablename' format */
+{
+	fil_system_t*	system		= fil_system;
+	fil_space_t*	namespace;
+	ulint		id		= ULINT_UNDEFINED;
+	char*		path;
+
+	ut_ad(system);
+
+	mutex_enter(&(system->mutex));
+
+	path = fil_make_ibd_name(name, FALSE);
+
+	/* Look if there is a space with the same name; the name is the
+	directory path to the file */
+
+	namespace = fil_space_get_by_name(path);
+
+	if (namespace) {
+		id = namespace->id;
+	}
+
+	mem_free(path);
+
+	mutex_exit(&(system->mutex));
+
+	return(id);
+}
+
+/**************************************************************************
+Tries to extend a data file so that it would accommodate the number of pages
+given. The tablespace must be cached in the memory cache. If the space is big
+enough already, does nothing. */
+UNIV_INTERN
+ibool
+fil_extend_space_to_desired_size(
+/*=============================*/
+				/* out: TRUE if success */
+	ulint*	actual_size,	/* out: size of the space after extension;
+				if we ran out of disk space this may be lower
+				than the desired size */
+	ulint	space_id,	/* in: space id */
+	ulint	size_after_extend)/* in: desired size in pages after the
+				extension; if the current space size is bigger
+				than this already, the function does nothing */
+{
+	fil_system_t*	system		= fil_system;
+	fil_node_t*	node;
+	fil_space_t*	space;
+	byte*		buf2;
+	byte*		buf;
+	ulint		buf_size;
+	ulint		start_page_no;
+	ulint		file_start_page_no;
+	ulint		offset_high;
+	ulint		offset_low;
+	ulint		page_size;
+	ibool		success		= TRUE;
+
+	fil_mutex_enter_and_prepare_for_io(space_id);
+
+	space = fil_space_get_by_id(space_id);
+	ut_a(space);
+
+	if (space->size >= size_after_extend) {
+		/* Space already big enough */
+
+		*actual_size = space->size;
+
+		mutex_exit(&(system->mutex));
+
+		return(TRUE);
+	}
+
+	page_size = dict_table_flags_to_zip_size(space->flags);
+	if (!page_size) {
+		page_size = UNIV_PAGE_SIZE;
+	}
+
+	node = UT_LIST_GET_LAST(space->chain);
+
+	fil_node_prepare_for_io(node, system, space);
+
+	start_page_no = space->size;
+	file_start_page_no = space->size - node->size;
+
+	/* Extend at most 64 pages at a time */
+	buf_size = ut_min(64, size_after_extend - start_page_no) * page_size;
+	buf2 = mem_alloc(buf_size + page_size);
+	buf = ut_align(buf2, page_size);
+
+	memset(buf, 0, buf_size);
+
+	while (start_page_no < size_after_extend) {
+		ulint	n_pages = ut_min(buf_size / page_size,
+					 size_after_extend - start_page_no);
+
+		offset_high = (start_page_no - file_start_page_no)
+			/ (4096 * ((1024 * 1024) / page_size));
+		offset_low  = ((start_page_no - file_start_page_no)
+			       % (4096 * ((1024 * 1024) / page_size)))
+			* page_size;
+#ifdef UNIV_HOTBACKUP
+		success = os_file_write(node->name, node->handle, buf,
+					offset_low, offset_high,
+					page_size * n_pages);
+#else
+		success = os_aio(OS_FILE_WRITE, OS_AIO_SYNC,
+				 node->name, node->handle, buf,
+				 offset_low, offset_high,
+				 page_size * n_pages,
+				 NULL, NULL);
+#endif
+		if (success) {
+			node->size += n_pages;
+			space->size += n_pages;
+
+			os_has_said_disk_full = FALSE;
+		} else {
+			/* Let us measure the size of the file to determine
+			how much we were able to extend it */
+
+			n_pages = ((ulint)
+				   (os_file_get_size_as_iblonglong(
+					   node->handle)
+				    / page_size)) - node->size;
+
+			node->size += n_pages;
+			space->size += n_pages;
+
+			break;
+		}
+
+		start_page_no += n_pages;
+	}
+
+	mem_free(buf2);
+
+	fil_node_complete_io(node, system, OS_FILE_WRITE);
+
+	*actual_size = space->size;
+
+#ifndef UNIV_HOTBACKUP
+	if (space_id == 0) {
+		ulint pages_per_mb = (1024 * 1024) / page_size;
+
+		/* Keep the last data file size info up to date, rounded to
+		full megabytes */
+
+		srv_data_file_sizes[srv_n_data_files - 1]
+			= (node->size / pages_per_mb) * pages_per_mb;
+	}
+#endif /* !UNIV_HOTBACKUP */
+
+	/*
+	printf("Extended %s to %lu, actual size %lu pages\n", space->name,
+	size_after_extend, *actual_size); */
+	mutex_exit(&(system->mutex));
+
+	fil_flush(space_id);
+
+	return(success);
+}
+
+#ifdef UNIV_HOTBACKUP
+/************************************************************************
+Extends all tablespaces to the size stored in the space header. During the
+ibbackup --apply-log phase we extended the spaces on-demand so that log records
+could be applied, but that may have left spaces still too small compared to
+the size stored in the space header. */
+UNIV_INTERN
+void
+fil_extend_tablespaces_to_stored_len(void)
+/*======================================*/
+{
+	fil_system_t*	system		= fil_system;
+	fil_space_t*	space;
+	byte*		buf;
+	ulint		actual_size;
+	ulint		size_in_header;
+	ulint		error;
+	ibool		success;
+
+	buf = mem_alloc(UNIV_PAGE_SIZE);
+
+	mutex_enter(&(system->mutex));
+
+	space = UT_LIST_GET_FIRST(system->space_list);
+
+	while (space) {
+		ut_a(space->purpose == FIL_TABLESPACE);
+
+		mutex_exit(&(system->mutex)); /* no need to protect with a
+					      mutex, because this is a
+					      single-threaded operation */
+		error = fil_read(TRUE, space->id, space->zip_size,
+				 0, 0, UNIV_PAGE_SIZE, buf, NULL);
+		ut_a(error == DB_SUCCESS);
+
+		size_in_header = fsp_get_size_low(buf);
+
+		success = fil_extend_space_to_desired_size(
+			&actual_size, space->id, size_in_header);
+		if (!success) {
+			fprintf(stderr,
+				"InnoDB: Error: could not extend the"
+				" tablespace of %s\n"
+				"InnoDB: to the size stored in header,"
+				" %lu pages;\n"
+				"InnoDB: size after extension %lu pages\n"
+				"InnoDB: Check that you have free disk space"
+				" and retry!\n",
+				space->name, size_in_header, actual_size);
+			exit(1);
+		}
+
+		mutex_enter(&(system->mutex));
+
+		space = UT_LIST_GET_NEXT(space_list, space);
+	}
+
+	mutex_exit(&(system->mutex));
+
+	mem_free(buf);
+}
+#endif
+
+/*========== RESERVE FREE EXTENTS (for a B-tree split, for example) ===*/
+
+/***********************************************************************
+Tries to reserve free extents in a file space. */
+UNIV_INTERN
+ibool
+fil_space_reserve_free_extents(
+/*===========================*/
+				/* out: TRUE if succeed */
+	ulint	id,		/* in: space id */
+	ulint	n_free_now,	/* in: number of free extents now */
+	ulint	n_to_reserve)	/* in: how many one wants to reserve */
+{
+	fil_system_t*	system		= fil_system;
+	fil_space_t*	space;
+	ibool		success;
+
+	ut_ad(system);
+
+	mutex_enter(&(system->mutex));
+
+	space = fil_space_get_by_id(id);
+
+	ut_a(space);
+
+	if (space->n_reserved_extents + n_to_reserve > n_free_now) {
+		success = FALSE;
+	} else {
+		space->n_reserved_extents += n_to_reserve;
+		success = TRUE;
+	}
+
+	mutex_exit(&(system->mutex));
+
+	return(success);
+}
+
+/***********************************************************************
+Releases free extents in a file space. */
+UNIV_INTERN
+void
+fil_space_release_free_extents(
+/*===========================*/
+	ulint	id,		/* in: space id */
+	ulint	n_reserved)	/* in: how many one reserved */
+{
+	fil_system_t*	system		= fil_system;
+	fil_space_t*	space;
+
+	ut_ad(system);
+
+	mutex_enter(&(system->mutex));
+
+	space = fil_space_get_by_id(id);
+
+	ut_a(space);
+	ut_a(space->n_reserved_extents >= n_reserved);
+
+	space->n_reserved_extents -= n_reserved;
+
+	mutex_exit(&(system->mutex));
+}
+
+/***********************************************************************
+Gets the number of reserved extents. If the database is silent, this number
+should be zero. */
+UNIV_INTERN
+ulint
+fil_space_get_n_reserved_extents(
+/*=============================*/
+	ulint	id)		/* in: space id */
+{
+	fil_system_t*	system		= fil_system;
+	fil_space_t*	space;
+	ulint		n;
+
+	ut_ad(system);
+
+	mutex_enter(&(system->mutex));
+
+	space = fil_space_get_by_id(id);
+
+	ut_a(space);
+
+	n = space->n_reserved_extents;
+
+	mutex_exit(&(system->mutex));
+
+	return(n);
+}
+
+/*============================ FILE I/O ================================*/
+
+/************************************************************************
+NOTE: you must call fil_mutex_enter_and_prepare_for_io() first!
+
+Prepares a file node for i/o. Opens the file if it is closed. Updates the
+pending i/o's field in the node and the system appropriately. Takes the node
+off the LRU list if it is in the LRU list. The caller must hold the fil_sys
+mutex. */
+static
+void
+fil_node_prepare_for_io(
+/*====================*/
+	fil_node_t*	node,	/* in: file node */
+	fil_system_t*	system,	/* in: tablespace memory cache */
+	fil_space_t*	space)	/* in: space */
+{
+	ut_ad(node && system && space);
+	ut_ad(mutex_own(&(system->mutex)));
+
+	if (system->n_open > system->max_n_open + 5) {
+		ut_print_timestamp(stderr);
+		fprintf(stderr,
+			"  InnoDB: Warning: open files %lu"
+			" exceeds the limit %lu\n",
+			(ulong) system->n_open,
+			(ulong) system->max_n_open);
+	}
+
+	if (node->open == FALSE) {
+		/* File is closed: open it */
+		ut_a(node->n_pending == 0);
+
+		fil_node_open_file(node, system, space);
+	}
+
+	if (node->n_pending == 0 && space->purpose == FIL_TABLESPACE
+	    && space->id != 0) {
+		/* The node is in the LRU list, remove it */
+
+		ut_a(UT_LIST_GET_LEN(system->LRU) > 0);
+
+		UT_LIST_REMOVE(LRU, system->LRU, node);
+	}
+
+	node->n_pending++;
+}
+
+/************************************************************************
+Updates the data structures when an i/o operation finishes. Updates the
+pending i/o's field in the node appropriately. */
+static
+void
+fil_node_complete_io(
+/*=================*/
+	fil_node_t*	node,	/* in: file node */
+	fil_system_t*	system,	/* in: tablespace memory cache */
+	ulint		type)	/* in: OS_FILE_WRITE or OS_FILE_READ; marks
+				the node as modified if
+				type == OS_FILE_WRITE */
+{
+	ut_ad(node);
+	ut_ad(system);
+	ut_ad(mutex_own(&(system->mutex)));
+
+	ut_a(node->n_pending > 0);
+
+	node->n_pending--;
+
+	if (type == OS_FILE_WRITE) {
+		system->modification_counter++;
+		node->modification_counter = system->modification_counter;
+
+		if (!node->space->is_in_unflushed_spaces) {
+
+			node->space->is_in_unflushed_spaces = TRUE;
+			UT_LIST_ADD_FIRST(unflushed_spaces,
+					  system->unflushed_spaces,
+					  node->space);
+		}
+	}
+
+	if (node->n_pending == 0 && node->space->purpose == FIL_TABLESPACE
+	    && node->space->id != 0) {
+		/* The node must be put back to the LRU list */
+		UT_LIST_ADD_FIRST(LRU, system->LRU, node);
+	}
+}
+
+/************************************************************************
+Report information about an invalid page access. */
+static
+void
+fil_report_invalid_page_access(
+/*===========================*/
+	ulint		block_offset,	/* in: block offset */
+	ulint		space_id,	/* in: space id */
+	const char*	space_name,	/* in: space name */
+	ulint		byte_offset,	/* in: byte offset */
+	ulint		len,		/* in: I/O length */
+	ulint		type)		/* in: I/O type */
+{
+	fprintf(stderr,
+		"InnoDB: Error: trying to access page number %lu"
+		" in space %lu,\n"
+		"InnoDB: space name %s,\n"
+		"InnoDB: which is outside the tablespace bounds.\n"
+		"InnoDB: Byte offset %lu, len %lu, i/o type %lu.\n"
+		"InnoDB: If you get this error at mysqld startup,"
+		" please check that\n"
+		"InnoDB: your my.cnf matches the ibdata files"
+		" that you have in the\n"
+		"InnoDB: MySQL server.\n",
+		(ulong) block_offset, (ulong) space_id, space_name,
+		(ulong) byte_offset, (ulong) len, (ulong) type);
+}
+
+/************************************************************************
+Reads or writes data. This operation is asynchronous (aio). */
+UNIV_INTERN
+ulint
+fil_io(
+/*===*/
+				/* out: DB_SUCCESS, or DB_TABLESPACE_DELETED
+				if we are trying to do i/o on a tablespace
+				which does not exist */
+	ulint	type,		/* in: OS_FILE_READ or OS_FILE_WRITE,
+				ORed to OS_FILE_LOG, if a log i/o
+				and ORed to OS_AIO_SIMULATED_WAKE_LATER
+				if simulated aio and we want to post a
+				batch of i/os; NOTE that a simulated batch
+				may introduce hidden chances of deadlocks,
+				because i/os are not actually handled until
+				all have been posted: use with great
+				caution! */
+	ibool	sync,		/* in: TRUE if synchronous aio is desired */
+	ulint	space_id,	/* in: space id */
+	ulint	zip_size,	/* in: compressed page size in bytes;
+				0 for uncompressed pages */
+	ulint	block_offset,	/* in: offset in number of blocks */
+	ulint	byte_offset,	/* in: remainder of offset in bytes; in
+				aio this must be divisible by the OS block
+				size */
+	ulint	len,		/* in: how many bytes to read or write; this
+				must not cross a file boundary; in aio this
+				must be a block size multiple */
+	void*	buf,		/* in/out: buffer where to store read data
+				or from where to write; in aio this must be
+				appropriately aligned */
+	void*	message)	/* in: message for aio handler if non-sync
+				aio used, else ignored */
+{
+	fil_system_t*	system		= fil_system;
+	ulint		mode;
+	fil_space_t*	space;
+	fil_node_t*	node;
+	ulint		offset_high;
+	ulint		offset_low;
+	ibool		ret;
+	ulint		is_log;
+	ulint		wake_later;
+
+	is_log = type & OS_FILE_LOG;
+	type = type & ~OS_FILE_LOG;
+
+	wake_later = type & OS_AIO_SIMULATED_WAKE_LATER;
+	type = type & ~OS_AIO_SIMULATED_WAKE_LATER;
+
+	ut_ad(byte_offset < UNIV_PAGE_SIZE);
+	ut_ad(!zip_size || !byte_offset);
+	ut_ad(ut_is_2pow(zip_size));
+	ut_ad(buf);
+	ut_ad(len > 0);
+#if (1 << UNIV_PAGE_SIZE_SHIFT) != UNIV_PAGE_SIZE
+# error "(1 << UNIV_PAGE_SIZE_SHIFT) != UNIV_PAGE_SIZE"
+#endif
+	ut_ad(fil_validate());
+#ifndef UNIV_LOG_DEBUG
+	/* ibuf bitmap pages must be read in the sync aio mode: */
+	ut_ad(recv_no_ibuf_operations || (type == OS_FILE_WRITE)
+	      || !ibuf_bitmap_page(zip_size, block_offset)
+	      || sync || is_log);
+	ut_ad(!ibuf_inside() || is_log || (type == OS_FILE_WRITE)
+	      || ibuf_page(space_id, zip_size, block_offset, NULL));
+#endif
+	if (sync) {
+		mode = OS_AIO_SYNC;
+	} else if (is_log) {
+		mode = OS_AIO_LOG;
+	} else if (type == OS_FILE_READ
+		   && !recv_no_ibuf_operations
+		   && ibuf_page(space_id, zip_size, block_offset, NULL)) {
+		mode = OS_AIO_IBUF;
+	} else {
+		mode = OS_AIO_NORMAL;
+	}
+
+	if (type == OS_FILE_READ) {
+		srv_data_read+= len;
+	} else if (type == OS_FILE_WRITE) {
+		srv_data_written+= len;
+	}
+
+	/* Reserve the fil_system mutex and make sure that we can open at
+	least one file while holding it, if the file is not already open */
+
+	fil_mutex_enter_and_prepare_for_io(space_id);
+
+	space = fil_space_get_by_id(space_id);
+
+	if (!space) {
+		mutex_exit(&(system->mutex));
+
+		ut_print_timestamp(stderr);
+		fprintf(stderr,
+			"  InnoDB: Error: trying to do i/o"
+			" to a tablespace which does not exist.\n"
+			"InnoDB: i/o type %lu, space id %lu,"
+			" page no. %lu, i/o length %lu bytes\n",
+			(ulong) type, (ulong) space_id, (ulong) block_offset,
+			(ulong) len);
+
+		return(DB_TABLESPACE_DELETED);
+	}
+
+	ut_ad((mode != OS_AIO_IBUF) || (space->purpose == FIL_TABLESPACE));
+
+	node = UT_LIST_GET_FIRST(space->chain);
+
+	for (;;) {
+		if (UNIV_UNLIKELY(node == NULL)) {
+			fil_report_invalid_page_access(
+				block_offset, space_id, space->name,
+				byte_offset, len, type);
+
+			ut_error;
+		}
+
+		if (space->id != 0 && node->size == 0) {
+			/* We do not know the size of a single-table tablespace
+			before we open the file */
+
+			break;
+		}
+
+		if (node->size > block_offset) {
+			/* Found! */
+			break;
+		} else {
+			block_offset -= node->size;
+			node = UT_LIST_GET_NEXT(chain, node);
+		}
+	}
+
+	/* Open file if closed */
+	fil_node_prepare_for_io(node, system, space);
+
+	/* Check that at least the start offset is within the bounds of a
+	single-table tablespace */
+	if (UNIV_UNLIKELY(node->size <= block_offset)
+	    && space->id != 0 && space->purpose == FIL_TABLESPACE) {
+
+		fil_report_invalid_page_access(
+			block_offset, space_id, space->name, byte_offset,
+			len, type);
+
+		ut_error;
+	}
+
+	/* Now we have made the changes in the data structures of system */
+	mutex_exit(&(system->mutex));
+
+	/* Calculate the low 32 bits and the high 32 bits of the file offset */
+
+	if (!zip_size) {
+		offset_high = (block_offset >> (32 - UNIV_PAGE_SIZE_SHIFT));
+		offset_low  = ((block_offset << UNIV_PAGE_SIZE_SHIFT)
+			       & 0xFFFFFFFFUL) + byte_offset;
+
+		ut_a(node->size - block_offset
+		     >= ((byte_offset + len + (UNIV_PAGE_SIZE - 1))
+			 / UNIV_PAGE_SIZE));
+	} else {
+		ulint	zip_size_shift;
+		switch (zip_size) {
+		case 1024: zip_size_shift = 10; break;
+		case 2048: zip_size_shift = 11; break;
+		case 4096: zip_size_shift = 12; break;
+		case 8192: zip_size_shift = 13; break;
+		case 16384: zip_size_shift = 14; break;
+		default: ut_error;
+		}
+		offset_high = block_offset >> (32 - zip_size_shift);
+		offset_low = (block_offset << zip_size_shift & 0xFFFFFFFFUL)
+			+ byte_offset;
+		ut_a(node->size - block_offset
+		     >= (len + (zip_size - 1)) / zip_size);
+	}
+
+	/* Do aio */
+
+	ut_a(byte_offset % OS_FILE_LOG_BLOCK_SIZE == 0);
+	ut_a((len % OS_FILE_LOG_BLOCK_SIZE) == 0);
+
+#ifdef UNIV_HOTBACKUP
+	/* In ibbackup do normal i/o, not aio */
+	if (type == OS_FILE_READ) {
+		ret = os_file_read(node->handle, buf, offset_low, offset_high,
+				   len);
+	} else {
+		ret = os_file_write(node->name, node->handle, buf,
+				    offset_low, offset_high, len);
+	}
+#else
+	/* Queue the aio request */
+	ret = os_aio(type, mode | wake_later, node->name, node->handle, buf,
+		     offset_low, offset_high, len, node, message);
+#endif
+	ut_a(ret);
+
+	if (mode == OS_AIO_SYNC) {
+		/* The i/o operation is already completed when we return from
+		os_aio: */
+
+		mutex_enter(&(system->mutex));
+
+		fil_node_complete_io(node, system, type);
+
+		mutex_exit(&(system->mutex));
+
+		ut_ad(fil_validate());
+	}
+
+	return(DB_SUCCESS);
+}
+
+/**************************************************************************
+Waits for an aio operation to complete. This function is used to write the
+handler for completed requests. The aio array of pending requests is divided
+into segments (see os0file.c for more info). The thread specifies which
+segment it wants to wait for. */
+UNIV_INTERN
+void
+fil_aio_wait(
+/*=========*/
+	ulint	segment)	/* in: the number of the segment in the aio
+				array to wait for */
+{
+	fil_system_t*	system		= fil_system;
+	ibool		ret;
+	fil_node_t*	fil_node;
+	void*		message;
+	ulint		type;
+
+	ut_ad(fil_validate());
+
+	if (os_aio_use_native_aio) {
+		srv_set_io_thread_op_info(segment, "native aio handle");
+#ifdef WIN_ASYNC_IO
+		ret = os_aio_windows_handle(segment, 0, &fil_node,
+					    &message, &type);
+#else
+		ret = 0; /* Eliminate compiler warning */
+		ut_error;
+#endif
+	} else {
+		srv_set_io_thread_op_info(segment, "simulated aio handle");
+
+		ret = os_aio_simulated_handle(segment, &fil_node,
+					      &message, &type);
+	}
+
+	ut_a(ret);
+
+	srv_set_io_thread_op_info(segment, "complete io for fil node");
+
+	mutex_enter(&(system->mutex));
+
+	fil_node_complete_io(fil_node, fil_system, type);
+
+	mutex_exit(&(system->mutex));
+
+	ut_ad(fil_validate());
+
+	/* Do the i/o handling */
+	/* IMPORTANT: since i/o handling for reads will read also the insert
+	buffer in tablespace 0, you have to be very careful not to introduce
+	deadlocks in the i/o system. We keep tablespace 0 data files always
+	open, and use a special i/o thread to serve insert buffer requests. */
+
+	if (fil_node->space->purpose == FIL_TABLESPACE) {
+		srv_set_io_thread_op_info(segment, "complete io for buf page");
+		buf_page_io_complete(message);
+	} else {
+		srv_set_io_thread_op_info(segment, "complete io for log");
+		log_io_complete(message);
+	}
+}
+
+/**************************************************************************
+Flushes to disk possible writes cached by the OS. If the space does not exist
+or is being dropped, does not do anything. */
+UNIV_INTERN
+void
+fil_flush(
+/*======*/
+	ulint	space_id)	/* in: file space id (this can be a group of
+				log files or a tablespace of the database) */
+{
+	fil_system_t*	system	= fil_system;
+	fil_space_t*	space;
+	fil_node_t*	node;
+	os_file_t	file;
+	ib_int64_t	old_mod_counter;
+
+	mutex_enter(&(system->mutex));
+
+	space = fil_space_get_by_id(space_id);
+
+	if (!space || space->is_being_deleted) {
+		mutex_exit(&(system->mutex));
+
+		return;
+	}
+
+	space->n_pending_flushes++;	/* prevent dropping of the space while
+					we are flushing */
+	node = UT_LIST_GET_FIRST(space->chain);
+
+	while (node) {
+		if (node->modification_counter > node->flush_counter) {
+			ut_a(node->open);
+
+			/* We want to flush the changes at least up to
+			old_mod_counter */
+			old_mod_counter = node->modification_counter;
+
+			if (space->purpose == FIL_TABLESPACE) {
+				fil_n_pending_tablespace_flushes++;
+			} else {
+				fil_n_pending_log_flushes++;
+				fil_n_log_flushes++;
+			}
+#ifdef __WIN__
+			if (node->is_raw_disk) {
+
+				goto skip_flush;
+			}
+#endif
+retry:
+			if (node->n_pending_flushes > 0) {
+				/* We want to avoid calling os_file_flush() on
+				the file twice at the same time, because we do
+				not know what bugs OS's may contain in file
+				i/o; sleep for a while */
+
+				mutex_exit(&(system->mutex));
+
+				os_thread_sleep(20000);
+
+				mutex_enter(&(system->mutex));
+
+				if (node->flush_counter >= old_mod_counter) {
+
+					goto skip_flush;
+				}
+
+				goto retry;
+			}
+
+			ut_a(node->open);
+			file = node->handle;
+			node->n_pending_flushes++;
+
+			mutex_exit(&(system->mutex));
+
+			/* fprintf(stderr, "Flushing to file %s\n",
+			node->name); */
+
+			os_file_flush(file);
+
+			mutex_enter(&(system->mutex));
+
+			node->n_pending_flushes--;
+skip_flush:
+			if (node->flush_counter < old_mod_counter) {
+				node->flush_counter = old_mod_counter;
+
+				if (space->is_in_unflushed_spaces
+				    && fil_space_is_flushed(space)) {
+
+					space->is_in_unflushed_spaces = FALSE;
+
+					UT_LIST_REMOVE(
+						unflushed_spaces,
+						system->unflushed_spaces,
+						space);
+				}
+			}
+
+			if (space->purpose == FIL_TABLESPACE) {
+				fil_n_pending_tablespace_flushes--;
+			} else {
+				fil_n_pending_log_flushes--;
+			}
+		}
+
+		node = UT_LIST_GET_NEXT(chain, node);
+	}
+
+	space->n_pending_flushes--;
+
+	mutex_exit(&(system->mutex));
+}
+
+/**************************************************************************
+Flushes to disk the writes in file spaces of the given type possibly cached by
+the OS. */
+UNIV_INTERN
+void
+fil_flush_file_spaces(
+/*==================*/
+	ulint	purpose)	/* in: FIL_TABLESPACE, FIL_LOG */
+{
+	fil_system_t*	system	= fil_system;
+	fil_space_t*	space;
+	ulint*		space_ids;
+	ulint		n_space_ids;
+	ulint		i;
+
+	mutex_enter(&(system->mutex));
+
+	n_space_ids = UT_LIST_GET_LEN(system->unflushed_spaces);
+	if (n_space_ids == 0) {
+
+		mutex_exit(&system->mutex);
+		return;
+	}
+
+	/* Assemble a list of space ids to flush.  Previously, we
+	traversed system->unflushed_spaces and called UT_LIST_GET_NEXT()
+	on a space that was just removed from the list by fil_flush().
+	Thus, the space could be dropped and the memory overwritten. */
+	space_ids = mem_alloc(n_space_ids * sizeof *space_ids);
+
+	n_space_ids = 0;
+
+	for (space = UT_LIST_GET_FIRST(system->unflushed_spaces);
+	     space;
+	     space = UT_LIST_GET_NEXT(unflushed_spaces, space)) {
+
+		if (space->purpose == purpose && !space->is_being_deleted) {
+
+			space_ids[n_space_ids++] = space->id;
+		}
+	}
+
+	mutex_exit(&system->mutex);
+
+	/* Flush the spaces.  It will not hurt to call fil_flush() on
+	a non-existing space id. */
+	for (i = 0; i < n_space_ids; i++) {
+
+		fil_flush(space_ids[i]);
+	}
+
+	mem_free(space_ids);
+}
+
+/**********************************************************************
+Checks the consistency of the tablespace cache. */
+UNIV_INTERN
+ibool
+fil_validate(void)
+/*==============*/
+			/* out: TRUE if ok */
+{
+	fil_system_t*	system		= fil_system;
+	fil_space_t*	space;
+	fil_node_t*	fil_node;
+	ulint		n_open		= 0;
+	ulint		i;
+
+	mutex_enter(&(system->mutex));
+
+	/* Look for spaces in the hash table */
+
+	for (i = 0; i < hash_get_n_cells(system->spaces); i++) {
+
+		space = HASH_GET_FIRST(system->spaces, i);
+
+		while (space != NULL) {
+			UT_LIST_VALIDATE(chain, fil_node_t, space->chain);
+
+			fil_node = UT_LIST_GET_FIRST(space->chain);
+
+			while (fil_node != NULL) {
+				if (fil_node->n_pending > 0) {
+					ut_a(fil_node->open);
+				}
+
+				if (fil_node->open) {
+					n_open++;
+				}
+				fil_node = UT_LIST_GET_NEXT(chain, fil_node);
+			}
+			space = HASH_GET_NEXT(hash, space);
+		}
+	}
+
+	ut_a(system->n_open == n_open);
+
+	UT_LIST_VALIDATE(LRU, fil_node_t, system->LRU);
+
+	fil_node = UT_LIST_GET_FIRST(system->LRU);
+
+	while (fil_node != NULL) {
+		ut_a(fil_node->n_pending == 0);
+		ut_a(fil_node->open);
+		ut_a(fil_node->space->purpose == FIL_TABLESPACE);
+		ut_a(fil_node->space->id != 0);
+
+		fil_node = UT_LIST_GET_NEXT(LRU, fil_node);
+	}
+
+	mutex_exit(&(system->mutex));
+
+	return(TRUE);
+}
+
+/************************************************************************
+Returns TRUE if file address is undefined. */
+UNIV_INTERN
+ibool
+fil_addr_is_null(
+/*=============*/
+				/* out: TRUE if undefined */
+	fil_addr_t	addr)	/* in: address */
+{
+	return(addr.page == FIL_NULL);
+}
+
+/************************************************************************
+Accessor functions for a file page */
+UNIV_INTERN
+ulint
+fil_page_get_prev(const byte*	page)
+{
+	return(mach_read_from_4(page + FIL_PAGE_PREV));
+}
+
+UNIV_INTERN
+ulint
+fil_page_get_next(const byte*	page)
+{
+	return(mach_read_from_4(page + FIL_PAGE_NEXT));
+}
+
+/*************************************************************************
+Sets the file page type. */
+UNIV_INTERN
+void
+fil_page_set_type(
+/*==============*/
+	byte*	page,	/* in: file page */
+	ulint	type)	/* in: type */
+{
+	ut_ad(page);
+
+	mach_write_to_2(page + FIL_PAGE_TYPE, type);
+}
+
+/*************************************************************************
+Gets the file page type. */
+UNIV_INTERN
+ulint
+fil_page_get_type(
+/*==============*/
+				/* out: type; NOTE that if the type
+				has not been written to page, the return value
+				not defined */
+	const byte*	page)	/* in: file page */
+{
+	ut_ad(page);
+
+	return(mach_read_from_2(page + FIL_PAGE_TYPE));
+}
+
+/*************************************************************************
+Return local hash table informations. */
+
+ulint
+fil_system_hash_cells(void)
+/*=======================*/
+{
+       if (fil_system) {
+               return (fil_system->spaces->n_cells
+                       + fil_system->name_hash->n_cells);
+       } else {
+               return 0;
+       }
+}
+
+ulint
+fil_system_hash_nodes(void)
+/*=======================*/
+{
+       if (fil_system) {
+               return (UT_LIST_GET_LEN(fil_system->space_list)
+                       * (sizeof(fil_space_t) + MEM_BLOCK_HEADER_SIZE));
+       } else {
+               return 0;
+       }
+}
diff --git a/storage/xtradb/fsp/fsp0fsp.c b/storage/xtradb/fsp/fsp0fsp.c
new file mode 100644
index 00000000000..25d260daeea
--- /dev/null
+++ b/storage/xtradb/fsp/fsp0fsp.c
@@ -0,0 +1,4284 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/**********************************************************************
+File space management
+
+Created 11/29/1995 Heikki Tuuri
+***********************************************************************/
+
+#include "fsp0fsp.h"
+
+#ifdef UNIV_NONINL
+#include "fsp0fsp.ic"
+#endif
+
+#include "buf0buf.h"
+#include "fil0fil.h"
+#include "sync0sync.h"
+#include "mtr0log.h"
+#include "fut0fut.h"
+#include "ut0byte.h"
+#include "srv0srv.h"
+#include "page0zip.h"
+#include "ibuf0ibuf.h"
+#include "btr0btr.h"
+#include "btr0sea.h"
+#include "dict0boot.h"
+#include "dict0mem.h"
+#include "log0log.h"
+
+
+#define FSP_HEADER_OFFSET	FIL_PAGE_DATA	/* Offset of the space header
+						within a file page */
+
+/* The data structures in files are defined just as byte strings in C */
+typedef	byte	fsp_header_t;
+typedef	byte	xdes_t;
+
+/*			SPACE HEADER
+			============
+
+File space header data structure: this data structure is contained in the
+first page of a space. The space for this header is reserved in every extent
+descriptor page, but used only in the first. */
+
+/*-------------------------------------*/
+#define FSP_SPACE_ID		0	/* space id */
+#define FSP_NOT_USED		4	/* this field contained a value up to
+					which we know that the modifications
+					in the database have been flushed to
+					the file space; not used now */
+#define	FSP_SIZE		8	/* Current size of the space in
+					pages */
+#define	FSP_FREE_LIMIT		12	/* Minimum page number for which the
+					free list has not been initialized:
+					the pages >= this limit are, by
+					definition, free; note that in a
+					single-table tablespace where size
+					< 64 pages, this number is 64, i.e.,
+					we have initialized the space
+					about the first extent, but have not
+					physically allocted those pages to the
+					file */
+#define	FSP_SPACE_FLAGS		16	/* table->flags & ~DICT_TF_COMPACT */
+#define	FSP_FRAG_N_USED		20	/* number of used pages in the
+					FSP_FREE_FRAG list */
+#define	FSP_FREE		24	/* list of free extents */
+#define	FSP_FREE_FRAG		(24 + FLST_BASE_NODE_SIZE)
+					/* list of partially free extents not
+					belonging to any segment */
+#define	FSP_FULL_FRAG		(24 + 2 * FLST_BASE_NODE_SIZE)
+					/* list of full extents not belonging
+					to any segment */
+#define FSP_SEG_ID		(24 + 3 * FLST_BASE_NODE_SIZE)
+					/* 8 bytes which give the first unused
+					segment id */
+#define FSP_SEG_INODES_FULL	(32 + 3 * FLST_BASE_NODE_SIZE)
+					/* list of pages containing segment
+					headers, where all the segment inode
+					slots are reserved */
+#define FSP_SEG_INODES_FREE	(32 + 4 * FLST_BASE_NODE_SIZE)
+					/* list of pages containing segment
+					headers, where not all the segment
+					header slots are reserved */
+/*-------------------------------------*/
+/* File space header size */
+#define	FSP_HEADER_SIZE		(32 + 5 * FLST_BASE_NODE_SIZE)
+
+#define	FSP_FREE_ADD		4	/* this many free extents are added
+					to the free list from above
+					FSP_FREE_LIMIT at a time */
+
+/*			FILE SEGMENT INODE
+			==================
+
+Segment inode which is created for each segment in a tablespace. NOTE: in
+purge we assume that a segment having only one currently used page can be
+freed in a few steps, so that the freeing cannot fill the file buffer with
+bufferfixed file pages. */
+
+typedef	byte	fseg_inode_t;
+
+#define FSEG_INODE_PAGE_NODE	FSEG_PAGE_DATA
+					/* the list node for linking
+					segment inode pages */
+
+#define FSEG_ARR_OFFSET		(FSEG_PAGE_DATA + FLST_NODE_SIZE)
+/*-------------------------------------*/
+#define	FSEG_ID			0	/* 8 bytes of segment id: if this is
+					ut_dulint_zero, it means that the
+					header is unused */
+#define FSEG_NOT_FULL_N_USED	8
+					/* number of used segment pages in
+					the FSEG_NOT_FULL list */
+#define	FSEG_FREE		12
+					/* list of free extents of this
+					segment */
+#define	FSEG_NOT_FULL		(12 + FLST_BASE_NODE_SIZE)
+					/* list of partially free extents */
+#define	FSEG_FULL		(12 + 2 * FLST_BASE_NODE_SIZE)
+					/* list of full extents */
+#define	FSEG_MAGIC_N		(12 + 3 * FLST_BASE_NODE_SIZE)
+					/* magic number used in debugging */
+#define	FSEG_FRAG_ARR		(16 + 3 * FLST_BASE_NODE_SIZE)
+					/* array of individual pages
+					belonging to this segment in fsp
+					fragment extent lists */
+#define FSEG_FRAG_ARR_N_SLOTS	(FSP_EXTENT_SIZE / 2)
+					/* number of slots in the array for
+					the fragment pages */
+#define	FSEG_FRAG_SLOT_SIZE	4	/* a fragment page slot contains its
+					page number within space, FIL_NULL
+					means that the slot is not in use */
+/*-------------------------------------*/
+#define FSEG_INODE_SIZE					\
+	(16 + 3 * FLST_BASE_NODE_SIZE			\
+	 + FSEG_FRAG_ARR_N_SLOTS * FSEG_FRAG_SLOT_SIZE)
+
+#define FSP_SEG_INODES_PER_PAGE(zip_size)		\
+	(((zip_size ? zip_size : UNIV_PAGE_SIZE)	\
+	  - FSEG_ARR_OFFSET - 10) / FSEG_INODE_SIZE)
+				/* Number of segment inodes which fit on a
+				single page */
+
+#define FSEG_MAGIC_N_VALUE	97937874
+
+#define	FSEG_FILLFACTOR		8	/* If this value is x, then if
+					the number of unused but reserved
+					pages in a segment is less than
+					reserved pages * 1/x, and there are
+					at least FSEG_FRAG_LIMIT used pages,
+					then we allow a new empty extent to
+					be added to the segment in
+					fseg_alloc_free_page. Otherwise, we
+					use unused pages of the segment. */
+
+#define FSEG_FRAG_LIMIT		FSEG_FRAG_ARR_N_SLOTS
+					/* If the segment has >= this many
+					used pages, it may be expanded by
+					allocating extents to the segment;
+					until that only individual fragment
+					pages are allocated from the space */
+
+#define	FSEG_FREE_LIST_LIMIT	40	/* If the reserved size of a segment
+					is at least this many extents, we
+					allow extents to be put to the free
+					list of the extent: at most
+					FSEG_FREE_LIST_MAX_LEN many */
+#define	FSEG_FREE_LIST_MAX_LEN	4
+
+
+/*			EXTENT DESCRIPTOR
+			=================
+
+File extent descriptor data structure: contains bits to tell which pages in
+the extent are free and which contain old tuple version to clean. */
+
+/*-------------------------------------*/
+#define	XDES_ID			0	/* The identifier of the segment
+					to which this extent belongs */
+#define XDES_FLST_NODE		8	/* The list node data structure
+					for the descriptors */
+#define	XDES_STATE		(FLST_NODE_SIZE + 8)
+					/* contains state information
+					of the extent */
+#define	XDES_BITMAP		(FLST_NODE_SIZE + 12)
+					/* Descriptor bitmap of the pages
+					in the extent */
+/*-------------------------------------*/
+
+#define	XDES_BITS_PER_PAGE	2	/* How many bits are there per page */
+#define	XDES_FREE_BIT		0	/* Index of the bit which tells if
+					the page is free */
+#define	XDES_CLEAN_BIT		1	/* NOTE: currently not used!
+					Index of the bit which tells if
+					there are old versions of tuples
+					on the page */
+/* States of a descriptor */
+#define	XDES_FREE		1	/* extent is in free list of space */
+#define	XDES_FREE_FRAG		2	/* extent is in free fragment list of
+					space */
+#define	XDES_FULL_FRAG		3	/* extent is in full fragment list of
+					space */
+#define	XDES_FSEG		4	/* extent belongs to a segment */
+
+/* File extent data structure size in bytes. */
+#define	XDES_SIZE							\
+	(XDES_BITMAP + UT_BITS_IN_BYTES(FSP_EXTENT_SIZE * XDES_BITS_PER_PAGE))
+
+/* Offset of the descriptor array on a descriptor page */
+#define	XDES_ARR_OFFSET		(FSP_HEADER_OFFSET + FSP_HEADER_SIZE)
+
+/**************************************************************************
+Returns an extent to the free list of a space. */
+static
+void
+fsp_free_extent(
+/*============*/
+	ulint		space,	/* in: space id */
+	ulint		zip_size,/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	ulint		page,	/* in: page offset in the extent */
+	mtr_t*		mtr);	/* in: mtr */
+/**************************************************************************
+Frees an extent of a segment to the space free list. */
+static
+void
+fseg_free_extent(
+/*=============*/
+	fseg_inode_t*	seg_inode, /* in: segment inode */
+	ulint		space,	/* in: space id */
+	ulint		zip_size,/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	ulint		page,	/* in: page offset in the extent */
+	mtr_t*		mtr);	/* in: mtr handle */
+/**************************************************************************
+Calculates the number of pages reserved by a segment, and how
+many pages are currently used. */
+static
+ulint
+fseg_n_reserved_pages_low(
+/*======================*/
+				/* out: number of reserved pages */
+	fseg_inode_t*	header,	/* in: segment inode */
+	ulint*		used,	/* out: number of pages used (<= reserved) */
+	mtr_t*		mtr);	/* in: mtr handle */
+/************************************************************************
+Marks a page used. The page must reside within the extents of the given
+segment. */
+static
+void
+fseg_mark_page_used(
+/*================*/
+	fseg_inode_t*	seg_inode,/* in: segment inode */
+	ulint		space,	/* in: space id */
+	ulint		zip_size,/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	ulint		page,	/* in: page offset */
+	mtr_t*		mtr);	/* in: mtr */
+/**************************************************************************
+Returns the first extent descriptor for a segment. We think of the extent
+lists of the segment catenated in the order FSEG_FULL -> FSEG_NOT_FULL
+-> FSEG_FREE. */
+static
+xdes_t*
+fseg_get_first_extent(
+/*==================*/
+				/* out: the first extent descriptor, or NULL if
+				none */
+	fseg_inode_t*	inode,	/* in: segment inode */
+	ulint		space,	/* in: space id */
+	ulint		zip_size,/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	mtr_t*		mtr);	/* in: mtr */
+/**************************************************************************
+Puts new extents to the free list if
+there are free extents above the free limit. If an extent happens
+to contain an extent descriptor page, the extent is put to
+the FSP_FREE_FRAG list with the page marked as used. */
+static
+void
+fsp_fill_free_list(
+/*===============*/
+	ibool		init_space,	/* in: TRUE if this is a single-table
+					tablespace and we are only initing
+					the tablespace's first extent
+					descriptor page and ibuf bitmap page;
+					then we do not allocate more extents */
+	ulint		space,		/* in: space */
+	fsp_header_t*	header,		/* in: space header */
+	mtr_t*		mtr);		/* in: mtr */
+/**************************************************************************
+Allocates a single free page from a segment. This function implements
+the intelligent allocation strategy which tries to minimize file space
+fragmentation. */
+static
+ulint
+fseg_alloc_free_page_low(
+/*=====================*/
+				/* out: the allocated page number, FIL_NULL
+				if no page could be allocated */
+	ulint		space,	/* in: space */
+	ulint		zip_size,/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	fseg_inode_t*	seg_inode, /* in: segment inode */
+	ulint		hint,	/* in: hint of which page would be desirable */
+	byte		direction, /* in: if the new page is needed because
+				of an index page split, and records are
+				inserted there in order, into which
+				direction they go alphabetically: FSP_DOWN,
+				FSP_UP, FSP_NO_DIR */
+	mtr_t*		mtr);	/* in: mtr handle */
+
+
+/**************************************************************************
+Reads the file space size stored in the header page. */
+UNIV_INTERN
+ulint
+fsp_get_size_low(
+/*=============*/
+			/* out: tablespace size stored in the space header */
+	page_t*	page)	/* in: header page (page 0 in the tablespace) */
+{
+	return(mach_read_from_4(page + FSP_HEADER_OFFSET + FSP_SIZE));
+}
+
+/**************************************************************************
+Gets a pointer to the space header and x-locks its page. */
+UNIV_INLINE
+fsp_header_t*
+fsp_get_space_header(
+/*=================*/
+			/* out: pointer to the space header, page x-locked */
+	ulint	id,	/* in: space id */
+	ulint	zip_size,/* in: compressed page size in bytes
+			or 0 for uncompressed pages */
+	mtr_t*	mtr)	/* in: mtr */
+{
+	buf_block_t*	block;
+	fsp_header_t*	header;
+
+	ut_ad(ut_is_2pow(zip_size));
+	ut_ad(zip_size <= UNIV_PAGE_SIZE);
+	ut_ad(!zip_size || zip_size >= PAGE_ZIP_MIN_SIZE);
+	ut_ad(id || !zip_size);
+
+	block = buf_page_get(id, zip_size, 0, RW_X_LATCH, mtr);
+	header = FSP_HEADER_OFFSET + buf_block_get_frame(block);
+	buf_block_dbg_add_level(block, SYNC_FSP_PAGE);
+
+	ut_ad(id == mach_read_from_4(FSP_SPACE_ID + header));
+	ut_ad(zip_size == dict_table_flags_to_zip_size(
+		      mach_read_from_4(FSP_SPACE_FLAGS + header)));
+	return(header);
+}
+
+/**************************************************************************
+Gets a descriptor bit of a page. */
+UNIV_INLINE
+ibool
+xdes_get_bit(
+/*=========*/
+			/* out: TRUE if free */
+	xdes_t*	descr,	/* in: descriptor */
+	ulint	bit,	/* in: XDES_FREE_BIT or XDES_CLEAN_BIT */
+	ulint	offset,	/* in: page offset within extent:
+			0 ... FSP_EXTENT_SIZE - 1 */
+	mtr_t*	mtr)	/* in: mtr */
+{
+	ulint	index;
+	ulint	byte_index;
+	ulint	bit_index;
+
+	ut_ad(mtr_memo_contains_page(mtr, descr, MTR_MEMO_PAGE_X_FIX));
+	ut_ad((bit == XDES_FREE_BIT) || (bit == XDES_CLEAN_BIT));
+	ut_ad(offset < FSP_EXTENT_SIZE);
+
+	index = bit + XDES_BITS_PER_PAGE * offset;
+
+	byte_index = index / 8;
+	bit_index = index % 8;
+
+	return(ut_bit_get_nth(mtr_read_ulint(descr + XDES_BITMAP + byte_index,
+					     MLOG_1BYTE, mtr),
+			      bit_index));
+}
+
+/**************************************************************************
+Sets a descriptor bit of a page. */
+UNIV_INLINE
+void
+xdes_set_bit(
+/*=========*/
+	xdes_t*	descr,	/* in: descriptor */
+	ulint	bit,	/* in: XDES_FREE_BIT or XDES_CLEAN_BIT */
+	ulint	offset,	/* in: page offset within extent:
+			0 ... FSP_EXTENT_SIZE - 1 */
+	ibool	val,	/* in: bit value */
+	mtr_t*	mtr)	/* in: mtr */
+{
+	ulint	index;
+	ulint	byte_index;
+	ulint	bit_index;
+	ulint	descr_byte;
+
+	ut_ad(mtr_memo_contains_page(mtr, descr, MTR_MEMO_PAGE_X_FIX));
+	ut_ad((bit == XDES_FREE_BIT) || (bit == XDES_CLEAN_BIT));
+	ut_ad(offset < FSP_EXTENT_SIZE);
+
+	index = bit + XDES_BITS_PER_PAGE * offset;
+
+	byte_index = index / 8;
+	bit_index = index % 8;
+
+	descr_byte = mtr_read_ulint(descr + XDES_BITMAP + byte_index,
+				    MLOG_1BYTE, mtr);
+	descr_byte = ut_bit_set_nth(descr_byte, bit_index, val);
+
+	mlog_write_ulint(descr + XDES_BITMAP + byte_index, descr_byte,
+			 MLOG_1BYTE, mtr);
+}
+
+/**************************************************************************
+Looks for a descriptor bit having the desired value. Starts from hint
+and scans upward; at the end of the extent the search is wrapped to
+the start of the extent. */
+UNIV_INLINE
+ulint
+xdes_find_bit(
+/*==========*/
+			/* out: bit index of the bit, ULINT_UNDEFINED if not
+			found */
+	xdes_t*	descr,	/* in: descriptor */
+	ulint	bit,	/* in: XDES_FREE_BIT or XDES_CLEAN_BIT */
+	ibool	val,	/* in: desired bit value */
+	ulint	hint,	/* in: hint of which bit position would be desirable */
+	mtr_t*	mtr)	/* in: mtr */
+{
+	ulint	i;
+
+	ut_ad(descr && mtr);
+	ut_ad(val <= TRUE);
+	ut_ad(hint < FSP_EXTENT_SIZE);
+	ut_ad(mtr_memo_contains_page(mtr, descr, MTR_MEMO_PAGE_X_FIX));
+	for (i = hint; i < FSP_EXTENT_SIZE; i++) {
+		if (val == xdes_get_bit(descr, bit, i, mtr)) {
+
+			return(i);
+		}
+	}
+
+	for (i = 0; i < hint; i++) {
+		if (val == xdes_get_bit(descr, bit, i, mtr)) {
+
+			return(i);
+		}
+	}
+
+	return(ULINT_UNDEFINED);
+}
+
+/**************************************************************************
+Looks for a descriptor bit having the desired value. Scans the extent in
+a direction opposite to xdes_find_bit. */
+UNIV_INLINE
+ulint
+xdes_find_bit_downward(
+/*===================*/
+			/* out: bit index of the bit, ULINT_UNDEFINED if not
+			found */
+	xdes_t*	descr,	/* in: descriptor */
+	ulint	bit,	/* in: XDES_FREE_BIT or XDES_CLEAN_BIT */
+	ibool	val,	/* in: desired bit value */
+	ulint	hint,	/* in: hint of which bit position would be desirable */
+	mtr_t*	mtr)	/* in: mtr */
+{
+	ulint	i;
+
+	ut_ad(descr && mtr);
+	ut_ad(val <= TRUE);
+	ut_ad(hint < FSP_EXTENT_SIZE);
+	ut_ad(mtr_memo_contains_page(mtr, descr, MTR_MEMO_PAGE_X_FIX));
+	for (i = hint + 1; i > 0; i--) {
+		if (val == xdes_get_bit(descr, bit, i - 1, mtr)) {
+
+			return(i - 1);
+		}
+	}
+
+	for (i = FSP_EXTENT_SIZE - 1; i > hint; i--) {
+		if (val == xdes_get_bit(descr, bit, i, mtr)) {
+
+			return(i);
+		}
+	}
+
+	return(ULINT_UNDEFINED);
+}
+
+/**************************************************************************
+Returns the number of used pages in a descriptor. */
+UNIV_INLINE
+ulint
+xdes_get_n_used(
+/*============*/
+			/* out: number of pages used */
+	xdes_t*	descr,	/* in: descriptor */
+	mtr_t*	mtr)	/* in: mtr */
+{
+	ulint	i;
+	ulint	count	= 0;
+
+	ut_ad(descr && mtr);
+	ut_ad(mtr_memo_contains_page(mtr, descr, MTR_MEMO_PAGE_X_FIX));
+	for (i = 0; i < FSP_EXTENT_SIZE; i++) {
+		if (FALSE == xdes_get_bit(descr, XDES_FREE_BIT, i, mtr)) {
+			count++;
+		}
+	}
+
+	return(count);
+}
+
+/**************************************************************************
+Returns true if extent contains no used pages. */
+UNIV_INLINE
+ibool
+xdes_is_free(
+/*=========*/
+			/* out: TRUE if totally free */
+	xdes_t*	descr,	/* in: descriptor */
+	mtr_t*	mtr)	/* in: mtr */
+{
+	if (0 == xdes_get_n_used(descr, mtr)) {
+
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
+
+/**************************************************************************
+Returns true if extent contains no free pages. */
+UNIV_INLINE
+ibool
+xdes_is_full(
+/*=========*/
+			/* out: TRUE if full */
+	xdes_t*	descr,	/* in: descriptor */
+	mtr_t*	mtr)	/* in: mtr */
+{
+	if (FSP_EXTENT_SIZE == xdes_get_n_used(descr, mtr)) {
+
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
+
+/**************************************************************************
+Sets the state of an xdes. */
+UNIV_INLINE
+void
+xdes_set_state(
+/*===========*/
+	xdes_t*	descr,	/* in: descriptor */
+	ulint	state,	/* in: state to set */
+	mtr_t*	mtr)	/* in: mtr handle */
+{
+	ut_ad(descr && mtr);
+	ut_ad(state >= XDES_FREE);
+	ut_ad(state <= XDES_FSEG);
+	ut_ad(mtr_memo_contains_page(mtr, descr, MTR_MEMO_PAGE_X_FIX));
+
+	mlog_write_ulint(descr + XDES_STATE, state, MLOG_4BYTES, mtr);
+}
+
+/**************************************************************************
+Gets the state of an xdes. */
+UNIV_INLINE
+ulint
+xdes_get_state(
+/*===========*/
+			/* out: state */
+	xdes_t*	descr,	/* in: descriptor */
+	mtr_t*	mtr)	/* in: mtr handle */
+{
+	ulint	state;
+
+	ut_ad(descr && mtr);
+	ut_ad(mtr_memo_contains_page(mtr, descr, MTR_MEMO_PAGE_X_FIX));
+
+	state = mtr_read_ulint(descr + XDES_STATE, MLOG_4BYTES, mtr);
+	ut_ad(state - 1 < XDES_FSEG);
+	return(state);
+}
+
+/**************************************************************************
+Inits an extent descriptor to the free and clean state. */
+UNIV_INLINE
+void
+xdes_init(
+/*======*/
+	xdes_t*	descr,	/* in: descriptor */
+	mtr_t*	mtr)	/* in: mtr */
+{
+	ulint	i;
+
+	ut_ad(descr && mtr);
+	ut_ad(mtr_memo_contains_page(mtr, descr, MTR_MEMO_PAGE_X_FIX));
+	ut_ad((XDES_SIZE - XDES_BITMAP) % 4 == 0);
+
+	for (i = XDES_BITMAP; i < XDES_SIZE; i += 4) {
+		mlog_write_ulint(descr + i, 0xFFFFFFFFUL, MLOG_4BYTES, mtr);
+	}
+
+	xdes_set_state(descr, XDES_FREE, mtr);
+}
+
+/************************************************************************
+Calculates the page where the descriptor of a page resides. */
+UNIV_INLINE
+ulint
+xdes_calc_descriptor_page(
+/*======================*/
+				/* out: descriptor page offset */
+	ulint	zip_size,	/* in: compressed page size in bytes;
+				0 for uncompressed pages */
+	ulint	offset)		/* in: page offset */
+{
+#if UNIV_PAGE_SIZE <= XDES_ARR_OFFSET \
+		+ (UNIV_PAGE_SIZE / FSP_EXTENT_SIZE) * XDES_SIZE
+# error
+#endif
+#if PAGE_ZIP_MIN_SIZE <= XDES_ARR_OFFSET \
+		+ (PAGE_ZIP_MIN_SIZE / FSP_EXTENT_SIZE) * XDES_SIZE
+# error
+#endif
+	ut_ad(ut_is_2pow(zip_size));
+
+	if (!zip_size) {
+		return(ut_2pow_round(offset, UNIV_PAGE_SIZE));
+	} else {
+		ut_ad(zip_size > XDES_ARR_OFFSET
+		      + (zip_size / FSP_EXTENT_SIZE) * XDES_SIZE);
+		return(ut_2pow_round(offset, zip_size));
+	}
+}
+
+/************************************************************************
+Calculates the descriptor index within a descriptor page. */
+UNIV_INLINE
+ulint
+xdes_calc_descriptor_index(
+/*=======================*/
+				/* out: descriptor index */
+	ulint	zip_size,	/* in: compressed page size in bytes;
+				0 for uncompressed pages */
+	ulint	offset)		/* in: page offset */
+{
+	ut_ad(ut_is_2pow(zip_size));
+
+	if (!zip_size) {
+		return(ut_2pow_remainder(offset, UNIV_PAGE_SIZE)
+		       / FSP_EXTENT_SIZE);
+	} else {
+		return(ut_2pow_remainder(offset, zip_size) / FSP_EXTENT_SIZE);
+	}
+}
+
+/************************************************************************
+Gets pointer to a the extent descriptor of a page. The page where the extent
+descriptor resides is x-locked. If the page offset is equal to the free limit
+of the space, adds new extents from above the free limit to the space free
+list, if not free limit == space size. This adding is necessary to make the
+descriptor defined, as they are uninitialized above the free limit. */
+UNIV_INLINE
+xdes_t*
+xdes_get_descriptor_with_space_hdr(
+/*===============================*/
+				/* out: pointer to the extent descriptor,
+				NULL if the page does not exist in the
+				space or if offset > free limit */
+	fsp_header_t*	sp_header,/* in: space header, x-latched */
+	ulint		space,	/* in: space id */
+	ulint		offset,	/* in: page offset;
+				if equal to the free limit,
+				we try to add new extents to
+				the space free list */
+	mtr_t*		mtr)	/* in: mtr handle */
+{
+	ulint	limit;
+	ulint	size;
+	ulint	zip_size;
+	ulint	descr_page_no;
+	page_t*	descr_page;
+
+	ut_ad(mtr);
+	ut_ad(mtr_memo_contains(mtr, fil_space_get_latch(space, NULL),
+				MTR_MEMO_X_LOCK));
+	ut_ad(mtr_memo_contains_page(mtr, sp_header, MTR_MEMO_PAGE_S_FIX)
+	      || mtr_memo_contains_page(mtr, sp_header, MTR_MEMO_PAGE_X_FIX));
+	ut_ad(page_offset(sp_header) == FSP_HEADER_OFFSET);
+	/* Read free limit and space size */
+	limit = mach_read_from_4(sp_header + FSP_FREE_LIMIT);
+	size  = mach_read_from_4(sp_header + FSP_SIZE);
+	zip_size = dict_table_flags_to_zip_size(
+		mach_read_from_4(sp_header + FSP_SPACE_FLAGS));
+
+	/* If offset is >= size or > limit, return NULL */
+
+	if ((offset >= size) || (offset > limit)) {
+
+		return(NULL);
+	}
+
+	/* If offset is == limit, fill free list of the space. */
+
+	if (offset == limit) {
+		fsp_fill_free_list(FALSE, space, sp_header, mtr);
+	}
+
+	descr_page_no = xdes_calc_descriptor_page(zip_size, offset);
+
+	if (descr_page_no == 0) {
+		/* It is on the space header page */
+
+		descr_page = page_align(sp_header);
+	} else {
+		buf_block_t*	block;
+
+		block = buf_page_get(space, zip_size, descr_page_no,
+				     RW_X_LATCH, mtr);
+		buf_block_dbg_add_level(block, SYNC_FSP_PAGE);
+
+		descr_page = buf_block_get_frame(block);
+	}
+
+	return(descr_page + XDES_ARR_OFFSET
+	       + XDES_SIZE * xdes_calc_descriptor_index(zip_size, offset));
+}
+
+/************************************************************************
+Gets pointer to a the extent descriptor of a page. The page where the
+extent descriptor resides is x-locked. If the page offset is equal to
+the free limit of the space, adds new extents from above the free limit
+to the space free list, if not free limit == space size. This adding
+is necessary to make the descriptor defined, as they are uninitialized
+above the free limit. */
+static
+xdes_t*
+xdes_get_descriptor(
+/*================*/
+			/* out: pointer to the extent descriptor, NULL if the
+			page does not exist in the space or if offset > free
+			limit */
+	ulint	space,	/* in: space id */
+	ulint	zip_size,/* in: compressed page size in bytes
+			or 0 for uncompressed pages */
+	ulint	offset,	/* in: page offset; if equal to the free limit,
+			we try to add new extents to the space free list */
+	mtr_t*	mtr)	/* in: mtr handle */
+{
+	buf_block_t*	block;
+	fsp_header_t*	sp_header;
+
+	block = buf_page_get(space, zip_size, 0, RW_X_LATCH, mtr);
+	buf_block_dbg_add_level(block, SYNC_FSP_PAGE);
+
+	sp_header = FSP_HEADER_OFFSET + buf_block_get_frame(block);
+	return(xdes_get_descriptor_with_space_hdr(sp_header, space, offset,
+						  mtr));
+}
+
+/************************************************************************
+Gets pointer to a the extent descriptor if the file address
+of the descriptor list node is known. The page where the
+extent descriptor resides is x-locked. */
+UNIV_INLINE
+xdes_t*
+xdes_lst_get_descriptor(
+/*====================*/
+				/* out: pointer to the extent descriptor */
+	ulint		space,	/* in: space id */
+	ulint		zip_size,/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	fil_addr_t	lst_node,/* in: file address of the list node
+				contained in the descriptor */
+	mtr_t*		mtr)	/* in: mtr handle */
+{
+	xdes_t*	descr;
+
+	ut_ad(mtr);
+	ut_ad(mtr_memo_contains(mtr, fil_space_get_latch(space, NULL),
+				MTR_MEMO_X_LOCK));
+	descr = fut_get_ptr(space, zip_size, lst_node, RW_X_LATCH, mtr)
+		- XDES_FLST_NODE;
+
+	return(descr);
+}
+
+/************************************************************************
+Returns page offset of the first page in extent described by a descriptor. */
+UNIV_INLINE
+ulint
+xdes_get_offset(
+/*============*/
+			/* out: offset of the first page in extent */
+	xdes_t*	descr)	/* in: extent descriptor */
+{
+	ut_ad(descr);
+
+	return(page_get_page_no(page_align(descr))
+	       + ((page_offset(descr) - XDES_ARR_OFFSET) / XDES_SIZE)
+	       * FSP_EXTENT_SIZE);
+}
+
+/***************************************************************
+Inits a file page whose prior contents should be ignored. */
+static
+void
+fsp_init_file_page_low(
+/*===================*/
+	buf_block_t*	block)	/* in: pointer to a page */
+{
+	page_t*		page	= buf_block_get_frame(block);
+	page_zip_des_t*	page_zip= buf_block_get_page_zip(block);
+
+	block->check_index_page_at_flush = FALSE;
+
+	if (UNIV_LIKELY_NULL(page_zip)) {
+		memset(page, 0, UNIV_PAGE_SIZE);
+		memset(page_zip->data, 0, page_zip_get_size(page_zip));
+		mach_write_to_4(page + FIL_PAGE_OFFSET,
+				buf_block_get_page_no(block));
+		mach_write_to_4(page
+				+ FIL_PAGE_ARCH_LOG_NO_OR_SPACE_ID,
+				buf_block_get_space(block));
+		memcpy(page_zip->data + FIL_PAGE_OFFSET,
+		       page + FIL_PAGE_OFFSET, 4);
+		memcpy(page_zip->data + FIL_PAGE_ARCH_LOG_NO_OR_SPACE_ID,
+		       page + FIL_PAGE_ARCH_LOG_NO_OR_SPACE_ID, 4);
+		return;
+	}
+
+#ifdef UNIV_BASIC_LOG_DEBUG
+	memset(page, 0xff, UNIV_PAGE_SIZE);
+#endif
+	mach_write_to_4(page + FIL_PAGE_OFFSET, buf_block_get_page_no(block));
+	memset(page + FIL_PAGE_LSN, 0, 8);
+	mach_write_to_4(page + FIL_PAGE_ARCH_LOG_NO_OR_SPACE_ID,
+			buf_block_get_space(block));
+	memset(page + UNIV_PAGE_SIZE - FIL_PAGE_END_LSN_OLD_CHKSUM, 0, 8);
+}
+
+/***************************************************************
+Inits a file page whose prior contents should be ignored. */
+static
+void
+fsp_init_file_page(
+/*===============*/
+	buf_block_t*	block,	/* in: pointer to a page */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	fsp_init_file_page_low(block);
+
+	mlog_write_initial_log_record(buf_block_get_frame(block),
+				      MLOG_INIT_FILE_PAGE, mtr);
+}
+
+/***************************************************************
+Parses a redo log record of a file page init. */
+UNIV_INTERN
+byte*
+fsp_parse_init_file_page(
+/*=====================*/
+				/* out: end of log record or NULL */
+	byte*		ptr,	/* in: buffer */
+	byte*		end_ptr __attribute__((unused)), /* in: buffer end */
+	buf_block_t*	block)	/* in: block or NULL */
+{
+	ut_ad(ptr && end_ptr);
+
+	if (block) {
+		fsp_init_file_page_low(block);
+	}
+
+	return(ptr);
+}
+
+/**************************************************************************
+Initializes the fsp system. */
+UNIV_INTERN
+void
+fsp_init(void)
+/*==========*/
+{
+	/* Does nothing at the moment */
+}
+
+/**************************************************************************
+Writes the space id and compressed page size to a tablespace header.
+This function is used past the buffer pool when we in fil0fil.c create
+a new single-table tablespace. */
+UNIV_INTERN
+void
+fsp_header_init_fields(
+/*===================*/
+	page_t*	page,		/* in/out: first page in the space */
+	ulint	space_id,	/* in: space id */
+	ulint	flags)		/* in: tablespace flags (FSP_SPACE_FLAGS):
+				0, or table->flags if newer than COMPACT */
+{
+	/* The tablespace flags (FSP_SPACE_FLAGS) should be 0 for
+	ROW_FORMAT=COMPACT (table->flags == DICT_TF_COMPACT) and
+	ROW_FORMAT=REDUNDANT (table->flags == 0).  For any other
+	format, the tablespace flags should equal table->flags. */
+	ut_a(flags != DICT_TF_COMPACT);
+
+	mach_write_to_4(FSP_HEADER_OFFSET + FSP_SPACE_ID + page,
+			space_id);
+	mach_write_to_4(FSP_HEADER_OFFSET + FSP_SPACE_FLAGS + page,
+			flags);
+}
+
+/**************************************************************************
+Initializes the space header of a new created space and creates also the
+insert buffer tree root if space == 0. */
+UNIV_INTERN
+void
+fsp_header_init(
+/*============*/
+	ulint	space,		/* in: space id */
+	ulint	size,		/* in: current size in blocks */
+	mtr_t*	mtr)		/* in: mini-transaction handle */
+{
+	fsp_header_t*	header;
+	buf_block_t*	block;
+	page_t*		page;
+	ulint		flags;
+	ulint		zip_size;
+
+	ut_ad(mtr);
+
+	mtr_x_lock(fil_space_get_latch(space, &flags), mtr);
+
+	zip_size = dict_table_flags_to_zip_size(flags);
+	block = buf_page_create(space, 0, zip_size, mtr);
+	buf_page_get(space, zip_size, 0, RW_X_LATCH, mtr);
+	buf_block_dbg_add_level(block, SYNC_FSP_PAGE);
+
+	/* The prior contents of the file page should be ignored */
+
+	fsp_init_file_page(block, mtr);
+	page = buf_block_get_frame(block);
+
+	mlog_write_ulint(page + FIL_PAGE_TYPE, FIL_PAGE_TYPE_FSP_HDR,
+			 MLOG_2BYTES, mtr);
+
+	header = FSP_HEADER_OFFSET + page;
+
+	mlog_write_ulint(header + FSP_SPACE_ID, space, MLOG_4BYTES, mtr);
+	mlog_write_ulint(header + FSP_NOT_USED, 0, MLOG_4BYTES, mtr);
+
+	mlog_write_ulint(header + FSP_SIZE, size, MLOG_4BYTES, mtr);
+	mlog_write_ulint(header + FSP_FREE_LIMIT, 0, MLOG_4BYTES, mtr);
+	mlog_write_ulint(header + FSP_SPACE_FLAGS, flags,
+			 MLOG_4BYTES, mtr);
+	mlog_write_ulint(header + FSP_FRAG_N_USED, 0, MLOG_4BYTES, mtr);
+
+	flst_init(header + FSP_FREE, mtr);
+	flst_init(header + FSP_FREE_FRAG, mtr);
+	flst_init(header + FSP_FULL_FRAG, mtr);
+	flst_init(header + FSP_SEG_INODES_FULL, mtr);
+	flst_init(header + FSP_SEG_INODES_FREE, mtr);
+
+	mlog_write_dulint(header + FSP_SEG_ID, ut_dulint_create(0, 1), mtr);
+	if (space == 0) {
+		fsp_fill_free_list(FALSE, space, header, mtr);
+		btr_create(DICT_CLUSTERED | DICT_UNIVERSAL | DICT_IBUF,
+			   0, 0, ut_dulint_add(DICT_IBUF_ID_MIN, space),
+			   srv_sys->dummy_ind1, mtr);
+	} else {
+		fsp_fill_free_list(TRUE, space, header, mtr);
+	}
+}
+
+/**************************************************************************
+Reads the space id from the first page of a tablespace. */
+UNIV_INTERN
+ulint
+fsp_header_get_space_id(
+/*====================*/
+				/* out: space id, ULINT UNDEFINED if error */
+	const page_t*	page)	/* in: first page of a tablespace */
+{
+	ulint	fsp_id;
+	ulint	id;
+
+	fsp_id = mach_read_from_4(FSP_HEADER_OFFSET + page + FSP_SPACE_ID);
+
+	id = mach_read_from_4(page + FIL_PAGE_ARCH_LOG_NO_OR_SPACE_ID);
+
+	if (id != fsp_id) {
+		fprintf(stderr,
+			"InnoDB: Error: space id in fsp header %lu,"
+			" but in the page header %lu\n",
+			(ulong) fsp_id, (ulong) id);
+
+		return(ULINT_UNDEFINED);
+	}
+
+	return(id);
+}
+
+/**************************************************************************
+Reads the space flags from the first page of a tablespace. */
+UNIV_INTERN
+ulint
+fsp_header_get_flags(
+/*=================*/
+				/* out: flags */
+	const page_t*	page)	/* in: first page of a tablespace */
+{
+	ut_ad(!page_offset(page));
+
+	return(mach_read_from_4(FSP_HEADER_OFFSET + FSP_SPACE_FLAGS + page));
+}
+
+/**************************************************************************
+Reads the compressed page size from the first page of a tablespace. */
+UNIV_INTERN
+ulint
+fsp_header_get_zip_size(
+/*====================*/
+				/* out: compressed page size in bytes,
+				or 0 if uncompressed */
+	const page_t*	page)	/* in: first page of a tablespace */
+{
+	ulint	flags = fsp_header_get_flags(page);
+
+	return(dict_table_flags_to_zip_size(flags));
+}
+
+/**************************************************************************
+Increases the space size field of a space. */
+UNIV_INTERN
+void
+fsp_header_inc_size(
+/*================*/
+	ulint	space,	/* in: space id */
+	ulint	size_inc,/* in: size increment in pages */
+	mtr_t*	mtr)	/* in: mini-transaction handle */
+{
+	fsp_header_t*	header;
+	ulint		size;
+	ulint		flags;
+
+	ut_ad(mtr);
+
+	mtr_x_lock(fil_space_get_latch(space, &flags), mtr);
+
+	header = fsp_get_space_header(space,
+				      dict_table_flags_to_zip_size(flags),
+				      mtr);
+
+	size = mtr_read_ulint(header + FSP_SIZE, MLOG_4BYTES, mtr);
+
+	mlog_write_ulint(header + FSP_SIZE, size + size_inc, MLOG_4BYTES,
+			 mtr);
+}
+
+/**************************************************************************
+Gets the current free limit of the system tablespace.  The free limit
+means the place of the first page which has never been put to the the
+free list for allocation.  The space above that address is initialized
+to zero.  Sets also the global variable log_fsp_current_free_limit. */
+UNIV_INTERN
+ulint
+fsp_header_get_free_limit(void)
+/*===========================*/
+			/* out: free limit in megabytes */
+{
+	fsp_header_t*	header;
+	ulint		limit;
+	mtr_t		mtr;
+
+	mtr_start(&mtr);
+
+	mtr_x_lock(fil_space_get_latch(0, NULL), &mtr);
+
+	header = fsp_get_space_header(0, 0, &mtr);
+
+	limit = mtr_read_ulint(header + FSP_FREE_LIMIT, MLOG_4BYTES, &mtr);
+
+	limit /= ((1024 * 1024) / UNIV_PAGE_SIZE);
+
+	log_fsp_current_free_limit_set_and_checkpoint(limit);
+
+	mtr_commit(&mtr);
+
+	return(limit);
+}
+
+/**************************************************************************
+Gets the size of the system tablespace from the tablespace header.  If
+we do not have an auto-extending data file, this should be equal to
+the size of the data files.  If there is an auto-extending data file,
+this can be smaller. */
+UNIV_INTERN
+ulint
+fsp_header_get_tablespace_size(void)
+/*================================*/
+			/* out: size in pages */
+{
+	fsp_header_t*	header;
+	ulint		size;
+	mtr_t		mtr;
+
+	mtr_start(&mtr);
+
+	mtr_x_lock(fil_space_get_latch(0, NULL), &mtr);
+
+	header = fsp_get_space_header(0, 0, &mtr);
+
+	size = mtr_read_ulint(header + FSP_SIZE, MLOG_4BYTES, &mtr);
+
+	mtr_commit(&mtr);
+
+	return(size);
+}
+
+/***************************************************************************
+Tries to extend a single-table tablespace so that a page would fit in the
+data file. */
+static
+ibool
+fsp_try_extend_data_file_with_pages(
+/*================================*/
+					/* out: TRUE if success */
+	ulint		space,		/* in: space */
+	ulint		page_no,	/* in: page number */
+	fsp_header_t*	header,		/* in: space header */
+	mtr_t*		mtr)		/* in: mtr */
+{
+	ibool	success;
+	ulint	actual_size;
+	ulint	size;
+
+	ut_a(space != 0);
+
+	size = mtr_read_ulint(header + FSP_SIZE, MLOG_4BYTES, mtr);
+
+	ut_a(page_no >= size);
+
+	success = fil_extend_space_to_desired_size(&actual_size, space,
+						   page_no + 1);
+	/* actual_size now has the space size in pages; it may be less than
+	we wanted if we ran out of disk space */
+
+	mlog_write_ulint(header + FSP_SIZE, actual_size, MLOG_4BYTES, mtr);
+
+	return(success);
+}
+
+/***************************************************************************
+Tries to extend the last data file of a tablespace if it is auto-extending. */
+static
+ibool
+fsp_try_extend_data_file(
+/*=====================*/
+					/* out: FALSE if not auto-extending */
+	ulint*		actual_increase,/* out: actual increase in pages, where
+					we measure the tablespace size from
+					what the header field says; it may be
+					the actual file size rounded down to
+					megabyte */
+	ulint		space,		/* in: space */
+	fsp_header_t*	header,		/* in: space header */
+	mtr_t*		mtr)		/* in: mtr */
+{
+	ulint	size;
+	ulint	zip_size;
+	ulint	new_size;
+	ulint	old_size;
+	ulint	size_increase;
+	ulint	actual_size;
+	ibool	success;
+
+	*actual_increase = 0;
+
+	if (space == 0 && !srv_auto_extend_last_data_file) {
+
+		return(FALSE);
+	}
+
+	size = mtr_read_ulint(header + FSP_SIZE, MLOG_4BYTES, mtr);
+	zip_size = dict_table_flags_to_zip_size(
+		mach_read_from_4(header + FSP_SPACE_FLAGS));
+
+	old_size = size;
+
+	if (space == 0) {
+		if (!srv_last_file_size_max) {
+			size_increase = SRV_AUTO_EXTEND_INCREMENT;
+		} else {
+			if (srv_last_file_size_max
+			    < srv_data_file_sizes[srv_n_data_files - 1]) {
+
+				fprintf(stderr,
+					"InnoDB: Error: Last data file size"
+					" is %lu, max size allowed %lu\n",
+					(ulong) srv_data_file_sizes[
+						srv_n_data_files - 1],
+					(ulong) srv_last_file_size_max);
+			}
+
+			size_increase = srv_last_file_size_max
+				- srv_data_file_sizes[srv_n_data_files - 1];
+			if (size_increase > SRV_AUTO_EXTEND_INCREMENT) {
+				size_increase = SRV_AUTO_EXTEND_INCREMENT;
+			}
+		}
+	} else {
+		/* We extend single-table tablespaces first one extent
+		at a time, but for bigger tablespaces more. It is not
+		enough to extend always by one extent, because some
+		extents are frag page extents. */
+		ulint	extent_size;	/* one megabyte, in pages */
+
+		if (!zip_size) {
+			extent_size = FSP_EXTENT_SIZE;
+		} else {
+			extent_size = FSP_EXTENT_SIZE
+				* UNIV_PAGE_SIZE / zip_size;
+		}
+
+		if (size < extent_size) {
+			/* Let us first extend the file to extent_size */
+			success = fsp_try_extend_data_file_with_pages(
+				space, extent_size - 1, header, mtr);
+			if (!success) {
+				new_size = mtr_read_ulint(header + FSP_SIZE,
+							  MLOG_4BYTES, mtr);
+
+				*actual_increase = new_size - old_size;
+
+				return(FALSE);
+			}
+
+			size = extent_size;
+		}
+
+		if (size < 32 * extent_size) {
+			size_increase = extent_size;
+		} else {
+			/* Below in fsp_fill_free_list() we assume
+			that we add at most FSP_FREE_ADD extents at
+			a time */
+			size_increase = FSP_FREE_ADD * extent_size;
+		}
+	}
+
+	if (size_increase == 0) {
+
+		return(TRUE);
+	}
+
+	success = fil_extend_space_to_desired_size(&actual_size, space,
+						   size + size_increase);
+	/* We ignore any fragments of a full megabyte when storing the size
+	to the space header */
+
+	if (!zip_size) {
+		new_size = ut_calc_align_down(actual_size,
+					      (1024 * 1024) / UNIV_PAGE_SIZE);
+	} else {
+		new_size = ut_calc_align_down(actual_size,
+					      (1024 * 1024) / zip_size);
+	}
+	mlog_write_ulint(header + FSP_SIZE, new_size, MLOG_4BYTES, mtr);
+
+	*actual_increase = new_size - old_size;
+
+	return(TRUE);
+}
+
+/**************************************************************************
+Puts new extents to the free list if there are free extents above the free
+limit. If an extent happens to contain an extent descriptor page, the extent
+is put to the FSP_FREE_FRAG list with the page marked as used. */
+static
+void
+fsp_fill_free_list(
+/*===============*/
+	ibool		init_space,	/* in: TRUE if this is a single-table
+					tablespace and we are only initing
+					the tablespace's first extent
+					descriptor page and ibuf bitmap page;
+					then we do not allocate more extents */
+	ulint		space,		/* in: space */
+	fsp_header_t*	header,		/* in: space header */
+	mtr_t*		mtr)		/* in: mtr */
+{
+	ulint	limit;
+	ulint	size;
+	ulint	zip_size;
+	xdes_t*	descr;
+	ulint	count		= 0;
+	ulint	frag_n_used;
+	ulint	actual_increase;
+	ulint	i;
+	mtr_t	ibuf_mtr;
+
+	ut_ad(header && mtr);
+	ut_ad(page_offset(header) == FSP_HEADER_OFFSET);
+
+	/* Check if we can fill free list from above the free list limit */
+	size = mtr_read_ulint(header + FSP_SIZE, MLOG_4BYTES, mtr);
+	limit = mtr_read_ulint(header + FSP_FREE_LIMIT, MLOG_4BYTES, mtr);
+
+	zip_size = dict_table_flags_to_zip_size(
+		mach_read_from_4(FSP_SPACE_FLAGS + header));
+	ut_a(ut_is_2pow(zip_size));
+	ut_a(zip_size <= UNIV_PAGE_SIZE);
+	ut_a(!zip_size || zip_size >= PAGE_ZIP_MIN_SIZE);
+
+	if (space == 0 && srv_auto_extend_last_data_file
+	    && size < limit + FSP_EXTENT_SIZE * FSP_FREE_ADD) {
+
+		/* Try to increase the last data file size */
+		fsp_try_extend_data_file(&actual_increase, space, header, mtr);
+		size = mtr_read_ulint(header + FSP_SIZE, MLOG_4BYTES, mtr);
+	}
+
+	if (space != 0 && !init_space
+	    && size < limit + FSP_EXTENT_SIZE * FSP_FREE_ADD) {
+
+		/* Try to increase the .ibd file size */
+		fsp_try_extend_data_file(&actual_increase, space, header, mtr);
+		size = mtr_read_ulint(header + FSP_SIZE, MLOG_4BYTES, mtr);
+	}
+
+	i = limit;
+
+	while ((init_space && i < 1)
+	       || ((i + FSP_EXTENT_SIZE <= size) && (count < FSP_FREE_ADD))) {
+
+		ibool	init_xdes;
+		if (zip_size) {
+			init_xdes = ut_2pow_remainder(i, zip_size) == 0;
+		} else {
+			init_xdes = ut_2pow_remainder(i, UNIV_PAGE_SIZE) == 0;
+		}
+
+		mlog_write_ulint(header + FSP_FREE_LIMIT, i + FSP_EXTENT_SIZE,
+				 MLOG_4BYTES, mtr);
+
+		/* Update the free limit info in the log system and make
+		a checkpoint */
+		if (space == 0) {
+			ut_a(!zip_size);
+			log_fsp_current_free_limit_set_and_checkpoint(
+				(i + FSP_EXTENT_SIZE)
+				/ ((1024 * 1024) / UNIV_PAGE_SIZE));
+		}
+
+		if (UNIV_UNLIKELY(init_xdes)) {
+
+			buf_block_t*	block;
+
+			/* We are going to initialize a new descriptor page
+			and a new ibuf bitmap page: the prior contents of the
+			pages should be ignored. */
+
+			if (i > 0) {
+				block = buf_page_create(
+					space, i, zip_size, mtr);
+				buf_page_get(space, zip_size, i,
+					     RW_X_LATCH, mtr);
+				buf_block_dbg_add_level(block,
+							SYNC_FSP_PAGE);
+
+				fsp_init_file_page(block, mtr);
+				mlog_write_ulint(buf_block_get_frame(block)
+						 + FIL_PAGE_TYPE,
+						 FIL_PAGE_TYPE_XDES,
+						 MLOG_2BYTES, mtr);
+			}
+
+			/* Initialize the ibuf bitmap page in a separate
+			mini-transaction because it is low in the latching
+			order, and we must be able to release its latch
+			before returning from the fsp routine */
+
+			mtr_start(&ibuf_mtr);
+
+			block = buf_page_create(space,
+						    i + FSP_IBUF_BITMAP_OFFSET,
+						    zip_size, &ibuf_mtr);
+			buf_page_get(space, zip_size,
+				     i + FSP_IBUF_BITMAP_OFFSET,
+				     RW_X_LATCH, &ibuf_mtr);
+			buf_block_dbg_add_level(block, SYNC_FSP_PAGE);
+
+			fsp_init_file_page(block, &ibuf_mtr);
+
+			ibuf_bitmap_page_init(block, &ibuf_mtr);
+
+			mtr_commit(&ibuf_mtr);
+		}
+
+		descr = xdes_get_descriptor_with_space_hdr(header, space, i,
+							   mtr);
+		xdes_init(descr, mtr);
+
+#if UNIV_PAGE_SIZE % FSP_EXTENT_SIZE
+# error "UNIV_PAGE_SIZE % FSP_EXTENT_SIZE != 0"
+#endif
+#if PAGE_ZIP_MIN_SIZE % FSP_EXTENT_SIZE
+# error "PAGE_ZIP_MIN_SIZE % FSP_EXTENT_SIZE != 0"
+#endif
+
+		if (UNIV_UNLIKELY(init_xdes)) {
+
+			/* The first page in the extent is a descriptor page
+			and the second is an ibuf bitmap page: mark them
+			used */
+
+			xdes_set_bit(descr, XDES_FREE_BIT, 0, FALSE, mtr);
+			xdes_set_bit(descr, XDES_FREE_BIT,
+				     FSP_IBUF_BITMAP_OFFSET, FALSE, mtr);
+			xdes_set_state(descr, XDES_FREE_FRAG, mtr);
+
+			flst_add_last(header + FSP_FREE_FRAG,
+				      descr + XDES_FLST_NODE, mtr);
+			frag_n_used = mtr_read_ulint(header + FSP_FRAG_N_USED,
+						     MLOG_4BYTES, mtr);
+			mlog_write_ulint(header + FSP_FRAG_N_USED,
+					 frag_n_used + 2, MLOG_4BYTES, mtr);
+		} else {
+			flst_add_last(header + FSP_FREE,
+				      descr + XDES_FLST_NODE, mtr);
+			count++;
+		}
+
+		i += FSP_EXTENT_SIZE;
+	}
+}
+
+/**************************************************************************
+Allocates a new free extent. */
+static
+xdes_t*
+fsp_alloc_free_extent(
+/*==================*/
+			/* out: extent descriptor, NULL if cannot be
+			allocated */
+	ulint	space,	/* in: space id */
+	ulint	zip_size,/* in: compressed page size in bytes
+			or 0 for uncompressed pages */
+	ulint	hint,	/* in: hint of which extent would be desirable: any
+			page offset in the extent goes; the hint must not
+			be > FSP_FREE_LIMIT */
+	mtr_t*	mtr)	/* in: mtr */
+{
+	fsp_header_t*	header;
+	fil_addr_t	first;
+	xdes_t*		descr;
+
+	ut_ad(mtr);
+
+	header = fsp_get_space_header(space, zip_size, mtr);
+
+	descr = xdes_get_descriptor_with_space_hdr(header, space, hint, mtr);
+
+	if (descr && (xdes_get_state(descr, mtr) == XDES_FREE)) {
+		/* Ok, we can take this extent */
+	} else {
+		/* Take the first extent in the free list */
+		first = flst_get_first(header + FSP_FREE, mtr);
+
+		if (fil_addr_is_null(first)) {
+			fsp_fill_free_list(FALSE, space, header, mtr);
+
+			first = flst_get_first(header + FSP_FREE, mtr);
+		}
+
+		if (fil_addr_is_null(first)) {
+
+			return(NULL);	/* No free extents left */
+		}
+
+		descr = xdes_lst_get_descriptor(space, zip_size, first, mtr);
+	}
+
+	flst_remove(header + FSP_FREE, descr + XDES_FLST_NODE, mtr);
+
+	return(descr);
+}
+
+/**************************************************************************
+Allocates a single free page from a space. The page is marked as used. */
+static
+ulint
+fsp_alloc_free_page(
+/*================*/
+			/* out: the page offset, FIL_NULL if no page could
+			be allocated */
+	ulint	space,	/* in: space id */
+	ulint	zip_size,/* in: compressed page size in bytes
+			or 0 for uncompressed pages */
+	ulint	hint,	/* in: hint of which page would be desirable */
+	mtr_t*	mtr)	/* in: mtr handle */
+{
+	fsp_header_t*	header;
+	fil_addr_t	first;
+	xdes_t*		descr;
+	buf_block_t*	block;
+	ulint		free;
+	ulint		frag_n_used;
+	ulint		page_no;
+	ulint		space_size;
+	ibool		success;
+
+	ut_ad(mtr);
+
+	header = fsp_get_space_header(space, zip_size, mtr);
+
+	/* Get the hinted descriptor */
+	descr = xdes_get_descriptor_with_space_hdr(header, space, hint, mtr);
+
+	if (descr && (xdes_get_state(descr, mtr) == XDES_FREE_FRAG)) {
+		/* Ok, we can take this extent */
+	} else {
+		/* Else take the first extent in free_frag list */
+		first = flst_get_first(header + FSP_FREE_FRAG, mtr);
+
+		if (fil_addr_is_null(first)) {
+			/* There are no partially full fragments: allocate
+			a free extent and add it to the FREE_FRAG list. NOTE
+			that the allocation may have as a side-effect that an
+			extent containing a descriptor page is added to the
+			FREE_FRAG list. But we will allocate our page from the
+			the free extent anyway. */
+
+			descr = fsp_alloc_free_extent(space, zip_size,
+						      hint, mtr);
+
+			if (descr == NULL) {
+				/* No free space left */
+
+				return(FIL_NULL);
+			}
+
+			xdes_set_state(descr, XDES_FREE_FRAG, mtr);
+			flst_add_last(header + FSP_FREE_FRAG,
+				      descr + XDES_FLST_NODE, mtr);
+		} else {
+			descr = xdes_lst_get_descriptor(space, zip_size,
+							first, mtr);
+		}
+
+		/* Reset the hint */
+		hint = 0;
+	}
+
+	/* Now we have in descr an extent with at least one free page. Look
+	for a free page in the extent. */
+
+	free = xdes_find_bit(descr, XDES_FREE_BIT, TRUE,
+			     hint % FSP_EXTENT_SIZE, mtr);
+	if (free == ULINT_UNDEFINED) {
+
+		ut_print_buf(stderr, ((byte*)descr) - 500, 1000);
+		putc('\n', stderr);
+
+		ut_error;
+	}
+
+	page_no = xdes_get_offset(descr) + free;
+
+	space_size = mtr_read_ulint(header + FSP_SIZE, MLOG_4BYTES, mtr);
+
+	if (space_size <= page_no) {
+		/* It must be that we are extending a single-table tablespace
+		whose size is still < 64 pages */
+
+		ut_a(space != 0);
+		if (page_no >= FSP_EXTENT_SIZE) {
+			fprintf(stderr,
+				"InnoDB: Error: trying to extend a"
+				" single-table tablespace %lu\n"
+				"InnoDB: by single page(s) though the"
+				" space size %lu. Page no %lu.\n",
+				(ulong) space, (ulong) space_size,
+				(ulong) page_no);
+			return(FIL_NULL);
+		}
+		success = fsp_try_extend_data_file_with_pages(space, page_no,
+							      header, mtr);
+		if (!success) {
+			/* No disk space left */
+			return(FIL_NULL);
+		}
+	}
+
+	xdes_set_bit(descr, XDES_FREE_BIT, free, FALSE, mtr);
+
+	/* Update the FRAG_N_USED field */
+	frag_n_used = mtr_read_ulint(header + FSP_FRAG_N_USED, MLOG_4BYTES,
+				     mtr);
+	frag_n_used++;
+	mlog_write_ulint(header + FSP_FRAG_N_USED, frag_n_used, MLOG_4BYTES,
+			 mtr);
+	if (xdes_is_full(descr, mtr)) {
+		/* The fragment is full: move it to another list */
+		flst_remove(header + FSP_FREE_FRAG, descr + XDES_FLST_NODE,
+			    mtr);
+		xdes_set_state(descr, XDES_FULL_FRAG, mtr);
+
+		flst_add_last(header + FSP_FULL_FRAG, descr + XDES_FLST_NODE,
+			      mtr);
+		mlog_write_ulint(header + FSP_FRAG_N_USED,
+				 frag_n_used - FSP_EXTENT_SIZE, MLOG_4BYTES,
+				 mtr);
+	}
+
+	/* Initialize the allocated page to the buffer pool, so that it can
+	be obtained immediately with buf_page_get without need for a disk
+	read. */
+
+	buf_page_create(space, page_no, zip_size, mtr);
+
+	block = buf_page_get(space, zip_size, page_no, RW_X_LATCH, mtr);
+	buf_block_dbg_add_level(block, SYNC_FSP_PAGE);
+
+	/* Prior contents of the page should be ignored */
+	fsp_init_file_page(block, mtr);
+
+	return(page_no);
+}
+
+/**************************************************************************
+Frees a single page of a space. The page is marked as free and clean. */
+static
+void
+fsp_free_page(
+/*==========*/
+	ulint	space,	/* in: space id */
+	ulint	zip_size,/* in: compressed page size in bytes
+			or 0 for uncompressed pages */
+	ulint	page,	/* in: page offset */
+	mtr_t*	mtr)	/* in: mtr handle */
+{
+	fsp_header_t*	header;
+	xdes_t*		descr;
+	ulint		state;
+	ulint		frag_n_used;
+
+	ut_ad(mtr);
+
+	/* fprintf(stderr, "Freeing page %lu in space %lu\n", page, space); */
+
+	header = fsp_get_space_header(space, zip_size, mtr);
+
+	descr = xdes_get_descriptor_with_space_hdr(header, space, page, mtr);
+
+	state = xdes_get_state(descr, mtr);
+
+	if (state != XDES_FREE_FRAG && state != XDES_FULL_FRAG) {
+		fprintf(stderr,
+			"InnoDB: Error: File space extent descriptor"
+			" of page %lu has state %lu\n",
+			(ulong) page,
+			(ulong) state);
+		fputs("InnoDB: Dump of descriptor: ", stderr);
+		ut_print_buf(stderr, ((byte*)descr) - 50, 200);
+		putc('\n', stderr);
+
+		if (state == XDES_FREE) {
+			/* We put here some fault tolerance: if the page
+			is already free, return without doing anything! */
+
+			return;
+		}
+
+		ut_error;
+	}
+
+	if (xdes_get_bit(descr, XDES_FREE_BIT, page % FSP_EXTENT_SIZE, mtr)) {
+		fprintf(stderr,
+			"InnoDB: Error: File space extent descriptor"
+			" of page %lu says it is free\n"
+			"InnoDB: Dump of descriptor: ", (ulong) page);
+		ut_print_buf(stderr, ((byte*)descr) - 50, 200);
+		putc('\n', stderr);
+
+		/* We put here some fault tolerance: if the page
+		is already free, return without doing anything! */
+
+		return;
+	}
+
+	xdes_set_bit(descr, XDES_FREE_BIT, page % FSP_EXTENT_SIZE, TRUE, mtr);
+	xdes_set_bit(descr, XDES_CLEAN_BIT, page % FSP_EXTENT_SIZE, TRUE, mtr);
+
+	frag_n_used = mtr_read_ulint(header + FSP_FRAG_N_USED, MLOG_4BYTES,
+				     mtr);
+	if (state == XDES_FULL_FRAG) {
+		/* The fragment was full: move it to another list */
+		flst_remove(header + FSP_FULL_FRAG, descr + XDES_FLST_NODE,
+			    mtr);
+		xdes_set_state(descr, XDES_FREE_FRAG, mtr);
+		flst_add_last(header + FSP_FREE_FRAG, descr + XDES_FLST_NODE,
+			      mtr);
+		mlog_write_ulint(header + FSP_FRAG_N_USED,
+				 frag_n_used + FSP_EXTENT_SIZE - 1,
+				 MLOG_4BYTES, mtr);
+	} else {
+		ut_a(frag_n_used > 0);
+		mlog_write_ulint(header + FSP_FRAG_N_USED, frag_n_used - 1,
+				 MLOG_4BYTES, mtr);
+	}
+
+	if (xdes_is_free(descr, mtr)) {
+		/* The extent has become free: move it to another list */
+		flst_remove(header + FSP_FREE_FRAG, descr + XDES_FLST_NODE,
+			    mtr);
+		fsp_free_extent(space, zip_size, page, mtr);
+	}
+}
+
+/**************************************************************************
+Returns an extent to the free list of a space. */
+static
+void
+fsp_free_extent(
+/*============*/
+	ulint	space,	/* in: space id */
+	ulint	zip_size,/* in: compressed page size in bytes
+			or 0 for uncompressed pages */
+	ulint	page,	/* in: page offset in the extent */
+	mtr_t*	mtr)	/* in: mtr */
+{
+	fsp_header_t*	header;
+	xdes_t*		descr;
+
+	ut_ad(mtr);
+
+	header = fsp_get_space_header(space, zip_size, mtr);
+
+	descr = xdes_get_descriptor_with_space_hdr(header, space, page, mtr);
+
+	if (xdes_get_state(descr, mtr) == XDES_FREE) {
+
+		ut_print_buf(stderr, (byte*)descr - 500, 1000);
+		putc('\n', stderr);
+
+		ut_error;
+	}
+
+	xdes_init(descr, mtr);
+
+	flst_add_last(header + FSP_FREE, descr + XDES_FLST_NODE, mtr);
+}
+
+/**************************************************************************
+Returns the nth inode slot on an inode page. */
+UNIV_INLINE
+fseg_inode_t*
+fsp_seg_inode_page_get_nth_inode(
+/*=============================*/
+			/* out: segment inode */
+	page_t*	page,	/* in: segment inode page */
+	ulint	i,	/* in: inode index on page */
+	ulint	zip_size __attribute__((unused)),
+			/* in: compressed page size, or 0 */
+	mtr_t*	mtr __attribute__((unused)))
+			/* in: mini-transaction handle */
+{
+	ut_ad(i < FSP_SEG_INODES_PER_PAGE(zip_size));
+	ut_ad(mtr_memo_contains_page(mtr, page, MTR_MEMO_PAGE_X_FIX));
+
+	return(page + FSEG_ARR_OFFSET + FSEG_INODE_SIZE * i);
+}
+
+/**************************************************************************
+Looks for a used segment inode on a segment inode page. */
+static
+ulint
+fsp_seg_inode_page_find_used(
+/*=========================*/
+			/* out: segment inode index, or ULINT_UNDEFINED
+			if not found */
+	page_t*	page,	/* in: segment inode page */
+	ulint	zip_size,/* in: compressed page size, or 0 */
+	mtr_t*	mtr)	/* in: mini-transaction handle */
+{
+	ulint		i;
+	fseg_inode_t*	inode;
+
+	for (i = 0; i < FSP_SEG_INODES_PER_PAGE(zip_size); i++) {
+
+		inode = fsp_seg_inode_page_get_nth_inode(
+			page, i, zip_size, mtr);
+
+		if (!ut_dulint_is_zero(mach_read_from_8(inode + FSEG_ID))) {
+			/* This is used */
+
+			return(i);
+		}
+	}
+
+	return(ULINT_UNDEFINED);
+}
+
+/**************************************************************************
+Looks for an unused segment inode on a segment inode page. */
+static
+ulint
+fsp_seg_inode_page_find_free(
+/*=========================*/
+			/* out: segment inode index, or ULINT_UNDEFINED
+			if not found */
+	page_t*	page,	/* in: segment inode page */
+	ulint	i,	/* in: search forward starting from this index */
+	ulint	zip_size,/* in: compressed page size, or 0 */
+	mtr_t*	mtr)	/* in: mini-transaction handle */
+{
+	fseg_inode_t*	inode;
+
+	for (; i < FSP_SEG_INODES_PER_PAGE(zip_size); i++) {
+
+		inode = fsp_seg_inode_page_get_nth_inode(
+			page, i, zip_size, mtr);
+
+		if (ut_dulint_is_zero(mach_read_from_8(inode + FSEG_ID))) {
+			/* This is unused */
+
+			return(i);
+		}
+	}
+
+	return(ULINT_UNDEFINED);
+}
+
+/**************************************************************************
+Allocates a new file segment inode page. */
+static
+ibool
+fsp_alloc_seg_inode_page(
+/*=====================*/
+					/* out: TRUE if could be allocated */
+	fsp_header_t*	space_header,	/* in: space header */
+	mtr_t*		mtr)		/* in: mini-transaction handle */
+{
+	fseg_inode_t*	inode;
+	buf_block_t*	block;
+	page_t*		page;
+	ulint		page_no;
+	ulint		space;
+	ulint		zip_size;
+	ulint		i;
+
+	ut_ad(page_offset(space_header) == FSP_HEADER_OFFSET);
+
+	space = page_get_space_id(page_align(space_header));
+	zip_size = dict_table_flags_to_zip_size(
+		mach_read_from_4(FSP_SPACE_FLAGS + space_header));
+
+	page_no = fsp_alloc_free_page(space, zip_size, 0, mtr);
+
+	if (page_no == FIL_NULL) {
+
+		return(FALSE);
+	}
+
+	block = buf_page_get(space, zip_size, page_no, RW_X_LATCH, mtr);
+	buf_block_dbg_add_level(block, SYNC_FSP_PAGE);
+
+	block->check_index_page_at_flush = FALSE;
+
+	page = buf_block_get_frame(block);
+
+	mlog_write_ulint(page + FIL_PAGE_TYPE, FIL_PAGE_INODE,
+			 MLOG_2BYTES, mtr);
+
+	for (i = 0; i < FSP_SEG_INODES_PER_PAGE(zip_size); i++) {
+
+		inode = fsp_seg_inode_page_get_nth_inode(page, i,
+							 zip_size, mtr);
+
+		mlog_write_dulint(inode + FSEG_ID, ut_dulint_zero, mtr);
+	}
+
+	flst_add_last(space_header + FSP_SEG_INODES_FREE,
+		      page + FSEG_INODE_PAGE_NODE, mtr);
+	return(TRUE);
+}
+
+/**************************************************************************
+Allocates a new file segment inode. */
+static
+fseg_inode_t*
+fsp_alloc_seg_inode(
+/*================*/
+					/* out: segment inode, or NULL if
+					not enough space */
+	fsp_header_t*	space_header,	/* in: space header */
+	mtr_t*		mtr)		/* in: mini-transaction handle */
+{
+	ulint		page_no;
+	buf_block_t*	block;
+	page_t*		page;
+	fseg_inode_t*	inode;
+	ibool		success;
+	ulint		zip_size;
+	ulint		n;
+
+	ut_ad(page_offset(space_header) == FSP_HEADER_OFFSET);
+
+	if (flst_get_len(space_header + FSP_SEG_INODES_FREE, mtr) == 0) {
+		/* Allocate a new segment inode page */
+
+		success = fsp_alloc_seg_inode_page(space_header, mtr);
+
+		if (!success) {
+
+			return(NULL);
+		}
+	}
+
+	page_no = flst_get_first(space_header + FSP_SEG_INODES_FREE, mtr).page;
+
+	zip_size = dict_table_flags_to_zip_size(
+		mach_read_from_4(FSP_SPACE_FLAGS + space_header));
+	block = buf_page_get(page_get_space_id(page_align(space_header)),
+			     zip_size, page_no, RW_X_LATCH, mtr);
+	buf_block_dbg_add_level(block, SYNC_FSP_PAGE);
+
+	page = buf_block_get_frame(block);
+
+	n = fsp_seg_inode_page_find_free(page, 0, zip_size, mtr);
+
+	ut_a(n != ULINT_UNDEFINED);
+
+	inode = fsp_seg_inode_page_get_nth_inode(page, n, zip_size, mtr);
+
+	if (ULINT_UNDEFINED == fsp_seg_inode_page_find_free(page, n + 1,
+							    zip_size, mtr)) {
+		/* There are no other unused headers left on the page: move it
+		to another list */
+
+		flst_remove(space_header + FSP_SEG_INODES_FREE,
+			    page + FSEG_INODE_PAGE_NODE, mtr);
+
+		flst_add_last(space_header + FSP_SEG_INODES_FULL,
+			      page + FSEG_INODE_PAGE_NODE, mtr);
+	}
+
+	return(inode);
+}
+
+/**************************************************************************
+Frees a file segment inode. */
+static
+void
+fsp_free_seg_inode(
+/*===============*/
+	ulint		space,	/* in: space id */
+	ulint		zip_size,/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	fseg_inode_t*	inode,	/* in: segment inode */
+	mtr_t*		mtr)	/* in: mini-transaction handle */
+{
+	page_t*		page;
+	fsp_header_t*	space_header;
+
+	page = page_align(inode);
+
+	space_header = fsp_get_space_header(space, zip_size, mtr);
+
+	ut_ad(mach_read_from_4(inode + FSEG_MAGIC_N) == FSEG_MAGIC_N_VALUE);
+
+	if (ULINT_UNDEFINED
+	    == fsp_seg_inode_page_find_free(page, 0, zip_size, mtr)) {
+
+		/* Move the page to another list */
+
+		flst_remove(space_header + FSP_SEG_INODES_FULL,
+			    page + FSEG_INODE_PAGE_NODE, mtr);
+
+		flst_add_last(space_header + FSP_SEG_INODES_FREE,
+			      page + FSEG_INODE_PAGE_NODE, mtr);
+	}
+
+	mlog_write_dulint(inode + FSEG_ID, ut_dulint_zero, mtr);
+	mlog_write_ulint(inode + FSEG_MAGIC_N, 0, MLOG_4BYTES, mtr);
+
+	if (ULINT_UNDEFINED
+	    == fsp_seg_inode_page_find_used(page, zip_size, mtr)) {
+
+		/* There are no other used headers left on the page: free it */
+
+		flst_remove(space_header + FSP_SEG_INODES_FREE,
+			    page + FSEG_INODE_PAGE_NODE, mtr);
+
+		fsp_free_page(space, zip_size, page_get_page_no(page), mtr);
+	}
+}
+
+/**************************************************************************
+Returns the file segment inode, page x-latched. */
+static
+fseg_inode_t*
+fseg_inode_get(
+/*===========*/
+				/* out: segment inode, page x-latched */
+	fseg_header_t*	header,	/* in: segment header */
+	ulint		space,	/* in: space id */
+	ulint		zip_size,/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	mtr_t*		mtr)	/* in: mtr handle */
+{
+	fil_addr_t	inode_addr;
+	fseg_inode_t*	inode;
+
+	inode_addr.page = mach_read_from_4(header + FSEG_HDR_PAGE_NO);
+	inode_addr.boffset = mach_read_from_2(header + FSEG_HDR_OFFSET);
+	ut_ad(space == mach_read_from_4(header + FSEG_HDR_SPACE));
+
+	inode = fut_get_ptr(space, zip_size, inode_addr, RW_X_LATCH, mtr);
+
+	ut_ad(mach_read_from_4(inode + FSEG_MAGIC_N) == FSEG_MAGIC_N_VALUE);
+
+	return(inode);
+}
+
+/**************************************************************************
+Gets the page number from the nth fragment page slot. */
+UNIV_INLINE
+ulint
+fseg_get_nth_frag_page_no(
+/*======================*/
+				/* out: page number, FIL_NULL if not in use */
+	fseg_inode_t*	inode,	/* in: segment inode */
+	ulint		n,	/* in: slot index */
+	mtr_t*		mtr __attribute__((unused))) /* in: mtr handle */
+{
+	ut_ad(inode && mtr);
+	ut_ad(n < FSEG_FRAG_ARR_N_SLOTS);
+	ut_ad(mtr_memo_contains_page(mtr, inode, MTR_MEMO_PAGE_X_FIX));
+	return(mach_read_from_4(inode + FSEG_FRAG_ARR
+				+ n * FSEG_FRAG_SLOT_SIZE));
+}
+
+/**************************************************************************
+Sets the page number in the nth fragment page slot. */
+UNIV_INLINE
+void
+fseg_set_nth_frag_page_no(
+/*======================*/
+	fseg_inode_t*	inode,	/* in: segment inode */
+	ulint		n,	/* in: slot index */
+	ulint		page_no,/* in: page number to set */
+	mtr_t*		mtr)	/* in: mtr handle */
+{
+	ut_ad(inode && mtr);
+	ut_ad(n < FSEG_FRAG_ARR_N_SLOTS);
+	ut_ad(mtr_memo_contains_page(mtr, inode, MTR_MEMO_PAGE_X_FIX));
+
+	mlog_write_ulint(inode + FSEG_FRAG_ARR + n * FSEG_FRAG_SLOT_SIZE,
+			 page_no, MLOG_4BYTES, mtr);
+}
+
+/**************************************************************************
+Finds a fragment page slot which is free. */
+static
+ulint
+fseg_find_free_frag_page_slot(
+/*==========================*/
+				/* out: slot index; ULINT_UNDEFINED if none
+				found */
+	fseg_inode_t*	inode,	/* in: segment inode */
+	mtr_t*		mtr)	/* in: mtr handle */
+{
+	ulint	i;
+	ulint	page_no;
+
+	ut_ad(inode && mtr);
+
+	for (i = 0; i < FSEG_FRAG_ARR_N_SLOTS; i++) {
+		page_no = fseg_get_nth_frag_page_no(inode, i, mtr);
+
+		if (page_no == FIL_NULL) {
+
+			return(i);
+		}
+	}
+
+	return(ULINT_UNDEFINED);
+}
+
+/**************************************************************************
+Finds a fragment page slot which is used and last in the array. */
+static
+ulint
+fseg_find_last_used_frag_page_slot(
+/*===============================*/
+				/* out: slot index; ULINT_UNDEFINED if none
+				found */
+	fseg_inode_t*	inode,	/* in: segment inode */
+	mtr_t*		mtr)	/* in: mtr handle */
+{
+	ulint	i;
+	ulint	page_no;
+
+	ut_ad(inode && mtr);
+
+	for (i = 0; i < FSEG_FRAG_ARR_N_SLOTS; i++) {
+		page_no = fseg_get_nth_frag_page_no(
+			inode, FSEG_FRAG_ARR_N_SLOTS - i - 1, mtr);
+
+		if (page_no != FIL_NULL) {
+
+			return(FSEG_FRAG_ARR_N_SLOTS - i - 1);
+		}
+	}
+
+	return(ULINT_UNDEFINED);
+}
+
+/**************************************************************************
+Calculates reserved fragment page slots. */
+static
+ulint
+fseg_get_n_frag_pages(
+/*==================*/
+				/* out: number of fragment pages */
+	fseg_inode_t*	inode,	/* in: segment inode */
+	mtr_t*		mtr)	/* in: mtr handle */
+{
+	ulint	i;
+	ulint	count	= 0;
+
+	ut_ad(inode && mtr);
+
+	for (i = 0; i < FSEG_FRAG_ARR_N_SLOTS; i++) {
+		if (FIL_NULL != fseg_get_nth_frag_page_no(inode, i, mtr)) {
+			count++;
+		}
+	}
+
+	return(count);
+}
+
+/**************************************************************************
+Creates a new segment. */
+UNIV_INTERN
+buf_block_t*
+fseg_create_general(
+/*================*/
+			/* out: the block where the segment header is placed,
+			x-latched, NULL if could not create segment
+			because of lack of space */
+	ulint	space,	/* in: space id */
+	ulint	page,	/* in: page where the segment header is placed: if
+			this is != 0, the page must belong to another segment,
+			if this is 0, a new page will be allocated and it
+			will belong to the created segment */
+	ulint	byte_offset, /* in: byte offset of the created segment header
+			on the page */
+	ibool	has_done_reservation, /* in: TRUE if the caller has already
+			done the reservation for the pages with
+			fsp_reserve_free_extents (at least 2 extents: one for
+			the inode and the other for the segment) then there is
+			no need to do the check for this individual
+			operation */
+	mtr_t*	mtr)	/* in: mtr */
+{
+	ulint		flags;
+	ulint		zip_size;
+	fsp_header_t*	space_header;
+	fseg_inode_t*	inode;
+	dulint		seg_id;
+	buf_block_t*	block	= 0; /* remove warning */
+	fseg_header_t*	header	= 0; /* remove warning */
+	rw_lock_t*	latch;
+	ibool		success;
+	ulint		n_reserved;
+	ulint		i;
+
+	ut_ad(mtr);
+	ut_ad(byte_offset + FSEG_HEADER_SIZE
+	      <= UNIV_PAGE_SIZE - FIL_PAGE_DATA_END);
+
+	latch = fil_space_get_latch(space, &flags);
+	zip_size = dict_table_flags_to_zip_size(flags);
+
+	if (page != 0) {
+		block = buf_page_get(space, zip_size, page, RW_X_LATCH, mtr);
+		header = byte_offset + buf_block_get_frame(block);
+	}
+
+	ut_ad(!mutex_own(&kernel_mutex)
+	      || mtr_memo_contains(mtr, latch, MTR_MEMO_X_LOCK));
+
+	mtr_x_lock(latch, mtr);
+
+	if (rw_lock_get_x_lock_count(latch) == 1) {
+		/* This thread did not own the latch before this call: free
+		excess pages from the insert buffer free list */
+
+		if (space == IBUF_SPACE_ID) {
+			ibuf_free_excess_pages();
+		}
+	}
+
+	if (!has_done_reservation) {
+		success = fsp_reserve_free_extents(&n_reserved, space, 2,
+						   FSP_NORMAL, mtr);
+		if (!success) {
+			return(NULL);
+		}
+	}
+
+	space_header = fsp_get_space_header(space, zip_size, mtr);
+
+	inode = fsp_alloc_seg_inode(space_header, mtr);
+
+	if (inode == NULL) {
+
+		goto funct_exit;
+	}
+
+	/* Read the next segment id from space header and increment the
+	value in space header */
+
+	seg_id = mtr_read_dulint(space_header + FSP_SEG_ID, mtr);
+
+	mlog_write_dulint(space_header + FSP_SEG_ID, ut_dulint_add(seg_id, 1),
+			  mtr);
+
+	mlog_write_dulint(inode + FSEG_ID, seg_id, mtr);
+	mlog_write_ulint(inode + FSEG_NOT_FULL_N_USED, 0, MLOG_4BYTES, mtr);
+
+	flst_init(inode + FSEG_FREE, mtr);
+	flst_init(inode + FSEG_NOT_FULL, mtr);
+	flst_init(inode + FSEG_FULL, mtr);
+
+	mlog_write_ulint(inode + FSEG_MAGIC_N, FSEG_MAGIC_N_VALUE,
+			 MLOG_4BYTES, mtr);
+	for (i = 0; i < FSEG_FRAG_ARR_N_SLOTS; i++) {
+		fseg_set_nth_frag_page_no(inode, i, FIL_NULL, mtr);
+	}
+
+	if (page == 0) {
+		page = fseg_alloc_free_page_low(space, zip_size,
+						inode, 0, FSP_UP, mtr);
+
+		if (page == FIL_NULL) {
+
+			fsp_free_seg_inode(space, zip_size, inode, mtr);
+
+			goto funct_exit;
+		}
+
+		block = buf_page_get(space, zip_size, page, RW_X_LATCH, mtr);
+		header = byte_offset + buf_block_get_frame(block);
+		mlog_write_ulint(header - byte_offset + FIL_PAGE_TYPE,
+				 FIL_PAGE_TYPE_SYS, MLOG_2BYTES, mtr);
+	}
+
+	mlog_write_ulint(header + FSEG_HDR_OFFSET,
+			 page_offset(inode), MLOG_2BYTES, mtr);
+
+	mlog_write_ulint(header + FSEG_HDR_PAGE_NO,
+			 page_get_page_no(page_align(inode)),
+			 MLOG_4BYTES, mtr);
+
+	mlog_write_ulint(header + FSEG_HDR_SPACE, space, MLOG_4BYTES, mtr);
+
+funct_exit:
+	if (!has_done_reservation) {
+
+		fil_space_release_free_extents(space, n_reserved);
+	}
+
+	return(block);
+}
+
+/**************************************************************************
+Creates a new segment. */
+UNIV_INTERN
+buf_block_t*
+fseg_create(
+/*========*/
+			/* out: the block where the segment header is placed,
+			x-latched, NULL if could not create segment
+			because of lack of space */
+	ulint	space,	/* in: space id */
+	ulint	page,	/* in: page where the segment header is placed: if
+			this is != 0, the page must belong to another segment,
+			if this is 0, a new page will be allocated and it
+			will belong to the created segment */
+	ulint	byte_offset, /* in: byte offset of the created segment header
+			on the page */
+	mtr_t*	mtr)	/* in: mtr */
+{
+	return(fseg_create_general(space, page, byte_offset, FALSE, mtr));
+}
+
+/**************************************************************************
+Calculates the number of pages reserved by a segment, and how many pages are
+currently used. */
+static
+ulint
+fseg_n_reserved_pages_low(
+/*======================*/
+				/* out: number of reserved pages */
+	fseg_inode_t*	inode,	/* in: segment inode */
+	ulint*		used,	/* out: number of pages used (<= reserved) */
+	mtr_t*		mtr)	/* in: mtr handle */
+{
+	ulint	ret;
+
+	ut_ad(inode && used && mtr);
+	ut_ad(mtr_memo_contains_page(mtr, inode, MTR_MEMO_PAGE_X_FIX));
+
+	*used = mtr_read_ulint(inode + FSEG_NOT_FULL_N_USED, MLOG_4BYTES, mtr)
+		+ FSP_EXTENT_SIZE * flst_get_len(inode + FSEG_FULL, mtr)
+		+ fseg_get_n_frag_pages(inode, mtr);
+
+	ret = fseg_get_n_frag_pages(inode, mtr)
+		+ FSP_EXTENT_SIZE * flst_get_len(inode + FSEG_FREE, mtr)
+		+ FSP_EXTENT_SIZE * flst_get_len(inode + FSEG_NOT_FULL, mtr)
+		+ FSP_EXTENT_SIZE * flst_get_len(inode + FSEG_FULL, mtr);
+
+	return(ret);
+}
+
+/**************************************************************************
+Calculates the number of pages reserved by a segment, and how many pages are
+currently used. */
+UNIV_INTERN
+ulint
+fseg_n_reserved_pages(
+/*==================*/
+				/* out: number of reserved pages */
+	fseg_header_t*	header,	/* in: segment header */
+	ulint*		used,	/* out: number of pages used (<= reserved) */
+	mtr_t*		mtr)	/* in: mtr handle */
+{
+	ulint		ret;
+	fseg_inode_t*	inode;
+	ulint		space;
+	ulint		flags;
+	ulint		zip_size;
+	rw_lock_t*	latch;
+
+	space = page_get_space_id(page_align(header));
+	latch = fil_space_get_latch(space, &flags);
+	zip_size = dict_table_flags_to_zip_size(flags);
+
+	ut_ad(!mutex_own(&kernel_mutex)
+	      || mtr_memo_contains(mtr, latch, MTR_MEMO_X_LOCK));
+
+	mtr_x_lock(latch, mtr);
+
+	inode = fseg_inode_get(header, space, zip_size, mtr);
+
+	ret = fseg_n_reserved_pages_low(inode, used, mtr);
+
+	return(ret);
+}
+
+/*************************************************************************
+Tries to fill the free list of a segment with consecutive free extents.
+This happens if the segment is big enough to allow extents in the free list,
+the free list is empty, and the extents can be allocated consecutively from
+the hint onward. */
+static
+void
+fseg_fill_free_list(
+/*================*/
+	fseg_inode_t*	inode,	/* in: segment inode */
+	ulint		space,	/* in: space id */
+	ulint		zip_size,/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	ulint		hint,	/* in: hint which extent would be good as
+				the first extent */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	xdes_t*	descr;
+	ulint	i;
+	dulint	seg_id;
+	ulint	reserved;
+	ulint	used;
+
+	ut_ad(inode && mtr);
+	ut_ad(!((page_offset(inode) - FSEG_ARR_OFFSET) % FSEG_INODE_SIZE));
+
+	reserved = fseg_n_reserved_pages_low(inode, &used, mtr);
+
+	if (reserved < FSEG_FREE_LIST_LIMIT * FSP_EXTENT_SIZE) {
+
+		/* The segment is too small to allow extents in free list */
+
+		return;
+	}
+
+	if (flst_get_len(inode + FSEG_FREE, mtr) > 0) {
+		/* Free list is not empty */
+
+		return;
+	}
+
+	for (i = 0; i < FSEG_FREE_LIST_MAX_LEN; i++) {
+		descr = xdes_get_descriptor(space, zip_size, hint, mtr);
+
+		if ((descr == NULL)
+		    || (XDES_FREE != xdes_get_state(descr, mtr))) {
+
+			/* We cannot allocate the desired extent: stop */
+
+			return;
+		}
+
+		descr = fsp_alloc_free_extent(space, zip_size, hint, mtr);
+
+		xdes_set_state(descr, XDES_FSEG, mtr);
+
+		seg_id = mtr_read_dulint(inode + FSEG_ID, mtr);
+		mlog_write_dulint(descr + XDES_ID, seg_id, mtr);
+
+		flst_add_last(inode + FSEG_FREE, descr + XDES_FLST_NODE, mtr);
+		hint += FSP_EXTENT_SIZE;
+	}
+}
+
+/*************************************************************************
+Allocates a free extent for the segment: looks first in the free list of the
+segment, then tries to allocate from the space free list. NOTE that the extent
+returned still resides in the segment free list, it is not yet taken off it! */
+static
+xdes_t*
+fseg_alloc_free_extent(
+/*===================*/
+				/* out: allocated extent, still placed in the
+				segment free list, NULL if could
+				not be allocated */
+	fseg_inode_t*	inode,	/* in: segment inode */
+	ulint		space,	/* in: space id */
+	ulint		zip_size,/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	xdes_t*		descr;
+	dulint		seg_id;
+	fil_addr_t	first;
+
+	ut_ad(!((page_offset(inode) - FSEG_ARR_OFFSET) % FSEG_INODE_SIZE));
+
+	if (flst_get_len(inode + FSEG_FREE, mtr) > 0) {
+		/* Segment free list is not empty, allocate from it */
+
+		first = flst_get_first(inode + FSEG_FREE, mtr);
+
+		descr = xdes_lst_get_descriptor(space, zip_size, first, mtr);
+	} else {
+		/* Segment free list was empty, allocate from space */
+		descr = fsp_alloc_free_extent(space, zip_size, 0, mtr);
+
+		if (descr == NULL) {
+
+			return(NULL);
+		}
+
+		seg_id = mtr_read_dulint(inode + FSEG_ID, mtr);
+
+		xdes_set_state(descr, XDES_FSEG, mtr);
+		mlog_write_dulint(descr + XDES_ID, seg_id, mtr);
+		flst_add_last(inode + FSEG_FREE, descr + XDES_FLST_NODE, mtr);
+
+		/* Try to fill the segment free list */
+		fseg_fill_free_list(inode, space, zip_size,
+				    xdes_get_offset(descr) + FSP_EXTENT_SIZE,
+				    mtr);
+	}
+
+	return(descr);
+}
+
+/**************************************************************************
+Allocates a single free page from a segment. This function implements
+the intelligent allocation strategy which tries to minimize file space
+fragmentation. */
+static
+ulint
+fseg_alloc_free_page_low(
+/*=====================*/
+				/* out: the allocated page number, FIL_NULL
+				if no page could be allocated */
+	ulint		space,	/* in: space */
+	ulint		zip_size,/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	fseg_inode_t*	seg_inode, /* in: segment inode */
+	ulint		hint,	/* in: hint of which page would be desirable */
+	byte		direction, /* in: if the new page is needed because
+				of an index page split, and records are
+				inserted there in order, into which
+				direction they go alphabetically: FSP_DOWN,
+				FSP_UP, FSP_NO_DIR */
+	mtr_t*		mtr)	/* in: mtr handle */
+{
+	fsp_header_t*	space_header;
+	ulint		space_size;
+	dulint		seg_id;
+	ulint		used;
+	ulint		reserved;
+	xdes_t*		descr;		/* extent of the hinted page */
+	ulint		ret_page;	/* the allocated page offset, FIL_NULL
+					if could not be allocated */
+	xdes_t*		ret_descr;	/* the extent of the allocated page */
+	ibool		frag_page_allocated = FALSE;
+	ibool		success;
+	ulint		n;
+
+	ut_ad(mtr);
+	ut_ad((direction >= FSP_UP) && (direction <= FSP_NO_DIR));
+	ut_ad(mach_read_from_4(seg_inode + FSEG_MAGIC_N)
+	      == FSEG_MAGIC_N_VALUE);
+	ut_ad(!((page_offset(seg_inode) - FSEG_ARR_OFFSET) % FSEG_INODE_SIZE));
+	seg_id = mtr_read_dulint(seg_inode + FSEG_ID, mtr);
+
+	ut_ad(!ut_dulint_is_zero(seg_id));
+
+	reserved = fseg_n_reserved_pages_low(seg_inode, &used, mtr);
+
+	space_header = fsp_get_space_header(space, zip_size, mtr);
+
+	descr = xdes_get_descriptor_with_space_hdr(space_header, space,
+						   hint, mtr);
+	if (descr == NULL) {
+		/* Hint outside space or too high above free limit: reset
+		hint */
+		hint = 0;
+		descr = xdes_get_descriptor(space, zip_size, hint, mtr);
+	}
+
+	/* In the big if-else below we look for ret_page and ret_descr */
+	/*-------------------------------------------------------------*/
+	if ((xdes_get_state(descr, mtr) == XDES_FSEG)
+	    && (0 == ut_dulint_cmp(mtr_read_dulint(descr + XDES_ID,
+						   mtr), seg_id))
+	    && (xdes_get_bit(descr, XDES_FREE_BIT,
+			     hint % FSP_EXTENT_SIZE, mtr) == TRUE)) {
+
+		/* 1. We can take the hinted page
+		=================================*/
+		ret_descr = descr;
+		ret_page = hint;
+		/*-----------------------------------------------------------*/
+	} else if ((xdes_get_state(descr, mtr) == XDES_FREE)
+		   && ((reserved - used) < reserved / FSEG_FILLFACTOR)
+		   && (used >= FSEG_FRAG_LIMIT)) {
+
+		/* 2. We allocate the free extent from space and can take
+		=========================================================
+		the hinted page
+		===============*/
+		ret_descr = fsp_alloc_free_extent(space, zip_size, hint, mtr);
+
+		ut_a(ret_descr == descr);
+
+		xdes_set_state(ret_descr, XDES_FSEG, mtr);
+		mlog_write_dulint(ret_descr + XDES_ID, seg_id, mtr);
+		flst_add_last(seg_inode + FSEG_FREE,
+			      ret_descr + XDES_FLST_NODE, mtr);
+
+		/* Try to fill the segment free list */
+		fseg_fill_free_list(seg_inode, space, zip_size,
+				    hint + FSP_EXTENT_SIZE, mtr);
+		ret_page = hint;
+		/*-----------------------------------------------------------*/
+	} else if ((direction != FSP_NO_DIR)
+		   && ((reserved - used) < reserved / FSEG_FILLFACTOR)
+		   && (used >= FSEG_FRAG_LIMIT)
+		   && (!!(ret_descr
+			  = fseg_alloc_free_extent(seg_inode,
+						   space, zip_size, mtr)))) {
+
+		/* 3. We take any free extent (which was already assigned above
+		===============================================================
+		in the if-condition to ret_descr) and take the lowest or
+		========================================================
+		highest page in it, depending on the direction
+		==============================================*/
+		ret_page = xdes_get_offset(ret_descr);
+
+		if (direction == FSP_DOWN) {
+			ret_page += FSP_EXTENT_SIZE - 1;
+		}
+		/*-----------------------------------------------------------*/
+	} else if ((xdes_get_state(descr, mtr) == XDES_FSEG)
+		   && (0 == ut_dulint_cmp(mtr_read_dulint(descr + XDES_ID,
+							  mtr), seg_id))
+		   && (!xdes_is_full(descr, mtr))) {
+
+		/* 4. We can take the page from the same extent as the
+		======================================================
+		hinted page (and the extent already belongs to the
+		==================================================
+		segment)
+		========*/
+		ret_descr = descr;
+		ret_page = xdes_get_offset(ret_descr)
+			+ xdes_find_bit(ret_descr, XDES_FREE_BIT, TRUE,
+					hint % FSP_EXTENT_SIZE, mtr);
+		/*-----------------------------------------------------------*/
+	} else if (reserved - used > 0) {
+		/* 5. We take any unused page from the segment
+		==============================================*/
+		fil_addr_t	first;
+
+		if (flst_get_len(seg_inode + FSEG_NOT_FULL, mtr) > 0) {
+			first = flst_get_first(seg_inode + FSEG_NOT_FULL,
+					       mtr);
+		} else if (flst_get_len(seg_inode + FSEG_FREE, mtr) > 0) {
+			first = flst_get_first(seg_inode + FSEG_FREE, mtr);
+		} else {
+			ut_error;
+			return(FIL_NULL);
+		}
+
+		ret_descr = xdes_lst_get_descriptor(space, zip_size,
+						    first, mtr);
+		ret_page = xdes_get_offset(ret_descr)
+			+ xdes_find_bit(ret_descr, XDES_FREE_BIT, TRUE,
+					0, mtr);
+		/*-----------------------------------------------------------*/
+	} else if (used < FSEG_FRAG_LIMIT) {
+		/* 6. We allocate an individual page from the space
+		===================================================*/
+		ret_page = fsp_alloc_free_page(space, zip_size, hint, mtr);
+		ret_descr = NULL;
+
+		frag_page_allocated = TRUE;
+
+		if (ret_page != FIL_NULL) {
+			/* Put the page in the fragment page array of the
+			segment */
+			n = fseg_find_free_frag_page_slot(seg_inode, mtr);
+			ut_a(n != FIL_NULL);
+
+			fseg_set_nth_frag_page_no(seg_inode, n, ret_page,
+						  mtr);
+		}
+		/*-----------------------------------------------------------*/
+	} else {
+		/* 7. We allocate a new extent and take its first page
+		======================================================*/
+		ret_descr = fseg_alloc_free_extent(seg_inode,
+						   space, zip_size, mtr);
+
+		if (ret_descr == NULL) {
+			ret_page = FIL_NULL;
+		} else {
+			ret_page = xdes_get_offset(ret_descr);
+		}
+	}
+
+	if (ret_page == FIL_NULL) {
+		/* Page could not be allocated */
+
+		return(FIL_NULL);
+	}
+
+	if (space != 0) {
+		space_size = fil_space_get_size(space);
+
+		if (space_size <= ret_page) {
+			/* It must be that we are extending a single-table
+			tablespace whose size is still < 64 pages */
+
+			if (ret_page >= FSP_EXTENT_SIZE) {
+				fprintf(stderr,
+					"InnoDB: Error (2): trying to extend"
+					" a single-table tablespace %lu\n"
+					"InnoDB: by single page(s) though"
+					" the space size %lu. Page no %lu.\n",
+					(ulong) space, (ulong) space_size,
+					(ulong) ret_page);
+				return(FIL_NULL);
+			}
+
+			success = fsp_try_extend_data_file_with_pages(
+				space, ret_page, space_header, mtr);
+			if (!success) {
+				/* No disk space left */
+				return(FIL_NULL);
+			}
+		}
+	}
+
+	if (!frag_page_allocated) {
+		/* Initialize the allocated page to buffer pool, so that it
+		can be obtained immediately with buf_page_get without need
+		for a disk read */
+		buf_block_t*	block;
+		ulint		zip_size = dict_table_flags_to_zip_size(
+			mach_read_from_4(FSP_SPACE_FLAGS + space_header));
+
+		block = buf_page_create(space, ret_page, zip_size, mtr);
+		buf_block_dbg_add_level(block, SYNC_FSP_PAGE);
+
+		if (UNIV_UNLIKELY(block != buf_page_get(space, zip_size,
+							ret_page, RW_X_LATCH,
+							mtr))) {
+			ut_error;
+		}
+
+		/* The prior contents of the page should be ignored */
+		fsp_init_file_page(block, mtr);
+
+		/* At this point we know the extent and the page offset.
+		The extent is still in the appropriate list (FSEG_NOT_FULL
+		or FSEG_FREE), and the page is not yet marked as used. */
+
+		ut_ad(xdes_get_descriptor(space, zip_size, ret_page, mtr)
+		      == ret_descr);
+		ut_ad(xdes_get_bit(ret_descr, XDES_FREE_BIT,
+				   ret_page % FSP_EXTENT_SIZE, mtr) == TRUE);
+
+		fseg_mark_page_used(seg_inode, space, zip_size, ret_page, mtr);
+	}
+
+	buf_reset_check_index_page_at_flush(space, ret_page);
+
+	return(ret_page);
+}
+
+/**************************************************************************
+Allocates a single free page from a segment. This function implements
+the intelligent allocation strategy which tries to minimize file space
+fragmentation. */
+UNIV_INTERN
+ulint
+fseg_alloc_free_page_general(
+/*=========================*/
+				/* out: allocated page offset, FIL_NULL if no
+				page could be allocated */
+	fseg_header_t*	seg_header,/* in: segment header */
+	ulint		hint,	/* in: hint of which page would be desirable */
+	byte		direction,/* in: if the new page is needed because
+				of an index page split, and records are
+				inserted there in order, into which
+				direction they go alphabetically: FSP_DOWN,
+				FSP_UP, FSP_NO_DIR */
+	ibool		has_done_reservation, /* in: TRUE if the caller has
+				already done the reservation for the page
+				with fsp_reserve_free_extents, then there
+				is no need to do the check for this individual
+				page */
+	mtr_t*		mtr)	/* in: mtr handle */
+{
+	fseg_inode_t*	inode;
+	ulint		space;
+	ulint		flags;
+	ulint		zip_size;
+	rw_lock_t*	latch;
+	ibool		success;
+	ulint		page_no;
+	ulint		n_reserved;
+
+	space = page_get_space_id(page_align(seg_header));
+
+	latch = fil_space_get_latch(space, &flags);
+
+	zip_size = dict_table_flags_to_zip_size(flags);
+
+	ut_ad(!mutex_own(&kernel_mutex)
+	      || mtr_memo_contains(mtr, latch, MTR_MEMO_X_LOCK));
+
+	mtr_x_lock(latch, mtr);
+
+	if (rw_lock_get_x_lock_count(latch) == 1) {
+		/* This thread did not own the latch before this call: free
+		excess pages from the insert buffer free list */
+
+		if (space == IBUF_SPACE_ID) {
+			ibuf_free_excess_pages();
+		}
+	}
+
+	inode = fseg_inode_get(seg_header, space, zip_size, mtr);
+
+	if (!has_done_reservation) {
+		success = fsp_reserve_free_extents(&n_reserved, space, 2,
+						   FSP_NORMAL, mtr);
+		if (!success) {
+			return(FIL_NULL);
+		}
+	}
+
+	page_no = fseg_alloc_free_page_low(space, zip_size,
+					   inode, hint, direction, mtr);
+	if (!has_done_reservation) {
+		fil_space_release_free_extents(space, n_reserved);
+	}
+
+	return(page_no);
+}
+
+/**************************************************************************
+Allocates a single free page from a segment. This function implements
+the intelligent allocation strategy which tries to minimize file space
+fragmentation. */
+UNIV_INTERN
+ulint
+fseg_alloc_free_page(
+/*=================*/
+				/* out: allocated page offset, FIL_NULL if no
+				page could be allocated */
+	fseg_header_t*	seg_header,/* in: segment header */
+	ulint		hint,	/* in: hint of which page would be desirable */
+	byte		direction,/* in: if the new page is needed because
+				of an index page split, and records are
+				inserted there in order, into which
+				direction they go alphabetically: FSP_DOWN,
+				FSP_UP, FSP_NO_DIR */
+	mtr_t*		mtr)	/* in: mtr handle */
+{
+	return(fseg_alloc_free_page_general(seg_header, hint, direction,
+					    FALSE, mtr));
+}
+
+/**************************************************************************
+Checks that we have at least 2 frag pages free in the first extent of a
+single-table tablespace, and they are also physically initialized to the data
+file. That is we have already extended the data file so that those pages are
+inside the data file. If not, this function extends the tablespace with
+pages. */
+static
+ibool
+fsp_reserve_free_pages(
+/*===================*/
+					/* out: TRUE if there were >= 3 free
+					pages, or we were able to extend */
+	ulint		space,		/* in: space id, must be != 0 */
+	fsp_header_t*	space_header,	/* in: header of that space,
+					x-latched */
+	ulint		size,		/* in: size of the tablespace in pages,
+					must be < FSP_EXTENT_SIZE / 2 */
+	mtr_t*		mtr)		/* in: mtr */
+{
+	xdes_t*	descr;
+	ulint	n_used;
+
+	ut_a(space != 0);
+	ut_a(size < FSP_EXTENT_SIZE / 2);
+
+	descr = xdes_get_descriptor_with_space_hdr(space_header, space, 0,
+						   mtr);
+	n_used = xdes_get_n_used(descr, mtr);
+
+	ut_a(n_used <= size);
+
+	if (size >= n_used + 2) {
+
+		return(TRUE);
+	}
+
+	return(fsp_try_extend_data_file_with_pages(space, n_used + 1,
+						   space_header, mtr));
+}
+
+/**************************************************************************
+Reserves free pages from a tablespace. All mini-transactions which may
+use several pages from the tablespace should call this function beforehand
+and reserve enough free extents so that they certainly will be able
+to do their operation, like a B-tree page split, fully. Reservations
+must be released with function fil_space_release_free_extents!
+
+The alloc_type below has the following meaning: FSP_NORMAL means an
+operation which will probably result in more space usage, like an
+insert in a B-tree; FSP_UNDO means allocation to undo logs: if we are
+deleting rows, then this allocation will in the long run result in
+less space usage (after a purge); FSP_CLEANING means allocation done
+in a physical record delete (like in a purge) or other cleaning operation
+which will result in less space usage in the long run. We prefer the latter
+two types of allocation: when space is scarce, FSP_NORMAL allocations
+will not succeed, but the latter two allocations will succeed, if possible.
+The purpose is to avoid dead end where the database is full but the
+user cannot free any space because these freeing operations temporarily
+reserve some space.
+
+Single-table tablespaces whose size is < 32 pages are a special case. In this
+function we would liberally reserve several 64 page extents for every page
+split or merge in a B-tree. But we do not want to waste disk space if the table
+only occupies < 32 pages. That is why we apply different rules in that special
+case, just ensuring that there are 3 free pages available. */
+UNIV_INTERN
+ibool
+fsp_reserve_free_extents(
+/*=====================*/
+			/* out: TRUE if we were able to make the reservation */
+	ulint*	n_reserved,/* out: number of extents actually reserved; if we
+			return TRUE and the tablespace size is < 64 pages,
+			then this can be 0, otherwise it is n_ext */
+	ulint	space,	/* in: space id */
+	ulint	n_ext,	/* in: number of extents to reserve */
+	ulint	alloc_type,/* in: FSP_NORMAL, FSP_UNDO, or FSP_CLEANING */
+	mtr_t*	mtr)	/* in: mtr */
+{
+	fsp_header_t*	space_header;
+	rw_lock_t*	latch;
+	ulint		n_free_list_ext;
+	ulint		free_limit;
+	ulint		size;
+	ulint		flags;
+	ulint		zip_size;
+	ulint		n_free;
+	ulint		n_free_up;
+	ulint		reserve;
+	ibool		success;
+	ulint		n_pages_added;
+
+	ut_ad(mtr);
+	*n_reserved = n_ext;
+
+	latch = fil_space_get_latch(space, &flags);
+	zip_size = dict_table_flags_to_zip_size(flags);
+
+	ut_ad(!mutex_own(&kernel_mutex)
+	      || mtr_memo_contains(mtr, latch, MTR_MEMO_X_LOCK));
+
+	mtr_x_lock(latch, mtr);
+
+	space_header = fsp_get_space_header(space, zip_size, mtr);
+try_again:
+	size = mtr_read_ulint(space_header + FSP_SIZE, MLOG_4BYTES, mtr);
+
+	if (size < FSP_EXTENT_SIZE / 2) {
+		/* Use different rules for small single-table tablespaces */
+		*n_reserved = 0;
+		return(fsp_reserve_free_pages(space, space_header, size, mtr));
+	}
+
+	n_free_list_ext = flst_get_len(space_header + FSP_FREE, mtr);
+
+	free_limit = mtr_read_ulint(space_header + FSP_FREE_LIMIT,
+				    MLOG_4BYTES, mtr);
+
+	/* Below we play safe when counting free extents above the free limit:
+	some of them will contain extent descriptor pages, and therefore
+	will not be free extents */
+
+	n_free_up = (size - free_limit) / FSP_EXTENT_SIZE;
+
+	if (n_free_up > 0) {
+		n_free_up--;
+		if (!zip_size) {
+			n_free_up -= n_free_up
+				/ (UNIV_PAGE_SIZE / FSP_EXTENT_SIZE);
+		} else {
+			n_free_up -= n_free_up
+				/ (zip_size / FSP_EXTENT_SIZE);
+		}
+	}
+
+	n_free = n_free_list_ext + n_free_up;
+
+	if (alloc_type == FSP_NORMAL) {
+		/* We reserve 1 extent + 0.5 % of the space size to undo logs
+		and 1 extent + 0.5 % to cleaning operations; NOTE: this source
+		code is duplicated in the function below! */
+
+		reserve = 2 + ((size / FSP_EXTENT_SIZE) * 2) / 200;
+
+		if (n_free <= reserve + n_ext) {
+
+			goto try_to_extend;
+		}
+	} else if (alloc_type == FSP_UNDO) {
+		/* We reserve 0.5 % of the space size to cleaning operations */
+
+		reserve = 1 + ((size / FSP_EXTENT_SIZE) * 1) / 200;
+
+		if (n_free <= reserve + n_ext) {
+
+			goto try_to_extend;
+		}
+	} else {
+		ut_a(alloc_type == FSP_CLEANING);
+	}
+
+	success = fil_space_reserve_free_extents(space, n_free, n_ext);
+
+	if (success) {
+		return(TRUE);
+	}
+try_to_extend:
+	success = fsp_try_extend_data_file(&n_pages_added, space,
+					   space_header, mtr);
+	if (success && n_pages_added > 0) {
+
+		goto try_again;
+	}
+
+	return(FALSE);
+}
+
+/**************************************************************************
+This function should be used to get information on how much we still
+will be able to insert new data to the database without running out the
+tablespace. Only free extents are taken into account and we also subtract
+the safety margin required by the above function fsp_reserve_free_extents. */
+UNIV_INTERN
+ullint
+fsp_get_available_space_in_free_extents(
+/*====================================*/
+			/* out: available space in kB */
+	ulint	space)	/* in: space id */
+{
+	fsp_header_t*	space_header;
+	ulint		n_free_list_ext;
+	ulint		free_limit;
+	ulint		size;
+	ulint		flags;
+	ulint		zip_size;
+	ulint		n_free;
+	ulint		n_free_up;
+	ulint		reserve;
+	rw_lock_t*	latch;
+	mtr_t		mtr;
+
+	ut_ad(!mutex_own(&kernel_mutex));
+
+	mtr_start(&mtr);
+
+	latch = fil_space_get_latch(space, &flags);
+	zip_size = dict_table_flags_to_zip_size(flags);
+
+	mtr_x_lock(latch, &mtr);
+
+	space_header = fsp_get_space_header(space, zip_size, &mtr);
+
+	size = mtr_read_ulint(space_header + FSP_SIZE, MLOG_4BYTES, &mtr);
+
+	n_free_list_ext = flst_get_len(space_header + FSP_FREE, &mtr);
+
+	free_limit = mtr_read_ulint(space_header + FSP_FREE_LIMIT,
+				    MLOG_4BYTES, &mtr);
+	mtr_commit(&mtr);
+
+	if (size < FSP_EXTENT_SIZE) {
+		ut_a(space != 0);	/* This must be a single-table
+					tablespace */
+
+		return(0);		/* TODO: count free frag pages and
+					return a value based on that */
+	}
+
+	/* Below we play safe when counting free extents above the free limit:
+	some of them will contain extent descriptor pages, and therefore
+	will not be free extents */
+
+	n_free_up = (size - free_limit) / FSP_EXTENT_SIZE;
+
+	if (n_free_up > 0) {
+		n_free_up--;
+		if (!zip_size) {
+			n_free_up -= n_free_up
+				/ (UNIV_PAGE_SIZE / FSP_EXTENT_SIZE);
+		} else {
+			n_free_up -= n_free_up
+				/ (zip_size / FSP_EXTENT_SIZE);
+		}
+	}
+
+	n_free = n_free_list_ext + n_free_up;
+
+	/* We reserve 1 extent + 0.5 % of the space size to undo logs
+	and 1 extent + 0.5 % to cleaning operations; NOTE: this source
+	code is duplicated in the function above! */
+
+	reserve = 2 + ((size / FSP_EXTENT_SIZE) * 2) / 200;
+
+	if (reserve > n_free) {
+		return(0);
+	}
+
+	if (!zip_size) {
+		return((ullint) (n_free - reserve)
+		       * FSP_EXTENT_SIZE
+		       * (UNIV_PAGE_SIZE / 1024));
+	} else {
+		return((ullint) (n_free - reserve)
+		       * FSP_EXTENT_SIZE
+		       * (zip_size / 1024));
+	}
+}
+
+/************************************************************************
+Marks a page used. The page must reside within the extents of the given
+segment. */
+static
+void
+fseg_mark_page_used(
+/*================*/
+	fseg_inode_t*	seg_inode,/* in: segment inode */
+	ulint		space,	/* in: space id */
+	ulint		zip_size,/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	ulint		page,	/* in: page offset */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	xdes_t*	descr;
+	ulint	not_full_n_used;
+
+	ut_ad(seg_inode && mtr);
+	ut_ad(!((page_offset(seg_inode) - FSEG_ARR_OFFSET) % FSEG_INODE_SIZE));
+
+	descr = xdes_get_descriptor(space, zip_size, page, mtr);
+
+	ut_ad(mtr_read_ulint(seg_inode + FSEG_ID, MLOG_4BYTES, mtr)
+	      == mtr_read_ulint(descr + XDES_ID, MLOG_4BYTES, mtr));
+
+	if (xdes_is_free(descr, mtr)) {
+		/* We move the extent from the free list to the
+		NOT_FULL list */
+		flst_remove(seg_inode + FSEG_FREE, descr + XDES_FLST_NODE,
+			    mtr);
+		flst_add_last(seg_inode + FSEG_NOT_FULL,
+			      descr + XDES_FLST_NODE, mtr);
+	}
+
+	ut_ad(xdes_get_bit(descr, XDES_FREE_BIT, page % FSP_EXTENT_SIZE, mtr)
+	      == TRUE);
+	/* We mark the page as used */
+	xdes_set_bit(descr, XDES_FREE_BIT, page % FSP_EXTENT_SIZE, FALSE, mtr);
+
+	not_full_n_used = mtr_read_ulint(seg_inode + FSEG_NOT_FULL_N_USED,
+					 MLOG_4BYTES, mtr);
+	not_full_n_used++;
+	mlog_write_ulint(seg_inode + FSEG_NOT_FULL_N_USED, not_full_n_used,
+			 MLOG_4BYTES, mtr);
+	if (xdes_is_full(descr, mtr)) {
+		/* We move the extent from the NOT_FULL list to the
+		FULL list */
+		flst_remove(seg_inode + FSEG_NOT_FULL,
+			    descr + XDES_FLST_NODE, mtr);
+		flst_add_last(seg_inode + FSEG_FULL,
+			      descr + XDES_FLST_NODE, mtr);
+
+		mlog_write_ulint(seg_inode + FSEG_NOT_FULL_N_USED,
+				 not_full_n_used - FSP_EXTENT_SIZE,
+				 MLOG_4BYTES, mtr);
+	}
+}
+
+/**************************************************************************
+Frees a single page of a segment. */
+static
+void
+fseg_free_page_low(
+/*===============*/
+	fseg_inode_t*	seg_inode, /* in: segment inode */
+	ulint		space,	/* in: space id */
+	ulint		zip_size,/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	ulint		page,	/* in: page offset */
+	mtr_t*		mtr)	/* in: mtr handle */
+{
+	xdes_t*	descr;
+	ulint	not_full_n_used;
+	ulint	state;
+	dulint	descr_id;
+	dulint	seg_id;
+	ulint	i;
+
+	ut_ad(seg_inode && mtr);
+	ut_ad(mach_read_from_4(seg_inode + FSEG_MAGIC_N)
+	      == FSEG_MAGIC_N_VALUE);
+	ut_ad(!((page_offset(seg_inode) - FSEG_ARR_OFFSET) % FSEG_INODE_SIZE));
+
+	/* Drop search system page hash index if the page is found in
+	the pool and is hashed */
+
+	btr_search_drop_page_hash_when_freed(space, zip_size, page);
+
+	descr = xdes_get_descriptor(space, zip_size, page, mtr);
+
+	ut_a(descr);
+	if (xdes_get_bit(descr, XDES_FREE_BIT, page % FSP_EXTENT_SIZE, mtr)) {
+		fputs("InnoDB: Dump of the tablespace extent descriptor: ",
+		      stderr);
+		ut_print_buf(stderr, descr, 40);
+
+		fprintf(stderr, "\n"
+			"InnoDB: Serious error! InnoDB is trying to"
+			" free page %lu\n"
+			"InnoDB: though it is already marked as free"
+			" in the tablespace!\n"
+			"InnoDB: The tablespace free space info is corrupt.\n"
+			"InnoDB: You may need to dump your"
+			" InnoDB tables and recreate the whole\n"
+			"InnoDB: database!\n", (ulong) page);
+crash:
+		fputs("InnoDB: Please refer to\n"
+		      "InnoDB: http://dev.mysql.com/doc/refman/5.1/en/"
+		      "forcing-recovery.html\n"
+		      "InnoDB: about forcing recovery.\n", stderr);
+		ut_error;
+	}
+
+	state = xdes_get_state(descr, mtr);
+
+	if (state != XDES_FSEG) {
+		/* The page is in the fragment pages of the segment */
+
+		for (i = 0;; i++) {
+			if (fseg_get_nth_frag_page_no(seg_inode, i, mtr)
+			    == page) {
+
+				fseg_set_nth_frag_page_no(seg_inode, i,
+							  FIL_NULL, mtr);
+				break;
+			}
+		}
+
+		fsp_free_page(space, zip_size, page, mtr);
+
+		return;
+	}
+
+	/* If we get here, the page is in some extent of the segment */
+
+	descr_id = mtr_read_dulint(descr + XDES_ID, mtr);
+	seg_id = mtr_read_dulint(seg_inode + FSEG_ID, mtr);
+#if 0
+	fprintf(stderr,
+		"InnoDB: InnoDB is freeing space %lu page %lu,\n"
+		"InnoDB: which belongs to descr seg %lu %lu\n"
+		"InnoDB: segment %lu %lu.\n",
+		(ulong) space, (ulong) page,
+		(ulong) ut_dulint_get_high(descr_id),
+		(ulong) ut_dulint_get_low(descr_id),
+		(ulong) ut_dulint_get_high(seg_id),
+		(ulong) ut_dulint_get_low(seg_id));
+#endif /* 0 */
+	if (0 != ut_dulint_cmp(descr_id, seg_id)) {
+		fputs("InnoDB: Dump of the tablespace extent descriptor: ",
+		      stderr);
+		ut_print_buf(stderr, descr, 40);
+		fputs("\nInnoDB: Dump of the segment inode: ", stderr);
+		ut_print_buf(stderr, seg_inode, 40);
+		putc('\n', stderr);
+
+		fprintf(stderr,
+			"InnoDB: Serious error: InnoDB is trying to"
+			" free space %lu page %lu,\n"
+			"InnoDB: which does not belong to"
+			" segment %lu %lu but belongs\n"
+			"InnoDB: to segment %lu %lu.\n",
+			(ulong) space, (ulong) page,
+			(ulong) ut_dulint_get_high(descr_id),
+			(ulong) ut_dulint_get_low(descr_id),
+			(ulong) ut_dulint_get_high(seg_id),
+			(ulong) ut_dulint_get_low(seg_id));
+		goto crash;
+	}
+
+	not_full_n_used = mtr_read_ulint(seg_inode + FSEG_NOT_FULL_N_USED,
+					 MLOG_4BYTES, mtr);
+	if (xdes_is_full(descr, mtr)) {
+		/* The fragment is full: move it to another list */
+		flst_remove(seg_inode + FSEG_FULL,
+			    descr + XDES_FLST_NODE, mtr);
+		flst_add_last(seg_inode + FSEG_NOT_FULL,
+			      descr + XDES_FLST_NODE, mtr);
+		mlog_write_ulint(seg_inode + FSEG_NOT_FULL_N_USED,
+				 not_full_n_used + FSP_EXTENT_SIZE - 1,
+				 MLOG_4BYTES, mtr);
+	} else {
+		ut_a(not_full_n_used > 0);
+		mlog_write_ulint(seg_inode + FSEG_NOT_FULL_N_USED,
+				 not_full_n_used - 1, MLOG_4BYTES, mtr);
+	}
+
+	xdes_set_bit(descr, XDES_FREE_BIT, page % FSP_EXTENT_SIZE, TRUE, mtr);
+	xdes_set_bit(descr, XDES_CLEAN_BIT, page % FSP_EXTENT_SIZE, TRUE, mtr);
+
+	if (xdes_is_free(descr, mtr)) {
+		/* The extent has become free: free it to space */
+		flst_remove(seg_inode + FSEG_NOT_FULL,
+			    descr + XDES_FLST_NODE, mtr);
+		fsp_free_extent(space, zip_size, page, mtr);
+	}
+}
+
+/**************************************************************************
+Frees a single page of a segment. */
+UNIV_INTERN
+void
+fseg_free_page(
+/*===========*/
+	fseg_header_t*	seg_header, /* in: segment header */
+	ulint		space,	/* in: space id */
+	ulint		page,	/* in: page offset */
+	mtr_t*		mtr)	/* in: mtr handle */
+{
+	ulint		flags;
+	ulint		zip_size;
+	fseg_inode_t*	seg_inode;
+	rw_lock_t*	latch;
+
+	latch = fil_space_get_latch(space, &flags);
+	zip_size = dict_table_flags_to_zip_size(flags);
+
+	ut_ad(!mutex_own(&kernel_mutex)
+	      || mtr_memo_contains(mtr, latch, MTR_MEMO_X_LOCK));
+
+	mtr_x_lock(latch, mtr);
+
+	seg_inode = fseg_inode_get(seg_header, space, zip_size, mtr);
+
+	fseg_free_page_low(seg_inode, space, zip_size, page, mtr);
+
+#ifdef UNIV_DEBUG_FILE_ACCESSES
+	buf_page_set_file_page_was_freed(space, page);
+#endif
+}
+
+/**************************************************************************
+Frees an extent of a segment to the space free list. */
+static
+void
+fseg_free_extent(
+/*=============*/
+	fseg_inode_t*	seg_inode, /* in: segment inode */
+	ulint		space,	/* in: space id */
+	ulint		zip_size,/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	ulint		page,	/* in: a page in the extent */
+	mtr_t*		mtr)	/* in: mtr handle */
+{
+	ulint	first_page_in_extent;
+	xdes_t*	descr;
+	ulint	not_full_n_used;
+	ulint	descr_n_used;
+	ulint	i;
+
+	ut_ad(seg_inode && mtr);
+
+	descr = xdes_get_descriptor(space, zip_size, page, mtr);
+
+	ut_a(xdes_get_state(descr, mtr) == XDES_FSEG);
+	ut_a(0 == ut_dulint_cmp(mtr_read_dulint(descr + XDES_ID, mtr),
+				mtr_read_dulint(seg_inode + FSEG_ID, mtr)));
+
+	first_page_in_extent = page - (page % FSP_EXTENT_SIZE);
+
+	for (i = 0; i < FSP_EXTENT_SIZE; i++) {
+		if (FALSE == xdes_get_bit(descr, XDES_FREE_BIT, i, mtr)) {
+
+			/* Drop search system page hash index if the page is
+			found in the pool and is hashed */
+
+			btr_search_drop_page_hash_when_freed(
+				space, zip_size, first_page_in_extent + i);
+		}
+	}
+
+	if (xdes_is_full(descr, mtr)) {
+		flst_remove(seg_inode + FSEG_FULL,
+			    descr + XDES_FLST_NODE, mtr);
+	} else if (xdes_is_free(descr, mtr)) {
+		flst_remove(seg_inode + FSEG_FREE,
+			    descr + XDES_FLST_NODE, mtr);
+	} else {
+		flst_remove(seg_inode + FSEG_NOT_FULL,
+			    descr + XDES_FLST_NODE, mtr);
+
+		not_full_n_used = mtr_read_ulint(
+			seg_inode + FSEG_NOT_FULL_N_USED, MLOG_4BYTES, mtr);
+
+		descr_n_used = xdes_get_n_used(descr, mtr);
+		ut_a(not_full_n_used >= descr_n_used);
+		mlog_write_ulint(seg_inode + FSEG_NOT_FULL_N_USED,
+				 not_full_n_used - descr_n_used,
+				 MLOG_4BYTES, mtr);
+	}
+
+	fsp_free_extent(space, zip_size, page, mtr);
+
+#ifdef UNIV_DEBUG_FILE_ACCESSES
+	for (i = 0; i < FSP_EXTENT_SIZE; i++) {
+
+		buf_page_set_file_page_was_freed(space,
+						 first_page_in_extent + i);
+	}
+#endif
+}
+
+/**************************************************************************
+Frees part of a segment. This function can be used to free a segment by
+repeatedly calling this function in different mini-transactions. Doing
+the freeing in a single mini-transaction might result in too big a
+mini-transaction. */
+UNIV_INTERN
+ibool
+fseg_free_step(
+/*===========*/
+				/* out: TRUE if freeing completed */
+	fseg_header_t*	header,	/* in, own: segment header; NOTE: if the header
+				resides on the first page of the frag list
+				of the segment, this pointer becomes obsolete
+				after the last freeing step */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	ulint		n;
+	ulint		page;
+	xdes_t*		descr;
+	fseg_inode_t*	inode;
+	ulint		space;
+	ulint		flags;
+	ulint		zip_size;
+	ulint		header_page;
+	rw_lock_t*	latch;
+
+	space = page_get_space_id(page_align(header));
+	header_page = page_get_page_no(page_align(header));
+
+	latch = fil_space_get_latch(space, &flags);
+	zip_size = dict_table_flags_to_zip_size(flags);
+
+	ut_ad(!mutex_own(&kernel_mutex)
+	      || mtr_memo_contains(mtr, latch, MTR_MEMO_X_LOCK));
+
+	mtr_x_lock(latch, mtr);
+
+	descr = xdes_get_descriptor(space, zip_size, header_page, mtr);
+
+	/* Check that the header resides on a page which has not been
+	freed yet */
+
+	ut_a(descr);
+	ut_a(xdes_get_bit(descr, XDES_FREE_BIT,
+			  header_page % FSP_EXTENT_SIZE, mtr) == FALSE);
+	inode = fseg_inode_get(header, space, zip_size, mtr);
+
+	descr = fseg_get_first_extent(inode, space, zip_size, mtr);
+
+	if (descr != NULL) {
+		/* Free the extent held by the segment */
+		page = xdes_get_offset(descr);
+
+		fseg_free_extent(inode, space, zip_size, page, mtr);
+
+		return(FALSE);
+	}
+
+	/* Free a frag page */
+	n = fseg_find_last_used_frag_page_slot(inode, mtr);
+
+	if (n == ULINT_UNDEFINED) {
+		/* Freeing completed: free the segment inode */
+		fsp_free_seg_inode(space, zip_size, inode, mtr);
+
+		return(TRUE);
+	}
+
+	fseg_free_page_low(inode, space, zip_size,
+			   fseg_get_nth_frag_page_no(inode, n, mtr), mtr);
+
+	n = fseg_find_last_used_frag_page_slot(inode, mtr);
+
+	if (n == ULINT_UNDEFINED) {
+		/* Freeing completed: free the segment inode */
+		fsp_free_seg_inode(space, zip_size, inode, mtr);
+
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
+
+/**************************************************************************
+Frees part of a segment. Differs from fseg_free_step because this function
+leaves the header page unfreed. */
+UNIV_INTERN
+ibool
+fseg_free_step_not_header(
+/*======================*/
+				/* out: TRUE if freeing completed, except the
+				header page */
+	fseg_header_t*	header,	/* in: segment header which must reside on
+				the first fragment page of the segment */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	ulint		n;
+	ulint		page;
+	xdes_t*		descr;
+	fseg_inode_t*	inode;
+	ulint		space;
+	ulint		flags;
+	ulint		zip_size;
+	ulint		page_no;
+	rw_lock_t*	latch;
+
+	space = page_get_space_id(page_align(header));
+
+	latch = fil_space_get_latch(space, &flags);
+	zip_size = dict_table_flags_to_zip_size(flags);
+
+	ut_ad(!mutex_own(&kernel_mutex)
+	      || mtr_memo_contains(mtr, latch, MTR_MEMO_X_LOCK));
+
+	mtr_x_lock(latch, mtr);
+
+	inode = fseg_inode_get(header, space, zip_size, mtr);
+
+	descr = fseg_get_first_extent(inode, space, zip_size, mtr);
+
+	if (descr != NULL) {
+		/* Free the extent held by the segment */
+		page = xdes_get_offset(descr);
+
+		fseg_free_extent(inode, space, zip_size, page, mtr);
+
+		return(FALSE);
+	}
+
+	/* Free a frag page */
+
+	n = fseg_find_last_used_frag_page_slot(inode, mtr);
+
+	if (n == ULINT_UNDEFINED) {
+		ut_error;
+	}
+
+	page_no = fseg_get_nth_frag_page_no(inode, n, mtr);
+
+	if (page_no == page_get_page_no(page_align(header))) {
+
+		return(TRUE);
+	}
+
+	fseg_free_page_low(inode, space, zip_size, page_no, mtr);
+
+	return(FALSE);
+}
+
+/***********************************************************************
+Frees a segment. The freeing is performed in several mini-transactions,
+so that there is no danger of bufferfixing too many buffer pages. */
+UNIV_INTERN
+void
+fseg_free(
+/*======*/
+	ulint	space,	/* in: space id */
+	ulint	zip_size,/* in: compressed page size in bytes
+			or 0 for uncompressed pages */
+	ulint	page_no,/* in: page number where the segment header is
+			placed */
+	ulint	offset) /* in: byte offset of the segment header on that
+			page */
+{
+	mtr_t		mtr;
+	ibool		finished;
+	fseg_header_t*	header;
+	fil_addr_t	addr;
+
+	addr.page = page_no;
+	addr.boffset = offset;
+
+	for (;;) {
+		mtr_start(&mtr);
+
+		header = fut_get_ptr(space, zip_size, addr, RW_X_LATCH, &mtr);
+
+		finished = fseg_free_step(header, &mtr);
+
+		mtr_commit(&mtr);
+
+		if (finished) {
+
+			return;
+		}
+	}
+}
+
+/**************************************************************************
+Returns the first extent descriptor for a segment. We think of the extent
+lists of the segment catenated in the order FSEG_FULL -> FSEG_NOT_FULL
+-> FSEG_FREE. */
+static
+xdes_t*
+fseg_get_first_extent(
+/*==================*/
+				/* out: the first extent descriptor, or NULL if
+				none */
+	fseg_inode_t*	inode,	/* in: segment inode */
+	ulint		space,	/* in: space id */
+	ulint		zip_size,/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	fil_addr_t	first;
+	xdes_t*		descr;
+
+	ut_ad(inode && mtr);
+
+	ut_ad(space == page_get_space_id(page_align(inode)));
+
+	first = fil_addr_null;
+
+	if (flst_get_len(inode + FSEG_FULL, mtr) > 0) {
+
+		first = flst_get_first(inode + FSEG_FULL, mtr);
+
+	} else if (flst_get_len(inode + FSEG_NOT_FULL, mtr) > 0) {
+
+		first = flst_get_first(inode + FSEG_NOT_FULL, mtr);
+
+	} else if (flst_get_len(inode + FSEG_FREE, mtr) > 0) {
+
+		first = flst_get_first(inode + FSEG_FREE, mtr);
+	}
+
+	if (first.page == FIL_NULL) {
+
+		return(NULL);
+	}
+	descr = xdes_lst_get_descriptor(space, zip_size, first, mtr);
+
+	return(descr);
+}
+
+/***********************************************************************
+Validates a segment. */
+static
+ibool
+fseg_validate_low(
+/*==============*/
+				/* out: TRUE if ok */
+	fseg_inode_t*	inode, /* in: segment inode */
+	mtr_t*		mtr2)	/* in: mtr */
+{
+	ulint		space;
+	dulint		seg_id;
+	mtr_t		mtr;
+	xdes_t*		descr;
+	fil_addr_t	node_addr;
+	ulint		n_used		= 0;
+	ulint		n_used2		= 0;
+
+	ut_ad(mtr_memo_contains_page(mtr2, inode, MTR_MEMO_PAGE_X_FIX));
+	ut_ad(mach_read_from_4(inode + FSEG_MAGIC_N) == FSEG_MAGIC_N_VALUE);
+
+	space = page_get_space_id(page_align(inode));
+
+	seg_id = mtr_read_dulint(inode + FSEG_ID, mtr2);
+	n_used = mtr_read_ulint(inode + FSEG_NOT_FULL_N_USED,
+				MLOG_4BYTES, mtr2);
+	flst_validate(inode + FSEG_FREE, mtr2);
+	flst_validate(inode + FSEG_NOT_FULL, mtr2);
+	flst_validate(inode + FSEG_FULL, mtr2);
+
+	/* Validate FSEG_FREE list */
+	node_addr = flst_get_first(inode + FSEG_FREE, mtr2);
+
+	while (!fil_addr_is_null(node_addr)) {
+		ulint	flags;
+		ulint	zip_size;
+
+		mtr_start(&mtr);
+		mtr_x_lock(fil_space_get_latch(space, &flags), &mtr);
+		zip_size = dict_table_flags_to_zip_size(flags);
+
+		descr = xdes_lst_get_descriptor(space, zip_size,
+						node_addr, &mtr);
+
+		ut_a(xdes_get_n_used(descr, &mtr) == 0);
+		ut_a(xdes_get_state(descr, &mtr) == XDES_FSEG);
+		ut_a(!ut_dulint_cmp(mtr_read_dulint(descr + XDES_ID, &mtr),
+				    seg_id));
+
+		node_addr = flst_get_next_addr(descr + XDES_FLST_NODE, &mtr);
+		mtr_commit(&mtr);
+	}
+
+	/* Validate FSEG_NOT_FULL list */
+
+	node_addr = flst_get_first(inode + FSEG_NOT_FULL, mtr2);
+
+	while (!fil_addr_is_null(node_addr)) {
+		ulint	flags;
+		ulint	zip_size;
+
+		mtr_start(&mtr);
+		mtr_x_lock(fil_space_get_latch(space, &flags), &mtr);
+		zip_size = dict_table_flags_to_zip_size(flags);
+
+		descr = xdes_lst_get_descriptor(space, zip_size,
+						node_addr, &mtr);
+
+		ut_a(xdes_get_n_used(descr, &mtr) > 0);
+		ut_a(xdes_get_n_used(descr, &mtr) < FSP_EXTENT_SIZE);
+		ut_a(xdes_get_state(descr, &mtr) == XDES_FSEG);
+		ut_a(!ut_dulint_cmp(mtr_read_dulint(descr + XDES_ID, &mtr),
+				    seg_id));
+
+		n_used2 += xdes_get_n_used(descr, &mtr);
+
+		node_addr = flst_get_next_addr(descr + XDES_FLST_NODE, &mtr);
+		mtr_commit(&mtr);
+	}
+
+	/* Validate FSEG_FULL list */
+
+	node_addr = flst_get_first(inode + FSEG_FULL, mtr2);
+
+	while (!fil_addr_is_null(node_addr)) {
+		ulint	flags;
+		ulint	zip_size;
+
+		mtr_start(&mtr);
+		mtr_x_lock(fil_space_get_latch(space, &flags), &mtr);
+		zip_size = dict_table_flags_to_zip_size(flags);
+
+		descr = xdes_lst_get_descriptor(space, zip_size,
+						node_addr, &mtr);
+
+		ut_a(xdes_get_n_used(descr, &mtr) == FSP_EXTENT_SIZE);
+		ut_a(xdes_get_state(descr, &mtr) == XDES_FSEG);
+		ut_a(!ut_dulint_cmp(mtr_read_dulint(descr + XDES_ID, &mtr),
+				    seg_id));
+
+		node_addr = flst_get_next_addr(descr + XDES_FLST_NODE, &mtr);
+		mtr_commit(&mtr);
+	}
+
+	ut_a(n_used == n_used2);
+
+	return(TRUE);
+}
+
+/***********************************************************************
+Validates a segment. */
+UNIV_INTERN
+ibool
+fseg_validate(
+/*==========*/
+				/* out: TRUE if ok */
+	fseg_header_t*	header, /* in: segment header */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	fseg_inode_t*	inode;
+	ibool		ret;
+	ulint		space;
+	ulint		flags;
+	ulint		zip_size;
+
+	space = page_get_space_id(page_align(header));
+
+	mtr_x_lock(fil_space_get_latch(space, &flags), mtr);
+	zip_size = dict_table_flags_to_zip_size(flags);
+
+	inode = fseg_inode_get(header, space, zip_size, mtr);
+
+	ret = fseg_validate_low(inode, mtr);
+
+	return(ret);
+}
+
+/***********************************************************************
+Writes info of a segment. */
+static
+void
+fseg_print_low(
+/*===========*/
+	fseg_inode_t*	inode, /* in: segment inode */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	ulint	space;
+	ulint	seg_id_low;
+	ulint	seg_id_high;
+	ulint	n_used;
+	ulint	n_frag;
+	ulint	n_free;
+	ulint	n_not_full;
+	ulint	n_full;
+	ulint	reserved;
+	ulint	used;
+	ulint	page_no;
+	dulint	 d_var;
+
+	ut_ad(mtr_memo_contains_page(mtr, inode, MTR_MEMO_PAGE_X_FIX));
+	space = page_get_space_id(page_align(inode));
+	page_no = page_get_page_no(page_align(inode));
+
+	reserved = fseg_n_reserved_pages_low(inode, &used, mtr);
+
+	d_var = mtr_read_dulint(inode + FSEG_ID, mtr);
+
+	seg_id_low = ut_dulint_get_low(d_var);
+	seg_id_high = ut_dulint_get_high(d_var);
+
+	n_used = mtr_read_ulint(inode + FSEG_NOT_FULL_N_USED,
+				MLOG_4BYTES, mtr);
+	n_frag = fseg_get_n_frag_pages(inode, mtr);
+	n_free = flst_get_len(inode + FSEG_FREE, mtr);
+	n_not_full = flst_get_len(inode + FSEG_NOT_FULL, mtr);
+	n_full = flst_get_len(inode + FSEG_FULL, mtr);
+
+	fprintf(stderr,
+		"SEGMENT id %lu %lu space %lu; page %lu;"
+		" res %lu used %lu; full ext %lu\n"
+		"fragm pages %lu; free extents %lu;"
+		" not full extents %lu: pages %lu\n",
+		(ulong) seg_id_high, (ulong) seg_id_low,
+		(ulong) space, (ulong) page_no,
+		(ulong) reserved, (ulong) used, (ulong) n_full,
+		(ulong) n_frag, (ulong) n_free, (ulong) n_not_full,
+		(ulong) n_used);
+}
+
+#ifdef UNIV_BTR_PRINT
+/***********************************************************************
+Writes info of a segment. */
+UNIV_INTERN
+void
+fseg_print(
+/*=======*/
+	fseg_header_t*	header, /* in: segment header */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	fseg_inode_t*	inode;
+	ulint		space;
+	ulint		flags;
+	ulint		zip_size;
+
+	space = page_get_space_id(page_align(header));
+
+	mtr_x_lock(fil_space_get_latch(space, &flags), mtr);
+	zip_size = dict_table_flags_to_zip_size(flags);
+
+	inode = fseg_inode_get(header, space, zip_size, mtr);
+
+	fseg_print_low(inode, mtr);
+}
+#endif /* UNIV_BTR_PRINT */
+
+/***********************************************************************
+Validates the file space system and its segments. */
+UNIV_INTERN
+ibool
+fsp_validate(
+/*=========*/
+			/* out: TRUE if ok */
+	ulint	space)	/* in: space id */
+{
+	fsp_header_t*	header;
+	fseg_inode_t*	seg_inode;
+	page_t*		seg_inode_page;
+	rw_lock_t*	latch;
+	ulint		size;
+	ulint		flags;
+	ulint		zip_size;
+	ulint		free_limit;
+	ulint		frag_n_used;
+	mtr_t		mtr;
+	mtr_t		mtr2;
+	xdes_t*		descr;
+	fil_addr_t	node_addr;
+	fil_addr_t	next_node_addr;
+	ulint		descr_count	= 0;
+	ulint		n_used		= 0;
+	ulint		n_used2		= 0;
+	ulint		n_full_frag_pages;
+	ulint		n;
+	ulint		seg_inode_len_free;
+	ulint		seg_inode_len_full;
+
+	latch = fil_space_get_latch(space, &flags);
+	zip_size = dict_table_flags_to_zip_size(flags);
+	ut_a(ut_is_2pow(zip_size));
+	ut_a(zip_size <= UNIV_PAGE_SIZE);
+	ut_a(!zip_size || zip_size >= PAGE_ZIP_MIN_SIZE);
+
+	/* Start first a mini-transaction mtr2 to lock out all other threads
+	from the fsp system */
+	mtr_start(&mtr2);
+	mtr_x_lock(latch, &mtr2);
+
+	mtr_start(&mtr);
+	mtr_x_lock(latch, &mtr);
+
+	header = fsp_get_space_header(space, zip_size, &mtr);
+
+	size = mtr_read_ulint(header + FSP_SIZE, MLOG_4BYTES, &mtr);
+	free_limit = mtr_read_ulint(header + FSP_FREE_LIMIT,
+				    MLOG_4BYTES, &mtr);
+	frag_n_used = mtr_read_ulint(header + FSP_FRAG_N_USED,
+				     MLOG_4BYTES, &mtr);
+
+	n_full_frag_pages = FSP_EXTENT_SIZE
+		* flst_get_len(header + FSP_FULL_FRAG, &mtr);
+
+	if (UNIV_UNLIKELY(free_limit > size)) {
+
+		ut_a(space != 0);
+		ut_a(size < FSP_EXTENT_SIZE);
+	}
+
+	flst_validate(header + FSP_FREE, &mtr);
+	flst_validate(header + FSP_FREE_FRAG, &mtr);
+	flst_validate(header + FSP_FULL_FRAG, &mtr);
+
+	mtr_commit(&mtr);
+
+	/* Validate FSP_FREE list */
+	mtr_start(&mtr);
+	mtr_x_lock(latch, &mtr);
+
+	header = fsp_get_space_header(space, zip_size, &mtr);
+	node_addr = flst_get_first(header + FSP_FREE, &mtr);
+
+	mtr_commit(&mtr);
+
+	while (!fil_addr_is_null(node_addr)) {
+		mtr_start(&mtr);
+		mtr_x_lock(latch, &mtr);
+
+		descr_count++;
+		descr = xdes_lst_get_descriptor(space, zip_size,
+						node_addr, &mtr);
+
+		ut_a(xdes_get_n_used(descr, &mtr) == 0);
+		ut_a(xdes_get_state(descr, &mtr) == XDES_FREE);
+
+		node_addr = flst_get_next_addr(descr + XDES_FLST_NODE, &mtr);
+		mtr_commit(&mtr);
+	}
+
+	/* Validate FSP_FREE_FRAG list */
+	mtr_start(&mtr);
+	mtr_x_lock(latch, &mtr);
+
+	header = fsp_get_space_header(space, zip_size, &mtr);
+	node_addr = flst_get_first(header + FSP_FREE_FRAG, &mtr);
+
+	mtr_commit(&mtr);
+
+	while (!fil_addr_is_null(node_addr)) {
+		mtr_start(&mtr);
+		mtr_x_lock(latch, &mtr);
+
+		descr_count++;
+		descr = xdes_lst_get_descriptor(space, zip_size,
+						node_addr, &mtr);
+
+		ut_a(xdes_get_n_used(descr, &mtr) > 0);
+		ut_a(xdes_get_n_used(descr, &mtr) < FSP_EXTENT_SIZE);
+		ut_a(xdes_get_state(descr, &mtr) == XDES_FREE_FRAG);
+
+		n_used += xdes_get_n_used(descr, &mtr);
+		node_addr = flst_get_next_addr(descr + XDES_FLST_NODE, &mtr);
+
+		mtr_commit(&mtr);
+	}
+
+	/* Validate FSP_FULL_FRAG list */
+	mtr_start(&mtr);
+	mtr_x_lock(latch, &mtr);
+
+	header = fsp_get_space_header(space, zip_size, &mtr);
+	node_addr = flst_get_first(header + FSP_FULL_FRAG, &mtr);
+
+	mtr_commit(&mtr);
+
+	while (!fil_addr_is_null(node_addr)) {
+		mtr_start(&mtr);
+		mtr_x_lock(latch, &mtr);
+
+		descr_count++;
+		descr = xdes_lst_get_descriptor(space, zip_size,
+						node_addr, &mtr);
+
+		ut_a(xdes_get_n_used(descr, &mtr) == FSP_EXTENT_SIZE);
+		ut_a(xdes_get_state(descr, &mtr) == XDES_FULL_FRAG);
+
+		node_addr = flst_get_next_addr(descr + XDES_FLST_NODE, &mtr);
+		mtr_commit(&mtr);
+	}
+
+	/* Validate segments */
+	mtr_start(&mtr);
+	mtr_x_lock(latch, &mtr);
+
+	header = fsp_get_space_header(space, zip_size, &mtr);
+
+	node_addr = flst_get_first(header + FSP_SEG_INODES_FULL, &mtr);
+
+	seg_inode_len_full = flst_get_len(header + FSP_SEG_INODES_FULL, &mtr);
+
+	mtr_commit(&mtr);
+
+	while (!fil_addr_is_null(node_addr)) {
+
+		n = 0;
+		do {
+			mtr_start(&mtr);
+			mtr_x_lock(latch, &mtr);
+
+			seg_inode_page = fut_get_ptr(
+				space, zip_size, node_addr, RW_X_LATCH, &mtr)
+				- FSEG_INODE_PAGE_NODE;
+
+			seg_inode = fsp_seg_inode_page_get_nth_inode(
+				seg_inode_page, n, zip_size, &mtr);
+			ut_a(!ut_dulint_is_zero(
+				     mach_read_from_8(seg_inode + FSEG_ID)));
+			fseg_validate_low(seg_inode, &mtr);
+
+			descr_count += flst_get_len(seg_inode + FSEG_FREE,
+						    &mtr);
+			descr_count += flst_get_len(seg_inode + FSEG_FULL,
+						    &mtr);
+			descr_count += flst_get_len(seg_inode + FSEG_NOT_FULL,
+						    &mtr);
+
+			n_used2 += fseg_get_n_frag_pages(seg_inode, &mtr);
+
+			next_node_addr = flst_get_next_addr(
+				seg_inode_page + FSEG_INODE_PAGE_NODE, &mtr);
+			mtr_commit(&mtr);
+		} while (++n < FSP_SEG_INODES_PER_PAGE(zip_size));
+
+		node_addr = next_node_addr;
+	}
+
+	mtr_start(&mtr);
+	mtr_x_lock(latch, &mtr);
+
+	header = fsp_get_space_header(space, zip_size, &mtr);
+
+	node_addr = flst_get_first(header + FSP_SEG_INODES_FREE, &mtr);
+
+	seg_inode_len_free = flst_get_len(header + FSP_SEG_INODES_FREE, &mtr);
+
+	mtr_commit(&mtr);
+
+	while (!fil_addr_is_null(node_addr)) {
+
+		n = 0;
+
+		do {
+			mtr_start(&mtr);
+			mtr_x_lock(latch, &mtr);
+
+			seg_inode_page = fut_get_ptr(
+				space, zip_size, node_addr, RW_X_LATCH, &mtr)
+				- FSEG_INODE_PAGE_NODE;
+
+			seg_inode = fsp_seg_inode_page_get_nth_inode(
+				seg_inode_page, n, zip_size, &mtr);
+			if (!ut_dulint_is_zero(
+				    mach_read_from_8(seg_inode + FSEG_ID))) {
+				fseg_validate_low(seg_inode, &mtr);
+
+				descr_count += flst_get_len(
+					seg_inode + FSEG_FREE, &mtr);
+				descr_count += flst_get_len(
+					seg_inode + FSEG_FULL, &mtr);
+				descr_count += flst_get_len(
+					seg_inode + FSEG_NOT_FULL, &mtr);
+				n_used2 += fseg_get_n_frag_pages(
+					seg_inode, &mtr);
+			}
+
+			next_node_addr = flst_get_next_addr(
+				seg_inode_page + FSEG_INODE_PAGE_NODE, &mtr);
+			mtr_commit(&mtr);
+		} while (++n < FSP_SEG_INODES_PER_PAGE(zip_size));
+
+		node_addr = next_node_addr;
+	}
+
+	ut_a(descr_count * FSP_EXTENT_SIZE == free_limit);
+	if (!zip_size) {
+		ut_a(n_used + n_full_frag_pages
+		     == n_used2 + 2 * ((free_limit + (UNIV_PAGE_SIZE - 1))
+				       / UNIV_PAGE_SIZE)
+		     + seg_inode_len_full + seg_inode_len_free);
+	} else {
+		ut_a(n_used + n_full_frag_pages
+		     == n_used2 + 2 * ((free_limit + (zip_size - 1))
+				       / zip_size)
+		     + seg_inode_len_full + seg_inode_len_free);
+	}
+	ut_a(frag_n_used == n_used);
+
+	mtr_commit(&mtr2);
+
+	return(TRUE);
+}
+
+/***********************************************************************
+Prints info of a file space. */
+UNIV_INTERN
+void
+fsp_print(
+/*======*/
+	ulint	space)	/* in: space id */
+{
+	fsp_header_t*	header;
+	fseg_inode_t*	seg_inode;
+	page_t*		seg_inode_page;
+	rw_lock_t*	latch;
+	ulint		flags;
+	ulint		zip_size;
+	ulint		size;
+	ulint		free_limit;
+	ulint		frag_n_used;
+	fil_addr_t	node_addr;
+	fil_addr_t	next_node_addr;
+	ulint		n_free;
+	ulint		n_free_frag;
+	ulint		n_full_frag;
+	ulint		seg_id_low;
+	ulint		seg_id_high;
+	ulint		n;
+	ulint		n_segs		= 0;
+	dulint		d_var;
+	mtr_t		mtr;
+	mtr_t		mtr2;
+
+	latch = fil_space_get_latch(space, &flags);
+	zip_size = dict_table_flags_to_zip_size(flags);
+
+	/* Start first a mini-transaction mtr2 to lock out all other threads
+	from the fsp system */
+
+	mtr_start(&mtr2);
+
+	mtr_x_lock(latch, &mtr2);
+
+	mtr_start(&mtr);
+
+	mtr_x_lock(latch, &mtr);
+
+	header = fsp_get_space_header(space, zip_size, &mtr);
+
+	size = mtr_read_ulint(header + FSP_SIZE, MLOG_4BYTES, &mtr);
+
+	free_limit = mtr_read_ulint(header + FSP_FREE_LIMIT, MLOG_4BYTES,
+				    &mtr);
+	frag_n_used = mtr_read_ulint(header + FSP_FRAG_N_USED, MLOG_4BYTES,
+				     &mtr);
+	n_free = flst_get_len(header + FSP_FREE, &mtr);
+	n_free_frag = flst_get_len(header + FSP_FREE_FRAG, &mtr);
+	n_full_frag = flst_get_len(header + FSP_FULL_FRAG, &mtr);
+
+	d_var = mtr_read_dulint(header + FSP_SEG_ID, &mtr);
+
+	seg_id_low = ut_dulint_get_low(d_var);
+	seg_id_high = ut_dulint_get_high(d_var);
+
+	fprintf(stderr,
+		"FILE SPACE INFO: id %lu\n"
+		"size %lu, free limit %lu, free extents %lu\n"
+		"not full frag extents %lu: used pages %lu,"
+		" full frag extents %lu\n"
+		"first seg id not used %lu %lu\n",
+		(ulong) space,
+		(ulong) size, (ulong) free_limit, (ulong) n_free,
+		(ulong) n_free_frag, (ulong) frag_n_used, (ulong) n_full_frag,
+		(ulong) seg_id_high, (ulong) seg_id_low);
+
+	mtr_commit(&mtr);
+
+	/* Print segments */
+
+	mtr_start(&mtr);
+	mtr_x_lock(latch, &mtr);
+
+	header = fsp_get_space_header(space, zip_size, &mtr);
+
+	node_addr = flst_get_first(header + FSP_SEG_INODES_FULL, &mtr);
+
+	mtr_commit(&mtr);
+
+	while (!fil_addr_is_null(node_addr)) {
+
+		n = 0;
+
+		do {
+
+			mtr_start(&mtr);
+			mtr_x_lock(latch, &mtr);
+
+			seg_inode_page = fut_get_ptr(
+				space, zip_size, node_addr, RW_X_LATCH, &mtr)
+				- FSEG_INODE_PAGE_NODE;
+
+			seg_inode = fsp_seg_inode_page_get_nth_inode(
+				seg_inode_page, n, zip_size, &mtr);
+			ut_a(!ut_dulint_is_zero(
+				     mach_read_from_8(seg_inode + FSEG_ID)));
+			fseg_print_low(seg_inode, &mtr);
+
+			n_segs++;
+
+			next_node_addr = flst_get_next_addr(
+				seg_inode_page + FSEG_INODE_PAGE_NODE, &mtr);
+			mtr_commit(&mtr);
+		} while (++n < FSP_SEG_INODES_PER_PAGE(zip_size));
+
+		node_addr = next_node_addr;
+	}
+
+	mtr_start(&mtr);
+	mtr_x_lock(latch, &mtr);
+
+	header = fsp_get_space_header(space, zip_size, &mtr);
+
+	node_addr = flst_get_first(header + FSP_SEG_INODES_FREE, &mtr);
+
+	mtr_commit(&mtr);
+
+	while (!fil_addr_is_null(node_addr)) {
+
+		n = 0;
+
+		do {
+
+			mtr_start(&mtr);
+			mtr_x_lock(latch, &mtr);
+
+			seg_inode_page = fut_get_ptr(
+				space, zip_size, node_addr, RW_X_LATCH, &mtr)
+				- FSEG_INODE_PAGE_NODE;
+
+			seg_inode = fsp_seg_inode_page_get_nth_inode(
+				seg_inode_page, n, zip_size, &mtr);
+			if (!ut_dulint_is_zero(
+				    mach_read_from_8(seg_inode + FSEG_ID))) {
+
+				fseg_print_low(seg_inode, &mtr);
+				n_segs++;
+			}
+
+			next_node_addr = flst_get_next_addr(
+				seg_inode_page + FSEG_INODE_PAGE_NODE, &mtr);
+			mtr_commit(&mtr);
+		} while (++n < FSP_SEG_INODES_PER_PAGE(zip_size));
+
+		node_addr = next_node_addr;
+	}
+
+	mtr_commit(&mtr2);
+
+	fprintf(stderr, "NUMBER of file segments: %lu\n", (ulong) n_segs);
+}
diff --git a/storage/xtradb/fut/fut0fut.c b/storage/xtradb/fut/fut0fut.c
new file mode 100644
index 00000000000..41ee0cb6715
--- /dev/null
+++ b/storage/xtradb/fut/fut0fut.c
@@ -0,0 +1,30 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/**********************************************************************
+File-based utilities
+
+Created 12/13/1995 Heikki Tuuri
+***********************************************************************/
+
+#include "fut0fut.h"
+
+#ifdef UNIV_NONINL
+#include "fut0fut.ic"
+#endif
+
diff --git a/storage/xtradb/fut/fut0lst.c b/storage/xtradb/fut/fut0lst.c
new file mode 100644
index 00000000000..bea27ab70d1
--- /dev/null
+++ b/storage/xtradb/fut/fut0lst.c
@@ -0,0 +1,529 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/**********************************************************************
+File-based list utilities
+
+Created 11/28/1995 Heikki Tuuri
+***********************************************************************/
+
+#include "fut0lst.h"
+
+#ifdef UNIV_NONINL
+#include "fut0lst.ic"
+#endif
+
+#include "buf0buf.h"
+#include "page0page.h"
+
+/************************************************************************
+Adds a node to an empty list. */
+static
+void
+flst_add_to_empty(
+/*==============*/
+	flst_base_node_t*	base,	/* in: pointer to base node of
+					empty list */
+	flst_node_t*		node,	/* in: node to add */
+	mtr_t*			mtr)	/* in: mini-transaction handle */
+{
+	ulint		space;
+	fil_addr_t	node_addr;
+	ulint		len;
+
+	ut_ad(mtr && base && node);
+	ut_ad(base != node);
+	ut_ad(mtr_memo_contains_page(mtr, base, MTR_MEMO_PAGE_X_FIX));
+	ut_ad(mtr_memo_contains_page(mtr, node, MTR_MEMO_PAGE_X_FIX));
+	len = flst_get_len(base, mtr);
+	ut_a(len == 0);
+
+	buf_ptr_get_fsp_addr(node, &space, &node_addr);
+
+	/* Update first and last fields of base node */
+	flst_write_addr(base + FLST_FIRST, node_addr, mtr);
+	flst_write_addr(base + FLST_LAST, node_addr, mtr);
+
+	/* Set prev and next fields of node to add */
+	flst_write_addr(node + FLST_PREV, fil_addr_null, mtr);
+	flst_write_addr(node + FLST_NEXT, fil_addr_null, mtr);
+
+	/* Update len of base node */
+	mlog_write_ulint(base + FLST_LEN, len + 1, MLOG_4BYTES, mtr);
+}
+
+/************************************************************************
+Adds a node as the last node in a list. */
+UNIV_INTERN
+void
+flst_add_last(
+/*==========*/
+	flst_base_node_t*	base,	/* in: pointer to base node of list */
+	flst_node_t*		node,	/* in: node to add */
+	mtr_t*			mtr)	/* in: mini-transaction handle */
+{
+	ulint		space;
+	fil_addr_t	node_addr;
+	ulint		len;
+	fil_addr_t	last_addr;
+	flst_node_t*	last_node;
+
+	ut_ad(mtr && base && node);
+	ut_ad(base != node);
+	ut_ad(mtr_memo_contains_page(mtr, base, MTR_MEMO_PAGE_X_FIX));
+	ut_ad(mtr_memo_contains_page(mtr, node, MTR_MEMO_PAGE_X_FIX));
+	len = flst_get_len(base, mtr);
+	last_addr = flst_get_last(base, mtr);
+
+	buf_ptr_get_fsp_addr(node, &space, &node_addr);
+
+	/* If the list is not empty, call flst_insert_after */
+	if (len != 0) {
+		if (last_addr.page == node_addr.page) {
+			last_node = page_align(node) + last_addr.boffset;
+		} else {
+			ulint	zip_size = fil_space_get_zip_size(space);
+
+			last_node = fut_get_ptr(space, zip_size, last_addr,
+						RW_X_LATCH, mtr);
+		}
+
+		flst_insert_after(base, last_node, node, mtr);
+	} else {
+		/* else call flst_add_to_empty */
+		flst_add_to_empty(base, node, mtr);
+	}
+}
+
+/************************************************************************
+Adds a node as the first node in a list. */
+UNIV_INTERN
+void
+flst_add_first(
+/*===========*/
+	flst_base_node_t*	base,	/* in: pointer to base node of list */
+	flst_node_t*		node,	/* in: node to add */
+	mtr_t*			mtr)	/* in: mini-transaction handle */
+{
+	ulint		space;
+	fil_addr_t	node_addr;
+	ulint		len;
+	fil_addr_t	first_addr;
+	flst_node_t*	first_node;
+
+	ut_ad(mtr && base && node);
+	ut_ad(base != node);
+	ut_ad(mtr_memo_contains_page(mtr, base, MTR_MEMO_PAGE_X_FIX));
+	ut_ad(mtr_memo_contains_page(mtr, node, MTR_MEMO_PAGE_X_FIX));
+	len = flst_get_len(base, mtr);
+	first_addr = flst_get_first(base, mtr);
+
+	buf_ptr_get_fsp_addr(node, &space, &node_addr);
+
+	/* If the list is not empty, call flst_insert_before */
+	if (len != 0) {
+		if (first_addr.page == node_addr.page) {
+			first_node = page_align(node) + first_addr.boffset;
+		} else {
+			ulint	zip_size = fil_space_get_zip_size(space);
+
+			first_node = fut_get_ptr(space, zip_size, first_addr,
+						 RW_X_LATCH, mtr);
+		}
+
+		flst_insert_before(base, node, first_node, mtr);
+	} else {
+		/* else call flst_add_to_empty */
+		flst_add_to_empty(base, node, mtr);
+	}
+}
+
+/************************************************************************
+Inserts a node after another in a list. */
+UNIV_INTERN
+void
+flst_insert_after(
+/*==============*/
+	flst_base_node_t*	base,	/* in: pointer to base node of list */
+	flst_node_t*		node1,	/* in: node to insert after */
+	flst_node_t*		node2,	/* in: node to add */
+	mtr_t*			mtr)	/* in: mini-transaction handle */
+{
+	ulint		space;
+	fil_addr_t	node1_addr;
+	fil_addr_t	node2_addr;
+	flst_node_t*	node3;
+	fil_addr_t	node3_addr;
+	ulint		len;
+
+	ut_ad(mtr && node1 && node2 && base);
+	ut_ad(base != node1);
+	ut_ad(base != node2);
+	ut_ad(node2 != node1);
+	ut_ad(mtr_memo_contains_page(mtr, base, MTR_MEMO_PAGE_X_FIX));
+	ut_ad(mtr_memo_contains_page(mtr, node1, MTR_MEMO_PAGE_X_FIX));
+	ut_ad(mtr_memo_contains_page(mtr, node2, MTR_MEMO_PAGE_X_FIX));
+
+	buf_ptr_get_fsp_addr(node1, &space, &node1_addr);
+	buf_ptr_get_fsp_addr(node2, &space, &node2_addr);
+
+	node3_addr = flst_get_next_addr(node1, mtr);
+
+	/* Set prev and next fields of node2 */
+	flst_write_addr(node2 + FLST_PREV, node1_addr, mtr);
+	flst_write_addr(node2 + FLST_NEXT, node3_addr, mtr);
+
+	if (!fil_addr_is_null(node3_addr)) {
+		/* Update prev field of node3 */
+		ulint	zip_size = fil_space_get_zip_size(space);
+
+		node3 = fut_get_ptr(space, zip_size,
+				    node3_addr, RW_X_LATCH, mtr);
+		flst_write_addr(node3 + FLST_PREV, node2_addr, mtr);
+	} else {
+		/* node1 was last in list: update last field in base */
+		flst_write_addr(base + FLST_LAST, node2_addr, mtr);
+	}
+
+	/* Set next field of node1 */
+	flst_write_addr(node1 + FLST_NEXT, node2_addr, mtr);
+
+	/* Update len of base node */
+	len = flst_get_len(base, mtr);
+	mlog_write_ulint(base + FLST_LEN, len + 1, MLOG_4BYTES, mtr);
+}
+
+/************************************************************************
+Inserts a node before another in a list. */
+UNIV_INTERN
+void
+flst_insert_before(
+/*===============*/
+	flst_base_node_t*	base,	/* in: pointer to base node of list */
+	flst_node_t*		node2,	/* in: node to insert */
+	flst_node_t*		node3,	/* in: node to insert before */
+	mtr_t*			mtr)	/* in: mini-transaction handle */
+{
+	ulint		space;
+	flst_node_t*	node1;
+	fil_addr_t	node1_addr;
+	fil_addr_t	node2_addr;
+	fil_addr_t	node3_addr;
+	ulint		len;
+
+	ut_ad(mtr && node2 && node3 && base);
+	ut_ad(base != node2);
+	ut_ad(base != node3);
+	ut_ad(node2 != node3);
+	ut_ad(mtr_memo_contains_page(mtr, base, MTR_MEMO_PAGE_X_FIX));
+	ut_ad(mtr_memo_contains_page(mtr, node2, MTR_MEMO_PAGE_X_FIX));
+	ut_ad(mtr_memo_contains_page(mtr, node3, MTR_MEMO_PAGE_X_FIX));
+
+	buf_ptr_get_fsp_addr(node2, &space, &node2_addr);
+	buf_ptr_get_fsp_addr(node3, &space, &node3_addr);
+
+	node1_addr = flst_get_prev_addr(node3, mtr);
+
+	/* Set prev and next fields of node2 */
+	flst_write_addr(node2 + FLST_PREV, node1_addr, mtr);
+	flst_write_addr(node2 + FLST_NEXT, node3_addr, mtr);
+
+	if (!fil_addr_is_null(node1_addr)) {
+		ulint	zip_size = fil_space_get_zip_size(space);
+		/* Update next field of node1 */
+		node1 = fut_get_ptr(space, zip_size, node1_addr,
+				    RW_X_LATCH, mtr);
+		flst_write_addr(node1 + FLST_NEXT, node2_addr, mtr);
+	} else {
+		/* node3 was first in list: update first field in base */
+		flst_write_addr(base + FLST_FIRST, node2_addr, mtr);
+	}
+
+	/* Set prev field of node3 */
+	flst_write_addr(node3 + FLST_PREV, node2_addr, mtr);
+
+	/* Update len of base node */
+	len = flst_get_len(base, mtr);
+	mlog_write_ulint(base + FLST_LEN, len + 1, MLOG_4BYTES, mtr);
+}
+
+/************************************************************************
+Removes a node. */
+UNIV_INTERN
+void
+flst_remove(
+/*========*/
+	flst_base_node_t*	base,	/* in: pointer to base node of list */
+	flst_node_t*		node2,	/* in: node to remove */
+	mtr_t*			mtr)	/* in: mini-transaction handle */
+{
+	ulint		space;
+	ulint		zip_size;
+	flst_node_t*	node1;
+	fil_addr_t	node1_addr;
+	fil_addr_t	node2_addr;
+	flst_node_t*	node3;
+	fil_addr_t	node3_addr;
+	ulint		len;
+
+	ut_ad(mtr && node2 && base);
+	ut_ad(mtr_memo_contains_page(mtr, base, MTR_MEMO_PAGE_X_FIX));
+	ut_ad(mtr_memo_contains_page(mtr, node2, MTR_MEMO_PAGE_X_FIX));
+
+	buf_ptr_get_fsp_addr(node2, &space, &node2_addr);
+	zip_size = fil_space_get_zip_size(space);
+
+	node1_addr = flst_get_prev_addr(node2, mtr);
+	node3_addr = flst_get_next_addr(node2, mtr);
+
+	if (!fil_addr_is_null(node1_addr)) {
+
+		/* Update next field of node1 */
+
+		if (node1_addr.page == node2_addr.page) {
+
+			node1 = page_align(node2) + node1_addr.boffset;
+		} else {
+			node1 = fut_get_ptr(space, zip_size,
+					    node1_addr, RW_X_LATCH, mtr);
+		}
+
+		ut_ad(node1 != node2);
+
+		flst_write_addr(node1 + FLST_NEXT, node3_addr, mtr);
+	} else {
+		/* node2 was first in list: update first field in base */
+		flst_write_addr(base + FLST_FIRST, node3_addr, mtr);
+	}
+
+	if (!fil_addr_is_null(node3_addr)) {
+		/* Update prev field of node3 */
+
+		if (node3_addr.page == node2_addr.page) {
+
+			node3 = page_align(node2) + node3_addr.boffset;
+		} else {
+			node3 = fut_get_ptr(space, zip_size,
+					    node3_addr, RW_X_LATCH, mtr);
+		}
+
+		ut_ad(node2 != node3);
+
+		flst_write_addr(node3 + FLST_PREV, node1_addr, mtr);
+	} else {
+		/* node2 was last in list: update last field in base */
+		flst_write_addr(base + FLST_LAST, node1_addr, mtr);
+	}
+
+	/* Update len of base node */
+	len = flst_get_len(base, mtr);
+	ut_ad(len > 0);
+
+	mlog_write_ulint(base + FLST_LEN, len - 1, MLOG_4BYTES, mtr);
+}
+
+/************************************************************************
+Cuts off the tail of the list, including the node given. The number of
+nodes which will be removed must be provided by the caller, as this function
+does not measure the length of the tail. */
+UNIV_INTERN
+void
+flst_cut_end(
+/*=========*/
+	flst_base_node_t*	base,	/* in: pointer to base node of list */
+	flst_node_t*		node2,	/* in: first node to remove */
+	ulint			n_nodes,/* in: number of nodes to remove,
+					must be >= 1 */
+	mtr_t*			mtr)	/* in: mini-transaction handle */
+{
+	ulint		space;
+	flst_node_t*	node1;
+	fil_addr_t	node1_addr;
+	fil_addr_t	node2_addr;
+	ulint		len;
+
+	ut_ad(mtr && node2 && base);
+	ut_ad(mtr_memo_contains_page(mtr, base, MTR_MEMO_PAGE_X_FIX));
+	ut_ad(mtr_memo_contains_page(mtr, node2, MTR_MEMO_PAGE_X_FIX));
+	ut_ad(n_nodes > 0);
+
+	buf_ptr_get_fsp_addr(node2, &space, &node2_addr);
+
+	node1_addr = flst_get_prev_addr(node2, mtr);
+
+	if (!fil_addr_is_null(node1_addr)) {
+
+		/* Update next field of node1 */
+
+		if (node1_addr.page == node2_addr.page) {
+
+			node1 = page_align(node2) + node1_addr.boffset;
+		} else {
+			node1 = fut_get_ptr(space,
+					    fil_space_get_zip_size(space),
+					    node1_addr, RW_X_LATCH, mtr);
+		}
+
+		flst_write_addr(node1 + FLST_NEXT, fil_addr_null, mtr);
+	} else {
+		/* node2 was first in list: update the field in base */
+		flst_write_addr(base + FLST_FIRST, fil_addr_null, mtr);
+	}
+
+	flst_write_addr(base + FLST_LAST, node1_addr, mtr);
+
+	/* Update len of base node */
+	len = flst_get_len(base, mtr);
+	ut_ad(len >= n_nodes);
+
+	mlog_write_ulint(base + FLST_LEN, len - n_nodes, MLOG_4BYTES, mtr);
+}
+
+/************************************************************************
+Cuts off the tail of the list, not including the given node. The number of
+nodes which will be removed must be provided by the caller, as this function
+does not measure the length of the tail. */
+UNIV_INTERN
+void
+flst_truncate_end(
+/*==============*/
+	flst_base_node_t*	base,	/* in: pointer to base node of list */
+	flst_node_t*		node2,	/* in: first node not to remove */
+	ulint			n_nodes,/* in: number of nodes to remove */
+	mtr_t*			mtr)	/* in: mini-transaction handle */
+{
+	fil_addr_t	node2_addr;
+	ulint		len;
+	ulint		space;
+
+	ut_ad(mtr && node2 && base);
+	ut_ad(mtr_memo_contains_page(mtr, base, MTR_MEMO_PAGE_X_FIX));
+	ut_ad(mtr_memo_contains_page(mtr, node2, MTR_MEMO_PAGE_X_FIX));
+	if (n_nodes == 0) {
+
+		ut_ad(fil_addr_is_null(flst_get_next_addr(node2, mtr)));
+
+		return;
+	}
+
+	buf_ptr_get_fsp_addr(node2, &space, &node2_addr);
+
+	/* Update next field of node2 */
+	flst_write_addr(node2 + FLST_NEXT, fil_addr_null, mtr);
+
+	flst_write_addr(base + FLST_LAST, node2_addr, mtr);
+
+	/* Update len of base node */
+	len = flst_get_len(base, mtr);
+	ut_ad(len >= n_nodes);
+
+	mlog_write_ulint(base + FLST_LEN, len - n_nodes, MLOG_4BYTES, mtr);
+}
+
+/************************************************************************
+Validates a file-based list. */
+UNIV_INTERN
+ibool
+flst_validate(
+/*==========*/
+					/* out: TRUE if ok */
+	const flst_base_node_t*	base,	/* in: pointer to base node of list */
+	mtr_t*			mtr1)	/* in: mtr */
+{
+	ulint			space;
+	ulint			zip_size;
+	const flst_node_t*	node;
+	fil_addr_t		node_addr;
+	fil_addr_t		base_addr;
+	ulint			len;
+	ulint			i;
+	mtr_t			mtr2;
+
+	ut_ad(base);
+	ut_ad(mtr_memo_contains_page(mtr1, base, MTR_MEMO_PAGE_X_FIX));
+
+	/* We use two mini-transaction handles: the first is used to
+	lock the base node, and prevent other threads from modifying the
+	list. The second is used to traverse the list. We cannot run the
+	second mtr without committing it at times, because if the list
+	is long, then the x-locked pages could fill the buffer resulting
+	in a deadlock. */
+
+	/* Find out the space id */
+	buf_ptr_get_fsp_addr(base, &space, &base_addr);
+	zip_size = fil_space_get_zip_size(space);
+
+	len = flst_get_len(base, mtr1);
+	node_addr = flst_get_first(base, mtr1);
+
+	for (i = 0; i < len; i++) {
+		mtr_start(&mtr2);
+
+		node = fut_get_ptr(space, zip_size,
+				   node_addr, RW_X_LATCH, &mtr2);
+		node_addr = flst_get_next_addr(node, &mtr2);
+
+		mtr_commit(&mtr2); /* Commit mtr2 each round to prevent buffer
+				   becoming full */
+	}
+
+	ut_a(fil_addr_is_null(node_addr));
+
+	node_addr = flst_get_last(base, mtr1);
+
+	for (i = 0; i < len; i++) {
+		mtr_start(&mtr2);
+
+		node = fut_get_ptr(space, zip_size,
+				   node_addr, RW_X_LATCH, &mtr2);
+		node_addr = flst_get_prev_addr(node, &mtr2);
+
+		mtr_commit(&mtr2); /* Commit mtr2 each round to prevent buffer
+				   becoming full */
+	}
+
+	ut_a(fil_addr_is_null(node_addr));
+
+	return(TRUE);
+}
+
+/************************************************************************
+Prints info of a file-based list. */
+UNIV_INTERN
+void
+flst_print(
+/*=======*/
+	const flst_base_node_t*	base,	/* in: pointer to base node of list */
+	mtr_t*			mtr)	/* in: mtr */
+{
+	const buf_frame_t*	frame;
+	ulint			len;
+
+	ut_ad(base && mtr);
+	ut_ad(mtr_memo_contains_page(mtr, base, MTR_MEMO_PAGE_X_FIX));
+	frame = page_align((byte*) base);
+
+	len = flst_get_len(base, mtr);
+
+	fprintf(stderr,
+		"FILE-BASED LIST:\n"
+		"Base node in space %lu page %lu byte offset %lu; len %lu\n",
+		(ulong) page_get_space_id(frame),
+		(ulong) page_get_page_no(frame),
+		(ulong) page_offset(base), (ulong) len);
+}
diff --git a/storage/xtradb/ha/ha0ha.c b/storage/xtradb/ha/ha0ha.c
new file mode 100644
index 00000000000..1ecba3df663
--- /dev/null
+++ b/storage/xtradb/ha/ha0ha.c
@@ -0,0 +1,442 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/************************************************************************
+The hash table with external chains
+
+Created 8/22/1994 Heikki Tuuri
+*************************************************************************/
+
+#include "ha0ha.h"
+#ifdef UNIV_NONINL
+#include "ha0ha.ic"
+#endif
+
+#ifdef UNIV_DEBUG
+# include "buf0buf.h"
+#endif /* UNIV_DEBUG */
+#ifdef UNIV_SYNC_DEBUG
+# include "btr0sea.h"
+#endif /* UNIV_SYNC_DEBUG */
+#include "page0page.h"
+
+/*****************************************************************
+Creates a hash table with >= n array cells. The actual number of cells is
+chosen to be a prime number slightly bigger than n. */
+UNIV_INTERN
+hash_table_t*
+ha_create_func(
+/*===========*/
+				/* out, own: created table */
+	ulint	n,		/* in: number of array cells */
+#ifdef UNIV_SYNC_DEBUG
+	ulint	mutex_level,	/* in: level of the mutexes in the latching
+				order: this is used in the debug version */
+#endif /* UNIV_SYNC_DEBUG */
+	ulint	n_mutexes)	/* in: number of mutexes to protect the
+				hash table: must be a power of 2, or 0 */
+{
+	hash_table_t*	table;
+	ulint		i;
+
+	table = hash_create(n);
+
+#if defined UNIV_AHI_DEBUG || defined UNIV_DEBUG
+	table->adaptive = TRUE;
+#endif /* UNIV_AHI_DEBUG || UNIV_DEBUG */
+	/* Creating MEM_HEAP_BTR_SEARCH type heaps can potentially fail,
+	but in practise it never should in this case, hence the asserts. */
+
+	if (n_mutexes == 0) {
+		table->heap = mem_heap_create_in_btr_search(
+			ut_min(4096, MEM_MAX_ALLOC_IN_BUF));
+		ut_a(table->heap);
+
+		return(table);
+	}
+
+	hash_create_mutexes(table, n_mutexes, mutex_level);
+
+	table->heaps = mem_alloc(n_mutexes * sizeof(void*));
+
+	for (i = 0; i < n_mutexes; i++) {
+		table->heaps[i] = mem_heap_create_in_btr_search(4096);
+		ut_a(table->heaps[i]);
+	}
+
+	return(table);
+}
+
+/*****************************************************************
+Empties a hash table and frees the memory heaps. */
+UNIV_INTERN
+void
+ha_clear(
+/*=====*/
+	hash_table_t*	table)	/* in, own: hash table */
+{
+	ulint	i;
+	ulint	n;
+
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(rw_lock_own(&btr_search_latch, RW_LOCK_EXCLUSIVE));
+#endif /* UNIV_SYNC_DEBUG */
+
+	/* Free the memory heaps. */
+	n = table->n_mutexes;
+
+	for (i = 0; i < n; i++) {
+		mem_heap_free(table->heaps[i]);
+	}
+
+	/* Clear the hash table. */
+	n = hash_get_n_cells(table);
+
+	for (i = 0; i < n; i++) {
+		hash_get_nth_cell(table, i)->node = NULL;
+	}
+}
+
+/*****************************************************************
+Inserts an entry into a hash table. If an entry with the same fold number
+is found, its node is updated to point to the new data, and no new node
+is inserted. */
+UNIV_INTERN
+ibool
+ha_insert_for_fold_func(
+/*====================*/
+				/* out: TRUE if succeed, FALSE if no more
+				memory could be allocated */
+	hash_table_t*	table,	/* in: hash table */
+	ulint		fold,	/* in: folded value of data; if a node with
+				the same fold value already exists, it is
+				updated to point to the same data, and no new
+				node is created! */
+#if defined UNIV_AHI_DEBUG || defined UNIV_DEBUG
+	buf_block_t*	block,	/* in: buffer block containing the data */
+#endif /* UNIV_AHI_DEBUG || UNIV_DEBUG */
+	void*		data)	/* in: data, must not be NULL */
+{
+	hash_cell_t*	cell;
+	ha_node_t*	node;
+	ha_node_t*	prev_node;
+	ulint		hash;
+
+	ut_ad(table && data);
+#if defined UNIV_AHI_DEBUG || defined UNIV_DEBUG
+	ut_a(block->frame == page_align(data));
+#endif /* UNIV_AHI_DEBUG || UNIV_DEBUG */
+	ut_ad(!table->mutexes || mutex_own(hash_get_mutex(table, fold)));
+
+	hash = hash_calc_hash(fold, table);
+
+	cell = hash_get_nth_cell(table, hash);
+
+	prev_node = cell->node;
+
+	while (prev_node != NULL) {
+		if (prev_node->fold == fold) {
+#if defined UNIV_AHI_DEBUG || defined UNIV_DEBUG
+			if (table->adaptive) {
+				buf_block_t* prev_block = prev_node->block;
+				ut_a(prev_block->frame
+				     == page_align(prev_node->data));
+				ut_a(prev_block->n_pointers > 0);
+				prev_block->n_pointers--;
+				block->n_pointers++;
+			}
+
+			prev_node->block = block;
+#endif /* UNIV_AHI_DEBUG || UNIV_DEBUG */
+			prev_node->data = data;
+
+			return(TRUE);
+		}
+
+		prev_node = prev_node->next;
+	}
+
+	/* We have to allocate a new chain node */
+
+	node = mem_heap_alloc(hash_get_heap(table, fold), sizeof(ha_node_t));
+
+	if (node == NULL) {
+		/* It was a btr search type memory heap and at the moment
+		no more memory could be allocated: return */
+
+		ut_ad(hash_get_heap(table, fold)->type & MEM_HEAP_BTR_SEARCH);
+
+		return(FALSE);
+	}
+
+	ha_node_set_data(node, block, data);
+
+#if defined UNIV_AHI_DEBUG || defined UNIV_DEBUG
+	if (table->adaptive) {
+		block->n_pointers++;
+	}
+#endif /* UNIV_AHI_DEBUG || UNIV_DEBUG */
+	node->fold = fold;
+
+	node->next = NULL;
+
+	prev_node = cell->node;
+
+	if (prev_node == NULL) {
+
+		cell->node = node;
+
+		return(TRUE);
+	}
+
+	while (prev_node->next != NULL) {
+
+		prev_node = prev_node->next;
+	}
+
+	prev_node->next = node;
+
+	return(TRUE);
+}
+
+/***************************************************************
+Deletes a hash node. */
+UNIV_INTERN
+void
+ha_delete_hash_node(
+/*================*/
+	hash_table_t*	table,		/* in: hash table */
+	ha_node_t*	del_node)	/* in: node to be deleted */
+{
+#if defined UNIV_AHI_DEBUG || defined UNIV_DEBUG
+	if (table->adaptive) {
+		ut_a(del_node->block->frame = page_align(del_node->data));
+		ut_a(del_node->block->n_pointers > 0);
+		del_node->block->n_pointers--;
+	}
+#endif /* UNIV_AHI_DEBUG || UNIV_DEBUG */
+	HASH_DELETE_AND_COMPACT(ha_node_t, next, table, del_node);
+}
+
+/*****************************************************************
+Deletes an entry from a hash table. */
+UNIV_INTERN
+void
+ha_delete(
+/*======*/
+	hash_table_t*	table,	/* in: hash table */
+	ulint		fold,	/* in: folded value of data */
+	void*		data)	/* in: data, must not be NULL and must exist
+				in the hash table */
+{
+	ha_node_t*	node;
+
+	ut_ad(!table->mutexes || mutex_own(hash_get_mutex(table, fold)));
+
+	node = ha_search_with_data(table, fold, data);
+
+	ut_a(node);
+
+	ha_delete_hash_node(table, node);
+}
+
+/*************************************************************
+Looks for an element when we know the pointer to the data, and updates
+the pointer to data, if found. */
+UNIV_INTERN
+void
+ha_search_and_update_if_found_func(
+/*===============================*/
+	hash_table_t*	table,	/* in: hash table */
+	ulint		fold,	/* in: folded value of the searched data */
+	void*		data,	/* in: pointer to the data */
+#if defined UNIV_AHI_DEBUG || defined UNIV_DEBUG
+	buf_block_t*	new_block,/* in: block containing new_data */
+#endif /* UNIV_AHI_DEBUG || UNIV_DEBUG */
+	void*		new_data)/* in: new pointer to the data */
+{
+	ha_node_t*	node;
+
+	ut_ad(!table->mutexes || mutex_own(hash_get_mutex(table, fold)));
+#if defined UNIV_AHI_DEBUG || defined UNIV_DEBUG
+	ut_a(new_block->frame == page_align(new_data));
+#endif /* UNIV_AHI_DEBUG || UNIV_DEBUG */
+
+	node = ha_search_with_data(table, fold, data);
+
+	if (node) {
+#if defined UNIV_AHI_DEBUG || defined UNIV_DEBUG
+		if (table->adaptive) {
+			ut_a(node->block->n_pointers > 0);
+			node->block->n_pointers--;
+			new_block->n_pointers++;
+		}
+
+		node->block = new_block;
+#endif /* UNIV_AHI_DEBUG || UNIV_DEBUG */
+		node->data = new_data;
+	}
+}
+
+/*********************************************************************
+Removes from the chain determined by fold all nodes whose data pointer
+points to the page given. */
+UNIV_INTERN
+void
+ha_remove_all_nodes_to_page(
+/*========================*/
+	hash_table_t*	table,	/* in: hash table */
+	ulint		fold,	/* in: fold value */
+	const page_t*	page)	/* in: buffer page */
+{
+	ha_node_t*	node;
+
+	ut_ad(!table->mutexes || mutex_own(hash_get_mutex(table, fold)));
+
+	node = ha_chain_get_first(table, fold);
+
+	while (node) {
+		if (page_align(ha_node_get_data(node)) == page) {
+
+			/* Remove the hash node */
+
+			ha_delete_hash_node(table, node);
+
+			/* Start again from the first node in the chain
+			because the deletion may compact the heap of
+			nodes and move other nodes! */
+
+			node = ha_chain_get_first(table, fold);
+		} else {
+			node = ha_chain_get_next(node);
+		}
+	}
+#ifdef UNIV_DEBUG
+	/* Check that all nodes really got deleted */
+
+	node = ha_chain_get_first(table, fold);
+
+	while (node) {
+		ut_a(page_align(ha_node_get_data(node)) != page);
+
+		node = ha_chain_get_next(node);
+	}
+#endif
+}
+
+/*****************************************************************
+Validates a given range of the cells in hash table. */
+UNIV_INTERN
+ibool
+ha_validate(
+/*========*/
+					/* out: TRUE if ok */
+	hash_table_t*	table,		/* in: hash table */
+	ulint		start_index,	/* in: start index */
+	ulint		end_index)	/* in: end index */
+{
+	hash_cell_t*	cell;
+	ha_node_t*	node;
+	ibool		ok	= TRUE;
+	ulint		i;
+
+	ut_a(start_index <= end_index);
+	ut_a(start_index < hash_get_n_cells(table));
+	ut_a(end_index < hash_get_n_cells(table));
+
+	for (i = start_index; i <= end_index; i++) {
+
+		cell = hash_get_nth_cell(table, i);
+
+		node = cell->node;
+
+		while (node) {
+			if (hash_calc_hash(node->fold, table) != i) {
+				ut_print_timestamp(stderr);
+				fprintf(stderr,
+					"InnoDB: Error: hash table node"
+					" fold value %lu does not\n"
+					"InnoDB: match the cell number %lu.\n",
+					(ulong) node->fold, (ulong) i);
+
+				ok = FALSE;
+			}
+
+			node = node->next;
+		}
+	}
+
+	return(ok);
+}
+
+/*****************************************************************
+Prints info of a hash table. */
+UNIV_INTERN
+void
+ha_print_info(
+/*==========*/
+	FILE*		file,	/* in: file where to print */
+	hash_table_t*	table)	/* in: hash table */
+{
+#ifdef UNIV_DEBUG
+/* Some of the code here is disabled for performance reasons in production
+builds, see http://bugs.mysql.com/36941 */
+#define PRINT_USED_CELLS
+#endif /* UNIV_DEBUG */
+
+#ifdef PRINT_USED_CELLS
+	hash_cell_t*	cell;
+	ulint		cells	= 0;
+	ulint		i;
+#endif /* PRINT_USED_CELLS */
+	ulint		n_bufs;
+
+#ifdef PRINT_USED_CELLS
+	for (i = 0; i < hash_get_n_cells(table); i++) {
+
+		cell = hash_get_nth_cell(table, i);
+
+		if (cell->node) {
+
+			cells++;
+		}
+	}
+#endif /* PRINT_USED_CELLS */
+
+	fprintf(file, "Hash table size %lu",
+		(ulong) hash_get_n_cells(table));
+
+#ifdef PRINT_USED_CELLS
+	fprintf(file, ", used cells %lu", (ulong) cells);
+#endif /* PRINT_USED_CELLS */
+
+	if (table->heaps == NULL && table->heap != NULL) {
+
+		/* This calculation is intended for the adaptive hash
+		index: how many buffer frames we have reserved? */
+
+		n_bufs = UT_LIST_GET_LEN(table->heap->base) - 1;
+
+		if (table->heap->free_block) {
+			n_bufs++;
+		}
+
+		fprintf(file, ", node heap has %lu buffer(s)\n",
+			(ulong) n_bufs);
+	}
+}
diff --git a/storage/xtradb/ha/ha0storage.c b/storage/xtradb/ha/ha0storage.c
new file mode 100644
index 00000000000..e7e09591193
--- /dev/null
+++ b/storage/xtradb/ha/ha0storage.c
@@ -0,0 +1,183 @@
+/*****************************************************************************
+
+Copyright (c) 2007, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Hash storage.
+Provides a data structure that stores chunks of data in
+its own storage, avoiding duplicates.
+
+Created September 22, 2007 Vasil Dimov
+*******************************************************/
+
+#include "univ.i"
+#include "ha0storage.h"
+#include "hash0hash.h"
+#include "mem0mem.h"
+#include "ut0rnd.h"
+
+#ifdef UNIV_NONINL
+#include "ha0storage.ic"
+#endif
+
+/***********************************************************************
+Retrieves a data from a storage. If it is present, a pointer to the
+stored copy of data is returned, otherwise NULL is returned. */
+static
+const void*
+ha_storage_get(
+/*===========*/
+	ha_storage_t*	storage,	/* in: hash storage */
+	const void*	data,		/* in: data to check for */
+	ulint		data_len)	/* in: data length */
+{
+	ha_storage_node_t*	node;
+	ulint			fold;
+
+	/* avoid repetitive calls to ut_fold_binary() in the HASH_SEARCH
+	macro */
+	fold = ut_fold_binary(data, data_len);
+
+#define IS_FOUND	\
+	node->data_len == data_len && memcmp(node->data, data, data_len) == 0
+
+	HASH_SEARCH(
+		next,			/* node->"next" */
+		storage->hash,		/* the hash table */
+		fold,			/* key */
+		ha_storage_node_t*,	/* type of node->next */
+		node,			/* auxiliary variable */
+		,			/* assertion */
+		IS_FOUND);		/* search criteria */
+
+	if (node == NULL) {
+
+		return(NULL);
+	}
+	/* else */
+
+	return(node->data);
+}
+
+/***********************************************************************
+Copies data into the storage and returns a pointer to the copy. If the
+same data chunk is already present, then pointer to it is returned.
+Data chunks are considered to be equal if len1 == len2 and
+memcmp(data1, data2, len1) == 0. If "data" is not present (and thus
+data_len bytes need to be allocated) and the size of storage is going to
+become more than "memlim" then "data" is not added and NULL is returned.
+To disable this behavior "memlim" can be set to 0, which stands for
+"no limit". */
+UNIV_INTERN
+const void*
+ha_storage_put_memlim(
+/*==================*/
+	ha_storage_t*	storage,	/* in/out: hash storage */
+	const void*	data,		/* in: data to store */
+	ulint		data_len,	/* in: data length */
+	ulint		memlim)		/* in: memory limit to obey */
+{
+	void*			raw;
+	ha_storage_node_t*	node;
+	const void*		data_copy;
+	ulint			fold;
+
+	/* check if data chunk is already present */
+	data_copy = ha_storage_get(storage, data, data_len);
+	if (data_copy != NULL) {
+
+		return(data_copy);
+	}
+
+	/* not present */
+
+	/* check if we are allowed to allocate data_len bytes */
+	if (memlim > 0
+	    && ha_storage_get_size(storage) + data_len > memlim) {
+
+		return(NULL);
+	}
+
+	/* we put the auxiliary node struct and the data itself in one
+	continuous block */
+	raw = mem_heap_alloc(storage->heap,
+			     sizeof(ha_storage_node_t) + data_len);
+
+	node = (ha_storage_node_t*) raw;
+	data_copy = (byte*) raw + sizeof(*node);
+
+	memcpy((byte*) raw + sizeof(*node), data, data_len);
+
+	node->data_len = data_len;
+	node->data = data_copy;
+
+	/* avoid repetitive calls to ut_fold_binary() in the HASH_INSERT
+	macro */
+	fold = ut_fold_binary(data, data_len);
+
+	HASH_INSERT(
+		ha_storage_node_t,	/* type used in the hash chain */
+		next,			/* node->"next" */
+		storage->hash,		/* the hash table */
+		fold,			/* key */
+		node);			/* add this data to the hash */
+
+	/* the output should not be changed because it will spoil the
+	hash table */
+	return(data_copy);
+}
+
+#ifdef UNIV_COMPILE_TEST_FUNCS
+
+void
+test_ha_storage()
+{
+	ha_storage_t*	storage;
+	char		buf[1024];
+	int		i;
+	const void*	stored[256];
+	const void*	p;
+
+	storage = ha_storage_create(0, 0);
+
+	for (i = 0; i < 256; i++) {
+
+		memset(buf, i, sizeof(buf));
+		stored[i] = ha_storage_put(storage, buf, sizeof(buf));
+	}
+
+	//ha_storage_empty(&storage);
+
+	for (i = 255; i >= 0; i--) {
+
+		memset(buf, i, sizeof(buf));
+		p = ha_storage_put(storage, buf, sizeof(buf));
+
+		if (p != stored[i]) {
+
+			fprintf(stderr, "ha_storage_put() returned %p "
+				"instead of %p, i=%d\n", p, stored[i], i);
+			return;
+		}
+	}
+
+	fprintf(stderr, "all ok\n");
+
+	ha_storage_free(storage);
+}
+
+#endif /* UNIV_COMPILE_TEST_FUNCS */
diff --git a/storage/xtradb/ha/hash0hash.c b/storage/xtradb/ha/hash0hash.c
new file mode 100644
index 00000000000..9694a288c99
--- /dev/null
+++ b/storage/xtradb/ha/hash0hash.c
@@ -0,0 +1,165 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The simple hash table utility
+
+Created 5/20/1997 Heikki Tuuri
+*******************************************************/
+
+#include "hash0hash.h"
+#ifdef UNIV_NONINL
+#include "hash0hash.ic"
+#endif
+
+#include "mem0mem.h"
+
+/****************************************************************
+Reserves the mutex for a fold value in a hash table. */
+UNIV_INTERN
+void
+hash_mutex_enter(
+/*=============*/
+	hash_table_t*	table,	/* in: hash table */
+	ulint		fold)	/* in: fold */
+{
+	mutex_enter(hash_get_mutex(table, fold));
+}
+
+/****************************************************************
+Releases the mutex for a fold value in a hash table. */
+UNIV_INTERN
+void
+hash_mutex_exit(
+/*============*/
+	hash_table_t*	table,	/* in: hash table */
+	ulint		fold)	/* in: fold */
+{
+	mutex_exit(hash_get_mutex(table, fold));
+}
+
+/****************************************************************
+Reserves all the mutexes of a hash table, in an ascending order. */
+UNIV_INTERN
+void
+hash_mutex_enter_all(
+/*=================*/
+	hash_table_t*	table)	/* in: hash table */
+{
+	ulint	i;
+
+	for (i = 0; i < table->n_mutexes; i++) {
+
+		mutex_enter(table->mutexes + i);
+	}
+}
+
+/****************************************************************
+Releases all the mutexes of a hash table. */
+UNIV_INTERN
+void
+hash_mutex_exit_all(
+/*================*/
+	hash_table_t*	table)	/* in: hash table */
+{
+	ulint	i;
+
+	for (i = 0; i < table->n_mutexes; i++) {
+
+		mutex_exit(table->mutexes + i);
+	}
+}
+
+/*****************************************************************
+Creates a hash table with >= n array cells. The actual number of cells is
+chosen to be a prime number slightly bigger than n. */
+UNIV_INTERN
+hash_table_t*
+hash_create(
+/*========*/
+			/* out, own: created table */
+	ulint	n)	/* in: number of array cells */
+{
+	hash_cell_t*	array;
+	ulint		prime;
+	hash_table_t*	table;
+
+	prime = ut_find_prime(n);
+
+	table = mem_alloc(sizeof(hash_table_t));
+
+	array = ut_malloc(sizeof(hash_cell_t) * prime);
+
+#if defined UNIV_AHI_DEBUG || defined UNIV_DEBUG
+	table->adaptive = FALSE;
+#endif /* UNIV_AHI_DEBUG || UNIV_DEBUG */
+	table->array = array;
+	table->n_cells = prime;
+	table->n_mutexes = 0;
+	table->mutexes = NULL;
+	table->heaps = NULL;
+	table->heap = NULL;
+	table->magic_n = HASH_TABLE_MAGIC_N;
+
+	/* Initialize the cell array */
+	hash_table_clear(table);
+
+	return(table);
+}
+
+/*****************************************************************
+Frees a hash table. */
+UNIV_INTERN
+void
+hash_table_free(
+/*============*/
+	hash_table_t*	table)	/* in, own: hash table */
+{
+	ut_a(table->mutexes == NULL);
+
+	ut_free(table->array);
+	mem_free(table);
+}
+
+/*****************************************************************
+Creates a mutex array to protect a hash table. */
+UNIV_INTERN
+void
+hash_create_mutexes_func(
+/*=====================*/
+	hash_table_t*	table,		/* in: hash table */
+#ifdef UNIV_SYNC_DEBUG
+	ulint		sync_level,	/* in: latching order level of the
+					mutexes: used in the debug version */
+#endif /* UNIV_SYNC_DEBUG */
+	ulint		n_mutexes)	/* in: number of mutexes, must be a
+					power of 2 */
+{
+	ulint	i;
+
+	ut_a(n_mutexes > 0);
+	ut_a(ut_is_2pow(n_mutexes));
+
+	table->mutexes = mem_alloc(n_mutexes * sizeof(mutex_t));
+
+	for (i = 0; i < n_mutexes; i++) {
+		mutex_create(table->mutexes + i, sync_level);
+	}
+
+	table->n_mutexes = n_mutexes;
+}
diff --git a/storage/xtradb/ha_innodb.def b/storage/xtradb/ha_innodb.def
new file mode 100644
index 00000000000..e0faa62deb1
--- /dev/null
+++ b/storage/xtradb/ha_innodb.def
@@ -0,0 +1,4 @@
+EXPORTS
+	_mysql_plugin_interface_version_
+	_mysql_sizeof_struct_st_plugin_
+	_mysql_plugin_declarations_
diff --git a/storage/xtradb/handler/ha_innodb.cc b/storage/xtradb/handler/ha_innodb.cc
new file mode 100644
index 00000000000..88ac35c1789
--- /dev/null
+++ b/storage/xtradb/handler/ha_innodb.cc
@@ -0,0 +1,10394 @@
+/*****************************************************************************
+
+Copyright (c) 2000, 2009, MySQL AB & Innobase Oy. All Rights Reserved.
+Copyright (c) 2008, Google Inc.
+
+Portions of this file contain modifications contributed and copyrighted by
+Google, Inc. Those modifications are gratefully acknowledged and are described
+briefly in the InnoDB documentation. The contributions by Google are
+incorporated with their permission, and subject to the conditions contained in
+the file COPYING.Google.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/* TODO list for the InnoDB handler in 5.0:
+  - Remove the flag trx->active_trans and look at trx->conc_state
+  - fix savepoint functions to use savepoint storage area
+  - Find out what kind of problems the OS X case-insensitivity causes to
+    table and database names; should we 'normalize' the names like we do
+    in Windows?
+*/
+
+#ifdef USE_PRAGMA_IMPLEMENTATION
+#pragma implementation				// gcc: Class implementation
+#endif
+
+#define MYSQL_SERVER
+
+#include <mysql_priv.h>
+#ifdef MYSQL_SERVER
+#include <log_event.h>
+#endif /* MYSQL_SERVER */
+
+#include <m_ctype.h>
+#include <mysys_err.h>
+#include <mysql/plugin.h>
+
+/* Include necessary InnoDB headers */
+extern "C" {
+#include "../storage/xtradb/include/univ.i"
+#include "../storage/xtradb/include/btr0sea.h"
+#include "../storage/xtradb/include/os0file.h"
+#include "../storage/xtradb/include/os0thread.h"
+#include "../storage/xtradb/include/srv0start.h"
+#include "../storage/xtradb/include/srv0srv.h"
+#include "../storage/xtradb/include/trx0roll.h"
+#include "../storage/xtradb/include/trx0trx.h"
+#include "../storage/xtradb/include/trx0sys.h"
+#include "../storage/xtradb/include/mtr0mtr.h"
+#include "../storage/xtradb/include/row0ins.h"
+#include "../storage/xtradb/include/row0mysql.h"
+#include "../storage/xtradb/include/row0sel.h"
+#include "../storage/xtradb/include/row0upd.h"
+#include "../storage/xtradb/include/log0log.h"
+#include "../storage/xtradb/include/lock0lock.h"
+#include "../storage/xtradb/include/dict0crea.h"
+#include "../storage/xtradb/include/btr0cur.h"
+#include "../storage/xtradb/include/btr0btr.h"
+#include "../storage/xtradb/include/fsp0fsp.h"
+#include "../storage/xtradb/include/sync0sync.h"
+#include "../storage/xtradb/include/fil0fil.h"
+#include "../storage/xtradb/include/trx0xa.h"
+#include "../storage/xtradb/include/row0merge.h"
+#include "../storage/xtradb/include/thr0loc.h"
+#include "../storage/xtradb/include/dict0boot.h"
+#include "../storage/xtradb/include/ha_prototypes.h"
+#include "../storage/xtradb/include/ut0mem.h"
+#include "../storage/xtradb/include/ibuf0ibuf.h"
+}
+
+#include "ha_innodb.h"
+#include "i_s.h"
+#include "handler0vars.h"
+
+#ifdef MYSQL_SERVER
+// Defined in trx0sys.c
+extern char		trx_sys_mysql_master_log_name[];
+extern ib_int64_t	trx_sys_mysql_master_log_pos;
+extern char		trx_sys_mysql_relay_log_name[];
+extern ib_int64_t	trx_sys_mysql_relay_log_pos;
+#endif /* MYSQL_SERVER */
+
+#ifndef MYSQL_SERVER
+/* This is needed because of Bug #3596.  Let us hope that pthread_mutex_t
+is defined the same in both builds: the MySQL server and the InnoDB plugin. */
+extern pthread_mutex_t LOCK_thread_count;
+
+#if MYSQL_VERSION_ID < 50124
+/* this is defined in mysql_priv.h inside #ifdef MYSQL_SERVER
+but we need it here */
+bool check_global_access(THD *thd, ulong want_access);
+#endif /* MYSQL_VERSION_ID < 50124 */
+#endif /* MYSQL_SERVER */
+
+/** to protect innobase_open_files */
+static pthread_mutex_t innobase_share_mutex;
+/** to force correct commit order in binlog */
+static pthread_mutex_t prepare_commit_mutex;
+static ulong commit_threads = 0;
+static pthread_mutex_t commit_threads_m;
+static pthread_cond_t commit_cond;
+static pthread_mutex_t commit_cond_m;
+static bool innodb_inited = 0;
+
+#define INSIDE_HA_INNOBASE_CC
+
+/* In the Windows plugin, the return value of current_thd is
+undefined.  Map it to NULL. */
+#if defined MYSQL_DYNAMIC_PLUGIN && defined __WIN__
+# undef current_thd
+# define current_thd NULL
+# define EQ_CURRENT_THD(thd) TRUE
+#else /* MYSQL_DYNAMIC_PLUGIN && __WIN__ */
+# define EQ_CURRENT_THD(thd) ((thd) == current_thd)
+#endif /* MYSQL_DYNAMIC_PLUGIN && __WIN__ */
+
+#ifdef MYSQL_DYNAMIC_PLUGIN
+/* These must be weak global variables in the dynamic plugin. */
+struct handlerton* innodb_hton_ptr;
+#ifdef __WIN__
+struct st_mysql_plugin*	builtin_innobase_plugin_ptr;
+#else
+int builtin_innobase_plugin;
+#endif /* __WIN__ */
+/********************************************************************
+Copy InnoDB system variables from the static InnoDB to the dynamic
+plugin. */
+static
+bool
+innodb_plugin_init(void);
+/*====================*/
+		/* out: TRUE if the dynamic InnoDB plugin should start */
+#else /* MYSQL_DYNAMIC_PLUGIN */
+/* This must be a global variable in the statically linked InnoDB. */
+struct handlerton* innodb_hton_ptr = NULL;
+#endif /* MYSQL_DYNAMIC_PLUGIN */
+
+static const long AUTOINC_OLD_STYLE_LOCKING = 0;
+static const long AUTOINC_NEW_STYLE_LOCKING = 1;
+static const long AUTOINC_NO_LOCKING = 2;
+
+static long innobase_mirrored_log_groups, innobase_log_files_in_group,
+	innobase_log_buffer_size,
+	innobase_additional_mem_pool_size, innobase_file_io_threads,
+	innobase_force_recovery, innobase_open_files,
+	innobase_autoinc_lock_mode;
+
+static unsigned long innobase_read_io_threads, innobase_write_io_threads;
+static long long innobase_buffer_pool_size, innobase_log_file_size;
+
+/* The default values for the following char* start-up parameters
+are determined in innobase_init below: */
+
+static char*	innobase_data_home_dir			= NULL;
+static char*	innobase_data_file_path			= NULL;
+static char*	innobase_log_group_home_dir		= NULL;
+static char*	innobase_file_format_name		= NULL;
+static char*	innobase_change_buffering		= NULL;
+
+/* Note: This variable can be set to on/off and any of the supported
+file formats in the configuration file, but can only be set to any
+of the supported file formats during runtime. */
+static char*	innobase_file_format_check		= NULL;
+
+/* The following has a misleading name: starting from 4.0.5, this also
+affects Windows: */
+static char*	innobase_unix_file_flush_method		= NULL;
+
+/* Below we have boolean-valued start-up parameters, and their default
+values */
+
+static ulong	innobase_fast_shutdown			= 1;
+#ifdef UNIV_LOG_ARCHIVE
+static my_bool	innobase_log_archive			= FALSE;
+static char*	innobase_log_arch_dir			= NULL;
+#endif /* UNIV_LOG_ARCHIVE */
+static my_bool	innobase_use_doublewrite		= TRUE;
+static my_bool	innobase_use_checksums			= TRUE;
+static my_bool	innobase_extra_undoslots		= FALSE;
+static my_bool	innobase_locks_unsafe_for_binlog	= FALSE;
+static my_bool	innobase_overwrite_relay_log_info	= FALSE;
+static my_bool	innobase_rollback_on_timeout		= FALSE;
+static my_bool	innobase_create_status_file		= FALSE;
+static my_bool	innobase_stats_on_metadata		= TRUE;
+
+static char*	internal_innobase_data_file_path	= NULL;
+
+static char*	innodb_version_str = (char*) INNODB_VERSION_STR;
+
+/* The following counter is used to convey information to InnoDB
+about server activity: in selects it is not sensible to call
+srv_active_wake_master_thread after each fetch or search, we only do
+it every INNOBASE_WAKE_INTERVAL'th step. */
+
+#define INNOBASE_WAKE_INTERVAL	32
+static ulong	innobase_active_counter	= 0;
+
+static hash_table_t*	innobase_open_tables;
+
+#ifdef __NETWARE__	/* some special cleanup for NetWare */
+bool nw_panic = FALSE;
+#endif
+
+/** Allowed values of innodb_change_buffering */
+static const char* innobase_change_buffering_values[IBUF_USE_COUNT] = {
+	"none",		/* IBUF_USE_NONE */
+	"inserts"	/* IBUF_USE_INSERT */
+};
+
+static INNOBASE_SHARE *get_share(const char *table_name);
+static void free_share(INNOBASE_SHARE *share);
+static int innobase_close_connection(handlerton *hton, THD* thd);
+static int innobase_commit(handlerton *hton, THD* thd, bool all);
+static int innobase_rollback(handlerton *hton, THD* thd, bool all);
+static int innobase_rollback_to_savepoint(handlerton *hton, THD* thd, 
+           void *savepoint);
+static int innobase_savepoint(handlerton *hton, THD* thd, void *savepoint);
+static int innobase_release_savepoint(handlerton *hton, THD* thd, 
+           void *savepoint);
+static handler *innobase_create_handler(handlerton *hton,
+                                        TABLE_SHARE *table,
+                                        MEM_ROOT *mem_root);
+
+/****************************************************************
+Validate the file format name and return its corresponding id. */
+static
+uint
+innobase_file_format_name_lookup(
+/*=============================*/
+						/* out: valid file format id */
+	const char*	format_name);		/* in: pointer to file format
+						name */
+/****************************************************************
+Validate the file format check config parameters, as a side effect it
+sets the srv_check_file_format_at_startup variable. */
+static
+bool
+innobase_file_format_check_on_off(
+/*==============================*/
+						/* out: true if one of 
+						"on" or "off" */
+	const char*	format_check);		/* in: parameter value */
+/****************************************************************
+Validate the file format check config parameters, as a side effect it
+sets the srv_check_file_format_at_startup variable. */
+static
+bool
+innobase_file_format_check_validate(
+/*================================*/
+						/* out: true if valid
+						config value */
+	const char*	format_check);		/* in: parameter value */
+/********************************************************************
+Return alter table flags supported in an InnoDB database. */
+static
+uint
+innobase_alter_table_flags(
+/*=======================*/
+	uint	flags);
+
+static const char innobase_hton_name[]= "InnoDB";
+
+static MYSQL_THDVAR_BOOL(support_xa, PLUGIN_VAR_OPCMDARG,
+  "Enable InnoDB support for the XA two-phase commit",
+  /* check_func */ NULL, /* update_func */ NULL,
+  /* default */ TRUE);
+
+static MYSQL_THDVAR_BOOL(table_locks, PLUGIN_VAR_OPCMDARG,
+  "Enable InnoDB locking in LOCK TABLES",
+  /* check_func */ NULL, /* update_func */ NULL,
+  /* default */ TRUE);
+
+static MYSQL_THDVAR_BOOL(strict_mode, PLUGIN_VAR_OPCMDARG,
+  "Use strict mode when evaluating create options.",
+  NULL, NULL, FALSE);
+
+static MYSQL_THDVAR_ULONG(lock_wait_timeout, PLUGIN_VAR_RQCMDARG,
+  "Timeout in seconds an InnoDB transaction may wait for a lock before being rolled back. Values above 100000000 disable the timeout.",
+  NULL, NULL, 50, 1, 1024 * 1024 * 1024, 0);
+
+
+static handler *innobase_create_handler(handlerton *hton,
+                                        TABLE_SHARE *table, 
+                                        MEM_ROOT *mem_root)
+{
+  return new (mem_root) ha_innobase(hton, table);
+}
+
+/***********************************************************************
+This function is used to prepare X/Open XA distributed transaction   */
+static
+int
+innobase_xa_prepare(
+/*================*/
+			/* out: 0 or error number */
+	handlerton* hton,
+	THD*	thd,	/* in: handle to the MySQL thread of the user
+			whose XA transaction should be prepared */
+	bool	all);	/* in: TRUE - commit transaction
+			FALSE - the current SQL statement ended */
+/***********************************************************************
+This function is used to recover X/Open XA distributed transactions   */
+static
+int
+innobase_xa_recover(
+/*================*/
+				/* out: number of prepared transactions
+				stored in xid_list */
+	handlerton* hton,
+	XID*	xid_list,	/* in/out: prepared transactions */
+	uint	len);		/* in: number of slots in xid_list */
+/***********************************************************************
+This function is used to commit one X/Open XA distributed transaction
+which is in the prepared state */
+static
+int
+innobase_commit_by_xid(
+/*===================*/
+			/* out: 0 or error number */
+	handlerton* hton,
+	XID*	xid);	/* in: X/Open XA transaction identification */
+/***********************************************************************
+This function is used to rollback one X/Open XA distributed transaction
+which is in the prepared state */
+static
+int
+innobase_rollback_by_xid(
+/*=====================*/
+			/* out: 0 or error number */
+	handlerton* hton,
+	XID	*xid);	/* in: X/Open XA transaction identification */
+/***********************************************************************
+Create a consistent view for a cursor based on current transaction
+which is created if the corresponding MySQL thread still lacks one.
+This consistent view is then used inside of MySQL when accessing records
+using a cursor. */
+static
+void*
+innobase_create_cursor_view(
+/*========================*/
+				/* out: pointer to cursor view or NULL */
+	handlerton*	hton,	/* in: innobase hton */
+	THD*		thd);	/* in: user thread handle */
+/***********************************************************************
+Set the given consistent cursor view to a transaction which is created
+if the corresponding MySQL thread still lacks one. If the given
+consistent cursor view is NULL global read view of a transaction is
+restored to a transaction read view. */
+static
+void
+innobase_set_cursor_view(
+/*=====================*/
+	handlerton* hton,
+	THD*	thd,	/* in: user thread handle */
+	void*	curview);/* in: Consistent cursor view to be set */
+/***********************************************************************
+Close the given consistent cursor view of a transaction and restore
+global read view to a transaction read view. Transaction is created if the
+corresponding MySQL thread still lacks one. */
+static
+void
+innobase_close_cursor_view(
+/*=======================*/
+	handlerton* hton,
+	THD*	thd,	/* in: user thread handle */
+	void*	curview);/* in: Consistent read view to be closed */
+/*********************************************************************
+Removes all tables in the named database inside InnoDB. */
+static
+void
+innobase_drop_database(
+/*===================*/
+			/* out: error number */
+	handlerton* hton, /* in: handlerton of Innodb */
+	char*	path);	/* in: database path; inside InnoDB the name
+			of the last directory in the path is used as
+			the database name: for example, in 'mysql/data/test'
+			the database name is 'test' */
+/***********************************************************************
+Closes an InnoDB database. */
+static
+int
+innobase_end(handlerton *hton, ha_panic_function type);
+
+/*********************************************************************
+Creates an InnoDB transaction struct for the thd if it does not yet have one.
+Starts a new InnoDB transaction if a transaction is not yet started. And
+assigns a new snapshot for a consistent read if the transaction does not yet
+have one. */
+static
+int
+innobase_start_trx_and_assign_read_view(
+/*====================================*/
+			/* out: 0 */
+	handlerton* hton, /* in: Innodb handlerton */ 
+	THD*	thd);	/* in: MySQL thread handle of the user for whom
+			the transaction should be committed */
+/********************************************************************
+Flushes InnoDB logs to disk and makes a checkpoint. Really, a commit flushes
+the logs, and the name of this function should be innobase_checkpoint. */
+static
+bool
+innobase_flush_logs(
+/*================*/
+				/* out: TRUE if error */
+	handlerton*	hton);	/* in: InnoDB handlerton */
+
+/****************************************************************************
+Implements the SHOW INNODB STATUS command. Sends the output of the InnoDB
+Monitor to the client. */
+static
+bool
+innodb_show_status(
+/*===============*/
+	handlerton*	hton,	/* in: the innodb handlerton */
+	THD*	thd,	/* in: the MySQL query thread of the caller */
+	stat_print_fn *stat_print);
+static
+bool innobase_show_status(handlerton *hton, THD* thd, 
+                          stat_print_fn* stat_print,
+                          enum ha_stat_type stat_type);
+
+/*********************************************************************
+Commits a transaction in an InnoDB database. */
+static
+void
+innobase_commit_low(
+/*================*/
+	trx_t*	trx);	/* in: transaction handle */
+
+static SHOW_VAR innodb_status_variables[]= {
+  {"buffer_pool_pages_data",
+  (char*) &export_vars.innodb_buffer_pool_pages_data,	  SHOW_LONG},
+  {"buffer_pool_pages_dirty",
+  (char*) &export_vars.innodb_buffer_pool_pages_dirty,	  SHOW_LONG},
+  {"buffer_pool_pages_flushed",
+  (char*) &export_vars.innodb_buffer_pool_pages_flushed,  SHOW_LONG},
+  {"buffer_pool_pages_free",
+  (char*) &export_vars.innodb_buffer_pool_pages_free,	  SHOW_LONG},
+#ifdef UNIV_DEBUG
+  {"buffer_pool_pages_latched",
+  (char*) &export_vars.innodb_buffer_pool_pages_latched,  SHOW_LONG},
+#endif /* UNIV_DEBUG */
+  {"buffer_pool_pages_misc",
+  (char*) &export_vars.innodb_buffer_pool_pages_misc,	  SHOW_LONG},
+  {"buffer_pool_pages_total",
+  (char*) &export_vars.innodb_buffer_pool_pages_total,	  SHOW_LONG},
+  {"buffer_pool_read_ahead_rnd",
+  (char*) &export_vars.innodb_buffer_pool_read_ahead_rnd, SHOW_LONG},
+  {"buffer_pool_read_ahead_seq",
+  (char*) &export_vars.innodb_buffer_pool_read_ahead_seq, SHOW_LONG},
+  {"buffer_pool_read_requests",
+  (char*) &export_vars.innodb_buffer_pool_read_requests,  SHOW_LONG},
+  {"buffer_pool_reads",
+  (char*) &export_vars.innodb_buffer_pool_reads,	  SHOW_LONG},
+  {"buffer_pool_wait_free",
+  (char*) &export_vars.innodb_buffer_pool_wait_free,	  SHOW_LONG},
+  {"buffer_pool_write_requests",
+  (char*) &export_vars.innodb_buffer_pool_write_requests, SHOW_LONG},
+  {"data_fsyncs",
+  (char*) &export_vars.innodb_data_fsyncs,		  SHOW_LONG},
+  {"data_pending_fsyncs",
+  (char*) &export_vars.innodb_data_pending_fsyncs,	  SHOW_LONG},
+  {"data_pending_reads",
+  (char*) &export_vars.innodb_data_pending_reads,	  SHOW_LONG},
+  {"data_pending_writes",
+  (char*) &export_vars.innodb_data_pending_writes,	  SHOW_LONG},
+  {"data_read",
+  (char*) &export_vars.innodb_data_read,		  SHOW_LONG},
+  {"data_reads",
+  (char*) &export_vars.innodb_data_reads,		  SHOW_LONG},
+  {"data_writes",
+  (char*) &export_vars.innodb_data_writes,		  SHOW_LONG},
+  {"data_written",
+  (char*) &export_vars.innodb_data_written,		  SHOW_LONG},
+  {"dblwr_pages_written",
+  (char*) &export_vars.innodb_dblwr_pages_written,	  SHOW_LONG},
+  {"dblwr_writes",
+  (char*) &export_vars.innodb_dblwr_writes,		  SHOW_LONG},
+  {"have_atomic_builtins",
+  (char*) &export_vars.innodb_have_atomic_builtins,	  SHOW_BOOL},
+  {"log_waits",
+  (char*) &export_vars.innodb_log_waits,		  SHOW_LONG},
+  {"log_write_requests",
+  (char*) &export_vars.innodb_log_write_requests,	  SHOW_LONG},
+  {"log_writes",
+  (char*) &export_vars.innodb_log_writes,		  SHOW_LONG},
+  {"os_log_fsyncs",
+  (char*) &export_vars.innodb_os_log_fsyncs,		  SHOW_LONG},
+  {"os_log_pending_fsyncs",
+  (char*) &export_vars.innodb_os_log_pending_fsyncs,	  SHOW_LONG},
+  {"os_log_pending_writes",
+  (char*) &export_vars.innodb_os_log_pending_writes,	  SHOW_LONG},
+  {"os_log_written",
+  (char*) &export_vars.innodb_os_log_written,		  SHOW_LONG},
+  {"page_size",
+  (char*) &export_vars.innodb_page_size,		  SHOW_LONG},
+  {"pages_created",
+  (char*) &export_vars.innodb_pages_created,		  SHOW_LONG},
+  {"pages_read",
+  (char*) &export_vars.innodb_pages_read,		  SHOW_LONG},
+  {"pages_written",
+  (char*) &export_vars.innodb_pages_written,		  SHOW_LONG},
+  {"row_lock_current_waits",
+  (char*) &export_vars.innodb_row_lock_current_waits,	  SHOW_LONG},
+  {"row_lock_time",
+  (char*) &export_vars.innodb_row_lock_time,		  SHOW_LONGLONG},
+  {"row_lock_time_avg",
+  (char*) &export_vars.innodb_row_lock_time_avg,	  SHOW_LONG},
+  {"row_lock_time_max",
+  (char*) &export_vars.innodb_row_lock_time_max,	  SHOW_LONG},
+  {"row_lock_waits",
+  (char*) &export_vars.innodb_row_lock_waits,		  SHOW_LONG},
+  {"rows_deleted",
+  (char*) &export_vars.innodb_rows_deleted,		  SHOW_LONG},
+  {"rows_inserted",
+  (char*) &export_vars.innodb_rows_inserted,		  SHOW_LONG},
+  {"rows_read",
+  (char*) &export_vars.innodb_rows_read,		  SHOW_LONG},
+  {"rows_updated",
+  (char*) &export_vars.innodb_rows_updated,		  SHOW_LONG},
+  {NullS, NullS, SHOW_LONG}
+};
+
+/* General functions */
+
+/**********************************************************************
+Returns true if the thread is the replication thread on the slave
+server. Used in srv_conc_enter_innodb() to determine if the thread
+should be allowed to enter InnoDB - the replication thread is treated
+differently than other threads. Also used in
+srv_conc_force_exit_innodb(). */
+extern "C" UNIV_INTERN
+ibool
+thd_is_replication_slave_thread(
+/*============================*/
+			/* out: true if thd is the replication thread */
+	void*	thd)	/* in: thread handle (THD*) */
+{
+	return((ibool) thd_slave_thread((THD*) thd));
+}
+
+/**********************************************************************
+Save some CPU by testing the value of srv_thread_concurrency in inline
+functions. */
+static inline
+void
+innodb_srv_conc_enter_innodb(
+/*=========================*/
+	trx_t*	trx)	/* in: transaction handle */
+{
+	if (UNIV_LIKELY(!srv_thread_concurrency)) {
+
+		return;
+	}
+
+	srv_conc_enter_innodb(trx);
+}
+
+/**********************************************************************
+Save some CPU by testing the value of srv_thread_concurrency in inline
+functions. */
+static inline
+void
+innodb_srv_conc_exit_innodb(
+/*========================*/
+	trx_t*	trx)	/* in: transaction handle */
+{
+	if (UNIV_LIKELY(!trx->declared_to_be_inside_innodb)) {
+
+		return;
+	}
+
+	srv_conc_exit_innodb(trx);
+}
+
+/**********************************************************************
+Releases possible search latch and InnoDB thread FIFO ticket. These should
+be released at each SQL statement end, and also when mysqld passes the
+control to the client. It does no harm to release these also in the middle
+of an SQL statement. */
+static inline
+void
+innobase_release_stat_resources(
+/*============================*/
+	trx_t*	trx)	/* in: transaction object */
+{
+	if (trx->has_search_latch) {
+		trx_search_latch_release_if_reserved(trx);
+	}
+
+	if (trx->declared_to_be_inside_innodb) {
+		/* Release our possible ticket in the FIFO */
+
+		srv_conc_force_exit_innodb(trx);
+	}
+}
+
+/**********************************************************************
+Returns true if the transaction this thread is processing has edited
+non-transactional tables. Used by the deadlock detector when deciding
+which transaction to rollback in case of a deadlock - we try to avoid
+rolling back transactions that have edited non-transactional tables. */
+extern "C" UNIV_INTERN
+ibool
+thd_has_edited_nontrans_tables(
+/*===========================*/
+			/* out: true if non-transactional tables have
+			been edited */
+	void*	thd)	/* in: thread handle (THD*) */
+{
+	return((ibool) thd_non_transactional_update((THD*) thd));
+}
+
+/**********************************************************************
+Returns true if the thread is executing a SELECT statement. */
+extern "C" UNIV_INTERN
+ibool
+thd_is_select(
+/*==========*/
+				/* out: true if thd is executing SELECT */
+	const void*	thd)	/* in: thread handle (THD*) */
+{
+	return(thd_sql_command((const THD*) thd) == SQLCOM_SELECT);
+}
+
+/**********************************************************************
+Returns true if the thread supports XA,
+global value of innodb_supports_xa if thd is NULL. */
+extern "C" UNIV_INTERN
+ibool
+thd_supports_xa(
+/*============*/
+			/* out: true if thd has XA support */
+	void*	thd)	/* in: thread handle (THD*), or NULL to query
+			the global innodb_supports_xa */
+{
+	return(THDVAR((THD*) thd, support_xa));
+}
+
+/**********************************************************************
+Returns the lock wait timeout for the current connection. */
+extern "C" UNIV_INTERN
+ulong
+thd_lock_wait_timeout(
+/*==================*/
+			/* out: the lock wait timeout, in seconds */
+	void*	thd)	/* in: thread handle (THD*), or NULL to query
+			the global innodb_lock_wait_timeout */
+{
+	/* According to <mysql/plugin.h>, passing thd == NULL
+	returns the global value of the session variable. */
+	return(THDVAR((THD*) thd, lock_wait_timeout));
+}
+
+/************************************************************************
+Obtain the InnoDB transaction of a MySQL thread. */
+static inline
+trx_t*&
+thd_to_trx(
+/*=======*/
+			/* out: reference to transaction pointer */
+	THD*	thd)	/* in: MySQL thread */
+{
+	return(*(trx_t**) thd_ha_data(thd, innodb_hton_ptr));
+}
+
+/************************************************************************
+Call this function when mysqld passes control to the client. That is to
+avoid deadlocks on the adaptive hash S-latch possibly held by thd. For more
+documentation, see handler.cc. */
+static
+int
+innobase_release_temporary_latches(
+/*===============================*/
+				/* out: 0 */
+	handlerton*	hton,	/* in: handlerton */
+	THD*		thd)	/* in: MySQL thread */
+{
+	trx_t*	trx;
+
+	DBUG_ASSERT(hton == innodb_hton_ptr);
+
+	if (!innodb_inited) {
+
+		return(0);
+	}
+
+	trx = thd_to_trx(thd);
+
+	if (trx) {
+		innobase_release_stat_resources(trx);
+	}
+	return(0);
+}
+
+/************************************************************************
+Increments innobase_active_counter and every INNOBASE_WAKE_INTERVALth
+time calls srv_active_wake_master_thread. This function should be used
+when a single database operation may introduce a small need for
+server utility activity, like checkpointing. */
+static inline
+void
+innobase_active_small(void)
+/*=======================*/
+{
+	innobase_active_counter++;
+
+	if ((innobase_active_counter % INNOBASE_WAKE_INTERVAL) == 0) {
+		srv_active_wake_master_thread();
+	}
+}
+
+/************************************************************************
+Converts an InnoDB error code to a MySQL error code and also tells to MySQL
+about a possible transaction rollback inside InnoDB caused by a lock wait
+timeout or a deadlock. */
+extern "C" UNIV_INTERN
+int
+convert_error_code_to_mysql(
+/*========================*/
+			/* out: MySQL error code */
+	int	error,	/* in: InnoDB error code */
+	ulint	flags,	/* in: InnoDB table flags, or 0 */
+	THD*	thd)	/* in: user thread handle or NULL */
+{
+	switch (error) {
+	case DB_SUCCESS:
+		return(0);
+
+	case DB_ERROR:
+	default:
+		return(-1); /* unspecified error */
+
+	case DB_DUPLICATE_KEY:
+		return(HA_ERR_FOUND_DUPP_KEY);
+
+	case DB_FOREIGN_DUPLICATE_KEY:
+		return(HA_ERR_FOREIGN_DUPLICATE_KEY);
+
+	case DB_RECORD_NOT_FOUND:
+		return(HA_ERR_NO_ACTIVE_RECORD);
+
+	case DB_DEADLOCK:
+		/* Since we rolled back the whole transaction, we must
+		tell it also to MySQL so that MySQL knows to empty the
+		cached binlog for this transaction */
+
+		if (thd) {
+			thd_mark_transaction_to_rollback(thd, TRUE);
+		}
+
+		return(HA_ERR_LOCK_DEADLOCK);
+
+	case DB_LOCK_WAIT_TIMEOUT:
+		/* Starting from 5.0.13, we let MySQL just roll back the
+		latest SQL statement in a lock wait timeout. Previously, we
+		rolled back the whole transaction. */
+
+		if (thd) {
+			thd_mark_transaction_to_rollback(
+				thd, (bool)row_rollback_on_timeout);
+		}
+
+		return(HA_ERR_LOCK_WAIT_TIMEOUT);
+
+	case DB_NO_REFERENCED_ROW:
+		return(HA_ERR_NO_REFERENCED_ROW);
+
+	case DB_ROW_IS_REFERENCED:
+		return(HA_ERR_ROW_IS_REFERENCED);
+
+	case DB_CANNOT_ADD_CONSTRAINT:
+		return(HA_ERR_CANNOT_ADD_FOREIGN);
+
+	case DB_CANNOT_DROP_CONSTRAINT:
+
+		return(HA_ERR_ROW_IS_REFERENCED); /* TODO: This is a bit
+						misleading, a new MySQL error
+						code should be introduced */
+
+	case DB_COL_APPEARS_TWICE_IN_INDEX:
+	case DB_CORRUPTION:
+		return(HA_ERR_CRASHED);
+
+	case DB_OUT_OF_FILE_SPACE:
+		return(HA_ERR_RECORD_FILE_FULL);
+
+	case DB_TABLE_IS_BEING_USED:
+		return(HA_ERR_WRONG_COMMAND);
+
+	case DB_TABLE_NOT_FOUND:
+		return(HA_ERR_NO_SUCH_TABLE);
+
+	case DB_TOO_BIG_RECORD:
+		my_error(ER_TOO_BIG_ROWSIZE, MYF(0),
+			 page_get_free_space_of_empty(flags
+						      & DICT_TF_COMPACT) / 2);
+		return(HA_ERR_TO_BIG_ROW);
+
+	case DB_NO_SAVEPOINT:
+		return(HA_ERR_NO_SAVEPOINT);
+
+	case DB_LOCK_TABLE_FULL:
+		/* Since we rolled back the whole transaction, we must
+		tell it also to MySQL so that MySQL knows to empty the
+		cached binlog for this transaction */
+
+		if (thd) {
+			thd_mark_transaction_to_rollback(thd, TRUE);
+		}
+
+		return(HA_ERR_LOCK_TABLE_FULL);
+
+	case DB_PRIMARY_KEY_IS_NULL:
+		return(ER_PRIMARY_CANT_HAVE_NULL);
+
+	case DB_TOO_MANY_CONCURRENT_TRXS:
+		/* Once MySQL add the appropriate code to errmsg.txt then
+		we can get rid of this #ifdef. NOTE: The code checked by
+		the #ifdef is the suggested name for the error condition
+		and the actual error code name could very well be different.
+		This will require some monitoring, ie. the status
+		of this request on our part.*/
+#ifdef ER_TOO_MANY_CONCURRENT_TRXS
+		return(ER_TOO_MANY_CONCURRENT_TRXS);
+#else
+		return(HA_ERR_RECORD_FILE_FULL);
+#endif
+	case DB_UNSUPPORTED:
+		return(HA_ERR_UNSUPPORTED);
+	}
+}
+
+/*****************************************************************
+If you want to print a thd that is not associated with the current thread,
+you must call this function before reserving the InnoDB kernel_mutex, to
+protect MySQL from setting thd->query NULL. If you print a thd of the current
+thread, we know that MySQL cannot modify thd->query, and it is not necessary
+to call this. Call innobase_mysql_end_print_arbitrary_thd() after you release
+the kernel_mutex. */
+extern "C" UNIV_INTERN
+void
+innobase_mysql_prepare_print_arbitrary_thd(void)
+/*============================================*/
+{
+	ut_ad(!mutex_own(&kernel_mutex));
+	VOID(pthread_mutex_lock(&LOCK_thread_count));
+}
+
+/*****************************************************************
+Releases the mutex reserved by innobase_mysql_prepare_print_arbitrary_thd().
+In the InnoDB latching order, the mutex sits right above the
+kernel_mutex.  In debug builds, we assert that the kernel_mutex is
+released before this function is invoked. */
+extern "C" UNIV_INTERN
+void
+innobase_mysql_end_print_arbitrary_thd(void)
+/*========================================*/
+{
+	ut_ad(!mutex_own(&kernel_mutex));
+	VOID(pthread_mutex_unlock(&LOCK_thread_count));
+}
+
+/*****************************************************************
+Prints info of a THD object (== user session thread) to the given file. */
+extern "C" UNIV_INTERN
+void
+innobase_mysql_print_thd(
+/*=====================*/
+	FILE*	f,		/* in: output stream */
+	void*	thd,		/* in: pointer to a MySQL THD object */
+	uint	max_query_len)	/* in: max query length to print, or 0 to
+				   use the default max length */
+{
+	char	buffer[1024];
+
+	fputs(thd_security_context((THD*) thd, buffer, sizeof buffer,
+				   max_query_len), f);
+	putc('\n', f);
+}
+
+/**********************************************************************
+Get the variable length bounds of the given character set. */
+extern "C" UNIV_INTERN
+void
+innobase_get_cset_width(
+/*====================*/
+	ulint	cset,		/* in: MySQL charset-collation code */
+	ulint*	mbminlen,	/* out: minimum length of a char (in bytes) */
+	ulint*	mbmaxlen)	/* out: maximum length of a char (in bytes) */
+{
+	CHARSET_INFO*	cs;
+	ut_ad(cset < 256);
+	ut_ad(mbminlen);
+	ut_ad(mbmaxlen);
+
+	cs = all_charsets[cset];
+	if (cs) {
+		*mbminlen = cs->mbminlen;
+		*mbmaxlen = cs->mbmaxlen;
+	} else {
+		ut_a(cset == 0);
+		*mbminlen = *mbmaxlen = 0;
+	}
+}
+
+/**********************************************************************
+Converts an identifier to a table name. */
+extern "C" UNIV_INTERN
+void
+innobase_convert_from_table_id(
+/*===========================*/
+	struct charset_info_st*	cs,	/* in: the 'from' character set */
+	char*			to,	/* out: converted identifier */
+	const char*		from,	/* in: identifier to convert */
+	ulint			len)	/* in: length of 'to', in bytes */
+{
+	uint	errors;
+
+	strconvert(cs, from, &my_charset_filename, to, (uint) len, &errors);
+}
+
+/**********************************************************************
+Converts an identifier to UTF-8. */
+extern "C" UNIV_INTERN
+void
+innobase_convert_from_id(
+/*=====================*/
+	struct charset_info_st*	cs,	/* in: the 'from' character set */
+	char*			to,	/* out: converted identifier */
+	const char*		from,	/* in: identifier to convert */
+	ulint			len)	/* in: length of 'to', in bytes */
+{
+	uint	errors;
+
+	strconvert(cs, from, system_charset_info, to, (uint) len, &errors);
+}
+
+/**********************************************************************
+Compares NUL-terminated UTF-8 strings case insensitively. */
+extern "C" UNIV_INTERN
+int
+innobase_strcasecmp(
+/*================*/
+				/* out: 0 if a=b, <0 if a<b, >1 if a>b */
+	const char*	a,	/* in: first string to compare */
+	const char*	b)	/* in: second string to compare */
+{
+	return(my_strcasecmp(system_charset_info, a, b));
+}
+
+/**********************************************************************
+Makes all characters in a NUL-terminated UTF-8 string lower case. */
+extern "C" UNIV_INTERN
+void
+innobase_casedn_str(
+/*================*/
+	char*	a)	/* in/out: string to put in lower case */
+{
+	my_casedn_str(system_charset_info, a);
+}
+
+/**************************************************************************
+Determines the connection character set. */
+extern "C" UNIV_INTERN
+struct charset_info_st*
+innobase_get_charset(
+/*=================*/
+				/* out: connection character set */
+	void*	mysql_thd)	/* in: MySQL thread handle */
+{
+	return(thd_charset((THD*) mysql_thd));
+}
+
+#if defined (__WIN__) && defined (MYSQL_DYNAMIC_PLUGIN)
+/***********************************************************************
+Map an OS error to an errno value. The OS error number is stored in
+_doserrno and the mapped value is stored in errno) */
+extern "C"
+void __cdecl
+_dosmaperr(
+	unsigned long);	/* in: OS error value */
+
+/*************************************************************************
+Creates a temporary file. */
+extern "C" UNIV_INTERN
+int
+innobase_mysql_tmpfile(void)
+/*========================*/
+			/* out: temporary file descriptor, or < 0 on error */
+{
+	int	fd;				/* handle of opened file */
+	HANDLE	osfh;				/* OS handle of opened file */
+	char*	tmpdir;				/* point to the directory
+						where to create file */
+	TCHAR	path_buf[MAX_PATH - 14];	/* buffer for tmp file path.
+						The length cannot be longer
+						than MAX_PATH - 14, or
+						GetTempFileName will fail. */
+	char	filename[MAX_PATH];		/* name of the tmpfile */
+	DWORD	fileaccess = GENERIC_READ	/* OS file access */
+			     | GENERIC_WRITE
+			     | DELETE;
+	DWORD	fileshare = FILE_SHARE_READ	/* OS file sharing mode */
+			    | FILE_SHARE_WRITE
+			    | FILE_SHARE_DELETE;
+	DWORD	filecreate = CREATE_ALWAYS;	/* OS method of open/create */
+	DWORD	fileattrib =			/* OS file attribute flags */
+			     FILE_ATTRIBUTE_NORMAL
+			     | FILE_FLAG_DELETE_ON_CLOSE
+			     | FILE_ATTRIBUTE_TEMPORARY
+			     | FILE_FLAG_SEQUENTIAL_SCAN;
+
+	DBUG_ENTER("innobase_mysql_tmpfile");
+
+	tmpdir = my_tmpdir(&mysql_tmpdir_list);
+
+	/* The tmpdir parameter can not be NULL for GetTempFileName. */
+	if (!tmpdir) {
+		uint	ret;
+
+		/* Use GetTempPath to determine path for temporary files. */
+		ret = GetTempPath(sizeof(path_buf), path_buf);
+		if (ret > sizeof(path_buf) || (ret == 0)) {
+
+			_dosmaperr(GetLastError());	/* map error */
+			DBUG_RETURN(-1);
+		}
+
+		tmpdir = path_buf;
+	}
+
+	/* Use GetTempFileName to generate a unique filename. */
+	if (!GetTempFileName(tmpdir, "ib", 0, filename)) {
+
+		_dosmaperr(GetLastError());	/* map error */
+		DBUG_RETURN(-1);
+	}
+
+	DBUG_PRINT("info", ("filename: %s", filename));
+
+	/* Open/Create the file. */
+	osfh = CreateFile(filename, fileaccess, fileshare, NULL,
+			  filecreate, fileattrib, NULL);
+	if (osfh == INVALID_HANDLE_VALUE) {
+
+		/* open/create file failed! */
+		_dosmaperr(GetLastError());	/* map error */
+		DBUG_RETURN(-1);
+	}
+
+	do {
+		/* Associates a CRT file descriptor with the OS file handle. */
+		fd = _open_osfhandle((intptr_t) osfh, 0);
+	} while (fd == -1 && errno == EINTR);
+
+	if (fd == -1) {
+		/* Open failed, close the file handle. */
+
+		_dosmaperr(GetLastError());	/* map error */
+		CloseHandle(osfh);		/* no need to check if
+						CloseHandle fails */
+	}
+
+	DBUG_RETURN(fd);
+}
+#else
+/*************************************************************************
+Creates a temporary file. */
+extern "C" UNIV_INTERN
+int
+innobase_mysql_tmpfile(void)
+/*========================*/
+			/* out: temporary file descriptor, or < 0 on error */
+{
+	int	fd2 = -1;
+	File	fd = mysql_tmpfile("ib");
+	if (fd >= 0) {
+		/* Copy the file descriptor, so that the additional resources
+		allocated by create_temp_file() can be freed by invoking
+		my_close().
+
+		Because the file descriptor returned by this function
+		will be passed to fdopen(), it will be closed by invoking
+		fclose(), which in turn will invoke close() instead of
+		my_close(). */
+		fd2 = dup(fd);
+		if (fd2 < 0) {
+			DBUG_PRINT("error",("Got error %d on dup",fd2));
+			my_errno=errno;
+			my_error(EE_OUT_OF_FILERESOURCES,
+				 MYF(ME_BELL+ME_WAITTANG),
+				 "ib*", my_errno);
+		}
+		my_close(fd, MYF(MY_WME));
+	}
+	return(fd2);
+}
+#endif /* defined (__WIN__) && defined (MYSQL_DYNAMIC_PLUGIN) */
+
+/*************************************************************************
+Wrapper around MySQL's copy_and_convert function, see it for
+documentation. */
+extern "C" UNIV_INTERN
+ulint
+innobase_convert_string(
+/*====================*/
+	void*		to,
+	ulint		to_length,
+	CHARSET_INFO*	to_cs,
+	const void*	from,
+	ulint		from_length,
+	CHARSET_INFO*	from_cs,
+	uint*		errors)
+{
+  return(copy_and_convert((char*)to, (uint32) to_length, to_cs,
+                          (const char*)from, (uint32) from_length, from_cs,
+                          errors));
+}
+
+/***********************************************************************
+Formats the raw data in "data" (in InnoDB on-disk format) that is of
+type DATA_(CHAR|VARCHAR|MYSQL|VARMYSQL) using "charset_coll" and writes
+the result to "buf". The result is converted to "system_charset_info".
+Not more than "buf_size" bytes are written to "buf".
+The result is always '\0'-terminated (provided buf_size > 0) and the
+number of bytes that were written to "buf" is returned (including the
+terminating '\0'). */
+extern "C" UNIV_INTERN
+ulint
+innobase_raw_format(
+/*================*/
+					/* out: number of bytes
+					that were written */
+	const char*	data,		/* in: raw data */
+	ulint		data_len,	/* in: raw data length
+					in bytes */
+	ulint		charset_coll,	/* in: charset collation */
+	char*		buf,		/* out: output buffer */
+	ulint		buf_size)	/* in: output buffer size
+					in bytes */
+{
+	/* XXX we use a hard limit instead of allocating
+	but_size bytes from the heap */
+	CHARSET_INFO*	data_cs;
+	char		buf_tmp[8192];
+	ulint		buf_tmp_used;
+	uint		num_errors;
+
+	data_cs = all_charsets[charset_coll];
+
+	buf_tmp_used = innobase_convert_string(buf_tmp, sizeof(buf_tmp),
+					       system_charset_info,
+					       data, data_len, data_cs,
+					       &num_errors);
+
+	return(ut_str_sql_format(buf_tmp, buf_tmp_used, buf, buf_size));
+}
+
+/*************************************************************************
+Compute the next autoinc value.
+
+For MySQL replication the autoincrement values can be partitioned among
+the nodes. The offset is the start or origin of the autoincrement value
+for a particular node. For n nodes the increment will be n and the offset
+will be in the interval [1, n]. The formula tries to allocate the next
+value for a particular node.
+
+Note: This function is also called with increment set to the number of
+values we want to reserve for multi-value inserts e.g.,
+
+	INSERT INTO T VALUES(), (), ();
+
+innobase_next_autoinc() will be called with increment set to
+n * 3 where autoinc_lock_mode != TRADITIONAL because we want
+to reserve 3 values for the multi-value INSERT above. */
+static
+ulonglong
+innobase_next_autoinc(
+/*==================*/
+					/* out: the next value */
+	ulonglong	current,	/* in: Current value */
+	ulonglong	increment,	/* in: increment current by */
+	ulonglong	offset,		/* in: AUTOINC offset */
+	ulonglong	max_value)	/* in: max value for type */
+{
+	ulonglong	next_value;
+
+	/* Should never be 0. */
+	ut_a(increment > 0);
+
+	/* According to MySQL documentation, if the offset is greater than
+	the increment then the offset is ignored. */
+	if (offset > increment) {
+		offset = 0;
+	}
+
+	if (max_value <= current) {
+		next_value = max_value;
+	} else if (offset <= 1) {
+		/* Offset 0 and 1 are the same, because there must be at
+		least one node in the system. */
+		if (max_value - current <= increment) {
+			next_value = max_value;
+		} else {
+			next_value = current + increment;
+		}
+	} else if (max_value > current) {
+		if (current > offset) {
+			next_value = ((current - offset) / increment) + 1;
+		} else {
+			next_value = ((offset - current) / increment) + 1;
+		}
+
+		ut_a(increment > 0);
+		ut_a(next_value > 0);
+
+		/* Check for multiplication overflow. */
+		if (increment > (max_value / next_value)) {
+
+			next_value = max_value;
+		} else {
+			next_value *= increment;
+
+			ut_a(max_value >= next_value);
+
+			/* Check for overflow. */
+			if (max_value - next_value <= offset) {
+				next_value = max_value;
+			} else {
+				next_value += offset;
+			}
+		}
+	} else {
+		next_value = max_value;
+	}
+
+	ut_a(next_value <= max_value);
+
+	return(next_value);
+}
+
+/*************************************************************************
+Initializes some fields in an InnoDB transaction object. */
+static
+void
+innobase_trx_init(
+/*==============*/
+	THD*	thd,	/* in: user thread handle */
+	trx_t*	trx)	/* in/out: InnoDB transaction handle */
+{
+	DBUG_ENTER("innobase_trx_init");
+	DBUG_ASSERT(EQ_CURRENT_THD(thd));
+	DBUG_ASSERT(thd == trx->mysql_thd);
+
+	trx->check_foreigns = !thd_test_options(
+		thd, OPTION_NO_FOREIGN_KEY_CHECKS);
+
+	trx->check_unique_secondary = !thd_test_options(
+		thd, OPTION_RELAXED_UNIQUE_CHECKS);
+
+	DBUG_VOID_RETURN;
+}
+
+/*************************************************************************
+Allocates an InnoDB transaction for a MySQL handler object. */
+extern "C" UNIV_INTERN
+trx_t*
+innobase_trx_allocate(
+/*==================*/
+			/* out: InnoDB transaction handle */
+	THD*	thd)	/* in: user thread handle */
+{
+	trx_t*	trx;
+
+	DBUG_ENTER("innobase_trx_allocate");
+	DBUG_ASSERT(thd != NULL);
+	DBUG_ASSERT(EQ_CURRENT_THD(thd));
+
+	trx = trx_allocate_for_mysql();
+
+	trx->mysql_thd = thd;
+	trx->mysql_query_str = thd_query(thd);
+
+	innobase_trx_init(thd, trx);
+
+	DBUG_RETURN(trx);
+}
+
+/*************************************************************************
+Gets the InnoDB transaction handle for a MySQL handler object, creates
+an InnoDB transaction struct if the corresponding MySQL thread struct still
+lacks one. */
+static
+trx_t*
+check_trx_exists(
+/*=============*/
+			/* out: InnoDB transaction handle */
+	THD*	thd)	/* in: user thread handle */
+{
+	trx_t*&	trx = thd_to_trx(thd);
+
+	ut_ad(EQ_CURRENT_THD(thd));
+
+	if (trx == NULL) {
+		trx = innobase_trx_allocate(thd);
+	} else if (UNIV_UNLIKELY(trx->magic_n != TRX_MAGIC_N)) {
+		mem_analyze_corruption(trx);
+		ut_error;
+	}
+
+	innobase_trx_init(thd, trx);
+
+	return(trx);
+}
+
+
+/*************************************************************************
+Construct ha_innobase handler. */
+UNIV_INTERN
+ha_innobase::ha_innobase(handlerton *hton, TABLE_SHARE *table_arg)
+  :handler(hton, table_arg),
+  int_table_flags(HA_REC_NOT_IN_SEQ |
+		  HA_NULL_IN_KEY |
+		  HA_CAN_INDEX_BLOBS |
+		  HA_CAN_SQL_HANDLER |
+		  HA_PRIMARY_KEY_REQUIRED_FOR_POSITION |
+		  HA_PRIMARY_KEY_IN_READ_INDEX |
+		  HA_BINLOG_ROW_CAPABLE |
+		  HA_CAN_GEOMETRY | HA_PARTIAL_COLUMN_READ |
+		  HA_TABLE_SCAN_ON_INDEX),
+  start_of_scan(0),
+  num_write_row(0)
+{}
+
+/*************************************************************************
+Destruct ha_innobase handler. */
+UNIV_INTERN
+ha_innobase::~ha_innobase()
+{
+}
+
+/*************************************************************************
+Updates the user_thd field in a handle and also allocates a new InnoDB
+transaction handle if needed, and updates the transaction fields in the
+prebuilt struct. */
+UNIV_INTERN inline
+void
+ha_innobase::update_thd(
+/*====================*/
+	THD*	thd)	/* in: thd to use the handle */
+{
+	trx_t*		trx;
+
+	trx = check_trx_exists(thd);
+
+	if (prebuilt->trx != trx) {
+
+		row_update_prebuilt_trx(prebuilt, trx);
+	}
+
+	user_thd = thd;
+}
+
+/*************************************************************************
+Updates the user_thd field in a handle and also allocates a new InnoDB
+transaction handle if needed, and updates the transaction fields in the
+prebuilt struct. */
+UNIV_INTERN
+void
+ha_innobase::update_thd()
+/*=====================*/
+{
+	THD*	thd = ha_thd();
+	ut_ad(EQ_CURRENT_THD(thd));
+	update_thd(thd);
+}
+
+/*************************************************************************
+Registers that InnoDB takes part in an SQL statement, so that MySQL knows to
+roll back the statement if the statement results in an error. This MUST be
+called for every SQL statement that may be rolled back by MySQL. Calling this
+several times to register the same statement is allowed, too. */
+static inline
+void
+innobase_register_stmt(
+/*===================*/
+        handlerton*	hton,	/* in: Innobase hton */
+	THD*	thd)	/* in: MySQL thd (connection) object */
+{
+	DBUG_ASSERT(hton == innodb_hton_ptr);
+	/* Register the statement */
+	trans_register_ha(thd, FALSE, hton);
+}
+
+/*************************************************************************
+Registers an InnoDB transaction in MySQL, so that the MySQL XA code knows
+to call the InnoDB prepare and commit, or rollback for the transaction. This
+MUST be called for every transaction for which the user may call commit or
+rollback. Calling this several times to register the same transaction is
+allowed, too.
+This function also registers the current SQL statement. */
+static inline
+void
+innobase_register_trx_and_stmt(
+/*===========================*/
+        handlerton *hton, /* in: Innobase handlerton */
+	THD*	thd)	/* in: MySQL thd (connection) object */
+{
+	/* NOTE that actually innobase_register_stmt() registers also
+	the transaction in the AUTOCOMMIT=1 mode. */
+
+	innobase_register_stmt(hton, thd);
+
+	if (thd_test_options(thd, OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN)) {
+
+		/* No autocommit mode, register for a transaction */
+		trans_register_ha(thd, TRUE, hton);
+	}
+}
+
+/*   BACKGROUND INFO: HOW THE MYSQL QUERY CACHE WORKS WITH INNODB
+     ------------------------------------------------------------
+
+1) The use of the query cache for TBL is disabled when there is an
+uncommitted change to TBL.
+
+2) When a change to TBL commits, InnoDB stores the current value of
+its global trx id counter, let us denote it by INV_TRX_ID, to the table object
+in the InnoDB data dictionary, and does only allow such transactions whose
+id <= INV_TRX_ID to use the query cache.
+
+3) When InnoDB does an INSERT/DELETE/UPDATE to a table TBL, or an implicit
+modification because an ON DELETE CASCADE, we invalidate the MySQL query cache
+of TBL immediately.
+
+How this is implemented inside InnoDB:
+
+1) Since every modification always sets an IX type table lock on the InnoDB
+table, it is easy to check if there can be uncommitted modifications for a
+table: just check if there are locks in the lock list of the table.
+
+2) When a transaction inside InnoDB commits, it reads the global trx id
+counter and stores the value INV_TRX_ID to the tables on which it had a lock.
+
+3) If there is an implicit table change from ON DELETE CASCADE or SET NULL,
+InnoDB calls an invalidate method for the MySQL query cache for that table.
+
+How this is implemented inside sql_cache.cc:
+
+1) The query cache for an InnoDB table TBL is invalidated immediately at an
+INSERT/UPDATE/DELETE, just like in the case of MyISAM. No need to delay
+invalidation to the transaction commit.
+
+2) To store or retrieve a value from the query cache of an InnoDB table TBL,
+any query must first ask InnoDB's permission. We must pass the thd as a
+parameter because InnoDB will look at the trx id, if any, associated with
+that thd.
+
+3) Use of the query cache for InnoDB tables is now allowed also when
+AUTOCOMMIT==0 or we are inside BEGIN ... COMMIT. Thus transactions no longer
+put restrictions on the use of the query cache.
+*/
+
+/**********************************************************************
+The MySQL query cache uses this to check from InnoDB if the query cache at
+the moment is allowed to operate on an InnoDB table. The SQL query must
+be a non-locking SELECT.
+
+The query cache is allowed to operate on certain query only if this function
+returns TRUE for all tables in the query.
+
+If thd is not in the autocommit state, this function also starts a new
+transaction for thd if there is no active trx yet, and assigns a consistent
+read view to it if there is no read view yet.
+
+Why a deadlock of threads is not possible: the query cache calls this function
+at the start of a SELECT processing. Then the calling thread cannot be
+holding any InnoDB semaphores. The calling thread is holding the
+query cache mutex, and this function will reserver the InnoDB kernel mutex.
+Thus, the 'rank' in sync0sync.h of the MySQL query cache mutex is above
+the InnoDB kernel mutex. */
+static
+my_bool
+innobase_query_caching_of_table_permitted(
+/*======================================*/
+				/* out: TRUE if permitted, FALSE if not;
+				note that the value FALSE does not mean
+				we should invalidate the query cache:
+				invalidation is called explicitly */
+	THD*	thd,		/* in: thd of the user who is trying to
+				store a result to the query cache or
+				retrieve it */
+	char*	full_name,	/* in: concatenation of database name,
+				the null character '\0', and the table
+				name */
+	uint	full_name_len,	/* in: length of the full name, i.e.
+				len(dbname) + len(tablename) + 1 */
+	ulonglong *unused)	/* unused for this engine */
+{
+	ibool	is_autocommit;
+	trx_t*	trx;
+	char	norm_name[1000];
+
+	ut_a(full_name_len < 999);
+
+	trx = check_trx_exists(thd);
+
+	if (trx->isolation_level == TRX_ISO_SERIALIZABLE) {
+		/* In the SERIALIZABLE mode we add LOCK IN SHARE MODE to every
+		plain SELECT if AUTOCOMMIT is not on. */
+
+		return((my_bool)FALSE);
+	}
+
+	if (trx->has_search_latch) {
+		sql_print_error("The calling thread is holding the adaptive "
+				"search, latch though calling "
+				"innobase_query_caching_of_table_permitted.");
+
+		mutex_enter(&kernel_mutex);
+		trx_print(stderr, trx, 1024);
+		mutex_exit(&kernel_mutex);
+	}
+
+	innobase_release_stat_resources(trx);
+
+	if (!thd_test_options(thd, OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN)) {
+
+		is_autocommit = TRUE;
+	} else {
+		is_autocommit = FALSE;
+
+	}
+
+	if (is_autocommit && trx->n_mysql_tables_in_use == 0) {
+		/* We are going to retrieve the query result from the query
+		cache. This cannot be a store operation to the query cache
+		because then MySQL would have locks on tables already.
+
+		TODO: if the user has used LOCK TABLES to lock the table,
+		then we open a transaction in the call of row_.. below.
+		That trx can stay open until UNLOCK TABLES. The same problem
+		exists even if we do not use the query cache. MySQL should be
+		modified so that it ALWAYS calls some cleanup function when
+		the processing of a query ends!
+
+		We can imagine we instantaneously serialize this consistent
+		read trx to the current trx id counter. If trx2 would have
+		changed the tables of a query result stored in the cache, and
+		trx2 would have already committed, making the result obsolete,
+		then trx2 would have already invalidated the cache. Thus we
+		can trust the result in the cache is ok for this query. */
+
+		return((my_bool)TRUE);
+	}
+
+	/* Normalize the table name to InnoDB format */
+
+	memcpy(norm_name, full_name, full_name_len);
+
+	norm_name[strlen(norm_name)] = '/'; /* InnoDB uses '/' as the
+					    separator between db and table */
+	norm_name[full_name_len] = '\0';
+#ifdef __WIN__
+	innobase_casedn_str(norm_name);
+#endif
+	/* The call of row_search_.. will start a new transaction if it is
+	not yet started */
+
+	if (trx->active_trans == 0) {
+
+		innobase_register_trx_and_stmt(innodb_hton_ptr, thd);
+		trx->active_trans = 1;
+	}
+
+	if (row_search_check_if_query_cache_permitted(trx, norm_name)) {
+
+		/* printf("Query cache for %s permitted\n", norm_name); */
+
+		return((my_bool)TRUE);
+	}
+
+	/* printf("Query cache for %s NOT permitted\n", norm_name); */
+
+	return((my_bool)FALSE);
+}
+
+/*********************************************************************
+Invalidates the MySQL query cache for the table.
+NOTE that the exact prototype of this function has to be in
+/xtradb/row/row0ins.c! */
+extern "C" UNIV_INTERN
+void
+innobase_invalidate_query_cache(
+/*============================*/
+	trx_t*	trx,		/* in: transaction which modifies the table */
+	char*	full_name,	/* in: concatenation of database name, null
+				char '\0', table name, null char'\0';
+				NOTE that in Windows this is always
+				in LOWER CASE! */
+	ulint	full_name_len)	/* in: full name length where also the null
+				chars count */
+{
+	/* Note that the sync0sync.h rank of the query cache mutex is just
+	above the InnoDB kernel mutex. The caller of this function must not
+	have latches of a lower rank. */
+
+	/* Argument TRUE below means we are using transactions */
+#ifdef HAVE_QUERY_CACHE
+	mysql_query_cache_invalidate4((THD*) trx->mysql_thd,
+				      (const char*) full_name,
+				      (uint32) full_name_len,
+				      TRUE);
+#endif
+}
+
+/*********************************************************************
+Convert an SQL identifier to the MySQL system_charset_info (UTF-8)
+and quote it if needed. */
+static
+char*
+innobase_convert_identifier(
+/*========================*/
+				/* out: pointer to the end of buf */
+	char*		buf,	/* out: buffer for converted identifier */
+	ulint		buflen,	/* in: length of buf, in bytes */
+	const char*	id,	/* in: identifier to convert */
+	ulint		idlen,	/* in: length of id, in bytes */
+	void*		thd,	/* in: MySQL connection thread, or NULL */
+	ibool		file_id)/* in: TRUE=id is a table or database name;
+				FALSE=id is an UTF-8 string */
+{
+	char nz[NAME_LEN + 1];
+	char nz2[NAME_LEN + 1 + sizeof srv_mysql50_table_name_prefix];
+
+	const char*	s	= id;
+	int		q;
+
+	if (file_id) {
+		/* Decode the table name.  The filename_to_tablename()
+		function expects a NUL-terminated string.  The input and
+		output strings buffers must not be shared. */
+
+		if (UNIV_UNLIKELY(idlen > (sizeof nz) - 1)) {
+			idlen = (sizeof nz) - 1;
+		}
+
+		memcpy(nz, id, idlen);
+		nz[idlen] = 0;
+
+		s = nz2;
+		idlen = filename_to_tablename(nz, nz2, sizeof nz2);
+	}
+
+	/* See if the identifier needs to be quoted. */
+	if (UNIV_UNLIKELY(!thd)) {
+		q = '"';
+	} else {
+		q = get_quote_char_for_identifier((THD*) thd, s, (int) idlen);
+	}
+
+	if (q == EOF) {
+		if (UNIV_UNLIKELY(idlen > buflen)) {
+			idlen = buflen;
+		}
+		memcpy(buf, s, idlen);
+		return(buf + idlen);
+	}
+
+	/* Quote the identifier. */
+	if (buflen < 2) {
+		return(buf);
+	}
+
+	*buf++ = q;
+	buflen--;
+
+	for (; idlen; idlen--) {
+		int	c = *s++;
+		if (UNIV_UNLIKELY(c == q)) {
+			if (UNIV_UNLIKELY(buflen < 3)) {
+				break;
+			}
+
+			*buf++ = c;
+			*buf++ = c;
+			buflen -= 2;
+		} else {
+			if (UNIV_UNLIKELY(buflen < 2)) {
+				break;
+			}
+
+			*buf++ = c;
+			buflen--;
+		}
+	}
+
+	*buf++ = q;
+	return(buf);
+}
+
+/*********************************************************************
+Convert a table or index name to the MySQL system_charset_info (UTF-8)
+and quote it if needed. */
+extern "C" UNIV_INTERN
+char*
+innobase_convert_name(
+/*==================*/
+				/* out: pointer to the end of buf */
+	char*		buf,	/* out: buffer for converted identifier */
+	ulint		buflen,	/* in: length of buf, in bytes */
+	const char*	id,	/* in: identifier to convert */
+	ulint		idlen,	/* in: length of id, in bytes */
+	void*		thd,	/* in: MySQL connection thread, or NULL */
+	ibool		table_id)/* in: TRUE=id is a table or database name;
+				FALSE=id is an index name */
+{
+	char*		s	= buf;
+	const char*	bufend	= buf + buflen;
+
+	if (table_id) {
+		const char*	slash = (const char*) memchr(id, '/', idlen);
+		if (!slash) {
+
+			goto no_db_name;
+		}
+
+		/* Print the database name and table name separately. */
+		s = innobase_convert_identifier(s, bufend - s, id, slash - id,
+						thd, TRUE);
+		if (UNIV_LIKELY(s < bufend)) {
+			*s++ = '.';
+			s = innobase_convert_identifier(s, bufend - s,
+							slash + 1, idlen
+							- (slash - id) - 1,
+							thd, TRUE);
+		}
+	} else if (UNIV_UNLIKELY(*id == TEMP_INDEX_PREFIX)) {
+		/* Temporary index name (smart ALTER TABLE) */
+		const char temp_index_suffix[]= "--temporary--";
+
+		s = innobase_convert_identifier(buf, buflen, id + 1, idlen - 1,
+						thd, FALSE);
+		if (s - buf + (sizeof temp_index_suffix - 1) < buflen) {
+			memcpy(s, temp_index_suffix,
+			       sizeof temp_index_suffix - 1);
+			s += sizeof temp_index_suffix - 1;
+		}
+	} else {
+no_db_name:
+		s = innobase_convert_identifier(buf, buflen, id, idlen,
+						thd, table_id);
+	}
+
+	return(s);
+
+}
+
+/**************************************************************************
+Determines if the currently running transaction has been interrupted. */
+extern "C" UNIV_INTERN
+ibool
+trx_is_interrupted(
+/*===============*/
+			/* out: TRUE if interrupted */
+	trx_t*	trx)	/* in: transaction */
+{
+	return(trx && trx->mysql_thd && thd_killed((THD*) trx->mysql_thd));
+}
+
+/******************************************************************
+Resets some fields of a prebuilt struct. The template is used in fast
+retrieval of just those column values MySQL needs in its processing. */
+static
+void
+reset_template(
+/*===========*/
+	row_prebuilt_t*	prebuilt)	/* in/out: prebuilt struct */
+{
+	prebuilt->keep_other_fields_on_keyread = 0;
+	prebuilt->read_just_key = 0;
+}
+
+/*********************************************************************
+Call this when you have opened a new table handle in HANDLER, before you
+call index_read_idx() etc. Actually, we can let the cursor stay open even
+over a transaction commit! Then you should call this before every operation,
+fetch next etc. This function inits the necessary things even after a
+transaction commit. */
+UNIV_INTERN
+void
+ha_innobase::init_table_handle_for_HANDLER(void)
+/*============================================*/
+{
+	/* If current thd does not yet have a trx struct, create one.
+	If the current handle does not yet have a prebuilt struct, create
+	one. Update the trx pointers in the prebuilt struct. Normally
+	this operation is done in external_lock. */
+
+	update_thd(ha_thd());
+
+	/* Initialize the prebuilt struct much like it would be inited in
+	external_lock */
+
+	innobase_release_stat_resources(prebuilt->trx);
+
+	/* If the transaction is not started yet, start it */
+
+	trx_start_if_not_started(prebuilt->trx);
+
+	/* Assign a read view if the transaction does not have it yet */
+
+	trx_assign_read_view(prebuilt->trx);
+
+	/* Set the MySQL flag to mark that there is an active transaction */
+
+	if (prebuilt->trx->active_trans == 0) {
+
+		innobase_register_trx_and_stmt(ht, user_thd);
+
+		prebuilt->trx->active_trans = 1;
+	}
+
+	/* We did the necessary inits in this function, no need to repeat them
+	in row_search_for_mysql */
+
+	prebuilt->sql_stat_start = FALSE;
+
+	/* We let HANDLER always to do the reads as consistent reads, even
+	if the trx isolation level would have been specified as SERIALIZABLE */
+
+	prebuilt->select_lock_type = LOCK_NONE;
+	prebuilt->stored_select_lock_type = LOCK_NONE;
+
+	/* Always fetch all columns in the index record */
+
+	prebuilt->hint_need_to_fetch_extra_cols = ROW_RETRIEVE_ALL_COLS;
+
+	/* We want always to fetch all columns in the whole row? Or do
+	we???? */
+
+	prebuilt->used_in_HANDLER = TRUE;
+	reset_template(prebuilt);
+}
+
+/*************************************************************************
+Opens an InnoDB database. */
+static
+int
+innobase_init(
+/*==========*/
+			/* out: 0 on success, error code on failure */
+	void	*p)	/* in: InnoDB handlerton */
+{
+	static char	current_dir[3];		/* Set if using current lib */
+	int		err;
+	bool		ret;
+	char		*default_path;
+	uint		format_id;
+
+	DBUG_ENTER("innobase_init");
+        handlerton *innobase_hton= (handlerton *)p;
+
+#ifdef MYSQL_DYNAMIC_PLUGIN
+	if (!innodb_plugin_init()) {
+		sql_print_error("InnoDB plugin init failed.");
+		DBUG_RETURN(-1);
+	}
+
+	if (innodb_hton_ptr) {
+		/* Patch the statically linked handlerton and variables */
+		innobase_hton = innodb_hton_ptr;
+	}
+#endif /* MYSQL_DYNAMIC_PLUGIN */
+
+        innodb_hton_ptr = innobase_hton;
+
+        innobase_hton->state = SHOW_OPTION_YES;
+        innobase_hton->db_type= DB_TYPE_INNODB;
+        innobase_hton->savepoint_offset=sizeof(trx_named_savept_t);
+        innobase_hton->close_connection=innobase_close_connection;
+        innobase_hton->savepoint_set=innobase_savepoint;
+        innobase_hton->savepoint_rollback=innobase_rollback_to_savepoint;
+        innobase_hton->savepoint_release=innobase_release_savepoint;
+        innobase_hton->commit=innobase_commit;
+        innobase_hton->rollback=innobase_rollback;
+        innobase_hton->prepare=innobase_xa_prepare;
+        innobase_hton->recover=innobase_xa_recover;
+        innobase_hton->commit_by_xid=innobase_commit_by_xid;
+        innobase_hton->rollback_by_xid=innobase_rollback_by_xid;
+        innobase_hton->create_cursor_read_view=innobase_create_cursor_view;
+        innobase_hton->set_cursor_read_view=innobase_set_cursor_view;
+        innobase_hton->close_cursor_read_view=innobase_close_cursor_view;
+        innobase_hton->create=innobase_create_handler;
+        innobase_hton->drop_database=innobase_drop_database;
+        innobase_hton->panic=innobase_end;
+        innobase_hton->start_consistent_snapshot=innobase_start_trx_and_assign_read_view;
+        innobase_hton->flush_logs=innobase_flush_logs;
+        innobase_hton->show_status=innobase_show_status;
+        innobase_hton->flags=HTON_NO_FLAGS;
+        innobase_hton->release_temporary_latches=innobase_release_temporary_latches;
+	innobase_hton->alter_table_flags = innobase_alter_table_flags;
+
+	ut_a(DATA_MYSQL_TRUE_VARCHAR == (ulint)MYSQL_TYPE_VARCHAR);
+
+#ifdef UNIV_DEBUG
+	static const char	test_filename[] = "-@";
+	char			test_tablename[sizeof test_filename
+				+ sizeof srv_mysql50_table_name_prefix];
+	if ((sizeof test_tablename) - 1
+			!= filename_to_tablename(test_filename, test_tablename,
+			sizeof test_tablename)
+			|| strncmp(test_tablename,
+			srv_mysql50_table_name_prefix,
+			sizeof srv_mysql50_table_name_prefix)
+			|| strcmp(test_tablename
+			+ sizeof srv_mysql50_table_name_prefix,
+			test_filename)) {
+		sql_print_error("tablename encoding has been changed");
+		goto error;
+	}
+#endif /* UNIV_DEBUG */
+
+#ifndef MYSQL_SERVER
+	innodb_overwrite_relay_log_info = FALSE;
+#endif
+
+#ifdef HAVE_REPLICATION
+#ifdef MYSQL_SERVER
+	/* read master log position from relay-log.info if exists */
+	char fname[FN_REFLEN+128];
+	int pos;
+	int info_fd;
+	IO_CACHE info_file;
+
+	fname[0] = '\0';
+
+	if(innobase_overwrite_relay_log_info) {
+
+	fprintf(stderr,
+		"InnoDB: Warning: innodb_overwrite_relay_log_info is enabled."
+		" Updates in other storage engines may have problem with consistency.\n");
+
+	bzero((char*) &info_file, sizeof(info_file));
+	fn_format(fname, relay_log_info_file, mysql_data_home, "", 4+32);
+
+	int error=0;
+
+	if (!access(fname,F_OK)) {
+		/* exist */
+		if ((info_fd = my_open(fname, O_RDWR|O_BINARY, MYF(MY_WME))) < 0) {
+			error=1;
+		} else if (init_io_cache(&info_file, info_fd, IO_SIZE*2,
+					READ_CACHE, 0L, 0, MYF(MY_WME))) {
+			error=1;
+		}
+
+		if (error) {
+relay_info_error:
+			if (info_fd >= 0)
+				my_close(info_fd, MYF(0));
+			fname[0] = '\0';
+			goto skip_relay;
+		}
+	} else {
+		fname[0] = '\0';
+		goto skip_relay;
+	}
+
+	if (init_strvar_from_file(fname, sizeof(fname), &info_file, "") || /* dummy (it is relay-log) */
+	    init_intvar_from_file(&pos, &info_file, BIN_LOG_HEADER_SIZE)) { 
+		end_io_cache(&info_file);
+		error=1;
+		goto relay_info_error;
+	}
+
+	fprintf(stderr,
+		"InnoDB: relay-log.info is detected.\n"
+		"InnoDB: relay log: position %u, file name %s\n",
+		pos, fname);
+
+	strncpy(trx_sys_mysql_relay_log_name, fname, TRX_SYS_MYSQL_MASTER_LOG_NAME_LEN);
+	trx_sys_mysql_relay_log_pos = (ib_int64_t) pos;
+
+	if (init_strvar_from_file(fname, sizeof(fname), &info_file, "") ||
+	    init_intvar_from_file(&pos, &info_file, 0)) {
+		end_io_cache(&info_file);
+		error=1;
+		goto relay_info_error;
+	}
+
+	fprintf(stderr,
+		"InnoDB: master log: position %u, file name %s\n",
+		pos, fname);
+
+	strncpy(trx_sys_mysql_master_log_name, fname, TRX_SYS_MYSQL_MASTER_LOG_NAME_LEN);
+	trx_sys_mysql_master_log_pos = (ib_int64_t) pos;
+
+	end_io_cache(&info_file);
+	if (info_fd >= 0)
+		my_close(info_fd, MYF(0));
+	}
+skip_relay:
+#endif /* MYSQL_SERVER */
+#endif /* HAVE_REPLICATION */
+
+	/* Check that values don't overflow on 32-bit systems. */
+	if (sizeof(ulint) == 4) {
+		if (innobase_buffer_pool_size > UINT_MAX32) {
+			sql_print_error(
+				"innobase_buffer_pool_size can't be over 4GB"
+				" on 32-bit systems");
+
+			goto error;
+		}
+
+		if (innobase_log_file_size > UINT_MAX32) {
+			sql_print_error(
+				"innobase_log_file_size can't be over 4GB"
+				" on 32-bit systems");
+
+			goto error;
+		}
+	}
+
+	os_innodb_umask = (ulint)my_umask;
+
+	/* First calculate the default path for innodb_data_home_dir etc.,
+	in case the user has not given any value.
+
+	Note that when using the embedded server, the datadirectory is not
+	necessarily the current directory of this program. */
+
+	if (mysqld_embedded) {
+		default_path = mysql_real_data_home;
+		fil_path_to_mysql_datadir = mysql_real_data_home;
+	} else {
+		/* It's better to use current lib, to keep paths short */
+		current_dir[0] = FN_CURLIB;
+		current_dir[1] = FN_LIBCHAR;
+		current_dir[2] = 0;
+		default_path = current_dir;
+	}
+
+	ut_a(default_path);
+
+	if (specialflag & SPECIAL_NO_PRIOR) {
+		srv_set_thread_priorities = FALSE;
+	} else {
+		srv_set_thread_priorities = TRUE;
+		srv_query_thread_priority = QUERY_PRIOR;
+	}
+
+	/* Set InnoDB initialization parameters according to the values
+	read from MySQL .cnf file */
+
+	/*--------------- Data files -------------------------*/
+
+	/* The default dir for data files is the datadir of MySQL */
+
+	srv_data_home = (innobase_data_home_dir ? innobase_data_home_dir :
+			 default_path);
+
+	/* Set default InnoDB data file size to 10 MB and let it be
+	auto-extending. Thus users can use InnoDB in >= 4.0 without having
+	to specify any startup options. */
+
+	if (!innobase_data_file_path) {
+		innobase_data_file_path = (char*) "ibdata1:10M:autoextend";
+	}
+
+	/* Since InnoDB edits the argument in the next call, we make another
+	copy of it: */
+
+	internal_innobase_data_file_path = my_strdup(innobase_data_file_path,
+						   MYF(MY_FAE));
+
+	ret = (bool) srv_parse_data_file_paths_and_sizes(
+		internal_innobase_data_file_path);
+	if (ret == FALSE) {
+		sql_print_error(
+			"InnoDB: syntax error in innodb_data_file_path");
+mem_free_and_error:
+		srv_free_paths_and_sizes();
+		my_free(internal_innobase_data_file_path,
+						MYF(MY_ALLOW_ZERO_PTR));
+		goto error;
+	}
+
+#ifdef HAVE_REPLICATION
+#ifdef MYSQL_SERVER
+	if(innobase_overwrite_relay_log_info) {
+	/* If InnoDB progressed from relay-log.info, overwrite it */
+	if (fname[0] == '\0') {
+		fprintf(stderr,
+			"InnoDB: something wrong with relay-info.log. InnoDB will not overwrite it.\n");
+	} else if (0 != strcmp(fname, trx_sys_mysql_master_log_name)
+		   || pos != trx_sys_mysql_master_log_pos) {
+		/* Overwrite relay-log.info */
+		bzero((char*) &info_file, sizeof(info_file));
+		fn_format(fname, relay_log_info_file, mysql_data_home, "", 4+32);
+
+		int error = 0;
+
+		if (!access(fname,F_OK)) {
+			/* exist */
+			if ((info_fd = my_open(fname, O_RDWR|O_BINARY, MYF(MY_WME))) < 0) {
+				error = 1;
+			} else if (init_io_cache(&info_file, info_fd, IO_SIZE*2,
+						WRITE_CACHE, 0L, 0, MYF(MY_WME))) {
+				error = 1;
+			}
+
+			if (error) {
+				if (info_fd >= 0)
+					my_close(info_fd, MYF(0));
+				goto skip_overwrite;
+			}
+		} else {
+			error = 1;
+			goto skip_overwrite;
+		}
+
+		char buff[FN_REFLEN*2+22*2+4], *pos;
+
+		my_b_seek(&info_file, 0L);
+		pos=strmov(buff, trx_sys_mysql_relay_log_name);
+		*pos++='\n';
+		pos=longlong2str(trx_sys_mysql_relay_log_pos, pos, 10);
+		*pos++='\n';
+		pos=strmov(pos, trx_sys_mysql_master_log_name);
+		*pos++='\n';
+		pos=longlong2str(trx_sys_mysql_master_log_pos, pos, 10);
+		*pos='\n';
+
+		if (my_b_write(&info_file, (uchar*) buff, (size_t) (pos-buff)+1))
+			error = 1;
+		if (flush_io_cache(&info_file))
+			error = 1;
+
+		end_io_cache(&info_file);
+		if (info_fd >= 0)
+			my_close(info_fd, MYF(0));
+skip_overwrite:
+		if (error) {
+			fprintf(stderr,
+				"InnoDB: ERROR: error occured during overwriting relay-log.info.\n");
+		} else {
+			fprintf(stderr,
+				"InnoDB: relay-log.info was overwritten.\n");
+		}
+	} else {
+		fprintf(stderr,
+			"InnoDB: InnoDB and relay-log.info are synchronized. InnoDB will not overwrite it.\n");
+	}
+	}
+#endif /* MYSQL_SERVER */
+#endif /* HAVE_REPLICATION */
+
+
+	srv_extra_undoslots = (ibool) innobase_extra_undoslots;
+
+	/* -------------- Log files ---------------------------*/
+
+	/* The default dir for log files is the datadir of MySQL */
+
+	if (!innobase_log_group_home_dir) {
+		innobase_log_group_home_dir = default_path;
+	}
+
+#ifdef UNIV_LOG_ARCHIVE
+	/* Since innodb_log_arch_dir has no relevance under MySQL,
+	starting from 4.0.6 we always set it the same as
+	innodb_log_group_home_dir: */
+
+	innobase_log_arch_dir = innobase_log_group_home_dir;
+
+	srv_arch_dir = innobase_log_arch_dir;
+#endif /* UNIG_LOG_ARCHIVE */
+
+	ret = (bool)
+		srv_parse_log_group_home_dirs(innobase_log_group_home_dir);
+
+	if (ret == FALSE || innobase_mirrored_log_groups != 1) {
+	  sql_print_error("syntax error in innodb_log_group_home_dir, or a "
+			  "wrong number of mirrored log groups");
+
+		goto mem_free_and_error;
+	}
+
+	/* Validate the file format by animal name */
+	if (innobase_file_format_name != NULL) {
+
+		format_id = innobase_file_format_name_lookup(
+			innobase_file_format_name);
+
+		if (format_id > DICT_TF_FORMAT_MAX) {
+
+			sql_print_error("InnoDB: wrong innodb_file_format.");
+
+			goto mem_free_and_error;
+		}
+	} else {
+		/* Set it to the default file format id. Though this
+		should never happen. */
+		format_id = 0;
+	}
+
+	srv_file_format = format_id;
+
+	/* Given the type of innobase_file_format_name we have little
+	choice but to cast away the constness from the returned name.
+	innobase_file_format_name is used in the MySQL set variable
+	interface and so can't be const. */
+
+	innobase_file_format_name = 
+		(char*) trx_sys_file_format_id_to_name(format_id);
+
+	/* Process innobase_file_format_check variable */
+	ut_a(innobase_file_format_check != NULL);
+
+	/* As a side effect it will set srv_check_file_format_at_startup
+	on valid input. First we check for "on"/"off". */
+	if (!innobase_file_format_check_on_off(innobase_file_format_check)) {
+
+		/* Did the user specify a format name that we support ?
+		As a side effect it will update the variable
+		srv_check_file_format_at_startup */
+		if (!innobase_file_format_check_validate(
+			innobase_file_format_check)) {
+
+			sql_print_error("InnoDB: invalid "
+					"innodb_file_format_check value: "
+					"should be either 'on' or 'off' or "
+					"any value up to %s or its "
+					"equivalent numeric id",
+					trx_sys_file_format_id_to_name(
+						DICT_TF_FORMAT_MAX));
+
+			goto mem_free_and_error;
+		}
+	}
+
+	ut_a((ulint) ibuf_use < UT_ARR_SIZE(innobase_change_buffering_values));
+	innobase_change_buffering = (char*)
+		innobase_change_buffering_values[ibuf_use];
+
+	/* --------------------------------------------------*/
+
+	srv_file_flush_method_str = innobase_unix_file_flush_method;
+
+	srv_n_log_groups = (ulint) innobase_mirrored_log_groups;
+	srv_n_log_files = (ulint) innobase_log_files_in_group;
+	srv_log_file_size = (ulint) innobase_log_file_size;
+
+#ifdef UNIV_LOG_ARCHIVE
+	srv_log_archive_on = (ulint) innobase_log_archive;
+#endif /* UNIV_LOG_ARCHIVE */
+	srv_log_buffer_size = (ulint) innobase_log_buffer_size;
+
+	srv_buf_pool_size = (ulint) innobase_buffer_pool_size;
+
+	srv_mem_pool_size = (ulint) innobase_additional_mem_pool_size;
+
+	srv_n_file_io_threads = (ulint) innobase_file_io_threads;
+	srv_n_read_io_threads = (ulint) innobase_read_io_threads;
+	srv_n_write_io_threads = (ulint) innobase_write_io_threads;
+
+	srv_read_ahead &= 3;
+
+	srv_force_recovery = (ulint) innobase_force_recovery;
+
+	srv_use_doublewrite_buf = (ibool) innobase_use_doublewrite;
+	srv_use_checksums = (ibool) innobase_use_checksums;
+
+#ifdef HAVE_LARGE_PAGES
+        if ((os_use_large_pages = (ibool) my_use_large_pages))
+		os_large_page_size = (ulint) opt_large_page_size;
+#endif
+
+	row_rollback_on_timeout = (ibool) innobase_rollback_on_timeout;
+
+	srv_locks_unsafe_for_binlog = (ibool) innobase_locks_unsafe_for_binlog;
+
+	srv_max_n_open_files = (ulint) innobase_open_files;
+	srv_innodb_status = (ibool) innobase_create_status_file;
+
+	srv_print_verbose_log = mysqld_embedded ? 0 : 1;
+
+	/* Store the default charset-collation number of this MySQL
+	installation */
+
+	data_mysql_default_charset_coll = (ulint)default_charset_info->number;
+
+	ut_a(DATA_MYSQL_LATIN1_SWEDISH_CHARSET_COLL ==
+					my_charset_latin1.number);
+	ut_a(DATA_MYSQL_BINARY_CHARSET_COLL == my_charset_bin.number);
+
+	/* Store the latin1_swedish_ci character ordering table to InnoDB. For
+	non-latin1_swedish_ci charsets we use the MySQL comparison functions,
+	and consequently we do not need to know the ordering internally in
+	InnoDB. */
+
+	ut_a(0 == strcmp(my_charset_latin1.name, "latin1_swedish_ci"));
+	srv_latin1_ordering = my_charset_latin1.sort_order;
+
+	/* Since we in this module access directly the fields of a trx
+	struct, and due to different headers and flags it might happen that
+	mutex_t has a different size in this module and in InnoDB
+	modules, we check at run time that the size is the same in
+	these compilation modules. */
+
+	err = innobase_start_or_create_for_mysql();
+
+	if (err != DB_SUCCESS) {
+		goto mem_free_and_error;
+	}
+
+	innobase_open_tables = hash_create(200);
+	pthread_mutex_init(&innobase_share_mutex, MY_MUTEX_INIT_FAST);
+	pthread_mutex_init(&prepare_commit_mutex, MY_MUTEX_INIT_FAST);
+	pthread_mutex_init(&commit_threads_m, MY_MUTEX_INIT_FAST);
+	pthread_mutex_init(&commit_cond_m, MY_MUTEX_INIT_FAST);
+	pthread_cond_init(&commit_cond, NULL);
+	innodb_inited= 1;
+#ifdef MYSQL_DYNAMIC_PLUGIN
+	if (innobase_hton != p) {
+		innobase_hton = reinterpret_cast<handlerton*>(p);
+		*innobase_hton = *innodb_hton_ptr;
+	}
+#endif /* MYSQL_DYNAMIC_PLUGIN */
+
+	/* Get the current high water mark format. */
+	innobase_file_format_check = (char*) trx_sys_file_format_max_get();
+
+	DBUG_RETURN(FALSE);
+error:
+	DBUG_RETURN(TRUE);
+}
+
+/***********************************************************************
+Closes an InnoDB database. */
+static
+int
+innobase_end(handlerton *hton, ha_panic_function type)
+/*==============*/
+				/* out: TRUE if error */
+{
+	int	err= 0;
+
+	DBUG_ENTER("innobase_end");
+	DBUG_ASSERT(hton == innodb_hton_ptr);
+
+#ifdef __NETWARE__	/* some special cleanup for NetWare */
+	if (nw_panic) {
+		set_panic_flag_for_netware();
+	}
+#endif
+	if (innodb_inited) {
+
+		srv_fast_shutdown = (ulint) innobase_fast_shutdown;
+		innodb_inited = 0;
+		hash_table_free(innobase_open_tables);
+		innobase_open_tables = NULL;
+		if (innobase_shutdown_for_mysql() != DB_SUCCESS) {
+			err = 1;
+		}
+		srv_free_paths_and_sizes();
+		my_free(internal_innobase_data_file_path,
+						MYF(MY_ALLOW_ZERO_PTR));
+		pthread_mutex_destroy(&innobase_share_mutex);
+		pthread_mutex_destroy(&prepare_commit_mutex);
+		pthread_mutex_destroy(&commit_threads_m);
+		pthread_mutex_destroy(&commit_cond_m);
+		pthread_cond_destroy(&commit_cond);
+	}
+
+	DBUG_RETURN(err);
+}
+
+/********************************************************************
+Flushes InnoDB logs to disk and makes a checkpoint. Really, a commit flushes
+the logs, and the name of this function should be innobase_checkpoint. */
+static
+bool
+innobase_flush_logs(handlerton *hton)
+/*=====================*/
+				/* out: TRUE if error */
+{
+	bool	result = 0;
+
+	DBUG_ENTER("innobase_flush_logs");
+	DBUG_ASSERT(hton == innodb_hton_ptr);
+
+	log_buffer_flush_to_disk();
+
+	DBUG_RETURN(result);
+}
+
+/********************************************************************
+Return alter table flags supported in an InnoDB database. */
+static
+uint
+innobase_alter_table_flags(
+/*=======================*/
+	uint	flags)
+{
+	return(HA_ONLINE_ADD_INDEX_NO_WRITES
+		| HA_ONLINE_DROP_INDEX_NO_WRITES
+		| HA_ONLINE_ADD_UNIQUE_INDEX_NO_WRITES
+		| HA_ONLINE_DROP_UNIQUE_INDEX_NO_WRITES
+		| HA_ONLINE_ADD_PK_INDEX_NO_WRITES);
+}
+
+/*********************************************************************
+Commits a transaction in an InnoDB database. */
+static
+void
+innobase_commit_low(
+/*================*/
+	trx_t*	trx)	/* in: transaction handle */
+{
+	if (trx->conc_state == TRX_NOT_STARTED) {
+
+		return;
+	}
+
+#ifdef HAVE_REPLICATION
+#ifdef MYSQL_SERVER
+	THD *thd=current_thd;
+
+	if (thd && thd->slave_thread) {
+		/* Update the replication position info inside InnoDB.
+		   In embedded server, does nothing. */
+		const char *log_file_name, *group_relay_log_name;
+		ulonglong log_pos, relay_log_pos;
+		bool res = rpl_get_position_info(&log_file_name, &log_pos,
+						 &group_relay_log_name,
+						 &relay_log_pos);
+		if (res) {
+			trx->mysql_master_log_file_name = log_file_name;
+			trx->mysql_master_log_pos = (ib_int64_t)log_pos;
+			trx->mysql_relay_log_file_name = group_relay_log_name;
+			trx->mysql_relay_log_pos = (ib_int64_t)relay_log_pos;
+		}
+	}
+#endif /* MYSQL_SERVER */
+#endif /* HAVE_REPLICATION */
+
+	trx_commit_for_mysql(trx);
+}
+
+/*********************************************************************
+Creates an InnoDB transaction struct for the thd if it does not yet have one.
+Starts a new InnoDB transaction if a transaction is not yet started. And
+assigns a new snapshot for a consistent read if the transaction does not yet
+have one. */
+static
+int
+innobase_start_trx_and_assign_read_view(
+/*====================================*/
+			/* out: 0 */
+        handlerton *hton, /* in: Innodb handlerton */ 
+	THD*	thd)	/* in: MySQL thread handle of the user for whom
+			the transaction should be committed */
+{
+	trx_t*	trx;
+
+	DBUG_ENTER("innobase_start_trx_and_assign_read_view");
+	DBUG_ASSERT(hton == innodb_hton_ptr);
+
+	/* Create a new trx struct for thd, if it does not yet have one */
+
+	trx = check_trx_exists(thd);
+
+	/* This is just to play safe: release a possible FIFO ticket and
+	search latch. Since we will reserve the kernel mutex, we have to
+	release the search system latch first to obey the latching order. */
+
+	innobase_release_stat_resources(trx);
+
+	/* If the transaction is not started yet, start it */
+
+	trx_start_if_not_started(trx);
+
+	/* Assign a read view if the transaction does not have it yet */
+
+	trx_assign_read_view(trx);
+
+	/* Set the MySQL flag to mark that there is an active transaction */
+
+	if (trx->active_trans == 0) {
+		innobase_register_trx_and_stmt(hton, thd);
+		trx->active_trans = 1;
+	}
+
+	DBUG_RETURN(0);
+}
+
+/*********************************************************************
+Commits a transaction in an InnoDB database or marks an SQL statement
+ended. */
+static
+int
+innobase_commit(
+/*============*/
+			/* out: 0 */
+        handlerton *hton, /* in: Innodb handlerton */ 
+	THD* 	thd,	/* in: MySQL thread handle of the user for whom
+			the transaction should be committed */
+	bool	all)	/* in:	TRUE - commit transaction
+				FALSE - the current SQL statement ended */
+{
+	trx_t*		trx;
+
+	DBUG_ENTER("innobase_commit");
+	DBUG_ASSERT(hton == innodb_hton_ptr);
+	DBUG_PRINT("trans", ("ending transaction"));
+
+	trx = check_trx_exists(thd);
+
+	/* Since we will reserve the kernel mutex, we have to release
+	the search system latch first to obey the latching order. */
+
+	if (trx->has_search_latch) {
+		trx_search_latch_release_if_reserved(trx);
+	}
+
+	/* The flag trx->active_trans is set to 1 in
+
+	1. ::external_lock(),
+	2. ::start_stmt(),
+	3. innobase_query_caching_of_table_permitted(),
+	4. innobase_savepoint(),
+	5. ::init_table_handle_for_HANDLER(),
+	6. innobase_start_trx_and_assign_read_view(),
+	7. ::transactional_table_lock()
+
+	and it is only set to 0 in a commit or a rollback. If it is 0 we know
+	there cannot be resources to be freed and we could return immediately.
+	For the time being, we play safe and do the cleanup though there should
+	be nothing to clean up. */
+
+	if (trx->active_trans == 0
+		&& trx->conc_state != TRX_NOT_STARTED) {
+
+		sql_print_error("trx->active_trans == 0, but"
+			" trx->conc_state != TRX_NOT_STARTED");
+	}
+	if (all
+		|| (!thd_test_options(thd, OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN))) {
+
+		/* We were instructed to commit the whole transaction, or
+		this is an SQL statement end and autocommit is on */
+
+		/* We need current binlog position for ibbackup to work.
+		Note, the position is current because of
+		prepare_commit_mutex */
+retry:
+		if (srv_commit_concurrency > 0) {
+			pthread_mutex_lock(&commit_cond_m);
+			commit_threads++;
+
+			if (commit_threads > srv_commit_concurrency) {
+				commit_threads--;
+				pthread_cond_wait(&commit_cond,
+					&commit_cond_m);
+				pthread_mutex_unlock(&commit_cond_m);
+				goto retry;
+			}
+			else {
+				pthread_mutex_unlock(&commit_cond_m);
+			}
+		}
+
+		trx->mysql_log_file_name = mysql_bin_log_file_name();
+		trx->mysql_log_offset = (ib_int64_t) mysql_bin_log_file_pos();
+
+		innobase_commit_low(trx);
+
+		if (srv_commit_concurrency > 0) {
+			pthread_mutex_lock(&commit_cond_m);
+			commit_threads--;
+			pthread_cond_signal(&commit_cond);
+			pthread_mutex_unlock(&commit_cond_m);
+		}
+
+		if (trx->active_trans == 2) {
+
+			pthread_mutex_unlock(&prepare_commit_mutex);
+		}
+
+		trx->active_trans = 0;
+
+	} else {
+		/* We just mark the SQL statement ended and do not do a
+		transaction commit */
+
+		/* If we had reserved the auto-inc lock for some
+		table in this SQL statement we release it now */
+
+		row_unlock_table_autoinc_for_mysql(trx);
+
+		/* Store the current undo_no of the transaction so that we
+		know where to roll back if we have to roll back the next
+		SQL statement */
+
+		trx_mark_sql_stat_end(trx);
+	}
+
+	trx->n_autoinc_rows = 0; /* Reset the number AUTO-INC rows required */
+
+	if (trx->declared_to_be_inside_innodb) {
+		/* Release our possible ticket in the FIFO */
+
+		srv_conc_force_exit_innodb(trx);
+	}
+
+	/* Tell the InnoDB server that there might be work for utility
+	threads: */
+	srv_active_wake_master_thread();
+
+	DBUG_RETURN(0);
+}
+
+/*********************************************************************
+Rolls back a transaction or the latest SQL statement. */
+static
+int
+innobase_rollback(
+/*==============*/
+			/* out: 0 or error number */
+        handlerton *hton, /* in: Innodb handlerton */ 
+	THD*	thd,	/* in: handle to the MySQL thread of the user
+			whose transaction should be rolled back */
+	bool	all)	/* in:	TRUE - commit transaction
+				FALSE - the current SQL statement ended */
+{
+	int	error = 0;
+	trx_t*	trx;
+
+	DBUG_ENTER("innobase_rollback");
+	DBUG_ASSERT(hton == innodb_hton_ptr);
+	DBUG_PRINT("trans", ("aborting transaction"));
+
+	trx = check_trx_exists(thd);
+
+	/* Release a possible FIFO ticket and search latch. Since we will
+	reserve the kernel mutex, we have to release the search system latch
+	first to obey the latching order. */
+
+	innobase_release_stat_resources(trx);
+
+	/* If we had reserved the auto-inc lock for some table (if
+	we come here to roll back the latest SQL statement) we
+	release it now before a possibly lengthy rollback */
+
+	row_unlock_table_autoinc_for_mysql(trx);
+
+	if (all
+		|| !thd_test_options(thd, OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN)) {
+
+		error = trx_rollback_for_mysql(trx);
+		trx->active_trans = 0;
+	} else {
+		error = trx_rollback_last_sql_stat_for_mysql(trx);
+	}
+
+	DBUG_RETURN(convert_error_code_to_mysql(error, 0, NULL));
+}
+
+/*********************************************************************
+Rolls back a transaction */
+static
+int
+innobase_rollback_trx(
+/*==================*/
+			/* out: 0 or error number */
+	trx_t*	trx)	/*  in: transaction */
+{
+	int	error = 0;
+
+	DBUG_ENTER("innobase_rollback_trx");
+	DBUG_PRINT("trans", ("aborting transaction"));
+
+	/* Release a possible FIFO ticket and search latch. Since we will
+	reserve the kernel mutex, we have to release the search system latch
+	first to obey the latching order. */
+
+	innobase_release_stat_resources(trx);
+
+	/* If we had reserved the auto-inc lock for some table (if
+	we come here to roll back the latest SQL statement) we
+	release it now before a possibly lengthy rollback */
+
+	row_unlock_table_autoinc_for_mysql(trx);
+
+	error = trx_rollback_for_mysql(trx);
+
+	DBUG_RETURN(convert_error_code_to_mysql(error, 0, NULL));
+}
+
+/*********************************************************************
+Rolls back a transaction to a savepoint. */
+static
+int
+innobase_rollback_to_savepoint(
+/*===========================*/
+				/* out: 0 if success, HA_ERR_NO_SAVEPOINT if
+				no savepoint with the given name */
+        handlerton *hton,       /* in: Innodb handlerton */ 
+	THD*	thd,		/* in: handle to the MySQL thread of the user
+				whose transaction should be rolled back */
+	void*	savepoint)	/* in: savepoint data */
+{
+	ib_int64_t	mysql_binlog_cache_pos;
+	int		error = 0;
+	trx_t*		trx;
+	char		name[64];
+
+	DBUG_ENTER("innobase_rollback_to_savepoint");
+	DBUG_ASSERT(hton == innodb_hton_ptr);
+
+	trx = check_trx_exists(thd);
+
+	/* Release a possible FIFO ticket and search latch. Since we will
+	reserve the kernel mutex, we have to release the search system latch
+	first to obey the latching order. */
+
+	innobase_release_stat_resources(trx);
+
+	/* TODO: use provided savepoint data area to store savepoint data */
+
+	longlong2str((ulint)savepoint, name, 36);
+
+	error = (int) trx_rollback_to_savepoint_for_mysql(trx, name,
+						&mysql_binlog_cache_pos);
+	DBUG_RETURN(convert_error_code_to_mysql(error, 0, NULL));
+}
+
+/*********************************************************************
+Release transaction savepoint name. */
+static
+int
+innobase_release_savepoint(
+/*=======================*/
+				/* out: 0 if success, HA_ERR_NO_SAVEPOINT if
+				no savepoint with the given name */
+        handlerton*	hton,	/* in: handlerton for Innodb */
+	THD*	thd,		/* in: handle to the MySQL thread of the user
+				whose transaction should be rolled back */
+	void*	savepoint)	/* in: savepoint data */
+{
+	int		error = 0;
+	trx_t*		trx;
+	char		name[64];
+
+	DBUG_ENTER("innobase_release_savepoint");
+	DBUG_ASSERT(hton == innodb_hton_ptr);
+
+	trx = check_trx_exists(thd);
+
+	/* TODO: use provided savepoint data area to store savepoint data */
+
+	longlong2str((ulint)savepoint, name, 36);
+
+	error = (int) trx_release_savepoint_for_mysql(trx, name);
+
+	DBUG_RETURN(convert_error_code_to_mysql(error, 0, NULL));
+}
+
+/*********************************************************************
+Sets a transaction savepoint. */
+static
+int
+innobase_savepoint(
+/*===============*/
+				/* out: always 0, that is, always succeeds */
+	handlerton*	hton,   /* in: handle to the Innodb handlerton */
+	THD*	thd,		/* in: handle to the MySQL thread */
+	void*	savepoint)	/* in: savepoint data */
+{
+	int	error = 0;
+	trx_t*	trx;
+
+	DBUG_ENTER("innobase_savepoint");
+	DBUG_ASSERT(hton == innodb_hton_ptr);
+
+	/*
+	  In the autocommit mode there is no sense to set a savepoint
+	  (unless we are in sub-statement), so SQL layer ensures that
+	  this method is never called in such situation.
+	*/
+#ifdef MYSQL_SERVER /* plugins cannot access thd->in_sub_stmt */
+	DBUG_ASSERT(thd_test_options(thd, OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN) ||
+		thd->in_sub_stmt);
+#endif /* MYSQL_SERVER */
+
+	trx = check_trx_exists(thd);
+
+	/* Release a possible FIFO ticket and search latch. Since we will
+	reserve the kernel mutex, we have to release the search system latch
+	first to obey the latching order. */
+
+	innobase_release_stat_resources(trx);
+
+	/* cannot happen outside of transaction */
+	DBUG_ASSERT(trx->active_trans);
+
+	/* TODO: use provided savepoint data area to store savepoint data */
+	char name[64];
+	longlong2str((ulint)savepoint,name,36);
+
+	error = (int) trx_savepoint_for_mysql(trx, name, (ib_int64_t)0);
+
+	DBUG_RETURN(convert_error_code_to_mysql(error, 0, NULL));
+}
+
+/*********************************************************************
+Frees a possible InnoDB trx object associated with the current THD. */
+static
+int
+innobase_close_connection(
+/*======================*/
+			/* out: 0 or error number */
+        handlerton*	hton,	/* in:  innobase handlerton */
+	THD*	thd)	/* in: handle to the MySQL thread of the user
+			whose resources should be free'd */
+{
+	trx_t*	trx;
+
+	DBUG_ENTER("innobase_close_connection");
+	DBUG_ASSERT(hton == innodb_hton_ptr);
+	trx = thd_to_trx(thd);
+
+	ut_a(trx);
+
+	if (trx->active_trans == 0
+		&& trx->conc_state != TRX_NOT_STARTED) {
+
+		sql_print_error("trx->active_trans == 0, but"
+			" trx->conc_state != TRX_NOT_STARTED");
+	}
+
+
+	if (trx->conc_state != TRX_NOT_STARTED &&
+		global_system_variables.log_warnings) {
+		sql_print_warning(
+			"MySQL is closing a connection that has an active "
+			"InnoDB transaction.  %lu row modifications will "
+			"roll back.",
+			(ulong) trx->undo_no.low);
+	}
+
+	innobase_rollback_trx(trx);
+
+	thr_local_free(trx->mysql_thread_id);
+	trx_free_for_mysql(trx);
+
+	DBUG_RETURN(0);
+}
+
+
+/*****************************************************************************
+** InnoDB database tables
+*****************************************************************************/
+
+/********************************************************************
+Get the record format from the data dictionary. */
+UNIV_INTERN
+enum row_type
+ha_innobase::get_row_type() const
+/*=============================*/
+			/* out: one of
+			ROW_TYPE_REDUNDANT,
+			ROW_TYPE_COMPACT,
+			ROW_TYPE_COMPRESSED,
+			ROW_TYPE_DYNAMIC */
+{
+	if (prebuilt && prebuilt->table) {
+		const ulint	flags = prebuilt->table->flags;
+
+		if (UNIV_UNLIKELY(!flags)) {
+			return(ROW_TYPE_REDUNDANT);
+		}
+
+		ut_ad(flags & DICT_TF_COMPACT);
+
+		switch (flags & DICT_TF_FORMAT_MASK) {
+		case DICT_TF_FORMAT_51 << DICT_TF_FORMAT_SHIFT:
+			return(ROW_TYPE_COMPACT);
+		case DICT_TF_FORMAT_ZIP << DICT_TF_FORMAT_SHIFT:
+			if (flags & DICT_TF_ZSSIZE_MASK) {
+				return(ROW_TYPE_COMPRESSED);
+			} else {
+				return(ROW_TYPE_DYNAMIC);
+			}
+#if DICT_TF_FORMAT_ZIP != DICT_TF_FORMAT_MAX
+# error "DICT_TF_FORMAT_ZIP != DICT_TF_FORMAT_MAX"
+#endif
+		}
+	}
+	ut_ad(0);
+	return(ROW_TYPE_NOT_USED);
+}
+
+
+
+/********************************************************************
+Get the table flags to use for the statement. */
+UNIV_INTERN
+handler::Table_flags
+ha_innobase::table_flags() const
+{
+       /* Need to use tx_isolation here since table flags is (also)
+          called before prebuilt is inited. */
+        ulong const tx_isolation = thd_tx_isolation(ha_thd());
+        if (tx_isolation <= ISO_READ_COMMITTED)
+                return int_table_flags;
+        return int_table_flags | HA_BINLOG_STMT_CAPABLE;
+}
+
+/********************************************************************
+Gives the file extension of an InnoDB single-table tablespace. */
+static const char* ha_innobase_exts[] = {
+  ".ibd",
+  NullS
+};
+
+UNIV_INTERN
+const char*
+ha_innobase::table_type() const
+/*===========================*/
+				/* out: table type */
+{
+	return(innobase_hton_name);
+}
+
+UNIV_INTERN
+const char*
+ha_innobase::index_type(uint)
+/*=========================*/
+				/* out: index type */
+{
+	return("BTREE");
+}
+
+UNIV_INTERN
+const char**
+ha_innobase::bas_ext() const
+/*========================*/
+				/* out: file extension string */
+{
+	return(ha_innobase_exts);
+}
+
+UNIV_INTERN
+ulong
+ha_innobase::index_flags(uint, uint, bool) const
+{
+	return(HA_READ_NEXT | HA_READ_PREV | HA_READ_ORDER
+	       | HA_READ_RANGE | HA_KEYREAD_ONLY);
+}
+
+UNIV_INTERN
+uint
+ha_innobase::max_supported_keys() const
+{
+	return(MAX_KEY);
+}
+
+UNIV_INTERN
+uint
+ha_innobase::max_supported_key_length() const
+{
+	/* An InnoDB page must store >= 2 keys; a secondary key record
+	must also contain the primary key value: max key length is
+	therefore set to slightly less than 1 / 4 of page size which
+	is 16 kB; but currently MySQL does not work with keys whose
+	size is > MAX_KEY_LENGTH */
+	return(3500);
+}
+
+UNIV_INTERN
+const key_map*
+ha_innobase::keys_to_use_for_scanning()
+{
+	return(&key_map_full);
+}
+
+UNIV_INTERN
+uint8
+ha_innobase::table_cache_type()
+{
+	return(HA_CACHE_TBL_ASKTRANSACT);
+}
+
+UNIV_INTERN
+bool
+ha_innobase::primary_key_is_clustered()
+{
+	return(true);
+}
+
+/*********************************************************************
+Normalizes a table name string. A normalized name consists of the
+database name catenated to '/' and table name. An example:
+test/mytable. On Windows normalization puts both the database name and the
+table name always to lower case. */
+static
+void
+normalize_table_name(
+/*=================*/
+	char*		norm_name,	/* out: normalized name as a
+					null-terminated string */
+	const char*	name)		/* in: table name string */
+{
+	char*	name_ptr;
+	char*	db_ptr;
+	char*	ptr;
+
+	/* Scan name from the end */
+
+	ptr = strend(name)-1;
+
+	while (ptr >= name && *ptr != '\\' && *ptr != '/') {
+		ptr--;
+	}
+
+	name_ptr = ptr + 1;
+
+	DBUG_ASSERT(ptr > name);
+
+	ptr--;
+
+	while (ptr >= name && *ptr != '\\' && *ptr != '/') {
+		ptr--;
+	}
+
+	db_ptr = ptr + 1;
+
+	memcpy(norm_name, db_ptr, strlen(name) + 1 - (db_ptr - name));
+
+	norm_name[name_ptr - db_ptr - 1] = '/';
+
+#ifdef __WIN__
+	innobase_casedn_str(norm_name);
+#endif
+}
+
+/************************************************************************
+Set the autoinc column max value. This should only be called once from
+ha_innobase::open(). Therefore there's no need for a covering lock. */
+UNIV_INTERN
+ulint
+ha_innobase::innobase_initialize_autoinc()
+/*======================================*/
+{
+	dict_index_t*	index;
+	ulonglong	auto_inc;
+	const char*	col_name;
+	ulint		error = DB_SUCCESS;
+	dict_table_t*	innodb_table = prebuilt->table;
+
+	col_name = table->found_next_number_field->field_name;
+	index = innobase_get_index(table->s->next_number_index);
+
+	/* Execute SELECT MAX(col_name) FROM TABLE; */
+	error = row_search_max_autoinc(index, col_name, &auto_inc);
+
+	if (error == DB_SUCCESS) {
+
+		/* At the this stage we dont' know the increment
+		or the offset, so use default inrement of 1. */
+		++auto_inc;
+
+		dict_table_autoinc_initialize(innodb_table, auto_inc);
+
+	} else {
+		ut_print_timestamp(stderr);
+		fprintf(stderr, "  InnoDB: Error: (%lu) Couldn't read "
+			"the MAX(%s) autoinc value from the "
+			"index (%s).\n", error, col_name, index->name);
+	}
+
+	return(error);
+}
+
+/*********************************************************************
+Creates and opens a handle to a table which already exists in an InnoDB
+database. */
+UNIV_INTERN
+int
+ha_innobase::open(
+/*==============*/
+					/* out: 1 if error, 0 if success */
+	const char*	name,		/* in: table name */
+	int		mode,		/* in: not used */
+	uint		test_if_locked)	/* in: not used */
+{
+	dict_table_t*	ib_table;
+	char		norm_name[1000];
+	THD*		thd;
+	ulint		retries = 0;
+	char*		is_part = NULL;
+
+	DBUG_ENTER("ha_innobase::open");
+
+	UT_NOT_USED(mode);
+	UT_NOT_USED(test_if_locked);
+
+	thd = ha_thd();
+
+	/* Under some cases MySQL seems to call this function while
+	holding btr_search_latch. This breaks the latching order as
+	we acquire dict_sys->mutex below and leads to a deadlock. */
+	if (thd != NULL) {
+		innobase_release_temporary_latches(ht, thd);
+	}
+
+	normalize_table_name(norm_name, name);
+
+	user_thd = NULL;
+
+	if (!(share=get_share(name))) {
+
+		DBUG_RETURN(1);
+	}
+
+	/* Create buffers for packing the fields of a record. Why
+	table->reclength did not work here? Obviously, because char
+	fields when packed actually became 1 byte longer, when we also
+	stored the string length as the first byte. */
+
+	upd_and_key_val_buff_len =
+				table->s->reclength + table->s->max_key_length
+							+ MAX_REF_PARTS * 3;
+	if (!(uchar*) my_multi_malloc(MYF(MY_WME),
+			&upd_buff, upd_and_key_val_buff_len,
+			&key_val_buff, upd_and_key_val_buff_len,
+			NullS)) {
+		free_share(share);
+
+		DBUG_RETURN(1);
+	}
+
+	/* We look for pattern #P# to see if the table is partitioned
+	MySQL table. The retry logic for partitioned tables is a
+	workaround for http://bugs.mysql.com/bug.php?id=33349. Look
+	at support issue https://support.mysql.com/view.php?id=21080
+	for more details. */
+	is_part = strstr(norm_name, "#P#");
+retry:
+	/* Get pointer to a table object in InnoDB dictionary cache */
+	ib_table = dict_table_get(norm_name, TRUE);
+	
+	if (NULL == ib_table) {
+		if (is_part && retries < 10) {
+			++retries;
+			os_thread_sleep(100000);
+			goto retry;
+		}
+
+		if (is_part) {
+			sql_print_error("Failed to open table %s after "
+					"%lu attemtps.\n", norm_name,
+					retries);
+		}
+
+		sql_print_error("Cannot find or open table %s from\n"
+				"the internal data dictionary of InnoDB "
+				"though the .frm file for the\n"
+				"table exists. Maybe you have deleted and "
+				"recreated InnoDB data\n"
+				"files but have forgotten to delete the "
+				"corresponding .frm files\n"
+				"of InnoDB tables, or you have moved .frm "
+				"files to another database?\n"
+				"or, the table contains indexes that this "
+				"version of the engine\n"
+				"doesn't support.\n"
+				"See http://dev.mysql.com/doc/refman/5.1/en/innodb-troubleshooting.html\n"
+				"how you can resolve the problem.\n",
+				norm_name);
+		free_share(share);
+		my_free(upd_buff, MYF(0));
+		my_errno = ENOENT;
+
+		DBUG_RETURN(HA_ERR_NO_SUCH_TABLE);
+	}
+
+	if (ib_table->ibd_file_missing && !thd_tablespace_op(thd)) {
+		sql_print_error("MySQL is trying to open a table handle but "
+				"the .ibd file for\ntable %s does not exist.\n"
+				"Have you deleted the .ibd file from the "
+				"database directory under\nthe MySQL datadir, "
+				"or have you used DISCARD TABLESPACE?\n"
+				"See http://dev.mysql.com/doc/refman/5.1/en/innodb-troubleshooting.html\n"
+				"how you can resolve the problem.\n",
+				norm_name);
+		free_share(share);
+		my_free(upd_buff, MYF(0));
+		my_errno = ENOENT;
+
+		dict_table_decrement_handle_count(ib_table, FALSE);
+		DBUG_RETURN(HA_ERR_NO_SUCH_TABLE);
+	}
+
+	prebuilt = row_create_prebuilt(ib_table);
+
+	prebuilt->mysql_row_len = table->s->reclength;
+	prebuilt->default_rec = table->s->default_values;
+	ut_ad(prebuilt->default_rec);
+
+	/* Looks like MySQL-3.23 sometimes has primary key number != 0 */
+
+	primary_key = table->s->primary_key;
+	key_used_on_scan = primary_key;
+
+	/* Allocate a buffer for a 'row reference'. A row reference is
+	a string of bytes of length ref_length which uniquely specifies
+	a row in our table. Note that MySQL may also compare two row
+	references for equality by doing a simple memcmp on the strings
+	of length ref_length! */
+
+	if (!row_table_got_default_clust_index(ib_table)) {
+		if (primary_key >= MAX_KEY) {
+		  sql_print_error("Table %s has a primary key in InnoDB data "
+				  "dictionary, but not in MySQL!", name);
+		}
+
+		prebuilt->clust_index_was_generated = FALSE;
+
+		/* MySQL allocates the buffer for ref. key_info->key_length
+		includes space for all key columns + one byte for each column
+		that may be NULL. ref_length must be as exact as possible to
+		save space, because all row reference buffers are allocated
+		based on ref_length. */
+
+		ref_length = table->key_info[primary_key].key_length;
+	} else {
+		if (primary_key != MAX_KEY) {
+		  sql_print_error("Table %s has no primary key in InnoDB data "
+				  "dictionary, but has one in MySQL! If you "
+				  "created the table with a MySQL version < "
+				  "3.23.54 and did not define a primary key, "
+				  "but defined a unique key with all non-NULL "
+				  "columns, then MySQL internally treats that "
+				  "key as the primary key. You can fix this "
+				  "error by dump + DROP + CREATE + reimport "
+				  "of the table.", name);
+		}
+
+		prebuilt->clust_index_was_generated = TRUE;
+
+		ref_length = DATA_ROW_ID_LEN;
+
+		/* If we automatically created the clustered index, then
+		MySQL does not know about it, and MySQL must NOT be aware
+		of the index used on scan, to make it avoid checking if we
+		update the column of the index. That is why we assert below
+		that key_used_on_scan is the undefined value MAX_KEY.
+		The column is the row id in the automatical generation case,
+		and it will never be updated anyway. */
+
+		if (key_used_on_scan != MAX_KEY) {
+			sql_print_warning(
+				"Table %s key_used_on_scan is %lu even "
+				"though there is no primary key inside "
+				"InnoDB.", name, (ulong) key_used_on_scan);
+		}
+	}
+
+	/* Index block size in InnoDB: used by MySQL in query optimization */
+	stats.block_size = 16 * 1024;
+
+	/* Init table lock structure */
+	thr_lock_data_init(&share->lock,&lock,(void*) 0);
+
+	if (prebuilt->table) {
+		/* We update the highest file format in the system table
+		space, if this table has higher file format setting. */
+
+		trx_sys_file_format_max_upgrade(
+			(const char**) &innobase_file_format_check,
+			dict_table_get_format(prebuilt->table));
+	}
+
+	info(HA_STATUS_NO_LOCK | HA_STATUS_VARIABLE | HA_STATUS_CONST);
+
+	/* Only if the table has an AUTOINC column. */
+	if (prebuilt->table != NULL && table->found_next_number_field != NULL) {
+		ulint	error;
+
+		dict_table_autoinc_lock(prebuilt->table);
+
+		/* Since a table can already be "open" in InnoDB's internal
+		data dictionary, we only init the autoinc counter once, the
+		first time the table is loaded. We can safely reuse the
+		autoinc value from a previous MySQL open. */
+		if (dict_table_autoinc_read(prebuilt->table) == 0) {
+
+			error = innobase_initialize_autoinc();
+			/* Should always succeed! */
+			ut_a(error == DB_SUCCESS);
+		}
+
+		dict_table_autoinc_unlock(prebuilt->table);
+	}
+
+	DBUG_RETURN(0);
+}
+
+UNIV_INTERN
+uint
+ha_innobase::max_supported_key_part_length() const
+{
+	return(DICT_MAX_INDEX_COL_LEN - 1);
+}
+
+/**********************************************************************
+Closes a handle to an InnoDB table. */
+UNIV_INTERN
+int
+ha_innobase::close(void)
+/*====================*/
+				/* out: 0 */
+{
+	THD*	thd;
+
+	DBUG_ENTER("ha_innobase::close");
+
+	thd = ha_thd();
+	if (thd != NULL) {
+		innobase_release_temporary_latches(ht, thd);
+	}
+
+	row_prebuilt_free(prebuilt, FALSE);
+
+	my_free(upd_buff, MYF(0));
+	free_share(share);
+
+	/* Tell InnoDB server that there might be work for
+	utility threads: */
+
+	srv_active_wake_master_thread();
+
+	DBUG_RETURN(0);
+}
+
+/* The following accessor functions should really be inside MySQL code! */
+
+/******************************************************************
+Gets field offset for a field in a table. */
+static inline
+uint
+get_field_offset(
+/*=============*/
+			/* out: offset */
+	TABLE*	table,	/* in: MySQL table object */
+	Field*	field)	/* in: MySQL field object */
+{
+	return((uint) (field->ptr - table->record[0]));
+}
+
+/******************************************************************
+Checks if a field in a record is SQL NULL. Uses the record format
+information in table to track the null bit in record. */
+static inline
+uint
+field_in_record_is_null(
+/*====================*/
+			/* out: 1 if NULL, 0 otherwise */
+	TABLE*	table,	/* in: MySQL table object */
+	Field*	field,	/* in: MySQL field object */
+	char*	record)	/* in: a row in MySQL format */
+{
+	int	null_offset;
+
+	if (!field->null_ptr) {
+
+		return(0);
+	}
+
+	null_offset = (uint) ((char*) field->null_ptr
+					- (char*) table->record[0]);
+
+	if (record[null_offset] & field->null_bit) {
+
+		return(1);
+	}
+
+	return(0);
+}
+
+/******************************************************************
+Sets a field in a record to SQL NULL. Uses the record format
+information in table to track the null bit in record. */
+static inline
+void
+set_field_in_record_to_null(
+/*========================*/
+	TABLE*	table,	/* in: MySQL table object */
+	Field*	field,	/* in: MySQL field object */
+	char*	record)	/* in: a row in MySQL format */
+{
+	int	null_offset;
+
+	null_offset = (uint) ((char*) field->null_ptr
+					- (char*) table->record[0]);
+
+	record[null_offset] = record[null_offset] | field->null_bit;
+}
+
+/*****************************************************************
+InnoDB uses this function to compare two data fields for which the data type
+is such that we must use MySQL code to compare them. NOTE that the prototype
+of this function is in rem0cmp.c in InnoDB source code! If you change this
+function, remember to update the prototype there! */
+extern "C" UNIV_INTERN
+int
+innobase_mysql_cmp(
+/*===============*/
+					/* out: 1, 0, -1, if a is greater,
+					equal, less than b, respectively */
+	int		mysql_type,	/* in: MySQL type */
+	uint		charset_number,	/* in: number of the charset */
+	const unsigned char* a,		/* in: data field */
+	unsigned int	a_length,	/* in: data field length,
+					not UNIV_SQL_NULL */
+	const unsigned char* b,		/* in: data field */
+	unsigned int	b_length)	/* in: data field length,
+					not UNIV_SQL_NULL */
+{
+	CHARSET_INFO*		charset;
+	enum_field_types	mysql_tp;
+	int			ret;
+
+	DBUG_ASSERT(a_length != UNIV_SQL_NULL);
+	DBUG_ASSERT(b_length != UNIV_SQL_NULL);
+
+	mysql_tp = (enum_field_types) mysql_type;
+
+	switch (mysql_tp) {
+
+	case MYSQL_TYPE_BIT:
+	case MYSQL_TYPE_STRING:
+	case MYSQL_TYPE_VAR_STRING:
+	case MYSQL_TYPE_TINY_BLOB:
+	case MYSQL_TYPE_MEDIUM_BLOB:
+	case MYSQL_TYPE_BLOB:
+	case MYSQL_TYPE_LONG_BLOB:
+	case MYSQL_TYPE_VARCHAR:
+		/* Use the charset number to pick the right charset struct for
+		the comparison. Since the MySQL function get_charset may be
+		slow before Bar removes the mutex operation there, we first
+		look at 2 common charsets directly. */
+
+		if (charset_number == default_charset_info->number) {
+			charset = default_charset_info;
+		} else if (charset_number == my_charset_latin1.number) {
+			charset = &my_charset_latin1;
+		} else {
+			charset = get_charset(charset_number, MYF(MY_WME));
+
+			if (charset == NULL) {
+			  sql_print_error("InnoDB needs charset %lu for doing "
+					  "a comparison, but MySQL cannot "
+					  "find that charset.",
+					  (ulong) charset_number);
+				ut_a(0);
+			}
+		}
+
+		/* Starting from 4.1.3, we use strnncollsp() in comparisons of
+		non-latin1_swedish_ci strings. NOTE that the collation order
+		changes then: 'b\0\0...' is ordered BEFORE 'b  ...'. Users
+		having indexes on such data need to rebuild their tables! */
+
+		ret = charset->coll->strnncollsp(charset,
+				  a, a_length,
+						 b, b_length, 0);
+		if (ret < 0) {
+			return(-1);
+		} else if (ret > 0) {
+			return(1);
+		} else {
+			return(0);
+		}
+	default:
+		ut_error;
+	}
+
+	return(0);
+}
+
+/******************************************************************
+Converts a MySQL type to an InnoDB type. Note that this function returns
+the 'mtype' of InnoDB. InnoDB differentiates between MySQL's old <= 4.1
+VARCHAR and the new true VARCHAR in >= 5.0.3 by the 'prtype'. */
+extern "C" UNIV_INTERN
+ulint
+get_innobase_type_from_mysql_type(
+/*==============================*/
+					/* out: DATA_BINARY,
+					DATA_VARCHAR, ... */
+	ulint*		unsigned_flag,	/* out: DATA_UNSIGNED if an
+					'unsigned type';
+					at least ENUM and SET,
+					and unsigned integer
+					types are 'unsigned types' */
+	const void*	f)		/* in: MySQL Field */
+{
+	const class Field* field = reinterpret_cast<const class Field*>(f);
+
+	/* The following asserts try to check that the MySQL type code fits in
+	8 bits: this is used in ibuf and also when DATA_NOT_NULL is ORed to
+	the type */
+
+	DBUG_ASSERT((ulint)MYSQL_TYPE_STRING < 256);
+	DBUG_ASSERT((ulint)MYSQL_TYPE_VAR_STRING < 256);
+	DBUG_ASSERT((ulint)MYSQL_TYPE_DOUBLE < 256);
+	DBUG_ASSERT((ulint)MYSQL_TYPE_FLOAT < 256);
+	DBUG_ASSERT((ulint)MYSQL_TYPE_DECIMAL < 256);
+
+	if (field->flags & UNSIGNED_FLAG) {
+
+		*unsigned_flag = DATA_UNSIGNED;
+	} else {
+		*unsigned_flag = 0;
+	}
+
+	if (field->real_type() == MYSQL_TYPE_ENUM
+		|| field->real_type() == MYSQL_TYPE_SET) {
+
+		/* MySQL has field->type() a string type for these, but the
+		data is actually internally stored as an unsigned integer
+		code! */
+
+		*unsigned_flag = DATA_UNSIGNED; /* MySQL has its own unsigned
+						flag set to zero, even though
+						internally this is an unsigned
+						integer type */
+		return(DATA_INT);
+	}
+
+	switch (field->type()) {
+		/* NOTE that we only allow string types in DATA_MYSQL and
+		DATA_VARMYSQL */
+	case MYSQL_TYPE_VAR_STRING: /* old <= 4.1 VARCHAR */
+	case MYSQL_TYPE_VARCHAR:    /* new >= 5.0.3 true VARCHAR */
+		if (field->binary()) {
+			return(DATA_BINARY);
+		} else if (strcmp(
+				   field->charset()->name,
+				   "latin1_swedish_ci") == 0) {
+			return(DATA_VARCHAR);
+		} else {
+			return(DATA_VARMYSQL);
+		}
+	case MYSQL_TYPE_BIT:
+	case MYSQL_TYPE_STRING: if (field->binary()) {
+
+			return(DATA_FIXBINARY);
+		} else if (strcmp(
+				   field->charset()->name,
+				   "latin1_swedish_ci") == 0) {
+			return(DATA_CHAR);
+		} else {
+			return(DATA_MYSQL);
+		}
+	case MYSQL_TYPE_NEWDECIMAL:
+		return(DATA_FIXBINARY);
+	case MYSQL_TYPE_LONG:
+	case MYSQL_TYPE_LONGLONG:
+	case MYSQL_TYPE_TINY:
+	case MYSQL_TYPE_SHORT:
+	case MYSQL_TYPE_INT24:
+	case MYSQL_TYPE_DATE:
+	case MYSQL_TYPE_DATETIME:
+	case MYSQL_TYPE_YEAR:
+	case MYSQL_TYPE_NEWDATE:
+	case MYSQL_TYPE_TIME:
+	case MYSQL_TYPE_TIMESTAMP:
+		return(DATA_INT);
+	case MYSQL_TYPE_FLOAT:
+		return(DATA_FLOAT);
+	case MYSQL_TYPE_DOUBLE:
+		return(DATA_DOUBLE);
+	case MYSQL_TYPE_DECIMAL:
+		return(DATA_DECIMAL);
+	case MYSQL_TYPE_GEOMETRY:
+	case MYSQL_TYPE_TINY_BLOB:
+	case MYSQL_TYPE_MEDIUM_BLOB:
+	case MYSQL_TYPE_BLOB:
+	case MYSQL_TYPE_LONG_BLOB:
+		return(DATA_BLOB);
+	default:
+		ut_error;
+	}
+
+	return(0);
+}
+
+/***********************************************************************
+Writes an unsigned integer value < 64k to 2 bytes, in the little-endian
+storage format. */
+static inline
+void
+innobase_write_to_2_little_endian(
+/*==============================*/
+	byte*	buf,	/* in: where to store */
+	ulint	val)	/* in: value to write, must be < 64k */
+{
+	ut_a(val < 256 * 256);
+
+	buf[0] = (byte)(val & 0xFF);
+	buf[1] = (byte)(val / 256);
+}
+
+/***********************************************************************
+Reads an unsigned integer value < 64k from 2 bytes, in the little-endian
+storage format. */
+static inline
+uint
+innobase_read_from_2_little_endian(
+/*===============================*/
+				/* out: value */
+	const uchar*	buf)	/* in: from where to read */
+{
+	return (uint) ((ulint)(buf[0]) + 256 * ((ulint)(buf[1])));
+}
+
+/***********************************************************************
+Stores a key value for a row to a buffer. */
+UNIV_INTERN
+uint
+ha_innobase::store_key_val_for_row(
+/*===============================*/
+				/* out: key value length as stored in buff */
+	uint		keynr,	/* in: key number */
+	char*		buff,	/* in/out: buffer for the key value (in MySQL
+				format) */
+	uint		buff_len,/* in: buffer length */
+	const uchar*	record)/* in: row in MySQL format */
+{
+	KEY*		key_info	= table->key_info + keynr;
+	KEY_PART_INFO*	key_part	= key_info->key_part;
+	KEY_PART_INFO*	end		= key_part + key_info->key_parts;
+	char*		buff_start	= buff;
+	enum_field_types mysql_type;
+	Field*		field;
+	ibool		is_null;
+
+	DBUG_ENTER("store_key_val_for_row");
+
+	/* The format for storing a key field in MySQL is the following:
+
+	1. If the column can be NULL, then in the first byte we put 1 if the
+	field value is NULL, 0 otherwise.
+
+	2. If the column is of a BLOB type (it must be a column prefix field
+	in this case), then we put the length of the data in the field to the
+	next 2 bytes, in the little-endian format. If the field is SQL NULL,
+	then these 2 bytes are set to 0. Note that the length of data in the
+	field is <= column prefix length.
+
+	3. In a column prefix field, prefix_len next bytes are reserved for
+	data. In a normal field the max field length next bytes are reserved
+	for data. For a VARCHAR(n) the max field length is n. If the stored
+	value is the SQL NULL then these data bytes are set to 0.
+
+	4. We always use a 2 byte length for a true >= 5.0.3 VARCHAR. Note that
+	in the MySQL row format, the length is stored in 1 or 2 bytes,
+	depending on the maximum allowed length. But in the MySQL key value
+	format, the length always takes 2 bytes.
+
+	We have to zero-fill the buffer so that MySQL is able to use a
+	simple memcmp to compare two key values to determine if they are
+	equal. MySQL does this to compare contents of two 'ref' values. */
+
+	bzero(buff, buff_len);
+
+	for (; key_part != end; key_part++) {
+		is_null = FALSE;
+
+		if (key_part->null_bit) {
+			if (record[key_part->null_offset]
+						& key_part->null_bit) {
+				*buff = 1;
+				is_null = TRUE;
+			} else {
+				*buff = 0;
+			}
+			buff++;
+		}
+
+		field = key_part->field;
+		mysql_type = field->type();
+
+		if (mysql_type == MYSQL_TYPE_VARCHAR) {
+						/* >= 5.0.3 true VARCHAR */
+			ulint		lenlen;
+			ulint		len;
+			const byte*	data;
+			ulint		key_len;
+			ulint		true_len;
+			CHARSET_INFO*	cs;
+			int		error=0;
+
+			key_len = key_part->length;
+
+			if (is_null) {
+				buff += key_len + 2;
+
+				continue;
+			}
+			cs = field->charset();
+
+			lenlen = (ulint)
+				(((Field_varstring*)field)->length_bytes);
+
+			data = row_mysql_read_true_varchar(&len,
+				(byte*) (record
+				+ (ulint)get_field_offset(table, field)),
+				lenlen);
+
+			true_len = len;
+
+			/* For multi byte character sets we need to calculate
+			the true length of the key */
+
+			if (len > 0 && cs->mbmaxlen > 1) {
+				true_len = (ulint) cs->cset->well_formed_len(cs,
+						(const char *) data,
+						(const char *) data + len,
+                                                (uint) (key_len /
+                                                        cs->mbmaxlen),
+						&error);
+			}
+
+			/* In a column prefix index, we may need to truncate
+			the stored value: */
+
+			if (true_len > key_len) {
+				true_len = key_len;
+			}
+
+			/* The length in a key value is always stored in 2
+			bytes */
+
+			row_mysql_store_true_var_len((byte*)buff, true_len, 2);
+			buff += 2;
+
+			memcpy(buff, data, true_len);
+
+			/* Note that we always reserve the maximum possible
+			length of the true VARCHAR in the key value, though
+			only len first bytes after the 2 length bytes contain
+			actual data. The rest of the space was reset to zero
+			in the bzero() call above. */
+
+			buff += key_len;
+
+		} else if (mysql_type == MYSQL_TYPE_TINY_BLOB
+			|| mysql_type == MYSQL_TYPE_MEDIUM_BLOB
+			|| mysql_type == MYSQL_TYPE_BLOB
+			|| mysql_type == MYSQL_TYPE_LONG_BLOB) {
+
+			CHARSET_INFO*	cs;
+			ulint		key_len;
+			ulint		true_len;
+			int		error=0;
+			ulint		blob_len;
+			const byte*	blob_data;
+
+			ut_a(key_part->key_part_flag & HA_PART_KEY_SEG);
+
+			key_len = key_part->length;
+
+			if (is_null) {
+				buff += key_len + 2;
+
+				continue;
+			}
+
+			cs = field->charset();
+
+			blob_data = row_mysql_read_blob_ref(&blob_len,
+				(byte*) (record
+				+ (ulint)get_field_offset(table, field)),
+					(ulint) field->pack_length());
+
+			true_len = blob_len;
+
+			ut_a(get_field_offset(table, field)
+				== key_part->offset);
+
+			/* For multi byte character sets we need to calculate
+			the true length of the key */
+
+			if (blob_len > 0 && cs->mbmaxlen > 1) {
+				true_len = (ulint) cs->cset->well_formed_len(cs,
+						(const char *) blob_data,
+						(const char *) blob_data
+							+ blob_len,
+                                                (uint) (key_len /
+                                                        cs->mbmaxlen),
+						&error);
+			}
+
+			/* All indexes on BLOB and TEXT are column prefix
+			indexes, and we may need to truncate the data to be
+			stored in the key value: */
+
+			if (true_len > key_len) {
+				true_len = key_len;
+			}
+
+			/* MySQL reserves 2 bytes for the length and the
+			storage of the number is little-endian */
+
+			innobase_write_to_2_little_endian(
+					(byte*)buff, true_len);
+			buff += 2;
+
+			memcpy(buff, blob_data, true_len);
+
+			/* Note that we always reserve the maximum possible
+			length of the BLOB prefix in the key value. */
+
+			buff += key_len;
+		} else {
+			/* Here we handle all other data types except the
+			true VARCHAR, BLOB and TEXT. Note that the column
+			value we store may be also in a column prefix
+			index. */
+
+			CHARSET_INFO*		cs;
+			ulint			true_len;
+			ulint			key_len;
+			const uchar*		src_start;
+			int			error=0;
+			enum_field_types	real_type;
+
+			key_len = key_part->length;
+
+			if (is_null) {
+				 buff += key_len;
+
+				 continue;
+			}
+
+			src_start = record + key_part->offset;
+			real_type = field->real_type();
+			true_len = key_len;
+
+			/* Character set for the field is defined only
+			to fields whose type is string and real field
+			type is not enum or set. For these fields check
+			if character set is multi byte. */
+
+			if (real_type != MYSQL_TYPE_ENUM
+				&& real_type != MYSQL_TYPE_SET
+				&& ( mysql_type == MYSQL_TYPE_VAR_STRING
+					|| mysql_type == MYSQL_TYPE_STRING)) {
+
+				cs = field->charset();
+
+				/* For multi byte character sets we need to
+				calculate the true length of the key */
+
+				if (key_len > 0 && cs->mbmaxlen > 1) {
+
+					true_len = (ulint)
+						cs->cset->well_formed_len(cs,
+							(const char *)src_start,
+							(const char *)src_start
+								+ key_len,
+                                                        (uint) (key_len /
+                                                                cs->mbmaxlen),
+							&error);
+				}
+			}
+
+			memcpy(buff, src_start, true_len);
+			buff += true_len;
+
+			/* Pad the unused space with spaces. Note that no
+			padding is ever needed for UCS-2 because in MySQL,
+			all UCS2 characters are 2 bytes, as MySQL does not
+			support surrogate pairs, which are needed to represent
+			characters in the range U+10000 to U+10FFFF. */
+
+			if (true_len < key_len) {
+				ulint pad_len = key_len - true_len;
+				memset(buff, ' ', pad_len);
+				buff += pad_len;
+			}
+		}
+	}
+
+	ut_a(buff <= buff_start + buff_len);
+
+	DBUG_RETURN((uint)(buff - buff_start));
+}
+
+/******************************************************************
+Builds a 'template' to the prebuilt struct. The template is used in fast
+retrieval of just those column values MySQL needs in its processing. */
+static
+void
+build_template(
+/*===========*/
+	row_prebuilt_t*	prebuilt,	/* in/out: prebuilt struct */
+	THD*		thd,		/* in: current user thread, used
+					only if templ_type is
+					ROW_MYSQL_REC_FIELDS */
+	TABLE*		table,		/* in: MySQL table */
+	uint		templ_type)	/* in: ROW_MYSQL_WHOLE_ROW or
+					ROW_MYSQL_REC_FIELDS */
+{
+	dict_index_t*	index;
+	dict_index_t*	clust_index;
+	mysql_row_templ_t* templ;
+	Field*		field;
+	ulint		n_fields;
+	ulint		n_requested_fields	= 0;
+	ibool		fetch_all_in_key	= FALSE;
+	ibool		fetch_primary_key_cols	= FALSE;
+	ulint		i;
+	/* byte offset of the end of last requested column */
+	ulint		mysql_prefix_len	= 0;
+
+	if (prebuilt->select_lock_type == LOCK_X) {
+		/* We always retrieve the whole clustered index record if we
+		use exclusive row level locks, for example, if the read is
+		done in an UPDATE statement. */
+
+		templ_type = ROW_MYSQL_WHOLE_ROW;
+	}
+
+	if (templ_type == ROW_MYSQL_REC_FIELDS) {
+		if (prebuilt->hint_need_to_fetch_extra_cols
+			== ROW_RETRIEVE_ALL_COLS) {
+
+			/* We know we must at least fetch all columns in the
+			key, or all columns in the table */
+
+			if (prebuilt->read_just_key) {
+				/* MySQL has instructed us that it is enough
+				to fetch the columns in the key; looks like
+				MySQL can set this flag also when there is
+				only a prefix of the column in the key: in
+				that case we retrieve the whole column from
+				the clustered index */
+
+				fetch_all_in_key = TRUE;
+			} else {
+				templ_type = ROW_MYSQL_WHOLE_ROW;
+			}
+		} else if (prebuilt->hint_need_to_fetch_extra_cols
+			== ROW_RETRIEVE_PRIMARY_KEY) {
+			/* We must at least fetch all primary key cols. Note
+			   that if the clustered index was internally generated
+			   by InnoDB on the row id (no primary key was
+			   defined), then row_search_for_mysql() will always
+			   retrieve the row id to a special buffer in the
+			   prebuilt struct. */
+
+			fetch_primary_key_cols = TRUE;
+		}
+	}
+
+	clust_index = dict_table_get_first_index(prebuilt->table);
+
+	if (templ_type == ROW_MYSQL_REC_FIELDS) {
+		index = prebuilt->index;
+	} else {
+		index = clust_index;
+	}
+
+	if (index == clust_index) {
+		prebuilt->need_to_access_clustered = TRUE;
+	} else {
+		prebuilt->need_to_access_clustered = FALSE;
+		/* Below we check column by column if we need to access
+		the clustered index */
+	}
+
+	n_fields = (ulint)table->s->fields; /* number of columns */
+
+	if (!prebuilt->mysql_template) {
+		prebuilt->mysql_template = (mysql_row_templ_t*)
+			mem_alloc(n_fields * sizeof(mysql_row_templ_t));
+	}
+
+	prebuilt->template_type = templ_type;
+	prebuilt->null_bitmap_len = table->s->null_bytes;
+
+	prebuilt->templ_contains_blob = FALSE;
+
+	/* Note that in InnoDB, i is the column number. MySQL calls columns
+	'fields'. */
+	for (i = 0; i < n_fields; i++) {
+		templ = prebuilt->mysql_template + n_requested_fields;
+		field = table->field[i];
+
+		if (UNIV_LIKELY(templ_type == ROW_MYSQL_REC_FIELDS)) {
+			/* Decide which columns we should fetch
+			and which we can skip. */
+			register const ibool	index_contains_field =
+				dict_index_contains_col_or_prefix(index, i);
+
+			if (!index_contains_field && prebuilt->read_just_key) {
+				/* If this is a 'key read', we do not need
+				columns that are not in the key */
+
+				goto skip_field;
+			}
+
+			if (index_contains_field && fetch_all_in_key) {
+				/* This field is needed in the query */
+
+				goto include_field;
+			}
+
+			if (bitmap_is_set(table->read_set, i) ||
+			    bitmap_is_set(table->write_set, i)) {
+				/* This field is needed in the query */
+
+				goto include_field;
+			}
+
+			if (fetch_primary_key_cols
+				&& dict_table_col_in_clustered_key(
+					index->table, i)) {
+				/* This field is needed in the query */
+
+				goto include_field;
+			}
+
+			/* This field is not needed in the query, skip it */
+
+			goto skip_field;
+		}
+include_field:
+		n_requested_fields++;
+
+		templ->col_no = i;
+
+		if (index == clust_index) {
+			templ->rec_field_no = dict_col_get_clust_pos(
+				&index->table->cols[i], index);
+		} else {
+			templ->rec_field_no = dict_index_get_nth_col_pos(
+								index, i);
+		}
+
+		if (templ->rec_field_no == ULINT_UNDEFINED) {
+			prebuilt->need_to_access_clustered = TRUE;
+		}
+
+		if (field->null_ptr) {
+			templ->mysql_null_byte_offset =
+				(ulint) ((char*) field->null_ptr
+					- (char*) table->record[0]);
+
+			templ->mysql_null_bit_mask = (ulint) field->null_bit;
+		} else {
+			templ->mysql_null_bit_mask = 0;
+		}
+
+		templ->mysql_col_offset = (ulint)
+					get_field_offset(table, field);
+
+		templ->mysql_col_len = (ulint) field->pack_length();
+		if (mysql_prefix_len < templ->mysql_col_offset
+				+ templ->mysql_col_len) {
+			mysql_prefix_len = templ->mysql_col_offset
+				+ templ->mysql_col_len;
+		}
+		templ->type = index->table->cols[i].mtype;
+		templ->mysql_type = (ulint)field->type();
+
+		if (templ->mysql_type == DATA_MYSQL_TRUE_VARCHAR) {
+			templ->mysql_length_bytes = (ulint)
+				(((Field_varstring*)field)->length_bytes);
+		}
+
+		templ->charset = dtype_get_charset_coll(
+			index->table->cols[i].prtype);
+		templ->mbminlen = index->table->cols[i].mbminlen;
+		templ->mbmaxlen = index->table->cols[i].mbmaxlen;
+		templ->is_unsigned = index->table->cols[i].prtype
+							& DATA_UNSIGNED;
+		if (templ->type == DATA_BLOB) {
+			prebuilt->templ_contains_blob = TRUE;
+		}
+skip_field:
+		;
+	}
+
+	prebuilt->n_template = n_requested_fields;
+	prebuilt->mysql_prefix_len = mysql_prefix_len;
+
+	if (index != clust_index && prebuilt->need_to_access_clustered) {
+		/* Change rec_field_no's to correspond to the clustered index
+		record */
+		for (i = 0; i < n_requested_fields; i++) {
+			templ = prebuilt->mysql_template + i;
+
+			templ->rec_field_no = dict_col_get_clust_pos(
+				&index->table->cols[templ->col_no],
+				clust_index);
+		}
+	}
+}
+
+/************************************************************************
+Get the upper limit of the MySQL integral and floating-point type. */
+UNIV_INTERN
+ulonglong
+ha_innobase::innobase_get_int_col_max_value(
+/*========================================*/
+	const Field*	field)
+{
+	ulonglong	max_value = 0;
+
+	switch(field->key_type()) {
+	/* TINY */
+	case HA_KEYTYPE_BINARY:
+		max_value = 0xFFULL;
+		break;
+	case HA_KEYTYPE_INT8:
+		max_value = 0x7FULL;
+		break;
+	/* SHORT */
+	case HA_KEYTYPE_USHORT_INT:
+		max_value = 0xFFFFULL;
+		break;
+	case HA_KEYTYPE_SHORT_INT:
+		max_value = 0x7FFFULL;
+		break;
+	/* MEDIUM */
+	case HA_KEYTYPE_UINT24:
+		max_value = 0xFFFFFFULL;
+		break;
+	case HA_KEYTYPE_INT24:
+		max_value = 0x7FFFFFULL;
+		break;
+	/* LONG */
+	case HA_KEYTYPE_ULONG_INT:
+		max_value = 0xFFFFFFFFULL;
+		break;
+	case HA_KEYTYPE_LONG_INT:
+		max_value = 0x7FFFFFFFULL;
+		break;
+	/* BIG */
+	case HA_KEYTYPE_ULONGLONG:
+		max_value = 0xFFFFFFFFFFFFFFFFULL;
+		break;
+	case HA_KEYTYPE_LONGLONG:
+		max_value = 0x7FFFFFFFFFFFFFFFULL;
+		break;
+	case HA_KEYTYPE_FLOAT:
+		/* We use the maximum as per IEEE754-2008 standard, 2^24 */
+		max_value = 0x1000000ULL;
+		break;
+	case HA_KEYTYPE_DOUBLE:
+		/* We use the maximum as per IEEE754-2008 standard, 2^53 */
+		max_value = 0x20000000000000ULL;
+		break;
+	default:
+		ut_error;
+	}
+
+	return(max_value);
+}
+
+/************************************************************************
+This special handling is really to overcome the limitations of MySQL's
+binlogging. We need to eliminate the non-determinism that will arise in
+INSERT ... SELECT type of statements, since MySQL binlog only stores the
+min value of the autoinc interval. Once that is fixed we can get rid of
+the special lock handling.*/
+UNIV_INTERN
+ulint
+ha_innobase::innobase_lock_autoinc(void)
+/*====================================*/
+					/* out: DB_SUCCESS if all OK else
+					error code */
+{
+	ulint		error = DB_SUCCESS;
+
+	switch (innobase_autoinc_lock_mode) {
+	case AUTOINC_NO_LOCKING:
+		/* Acquire only the AUTOINC mutex. */
+		dict_table_autoinc_lock(prebuilt->table);
+		break;
+
+	case AUTOINC_NEW_STYLE_LOCKING:
+		/* For simple (single/multi) row INSERTs, we fallback to the
+		old style only if another transaction has already acquired
+		the AUTOINC lock on behalf of a LOAD FILE or INSERT ... SELECT
+		etc. type of statement. */
+		if (thd_sql_command(user_thd) == SQLCOM_INSERT
+		    || thd_sql_command(user_thd) == SQLCOM_REPLACE) {
+			dict_table_t*	table = prebuilt->table;
+
+			/* Acquire the AUTOINC mutex. */
+			dict_table_autoinc_lock(table);
+
+			/* We need to check that another transaction isn't
+			already holding the AUTOINC lock on the table. */
+			if (table->n_waiting_or_granted_auto_inc_locks) {
+				/* Release the mutex to avoid deadlocks. */
+				dict_table_autoinc_unlock(table);
+			} else {
+				break;
+			}
+		}
+		/* Fall through to old style locking. */
+
+	case AUTOINC_OLD_STYLE_LOCKING:
+		error = row_lock_table_autoinc_for_mysql(prebuilt);
+
+		if (error == DB_SUCCESS) {
+
+			/* Acquire the AUTOINC mutex. */
+			dict_table_autoinc_lock(prebuilt->table);
+		}
+		break;
+
+	default:
+		ut_error;
+	}
+
+	return(ulong(error));
+}
+
+/************************************************************************
+Reset the autoinc value in the table.*/
+UNIV_INTERN
+ulint
+ha_innobase::innobase_reset_autoinc(
+/*================================*/
+					/* out: DB_SUCCESS if all went well
+					else error code */
+	ulonglong	autoinc)	/* in: value to store */
+{
+	ulint		error;
+
+	error = innobase_lock_autoinc();
+
+	if (error == DB_SUCCESS) {
+
+		dict_table_autoinc_initialize(prebuilt->table, autoinc);
+
+		dict_table_autoinc_unlock(prebuilt->table);
+	}
+
+	return(ulong(error));
+}
+
+/************************************************************************
+Store the autoinc value in the table. The autoinc value is only set if
+it's greater than the existing autoinc value in the table.*/
+UNIV_INTERN
+ulint
+ha_innobase::innobase_set_max_autoinc(
+/*==================================*/
+					/* out: DB_SUCCES if all went well
+					else error code */
+	ulonglong	auto_inc)	/* in: value to store */
+{
+	ulint		error;
+
+	error = innobase_lock_autoinc();
+
+	if (error == DB_SUCCESS) {
+
+		dict_table_autoinc_update_if_greater(prebuilt->table, auto_inc);
+
+		dict_table_autoinc_unlock(prebuilt->table);
+	}
+
+	return(ulong(error));
+}
+
+/************************************************************************
+Stores a row in an InnoDB database, to the table specified in this
+handle. */
+UNIV_INTERN
+int
+ha_innobase::write_row(
+/*===================*/
+			/* out: error code */
+	uchar*	record)	/* in: a row in MySQL format */
+{
+	ulint		error = 0;
+        int             error_result= 0;
+	ibool		auto_inc_used= FALSE;
+	ulint		sql_command;
+	trx_t*		trx = thd_to_trx(user_thd);
+
+	DBUG_ENTER("ha_innobase::write_row");
+
+	if (prebuilt->trx != trx) {
+	  sql_print_error("The transaction object for the table handle is at "
+			  "%p, but for the current thread it is at %p",
+			  (const void*) prebuilt->trx, (const void*) trx);
+
+		fputs("InnoDB: Dump of 200 bytes around prebuilt: ", stderr);
+		ut_print_buf(stderr, ((const byte*)prebuilt) - 100, 200);
+		fputs("\n"
+			"InnoDB: Dump of 200 bytes around ha_data: ",
+			stderr);
+		ut_print_buf(stderr, ((const byte*) trx) - 100, 200);
+		putc('\n', stderr);
+		ut_error;
+	}
+
+	ha_statistic_increment(&SSV::ha_write_count);
+
+	if (table->timestamp_field_type & TIMESTAMP_AUTO_SET_ON_INSERT)
+		table->timestamp_field->set_time();
+
+	sql_command = thd_sql_command(user_thd);
+
+	if ((sql_command == SQLCOM_ALTER_TABLE
+	     || sql_command == SQLCOM_OPTIMIZE
+	     || sql_command == SQLCOM_CREATE_INDEX
+	     || sql_command == SQLCOM_DROP_INDEX)
+	    && num_write_row >= 10000) {
+		/* ALTER TABLE is COMMITted at every 10000 copied rows.
+		The IX table lock for the original table has to be re-issued.
+		As this method will be called on a temporary table where the
+		contents of the original table is being copied to, it is
+		a bit tricky to determine the source table.  The cursor
+		position in the source table need not be adjusted after the
+		intermediate COMMIT, since writes by other transactions are
+		being blocked by a MySQL table lock TL_WRITE_ALLOW_READ. */
+
+		dict_table_t*	src_table;
+		enum lock_mode	mode;
+
+		num_write_row = 0;
+
+		/* Commit the transaction.  This will release the table
+		locks, so they have to be acquired again. */
+
+		/* Altering an InnoDB table */
+		/* Get the source table. */
+		src_table = lock_get_src_table(
+				prebuilt->trx, prebuilt->table, &mode);
+		if (!src_table) {
+no_commit:
+			/* Unknown situation: do not commit */
+			/*
+			ut_print_timestamp(stderr);
+			fprintf(stderr,
+				"  InnoDB: ALTER TABLE is holding lock"
+				" on %lu tables!\n",
+				prebuilt->trx->mysql_n_tables_locked);
+			*/
+			;
+		} else if (src_table == prebuilt->table) {
+			/* Source table is not in InnoDB format:
+			no need to re-acquire locks on it. */
+
+			/* Altering to InnoDB format */
+			innobase_commit(ht, user_thd, 1);
+			/* Note that this transaction is still active. */
+			prebuilt->trx->active_trans = 1;
+			/* We will need an IX lock on the destination table. */
+			prebuilt->sql_stat_start = TRUE;
+		} else {
+			/* Ensure that there are no other table locks than
+			LOCK_IX and LOCK_AUTO_INC on the destination table. */
+
+			if (!lock_is_table_exclusive(prebuilt->table,
+							prebuilt->trx)) {
+				goto no_commit;
+			}
+
+			/* Commit the transaction.  This will release the table
+			locks, so they have to be acquired again. */
+			innobase_commit(ht, user_thd, 1);
+			/* Note that this transaction is still active. */
+			prebuilt->trx->active_trans = 1;
+			/* Re-acquire the table lock on the source table. */
+			row_lock_table_for_mysql(prebuilt, src_table, mode);
+			/* We will need an IX lock on the destination table. */
+			prebuilt->sql_stat_start = TRUE;
+		}
+	}
+
+	num_write_row++;
+
+	/* This is the case where the table has an auto-increment column */
+	if (table->next_number_field && record == table->record[0]) {
+
+		/* Reset the error code before calling
+		innobase_get_auto_increment(). */
+		prebuilt->autoinc_error = DB_SUCCESS;
+
+		if ((error = update_auto_increment())) {
+
+			/* We don't want to mask autoinc overflow errors. */
+			if (prebuilt->autoinc_error != DB_SUCCESS) {
+				error = (int) prebuilt->autoinc_error;
+
+				goto report_error;
+			}
+
+			/* MySQL errors are passed straight back. */
+			error_result = (int) error;
+			goto func_exit;
+		}
+
+		auto_inc_used = TRUE;
+	}
+
+	if (prebuilt->mysql_template == NULL
+	    || prebuilt->template_type != ROW_MYSQL_WHOLE_ROW) {
+
+		/* Build the template used in converting quickly between
+		the two database formats */
+
+		build_template(prebuilt, NULL, table, ROW_MYSQL_WHOLE_ROW);
+	}
+
+	innodb_srv_conc_enter_innodb(prebuilt->trx);
+
+	error = row_insert_for_mysql((byte*) record, prebuilt);
+
+	/* Handle duplicate key errors */
+	if (auto_inc_used) {
+		ulint		err;
+		ulonglong	auto_inc;
+		ulonglong	col_max_value;
+
+		/* Note the number of rows processed for this statement, used
+		by get_auto_increment() to determine the number of AUTO-INC
+		values to reserve. This is only useful for a mult-value INSERT
+		and is a statement level counter.*/
+		if (trx->n_autoinc_rows > 0) {
+			--trx->n_autoinc_rows;
+		}
+
+		/* We need the upper limit of the col type to check for
+		whether we update the table autoinc counter or not. */
+		col_max_value = innobase_get_int_col_max_value(
+			table->next_number_field);
+
+		/* Get the value that MySQL attempted to store in the table.*/
+		auto_inc = table->next_number_field->val_int();
+
+		switch (error) {
+		case DB_DUPLICATE_KEY:
+
+			/* A REPLACE command and LOAD DATA INFILE REPLACE
+			handle a duplicate key error themselves, but we
+			must update the autoinc counter if we are performing
+			those statements. */
+
+			switch (sql_command) {
+			case SQLCOM_LOAD:
+				if ((trx->duplicates
+				    & (TRX_DUP_IGNORE | TRX_DUP_REPLACE))) {
+
+					goto set_max_autoinc;
+				}
+				break;
+
+			case SQLCOM_REPLACE:
+			case SQLCOM_INSERT_SELECT:
+			case SQLCOM_REPLACE_SELECT:
+				goto set_max_autoinc;
+
+			default:
+				break;
+			}
+
+			break;
+
+		case DB_SUCCESS:
+			/* If the actual value inserted is greater than
+			the upper limit of the interval, then we try and
+			update the table upper limit. Note: last_value
+			will be 0 if get_auto_increment() was not called.*/
+
+			if (auto_inc <= col_max_value
+			    && auto_inc >= prebuilt->autoinc_last_value) {
+set_max_autoinc:
+				ut_a(prebuilt->autoinc_increment > 0);
+
+				ulonglong	need;
+				ulonglong	offset;
+
+				offset = prebuilt->autoinc_offset;
+				need = prebuilt->autoinc_increment;
+
+				auto_inc = innobase_next_autoinc(
+					auto_inc, need, offset, col_max_value);
+
+				err = innobase_set_max_autoinc(auto_inc);
+
+				if (err != DB_SUCCESS) {
+					error = err;
+				}
+			}
+			break;
+		}
+	}
+
+	innodb_srv_conc_exit_innodb(prebuilt->trx);
+
+report_error:
+	error_result = convert_error_code_to_mysql((int) error,
+						   prebuilt->table->flags,
+						   user_thd);
+
+func_exit:
+	innobase_active_small();
+
+	DBUG_RETURN(error_result);
+}
+
+/**************************************************************************
+Checks which fields have changed in a row and stores information
+of them to an update vector. */
+static
+int
+calc_row_difference(
+/*================*/
+					/* out: error number or 0 */
+	upd_t*		uvect,		/* in/out: update vector */
+	uchar*		old_row,	/* in: old row in MySQL format */
+	uchar*		new_row,	/* in: new row in MySQL format */
+	struct st_table* table,		/* in: table in MySQL data
+					dictionary */
+	uchar*		upd_buff,	/* in: buffer to use */
+	ulint		buff_len,	/* in: buffer length */
+	row_prebuilt_t*	prebuilt,	/* in: InnoDB prebuilt struct */
+	THD*		thd)		/* in: user thread */
+{
+	uchar*		original_upd_buff = upd_buff;
+	Field*		field;
+	enum_field_types field_mysql_type;
+	uint		n_fields;
+	ulint		o_len;
+	ulint		n_len;
+	ulint		col_pack_len;
+	const byte*	new_mysql_row_col;
+	const byte*	o_ptr;
+	const byte*	n_ptr;
+	byte*		buf;
+	upd_field_t*	ufield;
+	ulint		col_type;
+	ulint		n_changed = 0;
+	dfield_t	dfield;
+	dict_index_t*	clust_index;
+	uint		i;
+
+	n_fields = table->s->fields;
+	clust_index = dict_table_get_first_index(prebuilt->table);
+
+	/* We use upd_buff to convert changed fields */
+	buf = (byte*) upd_buff;
+
+	for (i = 0; i < n_fields; i++) {
+		field = table->field[i];
+
+		o_ptr = (const byte*) old_row + get_field_offset(table, field);
+		n_ptr = (const byte*) new_row + get_field_offset(table, field);
+
+		/* Use new_mysql_row_col and col_pack_len save the values */
+
+		new_mysql_row_col = n_ptr;
+		col_pack_len = field->pack_length();
+
+		o_len = col_pack_len;
+		n_len = col_pack_len;
+
+		/* We use o_ptr and n_ptr to dig up the actual data for
+		comparison. */
+
+		field_mysql_type = field->type();
+
+		col_type = prebuilt->table->cols[i].mtype;
+
+		switch (col_type) {
+
+		case DATA_BLOB:
+			o_ptr = row_mysql_read_blob_ref(&o_len, o_ptr, o_len);
+			n_ptr = row_mysql_read_blob_ref(&n_len, n_ptr, n_len);
+
+			break;
+
+		case DATA_VARCHAR:
+		case DATA_BINARY:
+		case DATA_VARMYSQL:
+			if (field_mysql_type == MYSQL_TYPE_VARCHAR) {
+				/* This is a >= 5.0.3 type true VARCHAR where
+				the real payload data length is stored in
+				1 or 2 bytes */
+
+				o_ptr = row_mysql_read_true_varchar(
+					&o_len, o_ptr,
+					(ulint)
+					(((Field_varstring*)field)->length_bytes));
+
+				n_ptr = row_mysql_read_true_varchar(
+					&n_len, n_ptr,
+					(ulint)
+					(((Field_varstring*)field)->length_bytes));
+			}
+
+			break;
+		default:
+			;
+		}
+
+		if (field->null_ptr) {
+			if (field_in_record_is_null(table, field,
+							(char*) old_row)) {
+				o_len = UNIV_SQL_NULL;
+			}
+
+			if (field_in_record_is_null(table, field,
+							(char*) new_row)) {
+				n_len = UNIV_SQL_NULL;
+			}
+		}
+
+		if (o_len != n_len || (o_len != UNIV_SQL_NULL &&
+					0 != memcmp(o_ptr, n_ptr, o_len))) {
+			/* The field has changed */
+
+			ufield = uvect->fields + n_changed;
+
+			/* Let us use a dummy dfield to make the conversion
+			from the MySQL column format to the InnoDB format */
+
+			dict_col_copy_type(prebuilt->table->cols + i,
+					   dfield_get_type(&dfield));
+
+			if (n_len != UNIV_SQL_NULL) {
+				buf = row_mysql_store_col_in_innobase_format(
+					&dfield,
+					(byte*)buf,
+					TRUE,
+					new_mysql_row_col,
+					col_pack_len,
+					dict_table_is_comp(prebuilt->table));
+				dfield_copy_data(&ufield->new_val, &dfield);
+			} else {
+				dfield_set_null(&ufield->new_val);
+			}
+
+			ufield->exp = NULL;
+			ufield->orig_len = 0;
+			ufield->field_no = dict_col_get_clust_pos(
+				&prebuilt->table->cols[i], clust_index);
+			n_changed++;
+		}
+	}
+
+	uvect->n_fields = n_changed;
+	uvect->info_bits = 0;
+
+	ut_a(buf <= (byte*)original_upd_buff + buff_len);
+
+	return(0);
+}
+
+/**************************************************************************
+Updates a row given as a parameter to a new value. Note that we are given
+whole rows, not just the fields which are updated: this incurs some
+overhead for CPU when we check which fields are actually updated.
+TODO: currently InnoDB does not prevent the 'Halloween problem':
+in a searched update a single row can get updated several times
+if its index columns are updated! */
+UNIV_INTERN
+int
+ha_innobase::update_row(
+/*====================*/
+					/* out: error number or 0 */
+	const uchar*	old_row,	/* in: old row in MySQL format */
+	uchar*		new_row)	/* in: new row in MySQL format */
+{
+	upd_t*		uvect;
+	int		error = 0;
+	trx_t*		trx = thd_to_trx(user_thd);
+
+	DBUG_ENTER("ha_innobase::update_row");
+
+	ut_a(prebuilt->trx == trx);
+
+	ha_statistic_increment(&SSV::ha_update_count);
+
+	if (table->timestamp_field_type & TIMESTAMP_AUTO_SET_ON_UPDATE)
+		table->timestamp_field->set_time();
+
+	if (prebuilt->upd_node) {
+		uvect = prebuilt->upd_node->update;
+	} else {
+		uvect = row_get_prebuilt_update_vector(prebuilt);
+	}
+
+	/* Build an update vector from the modified fields in the rows
+	(uses upd_buff of the handle) */
+
+	calc_row_difference(uvect, (uchar*) old_row, new_row, table,
+			upd_buff, (ulint)upd_and_key_val_buff_len,
+			prebuilt, user_thd);
+
+	/* This is not a delete */
+	prebuilt->upd_node->is_delete = FALSE;
+
+	ut_a(prebuilt->template_type == ROW_MYSQL_WHOLE_ROW);
+
+	innodb_srv_conc_enter_innodb(trx);
+
+	error = row_update_for_mysql((byte*) old_row, prebuilt);
+
+	/* We need to do some special AUTOINC handling for the following case:
+
+	INSERT INTO t (c1,c2) VALUES(x,y) ON DUPLICATE KEY UPDATE ...
+
+	We need to use the AUTOINC counter that was actually used by
+	MySQL in the UPDATE statement, which can be different from the
+	value used in the INSERT statement.*/
+
+	if (error == DB_SUCCESS
+	    && table->next_number_field
+	    && new_row == table->record[0]
+	    && thd_sql_command(user_thd) == SQLCOM_INSERT
+	    && (trx->duplicates & (TRX_DUP_IGNORE | TRX_DUP_REPLACE))
+		== TRX_DUP_IGNORE)  {
+
+		ulonglong	auto_inc;
+		ulonglong	col_max_value;
+
+		auto_inc = table->next_number_field->val_int();
+
+		/* We need the upper limit of the col type to check for
+		whether we update the table autoinc counter or not. */
+		col_max_value = innobase_get_int_col_max_value(
+			table->next_number_field);
+
+		if (auto_inc <= col_max_value && auto_inc != 0) {
+
+			ulonglong	need;
+			ulonglong	offset;
+
+			offset = prebuilt->autoinc_offset;
+			need = prebuilt->autoinc_increment;
+
+			auto_inc = innobase_next_autoinc(
+				auto_inc, need, offset, col_max_value);
+
+			error = innobase_set_max_autoinc(auto_inc);
+		}
+	}
+
+	innodb_srv_conc_exit_innodb(trx);
+
+	error = convert_error_code_to_mysql(error,
+					    prebuilt->table->flags, user_thd);
+
+	if (error == 0 /* success */
+	    && uvect->n_fields == 0 /* no columns were updated */) {
+
+		/* This is the same as success, but instructs
+		MySQL that the row is not really updated and it
+		should not increase the count of updated rows.
+		This is fix for http://bugs.mysql.com/29157 */
+		error = HA_ERR_RECORD_IS_THE_SAME;
+	}
+
+	/* Tell InnoDB server that there might be work for
+	utility threads: */
+
+	innobase_active_small();
+
+	DBUG_RETURN(error);
+}
+
+/**************************************************************************
+Deletes a row given as the parameter. */
+UNIV_INTERN
+int
+ha_innobase::delete_row(
+/*====================*/
+				/* out: error number or 0 */
+	const uchar*	record)	/* in: a row in MySQL format */
+{
+	int		error = 0;
+	trx_t*		trx = thd_to_trx(user_thd);
+
+	DBUG_ENTER("ha_innobase::delete_row");
+
+	ut_a(prebuilt->trx == trx);
+
+	ha_statistic_increment(&SSV::ha_delete_count);
+
+	if (!prebuilt->upd_node) {
+		row_get_prebuilt_update_vector(prebuilt);
+	}
+
+	/* This is a delete */
+
+	prebuilt->upd_node->is_delete = TRUE;
+
+	innodb_srv_conc_enter_innodb(trx);
+
+	error = row_update_for_mysql((byte*) record, prebuilt);
+
+	innodb_srv_conc_exit_innodb(trx);
+
+	error = convert_error_code_to_mysql(
+		error, prebuilt->table->flags, user_thd);
+
+	/* Tell the InnoDB server that there might be work for
+	utility threads: */
+
+	innobase_active_small();
+
+	DBUG_RETURN(error);
+}
+
+/**************************************************************************
+Removes a new lock set on a row, if it was not read optimistically. This can
+be called after a row has been read in the processing of an UPDATE or a DELETE
+query, if the option innodb_locks_unsafe_for_binlog is set. */
+UNIV_INTERN
+void
+ha_innobase::unlock_row(void)
+/*=========================*/
+{
+	DBUG_ENTER("ha_innobase::unlock_row");
+
+	/* Consistent read does not take any locks, thus there is
+	nothing to unlock. */
+
+	if (prebuilt->select_lock_type == LOCK_NONE) {
+		DBUG_VOID_RETURN;
+	}
+
+	switch (prebuilt->row_read_type) {
+	case ROW_READ_WITH_LOCKS:
+		if (!srv_locks_unsafe_for_binlog
+		    && prebuilt->trx->isolation_level
+		    != TRX_ISO_READ_COMMITTED) {
+			break;
+		}
+		/* fall through */
+	case ROW_READ_TRY_SEMI_CONSISTENT:
+		row_unlock_for_mysql(prebuilt, FALSE);
+		break;
+	case ROW_READ_DID_SEMI_CONSISTENT:
+		prebuilt->row_read_type = ROW_READ_TRY_SEMI_CONSISTENT;
+		break;
+	}
+
+	DBUG_VOID_RETURN;
+}
+
+/* See handler.h and row0mysql.h for docs on this function. */
+UNIV_INTERN
+bool
+ha_innobase::was_semi_consistent_read(void)
+/*=======================================*/
+{
+	return(prebuilt->row_read_type == ROW_READ_DID_SEMI_CONSISTENT);
+}
+
+/* See handler.h and row0mysql.h for docs on this function. */
+UNIV_INTERN
+void
+ha_innobase::try_semi_consistent_read(bool yes)
+/*===========================================*/
+{
+	ut_a(prebuilt->trx == thd_to_trx(ha_thd()));
+
+	/* Row read type is set to semi consistent read if this was
+	requested by the MySQL and either innodb_locks_unsafe_for_binlog
+	option is used or this session is using READ COMMITTED isolation
+	level. */
+
+	if (yes
+	    && (srv_locks_unsafe_for_binlog
+		|| prebuilt->trx->isolation_level == TRX_ISO_READ_COMMITTED)) {
+		prebuilt->row_read_type = ROW_READ_TRY_SEMI_CONSISTENT;
+	} else {
+		prebuilt->row_read_type = ROW_READ_WITH_LOCKS;
+	}
+}
+
+#ifdef ROW_MERGE_IS_INDEX_USABLE
+/**********************************************************************
+Check if an index can be used by the optimizer. */
+UNIV_INTERN
+bool
+ha_innobase::is_index_available(
+/*============================*/
+					/* out: true if available else false*/
+	uint		keynr)		/* in: index number to check */
+{
+	DBUG_ENTER("ha_innobase::is_index_available");
+
+	if (table && keynr != MAX_KEY && table->s->keys > 0) {
+		const dict_index_t*	index;
+		const KEY*		key = table->key_info + keynr;
+
+		ut_ad(user_thd == ha_thd());
+		ut_a(prebuilt->trx == thd_to_trx(user_thd));
+
+		index = dict_table_get_index_on_name(
+			prebuilt->table, key->name);
+
+		if (!row_merge_is_index_usable(prebuilt->trx, index)) {
+
+			DBUG_RETURN(false);
+		}
+	}
+
+	DBUG_RETURN(true);
+}
+#endif /* ROW_MERGE_IS_INDEX_USABLE */
+
+/**********************************************************************
+Initializes a handle to use an index. */
+UNIV_INTERN
+int
+ha_innobase::index_init(
+/*====================*/
+			/* out: 0 or error number */
+	uint	keynr,	/* in: key (index) number */
+	bool sorted)	/* in: 1 if result MUST be sorted according to index */
+{
+	DBUG_ENTER("index_init");
+
+	DBUG_RETURN(change_active_index(keynr));
+}
+
+/**********************************************************************
+Currently does nothing. */
+UNIV_INTERN
+int
+ha_innobase::index_end(void)
+/*========================*/
+{
+	int	error	= 0;
+	DBUG_ENTER("index_end");
+	active_index=MAX_KEY;
+	DBUG_RETURN(error);
+}
+
+/*************************************************************************
+Converts a search mode flag understood by MySQL to a flag understood
+by InnoDB. */
+static inline
+ulint
+convert_search_mode_to_innobase(
+/*============================*/
+	enum ha_rkey_function	find_flag)
+{
+	switch (find_flag) {
+	case HA_READ_KEY_EXACT:
+		/* this does not require the index to be UNIQUE */
+		return(PAGE_CUR_GE);
+	case HA_READ_KEY_OR_NEXT:
+		return(PAGE_CUR_GE);
+	case HA_READ_KEY_OR_PREV:
+		return(PAGE_CUR_LE);
+	case HA_READ_AFTER_KEY:	
+		return(PAGE_CUR_G);
+	case HA_READ_BEFORE_KEY:
+		return(PAGE_CUR_L);
+	case HA_READ_PREFIX:
+		return(PAGE_CUR_GE);
+	case HA_READ_PREFIX_LAST:
+		return(PAGE_CUR_LE);
+	case HA_READ_PREFIX_LAST_OR_PREV:
+		return(PAGE_CUR_LE);
+		/* In MySQL-4.0 HA_READ_PREFIX and HA_READ_PREFIX_LAST always
+		pass a complete-field prefix of a key value as the search
+		tuple. I.e., it is not allowed that the last field would
+		just contain n first bytes of the full field value.
+		MySQL uses a 'padding' trick to convert LIKE 'abc%'
+		type queries so that it can use as a search tuple
+		a complete-field-prefix of a key value. Thus, the InnoDB
+		search mode PAGE_CUR_LE_OR_EXTENDS is never used.
+		TODO: when/if MySQL starts to use also partial-field
+		prefixes, we have to deal with stripping of spaces
+		and comparison of non-latin1 char type fields in
+		innobase_mysql_cmp() to get PAGE_CUR_LE_OR_EXTENDS to
+		work correctly. */
+	case HA_READ_MBR_CONTAIN:
+	case HA_READ_MBR_INTERSECT:
+	case HA_READ_MBR_WITHIN:
+	case HA_READ_MBR_DISJOINT:
+	case HA_READ_MBR_EQUAL:
+		my_error(ER_TABLE_CANT_HANDLE_SPKEYS, MYF(0));
+		return(PAGE_CUR_UNSUPP);
+	/* do not use "default:" in order to produce a gcc warning:
+	enumeration value '...' not handled in switch
+	(if -Wswitch or -Wall is used) */
+	}
+
+	my_error(ER_CHECK_NOT_IMPLEMENTED, MYF(0), "this functionality");
+
+	return(PAGE_CUR_UNSUPP);
+}
+
+/*
+   BACKGROUND INFO: HOW A SELECT SQL QUERY IS EXECUTED
+   ---------------------------------------------------
+The following does not cover all the details, but explains how we determine
+the start of a new SQL statement, and what is associated with it.
+
+For each table in the database the MySQL interpreter may have several
+table handle instances in use, also in a single SQL query. For each table
+handle instance there is an InnoDB  'prebuilt' struct which contains most
+of the InnoDB data associated with this table handle instance.
+
+  A) if the user has not explicitly set any MySQL table level locks:
+
+  1) MySQL calls ::external_lock to set an 'intention' table level lock on
+the table of the handle instance. There we set
+prebuilt->sql_stat_start = TRUE. The flag sql_stat_start should be set
+true if we are taking this table handle instance to use in a new SQL
+statement issued by the user. We also increment trx->n_mysql_tables_in_use.
+
+  2) If prebuilt->sql_stat_start == TRUE we 'pre-compile' the MySQL search
+instructions to prebuilt->template of the table handle instance in
+::index_read. The template is used to save CPU time in large joins.
+
+  3) In row_search_for_mysql, if prebuilt->sql_stat_start is true, we
+allocate a new consistent read view for the trx if it does not yet have one,
+or in the case of a locking read, set an InnoDB 'intention' table level
+lock on the table.
+
+  4) We do the SELECT. MySQL may repeatedly call ::index_read for the
+same table handle instance, if it is a join.
+
+  5) When the SELECT ends, MySQL removes its intention table level locks
+in ::external_lock. When trx->n_mysql_tables_in_use drops to zero,
+ (a) we execute a COMMIT there if the autocommit is on,
+ (b) we also release possible 'SQL statement level resources' InnoDB may
+have for this SQL statement. The MySQL interpreter does NOT execute
+autocommit for pure read transactions, though it should. That is why the
+table handler in that case has to execute the COMMIT in ::external_lock.
+
+  B) If the user has explicitly set MySQL table level locks, then MySQL
+does NOT call ::external_lock at the start of the statement. To determine
+when we are at the start of a new SQL statement we at the start of
+::index_read also compare the query id to the latest query id where the
+table handle instance was used. If it has changed, we know we are at the
+start of a new SQL statement. Since the query id can theoretically
+overwrap, we use this test only as a secondary way of determining the
+start of a new SQL statement. */
+
+
+/**************************************************************************
+Positions an index cursor to the index specified in the handle. Fetches the
+row if any. */
+UNIV_INTERN
+int
+ha_innobase::index_read(
+/*====================*/
+					/* out: 0, HA_ERR_KEY_NOT_FOUND,
+					or error number */
+	uchar*		buf,		/* in/out: buffer for the returned
+					row */
+	const uchar*	key_ptr,	/* in: key value; if this is NULL
+					we position the cursor at the
+					start or end of index; this can
+					also contain an InnoDB row id, in
+					which case key_len is the InnoDB
+					row id length; the key value can
+					also be a prefix of a full key value,
+					and the last column can be a prefix
+					of a full column */
+	uint			key_len,/* in: key value length */
+	enum ha_rkey_function find_flag)/* in: search flags from my_base.h */
+{
+	ulint		mode;
+	dict_index_t*	index;
+	ulint		match_mode	= 0;
+	int		error;
+	ulint		ret;
+
+	DBUG_ENTER("index_read");
+
+	ut_a(prebuilt->trx == thd_to_trx(user_thd));
+
+	ha_statistic_increment(&SSV::ha_read_key_count);
+
+	index = prebuilt->index;
+
+	/* Note that if the index for which the search template is built is not
+	necessarily prebuilt->index, but can also be the clustered index */
+
+	if (prebuilt->sql_stat_start) {
+		build_template(prebuilt, user_thd, table, ROW_MYSQL_REC_FIELDS);
+	}
+
+	if (key_ptr) {
+		/* Convert the search key value to InnoDB format into
+		prebuilt->search_tuple */
+
+		row_sel_convert_mysql_key_to_innobase(
+			prebuilt->search_tuple,
+			(byte*) key_val_buff,
+			(ulint)upd_and_key_val_buff_len,
+			index,
+			(byte*) key_ptr,
+			(ulint) key_len,
+			prebuilt->trx);
+	} else {
+		/* We position the cursor to the last or the first entry
+		in the index */
+
+		dtuple_set_n_fields(prebuilt->search_tuple, 0);
+	}
+
+	mode = convert_search_mode_to_innobase(find_flag);
+
+	match_mode = 0;
+
+	if (find_flag == HA_READ_KEY_EXACT) {
+
+		match_mode = ROW_SEL_EXACT;
+
+	} else if (find_flag == HA_READ_PREFIX
+		   || find_flag == HA_READ_PREFIX_LAST) {
+
+		match_mode = ROW_SEL_EXACT_PREFIX;
+	}
+
+	last_match_mode = (uint) match_mode;
+
+	if (mode != PAGE_CUR_UNSUPP) {
+
+		innodb_srv_conc_enter_innodb(prebuilt->trx);
+
+		ret = row_search_for_mysql((byte*) buf, mode, prebuilt,
+					   match_mode, 0);
+
+		innodb_srv_conc_exit_innodb(prebuilt->trx);
+	} else {
+
+		ret = DB_UNSUPPORTED;
+	}
+
+	switch (ret) {
+	case DB_SUCCESS:
+		error = 0;
+		table->status = 0;
+		break;
+	case DB_RECORD_NOT_FOUND:
+		error = HA_ERR_KEY_NOT_FOUND;
+		table->status = STATUS_NOT_FOUND;
+		break;
+	case DB_END_OF_INDEX:
+		error = HA_ERR_KEY_NOT_FOUND;
+		table->status = STATUS_NOT_FOUND;
+		break;
+	default:
+		error = convert_error_code_to_mysql((int) ret,
+						    prebuilt->table->flags,
+						    user_thd);
+		table->status = STATUS_NOT_FOUND;
+		break;
+	}
+
+	DBUG_RETURN(error);
+}
+
+/***********************************************************************
+The following functions works like index_read, but it find the last
+row with the current key value or prefix. */
+UNIV_INTERN
+int
+ha_innobase::index_read_last(
+/*=========================*/
+				/* out: 0, HA_ERR_KEY_NOT_FOUND, or an
+				error code */
+	uchar*		buf,	/* out: fetched row */
+	const uchar*	key_ptr,/* in: key value, or a prefix of a full
+				key value */
+	uint		key_len)/* in: length of the key val or prefix
+				in bytes */
+{
+	return(index_read(buf, key_ptr, key_len, HA_READ_PREFIX_LAST));
+}
+
+/************************************************************************
+Get the index for a handle. Does not change active index.*/
+UNIV_INTERN
+dict_index_t*
+ha_innobase::innobase_get_index(
+/*============================*/
+				/* out: NULL or index instance. */
+	uint		keynr)	/* in: use this index; MAX_KEY means always
+				clustered index, even if it was internally
+				generated by InnoDB */
+{
+	KEY*		key = 0;
+	dict_index_t*	index = 0;
+
+	DBUG_ENTER("innobase_get_index");
+	ha_statistic_increment(&SSV::ha_read_key_count);
+
+	ut_ad(user_thd == ha_thd());
+	ut_a(prebuilt->trx == thd_to_trx(user_thd));
+
+	if (keynr != MAX_KEY && table->s->keys > 0) {
+		key = table->key_info + keynr;
+
+		index = dict_table_get_index_on_name(prebuilt->table,
+						     key->name);
+	} else {
+		index = dict_table_get_first_index(prebuilt->table);
+	}
+
+	if (!index) {
+		sql_print_error(
+			"Innodb could not find key n:o %u with name %s "
+			"from dict cache for table %s",
+			keynr, key ? key->name : "NULL",
+			prebuilt->table->name);
+	}
+
+	DBUG_RETURN(index);
+}
+
+/************************************************************************
+Changes the active index of a handle. */
+UNIV_INTERN
+int
+ha_innobase::change_active_index(
+/*=============================*/
+			/* out: 0 or error code */
+	uint	keynr)	/* in: use this index; MAX_KEY means always clustered
+			index, even if it was internally generated by
+			InnoDB */
+{
+	DBUG_ENTER("change_active_index");
+
+	ut_ad(user_thd == ha_thd());
+	ut_a(prebuilt->trx == thd_to_trx(user_thd));
+
+	active_index = keynr;
+
+	prebuilt->index = innobase_get_index(keynr);
+
+	if (UNIV_UNLIKELY(!prebuilt->index)) {
+		sql_print_warning("InnoDB: change_active_index(%u) failed",
+				  keynr);
+		DBUG_RETURN(1);
+	}
+
+	ut_a(prebuilt->search_tuple != 0);
+
+	dtuple_set_n_fields(prebuilt->search_tuple, prebuilt->index->n_fields);
+
+	dict_index_copy_types(prebuilt->search_tuple, prebuilt->index,
+			prebuilt->index->n_fields);
+
+	/* MySQL changes the active index for a handle also during some
+	queries, for example SELECT MAX(a), SUM(a) first retrieves the MAX()
+	and then calculates the sum. Previously we played safe and used
+	the flag ROW_MYSQL_WHOLE_ROW below, but that caused unnecessary
+	copying. Starting from MySQL-4.1 we use a more efficient flag here. */
+
+	build_template(prebuilt, user_thd, table, ROW_MYSQL_REC_FIELDS);
+
+	DBUG_RETURN(0);
+}
+
+/**************************************************************************
+Positions an index cursor to the index specified in keynr. Fetches the
+row if any. */
+/* ??? This is only used to read whole keys ??? */
+UNIV_INTERN
+int
+ha_innobase::index_read_idx(
+/*========================*/
+					/* out: error number or 0 */
+	uchar*		buf,		/* in/out: buffer for the returned
+					row */
+	uint		keynr,		/* in: use this index */
+	const uchar*	key,		/* in: key value; if this is NULL
+					we position the cursor at the
+					start or end of index */
+	uint		key_len,	/* in: key value length */
+	enum ha_rkey_function find_flag)/* in: search flags from my_base.h */
+{
+	if (change_active_index(keynr)) {
+
+		return(1);
+	}
+
+	return(index_read(buf, key, key_len, find_flag));
+}
+
+/***************************************************************************
+Reads the next or previous row from a cursor, which must have previously been
+positioned using index_read. */
+UNIV_INTERN
+int
+ha_innobase::general_fetch(
+/*=======================*/
+				/* out: 0, HA_ERR_END_OF_FILE, or error
+				number */
+	uchar*	buf,		/* in/out: buffer for next row in MySQL
+				format */
+	uint	direction,	/* in: ROW_SEL_NEXT or ROW_SEL_PREV */
+	uint	match_mode)	/* in: 0, ROW_SEL_EXACT, or
+				ROW_SEL_EXACT_PREFIX */
+{
+	ulint		ret;
+	int		error	= 0;
+
+	DBUG_ENTER("general_fetch");
+
+	ut_a(prebuilt->trx == thd_to_trx(user_thd));
+
+	innodb_srv_conc_enter_innodb(prebuilt->trx);
+
+	ret = row_search_for_mysql(
+		(byte*)buf, 0, prebuilt, match_mode, direction);
+
+	innodb_srv_conc_exit_innodb(prebuilt->trx);
+
+	switch (ret) {
+	case DB_SUCCESS:
+		error = 0;
+		table->status = 0;
+		break;
+	case DB_RECORD_NOT_FOUND:
+		error = HA_ERR_END_OF_FILE;
+		table->status = STATUS_NOT_FOUND;
+		break;
+	case DB_END_OF_INDEX:
+		error = HA_ERR_END_OF_FILE;
+		table->status = STATUS_NOT_FOUND;
+		break;
+	default:
+		error = convert_error_code_to_mysql(
+			(int) ret, prebuilt->table->flags, user_thd);
+		table->status = STATUS_NOT_FOUND;
+		break;
+	}
+
+	DBUG_RETURN(error);
+}
+
+/***************************************************************************
+Reads the next row from a cursor, which must have previously been
+positioned using index_read. */
+UNIV_INTERN
+int
+ha_innobase::index_next(
+/*====================*/
+				/* out: 0, HA_ERR_END_OF_FILE, or error
+				number */
+	uchar*		buf)	/* in/out: buffer for next row in MySQL
+				format */
+{
+	ha_statistic_increment(&SSV::ha_read_next_count);
+
+	return(general_fetch(buf, ROW_SEL_NEXT, 0));
+}
+
+/***********************************************************************
+Reads the next row matching to the key value given as the parameter. */
+UNIV_INTERN
+int
+ha_innobase::index_next_same(
+/*=========================*/
+				/* out: 0, HA_ERR_END_OF_FILE, or error
+				number */
+	uchar*		buf,	/* in/out: buffer for the row */
+	const uchar*	key,	/* in: key value */
+	uint		keylen)	/* in: key value length */
+{
+	ha_statistic_increment(&SSV::ha_read_next_count);
+
+	return(general_fetch(buf, ROW_SEL_NEXT, last_match_mode));
+}
+
+/***************************************************************************
+Reads the previous row from a cursor, which must have previously been
+positioned using index_read. */
+UNIV_INTERN
+int
+ha_innobase::index_prev(
+/*====================*/
+			/* out: 0, HA_ERR_END_OF_FILE, or error number */
+	uchar*	buf)	/* in/out: buffer for previous row in MySQL format */
+{
+	ha_statistic_increment(&SSV::ha_read_prev_count);
+
+	return(general_fetch(buf, ROW_SEL_PREV, 0));
+}
+
+/************************************************************************
+Positions a cursor on the first record in an index and reads the
+corresponding row to buf. */
+UNIV_INTERN
+int
+ha_innobase::index_first(
+/*=====================*/
+			/* out: 0, HA_ERR_END_OF_FILE, or error code */
+	uchar*	buf)	/* in/out: buffer for the row */
+{
+	int	error;
+
+	DBUG_ENTER("index_first");
+	ha_statistic_increment(&SSV::ha_read_first_count);
+
+	error = index_read(buf, NULL, 0, HA_READ_AFTER_KEY);
+
+	/* MySQL does not seem to allow this to return HA_ERR_KEY_NOT_FOUND */
+
+	if (error == HA_ERR_KEY_NOT_FOUND) {
+		error = HA_ERR_END_OF_FILE;
+	}
+
+	DBUG_RETURN(error);
+}
+
+/************************************************************************
+Positions a cursor on the last record in an index and reads the
+corresponding row to buf. */
+UNIV_INTERN
+int
+ha_innobase::index_last(
+/*====================*/
+			/* out: 0, HA_ERR_END_OF_FILE, or error code */
+	uchar*	buf)	/* in/out: buffer for the row */
+{
+	int	error;
+
+	DBUG_ENTER("index_last");
+	ha_statistic_increment(&SSV::ha_read_last_count);
+
+	error = index_read(buf, NULL, 0, HA_READ_BEFORE_KEY);
+
+	/* MySQL does not seem to allow this to return HA_ERR_KEY_NOT_FOUND */
+
+	if (error == HA_ERR_KEY_NOT_FOUND) {
+		error = HA_ERR_END_OF_FILE;
+	}
+
+	DBUG_RETURN(error);
+}
+
+/********************************************************************
+Initialize a table scan. */
+UNIV_INTERN
+int
+ha_innobase::rnd_init(
+/*==================*/
+			/* out: 0 or error number */
+	bool	scan)	/* in: TRUE if table/index scan FALSE otherwise */
+{
+	int	err;
+
+	/* Store the active index value so that we can restore the original
+	value after a scan */
+
+	if (prebuilt->clust_index_was_generated) {
+		err = change_active_index(MAX_KEY);
+	} else {
+		err = change_active_index(primary_key);
+	}
+
+	/* Don't use semi-consistent read in random row reads (by position).
+	This means we must disable semi_consistent_read if scan is false */
+
+	if (!scan) {
+		try_semi_consistent_read(0);
+	}
+
+	start_of_scan = 1;
+
+	return(err);
+}
+
+/*********************************************************************
+Ends a table scan. */
+UNIV_INTERN
+int
+ha_innobase::rnd_end(void)
+/*======================*/
+				/* out: 0 or error number */
+{
+	return(index_end());
+}
+
+/*********************************************************************
+Reads the next row in a table scan (also used to read the FIRST row
+in a table scan). */
+UNIV_INTERN
+int
+ha_innobase::rnd_next(
+/*==================*/
+			/* out: 0, HA_ERR_END_OF_FILE, or error number */
+	uchar*	buf)	/* in/out: returns the row in this buffer,
+			in MySQL format */
+{
+	int	error;
+
+	DBUG_ENTER("rnd_next");
+	ha_statistic_increment(&SSV::ha_read_rnd_next_count);
+
+	if (start_of_scan) {
+		error = index_first(buf);
+
+		if (error == HA_ERR_KEY_NOT_FOUND) {
+			error = HA_ERR_END_OF_FILE;
+		}
+
+		start_of_scan = 0;
+	} else {
+		error = general_fetch(buf, ROW_SEL_NEXT, 0);
+	}
+
+	DBUG_RETURN(error);
+}
+
+/**************************************************************************
+Fetches a row from the table based on a row reference. */
+UNIV_INTERN
+int
+ha_innobase::rnd_pos(
+/*=================*/
+			/* out: 0, HA_ERR_KEY_NOT_FOUND, or error code */
+	uchar*	buf,	/* in/out: buffer for the row */
+	uchar*	pos)	/* in: primary key value of the row in the
+			MySQL format, or the row id if the clustered
+			index was internally generated by InnoDB; the
+			length of data in pos has to be ref_length */
+{
+	int		error;
+	uint		keynr	= active_index;
+	DBUG_ENTER("rnd_pos");
+	DBUG_DUMP("key", pos, ref_length);
+
+	ha_statistic_increment(&SSV::ha_read_rnd_count);
+
+	ut_a(prebuilt->trx == thd_to_trx(ha_thd()));
+
+	if (prebuilt->clust_index_was_generated) {
+		/* No primary key was defined for the table and we
+		generated the clustered index from the row id: the
+		row reference is the row id, not any key value
+		that MySQL knows of */
+
+		error = change_active_index(MAX_KEY);
+	} else {
+		error = change_active_index(primary_key);
+	}
+
+	if (error) {
+		DBUG_PRINT("error", ("Got error: %d", error));
+		DBUG_RETURN(error);
+	}
+
+	/* Note that we assume the length of the row reference is fixed
+	for the table, and it is == ref_length */
+
+	error = index_read(buf, pos, ref_length, HA_READ_KEY_EXACT);
+
+	if (error) {
+		DBUG_PRINT("error", ("Got error: %d", error));
+	}
+
+	change_active_index(keynr);
+
+	DBUG_RETURN(error);
+}
+
+/*************************************************************************
+Stores a reference to the current row to 'ref' field of the handle. Note
+that in the case where we have generated the clustered index for the
+table, the function parameter is illogical: we MUST ASSUME that 'record'
+is the current 'position' of the handle, because if row ref is actually
+the row id internally generated in InnoDB, then 'record' does not contain
+it. We just guess that the row id must be for the record where the handle
+was positioned the last time. */
+UNIV_INTERN
+void
+ha_innobase::position(
+/*==================*/
+	const uchar*	record)	/* in: row in MySQL format */
+{
+	uint		len;
+
+	ut_a(prebuilt->trx == thd_to_trx(ha_thd()));
+
+	if (prebuilt->clust_index_was_generated) {
+		/* No primary key was defined for the table and we
+		generated the clustered index from row id: the
+		row reference will be the row id, not any key value
+		that MySQL knows of */
+
+		len = DATA_ROW_ID_LEN;
+
+		memcpy(ref, prebuilt->row_id, len);
+	} else {
+		len = store_key_val_for_row(primary_key, (char*)ref,
+							 ref_length, record);
+	}
+
+	/* We assume that the 'ref' value len is always fixed for the same
+	table. */
+
+	if (len != ref_length) {
+	  sql_print_error("Stored ref len is %lu, but table ref len is %lu",
+			  (ulong) len, (ulong) ref_length);
+	}
+}
+
+/* limit innodb monitor access to users with PROCESS privilege.
+See http://bugs.mysql.com/32710 for expl. why we choose PROCESS. */
+#define IS_MAGIC_TABLE_AND_USER_DENIED_ACCESS(table_name, thd) \
+	(row_is_magic_monitor_table(table_name) \
+	 && check_global_access(thd, PROCESS_ACL))
+
+/*********************************************************************
+Creates a table definition to an InnoDB database. */
+static
+int
+create_table_def(
+/*=============*/
+	trx_t*		trx,		/* in: InnoDB transaction handle */
+	TABLE*		form,		/* in: information on table
+					columns and indexes */
+	const char*	table_name,	/* in: table name */
+	const char*	path_of_temp_table,/* in: if this is a table explicitly
+					created by the user with the
+					TEMPORARY keyword, then this
+					parameter is the dir path where the
+					table should be placed if we create
+					an .ibd file for it (no .ibd extension
+					in the path, though); otherwise this
+					is NULL */
+	ulint		flags)		/* in: table flags */
+{
+	Field*		field;
+	dict_table_t*	table;
+	ulint		n_cols;
+	int		error;
+	ulint		col_type;
+	ulint		col_len;
+	ulint		nulls_allowed;
+	ulint		unsigned_type;
+	ulint		binary_type;
+	ulint		long_true_varchar;
+	ulint		charset_no;
+	ulint		i;
+
+	DBUG_ENTER("create_table_def");
+	DBUG_PRINT("enter", ("table_name: %s", table_name));
+
+	ut_a(trx->mysql_thd != NULL);
+	if (IS_MAGIC_TABLE_AND_USER_DENIED_ACCESS(table_name,
+						  (THD*) trx->mysql_thd)) {
+		DBUG_RETURN(HA_ERR_GENERIC);
+	}
+
+	n_cols = form->s->fields;
+
+	/* We pass 0 as the space id, and determine at a lower level the space
+	id where to store the table */
+
+	table = dict_mem_table_create(table_name, 0, n_cols, flags);
+
+	if (path_of_temp_table) {
+		table->dir_path_of_temp_table =
+			mem_heap_strdup(table->heap, path_of_temp_table);
+	}
+
+	for (i = 0; i < n_cols; i++) {
+		field = form->field[i];
+
+		col_type = get_innobase_type_from_mysql_type(&unsigned_type,
+									field);
+		if (field->null_ptr) {
+			nulls_allowed = 0;
+		} else {
+			nulls_allowed = DATA_NOT_NULL;
+		}
+
+		if (field->binary()) {
+			binary_type = DATA_BINARY_TYPE;
+		} else {
+			binary_type = 0;
+		}
+
+		charset_no = 0;
+
+		if (dtype_is_string_type(col_type)) {
+
+			charset_no = (ulint)field->charset()->number;
+
+			if (UNIV_UNLIKELY(charset_no >= 256)) {
+				/* in data0type.h we assume that the
+				number fits in one byte in prtype */
+				push_warning_printf(
+					(THD*) trx->mysql_thd,
+					MYSQL_ERROR::WARN_LEVEL_ERROR,
+					ER_CANT_CREATE_TABLE,
+					"In InnoDB, charset-collation codes"
+					" must be below 256."
+					" Unsupported code %lu.",
+					(ulong) charset_no);
+				DBUG_RETURN(ER_CANT_CREATE_TABLE);
+			}
+		}
+
+		ut_a(field->type() < 256); /* we assume in dtype_form_prtype()
+					   that this fits in one byte */
+		col_len = field->pack_length();
+
+		/* The MySQL pack length contains 1 or 2 bytes length field
+		for a true VARCHAR. Let us subtract that, so that the InnoDB
+		column length in the InnoDB data dictionary is the real
+		maximum byte length of the actual data. */
+
+		long_true_varchar = 0;
+
+		if (field->type() == MYSQL_TYPE_VARCHAR) {
+			col_len -= ((Field_varstring*)field)->length_bytes;
+
+			if (((Field_varstring*)field)->length_bytes == 2) {
+				long_true_varchar = DATA_LONG_TRUE_VARCHAR;
+			}
+		}
+
+		dict_mem_table_add_col(table, table->heap,
+			(char*) field->field_name,
+			col_type,
+			dtype_form_prtype(
+				(ulint)field->type()
+				| nulls_allowed | unsigned_type
+				| binary_type | long_true_varchar,
+				charset_no),
+			col_len);
+	}
+
+	error = row_create_table_for_mysql(table, trx);
+
+	error = convert_error_code_to_mysql(error, flags, NULL);
+
+	DBUG_RETURN(error);
+}
+
+/*********************************************************************
+Creates an index in an InnoDB database. */
+static
+int
+create_index(
+/*=========*/
+	trx_t*		trx,		/* in: InnoDB transaction handle */
+	TABLE*		form,		/* in: information on table
+					columns and indexes */
+	ulint		flags,		/* in: InnoDB table flags */
+	const char*	table_name,	/* in: table name */
+	uint		key_num)	/* in: index number */
+{
+	Field*		field;
+	dict_index_t*	index;
+	int		error;
+	ulint		n_fields;
+	KEY*		key;
+	KEY_PART_INFO*	key_part;
+	ulint		ind_type;
+	ulint		col_type;
+	ulint		prefix_len;
+	ulint		is_unsigned;
+	ulint		i;
+	ulint		j;
+	ulint*		field_lengths;
+
+	DBUG_ENTER("create_index");
+
+	key = form->key_info + key_num;
+
+	n_fields = key->key_parts;
+
+	ind_type = 0;
+
+	if (key_num == form->s->primary_key) {
+		ind_type = ind_type | DICT_CLUSTERED;
+	}
+
+	if (key->flags & HA_NOSAME ) {
+		ind_type = ind_type | DICT_UNIQUE;
+	}
+
+	/* We pass 0 as the space id, and determine at a lower level the space
+	id where to store the table */
+
+	index = dict_mem_index_create(table_name, key->name, 0,
+				      ind_type, n_fields);
+
+	field_lengths = (ulint*) my_malloc(sizeof(ulint) * n_fields,
+		MYF(MY_FAE));
+
+	for (i = 0; i < n_fields; i++) {
+		key_part = key->key_part + i;
+
+		/* (The flag HA_PART_KEY_SEG denotes in MySQL a column prefix
+		field in an index: we only store a specified number of first
+		bytes of the column to the index field.) The flag does not
+		seem to be properly set by MySQL. Let us fall back on testing
+		the length of the key part versus the column. */
+
+		field = NULL;
+		for (j = 0; j < form->s->fields; j++) {
+
+			field = form->field[j];
+
+			if (0 == innobase_strcasecmp(
+					field->field_name,
+					key_part->field->field_name)) {
+				/* Found the corresponding column */
+
+				break;
+			}
+		}
+
+		ut_a(j < form->s->fields);
+
+		col_type = get_innobase_type_from_mysql_type(
+					&is_unsigned, key_part->field);
+
+		if (DATA_BLOB == col_type
+			|| (key_part->length < field->pack_length()
+				&& field->type() != MYSQL_TYPE_VARCHAR)
+			|| (field->type() == MYSQL_TYPE_VARCHAR
+				&& key_part->length < field->pack_length()
+				- ((Field_varstring*)field)->length_bytes)) {
+
+			prefix_len = key_part->length;
+
+			if (col_type == DATA_INT
+				|| col_type == DATA_FLOAT
+				|| col_type == DATA_DOUBLE
+				|| col_type == DATA_DECIMAL) {
+				sql_print_error(
+					"MySQL is trying to create a column "
+					"prefix index field, on an "
+					"inappropriate data type. Table "
+					"name %s, column name %s.",
+					table_name,
+					key_part->field->field_name);
+
+				prefix_len = 0;
+			}
+		} else {
+			prefix_len = 0;
+		}
+
+		field_lengths[i] = key_part->length;
+
+		dict_mem_index_add_field(index,
+			(char*) key_part->field->field_name, prefix_len);
+	}
+
+	/* Even though we've defined max_supported_key_part_length, we
+	still do our own checking using field_lengths to be absolutely
+	sure we don't create too long indexes. */
+	error = row_create_index_for_mysql(index, trx, field_lengths);
+
+	error = convert_error_code_to_mysql(error, flags, NULL);
+
+	my_free(field_lengths, MYF(0));
+
+	DBUG_RETURN(error);
+}
+
+/*********************************************************************
+Creates an index to an InnoDB table when the user has defined no
+primary index. */
+static
+int
+create_clustered_index_when_no_primary(
+/*===================================*/
+	trx_t*		trx,		/* in: InnoDB transaction handle */
+	ulint		flags,		/* in: InnoDB table flags */
+	const char*	table_name)	/* in: table name */
+{
+	dict_index_t*	index;
+	int		error;
+
+	/* We pass 0 as the space id, and determine at a lower level the space
+	id where to store the table */
+
+	index = dict_mem_index_create(table_name, "GEN_CLUST_INDEX",
+				      0, DICT_CLUSTERED, 0);
+
+	error = row_create_index_for_mysql(index, trx, NULL);
+
+	error = convert_error_code_to_mysql(error, flags, NULL);
+
+	return(error);
+}
+
+/*********************************************************************
+Validates the create options. We may build on this function
+in future. For now, it checks two specifiers:
+KEY_BLOCK_SIZE and ROW_FORMAT
+If innodb_strict_mode is not set then this function is a no-op */
+static
+ibool
+create_options_are_valid(
+/*=====================*/
+					/* out: TRUE if valid. */
+	THD*		thd,		/* in: connection thread. */
+	TABLE*		form,		/* in: information on table
+					columns and indexes */
+	HA_CREATE_INFO*	create_info)	/* in: create info. */
+{
+	ibool 	kbs_specified	= FALSE;
+	ibool	ret		= TRUE;
+
+
+	ut_ad(thd != NULL);
+
+	/* If innodb_strict_mode is not set don't do any validation. */
+	if (!(THDVAR(thd, strict_mode))) {
+		return(TRUE);
+	}
+
+	ut_ad(form != NULL);
+	ut_ad(create_info != NULL);
+
+	/* First check if KEY_BLOCK_SIZE was specified. */
+	if (create_info->key_block_size
+	    || (create_info->used_fields & HA_CREATE_USED_KEY_BLOCK_SIZE)) {
+
+		kbs_specified = TRUE;
+		switch (create_info->key_block_size) {
+		case 1:
+		case 2:
+		case 4:
+		case 8:
+		case 16:
+			/* Valid value. */
+			break;
+		default:
+			push_warning_printf(thd, MYSQL_ERROR::WARN_LEVEL_ERROR,
+					    ER_ILLEGAL_HA_CREATE_OPTION,
+					    "InnoDB: invalid"
+					    " KEY_BLOCK_SIZE = %lu."
+					    " Valid values are"
+					    " [1, 2, 4, 8, 16]",
+					    create_info->key_block_size);
+			ret = FALSE;
+		}
+	}
+	
+	/* If KEY_BLOCK_SIZE was specified, check for its
+	dependencies. */
+	if (kbs_specified && !srv_file_per_table) {
+		push_warning(thd, MYSQL_ERROR::WARN_LEVEL_ERROR,
+			     ER_ILLEGAL_HA_CREATE_OPTION,
+			     "InnoDB: KEY_BLOCK_SIZE"
+			     " requires innodb_file_per_table.");
+		ret = FALSE;
+	}
+
+	if (kbs_specified && srv_file_format < DICT_TF_FORMAT_ZIP) {
+		push_warning(thd, MYSQL_ERROR::WARN_LEVEL_ERROR,
+			     ER_ILLEGAL_HA_CREATE_OPTION,
+			     "InnoDB: KEY_BLOCK_SIZE"
+			     " requires innodb_file_format >"
+			     " Antelope.");
+		ret = FALSE;
+	}
+
+	/* Now check for ROW_FORMAT specifier. */
+	if (create_info->used_fields & HA_CREATE_USED_ROW_FORMAT) {
+		switch (form->s->row_type) {
+			const char* row_format_name;
+		case ROW_TYPE_COMPRESSED:
+		case ROW_TYPE_DYNAMIC:
+			row_format_name
+				= form->s->row_type == ROW_TYPE_COMPRESSED
+				? "COMPRESSED"
+				: "DYNAMIC";
+
+			/* These two ROW_FORMATs require
+			srv_file_per_table and srv_file_format */
+			if (!srv_file_per_table) {
+				push_warning_printf(
+					thd,
+					MYSQL_ERROR::WARN_LEVEL_ERROR,
+					ER_ILLEGAL_HA_CREATE_OPTION,
+					"InnoDB: ROW_FORMAT=%s"
+					" requires innodb_file_per_table.",
+					row_format_name);
+					ret = FALSE;
+
+			}
+
+			if (srv_file_format < DICT_TF_FORMAT_ZIP) {
+				push_warning_printf(
+					thd,
+					MYSQL_ERROR::WARN_LEVEL_ERROR,
+					ER_ILLEGAL_HA_CREATE_OPTION,
+					"InnoDB: ROW_FORMAT=%s"
+					" requires innodb_file_format >"
+					" Antelope.",
+					row_format_name);
+					ret = FALSE;
+			}
+
+			/* Cannot specify KEY_BLOCK_SIZE with
+			ROW_FORMAT = DYNAMIC.
+			However, we do allow COMPRESSED to be
+			specified with KEY_BLOCK_SIZE. */
+			if (kbs_specified
+			    && form->s->row_type == ROW_TYPE_DYNAMIC) {
+				push_warning_printf(
+					thd,
+					MYSQL_ERROR::WARN_LEVEL_ERROR,
+					ER_ILLEGAL_HA_CREATE_OPTION,
+					"InnoDB: cannot specify"
+					" ROW_FORMAT = DYNAMIC with"
+					" KEY_BLOCK_SIZE.");
+					ret = FALSE;
+			}
+
+			break;
+
+		case ROW_TYPE_REDUNDANT:
+		case ROW_TYPE_COMPACT:
+		case ROW_TYPE_DEFAULT:
+			/* Default is COMPACT. */
+			row_format_name
+				= form->s->row_type == ROW_TYPE_REDUNDANT
+				? "REDUNDANT"
+				: "COMPACT";
+
+			/* Cannot specify KEY_BLOCK_SIZE with these
+			format specifiers. */
+			if (kbs_specified) {
+				push_warning_printf(
+					thd,
+					MYSQL_ERROR::WARN_LEVEL_ERROR,
+					ER_ILLEGAL_HA_CREATE_OPTION,
+					"InnoDB: cannot specify"
+					" ROW_FORMAT = %s with"
+					" KEY_BLOCK_SIZE.",
+					row_format_name);
+					ret = FALSE;
+			}
+
+			break;
+
+		default:
+			push_warning(thd,
+				     MYSQL_ERROR::WARN_LEVEL_ERROR,
+				     ER_ILLEGAL_HA_CREATE_OPTION,
+				     "InnoDB: invalid ROW_FORMAT specifier.");
+			ret = FALSE;
+
+		}
+	}
+
+	return(ret);
+}
+
+/*********************************************************************
+Update create_info.  Used in SHOW CREATE TABLE et al. */
+UNIV_INTERN
+void
+ha_innobase::update_create_info(
+/*============================*/
+	HA_CREATE_INFO* create_info)	/* in/out: create info */
+{
+  if (!(create_info->used_fields & HA_CREATE_USED_AUTO)) {
+    ha_innobase::info(HA_STATUS_AUTO);
+    create_info->auto_increment_value = stats.auto_increment_value;
+  }
+}
+
+/*********************************************************************
+Creates a new table to an InnoDB database. */
+UNIV_INTERN
+int
+ha_innobase::create(
+/*================*/
+					/* out: error number */
+	const char*	name,		/* in: table name */
+	TABLE*		form,		/* in: information on table
+					columns and indexes */
+	HA_CREATE_INFO*	create_info)	/* in: more information of the
+					created table, contains also the
+					create statement string */
+{
+	int		error;
+	dict_table_t*	innobase_table;
+	trx_t*		parent_trx;
+	trx_t*		trx;
+	int		primary_key_no;
+	uint		i;
+	char		name2[FN_REFLEN];
+	char		norm_name[FN_REFLEN];
+	THD*		thd = ha_thd();
+	ib_int64_t	auto_inc_value;
+	ulint		flags;
+	/* Cache the value of innodb_file_format, in case it is
+	modified by another thread while the table is being created. */
+	const ulint	file_format = srv_file_format;
+
+	DBUG_ENTER("ha_innobase::create");
+
+	DBUG_ASSERT(thd != NULL);
+	DBUG_ASSERT(create_info != NULL);
+
+#ifdef __WIN__
+	/* Names passed in from server are in two formats:
+	1. <database_name>/<table_name>: for normal table creation
+	2. full path: for temp table creation, or sym link
+
+	When srv_file_per_table is on, check for full path pattern, i.e.
+	X:\dir\...,		X is a driver letter, or
+	\\dir1\dir2\...,	UNC path
+	returns error if it is in full path format, but not creating a temp.
+	table. Currently InnoDB does not support symbolic link on Windows. */
+
+	if (srv_file_per_table
+	    && (!create_info->options & HA_LEX_CREATE_TMP_TABLE)) {
+
+		if ((name[1] == ':')
+		    || (name[0] == '\\' && name[1] == '\\')) {
+			sql_print_error("Cannot create table %s\n", name);
+			DBUG_RETURN(HA_ERR_GENERIC);
+		}
+	}
+#endif
+
+	if (form->s->fields > 1000) {
+		/* The limit probably should be REC_MAX_N_FIELDS - 3 = 1020,
+		but we play safe here */
+
+		DBUG_RETURN(HA_ERR_TO_BIG_ROW);
+	}
+
+	/* Get the transaction associated with the current thd, or create one
+	if not yet created */
+
+	parent_trx = check_trx_exists(thd);
+
+	/* In case MySQL calls this in the middle of a SELECT query, release
+	possible adaptive hash latch to avoid deadlocks of threads */
+
+	trx_search_latch_release_if_reserved(parent_trx);
+
+	trx = innobase_trx_allocate(thd);
+
+	if (lower_case_table_names) {
+		srv_lower_case_table_names = TRUE;
+	} else {
+		srv_lower_case_table_names = FALSE;
+	}
+
+	strcpy(name2, name);
+
+	normalize_table_name(norm_name, name2);
+
+	/* Latch the InnoDB data dictionary exclusively so that no deadlocks
+	or lock waits can happen in it during a table create operation.
+	Drop table etc. do this latching in row0mysql.c. */
+
+	row_mysql_lock_data_dictionary(trx);
+
+	/* Create the table definition in InnoDB */
+
+	flags = 0;
+
+	/* Validate create options if innodb_strict_mode is set. */
+	if (!create_options_are_valid(thd, form, create_info)) {
+		error = ER_ILLEGAL_HA_CREATE_OPTION;
+		goto cleanup;
+	}
+
+	if (create_info->key_block_size
+	    || (create_info->used_fields & HA_CREATE_USED_KEY_BLOCK_SIZE)) {
+		/* Determine the page_zip.ssize corresponding to the
+		requested page size (key_block_size) in kilobytes. */
+
+		ulint	ssize, ksize;
+		ulint	key_block_size = create_info->key_block_size;
+
+		for (ssize = ksize = 1; ssize <= DICT_TF_ZSSIZE_MAX;
+		     ssize++, ksize <<= 1) {
+			if (key_block_size == ksize) {
+				flags = ssize << DICT_TF_ZSSIZE_SHIFT
+					| DICT_TF_COMPACT
+					| DICT_TF_FORMAT_ZIP
+					  << DICT_TF_FORMAT_SHIFT;
+				break;
+			}
+		}
+
+		if (!srv_file_per_table) {
+			push_warning(thd, MYSQL_ERROR::WARN_LEVEL_WARN,
+				     ER_ILLEGAL_HA_CREATE_OPTION,
+				     "InnoDB: KEY_BLOCK_SIZE"
+				     " requires innodb_file_per_table.");
+			flags = 0;
+		}
+
+		if (file_format < DICT_TF_FORMAT_ZIP) {
+			push_warning(thd, MYSQL_ERROR::WARN_LEVEL_WARN,
+				     ER_ILLEGAL_HA_CREATE_OPTION,
+				     "InnoDB: KEY_BLOCK_SIZE"
+				     " requires innodb_file_format >"
+				     " Antelope.");
+			flags = 0;
+		}
+
+		if (!flags) {
+			push_warning_printf(thd, MYSQL_ERROR::WARN_LEVEL_WARN,
+					    ER_ILLEGAL_HA_CREATE_OPTION,
+					    "InnoDB: ignoring"
+					    " KEY_BLOCK_SIZE=%lu.",
+					    create_info->key_block_size);
+		}
+	}
+
+	if (create_info->used_fields & HA_CREATE_USED_ROW_FORMAT) {
+		if (flags) {
+			/* KEY_BLOCK_SIZE was specified. */
+			if (form->s->row_type != ROW_TYPE_COMPRESSED) {
+				/* ROW_FORMAT other than COMPRESSED
+				ignores KEY_BLOCK_SIZE.  It does not
+				make sense to reject conflicting
+				KEY_BLOCK_SIZE and ROW_FORMAT, because
+				such combinations can be obtained
+				with ALTER TABLE anyway. */
+				push_warning_printf(
+					thd,
+					MYSQL_ERROR::WARN_LEVEL_WARN,
+					ER_ILLEGAL_HA_CREATE_OPTION,
+					"InnoDB: ignoring KEY_BLOCK_SIZE=%lu"
+					" unless ROW_FORMAT=COMPRESSED.",
+					create_info->key_block_size);
+				flags = 0;
+			}
+		} else {
+			/* No KEY_BLOCK_SIZE */
+			if (form->s->row_type == ROW_TYPE_COMPRESSED) {
+				/* ROW_FORMAT=COMPRESSED without
+				KEY_BLOCK_SIZE implies half the
+				maximum KEY_BLOCK_SIZE. */
+				flags = (DICT_TF_ZSSIZE_MAX - 1)
+					<< DICT_TF_ZSSIZE_SHIFT
+					| DICT_TF_COMPACT
+					| DICT_TF_FORMAT_ZIP
+					<< DICT_TF_FORMAT_SHIFT;
+#if DICT_TF_ZSSIZE_MAX < 1
+# error "DICT_TF_ZSSIZE_MAX < 1"
+#endif
+			}
+		}
+
+		switch (form->s->row_type) {
+			const char* row_format_name;
+		case ROW_TYPE_REDUNDANT:
+			break;
+		case ROW_TYPE_COMPRESSED:
+		case ROW_TYPE_DYNAMIC:
+			row_format_name
+				= form->s->row_type == ROW_TYPE_COMPRESSED
+				? "COMPRESSED"
+				: "DYNAMIC";
+
+			if (!srv_file_per_table) {
+				push_warning_printf(
+					thd,
+					MYSQL_ERROR::WARN_LEVEL_WARN,
+					ER_ILLEGAL_HA_CREATE_OPTION,
+					"InnoDB: ROW_FORMAT=%s"
+					" requires innodb_file_per_table.",
+					row_format_name);
+			} else if (file_format < DICT_TF_FORMAT_ZIP) {
+				push_warning_printf(
+					thd,
+					MYSQL_ERROR::WARN_LEVEL_WARN,
+					ER_ILLEGAL_HA_CREATE_OPTION,
+					"InnoDB: ROW_FORMAT=%s"
+					" requires innodb_file_format >"
+					" Antelope.",
+					row_format_name);
+			} else {
+				flags |= DICT_TF_COMPACT
+					| (DICT_TF_FORMAT_ZIP
+					   << DICT_TF_FORMAT_SHIFT);
+				break;
+			}
+
+			/* fall through */
+		case ROW_TYPE_NOT_USED:
+		case ROW_TYPE_FIXED:
+		default:
+			push_warning(thd,
+				     MYSQL_ERROR::WARN_LEVEL_WARN,
+				     ER_ILLEGAL_HA_CREATE_OPTION,
+				     "InnoDB: assuming ROW_FORMAT=COMPACT.");
+		case ROW_TYPE_DEFAULT:
+		case ROW_TYPE_COMPACT:
+			flags = DICT_TF_COMPACT;
+			break;
+		}
+	} else if (!flags) {
+		/* No KEY_BLOCK_SIZE or ROW_FORMAT specified:
+		use ROW_FORMAT=COMPACT by default. */
+		flags = DICT_TF_COMPACT;
+	}
+
+	error = create_table_def(trx, form, norm_name,
+		create_info->options & HA_LEX_CREATE_TMP_TABLE ? name2 : NULL,
+		flags);
+
+	if (error) {
+		goto cleanup;
+	}
+
+	/* Look for a primary key */
+
+	primary_key_no= (form->s->primary_key != MAX_KEY ?
+			 (int) form->s->primary_key :
+			 -1);
+
+	/* Our function row_get_mysql_key_number_for_index assumes
+	the primary key is always number 0, if it exists */
+
+	DBUG_ASSERT(primary_key_no == -1 || primary_key_no == 0);
+
+	/* Create the keys */
+
+	if (form->s->keys == 0 || primary_key_no == -1) {
+		/* Create an index which is used as the clustered index;
+		order the rows by their row id which is internally generated
+		by InnoDB */
+
+		error = create_clustered_index_when_no_primary(
+			trx, flags, norm_name);
+		if (error) {
+			goto cleanup;
+		}
+	}
+
+	if (primary_key_no != -1) {
+		/* In InnoDB the clustered index must always be created
+		first */
+		if ((error = create_index(trx, form, flags, norm_name,
+					  (uint) primary_key_no))) {
+			goto cleanup;
+		}
+	}
+
+	for (i = 0; i < form->s->keys; i++) {
+
+		if (i != (uint) primary_key_no) {
+
+			if ((error = create_index(trx, form, flags, norm_name,
+						  i))) {
+				goto cleanup;
+			}
+		}
+	}
+
+	if (*trx->mysql_query_str) {
+		error = row_table_add_foreign_constraints(trx,
+			*trx->mysql_query_str, norm_name,
+			create_info->options & HA_LEX_CREATE_TMP_TABLE);
+
+		error = convert_error_code_to_mysql(error, flags, NULL);
+
+		if (error) {
+			goto cleanup;
+		}
+	}
+
+	innobase_commit_low(trx);
+
+	row_mysql_unlock_data_dictionary(trx);
+
+	/* Flush the log to reduce probability that the .frm files and
+	the InnoDB data dictionary get out-of-sync if the user runs
+	with innodb_flush_log_at_trx_commit = 0 */
+
+	log_buffer_flush_to_disk();
+
+	innobase_table = dict_table_get(norm_name, FALSE);
+
+	DBUG_ASSERT(innobase_table != 0);
+
+	if (innobase_table) {
+		/* We update the highest file format in the system table
+		space, if this table has higher file format setting. */
+
+		trx_sys_file_format_max_upgrade(
+			(const char**) &innobase_file_format_check,
+			dict_table_get_format(innobase_table));
+	}
+
+	/* Note: We can't call update_thd() as prebuilt will not be
+	setup at this stage and so we use thd. */
+
+	/* We need to copy the AUTOINC value from the old table if
+	this is an ALTER TABLE. */
+
+	if (((create_info->used_fields & HA_CREATE_USED_AUTO)
+	    || thd_sql_command(thd) == SQLCOM_ALTER_TABLE)
+	    && create_info->auto_increment_value != 0) {
+
+		/* Query was ALTER TABLE...AUTO_INCREMENT = x; or
+		CREATE TABLE ...AUTO_INCREMENT = x; Find out a table
+		definition from the dictionary and get the current value
+		of the auto increment field. Set a new value to the
+		auto increment field if the value is greater than the
+		maximum value in the column. */
+
+		auto_inc_value = create_info->auto_increment_value;
+
+		dict_table_autoinc_lock(innobase_table);
+		dict_table_autoinc_initialize(innobase_table, auto_inc_value);
+		dict_table_autoinc_unlock(innobase_table);
+	}
+
+	/* Tell the InnoDB server that there might be work for
+	utility threads: */
+
+	srv_active_wake_master_thread();
+
+	trx_free_for_mysql(trx);
+
+	DBUG_RETURN(0);
+
+cleanup:
+	innobase_commit_low(trx);
+
+	row_mysql_unlock_data_dictionary(trx);
+
+	trx_free_for_mysql(trx);
+
+	DBUG_RETURN(error);
+}
+
+/*********************************************************************
+Discards or imports an InnoDB tablespace. */
+UNIV_INTERN
+int
+ha_innobase::discard_or_import_tablespace(
+/*======================================*/
+				/* out: 0 == success, -1 == error */
+	my_bool discard)	/* in: TRUE if discard, else import */
+{
+	dict_table_t*	dict_table;
+	trx_t*		trx;
+	int		err;
+
+	DBUG_ENTER("ha_innobase::discard_or_import_tablespace");
+
+	ut_a(prebuilt->trx);
+	ut_a(prebuilt->trx->magic_n == TRX_MAGIC_N);
+	ut_a(prebuilt->trx == thd_to_trx(ha_thd()));
+
+	dict_table = prebuilt->table;
+	trx = prebuilt->trx;
+
+	if (discard) {
+		err = row_discard_tablespace_for_mysql(dict_table->name, trx);
+	} else {
+		err = row_import_tablespace_for_mysql(dict_table->name, trx);
+	}
+
+	err = convert_error_code_to_mysql(err, dict_table->flags, NULL);
+
+	DBUG_RETURN(err);
+}
+
+/*********************************************************************
+Deletes all rows of an InnoDB table. */
+UNIV_INTERN
+int
+ha_innobase::delete_all_rows(void)
+/*==============================*/
+				/* out: error number */
+{
+	int		error;
+
+	DBUG_ENTER("ha_innobase::delete_all_rows");
+
+	/* Get the transaction associated with the current thd, or create one
+	if not yet created, and update prebuilt->trx */
+
+	update_thd(ha_thd());
+
+	if (thd_sql_command(user_thd) != SQLCOM_TRUNCATE) {
+	fallback:
+		/* We only handle TRUNCATE TABLE t as a special case.
+		DELETE FROM t will have to use ha_innobase::delete_row(),
+		because DELETE is transactional while TRUNCATE is not. */
+		DBUG_RETURN(my_errno=HA_ERR_WRONG_COMMAND);
+	}
+
+	/* Truncate the table in InnoDB */
+
+	error = row_truncate_table_for_mysql(prebuilt->table, prebuilt->trx);
+	if (error == DB_ERROR) {
+		/* Cannot truncate; resort to ha_innobase::delete_row() */
+		goto fallback;
+	}
+
+	error = convert_error_code_to_mysql(error, prebuilt->table->flags,
+					    NULL);
+
+	DBUG_RETURN(error);
+}
+
+/*********************************************************************
+Drops a table from an InnoDB database. Before calling this function,
+MySQL calls innobase_commit to commit the transaction of the current user.
+Then the current user cannot have locks set on the table. Drop table
+operation inside InnoDB will remove all locks any user has on the table
+inside InnoDB. */
+UNIV_INTERN
+int
+ha_innobase::delete_table(
+/*======================*/
+				/* out: error number */
+	const char*	name)	/* in: table name */
+{
+	ulint	name_len;
+	int	error;
+	trx_t*	parent_trx;
+	trx_t*	trx;
+	THD	*thd = ha_thd();
+	char	norm_name[1000];
+
+	DBUG_ENTER("ha_innobase::delete_table");
+
+	/* Strangely, MySQL passes the table name without the '.frm'
+	extension, in contrast to ::create */
+	normalize_table_name(norm_name, name);
+
+	if (IS_MAGIC_TABLE_AND_USER_DENIED_ACCESS(norm_name, thd)) {
+		DBUG_RETURN(HA_ERR_GENERIC);
+	}
+
+	/* Get the transaction associated with the current thd, or create one
+	if not yet created */
+
+	parent_trx = check_trx_exists(thd);
+
+	/* In case MySQL calls this in the middle of a SELECT query, release
+	possible adaptive hash latch to avoid deadlocks of threads */
+
+	trx_search_latch_release_if_reserved(parent_trx);
+
+	trx = innobase_trx_allocate(thd);
+
+	if (lower_case_table_names) {
+		srv_lower_case_table_names = TRUE;
+	} else {
+		srv_lower_case_table_names = FALSE;
+	}
+
+	name_len = strlen(name);
+
+	ut_a(name_len < 1000);
+
+	/* Drop the table in InnoDB */
+
+	error = row_drop_table_for_mysql(norm_name, trx,
+					 thd_sql_command(thd)
+					 == SQLCOM_DROP_DB);
+
+	/* Flush the log to reduce probability that the .frm files and
+	the InnoDB data dictionary get out-of-sync if the user runs
+	with innodb_flush_log_at_trx_commit = 0 */
+
+	log_buffer_flush_to_disk();
+
+	/* Tell the InnoDB server that there might be work for
+	utility threads: */
+
+	srv_active_wake_master_thread();
+
+	innobase_commit_low(trx);
+
+	trx_free_for_mysql(trx);
+
+	error = convert_error_code_to_mysql(error, 0, NULL);
+
+	DBUG_RETURN(error);
+}
+
+/*********************************************************************
+Removes all tables in the named database inside InnoDB. */
+static
+void
+innobase_drop_database(
+/*===================*/
+			/* out: error number */
+        handlerton *hton, /* in: handlerton of Innodb */
+	char*	path)	/* in: database path; inside InnoDB the name
+			of the last directory in the path is used as
+			the database name: for example, in 'mysql/data/test'
+			the database name is 'test' */
+{
+	ulint	len		= 0;
+	trx_t*	trx;
+	char*	ptr;
+	int	error;
+	char*	namebuf;
+	THD*	thd		= current_thd;
+
+	/* Get the transaction associated with the current thd, or create one
+	if not yet created */
+
+	DBUG_ASSERT(hton == innodb_hton_ptr);
+
+	/* In the Windows plugin, thd = current_thd is always NULL */
+	if (thd) {
+		trx_t*	parent_trx = check_trx_exists(thd);
+
+		/* In case MySQL calls this in the middle of a SELECT
+		query, release possible adaptive hash latch to avoid
+		deadlocks of threads */
+
+		trx_search_latch_release_if_reserved(parent_trx);
+	}
+
+	ptr = strend(path) - 2;
+
+	while (ptr >= path && *ptr != '\\' && *ptr != '/') {
+		ptr--;
+		len++;
+	}
+
+	ptr++;
+	namebuf = (char*) my_malloc((uint) len + 2, MYF(0));
+
+	memcpy(namebuf, ptr, len);
+	namebuf[len] = '/';
+	namebuf[len + 1] = '\0';
+#ifdef	__WIN__
+	innobase_casedn_str(namebuf);
+#endif
+#if defined __WIN__ && !defined MYSQL_SERVER
+	/* In the Windows plugin, thd = current_thd is always NULL */
+	trx = trx_allocate_for_mysql();
+	trx->mysql_thd = NULL;
+	trx->mysql_query_str = NULL;
+#else
+	trx = innobase_trx_allocate(thd);
+#endif
+	error = row_drop_database_for_mysql(namebuf, trx);
+	my_free(namebuf, MYF(0));
+
+	/* Flush the log to reduce probability that the .frm files and
+	the InnoDB data dictionary get out-of-sync if the user runs
+	with innodb_flush_log_at_trx_commit = 0 */
+
+	log_buffer_flush_to_disk();
+
+	/* Tell the InnoDB server that there might be work for
+	utility threads: */
+
+	srv_active_wake_master_thread();
+
+	innobase_commit_low(trx);
+	trx_free_for_mysql(trx);
+}
+/*************************************************************************
+Renames an InnoDB table. */
+static
+int
+innobase_rename_table(
+/*==================*/
+				/* out: 0 or error code */
+	trx_t*		trx,	/* in: transaction */
+	const char*	from,	/* in: old name of the table */
+	const char*	to,	/* in: new name of the table */
+	ibool		lock_and_commit)
+				/* in: TRUE=lock data dictionary and commit */
+{
+	int	error;
+	char*	norm_to;
+	char*	norm_from;
+
+	if (lower_case_table_names) {
+		srv_lower_case_table_names = TRUE;
+	} else {
+		srv_lower_case_table_names = FALSE;
+	}
+
+	// Magic number 64 arbitrary
+	norm_to = (char*) my_malloc(strlen(to) + 64, MYF(0));
+	norm_from = (char*) my_malloc(strlen(from) + 64, MYF(0));
+
+	normalize_table_name(norm_to, to);
+	normalize_table_name(norm_from, from);
+
+	/* Serialize data dictionary operations with dictionary mutex:
+	no deadlocks can occur then in these operations */
+
+	if (lock_and_commit) {
+		row_mysql_lock_data_dictionary(trx);
+	}
+
+	error = row_rename_table_for_mysql(
+		norm_from, norm_to, trx, lock_and_commit);
+
+	if (error != DB_SUCCESS) {
+		FILE* ef = dict_foreign_err_file;
+
+		fputs("InnoDB: Renaming table ", ef);
+		ut_print_name(ef, trx, TRUE, norm_from);
+		fputs(" to ", ef);
+		ut_print_name(ef, trx, TRUE, norm_to);
+		fputs(" failed!\n", ef);
+	}
+
+	if (lock_and_commit) {
+		row_mysql_unlock_data_dictionary(trx);
+
+		/* Flush the log to reduce probability that the .frm
+		files and the InnoDB data dictionary get out-of-sync
+		if the user runs with innodb_flush_log_at_trx_commit = 0 */
+
+		log_buffer_flush_to_disk();
+	}
+
+	my_free(norm_to, MYF(0));
+	my_free(norm_from, MYF(0));
+
+	return error;
+}
+/*************************************************************************
+Renames an InnoDB table. */
+UNIV_INTERN
+int
+ha_innobase::rename_table(
+/*======================*/
+				/* out: 0 or error code */
+	const char*	from,	/* in: old name of the table */
+	const char*	to)	/* in: new name of the table */
+{
+	trx_t*	trx;
+	int	error;
+	trx_t*	parent_trx;
+	THD*	thd		= ha_thd();
+
+	DBUG_ENTER("ha_innobase::rename_table");
+
+	/* Get the transaction associated with the current thd, or create one
+	if not yet created */
+
+	parent_trx = check_trx_exists(thd);
+
+	/* In case MySQL calls this in the middle of a SELECT query, release
+	possible adaptive hash latch to avoid deadlocks of threads */
+
+	trx_search_latch_release_if_reserved(parent_trx);
+
+	trx = innobase_trx_allocate(thd);
+
+	error = innobase_rename_table(trx, from, to, TRUE);
+
+	/* Tell the InnoDB server that there might be work for
+	utility threads: */
+
+	srv_active_wake_master_thread();
+
+	innobase_commit_low(trx);
+	trx_free_for_mysql(trx);
+
+	error = convert_error_code_to_mysql(error, 0, NULL);
+
+	DBUG_RETURN(error);
+}
+
+/*************************************************************************
+Estimates the number of index records in a range. */
+UNIV_INTERN
+ha_rows
+ha_innobase::records_in_range(
+/*==========================*/
+						/* out: estimated number of
+						rows */
+	uint			keynr,		/* in: index number */
+	key_range		*min_key,	/* in: start key value of the
+						   range, may also be 0 */
+	key_range		*max_key)	/* in: range end key val, may
+						   also be 0 */
+{
+	KEY*		key;
+	dict_index_t*	index;
+	uchar*		key_val_buff2	= (uchar*) my_malloc(
+						  table->s->reclength
+					+ table->s->max_key_length + 100,
+								MYF(MY_FAE));
+	ulint		buff2_len = table->s->reclength
+					+ table->s->max_key_length + 100;
+	dtuple_t*	range_start;
+	dtuple_t*	range_end;
+	ib_int64_t	n_rows;
+	ulint		mode1;
+	ulint		mode2;
+	mem_heap_t*	heap;
+
+	DBUG_ENTER("records_in_range");
+
+	ut_a(prebuilt->trx == thd_to_trx(ha_thd()));
+
+	prebuilt->trx->op_info = (char*)"estimating records in index range";
+
+	/* In case MySQL calls this in the middle of a SELECT query, release
+	possible adaptive hash latch to avoid deadlocks of threads */
+
+	trx_search_latch_release_if_reserved(prebuilt->trx);
+
+	active_index = keynr;
+
+	key = table->key_info + active_index;
+
+	index = dict_table_get_index_on_name(prebuilt->table, key->name);
+
+	/* MySQL knows about this index and so we must be able to find it.*/
+	ut_a(index);
+
+	heap = mem_heap_create(2 * (key->key_parts * sizeof(dfield_t)
+				    + sizeof(dtuple_t)));
+
+	range_start = dtuple_create(heap, key->key_parts);
+	dict_index_copy_types(range_start, index, key->key_parts);
+
+	range_end = dtuple_create(heap, key->key_parts);
+	dict_index_copy_types(range_end, index, key->key_parts);
+
+	row_sel_convert_mysql_key_to_innobase(
+				range_start, (byte*) key_val_buff,
+				(ulint)upd_and_key_val_buff_len,
+				index,
+				(byte*) (min_key ? min_key->key :
+					 (const uchar*) 0),
+				(ulint) (min_key ? min_key->length : 0),
+				prebuilt->trx);
+
+	row_sel_convert_mysql_key_to_innobase(
+				range_end, (byte*) key_val_buff2,
+				buff2_len, index,
+				(byte*) (max_key ? max_key->key :
+					 (const uchar*) 0),
+				(ulint) (max_key ? max_key->length : 0),
+				prebuilt->trx);
+
+	mode1 = convert_search_mode_to_innobase(min_key ? min_key->flag :
+						HA_READ_KEY_EXACT);
+	mode2 = convert_search_mode_to_innobase(max_key ? max_key->flag :
+						HA_READ_KEY_EXACT);
+
+	if (mode1 != PAGE_CUR_UNSUPP && mode2 != PAGE_CUR_UNSUPP) {
+
+		n_rows = btr_estimate_n_rows_in_range(index, range_start,
+						      mode1, range_end,
+						      mode2);
+	} else {
+
+		n_rows = 0;
+	}
+
+	mem_heap_free(heap);
+
+	my_free(key_val_buff2, MYF(0));
+
+	prebuilt->trx->op_info = (char*)"";
+
+	/* The MySQL optimizer seems to believe an estimate of 0 rows is
+	always accurate and may return the result 'Empty set' based on that.
+	The accuracy is not guaranteed, and even if it were, for a locking
+	read we should anyway perform the search to set the next-key lock.
+	Add 1 to the value to make sure MySQL does not make the assumption! */
+
+	if (n_rows == 0) {
+		n_rows = 1;
+	}
+
+	DBUG_RETURN((ha_rows) n_rows);
+}
+
+/*************************************************************************
+Gives an UPPER BOUND to the number of rows in a table. This is used in
+filesort.cc. */
+UNIV_INTERN
+ha_rows
+ha_innobase::estimate_rows_upper_bound(void)
+/*======================================*/
+			/* out: upper bound of rows */
+{
+	dict_index_t*	index;
+	ulonglong	estimate;
+	ulonglong	local_data_file_length;
+
+	DBUG_ENTER("estimate_rows_upper_bound");
+
+	/* We do not know if MySQL can call this function before calling
+	external_lock(). To be safe, update the thd of the current table
+	handle. */
+
+	update_thd(ha_thd());
+
+	prebuilt->trx->op_info = (char*)
+				 "calculating upper bound for table rows";
+
+	/* In case MySQL calls this in the middle of a SELECT query, release
+	possible adaptive hash latch to avoid deadlocks of threads */
+
+	trx_search_latch_release_if_reserved(prebuilt->trx);
+
+	index = dict_table_get_first_index(prebuilt->table);
+
+	ut_a(index->stat_n_leaf_pages > 0);
+
+	local_data_file_length =
+		((ulonglong) index->stat_n_leaf_pages) * UNIV_PAGE_SIZE;
+
+
+	/* Calculate a minimum length for a clustered index record and from
+	that an upper bound for the number of rows. Since we only calculate
+	new statistics in row0mysql.c when a table has grown by a threshold
+	factor, we must add a safety factor 2 in front of the formula below. */
+
+	estimate = 2 * local_data_file_length /
+					 dict_index_calc_min_rec_len(index);
+
+	prebuilt->trx->op_info = (char*)"";
+
+	DBUG_RETURN((ha_rows) estimate);
+}
+
+/*************************************************************************
+How many seeks it will take to read through the table. This is to be
+comparable to the number returned by records_in_range so that we can
+decide if we should scan the table or use keys. */
+UNIV_INTERN
+double
+ha_innobase::scan_time()
+/*====================*/
+			/* out: estimated time measured in disk seeks */
+{
+	/* Since MySQL seems to favor table scans too much over index
+	searches, we pretend that a sequential read takes the same time
+	as a random disk read, that is, we do not divide the following
+	by 10, which would be physically realistic. */
+
+	return((double) (prebuilt->table->stat_clustered_index_size));
+}
+
+/**********************************************************************
+Calculate the time it takes to read a set of ranges through an index
+This enables us to optimise reads for clustered indexes. */
+UNIV_INTERN
+double
+ha_innobase::read_time(
+/*===================*/
+			/* out: estimated time measured in disk seeks */
+	uint	index,	/* in: key number */
+	uint	ranges,	/* in: how many ranges */
+	ha_rows rows)	/* in: estimated number of rows in the ranges */
+{
+	ha_rows total_rows;
+	double	time_for_scan;
+
+	if (index != table->s->primary_key) {
+		/* Not clustered */
+		return(handler::read_time(index, ranges, rows));
+	}
+
+	if (rows <= 2) {
+
+		return((double) rows);
+	}
+
+	/* Assume that the read time is proportional to the scan time for all
+	rows + at most one seek per range. */
+
+	time_for_scan = scan_time();
+
+	if ((total_rows = estimate_rows_upper_bound()) < rows) {
+
+		return(time_for_scan);
+	}
+
+	return(ranges + (double) rows / (double) total_rows * time_for_scan);
+}
+
+/*************************************************************************
+Returns statistics information of the table to the MySQL interpreter,
+in various fields of the handle object. */
+UNIV_INTERN
+int
+ha_innobase::info(
+/*==============*/
+	uint flag)	/* in: what information MySQL requests */
+{
+	dict_table_t*	ib_table;
+	dict_index_t*	index;
+	ha_rows		rec_per_key;
+	ib_int64_t	n_rows;
+	ulong		j;
+	ulong		i;
+	char		path[FN_REFLEN];
+	os_file_stat_t	stat_info;
+
+	DBUG_ENTER("info");
+
+	/* If we are forcing recovery at a high level, we will suppress
+	statistics calculation on tables, because that may crash the
+	server if an index is badly corrupted. */
+
+	if (srv_force_recovery >= SRV_FORCE_NO_IBUF_MERGE) {
+
+		/* We return success (0) instead of HA_ERR_CRASHED,
+		because we want MySQL to process this query and not
+		stop, like it would do if it received the error code
+		HA_ERR_CRASHED. */
+
+		DBUG_RETURN(0);
+	}
+
+	/* We do not know if MySQL can call this function before calling
+	external_lock(). To be safe, update the thd of the current table
+	handle. */
+
+	update_thd(ha_thd());
+
+	/* In case MySQL calls this in the middle of a SELECT query, release
+	possible adaptive hash latch to avoid deadlocks of threads */
+
+	prebuilt->trx->op_info = (char*)"returning various info to MySQL";
+
+	trx_search_latch_release_if_reserved(prebuilt->trx);
+
+	ib_table = prebuilt->table;
+
+	if (flag & HA_STATUS_TIME) {
+		if (innobase_stats_on_metadata) {
+			/* In sql_show we call with this flag: update
+			then statistics so that they are up-to-date */
+
+			prebuilt->trx->op_info = "updating table statistics";
+
+			dict_update_statistics(ib_table);
+
+			prebuilt->trx->op_info = "returning various info to MySQL";
+		}
+
+		my_snprintf(path, sizeof(path), "%s/%s%s",
+				mysql_data_home, ib_table->name, reg_ext);
+
+		unpack_filename(path,path);
+
+		/* Note that we do not know the access time of the table,
+		nor the CHECK TABLE time, nor the UPDATE or INSERT time. */
+
+		if (os_file_get_status(path,&stat_info)) {
+			stats.create_time = stat_info.ctime;
+		}
+	}
+
+	if (flag & HA_STATUS_VARIABLE) {
+		n_rows = ib_table->stat_n_rows;
+
+		/* Because we do not protect stat_n_rows by any mutex in a
+		delete, it is theoretically possible that the value can be
+		smaller than zero! TODO: fix this race.
+
+		The MySQL optimizer seems to assume in a left join that n_rows
+		is an accurate estimate if it is zero. Of course, it is not,
+		since we do not have any locks on the rows yet at this phase.
+		Since SHOW TABLE STATUS seems to call this function with the
+		HA_STATUS_TIME flag set, while the left join optimizer does not
+		set that flag, we add one to a zero value if the flag is not
+		set. That way SHOW TABLE STATUS will show the best estimate,
+		while the optimizer never sees the table empty. */
+
+		if (n_rows < 0) {
+			n_rows = 0;
+		}
+
+		if (n_rows == 0 && !(flag & HA_STATUS_TIME)) {
+			n_rows++;
+		}
+
+		/* Fix bug#40386: Not flushing query cache after truncate.
+		n_rows can not be 0 unless the table is empty, set to 1
+		instead. The original problem of bug#29507 is actually
+		fixed in the server code. */
+		if (thd_sql_command(user_thd) == SQLCOM_TRUNCATE) {
+
+			n_rows = 1;
+
+			/* We need to reset the prebuilt value too, otherwise
+			checks for values greater than the last value written
+			to the table will fail and the autoinc counter will
+			not be updated. This will force write_row() into
+			attempting an update of the table's AUTOINC counter. */
+
+			prebuilt->autoinc_last_value = 0;
+		}
+
+		stats.records = (ha_rows)n_rows;
+		stats.deleted = 0;
+		stats.data_file_length = ((ulonglong)
+				ib_table->stat_clustered_index_size)
+					* UNIV_PAGE_SIZE;
+		stats.index_file_length = ((ulonglong)
+				ib_table->stat_sum_of_other_index_sizes)
+					* UNIV_PAGE_SIZE;
+
+		/* Since fsp_get_available_space_in_free_extents() is
+		acquiring latches inside InnoDB, we do not call it if we
+		are asked by MySQL to avoid locking. Another reason to
+		avoid the call is that it uses quite a lot of CPU.
+		See Bug#38185.
+		We do not update delete_length if no locking is requested
+		so the "old" value can remain. delete_length is initialized
+		to 0 in the ha_statistics' constructor. */
+		if (!(flag & HA_STATUS_NO_LOCK)) {
+
+			/* lock the data dictionary to avoid races with
+			ibd_file_missing and tablespace_discarded */
+			row_mysql_lock_data_dictionary(prebuilt->trx);
+
+			/* ib_table->space must be an existent tablespace */
+			if (!ib_table->ibd_file_missing
+			    && !ib_table->tablespace_discarded) {
+
+				stats.delete_length =
+					fsp_get_available_space_in_free_extents(
+						ib_table->space) * 1024;
+			} else {
+
+				THD*	thd;
+
+				thd = ha_thd();
+
+				push_warning_printf(
+					thd,
+					MYSQL_ERROR::WARN_LEVEL_WARN,
+					ER_CANT_GET_STAT,
+					"InnoDB: Trying to get the free "
+					"space for table %s but its "
+					"tablespace has been discarded or "
+					"the .ibd file is missing. Setting "
+					"the free space to zero.",
+					ib_table->name);
+
+				stats.delete_length = 0;
+			}
+
+			row_mysql_unlock_data_dictionary(prebuilt->trx);
+		}
+
+		stats.check_time = 0;
+
+		if (stats.records == 0) {
+			stats.mean_rec_length = 0;
+		} else {
+			stats.mean_rec_length = (ulong) (stats.data_file_length / stats.records);
+		}
+	}
+
+	if (flag & HA_STATUS_CONST) {
+		index = dict_table_get_first_index(ib_table);
+
+		if (prebuilt->clust_index_was_generated) {
+			index = dict_table_get_next_index(index);
+		}
+
+		for (i = 0; i < table->s->keys; i++) {
+			if (index == NULL) {
+				sql_print_error("Table %s contains fewer "
+						"indexes inside InnoDB than "
+						"are defined in the MySQL "
+						".frm file. Have you mixed up "
+						".frm files from different "
+						"installations? See "
+"http://dev.mysql.com/doc/refman/5.1/en/innodb-troubleshooting.html\n",
+
+						ib_table->name);
+				break;
+			}
+
+			for (j = 0; j < table->key_info[i].key_parts; j++) {
+
+				if (j + 1 > index->n_uniq) {
+					sql_print_error(
+"Index %s of %s has %lu columns unique inside InnoDB, but MySQL is asking "
+"statistics for %lu columns. Have you mixed up .frm files from different "
+"installations? "
+"See http://dev.mysql.com/doc/refman/5.1/en/innodb-troubleshooting.html\n",
+							index->name,
+							ib_table->name,
+							(unsigned long)
+							index->n_uniq, j + 1);
+					break;
+				}
+
+				if (index->stat_n_diff_key_vals[j + 1] == 0) {
+
+					rec_per_key = stats.records;
+				} else {
+					rec_per_key = (ha_rows)(stats.records /
+					 index->stat_n_diff_key_vals[j + 1]);
+				}
+
+				/* Since MySQL seems to favor table scans
+				too much over index searches, we pretend
+				index selectivity is 2 times better than
+				our estimate: */
+
+				rec_per_key = rec_per_key / 2;
+
+				if (rec_per_key == 0) {
+					rec_per_key = 1;
+				}
+
+				table->key_info[i].rec_per_key[j]=
+				  rec_per_key >= ~(ulong) 0 ? ~(ulong) 0 :
+				  (ulong) rec_per_key;
+			}
+
+			index = dict_table_get_next_index(index);
+		}
+	}
+
+	if (flag & HA_STATUS_ERRKEY) {
+		const dict_index_t*	err_index;
+
+		ut_a(prebuilt->trx);
+		ut_a(prebuilt->trx->magic_n == TRX_MAGIC_N);
+
+		err_index = trx_get_error_info(prebuilt->trx);
+
+		if (err_index) {
+			errkey = (unsigned int)
+				row_get_mysql_key_number_for_index(err_index);
+		} else {
+			errkey = (unsigned int) prebuilt->trx->error_key_num;
+		}
+	}
+
+	if ((flag & HA_STATUS_AUTO) && table->found_next_number_field) {
+		stats.auto_increment_value = innobase_peek_autoinc();
+	}
+
+	prebuilt->trx->op_info = (char*)"";
+
+	DBUG_RETURN(0);
+}
+
+/**************************************************************************
+Updates index cardinalities of the table, based on 8 random dives into
+each index tree. This does NOT calculate exact statistics on the table. */
+UNIV_INTERN
+int
+ha_innobase::analyze(
+/*=================*/
+					/* out: returns always 0 (success) */
+	THD*		thd,		/* in: connection thread handle */
+	HA_CHECK_OPT*	check_opt)	/* in: currently ignored */
+{
+	/* Simply call ::info() with all the flags */
+	info(HA_STATUS_TIME | HA_STATUS_CONST | HA_STATUS_VARIABLE);
+
+	return(0);
+}
+
+/**************************************************************************
+This is mapped to "ALTER TABLE tablename ENGINE=InnoDB", which rebuilds
+the table in MySQL. */
+UNIV_INTERN
+int
+ha_innobase::optimize(
+/*==================*/
+	THD*		thd,		/* in: connection thread handle */
+	HA_CHECK_OPT*	check_opt)	/* in: currently ignored */
+{
+	return(HA_ADMIN_TRY_ALTER);
+}
+
+/***********************************************************************
+Tries to check that an InnoDB table is not corrupted. If corruption is
+noticed, prints to stderr information about it. In case of corruption
+may also assert a failure and crash the server. */
+UNIV_INTERN
+int
+ha_innobase::check(
+/*===============*/
+					/* out: HA_ADMIN_CORRUPT or
+					HA_ADMIN_OK */
+	THD*		thd,		/* in: user thread handle */
+	HA_CHECK_OPT*	check_opt)	/* in: check options, currently
+					ignored */
+{
+	ulint		ret;
+
+	DBUG_ASSERT(thd == ha_thd());
+	ut_a(prebuilt->trx);
+	ut_a(prebuilt->trx->magic_n == TRX_MAGIC_N);
+	ut_a(prebuilt->trx == thd_to_trx(thd));
+
+	if (prebuilt->mysql_template == NULL) {
+		/* Build the template; we will use a dummy template
+		in index scans done in checking */
+
+		build_template(prebuilt, NULL, table, ROW_MYSQL_WHOLE_ROW);
+	}
+
+	ret = row_check_table_for_mysql(prebuilt);
+
+	if (ret == DB_SUCCESS) {
+		return(HA_ADMIN_OK);
+	}
+
+	return(HA_ADMIN_CORRUPT);
+}
+
+/*****************************************************************
+Adds information about free space in the InnoDB tablespace to a table comment
+which is printed out when a user calls SHOW TABLE STATUS. Adds also info on
+foreign keys. */
+UNIV_INTERN
+char*
+ha_innobase::update_table_comment(
+/*==============================*/
+				/* out: table comment + InnoDB free space +
+				info on foreign keys */
+	const char*	comment)/* in: table comment defined by user */
+{
+	uint	length = (uint) strlen(comment);
+	char*	str;
+	long	flen;
+
+	/* We do not know if MySQL can call this function before calling
+	external_lock(). To be safe, update the thd of the current table
+	handle. */
+
+	if (length > 64000 - 3) {
+		return((char*)comment); /* string too long */
+	}
+
+	update_thd(ha_thd());
+
+	prebuilt->trx->op_info = (char*)"returning table comment";
+
+	/* In case MySQL calls this in the middle of a SELECT query, release
+	possible adaptive hash latch to avoid deadlocks of threads */
+
+	trx_search_latch_release_if_reserved(prebuilt->trx);
+	str = NULL;
+
+	/* output the data to a temporary file */
+
+	mutex_enter(&srv_dict_tmpfile_mutex);
+	rewind(srv_dict_tmpfile);
+
+	fprintf(srv_dict_tmpfile, "InnoDB free: %llu kB",
+		fsp_get_available_space_in_free_extents(
+			prebuilt->table->space));
+
+	dict_print_info_on_foreign_keys(FALSE, srv_dict_tmpfile,
+				prebuilt->trx, prebuilt->table);
+	flen = ftell(srv_dict_tmpfile);
+	if (flen < 0) {
+		flen = 0;
+	} else if (length + flen + 3 > 64000) {
+		flen = 64000 - 3 - length;
+	}
+
+	/* allocate buffer for the full string, and
+	read the contents of the temporary file */
+
+	str = (char*) my_malloc(length + flen + 3, MYF(0));
+
+	if (str) {
+		char* pos	= str + length;
+		if (length) {
+			memcpy(str, comment, length);
+			*pos++ = ';';
+			*pos++ = ' ';
+		}
+		rewind(srv_dict_tmpfile);
+		flen = (uint) fread(pos, 1, flen, srv_dict_tmpfile);
+		pos[flen] = 0;
+	}
+
+	mutex_exit(&srv_dict_tmpfile_mutex);
+
+	prebuilt->trx->op_info = (char*)"";
+
+	return(str ? str : (char*) comment);
+}
+
+/***********************************************************************
+Gets the foreign key create info for a table stored in InnoDB. */
+UNIV_INTERN
+char*
+ha_innobase::get_foreign_key_create_info(void)
+/*==========================================*/
+			/* out, own: character string in the form which
+			can be inserted to the CREATE TABLE statement,
+			MUST be freed with ::free_foreign_key_create_info */
+{
+	char*	str	= 0;
+	long	flen;
+
+	ut_a(prebuilt != NULL);
+
+	/* We do not know if MySQL can call this function before calling
+	external_lock(). To be safe, update the thd of the current table
+	handle. */
+
+	update_thd(ha_thd());
+
+	prebuilt->trx->op_info = (char*)"getting info on foreign keys";
+
+	/* In case MySQL calls this in the middle of a SELECT query,
+	release possible adaptive hash latch to avoid
+	deadlocks of threads */
+
+	trx_search_latch_release_if_reserved(prebuilt->trx);
+
+	mutex_enter(&srv_dict_tmpfile_mutex);
+	rewind(srv_dict_tmpfile);
+
+	/* output the data to a temporary file */
+	dict_print_info_on_foreign_keys(TRUE, srv_dict_tmpfile,
+				prebuilt->trx, prebuilt->table);
+	prebuilt->trx->op_info = (char*)"";
+
+	flen = ftell(srv_dict_tmpfile);
+	if (flen < 0) {
+		flen = 0;
+	} else if (flen > 64000 - 1) {
+		flen = 64000 - 1;
+	}
+
+	/* allocate buffer for the string, and
+	read the contents of the temporary file */
+
+	str = (char*) my_malloc(flen + 1, MYF(0));
+
+	if (str) {
+		rewind(srv_dict_tmpfile);
+		flen = (uint) fread(str, 1, flen, srv_dict_tmpfile);
+		str[flen] = 0;
+	}
+
+	mutex_exit(&srv_dict_tmpfile_mutex);
+
+	return(str);
+}
+
+
+UNIV_INTERN
+int
+ha_innobase::get_foreign_key_list(THD *thd, List<FOREIGN_KEY_INFO> *f_key_list)
+{
+  dict_foreign_t* foreign;
+
+  DBUG_ENTER("get_foreign_key_list");
+  ut_a(prebuilt != NULL);
+  update_thd(ha_thd());
+  prebuilt->trx->op_info = (char*)"getting list of foreign keys";
+  trx_search_latch_release_if_reserved(prebuilt->trx);
+  mutex_enter(&(dict_sys->mutex));
+  foreign = UT_LIST_GET_FIRST(prebuilt->table->foreign_list);
+
+  while (foreign != NULL) {
+	  uint i;
+	  FOREIGN_KEY_INFO f_key_info;
+	  LEX_STRING *name= 0;
+          uint ulen;
+          char uname[NAME_LEN+1];           /* Unencoded name */
+          char db_name[NAME_LEN+1];
+	  const char *tmp_buff;
+
+	  tmp_buff= foreign->id;
+	  i= 0;
+	  while (tmp_buff[i] != '/')
+		  i++;
+	  tmp_buff+= i + 1;
+	  f_key_info.forein_id = thd_make_lex_string(thd, 0,
+		  tmp_buff, (uint) strlen(tmp_buff), 1);
+	  tmp_buff= foreign->referenced_table_name;
+
+          /* Database name */
+	  i= 0;
+	  while (tmp_buff[i] != '/')
+          {
+            db_name[i]= tmp_buff[i];
+            i++;
+          }
+          db_name[i]= 0;
+          ulen= filename_to_tablename(db_name, uname, sizeof(uname));
+	  f_key_info.referenced_db = thd_make_lex_string(thd, 0,
+		  uname, ulen, 1);
+
+          /* Table name */
+	  tmp_buff+= i + 1;
+          ulen= filename_to_tablename(tmp_buff, uname, sizeof(uname));
+	  f_key_info.referenced_table = thd_make_lex_string(thd, 0,
+		  uname, ulen, 1);
+
+	  for (i= 0;;) {
+		  tmp_buff= foreign->foreign_col_names[i];
+		  name = thd_make_lex_string(thd, name,
+			  tmp_buff, (uint) strlen(tmp_buff), 1);
+		  f_key_info.foreign_fields.push_back(name);
+		  tmp_buff= foreign->referenced_col_names[i];
+		  name = thd_make_lex_string(thd, name,
+			tmp_buff, (uint) strlen(tmp_buff), 1);
+		  f_key_info.referenced_fields.push_back(name);
+		  if (++i >= foreign->n_fields)
+			  break;
+	  }
+
+          ulong length;
+          if (foreign->type & DICT_FOREIGN_ON_DELETE_CASCADE)
+          {
+            length=7;
+            tmp_buff= "CASCADE";
+          }
+          else if (foreign->type & DICT_FOREIGN_ON_DELETE_SET_NULL)
+          {
+            length=8;
+            tmp_buff= "SET NULL";
+          }
+          else if (foreign->type & DICT_FOREIGN_ON_DELETE_NO_ACTION)
+          {
+            length=9;
+            tmp_buff= "NO ACTION";
+          }
+          else
+          {
+            length=8;
+            tmp_buff= "RESTRICT";
+          }
+	  f_key_info.delete_method = thd_make_lex_string(
+		  thd, f_key_info.delete_method, tmp_buff, length, 1);
+
+
+          if (foreign->type & DICT_FOREIGN_ON_UPDATE_CASCADE)
+          {
+            length=7;
+            tmp_buff= "CASCADE";
+          }
+          else if (foreign->type & DICT_FOREIGN_ON_UPDATE_SET_NULL)
+          {
+            length=8;
+            tmp_buff= "SET NULL";
+          }
+          else if (foreign->type & DICT_FOREIGN_ON_UPDATE_NO_ACTION)
+          {
+            length=9;
+            tmp_buff= "NO ACTION";
+          }
+          else
+          {
+            length=8;
+            tmp_buff= "RESTRICT";
+          }
+	  f_key_info.update_method = thd_make_lex_string(
+		  thd, f_key_info.update_method, tmp_buff, length, 1);
+          if (foreign->referenced_index &&
+              foreign->referenced_index->name)
+          {
+	    f_key_info.referenced_key_name = thd_make_lex_string(
+		    thd, f_key_info.referenced_key_name,
+		    foreign->referenced_index->name,
+		    strlen(foreign->referenced_index->name), 1);
+          }
+          else
+            f_key_info.referenced_key_name= 0;
+
+	  FOREIGN_KEY_INFO *pf_key_info = (FOREIGN_KEY_INFO *)
+		  thd_memdup(thd, &f_key_info, sizeof(FOREIGN_KEY_INFO));
+	  f_key_list->push_back(pf_key_info);
+	  foreign = UT_LIST_GET_NEXT(foreign_list, foreign);
+  }
+  mutex_exit(&(dict_sys->mutex));
+  prebuilt->trx->op_info = (char*)"";
+
+  DBUG_RETURN(0);
+}
+
+/*********************************************************************
+Checks if ALTER TABLE may change the storage engine of the table.
+Changing storage engines is not allowed for tables for which there
+are foreign key constraints (parent or child tables). */
+UNIV_INTERN
+bool
+ha_innobase::can_switch_engines(void)
+/*=================================*/
+{
+	bool	can_switch;
+
+	DBUG_ENTER("ha_innobase::can_switch_engines");
+
+	ut_a(prebuilt->trx == thd_to_trx(ha_thd()));
+
+	prebuilt->trx->op_info =
+			"determining if there are foreign key constraints";
+	row_mysql_lock_data_dictionary(prebuilt->trx);
+
+	can_switch = !UT_LIST_GET_FIRST(prebuilt->table->referenced_list)
+			&& !UT_LIST_GET_FIRST(prebuilt->table->foreign_list);
+
+	row_mysql_unlock_data_dictionary(prebuilt->trx);
+	prebuilt->trx->op_info = "";
+
+	DBUG_RETURN(can_switch);
+}
+
+/***********************************************************************
+Checks if a table is referenced by a foreign key. The MySQL manual states that
+a REPLACE is either equivalent to an INSERT, or DELETE(s) + INSERT. Only a
+delete is then allowed internally to resolve a duplicate key conflict in
+REPLACE, not an update. */
+UNIV_INTERN
+uint
+ha_innobase::referenced_by_foreign_key(void)
+/*========================================*/
+			/* out: > 0 if referenced by a FOREIGN KEY */
+{
+	if (dict_table_is_referenced_by_foreign_key(prebuilt->table)) {
+
+		return(1);
+	}
+
+	return(0);
+}
+
+/***********************************************************************
+Frees the foreign key create info for a table stored in InnoDB, if it is
+non-NULL. */
+UNIV_INTERN
+void
+ha_innobase::free_foreign_key_create_info(
+/*======================================*/
+	char*	str)	/* in, own: create info string to free	*/
+{
+	if (str) {
+		my_free(str, MYF(0));
+	}
+}
+
+/***********************************************************************
+Tells something additional to the handler about how to do things. */
+UNIV_INTERN
+int
+ha_innobase::extra(
+/*===============*/
+			   /* out: 0 or error number */
+	enum ha_extra_function operation)
+			   /* in: HA_EXTRA_FLUSH or some other flag */
+{
+	/* Warning: since it is not sure that MySQL calls external_lock
+	before calling this function, the trx field in prebuilt can be
+	obsolete! */
+
+	switch (operation) {
+		case HA_EXTRA_FLUSH:
+			if (prebuilt->blob_heap) {
+				row_mysql_prebuilt_free_blob_heap(prebuilt);
+			}
+			break;
+		case HA_EXTRA_RESET_STATE:
+			reset_template(prebuilt);
+			break;
+		case HA_EXTRA_NO_KEYREAD:
+			prebuilt->read_just_key = 0;
+			break;
+		case HA_EXTRA_KEYREAD:
+			prebuilt->read_just_key = 1;
+			break;
+		case HA_EXTRA_KEYREAD_PRESERVE_FIELDS:
+			prebuilt->keep_other_fields_on_keyread = 1;
+			break;
+
+			/* IMPORTANT: prebuilt->trx can be obsolete in
+			this method, because it is not sure that MySQL
+			calls external_lock before this method with the
+			parameters below.  We must not invoke update_thd()
+			either, because the calling threads may change.
+			CAREFUL HERE, OR MEMORY CORRUPTION MAY OCCUR! */
+		case HA_EXTRA_IGNORE_DUP_KEY:
+			thd_to_trx(ha_thd())->duplicates |= TRX_DUP_IGNORE;
+			break;
+		case HA_EXTRA_WRITE_CAN_REPLACE:
+			thd_to_trx(ha_thd())->duplicates |= TRX_DUP_REPLACE;
+			break;
+		case HA_EXTRA_WRITE_CANNOT_REPLACE:
+			thd_to_trx(ha_thd())->duplicates &= ~TRX_DUP_REPLACE;
+			break;
+		case HA_EXTRA_NO_IGNORE_DUP_KEY:
+			thd_to_trx(ha_thd())->duplicates &=
+				~(TRX_DUP_IGNORE | TRX_DUP_REPLACE);
+			break;
+		default:/* Do nothing */
+			;
+	}
+
+	return(0);
+}
+
+UNIV_INTERN
+int
+ha_innobase::reset()
+{
+	if (prebuilt->blob_heap) {
+		row_mysql_prebuilt_free_blob_heap(prebuilt);
+	}
+
+	reset_template(prebuilt);
+
+	/* TODO: This should really be reset in reset_template() but for now
+	it's safer to do it explicitly here. */
+
+	/* This is a statement level counter. */
+	prebuilt->autoinc_last_value = 0;
+
+	return(0);
+}
+
+/**********************************************************************
+MySQL calls this function at the start of each SQL statement inside LOCK
+TABLES. Inside LOCK TABLES the ::external_lock method does not work to
+mark SQL statement borders. Note also a special case: if a temporary table
+is created inside LOCK TABLES, MySQL has not called external_lock() at all
+on that table.
+MySQL-5.0 also calls this before each statement in an execution of a stored
+procedure. To make the execution more deterministic for binlogging, MySQL-5.0
+locks all tables involved in a stored procedure with full explicit table
+locks (thd_in_lock_tables(thd) holds in store_lock()) before executing the
+procedure. */
+UNIV_INTERN
+int
+ha_innobase::start_stmt(
+/*====================*/
+				/* out: 0 or error code */
+	THD*		thd,	/* in: handle to the user thread */
+	thr_lock_type	lock_type)
+{
+	trx_t*		trx;
+
+	update_thd(thd);
+
+	trx = prebuilt->trx;
+
+	/* Here we release the search latch and the InnoDB thread FIFO ticket
+	if they were reserved. They should have been released already at the
+	end of the previous statement, but because inside LOCK TABLES the
+	lock count method does not work to mark the end of a SELECT statement,
+	that may not be the case. We MUST release the search latch before an
+	INSERT, for example. */
+
+	innobase_release_stat_resources(trx);
+
+	/* Reset the AUTOINC statement level counter for multi-row INSERTs. */
+	trx->n_autoinc_rows = 0;
+
+	prebuilt->sql_stat_start = TRUE;
+	prebuilt->hint_need_to_fetch_extra_cols = 0;
+	reset_template(prebuilt);
+
+	if (!prebuilt->mysql_has_locked) {
+		/* This handle is for a temporary table created inside
+		this same LOCK TABLES; since MySQL does NOT call external_lock
+		in this case, we must use x-row locks inside InnoDB to be
+		prepared for an update of a row */
+
+		prebuilt->select_lock_type = LOCK_X;
+	} else {
+		if (trx->isolation_level != TRX_ISO_SERIALIZABLE
+			&& thd_sql_command(thd) == SQLCOM_SELECT
+			&& lock_type == TL_READ) {
+
+			/* For other than temporary tables, we obtain
+			no lock for consistent read (plain SELECT). */
+
+			prebuilt->select_lock_type = LOCK_NONE;
+		} else {
+			/* Not a consistent read: restore the
+			select_lock_type value. The value of
+			stored_select_lock_type was decided in:
+			1) ::store_lock(),
+			2) ::external_lock(),
+			3) ::init_table_handle_for_HANDLER(), and
+			4) ::transactional_table_lock(). */
+
+			prebuilt->select_lock_type =
+				prebuilt->stored_select_lock_type;
+		}
+	}
+
+	trx->detailed_error[0] = '\0';
+
+	/* Set the MySQL flag to mark that there is an active transaction */
+	if (trx->active_trans == 0) {
+
+		innobase_register_trx_and_stmt(ht, thd);
+		trx->active_trans = 1;
+	} else {
+		innobase_register_stmt(ht, thd);
+	}
+
+	return(0);
+}
+
+/**********************************************************************
+Maps a MySQL trx isolation level code to the InnoDB isolation level code */
+static inline
+ulint
+innobase_map_isolation_level(
+/*=========================*/
+					/* out: InnoDB isolation level */
+	enum_tx_isolation	iso)	/* in: MySQL isolation level code */
+{
+	switch(iso) {
+		case ISO_REPEATABLE_READ: return(TRX_ISO_REPEATABLE_READ);
+		case ISO_READ_COMMITTED: return(TRX_ISO_READ_COMMITTED);
+		case ISO_SERIALIZABLE: return(TRX_ISO_SERIALIZABLE);
+		case ISO_READ_UNCOMMITTED: return(TRX_ISO_READ_UNCOMMITTED);
+		default: ut_a(0); return(0);
+	}
+}
+
+/**********************************************************************
+As MySQL will execute an external lock for every new table it uses when it
+starts to process an SQL statement (an exception is when MySQL calls
+start_stmt for the handle) we can use this function to store the pointer to
+the THD in the handle. We will also use this function to communicate
+to InnoDB that a new SQL statement has started and that we must store a
+savepoint to our transaction handle, so that we are able to roll back
+the SQL statement in case of an error. */
+UNIV_INTERN
+int
+ha_innobase::external_lock(
+/*=======================*/
+				/* out: 0 */
+	THD*	thd,		/* in: handle to the user thread */
+	int	lock_type)	/* in: lock type */
+{
+	trx_t*		trx;
+
+	DBUG_ENTER("ha_innobase::external_lock");
+	DBUG_PRINT("enter",("lock_type: %d", lock_type));
+
+	update_thd(thd);
+
+	/* Statement based binlogging does not work in isolation level
+	READ UNCOMMITTED and READ COMMITTED since the necessary
+	locks cannot be taken. In this case, we print an
+	informative error message and return with an error. */
+	if (lock_type == F_WRLCK)
+	{
+		ulong const binlog_format= thd_binlog_format(thd);
+		ulong const tx_isolation = thd_tx_isolation(ha_thd());
+		if (tx_isolation <= ISO_READ_COMMITTED &&
+		    binlog_format == BINLOG_FORMAT_STMT)
+		{
+			char buf[256];
+			my_snprintf(buf, sizeof(buf),
+				    "Transaction level '%s' in"
+				    " InnoDB is not safe for binlog mode '%s'",
+				    tx_isolation_names[tx_isolation],
+				    binlog_format_names[binlog_format]);
+			my_error(ER_BINLOG_LOGGING_IMPOSSIBLE, MYF(0), buf);
+			DBUG_RETURN(HA_ERR_LOGGING_IMPOSSIBLE);
+		}
+	}
+
+
+	trx = prebuilt->trx;
+
+	prebuilt->sql_stat_start = TRUE;
+	prebuilt->hint_need_to_fetch_extra_cols = 0;
+
+	reset_template(prebuilt);
+
+	if (lock_type == F_WRLCK) {
+
+		/* If this is a SELECT, then it is in UPDATE TABLE ...
+		or SELECT ... FOR UPDATE */
+		prebuilt->select_lock_type = LOCK_X;
+		prebuilt->stored_select_lock_type = LOCK_X;
+	}
+
+	if (lock_type != F_UNLCK) {
+		/* MySQL is setting a new table lock */
+
+		trx->detailed_error[0] = '\0';
+
+		/* Set the MySQL flag to mark that there is an active
+		transaction */
+		if (trx->active_trans == 0) {
+
+			innobase_register_trx_and_stmt(ht, thd);
+			trx->active_trans = 1;
+		} else if (trx->n_mysql_tables_in_use == 0) {
+			innobase_register_stmt(ht, thd);
+		}
+
+		if (trx->isolation_level == TRX_ISO_SERIALIZABLE
+			&& prebuilt->select_lock_type == LOCK_NONE
+			&& thd_test_options(thd,
+				OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN)) {
+
+			/* To get serializable execution, we let InnoDB
+			conceptually add 'LOCK IN SHARE MODE' to all SELECTs
+			which otherwise would have been consistent reads. An
+			exception is consistent reads in the AUTOCOMMIT=1 mode:
+			we know that they are read-only transactions, and they
+			can be serialized also if performed as consistent
+			reads. */
+
+			prebuilt->select_lock_type = LOCK_S;
+			prebuilt->stored_select_lock_type = LOCK_S;
+		}
+
+		/* Starting from 4.1.9, no InnoDB table lock is taken in LOCK
+		TABLES if AUTOCOMMIT=1. It does not make much sense to acquire
+		an InnoDB table lock if it is released immediately at the end
+		of LOCK TABLES, and InnoDB's table locks in that case cause
+		VERY easily deadlocks.
+
+		We do not set InnoDB table locks if user has not explicitly
+		requested a table lock. Note that thd_in_lock_tables(thd)
+		can hold in some cases, e.g., at the start of a stored
+		procedure call (SQLCOM_CALL). */
+
+		if (prebuilt->select_lock_type != LOCK_NONE) {
+
+			if (thd_sql_command(thd) == SQLCOM_LOCK_TABLES
+			    && THDVAR(thd, table_locks)
+			    && thd_test_options(thd, OPTION_NOT_AUTOCOMMIT)
+			    && thd_in_lock_tables(thd)) {
+
+				ulint	error = row_lock_table_for_mysql(
+					prebuilt, NULL, 0);
+
+				if (error != DB_SUCCESS) {
+					error = convert_error_code_to_mysql(
+						(int) error, 0, thd);
+					DBUG_RETURN((int) error);
+				}
+			}
+
+			trx->mysql_n_tables_locked++;
+		}
+
+		trx->n_mysql_tables_in_use++;
+		prebuilt->mysql_has_locked = TRUE;
+
+		DBUG_RETURN(0);
+	}
+
+	/* MySQL is releasing a table lock */
+
+	trx->n_mysql_tables_in_use--;
+	prebuilt->mysql_has_locked = FALSE;
+
+	/* Release a possible FIFO ticket and search latch. Since we
+	may reserve the kernel mutex, we have to release the search
+	system latch first to obey the latching order. */
+
+	innobase_release_stat_resources(trx);
+
+	/* If the MySQL lock count drops to zero we know that the current SQL
+	statement has ended */
+
+	if (trx->n_mysql_tables_in_use == 0) {
+
+		trx->mysql_n_tables_locked = 0;
+		prebuilt->used_in_HANDLER = FALSE;
+
+		if (!thd_test_options(thd, OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN)) {
+			if (trx->active_trans != 0) {
+				innobase_commit(ht, thd, TRUE);
+			}
+		} else {
+			if (trx->isolation_level <= TRX_ISO_READ_COMMITTED
+						&& trx->global_read_view) {
+
+				/* At low transaction isolation levels we let
+				each consistent read set its own snapshot */
+
+				read_view_close_for_mysql(trx);
+			}
+		}
+	}
+
+	DBUG_RETURN(0);
+}
+
+/**********************************************************************
+With this function MySQL request a transactional lock to a table when
+user issued query LOCK TABLES..WHERE ENGINE = InnoDB. */
+UNIV_INTERN
+int
+ha_innobase::transactional_table_lock(
+/*==================================*/
+				/* out: error code */
+	THD*	thd,		/* in: handle to the user thread */
+	int	lock_type)	/* in: lock type */
+{
+	trx_t*		trx;
+
+	DBUG_ENTER("ha_innobase::transactional_table_lock");
+	DBUG_PRINT("enter",("lock_type: %d", lock_type));
+
+	/* We do not know if MySQL can call this function before calling
+	external_lock(). To be safe, update the thd of the current table
+	handle. */
+
+	update_thd(thd);
+
+	if (prebuilt->table->ibd_file_missing && !thd_tablespace_op(thd)) {
+		ut_print_timestamp(stderr);
+		fprintf(stderr,
+			"  InnoDB: MySQL is trying to use a table handle"
+			" but the .ibd file for\n"
+			"InnoDB: table %s does not exist.\n"
+			"InnoDB: Have you deleted the .ibd file"
+			" from the database directory under\n"
+			"InnoDB: the MySQL datadir?"
+			"InnoDB: See"
+			" http://dev.mysql.com/doc/refman/5.1/en/innodb-troubleshooting.html\n"
+			"InnoDB: how you can resolve the problem.\n",
+			prebuilt->table->name);
+		DBUG_RETURN(HA_ERR_CRASHED);
+	}
+
+	trx = prebuilt->trx;
+
+	prebuilt->sql_stat_start = TRUE;
+	prebuilt->hint_need_to_fetch_extra_cols = 0;
+
+	reset_template(prebuilt);
+
+	if (lock_type == F_WRLCK) {
+		prebuilt->select_lock_type = LOCK_X;
+		prebuilt->stored_select_lock_type = LOCK_X;
+	} else if (lock_type == F_RDLCK) {
+		prebuilt->select_lock_type = LOCK_S;
+		prebuilt->stored_select_lock_type = LOCK_S;
+	} else {
+		ut_print_timestamp(stderr);
+		fprintf(stderr, "  InnoDB error:\n"
+"MySQL is trying to set transactional table lock with corrupted lock type\n"
+"to table %s, lock type %d does not exist.\n",
+				prebuilt->table->name, lock_type);
+		DBUG_RETURN(HA_ERR_CRASHED);
+	}
+
+	/* MySQL is setting a new transactional table lock */
+
+	/* Set the MySQL flag to mark that there is an active transaction */
+	if (trx->active_trans == 0) {
+
+		innobase_register_trx_and_stmt(ht, thd);
+		trx->active_trans = 1;
+	}
+
+	if (THDVAR(thd, table_locks) && thd_in_lock_tables(thd)) {
+		ulint	error = DB_SUCCESS;
+
+		error = row_lock_table_for_mysql(prebuilt, NULL, 0);
+
+		if (error != DB_SUCCESS) {
+			error = convert_error_code_to_mysql(
+				(int) error, prebuilt->table->flags, thd);
+			DBUG_RETURN((int) error);
+		}
+
+		if (thd_test_options(thd, OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN)) {
+
+			/* Store the current undo_no of the transaction
+			so that we know where to roll back if we have
+			to roll back the next SQL statement */
+
+			trx_mark_sql_stat_end(trx);
+		}
+	}
+
+	DBUG_RETURN(0);
+}
+
+/****************************************************************************
+Here we export InnoDB status variables to MySQL.  */
+static
+int
+innodb_export_status()
+/*==================*/
+{
+	if (innodb_inited) {
+		srv_export_innodb_status();
+	}
+
+	return(0);
+}
+
+/****************************************************************************
+Implements the SHOW INNODB STATUS command. Sends the output of the InnoDB
+Monitor to the client. */
+static
+bool
+innodb_show_status(
+/*===============*/
+	handlerton*	hton,	/* in: the innodb handlerton */
+	THD*	thd,	/* in: the MySQL query thread of the caller */
+	stat_print_fn *stat_print)
+{
+	trx_t*			trx;
+	static const char	truncated_msg[] = "... truncated...\n";
+	const long		MAX_STATUS_SIZE = 64000;
+	ulint			trx_list_start = ULINT_UNDEFINED;
+	ulint			trx_list_end = ULINT_UNDEFINED;
+
+	DBUG_ENTER("innodb_show_status");
+	DBUG_ASSERT(hton == innodb_hton_ptr);
+
+	trx = check_trx_exists(thd);
+
+	innobase_release_stat_resources(trx);
+
+	/* We let the InnoDB Monitor to output at most MAX_STATUS_SIZE
+	bytes of text. */
+
+	long	flen, usable_len;
+	char*	str;
+
+	mutex_enter(&srv_monitor_file_mutex);
+	rewind(srv_monitor_file);
+	srv_printf_innodb_monitor(srv_monitor_file,
+				&trx_list_start, &trx_list_end);
+	flen = ftell(srv_monitor_file);
+	os_file_set_eof(srv_monitor_file);
+
+	if (flen < 0) {
+		flen = 0;
+	}
+
+	if (flen > MAX_STATUS_SIZE) {
+		usable_len = MAX_STATUS_SIZE;
+	} else {
+		usable_len = flen;
+	}
+
+	/* allocate buffer for the string, and
+	read the contents of the temporary file */
+
+	if (!(str = (char*) my_malloc(usable_len + 1, MYF(0)))) {
+	  mutex_exit(&srv_monitor_file_mutex);
+	  DBUG_RETURN(TRUE);
+	}
+
+	rewind(srv_monitor_file);
+	if (flen < MAX_STATUS_SIZE) {
+		/* Display the entire output. */
+		flen = (long) fread(str, 1, flen, srv_monitor_file);
+	} else if (trx_list_end < (ulint) flen
+			&& trx_list_start < trx_list_end
+			&& trx_list_start + (flen - trx_list_end)
+			< MAX_STATUS_SIZE - sizeof truncated_msg - 1) {
+		/* Omit the beginning of the list of active transactions. */
+		long len = (long) fread(str, 1, trx_list_start, srv_monitor_file);
+		memcpy(str + len, truncated_msg, sizeof truncated_msg - 1);
+		len += sizeof truncated_msg - 1;
+		usable_len = (MAX_STATUS_SIZE - 1) - len;
+		fseek(srv_monitor_file, flen - usable_len, SEEK_SET);
+		len += (long) fread(str + len, 1, usable_len, srv_monitor_file);
+		flen = len;
+	} else {
+		/* Omit the end of the output. */
+		flen = (long) fread(str, 1, MAX_STATUS_SIZE - 1, srv_monitor_file);
+	}
+
+	mutex_exit(&srv_monitor_file_mutex);
+
+	bool result = FALSE;
+
+	if (stat_print(thd, innobase_hton_name, strlen(innobase_hton_name),
+			STRING_WITH_LEN(""), str, flen)) {
+		result= TRUE;
+	}
+	my_free(str, MYF(0));
+
+	DBUG_RETURN(FALSE);
+}
+
+/****************************************************************************
+Implements the SHOW MUTEX STATUS command. . */
+static
+bool
+innodb_mutex_show_status(
+/*=====================*/
+	handlerton*	hton,	/* in: the innodb handlerton */
+	THD*		thd,		/* in: the MySQL query thread of the
+					caller */
+	stat_print_fn*	stat_print)
+{
+	char buf1[IO_SIZE], buf2[IO_SIZE];
+	mutex_t*	mutex;
+	rw_lock_t*	lock;
+#ifdef UNIV_DEBUG
+	ulint	  rw_lock_count= 0;
+	ulint	  rw_lock_count_spin_loop= 0;
+	ulint	  rw_lock_count_spin_rounds= 0;
+	ulint	  rw_lock_count_os_wait= 0;
+	ulint	  rw_lock_count_os_yield= 0;
+	ulonglong rw_lock_wait_time= 0;
+#endif /* UNIV_DEBUG */
+	uint	  hton_name_len= strlen(innobase_hton_name), buf1len, buf2len;
+	DBUG_ENTER("innodb_mutex_show_status");
+	DBUG_ASSERT(hton == innodb_hton_ptr);
+
+	mutex_enter(&mutex_list_mutex);
+
+	mutex = UT_LIST_GET_FIRST(mutex_list);
+
+	while (mutex != NULL) {
+#ifdef UNIV_DEBUG
+		if (mutex->mutex_type != 1) {
+			if (mutex->count_using > 0) {
+				buf1len= my_snprintf(buf1, sizeof(buf1),
+					"%s:%s",
+					mutex->cmutex_name, mutex->cfile_name);
+				buf2len= my_snprintf(buf2, sizeof(buf2),
+					"count=%lu, spin_waits=%lu,"
+					" spin_rounds=%lu, "
+					"os_waits=%lu, os_yields=%lu,"
+					" os_wait_times=%lu",
+					mutex->count_using,
+					mutex->count_spin_loop,
+					mutex->count_spin_rounds,
+					mutex->count_os_wait,
+					mutex->count_os_yield,
+					(ulong) (mutex->lspent_time/1000));
+
+				if (stat_print(thd, innobase_hton_name,
+						hton_name_len, buf1, buf1len,
+						buf2, buf2len)) {
+					mutex_exit(&mutex_list_mutex);
+					DBUG_RETURN(1);
+				}
+			}
+		}
+		else {
+			rw_lock_count += mutex->count_using;
+			rw_lock_count_spin_loop += mutex->count_spin_loop;
+			rw_lock_count_spin_rounds += mutex->count_spin_rounds;
+			rw_lock_count_os_wait += mutex->count_os_wait;
+			rw_lock_count_os_yield += mutex->count_os_yield;
+			rw_lock_wait_time += mutex->lspent_time;
+		}
+#else /* UNIV_DEBUG */
+		buf1len= my_snprintf(buf1, sizeof(buf1), "%s:%lu",
+				     mutex->cfile_name, (ulong) mutex->cline);
+		buf2len= my_snprintf(buf2, sizeof(buf2), "os_waits=%lu",
+				     mutex->count_os_wait);
+
+		if (stat_print(thd, innobase_hton_name,
+			       hton_name_len, buf1, buf1len,
+			       buf2, buf2len)) {
+			mutex_exit(&mutex_list_mutex);
+			DBUG_RETURN(1);
+		}
+#endif /* UNIV_DEBUG */
+
+		mutex = UT_LIST_GET_NEXT(list, mutex);
+	}
+
+	mutex_exit(&mutex_list_mutex);
+
+	mutex_enter(&rw_lock_list_mutex);
+
+	lock = UT_LIST_GET_FIRST(rw_lock_list);
+
+	while (lock != NULL) {
+		if (lock->count_os_wait) {
+			buf1len= my_snprintf(buf1, sizeof(buf1), "%s:%lu",
+                                    lock->cfile_name, (ulong) lock->cline);
+			buf2len= my_snprintf(buf2, sizeof(buf2),
+                                    "os_waits=%lu", lock->count_os_wait);
+
+			if (stat_print(thd, innobase_hton_name,
+				       hton_name_len, buf1, buf1len,
+				       buf2, buf2len)) {
+				mutex_exit(&rw_lock_list_mutex);
+				DBUG_RETURN(1);
+			}
+		}
+		lock = UT_LIST_GET_NEXT(list, lock);
+	}
+
+	mutex_exit(&rw_lock_list_mutex);
+
+#ifdef UNIV_DEBUG
+	buf2len= my_snprintf(buf2, sizeof(buf2),
+		"count=%lu, spin_waits=%lu, spin_rounds=%lu, "
+		"os_waits=%lu, os_yields=%lu, os_wait_times=%lu",
+		rw_lock_count, rw_lock_count_spin_loop,
+		rw_lock_count_spin_rounds,
+		rw_lock_count_os_wait, rw_lock_count_os_yield,
+		(ulong) (rw_lock_wait_time/1000));
+
+	if (stat_print(thd, innobase_hton_name, hton_name_len,
+			STRING_WITH_LEN("rw_lock_mutexes"), buf2, buf2len)) {
+		DBUG_RETURN(1);
+	}
+#endif /* UNIV_DEBUG */
+
+	DBUG_RETURN(FALSE);
+}
+
+static
+bool innobase_show_status(handlerton *hton, THD* thd, 
+                          stat_print_fn* stat_print,
+                          enum ha_stat_type stat_type)
+{
+	DBUG_ASSERT(hton == innodb_hton_ptr);
+
+	switch (stat_type) {
+	case HA_ENGINE_STATUS:
+		return innodb_show_status(hton, thd, stat_print);
+	case HA_ENGINE_MUTEX:
+		return innodb_mutex_show_status(hton, thd, stat_print);
+	default:
+		return(FALSE);
+	}
+}
+
+/****************************************************************************
+ Handling the shared INNOBASE_SHARE structure that is needed to provide table
+ locking.
+****************************************************************************/
+
+static INNOBASE_SHARE* get_share(const char* table_name)
+{
+	INNOBASE_SHARE *share;
+	pthread_mutex_lock(&innobase_share_mutex);
+
+	ulint	fold = ut_fold_string(table_name);
+
+	HASH_SEARCH(table_name_hash, innobase_open_tables, fold,
+		    INNOBASE_SHARE*, share,
+		    ut_ad(share->use_count > 0),
+		    !strcmp(share->table_name, table_name));
+
+	if (!share) {
+
+		uint length = (uint) strlen(table_name);
+
+		/* TODO: invoke HASH_MIGRATE if innobase_open_tables
+		grows too big */
+
+		share = (INNOBASE_SHARE *) my_malloc(sizeof(*share)+length+1,
+			MYF(MY_FAE | MY_ZEROFILL));
+
+		share->table_name = (char*) memcpy(share + 1,
+						   table_name, length + 1);
+
+		HASH_INSERT(INNOBASE_SHARE, table_name_hash,
+			    innobase_open_tables, fold, share);
+
+		thr_lock_init(&share->lock);
+		pthread_mutex_init(&share->mutex,MY_MUTEX_INIT_FAST);
+	}
+
+	share->use_count++;
+	pthread_mutex_unlock(&innobase_share_mutex);
+
+	return(share);
+}
+
+static void free_share(INNOBASE_SHARE* share)
+{
+	pthread_mutex_lock(&innobase_share_mutex);
+
+#ifdef UNIV_DEBUG
+	INNOBASE_SHARE* share2;
+	ulint	fold = ut_fold_string(share->table_name);
+
+	HASH_SEARCH(table_name_hash, innobase_open_tables, fold,
+		    INNOBASE_SHARE*, share2,
+		    ut_ad(share->use_count > 0),
+		    !strcmp(share->table_name, share2->table_name));
+
+	ut_a(share2 == share);
+#endif /* UNIV_DEBUG */
+
+	if (!--share->use_count) {
+		ulint	fold = ut_fold_string(share->table_name);
+
+		HASH_DELETE(INNOBASE_SHARE, table_name_hash,
+			    innobase_open_tables, fold, share);
+		thr_lock_delete(&share->lock);
+		pthread_mutex_destroy(&share->mutex);
+		my_free(share, MYF(0));
+
+		/* TODO: invoke HASH_MIGRATE if innobase_open_tables
+		shrinks too much */
+	}
+
+	pthread_mutex_unlock(&innobase_share_mutex);
+}
+
+/*********************************************************************
+Converts a MySQL table lock stored in the 'lock' field of the handle to
+a proper type before storing pointer to the lock into an array of pointers.
+MySQL also calls this if it wants to reset some table locks to a not-locked
+state during the processing of an SQL query. An example is that during a
+SELECT the read lock is released early on the 'const' tables where we only
+fetch one row. MySQL does not call this when it releases all locks at the
+end of an SQL statement. */
+UNIV_INTERN
+THR_LOCK_DATA**
+ha_innobase::store_lock(
+/*====================*/
+						/* out: pointer to the next
+						element in the 'to' array */
+	THD*			thd,		/* in: user thread handle */
+	THR_LOCK_DATA**		to,		/* in: pointer to an array
+						of pointers to lock structs;
+						pointer to the 'lock' field
+						of current handle is stored
+						next to this array */
+	enum thr_lock_type	lock_type)	/* in: lock type to store in
+						'lock'; this may also be
+						TL_IGNORE */
+{
+	trx_t*		trx;
+
+	/* Note that trx in this function is NOT necessarily prebuilt->trx
+	because we call update_thd() later, in ::external_lock()! Failure to
+	understand this caused a serious memory corruption bug in 5.1.11. */
+
+	trx = check_trx_exists(thd);
+
+	/* NOTE: MySQL can call this function with lock 'type' TL_IGNORE!
+	Be careful to ignore TL_IGNORE if we are going to do something with
+	only 'real' locks! */
+
+	/* If no MySQL table is in use, we need to set the isolation level
+	of the transaction. */
+
+	if (lock_type != TL_IGNORE
+	    && trx->n_mysql_tables_in_use == 0) {
+		trx->isolation_level = innobase_map_isolation_level(
+			(enum_tx_isolation) thd_tx_isolation(thd));
+
+		if (trx->isolation_level <= TRX_ISO_READ_COMMITTED
+		    && trx->global_read_view) {
+
+			/* At low transaction isolation levels we let
+			each consistent read set its own snapshot */
+
+			read_view_close_for_mysql(trx);
+		}
+	}
+
+	DBUG_ASSERT(EQ_CURRENT_THD(thd));
+	const bool in_lock_tables = thd_in_lock_tables(thd);
+	const uint sql_command = thd_sql_command(thd);
+
+	if (sql_command == SQLCOM_DROP_TABLE) {
+
+		/* MySQL calls this function in DROP TABLE though this table
+		handle may belong to another thd that is running a query. Let
+		us in that case skip any changes to the prebuilt struct. */ 
+
+	} else if ((lock_type == TL_READ && in_lock_tables)
+		   || (lock_type == TL_READ_HIGH_PRIORITY && in_lock_tables)
+		   || lock_type == TL_READ_WITH_SHARED_LOCKS
+		   || lock_type == TL_READ_NO_INSERT
+		   || (lock_type != TL_IGNORE
+		       && sql_command != SQLCOM_SELECT)) {
+
+		/* The OR cases above are in this order:
+		1) MySQL is doing LOCK TABLES ... READ LOCAL, or we
+		are processing a stored procedure or function, or
+		2) (we do not know when TL_READ_HIGH_PRIORITY is used), or
+		3) this is a SELECT ... IN SHARE MODE, or
+		4) we are doing a complex SQL statement like
+		INSERT INTO ... SELECT ... and the logical logging (MySQL
+		binlog) requires the use of a locking read, or
+		MySQL is doing LOCK TABLES ... READ.
+		5) we let InnoDB do locking reads for all SQL statements that
+		are not simple SELECTs; note that select_lock_type in this
+		case may get strengthened in ::external_lock() to LOCK_X.
+		Note that we MUST use a locking read in all data modifying
+		SQL statements, because otherwise the execution would not be
+		serializable, and also the results from the update could be
+		unexpected if an obsolete consistent read view would be
+		used. */
+
+		ulint	isolation_level;
+
+		isolation_level = trx->isolation_level;
+
+		if ((srv_locks_unsafe_for_binlog
+		     || isolation_level == TRX_ISO_READ_COMMITTED)
+		    && isolation_level != TRX_ISO_SERIALIZABLE
+		    && (lock_type == TL_READ || lock_type == TL_READ_NO_INSERT)
+		    && (sql_command == SQLCOM_INSERT_SELECT
+			|| sql_command == SQLCOM_UPDATE
+			|| sql_command == SQLCOM_CREATE_TABLE)) {
+
+			/* If we either have innobase_locks_unsafe_for_binlog
+			option set or this session is using READ COMMITTED
+			isolation level and isolation level of the transaction
+			is not set to serializable and MySQL is doing
+			INSERT INTO...SELECT or UPDATE ... = (SELECT ...) or
+			CREATE  ... SELECT... without FOR UPDATE or
+			IN SHARE MODE in select, then we use consistent
+			read for select. */
+
+			prebuilt->select_lock_type = LOCK_NONE;
+			prebuilt->stored_select_lock_type = LOCK_NONE;
+		} else if (sql_command == SQLCOM_CHECKSUM) {
+			/* Use consistent read for checksum table */
+
+			prebuilt->select_lock_type = LOCK_NONE;
+			prebuilt->stored_select_lock_type = LOCK_NONE;
+		} else {
+			prebuilt->select_lock_type = LOCK_S;
+			prebuilt->stored_select_lock_type = LOCK_S;
+		}
+
+	} else if (lock_type != TL_IGNORE) {
+
+		/* We set possible LOCK_X value in external_lock, not yet
+		here even if this would be SELECT ... FOR UPDATE */
+
+		prebuilt->select_lock_type = LOCK_NONE;
+		prebuilt->stored_select_lock_type = LOCK_NONE;
+	}
+
+	if (lock_type != TL_IGNORE && lock.type == TL_UNLOCK) {
+
+		/* Starting from 5.0.7, we weaken also the table locks
+		set at the start of a MySQL stored procedure call, just like
+		we weaken the locks set at the start of an SQL statement.
+		MySQL does set in_lock_tables TRUE there, but in reality
+		we do not need table locks to make the execution of a
+		single transaction stored procedure call deterministic
+		(if it does not use a consistent read). */
+
+		if (lock_type == TL_READ
+		    && sql_command == SQLCOM_LOCK_TABLES) {
+			/* We come here if MySQL is processing LOCK TABLES
+			... READ LOCAL. MyISAM under that table lock type
+			reads the table as it was at the time the lock was
+			granted (new inserts are allowed, but not seen by the
+			reader). To get a similar effect on an InnoDB table,
+			we must use LOCK TABLES ... READ. We convert the lock
+			type here, so that for InnoDB, READ LOCAL is
+			equivalent to READ. This will change the InnoDB
+			behavior in mysqldump, so that dumps of InnoDB tables
+			are consistent with dumps of MyISAM tables. */
+
+			lock_type = TL_READ_NO_INSERT;
+		}
+
+		/* If we are not doing a LOCK TABLE, DISCARD/IMPORT
+		TABLESPACE or TRUNCATE TABLE then allow multiple
+		writers. Note that ALTER TABLE uses a TL_WRITE_ALLOW_READ
+		< TL_WRITE_CONCURRENT_INSERT.
+
+		We especially allow multiple writers if MySQL is at the
+		start of a stored procedure call (SQLCOM_CALL) or a
+		stored function call (MySQL does have in_lock_tables
+		TRUE there). */
+
+		if ((lock_type >= TL_WRITE_CONCURRENT_INSERT
+		     && lock_type <= TL_WRITE)
+		    && !(in_lock_tables
+			 && sql_command == SQLCOM_LOCK_TABLES)
+		    && !thd_tablespace_op(thd)
+		    && sql_command != SQLCOM_TRUNCATE
+		    && sql_command != SQLCOM_OPTIMIZE
+		    && sql_command != SQLCOM_CREATE_TABLE) {
+
+			lock_type = TL_WRITE_ALLOW_WRITE;
+		}
+
+		/* In queries of type INSERT INTO t1 SELECT ... FROM t2 ...
+		MySQL would use the lock TL_READ_NO_INSERT on t2, and that
+		would conflict with TL_WRITE_ALLOW_WRITE, blocking all inserts
+		to t2. Convert the lock to a normal read lock to allow
+		concurrent inserts to t2.
+
+		We especially allow concurrent inserts if MySQL is at the
+		start of a stored procedure call (SQLCOM_CALL)
+		(MySQL does have thd_in_lock_tables() TRUE there). */
+
+		if (lock_type == TL_READ_NO_INSERT
+		    && sql_command != SQLCOM_LOCK_TABLES) {
+
+			lock_type = TL_READ;
+		}
+
+		lock.type = lock_type;
+	}
+
+	*to++= &lock;
+
+	return(to);
+}
+
+/*******************************************************************************
+Read the next autoinc value. Acquire the relevant locks before reading
+the AUTOINC value. If SUCCESS then the table AUTOINC mutex will be locked
+on return and all relevant locks acquired. */
+UNIV_INTERN
+ulint
+ha_innobase::innobase_get_autoinc(
+/*==============================*/
+					/* out: DB_SUCCESS or error code */
+	ulonglong*	value)		/* out: autoinc value */
+{
+ 	*value = 0;
+ 
+	prebuilt->autoinc_error = innobase_lock_autoinc();
+
+	if (prebuilt->autoinc_error == DB_SUCCESS) {
+
+		/* Determine the first value of the interval */
+		*value = dict_table_autoinc_read(prebuilt->table);
+
+		/* It should have been initialized during open. */
+		ut_a(*value != 0);
+	}
+
+	return(prebuilt->autoinc_error);
+}
+
+/***********************************************************************
+This function reads the global auto-inc counter. It doesn't use the 
+AUTOINC lock even if the lock mode is set to TRADITIONAL. */
+UNIV_INTERN
+ulonglong
+ha_innobase::innobase_peek_autoinc(void)
+/*====================================*/
+					/* out: the autoinc value */
+{
+	ulonglong	auto_inc;
+	dict_table_t*	innodb_table;
+
+	ut_a(prebuilt != NULL);
+	ut_a(prebuilt->table != NULL);
+
+	innodb_table = prebuilt->table;
+
+	dict_table_autoinc_lock(innodb_table);
+
+	auto_inc = dict_table_autoinc_read(innodb_table);
+
+	ut_a(auto_inc > 0);
+
+	dict_table_autoinc_unlock(innodb_table);
+ 
+	return(auto_inc);
+}
+  
+/*******************************************************************************
+This function initializes the auto-inc counter if it has not been
+initialized yet. This function does not change the value of the auto-inc
+counter if it already has been initialized. Returns the value of the
+auto-inc counter in *first_value, and ULONGLONG_MAX in *nb_reserved_values (as
+we have a table-level lock). offset, increment, nb_desired_values are ignored.
+*first_value is set to -1 if error (deadlock or lock wait timeout) */
+UNIV_INTERN
+void
+ha_innobase::get_auto_increment(
+/*============================*/
+        ulonglong	offset,              /* in: */
+        ulonglong	increment,           /* in: table autoinc increment */
+        ulonglong	nb_desired_values,   /* in: number of values reqd */
+        ulonglong	*first_value,        /* out: the autoinc value */
+        ulonglong	*nb_reserved_values) /* out: count of reserved values */
+{
+	trx_t*		trx;
+	ulint		error;
+	ulonglong	autoinc = 0;
+
+	/* Prepare prebuilt->trx in the table handle */
+	update_thd(ha_thd());
+
+	error = innobase_get_autoinc(&autoinc);
+
+	if (error != DB_SUCCESS) {
+		*first_value = (~(ulonglong) 0);
+		return;
+	}
+
+	/* This is a hack, since nb_desired_values seems to be accurate only
+	for the first call to get_auto_increment() for multi-row INSERT and
+	meaningless for other statements e.g, LOAD etc. Subsequent calls to
+	this method for the same statement results in different values which
+	don't make sense. Therefore we store the value the first time we are
+	called and count down from that as rows are written (see write_row()).
+	*/
+
+	trx = prebuilt->trx;
+
+	/* Note: We can't rely on *first_value since some MySQL engines,
+	in particular the partition engine, don't initialize it to 0 when
+	invoking this method. So we are not sure if it's guaranteed to
+	be 0 or not. */
+
+	/* Called for the first time ? */
+	if (trx->n_autoinc_rows == 0) {
+
+		trx->n_autoinc_rows = (ulint) nb_desired_values;
+
+		/* It's possible for nb_desired_values to be 0:
+		e.g., INSERT INTO T1(C) SELECT C FROM T2; */
+		if (nb_desired_values == 0) {
+
+			trx->n_autoinc_rows = 1;
+		}
+
+		set_if_bigger(*first_value, autoinc);
+	/* Not in the middle of a mult-row INSERT. */
+	} else if (prebuilt->autoinc_last_value == 0) {
+		set_if_bigger(*first_value, autoinc);
+	}
+
+	*nb_reserved_values = trx->n_autoinc_rows;
+
+	/* With old style AUTOINC locking we only update the table's
+	AUTOINC counter after attempting to insert the row. */
+	if (innobase_autoinc_lock_mode != AUTOINC_OLD_STYLE_LOCKING) {
+		ulonglong	need;
+		ulonglong	next_value;
+		ulonglong	col_max_value;
+
+		/* We need the upper limit of the col type to check for
+		whether we update the table autoinc counter or not. */
+		col_max_value = innobase_get_int_col_max_value(
+			table->next_number_field);
+
+		need = *nb_reserved_values * increment;
+
+		/* Compute the last value in the interval */
+		next_value = innobase_next_autoinc(
+			*first_value, need, offset, col_max_value);
+
+		prebuilt->autoinc_last_value = next_value;
+
+		if (prebuilt->autoinc_last_value < *first_value) {
+			*first_value = (~(ulonglong) 0);
+		} else {
+			/* Update the table autoinc variable */
+			dict_table_autoinc_update_if_greater(
+				prebuilt->table, prebuilt->autoinc_last_value);
+		}
+	} else {
+		/* This will force write_row() into attempting an update
+		of the table's AUTOINC counter. */
+		prebuilt->autoinc_last_value = 0;
+	}
+
+	/* The increment to be used to increase the AUTOINC value, we use
+	this in write_row() and update_row() to increase the autoinc counter
+	for columns that are filled by the user. We need the offset and
+	the increment. */
+	prebuilt->autoinc_offset = offset;
+	prebuilt->autoinc_increment = increment;
+
+	dict_table_autoinc_unlock(prebuilt->table);
+}
+
+/* See comment in handler.h */
+UNIV_INTERN
+int
+ha_innobase::reset_auto_increment(
+/*==============================*/
+	ulonglong	value)		/* in: new value for table autoinc */
+{
+	DBUG_ENTER("ha_innobase::reset_auto_increment");
+
+	int	error;
+
+	update_thd(ha_thd());
+
+	error = row_lock_table_autoinc_for_mysql(prebuilt);
+
+	if (error != DB_SUCCESS) {
+		error = convert_error_code_to_mysql(error,
+						    prebuilt->table->flags,
+						    user_thd);
+
+		DBUG_RETURN(error);
+	}
+
+	/* The next value can never be 0. */
+	if (value == 0) {
+		value = 1;
+	}
+
+	innobase_reset_autoinc(value);
+
+	DBUG_RETURN(0);
+}
+
+/* See comment in handler.cc */
+UNIV_INTERN
+bool
+ha_innobase::get_error_message(int error, String *buf)
+{
+	trx_t*	trx = check_trx_exists(ha_thd());
+
+	buf->copy(trx->detailed_error, strlen(trx->detailed_error),
+		system_charset_info);
+
+	return(FALSE);
+}
+
+/***********************************************************************
+Compares two 'refs'. A 'ref' is the (internal) primary key value of the row.
+If there is no explicitly declared non-null unique key or a primary key, then
+InnoDB internally uses the row id as the primary key. */
+UNIV_INTERN
+int
+ha_innobase::cmp_ref(
+/*=================*/
+				/* out: < 0 if ref1 < ref2, 0 if equal, else
+				> 0 */
+	const uchar*	ref1,	/* in: an (internal) primary key value in the
+				MySQL key value format */
+	const uchar*	ref2)	/* in: an (internal) primary key value in the
+				MySQL key value format */
+{
+	enum_field_types mysql_type;
+	Field*		field;
+	KEY_PART_INFO*	key_part;
+	KEY_PART_INFO*	key_part_end;
+	uint		len1;
+	uint		len2;
+	int		result;
+
+	if (prebuilt->clust_index_was_generated) {
+		/* The 'ref' is an InnoDB row id */
+
+		return(memcmp(ref1, ref2, DATA_ROW_ID_LEN));
+	}
+
+	/* Do a type-aware comparison of primary key fields. PK fields
+	are always NOT NULL, so no checks for NULL are performed. */
+
+	key_part = table->key_info[table->s->primary_key].key_part;
+
+	key_part_end = key_part
+			+ table->key_info[table->s->primary_key].key_parts;
+
+	for (; key_part != key_part_end; ++key_part) {
+		field = key_part->field;
+		mysql_type = field->type();
+
+		if (mysql_type == MYSQL_TYPE_TINY_BLOB
+			|| mysql_type == MYSQL_TYPE_MEDIUM_BLOB
+			|| mysql_type == MYSQL_TYPE_BLOB
+			|| mysql_type == MYSQL_TYPE_LONG_BLOB) {
+
+			/* In the MySQL key value format, a column prefix of
+			a BLOB is preceded by a 2-byte length field */
+
+			len1 = innobase_read_from_2_little_endian(ref1);
+			len2 = innobase_read_from_2_little_endian(ref2);
+
+			ref1 += 2;
+			ref2 += 2;
+			result = ((Field_blob*)field)->cmp( ref1, len1,
+                                                            ref2, len2);
+		} else {
+			result = field->key_cmp(ref1, ref2);
+		}
+
+		if (result) {
+
+			return(result);
+		}
+
+		ref1 += key_part->store_length;
+		ref2 += key_part->store_length;
+	}
+
+	return(0);
+}
+
+/***********************************************************************
+Ask InnoDB if a query to a table can be cached. */
+UNIV_INTERN
+my_bool
+ha_innobase::register_query_cache_table(
+/*====================================*/
+					/* out: TRUE if query caching
+					of the table is permitted */
+	THD*		thd,		/* in: user thread handle */
+	char*		table_key,	/* in: concatenation of database name,
+					the null character '\0',
+					and the table name */
+	uint		key_length,	/* in: length of the full name, i.e.
+					len(dbname) + len(tablename) + 1 */
+	qc_engine_callback*
+			call_back,	/* out: pointer to function for
+					checking if query caching
+					is permitted */
+	ulonglong	*engine_data)	/* in/out: data to call_back */
+{
+	*call_back = innobase_query_caching_of_table_permitted;
+	*engine_data = 0;
+	return(innobase_query_caching_of_table_permitted(thd, table_key,
+							 key_length,
+							 engine_data));
+}
+
+UNIV_INTERN
+char*
+ha_innobase::get_mysql_bin_log_name()
+{
+	return(trx_sys_mysql_bin_log_name);
+}
+
+UNIV_INTERN
+ulonglong
+ha_innobase::get_mysql_bin_log_pos()
+{
+	/* trx... is ib_int64_t, which is a typedef for a 64-bit integer
+	(__int64 or longlong) so it's ok to cast it to ulonglong. */
+
+	return(trx_sys_mysql_bin_log_pos);
+}
+
+/**********************************************************************
+This function is used to find the storage length in bytes of the first n
+characters for prefix indexes using a multibyte character set. The function
+finds charset information and returns length of prefix_len characters in the
+index field in bytes.
+
+NOTE: the prototype of this function is copied to data0type.c! If you change
+this function, you MUST change also data0type.c! */
+extern "C" UNIV_INTERN
+ulint
+innobase_get_at_most_n_mbchars(
+/*===========================*/
+				/* out: number of bytes occupied by the first
+				n characters */
+	ulint charset_id,	/* in: character set id */
+	ulint prefix_len,	/* in: prefix length in bytes of the index
+				(this has to be divided by mbmaxlen to get the
+				number of CHARACTERS n in the prefix) */
+	ulint data_len,		/* in: length of the string in bytes */
+	const char* str)	/* in: character string */
+{
+	ulint char_length;	/* character length in bytes */
+	ulint n_chars;		/* number of characters in prefix */
+	CHARSET_INFO* charset;	/* charset used in the field */
+
+	charset = get_charset((uint) charset_id, MYF(MY_WME));
+
+	ut_ad(charset);
+	ut_ad(charset->mbmaxlen);
+
+	/* Calculate how many characters at most the prefix index contains */
+
+	n_chars = prefix_len / charset->mbmaxlen;
+
+	/* If the charset is multi-byte, then we must find the length of the
+	first at most n chars in the string. If the string contains less
+	characters than n, then we return the length to the end of the last
+	character. */
+
+	if (charset->mbmaxlen > 1) {
+		/* my_charpos() returns the byte length of the first n_chars
+		characters, or a value bigger than the length of str, if
+		there were not enough full characters in str.
+
+		Why does the code below work:
+		Suppose that we are looking for n UTF-8 characters.
+
+		1) If the string is long enough, then the prefix contains at
+		least n complete UTF-8 characters + maybe some extra
+		characters + an incomplete UTF-8 character. No problem in
+		this case. The function returns the pointer to the
+		end of the nth character.
+
+		2) If the string is not long enough, then the string contains
+		the complete value of a column, that is, only complete UTF-8
+		characters, and we can store in the column prefix index the
+		whole string. */
+
+		char_length = my_charpos(charset, str,
+						str + data_len, (int) n_chars);
+		if (char_length > data_len) {
+			char_length = data_len;
+		}
+	} else {
+		if (data_len < prefix_len) {
+			char_length = data_len;
+		} else {
+			char_length = prefix_len;
+		}
+	}
+
+	return(char_length);
+}
+
+/***********************************************************************
+This function is used to prepare X/Open XA distributed transaction   */
+static
+int
+innobase_xa_prepare(
+/*================*/
+			/* out: 0 or error number */
+        handlerton *hton,
+	THD*	thd,	/* in: handle to the MySQL thread of the user
+			whose XA transaction should be prepared */
+	bool	all)	/* in: TRUE - commit transaction
+			FALSE - the current SQL statement ended */
+{
+	int error = 0;
+	trx_t* trx = check_trx_exists(thd);
+
+	DBUG_ASSERT(hton == innodb_hton_ptr);
+
+	if (thd_sql_command(thd) != SQLCOM_XA_PREPARE &&
+	    (all || !thd_test_options(thd, OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN)))
+	{
+		if (srv_enable_unsafe_group_commit && !THDVAR(thd, support_xa)) {
+			/* choose group commit rather than binlog order */
+			return(0);
+		}
+
+		/* For ibbackup to work the order of transactions in binlog
+		and InnoDB must be the same. Consider the situation
+
+		  thread1> prepare; write to binlog; ...
+			  <context switch>
+		  thread2> prepare; write to binlog; commit
+		  thread1>			     ... commit
+
+		To ensure this will not happen we're taking the mutex on
+		prepare, and releasing it on commit.
+
+		Note: only do it for normal commits, done via ha_commit_trans.
+		If 2pc protocol is executed by external transaction
+		coordinator, it will be just a regular MySQL client
+		executing XA PREPARE and XA COMMIT commands.
+		In this case we cannot know how many minutes or hours
+		will be between XA PREPARE and XA COMMIT, and we don't want
+		to block for undefined period of time.
+		*/
+		pthread_mutex_lock(&prepare_commit_mutex);
+		trx->active_trans = 2;
+	}
+
+	/* we use support_xa value as it was seen at transaction start
+	time, not the current session variable value. Any possible changes
+	to the session variable take effect only in the next transaction */
+	if (!trx->support_xa) {
+
+		return(0);
+	}
+
+	thd_get_xid(thd, (MYSQL_XID*) &trx->xid);
+
+	/* Release a possible FIFO ticket and search latch. Since we will
+	reserve the kernel mutex, we have to release the search system latch
+	first to obey the latching order. */
+
+	innobase_release_stat_resources(trx);
+
+	if (trx->active_trans == 0 && trx->conc_state != TRX_NOT_STARTED) {
+
+	  sql_print_error("trx->active_trans == 0, but trx->conc_state != "
+			  "TRX_NOT_STARTED");
+	}
+
+	if (all
+		|| (!thd_test_options(thd, OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN))) {
+
+		/* We were instructed to prepare the whole transaction, or
+		this is an SQL statement end and autocommit is on */
+
+		ut_ad(trx->active_trans);
+
+		error = (int) trx_prepare_for_mysql(trx);
+	} else {
+		/* We just mark the SQL statement ended and do not do a
+		transaction prepare */
+
+		/* If we had reserved the auto-inc lock for some
+		table in this SQL statement we release it now */
+
+		row_unlock_table_autoinc_for_mysql(trx);
+
+		/* Store the current undo_no of the transaction so that we
+		know where to roll back if we have to roll back the next
+		SQL statement */
+
+		trx_mark_sql_stat_end(trx);
+	}
+
+	/* Tell the InnoDB server that there might be work for utility
+	threads: */
+
+	srv_active_wake_master_thread();
+
+	return(error);
+}
+
+/***********************************************************************
+This function is used to recover X/Open XA distributed transactions   */
+static
+int
+innobase_xa_recover(
+/*================*/
+				/* out: number of prepared transactions
+				stored in xid_list */
+        handlerton *hton,
+	XID*	xid_list,	/* in/out: prepared transactions */
+	uint	len)		/* in: number of slots in xid_list */
+{
+	DBUG_ASSERT(hton == innodb_hton_ptr);
+
+	if (len == 0 || xid_list == NULL) {
+
+		return(0);
+	}
+
+	return(trx_recover_for_mysql(xid_list, len));
+}
+
+/***********************************************************************
+This function is used to commit one X/Open XA distributed transaction
+which is in the prepared state */
+static
+int
+innobase_commit_by_xid(
+/*===================*/
+			/* out: 0 or error number */
+        handlerton *hton,
+	XID*	xid)	/* in: X/Open XA transaction identification */
+{
+	trx_t*	trx;
+
+	DBUG_ASSERT(hton == innodb_hton_ptr);
+
+	trx = trx_get_trx_by_xid(xid);
+
+	if (trx) {
+		innobase_commit_low(trx);
+
+		return(XA_OK);
+	} else {
+		return(XAER_NOTA);
+	}
+}
+
+/***********************************************************************
+This function is used to rollback one X/Open XA distributed transaction
+which is in the prepared state */
+static
+int
+innobase_rollback_by_xid(
+/*=====================*/
+			/* out: 0 or error number */
+        handlerton *hton,
+	XID	*xid)	/* in: X/Open XA transaction identification */
+{
+	trx_t*	trx;
+
+	DBUG_ASSERT(hton == innodb_hton_ptr);
+
+	trx = trx_get_trx_by_xid(xid);
+
+	if (trx) {
+		return(innobase_rollback_trx(trx));
+	} else {
+		return(XAER_NOTA);
+	}
+}
+
+/***********************************************************************
+Create a consistent view for a cursor based on current transaction
+which is created if the corresponding MySQL thread still lacks one.
+This consistent view is then used inside of MySQL when accessing records
+using a cursor. */
+static
+void*
+innobase_create_cursor_view(
+/*========================*/
+                          /* out: pointer to cursor view or NULL */
+        handlerton *hton, /* in: innobase hton */
+	THD* thd)	  /* in: user thread handle */
+{
+	DBUG_ASSERT(hton == innodb_hton_ptr);
+
+	return(read_cursor_view_create_for_mysql(check_trx_exists(thd)));
+}
+
+/***********************************************************************
+Close the given consistent cursor view of a transaction and restore
+global read view to a transaction read view. Transaction is created if the
+corresponding MySQL thread still lacks one. */
+static
+void
+innobase_close_cursor_view(
+/*=======================*/
+        handlerton *hton,
+	THD*	thd,	/* in: user thread handle */
+	void*	curview)/* in: Consistent read view to be closed */
+{
+	DBUG_ASSERT(hton == innodb_hton_ptr);
+
+	read_cursor_view_close_for_mysql(check_trx_exists(thd),
+					 (cursor_view_t*) curview);
+}
+
+/***********************************************************************
+Set the given consistent cursor view to a transaction which is created
+if the corresponding MySQL thread still lacks one. If the given
+consistent cursor view is NULL global read view of a transaction is
+restored to a transaction read view. */
+static
+void
+innobase_set_cursor_view(
+/*=====================*/
+        handlerton *hton,
+	THD*	thd,	/* in: user thread handle */
+	void*	curview)/* in: Consistent cursor view to be set */
+{
+	DBUG_ASSERT(hton == innodb_hton_ptr);
+
+	read_cursor_set_for_mysql(check_trx_exists(thd),
+				  (cursor_view_t*) curview);
+}
+
+
+UNIV_INTERN
+bool
+ha_innobase::check_if_incompatible_data(
+	HA_CREATE_INFO*	info,
+	uint		table_changes)
+{
+	if (table_changes != IS_EQUAL_YES) {
+
+		return(COMPATIBLE_DATA_NO);
+	}
+
+	/* Check that auto_increment value was not changed */
+	if ((info->used_fields & HA_CREATE_USED_AUTO) &&
+		info->auto_increment_value != 0) {
+
+		return(COMPATIBLE_DATA_NO);
+	}
+
+	/* Check that row format didn't change */
+	if ((info->used_fields & HA_CREATE_USED_ROW_FORMAT) &&
+	    get_row_type() != info->row_type) {
+
+		return(COMPATIBLE_DATA_NO);
+	}
+
+	/* Specifying KEY_BLOCK_SIZE requests a rebuild of the table. */
+	if (info->used_fields & HA_CREATE_USED_KEY_BLOCK_SIZE) {
+		return(COMPATIBLE_DATA_NO);
+	}
+
+	return(COMPATIBLE_DATA_YES);
+}
+
+/****************************************************************
+Validate the file format name and return its corresponding id. */
+static
+uint
+innobase_file_format_name_lookup(
+/*=============================*/
+					/* out: valid file format id*/
+	const char*	format_name)	/* in: pointer to file format name */
+{
+	char*	endp;
+	uint	format_id;
+
+	ut_a(format_name != NULL);
+
+	/* The format name can contain the format id itself instead of
+	the name and we check for that. */
+	format_id = (uint) strtoul(format_name, &endp, 10);
+
+	/* Check for valid parse. */
+	if (*endp == '\0' && *format_name != '\0') {
+
+		if (format_id <= DICT_TF_FORMAT_MAX) {
+
+			return(format_id);
+		}
+	} else {
+
+		for (format_id = 0; format_id <= DICT_TF_FORMAT_MAX;
+		     format_id++) {
+			const char*	name;
+
+			name = trx_sys_file_format_id_to_name(format_id);
+
+			if (!innobase_strcasecmp(format_name, name)) {
+
+				return(format_id);
+			}
+		}
+	}
+
+	return(DICT_TF_FORMAT_MAX + 1);
+}
+
+/****************************************************************
+Validate the file format check value, is it one of "on" or "off",
+as a side effect it sets the srv_check_file_format_at_startup variable. */
+static
+bool
+innobase_file_format_check_on_off(
+/*==============================*/
+					/* out: true if config value one
+					of "on" or  "off" */
+	const char*	format_check)	/* in: parameter value */
+{
+	bool		ret = true;
+
+	if (!innobase_strcasecmp(format_check, "off")) {
+
+		/* Set the value to disable checking. */
+		srv_check_file_format_at_startup = DICT_TF_FORMAT_MAX + 1;
+
+	} else if (!innobase_strcasecmp(format_check, "on")) {
+
+		/* Set the value to the lowest supported format. */
+		srv_check_file_format_at_startup = DICT_TF_FORMAT_51;
+	} else {
+		ret = FALSE;
+	}
+
+	return(ret);
+}
+
+/****************************************************************
+Validate the file format check config parameters, as a side effect it
+sets the srv_check_file_format_at_startup variable. */
+static
+bool
+innobase_file_format_check_validate(
+/*================================*/
+					/* out: true if valid config value */
+	const char*	format_check)	/* in: parameter value */
+{
+	uint		format_id;
+	bool		ret = true;
+
+	format_id = innobase_file_format_name_lookup(format_check);
+
+	if (format_id < DICT_TF_FORMAT_MAX + 1) {
+		srv_check_file_format_at_startup = format_id;
+	} else {
+		ret = false;
+	}
+
+	return(ret);
+}
+
+/*****************************************************************
+Check if it is a valid file format. This function is registered as
+a callback with MySQL. */
+static
+int
+innodb_file_format_name_validate(
+/*=============================*/
+						/* out: 0 for valid file
+						format */
+	THD*				thd,	/* in: thread handle */
+	struct st_mysql_sys_var*	var,	/* in: pointer to system
+						variable */
+	void*				save,	/* out: immediate result
+						for update function */
+	struct st_mysql_value*		value)	/* in: incoming string */
+{
+	const char*	file_format_input;
+	char		buff[STRING_BUFFER_USUAL_SIZE];
+	int		len = sizeof(buff);
+
+	ut_a(save != NULL);
+	ut_a(value != NULL);
+
+	file_format_input = value->val_str(value, buff, &len);
+
+	if (file_format_input != NULL) {
+		uint	format_id;
+
+		format_id = innobase_file_format_name_lookup(
+			file_format_input);
+
+		if (format_id <= DICT_TF_FORMAT_MAX) {
+
+			*(uint*) save = format_id;
+			return(0);
+		}
+	}
+
+	return(1);
+}
+
+/********************************************************************
+Update the system variable innodb_file_format using the "saved"
+value. This function is registered as a callback with MySQL. */
+static
+void
+innodb_file_format_name_update(
+/*===========================*/
+	THD*				thd,		/* in: thread handle */
+	struct st_mysql_sys_var*	var,		/* in: pointer to
+							system variable */
+	void*				var_ptr,	/* out: where the
+							formal string goes */
+	const void*			save)		/* in: immediate result
+							from check function */
+{
+	ut_a(var_ptr != NULL);
+	ut_a(save != NULL);
+	ut_a((*(const uint*) save) <= DICT_TF_FORMAT_MAX);
+
+	srv_file_format = *(const uint*) save;
+
+	*(const char**) var_ptr
+		= trx_sys_file_format_id_to_name(srv_file_format);
+}
+
+/*****************************************************************
+Check if valid argument to innodb_file_format_check. This
+function is registered as a callback with MySQL. */
+static
+int
+innodb_file_format_check_validate(
+/*==============================*/
+						/* out: 0 for valid file
+						format */
+	THD*				thd,	/* in: thread handle */
+	struct st_mysql_sys_var*	var,	/* in: pointer to system
+						variable */
+	void*				save,	/* out: immediate result
+						for update function */
+	struct st_mysql_value*		value)	/* in: incoming string */
+{
+	const char*	file_format_input;
+	char		buff[STRING_BUFFER_USUAL_SIZE];
+	int		len = sizeof(buff);
+
+	ut_a(save != NULL);
+	ut_a(value != NULL);
+
+	file_format_input = value->val_str(value, buff, &len);
+
+	if (file_format_input != NULL) {
+
+		/* Check if user set on/off, we want to print a suitable
+		message if they did so. */
+
+		if (innobase_file_format_check_on_off(file_format_input)) {
+			sql_print_warning(
+				"InnoDB: invalid innodb_file_format_check "
+				"value; on/off can only be set at startup or "
+				"in the configuration file");
+		} else if (innobase_file_format_check_validate(
+				file_format_input)) {
+
+			uint	format_id;
+
+			format_id = innobase_file_format_name_lookup(
+				file_format_input);
+
+			ut_a(format_id <= DICT_TF_FORMAT_MAX);
+
+			*(uint*) save = format_id;
+
+			return(0);
+
+		} else {
+			sql_print_warning(
+				"InnoDB: invalid innodb_file_format_check "
+				"value; can be any format up to %s "
+				"or its equivalent numeric id",
+				trx_sys_file_format_id_to_name(
+					DICT_TF_FORMAT_MAX));
+		}
+	}
+
+	return(1);
+}
+
+/********************************************************************
+Update the system variable innodb_file_format_check using the "saved"
+value. This function is registered as a callback with MySQL. */
+static
+void
+innodb_file_format_check_update(
+/*============================*/
+	THD*				thd,		/* in: thread handle */
+	struct st_mysql_sys_var*	var,		/* in: pointer to
+							system variable */
+	void*				var_ptr,	/* out: where the
+							formal string goes */
+	const void*			save)		/* in: immediate result
+							from check function */
+{
+	uint	format_id;
+
+	ut_a(save != NULL);
+	ut_a(var_ptr != NULL);
+
+	format_id = *(const uint*) save;
+
+	/* Update the max format id in the system tablespace. */
+	if (trx_sys_file_format_max_set(format_id, (const char**) var_ptr)) {
+		ut_print_timestamp(stderr);
+		fprintf(stderr,
+			" [Info] InnoDB: the file format in the system "
+			"tablespace is now set to %s.\n", *(char**) var_ptr);
+	}
+}
+
+/********************************************************************
+Update the system variable innodb_adaptive_hash_index using the "saved"
+value. This function is registered as a callback with MySQL. */
+static
+void
+innodb_adaptive_hash_index_update(
+/*==============================*/
+	THD*				thd,		/* in: thread handle */
+	struct st_mysql_sys_var*	var,		/* in: pointer to
+							system variable */
+	void*				var_ptr,	/* out: where the
+							formal string goes */
+	const void*			save)		/* in: immediate result
+							from check function */
+{
+	if (*(my_bool*) save) {
+		btr_search_enable();
+	} else {
+		btr_search_disable();
+	}
+}
+
+/*****************************************************************
+Check if it is a valid value of innodb_change_buffering.  This function is
+registered as a callback with MySQL. */
+static
+int
+innodb_change_buffering_validate(
+/*=============================*/
+						/* out: 0 for valid
+						innodb_change_buffering */
+	THD*				thd,	/* in: thread handle */
+	struct st_mysql_sys_var*	var,	/* in: pointer to system
+						variable */
+	void*				save,	/* out: immediate result
+						for update function */
+	struct st_mysql_value*		value)	/* in: incoming string */
+{
+	const char*	change_buffering_input;
+	char		buff[STRING_BUFFER_USUAL_SIZE];
+	int		len = sizeof(buff);
+
+	ut_a(save != NULL);
+	ut_a(value != NULL);
+
+	change_buffering_input = value->val_str(value, buff, &len);
+
+	if (change_buffering_input != NULL) {
+		ulint	use;
+
+		for (use = 0; use < UT_ARR_SIZE(innobase_change_buffering_values);
+		     use++) {
+			if (!innobase_strcasecmp(
+				    change_buffering_input,
+				    innobase_change_buffering_values[use])) {
+				*(ibuf_use_t*) save = (ibuf_use_t) use;
+				return(0);
+			}
+		}
+	}
+
+	return(1);
+}
+
+/********************************************************************
+Update the system variable innodb_change_buffering using the "saved"
+value. This function is registered as a callback with MySQL. */
+static
+void
+innodb_change_buffering_update(
+/*===========================*/
+	THD*				thd,		/* in: thread handle */
+	struct st_mysql_sys_var*	var,		/* in: pointer to
+							system variable */
+	void*				var_ptr,	/* out: where the
+							formal string goes */
+	const void*			save)		/* in: immediate result
+							from check function */
+{
+	ut_a(var_ptr != NULL);
+	ut_a(save != NULL);
+	ut_a((*(ibuf_use_t*) save) < IBUF_USE_COUNT);
+
+	ibuf_use = *(const ibuf_use_t*) save;
+
+	*(const char**) var_ptr = innobase_change_buffering_values[ibuf_use];
+}
+
+static int show_innodb_vars(THD *thd, SHOW_VAR *var, char *buff)
+{
+  innodb_export_status();
+  var->type= SHOW_ARRAY;
+  var->value= (char *) &innodb_status_variables;
+  return 0;
+}
+
+static SHOW_VAR innodb_status_variables_export[]= {
+  {"Innodb",                   (char*) &show_innodb_vars, SHOW_FUNC},
+  {NullS, NullS, SHOW_LONG}
+};
+
+static struct st_mysql_storage_engine innobase_storage_engine=
+{ MYSQL_HANDLERTON_INTERFACE_VERSION };
+
+/* plugin options */
+static MYSQL_SYSVAR_BOOL(checksums, innobase_use_checksums,
+  PLUGIN_VAR_NOCMDARG | PLUGIN_VAR_READONLY,
+  "Enable InnoDB checksums validation (enabled by default). "
+  "Disable with --skip-innodb-checksums.",
+  NULL, NULL, TRUE);
+
+static MYSQL_SYSVAR_STR(data_home_dir, innobase_data_home_dir,
+  PLUGIN_VAR_READONLY,
+  "The common part for InnoDB table spaces.",
+  NULL, NULL, NULL);
+
+static MYSQL_SYSVAR_BOOL(extra_undoslots, innobase_extra_undoslots,
+  PLUGIN_VAR_NOCMDARG | PLUGIN_VAR_READONLY,
+  "Enable to use about 4000 undo slots instead of default 1024. "
+  "#### Attention: Once you enable this parameter, "
+  "don't use the datafile for normal mysqld or ibbackup! ####",
+  NULL, NULL, FALSE);
+
+static MYSQL_SYSVAR_BOOL(overwrite_relay_log_info, innobase_overwrite_relay_log_info,
+  PLUGIN_VAR_NOCMDARG | PLUGIN_VAR_READONLY,
+  "During InnoDB crash recovery on slave overwrite relay-log.info "
+  "to align master log file position if information in InnoDB and relay-log.info is different.",
+  NULL, NULL, FALSE);
+
+static MYSQL_SYSVAR_BOOL(doublewrite, innobase_use_doublewrite,
+  PLUGIN_VAR_NOCMDARG | PLUGIN_VAR_READONLY,
+  "Enable InnoDB doublewrite buffer (enabled by default). "
+  "Disable with --skip-innodb-doublewrite.",
+  NULL, NULL, TRUE);
+
+static MYSQL_SYSVAR_ULONG(fast_shutdown, innobase_fast_shutdown,
+  PLUGIN_VAR_OPCMDARG,
+  "Speeds up the shutdown process of the InnoDB storage engine. Possible "
+  "values are 0, 1 (faster)"
+  /*
+    NetWare can't close unclosed files, can't automatically kill remaining
+    threads, etc, so on this OS we disable the crash-like InnoDB shutdown.
+  */
+  IF_NETWARE("", " or 2 (fastest - crash-like)")
+  ".",
+  NULL, NULL, 1, 0, IF_NETWARE(1,2), 0);
+
+static MYSQL_SYSVAR_BOOL(file_per_table, srv_file_per_table,
+  PLUGIN_VAR_NOCMDARG,
+  "Stores each InnoDB table to an .ibd file in the database dir.",
+  NULL, NULL, FALSE);
+
+static MYSQL_SYSVAR_STR(file_format, innobase_file_format_name,
+  PLUGIN_VAR_RQCMDARG,
+  "File format to use for new tables in .ibd files.",
+  innodb_file_format_name_validate,
+  innodb_file_format_name_update, "Antelope");
+
+static MYSQL_SYSVAR_STR(file_format_check, innobase_file_format_check,
+  PLUGIN_VAR_OPCMDARG,
+  "The highest file format in the tablespace.",
+  innodb_file_format_check_validate,
+  innodb_file_format_check_update,
+  "on");
+
+static MYSQL_SYSVAR_ULONG(flush_log_at_trx_commit, srv_flush_log_at_trx_commit,
+  PLUGIN_VAR_OPCMDARG,
+  "Set to 0 (write and flush once per second),"
+  " 1 (write and flush at each commit)"
+  " or 2 (write at commit, flush once per second).",
+  NULL, NULL, 1, 0, 2, 0);
+
+static MYSQL_SYSVAR_STR(flush_method, innobase_unix_file_flush_method,
+  PLUGIN_VAR_RQCMDARG | PLUGIN_VAR_READONLY,
+  "With which method to flush data.", NULL, NULL, NULL);
+
+static MYSQL_SYSVAR_BOOL(locks_unsafe_for_binlog, innobase_locks_unsafe_for_binlog,
+  PLUGIN_VAR_NOCMDARG | PLUGIN_VAR_READONLY,
+  "Force InnoDB to not use next-key locking, to use only row-level locking.",
+  NULL, NULL, FALSE);
+
+static MYSQL_SYSVAR_ULONG(show_verbose_locks, srv_show_verbose_locks,
+  PLUGIN_VAR_OPCMDARG,
+  "Whether to show records locked in SHOW INNODB STATUS.",
+  NULL, NULL, 0, 0, 1, 0);
+
+static MYSQL_SYSVAR_ULONG(show_locks_held, srv_show_locks_held,
+  PLUGIN_VAR_RQCMDARG,
+  "Number of locks held to print for each InnoDB transaction in SHOW INNODB STATUS.",
+  NULL, NULL, 10, 0, 1000, 0);
+
+#ifdef UNIV_LOG_ARCHIVE
+static MYSQL_SYSVAR_STR(log_arch_dir, innobase_log_arch_dir,
+  PLUGIN_VAR_RQCMDARG | PLUGIN_VAR_READONLY,
+  "Where full logs should be archived.", NULL, NULL, NULL);
+
+static MYSQL_SYSVAR_BOOL(log_archive, innobase_log_archive,
+  PLUGIN_VAR_OPCMDARG | PLUGIN_VAR_READONLY,
+  "Set to 1 if you want to have logs archived.", NULL, NULL, FALSE);
+#endif /* UNIV_LOG_ARCHIVE */
+
+static MYSQL_SYSVAR_STR(log_group_home_dir, innobase_log_group_home_dir,
+  PLUGIN_VAR_RQCMDARG | PLUGIN_VAR_READONLY,
+  "Path to InnoDB log files.", NULL, NULL, NULL);
+
+static MYSQL_SYSVAR_ULONG(max_dirty_pages_pct, srv_max_buf_pool_modified_pct,
+  PLUGIN_VAR_RQCMDARG,
+  "Percentage of dirty pages allowed in bufferpool.",
+  NULL, NULL, 90, 0, 100, 0);
+
+static MYSQL_SYSVAR_ULONG(max_purge_lag, srv_max_purge_lag,
+  PLUGIN_VAR_RQCMDARG,
+  "Desired maximum length of the purge queue (0 = no limit)",
+  NULL, NULL, 0, 0, ~0L, 0);
+
+static MYSQL_SYSVAR_BOOL(rollback_on_timeout, innobase_rollback_on_timeout,
+  PLUGIN_VAR_OPCMDARG | PLUGIN_VAR_READONLY,
+  "Roll back the complete transaction on lock wait timeout, for 4.x compatibility (disabled by default)",
+  NULL, NULL, FALSE);
+
+static MYSQL_SYSVAR_BOOL(status_file, innobase_create_status_file,
+  PLUGIN_VAR_OPCMDARG | PLUGIN_VAR_NOSYSVAR,
+  "Enable SHOW INNODB STATUS output in the innodb_status.<pid> file",
+  NULL, NULL, FALSE);
+
+static MYSQL_SYSVAR_BOOL(stats_on_metadata, innobase_stats_on_metadata,
+  PLUGIN_VAR_OPCMDARG,
+  "Enable statistics gathering for metadata commands such as SHOW TABLE STATUS (on by default)",
+  NULL, NULL, TRUE);
+
+static MYSQL_SYSVAR_ULONGLONG(stats_sample_pages, srv_stats_sample_pages,
+  PLUGIN_VAR_RQCMDARG,
+  "The number of index pages to sample when calculating statistics (default 8)",
+  NULL, NULL, 8, 1, ~0ULL, 0);
+
+static MYSQL_SYSVAR_BOOL(adaptive_hash_index, btr_search_enabled,
+  PLUGIN_VAR_OPCMDARG,
+  "Enable InnoDB adaptive hash index (enabled by default).  "
+  "Disable with --skip-innodb-adaptive-hash-index.",
+  NULL, innodb_adaptive_hash_index_update, TRUE);
+
+static MYSQL_SYSVAR_ULONG(replication_delay, srv_replication_delay,
+  PLUGIN_VAR_RQCMDARG,
+  "Replication thread delay (ms) on the slave server if "
+  "innodb_thread_concurrency is reached (0 by default)",
+  NULL, NULL, 0, 0, ~0UL, 0);
+
+static MYSQL_SYSVAR_LONG(additional_mem_pool_size, innobase_additional_mem_pool_size,
+  PLUGIN_VAR_RQCMDARG | PLUGIN_VAR_READONLY,
+  "Size of a memory pool InnoDB uses to store data dictionary information and other internal data structures.",
+  NULL, NULL, 1*1024*1024L, 512*1024L, LONG_MAX, 1024);
+
+static MYSQL_SYSVAR_ULONG(autoextend_increment, srv_auto_extend_increment,
+  PLUGIN_VAR_RQCMDARG,
+  "Data file autoextend increment in megabytes",
+  NULL, NULL, 8L, 1L, 1000L, 0);
+
+static MYSQL_SYSVAR_LONGLONG(buffer_pool_size, innobase_buffer_pool_size,
+  PLUGIN_VAR_RQCMDARG | PLUGIN_VAR_READONLY,
+  "The size of the memory buffer InnoDB uses to cache data and indexes of its tables.",
+  NULL, NULL, 8*1024*1024L, 5*1024*1024L, LONGLONG_MAX, 1024*1024L);
+
+static MYSQL_SYSVAR_ULONG(commit_concurrency, srv_commit_concurrency,
+  PLUGIN_VAR_RQCMDARG,
+  "Helps in performance tuning in heavily concurrent environments.",
+  NULL, NULL, 0, 0, 1000, 0);
+
+static MYSQL_SYSVAR_ULONG(concurrency_tickets, srv_n_free_tickets_to_enter,
+  PLUGIN_VAR_RQCMDARG,
+  "Number of times a thread is allowed to enter InnoDB within the same SQL query after it has once got the ticket",
+  NULL, NULL, 500L, 1L, ~0L, 0);
+
+static MYSQL_SYSVAR_LONG(file_io_threads, innobase_file_io_threads,
+  PLUGIN_VAR_RQCMDARG | PLUGIN_VAR_READONLY,
+  "Number of file I/O threads in InnoDB.",
+  NULL, NULL, 4, 4, 64, 0);
+
+static MYSQL_SYSVAR_LONG(force_recovery, innobase_force_recovery,
+  PLUGIN_VAR_RQCMDARG | PLUGIN_VAR_READONLY,
+  "Helps to save your data in case the disk image of the database becomes corrupt.",
+  NULL, NULL, 0, 0, 6, 0);
+
+static MYSQL_SYSVAR_LONG(log_buffer_size, innobase_log_buffer_size,
+  PLUGIN_VAR_RQCMDARG | PLUGIN_VAR_READONLY,
+  "The size of the buffer which InnoDB uses to write log to the log files on disk.",
+  NULL, NULL, 1024*1024L, 256*1024L, LONG_MAX, 1024);
+
+static MYSQL_SYSVAR_LONGLONG(log_file_size, innobase_log_file_size,
+  PLUGIN_VAR_RQCMDARG | PLUGIN_VAR_READONLY,
+  "Size of each log file in a log group.",
+  NULL, NULL, 5*1024*1024L, 1*1024*1024L, LONGLONG_MAX, 1024*1024L);
+
+static MYSQL_SYSVAR_LONG(log_files_in_group, innobase_log_files_in_group,
+  PLUGIN_VAR_RQCMDARG | PLUGIN_VAR_READONLY,
+  "Number of log files in the log group. InnoDB writes to the files in a circular fashion. Value 3 is recommended here.",
+  NULL, NULL, 2, 2, 100, 0);
+
+static MYSQL_SYSVAR_LONG(mirrored_log_groups, innobase_mirrored_log_groups,
+  PLUGIN_VAR_RQCMDARG | PLUGIN_VAR_READONLY,
+  "Number of identical copies of log groups we keep for the database. Currently this should be set to 1.",
+  NULL, NULL, 1, 1, 10, 0);
+
+static MYSQL_SYSVAR_LONG(open_files, innobase_open_files,
+  PLUGIN_VAR_RQCMDARG | PLUGIN_VAR_READONLY,
+  "How many files at the maximum InnoDB keeps open at the same time.",
+  NULL, NULL, 300L, 10L, LONG_MAX, 0);
+
+static MYSQL_SYSVAR_ULONG(sync_spin_loops, srv_n_spin_wait_rounds,
+  PLUGIN_VAR_RQCMDARG,
+  "Count of spin-loop rounds in InnoDB mutexes",
+  NULL, NULL, 20L, 0L, ~0L, 0);
+
+static MYSQL_SYSVAR_ULONG(thread_concurrency, srv_thread_concurrency,
+  PLUGIN_VAR_RQCMDARG,
+  "Helps in performance tuning in heavily concurrent environments. Sets the maximum number of threads allowed inside InnoDB. Value 0 will disable the thread throttling.",
+  NULL, NULL, 0, 0, 1000, 0);
+
+static MYSQL_SYSVAR_ULONG(thread_sleep_delay, srv_thread_sleep_delay,
+  PLUGIN_VAR_RQCMDARG,
+  "Time of innodb thread sleeping before joining InnoDB queue (usec). Value 0 disable a sleep",
+  NULL, NULL, 10000L, 0L, ~0L, 0);
+
+static MYSQL_SYSVAR_STR(data_file_path, innobase_data_file_path,
+  PLUGIN_VAR_RQCMDARG | PLUGIN_VAR_READONLY,
+  "Path to individual files and their sizes.",
+  NULL, NULL, NULL);
+
+static MYSQL_SYSVAR_LONG(autoinc_lock_mode, innobase_autoinc_lock_mode,
+  PLUGIN_VAR_RQCMDARG | PLUGIN_VAR_READONLY,
+  "The AUTOINC lock modes supported by InnoDB:               "
+  "0 => Old style AUTOINC locking (for backward"
+  " compatibility)                                           "
+  "1 => New style AUTOINC locking                            "
+  "2 => No AUTOINC locking (unsafe for SBR)",
+  NULL, NULL,
+  AUTOINC_NEW_STYLE_LOCKING,	/* Default setting */
+  AUTOINC_OLD_STYLE_LOCKING,	/* Minimum value */
+  AUTOINC_NO_LOCKING, 0);	/* Maximum value */
+
+static MYSQL_SYSVAR_STR(version, innodb_version_str,
+  PLUGIN_VAR_NOCMDOPT | PLUGIN_VAR_READONLY,
+  "Percona-InnoDB-plugin version", NULL, NULL, INNODB_VERSION_STR);
+
+static MYSQL_SYSVAR_BOOL(use_sys_malloc, srv_use_sys_malloc,
+  PLUGIN_VAR_NOCMDARG | PLUGIN_VAR_READONLY,
+  "Use OS memory allocator instead of InnoDB's internal memory allocator",
+  NULL, NULL, TRUE);
+
+static MYSQL_SYSVAR_STR(change_buffering, innobase_change_buffering,
+  PLUGIN_VAR_RQCMDARG,
+  "Buffer changes to reduce random access: "
+  "OFF, ON, inserting, deleting, changing, or purging.",
+  innodb_change_buffering_validate,
+  innodb_change_buffering_update, NULL);
+
+static MYSQL_SYSVAR_ULONG(io_capacity, srv_io_capacity,
+  PLUGIN_VAR_RQCMDARG,
+  "Number of IO operations per second the server can do. Tunes background IO rate.",
+  NULL, NULL, 100, 100, 999999999, 0);
+
+static MYSQL_SYSVAR_LONGLONG(ibuf_max_size, srv_ibuf_max_size,
+  PLUGIN_VAR_RQCMDARG | PLUGIN_VAR_READONLY,
+  "The maximum size of the insert buffer. (in bytes)",
+  NULL, NULL, LONGLONG_MAX, 0, LONGLONG_MAX, 0);
+
+static MYSQL_SYSVAR_ULONG(ibuf_active_contract, srv_ibuf_active_contract,
+  PLUGIN_VAR_RQCMDARG,
+  "Enable/Disable active_contract of insert buffer. 0:disable 1:enable",
+  NULL, NULL, 0, 0, 1, 0);
+
+static MYSQL_SYSVAR_ULONG(ibuf_accel_rate, srv_ibuf_accel_rate,
+  PLUGIN_VAR_RQCMDARG,
+  "Tunes amount of insert buffer processing of background, in addition to innodb_io_capacity. (in percentage)",
+  NULL, NULL, 100, 100, 999999999, 0);
+
+static MYSQL_SYSVAR_ULONG(flush_neighbor_pages, srv_flush_neighbor_pages,
+  PLUGIN_VAR_RQCMDARG,
+  "Enable/Disable flushing also neighbor pages. 0:disable 1:enable",
+  NULL, NULL, 1, 0, 1, 0);
+
+static
+void
+innodb_read_ahead_update(
+  THD* thd,
+  struct st_mysql_sys_var*     var,
+  void*        var_ptr,
+  const void*  save)
+{
+  *(long *)var_ptr= (*(long *)save) & 3;
+}
+const char *read_ahead_names[]=
+{
+  "none", /* 0 */
+  "random",
+  "linear",
+  "both", /* 3 */
+  /* For compatibility of the older patch */
+  "0", /* 4 ("none" + 4) */
+  "1",
+  "2",
+  "3", /* 7 ("both" + 4) */
+  NullS
+};
+TYPELIB read_ahead_typelib=
+{
+  array_elements(read_ahead_names) - 1, "read_ahead_typelib",
+  read_ahead_names, NULL
+};
+static MYSQL_SYSVAR_ENUM(read_ahead, srv_read_ahead,
+  PLUGIN_VAR_RQCMDARG,
+  "Control read ahead activity. (none, random, linear, [both])",
+  NULL, innodb_read_ahead_update, 3, &read_ahead_typelib);
+
+static MYSQL_SYSVAR_ULONG(adaptive_checkpoint, srv_adaptive_checkpoint,
+  PLUGIN_VAR_RQCMDARG,
+  "Enable/Disable flushing along modified age. 0:disable 1:enable",
+  NULL, NULL, 0, 0, 1, 0);
+
+static MYSQL_SYSVAR_ULONG(enable_unsafe_group_commit, srv_enable_unsafe_group_commit,
+  PLUGIN_VAR_RQCMDARG,
+  "Enable/Disable unsafe group commit when support_xa=OFF and use with binlog or other XA storage engine.",
+  NULL, NULL, 0, 0, 1, 0);
+
+static MYSQL_SYSVAR_ULONG(read_io_threads, innobase_read_io_threads,
+  PLUGIN_VAR_RQCMDARG | PLUGIN_VAR_READONLY,
+  "Number of background read I/O threads in InnoDB.",
+  NULL, NULL, 1, 1, 64, 0);
+
+static MYSQL_SYSVAR_ULONG(write_io_threads, innobase_write_io_threads,
+  PLUGIN_VAR_RQCMDARG | PLUGIN_VAR_READONLY,
+  "Number of background write I/O threads in InnoDB.",
+  NULL, NULL, 1, 1, 64, 0);
+
+static MYSQL_SYSVAR_ULONG(extra_rsegments, srv_extra_rsegments,
+  PLUGIN_VAR_RQCMDARG | PLUGIN_VAR_READONLY,
+  "Number of extra user rollback segments when create new database.",
+  NULL, NULL, 0, 0, 127, 0);
+
+static struct st_mysql_sys_var* innobase_system_variables[]= {
+  MYSQL_SYSVAR(additional_mem_pool_size),
+  MYSQL_SYSVAR(autoextend_increment),
+  MYSQL_SYSVAR(buffer_pool_size),
+  MYSQL_SYSVAR(checksums),
+  MYSQL_SYSVAR(commit_concurrency),
+  MYSQL_SYSVAR(concurrency_tickets),
+  MYSQL_SYSVAR(data_file_path),
+  MYSQL_SYSVAR(data_home_dir),
+  MYSQL_SYSVAR(doublewrite),
+  MYSQL_SYSVAR(extra_undoslots),
+  MYSQL_SYSVAR(fast_shutdown),
+  MYSQL_SYSVAR(file_io_threads),
+  MYSQL_SYSVAR(file_per_table),
+  MYSQL_SYSVAR(file_format),
+  MYSQL_SYSVAR(file_format_check),
+  MYSQL_SYSVAR(flush_log_at_trx_commit),
+  MYSQL_SYSVAR(flush_method),
+  MYSQL_SYSVAR(force_recovery),
+  MYSQL_SYSVAR(locks_unsafe_for_binlog),
+  MYSQL_SYSVAR(lock_wait_timeout),
+#ifdef UNIV_LOG_ARCHIVE
+  MYSQL_SYSVAR(log_arch_dir),
+  MYSQL_SYSVAR(log_archive),
+#endif /* UNIV_LOG_ARCHIVE */
+  MYSQL_SYSVAR(log_buffer_size),
+  MYSQL_SYSVAR(log_file_size),
+  MYSQL_SYSVAR(log_files_in_group),
+  MYSQL_SYSVAR(log_group_home_dir),
+  MYSQL_SYSVAR(max_dirty_pages_pct),
+  MYSQL_SYSVAR(max_purge_lag),
+  MYSQL_SYSVAR(mirrored_log_groups),
+  MYSQL_SYSVAR(open_files),
+  MYSQL_SYSVAR(overwrite_relay_log_info),
+  MYSQL_SYSVAR(rollback_on_timeout),
+  MYSQL_SYSVAR(stats_on_metadata),
+  MYSQL_SYSVAR(stats_sample_pages),
+  MYSQL_SYSVAR(adaptive_hash_index),
+  MYSQL_SYSVAR(replication_delay),
+  MYSQL_SYSVAR(status_file),
+  MYSQL_SYSVAR(strict_mode),
+  MYSQL_SYSVAR(support_xa),
+  MYSQL_SYSVAR(sync_spin_loops),
+  MYSQL_SYSVAR(table_locks),
+  MYSQL_SYSVAR(thread_concurrency),
+  MYSQL_SYSVAR(thread_sleep_delay),
+  MYSQL_SYSVAR(autoinc_lock_mode),
+  MYSQL_SYSVAR(show_verbose_locks),
+  MYSQL_SYSVAR(show_locks_held),
+  MYSQL_SYSVAR(version),
+  MYSQL_SYSVAR(io_capacity),
+  MYSQL_SYSVAR(ibuf_max_size),
+  MYSQL_SYSVAR(ibuf_active_contract),
+  MYSQL_SYSVAR(ibuf_accel_rate),
+  MYSQL_SYSVAR(flush_neighbor_pages),
+  MYSQL_SYSVAR(read_ahead),
+  MYSQL_SYSVAR(adaptive_checkpoint),
+  MYSQL_SYSVAR(enable_unsafe_group_commit),
+  MYSQL_SYSVAR(read_io_threads),
+  MYSQL_SYSVAR(write_io_threads),
+  MYSQL_SYSVAR(extra_rsegments),
+  MYSQL_SYSVAR(use_sys_malloc),
+  MYSQL_SYSVAR(change_buffering),
+  NULL
+};
+
+#ifdef MYSQL_DYNAMIC_PLUGIN
+struct st_mysql_sys_var
+{
+	MYSQL_PLUGIN_VAR_HEADER;
+	void* value;
+};
+
+struct param_mapping
+{
+	const char*	server;		/* Parameter name in the server. */
+	const char*	plugin;		/* Paramater name in the plugin. */
+};
+
+/********************************************************************
+Match the parameters from the static and dynamic versions. */
+static
+bool
+innobase_match_parameter(
+/*=====================*/
+					/* out: true if names match */
+	const char*	from_server,	/* in: variable name from server */
+	const char*	from_plugin)	/* in: variable name from plugin */
+{
+	static const param_mapping param_map[] = {
+		{"use_adaptive_hash_indexes", "adaptive_hash_index"}
+	};
+
+	if (strcmp(from_server, from_plugin) == 0) {
+		return(true);
+	}
+
+	const param_mapping*	param = param_map;
+	int	n_elems = sizeof(param_map) / sizeof(param_map[0]);
+
+	for (int i = 0; i < n_elems; ++i, ++param) {
+
+		if (strcmp(param->server, from_server) == 0
+		    && strcmp(param->plugin, from_plugin) == 0) {
+
+			return(true);
+		}
+	}
+
+	return(false);
+}
+
+/********************************************************************
+Copy InnoDB system variables from the static InnoDB to the dynamic
+plugin. */
+static
+bool
+innodb_plugin_init(void)
+/*====================*/
+		/* out: TRUE if the dynamic InnoDB plugin should start */
+{
+#if !MYSQL_STORAGE_ENGINE_PLUGIN
+#error "MYSQL_STORAGE_ENGINE_PLUGIN must be nonzero."
+#endif
+
+	/* Copy the system variables. */
+
+	struct st_mysql_plugin*		builtin;
+	struct st_mysql_sys_var**	sta; /* static parameters */
+	struct st_mysql_sys_var**	dyn; /* dynamic parameters */
+
+#ifdef __WIN__
+	if (!builtin_innobase_plugin_ptr) {
+
+		return(true);
+	}
+
+	builtin = builtin_innobase_plugin_ptr;
+#else
+	switch (builtin_innobase_plugin) {
+	case 0:
+		return(true);
+	case MYSQL_STORAGE_ENGINE_PLUGIN:
+		break;
+	default:
+		return(false);
+	}
+
+	builtin = (struct st_mysql_plugin*) &builtin_innobase_plugin;
+#endif
+
+	for (sta = builtin->system_vars; *sta != NULL; sta++) {
+
+		for (dyn = innobase_system_variables; *dyn != NULL; dyn++) {
+
+			/* do not copy session variables */
+			if (((*sta)->flags | (*dyn)->flags)
+			    & PLUGIN_VAR_THDLOCAL) {
+				continue;
+			}
+
+			if (innobase_match_parameter((*sta)->name,
+						     (*dyn)->name)) {
+
+				/* found the corresponding parameter */
+
+				/* check if the flags are the same,
+				ignoring differences in the READONLY or
+				NOSYSVAR flags;
+				e.g. we are not copying string variable to
+				an integer one, but we do not care if it is
+				readonly in the static and not in the
+				dynamic */
+				if (((*sta)->flags ^ (*dyn)->flags)
+				    & ~(PLUGIN_VAR_READONLY
+					| PLUGIN_VAR_NOSYSVAR)) {
+
+					fprintf(stderr,
+						"InnoDB: %s in static InnoDB "
+						"(flags=0x%x) differs from "
+						"%s in dynamic InnoDB "
+						"(flags=0x%x)\n",
+						(*sta)->name, (*sta)->flags,
+						(*dyn)->name, (*dyn)->flags);
+
+					/* we could break; here leaving this
+					parameter uncopied */
+					return(false);
+				}
+
+				/* assign the value of the static parameter
+				to the dynamic one, according to their type */
+
+#define COPY_VAR(label, type)					\
+	case label:						\
+		*(type*)(*dyn)->value = *(type*)(*sta)->value;	\
+		break;
+
+				switch ((*sta)->flags
+					& ~(PLUGIN_VAR_MASK
+					    | PLUGIN_VAR_UNSIGNED)) {
+
+				COPY_VAR(PLUGIN_VAR_BOOL, char);
+				COPY_VAR(PLUGIN_VAR_INT, int);
+				COPY_VAR(PLUGIN_VAR_LONG, long);
+				COPY_VAR(PLUGIN_VAR_LONGLONG, long long);
+				COPY_VAR(PLUGIN_VAR_STR, char*);
+
+				default:
+					fprintf(stderr,
+						"InnoDB: unknown flags "
+						"0x%x for %s\n",
+						(*sta)->flags, (*sta)->name);
+				}
+
+				/* Make the static InnoDB variable point to
+				the dynamic one */
+				(*sta)->value = (*dyn)->value;
+
+				break;
+			}
+		}
+	}
+
+	return(true);
+}
+#endif /* MYSQL_DYNAMIC_PLUGIN */
+
+mysql_declare_plugin(innobase)
+{
+  MYSQL_STORAGE_ENGINE_PLUGIN,
+  &innobase_storage_engine,
+  innobase_hton_name,
+  "Innobase Oy",
+  "Supports transactions, row-level locking, and foreign keys",
+  PLUGIN_LICENSE_GPL,
+  innobase_init, /* Plugin Init */
+  NULL, /* Plugin Deinit */
+  INNODB_VERSION_SHORT,
+  innodb_status_variables_export,/* status variables             */
+  innobase_system_variables, /* system variables */
+  NULL /* reserved */
+},
+i_s_innodb_rseg,
+i_s_innodb_buffer_pool_pages,
+i_s_innodb_buffer_pool_pages_index,
+i_s_innodb_buffer_pool_pages_blob,
+i_s_innodb_trx,
+i_s_innodb_locks,
+i_s_innodb_lock_waits,
+i_s_innodb_cmp,
+i_s_innodb_cmp_reset,
+i_s_innodb_cmpmem,
+i_s_innodb_cmpmem_reset,
+i_s_innodb_patches
+mysql_declare_plugin_end;
+
+#ifdef UNIV_COMPILE_TEST_FUNCS
+
+typedef struct innobase_convert_name_test_struct {
+	char*		buf;
+	ulint		buflen;
+	const char*	id;
+	ulint		idlen;
+	void*		thd;
+	ibool		file_id;
+
+	const char*	expected;
+} innobase_convert_name_test_t;
+
+void
+test_innobase_convert_name()
+{
+	char	buf[1024];
+	ulint	i;
+
+	innobase_convert_name_test_t test_input[] = {
+		{buf, sizeof(buf), "abcd", 4, NULL, TRUE, "\"abcd\""},
+		{buf, 7, "abcd", 4, NULL, TRUE, "\"abcd\""},
+		{buf, 6, "abcd", 4, NULL, TRUE, "\"abcd\""},
+		{buf, 5, "abcd", 4, NULL, TRUE, "\"abc\""},
+		{buf, 4, "abcd", 4, NULL, TRUE, "\"ab\""},
+
+		{buf, sizeof(buf), "ab@0060cd", 9, NULL, TRUE, "\"ab`cd\""},
+		{buf, 9, "ab@0060cd", 9, NULL, TRUE, "\"ab`cd\""},
+		{buf, 8, "ab@0060cd", 9, NULL, TRUE, "\"ab`cd\""},
+		{buf, 7, "ab@0060cd", 9, NULL, TRUE, "\"ab`cd\""},
+		{buf, 6, "ab@0060cd", 9, NULL, TRUE, "\"ab`c\""},
+		{buf, 5, "ab@0060cd", 9, NULL, TRUE, "\"ab`\""},
+		{buf, 4, "ab@0060cd", 9, NULL, TRUE, "\"ab\""},
+
+		{buf, sizeof(buf), "ab\"cd", 5, NULL, TRUE,
+			"\"#mysql50#ab\"\"cd\""},
+		{buf, 17, "ab\"cd", 5, NULL, TRUE,
+			"\"#mysql50#ab\"\"cd\""},
+		{buf, 16, "ab\"cd", 5, NULL, TRUE,
+			"\"#mysql50#ab\"\"c\""},
+		{buf, 15, "ab\"cd", 5, NULL, TRUE,
+			"\"#mysql50#ab\"\"\""},
+		{buf, 14, "ab\"cd", 5, NULL, TRUE,
+			"\"#mysql50#ab\""},
+		{buf, 13, "ab\"cd", 5, NULL, TRUE,
+			"\"#mysql50#ab\""},
+		{buf, 12, "ab\"cd", 5, NULL, TRUE,
+			"\"#mysql50#a\""},
+		{buf, 11, "ab\"cd", 5, NULL, TRUE,
+			"\"#mysql50#\""},
+		{buf, 10, "ab\"cd", 5, NULL, TRUE,
+			"\"#mysql50\""},
+
+		{buf, sizeof(buf), "ab/cd", 5, NULL, TRUE, "\"ab\".\"cd\""},
+		{buf, 9, "ab/cd", 5, NULL, TRUE, "\"ab\".\"cd\""},
+		{buf, 8, "ab/cd", 5, NULL, TRUE, "\"ab\".\"c\""},
+		{buf, 7, "ab/cd", 5, NULL, TRUE, "\"ab\".\"\""},
+		{buf, 6, "ab/cd", 5, NULL, TRUE, "\"ab\"."},
+		{buf, 5, "ab/cd", 5, NULL, TRUE, "\"ab\"."},
+		{buf, 4, "ab/cd", 5, NULL, TRUE, "\"ab\""},
+		{buf, 3, "ab/cd", 5, NULL, TRUE, "\"a\""},
+		{buf, 2, "ab/cd", 5, NULL, TRUE, "\"\""},
+		/* XXX probably "" is a better result in this case
+		{buf, 1, "ab/cd", 5, NULL, TRUE, "."},
+		*/
+		{buf, 0, "ab/cd", 5, NULL, TRUE, ""},
+	};
+
+	for (i = 0; i < sizeof(test_input) / sizeof(test_input[0]); i++) {
+
+		char*	end;
+		ibool	ok = TRUE;
+		size_t	res_len;
+
+		fprintf(stderr, "TESTING %lu, %s, %lu, %s\n",
+			test_input[i].buflen,
+			test_input[i].id,
+			test_input[i].idlen,
+			test_input[i].expected);
+
+		end = innobase_convert_name(
+			test_input[i].buf,
+			test_input[i].buflen,
+			test_input[i].id,
+			test_input[i].idlen,
+			test_input[i].thd,
+			test_input[i].file_id);
+
+		res_len = (size_t) (end - test_input[i].buf);
+
+		if (res_len != strlen(test_input[i].expected)) {
+
+			fprintf(stderr, "unexpected len of the result: %u, "
+				"expected: %u\n", (unsigned) res_len,
+				(unsigned) strlen(test_input[i].expected));
+			ok = FALSE;
+		}
+
+		if (memcmp(test_input[i].buf,
+			   test_input[i].expected,
+			   strlen(test_input[i].expected)) != 0
+		    || !ok) {
+
+			fprintf(stderr, "unexpected result: %.*s, "
+				"expected: %s\n", (int) res_len,
+				test_input[i].buf,
+				test_input[i].expected);
+			ok = FALSE;
+		}
+
+		if (ok) {
+			fprintf(stderr, "OK: res: %.*s\n\n", (int) res_len,
+				buf);
+		} else {
+			fprintf(stderr, "FAILED\n\n");
+			return;
+		}
+	}
+}
+
+#endif /* UNIV_COMPILE_TEST_FUNCS */
diff --git a/storage/xtradb/handler/ha_innodb.h b/storage/xtradb/handler/ha_innodb.h
new file mode 100644
index 00000000000..c08dd3ed173
--- /dev/null
+++ b/storage/xtradb/handler/ha_innodb.h
@@ -0,0 +1,283 @@
+/*****************************************************************************
+
+Copyright (c) 2000, 2009, MySQL AB & Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/*
+  This file is based on ha_berkeley.h of MySQL distribution
+
+  This file defines the Innodb handler: the interface between MySQL and
+  Innodb
+*/
+
+#ifdef USE_PRAGMA_INTERFACE
+#pragma interface			/* gcc class implementation */
+#endif
+
+typedef struct st_innobase_share {
+  THR_LOCK lock;
+  pthread_mutex_t mutex;
+  const char* table_name;
+  uint use_count;
+  void* table_name_hash;
+} INNOBASE_SHARE;
+
+
+struct dict_index_struct;
+struct row_prebuilt_struct;
+
+typedef struct dict_index_struct dict_index_t;
+typedef struct row_prebuilt_struct row_prebuilt_t;
+
+/* The class defining a handle to an Innodb table */
+class ha_innobase: public handler
+{
+	row_prebuilt_t*	prebuilt;	/* prebuilt struct in InnoDB, used
+					to save CPU time with prebuilt data
+					structures*/
+	THD*		user_thd;	/* the thread handle of the user
+					currently using the handle; this is
+					set in external_lock function */
+	THR_LOCK_DATA	lock;
+	INNOBASE_SHARE	*share;
+
+	uchar*		upd_buff;	/* buffer used in updates */
+	uchar*		key_val_buff;	/* buffer used in converting
+					search key values from MySQL format
+					to Innodb format */
+	ulong		upd_and_key_val_buff_len;
+					/* the length of each of the previous
+					two buffers */
+	Table_flags	int_table_flags;
+	uint		primary_key;
+	ulong		start_of_scan;	/* this is set to 1 when we are
+					starting a table scan but have not
+					yet fetched any row, else 0 */
+	uint		last_match_mode;/* match mode of the latest search:
+					ROW_SEL_EXACT, ROW_SEL_EXACT_PREFIX,
+					or undefined */
+	uint		num_write_row;	/* number of write_row() calls */
+
+	uint store_key_val_for_row(uint keynr, char* buff, uint buff_len,
+                                   const uchar* record);
+	inline void update_thd(THD* thd);
+	void update_thd();
+	int change_active_index(uint keynr);
+	int general_fetch(uchar* buf, uint direction, uint match_mode);
+	ulint innobase_lock_autoinc();
+	ulonglong innobase_peek_autoinc();
+	ulint innobase_set_max_autoinc(ulonglong auto_inc);
+	ulint innobase_reset_autoinc(ulonglong auto_inc);
+	ulint innobase_get_autoinc(ulonglong* value);
+	ulint innobase_update_autoinc(ulonglong	auto_inc);
+	ulint innobase_initialize_autoinc();
+	dict_index_t* innobase_get_index(uint keynr);
+ 	ulonglong innobase_get_int_col_max_value(const Field* field);
+
+	/* Init values for the class: */
+ public:
+	ha_innobase(handlerton *hton, TABLE_SHARE *table_arg);
+	~ha_innobase();
+	/*
+	  Get the row type from the storage engine.  If this method returns
+	  ROW_TYPE_NOT_USED, the information in HA_CREATE_INFO should be used.
+	*/
+	enum row_type get_row_type() const;
+
+	const char* table_type() const;
+	const char* index_type(uint key_number);
+	const char** bas_ext() const;
+	Table_flags table_flags() const;
+	ulong index_flags(uint idx, uint part, bool all_parts) const;
+	uint max_supported_keys() const;
+	uint max_supported_key_length() const;
+	uint max_supported_key_part_length() const;
+	const key_map* keys_to_use_for_scanning();
+
+	int open(const char *name, int mode, uint test_if_locked);
+	int close(void);
+	double scan_time();
+	double read_time(uint index, uint ranges, ha_rows rows);
+
+	int write_row(uchar * buf);
+	int update_row(const uchar * old_data, uchar * new_data);
+	int delete_row(const uchar * buf);
+	bool was_semi_consistent_read();
+	void try_semi_consistent_read(bool yes);
+	void unlock_row();
+
+#ifdef ROW_MERGE_IS_INDEX_USABLE
+	/** Check if an index can be used by this transaction.
+	* @param keynr	key number to check
+	* @return	true if available, false if the index
+	*		does not contain old records that exist
+	*		in the read view of this transaction */
+	bool is_index_available(uint keynr);
+#endif /* ROW_MERGE_IS_INDEX_USABLE */
+	int index_init(uint index, bool sorted);
+	int index_end();
+	int index_read(uchar * buf, const uchar * key,
+		uint key_len, enum ha_rkey_function find_flag);
+	int index_read_idx(uchar * buf, uint index, const uchar * key,
+			   uint key_len, enum ha_rkey_function find_flag);
+	int index_read_last(uchar * buf, const uchar * key, uint key_len);
+	int index_next(uchar * buf);
+	int index_next_same(uchar * buf, const uchar *key, uint keylen);
+	int index_prev(uchar * buf);
+	int index_first(uchar * buf);
+	int index_last(uchar * buf);
+
+	int rnd_init(bool scan);
+	int rnd_end();
+	int rnd_next(uchar *buf);
+	int rnd_pos(uchar * buf, uchar *pos);
+
+	void position(const uchar *record);
+	int info(uint);
+	int analyze(THD* thd,HA_CHECK_OPT* check_opt);
+	int optimize(THD* thd,HA_CHECK_OPT* check_opt);
+	int discard_or_import_tablespace(my_bool discard);
+	int extra(enum ha_extra_function operation);
+        int reset();
+	int external_lock(THD *thd, int lock_type);
+	int transactional_table_lock(THD *thd, int lock_type);
+	int start_stmt(THD *thd, thr_lock_type lock_type);
+	void position(uchar *record);
+	ha_rows records_in_range(uint inx, key_range *min_key, key_range
+								*max_key);
+	ha_rows estimate_rows_upper_bound();
+
+	void update_create_info(HA_CREATE_INFO* create_info);
+	int create(const char *name, register TABLE *form,
+					HA_CREATE_INFO *create_info);
+	int delete_all_rows();
+	int delete_table(const char *name);
+	int rename_table(const char* from, const char* to);
+	int check(THD* thd, HA_CHECK_OPT* check_opt);
+	char* update_table_comment(const char* comment);
+	char* get_foreign_key_create_info();
+	int get_foreign_key_list(THD *thd, List<FOREIGN_KEY_INFO> *f_key_list);
+	bool can_switch_engines();
+	uint referenced_by_foreign_key();
+	void free_foreign_key_create_info(char* str);
+	THR_LOCK_DATA **store_lock(THD *thd, THR_LOCK_DATA **to,
+					enum thr_lock_type lock_type);
+	void init_table_handle_for_HANDLER();
+        virtual void get_auto_increment(ulonglong offset, ulonglong increment,
+                                        ulonglong nb_desired_values,
+                                        ulonglong *first_value,
+                                        ulonglong *nb_reserved_values);
+	int reset_auto_increment(ulonglong value);
+
+	virtual bool get_error_message(int error, String *buf);
+
+	uint8 table_cache_type();
+	/*
+	  ask handler about permission to cache table during query registration
+	*/
+	my_bool register_query_cache_table(THD *thd, char *table_key,
+					   uint key_length,
+					   qc_engine_callback *call_back,
+					   ulonglong *engine_data);
+	static char *get_mysql_bin_log_name();
+	static ulonglong get_mysql_bin_log_pos();
+	bool primary_key_is_clustered();
+	int cmp_ref(const uchar *ref1, const uchar *ref2);
+	/** Fast index creation (smart ALTER TABLE) @see handler0alter.cc @{ */
+	int add_index(TABLE *table_arg, KEY *key_info, uint num_of_keys);
+	int prepare_drop_index(TABLE *table_arg, uint *key_num,
+			       uint num_of_keys);
+	int final_drop_index(TABLE *table_arg);
+	/** @} */
+	bool check_if_incompatible_data(HA_CREATE_INFO *info,
+					uint table_changes);
+};
+
+/* Some accessor functions which the InnoDB plugin needs, but which
+can not be added to mysql/plugin.h as part of the public interface;
+the definitions are bracketed with #ifdef INNODB_COMPATIBILITY_HOOKS */
+
+#ifndef INNODB_COMPATIBILITY_HOOKS
+#error InnoDB needs MySQL to be built with #define INNODB_COMPATIBILITY_HOOKS
+#endif
+
+extern "C" {
+struct charset_info_st *thd_charset(MYSQL_THD thd);
+char **thd_query(MYSQL_THD thd);
+
+/** Get the file name of the MySQL binlog.
+ * @return the name of the binlog file
+ */
+const char* mysql_bin_log_file_name(void);
+
+/** Get the current position of the MySQL binlog.
+ * @return byte offset from the beginning of the binlog
+ */
+ulonglong mysql_bin_log_file_pos(void);
+
+/**
+  Check if a user thread is a replication slave thread
+  @param thd  user thread
+  @retval 0 the user thread is not a replication slave thread
+  @retval 1 the user thread is a replication slave thread
+*/
+int thd_slave_thread(const MYSQL_THD thd);
+
+/**
+  Check if a user thread is running a non-transactional update
+  @param thd  user thread
+  @retval 0 the user thread is not running a non-transactional update
+  @retval 1 the user thread is running a non-transactional update
+*/
+int thd_non_transactional_update(const MYSQL_THD thd);
+
+/**
+  Get the user thread's binary logging format
+  @param thd  user thread
+  @return Value to be used as index into the binlog_format_names array
+*/
+int thd_binlog_format(const MYSQL_THD thd);
+
+/**
+  Mark transaction to rollback and mark error as fatal to a sub-statement.
+  @param  thd   Thread handle
+  @param  all   TRUE <=> rollback main transaction.
+*/
+void thd_mark_transaction_to_rollback(MYSQL_THD thd, bool all);
+}
+
+typedef struct trx_struct trx_t;
+/************************************************************************
+Converts an InnoDB error code to a MySQL error code and also tells to MySQL
+about a possible transaction rollback inside InnoDB caused by a lock wait
+timeout or a deadlock. */
+extern "C"
+int
+convert_error_code_to_mysql(
+/*========================*/
+				/* out: MySQL error code */
+	int		error,	/* in: InnoDB error code */
+	ulint		flags,	/* in: InnoDB table flags, or 0 */
+	MYSQL_THD	thd);	/* in: user thread handle or NULL */
+
+/*************************************************************************
+Allocates an InnoDB transaction for a MySQL handler object. */
+extern "C"
+trx_t*
+innobase_trx_allocate(
+/*==================*/
+				/* out: InnoDB transaction handle */
+	MYSQL_THD	thd);	/* in: user thread handle */
diff --git a/storage/xtradb/handler/handler0alter.cc b/storage/xtradb/handler/handler0alter.cc
new file mode 100644
index 00000000000..1b5466e66eb
--- /dev/null
+++ b/storage/xtradb/handler/handler0alter.cc
@@ -0,0 +1,1214 @@
+/*****************************************************************************
+
+Copyright (c) 2005, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Smart ALTER TABLE
+*******************************************************/
+
+#include <mysql_priv.h>
+#include <mysqld_error.h>
+
+extern "C" {
+#include "log0log.h"
+#include "row0merge.h"
+#include "srv0srv.h"
+#include "trx0trx.h"
+#include "trx0roll.h"
+#include "ha_prototypes.h"
+#include "handler0alter.h"
+}
+
+#include "ha_innodb.h"
+#include "handler0vars.h"
+
+/*****************************************************************
+Copies an InnoDB column to a MySQL field.  This function is
+adapted from row_sel_field_store_in_mysql_format(). */
+static
+void
+innobase_col_to_mysql(
+/*==================*/
+	const dict_col_t*	col,	/* in: InnoDB column */
+	const uchar*		data,	/* in: InnoDB column data */
+	ulint			len,	/* in: length of data, in bytes */
+	Field*			field)	/* in/out: MySQL field */
+{
+	uchar*	ptr;
+	uchar*	dest	= field->ptr;
+	ulint	flen	= field->pack_length();
+
+	switch (col->mtype) {
+	case DATA_INT:
+		ut_ad(len == flen);
+
+		/* Convert integer data from Innobase to little-endian
+		format, sign bit restored to normal */
+
+		for (ptr = dest + len; ptr != dest; ) {
+			*--ptr = *data++;
+		}
+
+		if (!(field->flags & UNSIGNED_FLAG)) {
+			((byte*) dest)[len - 1] ^= 0x80;
+		}
+
+		break;
+
+	case DATA_VARCHAR:
+	case DATA_VARMYSQL:
+	case DATA_BINARY:
+		field->reset();
+
+		if (field->type() == MYSQL_TYPE_VARCHAR) {
+			/* This is a >= 5.0.3 type true VARCHAR. Store the
+			length of the data to the first byte or the first
+			two bytes of dest. */
+
+			dest = row_mysql_store_true_var_len(
+				dest, len, flen - field->key_length());
+		}
+
+		/* Copy the actual data */
+		memcpy(dest, data, len);
+		break;
+
+	case DATA_BLOB:
+		/* Store a pointer to the BLOB buffer to dest: the BLOB was
+		already copied to the buffer in row_sel_store_mysql_rec */
+
+		row_mysql_store_blob_ref(dest, flen, data, len);
+		break;
+
+#ifdef UNIV_DEBUG
+	case DATA_MYSQL:
+		ut_ad(flen >= len);
+		ut_ad(col->mbmaxlen >= col->mbminlen);
+		ut_ad(col->mbmaxlen > col->mbminlen || flen == len);
+		memcpy(dest, data, len);
+		break;
+
+	default:
+	case DATA_SYS_CHILD:
+	case DATA_SYS:
+		/* These column types should never be shipped to MySQL. */
+		ut_ad(0);
+
+	case DATA_CHAR:
+	case DATA_FIXBINARY:
+	case DATA_FLOAT:
+	case DATA_DOUBLE:
+	case DATA_DECIMAL:
+		/* Above are the valid column types for MySQL data. */
+		ut_ad(flen == len);
+#else /* UNIV_DEBUG */
+	default:
+#endif /* UNIV_DEBUG */
+		memcpy(dest, data, len);
+	}
+}
+
+/*****************************************************************
+Copies an InnoDB record to table->record[0]. */
+extern "C" UNIV_INTERN
+void
+innobase_rec_to_mysql(
+/*==================*/
+	TABLE*			table,		/* in/out: MySQL table */
+	const rec_t*		rec,		/* in: record */
+	const dict_index_t*	index,		/* in: index */
+	const ulint*		offsets)	/* in: rec_get_offsets(
+						rec, index, ...) */
+{
+	uint	n_fields	= table->s->fields;
+	uint	i;
+
+	ut_ad(n_fields == dict_table_get_n_user_cols(index->table));
+
+	for (i = 0; i < n_fields; i++) {
+		Field*		field	= table->field[i];
+		ulint		ipos;
+		ulint		ilen;
+		const uchar*	ifield;
+
+		field->reset();
+
+		ipos = dict_index_get_nth_col_pos(index, i);
+
+		if (UNIV_UNLIKELY(ipos == ULINT_UNDEFINED)) {
+null_field:
+			field->set_null();
+			continue;
+		}
+
+		ifield = rec_get_nth_field(rec, offsets, ipos, &ilen);
+
+		/* Assign the NULL flag */
+		if (ilen == UNIV_SQL_NULL) {
+			ut_ad(field->real_maybe_null());
+			goto null_field;
+		}
+
+		field->set_notnull();
+
+		innobase_col_to_mysql(
+			dict_field_get_col(
+				dict_index_get_nth_field(index, ipos)),
+			ifield, ilen, field);
+	}
+}
+
+/*****************************************************************
+Resets table->record[0]. */
+extern "C" UNIV_INTERN
+void
+innobase_rec_reset(
+/*===============*/
+	TABLE*			table)		/* in/out: MySQL table */
+{
+	uint	n_fields	= table->s->fields;
+	uint	i;
+
+	for (i = 0; i < n_fields; i++) {
+		table->field[i]->set_default();
+	}
+}
+
+/**********************************************************************
+Removes the filename encoding of a database and table name. */
+static
+void
+innobase_convert_tablename(
+/*=======================*/
+	char*	s)	/* in: identifier; out: decoded identifier */
+{
+	uint	errors;
+
+	char*	slash = strchr(s, '/');
+
+	if (slash) {
+		char*	t;
+		/* Temporarily replace the '/' with NUL. */
+		*slash = 0;
+		/* Convert the database name. */
+		strconvert(&my_charset_filename, s, system_charset_info,
+			   s, slash - s + 1, &errors);
+
+		t = s + strlen(s);
+		ut_ad(slash >= t);
+		/* Append a  '.' after the database name. */
+		*t++ = '.';
+		slash++;
+		/* Convert the table name. */
+		strconvert(&my_charset_filename, slash, system_charset_info,
+			   t, slash - t + strlen(slash), &errors);
+	} else {
+		strconvert(&my_charset_filename, s,
+			   system_charset_info, s, strlen(s), &errors);
+	}
+}
+
+/***********************************************************************
+This function checks that index keys are sensible. */
+static
+int
+innobase_check_index_keys(
+/*======================*/
+					/* out: 0 or error number */
+	const KEY*	key_info,	/* in: Indexes to be created */
+	ulint		num_of_keys)	/* in: Number of indexes to
+					be created */
+{
+	ulint		key_num;
+
+	ut_ad(key_info);
+	ut_ad(num_of_keys);
+
+	for (key_num = 0; key_num < num_of_keys; key_num++) {
+		const KEY&	key = key_info[key_num];
+
+		/* Check that the same index name does not appear
+		twice in indexes to be created. */
+
+		for (ulint i = 0; i < key_num; i++) {
+			const KEY&	key2 = key_info[i];
+
+			if (0 == strcmp(key.name, key2.name)) {
+				sql_print_error("InnoDB: key name `%s` appears"
+						" twice in CREATE INDEX\n",
+						key.name);
+
+				return(ER_WRONG_NAME_FOR_INDEX);
+			}
+		}
+
+		/* Check that MySQL does not try to create a column
+		prefix index field on an inappropriate data type and
+		that the same colum does not appear twice in the index. */
+
+		for (ulint i = 0; i < key.key_parts; i++) {
+			const KEY_PART_INFO&	key_part1
+				= key.key_part[i];
+			const Field*		field
+				= key_part1.field;
+			ibool			is_unsigned;
+
+			switch (get_innobase_type_from_mysql_type(
+					&is_unsigned, field)) {
+			default:
+				break;
+			case DATA_INT:
+			case DATA_FLOAT:
+			case DATA_DOUBLE:
+			case DATA_DECIMAL:
+				if (field->type() == MYSQL_TYPE_VARCHAR) {
+					if (key_part1.length
+					    >= field->pack_length()
+					    - ((Field_varstring*) field)
+					    ->length_bytes) {
+						break;
+					}
+				} else {
+					if (key_part1.length
+					    >= field->pack_length()) {
+						break;
+					}
+				}
+
+				sql_print_error("InnoDB: MySQL is trying to"
+						" create a column prefix"
+						" index field on an"
+						" inappropriate data type."
+						" column `%s`,"
+						" index `%s`.\n",
+						field->field_name,
+						key.name);
+				return(ER_WRONG_KEY_COLUMN);
+			}
+
+			for (ulint j = 0; j < i; j++) {
+				const KEY_PART_INFO&	key_part2
+					= key.key_part[j];
+
+				if (strcmp(key_part1.field->field_name,
+					   key_part2.field->field_name)) {
+					continue;
+				}
+
+				sql_print_error("InnoDB: column `%s`"
+						" is not allowed to occur"
+						" twice in index `%s`.\n",
+						key_part1.field->field_name,
+						key.name);
+				return(ER_WRONG_KEY_COLUMN);
+			}
+		}
+	}
+
+	return(0);
+}
+
+/***********************************************************************
+Create index field definition for key part */
+static
+void
+innobase_create_index_field_def(
+/*============================*/
+	KEY_PART_INFO*		key_part,	/* in: MySQL key definition */
+	mem_heap_t*		heap,		/* in: memory heap */
+	merge_index_field_t*	index_field)	/* out: index field
+						definition for key_part */
+{
+	Field*		field;
+	ibool		is_unsigned;
+	ulint		col_type;
+
+	DBUG_ENTER("innobase_create_index_field_def");
+
+	ut_ad(key_part);
+	ut_ad(index_field);
+
+	field = key_part->field;
+	ut_a(field);
+
+	col_type = get_innobase_type_from_mysql_type(&is_unsigned, field);
+
+	if (DATA_BLOB == col_type
+	    || (key_part->length < field->pack_length()
+		&& field->type() != MYSQL_TYPE_VARCHAR)
+	    || (field->type() == MYSQL_TYPE_VARCHAR
+		&& key_part->length < field->pack_length()
+			- ((Field_varstring*)field)->length_bytes)) {
+
+		index_field->prefix_len = key_part->length;
+	} else {
+		index_field->prefix_len = 0;
+	}
+
+	index_field->field_name = mem_heap_strdup(heap, field->field_name);
+
+	DBUG_VOID_RETURN;
+}
+
+/***********************************************************************
+Create index definition for key */
+static
+void
+innobase_create_index_def(
+/*======================*/
+	KEY*			key,		/* in: key definition */
+	bool			new_primary,	/* in: TRUE=generating
+						a new primary key
+						on the table */
+	bool			key_primary,	/* in: TRUE if this key
+						is a primary key */
+	merge_index_def_t*	index,		/* out: index definition */
+	mem_heap_t*		heap)		/* in: heap where memory
+						is allocated */
+{
+	ulint	i;
+	ulint	len;
+	ulint	n_fields = key->key_parts;
+	char*	index_name;
+
+	DBUG_ENTER("innobase_create_index_def");
+
+	index->fields = (merge_index_field_t*) mem_heap_alloc(
+		heap, n_fields * sizeof *index->fields);
+
+	index->ind_type = 0;
+	index->n_fields = n_fields;
+	len = strlen(key->name) + 1;
+	index->name = index_name = (char*) mem_heap_alloc(heap,
+							  len + !new_primary);
+
+	if (UNIV_LIKELY(!new_primary)) {
+		*index_name++ = TEMP_INDEX_PREFIX;
+	}
+
+	memcpy(index_name, key->name, len);
+
+	if (key->flags & HA_NOSAME) {
+		index->ind_type |= DICT_UNIQUE;
+	}
+
+	if (key_primary) {
+		index->ind_type |= DICT_CLUSTERED;
+	}
+
+	for (i = 0; i < n_fields; i++) {
+		innobase_create_index_field_def(&key->key_part[i], heap,
+						&index->fields[i]);
+	}
+
+	DBUG_VOID_RETURN;
+}
+
+/***********************************************************************
+Copy index field definition */
+static
+void
+innobase_copy_index_field_def(
+/*==========================*/
+	const dict_field_t*	field,		/* in: definition to copy */
+	merge_index_field_t*	index_field)	/* out: copied definition */
+{
+	DBUG_ENTER("innobase_copy_index_field_def");
+	DBUG_ASSERT(field != NULL);
+	DBUG_ASSERT(index_field != NULL);
+
+	index_field->field_name = field->name;
+	index_field->prefix_len = field->prefix_len;
+
+	DBUG_VOID_RETURN;
+}
+
+/***********************************************************************
+Copy index definition for the index */
+static
+void
+innobase_copy_index_def(
+/*====================*/
+	const dict_index_t*	index,	/* in: index definition to copy */
+	merge_index_def_t*	new_index,/* out: Index definition */
+	mem_heap_t*		heap)	/* in: heap where allocated */
+{
+	ulint	n_fields;
+	ulint	i;
+
+	DBUG_ENTER("innobase_copy_index_def");
+
+	/* Note that we take only those fields that user defined to be
+	in the index.  In the internal representation more colums were
+	added and those colums are not copied .*/
+
+	n_fields = index->n_user_defined_cols;
+
+	new_index->fields = (merge_index_field_t*) mem_heap_alloc(
+		heap, n_fields * sizeof *new_index->fields);
+
+	/* When adding a PRIMARY KEY, we may convert a previous
+	clustered index to a secondary index (UNIQUE NOT NULL). */
+	new_index->ind_type = index->type & ~DICT_CLUSTERED;
+	new_index->n_fields = n_fields;
+	new_index->name = index->name;
+
+	for (i = 0; i < n_fields; i++) {
+		innobase_copy_index_field_def(&index->fields[i],
+					      &new_index->fields[i]);
+	}
+
+	DBUG_VOID_RETURN;
+}
+
+/***********************************************************************
+Create an index table where indexes are ordered as follows:
+
+IF a new primary key is defined for the table THEN
+
+	1) New primary key
+	2) Original secondary indexes
+	3) New secondary indexes
+
+ELSE
+
+	1) All new indexes in the order they arrive from MySQL
+
+ENDIF
+
+*/
+static
+merge_index_def_t*
+innobase_create_key_def(
+/*====================*/
+					/* out: key definitions or NULL */
+	trx_t*		trx,		/* in: trx */
+	const dict_table_t*table,		/* in: table definition */
+	mem_heap_t*	heap,		/* in: heap where space for key
+					definitions are allocated */
+	KEY*		key_info,	/* in: Indexes to be created */
+	ulint&		n_keys)		/* in/out: Number of indexes to
+					be created */
+{
+	ulint			i = 0;
+	merge_index_def_t*	indexdef;
+	merge_index_def_t*	indexdefs;
+	bool			new_primary;
+
+	DBUG_ENTER("innobase_create_key_def");
+
+	indexdef = indexdefs = (merge_index_def_t*)
+		mem_heap_alloc(heap, sizeof *indexdef
+			       * (n_keys + UT_LIST_GET_LEN(table->indexes)));
+
+	/* If there is a primary key, it is always the first index
+	defined for the table. */
+
+	new_primary = !my_strcasecmp(system_charset_info,
+				     key_info->name, "PRIMARY");
+
+	/* If there is a UNIQUE INDEX consisting entirely of NOT NULL
+	columns, MySQL will treat it as a PRIMARY KEY unless the
+	table already has one. */
+
+	if (!new_primary && (key_info->flags & HA_NOSAME)
+	    && row_table_got_default_clust_index(table)) {
+		uint	key_part = key_info->key_parts;
+
+		new_primary = TRUE;
+
+		while (key_part--) {
+			if (key_info->key_part[key_part].key_type
+			    & FIELDFLAG_MAYBE_NULL) {
+				new_primary = FALSE;
+				break;
+			}
+		}
+	}
+
+	if (new_primary) {
+		const dict_index_t*	index;
+
+		/* Create the PRIMARY key index definition */
+		innobase_create_index_def(&key_info[i++], TRUE, TRUE,
+					  indexdef++, heap);
+
+		row_mysql_lock_data_dictionary(trx);
+
+		index = dict_table_get_first_index(table);
+
+		/* Copy the index definitions of the old table.  Skip
+		the old clustered index if it is a generated clustered
+		index or a PRIMARY KEY.  If the clustered index is a
+		UNIQUE INDEX, it must be converted to a secondary index. */
+
+		if (dict_index_get_nth_col(index, 0)->mtype == DATA_SYS
+		    || !my_strcasecmp(system_charset_info,
+				      index->name, "PRIMARY")) {
+			index = dict_table_get_next_index(index);
+		}
+
+		while (index) {
+			innobase_copy_index_def(index, indexdef++, heap);
+			index = dict_table_get_next_index(index);
+		}
+
+		row_mysql_unlock_data_dictionary(trx);
+	}
+
+	/* Create definitions for added secondary indexes. */
+
+	while (i < n_keys) {
+		innobase_create_index_def(&key_info[i++], new_primary, FALSE,
+					  indexdef++, heap);
+	}
+
+	n_keys = indexdef - indexdefs;
+
+	DBUG_RETURN(indexdefs);
+}
+
+/***********************************************************************
+Create a temporary tablename using query id, thread id, and id */
+static
+char*
+innobase_create_temporary_tablename(
+/*================================*/
+					/* out: temporary tablename */
+	mem_heap_t*	heap,		/* in: memory heap */
+	char		id,		/* in: identifier [0-9a-zA-Z] */
+	const char*     table_name)	/* in: table name */
+{
+	char*			name;
+	ulint			len;
+	static const char	suffix[] = "@0023 "; /* "# " */
+
+	len = strlen(table_name);
+
+	name = (char*) mem_heap_alloc(heap, len + sizeof suffix);
+	memcpy(name, table_name, len);
+	memcpy(name + len, suffix, sizeof suffix);
+	name[len + (sizeof suffix - 2)] = id;
+
+	return(name);
+}
+
+/***********************************************************************
+Create indexes. */
+UNIV_INTERN
+int
+ha_innobase::add_index(
+/*===================*/
+				/* out: 0 or error number */
+	TABLE*	table,		/* in: Table where indexes are created */
+	KEY*	key_info,	/* in: Indexes to be created */
+	uint	num_of_keys)	/* in: Number of indexes to be created */
+{
+	dict_index_t**	index;		/* Index to be created */
+	dict_table_t*	innodb_table;	/* InnoDB table in dictionary */
+	dict_table_t*	indexed_table;	/* Table where indexes are created */
+	merge_index_def_t* index_defs;	/* Index definitions */
+	mem_heap_t*     heap;		/* Heap for index definitions */
+	trx_t*		trx;		/* Transaction */
+	ulint		num_of_idx;
+	ulint		num_created	= 0;
+	ibool		dict_locked	= FALSE;
+	ulint		new_primary;
+	ulint		error;
+
+	DBUG_ENTER("ha_innobase::add_index");
+	ut_a(table);
+	ut_a(key_info);
+	ut_a(num_of_keys);
+
+	if (srv_created_new_raw || srv_force_recovery) {
+		DBUG_RETURN(HA_ERR_WRONG_COMMAND);
+	}
+
+	update_thd();
+
+	heap = mem_heap_create(1024);
+
+	/* In case MySQL calls this in the middle of a SELECT query, release
+	possible adaptive hash latch to avoid deadlocks of threads. */
+	trx_search_latch_release_if_reserved(prebuilt->trx);
+
+	/* Create a background transaction for the operations on
+	the data dictionary tables. */
+	trx = innobase_trx_allocate(user_thd);
+	trx_start_if_not_started(trx);
+
+	innodb_table = indexed_table
+		= dict_table_get(prebuilt->table->name, FALSE);
+
+	/* Check that index keys are sensible */
+
+	error = innobase_check_index_keys(key_info, num_of_keys);
+
+	if (UNIV_UNLIKELY(error)) {
+err_exit:
+		mem_heap_free(heap);
+		trx_general_rollback_for_mysql(trx, FALSE, NULL);
+		trx_free_for_mysql(trx);
+		trx_commit_for_mysql(prebuilt->trx);
+		DBUG_RETURN(error);
+	}
+
+	/* Create table containing all indexes to be built in this
+	alter table add index so that they are in the correct order
+	in the table. */
+
+	num_of_idx = num_of_keys;
+
+	index_defs = innobase_create_key_def(
+		trx, innodb_table, heap, key_info, num_of_idx);
+
+	new_primary = DICT_CLUSTERED & index_defs[0].ind_type;
+
+	/* Allocate memory for dictionary index definitions */
+
+	index = (dict_index_t**) mem_heap_alloc(
+		heap, num_of_idx * sizeof *index);
+
+	/* Flag this transaction as a dictionary operation, so that
+	the data dictionary will be locked in crash recovery. */
+	trx_set_dict_operation(trx, TRX_DICT_OP_INDEX);
+
+	/* Acquire a lock on the table before creating any indexes. */
+	error = row_merge_lock_table(prebuilt->trx, innodb_table,
+				     new_primary ? LOCK_X : LOCK_S);
+
+	if (UNIV_UNLIKELY(error != DB_SUCCESS)) {
+
+		goto error_handling;
+	}
+
+	/* Latch the InnoDB data dictionary exclusively so that no deadlocks
+	or lock waits can happen in it during an index create operation. */
+
+	row_mysql_lock_data_dictionary(trx);
+	dict_locked = TRUE;
+
+	/* If a new primary key is defined for the table we need
+	to drop the original table and rebuild all indexes. */
+
+	if (UNIV_UNLIKELY(new_primary)) {
+		/* This transaction should be the only one
+		operating on the table. */
+		ut_a(innodb_table->n_mysql_handles_opened == 1);
+
+		char*	new_table_name = innobase_create_temporary_tablename(
+			heap, '1', innodb_table->name);
+
+		/* Clone the table. */
+		trx_set_dict_operation(trx, TRX_DICT_OP_TABLE);
+		indexed_table = row_merge_create_temporary_table(
+			new_table_name, index_defs, innodb_table, trx);
+
+		if (!indexed_table) {
+
+			switch (trx->error_state) {
+			case DB_TABLESPACE_ALREADY_EXISTS:
+			case DB_DUPLICATE_KEY:
+				innobase_convert_tablename(new_table_name);
+				my_error(HA_ERR_TABLE_EXIST, MYF(0),
+					 new_table_name);
+				error = HA_ERR_TABLE_EXIST;
+				break;
+			default:
+				error = convert_error_code_to_mysql(
+					trx->error_state, innodb_table->flags,
+					user_thd);
+			}
+
+			row_mysql_unlock_data_dictionary(trx);
+			goto err_exit;
+		}
+
+		trx->table_id = indexed_table->id;
+	}
+
+	/* Create the indexes in SYS_INDEXES and load into dictionary. */
+
+	for (ulint i = 0; i < num_of_idx; i++) {
+
+		index[i] = row_merge_create_index(trx, indexed_table,
+						  &index_defs[i]);
+
+		if (!index[i]) {
+			error = trx->error_state;
+			goto error_handling;
+		}
+
+		num_created++;
+	}
+
+	ut_ad(error == DB_SUCCESS);
+
+	/* Commit the data dictionary transaction in order to release
+	the table locks on the system tables.  Unfortunately, this
+	means that if MySQL crashes while creating a new primary key
+	inside row_merge_build_indexes(), indexed_table will not be
+	dropped on crash recovery.  Thus, it will become orphaned. */
+	trx_commit_for_mysql(trx);
+
+	row_mysql_unlock_data_dictionary(trx);
+	dict_locked = FALSE;
+
+	ut_a(trx->n_active_thrs == 0);
+	ut_a(UT_LIST_GET_LEN(trx->signals) == 0);
+
+	if (UNIV_UNLIKELY(new_primary)) {
+		/* A primary key is to be built.  Acquire an exclusive
+		table lock also on the table that is being created. */
+		ut_ad(indexed_table != innodb_table);
+
+		error = row_merge_lock_table(prebuilt->trx, indexed_table,
+					     LOCK_X);
+
+		if (UNIV_UNLIKELY(error != DB_SUCCESS)) {
+
+			goto error_handling;
+		}
+	}
+
+	/* Read the clustered index of the table and build indexes
+	based on this information using temporary files and merge sort. */
+	error = row_merge_build_indexes(prebuilt->trx,
+					innodb_table, indexed_table,
+					index, num_of_idx, table);
+
+error_handling:
+#ifdef UNIV_DEBUG
+	/* TODO: At the moment we can't handle the following statement
+	in our debugging code below:
+
+	alter table t drop index b, add index (b);
+
+	The fix will have to parse the SQL and note that the index
+	being added has the same name as the the one being dropped and
+	ignore that in the dup index check.*/
+	//dict_table_check_for_dup_indexes(prebuilt->table);
+#endif
+
+	/* After an error, remove all those index definitions from the
+	dictionary which were defined. */
+
+	switch (error) {
+		const char*	old_name;
+		char*		tmp_name;
+	case DB_SUCCESS:
+		ut_a(!dict_locked);
+		row_mysql_lock_data_dictionary(trx);
+		dict_locked = TRUE;
+
+		if (!new_primary) {
+			error = row_merge_rename_indexes(trx, indexed_table);
+
+			if (error != DB_SUCCESS) {
+				row_merge_drop_indexes(trx, indexed_table,
+						       index, num_created);
+			}
+
+			goto convert_error;
+		}
+
+		/* If a new primary key was defined for the table and
+		there was no error at this point, we can now rename
+		the old table as a temporary table, rename the new
+		temporary table as the old table and drop the old table. */
+		old_name = innodb_table->name;
+		tmp_name = innobase_create_temporary_tablename(heap, '2',
+							       old_name);
+
+		error = row_merge_rename_tables(innodb_table, indexed_table,
+						tmp_name, trx);
+
+		if (error != DB_SUCCESS) {
+
+			row_merge_drop_table(trx, indexed_table);
+
+			switch (error) {
+			case DB_TABLESPACE_ALREADY_EXISTS:
+			case DB_DUPLICATE_KEY:
+				innobase_convert_tablename(tmp_name);
+				my_error(HA_ERR_TABLE_EXIST, MYF(0), tmp_name);
+				error = HA_ERR_TABLE_EXIST;
+				break;
+			default:
+				goto convert_error;
+			}
+			break;
+		}
+
+		trx_commit_for_mysql(prebuilt->trx);
+		row_prebuilt_free(prebuilt, TRUE);
+		prebuilt = row_create_prebuilt(indexed_table);
+
+		indexed_table->n_mysql_handles_opened++;
+
+		error = row_merge_drop_table(trx, innodb_table);
+		goto convert_error;
+
+	case DB_TOO_BIG_RECORD:
+		my_error(HA_ERR_TO_BIG_ROW, MYF(0));
+		goto error;
+	case DB_PRIMARY_KEY_IS_NULL:
+		my_error(ER_PRIMARY_CANT_HAVE_NULL, MYF(0));
+		/* fall through */
+	case DB_DUPLICATE_KEY:
+error:
+		prebuilt->trx->error_info = NULL;
+		/* fall through */
+	default:
+		if (new_primary) {
+			row_merge_drop_table(trx, indexed_table);
+		} else {
+			if (!dict_locked) {
+				row_mysql_lock_data_dictionary(trx);
+				dict_locked = TRUE;
+			}
+
+			row_merge_drop_indexes(trx, indexed_table,
+					       index, num_created);
+		}
+
+convert_error:
+		error = convert_error_code_to_mysql(error,
+						    innodb_table->flags,
+						    user_thd);
+	}
+
+	mem_heap_free(heap);
+	trx_commit_for_mysql(trx);
+	if (prebuilt->trx) {
+		trx_commit_for_mysql(prebuilt->trx);
+	}
+
+	if (dict_locked) {
+		row_mysql_unlock_data_dictionary(trx);
+	}
+
+	trx_free_for_mysql(trx);
+
+	/* There might be work for utility threads.*/
+	srv_active_wake_master_thread();
+
+	DBUG_RETURN(error);
+}
+
+/***********************************************************************
+Prepare to drop some indexes of a table. */
+UNIV_INTERN
+int
+ha_innobase::prepare_drop_index(
+/*============================*/
+				/* out: 0 or error number */
+	TABLE*	table,		/* in: Table where indexes are dropped */
+	uint*	key_num,	/* in: Key nums to be dropped */
+	uint	num_of_keys)	/* in: Number of keys to be dropped */
+{
+	trx_t*		trx;
+	int		err = 0;
+	uint 		n_key;
+
+	DBUG_ENTER("ha_innobase::prepare_drop_index");
+	ut_ad(table);
+	ut_ad(key_num);
+	ut_ad(num_of_keys);
+	if (srv_created_new_raw || srv_force_recovery) {
+		DBUG_RETURN(HA_ERR_WRONG_COMMAND);
+	}
+
+	update_thd();
+
+	trx_search_latch_release_if_reserved(prebuilt->trx);
+	trx = prebuilt->trx;
+
+	/* Test and mark all the indexes to be dropped */
+
+	row_mysql_lock_data_dictionary(trx);
+
+	/* Check that none of the indexes have previously been flagged
+	for deletion. */
+	{
+		const dict_index_t*	index
+			= dict_table_get_first_index(prebuilt->table);
+		do {
+			ut_a(!index->to_be_dropped);
+			index = dict_table_get_next_index(index);
+		} while (index);
+	}
+
+	for (n_key = 0; n_key < num_of_keys; n_key++) {
+		const KEY*	key;
+		dict_index_t*	index;
+
+		key = table->key_info + key_num[n_key];
+		index = dict_table_get_index_on_name_and_min_id(
+			prebuilt->table, key->name);
+
+		if (!index) {
+			sql_print_error("InnoDB could not find key n:o %u "
+					"with name %s for table %s",
+					key_num[n_key],
+					key ? key->name : "NULL",
+					prebuilt->table->name);
+
+			err = HA_ERR_KEY_NOT_FOUND;
+			goto func_exit;
+		}
+
+		/* Refuse to drop the clustered index.  It would be
+		better to automatically generate a clustered index,
+		but mysql_alter_table() will call this method only
+		after ha_innobase::add_index(). */
+
+		if (dict_index_is_clust(index)) {
+			my_error(ER_REQUIRES_PRIMARY_KEY, MYF(0));
+			err = -1;
+			goto func_exit;
+		}
+
+		index->to_be_dropped = TRUE;
+	}
+
+	/* If FOREIGN_KEY_CHECK = 1 you may not drop an index defined
+	for a foreign key constraint because InnoDB requires that both
+	tables contain indexes for the constraint.  Note that CREATE
+	INDEX id ON table does a CREATE INDEX and DROP INDEX, and we
+	can ignore here foreign keys because a new index for the
+	foreign key has already been created.
+
+	We check for the foreign key constraints after marking the
+	candidate indexes for deletion, because when we check for an
+	equivalent foreign index we don't want to select an index that
+	is later deleted. */
+
+	if (trx->check_foreigns
+	    && thd_sql_command(user_thd) != SQLCOM_CREATE_INDEX) {
+		dict_index_t*	index;
+
+		for (index = dict_table_get_first_index(prebuilt->table);
+		     index;
+		     index = dict_table_get_next_index(index)) {
+			dict_foreign_t*	foreign;
+
+			if (!index->to_be_dropped) {
+
+				continue;
+			}
+
+			/* Check if the index is referenced. */
+			foreign = dict_table_get_referenced_constraint(
+				prebuilt->table, index);
+
+			if (foreign) {
+index_needed:
+				trx_set_detailed_error(
+					trx,
+					"Index needed in foreign key "
+					"constraint");
+
+				trx->error_info = index;
+
+				err = HA_ERR_DROP_INDEX_FK;
+				break;
+			} else {
+				/* Check if this index references some
+				other table */
+				foreign = dict_table_get_foreign_constraint(
+					prebuilt->table, index);
+
+				if (foreign) {
+					ut_a(foreign->foreign_index == index);
+
+					/* Search for an equivalent index that
+					the foreign key constraint could use
+					if this index were to be deleted. */
+					if (!dict_foreign_find_equiv_index(
+						foreign)) {
+
+						goto index_needed;
+					}
+				}
+			}
+		}
+	} else if (thd_sql_command(user_thd) == SQLCOM_CREATE_INDEX) {
+		/* This is a drop of a foreign key constraint index that
+		was created by MySQL when the constraint was added.  MySQL
+		does this when the user creates an index explicitly which
+		can be used in place of the automatically generated index. */
+
+		dict_index_t*	index;
+
+		for (index = dict_table_get_first_index(prebuilt->table);
+		     index;
+		     index = dict_table_get_next_index(index)) {
+			dict_foreign_t*	foreign;
+
+			if (!index->to_be_dropped) {
+
+				continue;
+			}
+
+			/* Check if this index references some other table */
+			foreign = dict_table_get_foreign_constraint(
+				prebuilt->table, index);
+
+			if (foreign == NULL) {
+
+				continue;
+			}
+
+			ut_a(foreign->foreign_index == index);
+
+			/* Search for an equivalent index that the
+			foreign key constraint could use if this index
+			were to be deleted. */
+
+			if (!dict_foreign_find_equiv_index(foreign)) {
+				trx_set_detailed_error(
+					trx,
+					"Index needed in foreign key "
+					"constraint");
+
+				trx->error_info = foreign->foreign_index;
+
+				err = HA_ERR_DROP_INDEX_FK;
+				break;
+			}
+		}
+	}
+
+func_exit:
+	if (err) {
+		/* Undo our changes since there was some sort of error. */
+		dict_index_t*	index
+			= dict_table_get_first_index(prebuilt->table);
+
+		do {
+			index->to_be_dropped = FALSE;
+			index = dict_table_get_next_index(index);
+		} while (index);
+	}
+
+	row_mysql_unlock_data_dictionary(trx);
+
+	DBUG_RETURN(err);
+}
+
+/***********************************************************************
+Drop the indexes that were passed to a successful prepare_drop_index(). */
+UNIV_INTERN
+int
+ha_innobase::final_drop_index(
+/*==========================*/
+				/* out: 0 or error number */
+	TABLE*	table)		/* in: Table where indexes are dropped */
+{
+	dict_index_t*	index;		/* Index to be dropped */
+	trx_t*		trx;		/* Transaction */
+	int		err;
+
+	DBUG_ENTER("ha_innobase::final_drop_index");
+	ut_ad(table);
+
+	if (srv_created_new_raw || srv_force_recovery) {
+		DBUG_RETURN(HA_ERR_WRONG_COMMAND);
+	}
+
+	update_thd();
+
+	trx_search_latch_release_if_reserved(prebuilt->trx);
+
+	/* Create a background transaction for the operations on
+	the data dictionary tables. */
+	trx = innobase_trx_allocate(user_thd);
+	trx_start_if_not_started(trx);
+
+	/* Flag this transaction as a dictionary operation, so that
+	the data dictionary will be locked in crash recovery. */
+	trx_set_dict_operation(trx, TRX_DICT_OP_INDEX);
+
+	/* Lock the table exclusively, to ensure that no active
+	transaction depends on an index that is being dropped. */
+	err = convert_error_code_to_mysql(
+		row_merge_lock_table(prebuilt->trx, prebuilt->table, LOCK_X),
+		prebuilt->table->flags, user_thd);
+
+	row_mysql_lock_data_dictionary(trx);
+
+	if (UNIV_UNLIKELY(err)) {
+
+		/* Unmark the indexes to be dropped. */
+		for (index = dict_table_get_first_index(prebuilt->table);
+		     index; index = dict_table_get_next_index(index)) {
+
+			index->to_be_dropped = FALSE;
+		}
+
+		goto func_exit;
+	}
+
+	/* Drop indexes marked to be dropped */
+
+	index = dict_table_get_first_index(prebuilt->table);
+
+	while (index) {
+		dict_index_t*	next_index;
+
+		next_index = dict_table_get_next_index(index);
+
+		if (index->to_be_dropped) {
+
+			row_merge_drop_index(index, prebuilt->table, trx);
+		}
+
+		index = next_index;
+	}
+
+	/* Check that all flagged indexes were dropped. */
+	for (index = dict_table_get_first_index(prebuilt->table);
+	     index; index = dict_table_get_next_index(index)) {
+		ut_a(!index->to_be_dropped);
+	}
+
+#ifdef UNIV_DEBUG
+	dict_table_check_for_dup_indexes(prebuilt->table);
+#endif
+
+func_exit:
+	trx_commit_for_mysql(trx);
+	trx_commit_for_mysql(prebuilt->trx);
+	row_mysql_unlock_data_dictionary(trx);
+
+	/* Flush the log to reduce probability that the .frm files and
+	the InnoDB data dictionary get out-of-sync if the user runs
+	with innodb_flush_log_at_trx_commit = 0 */
+
+	log_buffer_flush_to_disk();
+
+	trx_free_for_mysql(trx);
+
+	/* Tell the InnoDB server that there might be work for
+	utility threads: */
+
+	srv_active_wake_master_thread();
+
+	DBUG_RETURN(err);
+}
diff --git a/storage/xtradb/handler/handler0vars.h b/storage/xtradb/handler/handler0vars.h
new file mode 100644
index 00000000000..ea9f305ce66
--- /dev/null
+++ b/storage/xtradb/handler/handler0vars.h
@@ -0,0 +1,68 @@
+/*****************************************************************************
+
+Copyright (c) 2008, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/***********************************************************************
+This file contains accessor functions for dynamic plugin on Windows.
+***********************************************************************/
+
+#if defined __WIN__ && defined MYSQL_DYNAMIC_PLUGIN
+/***********************************************************************
+This is a list of externals that can not be resolved by delay loading.
+They have to be resolved indirectly via their addresses in the .map file.
+All of them are external variables. */
+extern	CHARSET_INFO*		wdl_my_charset_bin;
+extern	CHARSET_INFO*		wdl_my_charset_latin1;
+extern	CHARSET_INFO*		wdl_my_charset_filename;
+extern	CHARSET_INFO**		wdl_system_charset_info;
+extern	CHARSET_INFO**		wdl_default_charset_info;
+extern	CHARSET_INFO**		wdl_all_charsets;
+extern	system_variables*	wdl_global_system_variables;
+extern	char*			wdl_mysql_real_data_home;
+extern	char**			wdl_mysql_data_home;
+extern	char**			wdl_tx_isolation_names;
+extern	char**			wdl_binlog_format_names;
+extern	char*			wdl_reg_ext;
+extern	pthread_mutex_t*	wdl_LOCK_thread_count;
+extern	key_map*		wdl_key_map_full;
+extern	MY_TMPDIR*		wdl_mysql_tmpdir_list;
+extern	bool*			wdl_mysqld_embedded;
+extern	uint*			wdl_lower_case_table_names;
+extern	ulong*			wdl_specialflag;
+extern	int*			wdl_my_umask;
+
+#define my_charset_bin		(*wdl_my_charset_bin)
+#define my_charset_latin1	(*wdl_my_charset_latin1)
+#define my_charset_filename	(*wdl_my_charset_filename)
+#define system_charset_info	(*wdl_system_charset_info)
+#define default_charset_info	(*wdl_default_charset_info)
+#define all_charsets		(wdl_all_charsets)
+#define global_system_variables	(*wdl_global_system_variables)
+#define mysql_real_data_home	(wdl_mysql_real_data_home)
+#define mysql_data_home		(*wdl_mysql_data_home)
+#define tx_isolation_names	(wdl_tx_isolation_names)
+#define binlog_format_names	(wdl_binlog_format_names)
+#define reg_ext			(wdl_reg_ext)
+#define LOCK_thread_count	(*wdl_LOCK_thread_count)
+#define key_map_full		(*wdl_key_map_full)
+#define mysql_tmpdir_list	(*wdl_mysql_tmpdir_list)
+#define mysqld_embedded		(*wdl_mysqld_embedded)
+#define lower_case_table_names	(*wdl_lower_case_table_names)
+#define specialflag		(*wdl_specialflag)
+#define my_umask		(*wdl_my_umask)
+
+#endif
diff --git a/storage/xtradb/handler/i_s.cc b/storage/xtradb/handler/i_s.cc
new file mode 100644
index 00000000000..a5404d067ca
--- /dev/null
+++ b/storage/xtradb/handler/i_s.cc
@@ -0,0 +1,2655 @@
+/*****************************************************************************
+
+Copyright (c) 2007, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+InnoDB INFORMATION SCHEMA tables interface to MySQL.
+
+Created July 18, 2007 Vasil Dimov
+*******************************************************/
+
+#include <mysql_priv.h>
+#include <mysqld_error.h>
+
+#include <m_ctype.h>
+#include <hash.h>
+#include <myisampack.h>
+#include <mysys_err.h>
+#include <my_sys.h>
+#include "i_s.h"
+#include "innodb_patch_info.h"
+#include <mysql/plugin.h>
+
+extern "C" {
+#include "trx0i_s.h"
+#include "trx0trx.h" /* for TRX_QUE_STATE_STR_MAX_LEN */
+#include "buf0buddy.h" /* for i_s_cmpmem */
+#include "buf0buf.h" /* for buf_pool and PAGE_ZIP_MIN_SIZE */
+#include "ha_prototypes.h" /* for innobase_convert_name() */
+#include "srv0start.h" /* for srv_was_started */
+#include "btr0btr.h" /* for btr_page_get_index_id */
+#include "dict0dict.h" /* for dict_index_get_if_in_cache */
+#include "trx0rseg.h" /* for trx_rseg_struct */
+#include "trx0sys.h" /* for trx_sys */
+/* from buf0buf.c */
+struct buf_chunk_struct{
+	ulint		mem_size;	/* allocated size of the chunk */
+	ulint		size;		/* size of frames[] and blocks[] */
+	void*		mem;		/* pointer to the memory area which
+					was allocated for the frames */
+	buf_block_t*	blocks;		/* array of buffer control blocks */
+};
+}
+#include "handler0vars.h"
+
+static const char plugin_author[] = "Innobase Oy";
+
+#define OK(expr)		\
+	if ((expr) != 0) {	\
+		DBUG_RETURN(1);	\
+	}
+
+#define RETURN_IF_INNODB_NOT_STARTED(plugin_name)			\
+do {									\
+	if (!srv_was_started) {						\
+		push_warning_printf(thd, MYSQL_ERROR::WARN_LEVEL_WARN,	\
+				    ER_CANT_FIND_SYSTEM_REC,		\
+				    "InnoDB: SELECTing from "		\
+				    "INFORMATION_SCHEMA.%s but "	\
+				    "the InnoDB storage engine "	\
+				    "is not installed", plugin_name);	\
+		DBUG_RETURN(0);						\
+	}								\
+} while (0)
+
+#if !defined __STRICT_ANSI__ && defined __GNUC__ && (__GNUC__) > 2 && !defined __INTEL_COMPILER
+#define STRUCT_FLD(name, value)	name: value
+#else
+#define STRUCT_FLD(name, value)	value
+#endif
+
+static const ST_FIELD_INFO END_OF_ST_FIELD_INFO =
+	{STRUCT_FLD(field_name,		NULL),
+	 STRUCT_FLD(field_length,	0),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_NULL),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	0),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)};
+
+/*
+Use the following types mapping:
+
+C type	ST_FIELD_INFO::field_type
+---------------------------------
+long			MYSQL_TYPE_LONGLONG
+(field_length=MY_INT64_NUM_DECIMAL_DIGITS)
+
+long unsigned		MYSQL_TYPE_LONGLONG
+(field_length=MY_INT64_NUM_DECIMAL_DIGITS, field_flags=MY_I_S_UNSIGNED)
+
+char*			MYSQL_TYPE_STRING
+(field_length=n)
+
+float			MYSQL_TYPE_FLOAT
+(field_length=0 is ignored)
+
+void*			MYSQL_TYPE_LONGLONG
+(field_length=MY_INT64_NUM_DECIMAL_DIGITS, field_flags=MY_I_S_UNSIGNED)
+
+boolean (if else)	MYSQL_TYPE_LONG
+(field_length=1)
+
+time_t			MYSQL_TYPE_DATETIME
+(field_length=0 ignored)
+---------------------------------
+*/
+
+/* XXX these are defined in mysql_priv.h inside #ifdef MYSQL_SERVER */
+bool schema_table_store_record(THD *thd, TABLE *table);
+void localtime_to_TIME(MYSQL_TIME *to, struct tm *from);
+bool check_global_access(THD *thd, ulong want_access);
+
+/***********************************************************************
+Common function to fill any of the dynamic tables:
+INFORMATION_SCHEMA.innodb_trx
+INFORMATION_SCHEMA.innodb_locks
+INFORMATION_SCHEMA.innodb_lock_waits */
+static
+int
+trx_i_s_common_fill_table(
+/*======================*/
+				/* out: 0 on success */
+	THD*		thd,	/* in: thread */
+	TABLE_LIST*	tables,	/* in/out: tables to fill */
+	COND*		cond);	/* in: condition (not used) */
+
+/***********************************************************************
+Unbind a dynamic INFORMATION_SCHEMA table. */
+static
+int
+i_s_common_deinit(
+/*==============*/
+			/* out: 0 on success */
+	void*	p);	/* in/out: table schema object */
+
+/***********************************************************************
+Auxiliary function to store time_t value in MYSQL_TYPE_DATETIME
+field. */
+static
+int
+field_store_time_t(
+/*===============*/
+			/* out: 0 on success */
+	Field*	field,	/* in/out: target field for storage */
+	time_t	time)	/* in: value to store */
+{
+	MYSQL_TIME	my_time;
+	struct tm	tm_time;
+
+#if 0
+	/* use this if you are sure that `variables' and `time_zone'
+	are always initialized */
+	thd->variables.time_zone->gmt_sec_to_TIME(
+		&my_time, (my_time_t) time);
+#else
+	localtime_r(&time, &tm_time);
+	localtime_to_TIME(&my_time, &tm_time);
+	my_time.time_type = MYSQL_TIMESTAMP_DATETIME;
+#endif
+
+	return(field->store_time(&my_time, MYSQL_TIMESTAMP_DATETIME));
+}
+
+/***********************************************************************
+Auxiliary function to store char* value in MYSQL_TYPE_STRING field. */
+static
+int
+field_store_string(
+/*===============*/
+				/* out: 0 on success */
+	Field*		field,	/* in/out: target field for storage */
+	const char*	str)	/* in: NUL-terminated utf-8 string,
+				or NULL */
+{
+	int	ret;
+
+	if (str != NULL) {
+
+		ret = field->store(str, strlen(str),
+				   system_charset_info);
+		field->set_notnull();
+	} else {
+
+		ret = 0; /* success */
+		field->set_null();
+	}
+
+	return(ret);
+}
+
+/***********************************************************************
+Auxiliary function to store ulint value in MYSQL_TYPE_LONGLONG field.
+If the value is ULINT_UNDEFINED then the field it set to NULL. */
+static
+int
+field_store_ulint(
+/*==============*/
+			/* out: 0 on success */
+	Field*	field,	/* in/out: target field for storage */
+	ulint	n)	/* in: value to store */
+{
+	int	ret;
+
+	if (n != ULINT_UNDEFINED) {
+
+		ret = field->store(n);
+		field->set_notnull();
+	} else {
+
+		ret = 0; /* success */
+		field->set_null();
+	}
+
+	return(ret);
+}
+
+/* Fields of the dynamic table INFORMATION_SCHEMA.innodb_patches */
+static ST_FIELD_INFO	innodb_patches_fields_info[] =
+{
+#define IDX_PATCH_NAME		0
+	{STRUCT_FLD(field_name,		"name"),
+	 STRUCT_FLD(field_length,	255),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_STRING),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	0),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+#define IDX_PATCH_DESCR		1
+	{STRUCT_FLD(field_name,		"description"),
+	 STRUCT_FLD(field_length,	255),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_STRING),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	0),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+#define IDX_PATCH_COMMENT		2
+	{STRUCT_FLD(field_name,		"comment"),
+	 STRUCT_FLD(field_length,	100),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_STRING),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	0),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+#define IDX_PATCH_LINK			3
+	{STRUCT_FLD(field_name,		"link"),
+	 STRUCT_FLD(field_length,	255),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_STRING),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	0),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	END_OF_ST_FIELD_INFO
+};
+
+static struct st_mysql_information_schema	i_s_info =
+{
+	MYSQL_INFORMATION_SCHEMA_INTERFACE_VERSION
+};
+
+/***********************************************************************
+Fill the dynamic table information_schema.innodb_patches */
+static
+int
+innodb_patches_fill(
+/*=============*/
+				/* out: 0 on success, 1 on failure */
+	THD*		thd,	/* in: thread */
+	TABLE_LIST*	tables,	/* in/out: tables to fill */
+	COND*		cond)	/* in: condition (ignored) */
+{
+	TABLE*	table	= (TABLE *) tables->table;
+	int	status	= 0;
+	int	i;
+	Field**	fields;
+
+
+	DBUG_ENTER("innodb_patches_fill");
+	fields = table->field;
+
+	/* deny access to non-superusers */
+	if (check_global_access(thd, PROCESS_ACL)) {
+
+		DBUG_RETURN(0);
+	}
+
+	RETURN_IF_INNODB_NOT_STARTED(tables->schema_table_name);
+	
+	for (i = 0; innodb_enhancements[i].file; i++) {
+
+   	field_store_string(fields[0],innodb_enhancements[i].file);
+   	field_store_string(fields[1],innodb_enhancements[i].name);
+   	field_store_string(fields[2],innodb_enhancements[i].comment);
+   	field_store_string(fields[3],innodb_enhancements[i].link);
+
+	if (schema_table_store_record(thd, table)) {
+		status = 1;
+		break;
+	}
+
+	}
+
+
+	DBUG_RETURN(status);
+}
+
+/***********************************************************************
+Bind the dynamic table information_schema.innodb_patches. */
+static
+int
+innodb_patches_init(
+/*=========*/
+			/* out: 0 on success */
+	void*	p)	/* in/out: table schema object */
+{
+	DBUG_ENTER("innodb_patches_init");
+	ST_SCHEMA_TABLE* schema = (ST_SCHEMA_TABLE*) p;
+
+	schema->fields_info = innodb_patches_fields_info;
+	schema->fill_table = innodb_patches_fill;
+
+	DBUG_RETURN(0);
+}
+
+
+UNIV_INTERN struct st_mysql_plugin      i_s_innodb_patches =
+{
+        /* the plugin type (a MYSQL_XXX_PLUGIN value) */
+        /* int */
+        STRUCT_FLD(type, MYSQL_INFORMATION_SCHEMA_PLUGIN),
+
+        /* pointer to type-specific plugin descriptor */
+        /* void* */
+        STRUCT_FLD(info, &i_s_info),
+
+        /* plugin name */
+        /* const char* */
+        STRUCT_FLD(name, "XTRADB_ENHANCEMENTS"),
+
+        /* plugin author (for SHOW PLUGINS) */
+        /* const char* */
+        STRUCT_FLD(author, "Percona"),
+
+        /* general descriptive text (for SHOW PLUGINS) */
+        /* const char* */
+        STRUCT_FLD(descr, "Enhancements applied to InnoDB plugin"),
+
+        /* the plugin license (PLUGIN_LICENSE_XXX) */
+        /* int */
+        STRUCT_FLD(license, PLUGIN_LICENSE_GPL),
+
+        /* the function to invoke when plugin is loaded */
+        /* int (*)(void*); */
+        STRUCT_FLD(init, innodb_patches_init),
+
+        /* the function to invoke when plugin is unloaded */
+        /* int (*)(void*); */
+        STRUCT_FLD(deinit, i_s_common_deinit),
+
+        /* plugin version (for SHOW PLUGINS) */
+        /* unsigned int */
+        STRUCT_FLD(version, INNODB_VERSION_SHORT),
+
+        /* struct st_mysql_show_var* */
+        STRUCT_FLD(status_vars, NULL),
+
+        /* struct st_mysql_sys_var** */
+        STRUCT_FLD(system_vars, NULL),
+
+        /* reserved for dependency checking */
+        /* void* */
+        STRUCT_FLD(__reserved1, NULL)
+};
+
+
+static ST_FIELD_INFO	i_s_innodb_buffer_pool_pages_fields_info[] =
+{
+	{STRUCT_FLD(field_name,		"page_type"),
+	 STRUCT_FLD(field_length,	64),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_STRING),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_MAYBE_NULL),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"space_id"),
+	 STRUCT_FLD(field_length,	MY_INT64_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONGLONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_UNSIGNED),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"page_no"),
+	 STRUCT_FLD(field_length,	MY_INT64_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONGLONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_UNSIGNED),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"lru_position"),
+	 STRUCT_FLD(field_length,	MY_INT64_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONGLONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_UNSIGNED),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"fix_count"),
+	 STRUCT_FLD(field_length,	MY_INT64_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONGLONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_UNSIGNED),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"flush_type"),
+	 STRUCT_FLD(field_length,	MY_INT64_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONGLONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_UNSIGNED),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	END_OF_ST_FIELD_INFO
+};
+
+static ST_FIELD_INFO	i_s_innodb_buffer_pool_pages_index_fields_info[] =
+{
+	{STRUCT_FLD(field_name,		"schema_name"),
+	 STRUCT_FLD(field_length,	64),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_STRING),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_MAYBE_NULL),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"table_name"),
+	 STRUCT_FLD(field_length,	64),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_STRING),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	0),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"index_name"),
+	 STRUCT_FLD(field_length,	64),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_STRING),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	0),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"space_id"),
+	 STRUCT_FLD(field_length,	MY_INT64_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONGLONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_UNSIGNED),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"page_no"),
+	 STRUCT_FLD(field_length,	MY_INT64_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONGLONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_UNSIGNED),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"n_recs"),
+	 STRUCT_FLD(field_length,	MY_INT64_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONGLONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_UNSIGNED),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"data_size"),
+	 STRUCT_FLD(field_length,	MY_INT64_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONGLONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_UNSIGNED),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"hashed"),
+	 STRUCT_FLD(field_length,	MY_INT64_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONGLONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_UNSIGNED),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"accessed"),
+	 STRUCT_FLD(field_length,	MY_INT64_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONGLONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_UNSIGNED),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"modified"),
+	 STRUCT_FLD(field_length,	MY_INT64_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONGLONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_UNSIGNED),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"dirty"),
+	 STRUCT_FLD(field_length,	MY_INT64_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONGLONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_UNSIGNED),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"old"),
+	 STRUCT_FLD(field_length,	MY_INT64_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONGLONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_UNSIGNED),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"lru_position"),
+	 STRUCT_FLD(field_length,	MY_INT64_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONGLONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_UNSIGNED),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"fix_count"),
+	 STRUCT_FLD(field_length,	MY_INT64_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONGLONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_UNSIGNED),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"flush_type"),
+	 STRUCT_FLD(field_length,	MY_INT64_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONGLONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_UNSIGNED),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	END_OF_ST_FIELD_INFO
+};
+
+static ST_FIELD_INFO	i_s_innodb_buffer_pool_pages_blob_fields_info[] =
+{
+	{STRUCT_FLD(field_name,		"space_id"),
+	 STRUCT_FLD(field_length,	MY_INT64_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONGLONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_UNSIGNED),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"page_no"),
+	 STRUCT_FLD(field_length,	MY_INT64_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONGLONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_UNSIGNED),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"compressed"),
+	 STRUCT_FLD(field_length,	MY_INT64_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONGLONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_UNSIGNED),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"part_len"),
+	 STRUCT_FLD(field_length,	MY_INT64_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONGLONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_UNSIGNED),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"next_page_no"),
+	 STRUCT_FLD(field_length,	MY_INT64_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONGLONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_UNSIGNED),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"lru_position"),
+	 STRUCT_FLD(field_length,	MY_INT64_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONGLONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_UNSIGNED),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"fix_count"),
+	 STRUCT_FLD(field_length,	MY_INT64_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONGLONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_UNSIGNED),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"flush_type"),
+	 STRUCT_FLD(field_length,	MY_INT64_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONGLONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_UNSIGNED),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	END_OF_ST_FIELD_INFO
+};
+
+/***********************************************************************
+Fill the dynamic table information_schema.innodb_buffer_pool_pages. */
+static
+int
+i_s_innodb_buffer_pool_pages_fill(
+/*================*/
+				/* out: 0 on success, 1 on failure */
+	THD*		thd,	/* in: thread */
+	TABLE_LIST*	tables,	/* in/out: tables to fill */
+	COND*		cond)	/* in: condition (ignored) */
+{
+	TABLE*	table	= (TABLE *) tables->table;
+	int	status	= 0;
+
+  ulint		n_chunks, n_blocks;
+
+	buf_chunk_t*	chunk;
+
+	DBUG_ENTER("i_s_innodb_buffer_pool_pages_fill");
+
+	/* deny access to non-superusers */
+	if (check_global_access(thd, PROCESS_ACL)) {
+
+		DBUG_RETURN(0);
+	}
+
+	RETURN_IF_INNODB_NOT_STARTED(tables->schema_table_name);
+
+	buf_pool_mutex_enter();
+  mutex_enter(&(dict_sys->mutex));
+	
+	chunk = buf_pool->chunks;
+  
+	for (n_chunks = buf_pool->n_chunks; n_chunks--; chunk++) {
+		buf_block_t*	block		= chunk->blocks;
+
+    for (n_blocks	= chunk->size; n_blocks--; block++) {
+      const buf_frame_t* frame = block->frame;
+  
+      char page_type[64];
+
+      switch(fil_page_get_type(frame))
+      {
+      case FIL_PAGE_INDEX:
+        strcpy(page_type, "index");
+        break;
+      case FIL_PAGE_UNDO_LOG:
+        strcpy(page_type, "undo_log");
+        break;
+      case FIL_PAGE_INODE:
+        strcpy(page_type, "inode");
+        break;
+      case FIL_PAGE_IBUF_FREE_LIST:
+        strcpy(page_type, "ibuf_free_list");
+        break;
+      case FIL_PAGE_TYPE_ALLOCATED:
+        strcpy(page_type, "allocated");
+        break;
+      case FIL_PAGE_IBUF_BITMAP:
+        strcpy(page_type, "bitmap");
+        break;
+      case FIL_PAGE_TYPE_SYS:
+        strcpy(page_type, "sys");
+        break;
+      case FIL_PAGE_TYPE_TRX_SYS:
+        strcpy(page_type, "trx_sys");
+        break;
+      case FIL_PAGE_TYPE_FSP_HDR:
+        strcpy(page_type, "fsp_hdr");
+        break;
+      case FIL_PAGE_TYPE_XDES:
+        strcpy(page_type, "xdes");
+        break;
+      case FIL_PAGE_TYPE_BLOB:
+        strcpy(page_type, "blob");
+        break;
+      case FIL_PAGE_TYPE_ZBLOB:
+        strcpy(page_type, "zblob");
+        break;
+      case FIL_PAGE_TYPE_ZBLOB2:
+        strcpy(page_type, "zblob2");
+        break;
+      default:
+        sprintf(page_type, "unknown (type=%li)", fil_page_get_type(frame));
+      }
+      
+      field_store_string(table->field[0], page_type);
+      table->field[1]->store(block->page.space);
+      table->field[2]->store(block->page.offset);
+      table->field[3]->store(block->page.LRU_position);
+      table->field[4]->store(block->page.buf_fix_count);
+      table->field[5]->store(block->page.flush_type);
+
+      if (schema_table_store_record(thd, table)) {
+        status = 1;
+        break;
+      }
+      
+    }      
+	}
+
+  mutex_exit(&(dict_sys->mutex));
+	buf_pool_mutex_exit();
+
+	DBUG_RETURN(status);
+}
+
+/***********************************************************************
+Fill the dynamic table information_schema.innodb_buffer_pool_pages_index. */
+static
+int
+i_s_innodb_buffer_pool_pages_index_fill(
+/*================*/
+				/* out: 0 on success, 1 on failure */
+	THD*		thd,	/* in: thread */
+	TABLE_LIST*	tables,	/* in/out: tables to fill */
+	COND*		cond)	/* in: condition (ignored) */
+{
+	TABLE*	table	= (TABLE *) tables->table;
+	int	status	= 0;
+
+  ulint		n_chunks, n_blocks;
+  dict_index_t*	index;
+  dulint		index_id;
+
+  char *p;
+  char db_name_raw[NAME_LEN*5+1], db_name[NAME_LEN+1];
+  char table_name_raw[NAME_LEN*5+1], table_name[NAME_LEN+1];
+
+	buf_chunk_t*	chunk;
+
+	DBUG_ENTER("i_s_innodb_buffer_pool_pages_index_fill");
+
+	/* deny access to non-superusers */
+	if (check_global_access(thd, PROCESS_ACL)) {
+
+		DBUG_RETURN(0);
+	}
+
+	RETURN_IF_INNODB_NOT_STARTED(tables->schema_table_name);
+
+	buf_pool_mutex_enter();
+  mutex_enter(&(dict_sys->mutex));
+	
+	chunk = buf_pool->chunks;
+  
+	for (n_chunks = buf_pool->n_chunks; n_chunks--; chunk++) {
+		buf_block_t*	block		= chunk->blocks;
+
+		for (n_blocks	= chunk->size; n_blocks--; block++) {
+			const buf_frame_t* frame = block->frame;
+  
+      if (fil_page_get_type(frame) == FIL_PAGE_INDEX) {
+        index_id = btr_page_get_index_id(frame);
+      	index = dict_index_get_if_in_cache_low(index_id);
+      	if(index)
+        {
+          if((p = strchr(index->table_name, '/')))
+          {
+            strncpy(db_name_raw, index->table_name, p-index->table_name);
+            db_name_raw[p-index->table_name] = 0;
+            filename_to_tablename(db_name_raw, db_name, sizeof(db_name));
+            field_store_string(table->field[0], db_name);
+            p++;            
+          } else {
+            field_store_string(table->field[0], NULL);
+            p = (char *)index->table_name;
+          }
+          strcpy(table_name_raw, p);
+          filename_to_tablename(table_name_raw, table_name, sizeof(table_name));
+          field_store_string(table->field[1], table_name);
+          field_store_string(table->field[2], index->name);
+          
+          table->field[3]->store(block->page.space);
+          table->field[4]->store(block->page.offset);
+          table->field[5]->store(page_get_n_recs(frame));
+          table->field[6]->store(page_get_data_size(frame));
+          table->field[7]->store(block->is_hashed);
+          table->field[8]->store(block->page.accessed);
+          table->field[9]->store(block->page.newest_modification != 0);
+          table->field[10]->store(block->page.oldest_modification != 0);
+          table->field[11]->store(block->page.old);
+          table->field[12]->store(block->page.LRU_position);
+          table->field[13]->store(block->page.buf_fix_count);
+          table->field[14]->store(block->page.flush_type);
+          
+          if (schema_table_store_record(thd, table)) {
+            status = 1;
+            break;
+          }
+        }
+      }      
+    }
+	}
+
+  mutex_exit(&(dict_sys->mutex));
+	buf_pool_mutex_exit();
+
+	DBUG_RETURN(status);
+}
+
+/***********************************************************************
+Fill the dynamic table information_schema.innodb_buffer_pool_pages_index. */
+static
+int
+i_s_innodb_buffer_pool_pages_blob_fill(
+/*================*/
+				/* out: 0 on success, 1 on failure */
+	THD*		thd,	/* in: thread */
+	TABLE_LIST*	tables,	/* in/out: tables to fill */
+	COND*		cond)	/* in: condition (ignored) */
+{
+	TABLE*	table	= (TABLE *) tables->table;
+	int	status	= 0;
+
+  ulint		n_chunks, n_blocks;
+	buf_chunk_t*	chunk;
+	page_zip_des_t*	block_page_zip;
+
+	ulint		part_len;
+	ulint		next_page_no;
+
+	DBUG_ENTER("i_s_innodb_buffer_pool_pages_blob_fill");
+
+	/* deny access to non-superusers */
+	if (check_global_access(thd, PROCESS_ACL)) {
+
+		DBUG_RETURN(0);
+	}
+
+	RETURN_IF_INNODB_NOT_STARTED(tables->schema_table_name);
+
+	buf_pool_mutex_enter();
+  mutex_enter(&(dict_sys->mutex));
+	
+	chunk = buf_pool->chunks;
+    
+	for (n_chunks = buf_pool->n_chunks; n_chunks--; chunk++) {
+		buf_block_t*	block		= chunk->blocks;
+    block_page_zip = buf_block_get_page_zip(block);
+
+    for (n_blocks	= chunk->size; n_blocks--; block++) {
+      const buf_frame_t* frame = block->frame;
+
+      if (fil_page_get_type(frame) == FIL_PAGE_TYPE_BLOB) {
+
+        if (UNIV_LIKELY_NULL(block_page_zip)) {
+          part_len = 0; /* hmm, can't figure it out */
+  
+          next_page_no = mach_read_from_4(
+            buf_block_get_frame(block)
+            + FIL_PAGE_NEXT);        
+        } else {
+          part_len = mach_read_from_4(
+            buf_block_get_frame(block)
+            + FIL_PAGE_DATA
+            + 0 /*BTR_BLOB_HDR_PART_LEN*/);
+  
+          next_page_no = mach_read_from_4(
+            buf_block_get_frame(block)
+            + FIL_PAGE_DATA
+            + 4 /*BTR_BLOB_HDR_NEXT_PAGE_NO*/);
+        }
+
+        table->field[0]->store(block->page.space);
+        table->field[1]->store(block->page.offset);
+        table->field[2]->store(block_page_zip != NULL);
+        table->field[3]->store(part_len);
+
+        if(next_page_no == FIL_NULL)
+        {
+          table->field[4]->store(0);
+        } else {
+          table->field[4]->store(block->page.offset);
+        }
+
+        table->field[5]->store(block->page.LRU_position);
+        table->field[6]->store(block->page.buf_fix_count);
+        table->field[7]->store(block->page.flush_type);
+  
+        if (schema_table_store_record(thd, table)) {
+          status = 1;
+          break;
+        }
+
+      }
+    }      
+	}
+
+  mutex_exit(&(dict_sys->mutex));
+	buf_pool_mutex_exit();
+
+	DBUG_RETURN(status);
+}
+
+/***********************************************************************
+Bind the dynamic table information_schema.innodb_buffer_pool_pages. */
+static
+int
+i_s_innodb_buffer_pool_pages_init(
+/*=========*/
+			/* out: 0 on success */
+	void*	p)	/* in/out: table schema object */
+{
+	DBUG_ENTER("i_s_innodb_buffer_pool_pages_init");
+	ST_SCHEMA_TABLE* schema = (ST_SCHEMA_TABLE*) p;
+
+	schema->fields_info = i_s_innodb_buffer_pool_pages_fields_info;
+	schema->fill_table = i_s_innodb_buffer_pool_pages_fill;
+
+	DBUG_RETURN(0);
+}
+
+/***********************************************************************
+Bind the dynamic table information_schema.innodb_buffer_pool_pages. */
+static
+int
+i_s_innodb_buffer_pool_pages_index_init(
+/*=========*/
+			/* out: 0 on success */
+	void*	p)	/* in/out: table schema object */
+{
+	DBUG_ENTER("i_s_innodb_buffer_pool_pages_index_init");
+	ST_SCHEMA_TABLE* schema = (ST_SCHEMA_TABLE*) p;
+
+	schema->fields_info = i_s_innodb_buffer_pool_pages_index_fields_info;
+	schema->fill_table = i_s_innodb_buffer_pool_pages_index_fill;
+
+	DBUG_RETURN(0);
+}
+
+/***********************************************************************
+Bind the dynamic table information_schema.innodb_buffer_pool_pages. */
+static
+int
+i_s_innodb_buffer_pool_pages_blob_init(
+/*=========*/
+			/* out: 0 on success */
+	void*	p)	/* in/out: table schema object */
+{
+	DBUG_ENTER("i_s_innodb_buffer_pool_pages_blob_init");
+	ST_SCHEMA_TABLE* schema = (ST_SCHEMA_TABLE*) p;
+
+	schema->fields_info = i_s_innodb_buffer_pool_pages_blob_fields_info;
+	schema->fill_table = i_s_innodb_buffer_pool_pages_blob_fill;
+
+	DBUG_RETURN(0);
+}
+
+
+UNIV_INTERN struct st_mysql_plugin	i_s_innodb_buffer_pool_pages =
+{
+	/* the plugin type (a MYSQL_XXX_PLUGIN value) */
+	/* int */
+	STRUCT_FLD(type, MYSQL_INFORMATION_SCHEMA_PLUGIN),
+
+	/* pointer to type-specific plugin descriptor */
+	/* void* */
+	STRUCT_FLD(info, &i_s_info),
+
+	/* plugin name */
+	/* const char* */
+	STRUCT_FLD(name, "INNODB_BUFFER_POOL_PAGES"),
+
+	/* plugin author (for SHOW PLUGINS) */
+	/* const char* */
+	STRUCT_FLD(author, plugin_author),
+
+	/* general descriptive text (for SHOW PLUGINS) */
+	/* const char* */
+	STRUCT_FLD(descr, "InnoDB buffer pool pages"),
+
+	/* the plugin license (PLUGIN_LICENSE_XXX) */
+	/* int */
+	STRUCT_FLD(license, PLUGIN_LICENSE_GPL),
+
+	/* the function to invoke when plugin is loaded */
+	/* int (*)(void*); */
+	STRUCT_FLD(init, i_s_innodb_buffer_pool_pages_init),
+
+	/* the function to invoke when plugin is unloaded */
+	/* int (*)(void*); */
+	STRUCT_FLD(deinit, i_s_common_deinit),
+
+	/* plugin version (for SHOW PLUGINS) */
+	/* unsigned int */
+	STRUCT_FLD(version, 0x0100 /* 1.0 */),
+
+	/* struct st_mysql_show_var* */
+	STRUCT_FLD(status_vars, NULL),
+
+	/* struct st_mysql_sys_var** */
+	STRUCT_FLD(system_vars, NULL),
+
+	/* reserved for dependency checking */
+	/* void* */
+	STRUCT_FLD(__reserved1, NULL)
+};
+
+UNIV_INTERN struct st_mysql_plugin	i_s_innodb_buffer_pool_pages_index =
+{
+	/* the plugin type (a MYSQL_XXX_PLUGIN value) */
+	/* int */
+	STRUCT_FLD(type, MYSQL_INFORMATION_SCHEMA_PLUGIN),
+
+	/* pointer to type-specific plugin descriptor */
+	/* void* */
+	STRUCT_FLD(info, &i_s_info),
+
+	/* plugin name */
+	/* const char* */
+	STRUCT_FLD(name, "INNODB_BUFFER_POOL_PAGES_INDEX"),
+
+	/* plugin author (for SHOW PLUGINS) */
+	/* const char* */
+	STRUCT_FLD(author, plugin_author),
+
+	/* general descriptive text (for SHOW PLUGINS) */
+	/* const char* */
+	STRUCT_FLD(descr, "InnoDB buffer pool index pages"),
+
+	/* the plugin license (PLUGIN_LICENSE_XXX) */
+	/* int */
+	STRUCT_FLD(license, PLUGIN_LICENSE_GPL),
+
+	/* the function to invoke when plugin is loaded */
+	/* int (*)(void*); */
+	STRUCT_FLD(init, i_s_innodb_buffer_pool_pages_index_init),
+
+	/* the function to invoke when plugin is unloaded */
+	/* int (*)(void*); */
+	STRUCT_FLD(deinit, i_s_common_deinit),
+
+	/* plugin version (for SHOW PLUGINS) */
+	/* unsigned int */
+	STRUCT_FLD(version, 0x0100 /* 1.0 */),
+
+	/* struct st_mysql_show_var* */
+	STRUCT_FLD(status_vars, NULL),
+
+	/* struct st_mysql_sys_var** */
+	STRUCT_FLD(system_vars, NULL),
+
+	/* reserved for dependency checking */
+	/* void* */
+	STRUCT_FLD(__reserved1, NULL)
+};
+
+UNIV_INTERN struct st_mysql_plugin	i_s_innodb_buffer_pool_pages_blob =
+{
+	/* the plugin type (a MYSQL_XXX_PLUGIN value) */
+	/* int */
+	STRUCT_FLD(type, MYSQL_INFORMATION_SCHEMA_PLUGIN),
+
+	/* pointer to type-specific plugin descriptor */
+	/* void* */
+	STRUCT_FLD(info, &i_s_info),
+
+	/* plugin name */
+	/* const char* */
+	STRUCT_FLD(name, "INNODB_BUFFER_POOL_PAGES_BLOB"),
+
+	/* plugin author (for SHOW PLUGINS) */
+	/* const char* */
+	STRUCT_FLD(author, plugin_author),
+
+	/* general descriptive text (for SHOW PLUGINS) */
+	/* const char* */
+	STRUCT_FLD(descr, "InnoDB buffer pool blob pages"),
+
+	/* the plugin license (PLUGIN_LICENSE_XXX) */
+	/* int */
+	STRUCT_FLD(license, PLUGIN_LICENSE_GPL),
+
+	/* the function to invoke when plugin is loaded */
+	/* int (*)(void*); */
+	STRUCT_FLD(init, i_s_innodb_buffer_pool_pages_blob_init),
+
+	/* the function to invoke when plugin is unloaded */
+	/* int (*)(void*); */
+	STRUCT_FLD(deinit, i_s_common_deinit),
+
+	/* plugin version (for SHOW PLUGINS) */
+	/* unsigned int */
+	STRUCT_FLD(version, 0x0100 /* 1.0 */),
+
+	/* struct st_mysql_show_var* */
+	STRUCT_FLD(status_vars, NULL),
+
+	/* struct st_mysql_sys_var** */
+	STRUCT_FLD(system_vars, NULL),
+
+	/* reserved for dependency checking */
+	/* void* */
+	STRUCT_FLD(__reserved1, NULL)
+};
+
+
+/* Fields of the dynamic table INFORMATION_SCHEMA.innodb_trx */
+static ST_FIELD_INFO	innodb_trx_fields_info[] =
+{
+#define IDX_TRX_ID		0
+	{STRUCT_FLD(field_name,		"trx_id"),
+	 STRUCT_FLD(field_length,	TRX_ID_MAX_LEN + 1),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_STRING),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	0),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+#define IDX_TRX_STATE		1
+	{STRUCT_FLD(field_name,		"trx_state"),
+	 STRUCT_FLD(field_length,	TRX_QUE_STATE_STR_MAX_LEN + 1),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_STRING),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	0),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+#define IDX_TRX_STARTED		2
+	{STRUCT_FLD(field_name,		"trx_started"),
+	 STRUCT_FLD(field_length,	0),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_DATETIME),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	0),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+#define IDX_TRX_REQUESTED_LOCK_ID	3
+	{STRUCT_FLD(field_name,		"trx_requested_lock_id"),
+	 STRUCT_FLD(field_length,	TRX_I_S_LOCK_ID_MAX_LEN + 1),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_STRING),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_MAYBE_NULL),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+#define IDX_TRX_WAIT_STARTED	4
+	{STRUCT_FLD(field_name,		"trx_wait_started"),
+	 STRUCT_FLD(field_length,	0),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_DATETIME),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_MAYBE_NULL),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+#define IDX_TRX_WEIGHT		5
+	{STRUCT_FLD(field_name,		"trx_weight"),
+	 STRUCT_FLD(field_length,	MY_INT64_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONGLONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_UNSIGNED),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+#define IDX_TRX_MYSQL_THREAD_ID	6
+	{STRUCT_FLD(field_name,		"trx_mysql_thread_id"),
+	 STRUCT_FLD(field_length,	MY_INT64_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONGLONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_UNSIGNED),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+#define IDX_TRX_QUERY		7
+	{STRUCT_FLD(field_name,		"trx_query"),
+	 STRUCT_FLD(field_length,	TRX_I_S_TRX_QUERY_MAX_LEN),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_STRING),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_MAYBE_NULL),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	END_OF_ST_FIELD_INFO
+};
+
+/***********************************************************************
+Read data from cache buffer and fill the INFORMATION_SCHEMA.innodb_trx
+table with it. */
+static
+int
+fill_innodb_trx_from_cache(
+/*=======================*/
+					/* out: 0 on success */
+	trx_i_s_cache_t*	cache,	/* in: cache to read from */
+	THD*			thd,	/* in: used to call
+					schema_table_store_record() */
+	TABLE*			table)	/* in/out: fill this table */
+{
+	Field**	fields;
+	ulint	rows_num;
+	char	lock_id[TRX_I_S_LOCK_ID_MAX_LEN + 1];
+	ulint	i;
+
+	DBUG_ENTER("fill_innodb_trx_from_cache");
+
+	fields = table->field;
+
+	rows_num = trx_i_s_cache_get_rows_used(cache,
+					       I_S_INNODB_TRX);
+
+	for (i = 0; i < rows_num; i++) {
+
+		i_s_trx_row_t*	row;
+		char		trx_id[TRX_ID_MAX_LEN + 1];
+
+		row = (i_s_trx_row_t*)
+			trx_i_s_cache_get_nth_row(
+				cache, I_S_INNODB_TRX, i);
+
+		/* trx_id */
+		ut_snprintf(trx_id, sizeof(trx_id), TRX_ID_FMT, row->trx_id);
+		OK(field_store_string(fields[IDX_TRX_ID], trx_id));
+
+		/* trx_state */
+		OK(field_store_string(fields[IDX_TRX_STATE],
+				      row->trx_state));
+
+		/* trx_started */
+		OK(field_store_time_t(fields[IDX_TRX_STARTED],
+				      (time_t) row->trx_started));
+
+		/* trx_requested_lock_id */
+		/* trx_wait_started */
+		if (row->trx_wait_started != 0) {
+
+			OK(field_store_string(
+				   fields[IDX_TRX_REQUESTED_LOCK_ID],
+				   trx_i_s_create_lock_id(
+					   row->requested_lock_row,
+					   lock_id, sizeof(lock_id))));
+			/* field_store_string() sets it no notnull */
+
+			OK(field_store_time_t(
+				   fields[IDX_TRX_WAIT_STARTED],
+				   (time_t) row->trx_wait_started));
+			fields[IDX_TRX_WAIT_STARTED]->set_notnull();
+		} else {
+
+			fields[IDX_TRX_REQUESTED_LOCK_ID]->set_null();
+			fields[IDX_TRX_WAIT_STARTED]->set_null();
+		}
+
+		/* trx_weight */
+		OK(fields[IDX_TRX_WEIGHT]->store((longlong) row->trx_weight,
+						 true));
+
+		/* trx_mysql_thread_id */
+		OK(fields[IDX_TRX_MYSQL_THREAD_ID]->store(
+			   row->trx_mysql_thread_id));
+
+		/* trx_query */
+		OK(field_store_string(fields[IDX_TRX_QUERY],
+				      row->trx_query));
+
+		OK(schema_table_store_record(thd, table));
+	}
+
+	DBUG_RETURN(0);
+}
+
+/***********************************************************************
+Bind the dynamic table INFORMATION_SCHEMA.innodb_trx */
+static
+int
+innodb_trx_init(
+/*============*/
+			/* out: 0 on success */
+	void*	p)	/* in/out: table schema object */
+{
+	ST_SCHEMA_TABLE*	schema;
+
+	DBUG_ENTER("innodb_trx_init");
+
+	schema = (ST_SCHEMA_TABLE*) p;
+
+	schema->fields_info = innodb_trx_fields_info;
+	schema->fill_table = trx_i_s_common_fill_table;
+
+	DBUG_RETURN(0);
+}
+
+
+UNIV_INTERN struct st_mysql_plugin	i_s_innodb_trx =
+{
+	/* the plugin type (a MYSQL_XXX_PLUGIN value) */
+	/* int */
+	STRUCT_FLD(type, MYSQL_INFORMATION_SCHEMA_PLUGIN),
+
+	/* pointer to type-specific plugin descriptor */
+	/* void* */
+	STRUCT_FLD(info, &i_s_info),
+
+	/* plugin name */
+	/* const char* */
+	STRUCT_FLD(name, "INNODB_TRX"),
+
+	/* plugin author (for SHOW PLUGINS) */
+	/* const char* */
+	STRUCT_FLD(author, plugin_author),
+
+	/* general descriptive text (for SHOW PLUGINS) */
+	/* const char* */
+	STRUCT_FLD(descr, "InnoDB transactions"),
+
+	/* the plugin license (PLUGIN_LICENSE_XXX) */
+	/* int */
+	STRUCT_FLD(license, PLUGIN_LICENSE_GPL),
+
+	/* the function to invoke when plugin is loaded */
+	/* int (*)(void*); */
+	STRUCT_FLD(init, innodb_trx_init),
+
+	/* the function to invoke when plugin is unloaded */
+	/* int (*)(void*); */
+	STRUCT_FLD(deinit, i_s_common_deinit),
+
+	/* plugin version (for SHOW PLUGINS) */
+	/* unsigned int */
+	STRUCT_FLD(version, INNODB_VERSION_SHORT),
+
+	/* struct st_mysql_show_var* */
+	STRUCT_FLD(status_vars, NULL),
+
+	/* struct st_mysql_sys_var** */
+	STRUCT_FLD(system_vars, NULL),
+
+	/* reserved for dependency checking */
+	/* void* */
+	STRUCT_FLD(__reserved1, NULL)
+};
+
+/* Fields of the dynamic table INFORMATION_SCHEMA.innodb_locks */
+static ST_FIELD_INFO	innodb_locks_fields_info[] =
+{
+#define IDX_LOCK_ID		0
+	{STRUCT_FLD(field_name,		"lock_id"),
+	 STRUCT_FLD(field_length,	TRX_I_S_LOCK_ID_MAX_LEN + 1),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_STRING),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	0),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+#define IDX_LOCK_TRX_ID		1
+	{STRUCT_FLD(field_name,		"lock_trx_id"),
+	 STRUCT_FLD(field_length,	TRX_ID_MAX_LEN + 1),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_STRING),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	0),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+#define IDX_LOCK_MODE		2
+	{STRUCT_FLD(field_name,		"lock_mode"),
+	 /* S[,GAP] X[,GAP] IS[,GAP] IX[,GAP] AUTO_INC UNKNOWN */
+	 STRUCT_FLD(field_length,	32),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_STRING),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	0),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+#define IDX_LOCK_TYPE		3
+	{STRUCT_FLD(field_name,		"lock_type"),
+	 STRUCT_FLD(field_length,	32 /* RECORD|TABLE|UNKNOWN */),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_STRING),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	0),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+#define IDX_LOCK_TABLE		4
+	{STRUCT_FLD(field_name,		"lock_table"),
+	 STRUCT_FLD(field_length,	1024),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_STRING),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	0),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+#define IDX_LOCK_INDEX		5
+	{STRUCT_FLD(field_name,		"lock_index"),
+	 STRUCT_FLD(field_length,	1024),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_STRING),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_MAYBE_NULL),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+#define IDX_LOCK_SPACE		6
+	{STRUCT_FLD(field_name,		"lock_space"),
+	 STRUCT_FLD(field_length,	MY_INT64_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONGLONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_UNSIGNED | MY_I_S_MAYBE_NULL),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+#define IDX_LOCK_PAGE		7
+	{STRUCT_FLD(field_name,		"lock_page"),
+	 STRUCT_FLD(field_length,	MY_INT64_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONGLONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_UNSIGNED | MY_I_S_MAYBE_NULL),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+#define IDX_LOCK_REC		8
+	{STRUCT_FLD(field_name,		"lock_rec"),
+	 STRUCT_FLD(field_length,	MY_INT64_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONGLONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_UNSIGNED | MY_I_S_MAYBE_NULL),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+#define IDX_LOCK_DATA		9
+	{STRUCT_FLD(field_name,		"lock_data"),
+	 STRUCT_FLD(field_length,	TRX_I_S_LOCK_DATA_MAX_LEN),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_STRING),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_MAYBE_NULL),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	END_OF_ST_FIELD_INFO
+};
+
+/***********************************************************************
+Read data from cache buffer and fill the INFORMATION_SCHEMA.innodb_locks
+table with it. */
+static
+int
+fill_innodb_locks_from_cache(
+/*=========================*/
+					/* out: 0 on success */
+	trx_i_s_cache_t*	cache,	/* in: cache to read from */
+	THD*			thd,	/* in: MySQL client connection */
+	TABLE*			table)	/* in/out: fill this table */
+{
+	Field**	fields;
+	ulint	rows_num;
+	char	lock_id[TRX_I_S_LOCK_ID_MAX_LEN + 1];
+	ulint	i;
+
+	DBUG_ENTER("fill_innodb_locks_from_cache");
+
+	fields = table->field;
+
+	rows_num = trx_i_s_cache_get_rows_used(cache,
+					       I_S_INNODB_LOCKS);
+
+	for (i = 0; i < rows_num; i++) {
+
+		i_s_locks_row_t*	row;
+
+		/* note that the decoded database or table name is
+		never expected to be longer than NAME_LEN;
+		NAME_LEN for database name
+		2 for surrounding quotes around database name
+		NAME_LEN for table name
+		2 for surrounding quotes around table name
+		1 for the separating dot (.)
+		9 for the #mysql50# prefix */
+		char			buf[2 * NAME_LEN + 14];
+		const char*		bufend;
+
+		char			lock_trx_id[TRX_ID_MAX_LEN + 1];
+
+		row = (i_s_locks_row_t*)
+			trx_i_s_cache_get_nth_row(
+				cache, I_S_INNODB_LOCKS, i);
+
+		/* lock_id */
+		trx_i_s_create_lock_id(row, lock_id, sizeof(lock_id));
+		OK(field_store_string(fields[IDX_LOCK_ID],
+				      lock_id));
+
+		/* lock_trx_id */
+		ut_snprintf(lock_trx_id, sizeof(lock_trx_id),
+			    TRX_ID_FMT, row->lock_trx_id);
+		OK(field_store_string(fields[IDX_LOCK_TRX_ID], lock_trx_id));
+
+		/* lock_mode */
+		OK(field_store_string(fields[IDX_LOCK_MODE],
+				      row->lock_mode));
+
+		/* lock_type */
+		OK(field_store_string(fields[IDX_LOCK_TYPE],
+				      row->lock_type));
+
+		/* lock_table */
+		bufend = innobase_convert_name(buf, sizeof(buf),
+					       row->lock_table,
+					       strlen(row->lock_table),
+					       thd, TRUE);
+		OK(fields[IDX_LOCK_TABLE]->store(buf, bufend - buf,
+						 system_charset_info));
+
+		/* lock_index */
+		if (row->lock_index != NULL) {
+
+			bufend = innobase_convert_name(buf, sizeof(buf),
+						       row->lock_index,
+						       strlen(row->lock_index),
+						       thd, FALSE);
+			OK(fields[IDX_LOCK_INDEX]->store(buf, bufend - buf,
+							 system_charset_info));
+			fields[IDX_LOCK_INDEX]->set_notnull();
+		} else {
+
+			fields[IDX_LOCK_INDEX]->set_null();
+		}
+
+		/* lock_space */
+		OK(field_store_ulint(fields[IDX_LOCK_SPACE],
+				     row->lock_space));
+
+		/* lock_page */
+		OK(field_store_ulint(fields[IDX_LOCK_PAGE],
+				     row->lock_page));
+
+		/* lock_rec */
+		OK(field_store_ulint(fields[IDX_LOCK_REC],
+				     row->lock_rec));
+
+		/* lock_data */
+		OK(field_store_string(fields[IDX_LOCK_DATA],
+				      row->lock_data));
+
+		OK(schema_table_store_record(thd, table));
+	}
+
+	DBUG_RETURN(0);
+}
+
+/***********************************************************************
+Bind the dynamic table INFORMATION_SCHEMA.innodb_locks */
+static
+int
+innodb_locks_init(
+/*==============*/
+			/* out: 0 on success */
+	void*	p)	/* in/out: table schema object */
+{
+	ST_SCHEMA_TABLE*	schema;
+
+	DBUG_ENTER("innodb_locks_init");
+
+	schema = (ST_SCHEMA_TABLE*) p;
+
+	schema->fields_info = innodb_locks_fields_info;
+	schema->fill_table = trx_i_s_common_fill_table;
+
+	DBUG_RETURN(0);
+}
+
+UNIV_INTERN struct st_mysql_plugin	i_s_innodb_locks =
+{
+	/* the plugin type (a MYSQL_XXX_PLUGIN value) */
+	/* int */
+	STRUCT_FLD(type, MYSQL_INFORMATION_SCHEMA_PLUGIN),
+
+	/* pointer to type-specific plugin descriptor */
+	/* void* */
+	STRUCT_FLD(info, &i_s_info),
+
+	/* plugin name */
+	/* const char* */
+	STRUCT_FLD(name, "INNODB_LOCKS"),
+
+	/* plugin author (for SHOW PLUGINS) */
+	/* const char* */
+	STRUCT_FLD(author, plugin_author),
+
+	/* general descriptive text (for SHOW PLUGINS) */
+	/* const char* */
+	STRUCT_FLD(descr, "InnoDB conflicting locks"),
+
+	/* the plugin license (PLUGIN_LICENSE_XXX) */
+	/* int */
+	STRUCT_FLD(license, PLUGIN_LICENSE_GPL),
+
+	/* the function to invoke when plugin is loaded */
+	/* int (*)(void*); */
+	STRUCT_FLD(init, innodb_locks_init),
+
+	/* the function to invoke when plugin is unloaded */
+	/* int (*)(void*); */
+	STRUCT_FLD(deinit, i_s_common_deinit),
+
+	/* plugin version (for SHOW PLUGINS) */
+	/* unsigned int */
+	STRUCT_FLD(version, INNODB_VERSION_SHORT),
+
+	/* struct st_mysql_show_var* */
+	STRUCT_FLD(status_vars, NULL),
+
+	/* struct st_mysql_sys_var** */
+	STRUCT_FLD(system_vars, NULL),
+
+	/* reserved for dependency checking */
+	/* void* */
+	STRUCT_FLD(__reserved1, NULL)
+};
+
+/* Fields of the dynamic table INFORMATION_SCHEMA.innodb_lock_waits */
+static ST_FIELD_INFO	innodb_lock_waits_fields_info[] =
+{
+#define IDX_REQUESTING_TRX_ID	0
+	{STRUCT_FLD(field_name,		"requesting_trx_id"),
+	 STRUCT_FLD(field_length,	TRX_ID_MAX_LEN + 1),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_STRING),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	0),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+#define IDX_REQUESTED_LOCK_ID	1
+	{STRUCT_FLD(field_name,		"requested_lock_id"),
+	 STRUCT_FLD(field_length,	TRX_I_S_LOCK_ID_MAX_LEN + 1),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_STRING),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	0),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+#define IDX_BLOCKING_TRX_ID	2
+	{STRUCT_FLD(field_name,		"blocking_trx_id"),
+	 STRUCT_FLD(field_length,	TRX_ID_MAX_LEN + 1),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_STRING),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	0),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+#define IDX_BLOCKING_LOCK_ID	3
+	{STRUCT_FLD(field_name,		"blocking_lock_id"),
+	 STRUCT_FLD(field_length,	TRX_I_S_LOCK_ID_MAX_LEN + 1),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_STRING),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	0),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	END_OF_ST_FIELD_INFO
+};
+
+/***********************************************************************
+Read data from cache buffer and fill the
+INFORMATION_SCHEMA.innodb_lock_waits table with it. */
+static
+int
+fill_innodb_lock_waits_from_cache(
+/*==============================*/
+					/* out: 0 on success */
+	trx_i_s_cache_t*	cache,	/* in: cache to read from */
+	THD*			thd,	/* in: used to call
+					schema_table_store_record() */
+	TABLE*			table)	/* in/out: fill this table */
+{
+	Field**	fields;
+	ulint	rows_num;
+	char	requested_lock_id[TRX_I_S_LOCK_ID_MAX_LEN + 1];
+	char	blocking_lock_id[TRX_I_S_LOCK_ID_MAX_LEN + 1];
+	ulint	i;
+
+	DBUG_ENTER("fill_innodb_lock_waits_from_cache");
+
+	fields = table->field;
+
+	rows_num = trx_i_s_cache_get_rows_used(cache,
+					       I_S_INNODB_LOCK_WAITS);
+
+	for (i = 0; i < rows_num; i++) {
+
+		i_s_lock_waits_row_t*	row;
+
+		char	requesting_trx_id[TRX_ID_MAX_LEN + 1];
+		char	blocking_trx_id[TRX_ID_MAX_LEN + 1];
+
+		row = (i_s_lock_waits_row_t*)
+			trx_i_s_cache_get_nth_row(
+				cache, I_S_INNODB_LOCK_WAITS, i);
+
+		/* requesting_trx_id */
+		ut_snprintf(requesting_trx_id, sizeof(requesting_trx_id),
+			    TRX_ID_FMT, row->requested_lock_row->lock_trx_id);
+		OK(field_store_string(fields[IDX_REQUESTING_TRX_ID],
+				      requesting_trx_id));
+
+		/* requested_lock_id */
+		OK(field_store_string(
+			   fields[IDX_REQUESTED_LOCK_ID],
+			   trx_i_s_create_lock_id(
+				   row->requested_lock_row,
+				   requested_lock_id,
+				   sizeof(requested_lock_id))));
+
+		/* blocking_trx_id */
+		ut_snprintf(blocking_trx_id, sizeof(blocking_trx_id),
+			    TRX_ID_FMT, row->blocking_lock_row->lock_trx_id);
+		OK(field_store_string(fields[IDX_BLOCKING_TRX_ID],
+				      blocking_trx_id));
+
+		/* blocking_lock_id */
+		OK(field_store_string(
+			   fields[IDX_BLOCKING_LOCK_ID],
+			   trx_i_s_create_lock_id(
+				   row->blocking_lock_row,
+				   blocking_lock_id,
+				   sizeof(blocking_lock_id))));
+
+		OK(schema_table_store_record(thd, table));
+	}
+
+	DBUG_RETURN(0);
+}
+
+/***********************************************************************
+Bind the dynamic table INFORMATION_SCHEMA.innodb_lock_waits */
+static
+int
+innodb_lock_waits_init(
+/*===================*/
+			/* out: 0 on success */
+	void*	p)	/* in/out: table schema object */
+{
+	ST_SCHEMA_TABLE*	schema;
+
+	DBUG_ENTER("innodb_lock_waits_init");
+
+	schema = (ST_SCHEMA_TABLE*) p;
+
+	schema->fields_info = innodb_lock_waits_fields_info;
+	schema->fill_table = trx_i_s_common_fill_table;
+
+	DBUG_RETURN(0);
+}
+
+UNIV_INTERN struct st_mysql_plugin	i_s_innodb_lock_waits =
+{
+	/* the plugin type (a MYSQL_XXX_PLUGIN value) */
+	/* int */
+	STRUCT_FLD(type, MYSQL_INFORMATION_SCHEMA_PLUGIN),
+
+	/* pointer to type-specific plugin descriptor */
+	/* void* */
+	STRUCT_FLD(info, &i_s_info),
+
+	/* plugin name */
+	/* const char* */
+	STRUCT_FLD(name, "INNODB_LOCK_WAITS"),
+
+	/* plugin author (for SHOW PLUGINS) */
+	/* const char* */
+	STRUCT_FLD(author, "Innobase Oy"),
+
+	/* general descriptive text (for SHOW PLUGINS) */
+	/* const char* */
+	STRUCT_FLD(descr, "InnoDB which lock is blocking which"),
+
+	/* the plugin license (PLUGIN_LICENSE_XXX) */
+	/* int */
+	STRUCT_FLD(license, PLUGIN_LICENSE_GPL),
+
+	/* the function to invoke when plugin is loaded */
+	/* int (*)(void*); */
+	STRUCT_FLD(init, innodb_lock_waits_init),
+
+	/* the function to invoke when plugin is unloaded */
+	/* int (*)(void*); */
+	STRUCT_FLD(deinit, i_s_common_deinit),
+
+	/* plugin version (for SHOW PLUGINS) */
+	/* unsigned int */
+	STRUCT_FLD(version, INNODB_VERSION_SHORT),
+
+	/* struct st_mysql_show_var* */
+	STRUCT_FLD(status_vars, NULL),
+
+	/* struct st_mysql_sys_var** */
+	STRUCT_FLD(system_vars, NULL),
+
+	/* reserved for dependency checking */
+	/* void* */
+	STRUCT_FLD(__reserved1, NULL)
+};
+
+/***********************************************************************
+Common function to fill any of the dynamic tables:
+INFORMATION_SCHEMA.innodb_trx
+INFORMATION_SCHEMA.innodb_locks
+INFORMATION_SCHEMA.innodb_lock_waits */
+static
+int
+trx_i_s_common_fill_table(
+/*======================*/
+				/* out: 0 on success */
+	THD*		thd,	/* in: thread */
+	TABLE_LIST*	tables,	/* in/out: tables to fill */
+	COND*		cond)	/* in: condition (not used) */
+{
+	const char*		table_name;
+	int			ret;
+	trx_i_s_cache_t*	cache;
+
+	DBUG_ENTER("trx_i_s_common_fill_table");
+
+	/* deny access to non-superusers */
+	if (check_global_access(thd, PROCESS_ACL)) {
+
+		DBUG_RETURN(0);
+	}
+
+	/* minimize the number of places where global variables are
+	referenced */
+	cache = trx_i_s_cache;
+
+	/* which table we have to fill? */
+	table_name = tables->schema_table_name;
+	/* or table_name = tables->schema_table->table_name; */
+
+	RETURN_IF_INNODB_NOT_STARTED(table_name);
+
+	/* update the cache */
+	trx_i_s_cache_start_write(cache);
+	trx_i_s_possibly_fetch_data_into_cache(cache);
+	trx_i_s_cache_end_write(cache);
+
+	if (trx_i_s_cache_is_truncated(cache)) {
+
+		/* XXX show warning to user if possible */
+		fprintf(stderr, "Warning: data in %s truncated due to "
+			"memory limit of %d bytes\n", table_name,
+			TRX_I_S_MEM_LIMIT);
+	}
+
+	ret = 0;
+
+	trx_i_s_cache_start_read(cache);
+
+	if (innobase_strcasecmp(table_name, "innodb_trx") == 0) {
+
+		if (fill_innodb_trx_from_cache(
+			cache, thd, tables->table) != 0) {
+
+			ret = 1;
+		}
+
+	} else if (innobase_strcasecmp(table_name, "innodb_locks") == 0) {
+
+		if (fill_innodb_locks_from_cache(
+			cache, thd, tables->table) != 0) {
+
+			ret = 1;
+		}
+
+	} else if (innobase_strcasecmp(table_name, "innodb_lock_waits") == 0) {
+
+		if (fill_innodb_lock_waits_from_cache(
+			cache, thd, tables->table) != 0) {
+
+			ret = 1;
+		}
+
+	} else {
+
+		/* huh! what happened!? */
+		fprintf(stderr,
+			"InnoDB: trx_i_s_common_fill_table() was "
+			"called to fill unknown table: %s.\n"
+			"This function only knows how to fill "
+			"innodb_trx, innodb_locks and "
+			"innodb_lock_waits tables.\n", table_name);
+
+		ret = 1;
+	}
+
+	trx_i_s_cache_end_read(cache);
+
+#if 0
+	DBUG_RETURN(ret);
+#else
+	/* if this function returns something else than 0 then a
+	deadlock occurs between the mysqld server and mysql client,
+	see http://bugs.mysql.com/29900 ; when that bug is resolved
+	we can enable the DBUG_RETURN(ret) above */
+	DBUG_RETURN(0);
+#endif
+}
+
+/* Fields of the dynamic table information_schema.innodb_cmp. */
+static ST_FIELD_INFO	i_s_cmp_fields_info[] =
+{
+	{STRUCT_FLD(field_name,		"page_size"),
+	 STRUCT_FLD(field_length,	5),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	0),
+	 STRUCT_FLD(old_name,		"Compressed Page Size"),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"compress_ops"),
+	 STRUCT_FLD(field_length,	MY_INT32_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	0),
+	 STRUCT_FLD(old_name,		"Total Number of Compressions"),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"compress_ops_ok"),
+	 STRUCT_FLD(field_length,	MY_INT32_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	0),
+	 STRUCT_FLD(old_name,		"Total Number of"
+					" Successful Compressions"),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"compress_time"),
+	 STRUCT_FLD(field_length,	MY_INT32_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	0),
+	 STRUCT_FLD(old_name,		"Total Duration of Compressions,"
+		    " in Seconds"),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"uncompress_ops"),
+	 STRUCT_FLD(field_length,	MY_INT32_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	0),
+	 STRUCT_FLD(old_name,		"Total Number of Decompressions"),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"uncompress_time"),
+	 STRUCT_FLD(field_length,	MY_INT32_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	0),
+	 STRUCT_FLD(old_name,		"Total Duration of Decompressions,"
+		    " in Seconds"),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	END_OF_ST_FIELD_INFO
+};
+
+
+/***********************************************************************
+Fill the dynamic table information_schema.innodb_cmp or
+innodb_cmp_reset. */
+static
+int
+i_s_cmp_fill_low(
+/*=============*/
+				/* out: 0 on success, 1 on failure */
+	THD*		thd,	/* in: thread */
+	TABLE_LIST*	tables,	/* in/out: tables to fill */
+	COND*		cond,	/* in: condition (ignored) */
+	ibool		reset)	/* in: TRUE=reset cumulated counts */
+{
+	TABLE*	table	= (TABLE *) tables->table;
+	int	status	= 0;
+
+	DBUG_ENTER("i_s_cmp_fill_low");
+
+	/* deny access to non-superusers */
+	if (check_global_access(thd, PROCESS_ACL)) {
+
+		DBUG_RETURN(0);
+	}
+
+	RETURN_IF_INNODB_NOT_STARTED(tables->schema_table_name);
+
+	for (uint i = 0; i < PAGE_ZIP_NUM_SSIZE - 1; i++) {
+		page_zip_stat_t*	zip_stat = &page_zip_stat[i];
+
+		table->field[0]->store(PAGE_ZIP_MIN_SIZE << i);
+
+		/* The cumulated counts are not protected by any
+		mutex.  Thus, some operation in page0zip.c could
+		increment a counter between the time we read it and
+		clear it.  We could introduce mutex protection, but it
+		could cause a measureable performance hit in
+		page0zip.c. */
+		table->field[1]->store(zip_stat->compressed);
+		table->field[2]->store(zip_stat->compressed_ok);
+		table->field[3]->store(
+			(ulong) (zip_stat->compressed_usec / 1000000));
+		table->field[4]->store(zip_stat->decompressed);
+		table->field[5]->store(
+			(ulong) (zip_stat->decompressed_usec / 1000000));
+
+		if (reset) {
+			memset(zip_stat, 0, sizeof *zip_stat);
+		}
+
+		if (schema_table_store_record(thd, table)) {
+			status = 1;
+			break;
+		}
+	}
+
+	DBUG_RETURN(status);
+}
+
+/***********************************************************************
+Fill the dynamic table information_schema.innodb_cmp. */
+static
+int
+i_s_cmp_fill(
+/*=========*/
+				/* out: 0 on success, 1 on failure */
+	THD*		thd,	/* in: thread */
+	TABLE_LIST*	tables,	/* in/out: tables to fill */
+	COND*		cond)	/* in: condition (ignored) */
+{
+	return(i_s_cmp_fill_low(thd, tables, cond, FALSE));
+}
+
+/***********************************************************************
+Fill the dynamic table information_schema.innodb_cmp_reset. */
+static
+int
+i_s_cmp_reset_fill(
+/*===============*/
+				/* out: 0 on success, 1 on failure */
+	THD*		thd,	/* in: thread */
+	TABLE_LIST*	tables,	/* in/out: tables to fill */
+	COND*		cond)	/* in: condition (ignored) */
+{
+	return(i_s_cmp_fill_low(thd, tables, cond, TRUE));
+}
+
+/***********************************************************************
+Bind the dynamic table information_schema.innodb_cmp. */
+static
+int
+i_s_cmp_init(
+/*=========*/
+			/* out: 0 on success */
+	void*	p)	/* in/out: table schema object */
+{
+	DBUG_ENTER("i_s_cmp_init");
+	ST_SCHEMA_TABLE* schema = (ST_SCHEMA_TABLE*) p;
+
+	schema->fields_info = i_s_cmp_fields_info;
+	schema->fill_table = i_s_cmp_fill;
+
+	DBUG_RETURN(0);
+}
+
+/***********************************************************************
+Bind the dynamic table information_schema.innodb_cmp_reset. */
+static
+int
+i_s_cmp_reset_init(
+/*===============*/
+			/* out: 0 on success */
+	void*	p)	/* in/out: table schema object */
+{
+	DBUG_ENTER("i_s_cmp_reset_init");
+	ST_SCHEMA_TABLE* schema = (ST_SCHEMA_TABLE*) p;
+
+	schema->fields_info = i_s_cmp_fields_info;
+	schema->fill_table = i_s_cmp_reset_fill;
+
+	DBUG_RETURN(0);
+}
+
+UNIV_INTERN struct st_mysql_plugin	i_s_innodb_cmp =
+{
+	/* the plugin type (a MYSQL_XXX_PLUGIN value) */
+	/* int */
+	STRUCT_FLD(type, MYSQL_INFORMATION_SCHEMA_PLUGIN),
+
+	/* pointer to type-specific plugin descriptor */
+	/* void* */
+	STRUCT_FLD(info, &i_s_info),
+
+	/* plugin name */
+	/* const char* */
+	STRUCT_FLD(name, "INNODB_CMP"),
+
+	/* plugin author (for SHOW PLUGINS) */
+	/* const char* */
+	STRUCT_FLD(author, plugin_author),
+
+	/* general descriptive text (for SHOW PLUGINS) */
+	/* const char* */
+	STRUCT_FLD(descr, "Statistics for the InnoDB compression"),
+
+	/* the plugin license (PLUGIN_LICENSE_XXX) */
+	/* int */
+	STRUCT_FLD(license, PLUGIN_LICENSE_GPL),
+
+	/* the function to invoke when plugin is loaded */
+	/* int (*)(void*); */
+	STRUCT_FLD(init, i_s_cmp_init),
+
+	/* the function to invoke when plugin is unloaded */
+	/* int (*)(void*); */
+	STRUCT_FLD(deinit, i_s_common_deinit),
+
+	/* plugin version (for SHOW PLUGINS) */
+	/* unsigned int */
+	STRUCT_FLD(version, INNODB_VERSION_SHORT),
+
+	/* struct st_mysql_show_var* */
+	STRUCT_FLD(status_vars, NULL),
+
+	/* struct st_mysql_sys_var** */
+	STRUCT_FLD(system_vars, NULL),
+
+	/* reserved for dependency checking */
+	/* void* */
+	STRUCT_FLD(__reserved1, NULL)
+};
+
+UNIV_INTERN struct st_mysql_plugin	i_s_innodb_cmp_reset =
+{
+	/* the plugin type (a MYSQL_XXX_PLUGIN value) */
+	/* int */
+	STRUCT_FLD(type, MYSQL_INFORMATION_SCHEMA_PLUGIN),
+
+	/* pointer to type-specific plugin descriptor */
+	/* void* */
+	STRUCT_FLD(info, &i_s_info),
+
+	/* plugin name */
+	/* const char* */
+	STRUCT_FLD(name, "INNODB_CMP_RESET"),
+
+	/* plugin author (for SHOW PLUGINS) */
+	/* const char* */
+	STRUCT_FLD(author, plugin_author),
+
+	/* general descriptive text (for SHOW PLUGINS) */
+	/* const char* */
+	STRUCT_FLD(descr, "Statistics for the InnoDB compression;"
+		   " reset cumulated counts"),
+
+	/* the plugin license (PLUGIN_LICENSE_XXX) */
+	/* int */
+	STRUCT_FLD(license, PLUGIN_LICENSE_GPL),
+
+	/* the function to invoke when plugin is loaded */
+	/* int (*)(void*); */
+	STRUCT_FLD(init, i_s_cmp_reset_init),
+
+	/* the function to invoke when plugin is unloaded */
+	/* int (*)(void*); */
+	STRUCT_FLD(deinit, i_s_common_deinit),
+
+	/* plugin version (for SHOW PLUGINS) */
+	/* unsigned int */
+	STRUCT_FLD(version, INNODB_VERSION_SHORT),
+
+	/* struct st_mysql_show_var* */
+	STRUCT_FLD(status_vars, NULL),
+
+	/* struct st_mysql_sys_var** */
+	STRUCT_FLD(system_vars, NULL),
+
+	/* reserved for dependency checking */
+	/* void* */
+	STRUCT_FLD(__reserved1, NULL)
+};
+
+/* Fields of the dynamic table information_schema.innodb_cmpmem. */
+static ST_FIELD_INFO	i_s_cmpmem_fields_info[] =
+{
+	{STRUCT_FLD(field_name,		"page_size"),
+	 STRUCT_FLD(field_length,	5),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	0),
+	 STRUCT_FLD(old_name,		"Buddy Block Size"),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"pages_used"),
+	 STRUCT_FLD(field_length,	MY_INT32_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	0),
+	 STRUCT_FLD(old_name,		"Currently in Use"),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"pages_free"),
+	 STRUCT_FLD(field_length,	MY_INT32_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	0),
+	 STRUCT_FLD(old_name,		"Currently Available"),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"relocation_ops"),
+	 STRUCT_FLD(field_length,	MY_INT64_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONGLONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	0),
+	 STRUCT_FLD(old_name,		"Total Number of Relocations"),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"relocation_time"),
+	 STRUCT_FLD(field_length,	MY_INT32_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	0),
+	 STRUCT_FLD(old_name,		"Total Duration of Relocations,"
+		    " in Seconds"),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	END_OF_ST_FIELD_INFO
+};
+
+/***********************************************************************
+Fill the dynamic table information_schema.innodb_cmpmem or
+innodb_cmpmem_reset. */
+static
+int
+i_s_cmpmem_fill_low(
+/*================*/
+				/* out: 0 on success, 1 on failure */
+	THD*		thd,	/* in: thread */
+	TABLE_LIST*	tables,	/* in/out: tables to fill */
+	COND*		cond,	/* in: condition (ignored) */
+	ibool		reset)	/* in: TRUE=reset cumulated counts */
+{
+	TABLE*	table	= (TABLE *) tables->table;
+	int	status	= 0;
+
+	DBUG_ENTER("i_s_cmpmem_fill_low");
+
+	/* deny access to non-superusers */
+	if (check_global_access(thd, PROCESS_ACL)) {
+
+		DBUG_RETURN(0);
+	}
+
+	RETURN_IF_INNODB_NOT_STARTED(tables->schema_table_name);
+
+	buf_pool_mutex_enter();
+
+	for (uint x = 0; x <= BUF_BUDDY_SIZES; x++) {
+		buf_buddy_stat_t*	buddy_stat = &buf_buddy_stat[x];
+
+		table->field[0]->store(BUF_BUDDY_LOW << x);
+		table->field[1]->store(buddy_stat->used);
+		table->field[2]->store(UNIV_LIKELY(x < BUF_BUDDY_SIZES)
+				       ? UT_LIST_GET_LEN(buf_pool->zip_free[x])
+				       : 0);
+		table->field[3]->store((longlong) buddy_stat->relocated, true);
+		table->field[4]->store(
+			(ulong) (buddy_stat->relocated_usec / 1000000));
+
+		if (reset) {
+			/* This is protected by buf_pool_mutex. */
+			buddy_stat->relocated = 0;
+			buddy_stat->relocated_usec = 0;
+		}
+
+		if (schema_table_store_record(thd, table)) {
+			status = 1;
+			break;
+		}
+	}
+
+	buf_pool_mutex_exit();
+	DBUG_RETURN(status);
+}
+
+/***********************************************************************
+Fill the dynamic table information_schema.innodb_cmpmem. */
+static
+int
+i_s_cmpmem_fill(
+/*============*/
+				/* out: 0 on success, 1 on failure */
+	THD*		thd,	/* in: thread */
+	TABLE_LIST*	tables,	/* in/out: tables to fill */
+	COND*		cond)	/* in: condition (ignored) */
+{
+	return(i_s_cmpmem_fill_low(thd, tables, cond, FALSE));
+}
+
+/***********************************************************************
+Fill the dynamic table information_schema.innodb_cmpmem_reset. */
+static
+int
+i_s_cmpmem_reset_fill(
+/*==================*/
+				/* out: 0 on success, 1 on failure */
+	THD*		thd,	/* in: thread */
+	TABLE_LIST*	tables,	/* in/out: tables to fill */
+	COND*		cond)	/* in: condition (ignored) */
+{
+	return(i_s_cmpmem_fill_low(thd, tables, cond, TRUE));
+}
+
+/***********************************************************************
+Bind the dynamic table information_schema.innodb_cmpmem. */
+static
+int
+i_s_cmpmem_init(
+/*============*/
+			/* out: 0 on success */
+	void*	p)	/* in/out: table schema object */
+{
+	DBUG_ENTER("i_s_cmpmem_init");
+	ST_SCHEMA_TABLE* schema = (ST_SCHEMA_TABLE*) p;
+
+	schema->fields_info = i_s_cmpmem_fields_info;
+	schema->fill_table = i_s_cmpmem_fill;
+
+	DBUG_RETURN(0);
+}
+
+/***********************************************************************
+Bind the dynamic table information_schema.innodb_cmpmem_reset. */
+static
+int
+i_s_cmpmem_reset_init(
+/*==================*/
+			/* out: 0 on success */
+	void*	p)	/* in/out: table schema object */
+{
+	DBUG_ENTER("i_s_cmpmem_reset_init");
+	ST_SCHEMA_TABLE* schema = (ST_SCHEMA_TABLE*) p;
+
+	schema->fields_info = i_s_cmpmem_fields_info;
+	schema->fill_table = i_s_cmpmem_reset_fill;
+
+	DBUG_RETURN(0);
+}
+
+UNIV_INTERN struct st_mysql_plugin	i_s_innodb_cmpmem =
+{
+	/* the plugin type (a MYSQL_XXX_PLUGIN value) */
+	/* int */
+	STRUCT_FLD(type, MYSQL_INFORMATION_SCHEMA_PLUGIN),
+
+	/* pointer to type-specific plugin descriptor */
+	/* void* */
+	STRUCT_FLD(info, &i_s_info),
+
+	/* plugin name */
+	/* const char* */
+	STRUCT_FLD(name, "INNODB_CMPMEM"),
+
+	/* plugin author (for SHOW PLUGINS) */
+	/* const char* */
+	STRUCT_FLD(author, plugin_author),
+
+	/* general descriptive text (for SHOW PLUGINS) */
+	/* const char* */
+	STRUCT_FLD(descr, "Statistics for the InnoDB compressed buffer pool"),
+
+	/* the plugin license (PLUGIN_LICENSE_XXX) */
+	/* int */
+	STRUCT_FLD(license, PLUGIN_LICENSE_GPL),
+
+	/* the function to invoke when plugin is loaded */
+	/* int (*)(void*); */
+	STRUCT_FLD(init, i_s_cmpmem_init),
+
+	/* the function to invoke when plugin is unloaded */
+	/* int (*)(void*); */
+	STRUCT_FLD(deinit, i_s_common_deinit),
+
+	/* plugin version (for SHOW PLUGINS) */
+	/* unsigned int */
+	STRUCT_FLD(version, INNODB_VERSION_SHORT),
+
+	/* struct st_mysql_show_var* */
+	STRUCT_FLD(status_vars, NULL),
+
+	/* struct st_mysql_sys_var** */
+	STRUCT_FLD(system_vars, NULL),
+
+	/* reserved for dependency checking */
+	/* void* */
+	STRUCT_FLD(__reserved1, NULL)
+};
+
+UNIV_INTERN struct st_mysql_plugin	i_s_innodb_cmpmem_reset =
+{
+	/* the plugin type (a MYSQL_XXX_PLUGIN value) */
+	/* int */
+	STRUCT_FLD(type, MYSQL_INFORMATION_SCHEMA_PLUGIN),
+
+	/* pointer to type-specific plugin descriptor */
+	/* void* */
+	STRUCT_FLD(info, &i_s_info),
+
+	/* plugin name */
+	/* const char* */
+	STRUCT_FLD(name, "INNODB_CMPMEM_RESET"),
+
+	/* plugin author (for SHOW PLUGINS) */
+	/* const char* */
+	STRUCT_FLD(author, plugin_author),
+
+	/* general descriptive text (for SHOW PLUGINS) */
+	/* const char* */
+	STRUCT_FLD(descr, "Statistics for the InnoDB compressed buffer pool;"
+		   " reset cumulated counts"),
+
+	/* the plugin license (PLUGIN_LICENSE_XXX) */
+	/* int */
+	STRUCT_FLD(license, PLUGIN_LICENSE_GPL),
+
+	/* the function to invoke when plugin is loaded */
+	/* int (*)(void*); */
+	STRUCT_FLD(init, i_s_cmpmem_reset_init),
+
+	/* the function to invoke when plugin is unloaded */
+	/* int (*)(void*); */
+	STRUCT_FLD(deinit, i_s_common_deinit),
+
+	/* plugin version (for SHOW PLUGINS) */
+	/* unsigned int */
+	STRUCT_FLD(version, INNODB_VERSION_SHORT),
+
+	/* struct st_mysql_show_var* */
+	STRUCT_FLD(status_vars, NULL),
+
+	/* struct st_mysql_sys_var** */
+	STRUCT_FLD(system_vars, NULL),
+
+	/* reserved for dependency checking */
+	/* void* */
+	STRUCT_FLD(__reserved1, NULL)
+};
+
+/***********************************************************************
+Unbind a dynamic INFORMATION_SCHEMA table. */
+static
+int
+i_s_common_deinit(
+/*==============*/
+			/* out: 0 on success */
+	void*	p)	/* in/out: table schema object */
+{
+	DBUG_ENTER("i_s_common_deinit");
+
+	/* Do nothing */
+
+	DBUG_RETURN(0);
+}
+
+/***********************************************************************
+*/
+static ST_FIELD_INFO	i_s_innodb_rseg_fields_info[] =
+{
+	{STRUCT_FLD(field_name,		"rseg_id"),
+	 STRUCT_FLD(field_length,	MY_INT64_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONGLONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_UNSIGNED),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"space_id"),
+	 STRUCT_FLD(field_length,	MY_INT64_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONGLONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_UNSIGNED),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"zip_size"),
+	 STRUCT_FLD(field_length,	MY_INT64_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONGLONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_UNSIGNED),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"page_no"),
+	 STRUCT_FLD(field_length,	MY_INT64_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONGLONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_UNSIGNED),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"max_size"),
+	 STRUCT_FLD(field_length,	MY_INT64_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONGLONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_UNSIGNED),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	{STRUCT_FLD(field_name,		"curr_size"),
+	 STRUCT_FLD(field_length,	MY_INT64_NUM_DECIMAL_DIGITS),
+	 STRUCT_FLD(field_type,		MYSQL_TYPE_LONGLONG),
+	 STRUCT_FLD(value,		0),
+	 STRUCT_FLD(field_flags,	MY_I_S_UNSIGNED),
+	 STRUCT_FLD(old_name,		""),
+	 STRUCT_FLD(open_method,	SKIP_OPEN_TABLE)},
+
+	END_OF_ST_FIELD_INFO
+};
+
+static
+int
+i_s_innodb_rseg_fill(
+/*=================*/
+	THD*		thd,	/* in: thread */
+	TABLE_LIST*	tables,	/* in/out: tables to fill */
+	COND*		cond)	/* in: condition (ignored) */
+{
+	TABLE*	table	= (TABLE *) tables->table;
+	int	status	= 0;
+	trx_rseg_t*	rseg;
+
+	DBUG_ENTER("i_s_innodb_rseg_fill");
+
+	/* deny access to non-superusers */
+	if (check_global_access(thd, PROCESS_ACL)) {
+
+		DBUG_RETURN(0);
+	}
+
+	RETURN_IF_INNODB_NOT_STARTED(tables->schema_table_name);
+
+	rseg = UT_LIST_GET_FIRST(trx_sys->rseg_list);
+
+	while (rseg) {
+		table->field[0]->store(rseg->id);
+		table->field[1]->store(rseg->space);
+		table->field[2]->store(rseg->zip_size);
+		table->field[3]->store(rseg->page_no);
+		table->field[4]->store(rseg->max_size);
+		table->field[5]->store(rseg->curr_size);
+
+		if (schema_table_store_record(thd, table)) {
+			status = 1;
+			break;
+		}
+
+		rseg = UT_LIST_GET_NEXT(rseg_list, rseg);
+	}
+
+	DBUG_RETURN(status);
+}
+
+static
+int
+i_s_innodb_rseg_init(
+/*=================*/
+			/* out: 0 on success */
+	void*	p)	/* in/out: table schema object */
+{
+	DBUG_ENTER("i_s_innodb_rseg_init");
+	ST_SCHEMA_TABLE* schema = (ST_SCHEMA_TABLE*) p;
+
+	schema->fields_info = i_s_innodb_rseg_fields_info;
+	schema->fill_table = i_s_innodb_rseg_fill;
+
+	DBUG_RETURN(0);
+}
+
+UNIV_INTERN struct st_mysql_plugin	i_s_innodb_rseg =
+{
+	/* the plugin type (a MYSQL_XXX_PLUGIN value) */
+	/* int */
+	STRUCT_FLD(type, MYSQL_INFORMATION_SCHEMA_PLUGIN),
+
+	/* pointer to type-specific plugin descriptor */
+	/* void* */
+	STRUCT_FLD(info, &i_s_info),
+
+	/* plugin name */
+	/* const char* */
+	STRUCT_FLD(name, "INNODB_RSEG"),
+
+	/* plugin author (for SHOW PLUGINS) */
+	/* const char* */
+	STRUCT_FLD(author, plugin_author),
+
+	/* general descriptive text (for SHOW PLUGINS) */
+	/* const char* */
+	STRUCT_FLD(descr, "InnoDB rollback segment information"),
+
+	/* the plugin license (PLUGIN_LICENSE_XXX) */
+	/* int */
+	STRUCT_FLD(license, PLUGIN_LICENSE_GPL),
+
+	/* the function to invoke when plugin is loaded */
+	/* int (*)(void*); */
+	STRUCT_FLD(init, i_s_innodb_rseg_init),
+
+	/* the function to invoke when plugin is unloaded */
+	/* int (*)(void*); */
+	STRUCT_FLD(deinit, i_s_common_deinit),
+
+	/* plugin version (for SHOW PLUGINS) */
+	/* unsigned int */
+	STRUCT_FLD(version, 0x0100 /* 1.0 */),
+
+	/* struct st_mysql_show_var* */
+	STRUCT_FLD(status_vars, NULL),
+
+	/* struct st_mysql_sys_var** */
+	STRUCT_FLD(system_vars, NULL),
+
+	/* reserved for dependency checking */
+	/* void* */
+	STRUCT_FLD(__reserved1, NULL)
+};
diff --git a/storage/xtradb/handler/i_s.h b/storage/xtradb/handler/i_s.h
new file mode 100644
index 00000000000..de0e7610646
--- /dev/null
+++ b/storage/xtradb/handler/i_s.h
@@ -0,0 +1,41 @@
+/*****************************************************************************
+
+Copyright (c) 2007, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+InnoDB INFORMATION SCHEMA tables interface to MySQL.
+
+Created July 18, 2007 Vasil Dimov
+*******************************************************/
+
+#ifndef i_s_h
+#define i_s_h
+
+extern struct st_mysql_plugin	i_s_innodb_buffer_pool_pages;
+extern struct st_mysql_plugin	i_s_innodb_buffer_pool_pages_index;
+extern struct st_mysql_plugin	i_s_innodb_buffer_pool_pages_blob;
+extern struct st_mysql_plugin	i_s_innodb_trx;
+extern struct st_mysql_plugin	i_s_innodb_locks;
+extern struct st_mysql_plugin	i_s_innodb_lock_waits;
+extern struct st_mysql_plugin	i_s_innodb_cmp;
+extern struct st_mysql_plugin	i_s_innodb_cmp_reset;
+extern struct st_mysql_plugin	i_s_innodb_cmpmem;
+extern struct st_mysql_plugin	i_s_innodb_cmpmem_reset;
+extern struct st_mysql_plugin	i_s_innodb_patches;
+extern struct st_mysql_plugin	i_s_innodb_rseg;
+
+#endif /* i_s_h */
diff --git a/storage/xtradb/handler/innodb_patch_info.h b/storage/xtradb/handler/innodb_patch_info.h
new file mode 100644
index 00000000000..976e6eefe1c
--- /dev/null
+++ b/storage/xtradb/handler/innodb_patch_info.h
@@ -0,0 +1,35 @@
+/* Copyright (C) 2002-2006 MySQL AB
+  
+   This program is free software; you can redistribute it and/or modify
+   it under the terms of the GNU General Public License as published by
+   the Free Software Foundation; version 2 of the License.
+  
+   This program is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+   GNU General Public License for more details.
+
+   You should have received a copy of the GNU General Public License
+   along with this program; if not, write to the Free Software
+   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA */
+
+#ifdef USE_PRAGMA_INTERFACE
+#pragma interface                      /* gcc class implementation */
+#endif
+
+struct innodb_enhancement {
+       const char *file;
+       const char *name;
+       const char *comment;
+       const char *link;
+}innodb_enhancements[] = {
+{"xtradb_show_enhancements","I_S.XTRADB_ENHANCEMENTS","","http://www.percona.com/docs/wiki/percona-xtradb"},
+{"innodb_show_status","Improvements to SHOW INNODB STATUS","Memory information and lock info fixes","http://www.percona.com/docs/wiki/percona-xtradb"},
+{"innodb_io","Improvements to InnoDB IO","","http://www.percona.com/docs/wiki/percona-xtradb"},
+{"innodb_opt_lru_count","Fix of buffer_pool mutex","Decreases contention on buffer_pool mutex on LRU operations","http://www.percona.com/docs/wiki/percona-xtradb"},
+{"innodb_buffer_pool_pages","Information of buffer pool content","","http://www.percona.com/docs/wiki/percona-xtradb"},
+{"innodb_expand_undo_slots","expandable maximum number of undo slots","from 1024 (default) to about 4000","http://www.percona.com/docs/wiki/percona-xtradb"},
+{"innodb_extra_rseg","allow to create extra rollback segments","When create new db, the new parameter allows to create more rollback segments","http://www.percona.com/docs/wiki/percona-xtradb"},
+{"innodb_overwrite_relay_log_info","overwrite relay-log.info when slave recovery","Building as plugin, it is not used.","http://www.percona.com/docs/wiki/percona-xtradb:innodb_overwrite_relay_log_info"},
+{NULL, NULL, NULL, NULL}
+};
diff --git a/storage/xtradb/handler/mysql_addons.cc b/storage/xtradb/handler/mysql_addons.cc
new file mode 100644
index 00000000000..a5d9c82c3e3
--- /dev/null
+++ b/storage/xtradb/handler/mysql_addons.cc
@@ -0,0 +1,41 @@
+/*****************************************************************************
+
+Copyright (c) 2007, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+This file contains functions that need to be added to
+MySQL code but have not been added yet.
+
+Whenever you add a function here submit a MySQL bug
+report (feature request) with the implementation. Then
+write the bug number in the comment before the
+function in this file.
+
+When MySQL commits the function it can be deleted from
+here. In a perfect world this file exists but is empty.
+
+Created November 07, 2007 Vasil Dimov
+*******************************************************/
+
+#ifndef MYSQL_SERVER
+#define MYSQL_SERVER
+#endif /* MYSQL_SERVER */
+
+#include <mysql_priv.h>
+
+#include "mysql_addons.h"
+#include "univ.i"
diff --git a/storage/xtradb/handler/win_delay_loader.cc b/storage/xtradb/handler/win_delay_loader.cc
new file mode 100644
index 00000000000..1572df42e30
--- /dev/null
+++ b/storage/xtradb/handler/win_delay_loader.cc
@@ -0,0 +1,1036 @@
+/*****************************************************************************
+
+Copyright (c) 2008, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/***********************************************************************
+This file contains functions that implement the delay loader on Windows.
+
+This is a customized version of delay loader with limited functionalities.
+It does not support:
+
+* (manual) unloading
+* multiple delay loaded DLLs
+* multiple loading of the same DLL
+
+This delay loader is used only by the InnoDB plugin. Other components (DLLs)
+can still use the default delay loader, provided by MSVC.
+
+Several acronyms used by Microsoft:
+ * IAT: import address table
+ * INT: import name table
+ * RVA: Relative Virtual Address
+
+See http://msdn.microsoft.com/en-us/magazine/bb985992.aspx for details of
+PE format.
+***********************************************************************/
+#if defined (__WIN__) && defined (MYSQL_DYNAMIC_PLUGIN)
+# define WIN32_LEAN_AND_MEAN
+# include <windows.h>
+# include <delayimp.h>
+# include <mysql_priv.h>
+
+extern "C" {
+# include "univ.i"
+# include "hash0hash.h"
+}
+
+/***********************************************************************
+This following contains a list of externals that can not be resolved by
+delay loading. They have to be resolved indirectly via their addresses
+in the .map file. All of them are external variables. */
+CHARSET_INFO*		wdl_my_charset_bin;
+CHARSET_INFO*		wdl_my_charset_latin1;
+CHARSET_INFO*		wdl_my_charset_filename;
+CHARSET_INFO**		wdl_system_charset_info;
+CHARSET_INFO**		wdl_default_charset_info;
+CHARSET_INFO**		wdl_all_charsets;
+system_variables*	wdl_global_system_variables;
+char*			wdl_mysql_real_data_home;
+char**			wdl_mysql_data_home;
+char**			wdl_tx_isolation_names;
+char**			wdl_binlog_format_names;
+char*			wdl_reg_ext;
+pthread_mutex_t*	wdl_LOCK_thread_count;
+key_map*		wdl_key_map_full;
+MY_TMPDIR*		wdl_mysql_tmpdir_list;
+bool*			wdl_mysqld_embedded;
+uint*			wdl_lower_case_table_names;
+ulong*			wdl_specialflag;
+int*			wdl_my_umask;
+
+/***********************************************************************
+The following is defined in ha_innodb.cc. It is used for copying the
+system variables from the builtin innodb plugin to the dynamic plugin.
+*/
+extern struct st_mysql_plugin*	builtin_innobase_plugin_ptr;
+
+/***********************************************************************
+The preffered load-address defined in PE (portable executable format).*/
+#if defined(_M_IA64)
+#pragma section(".base", long, read)
+extern "C"
+__declspec(allocate(".base"))
+const IMAGE_DOS_HEADER __ImageBase;
+#else
+extern "C"
+const IMAGE_DOS_HEADER __ImageBase;
+#endif
+
+/***********************************************************************
+A template function for converting a relative address (RVA) to an
+absolute address (VA). This is due to the pointers in the delay
+descriptor (ImgDelayDescr in delayimp.h) have been changed from
+VAs to RVAs to work on both 32- and 64-bit platforms. */
+template <class X>
+X PFromRva(RVA rva) {
+	return X(PBYTE(&__ImageBase) + rva);
+}
+
+/***********************************************************************
+Convert to the old format for convenience. The structure as well as its
+element names follow the definition of ImgDelayDescr in delayimp.h. */
+struct InternalImgDelayDescr {
+	DWORD		grAttrs;	/* attributes */
+	LPCSTR		szName;		/* pointer to dll name */
+	HMODULE*	phmod;		/* address of module handle */
+	PImgThunkData	pIAT;		/* address of the IAT */
+	PCImgThunkData	pINT;		/* address of the INT */
+	PCImgThunkData	pBoundIAT;	/* address of the optional bound IAT */
+	PCImgThunkData	pUnloadIAT;	/* address of optional copy of
+					   original IAT */
+	DWORD		dwTimeStamp;	/* 0 if not bound,
+					   otherwise date/time stamp of DLL
+					   bound to (Old BIND) */
+};
+
+typedef struct map_hash_chain_struct	map_hash_chain_t;
+
+struct map_hash_chain_struct {
+	char*			symbol;	/* pointer to a symbol */
+	ulint			value;	/* address of the symbol */
+	map_hash_chain_t*	next;	/* pointer to the next cell
+					in the same folder. */
+	map_hash_chain_t*	chain;	/* a linear chain used for
+					cleanup. */
+};
+
+static HMODULE				my_hmod = 0;
+static struct hash_table_struct*	m_htbl = NULL ;
+static map_hash_chain_t*		chain_header = NULL;
+static ibool				wdl_init = FALSE;
+const ulint				MAP_HASH_CELLS_NUM = 10000;
+
+#ifndef DBUG_OFF
+/***********************************************************************
+In the dynamic plugin, it is required to call the following dbug functions
+in the server:
+	_db_pargs_
+	_db_doprnt_
+	_db_enter_
+	_db_return_
+	_db_dump_
+
+The plugin will get those function pointers during the initialization.
+*/
+typedef void (__cdecl* pfn_db_enter_)(
+	const char*	_func_,
+	const char*	_file_,
+	uint		_line_,
+	const char**	_sfunc_,
+	const char**	_sfile_,
+	uint*		_slevel_,
+	char***);
+
+typedef void (__cdecl* pfn_db_return_)(
+	uint		_line_,
+	const char**	_sfunc_,
+	const char**	_sfile_,
+	uint*		_slevel_);
+
+typedef void (__cdecl* pfn_db_pargs_)(
+	uint		_line_,
+	const char*	keyword);
+
+typedef void (__cdecl* pfn_db_doprnt_)(
+	const char*	format,
+	...);
+
+typedef void (__cdecl* pfn_db_dump_)(
+	uint			_line_,
+	const char*		keyword,
+	const unsigned char*	memory,
+	size_t			length);
+
+static pfn_db_enter_	wdl_db_enter_;
+static pfn_db_return_	wdl_db_return_;
+static pfn_db_pargs_	wdl_db_pargs_;
+static pfn_db_doprnt_	wdl_db_doprnt_;
+static pfn_db_dump_	wdl_db_dump_;
+#endif /* !DBUG_OFF */
+
+/*****************************************************************
+Creates a hash table with >= n array cells. The actual number of cells is
+chosen to be a prime number slightly bigger than n.
+
+This is the same function as hash_create in hash0hash.c, except the
+memory allocation. This function is invoked before the engine is
+initialized, and buffer pools are not ready yet. */
+static
+hash_table_t*
+wdl_hash_create(
+/*============*/
+			/* out, own: created hash table */
+	ulint	n)	/* in: number of array cells */
+{
+	hash_cell_t*	array;
+	ulint		prime;
+	hash_table_t*	table;
+
+	prime = ut_find_prime(n);
+
+	table = (hash_table_t*) malloc(sizeof(hash_table_t));
+	if (table == NULL) {
+		return(NULL);
+	}
+
+	array = (hash_cell_t*) malloc(sizeof(hash_cell_t) * prime);
+	if (array == NULL) {
+		free(table);
+		return(NULL);
+	}
+
+	table->array = array;
+	table->n_cells = prime;
+	table->n_mutexes = 0;
+	table->mutexes = NULL;
+	table->heaps = NULL;
+	table->heap = NULL;
+	table->magic_n = HASH_TABLE_MAGIC_N;
+
+	/* Initialize the cell array */
+	hash_table_clear(table);
+
+	return(table);
+}
+
+/*****************************************************************
+Frees a hash table. */
+static
+void
+wdl_hash_table_free(
+/*================*/
+	hash_table_t*	table)	/* in, own: hash table */
+{
+	ut_a(table != NULL);
+	ut_a(table->mutexes == NULL);
+
+	free(table->array);
+	free(table);
+}
+
+/***********************************************************************
+Function for calculating the count of imports given the base of the IAT. */
+static
+ulint
+wdl_import_count(
+/*=============*/
+					/* out: number of imports */
+	PCImgThunkData	pitd_base)	/* in: base of the IAT */
+{
+	ulint		ret = 0;
+	PCImgThunkData	pitd = pitd_base;
+
+	while (pitd->u1.Function) {
+		pitd++;
+		ret++;
+	}
+
+	return(ret);
+}
+
+/***********************************************************************
+Read Mapfile to a hashtable for faster access */
+static
+ibool
+wdl_load_mapfile(
+/*=============*/
+					/* out: TRUE if the mapfile is
+					loaded successfully. */
+	const char*	filename)	/* in: name of the mapfile. */
+{
+	FILE*		fp;
+	const size_t	nSize = 256;
+	char		tmp_buf[nSize];
+	char*		func_name;
+	char*		func_addr;
+	ulint		load_addr = 0;
+	ibool		valid_load_addr = FALSE;
+#ifdef _WIN64
+	const char*	tmp_string = " Preferred load address is %16llx";
+#else
+	const char*	tmp_string = " Preferred load address is %08x";
+#endif
+
+	fp = fopen(filename, "r");
+	if (fp == NULL) {
+
+		return(FALSE);
+	}
+
+	/* Check whether to create the hashtable */
+	if (m_htbl == NULL) {
+
+		m_htbl = wdl_hash_create(MAP_HASH_CELLS_NUM);
+
+		if (m_htbl == NULL) {
+
+			fclose(fp);
+			return(FALSE);
+		}
+	}
+
+	/* Search start of symbol list and get the preferred load address */
+	while (fgets(tmp_buf, sizeof(tmp_buf), fp)) {
+
+		if (sscanf(tmp_buf, tmp_string, &load_addr) == 1) {
+
+			valid_load_addr = TRUE;
+		}
+
+		if (strstr(tmp_buf, "Rva+Base") != NULL) {
+
+			break;
+		}
+	}
+
+	if (valid_load_addr == FALSE) {
+
+		/* No "Preferred load address", the map file is wrong. */
+		fclose(fp);
+		return(FALSE);
+	}
+
+	/* Read symbol list */
+	while (fgets(tmp_buf, sizeof(tmp_buf), fp))
+	{
+		map_hash_chain_t*	map_cell;
+		ulint			map_fold;
+
+		if (*tmp_buf == 0) {
+
+			continue;
+		}
+
+		func_name = strtok(tmp_buf, " ");
+		func_name = strtok(NULL, " ");
+		func_addr = strtok(NULL, " ");
+
+		if (func_name && func_addr) {
+
+			ut_snprintf(tmp_buf, nSize, "0x%s", func_addr);
+			if (*func_name == '_') {
+
+				func_name++;
+			}
+
+			map_cell = (map_hash_chain_t*)
+				   malloc(sizeof(map_hash_chain_t));
+			if (map_cell == NULL) {
+				return(FALSE);
+			}
+
+			/* Chain all cells together */
+			map_cell->chain = chain_header;
+			chain_header = map_cell;
+
+			map_cell->symbol = strdup(func_name);
+			map_cell->value = (ulint) strtoull(tmp_buf, NULL, 0)
+					  - load_addr;
+			map_fold = ut_fold_string(map_cell->symbol);
+
+			HASH_INSERT(map_hash_chain_t,
+				    next,
+				    m_htbl,
+				    map_fold,
+				    map_cell);
+		}
+	}
+
+	fclose(fp);
+
+	return(TRUE);
+}
+
+/*****************************************************************
+Cleanup.during DLL unload */
+static
+void
+wdl_cleanup(void)
+/*=============*/
+{
+	while (chain_header != NULL) {
+		map_hash_chain_t*	tmp;
+
+		tmp = chain_header->chain;
+		free(chain_header->symbol);
+		free(chain_header);
+		chain_header = tmp;
+	}
+
+	if (m_htbl != NULL) {
+
+		wdl_hash_table_free(m_htbl);
+	}
+}
+
+/***********************************************************************
+Load the mapfile mysqld.map. */
+static
+HMODULE
+wdl_get_mysqld_mapfile(void)
+/*========================*/
+			/* out: the module handle */
+{
+	char	file_name[MAX_PATH];
+	char*	ext;
+	ulint	err;
+
+	if (my_hmod == 0) {
+
+		size_t	nSize = MAX_PATH - strlen(".map") -1;
+
+		/* First find out the name of current executable */
+		my_hmod = GetModuleHandle(NULL);
+		if (my_hmod == 0) {
+
+			return(my_hmod);
+		}
+
+		err = GetModuleFileName(my_hmod, file_name, nSize);
+		if (err == 0) {
+
+			my_hmod = 0;
+			return(my_hmod);
+		}
+
+		ext = strrchr(file_name, '.');
+		if (ext != NULL) {
+
+			*ext = 0;
+			strcat(file_name, ".map");
+
+			err = wdl_load_mapfile(file_name);
+			if (err == 0) {
+
+				my_hmod = 0;
+			}
+		} else {
+
+			my_hmod = 0;
+		}
+	}
+
+	return(my_hmod);
+}
+
+/***********************************************************************
+Retrieves the address of an exported function. It follows the convention
+of GetProcAddress(). */
+static
+FARPROC
+wdl_get_procaddr_from_map(
+/*======================*/
+					/* out: address of exported
+					function. */
+	HANDLE		m_handle,	/* in: module handle */
+	const char*	import_proc)	/* in: procedure name */
+{
+	map_hash_chain_t*	hash_chain;
+	ulint			map_fold;
+
+	map_fold = ut_fold_string(import_proc);
+	HASH_SEARCH(
+		next,
+		m_htbl,
+		map_fold,
+		map_hash_chain_t*,
+		hash_chain,
+		,
+		(ut_strcmp(hash_chain->symbol, import_proc) == 0));
+
+	if (hash_chain == NULL) {
+
+#ifdef _WIN64
+		/* On Win64, the leading '_' may not be taken out. In this
+		case, search again without the leading '_'. */
+		if (*import_proc == '_') {
+
+			import_proc++;
+		}
+
+		map_fold = ut_fold_string(import_proc);
+		HASH_SEARCH(
+			next,
+			m_htbl,
+			map_fold,
+			map_hash_chain_t*,
+			hash_chain,
+			,
+			(ut_strcmp(hash_chain->symbol, import_proc) == 0));
+
+		if (hash_chain == NULL) {
+#endif
+			if (wdl_init == TRUE) {
+
+				sql_print_error(
+					"InnoDB: the procedure pointer of %s"
+					" is not found.",
+					import_proc);
+			}
+
+			return(0);
+#ifdef _WIN64
+		}
+#endif
+	}
+
+	return((FARPROC) ((ulint) m_handle + hash_chain->value));
+}
+
+/***********************************************************************
+Retrieves the address of an exported variable.
+Note: It does not follow the Windows call convention FARPROC. */
+static
+void*
+wdl_get_varaddr_from_map(
+/*=====================*/
+						/* out: address of exported
+						variable. */
+	HANDLE		m_handle,		/* in: module handle */
+	const char*	import_variable)	/* in: variable name */
+{
+	map_hash_chain_t*	hash_chain;
+	ulint			map_fold;
+
+	map_fold = ut_fold_string(import_variable);
+	HASH_SEARCH(
+		next,
+		m_htbl,
+		map_fold,
+		map_hash_chain_t*,
+		hash_chain,
+		,
+		(ut_strcmp(hash_chain->symbol, import_variable) == 0));
+
+	if (hash_chain == NULL) {
+
+#ifdef _WIN64
+		/* On Win64, the leading '_' may not be taken out. In this
+		case, search again without the leading '_'. */
+		if (*import_variable == '_') {
+
+			import_variable++;
+		}
+
+		map_fold = ut_fold_string(import_variable);
+		HASH_SEARCH(
+			next,
+			m_htbl,
+			map_fold,
+			map_hash_chain_t*,
+			hash_chain,
+			,
+			(ut_strcmp(hash_chain->symbol, import_variable) == 0));
+
+		if (hash_chain == NULL) {
+#endif
+			if (wdl_init == TRUE) {
+
+				sql_print_error(
+					"InnoDB: the variable address of %s"
+					" is not found.",
+					import_variable);
+			}
+
+			return(0);
+#ifdef _WIN64
+		}
+#endif
+	}
+
+	return((void*) ((ulint) m_handle + hash_chain->value));
+}
+
+/***********************************************************************
+Bind all unresolved external variables from the MySQL executable. */
+static
+bool
+wdl_get_external_variables(void)
+/*============================*/
+			/* out: TRUE if successful */
+{
+	HMODULE	hmod = wdl_get_mysqld_mapfile();
+
+	if (hmod == 0) {
+
+		return(FALSE);
+	}
+
+#define GET_SYM(sym, var, type)					\
+	var = (type*) wdl_get_varaddr_from_map(hmod, sym);	\
+	if (var == NULL) return(FALSE)
+#ifdef _WIN64
+#define GET_SYM2(sym1, sym2, var, type)				\
+	var = (type*) wdl_get_varaddr_from_map(hmod, sym1);	\
+	if (var == NULL) return(FALSE)
+#else
+#define GET_SYM2(sym1, sym2, var, type)				\
+	var = (type*) wdl_get_varaddr_from_map(hmod, sym2);	\
+	if (var == NULL) return(FALSE)
+#endif // (_WIN64)
+#define GET_C_SYM(sym, type) GET_SYM(#sym, wdl_##sym, type)
+#define GET_PROC_ADDR(sym)					\
+	wdl##sym = (pfn##sym) wdl_get_procaddr_from_map(hmod, #sym)
+
+	GET_C_SYM(my_charset_bin, CHARSET_INFO);
+	GET_C_SYM(my_charset_latin1, CHARSET_INFO);
+	GET_C_SYM(my_charset_filename, CHARSET_INFO);
+	GET_C_SYM(default_charset_info, CHARSET_INFO*);
+	GET_C_SYM(all_charsets, CHARSET_INFO*);
+	GET_C_SYM(my_umask, int);
+
+	GET_SYM("?global_system_variables@@3Usystem_variables@@A",
+		wdl_global_system_variables, struct system_variables);
+	GET_SYM("?mysql_real_data_home@@3PADA",
+		wdl_mysql_real_data_home, char);
+	GET_SYM("?reg_ext@@3PADA", wdl_reg_ext, char);
+	GET_SYM("?LOCK_thread_count@@3U_RTL_CRITICAL_SECTION@@A",
+		wdl_LOCK_thread_count, pthread_mutex_t);
+	GET_SYM("?key_map_full@@3V?$Bitmap@$0EA@@@A",
+		wdl_key_map_full, key_map);
+	GET_SYM("?mysql_tmpdir_list@@3Ust_my_tmpdir@@A",
+		wdl_mysql_tmpdir_list, MY_TMPDIR);
+	GET_SYM("?mysqld_embedded@@3_NA",
+		wdl_mysqld_embedded, bool);
+	GET_SYM("?lower_case_table_names@@3IA",
+		wdl_lower_case_table_names, uint);
+	GET_SYM("?specialflag@@3KA", wdl_specialflag, ulong);
+
+	GET_SYM2("?system_charset_info@@3PEAUcharset_info_st@@EA",
+		 "?system_charset_info@@3PAUcharset_info_st@@A",
+		 wdl_system_charset_info, CHARSET_INFO*);
+	GET_SYM2("?mysql_data_home@@3PEADEA",
+		 "?mysql_data_home@@3PADA",
+		 wdl_mysql_data_home, char*);
+	GET_SYM2("?tx_isolation_names@@3PAPEBDA",
+		 "?tx_isolation_names@@3PAPBDA",
+		 wdl_tx_isolation_names, char*);
+	GET_SYM2("?binlog_format_names@@3PAPEBDA",
+		 "?binlog_format_names@@3PAPBDA",
+		 wdl_binlog_format_names, char*);
+
+	/* It is fine if builtin_innobase_plugin is not available. */
+	builtin_innobase_plugin_ptr = (struct st_mysql_plugin*)
+		wdl_get_varaddr_from_map(
+			hmod,
+			"?builtin_innobase_plugin@@3PAUst_mysql_plugin@@A");
+
+#ifndef DBUG_OFF
+	GET_PROC_ADDR(_db_enter_);
+	GET_PROC_ADDR(_db_return_);
+	GET_PROC_ADDR(_db_pargs_);
+	GET_PROC_ADDR(_db_doprnt_);
+	GET_PROC_ADDR(_db_dump_);
+
+	/* If any of the dbug functions is not available, just make them
+	all invalid. This is the case when working with a non-debug
+	version of the server. */
+	if (wdl_db_enter_ == NULL || wdl_db_return_ == NULL
+	    || wdl_db_pargs_ == NULL || wdl_db_doprnt_ == NULL
+	    || wdl_db_dump_ == NULL) {
+
+		wdl_db_enter_ = NULL;
+		wdl_db_return_ = NULL;
+		wdl_db_pargs_ = NULL;
+		wdl_db_doprnt_ = NULL;
+		wdl_db_dump_ = NULL;
+	}
+#endif /* !DBUG_OFF */
+
+	wdl_init = TRUE;
+	return(TRUE);
+
+#undef GET_SYM
+#undef GET_SYM2
+#undef GET_C_SYM
+#undef GET_PROC_ADDR
+}
+
+/***********************************************************************
+The DLL Delayed Loading Helper Function for resolving externals.
+
+The function may fail due to one of the three reasons:
+
+* Invalid parameter, which happens if the attributes in pidd aren't
+  specified correctly.
+* Failed to load the map file mysqld.map.
+* Failed to find an external name in the map file mysqld.map.
+
+Note: this function is called by run-time as well as __HrLoadAllImportsForDll.
+So, it has to follow Windows call convention. */
+extern "C"
+FARPROC WINAPI
+__delayLoadHelper2(
+/*===============*/
+					/* out: the address of the imported
+					function*/
+	PCImgDelayDescr	pidd,		/* in: a const pointer to a
+					ImgDelayDescr, see delayimp.h. */
+	FARPROC*	iat_entry)	/* in/out: A pointer to the slot in
+					the delay load import address table
+					to be updated with the address of the
+					imported function. */
+{
+	ulint		iIAT, iINT;
+	HMODULE		hmod;
+	PCImgThunkData	pitd;
+	FARPROC		fun = NULL;
+
+	/* Set up data used for the hook procs  */
+	InternalImgDelayDescr	idd = {
+				pidd->grAttrs,
+				PFromRva<LPCSTR>(pidd->rvaDLLName),
+				PFromRva<HMODULE*>(pidd->rvaHmod),
+				PFromRva<PImgThunkData>(pidd->rvaIAT),
+				PFromRva<PCImgThunkData>(pidd->rvaINT),
+				PFromRva<PCImgThunkData>(pidd->rvaBoundIAT),
+				PFromRva<PCImgThunkData>(pidd->rvaUnloadIAT),
+				pidd->dwTimeStamp
+	};
+
+	DelayLoadInfo		dli = {
+				sizeof(DelayLoadInfo),
+				pidd,
+				iat_entry,
+				idd.szName,
+				{0},
+				0,
+				0,
+				0
+	};
+
+	/* Check the Delay Load Attributes, log an error of invalid
+	parameter, which happens if the attributes in pidd are not
+	specified correctly. */
+	if ((idd.grAttrs & dlattrRva) == 0) {
+
+		sql_print_error("InnoDB: invalid parameter for delay loader.");
+		return(0);
+	}
+
+	hmod = *idd.phmod;
+
+	/* Calculate the index for the IAT entry in the import address table.
+	The INT entries are ordered the same as the IAT entries so the
+	calculation can be done on the IAT side. */
+	iIAT = (PCImgThunkData) iat_entry - idd.pIAT;
+	iINT = iIAT;
+
+	pitd = &(idd.pINT[iINT]);
+
+	dli.dlp.fImportByName = !IMAGE_SNAP_BY_ORDINAL(pitd->u1.Ordinal);
+
+	if (dli.dlp.fImportByName) {
+
+		dli.dlp.szProcName = (LPCSTR) (PFromRva<PIMAGE_IMPORT_BY_NAME>
+			((RVA) ((UINT_PTR) pitd->u1.AddressOfData))->Name);
+	} else {
+
+		dli.dlp.dwOrdinal = (ulint) IMAGE_ORDINAL(pitd->u1.Ordinal);
+	}
+
+	/* Now, load the mapfile, if it has not been done yet */
+	if (hmod == 0) {
+
+		hmod = wdl_get_mysqld_mapfile();
+	}
+
+	if (hmod == 0) {
+		/* LoadLibrary failed. */
+		PDelayLoadInfo	rgpdli[1] = {&dli};
+
+		dli.dwLastError = ::GetLastError();
+
+		sql_print_error(
+			"InnoDB: failed to load mysqld.map with error %d.",
+			dli.dwLastError);
+
+		return(0);
+	}
+
+	/* Store the library handle. */
+	idd.phmod = &hmod;
+
+	/* Go for the procedure now. */
+	dli.hmodCur = hmod;
+
+	if (pidd->rvaBoundIAT && pidd->dwTimeStamp) {
+
+		/* Bound imports exist, check the timestamp from the target
+		image */
+		PIMAGE_NT_HEADERS	pinh;
+
+		pinh = (PIMAGE_NT_HEADERS) ((byte*) hmod
+				+ ((PIMAGE_DOS_HEADER) hmod)->e_lfanew);
+
+		if (pinh->Signature == IMAGE_NT_SIGNATURE
+		    && pinh->FileHeader.TimeDateStamp == idd.dwTimeStamp
+		    && (DWORD) hmod == pinh->OptionalHeader.ImageBase) {
+
+			/* We have a decent address in the bound IAT. */
+			fun = (FARPROC) (UINT_PTR)
+					idd.pBoundIAT[iIAT].u1.Function;
+
+			if (fun) {
+
+				*iat_entry = fun;
+				return(fun);
+			}
+		}
+	}
+
+	fun = wdl_get_procaddr_from_map(hmod, dli.dlp.szProcName);
+
+	if (fun == 0) {
+
+		return(0);
+	}
+
+	*iat_entry = fun;
+	return(fun);
+}
+
+/***********************************************************************
+Unload a DLL that was delay loaded. This function is called by run-time. */
+extern "C"
+BOOL WINAPI
+__FUnloadDelayLoadedDLL2(
+/*=====================*/
+				/* out: TRUE is returned if the DLL is found
+				and the IAT matches the original one. */
+	LPCSTR	module_name)	/* in: DLL name */
+{
+	return(TRUE);
+}
+
+/******************************************************************
+Load all imports from a DLL that was specified with the /delayload linker
+option.
+Note: this function is called by run-time. So, it has to follow Windows call
+convention. */
+extern "C"
+HRESULT WINAPI
+__HrLoadAllImportsForDll(
+/*=====================*/
+				/* out: S_OK if the DLL matches, otherwise
+				ERROR_MOD_NOT_FOUND is returned. */
+	LPCSTR	module_name)	/* in: DLL name */
+{
+	PIMAGE_NT_HEADERS	img;
+	PCImgDelayDescr		pidd;
+	IMAGE_DATA_DIRECTORY*	image_data;
+	LPCSTR			current_module;
+	HRESULT			ret = ERROR_MOD_NOT_FOUND;
+	HMODULE			hmod = (HMODULE) &__ImageBase;
+
+	img = (PIMAGE_NT_HEADERS) ((byte*) hmod
+				   + ((PIMAGE_DOS_HEADER) hmod)->e_lfanew);
+	image_data =
+	 &img->OptionalHeader.DataDirectory[IMAGE_DIRECTORY_ENTRY_DELAY_IMPORT];
+
+	/* Scan the delay load IAT/INT for the DLL */
+	if (image_data->Size) {
+
+		pidd = PFromRva<PCImgDelayDescr>(image_data->VirtualAddress);
+
+		/* Check all of the listed DLLs we want to load. */
+		while (pidd->rvaDLLName) {
+
+			current_module = PFromRva<LPCSTR>(pidd->rvaDLLName);
+
+			if (stricmp(module_name, current_module) == 0) {
+
+				/* Found it, break out with pidd and
+				current_module set appropriately */
+				break;
+			}
+
+			/* To the next delay import descriptor */
+			pidd++;
+		}
+
+		if (pidd->rvaDLLName) {
+
+			/* Found a matching DLL, now process it. */
+			FARPROC*	iat_entry;
+			size_t		count;
+
+			iat_entry = PFromRva<FARPROC*>(pidd->rvaIAT);
+			count = wdl_import_count((PCImgThunkData) iat_entry);
+
+			/* now load all the imports from the DLL */
+			while (count > 0) {
+
+				/* No need to check the return value */
+				__delayLoadHelper2(pidd, iat_entry);
+				iat_entry++;
+				count--;
+			}
+
+			ret = S_OK;
+		}
+	}
+
+	return ret;
+}
+
+/******************************************************************
+The main function of a DLL */
+BOOL
+WINAPI
+DllMain(
+/*====*/
+					/* out: TRUE if the call succeeds */
+	HINSTANCE	hinstDLL,	/* in: handle to the DLL module */
+	DWORD		fdwReason,	/* Reason code that indicates why the
+					DLL entry-point function is being
+					called.*/
+	LPVOID		lpvReserved)	/* in: additional parameter based on
+					fdwReason */
+{
+	BOOL	success = TRUE;
+
+	switch (fdwReason) {
+
+	case DLL_PROCESS_ATTACH:
+		success = wdl_get_external_variables();
+		break;
+
+	case DLL_PROCESS_DETACH:
+		wdl_cleanup();
+		break;
+	}
+
+	return(success);
+}
+
+#ifndef DBUG_OFF
+/******************************************************************
+Process entry point to user function. It makes the call to _db_enter_
+in mysqld.exe. The DBUG functions are defined in my_dbug.h. */
+extern "C" UNIV_INTERN
+void
+_db_enter_(
+	const char*	_func_,		/* in: current function name */
+	const char*	_file_,		/* in: current file name */
+	uint		_line_,		/* in: current source line number */
+	const char**	_sfunc_,	/* out: previous _func_ */
+	const char**	_sfile_,	/* out: previous _file_ */
+	uint*		_slevel_,	/* out: previous nesting level */
+	char***		_sframep_)	/* out: previous frame pointer */
+{
+	if (wdl_db_enter_ != NULL) {
+
+		wdl_db_enter_(_func_, _file_, _line_, _sfunc_, _sfile_,
+			      _slevel_, _sframep_);
+	}
+}
+
+/******************************************************************
+Process exit from user function. It makes the call to _db_return_()
+in the server. */
+extern "C" UNIV_INTERN
+void
+_db_return_(
+	uint		_line_,		/* in: current source line number */
+	const char**	_sfunc_,	/* out: previous _func_ */
+	const char**	_sfile_,	/* out: previous _file_ */
+	uint*		_slevel_)	/* out: previous level */
+{
+	if (wdl_db_return_ != NULL) {
+
+		wdl_db_return_(_line_, _sfunc_, _sfile_, _slevel_);
+	}
+}
+
+/******************************************************************
+Log arguments for subsequent use. It makes the call to _db_pargs_()
+in the server. */
+extern "C" UNIV_INTERN
+void
+_db_pargs_(
+	uint		_line_,		/* in: current source line number */
+	const char*	keyword)	/* in: keyword for current macro */
+{
+	if (wdl_db_pargs_ != NULL) {
+
+		wdl_db_pargs_(_line_, keyword);
+	}
+}
+
+/******************************************************************
+Handle print of debug lines. It saves the text into a buffer first,
+then makes the call to _db_doprnt_() in the server. The text is
+truncated to the size of buffer. */
+extern "C" UNIV_INTERN
+void
+_db_doprnt_(
+	const char*	format,		/* in: the format string */
+	...)				/* in: list of arguments */
+{
+	va_list		argp;
+	char		buffer[512];
+
+	if (wdl_db_doprnt_ != NULL) {
+
+		va_start(argp, format);
+		/* it is ok to ignore the trunction. */
+		_vsnprintf(buffer, sizeof(buffer), format, argp);
+		wdl_db_doprnt_(buffer);
+		va_end(argp);
+	}
+}
+
+/******************************************************************
+Dump a string in hex. It makes the call to _db_dump_() in the server. */
+extern "C" UNIV_INTERN
+void
+_db_dump_(
+	uint			_line_,		/* in: current source line
+						number */
+	const char*		keyword,	/* in: keyword list */
+	const unsigned char*	memory,		/* in: memory to dump */
+	size_t			length)		/* in: bytes to dump */
+{
+	if (wdl_db_dump_ != NULL) {
+
+		wdl_db_dump_(_line_, keyword, memory, length);
+	}
+}
+
+#endif /* !DBUG_OFF */
+#endif /* defined (__WIN__) && defined (MYSQL_DYNAMIC_PLUGIN) */
diff --git a/storage/xtradb/ibuf/ibuf0ibuf.c b/storage/xtradb/ibuf/ibuf0ibuf.c
new file mode 100644
index 00000000000..ac678471312
--- /dev/null
+++ b/storage/xtradb/ibuf/ibuf0ibuf.c
@@ -0,0 +1,3598 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Insert buffer
+
+Created 7/19/1997 Heikki Tuuri
+*******************************************************/
+
+#include "ibuf0ibuf.h"
+
+#ifdef UNIV_NONINL
+#include "ibuf0ibuf.ic"
+#endif
+
+#include "buf0buf.h"
+#include "buf0rea.h"
+#include "fsp0fsp.h"
+#include "trx0sys.h"
+#include "fil0fil.h"
+#include "thr0loc.h"
+#include "rem0rec.h"
+#include "btr0cur.h"
+#include "btr0pcur.h"
+#include "btr0btr.h"
+#include "sync0sync.h"
+#include "dict0boot.h"
+#include "fut0lst.h"
+#include "lock0lock.h"
+#include "log0recv.h"
+#include "que0que.h"
+
+/*	STRUCTURE OF AN INSERT BUFFER RECORD
+
+In versions < 4.1.x:
+
+1. The first field is the page number.
+2. The second field is an array which stores type info for each subsequent
+   field. We store the information which affects the ordering of records, and
+   also the physical storage size of an SQL NULL value. E.g., for CHAR(10) it
+   is 10 bytes.
+3. Next we have the fields of the actual index record.
+
+In versions >= 4.1.x:
+
+Note that contary to what we planned in the 1990's, there will only be one
+insert buffer tree, and that is in the system tablespace of InnoDB.
+
+1. The first field is the space id.
+2. The second field is a one-byte marker (0) which differentiates records from
+   the < 4.1.x storage format.
+3. The third field is the page number.
+4. The fourth field contains the type info, where we have also added 2 bytes to
+   store the charset. In the compressed table format of 5.0.x we must add more
+   information here so that we can build a dummy 'index' struct which 5.0.x
+   can use in the binary search on the index page in the ibuf merge phase.
+5. The rest of the fields contain the fields of the actual index record.
+
+In versions >= 5.0.3:
+
+The first byte of the fourth field is an additional marker (0) if the record
+is in the compact format.  The presence of this marker can be detected by
+looking at the length of the field modulo DATA_NEW_ORDER_NULL_TYPE_BUF_SIZE.
+
+The high-order bit of the character set field in the type info is the
+"nullable" flag for the field. */
+
+
+/*	PREVENTING DEADLOCKS IN THE INSERT BUFFER SYSTEM
+
+If an OS thread performs any operation that brings in disk pages from
+non-system tablespaces into the buffer pool, or creates such a page there,
+then the operation may have as a side effect an insert buffer index tree
+compression. Thus, the tree latch of the insert buffer tree may be acquired
+in the x-mode, and also the file space latch of the system tablespace may
+be acquired in the x-mode.
+
+Also, an insert to an index in a non-system tablespace can have the same
+effect. How do we know this cannot lead to a deadlock of OS threads? There
+is a problem with the i\o-handler threads: they break the latching order
+because they own x-latches to pages which are on a lower level than the
+insert buffer tree latch, its page latches, and the tablespace latch an
+insert buffer operation can reserve.
+
+The solution is the following: Let all the tree and page latches connected
+with the insert buffer be later in the latching order than the fsp latch and
+fsp page latches.
+
+Insert buffer pages must be such that the insert buffer is never invoked
+when these pages are accessed as this would result in a recursion violating
+the latching order. We let a special i/o-handler thread take care of i/o to
+the insert buffer pages and the ibuf bitmap pages, as well as the fsp bitmap
+pages and the first inode page, which contains the inode of the ibuf tree: let
+us call all these ibuf pages. To prevent deadlocks, we do not let a read-ahead
+access both non-ibuf and ibuf pages.
+
+Then an i/o-handler for the insert buffer never needs to access recursively the
+insert buffer tree and thus obeys the latching order. On the other hand, other
+i/o-handlers for other tablespaces may require access to the insert buffer,
+but because all kinds of latches they need to access there are later in the
+latching order, no violation of the latching order occurs in this case,
+either.
+
+A problem is how to grow and contract an insert buffer tree. As it is later
+in the latching order than the fsp management, we have to reserve the fsp
+latch first, before adding or removing pages from the insert buffer tree.
+We let the insert buffer tree have its own file space management: a free
+list of pages linked to the tree root. To prevent recursive using of the
+insert buffer when adding pages to the tree, we must first load these pages
+to memory, obtaining a latch on them, and only after that add them to the
+free list of the insert buffer tree. More difficult is removing of pages
+from the free list. If there is an excess of pages in the free list of the
+ibuf tree, they might be needed if some thread reserves the fsp latch,
+intending to allocate more file space. So we do the following: if a thread
+reserves the fsp latch, we check the writer count field of the latch. If
+this field has value 1, it means that the thread did not own the latch
+before entering the fsp system, and the mtr of the thread contains no
+modifications to the fsp pages. Now we are free to reserve the ibuf latch,
+and check if there is an excess of pages in the free list. We can then, in a
+separate mini-transaction, take them out of the free list and free them to
+the fsp system.
+
+To avoid deadlocks in the ibuf system, we divide file pages into three levels:
+
+(1) non-ibuf pages,
+(2) ibuf tree pages and the pages in the ibuf tree free list, and
+(3) ibuf bitmap pages.
+
+No OS thread is allowed to access higher level pages if it has latches to
+lower level pages; even if the thread owns a B-tree latch it must not access
+the B-tree non-leaf pages if it has latches on lower level pages. Read-ahead
+is only allowed for level 1 and 2 pages. Dedicated i/o-handler threads handle
+exclusively level 1 i/o. A dedicated i/o handler thread handles exclusively
+level 2 i/o. However, if an OS thread does the i/o handling for itself, i.e.,
+it uses synchronous aio, it can access any pages, as long as it obeys the
+access order rules. */
+
+/* Buffer pool size per the maximum insert buffer size */
+#define IBUF_POOL_SIZE_PER_MAX_SIZE	2
+
+/* Table name for the insert buffer. */
+#define IBUF_TABLE_NAME		"SYS_IBUF_TABLE"
+
+/** Operations that can currently be buffered. */
+UNIV_INTERN ibuf_use_t	ibuf_use		= IBUF_USE_INSERT;
+
+/** The insert buffer control structure */
+UNIV_INTERN ibuf_t*	ibuf			= NULL;
+
+UNIV_INTERN ulint	ibuf_flush_count	= 0;
+
+#ifdef UNIV_IBUF_COUNT_DEBUG
+/* Dimensions for the ibuf_count array */
+#define IBUF_COUNT_N_SPACES	4
+#define IBUF_COUNT_N_PAGES	130000
+
+/* Buffered entry counts for file pages, used in debugging */
+static ulint	ibuf_counts[IBUF_COUNT_N_SPACES][IBUF_COUNT_N_PAGES];
+
+/**********************************************************************
+Checks that the indexes to ibuf_counts[][] are within limits. */
+UNIV_INLINE
+void
+ibuf_count_check(
+/*=============*/
+	ulint	space_id,	/* in: space identifier */
+	ulint	page_no)	/* in: page number */
+{
+	if (space_id < IBUF_COUNT_N_SPACES && page_no < IBUF_COUNT_N_PAGES) {
+		return;
+	}
+
+	fprintf(stderr,
+		"InnoDB: UNIV_IBUF_COUNT_DEBUG limits space_id and page_no\n"
+		"InnoDB: and breaks crash recovery.\n"
+		"InnoDB: space_id=%lu, should be 0<=space_id<%lu\n"
+		"InnoDB: page_no=%lu, should be 0<=page_no<%lu\n",
+		(ulint) space_id, (ulint) IBUF_COUNT_N_SPACES,
+		(ulint) page_no, (ulint) IBUF_COUNT_N_PAGES);
+	ut_error;
+}
+#endif
+
+/* The start address for an insert buffer bitmap page bitmap */
+#define IBUF_BITMAP		PAGE_DATA
+
+/* Offsets in bits for the bits describing a single page in the bitmap */
+#define	IBUF_BITMAP_FREE	0
+#define IBUF_BITMAP_BUFFERED	2
+#define IBUF_BITMAP_IBUF	3	/* TRUE if page is a part of the ibuf
+					tree, excluding the root page, or is
+					in the free list of the ibuf */
+
+/* Number of bits describing a single page */
+#define IBUF_BITS_PER_PAGE	4
+#if IBUF_BITS_PER_PAGE % 2
+# error "IBUF_BITS_PER_PAGE must be an even number!"
+#endif
+
+/* The mutex used to block pessimistic inserts to ibuf trees */
+static mutex_t	ibuf_pessimistic_insert_mutex;
+
+/* The mutex protecting the insert buffer structs */
+static mutex_t	ibuf_mutex;
+
+/* The mutex protecting the insert buffer bitmaps */
+static mutex_t	ibuf_bitmap_mutex;
+
+/* The area in pages from which contract looks for page numbers for merge */
+#define	IBUF_MERGE_AREA			8
+
+/* Inside the merge area, pages which have at most 1 per this number less
+buffered entries compared to maximum volume that can buffered for a single
+page are merged along with the page whose buffer became full */
+#define IBUF_MERGE_THRESHOLD		4
+
+/* In ibuf_contract at most this number of pages is read to memory in one
+batch, in order to merge the entries for them in the insert buffer */
+#define	IBUF_MAX_N_PAGES_MERGED		IBUF_MERGE_AREA
+
+/* If the combined size of the ibuf trees exceeds ibuf->max_size by this
+many pages, we start to contract it in connection to inserts there, using
+non-synchronous contract */
+#define IBUF_CONTRACT_ON_INSERT_NON_SYNC	0
+
+/* Same as above, but use synchronous contract */
+#define IBUF_CONTRACT_ON_INSERT_SYNC		5
+
+/* Same as above, but no insert is done, only contract is called */
+#define IBUF_CONTRACT_DO_NOT_INSERT		10
+
+/* TODO: how to cope with drop table if there are records in the insert
+buffer for the indexes of the table? Is there actually any problem,
+because ibuf merge is done to a page when it is read in, and it is
+still physically like the index page even if the index would have been
+dropped! So, there seems to be no problem. */
+
+/**********************************************************************
+Sets the flag in the current OS thread local storage denoting that it is
+inside an insert buffer routine. */
+UNIV_INLINE
+void
+ibuf_enter(void)
+/*============*/
+{
+	ibool*	ptr;
+
+	ptr = thr_local_get_in_ibuf_field();
+
+	ut_ad(*ptr == FALSE);
+
+	*ptr = TRUE;
+}
+
+/**********************************************************************
+Sets the flag in the current OS thread local storage denoting that it is
+exiting an insert buffer routine. */
+UNIV_INLINE
+void
+ibuf_exit(void)
+/*===========*/
+{
+	ibool*	ptr;
+
+	ptr = thr_local_get_in_ibuf_field();
+
+	ut_ad(*ptr == TRUE);
+
+	*ptr = FALSE;
+}
+
+/**********************************************************************
+Returns TRUE if the current OS thread is performing an insert buffer
+routine. */
+UNIV_INTERN
+ibool
+ibuf_inside(void)
+/*=============*/
+		/* out: TRUE if inside an insert buffer routine: for instance,
+		a read-ahead of non-ibuf pages is then forbidden */
+{
+	return(*thr_local_get_in_ibuf_field());
+}
+
+/**********************************************************************
+Gets the ibuf header page and x-latches it. */
+static
+page_t*
+ibuf_header_page_get(
+/*=================*/
+			/* out: insert buffer header page */
+	mtr_t*	mtr)	/* in: mtr */
+{
+	buf_block_t*	block;
+
+	ut_ad(!ibuf_inside());
+
+	block = buf_page_get(
+		IBUF_SPACE_ID, 0, FSP_IBUF_HEADER_PAGE_NO, RW_X_LATCH, mtr);
+	buf_block_dbg_add_level(block, SYNC_IBUF_HEADER);
+
+	return(buf_block_get_frame(block));
+}
+
+/**********************************************************************
+Gets the root page and x-latches it. */
+static
+page_t*
+ibuf_tree_root_get(
+/*===============*/
+				/* out: insert buffer tree root page */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	buf_block_t*	block;
+
+	ut_ad(ibuf_inside());
+
+	mtr_x_lock(dict_index_get_lock(ibuf->index), mtr);
+
+	block = buf_page_get(
+		IBUF_SPACE_ID, 0, FSP_IBUF_TREE_ROOT_PAGE_NO, RW_X_LATCH, mtr);
+
+	buf_block_dbg_add_level(block, SYNC_TREE_NODE);
+
+	return(buf_block_get_frame(block));
+}
+
+#ifdef UNIV_IBUF_COUNT_DEBUG
+/**********************************************************************
+Gets the ibuf count for a given page. */
+UNIV_INTERN
+ulint
+ibuf_count_get(
+/*===========*/
+			/* out: number of entries in the insert buffer
+			currently buffered for this page */
+	ulint	space,	/* in: space id */
+	ulint	page_no)/* in: page number */
+{
+	ibuf_count_check(space, page_no);
+
+	return(ibuf_counts[space][page_no]);
+}
+
+/**********************************************************************
+Sets the ibuf count for a given page. */
+static
+void
+ibuf_count_set(
+/*===========*/
+	ulint	space,	/* in: space id */
+	ulint	page_no,/* in: page number */
+	ulint	val)	/* in: value to set */
+{
+	ibuf_count_check(space, page_no);
+	ut_a(val < UNIV_PAGE_SIZE);
+
+	ibuf_counts[space][page_no] = val;
+}
+#endif
+
+/**********************************************************************
+Updates the size information of the ibuf, assuming the segment size has not
+changed. */
+static
+void
+ibuf_size_update(
+/*=============*/
+	const page_t*	root,	/* in: ibuf tree root */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	ut_ad(mutex_own(&ibuf_mutex));
+
+	ibuf->free_list_len = flst_get_len(root + PAGE_HEADER
+					   + PAGE_BTR_IBUF_FREE_LIST, mtr);
+
+	ibuf->height = 1 + btr_page_get_level(root, mtr);
+
+	/* the '1 +' is the ibuf header page */
+	ibuf->size = ibuf->seg_size - (1 + ibuf->free_list_len);
+
+	ibuf->empty = page_get_n_recs(root) == 0;
+}
+
+/**********************************************************************
+Creates the insert buffer data structure at a database startup and initializes
+the data structures for the insert buffer. */
+UNIV_INTERN
+void
+ibuf_init_at_db_start(void)
+/*=======================*/
+{
+	page_t*		root;
+	mtr_t		mtr;
+	dict_table_t*	table;
+	mem_heap_t*	heap;
+	dict_index_t*	index;
+	ulint		n_used;
+	page_t*		header_page;
+	ulint		error;
+
+	ibuf = mem_alloc(sizeof(ibuf_t));
+
+	memset(ibuf, 0, sizeof(*ibuf));
+
+	/* Note that also a pessimistic delete can sometimes make a B-tree
+	grow in size, as the references on the upper levels of the tree can
+	change */
+
+        /* The default for ibuf_max_size is calculated from the requested
+           buffer pool size srv_buf_pool_size, not the actual size as returned
+           by buf_pool_get_curr_size(). The latter can differ from the former
+           by one page due to alignment requirements, and we do not want a
+           user-visible variable like INNODB_IBUF_MAX_SIZE to vary at random. */
+	ibuf->max_size = ut_min( srv_buf_pool_size / UNIV_PAGE_SIZE
+		/ IBUF_POOL_SIZE_PER_MAX_SIZE, (ulint) srv_ibuf_max_size / UNIV_PAGE_SIZE);
+
+	srv_ibuf_max_size = (long long) ibuf->max_size * UNIV_PAGE_SIZE;
+
+	mutex_create(&ibuf_pessimistic_insert_mutex,
+		     SYNC_IBUF_PESS_INSERT_MUTEX);
+
+	mutex_create(&ibuf_mutex, SYNC_IBUF_MUTEX);
+
+	mutex_create(&ibuf_bitmap_mutex, SYNC_IBUF_BITMAP_MUTEX);
+
+	mtr_start(&mtr);
+
+	mutex_enter(&ibuf_mutex);
+
+	mtr_x_lock(fil_space_get_latch(IBUF_SPACE_ID, NULL), &mtr);
+
+	header_page = ibuf_header_page_get(&mtr);
+
+	fseg_n_reserved_pages(header_page + IBUF_HEADER + IBUF_TREE_SEG_HEADER,
+			      &n_used, &mtr);
+	ibuf_enter();
+
+	ut_ad(n_used >= 2);
+
+	ibuf->seg_size = n_used;
+
+	{
+		buf_block_t*	block;
+
+		block = buf_page_get(
+			IBUF_SPACE_ID, 0, FSP_IBUF_TREE_ROOT_PAGE_NO,
+			RW_X_LATCH, &mtr);
+		buf_block_dbg_add_level(block, SYNC_TREE_NODE);
+
+		root = buf_block_get_frame(block);
+	}
+
+	ibuf_size_update(root, &mtr);
+	mutex_exit(&ibuf_mutex);
+
+	mtr_commit(&mtr);
+
+	ibuf_exit();
+
+	heap = mem_heap_create(450);
+
+	/* Use old-style record format for the insert buffer. */
+	table = dict_mem_table_create(IBUF_TABLE_NAME, IBUF_SPACE_ID, 1, 0);
+
+	dict_mem_table_add_col(table, heap, "DUMMY_COLUMN", DATA_BINARY, 0, 0);
+
+	table->id = ut_dulint_add(DICT_IBUF_ID_MIN, IBUF_SPACE_ID);
+
+	dict_table_add_to_cache(table, heap);
+	mem_heap_free(heap);
+
+	index = dict_mem_index_create(
+		IBUF_TABLE_NAME, "CLUST_IND",
+		IBUF_SPACE_ID, DICT_CLUSTERED | DICT_UNIVERSAL | DICT_IBUF, 1);
+
+	dict_mem_index_add_field(index, "DUMMY_COLUMN", 0);
+
+	index->id = ut_dulint_add(DICT_IBUF_ID_MIN, IBUF_SPACE_ID);
+
+	error = dict_index_add_to_cache(table, index,
+					FSP_IBUF_TREE_ROOT_PAGE_NO, FALSE);
+	ut_a(error == DB_SUCCESS);
+
+	ibuf->index = dict_table_get_first_index(table);
+}
+
+/*************************************************************************
+Initializes an ibuf bitmap page. */
+UNIV_INTERN
+void
+ibuf_bitmap_page_init(
+/*==================*/
+	buf_block_t*	block,	/* in: bitmap page */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	page_t*	page;
+	ulint	byte_offset;
+	ulint	zip_size = buf_block_get_zip_size(block);
+
+	ut_a(ut_is_2pow(zip_size));
+
+	page = buf_block_get_frame(block);
+	fil_page_set_type(page, FIL_PAGE_IBUF_BITMAP);
+
+	/* Write all zeros to the bitmap */
+
+	if (!zip_size) {
+		byte_offset = UT_BITS_IN_BYTES(UNIV_PAGE_SIZE
+					       * IBUF_BITS_PER_PAGE);
+	} else {
+		byte_offset = UT_BITS_IN_BYTES(zip_size * IBUF_BITS_PER_PAGE);
+	}
+
+	memset(page + IBUF_BITMAP, 0, byte_offset);
+
+	/* The remaining area (up to the page trailer) is uninitialized. */
+
+	mlog_write_initial_log_record(page, MLOG_IBUF_BITMAP_INIT, mtr);
+}
+
+/*************************************************************************
+Parses a redo log record of an ibuf bitmap page init. */
+UNIV_INTERN
+byte*
+ibuf_parse_bitmap_init(
+/*===================*/
+				/* out: end of log record or NULL */
+	byte*		ptr,	/* in: buffer */
+	byte*		end_ptr __attribute__((unused)), /* in: buffer end */
+	buf_block_t*	block,	/* in: block or NULL */
+	mtr_t*		mtr)	/* in: mtr or NULL */
+{
+	ut_ad(ptr && end_ptr);
+
+	if (block) {
+		ibuf_bitmap_page_init(block, mtr);
+	}
+
+	return(ptr);
+}
+
+/************************************************************************
+Gets the desired bits for a given page from a bitmap page. */
+UNIV_INLINE
+ulint
+ibuf_bitmap_page_get_bits(
+/*======================*/
+				/* out: value of bits */
+	const page_t*	page,	/* in: bitmap page */
+	ulint		page_no,/* in: page whose bits to get */
+	ulint		zip_size,/* in: compressed page size in bytes;
+				0 for uncompressed pages */
+	ulint		bit,	/* in: IBUF_BITMAP_FREE,
+				IBUF_BITMAP_BUFFERED, ... */
+	mtr_t*		mtr __attribute__((unused)))
+				/* in: mtr containing an
+				x-latch to the bitmap page */
+{
+	ulint	byte_offset;
+	ulint	bit_offset;
+	ulint	map_byte;
+	ulint	value;
+
+	ut_ad(bit < IBUF_BITS_PER_PAGE);
+#if IBUF_BITS_PER_PAGE % 2
+# error "IBUF_BITS_PER_PAGE % 2 != 0"
+#endif
+	ut_ad(ut_is_2pow(zip_size));
+	ut_ad(mtr_memo_contains_page(mtr, page, MTR_MEMO_PAGE_X_FIX));
+
+	if (!zip_size) {
+		bit_offset = (page_no % UNIV_PAGE_SIZE) * IBUF_BITS_PER_PAGE
+			+ bit;
+	} else {
+		bit_offset = (page_no & (zip_size - 1)) * IBUF_BITS_PER_PAGE
+			+ bit;
+	}
+
+	byte_offset = bit_offset / 8;
+	bit_offset = bit_offset % 8;
+
+	ut_ad(byte_offset + IBUF_BITMAP < UNIV_PAGE_SIZE);
+
+	map_byte = mach_read_from_1(page + IBUF_BITMAP + byte_offset);
+
+	value = ut_bit_get_nth(map_byte, bit_offset);
+
+	if (bit == IBUF_BITMAP_FREE) {
+		ut_ad(bit_offset + 1 < 8);
+
+		value = value * 2 + ut_bit_get_nth(map_byte, bit_offset + 1);
+	}
+
+	return(value);
+}
+
+/************************************************************************
+Sets the desired bit for a given page in a bitmap page. */
+static
+void
+ibuf_bitmap_page_set_bits(
+/*======================*/
+	page_t*	page,	/* in: bitmap page */
+	ulint	page_no,/* in: page whose bits to set */
+	ulint	zip_size,/* in: compressed page size in bytes;
+			0 for uncompressed pages */
+	ulint	bit,	/* in: IBUF_BITMAP_FREE, IBUF_BITMAP_BUFFERED, ... */
+	ulint	val,	/* in: value to set */
+	mtr_t*	mtr)	/* in: mtr containing an x-latch to the bitmap page */
+{
+	ulint	byte_offset;
+	ulint	bit_offset;
+	ulint	map_byte;
+
+	ut_ad(bit < IBUF_BITS_PER_PAGE);
+#if IBUF_BITS_PER_PAGE % 2
+# error "IBUF_BITS_PER_PAGE % 2 != 0"
+#endif
+	ut_ad(ut_is_2pow(zip_size));
+	ut_ad(mtr_memo_contains_page(mtr, page, MTR_MEMO_PAGE_X_FIX));
+#ifdef UNIV_IBUF_COUNT_DEBUG
+	ut_a((bit != IBUF_BITMAP_BUFFERED) || (val != FALSE)
+	     || (0 == ibuf_count_get(page_get_space_id(page),
+				     page_no)));
+#endif
+	if (!zip_size) {
+		bit_offset = (page_no % UNIV_PAGE_SIZE) * IBUF_BITS_PER_PAGE
+			+ bit;
+	} else {
+		bit_offset = (page_no & (zip_size - 1)) * IBUF_BITS_PER_PAGE
+			+ bit;
+	}
+
+	byte_offset = bit_offset / 8;
+	bit_offset = bit_offset % 8;
+
+	ut_ad(byte_offset + IBUF_BITMAP < UNIV_PAGE_SIZE);
+
+	map_byte = mach_read_from_1(page + IBUF_BITMAP + byte_offset);
+
+	if (bit == IBUF_BITMAP_FREE) {
+		ut_ad(bit_offset + 1 < 8);
+		ut_ad(val <= 3);
+
+		map_byte = ut_bit_set_nth(map_byte, bit_offset, val / 2);
+		map_byte = ut_bit_set_nth(map_byte, bit_offset + 1, val % 2);
+	} else {
+		ut_ad(val <= 1);
+		map_byte = ut_bit_set_nth(map_byte, bit_offset, val);
+	}
+
+	mlog_write_ulint(page + IBUF_BITMAP + byte_offset, map_byte,
+			 MLOG_1BYTE, mtr);
+}
+
+/************************************************************************
+Calculates the bitmap page number for a given page number. */
+UNIV_INLINE
+ulint
+ibuf_bitmap_page_no_calc(
+/*=====================*/
+				/* out: the bitmap page number where
+				the file page is mapped */
+	ulint	zip_size,	/* in: compressed page size in bytes;
+				0 for uncompressed pages */
+	ulint	page_no)	/* in: tablespace page number */
+{
+	ut_ad(ut_is_2pow(zip_size));
+
+	if (!zip_size) {
+		return(FSP_IBUF_BITMAP_OFFSET
+		       + (page_no & ~(UNIV_PAGE_SIZE - 1)));
+	} else {
+		return(FSP_IBUF_BITMAP_OFFSET
+		       + (page_no & ~(zip_size - 1)));
+	}
+}
+
+/************************************************************************
+Gets the ibuf bitmap page where the bits describing a given file page are
+stored. */
+static
+page_t*
+ibuf_bitmap_get_map_page(
+/*=====================*/
+			/* out: bitmap page where the file page is mapped,
+			that is, the bitmap page containing the descriptor
+			bits for the file page; the bitmap page is
+			x-latched */
+	ulint	space,	/* in: space id of the file page */
+	ulint	page_no,/* in: page number of the file page */
+	ulint	zip_size,/* in: compressed page size in bytes;
+			0 for uncompressed pages */
+	mtr_t*	mtr)	/* in: mtr */
+{
+	buf_block_t*	block;
+
+	block = buf_page_get(space, zip_size,
+			     ibuf_bitmap_page_no_calc(zip_size, page_no),
+			     RW_X_LATCH, mtr);
+	buf_block_dbg_add_level(block, SYNC_IBUF_BITMAP);
+
+	return(buf_block_get_frame(block));
+}
+
+/****************************************************************************
+Sets the free bits of the page in the ibuf bitmap. This is done in a separate
+mini-transaction, hence this operation does not restrict further work to only
+ibuf bitmap operations, which would result if the latch to the bitmap page
+were kept. */
+UNIV_INLINE
+void
+ibuf_set_free_bits_low(
+/*===================*/
+	ulint			zip_size,/* in: compressed page size in bytes;
+					0 for uncompressed pages */
+	const buf_block_t*	block,	/* in: index page; free bits are set if
+					the index is non-clustered and page
+					level is 0 */
+	ulint			val,	/* in: value to set: < 4 */
+	mtr_t*			mtr)	/* in/out: mtr */
+{
+	page_t*	bitmap_page;
+	ulint	space;
+	ulint	page_no;
+
+	if (!page_is_leaf(buf_block_get_frame(block))) {
+
+		return;
+	}
+
+	space = buf_block_get_space(block);
+	page_no = buf_block_get_page_no(block);
+	bitmap_page = ibuf_bitmap_get_map_page(space, page_no, zip_size, mtr);
+#ifdef UNIV_IBUF_DEBUG
+# if 0
+	fprintf(stderr,
+		"Setting space %lu page %lu free bits to %lu should be %lu\n",
+		space, page_no, val,
+		ibuf_index_page_calc_free(zip_size, block));
+# endif
+
+	ut_a(val <= ibuf_index_page_calc_free(zip_size, block));
+#endif /* UNIV_IBUF_DEBUG */
+	ibuf_bitmap_page_set_bits(bitmap_page, page_no, zip_size,
+				  IBUF_BITMAP_FREE, val, mtr);
+}
+
+/****************************************************************************
+Sets the free bit of the page in the ibuf bitmap. This is done in a separate
+mini-transaction, hence this operation does not restrict further work to only
+ibuf bitmap operations, which would result if the latch to the bitmap page
+were kept. */
+UNIV_INTERN
+void
+ibuf_set_free_bits_func(
+/*====================*/
+	buf_block_t*	block,	/* in: index page of a non-clustered index;
+				free bit is reset if page level is 0 */
+#ifdef UNIV_IBUF_DEBUG
+	ulint		max_val,/* in: ULINT_UNDEFINED or a maximum
+				value which the bits must have before
+				setting; this is for debugging */
+#endif /* UNIV_IBUF_DEBUG */
+	ulint		val)	/* in: value to set: < 4 */
+{
+	mtr_t	mtr;
+	page_t*	page;
+	page_t*	bitmap_page;
+	ulint	space;
+	ulint	page_no;
+	ulint	zip_size;
+
+	page = buf_block_get_frame(block);
+
+	if (!page_is_leaf(page)) {
+
+		return;
+	}
+
+	mtr_start(&mtr);
+
+	space = buf_block_get_space(block);
+	page_no = buf_block_get_page_no(block);
+	zip_size = buf_block_get_zip_size(block);
+	bitmap_page = ibuf_bitmap_get_map_page(space, page_no, zip_size, &mtr);
+
+#ifdef UNIV_IBUF_DEBUG
+	if (max_val != ULINT_UNDEFINED) {
+		ulint	old_val;
+
+		old_val = ibuf_bitmap_page_get_bits(
+			bitmap_page, page_no, zip_size,
+			IBUF_BITMAP_FREE, &mtr);
+# if 0
+		if (old_val != max_val) {
+			fprintf(stderr,
+				"Ibuf: page %lu old val %lu max val %lu\n",
+				page_get_page_no(page),
+				old_val, max_val);
+		}
+# endif
+
+		ut_a(old_val <= max_val);
+	}
+# if 0
+	fprintf(stderr, "Setting page no %lu free bits to %lu should be %lu\n",
+		page_get_page_no(page), val,
+		ibuf_index_page_calc_free(zip_size, block));
+# endif
+
+	ut_a(val <= ibuf_index_page_calc_free(zip_size, block));
+#endif /* UNIV_IBUF_DEBUG */
+	ibuf_bitmap_page_set_bits(bitmap_page, page_no, zip_size,
+				  IBUF_BITMAP_FREE, val, &mtr);
+	mtr_commit(&mtr);
+}
+
+/****************************************************************************
+Resets the free bits of the page in the ibuf bitmap. This is done in a
+separate mini-transaction, hence this operation does not restrict
+further work to only ibuf bitmap operations, which would result if the
+latch to the bitmap page were kept.  NOTE: The free bits in the insert
+buffer bitmap must never exceed the free space on a page.  It is safe
+to decrement or reset the bits in the bitmap in a mini-transaction
+that is committed before the mini-transaction that affects the free
+space. */
+UNIV_INTERN
+void
+ibuf_reset_free_bits(
+/*=================*/
+	buf_block_t*	block)	/* in: index page; free bits are set to 0
+				if the index is a non-clustered
+				non-unique, and page level is 0 */
+{
+	ibuf_set_free_bits(block, 0, ULINT_UNDEFINED);
+}
+
+/**************************************************************************
+Updates the free bits for an uncompressed page to reflect the present
+state.  Does this in the mtr given, which means that the latching
+order rules virtually prevent any further operations for this OS
+thread until mtr is committed.  NOTE: The free bits in the insert
+buffer bitmap must never exceed the free space on a page.  It is safe
+to set the free bits in the same mini-transaction that updated the
+page. */
+UNIV_INTERN
+void
+ibuf_update_free_bits_low(
+/*======================*/
+	const buf_block_t*	block,		/* in: index page */
+	ulint			max_ins_size,	/* in: value of
+						maximum insert size
+						with reorganize before
+						the latest operation
+						performed to the page */
+	mtr_t*			mtr)		/* in/out: mtr */
+{
+	ulint	before;
+	ulint	after;
+
+	ut_a(!buf_block_get_page_zip(block));
+
+	before = ibuf_index_page_calc_free_bits(0, max_ins_size);
+
+	after = ibuf_index_page_calc_free(0, block);
+
+	/* This approach cannot be used on compressed pages, since the
+	computed value of "before" often does not match the current
+	state of the bitmap.  This is because the free space may
+	increase or decrease when a compressed page is reorganized. */
+	if (before != after) {
+		ibuf_set_free_bits_low(0, block, after, mtr);
+	}
+}
+
+/**************************************************************************
+Updates the free bits for a compressed page to reflect the present
+state.  Does this in the mtr given, which means that the latching
+order rules virtually prevent any further operations for this OS
+thread until mtr is committed.  NOTE: The free bits in the insert
+buffer bitmap must never exceed the free space on a page.  It is safe
+to set the free bits in the same mini-transaction that updated the
+page. */
+UNIV_INTERN
+void
+ibuf_update_free_bits_zip(
+/*======================*/
+	buf_block_t*	block,	/* in/out: index page */
+	mtr_t*		mtr)	/* in/out: mtr */
+{
+	page_t*	bitmap_page;
+	ulint	space;
+	ulint	page_no;
+	ulint	zip_size;
+	ulint	after;
+
+	space = buf_block_get_space(block);
+	page_no = buf_block_get_page_no(block);
+	zip_size = buf_block_get_zip_size(block);
+
+	ut_a(page_is_leaf(buf_block_get_frame(block)));
+	ut_a(zip_size);
+
+	bitmap_page = ibuf_bitmap_get_map_page(space, page_no, zip_size, mtr);
+
+	after = ibuf_index_page_calc_free_zip(zip_size, block);
+
+	if (after == 0) {
+		/* We move the page to the front of the buffer pool LRU list:
+		the purpose of this is to prevent those pages to which we
+		cannot make inserts using the insert buffer from slipping
+		out of the buffer pool */
+
+		buf_page_make_young(&block->page);
+	}
+
+	ibuf_bitmap_page_set_bits(bitmap_page, page_no, zip_size,
+				  IBUF_BITMAP_FREE, after, mtr);
+}
+
+/**************************************************************************
+Updates the free bits for the two pages to reflect the present state.
+Does this in the mtr given, which means that the latching order rules
+virtually prevent any further operations until mtr is committed.
+NOTE: The free bits in the insert buffer bitmap must never exceed the
+free space on a page.  It is safe to set the free bits in the same
+mini-transaction that updated the pages. */
+UNIV_INTERN
+void
+ibuf_update_free_bits_for_two_pages_low(
+/*====================================*/
+	ulint		zip_size,/* in: compressed page size in bytes;
+				0 for uncompressed pages */
+	buf_block_t*	block1,	/* in: index page */
+	buf_block_t*	block2,	/* in: index page */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	ulint	state;
+
+	/* As we have to x-latch two random bitmap pages, we have to acquire
+	the bitmap mutex to prevent a deadlock with a similar operation
+	performed by another OS thread. */
+
+	mutex_enter(&ibuf_bitmap_mutex);
+
+	state = ibuf_index_page_calc_free(zip_size, block1);
+
+	ibuf_set_free_bits_low(zip_size, block1, state, mtr);
+
+	state = ibuf_index_page_calc_free(zip_size, block2);
+
+	ibuf_set_free_bits_low(zip_size, block2, state, mtr);
+
+	mutex_exit(&ibuf_bitmap_mutex);
+}
+
+/**************************************************************************
+Returns TRUE if the page is one of the fixed address ibuf pages. */
+UNIV_INLINE
+ibool
+ibuf_fixed_addr_page(
+/*=================*/
+			/* out: TRUE if a fixed address ibuf i/o page */
+	ulint	space,	/* in: space id */
+	ulint	zip_size,/* in: compressed page size in bytes;
+			0 for uncompressed pages */
+	ulint	page_no)/* in: page number */
+{
+	return((space == IBUF_SPACE_ID && page_no == IBUF_TREE_ROOT_PAGE_NO)
+	       || ibuf_bitmap_page(zip_size, page_no));
+}
+
+/***************************************************************************
+Checks if a page is a level 2 or 3 page in the ibuf hierarchy of pages.
+Must not be called when recv_no_ibuf_operations==TRUE. */
+UNIV_INTERN
+ibool
+ibuf_page(
+/*======*/
+			/* out: TRUE if level 2 or level 3 page */
+	ulint	space,	/* in: space id */
+	ulint	zip_size,/* in: compressed page size in bytes, or 0 */
+	ulint	page_no,/* in: page number */
+	mtr_t*	mtr)	/* in: mtr which will contain an x-latch to the
+			bitmap page if the page is not one of the fixed
+			address ibuf pages, or NULL, in which case a new
+			transaction is created. */
+{
+	ibool	ret;
+	mtr_t	local_mtr;
+	page_t*	bitmap_page;
+
+	ut_ad(!recv_no_ibuf_operations);
+
+	if (ibuf_fixed_addr_page(space, zip_size, page_no)) {
+
+		return(TRUE);
+	} else if (space != IBUF_SPACE_ID) {
+
+		return(FALSE);
+	}
+
+	ut_ad(fil_space_get_type(IBUF_SPACE_ID) == FIL_TABLESPACE);
+
+	if (mtr == NULL) {
+		mtr = &local_mtr;
+		mtr_start(mtr);
+	}
+
+	bitmap_page = ibuf_bitmap_get_map_page(space, page_no, zip_size, mtr);
+
+	ret = ibuf_bitmap_page_get_bits(bitmap_page, page_no, zip_size,
+					IBUF_BITMAP_IBUF, mtr);
+
+	if (mtr == &local_mtr) {
+		mtr_commit(mtr);
+	}
+
+	return(ret);
+}
+
+/************************************************************************
+Returns the page number field of an ibuf record. */
+static
+ulint
+ibuf_rec_get_page_no(
+/*=================*/
+				/* out: page number */
+	const rec_t*	rec)	/* in: ibuf record */
+{
+	const byte*	field;
+	ulint		len;
+
+	ut_ad(ibuf_inside());
+	ut_ad(rec_get_n_fields_old(rec) > 2);
+
+	field = rec_get_nth_field_old(rec, 1, &len);
+
+	if (len == 1) {
+		/* This is of the >= 4.1.x record format */
+		ut_a(trx_sys_multiple_tablespace_format);
+
+		field = rec_get_nth_field_old(rec, 2, &len);
+	} else {
+		ut_a(trx_doublewrite_must_reset_space_ids);
+		ut_a(!trx_sys_multiple_tablespace_format);
+
+		field = rec_get_nth_field_old(rec, 0, &len);
+	}
+
+	ut_a(len == 4);
+
+	return(mach_read_from_4(field));
+}
+
+/************************************************************************
+Returns the space id field of an ibuf record. For < 4.1.x format records
+returns 0. */
+static
+ulint
+ibuf_rec_get_space(
+/*===============*/
+				/* out: space id */
+	const rec_t*	rec)	/* in: ibuf record */
+{
+	const byte*	field;
+	ulint		len;
+
+	ut_ad(ibuf_inside());
+	ut_ad(rec_get_n_fields_old(rec) > 2);
+
+	field = rec_get_nth_field_old(rec, 1, &len);
+
+	if (len == 1) {
+		/* This is of the >= 4.1.x record format */
+
+		ut_a(trx_sys_multiple_tablespace_format);
+		field = rec_get_nth_field_old(rec, 0, &len);
+		ut_a(len == 4);
+
+		return(mach_read_from_4(field));
+	}
+
+	ut_a(trx_doublewrite_must_reset_space_ids);
+	ut_a(!trx_sys_multiple_tablespace_format);
+
+	return(0);
+}
+
+/************************************************************************
+Creates a dummy index for inserting a record to a non-clustered index.
+*/
+static
+dict_index_t*
+ibuf_dummy_index_create(
+/*====================*/
+				/* out: dummy index */
+	ulint		n,	/* in: number of fields */
+	ibool		comp)	/* in: TRUE=use compact record format */
+{
+	dict_table_t*	table;
+	dict_index_t*	index;
+
+	table = dict_mem_table_create("IBUF_DUMMY",
+				      DICT_HDR_SPACE, n,
+				      comp ? DICT_TF_COMPACT : 0);
+
+	index = dict_mem_index_create("IBUF_DUMMY", "IBUF_DUMMY",
+				      DICT_HDR_SPACE, 0, n);
+
+	index->table = table;
+
+	/* avoid ut_ad(index->cached) in dict_index_get_n_unique_in_tree */
+	index->cached = TRUE;
+
+	return(index);
+}
+/************************************************************************
+Add a column to the dummy index */
+static
+void
+ibuf_dummy_index_add_col(
+/*=====================*/
+	dict_index_t*	index,	/* in: dummy index */
+	const dtype_t*	type,	/* in: the data type of the column */
+	ulint		len)	/* in: length of the column */
+{
+	ulint	i	= index->table->n_def;
+	dict_mem_table_add_col(index->table, NULL, NULL,
+			       dtype_get_mtype(type),
+			       dtype_get_prtype(type),
+			       dtype_get_len(type));
+	dict_index_add_col(index, index->table,
+			   dict_table_get_nth_col(index->table, i), len);
+}
+/************************************************************************
+Deallocates a dummy index for inserting a record to a non-clustered index.
+*/
+static
+void
+ibuf_dummy_index_free(
+/*==================*/
+	dict_index_t*	index)	/* in: dummy index */
+{
+	dict_table_t*	table = index->table;
+
+	dict_mem_index_free(index);
+	dict_mem_table_free(table);
+}
+
+/*************************************************************************
+Builds the entry to insert into a non-clustered index when we have the
+corresponding record in an ibuf index. */
+UNIV_INLINE
+dtuple_t*
+ibuf_build_entry_pre_4_1_x(
+/*=======================*/
+					/* out, own: entry to insert to
+					a non-clustered index; NOTE that
+					as we copy pointers to fields in
+					ibuf_rec, the caller must hold a
+					latch to the ibuf_rec page as long
+					as the entry is used! */
+	const rec_t*	ibuf_rec,	/* in: record in an insert buffer */
+	mem_heap_t*	heap,		/* in: heap where built */
+	dict_index_t**	pindex)		/* out, own: dummy index that
+					describes the entry */
+{
+	ulint		i;
+	ulint		len;
+	const byte*	types;
+	dtuple_t*	tuple;
+	ulint		n_fields;
+
+	ut_a(trx_doublewrite_must_reset_space_ids);
+	ut_a(!trx_sys_multiple_tablespace_format);
+
+	n_fields = rec_get_n_fields_old(ibuf_rec) - 2;
+	tuple = dtuple_create(heap, n_fields);
+	types = rec_get_nth_field_old(ibuf_rec, 1, &len);
+
+	ut_a(len == n_fields * DATA_ORDER_NULL_TYPE_BUF_SIZE);
+
+	for (i = 0; i < n_fields; i++) {
+		const byte*	data;
+		dfield_t*	field;
+
+		field = dtuple_get_nth_field(tuple, i);
+
+		data = rec_get_nth_field_old(ibuf_rec, i + 2, &len);
+
+		dfield_set_data(field, data, len);
+
+		dtype_read_for_order_and_null_size(
+			dfield_get_type(field),
+			types + i * DATA_ORDER_NULL_TYPE_BUF_SIZE);
+	}
+
+	*pindex = ibuf_dummy_index_create(n_fields, FALSE);
+
+	return(tuple);
+}
+
+/*************************************************************************
+Builds the entry to insert into a non-clustered index when we have the
+corresponding record in an ibuf index. */
+static
+dtuple_t*
+ibuf_build_entry_from_ibuf_rec(
+/*===========================*/
+					/* out, own: entry to insert to
+					a non-clustered index; NOTE that
+					as we copy pointers to fields in
+					ibuf_rec, the caller must hold a
+					latch to the ibuf_rec page as long
+					as the entry is used! */
+	const rec_t*	ibuf_rec,	/* in: record in an insert buffer */
+	mem_heap_t*	heap,		/* in: heap where built */
+	dict_index_t**	pindex)		/* out, own: dummy index that
+					describes the entry */
+{
+	dtuple_t*	tuple;
+	dfield_t*	field;
+	ulint		n_fields;
+	const byte*	types;
+	const byte*	data;
+	ulint		len;
+	ulint		i;
+	dict_index_t*	index;
+
+	data = rec_get_nth_field_old(ibuf_rec, 1, &len);
+
+	if (len > 1) {
+		/* This a < 4.1.x format record */
+
+		return(ibuf_build_entry_pre_4_1_x(ibuf_rec, heap, pindex));
+	}
+
+	/* This a >= 4.1.x format record */
+
+	ut_a(trx_sys_multiple_tablespace_format);
+	ut_a(*data == 0);
+	ut_a(rec_get_n_fields_old(ibuf_rec) > 4);
+
+	n_fields = rec_get_n_fields_old(ibuf_rec) - 4;
+
+	tuple = dtuple_create(heap, n_fields);
+
+	types = rec_get_nth_field_old(ibuf_rec, 3, &len);
+
+	ut_a(len % DATA_NEW_ORDER_NULL_TYPE_BUF_SIZE <= 1);
+	index = ibuf_dummy_index_create(
+		n_fields, len % DATA_NEW_ORDER_NULL_TYPE_BUF_SIZE);
+
+	if (len % DATA_NEW_ORDER_NULL_TYPE_BUF_SIZE) {
+		/* compact record format */
+		len--;
+		ut_a(*types == 0);
+		types++;
+	}
+
+	ut_a(len == n_fields * DATA_NEW_ORDER_NULL_TYPE_BUF_SIZE);
+
+	for (i = 0; i < n_fields; i++) {
+		field = dtuple_get_nth_field(tuple, i);
+
+		data = rec_get_nth_field_old(ibuf_rec, i + 4, &len);
+
+		dfield_set_data(field, data, len);
+
+		dtype_new_read_for_order_and_null_size(
+			dfield_get_type(field),
+			types + i * DATA_NEW_ORDER_NULL_TYPE_BUF_SIZE);
+
+		ibuf_dummy_index_add_col(index, dfield_get_type(field), len);
+	}
+
+	/* Prevent an ut_ad() failure in page_zip_write_rec() by
+	adding system columns to the dummy table pointed to by the
+	dummy secondary index.  The insert buffer is only used for
+	secondary indexes, whose records never contain any system
+	columns, such as DB_TRX_ID. */
+	ut_d(dict_table_add_system_columns(index->table, index->table->heap));
+
+	*pindex = index;
+
+	return(tuple);
+}
+
+/************************************************************************
+Returns the space taken by a stored non-clustered index entry if converted to
+an index record. */
+static
+ulint
+ibuf_rec_get_volume(
+/*================*/
+				/* out: size of index record in bytes
+				+ an upper limit of the space taken in the
+				page directory */
+	const rec_t*	ibuf_rec)/* in: ibuf record */
+{
+	dtype_t		dtype;
+	ibool		new_format	= FALSE;
+	ulint		data_size	= 0;
+	ulint		n_fields;
+	const byte*	types;
+	const byte*	data;
+	ulint		len;
+	ulint		i;
+
+	ut_ad(ibuf_inside());
+	ut_ad(rec_get_n_fields_old(ibuf_rec) > 2);
+
+	data = rec_get_nth_field_old(ibuf_rec, 1, &len);
+
+	if (len > 1) {
+		/* < 4.1.x format record */
+
+		ut_a(trx_doublewrite_must_reset_space_ids);
+		ut_a(!trx_sys_multiple_tablespace_format);
+
+		n_fields = rec_get_n_fields_old(ibuf_rec) - 2;
+
+		types = rec_get_nth_field_old(ibuf_rec, 1, &len);
+
+		ut_ad(len == n_fields * DATA_ORDER_NULL_TYPE_BUF_SIZE);
+	} else {
+		/* >= 4.1.x format record */
+
+		ut_a(trx_sys_multiple_tablespace_format);
+		ut_a(*data == 0);
+
+		types = rec_get_nth_field_old(ibuf_rec, 3, &len);
+
+		ut_a(len % DATA_NEW_ORDER_NULL_TYPE_BUF_SIZE <= 1);
+		if (len % DATA_NEW_ORDER_NULL_TYPE_BUF_SIZE) {
+			/* compact record format */
+			ulint		volume;
+			dict_index_t*	dummy_index;
+			mem_heap_t*	heap = mem_heap_create(500);
+			dtuple_t*	entry = ibuf_build_entry_from_ibuf_rec(
+				ibuf_rec, heap, &dummy_index);
+			volume = rec_get_converted_size(dummy_index, entry, 0);
+			ibuf_dummy_index_free(dummy_index);
+			mem_heap_free(heap);
+			return(volume + page_dir_calc_reserved_space(1));
+		}
+
+		n_fields = rec_get_n_fields_old(ibuf_rec) - 4;
+
+		new_format = TRUE;
+	}
+
+	for (i = 0; i < n_fields; i++) {
+		if (new_format) {
+			data = rec_get_nth_field_old(ibuf_rec, i + 4, &len);
+
+			dtype_new_read_for_order_and_null_size(
+				&dtype, types + i
+				* DATA_NEW_ORDER_NULL_TYPE_BUF_SIZE);
+		} else {
+			data = rec_get_nth_field_old(ibuf_rec, i + 2, &len);
+
+			dtype_read_for_order_and_null_size(
+				&dtype, types + i
+				* DATA_ORDER_NULL_TYPE_BUF_SIZE);
+		}
+
+		if (len == UNIV_SQL_NULL) {
+			data_size += dtype_get_sql_null_size(&dtype);
+		} else {
+			data_size += len;
+		}
+	}
+
+	return(data_size + rec_get_converted_extra_size(data_size, n_fields, 0)
+	       + page_dir_calc_reserved_space(1));
+}
+
+/*************************************************************************
+Builds the tuple to insert to an ibuf tree when we have an entry for a
+non-clustered index. */
+static
+dtuple_t*
+ibuf_entry_build(
+/*=============*/
+				/* out, own: entry to insert into an ibuf
+				index tree; NOTE that the original entry
+				must be kept because we copy pointers to its
+				fields */
+	dict_index_t*	index,	/* in: non-clustered index */
+	const dtuple_t*	entry,	/* in: entry for a non-clustered index */
+	ulint		space,	/* in: space id */
+	ulint		page_no,/* in: index page number where entry should
+				be inserted */
+	mem_heap_t*	heap)	/* in: heap into which to build */
+{
+	dtuple_t*	tuple;
+	dfield_t*	field;
+	const dfield_t*	entry_field;
+	ulint		n_fields;
+	byte*		buf;
+	byte*		buf2;
+	ulint		i;
+
+	/* Starting from 4.1.x, we have to build a tuple whose
+	(1) first field is the space id,
+	(2) the second field a single marker byte (0) to tell that this
+	is a new format record,
+	(3) the third contains the page number, and
+	(4) the fourth contains the relevent type information of each data
+	field; the length of this field % DATA_NEW_ORDER_NULL_TYPE_BUF_SIZE is
+	(a) 0 for b-trees in the old format, and
+	(b) 1 for b-trees in the compact format, the first byte of the field
+	being the marker (0);
+	(5) and the rest of the fields are copied from entry. All fields
+	in the tuple are ordered like the type binary in our insert buffer
+	tree. */
+
+	n_fields = dtuple_get_n_fields(entry);
+
+	tuple = dtuple_create(heap, n_fields + 4);
+
+	/* Store the space id in tuple */
+
+	field = dtuple_get_nth_field(tuple, 0);
+
+	buf = mem_heap_alloc(heap, 4);
+
+	mach_write_to_4(buf, space);
+
+	dfield_set_data(field, buf, 4);
+
+	/* Store the marker byte field in tuple */
+
+	field = dtuple_get_nth_field(tuple, 1);
+
+	buf = mem_heap_alloc(heap, 1);
+
+	/* We set the marker byte zero */
+
+	mach_write_to_1(buf, 0);
+
+	dfield_set_data(field, buf, 1);
+
+	/* Store the page number in tuple */
+
+	field = dtuple_get_nth_field(tuple, 2);
+
+	buf = mem_heap_alloc(heap, 4);
+
+	mach_write_to_4(buf, page_no);
+
+	dfield_set_data(field, buf, 4);
+
+	/* Store the type info in buf2, and add the fields from entry to
+	tuple */
+	buf2 = mem_heap_alloc(heap, n_fields
+			      * DATA_NEW_ORDER_NULL_TYPE_BUF_SIZE
+			      + dict_table_is_comp(index->table));
+	if (dict_table_is_comp(index->table)) {
+		*buf2++ = 0; /* write the compact format indicator */
+	}
+	for (i = 0; i < n_fields; i++) {
+		ulint			fixed_len;
+		const dict_field_t*	ifield;
+
+		/* We add 4 below because we have the 4 extra fields at the
+		start of an ibuf record */
+
+		field = dtuple_get_nth_field(tuple, i + 4);
+		entry_field = dtuple_get_nth_field(entry, i);
+		dfield_copy(field, entry_field);
+
+		ifield = dict_index_get_nth_field(index, i);
+		/* Prefix index columns of fixed-length columns are of
+		fixed length.  However, in the function call below,
+		dfield_get_type(entry_field) contains the fixed length
+		of the column in the clustered index.  Replace it with
+		the fixed length of the secondary index column. */
+		fixed_len = ifield->fixed_len;
+
+#ifdef UNIV_DEBUG
+		if (fixed_len) {
+			/* dict_index_add_col() should guarantee these */
+			ut_ad(fixed_len <= (ulint)
+			      dfield_get_type(entry_field)->len);
+			if (ifield->prefix_len) {
+				ut_ad(ifield->prefix_len == fixed_len);
+			} else {
+				ut_ad(fixed_len == (ulint)
+				      dfield_get_type(entry_field)->len);
+			}
+		}
+#endif /* UNIV_DEBUG */
+
+		dtype_new_store_for_order_and_null_size(
+			buf2 + i * DATA_NEW_ORDER_NULL_TYPE_BUF_SIZE,
+			dfield_get_type(entry_field), fixed_len);
+	}
+
+	/* Store the type info in buf2 to field 3 of tuple */
+
+	field = dtuple_get_nth_field(tuple, 3);
+
+	if (dict_table_is_comp(index->table)) {
+		buf2--;
+	}
+
+	dfield_set_data(field, buf2, n_fields
+			* DATA_NEW_ORDER_NULL_TYPE_BUF_SIZE
+			+ dict_table_is_comp(index->table));
+	/* Set all the types in the new tuple binary */
+
+	dtuple_set_types_binary(tuple, n_fields + 4);
+
+	return(tuple);
+}
+
+/*************************************************************************
+Builds a search tuple used to search buffered inserts for an index page.
+This is for < 4.1.x format records */
+static
+dtuple_t*
+ibuf_search_tuple_build(
+/*====================*/
+				/* out, own: search tuple */
+	ulint		space,	/* in: space id */
+	ulint		page_no,/* in: index page number */
+	mem_heap_t*	heap)	/* in: heap into which to build */
+{
+	dtuple_t*	tuple;
+	dfield_t*	field;
+	byte*		buf;
+
+	ut_a(space == 0);
+	ut_a(trx_doublewrite_must_reset_space_ids);
+	ut_a(!trx_sys_multiple_tablespace_format);
+
+	tuple = dtuple_create(heap, 1);
+
+	/* Store the page number in tuple */
+
+	field = dtuple_get_nth_field(tuple, 0);
+
+	buf = mem_heap_alloc(heap, 4);
+
+	mach_write_to_4(buf, page_no);
+
+	dfield_set_data(field, buf, 4);
+
+	dtuple_set_types_binary(tuple, 1);
+
+	return(tuple);
+}
+
+/*************************************************************************
+Builds a search tuple used to search buffered inserts for an index page.
+This is for >= 4.1.x format records. */
+static
+dtuple_t*
+ibuf_new_search_tuple_build(
+/*========================*/
+				/* out, own: search tuple */
+	ulint		space,	/* in: space id */
+	ulint		page_no,/* in: index page number */
+	mem_heap_t*	heap)	/* in: heap into which to build */
+{
+	dtuple_t*	tuple;
+	dfield_t*	field;
+	byte*		buf;
+
+	ut_a(trx_sys_multiple_tablespace_format);
+
+	tuple = dtuple_create(heap, 3);
+
+	/* Store the space id in tuple */
+
+	field = dtuple_get_nth_field(tuple, 0);
+
+	buf = mem_heap_alloc(heap, 4);
+
+	mach_write_to_4(buf, space);
+
+	dfield_set_data(field, buf, 4);
+
+	/* Store the new format record marker byte */
+
+	field = dtuple_get_nth_field(tuple, 1);
+
+	buf = mem_heap_alloc(heap, 1);
+
+	mach_write_to_1(buf, 0);
+
+	dfield_set_data(field, buf, 1);
+
+	/* Store the page number in tuple */
+
+	field = dtuple_get_nth_field(tuple, 2);
+
+	buf = mem_heap_alloc(heap, 4);
+
+	mach_write_to_4(buf, page_no);
+
+	dfield_set_data(field, buf, 4);
+
+	dtuple_set_types_binary(tuple, 3);
+
+	return(tuple);
+}
+
+/*************************************************************************
+Checks if there are enough pages in the free list of the ibuf tree that we
+dare to start a pessimistic insert to the insert buffer. */
+UNIV_INLINE
+ibool
+ibuf_data_enough_free_for_insert(void)
+/*==================================*/
+				/* out: TRUE if enough free pages in list */
+{
+	ut_ad(mutex_own(&ibuf_mutex));
+
+	/* We want a big margin of free pages, because a B-tree can sometimes
+	grow in size also if records are deleted from it, as the node pointers
+	can change, and we must make sure that we are able to delete the
+	inserts buffered for pages that we read to the buffer pool, without
+	any risk of running out of free space in the insert buffer. */
+
+	return(ibuf->free_list_len >= (ibuf->size / 2) + 3 * ibuf->height);
+}
+
+/*************************************************************************
+Checks if there are enough pages in the free list of the ibuf tree that we
+should remove them and free to the file space management. */
+UNIV_INLINE
+ibool
+ibuf_data_too_much_free(void)
+/*=========================*/
+				/* out: TRUE if enough free pages in list */
+{
+	ut_ad(mutex_own(&ibuf_mutex));
+
+	return(ibuf->free_list_len >= 3 + (ibuf->size / 2) + 3 * ibuf->height);
+}
+
+/*************************************************************************
+Allocates a new page from the ibuf file segment and adds it to the free
+list. */
+static
+ulint
+ibuf_add_free_page(void)
+/*====================*/
+					/* out: DB_SUCCESS, or DB_STRONG_FAIL
+					if no space left */
+{
+	mtr_t	mtr;
+	page_t*	header_page;
+	ulint	flags;
+	ulint	zip_size;
+	ulint	page_no;
+	page_t*	page;
+	page_t*	root;
+	page_t*	bitmap_page;
+
+	mtr_start(&mtr);
+
+	/* Acquire the fsp latch before the ibuf header, obeying the latching
+	order */
+	mtr_x_lock(fil_space_get_latch(IBUF_SPACE_ID, &flags), &mtr);
+	zip_size = dict_table_flags_to_zip_size(flags);
+
+	header_page = ibuf_header_page_get(&mtr);
+
+	/* Allocate a new page: NOTE that if the page has been a part of a
+	non-clustered index which has subsequently been dropped, then the
+	page may have buffered inserts in the insert buffer, and these
+	should be deleted from there. These get deleted when the page
+	allocation creates the page in buffer. Thus the call below may end
+	up calling the insert buffer routines and, as we yet have no latches
+	to insert buffer tree pages, these routines can run without a risk
+	of a deadlock. This is the reason why we created a special ibuf
+	header page apart from the ibuf tree. */
+
+	page_no = fseg_alloc_free_page(
+		header_page + IBUF_HEADER + IBUF_TREE_SEG_HEADER, 0, FSP_UP,
+		&mtr);
+
+	if (page_no == FIL_NULL) {
+		mtr_commit(&mtr);
+
+		return(DB_STRONG_FAIL);
+	}
+
+	{
+		buf_block_t*	block;
+
+		block = buf_page_get(
+			IBUF_SPACE_ID, 0, page_no, RW_X_LATCH, &mtr);
+
+		buf_block_dbg_add_level(block, SYNC_TREE_NODE_NEW);
+
+
+		page = buf_block_get_frame(block);
+	}
+
+	ibuf_enter();
+
+	mutex_enter(&ibuf_mutex);
+
+	root = ibuf_tree_root_get(&mtr);
+
+	/* Add the page to the free list and update the ibuf size data */
+
+	flst_add_last(root + PAGE_HEADER + PAGE_BTR_IBUF_FREE_LIST,
+		      page + PAGE_HEADER + PAGE_BTR_IBUF_FREE_LIST_NODE, &mtr);
+
+	mlog_write_ulint(page + FIL_PAGE_TYPE, FIL_PAGE_IBUF_FREE_LIST,
+			 MLOG_2BYTES, &mtr);
+
+	ibuf->seg_size++;
+	ibuf->free_list_len++;
+
+	/* Set the bit indicating that this page is now an ibuf tree page
+	(level 2 page) */
+
+	bitmap_page = ibuf_bitmap_get_map_page(
+		IBUF_SPACE_ID, page_no, zip_size, &mtr);
+
+	ibuf_bitmap_page_set_bits(
+		bitmap_page, page_no, zip_size, IBUF_BITMAP_IBUF, TRUE, &mtr);
+
+	mtr_commit(&mtr);
+
+	mutex_exit(&ibuf_mutex);
+
+	ibuf_exit();
+
+	return(DB_SUCCESS);
+}
+
+/*************************************************************************
+Removes a page from the free list and frees it to the fsp system. */
+static
+void
+ibuf_remove_free_page(void)
+/*=======================*/
+{
+	mtr_t	mtr;
+	mtr_t	mtr2;
+	page_t*	header_page;
+	ulint	flags;
+	ulint	zip_size;
+	ulint	page_no;
+	page_t*	page;
+	page_t*	root;
+	page_t*	bitmap_page;
+
+	mtr_start(&mtr);
+
+	/* Acquire the fsp latch before the ibuf header, obeying the latching
+	order */
+	mtr_x_lock(fil_space_get_latch(IBUF_SPACE_ID, &flags), &mtr);
+	zip_size = dict_table_flags_to_zip_size(flags);
+
+	header_page = ibuf_header_page_get(&mtr);
+
+	/* Prevent pessimistic inserts to insert buffer trees for a while */
+	mutex_enter(&ibuf_pessimistic_insert_mutex);
+
+	ibuf_enter();
+
+	mutex_enter(&ibuf_mutex);
+
+	if (!ibuf_data_too_much_free()) {
+
+		mutex_exit(&ibuf_mutex);
+
+		ibuf_exit();
+
+		mutex_exit(&ibuf_pessimistic_insert_mutex);
+
+		mtr_commit(&mtr);
+
+		return;
+	}
+
+	mtr_start(&mtr2);
+
+	root = ibuf_tree_root_get(&mtr2);
+
+	page_no = flst_get_last(root + PAGE_HEADER + PAGE_BTR_IBUF_FREE_LIST,
+				&mtr2).page;
+
+	/* NOTE that we must release the latch on the ibuf tree root
+	because in fseg_free_page we access level 1 pages, and the root
+	is a level 2 page. */
+
+	mtr_commit(&mtr2);
+	mutex_exit(&ibuf_mutex);
+
+	ibuf_exit();
+
+	/* Since pessimistic inserts were prevented, we know that the
+	page is still in the free list. NOTE that also deletes may take
+	pages from the free list, but they take them from the start, and
+	the free list was so long that they cannot have taken the last
+	page from it. */
+
+	fseg_free_page(header_page + IBUF_HEADER + IBUF_TREE_SEG_HEADER,
+		       IBUF_SPACE_ID, page_no, &mtr);
+
+#ifdef UNIV_DEBUG_FILE_ACCESSES
+	buf_page_reset_file_page_was_freed(IBUF_SPACE_ID, page_no);
+#endif
+
+	ibuf_enter();
+
+	mutex_enter(&ibuf_mutex);
+
+	root = ibuf_tree_root_get(&mtr);
+
+	ut_ad(page_no == flst_get_last(root + PAGE_HEADER
+				       + PAGE_BTR_IBUF_FREE_LIST, &mtr).page);
+
+	{
+		buf_block_t*	block;
+
+		block = buf_page_get(
+			IBUF_SPACE_ID, 0, page_no, RW_X_LATCH, &mtr);
+
+		buf_block_dbg_add_level(block, SYNC_TREE_NODE);
+
+
+		page = buf_block_get_frame(block);
+	}
+
+	/* Remove the page from the free list and update the ibuf size data */
+
+	flst_remove(root + PAGE_HEADER + PAGE_BTR_IBUF_FREE_LIST,
+		    page + PAGE_HEADER + PAGE_BTR_IBUF_FREE_LIST_NODE, &mtr);
+
+	ibuf->seg_size--;
+	ibuf->free_list_len--;
+
+	mutex_exit(&ibuf_pessimistic_insert_mutex);
+
+	/* Set the bit indicating that this page is no more an ibuf tree page
+	(level 2 page) */
+
+	bitmap_page = ibuf_bitmap_get_map_page(
+		IBUF_SPACE_ID, page_no, zip_size, &mtr);
+
+	ibuf_bitmap_page_set_bits(
+		bitmap_page, page_no, zip_size, IBUF_BITMAP_IBUF, FALSE, &mtr);
+
+#ifdef UNIV_DEBUG_FILE_ACCESSES
+	buf_page_set_file_page_was_freed(IBUF_SPACE_ID, page_no);
+#endif
+	mtr_commit(&mtr);
+
+	mutex_exit(&ibuf_mutex);
+
+	ibuf_exit();
+}
+
+/***************************************************************************
+Frees excess pages from the ibuf free list. This function is called when an OS
+thread calls fsp services to allocate a new file segment, or a new page to a
+file segment, and the thread did not own the fsp latch before this call. */
+UNIV_INTERN
+void
+ibuf_free_excess_pages(void)
+/*========================*/
+{
+	ulint		i;
+
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(rw_lock_own(fil_space_get_latch(IBUF_SPACE_ID, NULL),
+			  RW_LOCK_EX));
+#endif /* UNIV_SYNC_DEBUG */
+
+	ut_ad(rw_lock_get_x_lock_count(
+		fil_space_get_latch(IBUF_SPACE_ID, NULL)) == 1);
+
+	ut_ad(!ibuf_inside());
+
+	/* NOTE: We require that the thread did not own the latch before,
+	because then we know that we can obey the correct latching order
+	for ibuf latches */
+
+	if (!ibuf) {
+		/* Not yet initialized; not sure if this is possible, but
+		does no harm to check for it. */
+
+		return;
+	}
+
+	/* Free at most a few pages at a time, so that we do not delay the
+	requested service too much */
+
+	for (i = 0; i < 4; i++) {
+
+		mutex_enter(&ibuf_mutex);
+
+		if (!ibuf_data_too_much_free()) {
+
+			mutex_exit(&ibuf_mutex);
+
+			return;
+		}
+
+		mutex_exit(&ibuf_mutex);
+
+		ibuf_remove_free_page();
+	}
+}
+
+/*************************************************************************
+Reads page numbers from a leaf in an ibuf tree. */
+static
+ulint
+ibuf_get_merge_page_nos(
+/*====================*/
+				/* out: a lower limit for the combined volume
+				of records which will be merged */
+	ibool		contract,/* in: TRUE if this function is called to
+				contract the tree, FALSE if this is called
+				when a single page becomes full and we look
+				if it pays to read also nearby pages */
+	rec_t*		rec,	/* in: record from which we read up and down
+				in the chain of records */
+	ulint*		space_ids,/* in/out: space id's of the pages */
+	ib_int64_t*	space_versions,/* in/out: tablespace version
+				timestamps; used to prevent reading in old
+				pages after DISCARD + IMPORT tablespace */
+	ulint*		page_nos,/* in/out: buffer for at least
+				IBUF_MAX_N_PAGES_MERGED many page numbers;
+				the page numbers are in an ascending order */
+	ulint*		n_stored)/* out: number of page numbers stored to
+				page_nos in this function */
+{
+	ulint	prev_page_no;
+	ulint	prev_space_id;
+	ulint	first_page_no;
+	ulint	first_space_id;
+	ulint	rec_page_no;
+	ulint	rec_space_id;
+	ulint	sum_volumes;
+	ulint	volume_for_page;
+	ulint	rec_volume;
+	ulint	limit;
+	ulint	n_pages;
+
+	*n_stored = 0;
+
+	limit = ut_min(IBUF_MAX_N_PAGES_MERGED, buf_pool->curr_size / 4);
+
+	if (page_rec_is_supremum(rec)) {
+
+		rec = page_rec_get_prev(rec);
+	}
+
+	if (page_rec_is_infimum(rec)) {
+
+		rec = page_rec_get_next(rec);
+	}
+
+	if (page_rec_is_supremum(rec)) {
+
+		return(0);
+	}
+
+	first_page_no = ibuf_rec_get_page_no(rec);
+	first_space_id = ibuf_rec_get_space(rec);
+	n_pages = 0;
+	prev_page_no = 0;
+	prev_space_id = 0;
+
+	/* Go backwards from the first rec until we reach the border of the
+	'merge area', or the page start or the limit of storeable pages is
+	reached */
+
+	while (!page_rec_is_infimum(rec) && UNIV_LIKELY(n_pages < limit)) {
+
+		rec_page_no = ibuf_rec_get_page_no(rec);
+		rec_space_id = ibuf_rec_get_space(rec);
+
+		if (rec_space_id != first_space_id
+		    || (rec_page_no / IBUF_MERGE_AREA)
+		    != (first_page_no / IBUF_MERGE_AREA)) {
+
+			break;
+		}
+
+		if (rec_page_no != prev_page_no
+		    || rec_space_id != prev_space_id) {
+			n_pages++;
+		}
+
+		prev_page_no = rec_page_no;
+		prev_space_id = rec_space_id;
+
+		rec = page_rec_get_prev(rec);
+	}
+
+	rec = page_rec_get_next(rec);
+
+	/* At the loop start there is no prev page; we mark this with a pair
+	of space id, page no (0, 0) for which there can never be entries in
+	the insert buffer */
+
+	prev_page_no = 0;
+	prev_space_id = 0;
+	sum_volumes = 0;
+	volume_for_page = 0;
+
+	while (*n_stored < limit) {
+		if (page_rec_is_supremum(rec)) {
+			/* When no more records available, mark this with
+			another 'impossible' pair of space id, page no */
+			rec_page_no = 1;
+			rec_space_id = 0;
+		} else {
+			rec_page_no = ibuf_rec_get_page_no(rec);
+			rec_space_id = ibuf_rec_get_space(rec);
+			ut_ad(rec_page_no > IBUF_TREE_ROOT_PAGE_NO);
+		}
+
+#ifdef UNIV_IBUF_DEBUG
+		ut_a(*n_stored < IBUF_MAX_N_PAGES_MERGED);
+#endif
+		if ((rec_space_id != prev_space_id
+		     || rec_page_no != prev_page_no)
+		    && (prev_space_id != 0 || prev_page_no != 0)) {
+
+			if ((prev_page_no == first_page_no
+			     && prev_space_id == first_space_id)
+			    || contract
+			    || (volume_for_page
+				> ((IBUF_MERGE_THRESHOLD - 1)
+				   * 4 * UNIV_PAGE_SIZE
+				   / IBUF_PAGE_SIZE_PER_FREE_SPACE)
+				/ IBUF_MERGE_THRESHOLD)) {
+
+				space_ids[*n_stored] = prev_space_id;
+				space_versions[*n_stored]
+					= fil_space_get_version(prev_space_id);
+				page_nos[*n_stored] = prev_page_no;
+
+				(*n_stored)++;
+
+				sum_volumes += volume_for_page;
+			}
+
+			if (rec_space_id != first_space_id
+			    || rec_page_no / IBUF_MERGE_AREA
+			    != first_page_no / IBUF_MERGE_AREA) {
+
+				break;
+			}
+
+			volume_for_page = 0;
+		}
+
+		if (rec_page_no == 1 && rec_space_id == 0) {
+			/* Supremum record */
+
+			break;
+		}
+
+		rec_volume = ibuf_rec_get_volume(rec);
+
+		volume_for_page += rec_volume;
+
+		prev_page_no = rec_page_no;
+		prev_space_id = rec_space_id;
+
+		rec = page_rec_get_next(rec);
+	}
+
+#ifdef UNIV_IBUF_DEBUG
+	ut_a(*n_stored <= IBUF_MAX_N_PAGES_MERGED);
+#endif
+#if 0
+	fprintf(stderr, "Ibuf merge batch %lu pages %lu volume\n",
+		*n_stored, sum_volumes);
+#endif
+	return(sum_volumes);
+}
+
+/*************************************************************************
+Contracts insert buffer trees by reading pages to the buffer pool. */
+static
+ulint
+ibuf_contract_ext(
+/*==============*/
+			/* out: a lower limit for the combined size in bytes
+			of entries which will be merged from ibuf trees to the
+			pages read, 0 if ibuf is empty */
+	ulint*	n_pages,/* out: number of pages to which merged */
+	ibool	sync)	/* in: TRUE if the caller wants to wait for the
+			issued read with the highest tablespace address
+			to complete */
+{
+	btr_pcur_t	pcur;
+	ulint		page_nos[IBUF_MAX_N_PAGES_MERGED];
+	ulint		space_ids[IBUF_MAX_N_PAGES_MERGED];
+	ib_int64_t	space_versions[IBUF_MAX_N_PAGES_MERGED];
+	ulint		n_stored;
+	ulint		sum_sizes;
+	mtr_t		mtr;
+
+	*n_pages = 0;
+	ut_ad(!ibuf_inside());
+
+	mutex_enter(&ibuf_mutex);
+
+	if (ibuf->empty) {
+ibuf_is_empty:
+		mutex_exit(&ibuf_mutex);
+
+		return(0);
+	}
+
+	mtr_start(&mtr);
+
+	ibuf_enter();
+
+	/* Open a cursor to a randomly chosen leaf of the tree, at a random
+	position within the leaf */
+
+	btr_pcur_open_at_rnd_pos(ibuf->index, BTR_SEARCH_LEAF, &pcur, &mtr);
+
+	if (page_get_n_recs(btr_pcur_get_page(&pcur)) == 0) {
+		/* When the ibuf tree is emptied completely, the last record
+		is removed using an optimistic delete and ibuf_size_update
+		is not called, causing ibuf->empty to remain FALSE. If we do
+		not reset it to TRUE here then database shutdown will hang
+		in the loop in ibuf_contract_for_n_pages. */
+
+		ibuf->empty = TRUE;
+
+		ibuf_exit();
+
+		mtr_commit(&mtr);
+		btr_pcur_close(&pcur);
+
+		goto ibuf_is_empty;
+	}
+
+	mutex_exit(&ibuf_mutex);
+
+	sum_sizes = ibuf_get_merge_page_nos(TRUE, btr_pcur_get_rec(&pcur),
+					    space_ids, space_versions,
+					    page_nos, &n_stored);
+#if 0 /* defined UNIV_IBUF_DEBUG */
+	fprintf(stderr, "Ibuf contract sync %lu pages %lu volume %lu\n",
+		sync, n_stored, sum_sizes);
+#endif
+	ibuf_exit();
+
+	mtr_commit(&mtr);
+	btr_pcur_close(&pcur);
+
+	buf_read_ibuf_merge_pages(sync, space_ids, space_versions, page_nos,
+				  n_stored);
+	*n_pages = n_stored;
+
+	return(sum_sizes + 1);
+}
+
+/*************************************************************************
+Contracts insert buffer trees by reading pages to the buffer pool. */
+UNIV_INTERN
+ulint
+ibuf_contract(
+/*==========*/
+			/* out: a lower limit for the combined size in bytes
+			of entries which will be merged from ibuf trees to the
+			pages read, 0 if ibuf is empty */
+	ibool	sync)	/* in: TRUE if the caller wants to wait for the
+			issued read with the highest tablespace address
+			to complete */
+{
+	ulint	n_pages;
+
+	return(ibuf_contract_ext(&n_pages, sync));
+}
+
+/*************************************************************************
+Contracts insert buffer trees by reading pages to the buffer pool. */
+UNIV_INTERN
+ulint
+ibuf_contract_for_n_pages(
+/*======================*/
+			/* out: a lower limit for the combined size in bytes
+			of entries which will be merged from ibuf trees to the
+			pages read, 0 if ibuf is empty */
+	ibool	sync,	/* in: TRUE if the caller wants to wait for the
+			issued read with the highest tablespace address
+			to complete */
+	ulint	n_pages)/* in: try to read at least this many pages to
+			the buffer pool and merge the ibuf contents to
+			them */
+{
+	ulint	sum_bytes	= 0;
+	ulint	sum_pages	= 0;
+	ulint	n_bytes;
+	ulint	n_pag2;
+
+	while (sum_pages < n_pages) {
+		n_bytes = ibuf_contract_ext(&n_pag2, sync);
+
+		if (n_bytes == 0) {
+			return(sum_bytes);
+		}
+
+		sum_bytes += n_bytes;
+		sum_pages += n_pag2;
+	}
+
+	return(sum_bytes);
+}
+
+/*************************************************************************
+Contract insert buffer trees after insert if they are too big. */
+UNIV_INLINE
+void
+ibuf_contract_after_insert(
+/*=======================*/
+	ulint	entry_size)	/* in: size of a record which was inserted
+				into an ibuf tree */
+{
+	ibool	sync;
+	ulint	sum_sizes;
+	ulint	size;
+
+	mutex_enter(&ibuf_mutex);
+
+	if (!srv_ibuf_active_contract) {
+	if (ibuf->size < ibuf->max_size + IBUF_CONTRACT_ON_INSERT_NON_SYNC) {
+		mutex_exit(&ibuf_mutex);
+
+		return;
+	}
+	}
+
+	sync = FALSE;
+
+	if (ibuf->size >= ibuf->max_size + IBUF_CONTRACT_ON_INSERT_SYNC) {
+
+		sync = TRUE;
+	}
+
+	mutex_exit(&ibuf_mutex);
+
+	/* Contract at least entry_size many bytes */
+	sum_sizes = 0;
+	size = 1;
+
+	while ((size > 0) && (sum_sizes < entry_size)) {
+
+		size = ibuf_contract(sync);
+		sum_sizes += size;
+	}
+}
+
+/*************************************************************************
+Gets an upper limit for the combined size of entries buffered in the insert
+buffer for a given page. */
+static
+ulint
+ibuf_get_volume_buffered(
+/*=====================*/
+				/* out: upper limit for the volume of
+				buffered inserts for the index page, in bytes;
+				we may also return UNIV_PAGE_SIZE, if the
+				entries for the index page span on several
+				pages in the insert buffer */
+	btr_pcur_t*	pcur,	/* in: pcur positioned at a place in an
+				insert buffer tree where we would insert an
+				entry for the index page whose number is
+				page_no, latch mode has to be BTR_MODIFY_PREV
+				or BTR_MODIFY_TREE */
+	ulint		space,	/* in: space id */
+	ulint		page_no,/* in: page number of an index page */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	ulint	volume;
+	rec_t*	rec;
+	page_t*	page;
+	ulint	prev_page_no;
+	page_t*	prev_page;
+	ulint	next_page_no;
+	page_t*	next_page;
+
+	ut_a(trx_sys_multiple_tablespace_format);
+
+	ut_ad((pcur->latch_mode == BTR_MODIFY_PREV)
+	      || (pcur->latch_mode == BTR_MODIFY_TREE));
+
+	/* Count the volume of records earlier in the alphabetical order than
+	pcur */
+
+	volume = 0;
+
+	rec = btr_pcur_get_rec(pcur);
+	page = page_align(rec);
+
+	if (page_rec_is_supremum(rec)) {
+		rec = page_rec_get_prev(rec);
+	}
+
+	for (;;) {
+		if (page_rec_is_infimum(rec)) {
+
+			break;
+		}
+
+		if (page_no != ibuf_rec_get_page_no(rec)
+		    || space != ibuf_rec_get_space(rec)) {
+
+			goto count_later;
+		}
+
+		volume += ibuf_rec_get_volume(rec);
+
+		rec = page_rec_get_prev(rec);
+	}
+
+	/* Look at the previous page */
+
+	prev_page_no = btr_page_get_prev(page, mtr);
+
+	if (prev_page_no == FIL_NULL) {
+
+		goto count_later;
+	}
+
+	{
+		buf_block_t*	block;
+
+		block = buf_page_get(
+			IBUF_SPACE_ID, 0, prev_page_no, RW_X_LATCH, mtr);
+
+		buf_block_dbg_add_level(block, SYNC_TREE_NODE);
+
+
+		prev_page = buf_block_get_frame(block);
+	}
+
+#ifdef UNIV_BTR_DEBUG
+	ut_a(btr_page_get_next(prev_page, mtr)
+	     == page_get_page_no(page));
+#endif /* UNIV_BTR_DEBUG */
+
+	rec = page_get_supremum_rec(prev_page);
+	rec = page_rec_get_prev(rec);
+
+	for (;;) {
+		if (page_rec_is_infimum(rec)) {
+
+			/* We cannot go to yet a previous page, because we
+			do not have the x-latch on it, and cannot acquire one
+			because of the latching order: we have to give up */
+
+			return(UNIV_PAGE_SIZE);
+		}
+
+		if (page_no != ibuf_rec_get_page_no(rec)
+		    || space != ibuf_rec_get_space(rec)) {
+
+			goto count_later;
+		}
+
+		volume += ibuf_rec_get_volume(rec);
+
+		rec = page_rec_get_prev(rec);
+	}
+
+count_later:
+	rec = btr_pcur_get_rec(pcur);
+
+	if (!page_rec_is_supremum(rec)) {
+		rec = page_rec_get_next(rec);
+	}
+
+	for (;;) {
+		if (page_rec_is_supremum(rec)) {
+
+			break;
+		}
+
+		if (page_no != ibuf_rec_get_page_no(rec)
+		    || space != ibuf_rec_get_space(rec)) {
+
+			return(volume);
+		}
+
+		volume += ibuf_rec_get_volume(rec);
+
+		rec = page_rec_get_next(rec);
+	}
+
+	/* Look at the next page */
+
+	next_page_no = btr_page_get_next(page, mtr);
+
+	if (next_page_no == FIL_NULL) {
+
+		return(volume);
+	}
+
+	{
+		buf_block_t*	block;
+
+		block = buf_page_get(
+			IBUF_SPACE_ID, 0, next_page_no, RW_X_LATCH, mtr);
+
+		buf_block_dbg_add_level(block, SYNC_TREE_NODE);
+
+
+		next_page = buf_block_get_frame(block);
+	}
+
+#ifdef UNIV_BTR_DEBUG
+	ut_a(btr_page_get_prev(next_page, mtr) == page_get_page_no(page));
+#endif /* UNIV_BTR_DEBUG */
+
+	rec = page_get_infimum_rec(next_page);
+	rec = page_rec_get_next(rec);
+
+	for (;;) {
+		if (page_rec_is_supremum(rec)) {
+
+			/* We give up */
+
+			return(UNIV_PAGE_SIZE);
+		}
+
+		if (page_no != ibuf_rec_get_page_no(rec)
+		    || space != ibuf_rec_get_space(rec)) {
+
+			return(volume);
+		}
+
+		volume += ibuf_rec_get_volume(rec);
+
+		rec = page_rec_get_next(rec);
+	}
+}
+
+/*************************************************************************
+Reads the biggest tablespace id from the high end of the insert buffer
+tree and updates the counter in fil_system. */
+UNIV_INTERN
+void
+ibuf_update_max_tablespace_id(void)
+/*===============================*/
+{
+	ulint		max_space_id;
+	const rec_t*	rec;
+	const byte*	field;
+	ulint		len;
+	btr_pcur_t	pcur;
+	mtr_t		mtr;
+
+	ut_a(!dict_table_is_comp(ibuf->index->table));
+
+	ibuf_enter();
+
+	mtr_start(&mtr);
+
+	btr_pcur_open_at_index_side(
+		FALSE, ibuf->index, BTR_SEARCH_LEAF, &pcur, TRUE, &mtr);
+
+	btr_pcur_move_to_prev(&pcur, &mtr);
+
+	if (btr_pcur_is_before_first_on_page(&pcur)) {
+		/* The tree is empty */
+
+		max_space_id = 0;
+	} else {
+		rec = btr_pcur_get_rec(&pcur);
+
+		field = rec_get_nth_field_old(rec, 0, &len);
+
+		ut_a(len == 4);
+
+		max_space_id = mach_read_from_4(field);
+	}
+
+	mtr_commit(&mtr);
+	ibuf_exit();
+
+	/* printf("Maximum space id in insert buffer %lu\n", max_space_id); */
+
+	fil_set_max_space_id_if_bigger(max_space_id);
+}
+
+/*************************************************************************
+Makes an index insert to the insert buffer, instead of directly to the disk
+page, if this is possible. */
+static
+ulint
+ibuf_insert_low(
+/*============*/
+				/* out: DB_SUCCESS, DB_FAIL, DB_STRONG_FAIL */
+	ulint		mode,	/* in: BTR_MODIFY_PREV or BTR_MODIFY_TREE */
+	const dtuple_t*	entry,	/* in: index entry to insert */
+	ulint		entry_size,
+				/* in: rec_get_converted_size(index, entry) */
+	dict_index_t*	index,	/* in: index where to insert; must not be
+				unique or clustered */
+	ulint		space,	/* in: space id where to insert */
+	ulint		zip_size,/* in: compressed page size in bytes, or 0 */
+	ulint		page_no,/* in: page number where to insert */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	big_rec_t*	dummy_big_rec;
+	btr_pcur_t	pcur;
+	btr_cur_t*	cursor;
+	dtuple_t*	ibuf_entry;
+	mem_heap_t*	heap;
+	ulint		buffered;
+	rec_t*		ins_rec;
+	ibool		old_bit_value;
+	page_t*		bitmap_page;
+	page_t*		root;
+	ulint		err;
+	ibool		do_merge;
+	ulint		space_ids[IBUF_MAX_N_PAGES_MERGED];
+	ib_int64_t	space_versions[IBUF_MAX_N_PAGES_MERGED];
+	ulint		page_nos[IBUF_MAX_N_PAGES_MERGED];
+	ulint		n_stored;
+	ulint		bits;
+	mtr_t		mtr;
+	mtr_t		bitmap_mtr;
+
+	ut_a(!dict_index_is_clust(index));
+	ut_ad(dtuple_check_typed(entry));
+	ut_ad(ut_is_2pow(zip_size));
+
+	ut_a(trx_sys_multiple_tablespace_format);
+
+	do_merge = FALSE;
+
+	mutex_enter(&ibuf_mutex);
+
+	if (ibuf->size >= ibuf->max_size + IBUF_CONTRACT_DO_NOT_INSERT) {
+		/* Insert buffer is now too big, contract it but do not try
+		to insert */
+
+		mutex_exit(&ibuf_mutex);
+
+#ifdef UNIV_IBUF_DEBUG
+		fputs("Ibuf too big\n", stderr);
+#endif
+		/* Use synchronous contract (== TRUE) */
+		ibuf_contract(TRUE);
+
+		return(DB_STRONG_FAIL);
+	}
+
+	mutex_exit(&ibuf_mutex);
+
+	if (mode == BTR_MODIFY_TREE) {
+		mutex_enter(&ibuf_pessimistic_insert_mutex);
+
+		ibuf_enter();
+
+		mutex_enter(&ibuf_mutex);
+
+		while (!ibuf_data_enough_free_for_insert()) {
+
+			mutex_exit(&ibuf_mutex);
+
+			ibuf_exit();
+
+			mutex_exit(&ibuf_pessimistic_insert_mutex);
+
+			err = ibuf_add_free_page();
+
+			if (err == DB_STRONG_FAIL) {
+
+				return(err);
+			}
+
+			mutex_enter(&ibuf_pessimistic_insert_mutex);
+
+			ibuf_enter();
+
+			mutex_enter(&ibuf_mutex);
+		}
+	} else {
+		ibuf_enter();
+	}
+
+	heap = mem_heap_create(512);
+
+	/* Build the entry which contains the space id and the page number as
+	the first fields and the type information for other fields, and which
+	will be inserted to the insert buffer. */
+
+	ibuf_entry = ibuf_entry_build(index, entry, space, page_no, heap);
+
+	/* Open a cursor to the insert buffer tree to calculate if we can add
+	the new entry to it without exceeding the free space limit for the
+	page. */
+
+	mtr_start(&mtr);
+
+	btr_pcur_open(ibuf->index, ibuf_entry, PAGE_CUR_LE, mode, &pcur, &mtr);
+
+	/* Find out the volume of already buffered inserts for the same index
+	page */
+	buffered = ibuf_get_volume_buffered(&pcur, space, page_no, &mtr);
+
+#ifdef UNIV_IBUF_COUNT_DEBUG
+	ut_a((buffered == 0) || ibuf_count_get(space, page_no));
+#endif
+	mtr_start(&bitmap_mtr);
+
+	bitmap_page = ibuf_bitmap_get_map_page(space, page_no,
+					       zip_size, &bitmap_mtr);
+
+	/* We check if the index page is suitable for buffered entries */
+
+	if (buf_page_peek(space, page_no)
+	    || lock_rec_expl_exist_on_page(space, page_no)) {
+		err = DB_STRONG_FAIL;
+
+		mtr_commit(&bitmap_mtr);
+
+		goto function_exit;
+	}
+
+	bits = ibuf_bitmap_page_get_bits(bitmap_page, page_no, zip_size,
+					 IBUF_BITMAP_FREE, &bitmap_mtr);
+
+	if (buffered + entry_size + page_dir_calc_reserved_space(1)
+	    > ibuf_index_page_calc_free_from_bits(zip_size, bits)) {
+		mtr_commit(&bitmap_mtr);
+
+		/* It may not fit */
+		err = DB_STRONG_FAIL;
+
+		do_merge = TRUE;
+
+		ibuf_get_merge_page_nos(FALSE, btr_pcur_get_rec(&pcur),
+					space_ids, space_versions,
+					page_nos, &n_stored);
+		goto function_exit;
+	}
+
+	/* Set the bitmap bit denoting that the insert buffer contains
+	buffered entries for this index page, if the bit is not set yet */
+
+	old_bit_value = ibuf_bitmap_page_get_bits(
+		bitmap_page, page_no, zip_size,
+		IBUF_BITMAP_BUFFERED, &bitmap_mtr);
+
+	if (!old_bit_value) {
+		ibuf_bitmap_page_set_bits(bitmap_page, page_no, zip_size,
+					  IBUF_BITMAP_BUFFERED, TRUE,
+					  &bitmap_mtr);
+	}
+
+	mtr_commit(&bitmap_mtr);
+
+	cursor = btr_pcur_get_btr_cur(&pcur);
+
+	if (mode == BTR_MODIFY_PREV) {
+		err = btr_cur_optimistic_insert(BTR_NO_LOCKING_FLAG, cursor,
+						ibuf_entry, &ins_rec,
+						&dummy_big_rec, 0, thr, &mtr);
+		if (err == DB_SUCCESS) {
+			/* Update the page max trx id field */
+			page_update_max_trx_id(btr_cur_get_block(cursor), NULL,
+					       thr_get_trx(thr)->id);
+		}
+	} else {
+		ut_ad(mode == BTR_MODIFY_TREE);
+
+		/* We acquire an x-latch to the root page before the insert,
+		because a pessimistic insert releases the tree x-latch,
+		which would cause the x-latching of the root after that to
+		break the latching order. */
+
+		root = ibuf_tree_root_get(&mtr);
+
+		err = btr_cur_pessimistic_insert(BTR_NO_LOCKING_FLAG
+						 | BTR_NO_UNDO_LOG_FLAG,
+						 cursor,
+						 ibuf_entry, &ins_rec,
+						 &dummy_big_rec, 0, thr, &mtr);
+		if (err == DB_SUCCESS) {
+			/* Update the page max trx id field */
+			page_update_max_trx_id(btr_cur_get_block(cursor), NULL,
+					       thr_get_trx(thr)->id);
+		}
+
+		ibuf_size_update(root, &mtr);
+	}
+
+function_exit:
+#ifdef UNIV_IBUF_COUNT_DEBUG
+	if (err == DB_SUCCESS) {
+		fprintf(stderr,
+			"Incrementing ibuf count of space %lu page %lu\n"
+			"from %lu by 1\n", space, page_no,
+			ibuf_count_get(space, page_no));
+
+		ibuf_count_set(space, page_no,
+			       ibuf_count_get(space, page_no) + 1);
+	}
+#endif
+	if (mode == BTR_MODIFY_TREE) {
+
+		mutex_exit(&ibuf_mutex);
+		mutex_exit(&ibuf_pessimistic_insert_mutex);
+	}
+
+	mtr_commit(&mtr);
+	btr_pcur_close(&pcur);
+	ibuf_exit();
+
+	mem_heap_free(heap);
+
+	if (err == DB_SUCCESS) {
+		mutex_enter(&ibuf_mutex);
+
+		ibuf->empty = FALSE;
+		ibuf->n_inserts++;
+
+		mutex_exit(&ibuf_mutex);
+
+		if (mode == BTR_MODIFY_TREE) {
+			ibuf_contract_after_insert(entry_size);
+		}
+	}
+
+	if (do_merge) {
+#ifdef UNIV_IBUF_DEBUG
+		ut_a(n_stored <= IBUF_MAX_N_PAGES_MERGED);
+#endif
+		buf_read_ibuf_merge_pages(FALSE, space_ids, space_versions,
+					  page_nos, n_stored);
+	}
+
+	return(err);
+}
+
+/*************************************************************************
+Makes an index insert to the insert buffer, instead of directly to the disk
+page, if this is possible. Does not do insert if the index is clustered
+or unique. */
+UNIV_INTERN
+ibool
+ibuf_insert(
+/*========*/
+				/* out: TRUE if success */
+	const dtuple_t*	entry,	/* in: index entry to insert */
+	dict_index_t*	index,	/* in: index where to insert */
+	ulint		space,	/* in: space id where to insert */
+	ulint		zip_size,/* in: compressed page size in bytes, or 0 */
+	ulint		page_no,/* in: page number where to insert */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	ulint	err;
+	ulint	entry_size;
+
+	ut_a(trx_sys_multiple_tablespace_format);
+	ut_ad(dtuple_check_typed(entry));
+	ut_ad(ut_is_2pow(zip_size));
+
+	ut_a(!dict_index_is_clust(index));
+
+	switch (UNIV_EXPECT(ibuf_use, IBUF_USE_INSERT)) {
+	case IBUF_USE_NONE:
+		return(FALSE);
+	case IBUF_USE_INSERT:
+		goto do_insert;
+	case IBUF_USE_COUNT:
+		break;
+	}
+
+	ut_error; /* unknown value of ibuf_use */
+
+do_insert:
+	entry_size = rec_get_converted_size(index, entry, 0);
+
+	if (entry_size
+	    >= (page_get_free_space_of_empty(dict_table_is_comp(index->table))
+		/ 2)) {
+		return(FALSE);
+	}
+
+	err = ibuf_insert_low(BTR_MODIFY_PREV, entry, entry_size,
+			      index, space, zip_size, page_no, thr);
+	if (err == DB_FAIL) {
+		err = ibuf_insert_low(BTR_MODIFY_TREE, entry, entry_size,
+				      index, space, zip_size, page_no, thr);
+	}
+
+	if (err == DB_SUCCESS) {
+#ifdef UNIV_IBUF_DEBUG
+		/* fprintf(stderr, "Ibuf insert for page no %lu of index %s\n",
+		page_no, index->name); */
+#endif
+		return(TRUE);
+
+	} else {
+		ut_a(err == DB_STRONG_FAIL);
+
+		return(FALSE);
+	}
+}
+
+/************************************************************************
+During merge, inserts to an index page a secondary index entry extracted
+from the insert buffer. */
+static
+void
+ibuf_insert_to_index_page(
+/*======================*/
+	dtuple_t*	entry,	/* in: buffered entry to insert */
+	buf_block_t*	block,	/* in/out: index page where the buffered entry
+				should be placed */
+	dict_index_t*	index,	/* in: record descriptor */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	page_cur_t	page_cur;
+	ulint		low_match;
+	page_t*		page		= buf_block_get_frame(block);
+	rec_t*		rec;
+	page_t*		bitmap_page;
+	ulint		old_bits;
+
+	ut_ad(ibuf_inside());
+	ut_ad(dtuple_check_typed(entry));
+
+	if (UNIV_UNLIKELY(dict_table_is_comp(index->table)
+			  != (ibool)!!page_is_comp(page))) {
+		fputs("InnoDB: Trying to insert a record from"
+		      " the insert buffer to an index page\n"
+		      "InnoDB: but the 'compact' flag does not match!\n",
+		      stderr);
+		goto dump;
+	}
+
+	rec = page_rec_get_next(page_get_infimum_rec(page));
+
+	if (UNIV_UNLIKELY(rec_get_n_fields(rec, index)
+			  != dtuple_get_n_fields(entry))) {
+		fputs("InnoDB: Trying to insert a record from"
+		      " the insert buffer to an index page\n"
+		      "InnoDB: but the number of fields does not match!\n",
+		      stderr);
+dump:
+		buf_page_print(page, 0);
+
+		dtuple_print(stderr, entry);
+
+		fputs("InnoDB: The table where where"
+		      " this index record belongs\n"
+		      "InnoDB: is now probably corrupt."
+		      " Please run CHECK TABLE on\n"
+		      "InnoDB: your tables.\n"
+		      "InnoDB: Submit a detailed bug report to"
+		      " http://bugs.mysql.com!\n", stderr);
+
+		return;
+	}
+
+	low_match = page_cur_search(block, index, entry,
+				    PAGE_CUR_LE, &page_cur);
+
+	if (low_match == dtuple_get_n_fields(entry)) {
+		page_zip_des_t*	page_zip;
+
+		rec = page_cur_get_rec(&page_cur);
+		page_zip = buf_block_get_page_zip(block);
+
+		btr_cur_del_unmark_for_ibuf(rec, page_zip, mtr);
+	} else {
+		rec = page_cur_tuple_insert(&page_cur, entry, index, 0, mtr);
+
+		if (UNIV_LIKELY(rec != NULL)) {
+			return;
+		}
+
+		/* If the record did not fit, reorganize */
+
+		btr_page_reorganize(block, index, mtr);
+		page_cur_search(block, index, entry, PAGE_CUR_LE, &page_cur);
+
+		/* This time the record must fit */
+		if (UNIV_UNLIKELY
+		    (!page_cur_tuple_insert(&page_cur, entry, index,
+					    0, mtr))) {
+			ulint	space;
+			ulint	page_no;
+			ulint	zip_size;
+
+			ut_print_timestamp(stderr);
+
+			fprintf(stderr,
+				"  InnoDB: Error: Insert buffer insert"
+				" fails; page free %lu,"
+				" dtuple size %lu\n",
+				(ulong) page_get_max_insert_size(
+					page, 1),
+				(ulong) rec_get_converted_size(
+					index, entry, 0));
+			fputs("InnoDB: Cannot insert index record ",
+			      stderr);
+			dtuple_print(stderr, entry);
+			fputs("\nInnoDB: The table where"
+			      " this index record belongs\n"
+			      "InnoDB: is now probably corrupt."
+			      " Please run CHECK TABLE on\n"
+			      "InnoDB: that table.\n", stderr);
+
+			space = page_get_space_id(page);
+			zip_size = buf_block_get_zip_size(block);
+			page_no = page_get_page_no(page);
+
+			bitmap_page = ibuf_bitmap_get_map_page(
+				space, page_no, zip_size, mtr);
+			old_bits = ibuf_bitmap_page_get_bits(
+				bitmap_page, page_no, zip_size,
+				IBUF_BITMAP_FREE, mtr);
+
+			fprintf(stderr,
+				"InnoDB: space %lu, page %lu,"
+				" zip_size %lu, bitmap bits %lu\n",
+				(ulong) space, (ulong) page_no,
+				(ulong) zip_size, (ulong) old_bits);
+
+			fputs("InnoDB: Submit a detailed bug report"
+			      " to http://bugs.mysql.com\n", stderr);
+		}
+	}
+}
+
+/*************************************************************************
+Deletes from ibuf the record on which pcur is positioned. If we have to
+resort to a pessimistic delete, this function commits mtr and closes
+the cursor. */
+static
+ibool
+ibuf_delete_rec(
+/*============*/
+				/* out: TRUE if mtr was committed and pcur
+				closed in this operation */
+	ulint		space,	/* in: space id */
+	ulint		page_no,/* in: index page number where the record
+				should belong */
+	btr_pcur_t*	pcur,	/* in: pcur positioned on the record to
+				delete, having latch mode BTR_MODIFY_LEAF */
+	const dtuple_t*	search_tuple,
+				/* in: search tuple for entries of page_no */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	ibool		success;
+	page_t*		root;
+	ulint		err;
+
+	ut_ad(ibuf_inside());
+	ut_ad(page_rec_is_user_rec(btr_pcur_get_rec(pcur)));
+	ut_ad(ibuf_rec_get_page_no(btr_pcur_get_rec(pcur)) == page_no);
+	ut_ad(ibuf_rec_get_space(btr_pcur_get_rec(pcur)) == space);
+
+	success = btr_cur_optimistic_delete(btr_pcur_get_btr_cur(pcur), mtr);
+
+	if (success) {
+#ifdef UNIV_IBUF_COUNT_DEBUG
+		fprintf(stderr,
+			"Decrementing ibuf count of space %lu page %lu\n"
+			"from %lu by 1\n", space, page_no,
+			ibuf_count_get(space, page_no));
+		ibuf_count_set(space, page_no,
+			       ibuf_count_get(space, page_no) - 1);
+#endif
+		return(FALSE);
+	}
+
+	ut_ad(page_rec_is_user_rec(btr_pcur_get_rec(pcur)));
+	ut_ad(ibuf_rec_get_page_no(btr_pcur_get_rec(pcur)) == page_no);
+	ut_ad(ibuf_rec_get_space(btr_pcur_get_rec(pcur)) == space);
+
+	/* We have to resort to a pessimistic delete from ibuf */
+	btr_pcur_store_position(pcur, mtr);
+
+	btr_pcur_commit_specify_mtr(pcur, mtr);
+
+	mutex_enter(&ibuf_mutex);
+
+	mtr_start(mtr);
+
+	success = btr_pcur_restore_position(BTR_MODIFY_TREE, pcur, mtr);
+
+	if (!success) {
+		if (fil_space_get_flags(space) == ULINT_UNDEFINED) {
+			/* The tablespace has been dropped.  It is possible
+			that another thread has deleted the insert buffer
+			entry.  Do not complain. */
+			goto commit_and_exit;
+		}
+
+		fprintf(stderr,
+			"InnoDB: ERROR: Submit the output to"
+			" http://bugs.mysql.com\n"
+			"InnoDB: ibuf cursor restoration fails!\n"
+			"InnoDB: ibuf record inserted to page %lu\n",
+			(ulong) page_no);
+		fflush(stderr);
+
+		rec_print_old(stderr, btr_pcur_get_rec(pcur));
+		rec_print_old(stderr, pcur->old_rec);
+		dtuple_print(stderr, search_tuple);
+
+		rec_print_old(stderr,
+			      page_rec_get_next(btr_pcur_get_rec(pcur)));
+		fflush(stderr);
+
+		btr_pcur_commit_specify_mtr(pcur, mtr);
+
+		fputs("InnoDB: Validating insert buffer tree:\n", stderr);
+		if (!btr_validate_index(ibuf->index, NULL)) {
+			ut_error;
+		}
+
+		fprintf(stderr, "InnoDB: ibuf tree ok\n");
+		fflush(stderr);
+
+		goto func_exit;
+	}
+
+	root = ibuf_tree_root_get(mtr);
+
+	btr_cur_pessimistic_delete(&err, TRUE, btr_pcur_get_btr_cur(pcur),
+				   RB_NONE, mtr);
+	ut_a(err == DB_SUCCESS);
+
+#ifdef UNIV_IBUF_COUNT_DEBUG
+	ibuf_count_set(space, page_no, ibuf_count_get(space, page_no) - 1);
+#endif
+	ibuf_size_update(root, mtr);
+
+commit_and_exit:
+	btr_pcur_commit_specify_mtr(pcur, mtr);
+
+func_exit:
+	btr_pcur_close(pcur);
+
+	mutex_exit(&ibuf_mutex);
+
+	return(TRUE);
+}
+
+/*************************************************************************
+When an index page is read from a disk to the buffer pool, this function
+inserts to the page the possible index entries buffered in the insert buffer.
+The entries are deleted from the insert buffer. If the page is not read, but
+created in the buffer pool, this function deletes its buffered entries from
+the insert buffer; there can exist entries for such a page if the page
+belonged to an index which subsequently was dropped. */
+UNIV_INTERN
+void
+ibuf_merge_or_delete_for_page(
+/*==========================*/
+	buf_block_t*	block,	/* in: if page has been read from
+				disk, pointer to the page x-latched,
+				else NULL */
+	ulint		space,	/* in: space id of the index page */
+	ulint		page_no,/* in: page number of the index page */
+	ulint		zip_size,/* in: compressed page size in bytes,
+				or 0 */
+	ibool		update_ibuf_bitmap)/* in: normally this is set
+				to TRUE, but if we have deleted or are
+				deleting the tablespace, then we
+				naturally do not want to update a
+				non-existent bitmap page */
+{
+	mem_heap_t*	heap;
+	btr_pcur_t	pcur;
+	dtuple_t*	search_tuple;
+	ulint		n_inserts;
+#ifdef UNIV_IBUF_DEBUG
+	ulint		volume;
+#endif
+	page_zip_des_t*	page_zip		= NULL;
+	ibool		tablespace_being_deleted = FALSE;
+	ibool		corruption_noticed	= FALSE;
+	mtr_t		mtr;
+
+	ut_ad(!block || buf_block_get_space(block) == space);
+	ut_ad(!block || buf_block_get_page_no(block) == page_no);
+	ut_ad(!block || buf_block_get_zip_size(block) == zip_size);
+
+	if (srv_force_recovery >= SRV_FORCE_NO_IBUF_MERGE
+	    || trx_sys_hdr_page(space, page_no)) {
+		return;
+	}
+
+	/* We cannot refer to zip_size in the following, because
+	zip_size is passed as ULINT_UNDEFINED (it is unknown) when
+	buf_read_ibuf_merge_pages() is merging (discarding) changes
+	for a dropped tablespace.  When block != NULL or
+	update_ibuf_bitmap is specified, the zip_size must be known.
+	That is why we will repeat the check below, with zip_size in
+	place of 0.  Passing zip_size as 0 assumes that the
+	uncompressed page size always is a power-of-2 multiple of the
+	compressed page size. */
+
+	if (ibuf_fixed_addr_page(space, 0, page_no)
+	    || fsp_descr_page(0, page_no)) {
+		return;
+	}
+
+	if (UNIV_LIKELY(update_ibuf_bitmap)) {
+		ut_a(ut_is_2pow(zip_size));
+
+		if (ibuf_fixed_addr_page(space, zip_size, page_no)
+		    || fsp_descr_page(zip_size, page_no)) {
+			return;
+		}
+
+		/* If the following returns FALSE, we get the counter
+		incremented, and must decrement it when we leave this
+		function. When the counter is > 0, that prevents tablespace
+		from being dropped. */
+
+		tablespace_being_deleted = fil_inc_pending_ibuf_merges(space);
+
+		if (UNIV_UNLIKELY(tablespace_being_deleted)) {
+			/* Do not try to read the bitmap page from space;
+			just delete the ibuf records for the page */
+
+			block = NULL;
+			update_ibuf_bitmap = FALSE;
+		} else {
+			page_t*	bitmap_page;
+
+			mtr_start(&mtr);
+
+			bitmap_page = ibuf_bitmap_get_map_page(
+				space, page_no, zip_size, &mtr);
+
+			if (!ibuf_bitmap_page_get_bits(bitmap_page, page_no,
+						       zip_size,
+						       IBUF_BITMAP_BUFFERED,
+						       &mtr)) {
+				/* No inserts buffered for this page */
+				mtr_commit(&mtr);
+
+				if (!tablespace_being_deleted) {
+					fil_decr_pending_ibuf_merges(space);
+				}
+
+				return;
+			}
+			mtr_commit(&mtr);
+		}
+	} else if (block
+		   && (ibuf_fixed_addr_page(space, zip_size, page_no)
+		      || fsp_descr_page(zip_size, page_no))) {
+
+		return;
+	}
+
+	ibuf_enter();
+
+	heap = mem_heap_create(512);
+
+	if (!trx_sys_multiple_tablespace_format) {
+		ut_a(trx_doublewrite_must_reset_space_ids);
+		search_tuple = ibuf_search_tuple_build(space, page_no, heap);
+	} else {
+		search_tuple = ibuf_new_search_tuple_build(space, page_no,
+							   heap);
+	}
+
+	if (block) {
+		/* Move the ownership of the x-latch on the page to this OS
+		thread, so that we can acquire a second x-latch on it. This
+		is needed for the insert operations to the index page to pass
+		the debug checks. */
+
+		rw_lock_x_lock_move_ownership(&(block->lock));
+		page_zip = buf_block_get_page_zip(block);
+
+		if (UNIV_UNLIKELY(fil_page_get_type(block->frame)
+				  != FIL_PAGE_INDEX)
+		    || UNIV_UNLIKELY(!page_is_leaf(block->frame))) {
+
+			page_t*	bitmap_page;
+
+			corruption_noticed = TRUE;
+
+			ut_print_timestamp(stderr);
+
+			mtr_start(&mtr);
+
+			fputs("  InnoDB: Dump of the ibuf bitmap page:\n",
+			      stderr);
+
+			bitmap_page = ibuf_bitmap_get_map_page(space, page_no,
+							       zip_size, &mtr);
+			buf_page_print(bitmap_page, 0);
+
+			mtr_commit(&mtr);
+
+			fputs("\nInnoDB: Dump of the page:\n", stderr);
+
+			buf_page_print(block->frame, 0);
+
+			fprintf(stderr,
+				"InnoDB: Error: corruption in the tablespace."
+				" Bitmap shows insert\n"
+				"InnoDB: buffer records to page n:o %lu"
+				" though the page\n"
+				"InnoDB: type is %lu, which is"
+				" not an index leaf page!\n"
+				"InnoDB: We try to resolve the problem"
+				" by skipping the insert buffer\n"
+				"InnoDB: merge for this page."
+				" Please run CHECK TABLE on your tables\n"
+				"InnoDB: to determine if they are corrupt"
+				" after this.\n\n"
+				"InnoDB: Please submit a detailed bug report"
+				" to http://bugs.mysql.com\n\n",
+				(ulong) page_no,
+				(ulong)
+				fil_page_get_type(block->frame));
+		}
+	}
+
+	n_inserts = 0;
+#ifdef UNIV_IBUF_DEBUG
+	volume = 0;
+#endif
+loop:
+	mtr_start(&mtr);
+
+	if (block) {
+		ibool success;
+
+		success = buf_page_get_known_nowait(
+			RW_X_LATCH, block,
+			BUF_KEEP_OLD, __FILE__, __LINE__, &mtr);
+
+		ut_a(success);
+
+		buf_block_dbg_add_level(block, SYNC_TREE_NODE);
+	}
+
+	/* Position pcur in the insert buffer at the first entry for this
+	index page */
+	btr_pcur_open_on_user_rec(
+		ibuf->index, search_tuple, PAGE_CUR_GE, BTR_MODIFY_LEAF,
+		&pcur, &mtr);
+
+	if (!btr_pcur_is_on_user_rec(&pcur)) {
+		ut_ad(btr_pcur_is_after_last_in_tree(&pcur, &mtr));
+
+		goto reset_bit;
+	}
+
+	for (;;) {
+		rec_t*	rec;
+
+		ut_ad(btr_pcur_is_on_user_rec(&pcur));
+
+		rec = btr_pcur_get_rec(&pcur);
+
+		/* Check if the entry is for this index page */
+		if (ibuf_rec_get_page_no(rec) != page_no
+		    || ibuf_rec_get_space(rec) != space) {
+
+			if (block) {
+				page_header_reset_last_insert(
+					block->frame, page_zip, &mtr);
+			}
+
+			goto reset_bit;
+		}
+
+		if (UNIV_UNLIKELY(corruption_noticed)) {
+			fputs("InnoDB: Discarding record\n ", stderr);
+			rec_print_old(stderr, rec);
+			fputs("\nInnoDB: from the insert buffer!\n\n", stderr);
+		} else if (block) {
+			/* Now we have at pcur a record which should be
+			inserted to the index page; NOTE that the call below
+			copies pointers to fields in rec, and we must
+			keep the latch to the rec page until the
+			insertion is finished! */
+			dtuple_t*	entry;
+			dulint		max_trx_id;
+			dict_index_t*	dummy_index;
+
+			max_trx_id = page_get_max_trx_id(page_align(rec));
+			page_update_max_trx_id(block, page_zip, max_trx_id);
+
+			entry = ibuf_build_entry_from_ibuf_rec(
+				rec, heap, &dummy_index);
+#ifdef UNIV_IBUF_DEBUG
+			volume += rec_get_converted_size(dummy_index, entry, 0)
+				+ page_dir_calc_reserved_space(1);
+			ut_a(volume <= 4 * UNIV_PAGE_SIZE
+			     / IBUF_PAGE_SIZE_PER_FREE_SPACE);
+#endif
+			ibuf_insert_to_index_page(entry, block,
+						  dummy_index, &mtr);
+			ibuf_dummy_index_free(dummy_index);
+		}
+
+		n_inserts++;
+
+		/* Delete the record from ibuf */
+		if (ibuf_delete_rec(space, page_no, &pcur, search_tuple,
+				    &mtr)) {
+			/* Deletion was pessimistic and mtr was committed:
+			we start from the beginning again */
+
+			goto loop;
+		} else if (btr_pcur_is_after_last_on_page(&pcur)) {
+			mtr_commit(&mtr);
+			btr_pcur_close(&pcur);
+
+			goto loop;
+		}
+	}
+
+reset_bit:
+#ifdef UNIV_IBUF_COUNT_DEBUG
+	if (ibuf_count_get(space, page_no) > 0) {
+		/* btr_print_tree(ibuf_data->index->tree, 100);
+		ibuf_print(); */
+	}
+#endif
+	if (UNIV_LIKELY(update_ibuf_bitmap)) {
+		page_t*	bitmap_page;
+
+		bitmap_page = ibuf_bitmap_get_map_page(
+			space, page_no, zip_size, &mtr);
+
+		ibuf_bitmap_page_set_bits(
+			bitmap_page, page_no, zip_size,
+			IBUF_BITMAP_BUFFERED, FALSE, &mtr);
+
+		if (block) {
+			ulint old_bits = ibuf_bitmap_page_get_bits(
+				bitmap_page, page_no, zip_size,
+				IBUF_BITMAP_FREE, &mtr);
+
+			ulint new_bits = ibuf_index_page_calc_free(
+				zip_size, block);
+
+			if (old_bits != new_bits) {
+				ibuf_bitmap_page_set_bits(
+					bitmap_page, page_no, zip_size,
+					IBUF_BITMAP_FREE, new_bits, &mtr);
+			}
+		}
+	}
+
+	mtr_commit(&mtr);
+	btr_pcur_close(&pcur);
+	mem_heap_free(heap);
+
+	/* Protect our statistics keeping from race conditions */
+	mutex_enter(&ibuf_mutex);
+
+	ibuf->n_merges++;
+	ibuf->n_merged_recs += n_inserts;
+
+	mutex_exit(&ibuf_mutex);
+
+	if (update_ibuf_bitmap && !tablespace_being_deleted) {
+
+		fil_decr_pending_ibuf_merges(space);
+	}
+
+	ibuf_exit();
+
+#ifdef UNIV_IBUF_COUNT_DEBUG
+	ut_a(ibuf_count_get(space, page_no) == 0);
+#endif
+}
+
+/*************************************************************************
+Deletes all entries in the insert buffer for a given space id. This is used
+in DISCARD TABLESPACE and IMPORT TABLESPACE.
+NOTE: this does not update the page free bitmaps in the space. The space will
+become CORRUPT when you call this function! */
+UNIV_INTERN
+void
+ibuf_delete_for_discarded_space(
+/*============================*/
+	ulint	space)	/* in: space id */
+{
+	mem_heap_t*	heap;
+	btr_pcur_t	pcur;
+	dtuple_t*	search_tuple;
+	rec_t*		ibuf_rec;
+	ulint		page_no;
+	ibool		closed;
+	ulint		n_inserts;
+	mtr_t		mtr;
+
+	heap = mem_heap_create(512);
+
+	/* Use page number 0 to build the search tuple so that we get the
+	cursor positioned at the first entry for this space id */
+
+	search_tuple = ibuf_new_search_tuple_build(space, 0, heap);
+
+	n_inserts = 0;
+loop:
+	ibuf_enter();
+
+	mtr_start(&mtr);
+
+	/* Position pcur in the insert buffer at the first entry for the
+	space */
+	btr_pcur_open_on_user_rec(
+		ibuf->index, search_tuple, PAGE_CUR_GE, BTR_MODIFY_LEAF,
+		&pcur, &mtr);
+
+	if (!btr_pcur_is_on_user_rec(&pcur)) {
+		ut_ad(btr_pcur_is_after_last_in_tree(&pcur, &mtr));
+
+		goto leave_loop;
+	}
+
+	for (;;) {
+		ut_ad(btr_pcur_is_on_user_rec(&pcur));
+
+		ibuf_rec = btr_pcur_get_rec(&pcur);
+
+		/* Check if the entry is for this space */
+		if (ibuf_rec_get_space(ibuf_rec) != space) {
+
+			goto leave_loop;
+		}
+
+		page_no = ibuf_rec_get_page_no(ibuf_rec);
+
+		n_inserts++;
+
+		/* Delete the record from ibuf */
+		closed = ibuf_delete_rec(space, page_no, &pcur, search_tuple,
+					 &mtr);
+		if (closed) {
+			/* Deletion was pessimistic and mtr was committed:
+			we start from the beginning again */
+
+			ibuf_exit();
+
+			goto loop;
+		}
+
+		if (btr_pcur_is_after_last_on_page(&pcur)) {
+			mtr_commit(&mtr);
+			btr_pcur_close(&pcur);
+
+			ibuf_exit();
+
+			goto loop;
+		}
+	}
+
+leave_loop:
+	mtr_commit(&mtr);
+	btr_pcur_close(&pcur);
+
+	/* Protect our statistics keeping from race conditions */
+	mutex_enter(&ibuf_mutex);
+
+	ibuf->n_merges++;
+	ibuf->n_merged_recs += n_inserts;
+
+	mutex_exit(&ibuf_mutex);
+
+	ibuf_exit();
+
+	mem_heap_free(heap);
+}
+
+/**********************************************************************
+Looks if the insert buffer is empty. */
+UNIV_INTERN
+ibool
+ibuf_is_empty(void)
+/*===============*/
+			/* out: TRUE if empty */
+{
+	ibool		is_empty;
+	const page_t*	root;
+	mtr_t		mtr;
+
+	ibuf_enter();
+
+	mutex_enter(&ibuf_mutex);
+
+	mtr_start(&mtr);
+
+	root = ibuf_tree_root_get(&mtr);
+
+	if (page_get_n_recs(root) == 0) {
+
+		is_empty = TRUE;
+
+		if (ibuf->empty == FALSE) {
+			fprintf(stderr,
+				"InnoDB: Warning: insert buffer tree is empty"
+				" but the data struct does not\n"
+				"InnoDB: know it. This condition is legal"
+				" if the master thread has not yet\n"
+				"InnoDB: run to completion.\n");
+		}
+	} else {
+		ut_a(ibuf->empty == FALSE);
+
+		is_empty = FALSE;
+	}
+
+	mtr_commit(&mtr);
+
+	mutex_exit(&ibuf_mutex);
+
+	ibuf_exit();
+
+	return(is_empty);
+}
+
+/**********************************************************************
+Prints info of ibuf. */
+UNIV_INTERN
+void
+ibuf_print(
+/*=======*/
+	FILE*	file)	/* in: file where to print */
+{
+#ifdef UNIV_IBUF_COUNT_DEBUG
+	ulint		i;
+	ulint		j;
+#endif
+
+	mutex_enter(&ibuf_mutex);
+
+	fprintf(file,
+		"Ibuf: size %lu, free list len %lu, seg size %lu,\n"
+		"%lu inserts, %lu merged recs, %lu merges\n",
+		(ulong) ibuf->size,
+		(ulong) ibuf->free_list_len,
+		(ulong) ibuf->seg_size,
+		(ulong) ibuf->n_inserts,
+		(ulong) ibuf->n_merged_recs,
+		(ulong) ibuf->n_merges);
+#ifdef UNIV_IBUF_COUNT_DEBUG
+	for (i = 0; i < IBUF_COUNT_N_SPACES; i++) {
+		for (j = 0; j < IBUF_COUNT_N_PAGES; j++) {
+			ulint	count = ibuf_count_get(i, j);
+
+			if (count > 0) {
+				fprintf(stderr,
+					"Ibuf count for space/page %lu/%lu"
+					" is %lu\n",
+					(ulong) i, (ulong) j, (ulong) count);
+			}
+		}
+	}
+#endif /* UNIV_IBUF_COUNT_DEBUG */
+
+	mutex_exit(&ibuf_mutex);
+}
diff --git a/storage/xtradb/include/btr0btr.h b/storage/xtradb/include/btr0btr.h
new file mode 100644
index 00000000000..298942bd542
--- /dev/null
+++ b/storage/xtradb/include/btr0btr.h
@@ -0,0 +1,494 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The B-tree
+
+Created 6/2/1994 Heikki Tuuri
+*******************************************************/
+
+#ifndef btr0btr_h
+#define btr0btr_h
+
+#include "univ.i"
+
+#include "dict0dict.h"
+#include "data0data.h"
+#include "page0cur.h"
+#include "mtr0mtr.h"
+#include "btr0types.h"
+
+/* Maximum record size which can be stored on a page, without using the
+special big record storage structure */
+
+#define	BTR_PAGE_MAX_REC_SIZE	(UNIV_PAGE_SIZE / 2 - 200)
+
+/* Maximum depth of a B-tree in InnoDB. Note that this isn't a maximum as
+such; none of the tree operations avoid producing trees bigger than this. It
+is instead a "max depth that other code must work with", useful for e.g.
+fixed-size arrays that must store some information about each level in a
+tree. In other words: if a B-tree with bigger depth than this is
+encountered, it is not acceptable for it to lead to mysterious memory
+corruption, but it is acceptable for the program to die with a clear assert
+failure. */
+#define BTR_MAX_LEVELS		100
+
+/* Latching modes for btr_cur_search_to_nth_level(). */
+#define BTR_SEARCH_LEAF		RW_S_LATCH
+#define BTR_MODIFY_LEAF		RW_X_LATCH
+#define BTR_NO_LATCHES		RW_NO_LATCH
+#define	BTR_MODIFY_TREE		33
+#define	BTR_CONT_MODIFY_TREE	34
+#define	BTR_SEARCH_PREV		35
+#define	BTR_MODIFY_PREV		36
+
+/* If this is ORed to the latch mode, it means that the search tuple will be
+inserted to the index, at the searched position */
+#define BTR_INSERT		512
+
+/* This flag ORed to latch mode says that we do the search in query
+optimization */
+#define BTR_ESTIMATE		1024
+
+/* This flag ORed to latch mode says that we can ignore possible
+UNIQUE definition on secondary indexes when we decide if we can use the
+insert buffer to speed up inserts */
+#define BTR_IGNORE_SEC_UNIQUE	2048
+
+/******************************************************************
+Gets the root node of a tree and x-latches it. */
+UNIV_INTERN
+page_t*
+btr_root_get(
+/*=========*/
+				/* out: root page, x-latched */
+	dict_index_t*	index,	/* in: index tree */
+	mtr_t*		mtr);	/* in: mtr */
+/******************************************************************
+Gets a buffer page and declares its latching order level. */
+UNIV_INLINE
+buf_block_t*
+btr_block_get(
+/*==========*/
+	ulint	space,		/* in: space id */
+	ulint	zip_size,	/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	ulint	page_no,	/* in: page number */
+	ulint	mode,		/* in: latch mode */
+	mtr_t*	mtr);		/* in: mtr */
+/******************************************************************
+Gets a buffer page and declares its latching order level. */
+UNIV_INLINE
+page_t*
+btr_page_get(
+/*=========*/
+	ulint	space,		/* in: space id */
+	ulint	zip_size,	/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	ulint	page_no,	/* in: page number */
+	ulint	mode,		/* in: latch mode */
+	mtr_t*	mtr);		/* in: mtr */
+/******************************************************************
+Gets the index id field of a page. */
+UNIV_INLINE
+dulint
+btr_page_get_index_id(
+/*==================*/
+				/* out: index id */
+	const page_t*	page);	/* in: index page */
+/************************************************************
+Gets the node level field in an index page. */
+UNIV_INLINE
+ulint
+btr_page_get_level_low(
+/*===================*/
+				/* out: level, leaf level == 0 */
+	const page_t*	page);	/* in: index page */
+/************************************************************
+Gets the node level field in an index page. */
+UNIV_INLINE
+ulint
+btr_page_get_level(
+/*===============*/
+				/* out: level, leaf level == 0 */
+	const page_t*	page,	/* in: index page */
+	mtr_t*		mtr);	/* in: mini-transaction handle */
+/************************************************************
+Gets the next index page number. */
+UNIV_INLINE
+ulint
+btr_page_get_next(
+/*==============*/
+				/* out: next page number */
+	const page_t*	page,	/* in: index page */
+	mtr_t*		mtr);	/* in: mini-transaction handle */
+/************************************************************
+Gets the previous index page number. */
+UNIV_INLINE
+ulint
+btr_page_get_prev(
+/*==============*/
+				/* out: prev page number */
+	const page_t*	page,	/* in: index page */
+	mtr_t*		mtr);	/* in: mini-transaction handle */
+/*****************************************************************
+Gets pointer to the previous user record in the tree. It is assumed
+that the caller has appropriate latches on the page and its neighbor. */
+UNIV_INTERN
+rec_t*
+btr_get_prev_user_rec(
+/*==================*/
+			/* out: previous user record, NULL if there is none */
+	rec_t*	rec,	/* in: record on leaf level */
+	mtr_t*	mtr);	/* in: mtr holding a latch on the page, and if
+			needed, also to the previous page */
+/*****************************************************************
+Gets pointer to the next user record in the tree. It is assumed
+that the caller has appropriate latches on the page and its neighbor. */
+UNIV_INTERN
+rec_t*
+btr_get_next_user_rec(
+/*==================*/
+			/* out: next user record, NULL if there is none */
+	rec_t*	rec,	/* in: record on leaf level */
+	mtr_t*	mtr);	/* in: mtr holding a latch on the page, and if
+			needed, also to the next page */
+/******************************************************************
+Releases the latch on a leaf page and bufferunfixes it. */
+UNIV_INLINE
+void
+btr_leaf_page_release(
+/*==================*/
+	buf_block_t*	block,		/* in: buffer block */
+	ulint		latch_mode,	/* in: BTR_SEARCH_LEAF or
+					BTR_MODIFY_LEAF */
+	mtr_t*		mtr);		/* in: mtr */
+/******************************************************************
+Gets the child node file address in a node pointer. */
+UNIV_INLINE
+ulint
+btr_node_ptr_get_child_page_no(
+/*===========================*/
+				/* out: child node address */
+	const rec_t*	rec,	/* in: node pointer record */
+	const ulint*	offsets);/* in: array returned by rec_get_offsets() */
+/****************************************************************
+Creates the root node for a new index tree. */
+UNIV_INTERN
+ulint
+btr_create(
+/*=======*/
+				/* out: page number of the created root,
+				FIL_NULL if did not succeed */
+	ulint		type,	/* in: type of the index */
+	ulint		space,	/* in: space where created */
+	ulint		zip_size,/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	dulint		index_id,/* in: index id */
+	dict_index_t*	index,	/* in: index */
+	mtr_t*		mtr);	/* in: mini-transaction handle */
+/****************************************************************
+Frees a B-tree except the root page, which MUST be freed after this
+by calling btr_free_root. */
+UNIV_INTERN
+void
+btr_free_but_not_root(
+/*==================*/
+	ulint	space,		/* in: space where created */
+	ulint	zip_size,	/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	ulint	root_page_no);	/* in: root page number */
+/****************************************************************
+Frees the B-tree root page. Other tree MUST already have been freed. */
+UNIV_INTERN
+void
+btr_free_root(
+/*==========*/
+	ulint	space,		/* in: space where created */
+	ulint	zip_size,	/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	ulint	root_page_no,	/* in: root page number */
+	mtr_t*	mtr);		/* in: a mini-transaction which has already
+				been started */
+/*****************************************************************
+Makes tree one level higher by splitting the root, and inserts
+the tuple. It is assumed that mtr contains an x-latch on the tree.
+NOTE that the operation of this function must always succeed,
+we cannot reverse it: therefore enough free disk space must be
+guaranteed to be available before this function is called. */
+UNIV_INTERN
+rec_t*
+btr_root_raise_and_insert(
+/*======================*/
+				/* out: inserted record */
+	btr_cur_t*	cursor,	/* in: cursor at which to insert: must be
+				on the root page; when the function returns,
+				the cursor is positioned on the predecessor
+				of the inserted record */
+	const dtuple_t*	tuple,	/* in: tuple to insert */
+	ulint		n_ext,	/* in: number of externally stored columns */
+	mtr_t*		mtr);	/* in: mtr */
+/*****************************************************************
+Reorganizes an index page.
+IMPORTANT: if btr_page_reorganize() is invoked on a compressed leaf
+page of a non-clustered index, the caller must update the insert
+buffer free bits in the same mini-transaction in such a way that the
+modification will be redo-logged. */
+UNIV_INTERN
+ibool
+btr_page_reorganize(
+/*================*/
+				/* out: TRUE on success, FALSE on failure */
+	buf_block_t*	block,	/* in: page to be reorganized */
+	dict_index_t*	index,	/* in: record descriptor */
+	mtr_t*		mtr);	/* in: mtr */
+/*****************************************************************
+Decides if the page should be split at the convergence point of
+inserts converging to left. */
+UNIV_INTERN
+ibool
+btr_page_get_split_rec_to_left(
+/*===========================*/
+				/* out: TRUE if split recommended */
+	btr_cur_t*	cursor,	/* in: cursor at which to insert */
+	rec_t**		split_rec);/* out: if split recommended,
+				the first record on upper half page,
+				or NULL if tuple should be first */
+/*****************************************************************
+Decides if the page should be split at the convergence point of
+inserts converging to right. */
+UNIV_INTERN
+ibool
+btr_page_get_split_rec_to_right(
+/*============================*/
+				/* out: TRUE if split recommended */
+	btr_cur_t*	cursor,	/* in: cursor at which to insert */
+	rec_t**		split_rec);/* out: if split recommended,
+				the first record on upper half page,
+				or NULL if tuple should be first */
+/*****************************************************************
+Splits an index page to halves and inserts the tuple. It is assumed
+that mtr holds an x-latch to the index tree. NOTE: the tree x-latch
+is released within this function! NOTE that the operation of this
+function must always succeed, we cannot reverse it: therefore
+enough free disk space must be guaranteed to be available before
+this function is called. */
+UNIV_INTERN
+rec_t*
+btr_page_split_and_insert(
+/*======================*/
+				/* out: inserted record; NOTE: the tree
+				x-latch is released! NOTE: 2 free disk
+				pages must be available! */
+	btr_cur_t*	cursor,	/* in: cursor at which to insert; when the
+				function returns, the cursor is positioned
+				on the predecessor of the inserted record */
+	const dtuple_t*	tuple,	/* in: tuple to insert */
+	ulint		n_ext,	/* in: number of externally stored columns */
+	mtr_t*		mtr);	/* in: mtr */
+/***********************************************************
+Inserts a data tuple to a tree on a non-leaf level. It is assumed
+that mtr holds an x-latch on the tree. */
+UNIV_INTERN
+void
+btr_insert_on_non_leaf_level(
+/*=========================*/
+	dict_index_t*	index,	/* in: index */
+	ulint		level,	/* in: level, must be > 0 */
+	dtuple_t*	tuple,	/* in: the record to be inserted */
+	mtr_t*		mtr);	/* in: mtr */
+/********************************************************************
+Sets a record as the predefined minimum record. */
+UNIV_INTERN
+void
+btr_set_min_rec_mark(
+/*=================*/
+	rec_t*	rec,	/* in/out: record */
+	mtr_t*	mtr);	/* in: mtr */
+/*****************************************************************
+Deletes on the upper level the node pointer to a page. */
+UNIV_INTERN
+void
+btr_node_ptr_delete(
+/*================*/
+	dict_index_t*	index,	/* in: index tree */
+	buf_block_t*	block,	/* in: page whose node pointer is deleted */
+	mtr_t*		mtr);	/* in: mtr */
+#ifdef UNIV_DEBUG
+/****************************************************************
+Checks that the node pointer to a page is appropriate. */
+UNIV_INTERN
+ibool
+btr_check_node_ptr(
+/*===============*/
+				/* out: TRUE */
+	dict_index_t*	index,	/* in: index tree */
+	buf_block_t*	block,	/* in: index page */
+	mtr_t*		mtr);	/* in: mtr */
+#endif /* UNIV_DEBUG */
+/*****************************************************************
+Tries to merge the page first to the left immediate brother if such a
+brother exists, and the node pointers to the current page and to the
+brother reside on the same page. If the left brother does not satisfy these
+conditions, looks at the right brother. If the page is the only one on that
+level lifts the records of the page to the father page, thus reducing the
+tree height. It is assumed that mtr holds an x-latch on the tree and on the
+page. If cursor is on the leaf level, mtr must also hold x-latches to
+the brothers, if they exist. */
+UNIV_INTERN
+ibool
+btr_compress(
+/*=========*/
+				/* out: TRUE on success */
+	btr_cur_t*	cursor,	/* in: cursor on the page to merge or lift;
+				the page must not be empty: in record delete
+				use btr_discard_page if the page would become
+				empty */
+	mtr_t*		mtr);	/* in: mtr */
+/*****************************************************************
+Discards a page from a B-tree. This is used to remove the last record from
+a B-tree page: the whole page must be removed at the same time. This cannot
+be used for the root page, which is allowed to be empty. */
+UNIV_INTERN
+void
+btr_discard_page(
+/*=============*/
+	btr_cur_t*	cursor,	/* in: cursor on the page to discard: not on
+				the root page */
+	mtr_t*		mtr);	/* in: mtr */
+/********************************************************************
+Parses the redo log record for setting an index record as the predefined
+minimum record. */
+UNIV_INTERN
+byte*
+btr_parse_set_min_rec_mark(
+/*=======================*/
+			/* out: end of log record or NULL */
+	byte*	ptr,	/* in: buffer */
+	byte*	end_ptr,/* in: buffer end */
+	ulint	comp,	/* in: nonzero=compact page format */
+	page_t*	page,	/* in: page or NULL */
+	mtr_t*	mtr);	/* in: mtr or NULL */
+/***************************************************************
+Parses a redo log record of reorganizing a page. */
+UNIV_INTERN
+byte*
+btr_parse_page_reorganize(
+/*======================*/
+				/* out: end of log record or NULL */
+	byte*		ptr,	/* in: buffer */
+	byte*		end_ptr,/* in: buffer end */
+	dict_index_t*	index,	/* in: record descriptor */
+	buf_block_t*	block,	/* in: page to be reorganized, or NULL */
+	mtr_t*		mtr);	/* in: mtr or NULL */
+/******************************************************************
+Gets the number of pages in a B-tree. */
+UNIV_INTERN
+ulint
+btr_get_size(
+/*=========*/
+				/* out: number of pages */
+	dict_index_t*	index,	/* in: index */
+	ulint		flag);	/* in: BTR_N_LEAF_PAGES or BTR_TOTAL_SIZE */
+/******************************************************************
+Allocates a new file page to be used in an index tree. NOTE: we assume
+that the caller has made the reservation for free extents! */
+UNIV_INTERN
+buf_block_t*
+btr_page_alloc(
+/*===========*/
+					/* out: new allocated block, x-latched;
+					NULL if out of space */
+	dict_index_t*	index,		/* in: index tree */
+	ulint		hint_page_no,	/* in: hint of a good page */
+	byte		file_direction,	/* in: direction where a possible
+					page split is made */
+	ulint		level,		/* in: level where the page is placed
+					in the tree */
+	mtr_t*		mtr);		/* in: mtr */
+/******************************************************************
+Frees a file page used in an index tree. NOTE: cannot free field external
+storage pages because the page must contain info on its level. */
+UNIV_INTERN
+void
+btr_page_free(
+/*==========*/
+	dict_index_t*	index,	/* in: index tree */
+	buf_block_t*	block,	/* in: block to be freed, x-latched */
+	mtr_t*		mtr);	/* in: mtr */
+/******************************************************************
+Frees a file page used in an index tree. Can be used also to BLOB
+external storage pages, because the page level 0 can be given as an
+argument. */
+UNIV_INTERN
+void
+btr_page_free_low(
+/*==============*/
+	dict_index_t*	index,	/* in: index tree */
+	buf_block_t*	block,	/* in: block to be freed, x-latched */
+	ulint		level,	/* in: page level */
+	mtr_t*		mtr);	/* in: mtr */
+#ifdef UNIV_BTR_PRINT
+/*****************************************************************
+Prints size info of a B-tree. */
+UNIV_INTERN
+void
+btr_print_size(
+/*===========*/
+	dict_index_t*	index);	/* in: index tree */
+/******************************************************************
+Prints directories and other info of all nodes in the index. */
+UNIV_INTERN
+void
+btr_print_index(
+/*============*/
+	dict_index_t*	index,	/* in: index */
+	ulint		width);	/* in: print this many entries from start
+				and end */
+#endif /* UNIV_BTR_PRINT */
+/****************************************************************
+Checks the size and number of fields in a record based on the definition of
+the index. */
+UNIV_INTERN
+ibool
+btr_index_rec_validate(
+/*===================*/
+						/* out: TRUE if ok */
+	const rec_t*		rec,		/* in: index record */
+	const dict_index_t*	index,		/* in: index */
+	ibool			dump_on_error);	/* in: TRUE if the function
+						should print hex dump of record
+						and page on error */
+/******************************************************************
+Checks the consistency of an index tree. */
+UNIV_INTERN
+ibool
+btr_validate_index(
+/*===============*/
+				/* out: TRUE if ok */
+	dict_index_t*	index,	/* in: index */
+	trx_t*		trx);	/* in: transaction or NULL */
+
+#define BTR_N_LEAF_PAGES	1
+#define BTR_TOTAL_SIZE		2
+
+#ifndef UNIV_NONINL
+#include "btr0btr.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/btr0btr.ic b/storage/xtradb/include/btr0btr.ic
new file mode 100644
index 00000000000..a8d934ecc87
--- /dev/null
+++ b/storage/xtradb/include/btr0btr.ic
@@ -0,0 +1,301 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The B-tree
+
+Created 6/2/1994 Heikki Tuuri
+*******************************************************/
+
+#include "mach0data.h"
+#include "mtr0mtr.h"
+#include "mtr0log.h"
+#include "page0zip.h"
+
+#define BTR_MAX_NODE_LEVEL	50	/* used in debug checking */
+
+/******************************************************************
+Gets a buffer page and declares its latching order level. */
+UNIV_INLINE
+buf_block_t*
+btr_block_get(
+/*==========*/
+	ulint	space,		/* in: space id */
+	ulint	zip_size,	/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	ulint	page_no,	/* in: page number */
+	ulint	mode,		/* in: latch mode */
+	mtr_t*	mtr)		/* in: mtr */
+{
+	buf_block_t*	block;
+
+	block = buf_page_get(space, zip_size, page_no, mode, mtr);
+
+	if (mode != RW_NO_LATCH) {
+
+		buf_block_dbg_add_level(block, SYNC_TREE_NODE);
+	}
+
+	return(block);
+}
+
+/******************************************************************
+Gets a buffer page and declares its latching order level. */
+UNIV_INLINE
+page_t*
+btr_page_get(
+/*=========*/
+	ulint	space,		/* in: space id */
+	ulint	zip_size,	/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	ulint	page_no,	/* in: page number */
+	ulint	mode,		/* in: latch mode */
+	mtr_t*	mtr)		/* in: mtr */
+{
+	return(buf_block_get_frame(btr_block_get(space, zip_size, page_no,
+						 mode, mtr)));
+}
+
+/******************************************************************
+Sets the index id field of a page. */
+UNIV_INLINE
+void
+btr_page_set_index_id(
+/*==================*/
+	page_t*		page,	/* in: page to be created */
+	page_zip_des_t*	page_zip,/* in: compressed page whose uncompressed
+				part will be updated, or NULL */
+	dulint		id,	/* in: index id */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	if (UNIV_LIKELY_NULL(page_zip)) {
+		mach_write_to_8(page + (PAGE_HEADER + PAGE_INDEX_ID), id);
+		page_zip_write_header(page_zip,
+				      page + (PAGE_HEADER + PAGE_INDEX_ID),
+				      8, mtr);
+	} else {
+		mlog_write_dulint(page + (PAGE_HEADER + PAGE_INDEX_ID),
+				  id, mtr);
+	}
+}
+
+/******************************************************************
+Gets the index id field of a page. */
+UNIV_INLINE
+dulint
+btr_page_get_index_id(
+/*==================*/
+				/* out: index id */
+	const page_t*	page)	/* in: index page */
+{
+	return(mach_read_from_8(page + PAGE_HEADER + PAGE_INDEX_ID));
+}
+
+/************************************************************
+Gets the node level field in an index page. */
+UNIV_INLINE
+ulint
+btr_page_get_level_low(
+/*===================*/
+				/* out: level, leaf level == 0 */
+	const page_t*	page)	/* in: index page */
+{
+	ulint	level;
+
+	ut_ad(page);
+
+	level = mach_read_from_2(page + PAGE_HEADER + PAGE_LEVEL);
+
+	ut_ad(level <= BTR_MAX_NODE_LEVEL);
+
+	return(level);
+}
+
+/************************************************************
+Gets the node level field in an index page. */
+UNIV_INLINE
+ulint
+btr_page_get_level(
+/*===============*/
+				/* out: level, leaf level == 0 */
+	const page_t*	page,	/* in: index page */
+	mtr_t*		mtr __attribute__((unused)))
+				/* in: mini-transaction handle */
+{
+	ut_ad(page && mtr);
+
+	return(btr_page_get_level_low(page));
+}
+
+/************************************************************
+Sets the node level field in an index page. */
+UNIV_INLINE
+void
+btr_page_set_level(
+/*===============*/
+	page_t*		page,	/* in: index page */
+	page_zip_des_t*	page_zip,/* in: compressed page whose uncompressed
+				part will be updated, or NULL */
+	ulint		level,	/* in: level, leaf level == 0 */
+	mtr_t*		mtr)	/* in: mini-transaction handle */
+{
+	ut_ad(page && mtr);
+	ut_ad(level <= BTR_MAX_NODE_LEVEL);
+
+	if (UNIV_LIKELY_NULL(page_zip)) {
+		mach_write_to_2(page + (PAGE_HEADER + PAGE_LEVEL), level);
+		page_zip_write_header(page_zip,
+				      page + (PAGE_HEADER + PAGE_LEVEL),
+				      2, mtr);
+	} else {
+		mlog_write_ulint(page + (PAGE_HEADER + PAGE_LEVEL), level,
+				 MLOG_2BYTES, mtr);
+	}
+}
+
+/************************************************************
+Gets the next index page number. */
+UNIV_INLINE
+ulint
+btr_page_get_next(
+/*==============*/
+				/* out: next page number */
+	const page_t*	page,	/* in: index page */
+	mtr_t*		mtr __attribute__((unused)))
+				/* in: mini-transaction handle */
+{
+	ut_ad(page && mtr);
+	ut_ad(mtr_memo_contains_page(mtr, page, MTR_MEMO_PAGE_X_FIX)
+	      || mtr_memo_contains_page(mtr, page, MTR_MEMO_PAGE_S_FIX));
+
+	return(mach_read_from_4(page + FIL_PAGE_NEXT));
+}
+
+/************************************************************
+Sets the next index page field. */
+UNIV_INLINE
+void
+btr_page_set_next(
+/*==============*/
+	page_t*		page,	/* in: index page */
+	page_zip_des_t*	page_zip,/* in: compressed page whose uncompressed
+				part will be updated, or NULL */
+	ulint		next,	/* in: next page number */
+	mtr_t*		mtr)	/* in: mini-transaction handle */
+{
+	ut_ad(page && mtr);
+
+	if (UNIV_LIKELY_NULL(page_zip)) {
+		mach_write_to_4(page + FIL_PAGE_NEXT, next);
+		page_zip_write_header(page_zip, page + FIL_PAGE_NEXT, 4, mtr);
+	} else {
+		mlog_write_ulint(page + FIL_PAGE_NEXT, next, MLOG_4BYTES, mtr);
+	}
+}
+
+/************************************************************
+Gets the previous index page number. */
+UNIV_INLINE
+ulint
+btr_page_get_prev(
+/*==============*/
+				/* out: prev page number */
+	const page_t*	page,	/* in: index page */
+	mtr_t*	mtr __attribute__((unused))) /* in: mini-transaction handle */
+{
+	ut_ad(page && mtr);
+
+	return(mach_read_from_4(page + FIL_PAGE_PREV));
+}
+
+/************************************************************
+Sets the previous index page field. */
+UNIV_INLINE
+void
+btr_page_set_prev(
+/*==============*/
+	page_t*		page,	/* in: index page */
+	page_zip_des_t*	page_zip,/* in: compressed page whose uncompressed
+				part will be updated, or NULL */
+	ulint		prev,	/* in: previous page number */
+	mtr_t*		mtr)	/* in: mini-transaction handle */
+{
+	ut_ad(page && mtr);
+
+	if (UNIV_LIKELY_NULL(page_zip)) {
+		mach_write_to_4(page + FIL_PAGE_PREV, prev);
+		page_zip_write_header(page_zip, page + FIL_PAGE_PREV, 4, mtr);
+	} else {
+		mlog_write_ulint(page + FIL_PAGE_PREV, prev, MLOG_4BYTES, mtr);
+	}
+}
+
+/******************************************************************
+Gets the child node file address in a node pointer. */
+UNIV_INLINE
+ulint
+btr_node_ptr_get_child_page_no(
+/*===========================*/
+				/* out: child node address */
+	const rec_t*	rec,	/* in: node pointer record */
+	const ulint*	offsets)/* in: array returned by rec_get_offsets() */
+{
+	const byte*	field;
+	ulint		len;
+	ulint		page_no;
+
+	ut_ad(!rec_offs_comp(offsets) || rec_get_node_ptr_flag(rec));
+
+	/* The child address is in the last field */
+	field = rec_get_nth_field(rec, offsets,
+				  rec_offs_n_fields(offsets) - 1, &len);
+
+	ut_ad(len == 4);
+
+	page_no = mach_read_from_4(field);
+
+	if (UNIV_UNLIKELY(page_no == 0)) {
+		fprintf(stderr,
+			"InnoDB: a nonsensical page number 0"
+			" in a node ptr record at offset %lu\n",
+			(ulong) page_offset(rec));
+		buf_page_print(page_align(rec), 0);
+	}
+
+	return(page_no);
+}
+
+/******************************************************************
+Releases the latches on a leaf page and bufferunfixes it. */
+UNIV_INLINE
+void
+btr_leaf_page_release(
+/*==================*/
+	buf_block_t*	block,		/* in: buffer block */
+	ulint		latch_mode,	/* in: BTR_SEARCH_LEAF or
+					BTR_MODIFY_LEAF */
+	mtr_t*		mtr)		/* in: mtr */
+{
+	ut_ad(latch_mode == BTR_SEARCH_LEAF || latch_mode == BTR_MODIFY_LEAF);
+	ut_ad(!mtr_memo_contains(mtr, block, MTR_MEMO_MODIFY));
+
+	mtr_memo_release(mtr, block,
+			 latch_mode == BTR_SEARCH_LEAF
+			 ? MTR_MEMO_PAGE_S_FIX
+			 : MTR_MEMO_PAGE_X_FIX);
+}
diff --git a/storage/xtradb/include/btr0cur.h b/storage/xtradb/include/btr0cur.h
new file mode 100644
index 00000000000..c3a478c4bb7
--- /dev/null
+++ b/storage/xtradb/include/btr0cur.h
@@ -0,0 +1,742 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The index tree cursor
+
+Created 10/16/1994 Heikki Tuuri
+*******************************************************/
+
+#ifndef btr0cur_h
+#define btr0cur_h
+
+#include "univ.i"
+#include "dict0dict.h"
+#include "page0cur.h"
+#include "btr0types.h"
+#include "que0types.h"
+#include "row0types.h"
+#include "ha0ha.h"
+
+/* Mode flags for btr_cur operations; these can be ORed */
+#define BTR_NO_UNDO_LOG_FLAG	1	/* do no undo logging */
+#define BTR_NO_LOCKING_FLAG	2	/* do no record lock checking */
+#define BTR_KEEP_SYS_FLAG	4	/* sys fields will be found from the
+					update vector or inserted entry */
+
+#define BTR_CUR_ADAPT
+#define BTR_CUR_HASH_ADAPT
+
+#ifdef UNIV_DEBUG
+/*************************************************************
+Returns the page cursor component of a tree cursor. */
+UNIV_INLINE
+page_cur_t*
+btr_cur_get_page_cur(
+/*=================*/
+					/* out: pointer to page cursor
+					component */
+	const btr_cur_t*	cursor);/* in: tree cursor */
+#else /* UNIV_DEBUG */
+# define btr_cur_get_page_cur(cursor) (&(cursor)->page_cur)
+#endif /* UNIV_DEBUG */
+/*************************************************************
+Returns the buffer block on which the tree cursor is positioned. */
+UNIV_INLINE
+buf_block_t*
+btr_cur_get_block(
+/*==============*/
+				/* out: pointer to buffer block */
+	btr_cur_t*	cursor);/* in: tree cursor */
+/*************************************************************
+Returns the record pointer of a tree cursor. */
+UNIV_INLINE
+rec_t*
+btr_cur_get_rec(
+/*============*/
+				/* out: pointer to record */
+	btr_cur_t*	cursor);/* in: tree cursor */
+/*************************************************************
+Returns the compressed page on which the tree cursor is positioned. */
+UNIV_INLINE
+page_zip_des_t*
+btr_cur_get_page_zip(
+/*=================*/
+				/* out: pointer to compressed page,
+				or NULL if the page is not compressed */
+	btr_cur_t*	cursor);/* in: tree cursor */
+/*************************************************************
+Invalidates a tree cursor by setting record pointer to NULL. */
+UNIV_INLINE
+void
+btr_cur_invalidate(
+/*===============*/
+	btr_cur_t*	cursor);/* in: tree cursor */
+/*************************************************************
+Returns the page of a tree cursor. */
+UNIV_INLINE
+page_t*
+btr_cur_get_page(
+/*=============*/
+				/* out: pointer to page */
+	btr_cur_t*	cursor);/* in: tree cursor */
+/*************************************************************
+Returns the index of a cursor. */
+UNIV_INLINE
+dict_index_t*
+btr_cur_get_index(
+/*==============*/
+				/* out: index */
+	btr_cur_t*	cursor);/* in: B-tree cursor */
+/*************************************************************
+Positions a tree cursor at a given record. */
+UNIV_INLINE
+void
+btr_cur_position(
+/*=============*/
+	dict_index_t*	index,	/* in: index */
+	rec_t*		rec,	/* in: record in tree */
+	buf_block_t*	block,	/* in: buffer block of rec */
+	btr_cur_t*	cursor);/* in: cursor */
+/************************************************************************
+Searches an index tree and positions a tree cursor on a given level.
+NOTE: n_fields_cmp in tuple must be set so that it cannot be compared
+to node pointer page number fields on the upper levels of the tree!
+Note that if mode is PAGE_CUR_LE, which is used in inserts, then
+cursor->up_match and cursor->low_match both will have sensible values.
+If mode is PAGE_CUR_GE, then up_match will a have a sensible value. */
+UNIV_INTERN
+void
+btr_cur_search_to_nth_level(
+/*========================*/
+	dict_index_t*	index,	/* in: index */
+	ulint		level,	/* in: the tree level of search */
+	const dtuple_t*	tuple,	/* in: data tuple; NOTE: n_fields_cmp in
+				tuple must be set so that it cannot get
+				compared to the node ptr page number field! */
+	ulint		mode,	/* in: PAGE_CUR_L, ...;
+				NOTE that if the search is made using a unique
+				prefix of a record, mode should be PAGE_CUR_LE,
+				not PAGE_CUR_GE, as the latter may end up on
+				the previous page of the record! Inserts
+				should always be made using PAGE_CUR_LE to
+				search the position! */
+	ulint		latch_mode, /* in: BTR_SEARCH_LEAF, ..., ORed with
+				BTR_INSERT and BTR_ESTIMATE;
+				cursor->left_block is used to store a pointer
+				to the left neighbor page, in the cases
+				BTR_SEARCH_PREV and BTR_MODIFY_PREV;
+				NOTE that if has_search_latch
+				is != 0, we maybe do not have a latch set
+				on the cursor page, we assume
+				the caller uses his search latch
+				to protect the record! */
+	btr_cur_t*	cursor, /* in/out: tree cursor; the cursor page is
+				s- or x-latched, but see also above! */
+	ulint		has_search_latch,/* in: latch mode the caller
+				currently has on btr_search_latch:
+				RW_S_LATCH, or 0 */
+	mtr_t*		mtr);	/* in: mtr */
+/*********************************************************************
+Opens a cursor at either end of an index. */
+UNIV_INTERN
+void
+btr_cur_open_at_index_side(
+/*=======================*/
+	ibool		from_left,	/* in: TRUE if open to the low end,
+					FALSE if to the high end */
+	dict_index_t*	index,		/* in: index */
+	ulint		latch_mode,	/* in: latch mode */
+	btr_cur_t*	cursor,		/* in: cursor */
+	mtr_t*		mtr);		/* in: mtr */
+/**************************************************************************
+Positions a cursor at a randomly chosen position within a B-tree. */
+UNIV_INTERN
+void
+btr_cur_open_at_rnd_pos(
+/*====================*/
+	dict_index_t*	index,		/* in: index */
+	ulint		latch_mode,	/* in: BTR_SEARCH_LEAF, ... */
+	btr_cur_t*	cursor,		/* in/out: B-tree cursor */
+	mtr_t*		mtr);		/* in: mtr */
+/*****************************************************************
+Tries to perform an insert to a page in an index tree, next to cursor.
+It is assumed that mtr holds an x-latch on the page. The operation does
+not succeed if there is too little space on the page. If there is just
+one record on the page, the insert will always succeed; this is to
+prevent trying to split a page with just one record. */
+UNIV_INTERN
+ulint
+btr_cur_optimistic_insert(
+/*======================*/
+				/* out: DB_SUCCESS, DB_WAIT_LOCK,
+				DB_FAIL, or error number */
+	ulint		flags,	/* in: undo logging and locking flags: if not
+				zero, the parameters index and thr should be
+				specified */
+	btr_cur_t*	cursor,	/* in: cursor on page after which to insert;
+				cursor stays valid */
+	dtuple_t*	entry,	/* in/out: entry to insert */
+	rec_t**		rec,	/* out: pointer to inserted record if
+				succeed */
+	big_rec_t**	big_rec,/* out: big rec vector whose fields have to
+				be stored externally by the caller, or
+				NULL */
+	ulint		n_ext,	/* in: number of externally stored columns */
+	que_thr_t*	thr,	/* in: query thread or NULL */
+	mtr_t*		mtr);	/* in: mtr; if this function returns
+				DB_SUCCESS on a leaf page of a secondary
+				index in a compressed tablespace, the
+				mtr must be committed before latching
+				any further pages */
+/*****************************************************************
+Performs an insert on a page of an index tree. It is assumed that mtr
+holds an x-latch on the tree and on the cursor page. If the insert is
+made on the leaf level, to avoid deadlocks, mtr must also own x-latches
+to brothers of page, if those brothers exist. */
+UNIV_INTERN
+ulint
+btr_cur_pessimistic_insert(
+/*=======================*/
+				/* out: DB_SUCCESS or error number */
+	ulint		flags,	/* in: undo logging and locking flags: if not
+				zero, the parameter thr should be
+				specified; if no undo logging is specified,
+				then the caller must have reserved enough
+				free extents in the file space so that the
+				insertion will certainly succeed */
+	btr_cur_t*	cursor,	/* in: cursor after which to insert;
+				cursor stays valid */
+	dtuple_t*	entry,	/* in/out: entry to insert */
+	rec_t**		rec,	/* out: pointer to inserted record if
+				succeed */
+	big_rec_t**	big_rec,/* out: big rec vector whose fields have to
+				be stored externally by the caller, or
+				NULL */
+	ulint		n_ext,	/* in: number of externally stored columns */
+	que_thr_t*	thr,	/* in: query thread or NULL */
+	mtr_t*		mtr);	/* in: mtr */
+/*****************************************************************
+Updates a record when the update causes no size changes in its fields. */
+UNIV_INTERN
+ulint
+btr_cur_update_in_place(
+/*====================*/
+				/* out: DB_SUCCESS or error number */
+	ulint		flags,	/* in: undo logging and locking flags */
+	btr_cur_t*	cursor,	/* in: cursor on the record to update;
+				cursor stays valid and positioned on the
+				same record */
+	const upd_t*	update,	/* in: update vector */
+	ulint		cmpl_info,/* in: compiler info on secondary index
+				updates */
+	que_thr_t*	thr,	/* in: query thread */
+	mtr_t*		mtr);	/* in: mtr; must be committed before
+				latching any further pages */
+/*****************************************************************
+Tries to update a record on a page in an index tree. It is assumed that mtr
+holds an x-latch on the page. The operation does not succeed if there is too
+little space on the page or if the update would result in too empty a page,
+so that tree compression is recommended. */
+UNIV_INTERN
+ulint
+btr_cur_optimistic_update(
+/*======================*/
+				/* out: DB_SUCCESS, or DB_OVERFLOW if the
+				updated record does not fit, DB_UNDERFLOW
+				if the page would become too empty, or
+				DB_ZIP_OVERFLOW if there is not enough
+				space left on the compressed page */
+	ulint		flags,	/* in: undo logging and locking flags */
+	btr_cur_t*	cursor,	/* in: cursor on the record to update;
+				cursor stays valid and positioned on the
+				same record */
+	const upd_t*	update,	/* in: update vector; this must also
+				contain trx id and roll ptr fields */
+	ulint		cmpl_info,/* in: compiler info on secondary index
+				updates */
+	que_thr_t*	thr,	/* in: query thread */
+	mtr_t*		mtr);	/* in: mtr; must be committed before
+				latching any further pages */
+/*****************************************************************
+Performs an update of a record on a page of a tree. It is assumed
+that mtr holds an x-latch on the tree and on the cursor page. If the
+update is made on the leaf level, to avoid deadlocks, mtr must also
+own x-latches to brothers of page, if those brothers exist. */
+UNIV_INTERN
+ulint
+btr_cur_pessimistic_update(
+/*=======================*/
+				/* out: DB_SUCCESS or error code */
+	ulint		flags,	/* in: undo logging, locking, and rollback
+				flags */
+	btr_cur_t*	cursor,	/* in: cursor on the record to update */
+	mem_heap_t**	heap,	/* in/out: pointer to memory heap, or NULL */
+	big_rec_t**	big_rec,/* out: big rec vector whose fields have to
+				be stored externally by the caller, or NULL */
+	const upd_t*	update,	/* in: update vector; this is allowed also
+				contain trx id and roll ptr fields, but
+				the values in update vector have no effect */
+	ulint		cmpl_info,/* in: compiler info on secondary index
+				updates */
+	que_thr_t*	thr,	/* in: query thread */
+	mtr_t*		mtr);	/* in: mtr; must be committed before
+				latching any further pages */
+/***************************************************************
+Marks a clustered index record deleted. Writes an undo log record to
+undo log on this delete marking. Writes in the trx id field the id
+of the deleting transaction, and in the roll ptr field pointer to the
+undo log record created. */
+UNIV_INTERN
+ulint
+btr_cur_del_mark_set_clust_rec(
+/*===========================*/
+				/* out: DB_SUCCESS, DB_LOCK_WAIT, or error
+				number */
+	ulint		flags,	/* in: undo logging and locking flags */
+	btr_cur_t*	cursor,	/* in: cursor */
+	ibool		val,	/* in: value to set */
+	que_thr_t*	thr,	/* in: query thread */
+	mtr_t*		mtr);	/* in: mtr */
+/***************************************************************
+Sets a secondary index record delete mark to TRUE or FALSE. */
+UNIV_INTERN
+ulint
+btr_cur_del_mark_set_sec_rec(
+/*=========================*/
+				/* out: DB_SUCCESS, DB_LOCK_WAIT, or error
+				number */
+	ulint		flags,	/* in: locking flag */
+	btr_cur_t*	cursor,	/* in: cursor */
+	ibool		val,	/* in: value to set */
+	que_thr_t*	thr,	/* in: query thread */
+	mtr_t*		mtr);	/* in: mtr */
+/***************************************************************
+Clear a secondary index record's delete mark.  This function is only
+used by the insert buffer insert merge mechanism. */
+UNIV_INTERN
+void
+btr_cur_del_unmark_for_ibuf(
+/*========================*/
+	rec_t*		rec,		/* in/out: record to delete unmark */
+	page_zip_des_t*	page_zip,	/* in/out: compressed page
+					corresponding to rec, or NULL
+					when the tablespace is
+					uncompressed */
+	mtr_t*		mtr);		/* in: mtr */
+/*****************************************************************
+Tries to compress a page of the tree if it seems useful. It is assumed
+that mtr holds an x-latch on the tree and on the cursor page. To avoid
+deadlocks, mtr must also own x-latches to brothers of page, if those
+brothers exist. NOTE: it is assumed that the caller has reserved enough
+free extents so that the compression will always succeed if done! */
+UNIV_INTERN
+ibool
+btr_cur_compress_if_useful(
+/*=======================*/
+				/* out: TRUE if compression occurred */
+	btr_cur_t*	cursor,	/* in: cursor on the page to compress;
+				cursor does not stay valid if compression
+				occurs */
+	mtr_t*		mtr);	/* in: mtr */
+/***********************************************************
+Removes the record on which the tree cursor is positioned. It is assumed
+that the mtr has an x-latch on the page where the cursor is positioned,
+but no latch on the whole tree. */
+UNIV_INTERN
+ibool
+btr_cur_optimistic_delete(
+/*======================*/
+				/* out: TRUE if success, i.e., the page
+				did not become too empty */
+	btr_cur_t*	cursor,	/* in: cursor on the record to delete;
+				cursor stays valid: if deletion succeeds,
+				on function exit it points to the successor
+				of the deleted record */
+	mtr_t*		mtr);	/* in: mtr; if this function returns
+				TRUE on a leaf page of a secondary
+				index, the mtr must be committed
+				before latching any further pages */
+/*****************************************************************
+Removes the record on which the tree cursor is positioned. Tries
+to compress the page if its fillfactor drops below a threshold
+or if it is the only page on the level. It is assumed that mtr holds
+an x-latch on the tree and on the cursor page. To avoid deadlocks,
+mtr must also own x-latches to brothers of page, if those brothers
+exist. */
+UNIV_INTERN
+ibool
+btr_cur_pessimistic_delete(
+/*=======================*/
+				/* out: TRUE if compression occurred */
+	ulint*		err,	/* out: DB_SUCCESS or DB_OUT_OF_FILE_SPACE;
+				the latter may occur because we may have
+				to update node pointers on upper levels,
+				and in the case of variable length keys
+				these may actually grow in size */
+	ibool		has_reserved_extents, /* in: TRUE if the
+				caller has already reserved enough free
+				extents so that he knows that the operation
+				will succeed */
+	btr_cur_t*	cursor,	/* in: cursor on the record to delete;
+				if compression does not occur, the cursor
+				stays valid: it points to successor of
+				deleted record on function exit */
+	enum trx_rb_ctx	rb_ctx,	/* in: rollback context */
+	mtr_t*		mtr);	/* in: mtr */
+/***************************************************************
+Parses a redo log record of updating a record in-place. */
+UNIV_INTERN
+byte*
+btr_cur_parse_update_in_place(
+/*==========================*/
+				/* out: end of log record or NULL */
+	byte*		ptr,	/* in: buffer */
+	byte*		end_ptr,/* in: buffer end */
+	page_t*		page,	/* in/out: page or NULL */
+	page_zip_des_t*	page_zip,/* in/out: compressed page, or NULL */
+	dict_index_t*	index);	/* in: index corresponding to page */
+/********************************************************************
+Parses the redo log record for delete marking or unmarking of a clustered
+index record. */
+UNIV_INTERN
+byte*
+btr_cur_parse_del_mark_set_clust_rec(
+/*=================================*/
+				/* out: end of log record or NULL */
+	byte*		ptr,	/* in: buffer */
+	byte*		end_ptr,/* in: buffer end */
+	page_t*		page,	/* in/out: page or NULL */
+	page_zip_des_t*	page_zip,/* in/out: compressed page, or NULL */
+	dict_index_t*	index);	/* in: index corresponding to page */
+/********************************************************************
+Parses the redo log record for delete marking or unmarking of a secondary
+index record. */
+UNIV_INTERN
+byte*
+btr_cur_parse_del_mark_set_sec_rec(
+/*===============================*/
+				/* out: end of log record or NULL */
+	byte*		ptr,	/* in: buffer */
+	byte*		end_ptr,/* in: buffer end */
+	page_t*		page,	/* in/out: page or NULL */
+	page_zip_des_t*	page_zip);/* in/out: compressed page, or NULL */
+/***********************************************************************
+Estimates the number of rows in a given index range. */
+UNIV_INTERN
+ib_int64_t
+btr_estimate_n_rows_in_range(
+/*=========================*/
+				/* out: estimated number of rows */
+	dict_index_t*	index,	/* in: index */
+	const dtuple_t*	tuple1,	/* in: range start, may also be empty tuple */
+	ulint		mode1,	/* in: search mode for range start */
+	const dtuple_t*	tuple2,	/* in: range end, may also be empty tuple */
+	ulint		mode2);	/* in: search mode for range end */
+/***********************************************************************
+Estimates the number of different key values in a given index, for
+each n-column prefix of the index where n <= dict_index_get_n_unique(index).
+The estimates are stored in the array index->stat_n_diff_key_vals. */
+UNIV_INTERN
+void
+btr_estimate_number_of_different_key_vals(
+/*======================================*/
+	dict_index_t*	index);	/* in: index */
+/***********************************************************************
+Marks not updated extern fields as not-owned by this record. The ownership
+is transferred to the updated record which is inserted elsewhere in the
+index tree. In purge only the owner of externally stored field is allowed
+to free the field. */
+UNIV_INTERN
+void
+btr_cur_mark_extern_inherited_fields(
+/*=================================*/
+	page_zip_des_t*	page_zip,/* in/out: compressed page whose uncompressed
+				part will be updated, or NULL */
+	rec_t*		rec,	/* in/out: record in a clustered index */
+	dict_index_t*	index,	/* in: index of the page */
+	const ulint*	offsets,/* in: array returned by rec_get_offsets() */
+	const upd_t*	update,	/* in: update vector */
+	mtr_t*		mtr);	/* in: mtr, or NULL if not logged */
+/***********************************************************************
+The complement of the previous function: in an update entry may inherit
+some externally stored fields from a record. We must mark them as inherited
+in entry, so that they are not freed in a rollback. */
+UNIV_INTERN
+void
+btr_cur_mark_dtuple_inherited_extern(
+/*=================================*/
+	dtuple_t*	entry,		/* in/out: updated entry to be
+					inserted to clustered index */
+	const upd_t*	update);	/* in: update vector */
+/***********************************************************************
+Marks all extern fields in a dtuple as owned by the record. */
+UNIV_INTERN
+void
+btr_cur_unmark_dtuple_extern_fields(
+/*================================*/
+	dtuple_t*	entry);		/* in/out: clustered index entry */
+/***********************************************************************
+Stores the fields in big_rec_vec to the tablespace and puts pointers to
+them in rec.  The extern flags in rec will have to be set beforehand.
+The fields are stored on pages allocated from leaf node
+file segment of the index tree. */
+UNIV_INTERN
+ulint
+btr_store_big_rec_extern_fields(
+/*============================*/
+					/* out: DB_SUCCESS or error */
+	dict_index_t*	index,		/* in: index of rec; the index tree
+					MUST be X-latched */
+	buf_block_t*	rec_block,	/* in/out: block containing rec */
+	rec_t*		rec,		/* in: record */
+	const ulint*	offsets,	/* in: rec_get_offsets(rec, index);
+					the "external storage" flags in offsets
+					will not correspond to rec when
+					this function returns */
+	big_rec_t*	big_rec_vec,	/* in: vector containing fields
+					to be stored externally */
+	mtr_t*		local_mtr);	/* in: mtr containing the latch to
+					rec and to the tree */
+/***********************************************************************
+Frees the space in an externally stored field to the file space
+management if the field in data is owned the externally stored field,
+in a rollback we may have the additional condition that the field must
+not be inherited. */
+UNIV_INTERN
+void
+btr_free_externally_stored_field(
+/*=============================*/
+	dict_index_t*	index,		/* in: index of the data, the index
+					tree MUST be X-latched; if the tree
+					height is 1, then also the root page
+					must be X-latched! (this is relevant
+					in the case this function is called
+					from purge where 'data' is located on
+					an undo log page, not an index
+					page) */
+	byte*		field_ref,	/* in/out: field reference */
+	const rec_t*	rec,		/* in: record containing field_ref, for
+					page_zip_write_blob_ptr(), or NULL */
+	const ulint*	offsets,	/* in: rec_get_offsets(rec, index),
+					or NULL */
+	page_zip_des_t*	page_zip,	/* in: compressed page corresponding
+					to rec, or NULL if rec == NULL */
+	ulint		i,		/* in: field number of field_ref;
+					ignored if rec == NULL */
+	enum trx_rb_ctx	rb_ctx,		/* in: rollback context */
+	mtr_t*		local_mtr);	/* in: mtr containing the latch to
+					data an an X-latch to the index
+					tree */
+/***********************************************************************
+Copies the prefix of an externally stored field of a record.  The
+clustered index record must be protected by a lock or a page latch. */
+UNIV_INTERN
+ulint
+btr_copy_externally_stored_field_prefix(
+/*====================================*/
+				/* out: the length of the copied field,
+				or 0 if the column is being or has been
+				deleted */
+	byte*		buf,	/* out: the field, or a prefix of it */
+	ulint		len,	/* in: length of buf, in bytes */
+	ulint		zip_size,/* in: nonzero=compressed BLOB page size,
+				zero for uncompressed BLOBs */
+	const byte*	data,	/* in: 'internally' stored part of the
+				field containing also the reference to
+				the external part; must be protected by
+				a lock or a page latch */
+	ulint		local_len);/* in: length of data, in bytes */
+/***********************************************************************
+Copies an externally stored field of a record to mem heap. */
+UNIV_INTERN
+byte*
+btr_rec_copy_externally_stored_field(
+/*=================================*/
+				/* out: the field copied to heap */
+	const rec_t*	rec,	/* in: record in a clustered index;
+				must be protected by a lock or a page latch */
+	const ulint*	offsets,/* in: array returned by rec_get_offsets() */
+	ulint		zip_size,/* in: nonzero=compressed BLOB page size,
+				zero for uncompressed BLOBs */
+	ulint		no,	/* in: field number */
+	ulint*		len,	/* out: length of the field */
+	mem_heap_t*	heap);	/* in: mem heap */
+/***********************************************************************
+Flags the data tuple fields that are marked as extern storage in the
+update vector.  We use this function to remember which fields we must
+mark as extern storage in a record inserted for an update. */
+UNIV_INTERN
+ulint
+btr_push_update_extern_fields(
+/*==========================*/
+				/* out: number of flagged external columns */
+	dtuple_t*	tuple,	/* in/out: data tuple */
+	const upd_t*	update,	/* in: update vector */
+	mem_heap_t*	heap)	/* in: memory heap */
+	__attribute__((nonnull));
+
+/*######################################################################*/
+
+/* In the pessimistic delete, if the page data size drops below this
+limit, merging it to a neighbor is tried */
+
+#define BTR_CUR_PAGE_COMPRESS_LIMIT	(UNIV_PAGE_SIZE / 2)
+
+/* A slot in the path array. We store here info on a search path down the
+tree. Each slot contains data on a single level of the tree. */
+
+typedef struct btr_path_struct	btr_path_t;
+struct btr_path_struct{
+	ulint	nth_rec;	/* index of the record
+				where the page cursor stopped on
+				this level (index in alphabetical
+				order); value ULINT_UNDEFINED
+				denotes array end */
+	ulint	n_recs;		/* number of records on the page */
+};
+
+#define BTR_PATH_ARRAY_N_SLOTS	250	/* size of path array (in slots) */
+
+/* The tree cursor: the definition appears here only for the compiler
+to know struct size! */
+
+struct btr_cur_struct {
+	dict_index_t*	index;		/* index where positioned */
+	page_cur_t	page_cur;	/* page cursor */
+	buf_block_t*	left_block;	/* this field is used to store
+					a pointer to the left neighbor
+					page, in the cases
+					BTR_SEARCH_PREV and
+					BTR_MODIFY_PREV */
+	/*------------------------------*/
+	que_thr_t*	thr;		/* this field is only used when
+					btr_cur_search_... is called for an
+					index entry insertion: the calling
+					query thread is passed here to be
+					used in the insert buffer */
+	/*------------------------------*/
+	/* The following fields are used in btr_cur_search... to pass
+	information: */
+	ulint		flag;		/* BTR_CUR_HASH, BTR_CUR_HASH_FAIL,
+					BTR_CUR_BINARY, or
+					BTR_CUR_INSERT_TO_IBUF */
+	ulint		tree_height;	/* Tree height if the search is done
+					for a pessimistic insert or update
+					operation */
+	ulint		up_match;	/* If the search mode was PAGE_CUR_LE,
+					the number of matched fields to the
+					the first user record to the right of
+					the cursor record after
+					btr_cur_search_...;
+					for the mode PAGE_CUR_GE, the matched
+					fields to the first user record AT THE
+					CURSOR or to the right of it;
+					NOTE that the up_match and low_match
+					values may exceed the correct values
+					for comparison to the adjacent user
+					record if that record is on a
+					different leaf page! (See the note in
+					row_ins_duplicate_key.) */
+	ulint		up_bytes;	/* number of matched bytes to the
+					right at the time cursor positioned;
+					only used internally in searches: not
+					defined after the search */
+	ulint		low_match;	/* if search mode was PAGE_CUR_LE,
+					the number of matched fields to the
+					first user record AT THE CURSOR or
+					to the left of it after
+					btr_cur_search_...;
+					NOT defined for PAGE_CUR_GE or any
+					other search modes; see also the NOTE
+					in up_match! */
+	ulint		low_bytes;	/* number of matched bytes to the
+					right at the time cursor positioned;
+					only used internally in searches: not
+					defined after the search */
+	ulint		n_fields;	/* prefix length used in a hash
+					search if hash_node != NULL */
+	ulint		n_bytes;	/* hash prefix bytes if hash_node !=
+					NULL */
+	ulint		fold;		/* fold value used in the search if
+					flag is BTR_CUR_HASH */
+	/*------------------------------*/
+	btr_path_t*	path_arr;	/* in estimating the number of
+					rows in range, we store in this array
+					information of the path through
+					the tree */
+};
+
+/* Values for the flag documenting the used search method */
+#define BTR_CUR_HASH		1	/* successful shortcut using the hash
+					index */
+#define BTR_CUR_HASH_FAIL	2	/* failure using hash, success using
+					binary search: the misleading hash
+					reference is stored in the field
+					hash_node, and might be necessary to
+					update */
+#define BTR_CUR_BINARY		3	/* success using the binary search */
+#define BTR_CUR_INSERT_TO_IBUF	4	/* performed the intended insert to
+					the insert buffer */
+
+/* If pessimistic delete fails because of lack of file space,
+there is still a good change of success a little later: try this many times,
+and sleep this many microseconds in between */
+#define BTR_CUR_RETRY_DELETE_N_TIMES	100
+#define BTR_CUR_RETRY_SLEEP_TIME	50000
+
+/* The reference in a field for which data is stored on a different page.
+The reference is at the end of the 'locally' stored part of the field.
+'Locally' means storage in the index record.
+We store locally a long enough prefix of each column so that we can determine
+the ordering parts of each index record without looking into the externally
+stored part. */
+
+/*--------------------------------------*/
+#define BTR_EXTERN_SPACE_ID		0	/* space id where stored */
+#define BTR_EXTERN_PAGE_NO		4	/* page no where stored */
+#define BTR_EXTERN_OFFSET		8	/* offset of BLOB header
+						on that page */
+#define BTR_EXTERN_LEN			12	/* 8 bytes containing the
+						length of the externally
+						stored part of the BLOB.
+						The 2 highest bits are
+						reserved to the flags below. */
+/*--------------------------------------*/
+/* #define BTR_EXTERN_FIELD_REF_SIZE	20 // moved to btr0types.h */
+
+/* The highest bit of BTR_EXTERN_LEN (i.e., the highest bit of the byte
+at lowest address) is set to 1 if this field does not 'own' the externally
+stored field; only the owner field is allowed to free the field in purge!
+If the 2nd highest bit is 1 then it means that the externally stored field
+was inherited from an earlier version of the row. In rollback we are not
+allowed to free an inherited external field. */
+
+#define BTR_EXTERN_OWNER_FLAG		128
+#define BTR_EXTERN_INHERITED_FLAG	64
+
+extern ulint	btr_cur_n_non_sea;
+extern ulint	btr_cur_n_sea;
+extern ulint	btr_cur_n_non_sea_old;
+extern ulint	btr_cur_n_sea_old;
+
+#ifndef UNIV_NONINL
+#include "btr0cur.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/btr0cur.ic b/storage/xtradb/include/btr0cur.ic
new file mode 100644
index 00000000000..84a3a5cba0b
--- /dev/null
+++ b/storage/xtradb/include/btr0cur.ic
@@ -0,0 +1,200 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The index tree cursor
+
+Created 10/16/1994 Heikki Tuuri
+*******************************************************/
+
+#include "btr0btr.h"
+
+#ifdef UNIV_DEBUG
+/*************************************************************
+Returns the page cursor component of a tree cursor. */
+UNIV_INLINE
+page_cur_t*
+btr_cur_get_page_cur(
+/*=================*/
+					/* out: pointer to page cursor
+					component */
+	const btr_cur_t*	cursor)	/* in: tree cursor */
+{
+	return(&((btr_cur_t*) cursor)->page_cur);
+}
+#endif /* UNIV_DEBUG */
+/*************************************************************
+Returns the buffer block on which the tree cursor is positioned. */
+UNIV_INLINE
+buf_block_t*
+btr_cur_get_block(
+/*==============*/
+				/* out: pointer to buffer block */
+	btr_cur_t*	cursor)	/* in: tree cursor */
+{
+	return(page_cur_get_block(btr_cur_get_page_cur(cursor)));
+}
+
+/*************************************************************
+Returns the record pointer of a tree cursor. */
+UNIV_INLINE
+rec_t*
+btr_cur_get_rec(
+/*============*/
+				/* out: pointer to record */
+	btr_cur_t*	cursor)	/* in: tree cursor */
+{
+	return(page_cur_get_rec(&(cursor->page_cur)));
+}
+
+/*************************************************************
+Returns the compressed page on which the tree cursor is positioned. */
+UNIV_INLINE
+page_zip_des_t*
+btr_cur_get_page_zip(
+/*=================*/
+				/* out: pointer to compressed page,
+				or NULL if the page is not compressed */
+	btr_cur_t*	cursor)	/* in: tree cursor */
+{
+	return(buf_block_get_page_zip(btr_cur_get_block(cursor)));
+}
+
+/*************************************************************
+Invalidates a tree cursor by setting record pointer to NULL. */
+UNIV_INLINE
+void
+btr_cur_invalidate(
+/*===============*/
+	btr_cur_t*	cursor)	/* in: tree cursor */
+{
+	page_cur_invalidate(&(cursor->page_cur));
+}
+
+/*************************************************************
+Returns the page of a tree cursor. */
+UNIV_INLINE
+page_t*
+btr_cur_get_page(
+/*=============*/
+				/* out: pointer to page */
+	btr_cur_t*	cursor)	/* in: tree cursor */
+{
+	return(page_align(page_cur_get_rec(&(cursor->page_cur))));
+}
+
+/*************************************************************
+Returns the index of a cursor. */
+UNIV_INLINE
+dict_index_t*
+btr_cur_get_index(
+/*==============*/
+				/* out: index */
+	btr_cur_t*	cursor)	/* in: B-tree cursor */
+{
+	return(cursor->index);
+}
+
+/*************************************************************
+Positions a tree cursor at a given record. */
+UNIV_INLINE
+void
+btr_cur_position(
+/*=============*/
+	dict_index_t*	index,	/* in: index */
+	rec_t*		rec,	/* in: record in tree */
+	buf_block_t*	block,	/* in: buffer block of rec */
+	btr_cur_t*	cursor)	/* out: cursor */
+{
+	ut_ad(page_align(rec) == block->frame);
+
+	page_cur_position(rec, block, btr_cur_get_page_cur(cursor));
+
+	cursor->index = index;
+}
+
+/*************************************************************************
+Checks if compressing an index page where a btr cursor is placed makes
+sense. */
+UNIV_INLINE
+ibool
+btr_cur_compress_recommendation(
+/*============================*/
+				/* out: TRUE if compression is recommended */
+	btr_cur_t*	cursor,	/* in: btr cursor */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	page_t*		page;
+
+	ut_ad(mtr_memo_contains(mtr, btr_cur_get_block(cursor),
+				MTR_MEMO_PAGE_X_FIX));
+
+	page = btr_cur_get_page(cursor);
+
+	if ((page_get_data_size(page) < BTR_CUR_PAGE_COMPRESS_LIMIT)
+	    || ((btr_page_get_next(page, mtr) == FIL_NULL)
+		&& (btr_page_get_prev(page, mtr) == FIL_NULL))) {
+
+		/* The page fillfactor has dropped below a predefined
+		minimum value OR the level in the B-tree contains just
+		one page: we recommend compression if this is not the
+		root page. */
+
+		return(dict_index_get_page(cursor->index)
+		       != page_get_page_no(page));
+	}
+
+	return(FALSE);
+}
+
+/*************************************************************************
+Checks if the record on which the cursor is placed can be deleted without
+making tree compression necessary (or, recommended). */
+UNIV_INLINE
+ibool
+btr_cur_can_delete_without_compress(
+/*================================*/
+				/* out: TRUE if can be deleted without
+				recommended compression */
+	btr_cur_t*	cursor,	/* in: btr cursor */
+	ulint		rec_size,/* in: rec_get_size(btr_cur_get_rec(cursor))*/
+	mtr_t*		mtr)	/* in: mtr */
+{
+	page_t*		page;
+
+	ut_ad(mtr_memo_contains(mtr, btr_cur_get_block(cursor),
+				MTR_MEMO_PAGE_X_FIX));
+
+	page = btr_cur_get_page(cursor);
+
+	if ((page_get_data_size(page) - rec_size < BTR_CUR_PAGE_COMPRESS_LIMIT)
+	    || ((btr_page_get_next(page, mtr) == FIL_NULL)
+		&& (btr_page_get_prev(page, mtr) == FIL_NULL))
+	    || (page_get_n_recs(page) < 2)) {
+
+		/* The page fillfactor will drop below a predefined
+		minimum value, OR the level in the B-tree contains just
+		one page, OR the page will become empty: we recommend
+		compression if this is not the root page. */
+
+		return(dict_index_get_page(cursor->index)
+		       == page_get_page_no(page));
+	}
+
+	return(TRUE);
+}
diff --git a/storage/xtradb/include/btr0pcur.h b/storage/xtradb/include/btr0pcur.h
new file mode 100644
index 00000000000..1fdd102d32a
--- /dev/null
+++ b/storage/xtradb/include/btr0pcur.h
@@ -0,0 +1,545 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The index tree persistent cursor
+
+Created 2/23/1996 Heikki Tuuri
+*******************************************************/
+
+#ifndef btr0pcur_h
+#define btr0pcur_h
+
+#include "univ.i"
+#include "dict0dict.h"
+#include "data0data.h"
+#include "mtr0mtr.h"
+#include "page0cur.h"
+#include "btr0cur.h"
+#include "btr0btr.h"
+#include "btr0types.h"
+
+/* Relative positions for a stored cursor position */
+#define BTR_PCUR_ON			1
+#define BTR_PCUR_BEFORE			2
+#define BTR_PCUR_AFTER			3
+/* Note that if the tree is not empty, btr_pcur_store_position does not
+use the following, but only uses the above three alternatives, where the
+position is stored relative to a specific record: this makes implementation
+of a scroll cursor easier */
+#define BTR_PCUR_BEFORE_FIRST_IN_TREE	4	/* in an empty tree */
+#define BTR_PCUR_AFTER_LAST_IN_TREE	5	/* in an empty tree */
+
+/******************************************************************
+Allocates memory for a persistent cursor object and initializes the cursor. */
+UNIV_INTERN
+btr_pcur_t*
+btr_pcur_create_for_mysql(void);
+/*============================*/
+				/* out, own: persistent cursor */
+/******************************************************************
+Frees the memory for a persistent cursor object. */
+UNIV_INTERN
+void
+btr_pcur_free_for_mysql(
+/*====================*/
+	btr_pcur_t*	cursor);	/* in, own: persistent cursor */
+/******************************************************************
+Copies the stored position of a pcur to another pcur. */
+UNIV_INTERN
+void
+btr_pcur_copy_stored_position(
+/*==========================*/
+	btr_pcur_t*	pcur_receive,	/* in: pcur which will receive the
+					position info */
+	btr_pcur_t*	pcur_donate);	/* in: pcur from which the info is
+					copied */
+/******************************************************************
+Sets the old_rec_buf field to NULL. */
+UNIV_INLINE
+void
+btr_pcur_init(
+/*==========*/
+	btr_pcur_t*	pcur);	/* in: persistent cursor */
+/******************************************************************
+Initializes and opens a persistent cursor to an index tree. It should be
+closed with btr_pcur_close. */
+UNIV_INLINE
+void
+btr_pcur_open(
+/*==========*/
+	dict_index_t*	index,	/* in: index */
+	const dtuple_t*	tuple,	/* in: tuple on which search done */
+	ulint		mode,	/* in: PAGE_CUR_L, ...;
+				NOTE that if the search is made using a unique
+				prefix of a record, mode should be
+				PAGE_CUR_LE, not PAGE_CUR_GE, as the latter
+				may end up on the previous page from the
+				record! */
+	ulint		latch_mode,/* in: BTR_SEARCH_LEAF, ... */
+	btr_pcur_t*	cursor, /* in: memory buffer for persistent cursor */
+	mtr_t*		mtr);	/* in: mtr */
+/******************************************************************
+Opens an persistent cursor to an index tree without initializing the
+cursor. */
+UNIV_INLINE
+void
+btr_pcur_open_with_no_init(
+/*=======================*/
+	dict_index_t*	index,	/* in: index */
+	const dtuple_t*	tuple,	/* in: tuple on which search done */
+	ulint		mode,	/* in: PAGE_CUR_L, ...;
+				NOTE that if the search is made using a unique
+				prefix of a record, mode should be
+				PAGE_CUR_LE, not PAGE_CUR_GE, as the latter
+				may end up on the previous page of the
+				record! */
+	ulint		latch_mode,/* in: BTR_SEARCH_LEAF, ...;
+				NOTE that if has_search_latch != 0 then
+				we maybe do not acquire a latch on the cursor
+				page, but assume that the caller uses his
+				btr search latch to protect the record! */
+	btr_pcur_t*	cursor, /* in: memory buffer for persistent cursor */
+	ulint		has_search_latch,/* in: latch mode the caller
+				currently has on btr_search_latch:
+				RW_S_LATCH, or 0 */
+	mtr_t*		mtr);	/* in: mtr */
+/*********************************************************************
+Opens a persistent cursor at either end of an index. */
+UNIV_INLINE
+void
+btr_pcur_open_at_index_side(
+/*========================*/
+	ibool		from_left,	/* in: TRUE if open to the low end,
+					FALSE if to the high end */
+	dict_index_t*	index,		/* in: index */
+	ulint		latch_mode,	/* in: latch mode */
+	btr_pcur_t*	pcur,		/* in: cursor */
+	ibool		do_init,	/* in: TRUE if should be initialized */
+	mtr_t*		mtr);		/* in: mtr */
+/******************************************************************
+Gets the up_match value for a pcur after a search. */
+UNIV_INLINE
+ulint
+btr_pcur_get_up_match(
+/*==================*/
+				/* out: number of matched fields at the cursor
+				or to the right if search mode was PAGE_CUR_GE,
+				otherwise undefined */
+	btr_pcur_t*	cursor); /* in: memory buffer for persistent cursor */
+/******************************************************************
+Gets the low_match value for a pcur after a search. */
+UNIV_INLINE
+ulint
+btr_pcur_get_low_match(
+/*===================*/
+				/* out: number of matched fields at the cursor
+				or to the right if search mode was PAGE_CUR_LE,
+				otherwise undefined */
+	btr_pcur_t*	cursor); /* in: memory buffer for persistent cursor */
+/******************************************************************
+If mode is PAGE_CUR_G or PAGE_CUR_GE, opens a persistent cursor on the first
+user record satisfying the search condition, in the case PAGE_CUR_L or
+PAGE_CUR_LE, on the last user record. If no such user record exists, then
+in the first case sets the cursor after last in tree, and in the latter case
+before first in tree. The latching mode must be BTR_SEARCH_LEAF or
+BTR_MODIFY_LEAF. */
+UNIV_INTERN
+void
+btr_pcur_open_on_user_rec(
+/*======================*/
+	dict_index_t*	index,		/* in: index */
+	const dtuple_t*	tuple,		/* in: tuple on which search done */
+	ulint		mode,		/* in: PAGE_CUR_L, ... */
+	ulint		latch_mode,	/* in: BTR_SEARCH_LEAF or
+					BTR_MODIFY_LEAF */
+	btr_pcur_t*	cursor,		/* in: memory buffer for persistent
+					cursor */
+	mtr_t*		mtr);		/* in: mtr */
+/**************************************************************************
+Positions a cursor at a randomly chosen position within a B-tree. */
+UNIV_INLINE
+void
+btr_pcur_open_at_rnd_pos(
+/*=====================*/
+	dict_index_t*	index,		/* in: index */
+	ulint		latch_mode,	/* in: BTR_SEARCH_LEAF, ... */
+	btr_pcur_t*	cursor,		/* in/out: B-tree pcur */
+	mtr_t*		mtr);		/* in: mtr */
+/******************************************************************
+Frees the possible old_rec_buf buffer of a persistent cursor and sets the
+latch mode of the persistent cursor to BTR_NO_LATCHES. */
+UNIV_INLINE
+void
+btr_pcur_close(
+/*===========*/
+	btr_pcur_t*	cursor);	/* in: persistent cursor */
+/******************************************************************
+The position of the cursor is stored by taking an initial segment of the
+record the cursor is positioned on, before, or after, and copying it to the
+cursor data structure, or just setting a flag if the cursor id before the
+first in an EMPTY tree, or after the last in an EMPTY tree. NOTE that the
+page where the cursor is positioned must not be empty if the index tree is
+not totally empty! */
+UNIV_INTERN
+void
+btr_pcur_store_position(
+/*====================*/
+	btr_pcur_t*	cursor, /* in: persistent cursor */
+	mtr_t*		mtr);	/* in: mtr */
+/******************************************************************
+Restores the stored position of a persistent cursor bufferfixing the page and
+obtaining the specified latches. If the cursor position was saved when the
+(1) cursor was positioned on a user record: this function restores the position
+to the last record LESS OR EQUAL to the stored record;
+(2) cursor was positioned on a page infimum record: restores the position to
+the last record LESS than the user record which was the successor of the page
+infimum;
+(3) cursor was positioned on the page supremum: restores to the first record
+GREATER than the user record which was the predecessor of the supremum.
+(4) cursor was positioned before the first or after the last in an empty tree:
+restores to before first or after the last in the tree. */
+UNIV_INTERN
+ibool
+btr_pcur_restore_position(
+/*======================*/
+					/* out: TRUE if the cursor position
+					was stored when it was on a user record
+					and it can be restored on a user record
+					whose ordering fields are identical to
+					the ones of the original user record */
+	ulint		latch_mode,	/* in: BTR_SEARCH_LEAF, ... */
+	btr_pcur_t*	cursor,		/* in: detached persistent cursor */
+	mtr_t*		mtr);		/* in: mtr */
+/******************************************************************
+If the latch mode of the cursor is BTR_LEAF_SEARCH or BTR_LEAF_MODIFY,
+releases the page latch and bufferfix reserved by the cursor.
+NOTE! In the case of BTR_LEAF_MODIFY, there should not exist changes
+made by the current mini-transaction to the data protected by the
+cursor latch, as then the latch must not be released until mtr_commit. */
+UNIV_INTERN
+void
+btr_pcur_release_leaf(
+/*==================*/
+	btr_pcur_t*	cursor, /* in: persistent cursor */
+	mtr_t*		mtr);	/* in: mtr */
+/*************************************************************
+Gets the rel_pos field for a cursor whose position has been stored. */
+UNIV_INLINE
+ulint
+btr_pcur_get_rel_pos(
+/*=================*/
+					/* out: BTR_PCUR_ON, ... */
+	const btr_pcur_t*	cursor);/* in: persistent cursor */
+/*************************************************************
+Sets the mtr field for a pcur. */
+UNIV_INLINE
+void
+btr_pcur_set_mtr(
+/*=============*/
+	btr_pcur_t*	cursor,	/* in: persistent cursor */
+	mtr_t*		mtr);	/* in, own: mtr */
+/*************************************************************
+Gets the mtr field for a pcur. */
+UNIV_INLINE
+mtr_t*
+btr_pcur_get_mtr(
+/*=============*/
+				/* out: mtr */
+	btr_pcur_t*	cursor);	/* in: persistent cursor */
+/******************************************************************
+Commits the pcur mtr and sets the pcur latch mode to BTR_NO_LATCHES,
+that is, the cursor becomes detached. If there have been modifications
+to the page where pcur is positioned, this can be used instead of
+btr_pcur_release_leaf. Function btr_pcur_store_position should be used
+before calling this, if restoration of cursor is wanted later. */
+UNIV_INLINE
+void
+btr_pcur_commit(
+/*============*/
+	btr_pcur_t*	pcur);	/* in: persistent cursor */
+/******************************************************************
+Differs from btr_pcur_commit in that we can specify the mtr to commit. */
+UNIV_INLINE
+void
+btr_pcur_commit_specify_mtr(
+/*========================*/
+	btr_pcur_t*	pcur,	/* in: persistent cursor */
+	mtr_t*		mtr);	/* in: mtr to commit */
+/******************************************************************
+Tests if a cursor is detached: that is the latch mode is BTR_NO_LATCHES. */
+UNIV_INLINE
+ibool
+btr_pcur_is_detached(
+/*=================*/
+				/* out: TRUE if detached */
+	btr_pcur_t*	pcur);	/* in: persistent cursor */
+/*************************************************************
+Moves the persistent cursor to the next record in the tree. If no records are
+left, the cursor stays 'after last in tree'. */
+UNIV_INLINE
+ibool
+btr_pcur_move_to_next(
+/*==================*/
+				/* out: TRUE if the cursor was not after last
+				in tree */
+	btr_pcur_t*	cursor,	/* in: persistent cursor; NOTE that the
+				function may release the page latch */
+	mtr_t*		mtr);	/* in: mtr */
+/*************************************************************
+Moves the persistent cursor to the previous record in the tree. If no records
+are left, the cursor stays 'before first in tree'. */
+UNIV_INTERN
+ibool
+btr_pcur_move_to_prev(
+/*==================*/
+				/* out: TRUE if the cursor was not before first
+				in tree */
+	btr_pcur_t*	cursor,	/* in: persistent cursor; NOTE that the
+				function may release the page latch */
+	mtr_t*		mtr);	/* in: mtr */
+/*************************************************************
+Moves the persistent cursor to the last record on the same page. */
+UNIV_INLINE
+void
+btr_pcur_move_to_last_on_page(
+/*==========================*/
+	btr_pcur_t*	cursor,	/* in: persistent cursor */
+	mtr_t*		mtr);	/* in: mtr */
+/*************************************************************
+Moves the persistent cursor to the next user record in the tree. If no user
+records are left, the cursor ends up 'after last in tree'. */
+UNIV_INLINE
+ibool
+btr_pcur_move_to_next_user_rec(
+/*===========================*/
+				/* out: TRUE if the cursor moved forward,
+				ending on a user record */
+	btr_pcur_t*	cursor,	/* in: persistent cursor; NOTE that the
+				function may release the page latch */
+	mtr_t*		mtr);	/* in: mtr */
+/*************************************************************
+Moves the persistent cursor to the first record on the next page.
+Releases the latch on the current page, and bufferunfixes it.
+Note that there must not be modifications on the current page,
+as then the x-latch can be released only in mtr_commit. */
+UNIV_INTERN
+void
+btr_pcur_move_to_next_page(
+/*=======================*/
+	btr_pcur_t*	cursor,	/* in: persistent cursor; must be on the
+				last record of the current page */
+	mtr_t*		mtr);	/* in: mtr */
+/*************************************************************
+Moves the persistent cursor backward if it is on the first record
+of the page. Releases the latch on the current page, and bufferunfixes
+it. Note that to prevent a possible deadlock, the operation first
+stores the position of the cursor, releases the leaf latch, acquires
+necessary latches and restores the cursor position again before returning.
+The alphabetical position of the cursor is guaranteed to be sensible
+on return, but it may happen that the cursor is not positioned on the
+last record of any page, because the structure of the tree may have
+changed while the cursor had no latches. */
+UNIV_INTERN
+void
+btr_pcur_move_backward_from_page(
+/*=============================*/
+	btr_pcur_t*	cursor,	/* in: persistent cursor, must be on the
+				first record of the current page */
+	mtr_t*		mtr);	/* in: mtr */
+#ifdef UNIV_DEBUG
+/*************************************************************
+Returns the btr cursor component of a persistent cursor. */
+UNIV_INLINE
+btr_cur_t*
+btr_pcur_get_btr_cur(
+/*=================*/
+						/* out: pointer to
+						btr cursor component */
+	const btr_pcur_t*	cursor);	/* in: persistent cursor */
+/*************************************************************
+Returns the page cursor component of a persistent cursor. */
+UNIV_INLINE
+page_cur_t*
+btr_pcur_get_page_cur(
+/*==================*/
+						/* out: pointer to
+						page cursor component */
+	const btr_pcur_t*	cursor);	/* in: persistent cursor */
+#else /* UNIV_DEBUG */
+# define btr_pcur_get_btr_cur(cursor) (&(cursor)->btr_cur)
+# define btr_pcur_get_page_cur(cursor) (&(cursor)->btr_cur.page_cur)
+#endif /* UNIV_DEBUG */
+/*************************************************************
+Returns the page of a persistent cursor. */
+UNIV_INLINE
+page_t*
+btr_pcur_get_page(
+/*==============*/
+				/* out: pointer to the page */
+	btr_pcur_t*	cursor);/* in: persistent cursor */
+/*************************************************************
+Returns the buffer block of a persistent cursor. */
+UNIV_INLINE
+buf_block_t*
+btr_pcur_get_block(
+/*===============*/
+				/* out: pointer to the block */
+	btr_pcur_t*	cursor);/* in: persistent cursor */
+/*************************************************************
+Returns the record of a persistent cursor. */
+UNIV_INLINE
+rec_t*
+btr_pcur_get_rec(
+/*=============*/
+				/* out: pointer to the record */
+	btr_pcur_t*	cursor);/* in: persistent cursor */
+/*************************************************************
+Checks if the persistent cursor is on a user record. */
+UNIV_INLINE
+ibool
+btr_pcur_is_on_user_rec(
+/*====================*/
+	const btr_pcur_t*	cursor);/* in: persistent cursor */
+/*************************************************************
+Checks if the persistent cursor is after the last user record on
+a page. */
+UNIV_INLINE
+ibool
+btr_pcur_is_after_last_on_page(
+/*===========================*/
+	const btr_pcur_t*	cursor);/* in: persistent cursor */
+/*************************************************************
+Checks if the persistent cursor is before the first user record on
+a page. */
+UNIV_INLINE
+ibool
+btr_pcur_is_before_first_on_page(
+/*=============================*/
+	const btr_pcur_t*	cursor);/* in: persistent cursor */
+/*************************************************************
+Checks if the persistent cursor is before the first user record in
+the index tree. */
+UNIV_INLINE
+ibool
+btr_pcur_is_before_first_in_tree(
+/*=============================*/
+	btr_pcur_t*	cursor,	/* in: persistent cursor */
+	mtr_t*		mtr);	/* in: mtr */
+/*************************************************************
+Checks if the persistent cursor is after the last user record in
+the index tree. */
+UNIV_INLINE
+ibool
+btr_pcur_is_after_last_in_tree(
+/*===========================*/
+	btr_pcur_t*	cursor,	/* in: persistent cursor */
+	mtr_t*		mtr);	/* in: mtr */
+/*************************************************************
+Moves the persistent cursor to the next record on the same page. */
+UNIV_INLINE
+void
+btr_pcur_move_to_next_on_page(
+/*==========================*/
+	btr_pcur_t*	cursor);/* in/out: persistent cursor */
+/*************************************************************
+Moves the persistent cursor to the previous record on the same page. */
+UNIV_INLINE
+void
+btr_pcur_move_to_prev_on_page(
+/*==========================*/
+	btr_pcur_t*	cursor);/* in/out: persistent cursor */
+
+
+/* The persistent B-tree cursor structure. This is used mainly for SQL
+selects, updates, and deletes. */
+
+struct btr_pcur_struct{
+	btr_cur_t	btr_cur;	/* a B-tree cursor */
+	ulint		latch_mode;	/* see TODO note below!
+					BTR_SEARCH_LEAF, BTR_MODIFY_LEAF,
+					BTR_MODIFY_TREE, or BTR_NO_LATCHES,
+					depending on the latching state of
+					the page and tree where the cursor is
+					positioned; the last value means that
+					the cursor is not currently positioned:
+					we say then that the cursor is
+					detached; it can be restored to
+					attached if the old position was
+					stored in old_rec */
+	ulint		old_stored;	/* BTR_PCUR_OLD_STORED
+					or BTR_PCUR_OLD_NOT_STORED */
+	rec_t*		old_rec;	/* if cursor position is stored,
+					contains an initial segment of the
+					latest record cursor was positioned
+					either on, before, or after */
+	ulint		old_n_fields;	/* number of fields in old_rec */
+	ulint		rel_pos;	/* BTR_PCUR_ON, BTR_PCUR_BEFORE, or
+					BTR_PCUR_AFTER, depending on whether
+					cursor was on, before, or after the
+					old_rec record */
+	buf_block_t*	block_when_stored;/* buffer block when the position was
+					stored */
+	ib_uint64_t	modify_clock;	/* the modify clock value of the
+					buffer block when the cursor position
+					was stored */
+	ulint		pos_state;	/* see TODO note below!
+					BTR_PCUR_IS_POSITIONED,
+					BTR_PCUR_WAS_POSITIONED,
+					BTR_PCUR_NOT_POSITIONED */
+	ulint		search_mode;	/* PAGE_CUR_G, ... */
+	trx_t*		trx_if_known;	/* the transaction, if we know it;
+					otherwise this field is not defined;
+					can ONLY BE USED in error prints in
+					fatal assertion failures! */
+	/*-----------------------------*/
+	/* NOTE that the following fields may possess dynamically allocated
+	memory which should be freed if not needed anymore! */
+
+	mtr_t*		mtr;		/* NULL, or this field may contain
+					a mini-transaction which holds the
+					latch on the cursor page */
+	byte*		old_rec_buf;	/* NULL, or a dynamically allocated
+					buffer for old_rec */
+	ulint		buf_size;	/* old_rec_buf size if old_rec_buf
+					is not NULL */
+};
+
+#define BTR_PCUR_IS_POSITIONED	1997660512	/* TODO: currently, the state
+						can be BTR_PCUR_IS_POSITIONED,
+						though it really should be
+						BTR_PCUR_WAS_POSITIONED,
+						because we have no obligation
+						to commit the cursor with
+						mtr; similarly latch_mode may
+						be out of date. This can
+						lead to problems if btr_pcur
+						is not used the right way;
+						all current code should be
+						ok. */
+#define BTR_PCUR_WAS_POSITIONED	1187549791
+#define BTR_PCUR_NOT_POSITIONED 1328997689
+
+#define BTR_PCUR_OLD_STORED	908467085
+#define BTR_PCUR_OLD_NOT_STORED	122766467
+
+#ifndef UNIV_NONINL
+#include "btr0pcur.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/btr0pcur.ic b/storage/xtradb/include/btr0pcur.ic
new file mode 100644
index 00000000000..bde7413820a
--- /dev/null
+++ b/storage/xtradb/include/btr0pcur.ic
@@ -0,0 +1,656 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The index tree persistent cursor
+
+Created 2/23/1996 Heikki Tuuri
+*******************************************************/
+
+
+/*************************************************************
+Gets the rel_pos field for a cursor whose position has been stored. */
+UNIV_INLINE
+ulint
+btr_pcur_get_rel_pos(
+/*=================*/
+					/* out: BTR_PCUR_ON, ... */
+	const btr_pcur_t*	cursor)	/* in: persistent cursor */
+{
+	ut_ad(cursor);
+	ut_ad(cursor->old_rec);
+	ut_ad(cursor->old_stored == BTR_PCUR_OLD_STORED);
+	ut_ad(cursor->pos_state == BTR_PCUR_WAS_POSITIONED
+	      || cursor->pos_state == BTR_PCUR_IS_POSITIONED);
+
+	return(cursor->rel_pos);
+}
+
+/*************************************************************
+Sets the mtr field for a pcur. */
+UNIV_INLINE
+void
+btr_pcur_set_mtr(
+/*=============*/
+	btr_pcur_t*	cursor,	/* in: persistent cursor */
+	mtr_t*		mtr)	/* in, own: mtr */
+{
+	ut_ad(cursor);
+
+	cursor->mtr = mtr;
+}
+
+/*************************************************************
+Gets the mtr field for a pcur. */
+UNIV_INLINE
+mtr_t*
+btr_pcur_get_mtr(
+/*=============*/
+				/* out: mtr */
+	btr_pcur_t*	cursor)	/* in: persistent cursor */
+{
+	ut_ad(cursor);
+
+	return(cursor->mtr);
+}
+
+#ifdef UNIV_DEBUG
+/*************************************************************
+Returns the btr cursor component of a persistent cursor. */
+UNIV_INLINE
+btr_cur_t*
+btr_pcur_get_btr_cur(
+/*=================*/
+					/* out: pointer to
+					btr cursor component */
+	const btr_pcur_t*	cursor)	/* in: persistent cursor */
+{
+	const btr_cur_t*	btr_cur = &cursor->btr_cur;
+	return((btr_cur_t*) btr_cur);
+}
+
+/*************************************************************
+Returns the page cursor component of a persistent cursor. */
+UNIV_INLINE
+page_cur_t*
+btr_pcur_get_page_cur(
+/*==================*/
+					/* out: pointer to page cursor
+					component */
+	const btr_pcur_t*	cursor)	/* in: persistent cursor */
+{
+	return(btr_cur_get_page_cur(btr_pcur_get_btr_cur(cursor)));
+}
+#endif /* UNIV_DEBUG */
+/*************************************************************
+Returns the page of a persistent cursor. */
+UNIV_INLINE
+page_t*
+btr_pcur_get_page(
+/*==============*/
+				/* out: pointer to the page */
+	btr_pcur_t*	cursor)	/* in: persistent cursor */
+{
+	ut_ad(cursor->pos_state == BTR_PCUR_IS_POSITIONED);
+
+	return(btr_cur_get_page(btr_pcur_get_btr_cur(cursor)));
+}
+
+/*************************************************************
+Returns the buffer block of a persistent cursor. */
+UNIV_INLINE
+buf_block_t*
+btr_pcur_get_block(
+/*===============*/
+				/* out: pointer to the block */
+	btr_pcur_t*	cursor)	/* in: persistent cursor */
+{
+	ut_ad(cursor->pos_state == BTR_PCUR_IS_POSITIONED);
+
+	return(btr_cur_get_block(btr_pcur_get_btr_cur(cursor)));
+}
+
+/*************************************************************
+Returns the record of a persistent cursor. */
+UNIV_INLINE
+rec_t*
+btr_pcur_get_rec(
+/*=============*/
+				/* out: pointer to the record */
+	btr_pcur_t*	cursor)	/* in: persistent cursor */
+{
+	ut_ad(cursor->pos_state == BTR_PCUR_IS_POSITIONED);
+	ut_ad(cursor->latch_mode != BTR_NO_LATCHES);
+
+	return(btr_cur_get_rec(btr_pcur_get_btr_cur(cursor)));
+}
+
+/******************************************************************
+Gets the up_match value for a pcur after a search. */
+UNIV_INLINE
+ulint
+btr_pcur_get_up_match(
+/*==================*/
+				/* out: number of matched fields at the cursor
+				or to the right if search mode was PAGE_CUR_GE,
+				otherwise undefined */
+	btr_pcur_t*	cursor) /* in: memory buffer for persistent cursor */
+{
+	btr_cur_t*	btr_cursor;
+
+	ut_ad((cursor->pos_state == BTR_PCUR_WAS_POSITIONED)
+	      || (cursor->pos_state == BTR_PCUR_IS_POSITIONED));
+
+	btr_cursor = btr_pcur_get_btr_cur(cursor);
+
+	ut_ad(btr_cursor->up_match != ULINT_UNDEFINED);
+
+	return(btr_cursor->up_match);
+}
+
+/******************************************************************
+Gets the low_match value for a pcur after a search. */
+UNIV_INLINE
+ulint
+btr_pcur_get_low_match(
+/*===================*/
+				/* out: number of matched fields at the cursor
+				or to the right if search mode was PAGE_CUR_LE,
+				otherwise undefined */
+	btr_pcur_t*	cursor) /* in: memory buffer for persistent cursor */
+{
+	btr_cur_t*	btr_cursor;
+
+	ut_ad((cursor->pos_state == BTR_PCUR_WAS_POSITIONED)
+	      || (cursor->pos_state == BTR_PCUR_IS_POSITIONED));
+
+	btr_cursor = btr_pcur_get_btr_cur(cursor);
+	ut_ad(btr_cursor->low_match != ULINT_UNDEFINED);
+
+	return(btr_cursor->low_match);
+}
+
+/*************************************************************
+Checks if the persistent cursor is after the last user record on
+a page. */
+UNIV_INLINE
+ibool
+btr_pcur_is_after_last_on_page(
+/*===========================*/
+	const btr_pcur_t*	cursor)	/* in: persistent cursor */
+{
+	ut_ad(cursor->pos_state == BTR_PCUR_IS_POSITIONED);
+	ut_ad(cursor->latch_mode != BTR_NO_LATCHES);
+
+	return(page_cur_is_after_last(btr_pcur_get_page_cur(cursor)));
+}
+
+/*************************************************************
+Checks if the persistent cursor is before the first user record on
+a page. */
+UNIV_INLINE
+ibool
+btr_pcur_is_before_first_on_page(
+/*=============================*/
+	const btr_pcur_t*	cursor)	/* in: persistent cursor */
+{
+	ut_ad(cursor->pos_state == BTR_PCUR_IS_POSITIONED);
+	ut_ad(cursor->latch_mode != BTR_NO_LATCHES);
+
+	return(page_cur_is_before_first(btr_pcur_get_page_cur(cursor)));
+}
+
+/*************************************************************
+Checks if the persistent cursor is on a user record. */
+UNIV_INLINE
+ibool
+btr_pcur_is_on_user_rec(
+/*====================*/
+	const btr_pcur_t*	cursor)	/* in: persistent cursor */
+{
+	ut_ad(cursor->pos_state == BTR_PCUR_IS_POSITIONED);
+	ut_ad(cursor->latch_mode != BTR_NO_LATCHES);
+
+	if (btr_pcur_is_before_first_on_page(cursor)
+	    || btr_pcur_is_after_last_on_page(cursor)) {
+
+		return(FALSE);
+	}
+
+	return(TRUE);
+}
+
+/*************************************************************
+Checks if the persistent cursor is before the first user record in
+the index tree. */
+UNIV_INLINE
+ibool
+btr_pcur_is_before_first_in_tree(
+/*=============================*/
+	btr_pcur_t*	cursor,	/* in: persistent cursor */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	ut_ad(cursor->pos_state == BTR_PCUR_IS_POSITIONED);
+	ut_ad(cursor->latch_mode != BTR_NO_LATCHES);
+
+	if (btr_page_get_prev(btr_pcur_get_page(cursor), mtr) != FIL_NULL) {
+
+		return(FALSE);
+	}
+
+	return(page_cur_is_before_first(btr_pcur_get_page_cur(cursor)));
+}
+
+/*************************************************************
+Checks if the persistent cursor is after the last user record in
+the index tree. */
+UNIV_INLINE
+ibool
+btr_pcur_is_after_last_in_tree(
+/*===========================*/
+	btr_pcur_t*	cursor,	/* in: persistent cursor */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	ut_ad(cursor->pos_state == BTR_PCUR_IS_POSITIONED);
+	ut_ad(cursor->latch_mode != BTR_NO_LATCHES);
+
+	if (btr_page_get_next(btr_pcur_get_page(cursor), mtr) != FIL_NULL) {
+
+		return(FALSE);
+	}
+
+	return(page_cur_is_after_last(btr_pcur_get_page_cur(cursor)));
+}
+
+/*************************************************************
+Moves the persistent cursor to the next record on the same page. */
+UNIV_INLINE
+void
+btr_pcur_move_to_next_on_page(
+/*==========================*/
+	btr_pcur_t*	cursor)	/* in/out: persistent cursor */
+{
+	ut_ad(cursor->pos_state == BTR_PCUR_IS_POSITIONED);
+	ut_ad(cursor->latch_mode != BTR_NO_LATCHES);
+
+	page_cur_move_to_next(btr_pcur_get_page_cur(cursor));
+
+	cursor->old_stored = BTR_PCUR_OLD_NOT_STORED;
+}
+
+/*************************************************************
+Moves the persistent cursor to the previous record on the same page. */
+UNIV_INLINE
+void
+btr_pcur_move_to_prev_on_page(
+/*==========================*/
+	btr_pcur_t*	cursor)	/* in/out: persistent cursor */
+{
+	ut_ad(cursor->pos_state == BTR_PCUR_IS_POSITIONED);
+	ut_ad(cursor->latch_mode != BTR_NO_LATCHES);
+
+	page_cur_move_to_prev(btr_pcur_get_page_cur(cursor));
+
+	cursor->old_stored = BTR_PCUR_OLD_NOT_STORED;
+}
+
+/*************************************************************
+Moves the persistent cursor to the last record on the same page. */
+UNIV_INLINE
+void
+btr_pcur_move_to_last_on_page(
+/*==========================*/
+	btr_pcur_t*	cursor,	/* in: persistent cursor */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	UT_NOT_USED(mtr);
+	ut_ad(cursor->latch_mode != BTR_NO_LATCHES);
+
+	page_cur_set_after_last(btr_pcur_get_block(cursor),
+				btr_pcur_get_page_cur(cursor));
+
+	cursor->old_stored = BTR_PCUR_OLD_NOT_STORED;
+}
+
+/*************************************************************
+Moves the persistent cursor to the next user record in the tree. If no user
+records are left, the cursor ends up 'after last in tree'. */
+UNIV_INLINE
+ibool
+btr_pcur_move_to_next_user_rec(
+/*===========================*/
+				/* out: TRUE if the cursor moved forward,
+				ending on a user record */
+	btr_pcur_t*	cursor,	/* in: persistent cursor; NOTE that the
+				function may release the page latch */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	ut_ad(cursor->pos_state == BTR_PCUR_IS_POSITIONED);
+	ut_ad(cursor->latch_mode != BTR_NO_LATCHES);
+	cursor->old_stored = BTR_PCUR_OLD_NOT_STORED;
+loop:
+	if (btr_pcur_is_after_last_on_page(cursor)) {
+
+		if (btr_pcur_is_after_last_in_tree(cursor, mtr)) {
+
+			return(FALSE);
+		}
+
+		btr_pcur_move_to_next_page(cursor, mtr);
+	} else {
+		btr_pcur_move_to_next_on_page(cursor);
+	}
+
+	if (btr_pcur_is_on_user_rec(cursor)) {
+
+		return(TRUE);
+	}
+
+	goto loop;
+}
+
+/*************************************************************
+Moves the persistent cursor to the next record in the tree. If no records are
+left, the cursor stays 'after last in tree'. */
+UNIV_INLINE
+ibool
+btr_pcur_move_to_next(
+/*==================*/
+				/* out: TRUE if the cursor was not after last
+				in tree */
+	btr_pcur_t*	cursor,	/* in: persistent cursor; NOTE that the
+				function may release the page latch */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	ut_ad(cursor->pos_state == BTR_PCUR_IS_POSITIONED);
+	ut_ad(cursor->latch_mode != BTR_NO_LATCHES);
+
+	cursor->old_stored = BTR_PCUR_OLD_NOT_STORED;
+
+	if (btr_pcur_is_after_last_on_page(cursor)) {
+
+		if (btr_pcur_is_after_last_in_tree(cursor, mtr)) {
+
+			return(FALSE);
+		}
+
+		btr_pcur_move_to_next_page(cursor, mtr);
+
+		return(TRUE);
+	}
+
+	btr_pcur_move_to_next_on_page(cursor);
+
+	return(TRUE);
+}
+
+/******************************************************************
+Commits the pcur mtr and sets the pcur latch mode to BTR_NO_LATCHES,
+that is, the cursor becomes detached. If there have been modifications
+to the page where pcur is positioned, this can be used instead of
+btr_pcur_release_leaf. Function btr_pcur_store_position should be used
+before calling this, if restoration of cursor is wanted later. */
+UNIV_INLINE
+void
+btr_pcur_commit(
+/*============*/
+	btr_pcur_t*	pcur)	/* in: persistent cursor */
+{
+	ut_a(pcur->pos_state == BTR_PCUR_IS_POSITIONED);
+
+	pcur->latch_mode = BTR_NO_LATCHES;
+
+	mtr_commit(pcur->mtr);
+
+	pcur->pos_state = BTR_PCUR_WAS_POSITIONED;
+}
+
+/******************************************************************
+Differs from btr_pcur_commit in that we can specify the mtr to commit. */
+UNIV_INLINE
+void
+btr_pcur_commit_specify_mtr(
+/*========================*/
+	btr_pcur_t*	pcur,	/* in: persistent cursor */
+	mtr_t*		mtr)	/* in: mtr to commit */
+{
+	ut_a(pcur->pos_state == BTR_PCUR_IS_POSITIONED);
+
+	pcur->latch_mode = BTR_NO_LATCHES;
+
+	mtr_commit(mtr);
+
+	pcur->pos_state = BTR_PCUR_WAS_POSITIONED;
+}
+
+/******************************************************************
+Sets the pcur latch mode to BTR_NO_LATCHES. */
+UNIV_INLINE
+void
+btr_pcur_detach(
+/*============*/
+	btr_pcur_t*	pcur)	/* in: persistent cursor */
+{
+	ut_a(pcur->pos_state == BTR_PCUR_IS_POSITIONED);
+
+	pcur->latch_mode = BTR_NO_LATCHES;
+
+	pcur->pos_state = BTR_PCUR_WAS_POSITIONED;
+}
+
+/******************************************************************
+Tests if a cursor is detached: that is the latch mode is BTR_NO_LATCHES. */
+UNIV_INLINE
+ibool
+btr_pcur_is_detached(
+/*=================*/
+				/* out: TRUE if detached */
+	btr_pcur_t*	pcur)	/* in: persistent cursor */
+{
+	if (pcur->latch_mode == BTR_NO_LATCHES) {
+
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
+
+/******************************************************************
+Sets the old_rec_buf field to NULL. */
+UNIV_INLINE
+void
+btr_pcur_init(
+/*==========*/
+	btr_pcur_t*	pcur)	/* in: persistent cursor */
+{
+	pcur->old_stored = BTR_PCUR_OLD_NOT_STORED;
+	pcur->old_rec_buf = NULL;
+	pcur->old_rec = NULL;
+}
+
+/******************************************************************
+Initializes and opens a persistent cursor to an index tree. It should be
+closed with btr_pcur_close. */
+UNIV_INLINE
+void
+btr_pcur_open(
+/*==========*/
+	dict_index_t*	index,	/* in: index */
+	const dtuple_t*	tuple,	/* in: tuple on which search done */
+	ulint		mode,	/* in: PAGE_CUR_L, ...;
+				NOTE that if the search is made using a unique
+				prefix of a record, mode should be
+				PAGE_CUR_LE, not PAGE_CUR_GE, as the latter
+				may end up on the previous page from the
+				record! */
+	ulint		latch_mode,/* in: BTR_SEARCH_LEAF, ... */
+	btr_pcur_t*	cursor, /* in: memory buffer for persistent cursor */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	btr_cur_t*	btr_cursor;
+
+	/* Initialize the cursor */
+
+	btr_pcur_init(cursor);
+
+	cursor->latch_mode = latch_mode;
+	cursor->search_mode = mode;
+
+	/* Search with the tree cursor */
+
+	btr_cursor = btr_pcur_get_btr_cur(cursor);
+
+	btr_cur_search_to_nth_level(index, 0, tuple, mode, latch_mode,
+				    btr_cursor, 0, mtr);
+	cursor->pos_state = BTR_PCUR_IS_POSITIONED;
+
+	cursor->trx_if_known = NULL;
+}
+
+/******************************************************************
+Opens an persistent cursor to an index tree without initializing the
+cursor. */
+UNIV_INLINE
+void
+btr_pcur_open_with_no_init(
+/*=======================*/
+	dict_index_t*	index,	/* in: index */
+	const dtuple_t*	tuple,	/* in: tuple on which search done */
+	ulint		mode,	/* in: PAGE_CUR_L, ...;
+				NOTE that if the search is made using a unique
+				prefix of a record, mode should be
+				PAGE_CUR_LE, not PAGE_CUR_GE, as the latter
+				may end up on the previous page of the
+				record! */
+	ulint		latch_mode,/* in: BTR_SEARCH_LEAF, ...;
+				NOTE that if has_search_latch != 0 then
+				we maybe do not acquire a latch on the cursor
+				page, but assume that the caller uses his
+				btr search latch to protect the record! */
+	btr_pcur_t*	cursor, /* in: memory buffer for persistent cursor */
+	ulint		has_search_latch,/* in: latch mode the caller
+				currently has on btr_search_latch:
+				RW_S_LATCH, or 0 */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	btr_cur_t*	btr_cursor;
+
+	cursor->latch_mode = latch_mode;
+	cursor->search_mode = mode;
+
+	/* Search with the tree cursor */
+
+	btr_cursor = btr_pcur_get_btr_cur(cursor);
+
+	btr_cur_search_to_nth_level(index, 0, tuple, mode, latch_mode,
+				    btr_cursor, has_search_latch, mtr);
+	cursor->pos_state = BTR_PCUR_IS_POSITIONED;
+
+	cursor->old_stored = BTR_PCUR_OLD_NOT_STORED;
+
+	cursor->trx_if_known = NULL;
+}
+
+/*********************************************************************
+Opens a persistent cursor at either end of an index. */
+UNIV_INLINE
+void
+btr_pcur_open_at_index_side(
+/*========================*/
+	ibool		from_left,	/* in: TRUE if open to the low end,
+					FALSE if to the high end */
+	dict_index_t*	index,		/* in: index */
+	ulint		latch_mode,	/* in: latch mode */
+	btr_pcur_t*	pcur,		/* in: cursor */
+	ibool		do_init,	/* in: TRUE if should be initialized */
+	mtr_t*		mtr)		/* in: mtr */
+{
+	pcur->latch_mode = latch_mode;
+
+	if (from_left) {
+		pcur->search_mode = PAGE_CUR_G;
+	} else {
+		pcur->search_mode = PAGE_CUR_L;
+	}
+
+	if (do_init) {
+		btr_pcur_init(pcur);
+	}
+
+	btr_cur_open_at_index_side(from_left, index, latch_mode,
+				   btr_pcur_get_btr_cur(pcur), mtr);
+	pcur->pos_state = BTR_PCUR_IS_POSITIONED;
+
+	pcur->old_stored = BTR_PCUR_OLD_NOT_STORED;
+
+	pcur->trx_if_known = NULL;
+}
+
+/**************************************************************************
+Positions a cursor at a randomly chosen position within a B-tree. */
+UNIV_INLINE
+void
+btr_pcur_open_at_rnd_pos(
+/*=====================*/
+	dict_index_t*	index,		/* in: index */
+	ulint		latch_mode,	/* in: BTR_SEARCH_LEAF, ... */
+	btr_pcur_t*	cursor,		/* in/out: B-tree pcur */
+	mtr_t*		mtr)		/* in: mtr */
+{
+	/* Initialize the cursor */
+
+	cursor->latch_mode = latch_mode;
+	cursor->search_mode = PAGE_CUR_G;
+
+	btr_pcur_init(cursor);
+
+	btr_cur_open_at_rnd_pos(index, latch_mode,
+				btr_pcur_get_btr_cur(cursor), mtr);
+	cursor->pos_state = BTR_PCUR_IS_POSITIONED;
+	cursor->old_stored = BTR_PCUR_OLD_NOT_STORED;
+
+	cursor->trx_if_known = NULL;
+}
+
+/******************************************************************
+Frees the possible memory heap of a persistent cursor and sets the latch
+mode of the persistent cursor to BTR_NO_LATCHES. */
+UNIV_INLINE
+void
+btr_pcur_close(
+/*===========*/
+	btr_pcur_t*	cursor)	/* in: persistent cursor */
+{
+	if (cursor->old_rec_buf != NULL) {
+
+		mem_free(cursor->old_rec_buf);
+
+		cursor->old_rec = NULL;
+		cursor->old_rec_buf = NULL;
+	}
+
+	cursor->btr_cur.page_cur.rec = NULL;
+	cursor->btr_cur.page_cur.block = NULL;
+	cursor->old_rec = NULL;
+	cursor->old_stored = BTR_PCUR_OLD_NOT_STORED;
+
+	cursor->latch_mode = BTR_NO_LATCHES;
+	cursor->pos_state = BTR_PCUR_NOT_POSITIONED;
+
+	cursor->trx_if_known = NULL;
+}
diff --git a/storage/xtradb/include/btr0sea.h b/storage/xtradb/include/btr0sea.h
new file mode 100644
index 00000000000..074e6595258
--- /dev/null
+++ b/storage/xtradb/include/btr0sea.h
@@ -0,0 +1,298 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/************************************************************************
+The index tree adaptive search
+
+Created 2/17/1996 Heikki Tuuri
+*************************************************************************/
+
+#ifndef btr0sea_h
+#define btr0sea_h
+
+#include "univ.i"
+
+#include "rem0rec.h"
+#include "dict0dict.h"
+#include "btr0types.h"
+#include "mtr0mtr.h"
+#include "ha0ha.h"
+
+/*********************************************************************
+Creates and initializes the adaptive search system at a database start. */
+UNIV_INTERN
+void
+btr_search_sys_create(
+/*==================*/
+	ulint	hash_size);	/* in: hash index hash table size */
+
+/************************************************************************
+Disable the adaptive hash search system and empty the index. */
+UNIV_INTERN
+void
+btr_search_disable(void);
+/*====================*/
+/************************************************************************
+Enable the adaptive hash search system. */
+UNIV_INTERN
+void
+btr_search_enable(void);
+/*====================*/
+
+/************************************************************************
+Returns search info for an index. */
+UNIV_INLINE
+btr_search_t*
+btr_search_get_info(
+/*================*/
+				/* out: search info; search mutex reserved */
+	dict_index_t*	index);	/* in: index */
+/*********************************************************************
+Creates and initializes a search info struct. */
+UNIV_INTERN
+btr_search_t*
+btr_search_info_create(
+/*===================*/
+				/* out, own: search info struct */
+	mem_heap_t*	heap);	/* in: heap where created */
+/*********************************************************************
+Returns the value of ref_count. The value is protected by
+btr_search_latch. */
+UNIV_INTERN
+ulint
+btr_search_info_get_ref_count(
+/*==========================*/
+				/* out: ref_count value. */
+	btr_search_t*   info);	/* in: search info. */
+/*************************************************************************
+Updates the search info. */
+UNIV_INLINE
+void
+btr_search_info_update(
+/*===================*/
+	dict_index_t*	index,	/* in: index of the cursor */
+	btr_cur_t*	cursor);/* in: cursor which was just positioned */
+/**********************************************************************
+Tries to guess the right search position based on the hash search info
+of the index. Note that if mode is PAGE_CUR_LE, which is used in inserts,
+and the function returns TRUE, then cursor->up_match and cursor->low_match
+both have sensible values. */
+UNIV_INTERN
+ibool
+btr_search_guess_on_hash(
+/*=====================*/
+					/* out: TRUE if succeeded */
+	dict_index_t*	index,		/* in: index */
+	btr_search_t*	info,		/* in: index search info */
+	const dtuple_t*	tuple,		/* in: logical record */
+	ulint		mode,		/* in: PAGE_CUR_L, ... */
+	ulint		latch_mode,	/* in: BTR_SEARCH_LEAF, ... */
+	btr_cur_t*	cursor,		/* out: tree cursor */
+	ulint		has_search_latch,/* in: latch mode the caller
+					currently has on btr_search_latch:
+					RW_S_LATCH, RW_X_LATCH, or 0 */
+	mtr_t*		mtr);		/* in: mtr */
+/************************************************************************
+Moves or deletes hash entries for moved records. If new_page is already hashed,
+then the hash index for page, if any, is dropped. If new_page is not hashed,
+and page is hashed, then a new hash index is built to new_page with the same
+parameters as page (this often happens when a page is split). */
+UNIV_INTERN
+void
+btr_search_move_or_delete_hash_entries(
+/*===================================*/
+	buf_block_t*	new_block,	/* in: records are copied
+					to this page */
+	buf_block_t*	block,		/* in: index page from which
+					records were copied, and the
+					copied records will be deleted
+					from this page */
+	dict_index_t*	index);		/* in: record descriptor */
+/************************************************************************
+Drops a page hash index. */
+UNIV_INTERN
+void
+btr_search_drop_page_hash_index(
+/*============================*/
+	buf_block_t*	block);	/* in: block containing index page,
+				s- or x-latched, or an index page
+				for which we know that
+				block->buf_fix_count == 0 */
+/************************************************************************
+Drops a page hash index when a page is freed from a fseg to the file system.
+Drops possible hash index if the page happens to be in the buffer pool. */
+UNIV_INTERN
+void
+btr_search_drop_page_hash_when_freed(
+/*=================================*/
+	ulint	space,		/* in: space id */
+	ulint	zip_size,	/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	ulint	page_no);	/* in: page number */
+/************************************************************************
+Updates the page hash index when a single record is inserted on a page. */
+UNIV_INTERN
+void
+btr_search_update_hash_node_on_insert(
+/*==================================*/
+	btr_cur_t*	cursor);/* in: cursor which was positioned to the
+				place to insert using btr_cur_search_...,
+				and the new record has been inserted next
+				to the cursor */
+/************************************************************************
+Updates the page hash index when a single record is inserted on a page. */
+UNIV_INTERN
+void
+btr_search_update_hash_on_insert(
+/*=============================*/
+	btr_cur_t*	cursor);/* in: cursor which was positioned to the
+				place to insert using btr_cur_search_...,
+				and the new record has been inserted next
+				to the cursor */
+/************************************************************************
+Updates the page hash index when a single record is deleted from a page. */
+UNIV_INTERN
+void
+btr_search_update_hash_on_delete(
+/*=============================*/
+	btr_cur_t*	cursor);/* in: cursor which was positioned on the
+				record to delete using btr_cur_search_...,
+				the record is not yet deleted */
+/************************************************************************
+Validates the search system. */
+UNIV_INTERN
+ibool
+btr_search_validate(void);
+/*======================*/
+				/* out: TRUE if ok */
+
+/* Flag: has the search system been enabled?
+Protected by btr_search_latch and btr_search_enabled_mutex. */
+extern char btr_search_enabled;
+
+/* The search info struct in an index */
+
+struct btr_search_struct{
+	ulint	ref_count;	/* Number of blocks in this index tree
+				that have search index built
+				i.e. block->index points to this index.
+				Protected by btr_search_latch except
+				when during initialization in
+				btr_search_info_create(). */
+
+	/* The following fields are not protected by any latch.
+	Unfortunately, this means that they must be aligned to
+	the machine word, i.e., they cannot be turned into bit-fields. */
+	buf_block_t* root_guess;/* the root page frame when it was last time
+				fetched, or NULL */
+	ulint	hash_analysis;	/* when this exceeds BTR_SEARCH_HASH_ANALYSIS,
+				the hash analysis starts; this is reset if no
+				success noticed */
+	ibool	last_hash_succ;	/* TRUE if the last search would have
+				succeeded, or did succeed, using the hash
+				index; NOTE that the value here is not exact:
+				it is not calculated for every search, and the
+				calculation itself is not always accurate! */
+	ulint	n_hash_potential;
+				/* number of consecutive searches
+				which would have succeeded, or did succeed,
+				using the hash index;
+				the range is 0 .. BTR_SEARCH_BUILD_LIMIT + 5 */
+	/*----------------------*/
+	ulint	n_fields;	/* recommended prefix length for hash search:
+				number of full fields */
+	ulint	n_bytes;	/* recommended prefix: number of bytes in
+				an incomplete field;
+				see also BTR_PAGE_MAX_REC_SIZE */
+	ibool	left_side;	/* TRUE or FALSE, depending on whether
+				the leftmost record of several records with
+				the same prefix should be indexed in the
+				hash index */
+	/*----------------------*/
+#ifdef UNIV_SEARCH_PERF_STAT
+	ulint	n_hash_succ;	/* number of successful hash searches thus
+				far */
+	ulint	n_hash_fail;	/* number of failed hash searches */
+	ulint	n_patt_succ;	/* number of successful pattern searches thus
+				far */
+	ulint	n_searches;	/* number of searches */
+#endif /* UNIV_SEARCH_PERF_STAT */
+#ifdef UNIV_DEBUG
+	ulint	magic_n;	/* magic number */
+# define BTR_SEARCH_MAGIC_N	1112765
+#endif /* UNIV_DEBUG */
+};
+
+/* The hash index system */
+
+typedef struct btr_search_sys_struct	btr_search_sys_t;
+
+struct btr_search_sys_struct{
+	hash_table_t*	hash_index;
+};
+
+extern btr_search_sys_t*	btr_search_sys;
+
+/* The latch protecting the adaptive search system: this latch protects the
+(1) hash index;
+(2) columns of a record to which we have a pointer in the hash index;
+
+but does NOT protect:
+
+(3) next record offset field in a record;
+(4) next or previous records on the same page.
+
+Bear in mind (3) and (4) when using the hash index.
+*/
+
+extern rw_lock_t*	btr_search_latch_temp;
+
+#define btr_search_latch	(*btr_search_latch_temp)
+
+#ifdef UNIV_SEARCH_PERF_STAT
+extern ulint	btr_search_n_succ;
+extern ulint	btr_search_n_hash_fail;
+#endif /* UNIV_SEARCH_PERF_STAT */
+
+/* After change in n_fields or n_bytes in info, this many rounds are waited
+before starting the hash analysis again: this is to save CPU time when there
+is no hope in building a hash index. */
+
+#define BTR_SEARCH_HASH_ANALYSIS	17
+
+/* Limit of consecutive searches for trying a search shortcut on the search
+pattern */
+
+#define BTR_SEARCH_ON_PATTERN_LIMIT	3
+
+/* Limit of consecutive searches for trying a search shortcut using the hash
+index */
+
+#define BTR_SEARCH_ON_HASH_LIMIT	3
+
+/* We do this many searches before trying to keep the search latch over calls
+from MySQL. If we notice someone waiting for the latch, we again set this
+much timeout. This is to reduce contention. */
+
+#define BTR_SEA_TIMEOUT			10000
+
+#ifndef UNIV_NONINL
+#include "btr0sea.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/btr0sea.ic b/storage/xtradb/include/btr0sea.ic
new file mode 100644
index 00000000000..c948d7e92af
--- /dev/null
+++ b/storage/xtradb/include/btr0sea.ic
@@ -0,0 +1,83 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/************************************************************************
+The index tree adaptive search
+
+Created 2/17/1996 Heikki Tuuri
+*************************************************************************/
+
+#include "dict0mem.h"
+#include "btr0cur.h"
+#include "buf0buf.h"
+
+/*************************************************************************
+Updates the search info. */
+UNIV_INTERN
+void
+btr_search_info_update_slow(
+/*========================*/
+	btr_search_t*	info,	/* in/out: search info */
+	btr_cur_t*	cursor);/* in: cursor which was just positioned */
+
+/************************************************************************
+Returns search info for an index. */
+UNIV_INLINE
+btr_search_t*
+btr_search_get_info(
+/*================*/
+				/* out: search info; search mutex reserved */
+	dict_index_t*	index)	/* in: index */
+{
+	ut_ad(index);
+
+	return(index->search_info);
+}
+
+/*************************************************************************
+Updates the search info. */
+UNIV_INLINE
+void
+btr_search_info_update(
+/*===================*/
+	dict_index_t*	index,	/* in: index of the cursor */
+	btr_cur_t*	cursor)	/* in: cursor which was just positioned */
+{
+	btr_search_t*	info;
+
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(!rw_lock_own(&btr_search_latch, RW_LOCK_SHARED));
+	ut_ad(!rw_lock_own(&btr_search_latch, RW_LOCK_EX));
+#endif /* UNIV_SYNC_DEBUG */
+
+	info = btr_search_get_info(index);
+
+	info->hash_analysis++;
+
+	if (info->hash_analysis < BTR_SEARCH_HASH_ANALYSIS) {
+
+		/* Do nothing */
+
+		return;
+
+	}
+
+	ut_ad(cursor->flag != BTR_CUR_HASH);
+
+	btr_search_info_update_slow(info, cursor);
+}
diff --git a/storage/xtradb/include/btr0types.h b/storage/xtradb/include/btr0types.h
new file mode 100644
index 00000000000..074b15fa68d
--- /dev/null
+++ b/storage/xtradb/include/btr0types.h
@@ -0,0 +1,47 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/************************************************************************
+The index tree general types
+
+Created 2/17/1996 Heikki Tuuri
+*************************************************************************/
+
+#ifndef btr0types_h
+#define btr0types_h
+
+#include "univ.i"
+
+#include "rem0types.h"
+#include "page0types.h"
+
+typedef struct btr_pcur_struct		btr_pcur_t;
+typedef struct btr_cur_struct		btr_cur_t;
+typedef struct btr_search_struct	btr_search_t;
+
+/* The size of a reference to data stored on a different page.
+The reference is stored at the end of the prefix of the field
+in the index record. */
+#define BTR_EXTERN_FIELD_REF_SIZE	20
+
+/* A BLOB field reference full of zero, for use in assertions and tests.
+Initially, BLOB field references are set to zero, in
+dtuple_convert_big_rec(). */
+extern const byte field_ref_zero[BTR_EXTERN_FIELD_REF_SIZE];
+
+#endif
diff --git a/storage/xtradb/include/buf0buddy.h b/storage/xtradb/include/buf0buddy.h
new file mode 100644
index 00000000000..f3e593151b5
--- /dev/null
+++ b/storage/xtradb/include/buf0buddy.h
@@ -0,0 +1,89 @@
+/*****************************************************************************
+
+Copyright (c) 2006, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Binary buddy allocator for compressed pages
+
+Created December 2006 by Marko Makela
+*******************************************************/
+
+#ifndef buf0buddy_h
+#define buf0buddy_h
+
+#ifdef UNIV_MATERIALIZE
+# undef UNIV_INLINE
+# define UNIV_INLINE
+#endif
+
+#include "univ.i"
+#include "buf0types.h"
+
+/**************************************************************************
+Allocate a block.  The thread calling this function must hold
+buf_pool_mutex and must not hold buf_pool_zip_mutex or any
+block->mutex.  The buf_pool_mutex may only be released and reacquired
+if lru != NULL.  This function should only be used for allocating
+compressed page frames or control blocks (buf_page_t).  Allocated
+control blocks must be properly initialized immediately after
+buf_buddy_alloc() has returned the memory, before releasing
+buf_pool_mutex. */
+UNIV_INLINE
+void*
+buf_buddy_alloc(
+/*============*/
+			/* out: allocated block,
+			possibly NULL if lru == NULL */
+	ulint	size,	/* in: block size, up to UNIV_PAGE_SIZE */
+	ibool*	lru)	/* in: pointer to a variable that will be assigned
+			TRUE if storage was allocated from the LRU list
+			and buf_pool_mutex was temporarily released,
+			or NULL if the LRU list should not be used */
+	__attribute__((malloc));
+
+/**************************************************************************
+Release a block. */
+UNIV_INLINE
+void
+buf_buddy_free(
+/*===========*/
+	void*	buf,	/* in: block to be freed, must not be
+			pointed to by the buffer pool */
+	ulint	size)	/* in: block size, up to UNIV_PAGE_SIZE */
+	__attribute__((nonnull));
+
+/** Statistics of buddy blocks of a given size. */
+struct buf_buddy_stat_struct {
+	/** Number of blocks allocated from the buddy system. */
+	ulint		used;
+	/** Number of blocks relocated by the buddy system. */
+	ib_uint64_t	relocated;
+	/** Total duration of block relocations, in microseconds. */
+	ib_uint64_t	relocated_usec;
+};
+
+typedef struct buf_buddy_stat_struct buf_buddy_stat_t;
+
+/** Statistics of the buddy system, indexed by block size.
+Protected by buf_pool_mutex. */
+extern buf_buddy_stat_t buf_buddy_stat[BUF_BUDDY_SIZES + 1];
+
+#ifndef UNIV_NONINL
+# include "buf0buddy.ic"
+#endif
+
+#endif /* buf0buddy_h */
diff --git a/storage/xtradb/include/buf0buddy.ic b/storage/xtradb/include/buf0buddy.ic
new file mode 100644
index 00000000000..769b9d11d94
--- /dev/null
+++ b/storage/xtradb/include/buf0buddy.ic
@@ -0,0 +1,129 @@
+/*****************************************************************************
+
+Copyright (c) 2006, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Binary buddy allocator for compressed pages
+
+Created December 2006 by Marko Makela
+*******************************************************/
+
+#ifdef UNIV_MATERIALIZE
+# undef UNIV_INLINE
+# define UNIV_INLINE
+#endif
+
+#include "buf0buf.h"
+#include "buf0buddy.h"
+#include "ut0ut.h"
+#include "sync0sync.h"
+
+/**************************************************************************
+Allocate a block.  The thread calling this function must hold
+buf_pool_mutex and must not hold buf_pool_zip_mutex or any block->mutex.
+The buf_pool_mutex may only be released and reacquired if lru != NULL. */
+UNIV_INTERN
+void*
+buf_buddy_alloc_low(
+/*================*/
+			/* out: allocated block,
+			possibly NULL if lru==NULL */
+	ulint	i,	/* in: index of buf_pool->zip_free[],
+			or BUF_BUDDY_SIZES */
+	ibool*	lru)	/* in: pointer to a variable that will be assigned
+			TRUE if storage was allocated from the LRU list
+			and buf_pool_mutex was temporarily released,
+			or NULL if the LRU list should not be used */
+	__attribute__((malloc));
+
+/**************************************************************************
+Deallocate a block. */
+UNIV_INTERN
+void
+buf_buddy_free_low(
+/*===============*/
+	void*	buf,	/* in: block to be freed, must not be
+			pointed to by the buffer pool */
+	ulint	i)	/* in: index of buf_pool->zip_free[],
+			or BUF_BUDDY_SIZES */
+	__attribute__((nonnull));
+
+/**************************************************************************
+Get the index of buf_pool->zip_free[] for a given block size. */
+UNIV_INLINE
+ulint
+buf_buddy_get_slot(
+/*===============*/
+			/* out: index of buf_pool->zip_free[],
+			or BUF_BUDDY_SIZES */
+	ulint	size)	/* in: block size */
+{
+	ulint	i;
+	ulint	s;
+
+	for (i = 0, s = BUF_BUDDY_LOW; s < size; i++, s <<= 1) {
+	}
+
+	ut_ad(i <= BUF_BUDDY_SIZES);
+	return(i);
+}
+
+/**************************************************************************
+Allocate a block.  The thread calling this function must hold
+buf_pool_mutex and must not hold buf_pool_zip_mutex or any
+block->mutex.  The buf_pool_mutex may only be released and reacquired
+if lru != NULL.  This function should only be used for allocating
+compressed page frames or control blocks (buf_page_t).  Allocated
+control blocks must be properly initialized immediately after
+buf_buddy_alloc() has returned the memory, before releasing
+buf_pool_mutex. */
+UNIV_INLINE
+void*
+buf_buddy_alloc(
+/*============*/
+			/* out: allocated block,
+			possibly NULL if lru == NULL */
+	ulint	size,	/* in: block size, up to UNIV_PAGE_SIZE */
+	ibool*	lru)	/* in: pointer to a variable that will be assigned
+			TRUE if storage was allocated from the LRU list
+			and buf_pool_mutex was temporarily released,
+			or NULL if the LRU list should not be used */
+{
+	ut_ad(buf_pool_mutex_own());
+
+	return(buf_buddy_alloc_low(buf_buddy_get_slot(size), lru));
+}
+
+/**************************************************************************
+Deallocate a block. */
+UNIV_INLINE
+void
+buf_buddy_free(
+/*===========*/
+	void*	buf,	/* in: block to be freed, must not be
+			pointed to by the buffer pool */
+	ulint	size)	/* in: block size, up to UNIV_PAGE_SIZE */
+{
+	ut_ad(buf_pool_mutex_own());
+
+	buf_buddy_free_low(buf, buf_buddy_get_slot(size));
+}
+
+#ifdef UNIV_MATERIALIZE
+# undef UNIV_INLINE
+# define UNIV_INLINE	UNIV_INLINE_ORIGINAL
+#endif
diff --git a/storage/xtradb/include/buf0buf.h b/storage/xtradb/include/buf0buf.h
new file mode 100644
index 00000000000..65b7ad18da9
--- /dev/null
+++ b/storage/xtradb/include/buf0buf.h
@@ -0,0 +1,1411 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The database buffer pool high-level routines
+
+Created 11/5/1995 Heikki Tuuri
+*******************************************************/
+
+#ifndef buf0buf_h
+#define buf0buf_h
+
+#include "univ.i"
+#include "fil0fil.h"
+#include "mtr0types.h"
+#include "buf0types.h"
+#include "sync0rw.h"
+#include "hash0hash.h"
+#include "ut0byte.h"
+#include "os0proc.h"
+#include "page0types.h"
+
+/* Modes for buf_page_get_gen */
+#define BUF_GET			10	/* get always */
+#define	BUF_GET_IF_IN_POOL	11	/* get if in pool */
+#define BUF_GET_NO_LATCH	14	/* get and bufferfix, but set no latch;
+					we have separated this case, because
+					it is error-prone programming not to
+					set a latch, and it should be used
+					with care */
+/* Modes for buf_page_get_known_nowait */
+#define BUF_MAKE_YOUNG	51
+#define BUF_KEEP_OLD	52
+/* Magic value to use instead of checksums when they are disabled */
+#define BUF_NO_CHECKSUM_MAGIC 0xDEADBEEFUL
+
+extern buf_pool_t*	buf_pool;	/* The buffer pool of the database */
+#ifdef UNIV_DEBUG
+extern ibool		buf_debug_prints;/* If this is set TRUE, the program
+					prints info whenever read or flush
+					occurs */
+#endif /* UNIV_DEBUG */
+extern ulint srv_buf_pool_write_requests; /* variable to count write request
+					  issued */
+
+/* States of a control block (@see buf_page_struct).
+The enumeration values must be 0..7. */
+enum buf_page_state {
+	BUF_BLOCK_ZIP_FREE = 0,		/* contains a free compressed page */
+	BUF_BLOCK_ZIP_PAGE,		/* contains a clean compressed page */
+	BUF_BLOCK_ZIP_DIRTY,		/* contains a compressed page that is
+					in the buf_pool->flush_list */
+
+	/* The constants for compressed-only pages must precede
+	BUF_BLOCK_NOT_USED; @see buf_block_state_valid() */
+
+	BUF_BLOCK_NOT_USED,		/* is in the free list */
+	BUF_BLOCK_READY_FOR_USE,	/* when buf_LRU_get_free_block returns
+					a block, it is in this state */
+	BUF_BLOCK_FILE_PAGE,		/* contains a buffered file page */
+	BUF_BLOCK_MEMORY,		/* contains some main memory object */
+	BUF_BLOCK_REMOVE_HASH		/* hash index should be removed
+					before putting to the free list */
+};
+
+/************************************************************************
+Creates the buffer pool. */
+UNIV_INTERN
+buf_pool_t*
+buf_pool_init(void);
+/*===============*/
+				/* out, own: buf_pool object, NULL if not
+				enough memory or error */
+/************************************************************************
+Frees the buffer pool at shutdown.  This must not be invoked before
+freeing all mutexes. */
+UNIV_INTERN
+void
+buf_pool_free(void);
+/*===============*/
+
+/************************************************************************
+Drops the adaptive hash index.  To prevent a livelock, this function
+is only to be called while holding btr_search_latch and while
+btr_search_enabled == FALSE. */
+UNIV_INTERN
+void
+buf_pool_drop_hash_index(void);
+/*==========================*/
+
+/************************************************************************
+Relocate a buffer control block.  Relocates the block on the LRU list
+and in buf_pool->page_hash.  Does not relocate bpage->list.
+The caller must take care of relocating bpage->list. */
+UNIV_INTERN
+void
+buf_relocate(
+/*=========*/
+	buf_page_t*	bpage,	/* in/out: control block being relocated;
+				buf_page_get_state(bpage) must be
+				BUF_BLOCK_ZIP_DIRTY or BUF_BLOCK_ZIP_PAGE */
+	buf_page_t*	dpage)	/* in/out: destination control block */
+	__attribute__((nonnull));
+/************************************************************************
+Resizes the buffer pool. */
+UNIV_INTERN
+void
+buf_pool_resize(void);
+/*=================*/
+/*************************************************************************
+Gets the current size of buffer buf_pool in bytes. */
+UNIV_INLINE
+ulint
+buf_pool_get_curr_size(void);
+/*========================*/
+			/* out: size in bytes */
+/************************************************************************
+Gets the smallest oldest_modification lsn for any page in the pool. Returns
+zero if all modified pages have been flushed to disk. */
+UNIV_INLINE
+ib_uint64_t
+buf_pool_get_oldest_modification(void);
+/*==================================*/
+				/* out: oldest modification in pool,
+				zero if none */
+/************************************************************************
+Allocates a buffer block. */
+UNIV_INLINE
+buf_block_t*
+buf_block_alloc(
+/*============*/
+				/* out, own: the allocated block,
+				in state BUF_BLOCK_MEMORY */
+	ulint	zip_size);	/* in: compressed page size in bytes,
+				or 0 if uncompressed tablespace */
+/************************************************************************
+Frees a buffer block which does not contain a file page. */
+UNIV_INLINE
+void
+buf_block_free(
+/*===========*/
+	buf_block_t*	block);	/* in, own: block to be freed */
+/*************************************************************************
+Copies contents of a buffer frame to a given buffer. */
+UNIV_INLINE
+byte*
+buf_frame_copy(
+/*===========*/
+					/* out: buf */
+	byte*			buf,	/* in: buffer to copy to */
+	const buf_frame_t*	frame);	/* in: buffer frame */
+/******************************************************************
+NOTE! The following macros should be used instead of buf_page_get_gen,
+to improve debugging. Only values RW_S_LATCH and RW_X_LATCH are allowed
+in LA! */
+#define buf_page_get(SP, ZS, OF, LA, MTR)	 buf_page_get_gen(\
+				SP, ZS, OF, LA, NULL,\
+				BUF_GET, __FILE__, __LINE__, MTR)
+/******************************************************************
+Use these macros to bufferfix a page with no latching. Remember not to
+read the contents of the page unless you know it is safe. Do not modify
+the contents of the page! We have separated this case, because it is
+error-prone programming not to set a latch, and it should be used
+with care. */
+#define buf_page_get_with_no_latch(SP, ZS, OF, MTR)	   buf_page_get_gen(\
+				SP, ZS, OF, RW_NO_LATCH, NULL,\
+				BUF_GET_NO_LATCH, __FILE__, __LINE__, MTR)
+/******************************************************************
+NOTE! The following macros should be used instead of
+buf_page_optimistic_get_func, to improve debugging. Only values RW_S_LATCH and
+RW_X_LATCH are allowed as LA! */
+#define buf_page_optimistic_get(LA, BL, MC, MTR)			     \
+	buf_page_optimistic_get_func(LA, BL, MC, __FILE__, __LINE__, MTR)
+/************************************************************************
+This is the general function used to get optimistic access to a database
+page. */
+UNIV_INTERN
+ibool
+buf_page_optimistic_get_func(
+/*=========================*/
+				/* out: TRUE if success */
+	ulint		rw_latch,/* in: RW_S_LATCH, RW_X_LATCH */
+	buf_block_t*	block,	/* in: guessed block */
+	ib_uint64_t	modify_clock,/* in: modify clock value if mode is
+				..._GUESS_ON_CLOCK */
+	const char*	file,	/* in: file name */
+	ulint		line,	/* in: line where called */
+	mtr_t*		mtr);	/* in: mini-transaction */
+/************************************************************************
+This is used to get access to a known database page, when no waiting can be
+done. */
+UNIV_INTERN
+ibool
+buf_page_get_known_nowait(
+/*======================*/
+				/* out: TRUE if success */
+	ulint		rw_latch,/* in: RW_S_LATCH, RW_X_LATCH */
+	buf_block_t*	block,	/* in: the known page */
+	ulint		mode,	/* in: BUF_MAKE_YOUNG or BUF_KEEP_OLD */
+	const char*	file,	/* in: file name */
+	ulint		line,	/* in: line where called */
+	mtr_t*		mtr);	/* in: mini-transaction */
+
+/***********************************************************************
+Given a tablespace id and page number tries to get that page. If the
+page is not in the buffer pool it is not loaded and NULL is returned.
+Suitable for using when holding the kernel mutex. */
+
+const buf_block_t*
+buf_page_try_get_func(
+/*==================*/
+	ulint		space_id,/* in: tablespace id */
+	ulint		page_no,/* in: page number */
+	const char*	file,	/* in: file name */
+	ulint		line,	/* in: line where called */
+	mtr_t*		mtr);	/* in: mini-transaction */
+
+#define buf_page_try_get(space_id, page_no, mtr)	\
+	buf_page_try_get_func(space_id, page_no, __FILE__, __LINE__, mtr);
+
+/************************************************************************
+Get read access to a compressed page (usually of type
+FIL_PAGE_TYPE_ZBLOB or FIL_PAGE_TYPE_ZBLOB2).
+The page must be released with buf_page_release_zip().
+NOTE: the page is not protected by any latch.  Mutual exclusion has to
+be implemented at a higher level.  In other words, all possible
+accesses to a given page through this function must be protected by
+the same set of mutexes or latches. */
+UNIV_INTERN
+buf_page_t*
+buf_page_get_zip(
+/*=============*/
+				/* out: pointer to the block,
+				or NULL if not compressed */
+	ulint		space,	/* in: space id */
+	ulint		zip_size,/* in: compressed page size */
+	ulint		offset);/* in: page number */
+/************************************************************************
+This is the general function used to get access to a database page. */
+UNIV_INTERN
+buf_block_t*
+buf_page_get_gen(
+/*=============*/
+				/* out: pointer to the block or NULL */
+	ulint		space,	/* in: space id */
+	ulint		zip_size,/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	ulint		offset,	/* in: page number */
+	ulint		rw_latch,/* in: RW_S_LATCH, RW_X_LATCH, RW_NO_LATCH */
+	buf_block_t*	guess,	/* in: guessed block or NULL */
+	ulint		mode,	/* in: BUF_GET, BUF_GET_IF_IN_POOL,
+				BUF_GET_NO_LATCH */
+	const char*	file,	/* in: file name */
+	ulint		line,	/* in: line where called */
+	mtr_t*		mtr);	/* in: mini-transaction */
+/************************************************************************
+Initializes a page to the buffer buf_pool. The page is usually not read
+from a file even if it cannot be found in the buffer buf_pool. This is one
+of the functions which perform to a block a state transition NOT_USED =>
+FILE_PAGE (the other is buf_page_get_gen). */
+UNIV_INTERN
+buf_block_t*
+buf_page_create(
+/*============*/
+			/* out: pointer to the block, page bufferfixed */
+	ulint	space,	/* in: space id */
+	ulint	offset,	/* in: offset of the page within space in units of
+			a page */
+	ulint	zip_size,/* in: compressed page size, or 0 */
+	mtr_t*	mtr);	/* in: mini-transaction handle */
+#ifdef UNIV_HOTBACKUP
+/************************************************************************
+Inits a page to the buffer buf_pool, for use in ibbackup --restore. */
+UNIV_INTERN
+void
+buf_page_init_for_backup_restore(
+/*=============================*/
+	ulint		space,	/* in: space id */
+	ulint		offset,	/* in: offset of the page within space
+				in units of a page */
+	ulint		zip_size,/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	buf_block_t*	block);	/* in: block to init */
+#endif /* UNIV_HOTBACKUP */
+/************************************************************************
+Releases a compressed-only page acquired with buf_page_get_zip(). */
+UNIV_INLINE
+void
+buf_page_release_zip(
+/*=================*/
+	buf_page_t*	bpage);		/* in: buffer block */
+/************************************************************************
+Decrements the bufferfix count of a buffer control block and releases
+a latch, if specified. */
+UNIV_INLINE
+void
+buf_page_release(
+/*=============*/
+	buf_block_t*	block,		/* in: buffer block */
+	ulint		rw_latch,	/* in: RW_S_LATCH, RW_X_LATCH,
+					RW_NO_LATCH */
+	mtr_t*		mtr);		/* in: mtr */
+/************************************************************************
+Moves a page to the start of the buffer pool LRU list. This high-level
+function can be used to prevent an important page from from slipping out of
+the buffer pool. */
+UNIV_INTERN
+void
+buf_page_make_young(
+/*================*/
+	buf_page_t*	bpage);	/* in: buffer block of a file page */
+/************************************************************************
+Returns TRUE if the page can be found in the buffer pool hash table. NOTE
+that it is possible that the page is not yet read from disk, though. */
+UNIV_INLINE
+ibool
+buf_page_peek(
+/*==========*/
+			/* out: TRUE if found from page hash table,
+			NOTE that the page is not necessarily yet read
+			from disk! */
+	ulint	space,	/* in: space id */
+	ulint	offset);/* in: page number */
+/************************************************************************
+Resets the check_index_page_at_flush field of a page if found in the buffer
+pool. */
+UNIV_INTERN
+void
+buf_reset_check_index_page_at_flush(
+/*================================*/
+	ulint	space,	/* in: space id */
+	ulint	offset);/* in: page number */
+#ifdef UNIV_DEBUG_FILE_ACCESSES
+/************************************************************************
+Sets file_page_was_freed TRUE if the page is found in the buffer pool.
+This function should be called when we free a file page and want the
+debug version to check that it is not accessed any more unless
+reallocated. */
+UNIV_INTERN
+buf_page_t*
+buf_page_set_file_page_was_freed(
+/*=============================*/
+			/* out: control block if found in page hash table,
+			otherwise NULL */
+	ulint	space,	/* in: space id */
+	ulint	offset);/* in: page number */
+/************************************************************************
+Sets file_page_was_freed FALSE if the page is found in the buffer pool.
+This function should be called when we free a file page and want the
+debug version to check that it is not accessed any more unless
+reallocated. */
+UNIV_INTERN
+buf_page_t*
+buf_page_reset_file_page_was_freed(
+/*===============================*/
+			/* out: control block if found in page hash table,
+			otherwise NULL */
+	ulint	space,	/* in: space id */
+	ulint	offset);	/* in: page number */
+#endif /* UNIV_DEBUG_FILE_ACCESSES */
+/************************************************************************
+Reads the freed_page_clock of a buffer block. */
+UNIV_INLINE
+ulint
+buf_page_get_freed_page_clock(
+/*==========================*/
+					/* out: freed_page_clock */
+	const buf_page_t*	bpage)	/* in: block */
+	__attribute__((pure));
+/************************************************************************
+Reads the freed_page_clock of a buffer block. */
+UNIV_INLINE
+ulint
+buf_block_get_freed_page_clock(
+/*===========================*/
+					/* out: freed_page_clock */
+	const buf_block_t*	block)	/* in: block */
+	__attribute__((pure));
+
+/************************************************************************
+Recommends a move of a block to the start of the LRU list if there is danger
+of dropping from the buffer pool. NOTE: does not reserve the buffer pool
+mutex. */
+UNIV_INLINE
+ibool
+buf_page_peek_if_too_old(
+/*=====================*/
+					/* out: TRUE if should be made
+					younger */
+	const buf_page_t*	bpage);	/* in: block to make younger */
+/************************************************************************
+Returns the current state of is_hashed of a page. FALSE if the page is
+not in the pool. NOTE that this operation does not fix the page in the
+pool if it is found there. */
+UNIV_INTERN
+ibool
+buf_page_peek_if_search_hashed(
+/*===========================*/
+			/* out: TRUE if page hash index is built in search
+			system */
+	ulint	space,	/* in: space id */
+	ulint	offset);/* in: page number */
+/************************************************************************
+Gets the youngest modification log sequence number for a frame.
+Returns zero if not file page or no modification occurred yet. */
+UNIV_INLINE
+ib_uint64_t
+buf_page_get_newest_modification(
+/*=============================*/
+					/* out: newest modification to page */
+	const buf_page_t*	bpage);	/* in: block containing the
+					page frame */
+/************************************************************************
+Increments the modify clock of a frame by 1. The caller must (1) own the
+buf_pool mutex and block bufferfix count has to be zero, (2) or own an x-lock
+on the block. */
+UNIV_INLINE
+void
+buf_block_modify_clock_inc(
+/*=======================*/
+	buf_block_t*	block);	/* in: block */
+/************************************************************************
+Returns the value of the modify clock. The caller must have an s-lock
+or x-lock on the block. */
+UNIV_INLINE
+ib_uint64_t
+buf_block_get_modify_clock(
+/*=======================*/
+				/* out: value */
+	buf_block_t*	block);	/* in: block */
+/************************************************************************
+Calculates a page checksum which is stored to the page when it is written
+to a file. Note that we must be careful to calculate the same value
+on 32-bit and 64-bit architectures. */
+UNIV_INTERN
+ulint
+buf_calc_page_new_checksum(
+/*=======================*/
+				/* out: checksum */
+	const byte*	page);	/* in: buffer page */
+/************************************************************************
+In versions < 4.0.14 and < 4.1.1 there was a bug that the checksum only
+looked at the first few bytes of the page. This calculates that old
+checksum.
+NOTE: we must first store the new formula checksum to
+FIL_PAGE_SPACE_OR_CHKSUM before calculating and storing this old checksum
+because this takes that field as an input! */
+UNIV_INTERN
+ulint
+buf_calc_page_old_checksum(
+/*=======================*/
+				/* out: checksum */
+	const byte*	 page);	/* in: buffer page */
+/************************************************************************
+Checks if a page is corrupt. */
+UNIV_INTERN
+ibool
+buf_page_is_corrupted(
+/*==================*/
+					/* out: TRUE if corrupted */
+	const byte*	read_buf,	/* in: a database page */
+	ulint		zip_size);	/* in: size of compressed page;
+					0 for uncompressed pages */
+/**************************************************************************
+Gets the space id, page offset, and byte offset within page of a
+pointer pointing to a buffer frame containing a file page. */
+UNIV_INLINE
+void
+buf_ptr_get_fsp_addr(
+/*=================*/
+	const void*	ptr,	/* in: pointer to a buffer frame */
+	ulint*		space,	/* out: space id */
+	fil_addr_t*	addr);	/* out: page offset and byte offset */
+/**************************************************************************
+Gets the hash value of a block. This can be used in searches in the
+lock hash table. */
+UNIV_INLINE
+ulint
+buf_block_get_lock_hash_val(
+/*========================*/
+					/* out: lock hash value */
+	const buf_block_t*	block)	/* in: block */
+	__attribute__((pure));
+#ifdef UNIV_DEBUG
+/*************************************************************************
+Finds a block in the buffer pool that points to a
+given compressed page. */
+UNIV_INTERN
+buf_block_t*
+buf_pool_contains_zip(
+/*==================*/
+				/* out: buffer block pointing to
+				the compressed page, or NULL */
+	const void*	data);	/* in: pointer to compressed page */
+#endif /* UNIV_DEBUG */
+#if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
+/*************************************************************************
+Validates the buffer pool data structure. */
+UNIV_INTERN
+ibool
+buf_validate(void);
+/*==============*/
+#endif /* UNIV_DEBUG || UNIV_BUF_DEBUG */
+#if defined UNIV_DEBUG_PRINT || defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
+/*************************************************************************
+Prints info of the buffer pool data structure. */
+UNIV_INTERN
+void
+buf_print(void);
+/*============*/
+#endif /* UNIV_DEBUG_PRINT || UNIV_DEBUG || UNIV_BUF_DEBUG */
+/************************************************************************
+Prints a page to stderr. */
+UNIV_INTERN
+void
+buf_page_print(
+/*===========*/
+	const byte*	read_buf,	/* in: a database page */
+	ulint		zip_size);	/* in: compressed page size, or
+					0 for uncompressed pages */
+#ifdef UNIV_DEBUG
+/*************************************************************************
+Returns the number of latched pages in the buffer pool. */
+UNIV_INTERN
+ulint
+buf_get_latched_pages_number(void);
+/*==============================*/
+#endif /* UNIV_DEBUG */
+/*************************************************************************
+Returns the number of pending buf pool ios. */
+UNIV_INTERN
+ulint
+buf_get_n_pending_ios(void);
+/*=======================*/
+/*************************************************************************
+Prints info of the buffer i/o. */
+UNIV_INTERN
+void
+buf_print_io(
+/*=========*/
+	FILE*	file);	/* in: file where to print */
+/*************************************************************************
+Returns the ratio in percents of modified pages in the buffer pool /
+database pages in the buffer pool. */
+UNIV_INTERN
+ulint
+buf_get_modified_ratio_pct(void);
+/*============================*/
+/**************************************************************************
+Refreshes the statistics used to print per-second averages. */
+UNIV_INTERN
+void
+buf_refresh_io_stats(void);
+/*======================*/
+/*************************************************************************
+Checks that all file pages in the buffer are in a replaceable state. */
+UNIV_INTERN
+ibool
+buf_all_freed(void);
+/*===============*/
+/*************************************************************************
+Checks that there currently are no pending i/o-operations for the buffer
+pool. */
+UNIV_INTERN
+ibool
+buf_pool_check_no_pending_io(void);
+/*==============================*/
+				/* out: TRUE if there is no pending i/o */
+/*************************************************************************
+Invalidates the file pages in the buffer pool when an archive recovery is
+completed. All the file pages buffered must be in a replaceable state when
+this function is called: not latched and not modified. */
+UNIV_INTERN
+void
+buf_pool_invalidate(void);
+/*=====================*/
+
+/*========================================================================
+--------------------------- LOWER LEVEL ROUTINES -------------------------
+=========================================================================*/
+
+#ifdef UNIV_SYNC_DEBUG
+/*************************************************************************
+Adds latch level info for the rw-lock protecting the buffer frame. This
+should be called in the debug version after a successful latching of a
+page if we know the latching order level of the acquired latch. */
+UNIV_INLINE
+void
+buf_block_dbg_add_level(
+/*====================*/
+	buf_block_t*	block,	/* in: buffer page
+				where we have acquired latch */
+	ulint		level);	/* in: latching order level */
+#else /* UNIV_SYNC_DEBUG */
+# define buf_block_dbg_add_level(block, level) /* nothing */
+#endif /* UNIV_SYNC_DEBUG */
+/*************************************************************************
+Gets the state of a block. */
+UNIV_INLINE
+enum buf_page_state
+buf_page_get_state(
+/*===============*/
+					/* out: state */
+	const buf_page_t*	bpage);	/* in: pointer to the control block */
+/*************************************************************************
+Gets the state of a block. */
+UNIV_INLINE
+enum buf_page_state
+buf_block_get_state(
+/*================*/
+					/* out: state */
+	const buf_block_t*	block)	/* in: pointer to the control block */
+	__attribute__((pure));
+/*************************************************************************
+Sets the state of a block. */
+UNIV_INLINE
+void
+buf_page_set_state(
+/*===============*/
+	buf_page_t*		bpage,	/* in/out: pointer to control block */
+	enum buf_page_state	state);	/* in: state */
+/*************************************************************************
+Sets the state of a block. */
+UNIV_INLINE
+void
+buf_block_set_state(
+/*================*/
+	buf_block_t*		block,	/* in/out: pointer to control block */
+	enum buf_page_state	state);	/* in: state */
+/*************************************************************************
+Determines if a block is mapped to a tablespace. */
+UNIV_INLINE
+ibool
+buf_page_in_file(
+/*=============*/
+					/* out: TRUE if mapped */
+	const buf_page_t*	bpage)	/* in: pointer to control block */
+	__attribute__((pure));
+/*************************************************************************
+Determines if a block should be on unzip_LRU list. */
+UNIV_INLINE
+ibool
+buf_page_belongs_to_unzip_LRU(
+/*==========================*/
+					/* out: TRUE if block belongs
+					to unzip_LRU */
+	const buf_page_t*	bpage)	/* in: pointer to control block */
+	__attribute__((pure));
+/*************************************************************************
+Determine the approximate LRU list position of a block. */
+UNIV_INLINE
+ulint
+buf_page_get_LRU_position(
+/*======================*/
+					/* out: LRU list position */
+	const buf_page_t*	bpage)	/* in: control block */
+	__attribute__((pure));
+
+/*************************************************************************
+Gets the mutex of a block. */
+UNIV_INLINE
+mutex_t*
+buf_page_get_mutex(
+/*===============*/
+					/* out: pointer to mutex
+					protecting bpage */
+	const buf_page_t*	bpage)	/* in: pointer to control block */
+	__attribute__((pure));
+
+/*************************************************************************
+Get the flush type of a page. */
+UNIV_INLINE
+enum buf_flush
+buf_page_get_flush_type(
+/*====================*/
+					/* out: flush type */
+	const buf_page_t*	bpage)	/* in: buffer page */
+	__attribute__((pure));
+/*************************************************************************
+Set the flush type of a page. */
+UNIV_INLINE
+void
+buf_page_set_flush_type(
+/*====================*/
+	buf_page_t*	bpage,		/* in: buffer page */
+	enum buf_flush	flush_type);	/* in: flush type */
+/*************************************************************************
+Map a block to a file page. */
+UNIV_INLINE
+void
+buf_block_set_file_page(
+/*====================*/
+	buf_block_t*		block,	/* in/out: pointer to control block */
+	ulint			space,	/* in: tablespace id */
+	ulint			page_no);/* in: page number */
+/*************************************************************************
+Gets the io_fix state of a block. */
+UNIV_INLINE
+enum buf_io_fix
+buf_page_get_io_fix(
+/*================*/
+					/* out: io_fix state */
+	const buf_page_t*	bpage)	/* in: pointer to the control block */
+	__attribute__((pure));
+/*************************************************************************
+Gets the io_fix state of a block. */
+UNIV_INLINE
+enum buf_io_fix
+buf_block_get_io_fix(
+/*================*/
+					/* out: io_fix state */
+	const buf_block_t*	block)	/* in: pointer to the control block */
+	__attribute__((pure));
+/*************************************************************************
+Sets the io_fix state of a block. */
+UNIV_INLINE
+void
+buf_page_set_io_fix(
+/*================*/
+	buf_page_t*	bpage,	/* in/out: control block */
+	enum buf_io_fix	io_fix);/* in: io_fix state */
+/*************************************************************************
+Sets the io_fix state of a block. */
+UNIV_INLINE
+void
+buf_block_set_io_fix(
+/*=================*/
+	buf_block_t*	block,	/* in/out: control block */
+	enum buf_io_fix	io_fix);/* in: io_fix state */
+
+/************************************************************************
+Determine if a buffer block can be relocated in memory.  The block
+can be dirty, but it must not be I/O-fixed or bufferfixed. */
+UNIV_INLINE
+ibool
+buf_page_can_relocate(
+/*==================*/
+	const buf_page_t*	bpage)	/* control block being relocated */
+	__attribute__((pure));
+
+/*************************************************************************
+Determine if a block has been flagged old. */
+UNIV_INLINE
+ibool
+buf_page_is_old(
+/*============*/
+					/* out: TRUE if old */
+	const buf_page_t*	bpage)	/* in: control block */
+	__attribute__((pure));
+/*************************************************************************
+Flag a block old. */
+UNIV_INLINE
+void
+buf_page_set_old(
+/*=============*/
+	buf_page_t*	bpage,	/* in/out: control block */
+	ibool		old);	/* in: old */
+/*************************************************************************
+Determine if a block has been accessed in the buffer pool. */
+UNIV_INLINE
+ibool
+buf_page_is_accessed(
+/*=================*/
+					/* out: TRUE if accessed */
+	const buf_page_t*	bpage)	/* in: control block */
+	__attribute__((pure));
+/*************************************************************************
+Flag a block accessed. */
+UNIV_INLINE
+void
+buf_page_set_accessed(
+/*==================*/
+	buf_page_t*	bpage,		/* in/out: control block */
+	ibool		accessed);	/* in: accessed */
+/*************************************************************************
+Gets the buf_block_t handle of a buffered file block if an uncompressed
+page frame exists, or NULL. */
+UNIV_INLINE
+buf_block_t*
+buf_page_get_block(
+/*===============*/
+				/* out: control block, or NULL */
+	buf_page_t*	bpage)	/* in: control block, or NULL */
+	__attribute__((pure));
+#ifdef UNIV_DEBUG
+/*************************************************************************
+Gets a pointer to the memory frame of a block. */
+UNIV_INLINE
+buf_frame_t*
+buf_block_get_frame(
+/*================*/
+					/* out: pointer to the frame */
+	const buf_block_t*	block)	/* in: pointer to the control block */
+	__attribute__((pure));
+#else /* UNIV_DEBUG */
+# define buf_block_get_frame(block) (block)->frame
+#endif /* UNIV_DEBUG */
+/*************************************************************************
+Gets the space id of a block. */
+UNIV_INLINE
+ulint
+buf_page_get_space(
+/*===============*/
+					/* out: space id */
+	const buf_page_t*	bpage)	/* in: pointer to the control block */
+	__attribute__((pure));
+/*************************************************************************
+Gets the space id of a block. */
+UNIV_INLINE
+ulint
+buf_block_get_space(
+/*================*/
+					/* out: space id */
+	const buf_block_t*	block)	/* in: pointer to the control block */
+	__attribute__((pure));
+/*************************************************************************
+Gets the page number of a block. */
+UNIV_INLINE
+ulint
+buf_page_get_page_no(
+/*=================*/
+					/* out: page number */
+	const buf_page_t*	bpage)	/* in: pointer to the control block */
+	__attribute__((pure));
+/*************************************************************************
+Gets the page number of a block. */
+UNIV_INLINE
+ulint
+buf_block_get_page_no(
+/*==================*/
+					/* out: page number */
+	const buf_block_t*	block)	/* in: pointer to the control block */
+	__attribute__((pure));
+/*************************************************************************
+Gets the compressed page size of a block. */
+UNIV_INLINE
+ulint
+buf_page_get_zip_size(
+/*==================*/
+					/* out: compressed page size, or 0 */
+	const buf_page_t*	bpage)	/* in: pointer to the control block */
+	__attribute__((pure));
+/*************************************************************************
+Gets the compressed page size of a block. */
+UNIV_INLINE
+ulint
+buf_block_get_zip_size(
+/*===================*/
+					/* out: compressed page size, or 0 */
+	const buf_block_t*	block)	/* in: pointer to the control block */
+	__attribute__((pure));
+/*************************************************************************
+Gets the compressed page descriptor corresponding to an uncompressed page
+if applicable. */
+#define buf_block_get_page_zip(block) \
+	(UNIV_LIKELY_NULL((block)->page.zip.data) ? &(block)->page.zip : NULL)
+/***********************************************************************
+Gets the block to whose frame the pointer is pointing to. */
+UNIV_INTERN
+buf_block_t*
+buf_block_align(
+/*============*/
+				/* out: pointer to block, never NULL */
+	const byte*	ptr);	/* in: pointer to a frame */
+#if defined UNIV_DEBUG || defined UNIV_ZIP_DEBUG
+/*************************************************************************
+Gets the compressed page descriptor corresponding to an uncompressed page
+if applicable. */
+UNIV_INLINE
+const page_zip_des_t*
+buf_frame_get_page_zip(
+/*===================*/
+				/* out: compressed page descriptor, or NULL */
+	const byte*	ptr);	/* in: pointer to the page */
+#endif /* UNIV_DEBUG || UNIV_ZIP_DEBUG */
+/************************************************************************
+This function is used to get info if there is an io operation
+going on on a buffer page. */
+UNIV_INLINE
+ibool
+buf_page_io_query(
+/*==============*/
+				/* out: TRUE if io going on */
+	buf_page_t*	bpage);	/* in: pool block, must be bufferfixed */
+/************************************************************************
+Function which inits a page for read to the buffer buf_pool. If the page is
+(1) already in buf_pool, or
+(2) if we specify to read only ibuf pages and the page is not an ibuf page, or
+(3) if the space is deleted or being deleted,
+then this function does nothing.
+Sets the io_fix flag to BUF_IO_READ and sets a non-recursive exclusive lock
+on the buffer frame. The io-handler must take care that the flag is cleared
+and the lock released later. */
+UNIV_INTERN
+buf_page_t*
+buf_page_init_for_read(
+/*===================*/
+				/* out: pointer to the block or NULL */
+	ulint*		err,	/* out: DB_SUCCESS or DB_TABLESPACE_DELETED */
+	ulint		mode,	/* in: BUF_READ_IBUF_PAGES_ONLY, ... */
+	ulint		space,	/* in: space id */
+	ulint		zip_size,/* in: compressed page size, or 0 */
+	ibool		unzip,	/* in: TRUE=request uncompressed page */
+	ib_int64_t	tablespace_version,/* in: prevents reading from a wrong
+				version of the tablespace in case we have done
+				DISCARD + IMPORT */
+	ulint		offset);/* in: page number */
+/************************************************************************
+Completes an asynchronous read or write request of a file page to or from
+the buffer pool. */
+UNIV_INTERN
+void
+buf_page_io_complete(
+/*=================*/
+	buf_page_t*	bpage);	/* in: pointer to the block in question */
+/************************************************************************
+Calculates a folded value of a file page address to use in the page hash
+table. */
+UNIV_INLINE
+ulint
+buf_page_address_fold(
+/*==================*/
+			/* out: the folded value */
+	ulint	space,	/* in: space id */
+	ulint	offset)	/* in: offset of the page within space */
+	__attribute__((const));
+/**********************************************************************
+Returns the control block of a file page, NULL if not found. */
+UNIV_INLINE
+buf_page_t*
+buf_page_hash_get(
+/*==============*/
+			/* out: block, NULL if not found */
+	ulint	space,	/* in: space id */
+	ulint	offset);/* in: offset of the page within space */
+/**********************************************************************
+Returns the control block of a file page, NULL if not found
+or an uncompressed page frame does not exist. */
+UNIV_INLINE
+buf_block_t*
+buf_block_hash_get(
+/*===============*/
+			/* out: block, NULL if not found */
+	ulint	space,	/* in: space id */
+	ulint	offset);/* in: offset of the page within space */
+/***********************************************************************
+Increments the pool clock by one and returns its new value. Remember that
+in the 32 bit version the clock wraps around at 4 billion! */
+UNIV_INLINE
+ulint
+buf_pool_clock_tic(void);
+/*====================*/
+			/* out: new clock value */
+/*************************************************************************
+Gets the current length of the free list of buffer blocks. */
+UNIV_INTERN
+ulint
+buf_get_free_list_len(void);
+/*=======================*/
+
+
+
+/* The common buffer control block structure
+for compressed and uncompressed frames */
+
+struct buf_page_struct{
+	/* None of the following bit-fields must be modified without
+	holding buf_page_get_mutex() [block->mutex or buf_pool_zip_mutex],
+	since they can be stored in the same machine word.  Some of them are
+	additionally protected by buf_pool_mutex. */
+
+	unsigned	space:32;	/* tablespace id; also protected
+					by buf_pool_mutex. */
+	unsigned	offset:32;	/* page number; also protected
+					by buf_pool_mutex. */
+
+	unsigned	state:3;	/* state of the control block
+					(@see enum buf_page_state); also
+					protected by buf_pool_mutex.
+					State transitions from
+					BUF_BLOCK_READY_FOR_USE to
+					BUF_BLOCK_MEMORY need not be
+					protected by buf_page_get_mutex(). */
+	unsigned	flush_type:2;	/* if this block is currently being
+					flushed to disk, this tells the
+					flush_type (@see enum buf_flush) */
+	unsigned	accessed:1;	/* TRUE if the page has been accessed
+					while in the buffer pool: read-ahead
+					may read in pages which have not been
+					accessed yet; a thread is allowed to
+					read this for heuristic purposes
+					without holding any mutex or latch */
+	unsigned	io_fix:2;	/* type of pending I/O operation
+					(@see enum buf_io_fix); also
+					protected by buf_pool_mutex */
+	unsigned	buf_fix_count:24;/* count of how manyfold this block
+					is currently bufferfixed */
+
+	page_zip_des_t	zip;		/* compressed page; zip.data
+					(but not the data it points to) is
+					also protected by buf_pool_mutex */
+	buf_page_t*	hash;		/* node used in chaining to
+					buf_pool->page_hash or
+					buf_pool->zip_hash */
+#ifdef UNIV_DEBUG
+	ibool		in_page_hash;	/* TRUE if in buf_pool->page_hash */
+	ibool		in_zip_hash;	/* TRUE if in buf_pool->zip_hash */
+#endif /* UNIV_DEBUG */
+
+	/* 2. Page flushing fields; protected by buf_pool_mutex */
+
+	UT_LIST_NODE_T(buf_page_t) list;
+					/* based on state, this is a list
+					node in one of the following lists
+					in buf_pool:
+
+					BUF_BLOCK_NOT_USED:	free
+					BUF_BLOCK_FILE_PAGE:	flush_list
+					BUF_BLOCK_ZIP_DIRTY:	flush_list
+					BUF_BLOCK_ZIP_PAGE:	zip_clean
+					BUF_BLOCK_ZIP_FREE:	zip_free[] */
+#ifdef UNIV_DEBUG
+	ibool		in_flush_list;	/* TRUE if in buf_pool->flush_list;
+					when buf_pool_mutex is free, the
+					following should hold: in_flush_list
+					== (state == BUF_BLOCK_FILE_PAGE
+					    || state == BUF_BLOCK_ZIP_DIRTY) */
+	ibool		in_free_list;	/* TRUE if in buf_pool->free; when
+					buf_pool_mutex is free, the following
+					should hold: in_free_list
+					== (state == BUF_BLOCK_NOT_USED) */
+#endif /* UNIV_DEBUG */
+	ib_uint64_t	newest_modification;
+					/* log sequence number of the youngest
+					modification to this block, zero if
+					not modified */
+	ib_uint64_t	oldest_modification;
+					/* log sequence number of the START of
+					the log entry written of the oldest
+					modification to this block which has
+					not yet been flushed on disk; zero if
+					all modifications are on disk */
+
+	/* 3. LRU replacement algorithm fields; protected by
+	buf_pool_mutex only (not buf_pool_zip_mutex or block->mutex) */
+
+	UT_LIST_NODE_T(buf_page_t) LRU;
+					/* node of the LRU list */
+//#ifdef UNIV_DEBUG
+	ibool		in_LRU_list;	/* TRUE if the page is in the LRU list;
+					used in debugging */
+//#endif /* UNIV_DEBUG */
+	unsigned	old:1;		/* TRUE if the block is in the old
+					blocks in the LRU list */
+	unsigned	LRU_position:31;/* value which monotonically decreases
+					(or may stay constant if old==TRUE)
+					toward the end of the LRU list, if
+					buf_pool->ulint_clock has not wrapped
+					around: NOTE that this value can only
+					be used in heuristic algorithms,
+					because of the possibility of a
+					wrap-around! */
+	unsigned	freed_page_clock:32;/* the value of
+					buf_pool->freed_page_clock when this
+					block was the last time put to the
+					head of the LRU list; a thread is
+					allowed to read this for heuristic
+					purposes without holding any mutex or
+					latch */
+#ifdef UNIV_DEBUG_FILE_ACCESSES
+	ibool		file_page_was_freed;
+					/* this is set to TRUE when fsp
+					frees a page in buffer pool */
+#endif /* UNIV_DEBUG_FILE_ACCESSES */
+};
+
+/* The buffer control block structure */
+
+struct buf_block_struct{
+
+	/* 1. General fields */
+
+	buf_page_t	page;		/* page information; this must
+					be the first field, so that
+					buf_pool->page_hash can point
+					to buf_page_t or buf_block_t */
+	UT_LIST_NODE_T(buf_block_t) unzip_LRU;
+					/* node of the decompressed LRU list;
+					a block is in the unzip_LRU list
+					if page.state == BUF_BLOCK_FILE_PAGE
+					and page.zip.data != NULL */
+#ifdef UNIV_DEBUG
+	ibool		in_unzip_LRU_list;/* TRUE if the page is in the
+					decompressed LRU list;
+					used in debugging */
+#endif /* UNIV_DEBUG */
+	byte*		frame;		/* pointer to buffer frame which
+					is of size UNIV_PAGE_SIZE, and
+					aligned to an address divisible by
+					UNIV_PAGE_SIZE */
+	mutex_t		mutex;		/* mutex protecting this block:
+					state (also protected by the buffer
+					pool mutex), io_fix, buf_fix_count,
+					and accessed; we introduce this new
+					mutex in InnoDB-5.1 to relieve
+					contention on the buffer pool mutex */
+	rw_lock_t	lock;		/* read-write lock of the buffer
+					frame */
+	unsigned	lock_hash_val:32;/* hashed value of the page address
+					in the record lock hash table */
+	unsigned	check_index_page_at_flush:1;
+					/* TRUE if we know that this is
+					an index page, and want the database
+					to check its consistency before flush;
+					note that there may be pages in the
+					buffer pool which are index pages,
+					but this flag is not set because
+					we do not keep track of all pages */
+
+	/* 2. Optimistic search field */
+
+	ib_uint64_t	modify_clock;	/* this clock is incremented every
+					time a pointer to a record on the
+					page may become obsolete; this is
+					used in the optimistic cursor
+					positioning: if the modify clock has
+					not changed, we know that the pointer
+					is still valid; this field may be
+					changed if the thread (1) owns the
+					pool mutex and the page is not
+					bufferfixed, or (2) the thread has an
+					x-latch on the block */
+
+	/* 3. Hash search fields: NOTE that the first 4 fields are NOT
+	protected by any semaphore! */
+
+	ulint		n_hash_helps;	/* counter which controls building
+					of a new hash index for the page */
+	ulint		n_fields;	/* recommended prefix length for hash
+					search: number of full fields */
+	ulint		n_bytes;	/* recommended prefix: number of bytes
+					in an incomplete field */
+	ibool		left_side;	/* TRUE or FALSE, depending on
+					whether the leftmost record of several
+					records with the same prefix should be
+					indexed in the hash index */
+
+	/* These 6 fields may only be modified when we have
+	an x-latch on btr_search_latch AND
+	a) we are holding an s-latch or x-latch on block->lock or
+	b) we know that block->buf_fix_count == 0.
+
+	An exception to this is when we init or create a page
+	in the buffer pool in buf0buf.c. */
+
+#if defined UNIV_AHI_DEBUG || defined UNIV_DEBUG
+	ulint		n_pointers;	/* used in debugging: the number of
+					pointers in the adaptive hash index
+					pointing to this frame */
+#endif /* UNIV_AHI_DEBUG || UNIV_DEBUG */
+	unsigned	is_hashed:1;	/* TRUE if hash index has already been
+					built on this page; note that it does
+					not guarantee that the index is
+					complete, though: there may have been
+					hash collisions, record deletions,
+					etc. */
+	unsigned	curr_n_fields:10;/* prefix length for hash indexing:
+					number of full fields */
+	unsigned	curr_n_bytes:15;/* number of bytes in hash indexing */
+	unsigned	curr_left_side:1;/* TRUE or FALSE in hash indexing */
+	dict_index_t*	index;		/* Index for which the adaptive
+					hash index has been created. */
+	/* 4. Debug fields */
+#ifdef UNIV_SYNC_DEBUG
+	rw_lock_t	debug_latch;	/* in the debug version, each thread
+					which bufferfixes the block acquires
+					an s-latch here; so we can use the
+					debug utilities in sync0rw */
+#endif
+};
+
+/* Check if a buf_block_t object is in a valid state. */
+#define buf_block_state_valid(block)				\
+(buf_block_get_state(block) >= BUF_BLOCK_NOT_USED		\
+ && (buf_block_get_state(block) <= BUF_BLOCK_REMOVE_HASH))
+
+/**************************************************************************
+Compute the hash fold value for blocks in buf_pool->zip_hash. */
+#define BUF_POOL_ZIP_FOLD_PTR(ptr) ((ulint) (ptr) / UNIV_PAGE_SIZE)
+#define BUF_POOL_ZIP_FOLD(b) BUF_POOL_ZIP_FOLD_PTR((b)->frame)
+#define BUF_POOL_ZIP_FOLD_BPAGE(b) BUF_POOL_ZIP_FOLD((buf_block_t*) (b))
+
+/* The buffer pool structure. NOTE! The definition appears here only for
+other modules of this directory (buf) to see it. Do not use from outside! */
+
+struct buf_pool_struct{
+
+	/* 1. General fields */
+
+	ulint		n_chunks;	/* number of buffer pool chunks */
+	buf_chunk_t*	chunks;		/* buffer pool chunks */
+	ulint		curr_size;	/* current pool size in pages */
+	hash_table_t*	page_hash;	/* hash table of buf_page_t or
+					buf_block_t file pages,
+					buf_page_in_file() == TRUE,
+					indexed by (space_id, offset) */
+	hash_table_t*	zip_hash;	/* hash table of buf_block_t blocks
+					whose frames are allocated to the
+					zip buddy system,
+					indexed by block->frame */
+	ulint		n_pend_reads;	/* number of pending read operations */
+	ulint		n_pend_unzip;	/* number of pending decompressions */
+
+	time_t		last_printout_time; /* when buf_print was last time
+					called */
+	ulint		n_pages_read;	/* number read operations */
+	ulint		n_pages_written;/* number write operations */
+	ulint		n_pages_created;/* number of pages created in the pool
+					with no read */
+	ulint		n_page_gets;	/* number of page gets performed;
+					also successful searches through
+					the adaptive hash index are
+					counted as page gets; this field
+					is NOT protected by the buffer
+					pool mutex */
+	ulint		n_page_gets_old;/* n_page_gets when buf_print was
+					last time called: used to calculate
+					hit rate */
+	ulint		n_pages_read_old;/* n_pages_read when buf_print was
+					last time called */
+	ulint		n_pages_written_old;/* number write operations */
+	ulint		n_pages_created_old;/* number of pages created in
+					the pool with no read */
+	/* 2. Page flushing algorithm fields */
+
+	UT_LIST_BASE_NODE_T(buf_page_t) flush_list;
+					/* base node of the modified block
+					list */
+	ibool		init_flush[BUF_FLUSH_N_TYPES];
+					/* this is TRUE when a flush of the
+					given type is being initialized */
+	ulint		n_flush[BUF_FLUSH_N_TYPES];
+					/* this is the number of pending
+					writes in the given flush type */
+	os_event_t	no_flush[BUF_FLUSH_N_TYPES];
+					/* this is in the set state when there
+					is no flush batch of the given type
+					running */
+	ulint		ulint_clock;	/* a sequence number used to count
+					time. NOTE! This counter wraps
+					around at 4 billion (if ulint ==
+					32 bits)! */
+	ulint		freed_page_clock;/* a sequence number used to count the
+					number of buffer blocks removed from
+					the end of the LRU list; NOTE that
+					this counter may wrap around at 4
+					billion! A thread is allowed to
+					read this for heuristic purposes
+					without holding any mutex or latch */
+	ulint		LRU_flush_ended;/* when an LRU flush ends for a page,
+					this is incremented by one; this is
+					set to zero when a buffer block is
+					allocated */
+
+	/* 3. LRU replacement algorithm fields */
+
+	UT_LIST_BASE_NODE_T(buf_page_t) free;
+					/* base node of the free block list */
+	UT_LIST_BASE_NODE_T(buf_page_t) LRU;
+					/* base node of the LRU list */
+	buf_page_t*	LRU_old;	/* pointer to the about 3/8 oldest
+					blocks in the LRU list; NULL if LRU
+					length less than BUF_LRU_OLD_MIN_LEN;
+					NOTE: when LRU_old != NULL, its length
+					should always equal LRU_old_len */
+	ulint		LRU_old_len;	/* length of the LRU list from
+					the block to which LRU_old points
+					onward, including that block;
+					see buf0lru.c for the restrictions
+					on this value; not defined if
+					LRU_old == NULL;
+					NOTE: LRU_old_len must be adjusted
+					whenever LRU_old shrinks or grows! */
+
+	UT_LIST_BASE_NODE_T(buf_block_t) unzip_LRU;
+					/* base node of the unzip_LRU list */
+
+	/* 4. Fields for the buddy allocator of compressed pages */
+	UT_LIST_BASE_NODE_T(buf_page_t)	zip_clean;
+					/* unmodified compressed pages */
+	UT_LIST_BASE_NODE_T(buf_page_t) zip_free[BUF_BUDDY_SIZES];
+					/* buddy free lists */
+#if BUF_BUDDY_HIGH != UNIV_PAGE_SIZE
+# error "BUF_BUDDY_HIGH != UNIV_PAGE_SIZE"
+#endif
+#if BUF_BUDDY_LOW > PAGE_ZIP_MIN_SIZE
+# error "BUF_BUDDY_LOW > PAGE_ZIP_MIN_SIZE"
+#endif
+};
+
+/* mutex protecting the buffer pool struct and control blocks, except the
+read-write lock in them */
+extern mutex_t	buf_pool_mutex;
+/* mutex protecting the control blocks of compressed-only pages
+(of type buf_page_t, not buf_block_t) */
+extern mutex_t	buf_pool_zip_mutex;
+
+/* Accessors for buf_pool_mutex.  Use these instead of accessing
+buf_pool_mutex directly. */
+
+/* Test if buf_pool_mutex is owned. */
+#define buf_pool_mutex_own() mutex_own(&buf_pool_mutex)
+/* Acquire the buffer pool mutex. */
+#define buf_pool_mutex_enter() do {		\
+	ut_ad(!mutex_own(&buf_pool_zip_mutex));	\
+	mutex_enter(&buf_pool_mutex);		\
+} while (0)
+
+#if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
+/** Flag to forbid the release of the buffer pool mutex.
+Protected by buf_pool_mutex. */
+extern ulint	buf_pool_mutex_exit_forbidden;
+/* Forbid the release of the buffer pool mutex. */
+# define buf_pool_mutex_exit_forbid() do {	\
+	ut_ad(buf_pool_mutex_own());		\
+	buf_pool_mutex_exit_forbidden++;	\
+} while (0)
+/* Allow the release of the buffer pool mutex. */
+# define buf_pool_mutex_exit_allow() do {	\
+	ut_ad(buf_pool_mutex_own());		\
+	ut_a(buf_pool_mutex_exit_forbidden);	\
+	buf_pool_mutex_exit_forbidden--;	\
+} while (0)
+/* Release the buffer pool mutex. */
+# define buf_pool_mutex_exit() do {		\
+	ut_a(!buf_pool_mutex_exit_forbidden);	\
+	mutex_exit(&buf_pool_mutex);		\
+} while (0)
+#else
+/* Forbid the release of the buffer pool mutex. */
+# define buf_pool_mutex_exit_forbid() ((void) 0)
+/* Allow the release of the buffer pool mutex. */
+# define buf_pool_mutex_exit_allow() ((void) 0)
+/* Release the buffer pool mutex. */
+# define buf_pool_mutex_exit() mutex_exit(&buf_pool_mutex)
+#endif
+
+/************************************************************************
+Let us list the consistency conditions for different control block states.
+
+NOT_USED:	is in free list, not in LRU list, not in flush list, nor
+		page hash table
+READY_FOR_USE:	is not in free list, LRU list, or flush list, nor page
+		hash table
+MEMORY:		is not in free list, LRU list, or flush list, nor page
+		hash table
+FILE_PAGE:	space and offset are defined, is in page hash table
+		if io_fix == BUF_IO_WRITE,
+			pool: no_flush[flush_type] is in reset state,
+			pool: n_flush[flush_type] > 0
+
+		(1) if buf_fix_count == 0, then
+			is in LRU list, not in free list
+			is in flush list,
+				if and only if oldest_modification > 0
+			is x-locked,
+				if and only if io_fix == BUF_IO_READ
+			is s-locked,
+				if and only if io_fix == BUF_IO_WRITE
+
+		(2) if buf_fix_count > 0, then
+			is not in LRU list, not in free list
+			is in flush list,
+				if and only if oldest_modification > 0
+			if io_fix == BUF_IO_READ,
+				is x-locked
+			if io_fix == BUF_IO_WRITE,
+				is s-locked
+
+State transitions:
+
+NOT_USED => READY_FOR_USE
+READY_FOR_USE => MEMORY
+READY_FOR_USE => FILE_PAGE
+MEMORY => NOT_USED
+FILE_PAGE => NOT_USED	NOTE: This transition is allowed if and only if
+				(1) buf_fix_count == 0,
+				(2) oldest_modification == 0, and
+				(3) io_fix == 0.
+*/
+
+#ifndef UNIV_NONINL
+#include "buf0buf.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/buf0buf.ic b/storage/xtradb/include/buf0buf.ic
new file mode 100644
index 00000000000..a1dbfe2ebde
--- /dev/null
+++ b/storage/xtradb/include/buf0buf.ic
@@ -0,0 +1,1076 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+Copyright (c) 2008, Google Inc.
+
+Portions of this file contain modifications contributed and copyrighted by
+Google, Inc. Those modifications are gratefully acknowledged and are described
+briefly in the InnoDB documentation. The contributions by Google are
+incorporated with their permission, and subject to the conditions contained in
+the file COPYING.Google.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The database buffer buf_pool
+
+Created 11/5/1995 Heikki Tuuri
+*******************************************************/
+
+#include "buf0flu.h"
+#include "buf0lru.h"
+#include "buf0rea.h"
+#include "mtr0mtr.h"
+
+/************************************************************************
+Reads the freed_page_clock of a buffer block. */
+UNIV_INLINE
+ulint
+buf_page_get_freed_page_clock(
+/*==========================*/
+					/* out: freed_page_clock */
+	const buf_page_t*	bpage)	/* in: block */
+{
+	/* This is sometimes read without holding buf_pool_mutex. */
+	return(bpage->freed_page_clock);
+}
+
+/************************************************************************
+Reads the freed_page_clock of a buffer block. */
+UNIV_INLINE
+ulint
+buf_block_get_freed_page_clock(
+/*===========================*/
+					/* out: freed_page_clock */
+	const buf_block_t*	block)	/* in: block */
+{
+	return(buf_page_get_freed_page_clock(&block->page));
+}
+
+/************************************************************************
+Recommends a move of a block to the start of the LRU list if there is danger
+of dropping from the buffer pool. NOTE: does not reserve the buffer pool
+mutex. */
+UNIV_INLINE
+ibool
+buf_page_peek_if_too_old(
+/*=====================*/
+					/* out: TRUE if should be made
+					younger */
+	const buf_page_t*	bpage)	/* in: block to make younger */
+{
+	return(buf_pool->freed_page_clock
+	       >= buf_page_get_freed_page_clock(bpage)
+	       + 1 + (buf_pool->curr_size / 4));
+}
+
+/*************************************************************************
+Gets the current size of buffer buf_pool in bytes. */
+UNIV_INLINE
+ulint
+buf_pool_get_curr_size(void)
+/*========================*/
+			/* out: size in bytes */
+{
+	return(buf_pool->curr_size * UNIV_PAGE_SIZE);
+}
+
+/************************************************************************
+Gets the smallest oldest_modification lsn for any page in the pool. Returns
+zero if all modified pages have been flushed to disk. */
+UNIV_INLINE
+ib_uint64_t
+buf_pool_get_oldest_modification(void)
+/*==================================*/
+				/* out: oldest modification in pool,
+				zero if none */
+{
+	buf_page_t*	bpage;
+	ib_uint64_t	lsn;
+
+	buf_pool_mutex_enter();
+
+	bpage = UT_LIST_GET_LAST(buf_pool->flush_list);
+
+	if (bpage == NULL) {
+		lsn = 0;
+	} else {
+		ut_ad(bpage->in_flush_list);
+		lsn = bpage->oldest_modification;
+	}
+
+	buf_pool_mutex_exit();
+
+	/* The returned answer may be out of date: the flush_list can
+	change after the mutex has been released. */
+
+	return(lsn);
+}
+
+/***********************************************************************
+Increments the buf_pool clock by one and returns its new value. Remember
+that in the 32 bit version the clock wraps around at 4 billion! */
+UNIV_INLINE
+ulint
+buf_pool_clock_tic(void)
+/*====================*/
+			/* out: new clock value */
+{
+	ut_ad(buf_pool_mutex_own());
+
+	buf_pool->ulint_clock++;
+
+	return(buf_pool->ulint_clock);
+}
+
+/*************************************************************************
+Gets the state of a block. */
+UNIV_INLINE
+enum buf_page_state
+buf_page_get_state(
+/*===============*/
+					/* out: state */
+	const buf_page_t*	bpage)	/* in: pointer to the control block */
+{
+	enum buf_page_state	state = (enum buf_page_state) bpage->state;
+
+#ifdef UNIV_DEBUG
+	switch (state) {
+	case BUF_BLOCK_ZIP_FREE:
+	case BUF_BLOCK_ZIP_PAGE:
+	case BUF_BLOCK_ZIP_DIRTY:
+	case BUF_BLOCK_NOT_USED:
+	case BUF_BLOCK_READY_FOR_USE:
+	case BUF_BLOCK_FILE_PAGE:
+	case BUF_BLOCK_MEMORY:
+	case BUF_BLOCK_REMOVE_HASH:
+		break;
+	default:
+		ut_error;
+	}
+#endif /* UNIV_DEBUG */
+
+	return(state);
+}
+/*************************************************************************
+Gets the state of a block. */
+UNIV_INLINE
+enum buf_page_state
+buf_block_get_state(
+/*================*/
+					/* out: state */
+	const buf_block_t*	block)	/* in: pointer to the control block */
+{
+	return(buf_page_get_state(&block->page));
+}
+/*************************************************************************
+Sets the state of a block. */
+UNIV_INLINE
+void
+buf_page_set_state(
+/*===============*/
+	buf_page_t*		bpage,	/* in/out: pointer to control block */
+	enum buf_page_state	state)	/* in: state */
+{
+#ifdef UNIV_DEBUG
+	enum buf_page_state	old_state	= buf_page_get_state(bpage);
+
+	switch (old_state) {
+	case BUF_BLOCK_ZIP_FREE:
+		ut_error;
+		break;
+	case BUF_BLOCK_ZIP_PAGE:
+		ut_a(state == BUF_BLOCK_ZIP_DIRTY);
+		break;
+	case BUF_BLOCK_ZIP_DIRTY:
+		ut_a(state == BUF_BLOCK_ZIP_PAGE);
+		break;
+	case BUF_BLOCK_NOT_USED:
+		ut_a(state == BUF_BLOCK_READY_FOR_USE);
+		break;
+	case BUF_BLOCK_READY_FOR_USE:
+		ut_a(state == BUF_BLOCK_MEMORY
+		     || state == BUF_BLOCK_FILE_PAGE
+		     || state == BUF_BLOCK_NOT_USED);
+		break;
+	case BUF_BLOCK_MEMORY:
+		ut_a(state == BUF_BLOCK_NOT_USED);
+		break;
+	case BUF_BLOCK_FILE_PAGE:
+		ut_a(state == BUF_BLOCK_NOT_USED
+		     || state == BUF_BLOCK_REMOVE_HASH);
+		break;
+	case BUF_BLOCK_REMOVE_HASH:
+		ut_a(state == BUF_BLOCK_MEMORY);
+		break;
+	}
+#endif /* UNIV_DEBUG */
+	bpage->state = state;
+	ut_ad(buf_page_get_state(bpage) == state);
+}
+
+/*************************************************************************
+Sets the state of a block. */
+UNIV_INLINE
+void
+buf_block_set_state(
+/*================*/
+	buf_block_t*		block,	/* in/out: pointer to control block */
+	enum buf_page_state	state)	/* in: state */
+{
+	buf_page_set_state(&block->page, state);
+}
+
+/*************************************************************************
+Determines if a block is mapped to a tablespace. */
+UNIV_INLINE
+ibool
+buf_page_in_file(
+/*=============*/
+					/* out: TRUE if mapped */
+	const buf_page_t*	bpage)	/* in: pointer to control block */
+{
+	switch (buf_page_get_state(bpage)) {
+	case BUF_BLOCK_ZIP_FREE:
+		/* This is a free page in buf_pool->zip_free[].
+		Such pages should only be accessed by the buddy allocator. */
+		ut_error;
+		break;
+	case BUF_BLOCK_ZIP_PAGE:
+	case BUF_BLOCK_ZIP_DIRTY:
+	case BUF_BLOCK_FILE_PAGE:
+		return(TRUE);
+	case BUF_BLOCK_NOT_USED:
+	case BUF_BLOCK_READY_FOR_USE:
+	case BUF_BLOCK_MEMORY:
+	case BUF_BLOCK_REMOVE_HASH:
+		break;
+	}
+
+	return(FALSE);
+}
+
+/*************************************************************************
+Determines if a block should be on unzip_LRU list. */
+UNIV_INLINE
+ibool
+buf_page_belongs_to_unzip_LRU(
+/*==========================*/
+					/* out: TRUE if block belongs
+					to unzip_LRU */
+	const buf_page_t*	bpage)	/* in: pointer to control block */
+{
+	ut_ad(buf_page_in_file(bpage));
+
+	return(bpage->zip.data
+	       && buf_page_get_state(bpage) == BUF_BLOCK_FILE_PAGE);
+}
+
+/*************************************************************************
+Determine the approximate LRU list position of a block. */
+UNIV_INLINE
+ulint
+buf_page_get_LRU_position(
+/*======================*/
+					/* out: LRU list position */
+	const buf_page_t*	bpage)	/* in: control block */
+{
+	ut_ad(buf_page_in_file(bpage));
+	ut_ad(buf_pool_mutex_own());
+
+	return(bpage->LRU_position);
+}
+
+/*************************************************************************
+Gets the mutex of a block. */
+UNIV_INLINE
+mutex_t*
+buf_page_get_mutex(
+/*===============*/
+					/* out: pointer to mutex
+					protecting bpage */
+	const buf_page_t*	bpage)	/* in: pointer to control block */
+{
+	switch (buf_page_get_state(bpage)) {
+	case BUF_BLOCK_ZIP_FREE:
+		ut_error;
+		return(NULL);
+	case BUF_BLOCK_ZIP_PAGE:
+	case BUF_BLOCK_ZIP_DIRTY:
+		return(&buf_pool_zip_mutex);
+	default:
+		return(&((buf_block_t*) bpage)->mutex);
+	}
+}
+
+/*************************************************************************
+Get the flush type of a page. */
+UNIV_INLINE
+enum buf_flush
+buf_page_get_flush_type(
+/*====================*/
+					/* out: flush type */
+	const buf_page_t*	bpage)	/* in: buffer page */
+{
+	enum buf_flush	flush_type = (enum buf_flush) bpage->flush_type;
+
+#ifdef UNIV_DEBUG
+	switch (flush_type) {
+	case BUF_FLUSH_LRU:
+	case BUF_FLUSH_SINGLE_PAGE:
+	case BUF_FLUSH_LIST:
+		return(flush_type);
+	case BUF_FLUSH_N_TYPES:
+		break;
+	}
+	ut_error;
+#endif /* UNIV_DEBUG */
+	return(flush_type);
+}
+/*************************************************************************
+Set the flush type of a page. */
+UNIV_INLINE
+void
+buf_page_set_flush_type(
+/*====================*/
+	buf_page_t*	bpage,		/* in: buffer page */
+	enum buf_flush	flush_type)	/* in: flush type */
+{
+	bpage->flush_type = flush_type;
+	ut_ad(buf_page_get_flush_type(bpage) == flush_type);
+}
+
+/*************************************************************************
+Map a block to a file page. */
+UNIV_INLINE
+void
+buf_block_set_file_page(
+/*====================*/
+	buf_block_t*		block,	/* in/out: pointer to control block */
+	ulint			space,	/* in: tablespace id */
+	ulint			page_no)/* in: page number */
+{
+	buf_block_set_state(block, BUF_BLOCK_FILE_PAGE);
+	block->page.space = space;
+	block->page.offset = page_no;
+}
+
+/*************************************************************************
+Gets the io_fix state of a block. */
+UNIV_INLINE
+enum buf_io_fix
+buf_page_get_io_fix(
+/*================*/
+					/* out: io_fix state */
+	const buf_page_t*	bpage)	/* in: pointer to the control block */
+{
+	enum buf_io_fix	io_fix = (enum buf_io_fix) bpage->io_fix;
+#ifdef UNIV_DEBUG
+	switch (io_fix) {
+	case BUF_IO_NONE:
+	case BUF_IO_READ:
+	case BUF_IO_WRITE:
+		return(io_fix);
+	}
+	ut_error;
+#endif /* UNIV_DEBUG */
+	return(io_fix);
+}
+
+/*************************************************************************
+Gets the io_fix state of a block. */
+UNIV_INLINE
+enum buf_io_fix
+buf_block_get_io_fix(
+/*================*/
+					/* out: io_fix state */
+	const buf_block_t*	block)	/* in: pointer to the control block */
+{
+	return(buf_page_get_io_fix(&block->page));
+}
+
+/*************************************************************************
+Sets the io_fix state of a block. */
+UNIV_INLINE
+void
+buf_page_set_io_fix(
+/*================*/
+	buf_page_t*	bpage,	/* in/out: control block */
+	enum buf_io_fix	io_fix)	/* in: io_fix state */
+{
+	ut_ad(buf_pool_mutex_own());
+	ut_ad(mutex_own(buf_page_get_mutex(bpage)));
+
+	bpage->io_fix = io_fix;
+	ut_ad(buf_page_get_io_fix(bpage) == io_fix);
+}
+
+/*************************************************************************
+Sets the io_fix state of a block. */
+UNIV_INLINE
+void
+buf_block_set_io_fix(
+/*=================*/
+	buf_block_t*	block,	/* in/out: control block */
+	enum buf_io_fix	io_fix)	/* in: io_fix state */
+{
+	buf_page_set_io_fix(&block->page, io_fix);
+}
+
+/************************************************************************
+Determine if a buffer block can be relocated in memory.  The block
+can be dirty, but it must not be I/O-fixed or bufferfixed. */
+UNIV_INLINE
+ibool
+buf_page_can_relocate(
+/*==================*/
+	const buf_page_t*	bpage)	/* control block being relocated */
+{
+	ut_ad(buf_pool_mutex_own());
+	ut_ad(mutex_own(buf_page_get_mutex(bpage)));
+	ut_ad(buf_page_in_file(bpage));
+	ut_ad(bpage->in_LRU_list);
+
+	return(buf_page_get_io_fix(bpage) == BUF_IO_NONE
+	       && bpage->buf_fix_count == 0);
+}
+
+/*************************************************************************
+Determine if a block has been flagged old. */
+UNIV_INLINE
+ibool
+buf_page_is_old(
+/*============*/
+					/* out: TRUE if old */
+	const buf_page_t*	bpage)	/* in: control block */
+{
+	ut_ad(buf_page_in_file(bpage));
+	ut_ad(buf_pool_mutex_own());
+
+	return(bpage->old);
+}
+
+/*************************************************************************
+Flag a block old. */
+UNIV_INLINE
+void
+buf_page_set_old(
+/*=============*/
+	buf_page_t*	bpage,	/* in/out: control block */
+	ibool		old)	/* in: old */
+{
+	ut_a(buf_page_in_file(bpage));
+	ut_ad(buf_pool_mutex_own());
+	ut_ad(bpage->in_LRU_list);
+
+#ifdef UNIV_LRU_DEBUG
+	if (UT_LIST_GET_PREV(LRU, bpage) && UT_LIST_GET_NEXT(LRU, bpage)
+	    && UT_LIST_GET_PREV(LRU, bpage)->old
+	    == UT_LIST_GET_NEXT(LRU, bpage)->old) {
+		ut_a(UT_LIST_GET_PREV(LRU, bpage)->old == old);
+	}
+#endif /* UNIV_LRU_DEBUG */
+
+	bpage->old = old;
+}
+
+/*************************************************************************
+Determine if a block has been accessed in the buffer pool. */
+UNIV_INLINE
+ibool
+buf_page_is_accessed(
+/*=================*/
+					/* out: TRUE if accessed */
+	const buf_page_t*	bpage)	/* in: control block */
+{
+	ut_ad(buf_page_in_file(bpage));
+
+	return(bpage->accessed);
+}
+
+/*************************************************************************
+Flag a block accessed. */
+UNIV_INLINE
+void
+buf_page_set_accessed(
+/*==================*/
+	buf_page_t*	bpage,		/* in/out: control block */
+	ibool		accessed)	/* in: accessed */
+{
+	ut_a(buf_page_in_file(bpage));
+	ut_ad(mutex_own(buf_page_get_mutex(bpage)));
+
+	bpage->accessed = accessed;
+}
+
+/*************************************************************************
+Gets the buf_block_t handle of a buffered file block if an uncompressed
+page frame exists, or NULL. */
+UNIV_INLINE
+buf_block_t*
+buf_page_get_block(
+/*===============*/
+				/* out: control block, or NULL */
+	buf_page_t*	bpage)	/* in: control block, or NULL */
+{
+	if (UNIV_LIKELY(bpage != NULL)) {
+		ut_ad(buf_page_in_file(bpage));
+
+		if (buf_page_get_state(bpage) == BUF_BLOCK_FILE_PAGE) {
+			return((buf_block_t*) bpage);
+		}
+	}
+
+	return(NULL);
+}
+
+#ifdef UNIV_DEBUG
+/*************************************************************************
+Gets a pointer to the memory frame of a block. */
+UNIV_INLINE
+buf_frame_t*
+buf_block_get_frame(
+/*================*/
+					/* out: pointer to the frame */
+	const buf_block_t*	block)	/* in: pointer to the control block */
+{
+	ut_ad(block);
+
+	switch (buf_block_get_state(block)) {
+	case BUF_BLOCK_ZIP_FREE:
+	case BUF_BLOCK_ZIP_PAGE:
+	case BUF_BLOCK_ZIP_DIRTY:
+	case BUF_BLOCK_NOT_USED:
+		ut_error;
+		break;
+	case BUF_BLOCK_FILE_PAGE:
+		ut_a(block->page.buf_fix_count > 0);
+		/* fall through */
+	case BUF_BLOCK_READY_FOR_USE:
+	case BUF_BLOCK_MEMORY:
+	case BUF_BLOCK_REMOVE_HASH:
+		goto ok;
+	}
+	ut_error;
+ok:
+	return((buf_frame_t*) block->frame);
+}
+#endif /* UNIV_DEBUG */
+
+/*************************************************************************
+Gets the space id of a block. */
+UNIV_INLINE
+ulint
+buf_page_get_space(
+/*===============*/
+					/* out: space id */
+	const buf_page_t*	bpage)	/* in: pointer to the control block */
+{
+	ut_ad(bpage);
+	ut_a(buf_page_in_file(bpage));
+
+	return(bpage->space);
+}
+
+/*************************************************************************
+Gets the space id of a block. */
+UNIV_INLINE
+ulint
+buf_block_get_space(
+/*================*/
+					/* out: space id */
+	const buf_block_t*	block)	/* in: pointer to the control block */
+{
+	ut_ad(block);
+	ut_a(buf_block_get_state(block) == BUF_BLOCK_FILE_PAGE);
+
+	return(block->page.space);
+}
+
+/*************************************************************************
+Gets the page number of a block. */
+UNIV_INLINE
+ulint
+buf_page_get_page_no(
+/*=================*/
+					/* out: page number */
+	const buf_page_t*	bpage)	/* in: pointer to the control block */
+{
+	ut_ad(bpage);
+	ut_a(buf_page_in_file(bpage));
+
+	return(bpage->offset);
+}
+
+/*************************************************************************
+Gets the page number of a block. */
+UNIV_INLINE
+ulint
+buf_block_get_page_no(
+/*==================*/
+					/* out: page number */
+	const buf_block_t*	block)	/* in: pointer to the control block */
+{
+	ut_ad(block);
+	ut_a(buf_block_get_state(block) == BUF_BLOCK_FILE_PAGE);
+
+	return(block->page.offset);
+}
+
+/*************************************************************************
+Gets the compressed page size of a block. */
+UNIV_INLINE
+ulint
+buf_page_get_zip_size(
+/*==================*/
+					/* out: compressed page size, or 0 */
+	const buf_page_t*	bpage)	/* in: pointer to the control block */
+{
+	return(bpage->zip.ssize ? 512 << bpage->zip.ssize : 0);
+}
+
+/*************************************************************************
+Gets the compressed page size of a block. */
+UNIV_INLINE
+ulint
+buf_block_get_zip_size(
+/*===================*/
+					/* out: compressed page size, or 0 */
+	const buf_block_t*	block)	/* in: pointer to the control block */
+{
+	return(block->page.zip.ssize ? 512 << block->page.zip.ssize : 0);
+}
+
+#if defined UNIV_DEBUG || defined UNIV_ZIP_DEBUG
+/*************************************************************************
+Gets the compressed page descriptor corresponding to an uncompressed page
+if applicable. */
+UNIV_INLINE
+const page_zip_des_t*
+buf_frame_get_page_zip(
+/*===================*/
+				/* out: compressed page descriptor, or NULL */
+	const byte*	ptr)	/* in: pointer to the page */
+{
+	return(buf_block_get_page_zip(buf_block_align(ptr)));
+}
+#endif /* UNIV_DEBUG || UNIV_ZIP_DEBUG */
+
+/**************************************************************************
+Gets the space id, page offset, and byte offset within page of a
+pointer pointing to a buffer frame containing a file page. */
+UNIV_INLINE
+void
+buf_ptr_get_fsp_addr(
+/*=================*/
+	const void*	ptr,	/* in: pointer to a buffer frame */
+	ulint*		space,	/* out: space id */
+	fil_addr_t*	addr)	/* out: page offset and byte offset */
+{
+	const page_t*	page = (const page_t*) ut_align_down(ptr,
+							     UNIV_PAGE_SIZE);
+
+	*space = mach_read_from_4(page + FIL_PAGE_ARCH_LOG_NO_OR_SPACE_ID);
+	addr->page = mach_read_from_4(page + FIL_PAGE_OFFSET);
+	addr->boffset = ut_align_offset(ptr, UNIV_PAGE_SIZE);
+}
+
+/**************************************************************************
+Gets the hash value of the page the pointer is pointing to. This can be used
+in searches in the lock hash table. */
+UNIV_INLINE
+ulint
+buf_block_get_lock_hash_val(
+/*========================*/
+					/* out: lock hash value */
+	const buf_block_t*	block)	/* in: block */
+{
+	return(block->lock_hash_val);
+}
+
+/************************************************************************
+Allocates a buffer block. */
+UNIV_INLINE
+buf_block_t*
+buf_block_alloc(
+/*============*/
+				/* out, own: the allocated block,
+				in state BUF_BLOCK_MEMORY */
+	ulint	zip_size)	/* in: compressed page size in bytes,
+				or 0 if uncompressed tablespace */
+{
+	buf_block_t*	block;
+
+	block = buf_LRU_get_free_block(zip_size);
+
+	buf_block_set_state(block, BUF_BLOCK_MEMORY);
+
+	return(block);
+}
+
+/************************************************************************
+Frees a buffer block which does not contain a file page. */
+UNIV_INLINE
+void
+buf_block_free(
+/*===========*/
+	buf_block_t*	block)	/* in, own: block to be freed */
+{
+	buf_pool_mutex_enter();
+
+	mutex_enter(&block->mutex);
+
+	ut_a(buf_block_get_state(block) != BUF_BLOCK_FILE_PAGE);
+
+	buf_LRU_block_free_non_file_page(block);
+
+	mutex_exit(&block->mutex);
+
+	buf_pool_mutex_exit();
+}
+
+/*************************************************************************
+Copies contents of a buffer frame to a given buffer. */
+UNIV_INLINE
+byte*
+buf_frame_copy(
+/*===========*/
+					/* out: buf */
+	byte*			buf,	/* in: buffer to copy to */
+	const buf_frame_t*	frame)	/* in: buffer frame */
+{
+	ut_ad(buf && frame);
+
+	ut_memcpy(buf, frame, UNIV_PAGE_SIZE);
+
+	return(buf);
+}
+
+/************************************************************************
+Calculates a folded value of a file page address to use in the page hash
+table. */
+UNIV_INLINE
+ulint
+buf_page_address_fold(
+/*==================*/
+			/* out: the folded value */
+	ulint	space,	/* in: space id */
+	ulint	offset)	/* in: offset of the page within space */
+{
+	return((space << 20) + space + offset);
+}
+
+/************************************************************************
+This function is used to get info if there is an io operation
+going on on a buffer page. */
+UNIV_INLINE
+ibool
+buf_page_io_query(
+/*==============*/
+				/* out: TRUE if io going on */
+	buf_page_t*	bpage)	/* in: buf_pool block, must be bufferfixed */
+{
+	ibool	io_fixed;
+
+	buf_pool_mutex_enter();
+
+	ut_ad(buf_page_in_file(bpage));
+	ut_ad(bpage->buf_fix_count > 0);
+
+	io_fixed = buf_page_get_io_fix(bpage) != BUF_IO_NONE;
+	buf_pool_mutex_exit();
+
+	return(io_fixed);
+}
+
+/************************************************************************
+Gets the youngest modification log sequence number for a frame.
+Returns zero if not file page or no modification occurred yet. */
+UNIV_INLINE
+ib_uint64_t
+buf_page_get_newest_modification(
+/*=============================*/
+					/* out: newest modification to page */
+	const buf_page_t*	bpage)	/* in: block containing the
+					page frame */
+{
+	ib_uint64_t	lsn;
+	mutex_t*	block_mutex = buf_page_get_mutex(bpage);
+
+	mutex_enter(block_mutex);
+
+	if (buf_page_in_file(bpage)) {
+		lsn = bpage->newest_modification;
+	} else {
+		lsn = 0;
+	}
+
+	mutex_exit(block_mutex);
+
+	return(lsn);
+}
+
+/************************************************************************
+Increments the modify clock of a frame by 1. The caller must (1) own the
+buf_pool mutex and block bufferfix count has to be zero, (2) or own an x-lock
+on the block. */
+UNIV_INLINE
+void
+buf_block_modify_clock_inc(
+/*=======================*/
+	buf_block_t*	block)	/* in: block */
+{
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad((buf_pool_mutex_own()
+	       && (block->page.buf_fix_count == 0))
+	      || rw_lock_own(&(block->lock), RW_LOCK_EXCLUSIVE));
+#endif /* UNIV_SYNC_DEBUG */
+
+	block->modify_clock++;
+}
+
+/************************************************************************
+Returns the value of the modify clock. The caller must have an s-lock
+or x-lock on the block. */
+UNIV_INLINE
+ib_uint64_t
+buf_block_get_modify_clock(
+/*=======================*/
+				/* out: value */
+	buf_block_t*	block)	/* in: block */
+{
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(rw_lock_own(&(block->lock), RW_LOCK_SHARED)
+	      || rw_lock_own(&(block->lock), RW_LOCK_EXCLUSIVE));
+#endif /* UNIV_SYNC_DEBUG */
+
+	return(block->modify_clock);
+}
+
+/***********************************************************************
+Increments the bufferfix count. */
+UNIV_INLINE
+void
+buf_block_buf_fix_inc_func(
+/*=======================*/
+#ifdef UNIV_SYNC_DEBUG
+	const char*	file,	/* in: file name */
+	ulint		line,	/* in: line */
+#endif /* UNIV_SYNC_DEBUG */
+	buf_block_t*	block)	/* in: block to bufferfix */
+{
+#ifdef UNIV_SYNC_DEBUG
+	ibool	ret;
+
+	ret = rw_lock_s_lock_nowait(&(block->debug_latch), file, line);
+	ut_a(ret);
+#endif /* UNIV_SYNC_DEBUG */
+	ut_ad(mutex_own(&block->mutex));
+
+	block->page.buf_fix_count++;
+}
+#ifdef UNIV_SYNC_DEBUG
+# define buf_block_buf_fix_inc(b,f,l) buf_block_buf_fix_inc_func(f,l,b)
+#else /* UNIV_SYNC_DEBUG */
+# define buf_block_buf_fix_inc(b,f,l) buf_block_buf_fix_inc_func(b)
+#endif /* UNIV_SYNC_DEBUG */
+
+/***********************************************************************
+Decrements the bufferfix count. */
+UNIV_INLINE
+void
+buf_block_buf_fix_dec(
+/*==================*/
+	buf_block_t*	block)	/* in: block to bufferunfix */
+{
+	ut_ad(mutex_own(&block->mutex));
+
+	block->page.buf_fix_count--;
+#ifdef UNIV_SYNC_DEBUG
+	rw_lock_s_unlock(&block->debug_latch);
+#endif
+}
+
+/**********************************************************************
+Returns the control block of a file page, NULL if not found. */
+UNIV_INLINE
+buf_page_t*
+buf_page_hash_get(
+/*==============*/
+			/* out: block, NULL if not found */
+	ulint	space,	/* in: space id */
+	ulint	offset)	/* in: offset of the page within space */
+{
+	buf_page_t*	bpage;
+	ulint		fold;
+
+	ut_ad(buf_pool);
+	ut_ad(buf_pool_mutex_own());
+
+	/* Look for the page in the hash table */
+
+	fold = buf_page_address_fold(space, offset);
+
+	HASH_SEARCH(hash, buf_pool->page_hash, fold, buf_page_t*, bpage,
+		    ut_ad(bpage->in_page_hash && !bpage->in_zip_hash
+			  && buf_page_in_file(bpage)),
+		    bpage->space == space && bpage->offset == offset);
+	if (bpage) {
+		ut_a(buf_page_in_file(bpage));
+		ut_ad(bpage->in_page_hash);
+		ut_ad(!bpage->in_zip_hash);
+		UNIV_MEM_ASSERT_RW(bpage, sizeof *bpage);
+	}
+
+	return(bpage);
+}
+
+/**********************************************************************
+Returns the control block of a file page, NULL if not found
+or an uncompressed page frame does not exist. */
+UNIV_INLINE
+buf_block_t*
+buf_block_hash_get(
+/*===============*/
+			/* out: block, NULL if not found */
+	ulint	space,	/* in: space id */
+	ulint	offset)	/* in: offset of the page within space */
+{
+	return(buf_page_get_block(buf_page_hash_get(space, offset)));
+}
+
+/************************************************************************
+Returns TRUE if the page can be found in the buffer pool hash table. NOTE
+that it is possible that the page is not yet read from disk, though. */
+UNIV_INLINE
+ibool
+buf_page_peek(
+/*==========*/
+			/* out: TRUE if found from page hash table,
+			NOTE that the page is not necessarily yet read
+			from disk! */
+	ulint	space,	/* in: space id */
+	ulint	offset)	/* in: page number */
+{
+	const buf_page_t*	bpage;
+
+	buf_pool_mutex_enter();
+
+	bpage = buf_page_hash_get(space, offset);
+
+	buf_pool_mutex_exit();
+
+	return(bpage != NULL);
+}
+
+/************************************************************************
+Releases a compressed-only page acquired with buf_page_get_zip(). */
+UNIV_INLINE
+void
+buf_page_release_zip(
+/*=================*/
+	buf_page_t*	bpage)		/* in: buffer block */
+{
+	buf_block_t*	block;
+
+	ut_ad(bpage);
+	ut_a(bpage->buf_fix_count > 0);
+
+	switch (buf_page_get_state(bpage)) {
+	case BUF_BLOCK_ZIP_PAGE:
+	case BUF_BLOCK_ZIP_DIRTY:
+		mutex_enter(&buf_pool_zip_mutex);
+		bpage->buf_fix_count--;
+		mutex_exit(&buf_pool_zip_mutex);
+		return;
+	case BUF_BLOCK_FILE_PAGE:
+		block = (buf_block_t*) bpage;
+		mutex_enter(&block->mutex);
+#ifdef UNIV_SYNC_DEBUG
+		rw_lock_s_unlock(&block->debug_latch);
+#endif
+		bpage->buf_fix_count--;
+		mutex_exit(&block->mutex);
+		return;
+	case BUF_BLOCK_ZIP_FREE:
+	case BUF_BLOCK_NOT_USED:
+	case BUF_BLOCK_READY_FOR_USE:
+	case BUF_BLOCK_MEMORY:
+	case BUF_BLOCK_REMOVE_HASH:
+		break;
+	}
+
+	ut_error;
+}
+
+/************************************************************************
+Decrements the bufferfix count of a buffer control block and releases
+a latch, if specified. */
+UNIV_INLINE
+void
+buf_page_release(
+/*=============*/
+	buf_block_t*	block,		/* in: buffer block */
+	ulint		rw_latch,	/* in: RW_S_LATCH, RW_X_LATCH,
+					RW_NO_LATCH */
+	mtr_t*		mtr)		/* in: mtr */
+{
+	ut_ad(block);
+
+	ut_a(buf_block_get_state(block) == BUF_BLOCK_FILE_PAGE);
+	ut_a(block->page.buf_fix_count > 0);
+
+	if (rw_latch == RW_X_LATCH && mtr->modifications) {
+		buf_pool_mutex_enter();
+		buf_flush_note_modification(block, mtr);
+		buf_pool_mutex_exit();
+	}
+
+	mutex_enter(&block->mutex);
+
+#ifdef UNIV_SYNC_DEBUG
+	rw_lock_s_unlock(&(block->debug_latch));
+#endif
+	block->page.buf_fix_count--;
+
+	/* Dirty blocks should be in the flush list. */
+	ut_ad(!block->page.oldest_modification
+	      || block->page.in_flush_list);
+
+	mutex_exit(&block->mutex);
+
+	if (rw_latch == RW_S_LATCH) {
+		rw_lock_s_unlock(&(block->lock));
+	} else if (rw_latch == RW_X_LATCH) {
+		rw_lock_x_unlock(&(block->lock));
+	}
+}
+
+#ifdef UNIV_SYNC_DEBUG
+/*************************************************************************
+Adds latch level info for the rw-lock protecting the buffer frame. This
+should be called in the debug version after a successful latching of a
+page if we know the latching order level of the acquired latch. */
+UNIV_INLINE
+void
+buf_block_dbg_add_level(
+/*====================*/
+	buf_block_t*	block,	/* in: buffer page
+				where we have acquired latch */
+	ulint		level)	/* in: latching order level */
+{
+	sync_thread_add_level(&block->lock, level);
+}
+#endif /* UNIV_SYNC_DEBUG */
diff --git a/storage/xtradb/include/buf0flu.h b/storage/xtradb/include/buf0flu.h
new file mode 100644
index 00000000000..11a37351479
--- /dev/null
+++ b/storage/xtradb/include/buf0flu.h
@@ -0,0 +1,151 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The database buffer pool flush algorithm
+
+Created 11/5/1995 Heikki Tuuri
+*******************************************************/
+
+#ifndef buf0flu_h
+#define buf0flu_h
+
+#include "univ.i"
+#include "buf0types.h"
+#include "ut0byte.h"
+#include "mtr0types.h"
+
+/************************************************************************
+Remove a block from the flush list of modified blocks. */
+UNIV_INTERN
+void
+buf_flush_remove(
+/*=============*/
+	buf_page_t*	bpage);	/* in: pointer to the block in question */
+/************************************************************************
+Updates the flush system data structures when a write is completed. */
+UNIV_INTERN
+void
+buf_flush_write_complete(
+/*=====================*/
+	buf_page_t*	bpage);	/* in: pointer to the block in question */
+/*************************************************************************
+Flushes pages from the end of the LRU list if there is too small
+a margin of replaceable pages there. */
+UNIV_INTERN
+void
+buf_flush_free_margin(
+/*=======================*/
+	ibool	wait);
+/************************************************************************
+Initializes a page for writing to the tablespace. */
+UNIV_INTERN
+void
+buf_flush_init_for_writing(
+/*=======================*/
+	byte*		page,		/* in/out: page */
+	void*		page_zip_,	/* in/out: compressed page, or NULL */
+	ib_uint64_t	newest_lsn);	/* in: newest modification lsn
+					to the page */
+/***********************************************************************
+This utility flushes dirty blocks from the end of the LRU list or flush_list.
+NOTE 1: in the case of an LRU flush the calling thread may own latches to
+pages: to avoid deadlocks, this function must be written so that it cannot
+end up waiting for these latches! NOTE 2: in the case of a flush list flush,
+the calling thread is not allowed to own any latches on pages! */
+UNIV_INTERN
+ulint
+buf_flush_batch(
+/*============*/
+					/* out: number of blocks for which the
+					write request was queued;
+					ULINT_UNDEFINED if there was a flush
+					of the same type already running */
+	enum buf_flush	flush_type,	/* in: BUF_FLUSH_LRU or
+					BUF_FLUSH_LIST; if BUF_FLUSH_LIST,
+					then the caller must not own any
+					latches on pages */
+	ulint		min_n,		/* in: wished minimum mumber of blocks
+					flushed (it is not guaranteed that the
+					actual number is that big, though) */
+	ib_uint64_t	lsn_limit);	/* in the case BUF_FLUSH_LIST all
+					blocks whose oldest_modification is
+					smaller than this should be flushed
+					(if their number does not exceed
+					min_n), otherwise ignored */
+/**********************************************************************
+Waits until a flush batch of the given type ends */
+UNIV_INTERN
+void
+buf_flush_wait_batch_end(
+/*=====================*/
+	enum buf_flush	type);	/* in: BUF_FLUSH_LRU or BUF_FLUSH_LIST */
+/************************************************************************
+This function should be called at a mini-transaction commit, if a page was
+modified in it. Puts the block to the list of modified blocks, if it not
+already in it. */
+UNIV_INLINE
+void
+buf_flush_note_modification(
+/*========================*/
+	buf_block_t*	block,	/* in: block which is modified */
+	mtr_t*		mtr);	/* in: mtr */
+/************************************************************************
+This function should be called when recovery has modified a buffer page. */
+UNIV_INLINE
+void
+buf_flush_recv_note_modification(
+/*=============================*/
+	buf_block_t*	block,		/* in: block which is modified */
+	ib_uint64_t	start_lsn,	/* in: start lsn of the first mtr in a
+					set of mtr's */
+	ib_uint64_t	end_lsn);	/* in: end lsn of the last mtr in the
+					set of mtr's */
+/************************************************************************
+Returns TRUE if the file page block is immediately suitable for replacement,
+i.e., transition FILE_PAGE => NOT_USED allowed. */
+UNIV_INTERN
+ibool
+buf_flush_ready_for_replace(
+/*========================*/
+				/* out: TRUE if can replace immediately */
+	buf_page_t*	bpage);	/* in: buffer control block, must be
+				buf_page_in_file(bpage) and in the LRU list */
+#if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
+/**********************************************************************
+Validates the flush list. */
+UNIV_INTERN
+ibool
+buf_flush_validate(void);
+/*====================*/
+		/* out: TRUE if ok */
+#endif /* UNIV_DEBUG || UNIV_BUF_DEBUG */
+
+/* When buf_flush_free_margin is called, it tries to make this many blocks
+available to replacement in the free list and at the end of the LRU list (to
+make sure that a read-ahead batch can be read efficiently in a single
+sweep). */
+
+#define BUF_FLUSH_FREE_BLOCK_MARGIN	(5 + BUF_READ_AHEAD_AREA)
+#define BUF_FLUSH_EXTRA_MARGIN		(BUF_FLUSH_FREE_BLOCK_MARGIN / 4 + 100)
+
+#ifndef UNIV_NONINL
+#include "buf0flu.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/buf0flu.ic b/storage/xtradb/include/buf0flu.ic
new file mode 100644
index 00000000000..2dfa7e68d41
--- /dev/null
+++ b/storage/xtradb/include/buf0flu.ic
@@ -0,0 +1,120 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The database buffer pool flush algorithm
+
+Created 11/5/1995 Heikki Tuuri
+*******************************************************/
+
+#include "buf0buf.h"
+#include "mtr0mtr.h"
+
+/************************************************************************
+Inserts a modified block into the flush list. */
+UNIV_INTERN
+void
+buf_flush_insert_into_flush_list(
+/*=============================*/
+	buf_block_t*	block);	/* in/out: block which is modified */
+/************************************************************************
+Inserts a modified block into the flush list in the right sorted position.
+This function is used by recovery, because there the modifications do not
+necessarily come in the order of lsn's. */
+UNIV_INTERN
+void
+buf_flush_insert_sorted_into_flush_list(
+/*====================================*/
+	buf_block_t*	block);	/* in/out: block which is modified */
+
+/************************************************************************
+This function should be called at a mini-transaction commit, if a page was
+modified in it. Puts the block to the list of modified blocks, if it is not
+already in it. */
+UNIV_INLINE
+void
+buf_flush_note_modification(
+/*========================*/
+	buf_block_t*	block,	/* in: block which is modified */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	ut_ad(block);
+	ut_ad(buf_block_get_state(block) == BUF_BLOCK_FILE_PAGE);
+	ut_ad(block->page.buf_fix_count > 0);
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(rw_lock_own(&(block->lock), RW_LOCK_EX));
+#endif /* UNIV_SYNC_DEBUG */
+	ut_ad(buf_pool_mutex_own());
+
+	ut_ad(mtr->start_lsn != 0);
+	ut_ad(mtr->modifications);
+	ut_ad(block->page.newest_modification <= mtr->end_lsn);
+
+	block->page.newest_modification = mtr->end_lsn;
+
+	if (!block->page.oldest_modification) {
+
+		block->page.oldest_modification = mtr->start_lsn;
+		ut_ad(block->page.oldest_modification != 0);
+
+		buf_flush_insert_into_flush_list(block);
+	} else {
+		ut_ad(block->page.oldest_modification <= mtr->start_lsn);
+	}
+
+	++srv_buf_pool_write_requests;
+}
+
+/************************************************************************
+This function should be called when recovery has modified a buffer page. */
+UNIV_INLINE
+void
+buf_flush_recv_note_modification(
+/*=============================*/
+	buf_block_t*	block,		/* in: block which is modified */
+	ib_uint64_t	start_lsn,	/* in: start lsn of the first mtr in a
+					set of mtr's */
+	ib_uint64_t	end_lsn)	/* in: end lsn of the last mtr in the
+					set of mtr's */
+{
+	ut_ad(block);
+	ut_ad(buf_block_get_state(block) == BUF_BLOCK_FILE_PAGE);
+	ut_ad(block->page.buf_fix_count > 0);
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(rw_lock_own(&(block->lock), RW_LOCK_EX));
+#endif /* UNIV_SYNC_DEBUG */
+
+	buf_pool_mutex_enter();
+
+	ut_ad(block->page.newest_modification <= end_lsn);
+
+	block->page.newest_modification = end_lsn;
+
+	if (!block->page.oldest_modification) {
+
+		block->page.oldest_modification = start_lsn;
+
+		ut_ad(block->page.oldest_modification != 0);
+
+		buf_flush_insert_sorted_into_flush_list(block);
+	} else {
+		ut_ad(block->page.oldest_modification <= start_lsn);
+	}
+
+	buf_pool_mutex_exit();
+}
diff --git a/storage/xtradb/include/buf0lru.h b/storage/xtradb/include/buf0lru.h
new file mode 100644
index 00000000000..e73869580bd
--- /dev/null
+++ b/storage/xtradb/include/buf0lru.h
@@ -0,0 +1,265 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The database buffer pool LRU replacement algorithm
+
+Created 11/5/1995 Heikki Tuuri
+*******************************************************/
+
+#ifndef buf0lru_h
+#define buf0lru_h
+
+#include "univ.i"
+#include "ut0byte.h"
+#include "buf0types.h"
+
+/** The return type of buf_LRU_free_block() */
+enum buf_lru_free_block_status {
+	/** freed */
+	BUF_LRU_FREED = 0,
+	/** not freed because the caller asked to remove the
+	uncompressed frame but the control block cannot be
+	relocated */
+	BUF_LRU_CANNOT_RELOCATE,
+	/** not freed because of some other reason */
+	BUF_LRU_NOT_FREED
+};
+
+/**********************************************************************
+Tries to remove LRU flushed blocks from the end of the LRU list and put them
+to the free list. This is beneficial for the efficiency of the insert buffer
+operation, as flushed pages from non-unique non-clustered indexes are here
+taken out of the buffer pool, and their inserts redirected to the insert
+buffer. Otherwise, the flushed blocks could get modified again before read
+operations need new buffer blocks, and the i/o work done in flushing would be
+wasted. */
+UNIV_INTERN
+void
+buf_LRU_try_free_flushed_blocks(void);
+/*==================================*/
+/**********************************************************************
+Returns TRUE if less than 25 % of the buffer pool is available. This can be
+used in heuristics to prevent huge transactions eating up the whole buffer
+pool for their locks. */
+UNIV_INTERN
+ibool
+buf_LRU_buf_pool_running_out(void);
+/*==============================*/
+				/* out: TRUE if less than 25 % of buffer pool
+				left */
+
+/*#######################################################################
+These are low-level functions
+#########################################################################*/
+
+/* Minimum LRU list length for which the LRU_old pointer is defined */
+
+#define BUF_LRU_OLD_MIN_LEN	80
+
+#define BUF_LRU_FREE_SEARCH_LEN		(5 + 2 * BUF_READ_AHEAD_AREA)
+
+/**********************************************************************
+Invalidates all pages belonging to a given tablespace when we are deleting
+the data file(s) of that tablespace. A PROBLEM: if readahead is being started,
+what guarantees that it will not try to read in pages after this operation has
+completed? */
+UNIV_INTERN
+void
+buf_LRU_invalidate_tablespace(
+/*==========================*/
+	ulint	id);	/* in: space id */
+/**********************************************************************
+Gets the minimum LRU_position field for the blocks in an initial segment
+(determined by BUF_LRU_INITIAL_RATIO) of the LRU list. The limit is not
+guaranteed to be precise, because the ulint_clock may wrap around. */
+UNIV_INTERN
+ulint
+buf_LRU_get_recent_limit(void);
+/*==========================*/
+			/* out: the limit; zero if could not determine it */
+/************************************************************************
+Insert a compressed block into buf_pool->zip_clean in the LRU order. */
+UNIV_INTERN
+void
+buf_LRU_insert_zip_clean(
+/*=====================*/
+	buf_page_t*	bpage);	/* in: pointer to the block in question */
+
+/**********************************************************************
+Try to free a block.  If bpage is a descriptor of a compressed-only
+page, the descriptor object will be freed as well.
+
+NOTE: If this function returns BUF_LRU_FREED, it will not temporarily
+release buf_pool_mutex.  Furthermore, the page frame will no longer be
+accessible via bpage.
+
+The caller must hold buf_pool_mutex and buf_page_get_mutex(bpage) and
+release these two mutexes after the call.  No other
+buf_page_get_mutex() may be held when calling this function. */
+UNIV_INTERN
+enum buf_lru_free_block_status
+buf_LRU_free_block(
+/*===============*/
+				/* out: BUF_LRU_FREED if freed,
+				BUF_LRU_CANNOT_RELOCATE or
+				BUF_LRU_NOT_FREED otherwise. */
+	buf_page_t*	bpage,	/* in: block to be freed */
+	ibool		zip,	/* in: TRUE if should remove also the
+				compressed page of an uncompressed page */
+	ibool*		buf_pool_mutex_released);
+				/* in: pointer to a variable that will
+				be assigned TRUE if buf_pool_mutex
+				was temporarily released, or NULL */
+/**********************************************************************
+Try to free a replaceable block. */
+UNIV_INTERN
+ibool
+buf_LRU_search_and_free_block(
+/*==========================*/
+				/* out: TRUE if found and freed */
+	ulint	n_iterations);	/* in: how many times this has been called
+				repeatedly without result: a high value means
+				that we should search farther; if
+				n_iterations < 10, then we search
+				n_iterations / 10 * buf_pool->curr_size
+				pages from the end of the LRU list; if
+				n_iterations < 5, then we will also search
+				n_iterations / 5 of the unzip_LRU list. */
+/**********************************************************************
+Returns a free block from the buf_pool.  The block is taken off the
+free list.  If it is empty, returns NULL. */
+UNIV_INTERN
+buf_block_t*
+buf_LRU_get_free_only(void);
+/*=======================*/
+				/* out: a free control block, or NULL
+				if the buf_block->free list is empty */
+/**********************************************************************
+Returns a free block from the buf_pool. The block is taken off the
+free list. If it is empty, blocks are moved from the end of the
+LRU list to the free list. */
+UNIV_INTERN
+buf_block_t*
+buf_LRU_get_free_block(
+/*===================*/
+				/* out: the free control block,
+				in state BUF_BLOCK_READY_FOR_USE */
+	ulint	zip_size);	/* in: compressed page size in bytes,
+				or 0 if uncompressed tablespace */
+
+/**********************************************************************
+Puts a block back to the free list. */
+UNIV_INTERN
+void
+buf_LRU_block_free_non_file_page(
+/*=============================*/
+	buf_block_t*	block);	/* in: block, must not contain a file page */
+/**********************************************************************
+Adds a block to the LRU list. */
+UNIV_INTERN
+void
+buf_LRU_add_block(
+/*==============*/
+	buf_page_t*	bpage,	/* in: control block */
+	ibool		old);	/* in: TRUE if should be put to the old
+				blocks in the LRU list, else put to the
+				start; if the LRU list is very short, added to
+				the start regardless of this parameter */
+/**********************************************************************
+Adds a block to the LRU list of decompressed zip pages. */
+UNIV_INTERN
+void
+buf_unzip_LRU_add_block(
+/*====================*/
+	buf_block_t*	block,	/* in: control block */
+	ibool		old);	/* in: TRUE if should be put to the end
+				of the list, else put to the start */
+/**********************************************************************
+Moves a block to the start of the LRU list. */
+UNIV_INTERN
+void
+buf_LRU_make_block_young(
+/*=====================*/
+	buf_page_t*	bpage);	/* in: control block */
+/**********************************************************************
+Moves a block to the end of the LRU list. */
+UNIV_INTERN
+void
+buf_LRU_make_block_old(
+/*===================*/
+	buf_page_t*	bpage);	/* in: control block */
+/************************************************************************
+Update the historical stats that we are collecting for LRU eviction
+policy at the end of each interval. */
+UNIV_INTERN
+void
+buf_LRU_stat_update(void);
+/*=====================*/
+
+#if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
+/**************************************************************************
+Validates the LRU list. */
+UNIV_INTERN
+ibool
+buf_LRU_validate(void);
+/*==================*/
+#endif /* UNIV_DEBUG || UNIV_BUF_DEBUG */
+#if defined UNIV_DEBUG_PRINT || defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
+/**************************************************************************
+Prints the LRU list. */
+UNIV_INTERN
+void
+buf_LRU_print(void);
+/*===============*/
+#endif /* UNIV_DEBUG_PRINT || UNIV_DEBUG || UNIV_BUF_DEBUG */
+
+/**********************************************************************
+These statistics are not 'of' LRU but 'for' LRU.  We keep count of I/O
+and page_zip_decompress() operations.  Based on the statistics we decide
+if we want to evict from buf_pool->unzip_LRU or buf_pool->LRU. */
+
+/** Statistics for selecting the LRU list for eviction. */
+struct buf_LRU_stat_struct
+{
+	ulint	io;	/**< Counter of buffer pool I/O operations. */
+	ulint	unzip;	/**< Counter of page_zip_decompress operations. */
+};
+
+typedef struct buf_LRU_stat_struct buf_LRU_stat_t;
+
+/** Current operation counters.  Not protected by any mutex.
+Cleared by buf_LRU_stat_update(). */
+extern buf_LRU_stat_t	buf_LRU_stat_cur;
+
+/** Running sum of past values of buf_LRU_stat_cur.
+Updated by buf_LRU_stat_update().  Protected by buf_pool_mutex. */
+extern buf_LRU_stat_t	buf_LRU_stat_sum;
+
+/************************************************************************
+Increments the I/O counter in buf_LRU_stat_cur. */
+#define buf_LRU_stat_inc_io() buf_LRU_stat_cur.io++
+/************************************************************************
+Increments the page_zip_decompress() counter in buf_LRU_stat_cur. */
+#define buf_LRU_stat_inc_unzip() buf_LRU_stat_cur.unzip++
+
+#ifndef UNIV_NONINL
+#include "buf0lru.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/buf0lru.ic b/storage/xtradb/include/buf0lru.ic
new file mode 100644
index 00000000000..f4c40e0b606
--- /dev/null
+++ b/storage/xtradb/include/buf0lru.ic
@@ -0,0 +1,24 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The database buffer replacement algorithm
+
+Created 11/5/1995 Heikki Tuuri
+*******************************************************/
+
diff --git a/storage/xtradb/include/buf0rea.h b/storage/xtradb/include/buf0rea.h
new file mode 100644
index 00000000000..6d138a3a02b
--- /dev/null
+++ b/storage/xtradb/include/buf0rea.h
@@ -0,0 +1,136 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The database buffer read
+
+Created 11/5/1995 Heikki Tuuri
+*******************************************************/
+
+#ifndef buf0rea_h
+#define buf0rea_h
+
+#include "univ.i"
+#include "buf0types.h"
+
+/************************************************************************
+High-level function which reads a page asynchronously from a file to the
+buffer buf_pool if it is not already there. Sets the io_fix flag and sets
+an exclusive lock on the buffer frame. The flag is cleared and the x-lock
+released by the i/o-handler thread. Does a random read-ahead if it seems
+sensible. */
+UNIV_INTERN
+ulint
+buf_read_page(
+/*==========*/
+			/* out: number of page read requests issued: this can
+			be > 1 if read-ahead occurred */
+	ulint	space,	/* in: space id */
+	ulint	zip_size,/* in: compressed page size in bytes, or 0 */
+	ulint	offset);/* in: page number */
+/************************************************************************
+Applies linear read-ahead if in the buf_pool the page is a border page of
+a linear read-ahead area and all the pages in the area have been accessed.
+Does not read any page if the read-ahead mechanism is not activated. Note
+that the the algorithm looks at the 'natural' adjacent successor and
+predecessor of the page, which on the leaf level of a B-tree are the next
+and previous page in the chain of leaves. To know these, the page specified
+in (space, offset) must already be present in the buf_pool. Thus, the
+natural way to use this function is to call it when a page in the buf_pool
+is accessed the first time, calling this function just after it has been
+bufferfixed.
+NOTE 1: as this function looks at the natural predecessor and successor
+fields on the page, what happens, if these are not initialized to any
+sensible value? No problem, before applying read-ahead we check that the
+area to read is within the span of the space, if not, read-ahead is not
+applied. An uninitialized value may result in a useless read operation, but
+only very improbably.
+NOTE 2: the calling thread may own latches on pages: to avoid deadlocks this
+function must be written such that it cannot end up waiting for these
+latches!
+NOTE 3: the calling thread must want access to the page given: this rule is
+set to prevent unintended read-aheads performed by ibuf routines, a situation
+which could result in a deadlock if the OS does not support asynchronous io. */
+UNIV_INTERN
+ulint
+buf_read_ahead_linear(
+/*==================*/
+			/* out: number of page read requests issued */
+	ulint	space,	/* in: space id */
+	ulint	zip_size,/* in: compressed page size in bytes, or 0 */
+	ulint	offset);/* in: page number of a page; NOTE: the current thread
+			must want access to this page (see NOTE 3 above) */
+/************************************************************************
+Issues read requests for pages which the ibuf module wants to read in, in
+order to contract the insert buffer tree. Technically, this function is like
+a read-ahead function. */
+UNIV_INTERN
+void
+buf_read_ibuf_merge_pages(
+/*======================*/
+	ibool		sync,		/* in: TRUE if the caller
+					wants this function to wait
+					for the highest address page
+					to get read in, before this
+					function returns */
+	const ulint*	space_ids,	/* in: array of space ids */
+	const ib_int64_t* space_versions,/* in: the spaces must have
+					this version number
+					(timestamp), otherwise we
+					discard the read; we use this
+					to cancel reads if DISCARD +
+					IMPORT may have changed the
+					tablespace size */
+	const ulint*	page_nos,	/* in: array of page numbers
+					to read, with the highest page
+					number the last in the
+					array */
+	ulint		n_stored);	/* in: number of elements
+					in the arrays */
+/************************************************************************
+Issues read requests for pages which recovery wants to read in. */
+UNIV_INTERN
+void
+buf_read_recv_pages(
+/*================*/
+	ibool		sync,		/* in: TRUE if the caller
+					wants this function to wait
+					for the highest address page
+					to get read in, before this
+					function returns */
+	ulint		space,		/* in: space id */
+	ulint		zip_size,	/* in: compressed page size in
+					bytes, or 0 */
+	const ulint*	page_nos,	/* in: array of page numbers
+					to read, with the highest page
+					number the last in the
+					array */
+	ulint		n_stored);	/* in: number of page numbers
+					in the array */
+
+/* The size in pages of the area which the read-ahead algorithms read if
+invoked */
+
+#define	BUF_READ_AHEAD_AREA					\
+	ut_min(64, ut_2_power_up(buf_pool->curr_size / 32))
+
+/* Modes used in read-ahead */
+#define BUF_READ_IBUF_PAGES_ONLY	131
+#define BUF_READ_ANY_PAGE		132
+
+#endif
diff --git a/storage/xtradb/include/buf0types.h b/storage/xtradb/include/buf0types.h
new file mode 100644
index 00000000000..f2721da85f9
--- /dev/null
+++ b/storage/xtradb/include/buf0types.h
@@ -0,0 +1,70 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The database buffer pool global types for the directory
+
+Created 11/17/1995 Heikki Tuuri
+*******************************************************/
+
+#ifndef buf0types_h
+#define buf0types_h
+
+typedef	struct buf_page_struct		buf_page_t;
+typedef	struct buf_block_struct		buf_block_t;
+typedef struct buf_chunk_struct		buf_chunk_t;
+typedef	struct buf_pool_struct		buf_pool_t;
+
+/* The 'type' used of a buffer frame */
+typedef	byte	buf_frame_t;
+
+/* Flags for flush types */
+enum buf_flush {
+	BUF_FLUSH_LRU = 0,
+	BUF_FLUSH_SINGLE_PAGE,
+	BUF_FLUSH_LIST,
+	BUF_FLUSH_N_TYPES		/* index of last element + 1  */
+};
+
+/* Flags for io_fix types */
+enum buf_io_fix {
+	BUF_IO_NONE = 0,		/**< no pending I/O */
+	BUF_IO_READ,			/**< read pending */
+	BUF_IO_WRITE			/**< write pending */
+};
+
+/* Parameters of binary buddy system for compressed pages (buf0buddy.h) */
+#if UNIV_WORD_SIZE <= 4 /* 32-bit system */
+# define BUF_BUDDY_LOW_SHIFT	6
+#else /* 64-bit system */
+# define BUF_BUDDY_LOW_SHIFT	7
+#endif
+#define BUF_BUDDY_LOW		(1 << BUF_BUDDY_LOW_SHIFT)
+					/* minimum block size in the binary
+					buddy system; must be at least
+					sizeof(buf_page_t) */
+#define BUF_BUDDY_SIZES		(UNIV_PAGE_SIZE_SHIFT - BUF_BUDDY_LOW_SHIFT)
+					/* number of buddy sizes */
+
+/* twice the maximum block size of the buddy system;
+the underlying memory is aligned by this amount:
+this must be equal to UNIV_PAGE_SIZE */
+#define BUF_BUDDY_HIGH	(BUF_BUDDY_LOW << BUF_BUDDY_SIZES)
+
+#endif
+
diff --git a/storage/xtradb/include/data0data.h b/storage/xtradb/include/data0data.h
new file mode 100644
index 00000000000..1190a7ae45a
--- /dev/null
+++ b/storage/xtradb/include/data0data.h
@@ -0,0 +1,480 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/************************************************************************
+SQL data field and tuple
+
+Created 5/30/1994 Heikki Tuuri
+*************************************************************************/
+
+#ifndef data0data_h
+#define data0data_h
+
+#include "univ.i"
+
+#include "data0types.h"
+#include "data0type.h"
+#include "mem0mem.h"
+#include "dict0types.h"
+
+typedef struct big_rec_struct		big_rec_t;
+
+#ifdef UNIV_DEBUG
+/*************************************************************************
+Gets pointer to the type struct of SQL data field. */
+UNIV_INLINE
+dtype_t*
+dfield_get_type(
+/*============*/
+				/* out: pointer to the type struct */
+	const dfield_t*	field);	/* in: SQL data field */
+/*************************************************************************
+Gets pointer to the data in a field. */
+UNIV_INLINE
+void*
+dfield_get_data(
+/*============*/
+				/* out: pointer to data */
+	const dfield_t* field);	/* in: field */
+#else /* UNIV_DEBUG */
+# define dfield_get_type(field) (&(field)->type)
+# define dfield_get_data(field) ((field)->data)
+#endif /* UNIV_DEBUG */
+/*************************************************************************
+Sets the type struct of SQL data field. */
+UNIV_INLINE
+void
+dfield_set_type(
+/*============*/
+	dfield_t*	field,	/* in: SQL data field */
+	dtype_t*	type);	/* in: pointer to data type struct */
+/*************************************************************************
+Gets length of field data. */
+UNIV_INLINE
+ulint
+dfield_get_len(
+/*===========*/
+				/* out: length of data; UNIV_SQL_NULL if
+				SQL null data */
+	const dfield_t* field);	/* in: field */
+/*************************************************************************
+Sets length in a field. */
+UNIV_INLINE
+void
+dfield_set_len(
+/*===========*/
+	dfield_t*	field,	/* in: field */
+	ulint		len);	/* in: length or UNIV_SQL_NULL */
+/*************************************************************************
+Determines if a field is SQL NULL */
+UNIV_INLINE
+ulint
+dfield_is_null(
+/*===========*/
+				/* out: nonzero if SQL null data */
+	const dfield_t* field);	/* in: field */
+/*************************************************************************
+Determines if a field is externally stored */
+UNIV_INLINE
+ulint
+dfield_is_ext(
+/*==========*/
+				/* out: nonzero if externally stored */
+	const dfield_t* field);	/* in: field */
+/*************************************************************************
+Sets the "external storage" flag */
+UNIV_INLINE
+void
+dfield_set_ext(
+/*===========*/
+	dfield_t*	field);	/* in/out: field */
+/*************************************************************************
+Sets pointer to the data and length in a field. */
+UNIV_INLINE
+void
+dfield_set_data(
+/*============*/
+	dfield_t*	field,	/* in: field */
+	const void*	data,	/* in: data */
+	ulint		len);	/* in: length or UNIV_SQL_NULL */
+/*************************************************************************
+Sets a data field to SQL NULL. */
+UNIV_INLINE
+void
+dfield_set_null(
+/*============*/
+	dfield_t*	field);	/* in/out: field */
+/**************************************************************************
+Writes an SQL null field full of zeros. */
+UNIV_INLINE
+void
+data_write_sql_null(
+/*================*/
+	byte*	data,	/* in: pointer to a buffer of size len */
+	ulint	len);	/* in: SQL null size in bytes */
+/*************************************************************************
+Copies the data and len fields. */
+UNIV_INLINE
+void
+dfield_copy_data(
+/*=============*/
+	dfield_t*	field1,	/* out: field to copy to */
+	const dfield_t*	field2);/* in: field to copy from */
+/*************************************************************************
+Copies a data field to another. */
+UNIV_INLINE
+void
+dfield_copy(
+/*========*/
+	dfield_t*	field1,	/* out: field to copy to */
+	const dfield_t*	field2);/* in: field to copy from */
+/*************************************************************************
+Copies the data pointed to by a data field. */
+UNIV_INLINE
+void
+dfield_dup(
+/*=======*/
+	dfield_t*	field,	/* in/out: data field */
+	mem_heap_t*	heap);	/* in: memory heap where allocated */
+/*************************************************************************
+Tests if data length and content is equal for two dfields. */
+UNIV_INLINE
+ibool
+dfield_datas_are_binary_equal(
+/*==========================*/
+				/* out: TRUE if equal */
+	const dfield_t*	field1,	/* in: field */
+	const dfield_t*	field2);/* in: field */
+/*************************************************************************
+Tests if dfield data length and content is equal to the given. */
+UNIV_INTERN
+ibool
+dfield_data_is_binary_equal(
+/*========================*/
+				/* out: TRUE if equal */
+	const dfield_t*	field,	/* in: field */
+	ulint		len,	/* in: data length or UNIV_SQL_NULL */
+	const byte*	data);	/* in: data */
+/*************************************************************************
+Gets number of fields in a data tuple. */
+UNIV_INLINE
+ulint
+dtuple_get_n_fields(
+/*================*/
+				/* out: number of fields */
+	const dtuple_t*	tuple);	/* in: tuple */
+#ifdef UNIV_DEBUG
+/*************************************************************************
+Gets nth field of a tuple. */
+UNIV_INLINE
+dfield_t*
+dtuple_get_nth_field(
+/*=================*/
+				/* out: nth field */
+	const dtuple_t*	tuple,	/* in: tuple */
+	ulint		n);	/* in: index of field */
+#else /* UNIV_DEBUG */
+# define dtuple_get_nth_field(tuple, n) ((tuple)->fields + (n))
+#endif /* UNIV_DEBUG */
+/*************************************************************************
+Gets info bits in a data tuple. */
+UNIV_INLINE
+ulint
+dtuple_get_info_bits(
+/*=================*/
+				/* out: info bits */
+	const dtuple_t*	tuple);	/* in: tuple */
+/*************************************************************************
+Sets info bits in a data tuple. */
+UNIV_INLINE
+void
+dtuple_set_info_bits(
+/*=================*/
+	dtuple_t*	tuple,		/* in: tuple */
+	ulint		info_bits);	/* in: info bits */
+/*************************************************************************
+Gets number of fields used in record comparisons. */
+UNIV_INLINE
+ulint
+dtuple_get_n_fields_cmp(
+/*====================*/
+				/* out: number of fields used in comparisons
+				in rem0cmp.* */
+	const dtuple_t*	tuple);	/* in: tuple */
+/*************************************************************************
+Gets number of fields used in record comparisons. */
+UNIV_INLINE
+void
+dtuple_set_n_fields_cmp(
+/*====================*/
+	dtuple_t*	tuple,		/* in: tuple */
+	ulint		n_fields_cmp);	/* in: number of fields used in
+					comparisons in rem0cmp.* */
+/**************************************************************
+Creates a data tuple to a memory heap. The default value for number
+of fields used in record comparisons for this tuple is n_fields. */
+UNIV_INLINE
+dtuple_t*
+dtuple_create(
+/*==========*/
+				/* out, own: created tuple */
+	mem_heap_t*	heap,	/* in: memory heap where the tuple
+				is created */
+	ulint		n_fields); /* in: number of fields */
+
+/**************************************************************
+Wrap data fields in a tuple. The default value for number
+of fields used in record comparisons for this tuple is n_fields. */
+UNIV_INLINE
+const dtuple_t*
+dtuple_from_fields(
+/*===============*/
+					/* out: data tuple */
+	dtuple_t*	tuple,		/* in: storage for data tuple */
+	const dfield_t*	fields,		/* in: fields */
+	ulint		n_fields);	/* in: number of fields */
+
+/*************************************************************************
+Sets number of fields used in a tuple. Normally this is set in
+dtuple_create, but if you want later to set it smaller, you can use this. */
+UNIV_INTERN
+void
+dtuple_set_n_fields(
+/*================*/
+	dtuple_t*	tuple,		/* in: tuple */
+	ulint		n_fields);	/* in: number of fields */
+/*************************************************************************
+Copies a data tuple to another.  This is a shallow copy; if a deep copy
+is desired, dfield_dup() will have to be invoked on each field. */
+UNIV_INLINE
+dtuple_t*
+dtuple_copy(
+/*========*/
+				/* out, own: copy of tuple */
+	const dtuple_t*	tuple,	/* in: tuple to copy from */
+	mem_heap_t*	heap);	/* in: memory heap
+				where the tuple is created */
+/**************************************************************
+The following function returns the sum of data lengths of a tuple. The space
+occupied by the field structs or the tuple struct is not counted. */
+UNIV_INLINE
+ulint
+dtuple_get_data_size(
+/*=================*/
+				/* out: sum of data lens */
+	const dtuple_t*	tuple);	/* in: typed data tuple */
+/*************************************************************************
+Computes the number of externally stored fields in a data tuple. */
+UNIV_INLINE
+ulint
+dtuple_get_n_ext(
+/*=============*/
+				/* out: number of fields */
+	const dtuple_t*	tuple);	/* in: tuple */
+/****************************************************************
+Compare two data tuples, respecting the collation of character fields. */
+UNIV_INTERN
+int
+dtuple_coll_cmp(
+/*============*/
+				/* out: 1, 0 , -1 if tuple1 is greater, equal,
+				less, respectively, than tuple2 */
+	const dtuple_t*	tuple1,	/* in: tuple 1 */
+	const dtuple_t*	tuple2);/* in: tuple 2 */
+/****************************************************************
+Folds a prefix given as the number of fields of a tuple. */
+UNIV_INLINE
+ulint
+dtuple_fold(
+/*========*/
+				/* out: the folded value */
+	const dtuple_t*	tuple,	/* in: the tuple */
+	ulint		n_fields,/* in: number of complete fields to fold */
+	ulint		n_bytes,/* in: number of bytes to fold in an
+				incomplete last field */
+	dulint		tree_id)/* in: index tree id */
+	__attribute__((pure));
+/***********************************************************************
+Sets types of fields binary in a tuple. */
+UNIV_INLINE
+void
+dtuple_set_types_binary(
+/*====================*/
+	dtuple_t*	tuple,	/* in: data tuple */
+	ulint		n);	/* in: number of fields to set */
+/**************************************************************************
+Checks if a dtuple contains an SQL null value. */
+UNIV_INLINE
+ibool
+dtuple_contains_null(
+/*=================*/
+				/* out: TRUE if some field is SQL null */
+	const dtuple_t*	tuple);	/* in: dtuple */
+/**************************************************************
+Checks that a data field is typed. Asserts an error if not. */
+UNIV_INTERN
+ibool
+dfield_check_typed(
+/*===============*/
+				/* out: TRUE if ok */
+	const dfield_t*	field);	/* in: data field */
+/**************************************************************
+Checks that a data tuple is typed. Asserts an error if not. */
+UNIV_INTERN
+ibool
+dtuple_check_typed(
+/*===============*/
+				/* out: TRUE if ok */
+	const dtuple_t*	tuple);	/* in: tuple */
+/**************************************************************
+Checks that a data tuple is typed. */
+UNIV_INTERN
+ibool
+dtuple_check_typed_no_assert(
+/*=========================*/
+				/* out: TRUE if ok */
+	const dtuple_t*	tuple);	/* in: tuple */
+#ifdef UNIV_DEBUG
+/**************************************************************
+Validates the consistency of a tuple which must be complete, i.e,
+all fields must have been set. */
+UNIV_INTERN
+ibool
+dtuple_validate(
+/*============*/
+				/* out: TRUE if ok */
+	const dtuple_t*	tuple);	/* in: tuple */
+#endif /* UNIV_DEBUG */
+/*****************************************************************
+Pretty prints a dfield value according to its data type. */
+UNIV_INTERN
+void
+dfield_print(
+/*=========*/
+	const dfield_t*	dfield);/* in: dfield */
+/*****************************************************************
+Pretty prints a dfield value according to its data type. Also the hex string
+is printed if a string contains non-printable characters. */
+UNIV_INTERN
+void
+dfield_print_also_hex(
+/*==================*/
+	const dfield_t*	dfield);	 /* in: dfield */
+/**************************************************************
+The following function prints the contents of a tuple. */
+UNIV_INTERN
+void
+dtuple_print(
+/*=========*/
+	FILE*		f,	/* in: output stream */
+	const dtuple_t*	tuple);	/* in: tuple */
+/******************************************************************
+Moves parts of long fields in entry to the big record vector so that
+the size of tuple drops below the maximum record size allowed in the
+database. Moves data only from those fields which are not necessary
+to determine uniquely the insertion place of the tuple in the index. */
+UNIV_INTERN
+big_rec_t*
+dtuple_convert_big_rec(
+/*===================*/
+				/* out, own: created big record vector,
+				NULL if we are not able to shorten
+				the entry enough, i.e., if there are
+				too many fixed-length or short fields
+				in entry or the index is clustered */
+	dict_index_t*	index,	/* in: index */
+	dtuple_t*	entry,	/* in/out: index entry */
+	ulint*		n_ext);	/* in/out: number of
+				externally stored columns */
+/******************************************************************
+Puts back to entry the data stored in vector. Note that to ensure the
+fields in entry can accommodate the data, vector must have been created
+from entry with dtuple_convert_big_rec. */
+UNIV_INTERN
+void
+dtuple_convert_back_big_rec(
+/*========================*/
+	dict_index_t*	index,	/* in: index */
+	dtuple_t*	entry,	/* in: entry whose data was put to vector */
+	big_rec_t*	vector);/* in, own: big rec vector; it is
+				freed in this function */
+/******************************************************************
+Frees the memory in a big rec vector. */
+UNIV_INLINE
+void
+dtuple_big_rec_free(
+/*================*/
+	big_rec_t*	vector);	/* in, own: big rec vector; it is
+				freed in this function */
+
+/*######################################################################*/
+
+/* Structure for an SQL data field */
+struct dfield_struct{
+	void*		data;	/* pointer to data */
+	unsigned	ext:1;	/* TRUE=externally stored, FALSE=local */
+	unsigned	len:32;	/* data length; UNIV_SQL_NULL if SQL null */
+	dtype_t		type;	/* type of data */
+};
+
+struct dtuple_struct {
+	ulint		info_bits;	/* info bits of an index record:
+					the default is 0; this field is used
+					if an index record is built from
+					a data tuple */
+	ulint		n_fields;	/* number of fields in dtuple */
+	ulint		n_fields_cmp;	/* number of fields which should
+					be used in comparison services
+					of rem0cmp.*; the index search
+					is performed by comparing only these
+					fields, others are ignored; the
+					default value in dtuple creation is
+					the same value as n_fields */
+	dfield_t*	fields;		/* fields */
+	UT_LIST_NODE_T(dtuple_t) tuple_list;
+					/* data tuples can be linked into a
+					list using this field */
+#ifdef UNIV_DEBUG
+	ulint		magic_n;
+# define		DATA_TUPLE_MAGIC_N	65478679
+#endif /* UNIV_DEBUG */
+};
+
+/* A slot for a field in a big rec vector */
+
+typedef struct big_rec_field_struct	big_rec_field_t;
+struct big_rec_field_struct {
+	ulint		field_no;	/* field number in record */
+	ulint		len;		/* stored data len */
+	const void*	data;		/* stored data */
+};
+
+/* Storage format for overflow data in a big record, that is, a record
+which needs external storage of data fields */
+
+struct big_rec_struct {
+	mem_heap_t*	heap;		/* memory heap from which allocated */
+	ulint		n_fields;	/* number of stored fields */
+	big_rec_field_t* fields;	/* stored fields */
+};
+
+#ifndef UNIV_NONINL
+#include "data0data.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/data0data.ic b/storage/xtradb/include/data0data.ic
new file mode 100644
index 00000000000..f11dbd9fce6
--- /dev/null
+++ b/storage/xtradb/include/data0data.ic
@@ -0,0 +1,608 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/************************************************************************
+SQL data field and tuple
+
+Created 5/30/1994 Heikki Tuuri
+*************************************************************************/
+
+#include "mem0mem.h"
+#include "ut0rnd.h"
+
+#ifdef UNIV_DEBUG
+extern byte data_error;
+
+/*************************************************************************
+Gets pointer to the type struct of SQL data field. */
+UNIV_INLINE
+dtype_t*
+dfield_get_type(
+/*============*/
+				/* out: pointer to the type struct */
+	const dfield_t*	field)	/* in: SQL data field */
+{
+	ut_ad(field);
+
+	return((dtype_t*) &(field->type));
+}
+#endif /* UNIV_DEBUG */
+
+/*************************************************************************
+Sets the type struct of SQL data field. */
+UNIV_INLINE
+void
+dfield_set_type(
+/*============*/
+	dfield_t*	field,	/* in: SQL data field */
+	dtype_t*	type)	/* in: pointer to data type struct */
+{
+	ut_ad(field && type);
+
+	field->type = *type;
+}
+
+#ifdef UNIV_DEBUG
+/*************************************************************************
+Gets pointer to the data in a field. */
+UNIV_INLINE
+void*
+dfield_get_data(
+/*============*/
+				/* out: pointer to data */
+	const dfield_t* field)	/* in: field */
+{
+	ut_ad(field);
+	ut_ad((field->len == UNIV_SQL_NULL)
+	      || (field->data != &data_error));
+
+	return((void*) field->data);
+}
+#endif /* UNIV_DEBUG */
+
+/*************************************************************************
+Gets length of field data. */
+UNIV_INLINE
+ulint
+dfield_get_len(
+/*===========*/
+				/* out: length of data; UNIV_SQL_NULL if
+				SQL null data */
+	const dfield_t*	field)	/* in: field */
+{
+	ut_ad(field);
+	ut_ad((field->len == UNIV_SQL_NULL)
+	      || (field->data != &data_error));
+
+	return(field->len);
+}
+
+/*************************************************************************
+Sets length in a field. */
+UNIV_INLINE
+void
+dfield_set_len(
+/*===========*/
+	dfield_t*	field,	/* in: field */
+	ulint		len)	/* in: length or UNIV_SQL_NULL */
+{
+	ut_ad(field);
+#ifdef UNIV_VALGRIND_DEBUG
+	if (len != UNIV_SQL_NULL) UNIV_MEM_ASSERT_RW(field->data, len);
+#endif /* UNIV_VALGRIND_DEBUG */
+
+	field->ext = 0;
+	field->len = len;
+}
+
+/*************************************************************************
+Determines if a field is SQL NULL */
+UNIV_INLINE
+ulint
+dfield_is_null(
+/*===========*/
+				/* out: nonzero if SQL null data */
+	const dfield_t* field)	/* in: field */
+{
+	ut_ad(field);
+
+	return(field->len == UNIV_SQL_NULL);
+}
+
+/*************************************************************************
+Determines if a field is externally stored */
+UNIV_INLINE
+ulint
+dfield_is_ext(
+/*==========*/
+				/* out: nonzero if externally stored */
+	const dfield_t* field)	/* in: field */
+{
+	ut_ad(field);
+
+	return(UNIV_UNLIKELY(field->ext));
+}
+
+/*************************************************************************
+Sets the "external storage" flag */
+UNIV_INLINE
+void
+dfield_set_ext(
+/*===========*/
+	dfield_t*	field)	/* in/out: field */
+{
+	ut_ad(field);
+
+	field->ext = 1;
+}
+
+/*************************************************************************
+Sets pointer to the data and length in a field. */
+UNIV_INLINE
+void
+dfield_set_data(
+/*============*/
+	dfield_t*	field,	/* in: field */
+	const void*	data,	/* in: data */
+	ulint		len)	/* in: length or UNIV_SQL_NULL */
+{
+	ut_ad(field);
+
+#ifdef UNIV_VALGRIND_DEBUG
+	if (len != UNIV_SQL_NULL) UNIV_MEM_ASSERT_RW(data, len);
+#endif /* UNIV_VALGRIND_DEBUG */
+	field->data = (void*) data;
+	field->ext = 0;
+	field->len = len;
+}
+
+/*************************************************************************
+Sets a data field to SQL NULL. */
+UNIV_INLINE
+void
+dfield_set_null(
+/*============*/
+	dfield_t*	field)	/* in/out: field */
+{
+	dfield_set_data(field, NULL, UNIV_SQL_NULL);
+}
+
+/*************************************************************************
+Copies the data and len fields. */
+UNIV_INLINE
+void
+dfield_copy_data(
+/*=============*/
+	dfield_t*	field1,	/* out: field to copy to */
+	const dfield_t*	field2)	/* in: field to copy from */
+{
+	ut_ad(field1 && field2);
+
+	field1->data = field2->data;
+	field1->len = field2->len;
+	field1->ext = field2->ext;
+}
+
+/*************************************************************************
+Copies a data field to another. */
+UNIV_INLINE
+void
+dfield_copy(
+/*========*/
+	dfield_t*	field1,	/* out: field to copy to */
+	const dfield_t*	field2)	/* in: field to copy from */
+{
+	*field1 = *field2;
+}
+
+/*************************************************************************
+Copies the data pointed to by a data field. */
+UNIV_INLINE
+void
+dfield_dup(
+/*=======*/
+	dfield_t*	field,	/* in/out: data field */
+	mem_heap_t*	heap)	/* in: memory heap where allocated */
+{
+	if (!dfield_is_null(field)) {
+		UNIV_MEM_ASSERT_RW(field->data, field->len);
+		field->data = mem_heap_dup(heap, field->data, field->len);
+	}
+}
+
+/*************************************************************************
+Tests if data length and content is equal for two dfields. */
+UNIV_INLINE
+ibool
+dfield_datas_are_binary_equal(
+/*==========================*/
+				/* out: TRUE if equal */
+	const dfield_t*	field1,	/* in: field */
+	const dfield_t*	field2)	/* in: field */
+{
+	ulint	len;
+
+	len = field1->len;
+
+	return(len == field2->len
+	       && (len == UNIV_SQL_NULL
+		   || !memcmp(field1->data, field2->data, len)));
+}
+
+/*************************************************************************
+Gets info bits in a data tuple. */
+UNIV_INLINE
+ulint
+dtuple_get_info_bits(
+/*=================*/
+				/* out: info bits */
+	const dtuple_t*	tuple)	/* in: tuple */
+{
+	ut_ad(tuple);
+
+	return(tuple->info_bits);
+}
+
+/*************************************************************************
+Sets info bits in a data tuple. */
+UNIV_INLINE
+void
+dtuple_set_info_bits(
+/*=================*/
+	dtuple_t*	tuple,		/* in: tuple */
+	ulint		info_bits)	/* in: info bits */
+{
+	ut_ad(tuple);
+
+	tuple->info_bits = info_bits;
+}
+
+/*************************************************************************
+Gets number of fields used in record comparisons. */
+UNIV_INLINE
+ulint
+dtuple_get_n_fields_cmp(
+/*====================*/
+				/* out: number of fields used in comparisons
+				in rem0cmp.* */
+	const dtuple_t*	tuple)	/* in: tuple */
+{
+	ut_ad(tuple);
+
+	return(tuple->n_fields_cmp);
+}
+
+/*************************************************************************
+Sets number of fields used in record comparisons. */
+UNIV_INLINE
+void
+dtuple_set_n_fields_cmp(
+/*====================*/
+	dtuple_t*	tuple,		/* in: tuple */
+	ulint		n_fields_cmp)	/* in: number of fields used in
+					comparisons in rem0cmp.* */
+{
+	ut_ad(tuple);
+	ut_ad(n_fields_cmp <= tuple->n_fields);
+
+	tuple->n_fields_cmp = n_fields_cmp;
+}
+
+/*************************************************************************
+Gets number of fields in a data tuple. */
+UNIV_INLINE
+ulint
+dtuple_get_n_fields(
+/*================*/
+				/* out: number of fields */
+	const dtuple_t*	tuple)	/* in: tuple */
+{
+	ut_ad(tuple);
+
+	return(tuple->n_fields);
+}
+
+#ifdef UNIV_DEBUG
+/*************************************************************************
+Gets nth field of a tuple. */
+UNIV_INLINE
+dfield_t*
+dtuple_get_nth_field(
+/*=================*/
+				/* out: nth field */
+	const dtuple_t*	tuple,	/* in: tuple */
+	ulint		n)	/* in: index of field */
+{
+	ut_ad(tuple);
+	ut_ad(n < tuple->n_fields);
+
+	return((dfield_t*) tuple->fields + n);
+}
+#endif /* UNIV_DEBUG */
+
+/**************************************************************
+Creates a data tuple to a memory heap. The default value for number
+of fields used in record comparisons for this tuple is n_fields. */
+UNIV_INLINE
+dtuple_t*
+dtuple_create(
+/*==========*/
+				/* out, own: created tuple */
+	mem_heap_t*	heap,	/* in: memory heap where the tuple
+				is created */
+	ulint		n_fields) /* in: number of fields */
+{
+	dtuple_t*	tuple;
+
+	ut_ad(heap);
+
+	tuple = (dtuple_t*) mem_heap_alloc(heap, sizeof(dtuple_t)
+					   + n_fields * sizeof(dfield_t));
+	tuple->info_bits = 0;
+	tuple->n_fields = n_fields;
+	tuple->n_fields_cmp = n_fields;
+	tuple->fields = (dfield_t*) &tuple[1];
+
+#ifdef UNIV_DEBUG
+	tuple->magic_n = DATA_TUPLE_MAGIC_N;
+
+	{	/* In the debug version, initialize fields to an error value */
+		ulint	i;
+
+		for (i = 0; i < n_fields; i++) {
+			dfield_t*       field;
+
+			field = dtuple_get_nth_field(tuple, i);
+
+			dfield_set_len(field, UNIV_SQL_NULL);
+			field->data = &data_error;
+			dfield_get_type(field)->mtype = DATA_ERROR;
+		}
+	}
+
+	UNIV_MEM_INVALID(tuple->fields, n_fields * sizeof *tuple->fields);
+#endif
+	return(tuple);
+}
+
+/**************************************************************
+Wrap data fields in a tuple. The default value for number
+of fields used in record comparisons for this tuple is n_fields. */
+UNIV_INLINE
+const dtuple_t*
+dtuple_from_fields(
+/*===============*/
+					/* out: data tuple */
+	dtuple_t*	tuple,		/* in: storage for data tuple */
+	const dfield_t*	fields,		/* in: fields */
+	ulint		n_fields)	/* in: number of fields */
+{
+	tuple->info_bits = 0;
+	tuple->n_fields = tuple->n_fields_cmp = n_fields;
+	tuple->fields = (dfield_t*) fields;
+	ut_d(tuple->magic_n = DATA_TUPLE_MAGIC_N);
+
+	return(tuple);
+}
+
+/*************************************************************************
+Copies a data tuple to another.  This is a shallow copy; if a deep copy
+is desired, dfield_dup() will have to be invoked on each field. */
+UNIV_INLINE
+dtuple_t*
+dtuple_copy(
+/*========*/
+				/* out, own: copy of tuple */
+	const dtuple_t*	tuple,	/* in: tuple to copy from */
+	mem_heap_t*	heap)	/* in: memory heap
+				where the tuple is created */
+{
+	ulint		n_fields	= dtuple_get_n_fields(tuple);
+	dtuple_t*	new_tuple	= dtuple_create(heap, n_fields);
+	ulint		i;
+
+	for (i = 0; i < n_fields; i++) {
+		dfield_copy(dtuple_get_nth_field(new_tuple, i),
+			    dtuple_get_nth_field(tuple, i));
+	}
+
+	return(new_tuple);
+}
+
+/**************************************************************
+The following function returns the sum of data lengths of a tuple. The space
+occupied by the field structs or the tuple struct is not counted. Neither
+is possible space in externally stored parts of the field. */
+UNIV_INLINE
+ulint
+dtuple_get_data_size(
+/*=================*/
+				/* out: sum of data lengths */
+	const dtuple_t*	tuple)	/* in: typed data tuple */
+{
+	const dfield_t*	field;
+	ulint		n_fields;
+	ulint		len;
+	ulint		i;
+	ulint		sum	= 0;
+
+	ut_ad(tuple);
+	ut_ad(dtuple_check_typed(tuple));
+	ut_ad(tuple->magic_n == DATA_TUPLE_MAGIC_N);
+
+	n_fields = tuple->n_fields;
+
+	for (i = 0; i < n_fields; i++) {
+		field = dtuple_get_nth_field(tuple,  i);
+		len = dfield_get_len(field);
+
+		if (len == UNIV_SQL_NULL) {
+			len = dtype_get_sql_null_size(dfield_get_type(field));
+		}
+
+		sum += len;
+	}
+
+	return(sum);
+}
+
+/*************************************************************************
+Computes the number of externally stored fields in a data tuple. */
+UNIV_INLINE
+ulint
+dtuple_get_n_ext(
+/*=============*/
+				/* out: number of externally stored fields */
+	const dtuple_t*	tuple)	/* in: tuple */
+{
+	ulint	n_ext		= 0;
+	ulint	n_fields	= tuple->n_fields;
+	ulint	i;
+
+	ut_ad(tuple);
+	ut_ad(dtuple_check_typed(tuple));
+	ut_ad(tuple->magic_n == DATA_TUPLE_MAGIC_N);
+
+	for (i = 0; i < n_fields; i++) {
+		n_ext += dtuple_get_nth_field(tuple, i)->ext;
+	}
+
+	return(n_ext);
+}
+
+/***********************************************************************
+Sets types of fields binary in a tuple. */
+UNIV_INLINE
+void
+dtuple_set_types_binary(
+/*====================*/
+	dtuple_t*	tuple,	/* in: data tuple */
+	ulint		n)	/* in: number of fields to set */
+{
+	dtype_t*	dfield_type;
+	ulint		i;
+
+	for (i = 0; i < n; i++) {
+		dfield_type = dfield_get_type(dtuple_get_nth_field(tuple, i));
+		dtype_set(dfield_type, DATA_BINARY, 0, 0);
+	}
+}
+
+/****************************************************************
+Folds a prefix given as the number of fields of a tuple. */
+UNIV_INLINE
+ulint
+dtuple_fold(
+/*========*/
+				/* out: the folded value */
+	const dtuple_t*	tuple,	/* in: the tuple */
+	ulint		n_fields,/* in: number of complete fields to fold */
+	ulint		n_bytes,/* in: number of bytes to fold in an
+				incomplete last field */
+	dulint		tree_id)/* in: index tree id */
+{
+	const dfield_t*	field;
+	ulint		i;
+	const byte*	data;
+	ulint		len;
+	ulint		fold;
+
+	ut_ad(tuple);
+	ut_ad(tuple->magic_n == DATA_TUPLE_MAGIC_N);
+	ut_ad(dtuple_check_typed(tuple));
+
+	fold = ut_fold_dulint(tree_id);
+
+	for (i = 0; i < n_fields; i++) {
+		field = dtuple_get_nth_field(tuple, i);
+
+		data = (const byte*) dfield_get_data(field);
+		len = dfield_get_len(field);
+
+		if (len != UNIV_SQL_NULL) {
+			fold = ut_fold_ulint_pair(fold,
+						  ut_fold_binary(data, len));
+		}
+	}
+
+	if (n_bytes > 0) {
+		field = dtuple_get_nth_field(tuple, i);
+
+		data = (const byte*) dfield_get_data(field);
+		len = dfield_get_len(field);
+
+		if (len != UNIV_SQL_NULL) {
+			if (len > n_bytes) {
+				len = n_bytes;
+			}
+
+			fold = ut_fold_ulint_pair(fold,
+						  ut_fold_binary(data, len));
+		}
+	}
+
+	return(fold);
+}
+
+/**************************************************************************
+Writes an SQL null field full of zeros. */
+UNIV_INLINE
+void
+data_write_sql_null(
+/*================*/
+	byte*	data,	/* in: pointer to a buffer of size len */
+	ulint	len)	/* in: SQL null size in bytes */
+{
+	memset(data, 0, len);
+}
+
+/**************************************************************************
+Checks if a dtuple contains an SQL null value. */
+UNIV_INLINE
+ibool
+dtuple_contains_null(
+/*=================*/
+				/* out: TRUE if some field is SQL null */
+	const dtuple_t*	tuple)	/* in: dtuple */
+{
+	ulint	n;
+	ulint	i;
+
+	n = dtuple_get_n_fields(tuple);
+
+	for (i = 0; i < n; i++) {
+		if (dfield_is_null(dtuple_get_nth_field(tuple, i))) {
+
+			return(TRUE);
+		}
+	}
+
+	return(FALSE);
+}
+
+/******************************************************************
+Frees the memory in a big rec vector. */
+UNIV_INLINE
+void
+dtuple_big_rec_free(
+/*================*/
+	big_rec_t*	vector)	/* in, own: big rec vector; it is
+				freed in this function */
+{
+	mem_heap_free(vector->heap);
+}
diff --git a/storage/xtradb/include/data0type.h b/storage/xtradb/include/data0type.h
new file mode 100644
index 00000000000..1f10878984b
--- /dev/null
+++ b/storage/xtradb/include/data0type.h
@@ -0,0 +1,471 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Data types
+
+Created 1/16/1996 Heikki Tuuri
+*******************************************************/
+
+#ifndef data0type_h
+#define data0type_h
+
+#include "univ.i"
+
+extern ulint	data_mysql_default_charset_coll;
+#define DATA_MYSQL_LATIN1_SWEDISH_CHARSET_COLL 8
+#define DATA_MYSQL_BINARY_CHARSET_COLL 63
+
+/* SQL data type struct */
+typedef struct dtype_struct		dtype_t;
+
+/*-------------------------------------------*/
+/* The 'MAIN TYPE' of a column */
+#define	DATA_VARCHAR	1	/* character varying of the
+				latin1_swedish_ci charset-collation; note
+				that the MySQL format for this, DATA_BINARY,
+				DATA_VARMYSQL, is also affected by whether the
+				'precise type' contains
+				DATA_MYSQL_TRUE_VARCHAR */
+#define DATA_CHAR	2	/* fixed length character of the
+				latin1_swedish_ci charset-collation */
+#define DATA_FIXBINARY	3	/* binary string of fixed length */
+#define DATA_BINARY	4	/* binary string */
+#define DATA_BLOB	5	/* binary large object, or a TEXT type;
+				if prtype & DATA_BINARY_TYPE == 0, then this is
+				actually a TEXT column (or a BLOB created
+				with < 4.0.14; since column prefix indexes
+				came only in 4.0.14, the missing flag in BLOBs
+				created before that does not cause any harm) */
+#define	DATA_INT	6	/* integer: can be any size 1 - 8 bytes */
+#define	DATA_SYS_CHILD	7	/* address of the child page in node pointer */
+#define	DATA_SYS	8	/* system column */
+
+/* Data types >= DATA_FLOAT must be compared using the whole field, not as
+binary strings */
+
+#define DATA_FLOAT	9
+#define DATA_DOUBLE	10
+#define DATA_DECIMAL	11	/* decimal number stored as an ASCII string */
+#define	DATA_VARMYSQL	12	/* any charset varying length char */
+#define	DATA_MYSQL	13	/* any charset fixed length char */
+				/* NOTE that 4.1.1 used DATA_MYSQL and
+				DATA_VARMYSQL for all character sets, and the
+				charset-collation for tables created with it
+				can also be latin1_swedish_ci */
+#define DATA_MTYPE_MAX	63	/* dtype_store_for_order_and_null_size()
+				requires the values are <= 63 */
+/*-------------------------------------------*/
+/* The 'PRECISE TYPE' of a column */
+/*
+Tables created by a MySQL user have the following convention:
+
+- In the least significant byte in the precise type we store the MySQL type
+code (not applicable for system columns).
+
+- In the second least significant byte we OR flags DATA_NOT_NULL,
+DATA_UNSIGNED, DATA_BINARY_TYPE.
+
+- In the third least significant byte of the precise type of string types we
+store the MySQL charset-collation code. In DATA_BLOB columns created with
+< 4.0.14 we do not actually know if it is a BLOB or a TEXT column. Since there
+are no indexes on prefixes of BLOB or TEXT columns in < 4.0.14, this is no
+problem, though.
+
+Note that versions < 4.1.2 or < 5.0.1 did not store the charset code to the
+precise type, since the charset was always the default charset of the MySQL
+installation. If the stored charset code is 0 in the system table SYS_COLUMNS
+of InnoDB, that means that the default charset of this MySQL installation
+should be used.
+
+When loading a table definition from the system tables to the InnoDB data
+dictionary cache in main memory, InnoDB versions >= 4.1.2 and >= 5.0.1 check
+if the stored charset-collation is 0, and if that is the case and the type is
+a non-binary string, replace that 0 by the default charset-collation code of
+this MySQL installation. In short, in old tables, the charset-collation code
+in the system tables on disk can be 0, but in in-memory data structures
+(dtype_t), the charset-collation code is always != 0 for non-binary string
+types.
+
+In new tables, in binary string types, the charset-collation code is the
+MySQL code for the 'binary charset', that is, != 0.
+
+For binary string types and for DATA_CHAR, DATA_VARCHAR, and for those
+DATA_BLOB which are binary or have the charset-collation latin1_swedish_ci,
+InnoDB performs all comparisons internally, without resorting to the MySQL
+comparison functions. This is to save CPU time.
+
+InnoDB's own internal system tables have different precise types for their
+columns, and for them the precise type is usually not used at all.
+*/
+
+#define DATA_ENGLISH	4	/* English language character string: this
+				is a relic from pre-MySQL time and only used
+				for InnoDB's own system tables */
+#define DATA_ERROR	111	/* another relic from pre-MySQL time */
+
+#define DATA_MYSQL_TYPE_MASK 255 /* AND with this mask to extract the MySQL
+				 type from the precise type */
+#define DATA_MYSQL_TRUE_VARCHAR 15 /* MySQL type code for the >= 5.0.3
+				   format true VARCHAR */
+
+/* Precise data types for system columns and the length of those columns;
+NOTE: the values must run from 0 up in the order given! All codes must
+be less than 256 */
+#define	DATA_ROW_ID	0	/* row id: a dulint */
+#define DATA_ROW_ID_LEN	6	/* stored length for row id */
+
+#define DATA_TRX_ID	1	/* transaction id: 6 bytes */
+#define DATA_TRX_ID_LEN	6
+
+#define	DATA_ROLL_PTR	2	/* rollback data pointer: 7 bytes */
+#define DATA_ROLL_PTR_LEN 7
+
+#define	DATA_N_SYS_COLS 3	/* number of system columns defined above */
+
+#define DATA_SYS_PRTYPE_MASK 0xF /* mask to extract the above from prtype */
+
+/* Flags ORed to the precise data type */
+#define DATA_NOT_NULL	256	/* this is ORed to the precise type when
+				the column is declared as NOT NULL */
+#define DATA_UNSIGNED	512	/* this id ORed to the precise type when
+				we have an unsigned integer type */
+#define	DATA_BINARY_TYPE 1024	/* if the data type is a binary character
+				string, this is ORed to the precise type:
+				this only holds for tables created with
+				>= MySQL-4.0.14 */
+/* #define	DATA_NONLATIN1	2048 This is a relic from < 4.1.2 and < 5.0.1.
+				In earlier versions this was set for some
+				BLOB columns.
+*/
+#define	DATA_LONG_TRUE_VARCHAR 4096	/* this is ORed to the precise data
+				type when the column is true VARCHAR where
+				MySQL uses 2 bytes to store the data len;
+				for shorter VARCHARs MySQL uses only 1 byte */
+/*-------------------------------------------*/
+
+/* This many bytes we need to store the type information affecting the
+alphabetical order for a single field and decide the storage size of an
+SQL null*/
+#define DATA_ORDER_NULL_TYPE_BUF_SIZE		4
+/* In the >= 4.1.x storage format we add 2 bytes more so that we can also
+store the charset-collation number; one byte is left unused, though */
+#define DATA_NEW_ORDER_NULL_TYPE_BUF_SIZE	6
+
+/*************************************************************************
+Gets the MySQL type code from a dtype. */
+UNIV_INLINE
+ulint
+dtype_get_mysql_type(
+/*=================*/
+				/* out: MySQL type code; this is NOT an InnoDB
+				type code! */
+	const dtype_t*	type);	/* in: type struct */
+/*************************************************************************
+Determine how many bytes the first n characters of the given string occupy.
+If the string is shorter than n characters, returns the number of bytes
+the characters in the string occupy. */
+UNIV_INTERN
+ulint
+dtype_get_at_most_n_mbchars(
+/*========================*/
+					/* out: length of the prefix,
+					in bytes */
+	ulint		prtype,		/* in: precise type */
+	ulint		mbminlen,	/* in: minimum length of a
+					multi-byte character */
+	ulint		mbmaxlen,	/* in: maximum length of a
+					multi-byte character */
+	ulint		prefix_len,	/* in: length of the requested
+					prefix, in characters, multiplied by
+					dtype_get_mbmaxlen(dtype) */
+	ulint		data_len,	/* in: length of str (in bytes) */
+	const char*	str);		/* in: the string whose prefix
+					length is being determined */
+/*************************************************************************
+Checks if a data main type is a string type. Also a BLOB is considered a
+string type. */
+UNIV_INTERN
+ibool
+dtype_is_string_type(
+/*=================*/
+			/* out: TRUE if string type */
+	ulint	mtype);	/* in: InnoDB main data type code: DATA_CHAR, ... */
+/*************************************************************************
+Checks if a type is a binary string type. Note that for tables created with
+< 4.0.14, we do not know if a DATA_BLOB column is a BLOB or a TEXT column. For
+those DATA_BLOB columns this function currently returns FALSE. */
+UNIV_INTERN
+ibool
+dtype_is_binary_string_type(
+/*========================*/
+			/* out: TRUE if binary string type */
+	ulint	mtype,	/* in: main data type */
+	ulint	prtype);/* in: precise type */
+/*************************************************************************
+Checks if a type is a non-binary string type. That is, dtype_is_string_type is
+TRUE and dtype_is_binary_string_type is FALSE. Note that for tables created
+with < 4.0.14, we do not know if a DATA_BLOB column is a BLOB or a TEXT column.
+For those DATA_BLOB columns this function currently returns TRUE. */
+UNIV_INTERN
+ibool
+dtype_is_non_binary_string_type(
+/*============================*/
+			/* out: TRUE if non-binary string type */
+	ulint	mtype,	/* in: main data type */
+	ulint	prtype);/* in: precise type */
+/*************************************************************************
+Sets a data type structure. */
+UNIV_INLINE
+void
+dtype_set(
+/*======*/
+	dtype_t*	type,	/* in: type struct to init */
+	ulint		mtype,	/* in: main data type */
+	ulint		prtype,	/* in: precise type */
+	ulint		len);	/* in: precision of type */
+/*************************************************************************
+Copies a data type structure. */
+UNIV_INLINE
+void
+dtype_copy(
+/*=======*/
+	dtype_t*	type1,	/* in: type struct to copy to */
+	const dtype_t*	type2);	/* in: type struct to copy from */
+/*************************************************************************
+Gets the SQL main data type. */
+UNIV_INLINE
+ulint
+dtype_get_mtype(
+/*============*/
+	const dtype_t*	type);
+/*************************************************************************
+Gets the precise data type. */
+UNIV_INLINE
+ulint
+dtype_get_prtype(
+/*=============*/
+	const dtype_t*	type);
+/*************************************************************************
+Compute the mbminlen and mbmaxlen members of a data type structure. */
+UNIV_INLINE
+void
+dtype_get_mblen(
+/*============*/
+	ulint	mtype,		/* in: main type */
+	ulint	prtype,		/* in: precise type (and collation) */
+	ulint*	mbminlen,	/* out: minimum length of a
+				multi-byte character */
+	ulint*	mbmaxlen);	/* out: maximum length of a
+				multi-byte character */
+/*************************************************************************
+Gets the MySQL charset-collation code for MySQL string types. */
+UNIV_INLINE
+ulint
+dtype_get_charset_coll(
+/*===================*/
+	ulint	prtype);/* in: precise data type */
+/*************************************************************************
+Forms a precise type from the < 4.1.2 format precise type plus the
+charset-collation code. */
+UNIV_INTERN
+ulint
+dtype_form_prtype(
+/*==============*/
+	ulint	old_prtype,	/* in: the MySQL type code and the flags
+				DATA_BINARY_TYPE etc. */
+	ulint	charset_coll);	/* in: MySQL charset-collation code */
+/*************************************************************************
+Determines if a MySQL string type is a subset of UTF-8.  This function
+may return false negatives, in case further character-set collation
+codes are introduced in MySQL later. */
+UNIV_INLINE
+ibool
+dtype_is_utf8(
+/*==========*/
+			/* out: TRUE if a subset of UTF-8 */
+	ulint	prtype);/* in: precise data type */
+/*************************************************************************
+Gets the type length. */
+UNIV_INLINE
+ulint
+dtype_get_len(
+/*==========*/
+	const dtype_t*	type);
+/*************************************************************************
+Gets the minimum length of a character, in bytes. */
+UNIV_INLINE
+ulint
+dtype_get_mbminlen(
+/*===============*/
+				/* out: minimum length of a char, in bytes,
+				or 0 if this is not a character type */
+	const dtype_t*	type);	/* in: type */
+/*************************************************************************
+Gets the maximum length of a character, in bytes. */
+UNIV_INLINE
+ulint
+dtype_get_mbmaxlen(
+/*===============*/
+				/* out: maximum length of a char, in bytes,
+				or 0 if this is not a character type */
+	const dtype_t*	type);	/* in: type */
+/*************************************************************************
+Gets the padding character code for the type. */
+UNIV_INLINE
+ulint
+dtype_get_pad_char(
+/*===============*/
+				/* out: padding character code, or
+				ULINT_UNDEFINED if no padding specified */
+	ulint	mtype,		/* in: main type */
+	ulint	prtype);	/* in: precise type */
+/***************************************************************************
+Returns the size of a fixed size data type, 0 if not a fixed size type. */
+UNIV_INLINE
+ulint
+dtype_get_fixed_size_low(
+/*=====================*/
+				/* out: fixed size, or 0 */
+	ulint	mtype,		/* in: main type */
+	ulint	prtype,		/* in: precise type */
+	ulint	len,		/* in: length */
+	ulint	mbminlen,	/* in: minimum length of a multibyte char */
+	ulint	mbmaxlen);	/* in: maximum length of a multibyte char */
+/***************************************************************************
+Returns the minimum size of a data type. */
+UNIV_INLINE
+ulint
+dtype_get_min_size_low(
+/*===================*/
+				/* out: minimum size */
+	ulint	mtype,		/* in: main type */
+	ulint	prtype,		/* in: precise type */
+	ulint	len,		/* in: length */
+	ulint	mbminlen,	/* in: minimum length of a multibyte char */
+	ulint	mbmaxlen);	/* in: maximum length of a multibyte char */
+/***************************************************************************
+Returns the maximum size of a data type. Note: types in system tables may be
+incomplete and return incorrect information. */
+UNIV_INLINE
+ulint
+dtype_get_max_size_low(
+/*===================*/
+				/* out: maximum size */
+	ulint	mtype,		/* in: main type */
+	ulint	len);		/* in: length */
+/***************************************************************************
+Returns the ROW_FORMAT=REDUNDANT stored SQL NULL size of a type.
+For fixed length types it is the fixed length of the type, otherwise 0. */
+UNIV_INLINE
+ulint
+dtype_get_sql_null_size(
+/*====================*/
+				/* out: SQL null storage size
+				in ROW_FORMAT=REDUNDANT */
+	const dtype_t*	type);	/* in: type */
+/**************************************************************************
+Reads to a type the stored information which determines its alphabetical
+ordering and the storage size of an SQL NULL value. */
+UNIV_INLINE
+void
+dtype_read_for_order_and_null_size(
+/*===============================*/
+	dtype_t*	type,	/* in: type struct */
+	const byte*	buf);	/* in: buffer for the stored order info */
+/**************************************************************************
+Stores for a type the information which determines its alphabetical ordering
+and the storage size of an SQL NULL value. This is the >= 4.1.x storage
+format. */
+UNIV_INLINE
+void
+dtype_new_store_for_order_and_null_size(
+/*====================================*/
+	byte*		buf,	/* in: buffer for
+				DATA_NEW_ORDER_NULL_TYPE_BUF_SIZE
+				bytes where we store the info */
+	const dtype_t*	type,	/* in: type struct */
+	ulint		prefix_len);/* in: prefix length to
+				replace type->len, or 0 */
+/**************************************************************************
+Reads to a type the stored information which determines its alphabetical
+ordering and the storage size of an SQL NULL value. This is the 4.1.x storage
+format. */
+UNIV_INLINE
+void
+dtype_new_read_for_order_and_null_size(
+/*===================================*/
+	dtype_t*	type,	/* in: type struct */
+	const byte*	buf);	/* in: buffer for stored type order info */
+
+/*************************************************************************
+Validates a data type structure. */
+UNIV_INTERN
+ibool
+dtype_validate(
+/*===========*/
+				/* out: TRUE if ok */
+	const dtype_t*	type);	/* in: type struct to validate */
+/*************************************************************************
+Prints a data type structure. */
+UNIV_INTERN
+void
+dtype_print(
+/*========*/
+	const dtype_t*	type);	/* in: type */
+
+/* Structure for an SQL data type.
+If you add fields to this structure, be sure to initialize them everywhere.
+This structure is initialized in the following functions:
+dtype_set()
+dtype_read_for_order_and_null_size()
+dtype_new_read_for_order_and_null_size()
+sym_tab_add_null_lit() */
+
+struct dtype_struct{
+	unsigned	mtype:8;	/* main data type */
+	unsigned	prtype:24;	/* precise type; MySQL data
+					type, charset code, flags to
+					indicate nullability,
+					signedness, whether this is a
+					binary string, whether this is
+					a true VARCHAR where MySQL
+					uses 2 bytes to store the length */
+
+	/* the remaining fields do not affect alphabetical ordering: */
+
+	unsigned	len:16;		/* length; for MySQL data this
+					is field->pack_length(),
+					except that for a >= 5.0.3
+					type true VARCHAR this is the
+					maximum byte length of the
+					string data (in addition to
+					the string, MySQL uses 1 or 2
+					bytes to store the string length) */
+
+	unsigned	mbminlen:2;	/* minimum length of a
+					character, in bytes */
+	unsigned	mbmaxlen:3;	/* maximum length of a
+					character, in bytes */
+};
+
+#ifndef UNIV_NONINL
+#include "data0type.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/data0type.ic b/storage/xtradb/include/data0type.ic
new file mode 100644
index 00000000000..d4c1080bebe
--- /dev/null
+++ b/storage/xtradb/include/data0type.ic
@@ -0,0 +1,587 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Data types
+
+Created 1/16/1996 Heikki Tuuri
+*******************************************************/
+
+#include "mach0data.h"
+#include "ha_prototypes.h"
+
+/*************************************************************************
+Gets the MySQL charset-collation code for MySQL string types. */
+UNIV_INLINE
+ulint
+dtype_get_charset_coll(
+/*===================*/
+	ulint	prtype)	/* in: precise data type */
+{
+	return((prtype >> 16) & 0xFFUL);
+}
+
+/*************************************************************************
+Determines if a MySQL string type is a subset of UTF-8.  This function
+may return false negatives, in case further character-set collation
+codes are introduced in MySQL later. */
+UNIV_INLINE
+ibool
+dtype_is_utf8(
+/*==========*/
+			/* out: TRUE if a subset of UTF-8 */
+	ulint	prtype)	/* in: precise data type */
+{
+	/* These codes have been copied from strings/ctype-extra.c
+	and strings/ctype-utf8.c. */
+	switch (dtype_get_charset_coll(prtype)) {
+	case 11: /* ascii_general_ci */
+	case 65: /* ascii_bin */
+	case 33: /* utf8_general_ci */
+	case 83: /* utf8_bin */
+	case 254: /* utf8_general_cs */
+			return(TRUE);
+	}
+
+	return(FALSE);
+}
+
+/*************************************************************************
+Gets the MySQL type code from a dtype. */
+UNIV_INLINE
+ulint
+dtype_get_mysql_type(
+/*=================*/
+				/* out: MySQL type code; this is NOT an InnoDB
+				type code! */
+	const dtype_t*	type)	/* in: type struct */
+{
+	return(type->prtype & 0xFFUL);
+}
+
+/*************************************************************************
+Compute the mbminlen and mbmaxlen members of a data type structure. */
+UNIV_INLINE
+void
+dtype_get_mblen(
+/*============*/
+	ulint	mtype,		/* in: main type */
+	ulint	prtype,		/* in: precise type (and collation) */
+	ulint*	mbminlen,	/* out: minimum length of a
+				multi-byte character */
+	ulint*	mbmaxlen)	/* out: maximum length of a
+				multi-byte character */
+{
+	if (dtype_is_string_type(mtype)) {
+#ifndef UNIV_HOTBACKUP
+		innobase_get_cset_width(dtype_get_charset_coll(prtype),
+					mbminlen, mbmaxlen);
+		ut_ad(*mbminlen <= *mbmaxlen);
+		ut_ad(*mbminlen <= 2); /* mbminlen in dtype_t is 0..3 */
+		ut_ad(*mbmaxlen < 1 << 3); /* mbmaxlen in dtype_t is 0..7 */
+#else /* !UNIV_HOTBACKUP */
+		ut_a(mtype <= DATA_BINARY);
+		*mbminlen = *mbmaxlen = 1;
+#endif /* !UNIV_HOTBACKUP */
+	} else {
+		*mbminlen = *mbmaxlen = 0;
+	}
+}
+
+/*************************************************************************
+Compute the mbminlen and mbmaxlen members of a data type structure. */
+UNIV_INLINE
+void
+dtype_set_mblen(
+/*============*/
+	dtype_t*	type)	/* in/out: type */
+{
+	ulint	mbminlen;
+	ulint	mbmaxlen;
+
+	dtype_get_mblen(type->mtype, type->prtype, &mbminlen, &mbmaxlen);
+	type->mbminlen = mbminlen;
+	type->mbmaxlen = mbmaxlen;
+
+	ut_ad(dtype_validate(type));
+}
+
+/*************************************************************************
+Sets a data type structure. */
+UNIV_INLINE
+void
+dtype_set(
+/*======*/
+	dtype_t*	type,	/* in: type struct to init */
+	ulint		mtype,	/* in: main data type */
+	ulint		prtype,	/* in: precise type */
+	ulint		len)	/* in: precision of type */
+{
+	ut_ad(type);
+	ut_ad(mtype <= DATA_MTYPE_MAX);
+
+	type->mtype = mtype;
+	type->prtype = prtype;
+	type->len = len;
+
+	dtype_set_mblen(type);
+}
+
+/*************************************************************************
+Copies a data type structure. */
+UNIV_INLINE
+void
+dtype_copy(
+/*=======*/
+	dtype_t*	type1,	/* in: type struct to copy to */
+	const dtype_t*	type2)	/* in: type struct to copy from */
+{
+	*type1 = *type2;
+
+	ut_ad(dtype_validate(type1));
+}
+
+/*************************************************************************
+Gets the SQL main data type. */
+UNIV_INLINE
+ulint
+dtype_get_mtype(
+/*============*/
+	const dtype_t*	type)
+{
+	ut_ad(type);
+
+	return(type->mtype);
+}
+
+/*************************************************************************
+Gets the precise data type. */
+UNIV_INLINE
+ulint
+dtype_get_prtype(
+/*=============*/
+	const dtype_t*	type)
+{
+	ut_ad(type);
+
+	return(type->prtype);
+}
+
+/*************************************************************************
+Gets the type length. */
+UNIV_INLINE
+ulint
+dtype_get_len(
+/*==========*/
+	const dtype_t*	type)
+{
+	ut_ad(type);
+
+	return(type->len);
+}
+
+/*************************************************************************
+Gets the minimum length of a character, in bytes. */
+UNIV_INLINE
+ulint
+dtype_get_mbminlen(
+/*===============*/
+				/* out: minimum length of a char, in bytes,
+				or 0 if this is not a character type */
+	const dtype_t*	type)	/* in: type */
+{
+	ut_ad(type);
+	return(type->mbminlen);
+}
+/*************************************************************************
+Gets the maximum length of a character, in bytes. */
+UNIV_INLINE
+ulint
+dtype_get_mbmaxlen(
+/*===============*/
+				/* out: maximum length of a char, in bytes,
+				or 0 if this is not a character type */
+	const dtype_t*	type)	/* in: type */
+{
+	ut_ad(type);
+	return(type->mbmaxlen);
+}
+
+/*************************************************************************
+Gets the padding character code for a type. */
+UNIV_INLINE
+ulint
+dtype_get_pad_char(
+/*===============*/
+				/* out: padding character code, or
+				ULINT_UNDEFINED if no padding specified */
+	ulint	mtype,		/* in: main type */
+	ulint	prtype)		/* in: precise type */
+{
+	switch (mtype) {
+	case DATA_FIXBINARY:
+	case DATA_BINARY:
+		if (UNIV_UNLIKELY(dtype_get_charset_coll(prtype)
+				  == DATA_MYSQL_BINARY_CHARSET_COLL)) {
+			/* Starting from 5.0.18, do not pad
+			VARBINARY or BINARY columns. */
+			return(ULINT_UNDEFINED);
+		}
+		/* Fall through */
+	case DATA_CHAR:
+	case DATA_VARCHAR:
+	case DATA_MYSQL:
+	case DATA_VARMYSQL:
+		/* Space is the padding character for all char and binary
+		strings, and starting from 5.0.3, also for TEXT strings. */
+
+		return(0x20);
+	case DATA_BLOB:
+		if (!(prtype & DATA_BINARY_TYPE)) {
+			return(0x20);
+		}
+		/* Fall through */
+	default:
+		/* No padding specified */
+		return(ULINT_UNDEFINED);
+	}
+}
+
+/**************************************************************************
+Stores for a type the information which determines its alphabetical ordering
+and the storage size of an SQL NULL value. This is the >= 4.1.x storage
+format. */
+UNIV_INLINE
+void
+dtype_new_store_for_order_and_null_size(
+/*====================================*/
+	byte*		buf,	/* in: buffer for
+				DATA_NEW_ORDER_NULL_TYPE_BUF_SIZE
+				bytes where we store the info */
+	const dtype_t*	type,	/* in: type struct */
+	ulint		prefix_len)/* in: prefix length to
+				replace type->len, or 0 */
+{
+#if 6 != DATA_NEW_ORDER_NULL_TYPE_BUF_SIZE
+#error "6 != DATA_NEW_ORDER_NULL_TYPE_BUF_SIZE"
+#endif
+	ulint	len;
+
+	buf[0] = (byte)(type->mtype & 0xFFUL);
+
+	if (type->prtype & DATA_BINARY_TYPE) {
+		buf[0] = buf[0] | 128;
+	}
+
+	/* In versions < 4.1.2 we had:	if (type->prtype & DATA_NONLATIN1) {
+	buf[0] = buf[0] | 64;
+	}
+	*/
+
+	buf[1] = (byte)(type->prtype & 0xFFUL);
+
+	len = prefix_len ? prefix_len : type->len;
+
+	mach_write_to_2(buf + 2, len & 0xFFFFUL);
+
+	ut_ad(dtype_get_charset_coll(type->prtype) < 256);
+	mach_write_to_2(buf + 4, dtype_get_charset_coll(type->prtype));
+
+	if (type->prtype & DATA_NOT_NULL) {
+		buf[4] |= 128;
+	}
+}
+
+/**************************************************************************
+Reads to a type the stored information which determines its alphabetical
+ordering and the storage size of an SQL NULL value. This is the < 4.1.x
+storage format. */
+UNIV_INLINE
+void
+dtype_read_for_order_and_null_size(
+/*===============================*/
+	dtype_t*	type,	/* in: type struct */
+	const byte*	buf)	/* in: buffer for stored type order info */
+{
+#if 4 != DATA_ORDER_NULL_TYPE_BUF_SIZE
+# error "4 != DATA_ORDER_NULL_TYPE_BUF_SIZE"
+#endif
+
+	type->mtype = buf[0] & 63;
+	type->prtype = buf[1];
+
+	if (buf[0] & 128) {
+		type->prtype = type->prtype | DATA_BINARY_TYPE;
+	}
+
+	type->len = mach_read_from_2(buf + 2);
+
+	type->prtype = dtype_form_prtype(type->prtype,
+					 data_mysql_default_charset_coll);
+	dtype_set_mblen(type);
+}
+
+/**************************************************************************
+Reads to a type the stored information which determines its alphabetical
+ordering and the storage size of an SQL NULL value. This is the >= 4.1.x
+storage format. */
+UNIV_INLINE
+void
+dtype_new_read_for_order_and_null_size(
+/*===================================*/
+	dtype_t*	type,	/* in: type struct */
+	const byte*	buf)	/* in: buffer for stored type order info */
+{
+	ulint	charset_coll;
+
+#if 6 != DATA_NEW_ORDER_NULL_TYPE_BUF_SIZE
+#error "6 != DATA_NEW_ORDER_NULL_TYPE_BUF_SIZE"
+#endif
+
+	type->mtype = buf[0] & 63;
+	type->prtype = buf[1];
+
+	if (buf[0] & 128) {
+		type->prtype |= DATA_BINARY_TYPE;
+	}
+
+	if (buf[4] & 128) {
+		type->prtype |= DATA_NOT_NULL;
+	}
+
+	type->len = mach_read_from_2(buf + 2);
+
+	charset_coll = mach_read_from_2(buf + 4) & 0x7fff;
+
+	if (dtype_is_string_type(type->mtype)) {
+		ut_a(charset_coll < 256);
+
+		if (charset_coll == 0) {
+			/* This insert buffer record was inserted with MySQL
+			version < 4.1.2, and the charset-collation code was not
+			explicitly stored to dtype->prtype at that time. It
+			must be the default charset-collation of this MySQL
+			installation. */
+
+			charset_coll = data_mysql_default_charset_coll;
+		}
+
+		type->prtype = dtype_form_prtype(type->prtype, charset_coll);
+	}
+	dtype_set_mblen(type);
+}
+
+/***************************************************************************
+Returns the size of a fixed size data type, 0 if not a fixed size type. */
+UNIV_INLINE
+ulint
+dtype_get_fixed_size_low(
+/*=====================*/
+				/* out: fixed size, or 0 */
+	ulint	mtype,		/* in: main type */
+	ulint	prtype,		/* in: precise type */
+	ulint	len,		/* in: length */
+	ulint	mbminlen,	/* in: minimum length of a multibyte char */
+	ulint	mbmaxlen)	/* in: maximum length of a multibyte char */
+{
+	switch (mtype) {
+	case DATA_SYS:
+#ifdef UNIV_DEBUG
+		switch (prtype & DATA_MYSQL_TYPE_MASK) {
+		case DATA_ROW_ID:
+			ut_ad(len == DATA_ROW_ID_LEN);
+			break;
+		case DATA_TRX_ID:
+			ut_ad(len == DATA_TRX_ID_LEN);
+			break;
+		case DATA_ROLL_PTR:
+			ut_ad(len == DATA_ROLL_PTR_LEN);
+			break;
+		default:
+			ut_ad(0);
+			return(0);
+		}
+#endif /* UNIV_DEBUG */
+	case DATA_CHAR:
+	case DATA_FIXBINARY:
+	case DATA_INT:
+	case DATA_FLOAT:
+	case DATA_DOUBLE:
+		return(len);
+	case DATA_MYSQL:
+		if (prtype & DATA_BINARY_TYPE) {
+			return(len);
+		} else {
+#ifdef UNIV_HOTBACKUP
+			if (mbminlen == mbmaxlen) {
+				return(len);
+			}
+#else /* UNIV_HOTBACKUP */
+			/* We play it safe here and ask MySQL for
+			mbminlen and mbmaxlen.	Although
+			mbminlen and mbmaxlen are
+			initialized if and only if prtype
+			is (in one of the 3 functions in this file),
+			it could be that none of these functions
+			has been called. */
+
+			ulint	i_mbminlen, i_mbmaxlen;
+
+			innobase_get_cset_width(
+				dtype_get_charset_coll(prtype),
+				&i_mbminlen, &i_mbmaxlen);
+
+			if (UNIV_UNLIKELY(mbminlen != i_mbminlen)
+			    || UNIV_UNLIKELY(mbmaxlen != i_mbmaxlen)) {
+
+				ut_print_timestamp(stderr);
+				fprintf(stderr, "  InnoDB: "
+					"mbminlen=%lu, "
+					"mbmaxlen=%lu, "
+					"type->mbminlen=%lu, "
+					"type->mbmaxlen=%lu\n",
+					(ulong) i_mbminlen,
+					(ulong) i_mbmaxlen,
+					(ulong) mbminlen,
+					(ulong) mbmaxlen);
+			}
+			if (mbminlen == mbmaxlen) {
+				return(len);
+			}
+#endif /* !UNIV_HOTBACKUP */
+		}
+		/* fall through for variable-length charsets */
+	case DATA_VARCHAR:
+	case DATA_BINARY:
+	case DATA_DECIMAL:
+	case DATA_VARMYSQL:
+	case DATA_BLOB:
+		return(0);
+	default:
+		ut_error;
+	}
+
+	return(0);
+}
+
+/***************************************************************************
+Returns the minimum size of a data type. */
+UNIV_INLINE
+ulint
+dtype_get_min_size_low(
+/*===================*/
+				/* out: minimum size */
+	ulint	mtype,		/* in: main type */
+	ulint	prtype,		/* in: precise type */
+	ulint	len,		/* in: length */
+	ulint	mbminlen,	/* in: minimum length of a multibyte char */
+	ulint	mbmaxlen)	/* in: maximum length of a multibyte char */
+{
+	switch (mtype) {
+	case DATA_SYS:
+#ifdef UNIV_DEBUG
+		switch (prtype & DATA_MYSQL_TYPE_MASK) {
+		case DATA_ROW_ID:
+			ut_ad(len == DATA_ROW_ID_LEN);
+			break;
+		case DATA_TRX_ID:
+			ut_ad(len == DATA_TRX_ID_LEN);
+			break;
+		case DATA_ROLL_PTR:
+			ut_ad(len == DATA_ROLL_PTR_LEN);
+			break;
+		default:
+			ut_ad(0);
+			return(0);
+		}
+#endif /* UNIV_DEBUG */
+	case DATA_CHAR:
+	case DATA_FIXBINARY:
+	case DATA_INT:
+	case DATA_FLOAT:
+	case DATA_DOUBLE:
+		return(len);
+	case DATA_MYSQL:
+		if ((prtype & DATA_BINARY_TYPE) || mbminlen == mbmaxlen) {
+			return(len);
+		}
+		/* this is a variable-length character set */
+		ut_a(mbminlen > 0);
+		ut_a(mbmaxlen > mbminlen);
+		ut_a(len % mbmaxlen == 0);
+		return(len * mbminlen / mbmaxlen);
+	case DATA_VARCHAR:
+	case DATA_BINARY:
+	case DATA_DECIMAL:
+	case DATA_VARMYSQL:
+	case DATA_BLOB:
+		return(0);
+	default:
+		ut_error;
+	}
+
+	return(0);
+}
+
+/***************************************************************************
+Returns the maximum size of a data type. Note: types in system tables may be
+incomplete and return incorrect information. */
+UNIV_INLINE
+ulint
+dtype_get_max_size_low(
+/*===================*/
+				/* out: maximum size */
+	ulint	mtype,		/* in: main type */
+	ulint	len)		/* in: length */
+{
+	switch (mtype) {
+	case DATA_SYS:
+	case DATA_CHAR:
+	case DATA_FIXBINARY:
+	case DATA_INT:
+	case DATA_FLOAT:
+	case DATA_DOUBLE:
+	case DATA_MYSQL:
+	case DATA_VARCHAR:
+	case DATA_BINARY:
+	case DATA_DECIMAL:
+	case DATA_VARMYSQL:
+		return(len);
+	case DATA_BLOB:
+		break;
+	default:
+		ut_error;
+	}
+
+	return(ULINT_MAX);
+}
+
+/***************************************************************************
+Returns the ROW_FORMAT=REDUNDANT stored SQL NULL size of a type.
+For fixed length types it is the fixed length of the type, otherwise 0. */
+UNIV_INLINE
+ulint
+dtype_get_sql_null_size(
+/*====================*/
+				/* out: SQL null storage size
+				in ROW_FORMAT=REDUNDANT */
+	const dtype_t*	type)	/* in: type */
+{
+	return(dtype_get_fixed_size_low(type->mtype, type->prtype, type->len,
+					type->mbminlen, type->mbmaxlen));
+}
diff --git a/storage/xtradb/include/data0types.h b/storage/xtradb/include/data0types.h
new file mode 100644
index 00000000000..9e536478d68
--- /dev/null
+++ b/storage/xtradb/include/data0types.h
@@ -0,0 +1,35 @@
+/*****************************************************************************
+
+Copyright (c) 2000, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/************************************************************************
+Some type definitions
+
+Created 9/21/2000 Heikki Tuuri
+*************************************************************************/
+
+#ifndef data0types_h
+#define data0types_h
+
+/* SQL data field struct */
+typedef struct dfield_struct	dfield_t;
+
+/* SQL data tuple struct */
+typedef struct dtuple_struct	dtuple_t;
+
+#endif
+
diff --git a/storage/xtradb/include/db0err.h b/storage/xtradb/include/db0err.h
new file mode 100644
index 00000000000..d6d2a9785a5
--- /dev/null
+++ b/storage/xtradb/include/db0err.h
@@ -0,0 +1,104 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Global error codes for the database
+
+Created 5/24/1996 Heikki Tuuri
+*******************************************************/
+
+#ifndef db0err_h
+#define db0err_h
+
+
+enum db_err {
+	DB_SUCCESS = 10,
+
+	/* The following are error codes */
+	DB_ERROR,
+	DB_OUT_OF_MEMORY,
+	DB_OUT_OF_FILE_SPACE,
+	DB_LOCK_WAIT,
+	DB_DEADLOCK,
+	DB_ROLLBACK,
+	DB_DUPLICATE_KEY,
+	DB_QUE_THR_SUSPENDED,
+	DB_MISSING_HISTORY,		/* required history data has been
+					deleted due to lack of space in
+					rollback segment */
+	DB_CLUSTER_NOT_FOUND = 30,
+	DB_TABLE_NOT_FOUND,
+	DB_MUST_GET_MORE_FILE_SPACE,	/* the database has to be stopped
+					and restarted with more file space */
+	DB_TABLE_IS_BEING_USED,
+	DB_TOO_BIG_RECORD,		/* a record in an index would not fit
+					on a compressed page, or it would
+					become bigger than 1/2 free space in
+					an uncompressed page frame */
+	DB_LOCK_WAIT_TIMEOUT,		/* lock wait lasted too long */
+	DB_NO_REFERENCED_ROW,		/* referenced key value not found
+					for a foreign key in an insert or
+					update of a row */
+	DB_ROW_IS_REFERENCED,		/* cannot delete or update a row
+					because it contains a key value
+					which is referenced */
+	DB_CANNOT_ADD_CONSTRAINT,	/* adding a foreign key constraint
+					to a table failed */
+	DB_CORRUPTION,			/* data structure corruption noticed */
+	DB_COL_APPEARS_TWICE_IN_INDEX,	/* InnoDB cannot handle an index
+					where same column appears twice */
+	DB_CANNOT_DROP_CONSTRAINT,	/* dropping a foreign key constraint
+					from a table failed */
+	DB_NO_SAVEPOINT,		/* no savepoint exists with the given
+					name */
+	DB_TABLESPACE_ALREADY_EXISTS,	/* we cannot create a new single-table
+					tablespace because a file of the same
+					name already exists */
+	DB_TABLESPACE_DELETED,		/* tablespace does not exist or is
+					being dropped right now */
+	DB_LOCK_TABLE_FULL,		/* lock structs have exhausted the
+					buffer pool (for big transactions,
+					InnoDB stores the lock structs in the
+					buffer pool) */
+	DB_FOREIGN_DUPLICATE_KEY,	/* foreign key constraints
+					activated by the operation would
+					lead to a duplicate key in some
+					table */
+	DB_TOO_MANY_CONCURRENT_TRXS,	/* when InnoDB runs out of the
+					preconfigured undo slots, this can
+					only happen when there are too many
+					concurrent transactions */
+	DB_UNSUPPORTED,			/* when InnoDB sees any artefact or
+					a feature that it can't recoginize or
+					work with e.g., FT indexes created by
+					a later version of the engine. */
+
+	DB_PRIMARY_KEY_IS_NULL,		/* a column in the PRIMARY KEY
+					was found to be NULL */
+
+	/* The following are partial failure codes */
+	DB_FAIL = 1000,
+	DB_OVERFLOW,
+	DB_UNDERFLOW,
+	DB_STRONG_FAIL,
+	DB_ZIP_OVERFLOW,
+	DB_RECORD_NOT_FOUND = 1500,
+	DB_END_OF_INDEX
+};
+
+#endif
diff --git a/storage/xtradb/include/dict0boot.h b/storage/xtradb/include/dict0boot.h
new file mode 100644
index 00000000000..e1556bdb16e
--- /dev/null
+++ b/storage/xtradb/include/dict0boot.h
@@ -0,0 +1,150 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Data dictionary creation and booting
+
+Created 4/18/1996 Heikki Tuuri
+*******************************************************/
+
+#ifndef dict0boot_h
+#define dict0boot_h
+
+#include "univ.i"
+
+#include "mtr0mtr.h"
+#include "mtr0log.h"
+#include "ut0byte.h"
+#include "buf0buf.h"
+#include "fsp0fsp.h"
+#include "dict0dict.h"
+
+typedef	byte	dict_hdr_t;
+
+/**************************************************************************
+Gets a pointer to the dictionary header and x-latches its page. */
+UNIV_INTERN
+dict_hdr_t*
+dict_hdr_get(
+/*=========*/
+			/* out: pointer to the dictionary header,
+			page x-latched */
+	mtr_t*	mtr);	/* in: mtr */
+/**************************************************************************
+Returns a new row, table, index, or tree id. */
+UNIV_INTERN
+dulint
+dict_hdr_get_new_id(
+/*================*/
+			/* out: the new id */
+	ulint	type);	/* in: DICT_HDR_ROW_ID, ... */
+/**************************************************************************
+Returns a new row id. */
+UNIV_INLINE
+dulint
+dict_sys_get_new_row_id(void);
+/*=========================*/
+			/* out: the new id */
+/**************************************************************************
+Reads a row id from a record or other 6-byte stored form. */
+UNIV_INLINE
+dulint
+dict_sys_read_row_id(
+/*=================*/
+			/* out: row id */
+	byte*	field);	/* in: record field */
+/**************************************************************************
+Writes a row id to a record or other 6-byte stored form. */
+UNIV_INLINE
+void
+dict_sys_write_row_id(
+/*==================*/
+	byte*	field,	/* in: record field */
+	dulint	row_id);/* in: row id */
+/*********************************************************************
+Initializes the data dictionary memory structures when the database is
+started. This function is also called when the data dictionary is created. */
+UNIV_INTERN
+void
+dict_boot(void);
+/*===========*/
+/*********************************************************************
+Creates and initializes the data dictionary at the database creation. */
+UNIV_INTERN
+void
+dict_create(void);
+/*=============*/
+
+
+/* Space id and page no where the dictionary header resides */
+#define	DICT_HDR_SPACE		0	/* the SYSTEM tablespace */
+#define	DICT_HDR_PAGE_NO	FSP_DICT_HDR_PAGE_NO
+
+/* The ids for the basic system tables and their indexes */
+#define DICT_TABLES_ID		ut_dulint_create(0, 1)
+#define DICT_COLUMNS_ID		ut_dulint_create(0, 2)
+#define DICT_INDEXES_ID		ut_dulint_create(0, 3)
+#define DICT_FIELDS_ID		ut_dulint_create(0, 4)
+/* The following is a secondary index on SYS_TABLES */
+#define DICT_TABLE_IDS_ID	ut_dulint_create(0, 5)
+
+#define	DICT_HDR_FIRST_ID	10	/* the ids for tables etc. start
+					from this number, except for basic
+					system tables and their above defined
+					indexes; ibuf tables and indexes are
+					assigned as the id the number
+					DICT_IBUF_ID_MIN plus the space id */
+#define DICT_IBUF_ID_MIN	ut_dulint_create(0xFFFFFFFFUL, 0)
+
+/* The offset of the dictionary header on the page */
+#define	DICT_HDR		FSEG_PAGE_DATA
+
+/*-------------------------------------------------------------*/
+/* Dictionary header offsets */
+#define DICT_HDR_ROW_ID		0	/* The latest assigned row id */
+#define	DICT_HDR_TABLE_ID	8	/* The latest assigned table id */
+#define	DICT_HDR_INDEX_ID	16	/* The latest assigned index id */
+#define	DICT_HDR_MIX_ID		24	/* Obsolete, always 0. */
+#define	DICT_HDR_TABLES		32	/* Root of the table index tree */
+#define	DICT_HDR_TABLE_IDS	36	/* Root of the table index tree */
+#define	DICT_HDR_COLUMNS	40	/* Root of the column index tree */
+#define	DICT_HDR_INDEXES	44	/* Root of the index index tree */
+#define	DICT_HDR_FIELDS		48	/* Root of the index field
+					index tree */
+
+#define DICT_HDR_FSEG_HEADER	56	/* Segment header for the tablespace
+					segment into which the dictionary
+					header is created */
+/*-------------------------------------------------------------*/
+
+/* The field number of the page number field in the sys_indexes table
+clustered index */
+#define DICT_SYS_INDEXES_PAGE_NO_FIELD	 8
+#define DICT_SYS_INDEXES_SPACE_NO_FIELD	 7
+#define DICT_SYS_INDEXES_TYPE_FIELD	 6
+
+/* When a row id which is zero modulo this number (which must be a power of
+two) is assigned, the field DICT_HDR_ROW_ID on the dictionary header page is
+updated */
+#define DICT_HDR_ROW_ID_WRITE_MARGIN	256
+
+#ifndef UNIV_NONINL
+#include "dict0boot.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/dict0boot.ic b/storage/xtradb/include/dict0boot.ic
new file mode 100644
index 00000000000..9b45f9e84be
--- /dev/null
+++ b/storage/xtradb/include/dict0boot.ic
@@ -0,0 +1,92 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Data dictionary creation and booting
+
+Created 4/18/1996 Heikki Tuuri
+*******************************************************/
+
+/**************************************************************************
+Writes the current value of the row id counter to the dictionary header file
+page. */
+UNIV_INTERN
+void
+dict_hdr_flush_row_id(void);
+/*=======================*/
+
+
+/**************************************************************************
+Returns a new row id. */
+UNIV_INLINE
+dulint
+dict_sys_get_new_row_id(void)
+/*=========================*/
+			/* out: the new id */
+{
+	dulint	id;
+
+	mutex_enter(&(dict_sys->mutex));
+
+	id = dict_sys->row_id;
+
+	if (0 == (ut_dulint_get_low(id) % DICT_HDR_ROW_ID_WRITE_MARGIN)) {
+
+		dict_hdr_flush_row_id();
+	}
+
+	UT_DULINT_INC(dict_sys->row_id);
+
+	mutex_exit(&(dict_sys->mutex));
+
+	return(id);
+}
+
+/**************************************************************************
+Reads a row id from a record or other 6-byte stored form. */
+UNIV_INLINE
+dulint
+dict_sys_read_row_id(
+/*=================*/
+			/* out: row id */
+	byte*	field)	/* in: record field */
+{
+#if DATA_ROW_ID_LEN != 6
+# error "DATA_ROW_ID_LEN != 6"
+#endif
+
+	return(mach_read_from_6(field));
+}
+
+/**************************************************************************
+Writes a row id to a record or other 6-byte stored form. */
+UNIV_INLINE
+void
+dict_sys_write_row_id(
+/*==================*/
+	byte*	field,	/* in: record field */
+	dulint	row_id)	/* in: row id */
+{
+#if DATA_ROW_ID_LEN != 6
+# error "DATA_ROW_ID_LEN != 6"
+#endif
+
+	mach_write_to_6(field, row_id);
+}
+
+
diff --git a/storage/xtradb/include/dict0crea.h b/storage/xtradb/include/dict0crea.h
new file mode 100644
index 00000000000..9ac3e408f1f
--- /dev/null
+++ b/storage/xtradb/include/dict0crea.h
@@ -0,0 +1,199 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Database object creation
+
+Created 1/8/1996 Heikki Tuuri
+*******************************************************/
+
+#ifndef dict0crea_h
+#define dict0crea_h
+
+#include "univ.i"
+#include "dict0types.h"
+#include "dict0dict.h"
+#include "que0types.h"
+#include "row0types.h"
+#include "mtr0mtr.h"
+
+/*************************************************************************
+Creates a table create graph. */
+UNIV_INTERN
+tab_node_t*
+tab_create_graph_create(
+/*====================*/
+				/* out, own: table create node */
+	dict_table_t*	table,	/* in: table to create, built as a memory data
+				structure */
+	mem_heap_t*	heap);	/* in: heap where created */
+/*************************************************************************
+Creates an index create graph. */
+UNIV_INTERN
+ind_node_t*
+ind_create_graph_create(
+/*====================*/
+				/* out, own: index create node */
+	dict_index_t*	index,	/* in: index to create, built as a memory data
+				structure */
+	mem_heap_t*	heap);	/* in: heap where created */
+/***************************************************************
+Creates a table. This is a high-level function used in SQL execution graphs. */
+UNIV_INTERN
+que_thr_t*
+dict_create_table_step(
+/*===================*/
+				/* out: query thread to run next or NULL */
+	que_thr_t*	thr);	/* in: query thread */
+/***************************************************************
+Creates an index. This is a high-level function used in SQL execution
+graphs. */
+UNIV_INTERN
+que_thr_t*
+dict_create_index_step(
+/*===================*/
+				/* out: query thread to run next or NULL */
+	que_thr_t*	thr);	/* in: query thread */
+/***********************************************************************
+Truncates the index tree associated with a row in SYS_INDEXES table. */
+UNIV_INTERN
+ulint
+dict_truncate_index_tree(
+/*=====================*/
+				/* out: new root page number, or
+				FIL_NULL on failure */
+	dict_table_t*	table,	/* in: the table the index belongs to */
+	ulint		space,	/* in: 0=truncate,
+				nonzero=create the index tree in the
+				given tablespace */
+	btr_pcur_t*	pcur,	/* in/out: persistent cursor pointing to
+				record in the clustered index of
+				SYS_INDEXES table. The cursor may be
+				repositioned in this call. */
+	mtr_t*		mtr);	/* in: mtr having the latch
+				on the record page. The mtr may be
+				committed and restarted in this call. */
+/***********************************************************************
+Drops the index tree associated with a row in SYS_INDEXES table. */
+UNIV_INTERN
+void
+dict_drop_index_tree(
+/*=================*/
+	rec_t*	rec,	/* in/out: record in the clustered index
+			of SYS_INDEXES table */
+	mtr_t*	mtr);	/* in: mtr having the latch on the record page */
+#ifndef UNIV_HOTBACKUP
+/********************************************************************
+Creates the foreign key constraints system tables inside InnoDB
+at database creation or database start if they are not found or are
+not of the right form. */
+UNIV_INTERN
+ulint
+dict_create_or_check_foreign_constraint_tables(void);
+/*================================================*/
+				/* out: DB_SUCCESS or error code */
+/************************************************************************
+Adds foreign key definitions to data dictionary tables in the database. We
+look at table->foreign_list, and also generate names to constraints that were
+not named by the user. A generated constraint has a name of the format
+databasename/tablename_ibfk_<number>, where the numbers start from 1, and are
+given locally for this table, that is, the number is not global, as in the
+old format constraints < 4.0.18 it used to be. */
+UNIV_INTERN
+ulint
+dict_create_add_foreigns_to_dictionary(
+/*===================================*/
+				/* out: error code or DB_SUCCESS */
+	ulint		start_id,/* in: if we are actually doing ALTER TABLE
+				ADD CONSTRAINT, we want to generate constraint
+				numbers which are bigger than in the table so
+				far; we number the constraints from
+				start_id + 1 up; start_id should be set to 0 if
+				we are creating a new table, or if the table
+				so far has no constraints for which the name
+				was generated here */
+	dict_table_t*	table,	/* in: table */
+	trx_t*		trx);	/* in: transaction */
+#endif /* !UNIV_HOTBACKUP */
+
+/* Table create node structure */
+
+struct tab_node_struct{
+	que_common_t	common;	/* node type: QUE_NODE_TABLE_CREATE */
+	dict_table_t*	table;	/* table to create, built as a memory data
+				structure with dict_mem_... functions */
+	ins_node_t*	tab_def; /* child node which does the insert of
+				the table definition; the row to be inserted
+				is built by the parent node  */
+	ins_node_t*	col_def; /* child node which does the inserts of
+				the column definitions; the row to be inserted
+				is built by the parent node  */
+	commit_node_t*	commit_node;
+				/* child node which performs a commit after
+				a successful table creation */
+	/*----------------------*/
+	/* Local storage for this graph node */
+	ulint		state;	/* node execution state */
+	ulint		col_no;	/* next column definition to insert */
+	mem_heap_t*	heap;	/* memory heap used as auxiliary storage */
+};
+
+/* Table create node states */
+#define	TABLE_BUILD_TABLE_DEF	1
+#define	TABLE_BUILD_COL_DEF	2
+#define	TABLE_COMMIT_WORK	3
+#define	TABLE_ADD_TO_CACHE	4
+#define	TABLE_COMPLETED		5
+
+/* Index create node struct */
+
+struct ind_node_struct{
+	que_common_t	common;	/* node type: QUE_NODE_INDEX_CREATE */
+	dict_index_t*	index;	/* index to create, built as a memory data
+				structure with dict_mem_... functions */
+	ins_node_t*	ind_def; /* child node which does the insert of
+				the index definition; the row to be inserted
+				is built by the parent node  */
+	ins_node_t*	field_def; /* child node which does the inserts of
+				the field definitions; the row to be inserted
+				is built by the parent node  */
+	commit_node_t*	commit_node;
+				/* child node which performs a commit after
+				a successful index creation */
+	/*----------------------*/
+	/* Local storage for this graph node */
+	ulint		state;	/* node execution state */
+	ulint		page_no;/* root page number of the index */
+	dict_table_t*	table;	/* table which owns the index */
+	dtuple_t*	ind_row;/* index definition row built */
+	ulint		field_no;/* next field definition to insert */
+	mem_heap_t*	heap;	/* memory heap used as auxiliary storage */
+};
+
+/* Index create node states */
+#define	INDEX_BUILD_INDEX_DEF	1
+#define	INDEX_BUILD_FIELD_DEF	2
+#define	INDEX_CREATE_INDEX_TREE	3
+#define	INDEX_COMMIT_WORK	4
+#define	INDEX_ADD_TO_CACHE	5
+
+#ifndef UNIV_NONINL
+#include "dict0crea.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/dict0crea.ic b/storage/xtradb/include/dict0crea.ic
new file mode 100644
index 00000000000..b05385fa121
--- /dev/null
+++ b/storage/xtradb/include/dict0crea.ic
@@ -0,0 +1,24 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Database object creation
+
+Created 1/8/1996 Heikki Tuuri
+*******************************************************/
+
diff --git a/storage/xtradb/include/dict0dict.h b/storage/xtradb/include/dict0dict.h
new file mode 100644
index 00000000000..82a139a7ff9
--- /dev/null
+++ b/storage/xtradb/include/dict0dict.h
@@ -0,0 +1,1147 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Data dictionary system
+
+Created 1/8/1996 Heikki Tuuri
+*******************************************************/
+
+#ifndef dict0dict_h
+#define dict0dict_h
+
+#include "univ.i"
+#include "dict0types.h"
+#include "dict0mem.h"
+#include "data0type.h"
+#include "data0data.h"
+#include "sync0sync.h"
+#include "sync0rw.h"
+#include "mem0mem.h"
+#include "rem0types.h"
+#include "ut0mem.h"
+#include "ut0lst.h"
+#include "hash0hash.h"
+#include "ut0rnd.h"
+#include "ut0byte.h"
+#include "trx0types.h"
+
+#ifndef UNIV_HOTBACKUP
+/**********************************************************************
+Makes all characters in a NUL-terminated UTF-8 string lower case. */
+UNIV_INTERN
+void
+dict_casedn_str(
+/*============*/
+	char*	a);	/* in/out: string to put in lower case */
+#endif /* !UNIV_HOTBACKUP */
+/************************************************************************
+Get the database name length in a table name. */
+UNIV_INTERN
+ulint
+dict_get_db_name_len(
+/*=================*/
+				/* out: database name length */
+	const char*	name);	/* in: table name in the form
+				dbname '/' tablename */
+/************************************************************************
+Return the end of table name where we have removed dbname and '/'. */
+
+const char*
+dict_remove_db_name(
+/*================*/
+				/* out: table name */
+	const char*	name);	/* in: table name in the form
+				dbname '/' tablename */
+/**************************************************************************
+Returns a table object based on table id. */
+UNIV_INTERN
+dict_table_t*
+dict_table_get_on_id(
+/*=================*/
+                                /* out: table, NULL if does not exist */
+        dulint  table_id,       /* in: table id */
+        trx_t*  trx);           /* in: transaction handle */
+/************************************************************************
+Decrements the count of open MySQL handles to a table. */
+UNIV_INTERN
+void
+dict_table_decrement_handle_count(
+/*==============================*/
+	dict_table_t*	table,		/* in/out: table */
+	ibool		dict_locked);	/* in: TRUE=data dictionary locked */
+/**************************************************************************
+Inits the data dictionary module. */
+UNIV_INTERN
+void
+dict_init(void);
+/*===========*/
+/************************************************************************
+Gets the space id of every table of the data dictionary and makes a linear
+list and a hash table of them to the data dictionary cache. This function
+can be called at database startup if we did not need to do a crash recovery.
+In crash recovery we must scan the space id's from the .ibd files in MySQL
+database directories. */
+UNIV_INTERN
+void
+dict_load_space_id_list(void);
+/*=========================*/
+/*************************************************************************
+Gets the column data type. */
+UNIV_INLINE
+void
+dict_col_copy_type(
+/*===============*/
+	const dict_col_t*	col,	/* in: column */
+	dtype_t*		type);	/* out: data type */
+#ifdef UNIV_DEBUG
+/*************************************************************************
+Assert that a column and a data type match. */
+UNIV_INLINE
+ibool
+dict_col_type_assert_equal(
+/*=======================*/
+					/* out: TRUE */
+	const dict_col_t*	col,	/* in: column */
+	const dtype_t*		type);	/* in: data type */
+#endif /* UNIV_DEBUG */
+/***************************************************************************
+Returns the minimum size of the column. */
+UNIV_INLINE
+ulint
+dict_col_get_min_size(
+/*==================*/
+					/* out: minimum size */
+	const dict_col_t*	col);	/* in: column */
+/***************************************************************************
+Returns the maximum size of the column. */
+UNIV_INLINE
+ulint
+dict_col_get_max_size(
+/*==================*/
+					/* out: maximum size */
+	const dict_col_t*	col);	/* in: column */
+/***************************************************************************
+Returns the size of a fixed size column, 0 if not a fixed size column. */
+UNIV_INLINE
+ulint
+dict_col_get_fixed_size(
+/*====================*/
+					/* out: fixed size, or 0 */
+	const dict_col_t*	col);	/* in: column */
+/***************************************************************************
+Returns the ROW_FORMAT=REDUNDANT stored SQL NULL size of a column.
+For fixed length types it is the fixed length of the type, otherwise 0. */
+UNIV_INLINE
+ulint
+dict_col_get_sql_null_size(
+/*=======================*/
+					/* out: SQL null storage size
+					in ROW_FORMAT=REDUNDANT */
+	const dict_col_t*	col);	/* in: column */
+
+/*************************************************************************
+Gets the column number. */
+UNIV_INLINE
+ulint
+dict_col_get_no(
+/*============*/
+	const dict_col_t*	col);
+/*************************************************************************
+Gets the column position in the clustered index. */
+UNIV_INLINE
+ulint
+dict_col_get_clust_pos(
+/*===================*/
+	const dict_col_t*	col,		/* in: table column */
+	const dict_index_t*	clust_index);	/* in: clustered index */
+/********************************************************************
+If the given column name is reserved for InnoDB system columns, return
+TRUE. */
+UNIV_INTERN
+ibool
+dict_col_name_is_reserved(
+/*======================*/
+				/* out: TRUE if name is reserved */
+	const char*	name);	/* in: column name */
+/************************************************************************
+Acquire the autoinc lock.*/
+UNIV_INTERN
+void
+dict_table_autoinc_lock(
+/*====================*/
+	dict_table_t*	table);	/* in/out: table */
+/************************************************************************
+Unconditionally set the autoinc counter. */
+UNIV_INTERN
+void
+dict_table_autoinc_initialize(
+/*==========================*/
+	dict_table_t*	table,	/* in/out: table */
+	ib_uint64_t	value);	/* in: next value to assign to a row */
+/************************************************************************
+Reads the next autoinc value (== autoinc counter value), 0 if not yet
+initialized. */
+UNIV_INTERN
+ib_uint64_t
+dict_table_autoinc_read(
+/*====================*/
+					/* out: value for a new row, or 0 */
+	const dict_table_t*	table);	/* in: table */
+/************************************************************************
+Updates the autoinc counter if the value supplied is greater than the
+current value. */
+UNIV_INTERN
+void
+dict_table_autoinc_update_if_greater(
+/*=================================*/
+
+	dict_table_t*	table,	/* in/out: table */
+	ib_uint64_t	value);	/* in: value which was assigned to a row */
+/************************************************************************
+Release the autoinc lock.*/
+UNIV_INTERN
+void
+dict_table_autoinc_unlock(
+/*======================*/
+	dict_table_t*	table);	/* in/out: table */
+/**************************************************************************
+Adds system columns to a table object. */
+UNIV_INTERN
+void
+dict_table_add_system_columns(
+/*==========================*/
+	dict_table_t*	table,	/* in/out: table */
+	mem_heap_t*	heap);	/* in: temporary heap */
+/**************************************************************************
+Adds a table object to the dictionary cache. */
+UNIV_INTERN
+void
+dict_table_add_to_cache(
+/*====================*/
+	dict_table_t*	table,	/* in: table */
+	mem_heap_t*	heap);	/* in: temporary heap */
+/**************************************************************************
+Removes a table object from the dictionary cache. */
+UNIV_INTERN
+void
+dict_table_remove_from_cache(
+/*=========================*/
+	dict_table_t*	table);	/* in, own: table */
+/**************************************************************************
+Renames a table object. */
+UNIV_INTERN
+ibool
+dict_table_rename_in_cache(
+/*=======================*/
+					/* out: TRUE if success */
+	dict_table_t*	table,		/* in/out: table */
+	const char*	new_name,	/* in: new name */
+	ibool		rename_also_foreigns);/* in: in ALTER TABLE we want
+					to preserve the original table name
+					in constraints which reference it */
+/**************************************************************************
+Removes an index from the dictionary cache. */
+UNIV_INTERN
+void
+dict_index_remove_from_cache(
+/*=========================*/
+	dict_table_t*	table,	/* in/out: table */
+	dict_index_t*	index);	/* in, own: index */
+/**************************************************************************
+Change the id of a table object in the dictionary cache. This is used in
+DISCARD TABLESPACE. */
+UNIV_INTERN
+void
+dict_table_change_id_in_cache(
+/*==========================*/
+	dict_table_t*	table,	/* in/out: table object already in cache */
+	dulint		new_id);/* in: new id to set */
+/**************************************************************************
+Adds a foreign key constraint object to the dictionary cache. May free
+the object if there already is an object with the same identifier in.
+At least one of foreign table or referenced table must already be in
+the dictionary cache! */
+UNIV_INTERN
+ulint
+dict_foreign_add_to_cache(
+/*======================*/
+					/* out: DB_SUCCESS or error code */
+	dict_foreign_t*	foreign,	/* in, own: foreign key constraint */
+	ibool		check_charsets);/* in: TRUE=check charset
+					compatibility */
+/*************************************************************************
+Check if the index is referenced by a foreign key, if TRUE return the
+matching instance NULL otherwise. */
+UNIV_INTERN
+dict_foreign_t*
+dict_table_get_referenced_constraint(
+/*=================================*/
+				/* out: pointer to foreign key struct if index
+				is defined for foreign key, otherwise NULL */
+	dict_table_t*	table,	/* in: InnoDB table */
+	dict_index_t*	index);	/* in: InnoDB index */
+/*************************************************************************
+Checks if a table is referenced by foreign keys. */
+UNIV_INTERN
+ibool
+dict_table_is_referenced_by_foreign_key(
+/*====================================*/
+					/* out: TRUE if table is referenced
+					by a foreign key */
+	const dict_table_t*	table);	/* in: InnoDB table */
+/**************************************************************************
+Replace the index in the foreign key list that matches this index's
+definition with an equivalent index. */
+UNIV_INTERN
+void
+dict_table_replace_index_in_foreign_list(
+/*=====================================*/
+	dict_table_t*	table,  /* in/out: table */
+	dict_index_t*	index);	/* in: index to be replaced */
+/*************************************************************************
+Checks if a index is defined for a foreign key constraint. Index is a part
+of a foreign key constraint if the index is referenced by foreign key
+or index is a foreign key index */
+UNIV_INTERN
+dict_foreign_t*
+dict_table_get_foreign_constraint(
+/*==============================*/
+				/* out: pointer to foreign key struct if index
+				is defined for foreign key, otherwise NULL */
+	dict_table_t*	table,	/* in: InnoDB table */
+	dict_index_t*	index);	/* in: InnoDB index */
+/*************************************************************************
+Scans a table create SQL string and adds to the data dictionary
+the foreign key constraints declared in the string. This function
+should be called after the indexes for a table have been created.
+Each foreign key constraint must be accompanied with indexes in
+bot participating tables. The indexes are allowed to contain more
+fields than mentioned in the constraint. */
+UNIV_INTERN
+ulint
+dict_create_foreign_constraints(
+/*============================*/
+					/* out: error code or DB_SUCCESS */
+	trx_t*		trx,		/* in: transaction */
+	const char*	sql_string,	/* in: table create statement where
+					foreign keys are declared like:
+					FOREIGN KEY (a, b) REFERENCES
+					table2(c, d), table2 can be written
+					also with the database
+					name before it: test.table2; the
+					default database id the database of
+					parameter name */
+	const char*	name,		/* in: table full name in the
+					normalized form
+					database_name/table_name */
+	ibool		reject_fks);	/* in: if TRUE, fail with error
+					code DB_CANNOT_ADD_CONSTRAINT if
+					any foreign keys are found. */
+/**************************************************************************
+Parses the CONSTRAINT id's to be dropped in an ALTER TABLE statement. */
+UNIV_INTERN
+ulint
+dict_foreign_parse_drop_constraints(
+/*================================*/
+						/* out: DB_SUCCESS or
+						DB_CANNOT_DROP_CONSTRAINT if
+						syntax error or the constraint
+						id does not match */
+	mem_heap_t*	heap,			/* in: heap from which we can
+						allocate memory */
+	trx_t*		trx,			/* in: transaction */
+	dict_table_t*	table,			/* in: table */
+	ulint*		n,			/* out: number of constraints
+						to drop */
+	const char***	constraints_to_drop);	/* out: id's of the
+						constraints to drop */
+/**************************************************************************
+Returns a table object and optionally increment its MySQL open handle count.
+NOTE! This is a high-level function to be used mainly from outside the
+'dict' directory. Inside this directory dict_table_get_low is usually the
+appropriate function. */
+UNIV_INTERN
+dict_table_t*
+dict_table_get(
+/*===========*/
+					/* out: table, NULL if
+					does not exist */
+	const char*	table_name,	/* in: table name */
+	ibool		inc_mysql_count);
+					/* in: whether to increment the open
+					handle count on the table */
+/**************************************************************************
+Returns a index object, based on table and index id, and memoryfixes it. */
+UNIV_INTERN
+dict_index_t*
+dict_index_get_on_id_low(
+/*=====================*/
+					/* out: index, NULL if does not
+					exist */
+	dict_table_t*	table,		/* in: table */
+	dulint		index_id);	/* in: index id */
+/**************************************************************************
+Checks if a table is in the dictionary cache. */
+
+UNIV_INLINE
+dict_table_t*
+dict_table_check_if_in_cache_low(
+/*=============================*/
+					/* out: table, NULL if not found */
+	const char*	table_name);	/* in: table name */
+/**************************************************************************
+Gets a table; loads it to the dictionary cache if necessary. A low-level
+function. */
+UNIV_INLINE
+dict_table_t*
+dict_table_get_low(
+/*===============*/
+					/* out: table, NULL if not found */
+	const char*	table_name);	/* in: table name */
+/**************************************************************************
+Returns a table object based on table id. */
+UNIV_INLINE
+dict_table_t*
+dict_table_get_on_id_low(
+/*=====================*/
+				/* out: table, NULL if does not exist */
+	dulint	table_id);	/* in: table id */
+/**************************************************************************
+Find an index that is equivalent to the one passed in and is not marked
+for deletion. */
+UNIV_INTERN
+dict_index_t*
+dict_foreign_find_equiv_index(
+/*==========================*/
+				/* out: index equivalent to
+				foreign->foreign_index, or NULL */
+	dict_foreign_t*	foreign);/* in: foreign key */
+/**************************************************************************
+Returns an index object by matching on the name and column names and if
+more than index is found return the index with the higher id.*/
+UNIV_INTERN
+dict_index_t*
+dict_table_get_index_by_max_id(
+/*===========================*/
+				/* out: matching index, NULL if not found */
+	dict_table_t*	table,	/* in: table */
+	const char*	name,	/* in: the index name to find */
+	const char**	columns,/* in: array of column names */
+	ulint		n_cols);/* in: number of columns */
+/**************************************************************************
+Returns a column's name. */
+
+const char*
+dict_table_get_col_name(
+/*====================*/
+					/* out: column name. NOTE: not
+					guaranteed to stay valid if table is
+					modified in any way (columns added,
+					etc.). */
+	const dict_table_t*	table,	/* in: table */
+	ulint			col_nr);/* in: column number */
+
+/**************************************************************************
+Prints a table definition. */
+UNIV_INTERN
+void
+dict_table_print(
+/*=============*/
+	dict_table_t*	table);	/* in: table */
+/**************************************************************************
+Prints a table data. */
+UNIV_INTERN
+void
+dict_table_print_low(
+/*=================*/
+	dict_table_t*	table);	/* in: table */
+/**************************************************************************
+Prints a table data when we know the table name. */
+UNIV_INTERN
+void
+dict_table_print_by_name(
+/*=====================*/
+	const char*	name);
+/**************************************************************************
+Outputs info on foreign keys of a table. */
+UNIV_INTERN
+void
+dict_print_info_on_foreign_keys(
+/*============================*/
+	ibool		create_table_format, /* in: if TRUE then print in
+				a format suitable to be inserted into
+				a CREATE TABLE, otherwise in the format
+				of SHOW TABLE STATUS */
+	FILE*		file,	/* in: file where to print */
+	trx_t*		trx,	/* in: transaction */
+	dict_table_t*	table);	/* in: table */
+/**************************************************************************
+Outputs info on a foreign key of a table in a format suitable for
+CREATE TABLE. */
+UNIV_INTERN
+void
+dict_print_info_on_foreign_key_in_create_format(
+/*============================================*/
+	FILE*		file,		/* in: file where to print */
+	trx_t*		trx,		/* in: transaction */
+	dict_foreign_t*	foreign,	/* in: foreign key constraint */
+	ibool		add_newline);	/* in: whether to add a newline */
+/************************************************************************
+Displays the names of the index and the table. */
+UNIV_INTERN
+void
+dict_index_name_print(
+/*==================*/
+	FILE*			file,	/* in: output stream */
+	trx_t*			trx,	/* in: transaction */
+	const dict_index_t*	index);	/* in: index to print */
+#ifdef UNIV_DEBUG
+/************************************************************************
+Gets the first index on the table (the clustered index). */
+UNIV_INLINE
+dict_index_t*
+dict_table_get_first_index(
+/*=======================*/
+					/* out: index, NULL if none exists */
+	const dict_table_t*	table);	/* in: table */
+/************************************************************************
+Gets the next index on the table. */
+UNIV_INLINE
+dict_index_t*
+dict_table_get_next_index(
+/*======================*/
+					/* out: index, NULL if none left */
+	const dict_index_t*	index);	/* in: index */
+#else /* UNIV_DEBUG */
+# define dict_table_get_first_index(table) UT_LIST_GET_FIRST((table)->indexes)
+# define dict_table_get_next_index(index) UT_LIST_GET_NEXT(indexes, index)
+#endif /* UNIV_DEBUG */
+/************************************************************************
+Check whether the index is the clustered index. */
+UNIV_INLINE
+ulint
+dict_index_is_clust(
+/*================*/
+					/* out: nonzero for clustered index,
+					zero for other indexes */
+	const dict_index_t*	index)	/* in: index */
+	__attribute__((pure));
+/************************************************************************
+Check whether the index is unique. */
+UNIV_INLINE
+ulint
+dict_index_is_unique(
+/*=================*/
+					/* out: nonzero for unique index,
+					zero for other indexes */
+	const dict_index_t*	index)	/* in: index */
+	__attribute__((pure));
+/************************************************************************
+Check whether the index is the insert buffer tree. */
+UNIV_INLINE
+ulint
+dict_index_is_ibuf(
+/*===============*/
+					/* out: nonzero for insert buffer,
+					zero for other indexes */
+	const dict_index_t*	index)	/* in: index */
+	__attribute__((pure));
+
+/************************************************************************
+Gets the number of user-defined columns in a table in the dictionary
+cache. */
+UNIV_INLINE
+ulint
+dict_table_get_n_user_cols(
+/*=======================*/
+					/* out: number of user-defined
+					(e.g., not ROW_ID)
+					columns of a table */
+	const dict_table_t*	table);	/* in: table */
+/************************************************************************
+Gets the number of system columns in a table in the dictionary cache. */
+UNIV_INLINE
+ulint
+dict_table_get_n_sys_cols(
+/*======================*/
+					/* out: number of system (e.g.,
+					ROW_ID) columns of a table */
+	const dict_table_t*	table);	/* in: table */
+/************************************************************************
+Gets the number of all columns (also system) in a table in the dictionary
+cache. */
+UNIV_INLINE
+ulint
+dict_table_get_n_cols(
+/*==================*/
+					/* out: number of columns of a table */
+	const dict_table_t*	table);	/* in: table */
+#ifdef UNIV_DEBUG
+/************************************************************************
+Gets the nth column of a table. */
+UNIV_INLINE
+dict_col_t*
+dict_table_get_nth_col(
+/*===================*/
+					/* out: pointer to column object */
+	const dict_table_t*	table,	/* in: table */
+	ulint			pos);	/* in: position of column */
+/************************************************************************
+Gets the given system column of a table. */
+UNIV_INLINE
+dict_col_t*
+dict_table_get_sys_col(
+/*===================*/
+					/* out: pointer to column object */
+	const dict_table_t*	table,	/* in: table */
+	ulint			sys);	/* in: DATA_ROW_ID, ... */
+#else /* UNIV_DEBUG */
+#define dict_table_get_nth_col(table, pos) \
+((table)->cols + (pos))
+#define dict_table_get_sys_col(table, sys) \
+((table)->cols + (table)->n_cols + (sys) - DATA_N_SYS_COLS)
+#endif /* UNIV_DEBUG */
+/************************************************************************
+Gets the given system column number of a table. */
+UNIV_INLINE
+ulint
+dict_table_get_sys_col_no(
+/*======================*/
+					/* out: column number */
+	const dict_table_t*	table,	/* in: table */
+	ulint			sys);	/* in: DATA_ROW_ID, ... */
+/************************************************************************
+Returns the minimum data size of an index record. */
+UNIV_INLINE
+ulint
+dict_index_get_min_size(
+/*====================*/
+					/* out: minimum data size in bytes */
+	const dict_index_t*	index);	/* in: index */
+/************************************************************************
+Check whether the table uses the compact page format. */
+UNIV_INLINE
+ibool
+dict_table_is_comp(
+/*===============*/
+					/* out: TRUE if table uses the
+					compact page format */
+	const dict_table_t*	table);	/* in: table */
+/************************************************************************
+Determine the file format of a table. */
+UNIV_INLINE
+ulint
+dict_table_get_format(
+/*==================*/
+					/* out: file format version */
+	const dict_table_t*	table);	/* in: table */
+/************************************************************************
+Set the file format of a table. */
+UNIV_INLINE
+void
+dict_table_set_format(
+/*==================*/
+	dict_table_t*	table,	/* in/out: table */
+	ulint		format);/* in: file format version */
+/************************************************************************
+Extract the compressed page size from table flags. */
+UNIV_INLINE
+ulint
+dict_table_flags_to_zip_size(
+/*=========================*/
+			/* out: compressed page size,
+			or 0 if not compressed */
+	ulint	flags)	/* in: flags */
+	__attribute__((const));
+/************************************************************************
+Check whether the table uses the compressed compact page format. */
+UNIV_INLINE
+ulint
+dict_table_zip_size(
+/*================*/
+					/* out: compressed page size,
+					or 0 if not compressed */
+	const dict_table_t*	table);	/* in: table */
+/************************************************************************
+Checks if a column is in the ordering columns of the clustered index of a
+table. Column prefixes are treated like whole columns. */
+UNIV_INTERN
+ibool
+dict_table_col_in_clustered_key(
+/*============================*/
+					/* out: TRUE if the column, or its
+					prefix, is in the clustered key */
+	const dict_table_t*	table,	/* in: table */
+	ulint			n);	/* in: column number */
+/***********************************************************************
+Copies types of columns contained in table to tuple and sets all
+fields of the tuple to the SQL NULL value.  This function should
+be called right after dtuple_create(). */
+UNIV_INTERN
+void
+dict_table_copy_types(
+/*==================*/
+	dtuple_t*		tuple,	/* in/out: data tuple */
+	const dict_table_t*	table);	/* in: table */
+/**************************************************************************
+Looks for an index with the given id. NOTE that we do not reserve
+the dictionary mutex: this function is for emergency purposes like
+printing info of a corrupt database page! */
+UNIV_INTERN
+dict_index_t*
+dict_index_find_on_id_low(
+/*======================*/
+			/* out: index or NULL if not found from cache */
+	dulint	id);	/* in: index id */
+/**************************************************************************
+Adds an index to the dictionary cache. */
+UNIV_INTERN
+ulint
+dict_index_add_to_cache(
+/*====================*/
+				/* out: DB_SUCCESS or error code */
+	dict_table_t*	table,	/* in: table on which the index is */
+	dict_index_t*	index,	/* in, own: index; NOTE! The index memory
+				object is freed in this function! */
+	ulint		page_no,/* in: root page number of the index */
+	ibool		strict);/* in: TRUE=refuse to create the index
+				if records could be too big to fit in
+				an B-tree page */
+/**************************************************************************
+Removes an index from the dictionary cache. */
+UNIV_INTERN
+void
+dict_index_remove_from_cache(
+/*=========================*/
+	dict_table_t*	table,	/* in/out: table */
+	dict_index_t*	index);	/* in, own: index */
+/************************************************************************
+Gets the number of fields in the internal representation of an index,
+including fields added by the dictionary system. */
+UNIV_INLINE
+ulint
+dict_index_get_n_fields(
+/*====================*/
+					/* out: number of fields */
+	const dict_index_t*	index);	/* in: an internal
+					representation of index (in
+					the dictionary cache) */
+/************************************************************************
+Gets the number of fields in the internal representation of an index
+that uniquely determine the position of an index entry in the index, if
+we do not take multiversioning into account: in the B-tree use the value
+returned by dict_index_get_n_unique_in_tree. */
+UNIV_INLINE
+ulint
+dict_index_get_n_unique(
+/*====================*/
+					/* out: number of fields */
+	const dict_index_t*	index);	/* in: an internal representation
+					of index (in the dictionary cache) */
+/************************************************************************
+Gets the number of fields in the internal representation of an index
+which uniquely determine the position of an index entry in the index, if
+we also take multiversioning into account. */
+UNIV_INLINE
+ulint
+dict_index_get_n_unique_in_tree(
+/*============================*/
+					/* out: number of fields */
+	const dict_index_t*	index);	/* in: an internal representation
+					of index (in the dictionary cache) */
+/************************************************************************
+Gets the number of user-defined ordering fields in the index. In the internal
+representation we add the row id to the ordering fields to make all indexes
+unique, but this function returns the number of fields the user defined
+in the index as ordering fields. */
+UNIV_INLINE
+ulint
+dict_index_get_n_ordering_defined_by_user(
+/*======================================*/
+					/* out: number of fields */
+	const dict_index_t*	index);	/* in: an internal representation
+					of index (in the dictionary cache) */
+#ifdef UNIV_DEBUG
+/************************************************************************
+Gets the nth field of an index. */
+UNIV_INLINE
+dict_field_t*
+dict_index_get_nth_field(
+/*=====================*/
+					/* out: pointer to field object */
+	const dict_index_t*	index,	/* in: index */
+	ulint			pos);	/* in: position of field */
+#else /* UNIV_DEBUG */
+# define dict_index_get_nth_field(index, pos) ((index)->fields + (pos))
+#endif /* UNIV_DEBUG */
+/************************************************************************
+Gets pointer to the nth column in an index. */
+UNIV_INLINE
+const dict_col_t*
+dict_index_get_nth_col(
+/*===================*/
+					/* out: column */
+	const dict_index_t*	index,	/* in: index */
+	ulint			pos);	/* in: position of the field */
+/************************************************************************
+Gets the column number of the nth field in an index. */
+UNIV_INLINE
+ulint
+dict_index_get_nth_col_no(
+/*======================*/
+					/* out: column number */
+	const dict_index_t*	index,	/* in: index */
+	ulint			pos);	/* in: position of the field */
+/************************************************************************
+Looks for column n in an index. */
+UNIV_INTERN
+ulint
+dict_index_get_nth_col_pos(
+/*=======================*/
+					/* out: position in internal
+					representation of the index;
+					if not contained, returns
+					ULINT_UNDEFINED */
+	const dict_index_t*	index,	/* in: index */
+	ulint			n);	/* in: column number */
+/************************************************************************
+Returns TRUE if the index contains a column or a prefix of that column. */
+UNIV_INTERN
+ibool
+dict_index_contains_col_or_prefix(
+/*==============================*/
+					/* out: TRUE if contains the column
+					or its prefix */
+	const dict_index_t*	index,	/* in: index */
+	ulint			n);	/* in: column number */
+/************************************************************************
+Looks for a matching field in an index. The column has to be the same. The
+column in index must be complete, or must contain a prefix longer than the
+column in index2. That is, we must be able to construct the prefix in index2
+from the prefix in index. */
+UNIV_INTERN
+ulint
+dict_index_get_nth_field_pos(
+/*=========================*/
+					/* out: position in internal
+					representation of the index;
+					if not contained, returns
+					ULINT_UNDEFINED */
+	const dict_index_t*	index,	/* in: index from which to search */
+	const dict_index_t*	index2,	/* in: index */
+	ulint			n);	/* in: field number in index2 */
+/************************************************************************
+Looks for column n position in the clustered index. */
+UNIV_INTERN
+ulint
+dict_table_get_nth_col_pos(
+/*=======================*/
+					/* out: position in internal
+					representation of
+					the clustered index */
+	const dict_table_t*	table,	/* in: table */
+	ulint			n);	/* in: column number */
+/************************************************************************
+Returns the position of a system column in an index. */
+UNIV_INLINE
+ulint
+dict_index_get_sys_col_pos(
+/*=======================*/
+					/* out: position,
+					ULINT_UNDEFINED if not contained */
+	const dict_index_t*	index,	/* in: index */
+	ulint			type);	/* in: DATA_ROW_ID, ... */
+/***********************************************************************
+Adds a column to index. */
+UNIV_INTERN
+void
+dict_index_add_col(
+/*===============*/
+	dict_index_t*		index,		/* in/out: index */
+	const dict_table_t*	table,		/* in: table */
+	dict_col_t*		col,		/* in: column */
+	ulint			prefix_len);	/* in: column prefix length */
+/***********************************************************************
+Copies types of fields contained in index to tuple. */
+UNIV_INTERN
+void
+dict_index_copy_types(
+/*==================*/
+	dtuple_t*		tuple,		/* in/out: data tuple */
+	const dict_index_t*	index,		/* in: index */
+	ulint			n_fields);	/* in: number of
+						field types to copy */
+/*************************************************************************
+Gets the field column. */
+UNIV_INLINE
+const dict_col_t*
+dict_field_get_col(
+/*===============*/
+	const dict_field_t*	field);
+
+/**************************************************************************
+Returns an index object if it is found in the dictionary cache.
+Assumes that dict_sys->mutex is already being held. */
+UNIV_INTERN
+dict_index_t*
+dict_index_get_if_in_cache_low(
+/*===========================*/
+				/* out: index, NULL if not found */
+	dulint	index_id);	/* in: index id */
+#if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
+/**************************************************************************
+Returns an index object if it is found in the dictionary cache. */
+UNIV_INTERN
+dict_index_t*
+dict_index_get_if_in_cache(
+/*=======================*/
+				/* out: index, NULL if not found */
+	dulint	index_id);	/* in: index id */
+#endif /* UNIV_DEBUG || UNIV_BUF_DEBUG */
+#ifdef UNIV_DEBUG
+/**************************************************************************
+Checks that a tuple has n_fields_cmp value in a sensible range, so that
+no comparison can occur with the page number field in a node pointer. */
+UNIV_INTERN
+ibool
+dict_index_check_search_tuple(
+/*==========================*/
+					/* out: TRUE if ok */
+	const dict_index_t*	index,	/* in: index tree */
+	const dtuple_t*		tuple);	/* in: tuple used in a search */
+/**************************************************************************
+Check for duplicate index entries in a table [using the index name] */
+UNIV_INTERN
+void
+dict_table_check_for_dup_indexes(
+/*=============================*/
+	const dict_table_t*	table);	/* in: Check for dup indexes
+					in this table */
+
+#endif /* UNIV_DEBUG */
+/**************************************************************************
+Builds a node pointer out of a physical record and a page number. */
+UNIV_INTERN
+dtuple_t*
+dict_index_build_node_ptr(
+/*======================*/
+					/* out, own: node pointer */
+	const dict_index_t*	index,	/* in: index */
+	const rec_t*		rec,	/* in: record for which to build node
+					pointer */
+	ulint			page_no,/* in: page number to put in node
+					pointer */
+	mem_heap_t*		heap,	/* in: memory heap where pointer
+					created */
+	ulint			level);	/* in: level of rec in tree:
+					0 means leaf level */
+/**************************************************************************
+Copies an initial segment of a physical record, long enough to specify an
+index entry uniquely. */
+UNIV_INTERN
+rec_t*
+dict_index_copy_rec_order_prefix(
+/*=============================*/
+					/* out: pointer to the prefix record */
+	const dict_index_t*	index,	/* in: index */
+	const rec_t*		rec,	/* in: record for which to
+					copy prefix */
+	ulint*			n_fields,/* out: number of fields copied */
+	byte**			buf,	/* in/out: memory buffer for the
+					copied prefix, or NULL */
+	ulint*			buf_size);/* in/out: buffer size */
+/**************************************************************************
+Builds a typed data tuple out of a physical record. */
+UNIV_INTERN
+dtuple_t*
+dict_index_build_data_tuple(
+/*========================*/
+				/* out, own: data tuple */
+	dict_index_t*	index,	/* in: index */
+	rec_t*		rec,	/* in: record for which to build data tuple */
+	ulint		n_fields,/* in: number of data fields */
+	mem_heap_t*	heap);	/* in: memory heap where tuple created */
+/*************************************************************************
+Gets the space id of the root of the index tree. */
+UNIV_INLINE
+ulint
+dict_index_get_space(
+/*=================*/
+					/* out: space id */
+	const dict_index_t*	index);	/* in: index */
+/*************************************************************************
+Sets the space id of the root of the index tree. */
+UNIV_INLINE
+void
+dict_index_set_space(
+/*=================*/
+	dict_index_t*	index,	/* in/out: index */
+	ulint		space);	/* in: space id */
+/*************************************************************************
+Gets the page number of the root of the index tree. */
+UNIV_INLINE
+ulint
+dict_index_get_page(
+/*================*/
+					/* out: page number */
+	const dict_index_t*	tree);	/* in: index */
+/*************************************************************************
+Sets the page number of the root of index tree. */
+UNIV_INLINE
+void
+dict_index_set_page(
+/*================*/
+	dict_index_t*	index,	/* in/out: index */
+	ulint		page);	/* in: page number */
+/*************************************************************************
+Gets the read-write lock of the index tree. */
+UNIV_INLINE
+rw_lock_t*
+dict_index_get_lock(
+/*================*/
+				/* out: read-write lock */
+	dict_index_t*	index);	/* in: index */
+/************************************************************************
+Returns free space reserved for future updates of records. This is
+relevant only in the case of many consecutive inserts, as updates
+which make the records bigger might fragment the index. */
+UNIV_INLINE
+ulint
+dict_index_get_space_reserve(void);
+/*==============================*/
+				/* out: number of free bytes on page,
+				reserved for updates */
+/*************************************************************************
+Calculates the minimum record length in an index. */
+UNIV_INTERN
+ulint
+dict_index_calc_min_rec_len(
+/*========================*/
+	const dict_index_t*	index);	/* in: index */
+/*************************************************************************
+Calculates new estimates for table and index statistics. The statistics
+are used in query optimization. */
+UNIV_INTERN
+void
+dict_update_statistics_low(
+/*=======================*/
+	dict_table_t*	table,		/* in/out: table */
+	ibool		has_dict_mutex);/* in: TRUE if the caller has the
+					dictionary mutex */
+/*************************************************************************
+Calculates new estimates for table and index statistics. The statistics
+are used in query optimization. */
+UNIV_INTERN
+void
+dict_update_statistics(
+/*===================*/
+	dict_table_t*	table);	/* in/out: table */
+/************************************************************************
+Reserves the dictionary system mutex for MySQL. */
+UNIV_INTERN
+void
+dict_mutex_enter_for_mysql(void);
+/*============================*/
+/************************************************************************
+Releases the dictionary system mutex for MySQL. */
+UNIV_INTERN
+void
+dict_mutex_exit_for_mysql(void);
+/*===========================*/
+/************************************************************************
+Checks if the database name in two table names is the same. */
+UNIV_INTERN
+ibool
+dict_tables_have_same_db(
+/*=====================*/
+				/* out: TRUE if same db name */
+	const char*	name1,	/* in: table name in the form
+				dbname '/' tablename */
+	const char*	name2);	/* in: table name in the form
+				dbname '/' tablename */
+/*************************************************************************
+Removes an index from the cache */
+UNIV_INTERN
+void
+dict_index_remove_from_cache(
+/*=========================*/
+	dict_table_t*	table,	/* in/out: table */
+	dict_index_t*	index);	/* in, own: index */
+/**************************************************************************
+Get index by name */
+UNIV_INTERN
+dict_index_t*
+dict_table_get_index_on_name(
+/*=========================*/
+				/* out: index, NULL if does not exist */
+	dict_table_t*	table,	/* in: table */
+	const char*	name);	/* in: name of the index to find */
+/**************************************************************************
+In case there is more than one index with the same name return the index
+with the min(id). */
+UNIV_INTERN
+dict_index_t*
+dict_table_get_index_on_name_and_min_id(
+/*====================================*/
+				/* out: index, NULL if does not exist */
+	dict_table_t*	table,	/* in: table */
+	const char*	name);	/* in: name of the index to find */
+/* Buffers for storing detailed information about the latest foreign key
+and unique key errors */
+extern FILE*	dict_foreign_err_file;
+extern mutex_t	dict_foreign_err_mutex; /* mutex protecting the buffers */
+
+extern dict_sys_t*	dict_sys;	/* the dictionary system */
+extern rw_lock_t	dict_operation_lock;
+
+/* Dictionary system struct */
+struct dict_sys_struct{
+	mutex_t		mutex;		/* mutex protecting the data
+					dictionary; protects also the
+					disk-based dictionary system tables;
+					this mutex serializes CREATE TABLE
+					and DROP TABLE, as well as reading
+					the dictionary data for a table from
+					system tables */
+	dulint		row_id;		/* the next row id to assign;
+					NOTE that at a checkpoint this
+					must be written to the dict system
+					header and flushed to a file; in
+					recovery this must be derived from
+					the log records */
+	hash_table_t*	table_hash;	/* hash table of the tables, based
+					on name */
+	hash_table_t*	table_id_hash;	/* hash table of the tables, based
+					on id */
+	UT_LIST_BASE_NODE_T(dict_table_t)
+			table_LRU;	/* LRU list of tables */
+	ulint		size;		/* varying space in bytes occupied
+					by the data dictionary table and
+					index objects */
+	dict_table_t*	sys_tables;	/* SYS_TABLES table */
+	dict_table_t*	sys_columns;	/* SYS_COLUMNS table */
+	dict_table_t*	sys_indexes;	/* SYS_INDEXES table */
+	dict_table_t*	sys_fields;	/* SYS_FIELDS table */
+};
+
+#ifndef UNIV_NONINL
+#include "dict0dict.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/dict0dict.ic b/storage/xtradb/include/dict0dict.ic
new file mode 100644
index 00000000000..628d207b329
--- /dev/null
+++ b/storage/xtradb/include/dict0dict.ic
@@ -0,0 +1,785 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/**********************************************************************
+Data dictionary system
+
+Created 1/8/1996 Heikki Tuuri
+***********************************************************************/
+
+#include "dict0load.h"
+#include "rem0types.h"
+#include "data0type.h"
+
+/*************************************************************************
+Gets the column data type. */
+UNIV_INLINE
+void
+dict_col_copy_type(
+/*===============*/
+	const dict_col_t*	col,	/* in: column */
+	dtype_t*		type)	/* out: data type */
+{
+	ut_ad(col && type);
+
+	type->mtype = col->mtype;
+	type->prtype = col->prtype;
+	type->len = col->len;
+	type->mbminlen = col->mbminlen;
+	type->mbmaxlen = col->mbmaxlen;
+}
+
+#ifdef UNIV_DEBUG
+/*************************************************************************
+Assert that a column and a data type match. */
+UNIV_INLINE
+ibool
+dict_col_type_assert_equal(
+/*=======================*/
+					/* out: TRUE */
+	const dict_col_t*	col,	/* in: column */
+	const dtype_t*		type)	/* in: data type */
+{
+	ut_ad(col);
+	ut_ad(type);
+
+	ut_ad(col->mtype == type->mtype);
+	ut_ad(col->prtype == type->prtype);
+	ut_ad(col->len == type->len);
+	ut_ad(col->mbminlen == type->mbminlen);
+	ut_ad(col->mbmaxlen == type->mbmaxlen);
+
+	return(TRUE);
+}
+#endif /* UNIV_DEBUG */
+
+/***************************************************************************
+Returns the minimum size of the column. */
+UNIV_INLINE
+ulint
+dict_col_get_min_size(
+/*==================*/
+					/* out: minimum size */
+	const dict_col_t*	col)	/* in: column */
+{
+	return(dtype_get_min_size_low(col->mtype, col->prtype, col->len,
+				      col->mbminlen, col->mbmaxlen));
+}
+/***************************************************************************
+Returns the maximum size of the column. */
+UNIV_INLINE
+ulint
+dict_col_get_max_size(
+/*==================*/
+					/* out: maximum size */
+	const dict_col_t*	col)	/* in: column */
+{
+	return(dtype_get_max_size_low(col->mtype, col->len));
+}
+/***************************************************************************
+Returns the size of a fixed size column, 0 if not a fixed size column. */
+UNIV_INLINE
+ulint
+dict_col_get_fixed_size(
+/*====================*/
+					/* out: fixed size, or 0 */
+	const dict_col_t*	col)	/* in: column */
+{
+	return(dtype_get_fixed_size_low(col->mtype, col->prtype, col->len,
+					col->mbminlen, col->mbmaxlen));
+}
+/***************************************************************************
+Returns the ROW_FORMAT=REDUNDANT stored SQL NULL size of a column.
+For fixed length types it is the fixed length of the type, otherwise 0. */
+UNIV_INLINE
+ulint
+dict_col_get_sql_null_size(
+/*=======================*/
+					/* out: SQL null storage size
+					in ROW_FORMAT=REDUNDANT */
+	const dict_col_t*	col)	/* in: column */
+{
+	return(dict_col_get_fixed_size(col));
+}
+
+/*************************************************************************
+Gets the column number. */
+UNIV_INLINE
+ulint
+dict_col_get_no(
+/*============*/
+	const dict_col_t*	col)
+{
+	ut_ad(col);
+
+	return(col->ind);
+}
+
+/*************************************************************************
+Gets the column position in the clustered index. */
+UNIV_INLINE
+ulint
+dict_col_get_clust_pos(
+/*===================*/
+	const dict_col_t*	col,		/* in: table column */
+	const dict_index_t*	clust_index)	/* in: clustered index */
+{
+	ulint	i;
+
+	ut_ad(col);
+	ut_ad(clust_index);
+	ut_ad(dict_index_is_clust(clust_index));
+
+	for (i = 0; i < clust_index->n_def; i++) {
+		const dict_field_t*	field = &clust_index->fields[i];
+
+		if (!field->prefix_len && field->col == col) {
+			return(i);
+		}
+	}
+
+	return(ULINT_UNDEFINED);
+}
+
+#ifdef UNIV_DEBUG
+/************************************************************************
+Gets the first index on the table (the clustered index). */
+UNIV_INLINE
+dict_index_t*
+dict_table_get_first_index(
+/*=======================*/
+					/* out: index, NULL if none exists */
+	const dict_table_t*	table)	/* in: table */
+{
+	ut_ad(table);
+	ut_ad(table->magic_n == DICT_TABLE_MAGIC_N);
+
+	return(UT_LIST_GET_FIRST(((dict_table_t*) table)->indexes));
+}
+
+/************************************************************************
+Gets the next index on the table. */
+UNIV_INLINE
+dict_index_t*
+dict_table_get_next_index(
+/*======================*/
+					/* out: index, NULL if none left */
+	const dict_index_t*	index)	/* in: index */
+{
+	ut_ad(index);
+	ut_ad(index->magic_n == DICT_INDEX_MAGIC_N);
+
+	return(UT_LIST_GET_NEXT(indexes, (dict_index_t*) index));
+}
+#endif /* UNIV_DEBUG */
+
+/************************************************************************
+Check whether the index is the clustered index. */
+UNIV_INLINE
+ulint
+dict_index_is_clust(
+/*================*/
+					/* out: nonzero for clustered index,
+					zero for other indexes */
+	const dict_index_t*	index)	/* in: index */
+{
+	ut_ad(index);
+	ut_ad(index->magic_n == DICT_INDEX_MAGIC_N);
+
+	return(UNIV_UNLIKELY(index->type & DICT_CLUSTERED));
+}
+/************************************************************************
+Check whether the index is unique. */
+UNIV_INLINE
+ulint
+dict_index_is_unique(
+/*=================*/
+					/* out: nonzero for unique index,
+					zero for other indexes */
+	const dict_index_t*	index)	/* in: index */
+{
+	ut_ad(index);
+	ut_ad(index->magic_n == DICT_INDEX_MAGIC_N);
+
+	return(UNIV_UNLIKELY(index->type & DICT_UNIQUE));
+}
+
+/************************************************************************
+Check whether the index is the insert buffer tree. */
+UNIV_INLINE
+ulint
+dict_index_is_ibuf(
+/*===============*/
+					/* out: nonzero for insert buffer,
+					zero for other indexes */
+	const dict_index_t*	index)	/* in: index */
+{
+	ut_ad(index);
+	ut_ad(index->magic_n == DICT_INDEX_MAGIC_N);
+
+	return(UNIV_UNLIKELY(index->type & DICT_IBUF));
+}
+
+/************************************************************************
+Gets the number of user-defined columns in a table in the dictionary
+cache. */
+UNIV_INLINE
+ulint
+dict_table_get_n_user_cols(
+/*=======================*/
+					/* out: number of user-defined
+					(e.g., not ROW_ID)
+					columns of a table */
+	const dict_table_t*	table)	/* in: table */
+{
+	ut_ad(table);
+	ut_ad(table->magic_n == DICT_TABLE_MAGIC_N);
+
+	return(table->n_cols - DATA_N_SYS_COLS);
+}
+
+/************************************************************************
+Gets the number of system columns in a table in the dictionary cache. */
+UNIV_INLINE
+ulint
+dict_table_get_n_sys_cols(
+/*======================*/
+					/* out: number of system (e.g.,
+					ROW_ID) columns of a table */
+	const dict_table_t*	table __attribute__((unused)))	/* in: table */
+{
+	ut_ad(table);
+	ut_ad(table->magic_n == DICT_TABLE_MAGIC_N);
+	ut_ad(table->cached);
+
+	return(DATA_N_SYS_COLS);
+}
+
+/************************************************************************
+Gets the number of all columns (also system) in a table in the dictionary
+cache. */
+UNIV_INLINE
+ulint
+dict_table_get_n_cols(
+/*==================*/
+					/* out: number of columns of a table */
+	const dict_table_t*	table)	/* in: table */
+{
+	ut_ad(table);
+	ut_ad(table->magic_n == DICT_TABLE_MAGIC_N);
+
+	return(table->n_cols);
+}
+
+#ifdef UNIV_DEBUG
+/************************************************************************
+Gets the nth column of a table. */
+UNIV_INLINE
+dict_col_t*
+dict_table_get_nth_col(
+/*===================*/
+					/* out: pointer to column object */
+	const dict_table_t*	table,	/* in: table */
+	ulint			pos)	/* in: position of column */
+{
+	ut_ad(table);
+	ut_ad(pos < table->n_def);
+	ut_ad(table->magic_n == DICT_TABLE_MAGIC_N);
+
+	return((dict_col_t*) (table->cols) + pos);
+}
+
+/************************************************************************
+Gets the given system column of a table. */
+UNIV_INLINE
+dict_col_t*
+dict_table_get_sys_col(
+/*===================*/
+					/* out: pointer to column object */
+	const dict_table_t*	table,	/* in: table */
+	ulint			sys)	/* in: DATA_ROW_ID, ... */
+{
+	dict_col_t*	col;
+
+	ut_ad(table);
+	ut_ad(sys < DATA_N_SYS_COLS);
+	ut_ad(table->magic_n == DICT_TABLE_MAGIC_N);
+
+	col = dict_table_get_nth_col(table, table->n_cols
+				     - DATA_N_SYS_COLS + sys);
+	ut_ad(col->mtype == DATA_SYS);
+	ut_ad(col->prtype == (sys | DATA_NOT_NULL));
+
+	return(col);
+}
+#endif /* UNIV_DEBUG */
+
+/************************************************************************
+Gets the given system column number of a table. */
+UNIV_INLINE
+ulint
+dict_table_get_sys_col_no(
+/*======================*/
+					/* out: column number */
+	const dict_table_t*	table,	/* in: table */
+	ulint			sys)	/* in: DATA_ROW_ID, ... */
+{
+	ut_ad(table);
+	ut_ad(sys < DATA_N_SYS_COLS);
+	ut_ad(table->magic_n == DICT_TABLE_MAGIC_N);
+
+	return(table->n_cols - DATA_N_SYS_COLS + sys);
+}
+
+/************************************************************************
+Check whether the table uses the compact page format. */
+UNIV_INLINE
+ibool
+dict_table_is_comp(
+/*===============*/
+					/* out: TRUE if table uses the
+					compact page format */
+	const dict_table_t*	table)	/* in: table */
+{
+	ut_ad(table);
+
+#if DICT_TF_COMPACT != TRUE
+#error
+#endif
+
+	return(UNIV_LIKELY(table->flags & DICT_TF_COMPACT));
+}
+
+/************************************************************************
+Determine the file format of a table. */
+UNIV_INLINE
+ulint
+dict_table_get_format(
+/*==================*/
+					/* out: file format version */
+	const dict_table_t*	table)	/* in: table */
+{
+	ut_ad(table);
+
+	return((table->flags & DICT_TF_FORMAT_MASK) >> DICT_TF_FORMAT_SHIFT);
+}
+
+/************************************************************************
+Determine the file format of a table. */
+UNIV_INLINE
+void
+dict_table_set_format(
+/*==================*/
+	dict_table_t*	table,	/* in/out: table */
+	ulint		format)	/* in: file format version */
+{
+	ut_ad(table);
+
+	table->flags = (table->flags & ~DICT_TF_FORMAT_MASK)
+		| (format << DICT_TF_FORMAT_SHIFT);
+}
+
+/************************************************************************
+Extract the compressed page size from table flags. */
+UNIV_INLINE
+ulint
+dict_table_flags_to_zip_size(
+/*=========================*/
+			/* out: compressed page size,
+			or 0 if not compressed */
+	ulint	flags)	/* in: flags */
+{
+	ulint	zip_size = flags & DICT_TF_ZSSIZE_MASK;
+
+	if (UNIV_UNLIKELY(zip_size)) {
+		zip_size = ((PAGE_ZIP_MIN_SIZE >> 1)
+			 << (zip_size >> DICT_TF_ZSSIZE_SHIFT));
+
+		ut_ad(zip_size <= UNIV_PAGE_SIZE);
+	}
+
+	return(zip_size);
+}
+
+/************************************************************************
+Check whether the table uses the compressed compact page format. */
+UNIV_INLINE
+ulint
+dict_table_zip_size(
+/*================*/
+					/* out: compressed page size,
+					or 0 if not compressed */
+	const dict_table_t*	table)	/* in: table */
+{
+	ut_ad(table);
+
+	return(dict_table_flags_to_zip_size(table->flags));
+}
+
+/************************************************************************
+Gets the number of fields in the internal representation of an index,
+including fields added by the dictionary system. */
+UNIV_INLINE
+ulint
+dict_index_get_n_fields(
+/*====================*/
+					/* out: number of fields */
+	const dict_index_t*	index)	/* in: an internal
+					representation of index (in
+					the dictionary cache) */
+{
+	ut_ad(index);
+	ut_ad(index->magic_n == DICT_INDEX_MAGIC_N);
+
+	return(index->n_fields);
+}
+
+/************************************************************************
+Gets the number of fields in the internal representation of an index
+that uniquely determine the position of an index entry in the index, if
+we do not take multiversioning into account: in the B-tree use the value
+returned by dict_index_get_n_unique_in_tree. */
+UNIV_INLINE
+ulint
+dict_index_get_n_unique(
+/*====================*/
+					/* out: number of fields */
+	const dict_index_t*	index)	/* in: an internal representation
+					of index (in the dictionary cache) */
+{
+	ut_ad(index);
+	ut_ad(index->magic_n == DICT_INDEX_MAGIC_N);
+	ut_ad(index->cached);
+
+	return(index->n_uniq);
+}
+
+/************************************************************************
+Gets the number of fields in the internal representation of an index
+which uniquely determine the position of an index entry in the index, if
+we also take multiversioning into account. */
+UNIV_INLINE
+ulint
+dict_index_get_n_unique_in_tree(
+/*============================*/
+					/* out: number of fields */
+	const dict_index_t*	index)	/* in: an internal representation
+					of index (in the dictionary cache) */
+{
+	ut_ad(index);
+	ut_ad(index->magic_n == DICT_INDEX_MAGIC_N);
+	ut_ad(index->cached);
+
+	if (dict_index_is_clust(index)) {
+
+		return(dict_index_get_n_unique(index));
+	}
+
+	return(dict_index_get_n_fields(index));
+}
+
+/************************************************************************
+Gets the number of user-defined ordering fields in the index. In the internal
+representation of clustered indexes we add the row id to the ordering fields
+to make a clustered index unique, but this function returns the number of
+fields the user defined in the index as ordering fields. */
+UNIV_INLINE
+ulint
+dict_index_get_n_ordering_defined_by_user(
+/*======================================*/
+					/* out: number of fields */
+	const dict_index_t*	index)	/* in: an internal representation
+					of index (in the dictionary cache) */
+{
+	return(index->n_user_defined_cols);
+}
+
+#ifdef UNIV_DEBUG
+/************************************************************************
+Gets the nth field of an index. */
+UNIV_INLINE
+dict_field_t*
+dict_index_get_nth_field(
+/*=====================*/
+					/* out: pointer to field object */
+	const dict_index_t*	index,	/* in: index */
+	ulint			pos)	/* in: position of field */
+{
+	ut_ad(index);
+	ut_ad(pos < index->n_def);
+	ut_ad(index->magic_n == DICT_INDEX_MAGIC_N);
+
+	return((dict_field_t*) (index->fields) + pos);
+}
+#endif /* UNIV_DEBUG */
+
+/************************************************************************
+Returns the position of a system column in an index. */
+UNIV_INLINE
+ulint
+dict_index_get_sys_col_pos(
+/*=======================*/
+					/* out: position,
+					ULINT_UNDEFINED if not contained */
+	const dict_index_t*	index,	/* in: index */
+	ulint			type)	/* in: DATA_ROW_ID, ... */
+{
+	ut_ad(index);
+	ut_ad(index->magic_n == DICT_INDEX_MAGIC_N);
+	ut_ad(!(index->type & DICT_UNIVERSAL));
+
+	if (dict_index_is_clust(index)) {
+
+		return(dict_col_get_clust_pos(
+			       dict_table_get_sys_col(index->table, type),
+			       index));
+	}
+
+	return(dict_index_get_nth_col_pos(
+		       index, dict_table_get_sys_col_no(index->table, type)));
+}
+
+/*************************************************************************
+Gets the field column. */
+UNIV_INLINE
+const dict_col_t*
+dict_field_get_col(
+/*===============*/
+	const dict_field_t*	field)
+{
+	ut_ad(field);
+
+	return(field->col);
+}
+
+/************************************************************************
+Gets pointer to the nth column in an index. */
+UNIV_INLINE
+const dict_col_t*
+dict_index_get_nth_col(
+/*===================*/
+					/* out: column */
+	const dict_index_t*	index,	/* in: index */
+	ulint			pos)	/* in: position of the field */
+{
+	return(dict_field_get_col(dict_index_get_nth_field(index, pos)));
+}
+
+/************************************************************************
+Gets the column number the nth field in an index. */
+UNIV_INLINE
+ulint
+dict_index_get_nth_col_no(
+/*======================*/
+					/* out: column number */
+	const dict_index_t*	index,	/* in: index */
+	ulint			pos)	/* in: position of the field */
+{
+	return(dict_col_get_no(dict_index_get_nth_col(index, pos)));
+}
+
+/************************************************************************
+Returns the minimum data size of an index record. */
+UNIV_INLINE
+ulint
+dict_index_get_min_size(
+/*====================*/
+					/* out: minimum data size in bytes */
+	const dict_index_t*	index)	/* in: index */
+{
+	ulint	n	= dict_index_get_n_fields(index);
+	ulint	size	= 0;
+
+	while (n--) {
+		size += dict_col_get_min_size(dict_index_get_nth_col(index,
+								     n));
+	}
+
+	return(size);
+}
+
+/*************************************************************************
+Gets the space id of the root of the index tree. */
+UNIV_INLINE
+ulint
+dict_index_get_space(
+/*=================*/
+					/* out: space id */
+	const dict_index_t*	index)	/* in: index */
+{
+	ut_ad(index);
+	ut_ad(index->magic_n == DICT_INDEX_MAGIC_N);
+
+	return(index->space);
+}
+
+/*************************************************************************
+Sets the space id of the root of the index tree. */
+UNIV_INLINE
+void
+dict_index_set_space(
+/*=================*/
+	dict_index_t*	index,	/* in/out: index */
+	ulint		space)	/* in: space id */
+{
+	ut_ad(index);
+	ut_ad(index->magic_n == DICT_INDEX_MAGIC_N);
+
+	index->space = space;
+}
+
+/*************************************************************************
+Gets the page number of the root of the index tree. */
+UNIV_INLINE
+ulint
+dict_index_get_page(
+/*================*/
+					/* out: page number */
+	const dict_index_t*	index)	/* in: index */
+{
+	ut_ad(index);
+	ut_ad(index->magic_n == DICT_INDEX_MAGIC_N);
+
+	return(index->page);
+}
+
+/*************************************************************************
+Sets the page number of the root of index tree. */
+UNIV_INLINE
+void
+dict_index_set_page(
+/*================*/
+	dict_index_t*	index,	/* in/out: index */
+	ulint		page)	/* in: page number */
+{
+	ut_ad(index);
+	ut_ad(index->magic_n == DICT_INDEX_MAGIC_N);
+
+	index->page = page;
+}
+
+/*************************************************************************
+Gets the read-write lock of the index tree. */
+UNIV_INLINE
+rw_lock_t*
+dict_index_get_lock(
+/*================*/
+				/* out: read-write lock */
+	dict_index_t*	index)	/* in: index */
+{
+	ut_ad(index);
+	ut_ad(index->magic_n == DICT_INDEX_MAGIC_N);
+
+	return(&(index->lock));
+}
+
+/************************************************************************
+Returns free space reserved for future updates of records. This is
+relevant only in the case of many consecutive inserts, as updates
+which make the records bigger might fragment the index. */
+UNIV_INLINE
+ulint
+dict_index_get_space_reserve(void)
+/*==============================*/
+				/* out: number of free bytes on page,
+				reserved for updates */
+{
+	return(UNIV_PAGE_SIZE / 16);
+}
+
+/**************************************************************************
+Checks if a table is in the dictionary cache. */
+UNIV_INLINE
+dict_table_t*
+dict_table_check_if_in_cache_low(
+/*=============================*/
+					/* out: table, NULL if not found */
+	const char*	table_name)	/* in: table name */
+{
+	dict_table_t*	table;
+	ulint		table_fold;
+
+	ut_ad(table_name);
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+
+	/* Look for the table name in the hash table */
+	table_fold = ut_fold_string(table_name);
+
+	HASH_SEARCH(name_hash, dict_sys->table_hash, table_fold,
+		    dict_table_t*, table, ut_ad(table->cached),
+		    !strcmp(table->name, table_name));
+	return(table);
+}
+
+/**************************************************************************
+Gets a table; loads it to the dictionary cache if necessary. A low-level
+function. */
+UNIV_INLINE
+dict_table_t*
+dict_table_get_low(
+/*===============*/
+					/* out: table, NULL if not found */
+	const char*	table_name)	/* in: table name */
+{
+	dict_table_t*	table;
+
+	ut_ad(table_name);
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+
+	table = dict_table_check_if_in_cache_low(table_name);
+
+	if (table == NULL) {
+		table = dict_load_table(table_name);
+	}
+
+	ut_ad(!table || table->cached);
+
+	return(table);
+}
+
+/**************************************************************************
+Returns a table object based on table id. */
+UNIV_INLINE
+dict_table_t*
+dict_table_get_on_id_low(
+/*=====================*/
+				/* out: table, NULL if does not exist */
+	dulint	table_id)	/* in: table id */
+{
+	dict_table_t*	table;
+	ulint		fold;
+
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+
+	/* Look for the table name in the hash table */
+	fold = ut_fold_dulint(table_id);
+
+	HASH_SEARCH(id_hash, dict_sys->table_id_hash, fold,
+		    dict_table_t*, table, ut_ad(table->cached),
+		    !ut_dulint_cmp(table->id, table_id));
+	if (table == NULL) {
+		table = dict_load_table_on_id(table_id);
+	}
+
+	ut_ad(!table || table->cached);
+
+	/* TODO: should get the type information from MySQL */
+
+	return(table);
+}
+
diff --git a/storage/xtradb/include/dict0load.h b/storage/xtradb/include/dict0load.h
new file mode 100644
index 00000000000..759cbcdb14a
--- /dev/null
+++ b/storage/xtradb/include/dict0load.h
@@ -0,0 +1,119 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Loads to the memory cache database object definitions
+from dictionary tables
+
+Created 4/24/1996 Heikki Tuuri
+*******************************************************/
+
+#ifndef dict0load_h
+#define dict0load_h
+
+#include "univ.i"
+#include "dict0types.h"
+#include "ut0byte.h"
+#include "mem0mem.h"
+
+/************************************************************************
+In a crash recovery we already have all the tablespace objects created.
+This function compares the space id information in the InnoDB data dictionary
+to what we already read with fil_load_single_table_tablespaces().
+
+In a normal startup, we create the tablespace objects for every table in
+InnoDB's data dictionary, if the corresponding .ibd file exists.
+We also scan the biggest space id, and store it to fil_system. */
+UNIV_INTERN
+void
+dict_check_tablespaces_and_store_max_id(
+/*====================================*/
+	ibool	in_crash_recovery);	/* in: are we doing a crash recovery */
+/************************************************************************
+Finds the first table name in the given database. */
+UNIV_INTERN
+char*
+dict_get_first_table_name_in_db(
+/*============================*/
+				/* out, own: table name, NULL if
+				does not exist; the caller must free
+				the memory in the string! */
+	const char*	name);	/* in: database name which ends to '/' */
+/************************************************************************
+Loads a table definition and also all its index definitions, and also
+the cluster definition if the table is a member in a cluster. Also loads
+all foreign key constraints where the foreign key is in the table or where
+a foreign key references columns in this table. */
+UNIV_INTERN
+dict_table_t*
+dict_load_table(
+/*============*/
+				/* out: table, NULL if does not exist;
+				if the table is stored in an .ibd file,
+				but the file does not exist,
+				then we set the ibd_file_missing flag TRUE
+				in the table object we return */
+	const char*	name);	/* in: table name in the
+				databasename/tablename format */
+/***************************************************************************
+Loads a table object based on the table id. */
+UNIV_INTERN
+dict_table_t*
+dict_load_table_on_id(
+/*==================*/
+				/* out: table; NULL if table does not exist */
+	dulint	table_id);	/* in: table id */
+/************************************************************************
+This function is called when the database is booted.
+Loads system table index definitions except for the clustered index which
+is added to the dictionary cache at booting before calling this function. */
+UNIV_INTERN
+void
+dict_load_sys_table(
+/*================*/
+	dict_table_t*	table);	/* in: system table */
+#ifndef UNIV_HOTBACKUP
+/***************************************************************************
+Loads foreign key constraints where the table is either the foreign key
+holder or where the table is referenced by a foreign key. Adds these
+constraints to the data dictionary. Note that we know that the dictionary
+cache already contains all constraints where the other relevant table is
+already in the dictionary cache. */
+UNIV_INTERN
+ulint
+dict_load_foreigns(
+/*===============*/
+					/* out: DB_SUCCESS or error code */
+	const char*	table_name,	/* in: table name */
+	ibool		check_charsets);/* in: TRUE=check charsets
+					compatibility */
+#endif /* !UNIV_HOTBACKUP */
+/************************************************************************
+Prints to the standard output information on all tables found in the data
+dictionary system table. */
+UNIV_INTERN
+void
+dict_print(void);
+/*============*/
+
+
+#ifndef UNIV_NONINL
+#include "dict0load.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/dict0load.ic b/storage/xtradb/include/dict0load.ic
new file mode 100644
index 00000000000..72eac2f621a
--- /dev/null
+++ b/storage/xtradb/include/dict0load.ic
@@ -0,0 +1,25 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Loads to the memory cache database object definitions
+from dictionary tables
+
+Created 4/24/1996 Heikki Tuuri
+*******************************************************/
+
diff --git a/storage/xtradb/include/dict0mem.h b/storage/xtradb/include/dict0mem.h
new file mode 100644
index 00000000000..e2b3cfa3679
--- /dev/null
+++ b/storage/xtradb/include/dict0mem.h
@@ -0,0 +1,501 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Data dictionary memory object creation
+
+Created 1/8/1996 Heikki Tuuri
+*******************************************************/
+
+#ifndef dict0mem_h
+#define dict0mem_h
+
+#include "univ.i"
+#include "dict0types.h"
+#include "data0type.h"
+#include "mem0mem.h"
+#include "rem0types.h"
+#include "btr0types.h"
+#include "ut0mem.h"
+#include "ut0lst.h"
+#include "ut0rnd.h"
+#include "ut0byte.h"
+#include "sync0rw.h"
+#include "lock0types.h"
+#include "hash0hash.h"
+#include "que0types.h"
+#include "trx0types.h"
+
+/* Type flags of an index: OR'ing of the flags is allowed to define a
+combination of types */
+#define DICT_CLUSTERED	1	/* clustered index */
+#define DICT_UNIQUE	2	/* unique index */
+#define	DICT_UNIVERSAL	4	/* index which can contain records from any
+				other index */
+#define	DICT_IBUF 	8	/* insert buffer tree */
+
+/* Types for a table object */
+#define DICT_TABLE_ORDINARY		1
+#if 0 /* not implemented */
+#define	DICT_TABLE_CLUSTER_MEMBER	2
+#define	DICT_TABLE_CLUSTER		3 /* this means that the table is
+					  really a cluster definition */
+#endif
+
+/* Table flags.  All unused bits must be 0. */
+#define DICT_TF_COMPACT			1	/* Compact page format.
+						This must be set for
+						new file formats
+						(later than
+						DICT_TF_FORMAT_51). */
+
+/* compressed page size (0=uncompressed, up to 15 compressed sizes) */
+#define DICT_TF_ZSSIZE_SHIFT		1
+#define DICT_TF_ZSSIZE_MASK		(15 << DICT_TF_ZSSIZE_SHIFT)
+#define DICT_TF_ZSSIZE_MAX (UNIV_PAGE_SIZE_SHIFT - PAGE_ZIP_MIN_SIZE_SHIFT + 1)
+
+
+#define DICT_TF_FORMAT_SHIFT		5	/* file format */
+#define DICT_TF_FORMAT_MASK		(127 << DICT_TF_FORMAT_SHIFT)
+#define DICT_TF_FORMAT_51		0	/* InnoDB/MySQL up to 5.1 */
+#define DICT_TF_FORMAT_ZIP		1	/* InnoDB plugin for 5.1:
+						compressed tables,
+						new BLOB treatment */
+#define DICT_TF_FORMAT_MAX		DICT_TF_FORMAT_ZIP
+
+#define DICT_TF_BITS			6	/* number of flag bits */
+#if (1 << (DICT_TF_BITS - DICT_TF_FORMAT_SHIFT)) <= DICT_TF_FORMAT_MAX
+# error "DICT_TF_BITS is insufficient for DICT_TF_FORMAT_MAX"
+#endif
+
+/**************************************************************************
+Creates a table memory object. */
+UNIV_INTERN
+dict_table_t*
+dict_mem_table_create(
+/*==================*/
+					/* out, own: table object */
+	const char*	name,		/* in: table name */
+	ulint		space,		/* in: space where the clustered index
+					of the table is placed; this parameter
+					is ignored if the table is made
+					a member of a cluster */
+	ulint		n_cols,		/* in: number of columns */
+	ulint		flags);		/* in: table flags */
+/********************************************************************
+Free a table memory object. */
+UNIV_INTERN
+void
+dict_mem_table_free(
+/*================*/
+	dict_table_t*	table);		/* in: table */
+/**************************************************************************
+Adds a column definition to a table. */
+UNIV_INTERN
+void
+dict_mem_table_add_col(
+/*===================*/
+	dict_table_t*	table,	/* in: table */
+	mem_heap_t*	heap,	/* in: temporary memory heap, or NULL */
+	const char*	name,	/* in: column name, or NULL */
+	ulint		mtype,	/* in: main datatype */
+	ulint		prtype,	/* in: precise type */
+	ulint		len);	/* in: precision */
+/**************************************************************************
+Creates an index memory object. */
+UNIV_INTERN
+dict_index_t*
+dict_mem_index_create(
+/*==================*/
+					/* out, own: index object */
+	const char*	table_name,	/* in: table name */
+	const char*	index_name,	/* in: index name */
+	ulint		space,		/* in: space where the index tree is
+					placed, ignored if the index is of
+					the clustered type */
+	ulint		type,		/* in: DICT_UNIQUE,
+					DICT_CLUSTERED, ... ORed */
+	ulint		n_fields);	/* in: number of fields */
+/**************************************************************************
+Adds a field definition to an index. NOTE: does not take a copy
+of the column name if the field is a column. The memory occupied
+by the column name may be released only after publishing the index. */
+UNIV_INTERN
+void
+dict_mem_index_add_field(
+/*=====================*/
+	dict_index_t*	index,		/* in: index */
+	const char*	name,		/* in: column name */
+	ulint		prefix_len);	/* in: 0 or the column prefix length
+					in a MySQL index like
+					INDEX (textcol(25)) */
+/**************************************************************************
+Frees an index memory object. */
+UNIV_INTERN
+void
+dict_mem_index_free(
+/*================*/
+	dict_index_t*	index);	/* in: index */
+/**************************************************************************
+Creates and initializes a foreign constraint memory object. */
+UNIV_INTERN
+dict_foreign_t*
+dict_mem_foreign_create(void);
+/*=========================*/
+				/* out, own: foreign constraint struct */
+
+/* Data structure for a column in a table */
+struct dict_col_struct{
+	/*----------------------*/
+	/* The following are copied from dtype_t,
+	so that all bit-fields can be packed tightly. */
+	unsigned	mtype:8;	/* main data type */
+	unsigned	prtype:24;	/* precise type; MySQL data
+					type, charset code, flags to
+					indicate nullability,
+					signedness, whether this is a
+					binary string, whether this is
+					a true VARCHAR where MySQL
+					uses 2 bytes to store the length */
+
+	/* the remaining fields do not affect alphabetical ordering: */
+
+	unsigned	len:16;		/* length; for MySQL data this
+					is field->pack_length(),
+					except that for a >= 5.0.3
+					type true VARCHAR this is the
+					maximum byte length of the
+					string data (in addition to
+					the string, MySQL uses 1 or 2
+					bytes to store the string length) */
+
+	unsigned	mbminlen:2;	/* minimum length of a
+					character, in bytes */
+	unsigned	mbmaxlen:3;	/* maximum length of a
+					character, in bytes */
+	/*----------------------*/
+	/* End of definitions copied from dtype_t */
+
+	unsigned	ind:10;		/* table column position
+					(starting from 0) */
+	unsigned	ord_part:1;	/* nonzero if this column
+					appears in the ordering fields
+					of an index */
+};
+
+/* DICT_MAX_INDEX_COL_LEN is measured in bytes and is the maximum
+indexed column length (or indexed prefix length). It is set to 3*256,
+so that one can create a column prefix index on 256 characters of a
+TEXT or VARCHAR column also in the UTF-8 charset. In that charset,
+a character may take at most 3 bytes.
+This constant MUST NOT BE CHANGED, or the compatibility of InnoDB data
+files would be at risk! */
+
+#define DICT_MAX_INDEX_COL_LEN		REC_MAX_INDEX_COL_LEN
+
+/* Data structure for a field in an index */
+struct dict_field_struct{
+	dict_col_t*	col;		/* pointer to the table column */
+	const char*	name;		/* name of the column */
+	unsigned	prefix_len:10;	/* 0 or the length of the column
+					prefix in bytes in a MySQL index of
+					type, e.g., INDEX (textcol(25));
+					must be smaller than
+					DICT_MAX_INDEX_COL_LEN; NOTE that
+					in the UTF-8 charset, MySQL sets this
+					to 3 * the prefix len in UTF-8 chars */
+	unsigned	fixed_len:10;	/* 0 or the fixed length of the
+					column if smaller than
+					DICT_MAX_INDEX_COL_LEN */
+};
+
+/* Data structure for an index.  Most fields will be
+initialized to 0, NULL or FALSE in dict_mem_index_create(). */
+struct dict_index_struct{
+	dulint		id;	/* id of the index */
+	mem_heap_t*	heap;	/* memory heap */
+	const char*	name;	/* index name */
+	const char*	table_name; /* table name */
+	dict_table_t*	table;	/* back pointer to table */
+	unsigned	space:32;
+				/* space where the index tree is placed */
+	unsigned	page:32;/* index tree root page number */
+	unsigned	type:4;	/* index type (DICT_CLUSTERED, DICT_UNIQUE,
+				DICT_UNIVERSAL, DICT_IBUF) */
+	unsigned	trx_id_offset:10;/* position of the trx id column
+				in a clustered index record, if the fields
+				before it are known to be of a fixed size,
+				0 otherwise */
+	unsigned	n_user_defined_cols:10;
+				/* number of columns the user defined to
+				be in the index: in the internal
+				representation we add more columns */
+	unsigned	n_uniq:10;/* number of fields from the beginning
+				which are enough to determine an index
+				entry uniquely */
+	unsigned	n_def:10;/* number of fields defined so far */
+	unsigned	n_fields:10;/* number of fields in the index */
+	unsigned	n_nullable:10;/* number of nullable fields */
+	unsigned	cached:1;/* TRUE if the index object is in the
+				dictionary cache */
+	unsigned	to_be_dropped:1;
+				/* TRUE if this index is marked to be
+				dropped in ha_innobase::prepare_drop_index(),
+				otherwise FALSE */
+	dict_field_t*	fields;	/* array of field descriptions */
+	UT_LIST_NODE_T(dict_index_t)
+			indexes;/* list of indexes of the table */
+	btr_search_t*	search_info; /* info used in optimistic searches */
+	/*----------------------*/
+	ib_int64_t*	stat_n_diff_key_vals;
+				/* approximate number of different key values
+				for this index, for each n-column prefix
+				where n <= dict_get_n_unique(index); we
+				periodically calculate new estimates */
+	ulint		stat_index_size;
+				/* approximate index size in database pages */
+	ulint		stat_n_leaf_pages;
+				/* approximate number of leaf pages in the
+				index tree */
+	rw_lock_t	lock;	/* read-write lock protecting the upper levels
+				of the index tree */
+#ifdef ROW_MERGE_IS_INDEX_USABLE
+	dulint		trx_id; /* id of the transaction that created this
+				index, or ut_dulint_zero if the index existed
+				when InnoDB was started up */
+#endif /* ROW_MERGE_IS_INDEX_USABLE */
+#ifdef UNIV_DEBUG
+	ulint		magic_n;/* magic number */
+# define DICT_INDEX_MAGIC_N	76789786
+#endif
+};
+
+/* Data structure for a foreign key constraint; an example:
+FOREIGN KEY (A, B) REFERENCES TABLE2 (C, D).  Most fields will be
+initialized to 0, NULL or FALSE in dict_mem_foreign_create(). */
+
+struct dict_foreign_struct{
+	mem_heap_t*	heap;		/* this object is allocated from
+					this memory heap */
+	char*		id;		/* id of the constraint as a
+					null-terminated string */
+	unsigned	n_fields:10;	/* number of indexes' first fields
+					for which the the foreign key
+					constraint is defined: we allow the
+					indexes to contain more fields than
+					mentioned in the constraint, as long
+					as the first fields are as mentioned */
+	unsigned	type:6;		/* 0 or DICT_FOREIGN_ON_DELETE_CASCADE
+					or DICT_FOREIGN_ON_DELETE_SET_NULL */
+	char*		foreign_table_name;/* foreign table name */
+	dict_table_t*	foreign_table;	/* table where the foreign key is */
+	const char**	foreign_col_names;/* names of the columns in the
+					foreign key */
+	char*		referenced_table_name;/* referenced table name */
+	dict_table_t*	referenced_table;/* table where the referenced key
+					is */
+	const char**	referenced_col_names;/* names of the referenced
+					columns in the referenced table */
+	dict_index_t*	foreign_index;	/* foreign index; we require that
+					both tables contain explicitly defined
+					indexes for the constraint: InnoDB
+					does not generate new indexes
+					implicitly */
+	dict_index_t*	referenced_index;/* referenced index */
+	UT_LIST_NODE_T(dict_foreign_t)
+			foreign_list;	/* list node for foreign keys of the
+					table */
+	UT_LIST_NODE_T(dict_foreign_t)
+			referenced_list;/* list node for referenced keys of the
+					table */
+};
+
+/* The flags for ON_UPDATE and ON_DELETE can be ORed; the default is that
+a foreign key constraint is enforced, therefore RESTRICT just means no flag */
+#define DICT_FOREIGN_ON_DELETE_CASCADE	1
+#define DICT_FOREIGN_ON_DELETE_SET_NULL	2
+#define DICT_FOREIGN_ON_UPDATE_CASCADE	4
+#define DICT_FOREIGN_ON_UPDATE_SET_NULL	8
+#define DICT_FOREIGN_ON_DELETE_NO_ACTION 16
+#define DICT_FOREIGN_ON_UPDATE_NO_ACTION 32
+
+
+/* Data structure for a database table.  Most fields will be
+initialized to 0, NULL or FALSE in dict_mem_table_create(). */
+struct dict_table_struct{
+	dulint		id;	/* id of the table */
+	mem_heap_t*	heap;	/* memory heap */
+	const char*	name;	/* table name */
+	const char*	dir_path_of_temp_table;/* NULL or the directory path
+				where a TEMPORARY table that was explicitly
+				created by a user should be placed if
+				innodb_file_per_table is defined in my.cnf;
+				in Unix this is usually /tmp/..., in Windows
+				\temp\... */
+	unsigned	space:32;
+				/* space where the clustered index of the
+				table is placed */
+	unsigned	flags:DICT_TF_BITS;/* DICT_TF_COMPACT, ... */
+	unsigned	ibd_file_missing:1;
+				/* TRUE if this is in a single-table
+				tablespace and the .ibd file is missing; then
+				we must return in ha_innodb.cc an error if the
+				user tries to query such an orphaned table */
+	unsigned	tablespace_discarded:1;
+				/* this flag is set TRUE when the user
+				calls DISCARD TABLESPACE on this
+				table, and reset to FALSE in IMPORT
+				TABLESPACE */
+	unsigned	cached:1;/* TRUE if the table object has been added
+				to the dictionary cache */
+	unsigned	n_def:10;/* number of columns defined so far */
+	unsigned	n_cols:10;/* number of columns */
+	dict_col_t*	cols;	/* array of column descriptions */
+	const char*	col_names;
+				/* Column names packed in a character string
+				"name1\0name2\0...nameN\0".  Until
+				the string contains n_cols, it will be
+				allocated from a temporary heap.  The final
+				string will be allocated from table->heap. */
+	hash_node_t	name_hash; /* hash chain node */
+	hash_node_t	id_hash; /* hash chain node */
+	UT_LIST_BASE_NODE_T(dict_index_t)
+			indexes; /* list of indexes of the table */
+	UT_LIST_BASE_NODE_T(dict_foreign_t)
+			foreign_list;/* list of foreign key constraints
+				in the table; these refer to columns
+				in other tables */
+	UT_LIST_BASE_NODE_T(dict_foreign_t)
+			referenced_list;/* list of foreign key constraints
+				which refer to this table */
+	UT_LIST_NODE_T(dict_table_t)
+			table_LRU; /* node of the LRU list of tables */
+	ulint		n_mysql_handles_opened;
+				/* count of how many handles MySQL has opened
+				to this table; dropping of the table is
+				NOT allowed until this count gets to zero;
+				MySQL does NOT itself check the number of
+				open handles at drop */
+	ulint		n_foreign_key_checks_running;
+				/* count of how many foreign key check
+				operations are currently being performed
+				on the table: we cannot drop the table while
+				there are foreign key checks running on
+				it! */
+	dulint		query_cache_inv_trx_id;
+				/* transactions whose trx id < than this
+				number are not allowed to store to the MySQL
+				query cache or retrieve from it; when a trx
+				with undo logs commits, it sets this to the
+				value of the trx id counter for the tables it
+				had an IX lock on */
+	UT_LIST_BASE_NODE_T(lock_t)
+			locks; /* list of locks on the table */
+#ifdef UNIV_DEBUG
+	/*----------------------*/
+	ibool		does_not_fit_in_memory;
+				/* this field is used to specify in simulations
+				tables which are so big that disk should be
+				accessed: disk access is simulated by
+				putting the thread to sleep for a while;
+				NOTE that this flag is not stored to the data
+				dictionary on disk, and the database will
+				forget about value TRUE if it has to reload
+				the table definition from disk */
+#endif /* UNIV_DEBUG */
+	/*----------------------*/
+	unsigned	big_rows:1;
+				/* flag: TRUE if the maximum length of
+				a single row exceeds BIG_ROW_SIZE;
+				initialized in dict_table_add_to_cache() */
+	unsigned	stat_initialized:1; /* TRUE if statistics have
+				been calculated the first time
+				after database startup or table creation */
+	ib_int64_t	stat_n_rows;
+				/* approximate number of rows in the table;
+				we periodically calculate new estimates */
+	ulint		stat_clustered_index_size;
+				/* approximate clustered index size in
+				database pages */
+	ulint		stat_sum_of_other_index_sizes;
+				/* other indexes in database pages */
+	ulint		stat_modified_counter;
+				/* when a row is inserted, updated, or deleted,
+				we add 1 to this number; we calculate new
+				estimates for the stat_... values for the
+				table and the indexes at an interval of 2 GB
+				or when about 1 / 16 of table has been
+				modified; also when the estimate operation is
+				called for MySQL SHOW TABLE STATUS; the
+				counter is reset to zero at statistics
+				calculation; this counter is not protected by
+				any latch, because this is only used for
+				heuristics */
+	/*----------------------*/
+				/* The following fields are used by the
+				AUTOINC code.  The actual collection of
+				tables locked during AUTOINC read/write is
+				kept in trx_t. In order to quickly determine
+				whether a transaction has locked the AUTOINC
+				lock we keep a pointer to the transaction
+				here in the autoinc_trx variable. This is to
+				avoid acquiring the kernel mutex and scanning
+				the vector in trx_t.
+
+				When an AUTOINC lock has to wait, the
+				corresponding lock instance is created on
+				the trx lock heap rather than use the
+				pre-allocated instance in autoinc_lock below.*/
+	lock_t*		autoinc_lock;
+				/* a buffer for an AUTOINC lock
+				for this table: we allocate the memory here
+				so that individual transactions can get it
+				and release it without a need to allocate
+				space from the lock heap of the trx:
+				otherwise the lock heap would grow rapidly
+				if we do a large insert from a select */
+	mutex_t		autoinc_mutex;
+				/* mutex protecting the autoincrement
+				counter */
+	ib_uint64_t	autoinc;/* autoinc counter value to give to the
+				next inserted row */
+	ulong		n_waiting_or_granted_auto_inc_locks;
+				/* This counter is used to track the number
+				of granted and pending autoinc locks on this
+				table. This value is set after acquiring the
+				kernel mutex but we peek the contents to
+				determine whether other transactions have
+				acquired the AUTOINC lock or not. Of course
+				only one transaction can be granted the
+				lock but there can be multiple waiters. */
+	const trx_t*		autoinc_trx;
+				/* The transaction that currently holds the
+				the AUTOINC lock on this table. */
+	/*----------------------*/
+
+#ifdef UNIV_DEBUG
+	ulint		magic_n;/* magic number */
+# define DICT_TABLE_MAGIC_N	76333786
+#endif /* UNIV_DEBUG */
+};
+
+#ifndef UNIV_NONINL
+#include "dict0mem.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/dict0mem.ic b/storage/xtradb/include/dict0mem.ic
new file mode 100644
index 00000000000..6916393a9cd
--- /dev/null
+++ b/storage/xtradb/include/dict0mem.ic
@@ -0,0 +1,25 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/**********************************************************************
+Data dictionary memory object creation
+
+Created 1/8/1996 Heikki Tuuri
+***********************************************************************/
+
+
diff --git a/storage/xtradb/include/dict0types.h b/storage/xtradb/include/dict0types.h
new file mode 100644
index 00000000000..b93e995e01b
--- /dev/null
+++ b/storage/xtradb/include/dict0types.h
@@ -0,0 +1,45 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Data dictionary global types
+
+Created 1/8/1996 Heikki Tuuri
+*******************************************************/
+
+#ifndef dict0types_h
+#define dict0types_h
+
+#include "ut0list.h"
+
+typedef struct dict_sys_struct		dict_sys_t;
+typedef struct dict_col_struct		dict_col_t;
+typedef struct dict_field_struct	dict_field_t;
+typedef struct dict_index_struct	dict_index_t;
+typedef struct dict_table_struct	dict_table_t;
+typedef struct dict_foreign_struct	dict_foreign_t;
+
+/* A cluster object is a table object with the type field set to
+DICT_CLUSTERED */
+
+typedef dict_table_t			dict_cluster_t;
+
+typedef struct ind_node_struct		ind_node_t;
+typedef struct tab_node_struct		tab_node_t;
+
+#endif
diff --git a/storage/xtradb/include/dyn0dyn.h b/storage/xtradb/include/dyn0dyn.h
new file mode 100644
index 00000000000..c06d6b88d2f
--- /dev/null
+++ b/storage/xtradb/include/dyn0dyn.h
@@ -0,0 +1,182 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The dynamically allocated array
+
+Created 2/5/1996 Heikki Tuuri
+*******************************************************/
+
+#ifndef dyn0dyn_h
+#define dyn0dyn_h
+
+#include "univ.i"
+#include "ut0lst.h"
+#include "mem0mem.h"
+
+typedef struct dyn_block_struct		dyn_block_t;
+typedef dyn_block_t			dyn_array_t;
+
+
+/* This is the initial 'payload' size of a dynamic array;
+this must be > MLOG_BUF_MARGIN + 30! */
+#define	DYN_ARRAY_DATA_SIZE	512
+
+/*************************************************************************
+Initializes a dynamic array. */
+UNIV_INLINE
+dyn_array_t*
+dyn_array_create(
+/*=============*/
+				/* out: initialized dyn array */
+	dyn_array_t*	arr);	/* in: pointer to a memory buffer of
+				size sizeof(dyn_array_t) */
+/****************************************************************
+Frees a dynamic array. */
+UNIV_INLINE
+void
+dyn_array_free(
+/*===========*/
+	dyn_array_t*	arr);	/* in: dyn array */
+/*************************************************************************
+Makes room on top of a dyn array and returns a pointer to a buffer in it.
+After copying the elements, the caller must close the buffer using
+dyn_array_close. */
+UNIV_INLINE
+byte*
+dyn_array_open(
+/*===========*/
+				/* out: pointer to the buffer */
+	dyn_array_t*	arr,	/* in: dynamic array */
+	ulint		size);	/* in: size in bytes of the buffer; MUST be
+				smaller than DYN_ARRAY_DATA_SIZE! */
+/*************************************************************************
+Closes the buffer returned by dyn_array_open. */
+UNIV_INLINE
+void
+dyn_array_close(
+/*============*/
+	dyn_array_t*	arr,	/* in: dynamic array */
+	byte*		ptr);	/* in: buffer space from ptr up was not used */
+/*************************************************************************
+Makes room on top of a dyn array and returns a pointer to
+the added element. The caller must copy the element to
+the pointer returned. */
+UNIV_INLINE
+void*
+dyn_array_push(
+/*===========*/
+				/* out: pointer to the element */
+	dyn_array_t*	arr,	/* in: dynamic array */
+	ulint		size);	/* in: size in bytes of the element */
+/****************************************************************
+Returns pointer to an element in dyn array. */
+UNIV_INLINE
+void*
+dyn_array_get_element(
+/*==================*/
+				/* out: pointer to element */
+	dyn_array_t*	arr,	/* in: dyn array */
+	ulint		pos);	/* in: position of element as bytes
+				from array start */
+/****************************************************************
+Returns the size of stored data in a dyn array. */
+UNIV_INLINE
+ulint
+dyn_array_get_data_size(
+/*====================*/
+				/* out: data size in bytes */
+	dyn_array_t*	arr);	/* in: dyn array */
+/****************************************************************
+Gets the first block in a dyn array. */
+UNIV_INLINE
+dyn_block_t*
+dyn_array_get_first_block(
+/*======================*/
+	dyn_array_t*	arr);	/* in: dyn array */
+/****************************************************************
+Gets the last block in a dyn array. */
+UNIV_INLINE
+dyn_block_t*
+dyn_array_get_last_block(
+/*=====================*/
+	dyn_array_t*	arr);	/* in: dyn array */
+/************************************************************************
+Gets the next block in a dyn array. */
+UNIV_INLINE
+dyn_block_t*
+dyn_array_get_next_block(
+/*=====================*/
+				/* out: pointer to next, NULL if end of list */
+	dyn_array_t*	arr,	/* in: dyn array */
+	dyn_block_t*	block);	/* in: dyn array block */
+/************************************************************************
+Gets the number of used bytes in a dyn array block. */
+UNIV_INLINE
+ulint
+dyn_block_get_used(
+/*===============*/
+				/* out: number of bytes used */
+	dyn_block_t*	block);	/* in: dyn array block */
+/************************************************************************
+Gets pointer to the start of data in a dyn array block. */
+UNIV_INLINE
+byte*
+dyn_block_get_data(
+/*===============*/
+				/* out: pointer to data */
+	dyn_block_t*	block);	/* in: dyn array block */
+/************************************************************
+Pushes n bytes to a dyn array. */
+UNIV_INLINE
+void
+dyn_push_string(
+/*============*/
+	dyn_array_t*	arr,	/* in: dyn array */
+	const byte*	str,	/* in: string to write */
+	ulint		len);	/* in: string length */
+
+/*#################################################################*/
+
+/* NOTE! Do not use the fields of the struct directly: the definition
+appears here only for the compiler to know its size! */
+struct dyn_block_struct{
+	mem_heap_t*	heap;	/* in the first block this is != NULL
+				if dynamic allocation has been needed */
+	ulint		used;	/* number of data bytes used in this block */
+	byte		data[DYN_ARRAY_DATA_SIZE];
+				/* storage for array elements */
+	UT_LIST_BASE_NODE_T(dyn_block_t) base;
+				/* linear list of dyn blocks: this node is
+				used only in the first block */
+	UT_LIST_NODE_T(dyn_block_t) list;
+				/* linear list node: used in all blocks */
+#ifdef UNIV_DEBUG
+	ulint		buf_end;/* only in the debug version: if dyn array is
+				opened, this is the buffer end offset, else
+				this is 0 */
+	ulint		magic_n;
+#endif
+};
+
+
+#ifndef UNIV_NONINL
+#include "dyn0dyn.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/dyn0dyn.ic b/storage/xtradb/include/dyn0dyn.ic
new file mode 100644
index 00000000000..1ef8b284a99
--- /dev/null
+++ b/storage/xtradb/include/dyn0dyn.ic
@@ -0,0 +1,362 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The dynamically allocated array
+
+Created 2/5/1996 Heikki Tuuri
+*******************************************************/
+
+#define DYN_BLOCK_MAGIC_N	375767
+#define DYN_BLOCK_FULL_FLAG	0x1000000UL
+
+/****************************************************************
+Adds a new block to a dyn array. */
+UNIV_INTERN
+dyn_block_t*
+dyn_array_add_block(
+/*================*/
+				/* out: created block */
+	dyn_array_t*	arr);	/* in: dyn array */
+
+
+/****************************************************************
+Gets the first block in a dyn array. */
+UNIV_INLINE
+dyn_block_t*
+dyn_array_get_first_block(
+/*======================*/
+	dyn_array_t*	arr)	/* in: dyn array */
+{
+	return(arr);
+}
+
+/****************************************************************
+Gets the last block in a dyn array. */
+UNIV_INLINE
+dyn_block_t*
+dyn_array_get_last_block(
+/*=====================*/
+	dyn_array_t*	arr)	/* in: dyn array */
+{
+	if (arr->heap == NULL) {
+
+		return(arr);
+	}
+
+	return(UT_LIST_GET_LAST(arr->base));
+}
+
+/************************************************************************
+Gets the next block in a dyn array. */
+UNIV_INLINE
+dyn_block_t*
+dyn_array_get_next_block(
+/*=====================*/
+				/* out: pointer to next, NULL if end of list */
+	dyn_array_t*	arr,	/* in: dyn array */
+	dyn_block_t*	block)	/* in: dyn array block */
+{
+	ut_ad(arr && block);
+
+	if (arr->heap == NULL) {
+		ut_ad(arr == block);
+
+		return(NULL);
+	}
+
+	return(UT_LIST_GET_NEXT(list, block));
+}
+
+/************************************************************************
+Gets the number of used bytes in a dyn array block. */
+UNIV_INLINE
+ulint
+dyn_block_get_used(
+/*===============*/
+				/* out: number of bytes used */
+	dyn_block_t*	block)	/* in: dyn array block */
+{
+	ut_ad(block);
+
+	return((block->used) & ~DYN_BLOCK_FULL_FLAG);
+}
+
+/************************************************************************
+Gets pointer to the start of data in a dyn array block. */
+UNIV_INLINE
+byte*
+dyn_block_get_data(
+/*===============*/
+				/* out: pointer to data */
+	dyn_block_t*	block)	/* in: dyn array block */
+{
+	ut_ad(block);
+
+	return(block->data);
+}
+
+/*************************************************************************
+Initializes a dynamic array. */
+UNIV_INLINE
+dyn_array_t*
+dyn_array_create(
+/*=============*/
+				/* out: initialized dyn array */
+	dyn_array_t*	arr)	/* in: pointer to a memory buffer of
+				size sizeof(dyn_array_t) */
+{
+	ut_ad(arr);
+#if DYN_ARRAY_DATA_SIZE >= DYN_BLOCK_FULL_FLAG
+# error "DYN_ARRAY_DATA_SIZE >= DYN_BLOCK_FULL_FLAG"
+#endif
+
+	arr->heap = NULL;
+	arr->used = 0;
+
+#ifdef UNIV_DEBUG
+	arr->buf_end = 0;
+	arr->magic_n = DYN_BLOCK_MAGIC_N;
+#endif
+	return(arr);
+}
+
+/****************************************************************
+Frees a dynamic array. */
+UNIV_INLINE
+void
+dyn_array_free(
+/*===========*/
+	dyn_array_t*	arr)	/* in: dyn array */
+{
+	if (arr->heap != NULL) {
+		mem_heap_free(arr->heap);
+	}
+
+#ifdef UNIV_DEBUG
+	arr->magic_n = 0;
+#endif
+}
+
+/*************************************************************************
+Makes room on top of a dyn array and returns a pointer to the added element.
+The caller must copy the element to the pointer returned. */
+UNIV_INLINE
+void*
+dyn_array_push(
+/*===========*/
+				/* out: pointer to the element */
+	dyn_array_t*	arr,	/* in: dynamic array */
+	ulint		size)	/* in: size in bytes of the element */
+{
+	dyn_block_t*	block;
+	ulint		used;
+
+	ut_ad(arr);
+	ut_ad(arr->magic_n == DYN_BLOCK_MAGIC_N);
+	ut_ad(size <= DYN_ARRAY_DATA_SIZE);
+	ut_ad(size);
+
+	block = arr;
+	used = block->used;
+
+	if (used + size > DYN_ARRAY_DATA_SIZE) {
+		/* Get the last array block */
+
+		block = dyn_array_get_last_block(arr);
+		used = block->used;
+
+		if (used + size > DYN_ARRAY_DATA_SIZE) {
+			block = dyn_array_add_block(arr);
+			used = block->used;
+		}
+	}
+
+	block->used = used + size;
+	ut_ad(block->used <= DYN_ARRAY_DATA_SIZE);
+
+	return((block->data) + used);
+}
+
+/*************************************************************************
+Makes room on top of a dyn array and returns a pointer to a buffer in it.
+After copying the elements, the caller must close the buffer using
+dyn_array_close. */
+UNIV_INLINE
+byte*
+dyn_array_open(
+/*===========*/
+				/* out: pointer to the buffer */
+	dyn_array_t*	arr,	/* in: dynamic array */
+	ulint		size)	/* in: size in bytes of the buffer; MUST be
+				smaller than DYN_ARRAY_DATA_SIZE! */
+{
+	dyn_block_t*	block;
+	ulint		used;
+
+	ut_ad(arr);
+	ut_ad(arr->magic_n == DYN_BLOCK_MAGIC_N);
+	ut_ad(size <= DYN_ARRAY_DATA_SIZE);
+	ut_ad(size);
+
+	block = arr;
+	used = block->used;
+
+	if (used + size > DYN_ARRAY_DATA_SIZE) {
+		/* Get the last array block */
+
+		block = dyn_array_get_last_block(arr);
+		used = block->used;
+
+		if (used + size > DYN_ARRAY_DATA_SIZE) {
+			block = dyn_array_add_block(arr);
+			used = block->used;
+			ut_a(size <= DYN_ARRAY_DATA_SIZE);
+		}
+	}
+
+	ut_ad(block->used <= DYN_ARRAY_DATA_SIZE);
+#ifdef UNIV_DEBUG
+	ut_ad(arr->buf_end == 0);
+
+	arr->buf_end = used + size;
+#endif
+	return((block->data) + used);
+}
+
+/*************************************************************************
+Closes the buffer returned by dyn_array_open. */
+UNIV_INLINE
+void
+dyn_array_close(
+/*============*/
+	dyn_array_t*	arr,	/* in: dynamic array */
+	byte*		ptr)	/* in: buffer space from ptr up was not used */
+{
+	dyn_block_t*	block;
+
+	ut_ad(arr);
+	ut_ad(arr->magic_n == DYN_BLOCK_MAGIC_N);
+
+	block = dyn_array_get_last_block(arr);
+
+	ut_ad(arr->buf_end + block->data >= ptr);
+
+	block->used = ptr - block->data;
+
+	ut_ad(block->used <= DYN_ARRAY_DATA_SIZE);
+
+#ifdef UNIV_DEBUG
+	arr->buf_end = 0;
+#endif
+}
+
+/****************************************************************
+Returns pointer to an element in dyn array. */
+UNIV_INLINE
+void*
+dyn_array_get_element(
+/*==================*/
+				/* out: pointer to element */
+	dyn_array_t*	arr,	/* in: dyn array */
+	ulint		pos)	/* in: position of element as bytes
+				from array start */
+{
+	dyn_block_t*	block;
+	ulint		used;
+
+	ut_ad(arr);
+	ut_ad(arr->magic_n == DYN_BLOCK_MAGIC_N);
+
+	/* Get the first array block */
+	block = dyn_array_get_first_block(arr);
+
+	if (arr->heap != NULL) {
+		used = dyn_block_get_used(block);
+
+		while (pos >= used) {
+			pos -= used;
+			block = UT_LIST_GET_NEXT(list, block);
+			ut_ad(block);
+
+			used = dyn_block_get_used(block);
+		}
+	}
+
+	ut_ad(block);
+	ut_ad(dyn_block_get_used(block) >= pos);
+
+	return(block->data + pos);
+}
+
+/****************************************************************
+Returns the size of stored data in a dyn array. */
+UNIV_INLINE
+ulint
+dyn_array_get_data_size(
+/*====================*/
+				/* out: data size in bytes */
+	dyn_array_t*	arr)	/* in: dyn array */
+{
+	dyn_block_t*	block;
+	ulint		sum	= 0;
+
+	ut_ad(arr);
+	ut_ad(arr->magic_n == DYN_BLOCK_MAGIC_N);
+
+	if (arr->heap == NULL) {
+
+		return(arr->used);
+	}
+
+	/* Get the first array block */
+	block = dyn_array_get_first_block(arr);
+
+	while (block != NULL) {
+		sum += dyn_block_get_used(block);
+		block = dyn_array_get_next_block(arr, block);
+	}
+
+	return(sum);
+}
+
+/************************************************************
+Pushes n bytes to a dyn array. */
+UNIV_INLINE
+void
+dyn_push_string(
+/*============*/
+	dyn_array_t*	arr,	/* in: dyn array */
+	const byte*	str,	/* in: string to write */
+	ulint		len)	/* in: string length */
+{
+	ulint	n_copied;
+
+	while (len > 0) {
+		if (len > DYN_ARRAY_DATA_SIZE) {
+			n_copied = DYN_ARRAY_DATA_SIZE;
+		} else {
+			n_copied = len;
+		}
+
+		memcpy(dyn_array_push(arr, n_copied), str, n_copied);
+
+		str += n_copied;
+		len -= n_copied;
+	}
+}
diff --git a/storage/xtradb/include/eval0eval.h b/storage/xtradb/include/eval0eval.h
new file mode 100644
index 00000000000..75cf9b38c3a
--- /dev/null
+++ b/storage/xtradb/include/eval0eval.h
@@ -0,0 +1,113 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+SQL evaluator: evaluates simple data structures, like expressions, in
+a query graph
+
+Created 12/29/1997 Heikki Tuuri
+*******************************************************/
+
+#ifndef eval0eval_h
+#define eval0eval_h
+
+#include "univ.i"
+#include "que0types.h"
+#include "pars0sym.h"
+#include "pars0pars.h"
+
+/*********************************************************************
+Free the buffer from global dynamic memory for a value of a que_node,
+if it has been allocated in the above function. The freeing for pushed
+column values is done in sel_col_prefetch_buf_free. */
+UNIV_INTERN
+void
+eval_node_free_val_buf(
+/*===================*/
+	que_node_t*	node);	/* in: query graph node */
+/*********************************************************************
+Evaluates a symbol table symbol. */
+UNIV_INLINE
+void
+eval_sym(
+/*=====*/
+	sym_node_t*	sym_node);	/* in: symbol table node */
+/*********************************************************************
+Evaluates an expression. */
+UNIV_INLINE
+void
+eval_exp(
+/*=====*/
+	que_node_t*	exp_node);	/* in: expression */
+/*********************************************************************
+Sets an integer value as the value of an expression node. */
+UNIV_INLINE
+void
+eval_node_set_int_val(
+/*==================*/
+	que_node_t*	node,	/* in: expression node */
+	lint		val);	/* in: value to set */
+/*********************************************************************
+Gets an integer value from an expression node. */
+UNIV_INLINE
+lint
+eval_node_get_int_val(
+/*==================*/
+				/* out: integer value */
+	que_node_t*	node);	/* in: expression node */
+/*********************************************************************
+Copies a binary string value as the value of a query graph node. Allocates a
+new buffer if necessary. */
+UNIV_INLINE
+void
+eval_node_copy_and_alloc_val(
+/*=========================*/
+	que_node_t*	node,	/* in: query graph node */
+	const byte*	str,	/* in: binary string */
+	ulint		len);	/* in: string length or UNIV_SQL_NULL */
+/*********************************************************************
+Copies a query node value to another node. */
+UNIV_INLINE
+void
+eval_node_copy_val(
+/*===============*/
+	que_node_t*	node1,	/* in: node to copy to */
+	que_node_t*	node2);	/* in: node to copy from */
+/*********************************************************************
+Gets a iboolean value from a query node. */
+UNIV_INLINE
+ibool
+eval_node_get_ibool_val(
+/*====================*/
+				/* out: iboolean value */
+	que_node_t*	node);	/* in: query graph node */
+/*********************************************************************
+Evaluates a comparison node. */
+UNIV_INTERN
+ibool
+eval_cmp(
+/*=====*/
+					/* out: the result of the comparison */
+	func_node_t*	cmp_node);	/* in: comparison node */
+
+
+#ifndef UNIV_NONINL
+#include "eval0eval.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/eval0eval.ic b/storage/xtradb/include/eval0eval.ic
new file mode 100644
index 00000000000..a6330ae441f
--- /dev/null
+++ b/storage/xtradb/include/eval0eval.ic
@@ -0,0 +1,250 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+SQL evaluator: evaluates simple data structures, like expressions, in
+a query graph
+
+Created 12/29/1997 Heikki Tuuri
+*******************************************************/
+
+#include "que0que.h"
+#include "rem0cmp.h"
+#include "pars0grm.h"
+
+/*********************************************************************
+Evaluates a function node. */
+UNIV_INTERN
+void
+eval_func(
+/*======*/
+	func_node_t*	func_node);	/* in: function node */
+/*********************************************************************
+Allocate a buffer from global dynamic memory for a value of a que_node.
+NOTE that this memory must be explicitly freed when the query graph is
+freed. If the node already has allocated buffer, that buffer is freed
+here. NOTE that this is the only function where dynamic memory should be
+allocated for a query node val field. */
+UNIV_INTERN
+byte*
+eval_node_alloc_val_buf(
+/*====================*/
+				/* out: pointer to allocated buffer */
+	que_node_t*	node,	/* in: query graph node; sets the val field
+				data field to point to the new buffer, and
+				len field equal to size */
+	ulint		size);	/* in: buffer size */
+
+
+/*********************************************************************
+Allocates a new buffer if needed. */
+UNIV_INLINE
+byte*
+eval_node_ensure_val_buf(
+/*=====================*/
+				/* out: pointer to buffer */
+	que_node_t*	node,	/* in: query graph node; sets the val field
+				data field to point to the new buffer, and
+				len field equal to size */
+	ulint		size)	/* in: buffer size */
+{
+	dfield_t*	dfield;
+	byte*		data;
+
+	dfield = que_node_get_val(node);
+	dfield_set_len(dfield, size);
+
+	data = dfield_get_data(dfield);
+
+	if (!data || que_node_get_val_buf_size(node) < size) {
+
+		data = eval_node_alloc_val_buf(node, size);
+	}
+
+	return(data);
+}
+
+/*********************************************************************
+Evaluates a symbol table symbol. */
+UNIV_INLINE
+void
+eval_sym(
+/*=====*/
+	sym_node_t*	sym_node)	/* in: symbol table node */
+{
+
+	ut_ad(que_node_get_type(sym_node) == QUE_NODE_SYMBOL);
+
+	if (sym_node->indirection) {
+		/* The symbol table node is an alias for a variable or a
+		column */
+
+		dfield_copy_data(que_node_get_val(sym_node),
+				 que_node_get_val(sym_node->indirection));
+	}
+}
+
+/*********************************************************************
+Evaluates an expression. */
+UNIV_INLINE
+void
+eval_exp(
+/*=====*/
+	que_node_t*	exp_node)	/* in: expression */
+{
+	if (que_node_get_type(exp_node) == QUE_NODE_SYMBOL) {
+
+		eval_sym((sym_node_t*)exp_node);
+
+		return;
+	}
+
+	eval_func(exp_node);
+}
+
+/*********************************************************************
+Sets an integer value as the value of an expression node. */
+UNIV_INLINE
+void
+eval_node_set_int_val(
+/*==================*/
+	que_node_t*	node,	/* in: expression node */
+	lint		val)	/* in: value to set */
+{
+	dfield_t*	dfield;
+	byte*		data;
+
+	dfield = que_node_get_val(node);
+
+	data = dfield_get_data(dfield);
+
+	if (data == NULL) {
+		data = eval_node_alloc_val_buf(node, 4);
+	}
+
+	ut_ad(dfield_get_len(dfield) == 4);
+
+	mach_write_to_4(data, (ulint)val);
+}
+
+/*********************************************************************
+Gets an integer non-SQL null value from an expression node. */
+UNIV_INLINE
+lint
+eval_node_get_int_val(
+/*==================*/
+				/* out: integer value */
+	que_node_t*	node)	/* in: expression node */
+{
+	dfield_t*	dfield;
+
+	dfield = que_node_get_val(node);
+
+	ut_ad(dfield_get_len(dfield) == 4);
+
+	return((int)mach_read_from_4(dfield_get_data(dfield)));
+}
+
+/*********************************************************************
+Gets a iboolean value from a query node. */
+UNIV_INLINE
+ibool
+eval_node_get_ibool_val(
+/*====================*/
+				/* out: iboolean value */
+	que_node_t*	node)	/* in: query graph node */
+{
+	dfield_t*	dfield;
+	byte*		data;
+
+	dfield = que_node_get_val(node);
+
+	data = dfield_get_data(dfield);
+
+	ut_ad(data != NULL);
+
+	return(mach_read_from_1(data));
+}
+
+/*********************************************************************
+Sets a iboolean value as the value of a function node. */
+UNIV_INLINE
+void
+eval_node_set_ibool_val(
+/*====================*/
+	func_node_t*	func_node,	/* in: function node */
+	ibool		val)		/* in: value to set */
+{
+	dfield_t*	dfield;
+	byte*		data;
+
+	dfield = que_node_get_val(func_node);
+
+	data = dfield_get_data(dfield);
+
+	if (data == NULL) {
+		/* Allocate 1 byte to hold the value */
+
+		data = eval_node_alloc_val_buf(func_node, 1);
+	}
+
+	ut_ad(dfield_get_len(dfield) == 1);
+
+	mach_write_to_1(data, val);
+}
+
+/*********************************************************************
+Copies a binary string value as the value of a query graph node. Allocates a
+new buffer if necessary. */
+UNIV_INLINE
+void
+eval_node_copy_and_alloc_val(
+/*=========================*/
+	que_node_t*	node,	/* in: query graph node */
+	const byte*	str,	/* in: binary string */
+	ulint		len)	/* in: string length or UNIV_SQL_NULL */
+{
+	byte*		data;
+
+	if (len == UNIV_SQL_NULL) {
+		dfield_set_len(que_node_get_val(node), len);
+
+		return;
+	}
+
+	data = eval_node_ensure_val_buf(node, len);
+
+	ut_memcpy(data, str, len);
+}
+
+/*********************************************************************
+Copies a query node value to another node. */
+UNIV_INLINE
+void
+eval_node_copy_val(
+/*===============*/
+	que_node_t*	node1,	/* in: node to copy to */
+	que_node_t*	node2)	/* in: node to copy from */
+{
+	dfield_t*	dfield2;
+
+	dfield2 = que_node_get_val(node2);
+
+	eval_node_copy_and_alloc_val(node1, dfield_get_data(dfield2),
+				     dfield_get_len(dfield2));
+}
diff --git a/storage/xtradb/include/eval0proc.h b/storage/xtradb/include/eval0proc.h
new file mode 100644
index 00000000000..58937c18124
--- /dev/null
+++ b/storage/xtradb/include/eval0proc.h
@@ -0,0 +1,103 @@
+/*****************************************************************************
+
+Copyright (c) 1998, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Executes SQL stored procedures and their control structures
+
+Created 1/20/1998 Heikki Tuuri
+*******************************************************/
+
+#ifndef eval0proc_h
+#define eval0proc_h
+
+#include "univ.i"
+#include "que0types.h"
+#include "pars0sym.h"
+#include "pars0pars.h"
+
+/**************************************************************************
+Performs an execution step of a procedure node. */
+UNIV_INLINE
+que_thr_t*
+proc_step(
+/*======*/
+				/* out: query thread to run next or NULL */
+	que_thr_t*	thr);	/* in: query thread */
+/**************************************************************************
+Performs an execution step of an if-statement node. */
+UNIV_INTERN
+que_thr_t*
+if_step(
+/*====*/
+				/* out: query thread to run next or NULL */
+	que_thr_t*	thr);	/* in: query thread */
+/**************************************************************************
+Performs an execution step of a while-statement node. */
+UNIV_INTERN
+que_thr_t*
+while_step(
+/*=======*/
+				/* out: query thread to run next or NULL */
+	que_thr_t*	thr);	/* in: query thread */
+/**************************************************************************
+Performs an execution step of a for-loop node. */
+UNIV_INTERN
+que_thr_t*
+for_step(
+/*=====*/
+				/* out: query thread to run next or NULL */
+	que_thr_t*	thr);	/* in: query thread */
+/**************************************************************************
+Performs an execution step of an assignment statement node. */
+UNIV_INTERN
+que_thr_t*
+assign_step(
+/*========*/
+				/* out: query thread to run next or NULL */
+	que_thr_t*	thr);	/* in: query thread */
+/**************************************************************************
+Performs an execution step of a procedure call node. */
+UNIV_INLINE
+que_thr_t*
+proc_eval_step(
+/*===========*/
+				/* out: query thread to run next or NULL */
+	que_thr_t*	thr);	/* in: query thread */
+/**************************************************************************
+Performs an execution step of an exit statement node. */
+UNIV_INTERN
+que_thr_t*
+exit_step(
+/*======*/
+				/* out: query thread to run next or NULL */
+	que_thr_t*	thr);	/* in: query thread */
+/**************************************************************************
+Performs an execution step of a return-statement node. */
+UNIV_INTERN
+que_thr_t*
+return_step(
+/*========*/
+				/* out: query thread to run next or NULL */
+	que_thr_t*	thr);	/* in: query thread */
+
+
+#ifndef UNIV_NONINL
+#include "eval0proc.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/eval0proc.ic b/storage/xtradb/include/eval0proc.ic
new file mode 100644
index 00000000000..6bd978ad3fc
--- /dev/null
+++ b/storage/xtradb/include/eval0proc.ic
@@ -0,0 +1,87 @@
+/*****************************************************************************
+
+Copyright (c) 1998, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Executes SQL stored procedures and their control structures
+
+Created 1/20/1998 Heikki Tuuri
+*******************************************************/
+
+#include "pars0pars.h"
+#include "que0que.h"
+#include "eval0eval.h"
+
+/**************************************************************************
+Performs an execution step of a procedure node. */
+UNIV_INLINE
+que_thr_t*
+proc_step(
+/*======*/
+				/* out: query thread to run next or NULL */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	proc_node_t*	node;
+
+	ut_ad(thr);
+
+	node = thr->run_node;
+	ut_ad(que_node_get_type(node) == QUE_NODE_PROC);
+
+	if (thr->prev_node == que_node_get_parent(node)) {
+		/* Start execution from the first statement in the statement
+		list */
+
+		thr->run_node = node->stat_list;
+	} else {
+		/* Move to the next statement */
+		ut_ad(que_node_get_next(thr->prev_node) == NULL);
+
+		thr->run_node = NULL;
+	}
+
+	if (thr->run_node == NULL) {
+		thr->run_node = que_node_get_parent(node);
+	}
+
+	return(thr);
+}
+
+/**************************************************************************
+Performs an execution step of a procedure call node. */
+UNIV_INLINE
+que_thr_t*
+proc_eval_step(
+/*===========*/
+				/* out: query thread to run next or NULL */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	func_node_t*	node;
+
+	ut_ad(thr);
+
+	node = thr->run_node;
+	ut_ad(que_node_get_type(node) == QUE_NODE_FUNC);
+
+	/* Evaluate the procedure */
+
+	eval_exp(node);
+
+	thr->run_node = que_node_get_parent(node);
+
+	return(thr);
+}
diff --git a/storage/xtradb/include/fil0fil.h b/storage/xtradb/include/fil0fil.h
new file mode 100644
index 00000000000..587e5ee48a8
--- /dev/null
+++ b/storage/xtradb/include/fil0fil.h
@@ -0,0 +1,711 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The low-level file system
+
+Created 10/25/1995 Heikki Tuuri
+*******************************************************/
+
+#ifndef fil0fil_h
+#define fil0fil_h
+
+#include "univ.i"
+#include "sync0rw.h"
+#include "dict0types.h"
+#include "ut0byte.h"
+#include "os0file.h"
+
+/* When mysqld is run, the default directory "." is the mysqld datadir, but in
+ibbackup we must set it explicitly; the patgh must NOT contain the trailing
+'/' or '\' */
+extern const char*	fil_path_to_mysql_datadir;
+
+/* Initial size of a single-table tablespace in pages */
+#define FIL_IBD_FILE_INITIAL_SIZE	4
+
+/* 'null' (undefined) page offset in the context of file spaces */
+#define	FIL_NULL	ULINT32_UNDEFINED
+
+/* Space address data type; this is intended to be used when
+addresses accurate to a byte are stored in file pages. If the page part
+of the address is FIL_NULL, the address is considered undefined. */
+
+typedef	byte	fil_faddr_t;	/* 'type' definition in C: an address
+				stored in a file page is a string of bytes */
+#define FIL_ADDR_PAGE	0	/* first in address is the page offset */
+#define	FIL_ADDR_BYTE	4	/* then comes 2-byte byte offset within page*/
+
+#define	FIL_ADDR_SIZE	6	/* address size is 6 bytes */
+
+/* A struct for storing a space address FIL_ADDR, when it is used
+in C program data structures. */
+
+typedef struct fil_addr_struct	fil_addr_t;
+struct fil_addr_struct{
+	ulint	page;		/* page number within a space */
+	ulint	boffset;	/* byte offset within the page */
+};
+
+/* Null file address */
+extern fil_addr_t	fil_addr_null;
+
+/* The byte offsets on a file page for various variables */
+#define FIL_PAGE_SPACE_OR_CHKSUM 0	/* in < MySQL-4.0.14 space id the
+					page belongs to (== 0) but in later
+					versions the 'new' checksum of the
+					page */
+#define FIL_PAGE_OFFSET		4	/* page offset inside space */
+#define FIL_PAGE_PREV		8	/* if there is a 'natural' predecessor
+					of the page, its offset.
+					Otherwise FIL_NULL.
+					This field is not set on BLOB pages,
+					which are stored as a singly-linked
+					list.  See also FIL_PAGE_NEXT. */
+#define FIL_PAGE_NEXT		12	/* if there is a 'natural' successor
+					of the page, its offset.
+					Otherwise FIL_NULL.
+					B-tree index pages
+					(FIL_PAGE_TYPE contains FIL_PAGE_INDEX)
+					on the same PAGE_LEVEL are maintained
+					as a doubly linked list via
+					FIL_PAGE_PREV and FIL_PAGE_NEXT
+					in the collation order of the
+					smallest user record on each page. */
+#define FIL_PAGE_LSN		16	/* lsn of the end of the newest
+					modification log record to the page */
+#define	FIL_PAGE_TYPE		24	/* file page type: FIL_PAGE_INDEX,...,
+					2 bytes.
+
+					The contents of this field can only
+					be trusted in the following case:
+					if the page is an uncompressed
+					B-tree index page, then it is
+					guaranteed that the value is
+					FIL_PAGE_INDEX.
+					The opposite does not hold.
+
+					In tablespaces created by
+					MySQL/InnoDB 5.1.7 or later, the
+					contents of this field is valid
+					for all uncompressed pages. */
+#define FIL_PAGE_FILE_FLUSH_LSN	26	/* this is only defined for the
+					first page in a data file: the file
+					has been flushed to disk at least up
+					to this lsn */
+#define FIL_PAGE_ARCH_LOG_NO_OR_SPACE_ID  34 /* starting from 4.1.x this
+					contains the space id of the page */
+#define FIL_PAGE_DATA		38	/* start of the data on the page */
+
+/* File page trailer */
+#define FIL_PAGE_END_LSN_OLD_CHKSUM 8	/* the low 4 bytes of this are used
+					to store the page checksum, the
+					last 4 bytes should be identical
+					to the last 4 bytes of FIL_PAGE_LSN */
+#define FIL_PAGE_DATA_END	8
+
+/* File page types (values of FIL_PAGE_TYPE) */
+#define FIL_PAGE_INDEX		17855	/* B-tree node */
+#define FIL_PAGE_UNDO_LOG	2	/* Undo log page */
+#define FIL_PAGE_INODE		3	/* Index node */
+#define FIL_PAGE_IBUF_FREE_LIST	4	/* Insert buffer free list */
+/* File page types introduced in MySQL/InnoDB 5.1.7 */
+#define FIL_PAGE_TYPE_ALLOCATED	0	/* Freshly allocated page */
+#define FIL_PAGE_IBUF_BITMAP	5	/* Insert buffer bitmap */
+#define FIL_PAGE_TYPE_SYS	6	/* System page */
+#define FIL_PAGE_TYPE_TRX_SYS	7	/* Transaction system data */
+#define FIL_PAGE_TYPE_FSP_HDR	8	/* File space header */
+#define FIL_PAGE_TYPE_XDES	9	/* Extent descriptor page */
+#define FIL_PAGE_TYPE_BLOB	10	/* Uncompressed BLOB page */
+#define FIL_PAGE_TYPE_ZBLOB	11	/* First compressed BLOB page */
+#define FIL_PAGE_TYPE_ZBLOB2	12	/* Subsequent compressed BLOB page */
+
+/* Space types */
+#define FIL_TABLESPACE		501
+#define FIL_LOG			502
+
+extern ulint	fil_n_log_flushes;
+
+extern ulint	fil_n_pending_log_flushes;
+extern ulint	fil_n_pending_tablespace_flushes;
+
+
+/***********************************************************************
+Returns the version number of a tablespace, -1 if not found. */
+UNIV_INTERN
+ib_int64_t
+fil_space_get_version(
+/*==================*/
+			/* out: version number, -1 if the tablespace does not
+			exist in the memory cache */
+	ulint	id);	/* in: space id */
+/***********************************************************************
+Returns the latch of a file space. */
+UNIV_INTERN
+rw_lock_t*
+fil_space_get_latch(
+/*================*/
+			/* out: latch protecting storage allocation */
+	ulint	id,	/* in: space id */
+	ulint*	zip_size);/* out: compressed page size, or
+			0 for uncompressed tablespaces */
+/***********************************************************************
+Returns the type of a file space. */
+UNIV_INTERN
+ulint
+fil_space_get_type(
+/*===============*/
+			/* out: FIL_TABLESPACE or FIL_LOG */
+	ulint	id);	/* in: space id */
+/***********************************************************************
+Appends a new file to the chain of files of a space. File must be closed. */
+UNIV_INTERN
+void
+fil_node_create(
+/*============*/
+	const char*	name,	/* in: file name (file must be closed) */
+	ulint		size,	/* in: file size in database blocks, rounded
+				downwards to an integer */
+	ulint		id,	/* in: space id where to append */
+	ibool		is_raw);/* in: TRUE if a raw device or
+				a raw disk partition */
+#ifdef UNIV_LOG_ARCHIVE
+/********************************************************************
+Drops files from the start of a file space, so that its size is cut by
+the amount given. */
+UNIV_INTERN
+void
+fil_space_truncate_start(
+/*=====================*/
+	ulint	id,		/* in: space id */
+	ulint	trunc_len);	/* in: truncate by this much; it is an error
+				if this does not equal to the combined size of
+				some initial files in the space */
+#endif /* UNIV_LOG_ARCHIVE */
+/***********************************************************************
+Creates a space memory object and puts it to the 'fil system' hash table. If
+there is an error, prints an error message to the .err log. */
+UNIV_INTERN
+ibool
+fil_space_create(
+/*=============*/
+				/* out: TRUE if success */
+	const char*	name,	/* in: space name */
+	ulint		id,	/* in: space id */
+	ulint		zip_size,/* in: compressed page size, or
+				0 for uncompressed tablespaces */
+	ulint		purpose);/* in: FIL_TABLESPACE, or FIL_LOG if log */
+/***********************************************************************
+Frees a space object from a the tablespace memory cache. Closes the files in
+the chain but does not delete them. */
+UNIV_INTERN
+ibool
+fil_space_free(
+/*===========*/
+			/* out: TRUE if success */
+	ulint	id);	/* in: space id */
+/***********************************************************************
+Returns the size of the space in pages. The tablespace must be cached in the
+memory cache. */
+UNIV_INTERN
+ulint
+fil_space_get_size(
+/*===============*/
+			/* out: space size, 0 if space not found */
+	ulint	id);	/* in: space id */
+/***********************************************************************
+Returns the flags of the space. The tablespace must be cached
+in the memory cache. */
+UNIV_INTERN
+ulint
+fil_space_get_flags(
+/*================*/
+			/* out: flags, ULINT_UNDEFINED if space not found */
+	ulint	id);	/* in: space id */
+/***********************************************************************
+Returns the compressed page size of the space, or 0 if the space
+is not compressed. The tablespace must be cached in the memory cache. */
+UNIV_INTERN
+ulint
+fil_space_get_zip_size(
+/*===================*/
+			/* out: compressed page size, ULINT_UNDEFINED
+			if space not found */
+	ulint	id);	/* in: space id */
+/***********************************************************************
+Checks if the pair space, page_no refers to an existing page in a tablespace
+file space. The tablespace must be cached in the memory cache. */
+UNIV_INTERN
+ibool
+fil_check_adress_in_tablespace(
+/*===========================*/
+			/* out: TRUE if the address is meaningful */
+	ulint	id,	/* in: space id */
+	ulint	page_no);/* in: page number */
+/********************************************************************
+Initializes the tablespace memory cache. */
+UNIV_INTERN
+void
+fil_init(
+/*=====*/
+	ulint	max_n_open);	/* in: max number of open files */
+/***********************************************************************
+Opens all log files and system tablespace data files. They stay open until the
+database server shutdown. This should be called at a server startup after the
+space objects for the log and the system tablespace have been created. The
+purpose of this operation is to make sure we never run out of file descriptors
+if we need to read from the insert buffer or to write to the log. */
+UNIV_INTERN
+void
+fil_open_log_and_system_tablespace_files(void);
+/*==========================================*/
+/***********************************************************************
+Closes all open files. There must not be any pending i/o's or not flushed
+modifications in the files. */
+UNIV_INTERN
+void
+fil_close_all_files(void);
+/*=====================*/
+/***********************************************************************
+Sets the max tablespace id counter if the given number is bigger than the
+previous value. */
+UNIV_INTERN
+void
+fil_set_max_space_id_if_bigger(
+/*===========================*/
+	ulint	max_id);/* in: maximum known id */
+/********************************************************************
+Writes the flushed lsn and the latest archived log number to the page
+header of the first page of each data file in the system tablespace. */
+UNIV_INTERN
+ulint
+fil_write_flushed_lsn_to_data_files(
+/*================================*/
+					/* out: DB_SUCCESS or error number */
+	ib_uint64_t	lsn,		/* in: lsn to write */
+	ulint		arch_log_no);	/* in: latest archived log
+					file number */
+/***********************************************************************
+Reads the flushed lsn and arch no fields from a data file at database
+startup. */
+UNIV_INTERN
+void
+fil_read_flushed_lsn_and_arch_log_no(
+/*=================================*/
+	os_file_t	data_file,		/* in: open data file */
+	ibool		one_read_already,	/* in: TRUE if min and max
+						parameters below already
+						contain sensible data */
+#ifdef UNIV_LOG_ARCHIVE
+	ulint*		min_arch_log_no,	/* in/out: */
+	ulint*		max_arch_log_no,	/* in/out: */
+#endif /* UNIV_LOG_ARCHIVE */
+	ib_uint64_t*	min_flushed_lsn,	/* in/out: */
+	ib_uint64_t*	max_flushed_lsn);	/* in/out: */
+/***********************************************************************
+Increments the count of pending insert buffer page merges, if space is not
+being deleted. */
+UNIV_INTERN
+ibool
+fil_inc_pending_ibuf_merges(
+/*========================*/
+			/* out: TRUE if being deleted, and ibuf merges should
+			be skipped */
+	ulint	id);	/* in: space id */
+/***********************************************************************
+Decrements the count of pending insert buffer page merges. */
+UNIV_INTERN
+void
+fil_decr_pending_ibuf_merges(
+/*=========================*/
+	ulint	id);	/* in: space id */
+/***********************************************************************
+Parses the body of a log record written about an .ibd file operation. That is,
+the log record part after the standard (type, space id, page no) header of the
+log record.
+
+If desired, also replays the delete or rename operation if the .ibd file
+exists and the space id in it matches. Replays the create operation if a file
+at that path does not exist yet. If the database directory for the file to be
+created does not exist, then we create the directory, too.
+
+Note that ibbackup --apply-log sets fil_path_to_mysql_datadir to point to the
+datadir that we should use in replaying the file operations. */
+UNIV_INTERN
+byte*
+fil_op_log_parse_or_replay(
+/*=======================*/
+				/* out: end of log record, or NULL if the
+				record was not completely contained between
+				ptr and end_ptr */
+	byte*	ptr,		/* in: buffer containing the log record body,
+				or an initial segment of it, if the record does
+				not fir completely between ptr and end_ptr */
+	byte*	end_ptr,	/* in: buffer end */
+	ulint	type,		/* in: the type of this log record */
+	ulint	space_id);	/* in: the space id of the tablespace in
+				question, or 0 if the log record should
+				only be parsed but not replayed */
+/***********************************************************************
+Deletes a single-table tablespace. The tablespace must be cached in the
+memory cache. */
+UNIV_INTERN
+ibool
+fil_delete_tablespace(
+/*==================*/
+			/* out: TRUE if success */
+	ulint	id);	/* in: space id */
+/***********************************************************************
+Discards a single-table tablespace. The tablespace must be cached in the
+memory cache. Discarding is like deleting a tablespace, but
+1) we do not drop the table from the data dictionary;
+2) we remove all insert buffer entries for the tablespace immediately; in DROP
+TABLE they are only removed gradually in the background;
+3) when the user does IMPORT TABLESPACE, the tablespace will have the same id
+as it originally had. */
+UNIV_INTERN
+ibool
+fil_discard_tablespace(
+/*===================*/
+			/* out: TRUE if success */
+	ulint	id);	/* in: space id */
+/***********************************************************************
+Renames a single-table tablespace. The tablespace must be cached in the
+tablespace memory cache. */
+UNIV_INTERN
+ibool
+fil_rename_tablespace(
+/*==================*/
+					/* out: TRUE if success */
+	const char*	old_name,	/* in: old table name in the standard
+					databasename/tablename format of
+					InnoDB, or NULL if we do the rename
+					based on the space id only */
+	ulint		id,		/* in: space id */
+	const char*	new_name);	/* in: new table name in the standard
+					databasename/tablename format
+					of InnoDB */
+
+/***********************************************************************
+Creates a new single-table tablespace to a database directory of MySQL.
+Database directories are under the 'datadir' of MySQL. The datadir is the
+directory of a running mysqld program. We can refer to it by simply the
+path '.'. Tables created with CREATE TEMPORARY TABLE we place in the temp
+dir of the mysqld server. */
+UNIV_INTERN
+ulint
+fil_create_new_single_table_tablespace(
+/*===================================*/
+					/* out: DB_SUCCESS or error code */
+	ulint*		space_id,	/* in/out: space id; if this is != 0,
+					then this is an input parameter,
+					otherwise output */
+	const char*	tablename,	/* in: the table name in the usual
+					databasename/tablename format
+					of InnoDB, or a dir path to a temp
+					table */
+	ibool		is_temp,	/* in: TRUE if a table created with
+					CREATE TEMPORARY TABLE */
+	ulint		flags,		/* in: tablespace flags */
+	ulint		size);		/* in: the initial size of the
+					tablespace file in pages,
+					must be >= FIL_IBD_FILE_INITIAL_SIZE */
+/************************************************************************
+Tries to open a single-table tablespace and optionally checks the space id is
+right in it. If does not succeed, prints an error message to the .err log. This
+function is used to open a tablespace when we start up mysqld, and also in
+IMPORT TABLESPACE.
+NOTE that we assume this operation is used either at the database startup
+or under the protection of the dictionary mutex, so that two users cannot
+race here. This operation does not leave the file associated with the
+tablespace open, but closes it after we have looked at the space id in it. */
+UNIV_INTERN
+ibool
+fil_open_single_table_tablespace(
+/*=============================*/
+					/* out: TRUE if success */
+	ibool		check_space_id,	/* in: should we check that the space
+					id in the file is right; we assume
+					that this function runs much faster
+					if no check is made, since accessing
+					the file inode probably is much
+					faster (the OS caches them) than
+					accessing the first page of the file */
+	ulint		id,		/* in: space id */
+	ulint		flags,		/* in: tablespace flags */
+	const char*	name);		/* in: table name in the
+					databasename/tablename format */
+/************************************************************************
+It is possible, though very improbable, that the lsn's in the tablespace to be
+imported have risen above the current system lsn, if a lengthy purge, ibuf
+merge, or rollback was performed on a backup taken with ibbackup. If that is
+the case, reset page lsn's in the file. We assume that mysqld was shut down
+after it performed these cleanup operations on the .ibd file, so that it at
+the shutdown stamped the latest lsn to the FIL_PAGE_FILE_FLUSH_LSN in the
+first page of the .ibd file, and we can determine whether we need to reset the
+lsn's just by looking at that flush lsn. */
+UNIV_INTERN
+ibool
+fil_reset_too_high_lsns(
+/*====================*/
+					/* out: TRUE if success */
+	const char*	name,		/* in: table name in the
+					databasename/tablename format */
+	ib_uint64_t	current_lsn);	/* in: reset lsn's if the lsn stamped
+					to FIL_PAGE_FILE_FLUSH_LSN in the
+					first page is too high */
+/************************************************************************
+At the server startup, if we need crash recovery, scans the database
+directories under the MySQL datadir, looking for .ibd files. Those files are
+single-table tablespaces. We need to know the space id in each of them so that
+we know into which file we should look to check the contents of a page stored
+in the doublewrite buffer, also to know where to apply log records where the
+space id is != 0. */
+UNIV_INTERN
+ulint
+fil_load_single_table_tablespaces(void);
+/*===================================*/
+			/* out: DB_SUCCESS or error number */
+/************************************************************************
+If we need crash recovery, and we have called
+fil_load_single_table_tablespaces() and dict_load_single_table_tablespaces(),
+we can call this function to print an error message of orphaned .ibd files
+for which there is not a data dictionary entry with a matching table name
+and space id. */
+UNIV_INTERN
+void
+fil_print_orphaned_tablespaces(void);
+/*================================*/
+/***********************************************************************
+Returns TRUE if a single-table tablespace does not exist in the memory cache,
+or is being deleted there. */
+UNIV_INTERN
+ibool
+fil_tablespace_deleted_or_being_deleted_in_mem(
+/*===========================================*/
+				/* out: TRUE if does not exist or is being\
+				deleted */
+	ulint		id,	/* in: space id */
+	ib_int64_t	version);/* in: tablespace_version should be this; if
+				you pass -1 as the value of this, then this
+				parameter is ignored */
+/***********************************************************************
+Returns TRUE if a single-table tablespace exists in the memory cache. */
+UNIV_INTERN
+ibool
+fil_tablespace_exists_in_mem(
+/*=========================*/
+			/* out: TRUE if exists */
+	ulint	id);	/* in: space id */
+/***********************************************************************
+Returns TRUE if a matching tablespace exists in the InnoDB tablespace memory
+cache. Note that if we have not done a crash recovery at the database startup,
+there may be many tablespaces which are not yet in the memory cache. */
+UNIV_INTERN
+ibool
+fil_space_for_table_exists_in_mem(
+/*==============================*/
+					/* out: TRUE if a matching tablespace
+					exists in the memory cache */
+	ulint		id,		/* in: space id */
+	const char*	name,		/* in: table name in the standard
+					'databasename/tablename' format or
+					the dir path to a temp table */
+	ibool		is_temp,	/* in: TRUE if created with CREATE
+					TEMPORARY TABLE */
+	ibool		mark_space,	/* in: in crash recovery, at database
+					startup we mark all spaces which have
+					an associated table in the InnoDB
+					data dictionary, so that
+					we can print a warning about orphaned
+					tablespaces */
+	ibool		print_error_if_does_not_exist);
+					/* in: print detailed error
+					information to the .err log if a
+					matching tablespace is not found from
+					memory */
+/**************************************************************************
+Tries to extend a data file so that it would accommodate the number of pages
+given. The tablespace must be cached in the memory cache. If the space is big
+enough already, does nothing. */
+UNIV_INTERN
+ibool
+fil_extend_space_to_desired_size(
+/*=============================*/
+				/* out: TRUE if success */
+	ulint*	actual_size,	/* out: size of the space after extension;
+				if we ran out of disk space this may be lower
+				than the desired size */
+	ulint	space_id,	/* in: space id */
+	ulint	size_after_extend);/* in: desired size in pages after the
+				extension; if the current space size is bigger
+				than this already, the function does nothing */
+#ifdef UNIV_HOTBACKUP
+/************************************************************************
+Extends all tablespaces to the size stored in the space header. During the
+ibbackup --apply-log phase we extended the spaces on-demand so that log records
+could be appllied, but that may have left spaces still too small compared to
+the size stored in the space header. */
+UNIV_INTERN
+void
+fil_extend_tablespaces_to_stored_len(void);
+/*======================================*/
+#endif
+/***********************************************************************
+Tries to reserve free extents in a file space. */
+UNIV_INTERN
+ibool
+fil_space_reserve_free_extents(
+/*===========================*/
+				/* out: TRUE if succeed */
+	ulint	id,		/* in: space id */
+	ulint	n_free_now,	/* in: number of free extents now */
+	ulint	n_to_reserve);	/* in: how many one wants to reserve */
+/***********************************************************************
+Releases free extents in a file space. */
+UNIV_INTERN
+void
+fil_space_release_free_extents(
+/*===========================*/
+	ulint	id,		/* in: space id */
+	ulint	n_reserved);	/* in: how many one reserved */
+/***********************************************************************
+Gets the number of reserved extents. If the database is silent, this number
+should be zero. */
+UNIV_INTERN
+ulint
+fil_space_get_n_reserved_extents(
+/*=============================*/
+	ulint	id);		/* in: space id */
+/************************************************************************
+Reads or writes data. This operation is asynchronous (aio). */
+UNIV_INTERN
+ulint
+fil_io(
+/*===*/
+				/* out: DB_SUCCESS, or DB_TABLESPACE_DELETED
+				if we are trying to do i/o on a tablespace
+				which does not exist */
+	ulint	type,		/* in: OS_FILE_READ or OS_FILE_WRITE,
+				ORed to OS_FILE_LOG, if a log i/o
+				and ORed to OS_AIO_SIMULATED_WAKE_LATER
+				if simulated aio and we want to post a
+				batch of i/os; NOTE that a simulated batch
+				may introduce hidden chances of deadlocks,
+				because i/os are not actually handled until
+				all have been posted: use with great
+				caution! */
+	ibool	sync,		/* in: TRUE if synchronous aio is desired */
+	ulint	space_id,	/* in: space id */
+	ulint	zip_size,	/* in: compressed page size in bytes;
+				0 for uncompressed pages */
+	ulint	block_offset,	/* in: offset in number of blocks */
+	ulint	byte_offset,	/* in: remainder of offset in bytes; in
+				aio this must be divisible by the OS block
+				size */
+	ulint	len,		/* in: how many bytes to read or write; this
+				must not cross a file boundary; in aio this
+				must be a block size multiple */
+	void*	buf,		/* in/out: buffer where to store read data
+				or from where to write; in aio this must be
+				appropriately aligned */
+	void*	message);	/* in: message for aio handler if non-sync
+				aio used, else ignored */
+/**************************************************************************
+Waits for an aio operation to complete. This function is used to write the
+handler for completed requests. The aio array of pending requests is divided
+into segments (see os0file.c for more info). The thread specifies which
+segment it wants to wait for. */
+UNIV_INTERN
+void
+fil_aio_wait(
+/*=========*/
+	ulint	segment);	/* in: the number of the segment in the aio
+				array to wait for */
+/**************************************************************************
+Flushes to disk possible writes cached by the OS. If the space does not exist
+or is being dropped, does not do anything. */
+UNIV_INTERN
+void
+fil_flush(
+/*======*/
+	ulint	space_id);	/* in: file space id (this can be a group of
+				log files or a tablespace of the database) */
+/**************************************************************************
+Flushes to disk writes in file spaces of the given type possibly cached by
+the OS. */
+UNIV_INTERN
+void
+fil_flush_file_spaces(
+/*==================*/
+	ulint	purpose);	/* in: FIL_TABLESPACE, FIL_LOG */
+/**********************************************************************
+Checks the consistency of the tablespace cache. */
+UNIV_INTERN
+ibool
+fil_validate(void);
+/*==============*/
+			/* out: TRUE if ok */
+/************************************************************************
+Returns TRUE if file address is undefined. */
+UNIV_INTERN
+ibool
+fil_addr_is_null(
+/*=============*/
+				/* out: TRUE if undefined */
+	fil_addr_t	addr);	/* in: address */
+/************************************************************************
+Accessor functions for a file page */
+UNIV_INTERN
+ulint
+fil_page_get_prev(const byte*	page);
+ulint
+fil_page_get_next(const byte*	page);
+/*************************************************************************
+Sets the file page type. */
+UNIV_INTERN
+void
+fil_page_set_type(
+/*==============*/
+	byte*	page,	/* in: file page */
+	ulint	type);	/* in: type */
+/*************************************************************************
+Gets the file page type. */
+UNIV_INTERN
+ulint
+fil_page_get_type(
+/*==============*/
+				/* out: type; NOTE that if the type
+				has not been written to page, the
+				return value not defined */
+	const byte*	page);	/* in: file page */
+
+/*************************************************************************
+Return local hash table informations. */
+
+ulint
+fil_system_hash_cells(void);
+/*========================*/
+
+ulint
+fil_system_hash_nodes(void);
+/*========================*/
+
+typedef	struct fil_space_struct	fil_space_t;
+
+#endif
diff --git a/storage/xtradb/include/fsp0fsp.h b/storage/xtradb/include/fsp0fsp.h
new file mode 100644
index 00000000000..1f6ae4b614b
--- /dev/null
+++ b/storage/xtradb/include/fsp0fsp.h
@@ -0,0 +1,433 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+File space management
+
+Created 12/18/1995 Heikki Tuuri
+*******************************************************/
+
+#ifndef fsp0fsp_h
+#define fsp0fsp_h
+
+#include "univ.i"
+
+#include "mtr0mtr.h"
+#include "fut0lst.h"
+#include "ut0byte.h"
+#include "page0types.h"
+
+/* If records are inserted in order, there are the following
+flags to tell this (their type is made byte for the compiler
+to warn if direction and hint parameters are switched in
+fseg_alloc_free_page): */
+#define	FSP_UP		((byte)111)	/* alphabetically upwards */
+#define	FSP_DOWN	((byte)112)	/* alphabetically downwards */
+#define	FSP_NO_DIR	((byte)113)	/* no order */
+
+/* File space extent size (one megabyte) in pages */
+#define	FSP_EXTENT_SIZE		(1 << (20 - UNIV_PAGE_SIZE_SHIFT))
+
+/* On a page of any file segment, data may be put starting from this offset: */
+#define FSEG_PAGE_DATA		FIL_PAGE_DATA
+
+/* File segment header which points to the inode describing the file segment */
+typedef	byte	fseg_header_t;
+
+#define FSEG_HDR_SPACE		0	/* space id of the inode */
+#define FSEG_HDR_PAGE_NO	4	/* page number of the inode */
+#define FSEG_HDR_OFFSET		8	/* byte offset of the inode */
+
+#define FSEG_HEADER_SIZE	10
+
+/**************************************************************************
+Initializes the file space system. */
+UNIV_INTERN
+void
+fsp_init(void);
+/*==========*/
+/**************************************************************************
+Gets the current free limit of the system tablespace.  The free limit
+means the place of the first page which has never been put to the the
+free list for allocation.  The space above that address is initialized
+to zero.  Sets also the global variable log_fsp_current_free_limit. */
+UNIV_INTERN
+ulint
+fsp_header_get_free_limit(void);
+/*===========================*/
+			/* out: free limit in megabytes */
+/**************************************************************************
+Gets the size of the system tablespace from the tablespace header.  If
+we do not have an auto-extending data file, this should be equal to
+the size of the data files.  If there is an auto-extending data file,
+this can be smaller. */
+UNIV_INTERN
+ulint
+fsp_header_get_tablespace_size(void);
+/*================================*/
+			/* out: size in pages */
+/**************************************************************************
+Reads the file space size stored in the header page. */
+UNIV_INTERN
+ulint
+fsp_get_size_low(
+/*=============*/
+			/* out: tablespace size stored in the space header */
+	page_t*	page);	/* in: header page (page 0 in the tablespace) */
+/**************************************************************************
+Reads the space id from the first page of a tablespace. */
+UNIV_INTERN
+ulint
+fsp_header_get_space_id(
+/*====================*/
+				/* out: space id, ULINT UNDEFINED if error */
+	const page_t*	page);	/* in: first page of a tablespace */
+/**************************************************************************
+Reads the space flags from the first page of a tablespace. */
+UNIV_INTERN
+ulint
+fsp_header_get_flags(
+/*=================*/
+				/* out: flags */
+	const page_t*	page);	/* in: first page of a tablespace */
+/**************************************************************************
+Reads the compressed page size from the first page of a tablespace. */
+UNIV_INTERN
+ulint
+fsp_header_get_zip_size(
+/*====================*/
+				/* out: compressed page size in bytes,
+				or 0 if uncompressed */
+	const page_t*	page);	/* in: first page of a tablespace */
+/**************************************************************************
+Writes the space id and compressed page size to a tablespace header.
+This function is used past the buffer pool when we in fil0fil.c create
+a new single-table tablespace. */
+UNIV_INTERN
+void
+fsp_header_init_fields(
+/*===================*/
+	page_t*	page,		/* in/out: first page in the space */
+	ulint	space_id,	/* in: space id */
+	ulint	flags);		/* in: tablespace flags (FSP_SPACE_FLAGS):
+				0, or table->flags if newer than COMPACT */
+/**************************************************************************
+Initializes the space header of a new created space and creates also the
+insert buffer tree root if space == 0. */
+UNIV_INTERN
+void
+fsp_header_init(
+/*============*/
+	ulint	space,		/* in: space id */
+	ulint	size,		/* in: current size in blocks */
+	mtr_t*	mtr);		/* in: mini-transaction handle */
+/**************************************************************************
+Increases the space size field of a space. */
+UNIV_INTERN
+void
+fsp_header_inc_size(
+/*================*/
+	ulint	space,	/* in: space id */
+	ulint	size_inc,/* in: size increment in pages */
+	mtr_t*	mtr);	/* in: mini-transaction handle */
+/**************************************************************************
+Creates a new segment. */
+UNIV_INTERN
+buf_block_t*
+fseg_create(
+/*========*/
+			/* out: the block where the segment header is placed,
+			x-latched, NULL if could not create segment
+			because of lack of space */
+	ulint	space,	/* in: space id */
+	ulint	page,	/* in: page where the segment header is placed: if
+			this is != 0, the page must belong to another segment,
+			if this is 0, a new page will be allocated and it
+			will belong to the created segment */
+	ulint	byte_offset, /* in: byte offset of the created segment header
+			on the page */
+	mtr_t*	mtr);	/* in: mtr */
+/**************************************************************************
+Creates a new segment. */
+UNIV_INTERN
+buf_block_t*
+fseg_create_general(
+/*================*/
+			/* out: the block where the segment header is placed,
+			x-latched, NULL if could not create segment
+			because of lack of space */
+	ulint	space,	/* in: space id */
+	ulint	page,	/* in: page where the segment header is placed: if
+			this is != 0, the page must belong to another segment,
+			if this is 0, a new page will be allocated and it
+			will belong to the created segment */
+	ulint	byte_offset, /* in: byte offset of the created segment header
+			on the page */
+	ibool	has_done_reservation, /* in: TRUE if the caller has already
+			done the reservation for the pages with
+			fsp_reserve_free_extents (at least 2 extents: one for
+			the inode and the other for the segment) then there is
+			no need to do the check for this individual
+			operation */
+	mtr_t*	mtr);	/* in: mtr */
+/**************************************************************************
+Calculates the number of pages reserved by a segment, and how many pages are
+currently used. */
+UNIV_INTERN
+ulint
+fseg_n_reserved_pages(
+/*==================*/
+				/* out: number of reserved pages */
+	fseg_header_t*	header,	/* in: segment header */
+	ulint*		used,	/* out: number of pages used (<= reserved) */
+	mtr_t*		mtr);	/* in: mtr handle */
+/**************************************************************************
+Allocates a single free page from a segment. This function implements
+the intelligent allocation strategy which tries to minimize
+file space fragmentation. */
+UNIV_INTERN
+ulint
+fseg_alloc_free_page(
+/*=================*/
+				/* out: the allocated page offset
+				FIL_NULL if no page could be allocated */
+	fseg_header_t*	seg_header, /* in: segment header */
+	ulint		hint,	/* in: hint of which page would be desirable */
+	byte		direction, /* in: if the new page is needed because
+				of an index page split, and records are
+				inserted there in order, into which
+				direction they go alphabetically: FSP_DOWN,
+				FSP_UP, FSP_NO_DIR */
+	mtr_t*		mtr);	/* in: mtr handle */
+/**************************************************************************
+Allocates a single free page from a segment. This function implements
+the intelligent allocation strategy which tries to minimize file space
+fragmentation. */
+UNIV_INTERN
+ulint
+fseg_alloc_free_page_general(
+/*=========================*/
+				/* out: allocated page offset, FIL_NULL if no
+				page could be allocated */
+	fseg_header_t*	seg_header,/* in: segment header */
+	ulint		hint,	/* in: hint of which page would be desirable */
+	byte		direction,/* in: if the new page is needed because
+				of an index page split, and records are
+				inserted there in order, into which
+				direction they go alphabetically: FSP_DOWN,
+				FSP_UP, FSP_NO_DIR */
+	ibool		has_done_reservation, /* in: TRUE if the caller has
+				already done the reservation for the page
+				with fsp_reserve_free_extents, then there
+				is no need to do the check for this individual
+				page */
+	mtr_t*		mtr);	/* in: mtr handle */
+/**************************************************************************
+Reserves free pages from a tablespace. All mini-transactions which may
+use several pages from the tablespace should call this function beforehand
+and reserve enough free extents so that they certainly will be able
+to do their operation, like a B-tree page split, fully. Reservations
+must be released with function fil_space_release_free_extents!
+
+The alloc_type below has the following meaning: FSP_NORMAL means an
+operation which will probably result in more space usage, like an
+insert in a B-tree; FSP_UNDO means allocation to undo logs: if we are
+deleting rows, then this allocation will in the long run result in
+less space usage (after a purge); FSP_CLEANING means allocation done
+in a physical record delete (like in a purge) or other cleaning operation
+which will result in less space usage in the long run. We prefer the latter
+two types of allocation: when space is scarce, FSP_NORMAL allocations
+will not succeed, but the latter two allocations will succeed, if possible.
+The purpose is to avoid dead end where the database is full but the
+user cannot free any space because these freeing operations temporarily
+reserve some space.
+
+Single-table tablespaces whose size is < 32 pages are a special case. In this
+function we would liberally reserve several 64 page extents for every page
+split or merge in a B-tree. But we do not want to waste disk space if the table
+only occupies < 32 pages. That is why we apply different rules in that special
+case, just ensuring that there are 3 free pages available. */
+UNIV_INTERN
+ibool
+fsp_reserve_free_extents(
+/*=====================*/
+			/* out: TRUE if we were able to make the reservation */
+	ulint*	n_reserved,/* out: number of extents actually reserved; if we
+			return TRUE and the tablespace size is < 64 pages,
+			then this can be 0, otherwise it is n_ext */
+	ulint	space,	/* in: space id */
+	ulint	n_ext,	/* in: number of extents to reserve */
+	ulint	alloc_type,/* in: FSP_NORMAL, FSP_UNDO, or FSP_CLEANING */
+	mtr_t*	mtr);	/* in: mtr */
+/**************************************************************************
+This function should be used to get information on how much we still
+will be able to insert new data to the database without running out the
+tablespace. Only free extents are taken into account and we also subtract
+the safety margin required by the above function fsp_reserve_free_extents. */
+UNIV_INTERN
+ullint
+fsp_get_available_space_in_free_extents(
+/*====================================*/
+			/* out: available space in kB */
+	ulint	space);	/* in: space id */
+/**************************************************************************
+Frees a single page of a segment. */
+UNIV_INTERN
+void
+fseg_free_page(
+/*===========*/
+	fseg_header_t*	seg_header, /* in: segment header */
+	ulint		space,	/* in: space id */
+	ulint		page,	/* in: page offset */
+	mtr_t*		mtr);	/* in: mtr handle */
+/***********************************************************************
+Frees a segment. The freeing is performed in several mini-transactions,
+so that there is no danger of bufferfixing too many buffer pages. */
+UNIV_INTERN
+void
+fseg_free(
+/*======*/
+	ulint	space,	/* in: space id */
+	ulint	zip_size,/* in: compressed page size in bytes
+			or 0 for uncompressed pages */
+	ulint	page_no,/* in: page number where the segment header is
+			placed */
+	ulint	offset);/* in: byte offset of the segment header on that
+			page */
+/**************************************************************************
+Frees part of a segment. This function can be used to free a segment
+by repeatedly calling this function in different mini-transactions.
+Doing the freeing in a single mini-transaction might result in
+too big a mini-transaction. */
+UNIV_INTERN
+ibool
+fseg_free_step(
+/*===========*/
+				/* out: TRUE if freeing completed */
+	fseg_header_t*	header,	/* in, own: segment header; NOTE: if the header
+				resides on the first page of the frag list
+				of the segment, this pointer becomes obsolete
+				after the last freeing step */
+	mtr_t*		mtr);	/* in: mtr */
+/**************************************************************************
+Frees part of a segment. Differs from fseg_free_step because this function
+leaves the header page unfreed. */
+UNIV_INTERN
+ibool
+fseg_free_step_not_header(
+/*======================*/
+				/* out: TRUE if freeing completed, except the
+				header page */
+	fseg_header_t*	header,	/* in: segment header which must reside on
+				the first fragment page of the segment */
+	mtr_t*		mtr);	/* in: mtr */
+/***************************************************************************
+Checks if a page address is an extent descriptor page address. */
+UNIV_INLINE
+ibool
+fsp_descr_page(
+/*===========*/
+			/* out: TRUE if a descriptor page */
+	ulint	zip_size,/* in: compressed page size in bytes;
+			0 for uncompressed pages */
+	ulint	page_no);/* in: page number */
+/***************************************************************
+Parses a redo log record of a file page init. */
+UNIV_INTERN
+byte*
+fsp_parse_init_file_page(
+/*=====================*/
+				/* out: end of log record or NULL */
+	byte*		ptr,	/* in: buffer */
+	byte*		end_ptr, /* in: buffer end */
+	buf_block_t*	block);	/* in: block or NULL */
+/***********************************************************************
+Validates the file space system and its segments. */
+UNIV_INTERN
+ibool
+fsp_validate(
+/*=========*/
+			/* out: TRUE if ok */
+	ulint	space);	/* in: space id */
+/***********************************************************************
+Prints info of a file space. */
+UNIV_INTERN
+void
+fsp_print(
+/*======*/
+	ulint	space);	/* in: space id */
+/***********************************************************************
+Validates a segment. */
+UNIV_INTERN
+ibool
+fseg_validate(
+/*==========*/
+				/* out: TRUE if ok */
+	fseg_header_t*	header, /* in: segment header */
+	mtr_t*		mtr2);	/* in: mtr */
+#ifdef UNIV_BTR_PRINT
+/***********************************************************************
+Writes info of a segment. */
+UNIV_INTERN
+void
+fseg_print(
+/*=======*/
+	fseg_header_t*	header, /* in: segment header */
+	mtr_t*		mtr);	/* in: mtr */
+#endif /* UNIV_BTR_PRINT */
+
+/* Flags for fsp_reserve_free_extents */
+#define FSP_NORMAL	1000000
+#define	FSP_UNDO	2000000
+#define FSP_CLEANING	3000000
+
+/* Number of pages described in a single descriptor page: currently each page
+description takes less than 1 byte; a descriptor page is repeated every
+this many file pages */
+/* #define XDES_DESCRIBED_PER_PAGE		UNIV_PAGE_SIZE */
+/* This has been replaced with either UNIV_PAGE_SIZE or page_zip->size. */
+
+/* The space low address page map */
+/*--------------------------------------*/
+				/* The following two pages are repeated
+				every XDES_DESCRIBED_PER_PAGE pages in
+				every tablespace. */
+#define FSP_XDES_OFFSET			0	/* extent descriptor */
+#define FSP_IBUF_BITMAP_OFFSET		1	/* insert buffer bitmap */
+				/* The ibuf bitmap pages are the ones whose
+				page number is the number above plus a
+				multiple of XDES_DESCRIBED_PER_PAGE */
+
+#define FSP_FIRST_INODE_PAGE_NO		2	/* in every tablespace */
+				/* The following pages exist
+				in the system tablespace (space 0). */
+#define FSP_IBUF_HEADER_PAGE_NO		3	/* in tablespace 0 */
+#define FSP_IBUF_TREE_ROOT_PAGE_NO	4	/* in tablespace 0 */
+				/* The ibuf tree root page number in
+				tablespace 0; its fseg inode is on the page
+				number FSP_FIRST_INODE_PAGE_NO */
+#define FSP_TRX_SYS_PAGE_NO		5	/* in tablespace 0 */
+#define	FSP_FIRST_RSEG_PAGE_NO		6	/* in tablespace 0 */
+#define FSP_DICT_HDR_PAGE_NO		7	/* in tablespace 0 */
+/*--------------------------------------*/
+
+#ifndef UNIV_NONINL
+#include "fsp0fsp.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/fsp0fsp.ic b/storage/xtradb/include/fsp0fsp.ic
new file mode 100644
index 00000000000..f0301cc5e18
--- /dev/null
+++ b/storage/xtradb/include/fsp0fsp.ic
@@ -0,0 +1,44 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+File space management
+
+Created 12/18/1995 Heikki Tuuri
+*******************************************************/
+
+/***************************************************************************
+Checks if a page address is an extent descriptor page address. */
+UNIV_INLINE
+ibool
+fsp_descr_page(
+/*===========*/
+			/* out: TRUE if a descriptor page */
+	ulint	zip_size,/* in: compressed page size in bytes;
+			0 for uncompressed pages */
+	ulint	page_no)/* in: page number */
+{
+	ut_ad(ut_is_2pow(zip_size));
+
+	if (!zip_size) {
+		return(UNIV_UNLIKELY((page_no & (UNIV_PAGE_SIZE - 1))
+				     == FSP_XDES_OFFSET));
+	}
+
+	return(UNIV_UNLIKELY((page_no & (zip_size - 1)) == FSP_XDES_OFFSET));
+}
diff --git a/storage/xtradb/include/fut0fut.h b/storage/xtradb/include/fut0fut.h
new file mode 100644
index 00000000000..4de0c97294c
--- /dev/null
+++ b/storage/xtradb/include/fut0fut.h
@@ -0,0 +1,54 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/**********************************************************************
+File-based utilities
+
+Created 12/13/1995 Heikki Tuuri
+***********************************************************************/
+
+
+#ifndef fut0fut_h
+#define fut0fut_h
+
+#include "univ.i"
+
+#include "fil0fil.h"
+#include "mtr0mtr.h"
+
+/************************************************************************
+Gets a pointer to a file address and latches the page. */
+UNIV_INLINE
+byte*
+fut_get_ptr(
+/*========*/
+				/* out: pointer to a byte in a frame; the file
+				page in the frame is bufferfixed and latched */
+	ulint		space,	/* in: space id */
+	ulint		zip_size,/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	fil_addr_t	addr,	/* in: file address */
+	ulint		rw_latch, /* in: RW_S_LATCH, RW_X_LATCH */
+	mtr_t*		mtr);	/* in: mtr handle */
+
+#ifndef UNIV_NONINL
+#include "fut0fut.ic"
+#endif
+
+#endif
+
diff --git a/storage/xtradb/include/fut0fut.ic b/storage/xtradb/include/fut0fut.ic
new file mode 100644
index 00000000000..f7e820da008
--- /dev/null
+++ b/storage/xtradb/include/fut0fut.ic
@@ -0,0 +1,55 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/**********************************************************************
+File-based utilities
+
+Created 12/13/1995 Heikki Tuuri
+***********************************************************************/
+
+#include "sync0rw.h"
+#include "buf0buf.h"
+
+/************************************************************************
+Gets a pointer to a file address and latches the page. */
+UNIV_INLINE
+byte*
+fut_get_ptr(
+/*========*/
+				/* out: pointer to a byte in a frame; the file
+				page in the frame is bufferfixed and latched */
+	ulint		space,	/* in: space id */
+	ulint		zip_size,/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	fil_addr_t	addr,	/* in: file address */
+	ulint		rw_latch, /* in: RW_S_LATCH, RW_X_LATCH */
+	mtr_t*		mtr)	/* in: mtr handle */
+{
+	buf_block_t*	block;
+	byte*		ptr;
+
+	ut_ad(addr.boffset < UNIV_PAGE_SIZE);
+	ut_ad((rw_latch == RW_S_LATCH) || (rw_latch == RW_X_LATCH));
+
+	block = buf_page_get(space, zip_size, addr.page, rw_latch, mtr);
+	ptr = buf_block_get_frame(block) + addr.boffset;
+
+	buf_block_dbg_add_level(block, SYNC_NO_ORDER_CHECK);
+
+	return(ptr);
+}
diff --git a/storage/xtradb/include/fut0lst.h b/storage/xtradb/include/fut0lst.h
new file mode 100644
index 00000000000..f812874fe00
--- /dev/null
+++ b/storage/xtradb/include/fut0lst.h
@@ -0,0 +1,214 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/**********************************************************************
+File-based list utilities
+
+Created 11/28/1995 Heikki Tuuri
+***********************************************************************/
+
+#ifndef fut0lst_h
+#define fut0lst_h
+
+#include "univ.i"
+
+#include "fil0fil.h"
+#include "mtr0mtr.h"
+
+
+/* The C 'types' of base node and list node: these should be used to
+write self-documenting code. Of course, the sizeof macro cannot be
+applied to these types! */
+
+typedef	byte	flst_base_node_t;
+typedef	byte	flst_node_t;
+
+/* The physical size of a list base node in bytes */
+#define	FLST_BASE_NODE_SIZE	(4 + 2 * FIL_ADDR_SIZE)
+
+/* The physical size of a list node in bytes */
+#define	FLST_NODE_SIZE		(2 * FIL_ADDR_SIZE)
+
+
+/************************************************************************
+Initializes a list base node. */
+UNIV_INLINE
+void
+flst_init(
+/*======*/
+	flst_base_node_t*	base,	/* in: pointer to base node */
+	mtr_t*			mtr);	/* in: mini-transaction handle */
+/************************************************************************
+Adds a node as the last node in a list. */
+UNIV_INTERN
+void
+flst_add_last(
+/*==========*/
+	flst_base_node_t*	base,	/* in: pointer to base node of list */
+	flst_node_t*		node,	/* in: node to add */
+	mtr_t*			mtr);	/* in: mini-transaction handle */
+/************************************************************************
+Adds a node as the first node in a list. */
+UNIV_INTERN
+void
+flst_add_first(
+/*===========*/
+	flst_base_node_t*	base,	/* in: pointer to base node of list */
+	flst_node_t*		node,	/* in: node to add */
+	mtr_t*			mtr);	/* in: mini-transaction handle */
+/************************************************************************
+Inserts a node after another in a list. */
+UNIV_INTERN
+void
+flst_insert_after(
+/*==============*/
+	flst_base_node_t*	base,	/* in: pointer to base node of list */
+	flst_node_t*		node1,	/* in: node to insert after */
+	flst_node_t*		node2,	/* in: node to add */
+	mtr_t*			mtr);	/* in: mini-transaction handle */
+/************************************************************************
+Inserts a node before another in a list. */
+UNIV_INTERN
+void
+flst_insert_before(
+/*===============*/
+	flst_base_node_t*	base,	/* in: pointer to base node of list */
+	flst_node_t*		node2,	/* in: node to insert */
+	flst_node_t*		node3,	/* in: node to insert before */
+	mtr_t*			mtr);	/* in: mini-transaction handle */
+/************************************************************************
+Removes a node. */
+UNIV_INTERN
+void
+flst_remove(
+/*========*/
+	flst_base_node_t*	base,	/* in: pointer to base node of list */
+	flst_node_t*		node2,	/* in: node to remove */
+	mtr_t*			mtr);	/* in: mini-transaction handle */
+/************************************************************************
+Cuts off the tail of the list, including the node given. The number of
+nodes which will be removed must be provided by the caller, as this function
+does not measure the length of the tail. */
+UNIV_INTERN
+void
+flst_cut_end(
+/*=========*/
+	flst_base_node_t*	base,	/* in: pointer to base node of list */
+	flst_node_t*		node2,	/* in: first node to remove */
+	ulint			n_nodes,/* in: number of nodes to remove,
+					must be >= 1 */
+	mtr_t*			mtr);	/* in: mini-transaction handle */
+/************************************************************************
+Cuts off the tail of the list, not including the given node. The number of
+nodes which will be removed must be provided by the caller, as this function
+does not measure the length of the tail. */
+UNIV_INTERN
+void
+flst_truncate_end(
+/*==============*/
+	flst_base_node_t*	base,	/* in: pointer to base node of list */
+	flst_node_t*		node2,	/* in: first node not to remove */
+	ulint			n_nodes,/* in: number of nodes to remove */
+	mtr_t*			mtr);	/* in: mini-transaction handle */
+/************************************************************************
+Gets list length. */
+UNIV_INLINE
+ulint
+flst_get_len(
+/*=========*/
+					/* out: length */
+	const flst_base_node_t*	base,	/* in: pointer to base node */
+	mtr_t*			mtr);	/* in: mini-transaction handle */
+/************************************************************************
+Gets list first node address. */
+UNIV_INLINE
+fil_addr_t
+flst_get_first(
+/*===========*/
+					/* out: file address */
+	const flst_base_node_t*	base,	/* in: pointer to base node */
+	mtr_t*			mtr);	/* in: mini-transaction handle */
+/************************************************************************
+Gets list last node address. */
+UNIV_INLINE
+fil_addr_t
+flst_get_last(
+/*==========*/
+					/* out: file address */
+	const flst_base_node_t*	base,	/* in: pointer to base node */
+	mtr_t*			mtr);	/* in: mini-transaction handle */
+/************************************************************************
+Gets list next node address. */
+UNIV_INLINE
+fil_addr_t
+flst_get_next_addr(
+/*===============*/
+					/* out: file address */
+	const flst_node_t*	node,	/* in: pointer to node */
+	mtr_t*			mtr);	/* in: mini-transaction handle */
+/************************************************************************
+Gets list prev node address. */
+UNIV_INLINE
+fil_addr_t
+flst_get_prev_addr(
+/*===============*/
+					/* out: file address */
+	const flst_node_t*	node,	/* in: pointer to node */
+	mtr_t*			mtr);	/* in: mini-transaction handle */
+/************************************************************************
+Writes a file address. */
+UNIV_INLINE
+void
+flst_write_addr(
+/*============*/
+	fil_faddr_t*	faddr,	/* in: pointer to file faddress */
+	fil_addr_t	addr,	/* in: file address */
+	mtr_t*		mtr);	/* in: mini-transaction handle */
+/************************************************************************
+Reads a file address. */
+UNIV_INLINE
+fil_addr_t
+flst_read_addr(
+/*===========*/
+					/* out: file address */
+	const fil_faddr_t*	faddr,	/* in: pointer to file faddress */
+	mtr_t*			mtr);	/* in: mini-transaction handle */
+/************************************************************************
+Validates a file-based list. */
+UNIV_INTERN
+ibool
+flst_validate(
+/*==========*/
+					/* out: TRUE if ok */
+	const flst_base_node_t*	base,	/* in: pointer to base node of list */
+	mtr_t*			mtr1);	/* in: mtr */
+/************************************************************************
+Prints info of a file-based list. */
+UNIV_INTERN
+void
+flst_print(
+/*=======*/
+	const flst_base_node_t*	base,	/* in: pointer to base node of list */
+	mtr_t*			mtr);	/* in: mtr */
+
+
+#ifndef UNIV_NONINL
+#include "fut0lst.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/fut0lst.ic b/storage/xtradb/include/fut0lst.ic
new file mode 100644
index 00000000000..5899e996059
--- /dev/null
+++ b/storage/xtradb/include/fut0lst.ic
@@ -0,0 +1,166 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/**********************************************************************
+File-based list utilities
+
+Created 11/28/1995 Heikki Tuuri
+***********************************************************************/
+
+#include "fut0fut.h"
+#include "mtr0log.h"
+#include "buf0buf.h"
+
+/* We define the field offsets of a node for the list */
+#define FLST_PREV	0	/* 6-byte address of the previous list element;
+				the page part of address is FIL_NULL, if no
+				previous element */
+#define FLST_NEXT	FIL_ADDR_SIZE	/* 6-byte address of the next
+				list element; the page part of address
+				is FIL_NULL, if no next element */
+
+/* We define the field offsets of a base node for the list */
+#define FLST_LEN	0	/* 32-bit list length field */
+#define	FLST_FIRST	4	/* 6-byte address of the first element
+				of the list; undefined if empty list */
+#define	FLST_LAST	(4 + FIL_ADDR_SIZE) /* 6-byte address of the
+				last element of the list; undefined
+				if empty list */
+
+/************************************************************************
+Writes a file address. */
+UNIV_INLINE
+void
+flst_write_addr(
+/*============*/
+	fil_faddr_t*	faddr,	/* in: pointer to file faddress */
+	fil_addr_t	addr,	/* in: file address */
+	mtr_t*		mtr)	/* in: mini-transaction handle */
+{
+	ut_ad(faddr && mtr);
+	ut_ad(mtr_memo_contains_page(mtr, faddr, MTR_MEMO_PAGE_X_FIX));
+	ut_a(addr.page == FIL_NULL || addr.boffset >= FIL_PAGE_DATA);
+	ut_a(ut_align_offset(faddr, UNIV_PAGE_SIZE) >= FIL_PAGE_DATA);
+
+	mlog_write_ulint(faddr + FIL_ADDR_PAGE, addr.page, MLOG_4BYTES, mtr);
+	mlog_write_ulint(faddr + FIL_ADDR_BYTE, addr.boffset,
+			 MLOG_2BYTES, mtr);
+}
+
+/************************************************************************
+Reads a file address. */
+UNIV_INLINE
+fil_addr_t
+flst_read_addr(
+/*===========*/
+					/* out: file address */
+	const fil_faddr_t*	faddr,	/* in: pointer to file faddress */
+	mtr_t*			mtr)	/* in: mini-transaction handle */
+{
+	fil_addr_t	addr;
+
+	ut_ad(faddr && mtr);
+
+	addr.page = mtr_read_ulint(faddr + FIL_ADDR_PAGE, MLOG_4BYTES, mtr);
+	addr.boffset = mtr_read_ulint(faddr + FIL_ADDR_BYTE, MLOG_2BYTES,
+				      mtr);
+	ut_a(addr.page == FIL_NULL || addr.boffset >= FIL_PAGE_DATA);
+	ut_a(ut_align_offset(faddr, UNIV_PAGE_SIZE) >= FIL_PAGE_DATA);
+	return(addr);
+}
+
+/************************************************************************
+Initializes a list base node. */
+UNIV_INLINE
+void
+flst_init(
+/*======*/
+	flst_base_node_t*	base,	/* in: pointer to base node */
+	mtr_t*			mtr)	/* in: mini-transaction handle */
+{
+	ut_ad(mtr_memo_contains_page(mtr, base, MTR_MEMO_PAGE_X_FIX));
+
+	mlog_write_ulint(base + FLST_LEN, 0, MLOG_4BYTES, mtr);
+	flst_write_addr(base + FLST_FIRST, fil_addr_null, mtr);
+	flst_write_addr(base + FLST_LAST, fil_addr_null, mtr);
+}
+
+/************************************************************************
+Gets list length. */
+UNIV_INLINE
+ulint
+flst_get_len(
+/*=========*/
+					/* out: length */
+	const flst_base_node_t*	base,	/* in: pointer to base node */
+	mtr_t*			mtr)	/* in: mini-transaction handle */
+{
+	return(mtr_read_ulint(base + FLST_LEN, MLOG_4BYTES, mtr));
+}
+
+/************************************************************************
+Gets list first node address. */
+UNIV_INLINE
+fil_addr_t
+flst_get_first(
+/*===========*/
+					/* out: file address */
+	const flst_base_node_t*	base,	/* in: pointer to base node */
+	mtr_t*			mtr)	/* in: mini-transaction handle */
+{
+	return(flst_read_addr(base + FLST_FIRST, mtr));
+}
+
+/************************************************************************
+Gets list last node address. */
+UNIV_INLINE
+fil_addr_t
+flst_get_last(
+/*==========*/
+					/* out: file address */
+	const flst_base_node_t*	base,	/* in: pointer to base node */
+	mtr_t*			mtr)	/* in: mini-transaction handle */
+{
+	return(flst_read_addr(base + FLST_LAST, mtr));
+}
+
+/************************************************************************
+Gets list next node address. */
+UNIV_INLINE
+fil_addr_t
+flst_get_next_addr(
+/*===============*/
+					/* out: file address */
+	const flst_node_t*	node,	/* in: pointer to node */
+	mtr_t*			mtr)	/* in: mini-transaction handle */
+{
+	return(flst_read_addr(node + FLST_NEXT, mtr));
+}
+
+/************************************************************************
+Gets list prev node address. */
+UNIV_INLINE
+fil_addr_t
+flst_get_prev_addr(
+/*===============*/
+					/* out: file address */
+	const flst_node_t*	node,	/* in: pointer to node */
+	mtr_t*			mtr)	/* in: mini-transaction handle */
+{
+	return(flst_read_addr(node + FLST_PREV, mtr));
+}
diff --git a/storage/xtradb/include/ha0ha.h b/storage/xtradb/include/ha0ha.h
new file mode 100644
index 00000000000..768f3d7aca3
--- /dev/null
+++ b/storage/xtradb/include/ha0ha.h
@@ -0,0 +1,188 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The hash table with external chains
+
+Created 8/18/1994 Heikki Tuuri
+*******************************************************/
+
+#ifndef ha0ha_h
+#define ha0ha_h
+
+#include "univ.i"
+
+#include "hash0hash.h"
+#include "page0types.h"
+#include "buf0types.h"
+
+/*****************************************************************
+Looks for an element in a hash table. */
+UNIV_INLINE
+void*
+ha_search_and_get_data(
+/*===================*/
+				/* out: pointer to the data of the first hash
+				table node in chain having the fold number,
+				NULL if not found */
+	hash_table_t*	table,	/* in: hash table */
+	ulint		fold);	/* in: folded value of the searched data */
+/*************************************************************
+Looks for an element when we know the pointer to the data and updates
+the pointer to data if found. */
+UNIV_INTERN
+void
+ha_search_and_update_if_found_func(
+/*===============================*/
+	hash_table_t*	table,	/* in: hash table */
+	ulint		fold,	/* in: folded value of the searched data */
+	void*		data,	/* in: pointer to the data */
+#if defined UNIV_AHI_DEBUG || defined UNIV_DEBUG
+	buf_block_t*	new_block,/* in: block containing new_data */
+#endif /* UNIV_AHI_DEBUG || UNIV_DEBUG */
+	void*		new_data);/* in: new pointer to the data */
+
+#if defined UNIV_AHI_DEBUG || defined UNIV_DEBUG
+# define ha_search_and_update_if_found(table,fold,data,new_block,new_data) \
+	ha_search_and_update_if_found_func(table,fold,data,new_block,new_data)
+#else /* UNIV_AHI_DEBUG || UNIV_DEBUG */
+# define ha_search_and_update_if_found(table,fold,data,new_block,new_data) \
+	ha_search_and_update_if_found_func(table,fold,data,new_data)
+#endif /* UNIV_AHI_DEBUG || UNIV_DEBUG */
+/*****************************************************************
+Creates a hash table with >= n array cells. The actual number of cells is
+chosen to be a prime number slightly bigger than n. */
+UNIV_INTERN
+hash_table_t*
+ha_create_func(
+/*===========*/
+				/* out, own: created table */
+	ulint	n,		/* in: number of array cells */
+#ifdef UNIV_SYNC_DEBUG
+	ulint	mutex_level,	/* in: level of the mutexes in the latching
+				order: this is used in the debug version */
+#endif /* UNIV_SYNC_DEBUG */
+	ulint	n_mutexes);	/* in: number of mutexes to protect the
+				hash table: must be a power of 2 */
+#ifdef UNIV_SYNC_DEBUG
+# define ha_create(n_c,n_m,level) ha_create_func(n_c,level,n_m)
+#else /* UNIV_SYNC_DEBUG */
+# define ha_create(n_c,n_m,level) ha_create_func(n_c,n_m)
+#endif /* UNIV_SYNC_DEBUG */
+
+/*****************************************************************
+Empties a hash table and frees the memory heaps. */
+UNIV_INTERN
+void
+ha_clear(
+/*=====*/
+	hash_table_t*	table);	/* in, own: hash table */
+
+/*****************************************************************
+Inserts an entry into a hash table. If an entry with the same fold number
+is found, its node is updated to point to the new data, and no new node
+is inserted. */
+UNIV_INTERN
+ibool
+ha_insert_for_fold_func(
+/*====================*/
+				/* out: TRUE if succeed, FALSE if no more
+				memory could be allocated */
+	hash_table_t*	table,	/* in: hash table */
+	ulint		fold,	/* in: folded value of data; if a node with
+				the same fold value already exists, it is
+				updated to point to the same data, and no new
+				node is created! */
+#if defined UNIV_AHI_DEBUG || defined UNIV_DEBUG
+	buf_block_t*	block,	/* in: buffer block containing the data */
+#endif /* UNIV_AHI_DEBUG || UNIV_DEBUG */
+	void*		data);	/* in: data, must not be NULL */
+
+#if defined UNIV_AHI_DEBUG || defined UNIV_DEBUG
+# define ha_insert_for_fold(t,f,b,d) ha_insert_for_fold_func(t,f,b,d)
+#else /* UNIV_AHI_DEBUG || UNIV_DEBUG */
+# define ha_insert_for_fold(t,f,b,d) ha_insert_for_fold_func(t,f,d)
+#endif /* UNIV_AHI_DEBUG || UNIV_DEBUG */
+
+/*****************************************************************
+Deletes an entry from a hash table. */
+UNIV_INTERN
+void
+ha_delete(
+/*======*/
+	hash_table_t*	table,	/* in: hash table */
+	ulint		fold,	/* in: folded value of data */
+	void*		data);	/* in: data, must not be NULL and must exist
+				in the hash table */
+/*************************************************************
+Looks for an element when we know the pointer to the data and deletes
+it from the hash table if found. */
+UNIV_INLINE
+ibool
+ha_search_and_delete_if_found(
+/*==========================*/
+				/* out: TRUE if found */
+	hash_table_t*	table,	/* in: hash table */
+	ulint		fold,	/* in: folded value of the searched data */
+	void*		data);	/* in: pointer to the data */
+/*********************************************************************
+Removes from the chain determined by fold all nodes whose data pointer
+points to the page given. */
+UNIV_INTERN
+void
+ha_remove_all_nodes_to_page(
+/*========================*/
+	hash_table_t*	table,	/* in: hash table */
+	ulint		fold,	/* in: fold value */
+	const page_t*	page);	/* in: buffer page */
+/*****************************************************************
+Validates a given range of the cells in hash table. */
+UNIV_INTERN
+ibool
+ha_validate(
+/*========*/
+					/* out: TRUE if ok */
+	hash_table_t*	table,		/* in: hash table */
+	ulint		start_index,	/* in: start index */
+	ulint		end_index);	/* in: end index */
+/*****************************************************************
+Prints info of a hash table. */
+UNIV_INTERN
+void
+ha_print_info(
+/*==========*/
+	FILE*		file,	/* in: file where to print */
+	hash_table_t*	table);	/* in: hash table */
+
+/* The hash table external chain node */
+
+typedef struct ha_node_struct ha_node_t;
+struct ha_node_struct {
+	ha_node_t*	next;	/* next chain node or NULL if none */
+#if defined UNIV_AHI_DEBUG || defined UNIV_DEBUG
+	buf_block_t*	block;	/* buffer block containing the data, or NULL */
+#endif /* UNIV_AHI_DEBUG || UNIV_DEBUG */
+	void*		data;	/* pointer to the data */
+	ulint		fold;	/* fold value for the data */
+};
+
+#ifndef UNIV_NONINL
+#include "ha0ha.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/ha0ha.ic b/storage/xtradb/include/ha0ha.ic
new file mode 100644
index 00000000000..35fd802eaef
--- /dev/null
+++ b/storage/xtradb/include/ha0ha.ic
@@ -0,0 +1,214 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/************************************************************************
+The hash table with external chains
+
+Created 8/18/1994 Heikki Tuuri
+*************************************************************************/
+
+#include "ut0rnd.h"
+#include "mem0mem.h"
+
+/***************************************************************
+Deletes a hash node. */
+UNIV_INTERN
+void
+ha_delete_hash_node(
+/*================*/
+	hash_table_t*	table,		/* in: hash table */
+	ha_node_t*	del_node);	/* in: node to be deleted */
+
+/**********************************************************************
+Gets a hash node data. */
+UNIV_INLINE
+void*
+ha_node_get_data(
+/*=============*/
+				/* out: pointer to the data */
+	ha_node_t*	node)	/* in: hash chain node */
+{
+	return(node->data);
+}
+
+/**********************************************************************
+Sets hash node data. */
+UNIV_INLINE
+void
+ha_node_set_data_func(
+/*==================*/
+	ha_node_t*	node,	/* in: hash chain node */
+#if defined UNIV_AHI_DEBUG || defined UNIV_DEBUG
+	buf_block_t*	block,	/* in: buffer block containing the data */
+#endif /* UNIV_AHI_DEBUG || UNIV_DEBUG */
+	void*		data)	/* in: pointer to the data */
+{
+#if defined UNIV_AHI_DEBUG || defined UNIV_DEBUG
+	node->block = block;
+#endif /* UNIV_AHI_DEBUG || UNIV_DEBUG */
+	node->data = data;
+}
+
+#if defined UNIV_AHI_DEBUG || defined UNIV_DEBUG
+# define ha_node_set_data(n,b,d) ha_node_set_data_func(n,b,d)
+#else /* UNIV_AHI_DEBUG || UNIV_DEBUG */
+# define ha_node_set_data(n,b,d) ha_node_set_data_func(n,d)
+#endif /* UNIV_AHI_DEBUG || UNIV_DEBUG */
+
+/**********************************************************************
+Gets the next node in a hash chain. */
+UNIV_INLINE
+ha_node_t*
+ha_chain_get_next(
+/*==============*/
+				/* out: next node, NULL if none */
+	ha_node_t*	node)	/* in: hash chain node */
+{
+	return(node->next);
+}
+
+/**********************************************************************
+Gets the first node in a hash chain. */
+UNIV_INLINE
+ha_node_t*
+ha_chain_get_first(
+/*===============*/
+				/* out: first node, NULL if none */
+	hash_table_t*	table,	/* in: hash table */
+	ulint		fold)	/* in: fold value determining the chain */
+{
+	return((ha_node_t*)
+	       hash_get_nth_cell(table, hash_calc_hash(fold, table))->node);
+}
+
+/*****************************************************************
+Looks for an element in a hash table. */
+UNIV_INLINE
+ha_node_t*
+ha_search(
+/*======*/
+				/* out: pointer to the first hash table node
+				in chain having the fold number, NULL if not
+				found */
+	hash_table_t*	table,	/* in: hash table */
+	ulint		fold)	/* in: folded value of the searched data */
+{
+	ha_node_t*	node;
+
+	ut_ad(!table->mutexes || mutex_own(hash_get_mutex(table, fold)));
+
+	node = ha_chain_get_first(table, fold);
+
+	while (node) {
+		if (node->fold == fold) {
+
+			return(node);
+		}
+
+		node = ha_chain_get_next(node);
+	}
+
+	return(NULL);
+}
+
+/*****************************************************************
+Looks for an element in a hash table. */
+UNIV_INLINE
+void*
+ha_search_and_get_data(
+/*===================*/
+				/* out: pointer to the data of the first hash
+				table node in chain having the fold number,
+				NULL if not found */
+	hash_table_t*	table,	/* in: hash table */
+	ulint		fold)	/* in: folded value of the searched data */
+{
+	ha_node_t*	node;
+
+	ut_ad(!table->mutexes || mutex_own(hash_get_mutex(table, fold)));
+
+	node = ha_chain_get_first(table, fold);
+
+	while (node) {
+		if (node->fold == fold) {
+
+			return(node->data);
+		}
+
+		node = ha_chain_get_next(node);
+	}
+
+	return(NULL);
+}
+
+/*************************************************************
+Looks for an element when we know the pointer to the data. */
+UNIV_INLINE
+ha_node_t*
+ha_search_with_data(
+/*================*/
+				/* out: pointer to the hash table node, NULL
+				if not found in the table */
+	hash_table_t*	table,	/* in: hash table */
+	ulint		fold,	/* in: folded value of the searched data */
+	void*		data)	/* in: pointer to the data */
+{
+	ha_node_t*	node;
+
+	ut_ad(!table->mutexes || mutex_own(hash_get_mutex(table, fold)));
+
+	node = ha_chain_get_first(table, fold);
+
+	while (node) {
+		if (node->data == data) {
+
+			return(node);
+		}
+
+		node = ha_chain_get_next(node);
+	}
+
+	return(NULL);
+}
+
+/*************************************************************
+Looks for an element when we know the pointer to the data, and deletes
+it from the hash table, if found. */
+UNIV_INLINE
+ibool
+ha_search_and_delete_if_found(
+/*==========================*/
+				/* out: TRUE if found */
+	hash_table_t*	table,	/* in: hash table */
+	ulint		fold,	/* in: folded value of the searched data */
+	void*		data)	/* in: pointer to the data */
+{
+	ha_node_t*	node;
+
+	ut_ad(!table->mutexes || mutex_own(hash_get_mutex(table, fold)));
+
+	node = ha_search_with_data(table, fold, data);
+
+	if (node) {
+		ha_delete_hash_node(table, node);
+
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
diff --git a/storage/xtradb/include/ha0storage.h b/storage/xtradb/include/ha0storage.h
new file mode 100644
index 00000000000..f5a3938f434
--- /dev/null
+++ b/storage/xtradb/include/ha0storage.h
@@ -0,0 +1,129 @@
+/*****************************************************************************
+
+Copyright (c) 2007, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Hash storage.
+Provides a data structure that stores chunks of data in
+its own storage, avoiding duplicates.
+
+Created September 22, 2007 Vasil Dimov
+*******************************************************/
+
+#ifndef ha0storage_h
+#define ha0storage_h
+
+#include "univ.i"
+
+/* This value is used by default by ha_storage_create(). More memory
+is allocated later when/if it is needed. */
+#define HA_STORAGE_DEFAULT_HEAP_BYTES	1024
+
+/* This value is used by default by ha_storage_create(). It is a
+constant per ha_storage's lifetime. */
+#define HA_STORAGE_DEFAULT_HASH_CELLS	4096
+
+typedef struct ha_storage_struct	ha_storage_t;
+
+/***********************************************************************
+Creates a hash storage. If any of the parameters is 0, then a default
+value is used. */
+UNIV_INLINE
+ha_storage_t*
+ha_storage_create(
+/*==============*/
+					/* out, own: hash storage */
+	ulint	initial_heap_bytes,	/* in: initial heap's size */
+	ulint	initial_hash_cells);	/* in: initial number of cells
+					in the hash table */
+
+/***********************************************************************
+Copies data into the storage and returns a pointer to the copy. If the
+same data chunk is already present, then pointer to it is returned.
+Data chunks are considered to be equal if len1 == len2 and
+memcmp(data1, data2, len1) == 0. If "data" is not present (and thus
+data_len bytes need to be allocated) and the size of storage is going to
+become more than "memlim" then "data" is not added and NULL is returned.
+To disable this behavior "memlim" can be set to 0, which stands for
+"no limit". */
+
+const void*
+ha_storage_put_memlim(
+/*==================*/
+					/* out: pointer to the copy */
+	ha_storage_t*	storage,	/* in/out: hash storage */
+	const void*	data,		/* in: data to store */
+	ulint		data_len,	/* in: data length */
+	ulint		memlim);	/* in: memory limit to obey */
+
+/***********************************************************************
+Same as ha_storage_put_memlim() but without memory limit. */
+
+#define ha_storage_put(storage, data, data_len)	\
+	ha_storage_put_memlim((storage), (data), (data_len), 0)
+
+/***********************************************************************
+Copies string into the storage and returns a pointer to the copy. If the
+same string is already present, then pointer to it is returned.
+Strings are considered to be equal if strcmp(str1, str2) == 0. */
+
+#define ha_storage_put_str(storage, str)	\
+	((const char*) ha_storage_put((storage), (str), strlen(str) + 1))
+
+/***********************************************************************
+Copies string into the storage and returns a pointer to the copy obeying
+a memory limit. */
+
+#define ha_storage_put_str_memlim(storage, str, memlim)	\
+	((const char*) ha_storage_put_memlim((storage), (str),	\
+					     strlen(str) + 1, (memlim)))
+
+/***********************************************************************
+Empties a hash storage, freeing memory occupied by data chunks.
+This invalidates any pointers previously returned by ha_storage_put().
+The hash storage is not invalidated itself and can be used again. */
+UNIV_INLINE
+void
+ha_storage_empty(
+/*=============*/
+	ha_storage_t**	storage);	/* in/out: hash storage */
+
+/***********************************************************************
+Frees a hash storage and everything it contains, it cannot be used after
+this call.
+This invalidates any pointers previously returned by ha_storage_put().
+*/
+UNIV_INLINE
+void
+ha_storage_free(
+/*============*/
+	ha_storage_t*	storage);	/* in/out: hash storage */
+
+/***********************************************************************
+Gets the size of the memory used by a storage. */
+UNIV_INLINE
+ulint
+ha_storage_get_size(
+/*================*/
+						/* out: bytes used */
+	const ha_storage_t*	storage);	/* in: hash storage */
+
+#ifndef UNIV_NONINL
+#include "ha0storage.ic"
+#endif
+
+#endif /* ha0storage_h */
diff --git a/storage/xtradb/include/ha0storage.ic b/storage/xtradb/include/ha0storage.ic
new file mode 100644
index 00000000000..7ab43bc00ba
--- /dev/null
+++ b/storage/xtradb/include/ha0storage.ic
@@ -0,0 +1,146 @@
+/*****************************************************************************
+
+Copyright (c) 2007, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Hash storage.
+Provides a data structure that stores chunks of data in
+its own storage, avoiding duplicates.
+
+Created September 24, 2007 Vasil Dimov
+*******************************************************/
+
+#include "univ.i"
+#include "ha0storage.h"
+#include "hash0hash.h"
+#include "mem0mem.h"
+
+struct ha_storage_struct {
+	mem_heap_t*	heap;	/* storage from which memory is
+				allocated */
+	hash_table_t*	hash;	/* hash table used to avoid
+				duplicates */
+};
+
+/* Objects of this type are put in the hash */
+typedef struct ha_storage_node_struct ha_storage_node_t;
+struct ha_storage_node_struct {
+	ulint			data_len;/* length of the data */
+	const void*		data;	/* pointer to data */
+	ha_storage_node_t*	next;	/* next node in hash chain */
+};
+
+/***********************************************************************
+Creates a hash storage. If any of the parameters is 0, then a default
+value is used. */
+UNIV_INLINE
+ha_storage_t*
+ha_storage_create(
+/*==============*/
+					/* out, own: hash storage */
+	ulint	initial_heap_bytes,	/* in: initial heap's size */
+	ulint	initial_hash_cells)	/* in: initial number of cells
+					in the hash table */
+{
+	ha_storage_t*	storage;
+	mem_heap_t*	heap;
+
+	if (initial_heap_bytes == 0) {
+
+		initial_heap_bytes = HA_STORAGE_DEFAULT_HEAP_BYTES;
+	}
+
+	if (initial_hash_cells == 0) {
+
+		initial_hash_cells = HA_STORAGE_DEFAULT_HASH_CELLS;
+	}
+
+	/* we put "storage" within "storage->heap" */
+
+	heap = mem_heap_create(sizeof(ha_storage_t)
+			       + initial_heap_bytes);
+
+	storage = (ha_storage_t*) mem_heap_alloc(heap,
+						 sizeof(ha_storage_t));
+
+	storage->heap = heap;
+	storage->hash = hash_create(initial_hash_cells);
+
+	return(storage);
+}
+
+/***********************************************************************
+Empties a hash storage, freeing memory occupied by data chunks.
+This invalidates any pointers previously returned by ha_storage_put().
+The hash storage is not invalidated itself and can be used again. */
+UNIV_INLINE
+void
+ha_storage_empty(
+/*=============*/
+	ha_storage_t**	storage)	/* in/out: hash storage */
+{
+	ha_storage_t	temp_storage;
+
+	temp_storage.heap = (*storage)->heap;
+	temp_storage.hash = (*storage)->hash;
+
+	hash_table_clear(temp_storage.hash);
+	mem_heap_empty(temp_storage.heap);
+
+	*storage = (ha_storage_t*) mem_heap_alloc(temp_storage.heap,
+						  sizeof(ha_storage_t));
+
+	(*storage)->heap = temp_storage.heap;
+	(*storage)->hash = temp_storage.hash;
+}
+
+/***********************************************************************
+Frees a hash storage and everything it contains, it cannot be used after
+this call.
+This invalidates any pointers previously returned by ha_storage_put().
+*/
+UNIV_INLINE
+void
+ha_storage_free(
+/*============*/
+	ha_storage_t*	storage)	/* in/out: hash storage */
+{
+	/* order is important because the pointer storage->hash is
+	within the heap */
+	hash_table_free(storage->hash);
+	mem_heap_free(storage->heap);
+}
+
+/***********************************************************************
+Gets the size of the memory used by a storage. */
+UNIV_INLINE
+ulint
+ha_storage_get_size(
+/*================*/
+						/* out: bytes used */
+	const ha_storage_t*	storage)	/* in: hash storage */
+{
+	ulint	ret;
+
+	ret = mem_heap_get_size(storage->heap);
+
+	/* this assumes hash->heap and hash->heaps are NULL */
+	ret += sizeof(hash_table_t);
+	ret += sizeof(hash_cell_t) * hash_get_n_cells(storage->hash);
+
+	return(ret);
+}
diff --git a/storage/xtradb/include/ha_prototypes.h b/storage/xtradb/include/ha_prototypes.h
new file mode 100644
index 00000000000..116242b32e4
--- /dev/null
+++ b/storage/xtradb/include/ha_prototypes.h
@@ -0,0 +1,249 @@
+/*****************************************************************************
+
+Copyright (c) 2006, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+#ifndef HA_INNODB_PROTOTYPES_H
+#define HA_INNODB_PROTOTYPES_H
+
+#ifndef UNIV_HOTBACKUP
+
+#include "univ.i" /* ulint, uint */
+#include "m_ctype.h" /* CHARSET_INFO */
+
+/* Prototypes for global functions in ha_innodb.cc that are called by
+InnoDB's C-code. */
+
+/*************************************************************************
+Wrapper around MySQL's copy_and_convert function, see it for
+documentation. */
+UNIV_INTERN
+ulint
+innobase_convert_string(
+/*====================*/
+	void*		to,
+	ulint		to_length,
+	CHARSET_INFO*	to_cs,
+	const void*	from,
+	ulint		from_length,
+	CHARSET_INFO*	from_cs,
+	uint*		errors);
+
+/***********************************************************************
+Formats the raw data in "data" (in InnoDB on-disk format) that is of
+type DATA_(CHAR|VARCHAR|MYSQL|VARMYSQL) using "charset_coll" and writes
+the result to "buf". The result is converted to "system_charset_info".
+Not more than "buf_size" bytes are written to "buf".
+The result is always '\0'-terminated (provided buf_size > 0) and the
+number of bytes that were written to "buf" is returned (including the
+terminating '\0'). */
+UNIV_INTERN
+ulint
+innobase_raw_format(
+/*================*/
+					/* out: number of bytes
+					that were written */
+	const char*	data,		/* in: raw data */
+	ulint		data_len,	/* in: raw data length
+					in bytes */
+	ulint		charset_coll,	/* in: charset collation */
+	char*		buf,		/* out: output buffer */
+	ulint		buf_size);	/* in: output buffer size
+					in bytes */
+                    
+/*********************************************************************
+Convert a table or index name to the MySQL system_charset_info (UTF-8)
+and quote it if needed. */
+UNIV_INTERN
+char*
+innobase_convert_name(
+/*==================*/
+				/* out: pointer to the end of buf */
+	char*		buf,	/* out: buffer for converted identifier */
+	ulint		buflen,	/* in: length of buf, in bytes */
+	const char*	id,	/* in: identifier to convert */
+	ulint		idlen,	/* in: length of id, in bytes */
+	void*		thd,	/* in: MySQL connection thread, or NULL */
+	ibool		table_id);/* in: TRUE=id is a table or database name;
+				FALSE=id is an index name */
+
+/**********************************************************************
+Returns true if the thread is the replication thread on the slave
+server. Used in srv_conc_enter_innodb() to determine if the thread
+should be allowed to enter InnoDB - the replication thread is treated
+differently than other threads. Also used in
+srv_conc_force_exit_innodb(). */
+UNIV_INTERN
+ibool
+thd_is_replication_slave_thread(
+/*============================*/
+			/* out: true if thd is the replication thread */
+	void*	thd);	/* in: thread handle (THD*) */
+
+/**********************************************************************
+Returns true if the transaction this thread is processing has edited
+non-transactional tables. Used by the deadlock detector when deciding
+which transaction to rollback in case of a deadlock - we try to avoid
+rolling back transactions that have edited non-transactional tables. */
+UNIV_INTERN
+ibool
+thd_has_edited_nontrans_tables(
+/*===========================*/
+			/* out: true if non-transactional tables have
+			been edited */
+	void*	thd);	/* in: thread handle (THD*) */
+
+/*****************************************************************
+Prints info of a THD object (== user session thread) to the given file. */
+UNIV_INTERN
+void
+innobase_mysql_print_thd(
+/*=====================*/
+	FILE*	f,		/* in: output stream */
+	void*	thd,		/* in: pointer to a MySQL THD object */
+	uint	max_query_len);	/* in: max query length to print, or 0 to
+				   use the default max length */
+
+/******************************************************************
+Converts a MySQL type to an InnoDB type. Note that this function returns
+the 'mtype' of InnoDB. InnoDB differentiates between MySQL's old <= 4.1
+VARCHAR and the new true VARCHAR in >= 5.0.3 by the 'prtype'. */
+UNIV_INTERN
+ulint
+get_innobase_type_from_mysql_type(
+/*==============================*/
+					/* out: DATA_BINARY,
+					DATA_VARCHAR, ... */
+	ulint*		unsigned_flag,	/* out: DATA_UNSIGNED if an
+					'unsigned type';
+					at least ENUM and SET,
+					and unsigned integer
+					types are 'unsigned types' */
+	const void*	field)		/* in: MySQL Field */
+	__attribute__((nonnull));
+
+/*****************************************************************
+If you want to print a thd that is not associated with the current thread,
+you must call this function before reserving the InnoDB kernel_mutex, to
+protect MySQL from setting thd->query NULL. If you print a thd of the current
+thread, we know that MySQL cannot modify thd->query, and it is not necessary
+to call this. Call innobase_mysql_end_print_arbitrary_thd() after you release
+the kernel_mutex. */
+UNIV_INTERN
+void
+innobase_mysql_prepare_print_arbitrary_thd(void);
+/*============================================*/
+
+/*****************************************************************
+Releases the mutex reserved by innobase_mysql_prepare_print_arbitrary_thd().
+In the InnoDB latching order, the mutex sits right above the
+kernel_mutex.  In debug builds, we assert that the kernel_mutex is
+released before this function is invoked. */
+UNIV_INTERN
+void
+innobase_mysql_end_print_arbitrary_thd(void);
+/*========================================*/
+
+/**********************************************************************
+Get the variable length bounds of the given character set. */
+UNIV_INTERN
+void
+innobase_get_cset_width(
+/*====================*/
+	ulint	cset,		/* in: MySQL charset-collation code */
+	ulint*	mbminlen,	/* out: minimum length of a char (in bytes) */
+	ulint*	mbmaxlen);	/* out: maximum length of a char (in bytes) */
+
+/**********************************************************************
+Compares NUL-terminated UTF-8 strings case insensitively. */
+UNIV_INTERN
+int
+innobase_strcasecmp(
+/*================*/
+				/* out: 0 if a=b, <0 if a<b, >1 if a>b */
+	const char*	a,	/* in: first string to compare */
+	const char*	b);	/* in: second string to compare */
+
+/**********************************************************************
+Returns true if the thread is executing a SELECT statement. */
+
+ibool
+thd_is_select(
+/*==========*/
+				/* out: true if thd is executing SELECT */
+	const void*	thd);	/* in: thread handle (THD*) */
+
+/**********************************************************************
+Converts an identifier to a table name. */
+UNIV_INTERN
+void
+innobase_convert_from_table_id(
+/*===========================*/
+	struct charset_info_st*	cs,	/* in: the 'from' character set */
+	char*			to,	/* out: converted identifier */
+	const char*		from,	/* in: identifier to convert */
+	ulint			len);	/* in: length of 'to', in bytes; should
+					be at least 5 * strlen(to) + 1 */
+/**********************************************************************
+Converts an identifier to UTF-8. */
+UNIV_INTERN
+void
+innobase_convert_from_id(
+/*=====================*/
+	struct charset_info_st*	cs,	/* in: the 'from' character set */
+	char*			to,	/* out: converted identifier */
+	const char*		from,	/* in: identifier to convert */
+	ulint			len);	/* in: length of 'to', in bytes; should
+					be at least 3 * strlen(to) + 1 */
+/**********************************************************************
+Makes all characters in a NUL-terminated UTF-8 string lower case. */
+UNIV_INTERN
+void
+innobase_casedn_str(
+/*================*/
+	char*	a);	/* in/out: string to put in lower case */
+
+/**************************************************************************
+Determines the connection character set. */
+struct charset_info_st*
+innobase_get_charset(
+/*=================*/
+				/* out: connection character set */
+	void*	mysql_thd);	/* in: MySQL thread handle */
+
+/**********************************************************************
+Returns true if the thread supports XA,
+global value of innodb_supports_xa if thd is NULL. */
+
+ibool
+thd_supports_xa(
+/*============*/
+			/* out: true if thd supports XA */
+	void*	thd);	/* in: thread handle (THD*), or NULL to query
+			the global innodb_supports_xa */
+
+/**********************************************************************
+Returns the lock wait timeout for the current connection. */
+
+ulong
+thd_lock_wait_timeout(
+/*==================*/
+			/* out: the lock wait timeout, in seconds */
+	void*	thd);	/* in: thread handle (THD*), or NULL to query
+			the global innodb_lock_wait_timeout */
+
+#endif
+#endif
diff --git a/storage/xtradb/include/handler0alter.h b/storage/xtradb/include/handler0alter.h
new file mode 100644
index 00000000000..69488b67b2b
--- /dev/null
+++ b/storage/xtradb/include/handler0alter.h
@@ -0,0 +1,41 @@
+/*****************************************************************************
+
+Copyright (c) 2005, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Smart ALTER TABLE
+*******************************************************/
+
+/*****************************************************************
+Copies an InnoDB record to table->record[0]. */
+UNIV_INTERN
+void
+innobase_rec_to_mysql(
+/*==================*/
+	TABLE*			table,		/* in/out: MySQL table */
+	const rec_t*		rec,		/* in: record */
+	const dict_index_t*	index,		/* in: index */
+	const ulint*		offsets);	/* in: rec_get_offsets(
+						rec, index, ...) */
+
+/*****************************************************************
+Resets table->record[0]. */
+UNIV_INTERN
+void
+innobase_rec_reset(
+/*===============*/
+	TABLE*			table);		/* in/out: MySQL table */
diff --git a/storage/xtradb/include/hash0hash.h b/storage/xtradb/include/hash0hash.h
new file mode 100644
index 00000000000..2b3eea62754
--- /dev/null
+++ b/storage/xtradb/include/hash0hash.h
@@ -0,0 +1,429 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The simple hash table utility
+
+Created 5/20/1997 Heikki Tuuri
+*******************************************************/
+
+#ifndef hash0hash_h
+#define hash0hash_h
+
+#include "univ.i"
+#include "mem0mem.h"
+#include "sync0sync.h"
+
+typedef struct hash_table_struct hash_table_t;
+typedef struct hash_cell_struct hash_cell_t;
+
+typedef void*	hash_node_t;
+
+/* Fix Bug #13859: symbol collision between imap/mysql */
+#define hash_create hash0_create
+
+/*****************************************************************
+Creates a hash table with >= n array cells. The actual number
+of cells is chosen to be a prime number slightly bigger than n. */
+UNIV_INTERN
+hash_table_t*
+hash_create(
+/*========*/
+			/* out, own: created table */
+	ulint	n);	/* in: number of array cells */
+/*****************************************************************
+Creates a mutex array to protect a hash table. */
+UNIV_INTERN
+void
+hash_create_mutexes_func(
+/*=====================*/
+	hash_table_t*	table,		/* in: hash table */
+#ifdef UNIV_SYNC_DEBUG
+	ulint		sync_level,	/* in: latching order level of the
+					mutexes: used in the debug version */
+#endif /* UNIV_SYNC_DEBUG */
+	ulint		n_mutexes);	/* in: number of mutexes */
+#ifdef UNIV_SYNC_DEBUG
+# define hash_create_mutexes(t,n,level) hash_create_mutexes_func(t,level,n)
+#else /* UNIV_SYNC_DEBUG */
+# define hash_create_mutexes(t,n,level) hash_create_mutexes_func(t,n)
+#endif /* UNIV_SYNC_DEBUG */
+
+/*****************************************************************
+Frees a hash table. */
+UNIV_INTERN
+void
+hash_table_free(
+/*============*/
+	hash_table_t*	table);	/* in, own: hash table */
+/******************************************************************
+Calculates the hash value from a folded value. */
+UNIV_INLINE
+ulint
+hash_calc_hash(
+/*===========*/
+				/* out: hashed value */
+	ulint		fold,	/* in: folded value */
+	hash_table_t*	table);	/* in: hash table */
+/************************************************************************
+Assert that the mutex for the table in a hash operation is owned. */
+#define HASH_ASSERT_OWNED(TABLE, FOLD)					\
+ut_ad(!(TABLE)->mutexes || mutex_own(hash_get_mutex(TABLE, FOLD)));
+
+/***********************************************************************
+Inserts a struct to a hash table. */
+
+#define HASH_INSERT(TYPE, NAME, TABLE, FOLD, DATA)\
+do {\
+	hash_cell_t*	cell3333;\
+	TYPE*		struct3333;\
+\
+	HASH_ASSERT_OWNED(TABLE, FOLD)\
+\
+	(DATA)->NAME = NULL;\
+\
+	cell3333 = hash_get_nth_cell(TABLE, hash_calc_hash(FOLD, TABLE));\
+\
+	if (cell3333->node == NULL) {\
+		cell3333->node = DATA;\
+	} else {\
+		struct3333 = (TYPE*) cell3333->node;\
+\
+		while (struct3333->NAME != NULL) {\
+\
+			struct3333 = (TYPE*) struct3333->NAME;\
+		}\
+\
+		struct3333->NAME = DATA;\
+	}\
+} while (0)
+
+#ifdef UNIV_HASH_DEBUG
+# define HASH_ASSERT_VALID(DATA) ut_a((void*) (DATA) != (void*) -1)
+# define HASH_INVALIDATE(DATA, NAME) DATA->NAME = (void*) -1
+#else
+# define HASH_ASSERT_VALID(DATA) do {} while (0)
+# define HASH_INVALIDATE(DATA, NAME) do {} while (0)
+#endif
+
+/***********************************************************************
+Deletes a struct from a hash table. */
+
+#define HASH_DELETE(TYPE, NAME, TABLE, FOLD, DATA)\
+do {\
+	hash_cell_t*	cell3333;\
+	TYPE*		struct3333;\
+\
+	HASH_ASSERT_OWNED(TABLE, FOLD)\
+\
+	cell3333 = hash_get_nth_cell(TABLE, hash_calc_hash(FOLD, TABLE));\
+\
+	if (cell3333->node == DATA) {\
+		HASH_ASSERT_VALID(DATA->NAME);\
+		cell3333->node = DATA->NAME;\
+	} else {\
+		struct3333 = (TYPE*) cell3333->node;\
+\
+		while (struct3333->NAME != DATA) {\
+\
+			struct3333 = (TYPE*) struct3333->NAME;\
+			ut_a(struct3333);\
+		}\
+\
+		struct3333->NAME = DATA->NAME;\
+	}\
+	HASH_INVALIDATE(DATA, NAME);\
+} while (0)
+
+/***********************************************************************
+Gets the first struct in a hash chain, NULL if none. */
+
+#define HASH_GET_FIRST(TABLE, HASH_VAL)\
+	(hash_get_nth_cell(TABLE, HASH_VAL)->node)
+
+/***********************************************************************
+Gets the next struct in a hash chain, NULL if none. */
+
+#define HASH_GET_NEXT(NAME, DATA)	((DATA)->NAME)
+
+/************************************************************************
+Looks for a struct in a hash table. */
+#define HASH_SEARCH(NAME, TABLE, FOLD, TYPE, DATA, ASSERTION, TEST)\
+{\
+\
+	HASH_ASSERT_OWNED(TABLE, FOLD)\
+\
+	(DATA) = (TYPE) HASH_GET_FIRST(TABLE, hash_calc_hash(FOLD, TABLE));\
+	HASH_ASSERT_VALID(DATA);\
+\
+	while ((DATA) != NULL) {\
+		ASSERTION;\
+		if (TEST) {\
+			break;\
+		} else {\
+			HASH_ASSERT_VALID(HASH_GET_NEXT(NAME, DATA));\
+			(DATA) = (TYPE) HASH_GET_NEXT(NAME, DATA);\
+		}\
+	}\
+}
+
+/************************************************************************
+Looks for an item in all hash buckets. */
+#define HASH_SEARCH_ALL(NAME, TABLE, TYPE, DATA, ASSERTION, TEST)	\
+do {									\
+	ulint	i3333;							\
+									\
+	for (i3333 = (TABLE)->n_cells; i3333--; ) {			\
+		(DATA) = (TYPE) HASH_GET_FIRST(TABLE, i3333);		\
+									\
+		while ((DATA) != NULL) {				\
+			HASH_ASSERT_VALID(DATA);			\
+			ASSERTION;					\
+									\
+			if (TEST) {					\
+				break;					\
+			}						\
+									\
+			(DATA) = (TYPE) HASH_GET_NEXT(NAME, DATA);	\
+		}							\
+									\
+		if ((DATA) != NULL) {					\
+			break;						\
+		}							\
+	}								\
+} while (0)
+
+/****************************************************************
+Gets the nth cell in a hash table. */
+UNIV_INLINE
+hash_cell_t*
+hash_get_nth_cell(
+/*==============*/
+				/* out: pointer to cell */
+	hash_table_t*	table,	/* in: hash table */
+	ulint		n);	/* in: cell index */
+
+/*****************************************************************
+Clears a hash table so that all the cells become empty. */
+UNIV_INLINE
+void
+hash_table_clear(
+/*=============*/
+	hash_table_t*	table);	/* in/out: hash table */
+
+/*****************************************************************
+Returns the number of cells in a hash table. */
+UNIV_INLINE
+ulint
+hash_get_n_cells(
+/*=============*/
+				/* out: number of cells */
+	hash_table_t*	table);	/* in: table */
+/***********************************************************************
+Deletes a struct which is stored in the heap of the hash table, and compacts
+the heap. The fold value must be stored in the struct NODE in a field named
+'fold'. */
+
+#define HASH_DELETE_AND_COMPACT(TYPE, NAME, TABLE, NODE)\
+do {\
+	TYPE*		node111;\
+	TYPE*		top_node111;\
+	hash_cell_t*	cell111;\
+	ulint		fold111;\
+\
+	fold111 = (NODE)->fold;\
+\
+	HASH_DELETE(TYPE, NAME, TABLE, fold111, NODE);\
+\
+	top_node111 = (TYPE*)mem_heap_get_top(\
+				hash_get_heap(TABLE, fold111),\
+							sizeof(TYPE));\
+\
+	/* If the node to remove is not the top node in the heap, compact the\
+	heap of nodes by moving the top node in the place of NODE. */\
+\
+	if (NODE != top_node111) {\
+\
+		/* Copy the top node in place of NODE */\
+\
+		*(NODE) = *top_node111;\
+\
+		cell111 = hash_get_nth_cell(TABLE,\
+				hash_calc_hash(top_node111->fold, TABLE));\
+\
+		/* Look for the pointer to the top node, to update it */\
+\
+		if (cell111->node == top_node111) {\
+			/* The top node is the first in the chain */\
+\
+			cell111->node = NODE;\
+		} else {\
+			/* We have to look for the predecessor of the top\
+			node */\
+			node111 = cell111->node;\
+\
+			while (top_node111 != HASH_GET_NEXT(NAME, node111)) {\
+\
+				node111 = HASH_GET_NEXT(NAME, node111);\
+			}\
+\
+			/* Now we have the predecessor node */\
+\
+			node111->NAME = NODE;\
+		}\
+	}\
+\
+	/* Free the space occupied by the top node */\
+\
+	mem_heap_free_top(hash_get_heap(TABLE, fold111), sizeof(TYPE));\
+} while (0)
+
+/********************************************************************
+Move all hash table entries from OLD_TABLE to NEW_TABLE.*/
+
+#define HASH_MIGRATE(OLD_TABLE, NEW_TABLE, NODE_TYPE, PTR_NAME, FOLD_FUNC) \
+do {\
+	ulint		i2222;\
+	ulint		cell_count2222;\
+\
+	cell_count2222 = hash_get_n_cells(OLD_TABLE);\
+\
+	for (i2222 = 0; i2222 < cell_count2222; i2222++) {\
+		NODE_TYPE*	node2222 = HASH_GET_FIRST((OLD_TABLE), i2222);\
+\
+		while (node2222) {\
+			NODE_TYPE*	next2222 = node2222->PTR_NAME;\
+			ulint		fold2222 = FOLD_FUNC(node2222);\
+\
+			HASH_INSERT(NODE_TYPE, PTR_NAME, (NEW_TABLE),\
+				fold2222, node2222);\
+\
+			node2222 = next2222;\
+		}\
+	}\
+} while (0)
+
+
+/****************************************************************
+Gets the mutex index for a fold value in a hash table. */
+UNIV_INLINE
+ulint
+hash_get_mutex_no(
+/*==============*/
+				/* out: mutex number */
+	hash_table_t*	table,	/* in: hash table */
+	ulint		fold);	/* in: fold */
+/****************************************************************
+Gets the nth heap in a hash table. */
+UNIV_INLINE
+mem_heap_t*
+hash_get_nth_heap(
+/*==============*/
+				/* out: mem heap */
+	hash_table_t*	table,	/* in: hash table */
+	ulint		i);	/* in: index of the heap */
+/****************************************************************
+Gets the heap for a fold value in a hash table. */
+UNIV_INLINE
+mem_heap_t*
+hash_get_heap(
+/*==========*/
+				/* out: mem heap */
+	hash_table_t*	table,	/* in: hash table */
+	ulint		fold);	/* in: fold */
+/****************************************************************
+Gets the nth mutex in a hash table. */
+UNIV_INLINE
+mutex_t*
+hash_get_nth_mutex(
+/*===============*/
+				/* out: mutex */
+	hash_table_t*	table,	/* in: hash table */
+	ulint		i);	/* in: index of the mutex */
+/****************************************************************
+Gets the mutex for a fold value in a hash table. */
+UNIV_INLINE
+mutex_t*
+hash_get_mutex(
+/*===========*/
+				/* out: mutex */
+	hash_table_t*	table,	/* in: hash table */
+	ulint		fold);	/* in: fold */
+/****************************************************************
+Reserves the mutex for a fold value in a hash table. */
+UNIV_INTERN
+void
+hash_mutex_enter(
+/*=============*/
+	hash_table_t*	table,	/* in: hash table */
+	ulint		fold);	/* in: fold */
+/****************************************************************
+Releases the mutex for a fold value in a hash table. */
+UNIV_INTERN
+void
+hash_mutex_exit(
+/*============*/
+	hash_table_t*	table,	/* in: hash table */
+	ulint		fold);	/* in: fold */
+/****************************************************************
+Reserves all the mutexes of a hash table, in an ascending order. */
+UNIV_INTERN
+void
+hash_mutex_enter_all(
+/*=================*/
+	hash_table_t*	table);	/* in: hash table */
+/****************************************************************
+Releases all the mutexes of a hash table. */
+UNIV_INTERN
+void
+hash_mutex_exit_all(
+/*================*/
+	hash_table_t*	table);	/* in: hash table */
+
+
+struct hash_cell_struct{
+	void*	node;	/* hash chain node, NULL if none */
+};
+
+/* The hash table structure */
+struct hash_table_struct {
+#if defined UNIV_AHI_DEBUG || defined UNIV_DEBUG
+	ibool		adaptive;/* TRUE if this is the hash table of the
+				adaptive hash index */
+#endif /* UNIV_AHI_DEBUG || UNIV_DEBUG */
+	ulint		n_cells;/* number of cells in the hash table */
+	hash_cell_t*	array;	/* pointer to cell array */
+	ulint		n_mutexes;/* if mutexes != NULL, then the number of
+				mutexes, must be a power of 2 */
+	mutex_t*	mutexes;/* NULL, or an array of mutexes used to
+				protect segments of the hash table */
+	mem_heap_t**	heaps;	/* if this is non-NULL, hash chain nodes for
+				external chaining can be allocated from these
+				memory heaps; there are then n_mutexes many of
+				these heaps */
+	mem_heap_t*	heap;
+	ulint		magic_n;
+};
+
+#define HASH_TABLE_MAGIC_N	76561114
+
+#ifndef UNIV_NONINL
+#include "hash0hash.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/hash0hash.ic b/storage/xtradb/include/hash0hash.ic
new file mode 100644
index 00000000000..792fdcbf4f8
--- /dev/null
+++ b/storage/xtradb/include/hash0hash.ic
@@ -0,0 +1,160 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The simple hash table utility
+
+Created 5/20/1997 Heikki Tuuri
+*******************************************************/
+
+#include "ut0rnd.h"
+
+/****************************************************************
+Gets the nth cell in a hash table. */
+UNIV_INLINE
+hash_cell_t*
+hash_get_nth_cell(
+/*==============*/
+				/* out: pointer to cell */
+	hash_table_t*	table,	/* in: hash table */
+	ulint		n)	/* in: cell index */
+{
+	ut_ad(n < table->n_cells);
+
+	return(table->array + n);
+}
+
+/*****************************************************************
+Clears a hash table so that all the cells become empty. */
+UNIV_INLINE
+void
+hash_table_clear(
+/*=============*/
+	hash_table_t*	table)	/* in/out: hash table */
+{
+	memset(table->array, 0x0,
+	       table->n_cells * sizeof(*table->array));
+}
+
+/*****************************************************************
+Returns the number of cells in a hash table. */
+UNIV_INLINE
+ulint
+hash_get_n_cells(
+/*=============*/
+				/* out: number of cells */
+	hash_table_t*	table)	/* in: table */
+{
+	return(table->n_cells);
+}
+
+/******************************************************************
+Calculates the hash value from a folded value. */
+UNIV_INLINE
+ulint
+hash_calc_hash(
+/*===========*/
+				/* out: hashed value */
+	ulint		fold,	/* in: folded value */
+	hash_table_t*	table)	/* in: hash table */
+{
+	return(ut_hash_ulint(fold, table->n_cells));
+}
+
+/****************************************************************
+Gets the mutex index for a fold value in a hash table. */
+UNIV_INLINE
+ulint
+hash_get_mutex_no(
+/*==============*/
+				/* out: mutex number */
+	hash_table_t*	table,	/* in: hash table */
+	ulint		fold)	/* in: fold */
+{
+	ut_ad(ut_is_2pow(table->n_mutexes));
+	return(ut_2pow_remainder(hash_calc_hash(fold, table),
+				 table->n_mutexes));
+}
+
+/****************************************************************
+Gets the nth heap in a hash table. */
+UNIV_INLINE
+mem_heap_t*
+hash_get_nth_heap(
+/*==============*/
+				/* out: mem heap */
+	hash_table_t*	table,	/* in: hash table */
+	ulint		i)	/* in: index of the heap */
+{
+	ut_ad(i < table->n_mutexes);
+
+	return(table->heaps[i]);
+}
+
+/****************************************************************
+Gets the heap for a fold value in a hash table. */
+UNIV_INLINE
+mem_heap_t*
+hash_get_heap(
+/*==========*/
+				/* out: mem heap */
+	hash_table_t*	table,	/* in: hash table */
+	ulint		fold)	/* in: fold */
+{
+	ulint	i;
+
+	if (table->heap) {
+		return(table->heap);
+	}
+
+	i = hash_get_mutex_no(table, fold);
+
+	return(hash_get_nth_heap(table, i));
+}
+
+/****************************************************************
+Gets the nth mutex in a hash table. */
+UNIV_INLINE
+mutex_t*
+hash_get_nth_mutex(
+/*===============*/
+				/* out: mutex */
+	hash_table_t*	table,	/* in: hash table */
+	ulint		i)	/* in: index of the mutex */
+{
+	ut_ad(i < table->n_mutexes);
+
+	return(table->mutexes + i);
+}
+
+/****************************************************************
+Gets the mutex for a fold value in a hash table. */
+UNIV_INLINE
+mutex_t*
+hash_get_mutex(
+/*===========*/
+				/* out: mutex */
+	hash_table_t*	table,	/* in: hash table */
+	ulint		fold)	/* in: fold */
+{
+	ulint	i;
+
+	i = hash_get_mutex_no(table, fold);
+
+	return(hash_get_nth_mutex(table, i));
+}
diff --git a/storage/xtradb/include/ibuf0ibuf.h b/storage/xtradb/include/ibuf0ibuf.h
new file mode 100644
index 00000000000..41e2392cc4a
--- /dev/null
+++ b/storage/xtradb/include/ibuf0ibuf.h
@@ -0,0 +1,369 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Insert buffer
+
+Created 7/19/1997 Heikki Tuuri
+*******************************************************/
+
+#ifndef ibuf0ibuf_h
+#define ibuf0ibuf_h
+
+#include "univ.i"
+
+#include "dict0mem.h"
+#include "mtr0mtr.h"
+#include "que0types.h"
+#include "ibuf0types.h"
+#include "fsp0fsp.h"
+
+/** Combinations of operations that can be buffered.  Because the enum
+values are used for indexing innobase_change_buffering_values[], they
+should start at 0 and there should not be any gaps. */
+typedef enum {
+	IBUF_USE_NONE = 0,
+	IBUF_USE_INSERT,	/* insert */
+
+	IBUF_USE_COUNT		/* number of entries in ibuf_use_t */
+} ibuf_use_t;
+
+/** Operations that can currently be buffered. */
+extern ibuf_use_t	ibuf_use;
+
+/** The insert buffer control structure */
+extern ibuf_t*		ibuf;
+
+/* The purpose of the insert buffer is to reduce random disk access.
+When we wish to insert a record into a non-unique secondary index and
+the B-tree leaf page where the record belongs to is not in the buffer
+pool, we insert the record into the insert buffer B-tree, indexed by
+(space_id, page_no).  When the page is eventually read into the buffer
+pool, we look up the insert buffer B-tree for any modifications to the
+page, and apply these upon the completion of the read operation.  This
+is called the insert buffer merge. */
+
+/* The insert buffer merge must always succeed.  To guarantee this,
+the insert buffer subsystem keeps track of the free space in pages for
+which it can buffer operations.  Two bits per page in the insert
+buffer bitmap indicate the available space in coarse increments.  The
+free bits in the insert buffer bitmap must never exceed the free space
+on a page.  It is safe to decrement or reset the bits in the bitmap in
+a mini-transaction that is committed before the mini-transaction that
+affects the free space.  It is unsafe to increment the bits in a
+separately committed mini-transaction, because in crash recovery, the
+free bits could momentarily be set too high. */
+
+/**********************************************************************
+Creates the insert buffer data structure at a database startup and
+initializes the data structures for the insert buffer of each tablespace. */
+UNIV_INTERN
+void
+ibuf_init_at_db_start(void);
+/*=======================*/
+/*************************************************************************
+Reads the biggest tablespace id from the high end of the insert buffer
+tree and updates the counter in fil_system. */
+UNIV_INTERN
+void
+ibuf_update_max_tablespace_id(void);
+/*===============================*/
+/*************************************************************************
+Initializes an ibuf bitmap page. */
+UNIV_INTERN
+void
+ibuf_bitmap_page_init(
+/*==================*/
+	buf_block_t*	block,	/* in: bitmap page */
+	mtr_t*		mtr);	/* in: mtr */
+/****************************************************************************
+Resets the free bits of the page in the ibuf bitmap. This is done in a
+separate mini-transaction, hence this operation does not restrict
+further work to only ibuf bitmap operations, which would result if the
+latch to the bitmap page were kept.  NOTE: The free bits in the insert
+buffer bitmap must never exceed the free space on a page.  It is safe
+to decrement or reset the bits in the bitmap in a mini-transaction
+that is committed before the mini-transaction that affects the free
+space. */
+UNIV_INTERN
+void
+ibuf_reset_free_bits(
+/*=================*/
+	buf_block_t*	block);	/* in: index page; free bits are set to 0
+				if the index is a non-clustered
+				non-unique, and page level is 0 */
+/****************************************************************************
+Updates the free bits of an uncompressed page in the ibuf bitmap if
+there is not enough free on the page any more.  This is done in a
+separate mini-transaction, hence this operation does not restrict
+further work to only ibuf bitmap operations, which would result if the
+latch to the bitmap page were kept.  NOTE: The free bits in the insert
+buffer bitmap must never exceed the free space on a page.  It is
+unsafe to increment the bits in a separately committed
+mini-transaction, because in crash recovery, the free bits could
+momentarily be set too high.  It is only safe to use this function for
+decrementing the free bits.  Should more free space become available,
+we must not update the free bits here, because that would break crash
+recovery. */
+UNIV_INLINE
+void
+ibuf_update_free_bits_if_full(
+/*==========================*/
+	buf_block_t*	block,	/* in: index page to which we have added new
+				records; the free bits are updated if the
+				index is non-clustered and non-unique and
+				the page level is 0, and the page becomes
+				fuller */
+	ulint		max_ins_size,/* in: value of maximum insert size with
+				reorganize before the latest operation
+				performed to the page */
+	ulint		increase);/* in: upper limit for the additional space
+				used in the latest operation, if known, or
+				ULINT_UNDEFINED */
+/**************************************************************************
+Updates the free bits for an uncompressed page to reflect the present
+state.  Does this in the mtr given, which means that the latching
+order rules virtually prevent any further operations for this OS
+thread until mtr is committed.  NOTE: The free bits in the insert
+buffer bitmap must never exceed the free space on a page.  It is safe
+to set the free bits in the same mini-transaction that updated the
+page. */
+UNIV_INTERN
+void
+ibuf_update_free_bits_low(
+/*======================*/
+	const buf_block_t*	block,		/* in: index page */
+	ulint			max_ins_size,	/* in: value of
+						maximum insert size
+						with reorganize before
+						the latest operation
+						performed to the page */
+	mtr_t*			mtr);		/* in/out: mtr */
+/**************************************************************************
+Updates the free bits for a compressed page to reflect the present
+state.  Does this in the mtr given, which means that the latching
+order rules virtually prevent any further operations for this OS
+thread until mtr is committed.  NOTE: The free bits in the insert
+buffer bitmap must never exceed the free space on a page.  It is safe
+to set the free bits in the same mini-transaction that updated the
+page. */
+UNIV_INTERN
+void
+ibuf_update_free_bits_zip(
+/*======================*/
+	buf_block_t*	block,	/* in/out: index page */
+	mtr_t*		mtr);	/* in/out: mtr */
+/**************************************************************************
+Updates the free bits for the two pages to reflect the present state.
+Does this in the mtr given, which means that the latching order rules
+virtually prevent any further operations until mtr is committed.
+NOTE: The free bits in the insert buffer bitmap must never exceed the
+free space on a page.  It is safe to set the free bits in the same
+mini-transaction that updated the pages. */
+UNIV_INTERN
+void
+ibuf_update_free_bits_for_two_pages_low(
+/*====================================*/
+	ulint		zip_size,/* in: compressed page size in bytes;
+				0 for uncompressed pages */
+	buf_block_t*	block1,	/* in: index page */
+	buf_block_t*	block2,	/* in: index page */
+	mtr_t*		mtr);	/* in: mtr */
+/**************************************************************************
+A basic partial test if an insert to the insert buffer could be possible and
+recommended. */
+UNIV_INLINE
+ibool
+ibuf_should_try(
+/*============*/
+	dict_index_t*	index,			/* in: index where to insert */
+	ulint		ignore_sec_unique);	/* in: if != 0, we should
+						ignore UNIQUE constraint on
+						a secondary index when we
+						decide */
+/**********************************************************************
+Returns TRUE if the current OS thread is performing an insert buffer
+routine. */
+UNIV_INTERN
+ibool
+ibuf_inside(void);
+/*=============*/
+		/* out: TRUE if inside an insert buffer routine: for instance,
+		a read-ahead of non-ibuf pages is then forbidden */
+/***************************************************************************
+Checks if a page address is an ibuf bitmap page (level 3 page) address. */
+UNIV_INLINE
+ibool
+ibuf_bitmap_page(
+/*=============*/
+			/* out: TRUE if a bitmap page */
+	ulint	zip_size,/* in: compressed page size in bytes;
+			0 for uncompressed pages */
+	ulint	page_no);/* in: page number */
+/***************************************************************************
+Checks if a page is a level 2 or 3 page in the ibuf hierarchy of pages.
+Must not be called when recv_no_ibuf_operations==TRUE. */
+UNIV_INTERN
+ibool
+ibuf_page(
+/*======*/
+			/* out: TRUE if level 2 or level 3 page */
+	ulint	space,	/* in: space id */
+	ulint	zip_size,/* in: compressed page size in bytes, or 0 */
+	ulint	page_no,/* in: page number */
+	mtr_t*	mtr);	/* in: mtr which will contain an x-latch to the
+			bitmap page if the page is not one of the fixed
+			address ibuf pages, or NULL, in which case a new
+			transaction is created. */
+/***************************************************************************
+Frees excess pages from the ibuf free list. This function is called when an OS
+thread calls fsp services to allocate a new file segment, or a new page to a
+file segment, and the thread did not own the fsp latch before this call. */
+UNIV_INTERN
+void
+ibuf_free_excess_pages(void);
+/*========================*/
+/*************************************************************************
+Makes an index insert to the insert buffer, instead of directly to the disk
+page, if this is possible. Does not do insert if the index is clustered
+or unique. */
+UNIV_INTERN
+ibool
+ibuf_insert(
+/*========*/
+				/* out: TRUE if success */
+	const dtuple_t*	entry,	/* in: index entry to insert */
+	dict_index_t*	index,	/* in: index where to insert */
+	ulint		space,	/* in: space id where to insert */
+	ulint		zip_size,/* in: compressed page size in bytes, or 0 */
+	ulint		page_no,/* in: page number where to insert */
+	que_thr_t*	thr);	/* in: query thread */
+/*************************************************************************
+When an index page is read from a disk to the buffer pool, this function
+inserts to the page the possible index entries buffered in the insert buffer.
+The entries are deleted from the insert buffer. If the page is not read, but
+created in the buffer pool, this function deletes its buffered entries from
+the insert buffer; there can exist entries for such a page if the page
+belonged to an index which subsequently was dropped. */
+UNIV_INTERN
+void
+ibuf_merge_or_delete_for_page(
+/*==========================*/
+	buf_block_t*	block,	/* in: if page has been read from
+				disk, pointer to the page x-latched,
+				else NULL */
+	ulint		space,	/* in: space id of the index page */
+	ulint		page_no,/* in: page number of the index page */
+	ulint		zip_size,/* in: compressed page size in bytes,
+				or 0 */
+	ibool		update_ibuf_bitmap);/* in: normally this is set
+				to TRUE, but if we have deleted or are
+				deleting the tablespace, then we
+				naturally do not want to update a
+				non-existent bitmap page */
+/*************************************************************************
+Deletes all entries in the insert buffer for a given space id. This is used
+in DISCARD TABLESPACE and IMPORT TABLESPACE.
+NOTE: this does not update the page free bitmaps in the space. The space will
+become CORRUPT when you call this function! */
+UNIV_INTERN
+void
+ibuf_delete_for_discarded_space(
+/*============================*/
+	ulint	space);	/* in: space id */
+/*************************************************************************
+Contracts insert buffer trees by reading pages to the buffer pool. */
+UNIV_INTERN
+ulint
+ibuf_contract(
+/*==========*/
+			/* out: a lower limit for the combined size in bytes
+			of entries which will be merged from ibuf trees to the
+			pages read, 0 if ibuf is empty */
+	ibool	sync);	/* in: TRUE if the caller wants to wait for the
+			issued read with the highest tablespace address
+			to complete */
+/*************************************************************************
+Contracts insert buffer trees by reading pages to the buffer pool. */
+UNIV_INTERN
+ulint
+ibuf_contract_for_n_pages(
+/*======================*/
+			/* out: a lower limit for the combined size in bytes
+			of entries which will be merged from ibuf trees to the
+			pages read, 0 if ibuf is empty */
+	ibool	sync,	/* in: TRUE if the caller wants to wait for the
+			issued read with the highest tablespace address
+			to complete */
+	ulint	n_pages);/* in: try to read at least this many pages to
+			the buffer pool and merge the ibuf contents to
+			them */
+/*************************************************************************
+Parses a redo log record of an ibuf bitmap page init. */
+UNIV_INTERN
+byte*
+ibuf_parse_bitmap_init(
+/*===================*/
+				/* out: end of log record or NULL */
+	byte*		ptr,	/* in: buffer */
+	byte*		end_ptr,/* in: buffer end */
+	buf_block_t*	block,	/* in: block or NULL */
+	mtr_t*		mtr);	/* in: mtr or NULL */
+#ifdef UNIV_IBUF_COUNT_DEBUG
+/**********************************************************************
+Gets the ibuf count for a given page. */
+UNIV_INTERN
+ulint
+ibuf_count_get(
+/*===========*/
+			/* out: number of entries in the insert buffer
+			currently buffered for this page */
+	ulint	space,	/* in: space id */
+	ulint	page_no);/* in: page number */
+#endif
+/**********************************************************************
+Looks if the insert buffer is empty. */
+UNIV_INTERN
+ibool
+ibuf_is_empty(void);
+/*===============*/
+			/* out: TRUE if empty */
+/**********************************************************************
+Prints info of ibuf. */
+UNIV_INTERN
+void
+ibuf_print(
+/*=======*/
+	FILE*	file);	/* in: file where to print */
+
+#define IBUF_HEADER_PAGE_NO	FSP_IBUF_HEADER_PAGE_NO
+#define IBUF_TREE_ROOT_PAGE_NO	FSP_IBUF_TREE_ROOT_PAGE_NO
+
+/* The ibuf header page currently contains only the file segment header
+for the file segment from which the pages for the ibuf tree are allocated */
+#define IBUF_HEADER		PAGE_DATA
+#define	IBUF_TREE_SEG_HEADER	0	/* fseg header for ibuf tree */
+
+/* The insert buffer tree itself is always located in space 0. */
+#define IBUF_SPACE_ID		0
+
+#ifndef UNIV_NONINL
+#include "ibuf0ibuf.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/ibuf0ibuf.ic b/storage/xtradb/include/ibuf0ibuf.ic
new file mode 100644
index 00000000000..170e5dba473
--- /dev/null
+++ b/storage/xtradb/include/ibuf0ibuf.ic
@@ -0,0 +1,325 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Insert buffer
+
+Created 7/19/1997 Heikki Tuuri
+*******************************************************/
+
+#include "buf0lru.h"
+#include "page0page.h"
+#include "page0zip.h"
+
+extern ulint	ibuf_flush_count;
+
+/* If this number is n, an index page must contain at least the page size
+per n bytes of free space for ibuf to try to buffer inserts to this page.
+If there is this much of free space, the corresponding bits are set in the
+ibuf bitmap. */
+#define IBUF_PAGE_SIZE_PER_FREE_SPACE	32
+
+/* Insert buffer struct */
+
+struct ibuf_struct{
+	ulint		size;		/* current size of the ibuf index
+					tree, in pages */
+	ulint		max_size;	/* recommended maximum size of the
+					ibuf index tree, in pages */
+	ulint		seg_size;	/* allocated pages of the file
+					segment containing ibuf header and
+					tree */
+	ibool		empty;		/* after an insert to the ibuf tree
+					is performed, this is set to FALSE,
+					and if a contract operation finds
+					the tree empty, this is set to
+					TRUE */
+	ulint		free_list_len;	/* length of the free list */
+	ulint		height;		/* tree height */
+	dict_index_t*	index;		/* insert buffer index */
+
+	ulint		n_inserts;	/* number of inserts made to
+					the insert buffer */
+	ulint		n_merges;	/* number of pages merged */
+	ulint		n_merged_recs;	/* number of records merged */
+};
+
+/****************************************************************************
+Sets the free bit of the page in the ibuf bitmap. This is done in a separate
+mini-transaction, hence this operation does not restrict further work to only
+ibuf bitmap operations, which would result if the latch to the bitmap page
+were kept. */
+UNIV_INTERN
+void
+ibuf_set_free_bits_func(
+/*====================*/
+	buf_block_t*	block,	/* in: index page of a non-clustered index;
+				free bit is reset if page level is 0 */
+#ifdef UNIV_IBUF_DEBUG
+	ulint		max_val,/* in: ULINT_UNDEFINED or a maximum
+				value which the bits must have before
+				setting; this is for debugging */
+#endif /* UNIV_IBUF_DEBUG */
+	ulint		val);	/* in: value to set: < 4 */
+#ifdef UNIV_IBUF_DEBUG
+# define ibuf_set_free_bits(b,v,max) ibuf_set_free_bits_func(b,max,v)
+#else /* UNIV_IBUF_DEBUG */
+# define ibuf_set_free_bits(b,v,max) ibuf_set_free_bits_func(b,v)
+#endif /* UNIV_IBUF_DEBUG */
+
+/**************************************************************************
+A basic partial test if an insert to the insert buffer could be possible and
+recommended. */
+UNIV_INLINE
+ibool
+ibuf_should_try(
+/*============*/
+	dict_index_t*	index,			/* in: index where to insert */
+	ulint		ignore_sec_unique)	/* in: if != 0, we should
+						ignore UNIQUE constraint on
+						a secondary index when we
+						decide */
+{
+	if (ibuf_use != IBUF_USE_NONE
+	    && !dict_index_is_clust(index)
+	    && (ignore_sec_unique || !dict_index_is_unique(index))) {
+
+		ibuf_flush_count++;
+
+		if (ibuf_flush_count % 4 == 0) {
+
+			buf_LRU_try_free_flushed_blocks();
+		}
+
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
+
+/***************************************************************************
+Checks if a page address is an ibuf bitmap page address. */
+UNIV_INLINE
+ibool
+ibuf_bitmap_page(
+/*=============*/
+			/* out: TRUE if a bitmap page */
+	ulint	zip_size,/* in: compressed page size in bytes;
+			0 for uncompressed pages */
+	ulint	page_no)/* in: page number */
+{
+	ut_ad(ut_is_2pow(zip_size));
+
+	if (!zip_size) {
+		return(UNIV_UNLIKELY((page_no & (UNIV_PAGE_SIZE - 1))
+				     == FSP_IBUF_BITMAP_OFFSET));
+	}
+
+	return(UNIV_UNLIKELY((page_no & (zip_size - 1))
+			     == FSP_IBUF_BITMAP_OFFSET));
+}
+
+/*************************************************************************
+Translates the free space on a page to a value in the ibuf bitmap.*/
+UNIV_INLINE
+ulint
+ibuf_index_page_calc_free_bits(
+/*===========================*/
+				/* out: value for ibuf bitmap bits */
+	ulint	zip_size,	/* in: compressed page size in bytes;
+				0 for uncompressed pages */
+	ulint	max_ins_size)	/* in: maximum insert size after reorganize
+				for the page */
+{
+	ulint	n;
+	ut_ad(ut_is_2pow(zip_size));
+	ut_ad(!zip_size || zip_size > IBUF_PAGE_SIZE_PER_FREE_SPACE);
+	ut_ad(zip_size <= UNIV_PAGE_SIZE);
+
+	if (zip_size) {
+		n = max_ins_size
+			/ (zip_size / IBUF_PAGE_SIZE_PER_FREE_SPACE);
+	} else {
+		n = max_ins_size
+			/ (UNIV_PAGE_SIZE / IBUF_PAGE_SIZE_PER_FREE_SPACE);
+	}
+
+	if (n == 3) {
+		n = 2;
+	}
+
+	if (n > 3) {
+		n = 3;
+	}
+
+	return(n);
+}
+
+/*************************************************************************
+Translates the ibuf free bits to the free space on a page in bytes. */
+UNIV_INLINE
+ulint
+ibuf_index_page_calc_free_from_bits(
+/*================================*/
+			/* out: maximum insert size after reorganize for the
+			page */
+	ulint	zip_size,/* in: compressed page size in bytes;
+			0 for uncompressed pages */
+	ulint	bits)	/* in: value for ibuf bitmap bits */
+{
+	ut_ad(bits < 4);
+	ut_ad(ut_is_2pow(zip_size));
+	ut_ad(!zip_size || zip_size > IBUF_PAGE_SIZE_PER_FREE_SPACE);
+	ut_ad(zip_size <= UNIV_PAGE_SIZE);
+
+	if (zip_size) {
+		if (bits == 3) {
+			return(4 * zip_size / IBUF_PAGE_SIZE_PER_FREE_SPACE);
+		}
+
+		return(bits * zip_size / IBUF_PAGE_SIZE_PER_FREE_SPACE);
+	}
+
+	if (bits == 3) {
+		return(4 * UNIV_PAGE_SIZE / IBUF_PAGE_SIZE_PER_FREE_SPACE);
+	}
+
+	return(bits * (UNIV_PAGE_SIZE / IBUF_PAGE_SIZE_PER_FREE_SPACE));
+}
+
+/*************************************************************************
+Translates the free space on a compressed page to a value in the ibuf bitmap.*/
+UNIV_INLINE
+ulint
+ibuf_index_page_calc_free_zip(
+/*==========================*/
+					/* out: value for ibuf bitmap bits */
+	ulint			zip_size,
+					/* in: compressed page size in bytes */
+	const buf_block_t*	block)	/* in: buffer block */
+{
+	ulint			max_ins_size;
+	const page_zip_des_t*	page_zip;
+	lint			zip_max_ins;
+
+	ut_ad(zip_size == buf_block_get_zip_size(block));
+	ut_ad(zip_size);
+
+	max_ins_size = page_get_max_insert_size_after_reorganize(
+		buf_block_get_frame(block), 1);
+
+	page_zip = buf_block_get_page_zip(block);
+	zip_max_ins = page_zip_max_ins_size(page_zip,
+					    FALSE/* not clustered */);
+
+	if (UNIV_UNLIKELY(zip_max_ins < 0)) {
+		return(0);
+	} else if (UNIV_LIKELY(max_ins_size > (ulint) zip_max_ins)) {
+		max_ins_size = (ulint) zip_max_ins;
+	}
+
+	return(ibuf_index_page_calc_free_bits(zip_size, max_ins_size));
+}
+
+/*************************************************************************
+Translates the free space on a page to a value in the ibuf bitmap.*/
+UNIV_INLINE
+ulint
+ibuf_index_page_calc_free(
+/*======================*/
+					/* out: value for ibuf bitmap bits */
+	ulint			zip_size,/* in: compressed page size in bytes;
+					0 for uncompressed pages */
+	const buf_block_t*	block)	/* in: buffer block */
+{
+	ut_ad(zip_size == buf_block_get_zip_size(block));
+
+	if (!zip_size) {
+		ulint	max_ins_size;
+
+		max_ins_size = page_get_max_insert_size_after_reorganize(
+			buf_block_get_frame(block), 1);
+
+		return(ibuf_index_page_calc_free_bits(0, max_ins_size));
+	} else {
+		return(ibuf_index_page_calc_free_zip(zip_size, block));
+	}
+}
+
+/****************************************************************************
+Updates the free bits of an uncompressed page in the ibuf bitmap if
+there is not enough free on the page any more.  This is done in a
+separate mini-transaction, hence this operation does not restrict
+further work to only ibuf bitmap operations, which would result if the
+latch to the bitmap page were kept.  NOTE: The free bits in the insert
+buffer bitmap must never exceed the free space on a page.  It is
+unsafe to increment the bits in a separately committed
+mini-transaction, because in crash recovery, the free bits could
+momentarily be set too high.  It is only safe to use this function for
+decrementing the free bits.  Should more free space become available,
+we must not update the free bits here, because that would break crash
+recovery. */
+UNIV_INLINE
+void
+ibuf_update_free_bits_if_full(
+/*==========================*/
+	buf_block_t*	block,	/* in: index page to which we have added new
+				records; the free bits are updated if the
+				index is non-clustered and non-unique and
+				the page level is 0, and the page becomes
+				fuller */
+	ulint		max_ins_size,/* in: value of maximum insert size with
+				reorganize before the latest operation
+				performed to the page */
+	ulint		increase)/* in: upper limit for the additional space
+				used in the latest operation, if known, or
+				ULINT_UNDEFINED */
+{
+	ulint	before;
+	ulint	after;
+
+	ut_ad(!buf_block_get_page_zip(block));
+
+	before = ibuf_index_page_calc_free_bits(0, max_ins_size);
+
+	if (max_ins_size >= increase) {
+#if ULINT32_UNDEFINED <= UNIV_PAGE_SIZE
+# error "ULINT32_UNDEFINED <= UNIV_PAGE_SIZE"
+#endif
+		after = ibuf_index_page_calc_free_bits(0, max_ins_size
+						       - increase);
+#ifdef UNIV_IBUF_DEBUG
+		ut_a(after <= ibuf_index_page_calc_free(0, block));
+#endif
+	} else {
+		after = ibuf_index_page_calc_free(0, block);
+	}
+
+	if (after == 0) {
+		/* We move the page to the front of the buffer pool LRU list:
+		the purpose of this is to prevent those pages to which we
+		cannot make inserts using the insert buffer from slipping
+		out of the buffer pool */
+
+		buf_page_make_young(&block->page);
+	}
+
+	if (before > after) {
+		ibuf_set_free_bits(block, after, before);
+	}
+}
diff --git a/storage/xtradb/include/ibuf0types.h b/storage/xtradb/include/ibuf0types.h
new file mode 100644
index 00000000000..264415196a1
--- /dev/null
+++ b/storage/xtradb/include/ibuf0types.h
@@ -0,0 +1,30 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Insert buffer global types
+
+Created 7/29/1997 Heikki Tuuri
+*******************************************************/
+
+#ifndef ibuf0types_h
+#define ibuf0types_h
+
+typedef	struct ibuf_struct	ibuf_t;
+
+#endif
diff --git a/storage/xtradb/include/lock0iter.h b/storage/xtradb/include/lock0iter.h
new file mode 100644
index 00000000000..3cd47bb95d2
--- /dev/null
+++ b/storage/xtradb/include/lock0iter.h
@@ -0,0 +1,68 @@
+/*****************************************************************************
+
+Copyright (c) 2007, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Lock queue iterator type and function prototypes.
+
+Created July 16, 2007 Vasil Dimov
+*******************************************************/
+
+#ifndef lock0iter_h
+#define lock0iter_h
+
+#include "univ.i"
+#include "lock0types.h"
+
+typedef struct lock_queue_iterator_struct {
+	const lock_t*	current_lock;
+	/* In case this is a record lock queue (not table lock queue)
+	then bit_no is the record number within the heap in which the
+	record is stored. */
+	ulint		bit_no;
+} lock_queue_iterator_t;
+
+/***********************************************************************
+Initialize lock queue iterator so that it starts to iterate from
+"lock". bit_no specifies the record number within the heap where the
+record is stored. It can be undefined (ULINT_UNDEFINED) in two cases:
+1. If the lock is a table lock, thus we have a table lock queue;
+2. If the lock is a record lock and it is a wait lock. In this case
+   bit_no is calculated in this function by using
+   lock_rec_find_set_bit(). There is exactly one bit set in the bitmap
+   of a wait lock. */
+UNIV_INTERN
+void
+lock_queue_iterator_reset(
+/*======================*/
+	lock_queue_iterator_t*	iter,	/* out: iterator */
+	const lock_t*		lock,	/* in: lock to start from */
+	ulint			bit_no);/* in: record number in the
+					heap */
+
+/***********************************************************************
+Gets the previous lock in the lock queue, returns NULL if there are no
+more locks (i.e. the current lock is the first one). The iterator is
+receded (if not-NULL is returned). */
+
+const lock_t*
+lock_queue_iterator_get_prev(
+/*=========================*/
+					/* out: previous lock or NULL */
+	lock_queue_iterator_t*	iter);	/* in/out: iterator */
+
+#endif /* lock0iter_h */
diff --git a/storage/xtradb/include/lock0lock.h b/storage/xtradb/include/lock0lock.h
new file mode 100644
index 00000000000..2deeb804737
--- /dev/null
+++ b/storage/xtradb/include/lock0lock.h
@@ -0,0 +1,838 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The transaction lock system
+
+Created 5/7/1996 Heikki Tuuri
+*******************************************************/
+
+#ifndef lock0lock_h
+#define lock0lock_h
+
+#include "univ.i"
+#include "buf0types.h"
+#include "trx0types.h"
+#include "rem0types.h"
+#include "dict0types.h"
+#include "que0types.h"
+#include "lock0types.h"
+#include "read0types.h"
+#include "hash0hash.h"
+#include "ut0vec.h"
+
+#ifdef UNIV_DEBUG
+extern ibool	lock_print_waits;
+#endif /* UNIV_DEBUG */
+/* Buffer for storing information about the most recent deadlock error */
+extern FILE*	lock_latest_err_file;
+
+/*************************************************************************
+Gets the size of a lock struct. */
+UNIV_INTERN
+ulint
+lock_get_size(void);
+/*===============*/
+			/* out: size in bytes */
+/*************************************************************************
+Creates the lock system at database start. */
+UNIV_INTERN
+void
+lock_sys_create(
+/*============*/
+	ulint	n_cells);	/* in: number of slots in lock hash table */
+/*************************************************************************
+Checks if some transaction has an implicit x-lock on a record in a clustered
+index. */
+UNIV_INLINE
+trx_t*
+lock_clust_rec_some_has_impl(
+/*=========================*/
+				/* out: transaction which has the x-lock, or
+				NULL */
+	const rec_t*	rec,	/* in: user record */
+	dict_index_t*	index,	/* in: clustered index */
+	const ulint*	offsets);/* in: rec_get_offsets(rec, index) */
+/*************************************************************************
+Gets the heap_no of the smallest user record on a page. */
+UNIV_INLINE
+ulint
+lock_get_min_heap_no(
+/*=================*/
+					/* out: heap_no of smallest
+					user record, or
+					PAGE_HEAP_NO_SUPREMUM */
+	const buf_block_t*	block);	/* in: buffer block */
+/*****************************************************************
+Updates the lock table when we have reorganized a page. NOTE: we copy
+also the locks set on the infimum of the page; the infimum may carry
+locks if an update of a record is occurring on the page, and its locks
+were temporarily stored on the infimum. */
+UNIV_INTERN
+void
+lock_move_reorganize_page(
+/*======================*/
+	const buf_block_t*	block,	/* in: old index page, now
+					reorganized */
+	const buf_block_t*	oblock);/* in: copy of the old, not
+					reorganized page */
+/*****************************************************************
+Moves the explicit locks on user records to another page if a record
+list end is moved to another page. */
+UNIV_INTERN
+void
+lock_move_rec_list_end(
+/*===================*/
+	const buf_block_t*	new_block,	/* in: index page to move to */
+	const buf_block_t*	block,		/* in: index page */
+	const rec_t*		rec);		/* in: record on page: this
+						is the first record moved */
+/*****************************************************************
+Moves the explicit locks on user records to another page if a record
+list start is moved to another page. */
+UNIV_INTERN
+void
+lock_move_rec_list_start(
+/*=====================*/
+	const buf_block_t*	new_block,	/* in: index page to move to */
+	const buf_block_t*	block,		/* in: index page */
+	const rec_t*		rec,		/* in: record on page:
+						this is the first
+						record NOT copied */
+	const rec_t*		old_end);	/* in: old
+						previous-to-last
+						record on new_page
+						before the records
+						were copied */
+/*****************************************************************
+Updates the lock table when a page is split to the right. */
+UNIV_INTERN
+void
+lock_update_split_right(
+/*====================*/
+	const buf_block_t*	right_block,	/* in: right page */
+	const buf_block_t*	left_block);	/* in: left page */
+/*****************************************************************
+Updates the lock table when a page is merged to the right. */
+UNIV_INTERN
+void
+lock_update_merge_right(
+/*====================*/
+	const buf_block_t*	right_block,	/* in: right page to
+						which merged */
+	const rec_t*		orig_succ,	/* in: original
+						successor of infimum
+						on the right page
+						before merge */
+	const buf_block_t*	left_block);	/* in: merged index
+						page which will be
+						discarded */
+/*****************************************************************
+Updates the lock table when the root page is copied to another in
+btr_root_raise_and_insert. Note that we leave lock structs on the
+root page, even though they do not make sense on other than leaf
+pages: the reason is that in a pessimistic update the infimum record
+of the root page will act as a dummy carrier of the locks of the record
+to be updated. */
+UNIV_INTERN
+void
+lock_update_root_raise(
+/*===================*/
+	const buf_block_t*	block,	/* in: index page to which copied */
+	const buf_block_t*	root);	/* in: root page */
+/*****************************************************************
+Updates the lock table when a page is copied to another and the original page
+is removed from the chain of leaf pages, except if page is the root! */
+UNIV_INTERN
+void
+lock_update_copy_and_discard(
+/*=========================*/
+	const buf_block_t*	new_block,	/* in: index page to
+						which copied */
+	const buf_block_t*	block);		/* in: index page;
+						NOT the root! */
+/*****************************************************************
+Updates the lock table when a page is split to the left. */
+UNIV_INTERN
+void
+lock_update_split_left(
+/*===================*/
+	const buf_block_t*	right_block,	/* in: right page */
+	const buf_block_t*	left_block);	/* in: left page */
+/*****************************************************************
+Updates the lock table when a page is merged to the left. */
+UNIV_INTERN
+void
+lock_update_merge_left(
+/*===================*/
+	const buf_block_t*	left_block,	/* in: left page to
+						which merged */
+	const rec_t*		orig_pred,	/* in: original predecessor
+						of supremum on the left page
+						before merge */
+	const buf_block_t*	right_block);	/* in: merged index page
+						which will be discarded */
+/*****************************************************************
+Resets the original locks on heir and replaces them with gap type locks
+inherited from rec. */
+UNIV_INTERN
+void
+lock_rec_reset_and_inherit_gap_locks(
+/*=================================*/
+	const buf_block_t*	heir_block,	/* in: block containing the
+						record which inherits */
+	const buf_block_t*	block,		/* in: block containing the
+						record from which inherited;
+						does NOT reset the locks on
+						this record */
+	ulint			heir_heap_no,	/* in: heap_no of the
+						inheriting record */
+	ulint			heap_no);	/* in: heap_no of the
+						donating record */
+/*****************************************************************
+Updates the lock table when a page is discarded. */
+UNIV_INTERN
+void
+lock_update_discard(
+/*================*/
+	const buf_block_t*	heir_block,	/* in: index page
+						which will inherit the locks */
+	ulint			heir_heap_no,	/* in: heap_no of the record
+						which will inherit the locks */
+	const buf_block_t*	block);		/* in: index page
+						which will be discarded */
+/*****************************************************************
+Updates the lock table when a new user record is inserted. */
+UNIV_INTERN
+void
+lock_update_insert(
+/*===============*/
+	const buf_block_t*	block,	/* in: buffer block containing rec */
+	const rec_t*		rec);	/* in: the inserted record */
+/*****************************************************************
+Updates the lock table when a record is removed. */
+UNIV_INTERN
+void
+lock_update_delete(
+/*===============*/
+	const buf_block_t*	block,	/* in: buffer block containing rec */
+	const rec_t*		rec);	/* in: the record to be removed */
+/*************************************************************************
+Stores on the page infimum record the explicit locks of another record.
+This function is used to store the lock state of a record when it is
+updated and the size of the record changes in the update. The record
+is in such an update moved, perhaps to another page. The infimum record
+acts as a dummy carrier record, taking care of lock releases while the
+actual record is being moved. */
+UNIV_INTERN
+void
+lock_rec_store_on_page_infimum(
+/*===========================*/
+	const buf_block_t*	block,	/* in: buffer block containing rec */
+	const rec_t*		rec);	/* in: record whose lock state
+					is stored on the infimum
+					record of the same page; lock
+					bits are reset on the
+					record */
+/*************************************************************************
+Restores the state of explicit lock requests on a single record, where the
+state was stored on the infimum of the page. */
+UNIV_INTERN
+void
+lock_rec_restore_from_page_infimum(
+/*===============================*/
+	const buf_block_t*	block,	/* in: buffer block containing rec */
+	const rec_t*		rec,	/* in: record whose lock state
+					is restored */
+	const buf_block_t*	donator);/* in: page (rec is not
+					necessarily on this page)
+					whose infimum stored the lock
+					state; lock bits are reset on
+					the infimum */
+/*************************************************************************
+Returns TRUE if there are explicit record locks on a page. */
+UNIV_INTERN
+ibool
+lock_rec_expl_exist_on_page(
+/*========================*/
+			/* out: TRUE if there are explicit record locks on
+			the page */
+	ulint	space,	/* in: space id */
+	ulint	page_no);/* in: page number */
+/*************************************************************************
+Checks if locks of other transactions prevent an immediate insert of
+a record. If they do, first tests if the query thread should anyway
+be suspended for some reason; if not, then puts the transaction and
+the query thread to the lock wait state and inserts a waiting request
+for a gap x-lock to the lock queue. */
+UNIV_INTERN
+ulint
+lock_rec_insert_check_and_lock(
+/*===========================*/
+				/* out: DB_SUCCESS, DB_LOCK_WAIT,
+				DB_DEADLOCK, or DB_QUE_THR_SUSPENDED */
+	ulint		flags,	/* in: if BTR_NO_LOCKING_FLAG bit is
+				set, does nothing */
+	rec_t*		rec,	/* in: record after which to insert */
+	buf_block_t*	block,	/* in/out: buffer block of rec */
+	dict_index_t*	index,	/* in: index */
+	que_thr_t*	thr,	/* in: query thread */
+	ibool*		inherit);/* out: set to TRUE if the new
+				inserted record maybe should inherit
+				LOCK_GAP type locks from the successor
+				record */
+/*************************************************************************
+Checks if locks of other transactions prevent an immediate modify (update,
+delete mark, or delete unmark) of a clustered index record. If they do,
+first tests if the query thread should anyway be suspended for some
+reason; if not, then puts the transaction and the query thread to the
+lock wait state and inserts a waiting request for a record x-lock to the
+lock queue. */
+UNIV_INTERN
+ulint
+lock_clust_rec_modify_check_and_lock(
+/*=================================*/
+					/* out: DB_SUCCESS,
+					DB_LOCK_WAIT, DB_DEADLOCK, or
+					DB_QUE_THR_SUSPENDED */
+	ulint			flags,	/* in: if BTR_NO_LOCKING_FLAG
+					bit is set, does nothing */
+	const buf_block_t*	block,	/* in: buffer block of rec */
+	const rec_t*		rec,	/* in: record which should be
+					modified */
+	dict_index_t*		index,	/* in: clustered index */
+	const ulint*		offsets,/* in: rec_get_offsets(rec, index) */
+	que_thr_t*		thr);	/* in: query thread */
+/*************************************************************************
+Checks if locks of other transactions prevent an immediate modify
+(delete mark or delete unmark) of a secondary index record. */
+UNIV_INTERN
+ulint
+lock_sec_rec_modify_check_and_lock(
+/*===============================*/
+				/* out: DB_SUCCESS, DB_LOCK_WAIT,
+				DB_DEADLOCK, or DB_QUE_THR_SUSPENDED */
+	ulint		flags,	/* in: if BTR_NO_LOCKING_FLAG
+				bit is set, does nothing */
+	buf_block_t*	block,	/* in/out: buffer block of rec */
+	rec_t*		rec,	/* in: record which should be
+				modified; NOTE: as this is a secondary
+				index, we always have to modify the
+				clustered index record first: see the
+				comment below */
+	dict_index_t*	index,	/* in: secondary index */
+	que_thr_t*	thr);	/* in: query thread */
+/*************************************************************************
+Like the counterpart for a clustered index below, but now we read a
+secondary index record. */
+UNIV_INTERN
+ulint
+lock_sec_rec_read_check_and_lock(
+/*=============================*/
+					/* out: DB_SUCCESS,
+					DB_LOCK_WAIT, DB_DEADLOCK, or
+					DB_QUE_THR_SUSPENDED */
+	ulint			flags,	/* in: if BTR_NO_LOCKING_FLAG
+					bit is set, does nothing */
+	const buf_block_t*	block,	/* in: buffer block of rec */
+	const rec_t*		rec,	/* in: user record or page
+					supremum record which should
+					be read or passed over by a
+					read cursor */
+	dict_index_t*		index,	/* in: secondary index */
+	const ulint*		offsets,/* in: rec_get_offsets(rec, index) */
+	enum lock_mode		mode,	/* in: mode of the lock which
+					the read cursor should set on
+					records: LOCK_S or LOCK_X; the
+					latter is possible in
+					SELECT FOR UPDATE */
+	ulint			gap_mode,/* in: LOCK_ORDINARY, LOCK_GAP, or
+					LOCK_REC_NOT_GAP */
+	que_thr_t*		thr);	/* in: query thread */
+/*************************************************************************
+Checks if locks of other transactions prevent an immediate read, or passing
+over by a read cursor, of a clustered index record. If they do, first tests
+if the query thread should anyway be suspended for some reason; if not, then
+puts the transaction and the query thread to the lock wait state and inserts a
+waiting request for a record lock to the lock queue. Sets the requested mode
+lock on the record. */
+UNIV_INTERN
+ulint
+lock_clust_rec_read_check_and_lock(
+/*===============================*/
+					/* out: DB_SUCCESS,
+					DB_LOCK_WAIT, DB_DEADLOCK, or
+					DB_QUE_THR_SUSPENDED */
+	ulint			flags,	/* in: if BTR_NO_LOCKING_FLAG
+					bit is set, does nothing */
+	const buf_block_t*	block,	/* in: buffer block of rec */
+	const rec_t*		rec,	/* in: user record or page
+					supremum record which should
+					be read or passed over by a
+					read cursor */
+	dict_index_t*		index,	/* in: clustered index */
+	const ulint*		offsets,/* in: rec_get_offsets(rec, index) */
+	enum lock_mode		mode,	/* in: mode of the lock which
+					the read cursor should set on
+					records: LOCK_S or LOCK_X; the
+					latter is possible in
+					SELECT FOR UPDATE */
+	ulint			gap_mode,/* in: LOCK_ORDINARY, LOCK_GAP, or
+					LOCK_REC_NOT_GAP */
+	que_thr_t*		thr);	/* in: query thread */
+/*************************************************************************
+Checks if locks of other transactions prevent an immediate read, or passing
+over by a read cursor, of a clustered index record. If they do, first tests
+if the query thread should anyway be suspended for some reason; if not, then
+puts the transaction and the query thread to the lock wait state and inserts a
+waiting request for a record lock to the lock queue. Sets the requested mode
+lock on the record. This is an alternative version of
+lock_clust_rec_read_check_and_lock() that does not require the parameter
+"offsets". */
+UNIV_INTERN
+ulint
+lock_clust_rec_read_check_and_lock_alt(
+/*===================================*/
+					/* out: DB_SUCCESS,
+					DB_LOCK_WAIT, DB_DEADLOCK, or
+					DB_QUE_THR_SUSPENDED */
+	ulint			flags,	/* in: if BTR_NO_LOCKING_FLAG
+					bit is set, does nothing */
+	const buf_block_t*	block,	/* in: buffer block of rec */
+	const rec_t*		rec,	/* in: user record or page
+					supremum record which should
+					be read or passed over by a
+					read cursor */
+	dict_index_t*		index,	/* in: clustered index */
+	enum lock_mode		mode,	/* in: mode of the lock which
+					the read cursor should set on
+					records: LOCK_S or LOCK_X; the
+					latter is possible in
+					SELECT FOR UPDATE */
+	ulint			gap_mode,/* in: LOCK_ORDINARY, LOCK_GAP, or
+					LOCK_REC_NOT_GAP */
+	que_thr_t*		thr);	/* in: query thread */
+/*************************************************************************
+Checks that a record is seen in a consistent read. */
+UNIV_INTERN
+ibool
+lock_clust_rec_cons_read_sees(
+/*==========================*/
+				/* out: TRUE if sees, or FALSE if an earlier
+				version of the record should be retrieved */
+	const rec_t*	rec,	/* in: user record which should be read or
+				passed over by a read cursor */
+	dict_index_t*	index,	/* in: clustered index */
+	const ulint*	offsets,/* in: rec_get_offsets(rec, index) */
+	read_view_t*	view);	/* in: consistent read view */
+/*************************************************************************
+Checks that a non-clustered index record is seen in a consistent read. */
+UNIV_INTERN
+ulint
+lock_sec_rec_cons_read_sees(
+/*========================*/
+					/* out: TRUE if certainly
+					sees, or FALSE if an earlier
+					version of the clustered index
+					record might be needed: NOTE
+					that a non-clustered index
+					page contains so little
+					information on its
+					modifications that also in the
+					case FALSE, the present
+					version of rec may be the
+					right, but we must check this
+					from the clustered index
+					record */
+	const rec_t*		rec,	/* in: user record which
+					should be read or passed over
+					by a read cursor */
+	const read_view_t*	view);	/* in: consistent read view */
+/*************************************************************************
+Locks the specified database table in the mode given. If the lock cannot
+be granted immediately, the query thread is put to wait. */
+UNIV_INTERN
+ulint
+lock_table(
+/*=======*/
+				/* out: DB_SUCCESS, DB_LOCK_WAIT,
+				DB_DEADLOCK, or DB_QUE_THR_SUSPENDED */
+	ulint		flags,	/* in: if BTR_NO_LOCKING_FLAG bit is set,
+				does nothing */
+	dict_table_t*	table,	/* in: database table in dictionary cache */
+	enum lock_mode	mode,	/* in: lock mode */
+	que_thr_t*	thr);	/* in: query thread */
+/*************************************************************************
+Checks if there are any locks set on the table. */
+UNIV_INTERN
+ibool
+lock_is_on_table(
+/*=============*/
+				/* out: TRUE if there are lock(s) */
+	dict_table_t*	table);	/* in: database table in dictionary cache */
+/*****************************************************************
+Removes a granted record lock of a transaction from the queue and grants
+locks to other transactions waiting in the queue if they now are entitled
+to a lock. */
+UNIV_INTERN
+void
+lock_rec_unlock(
+/*============*/
+	trx_t*			trx,	/* in: transaction that has
+					set a record lock */
+	const buf_block_t*	block,	/* in: buffer block containing rec */
+	const rec_t*		rec,	/* in: record */
+	enum lock_mode		lock_mode);/* in: LOCK_S or LOCK_X */
+/*************************************************************************
+Releases a table lock.
+Releases possible other transactions waiting for this lock. */
+UNIV_INTERN
+void
+lock_table_unlock(
+/*==============*/
+	lock_t*	lock);	/* in: lock */
+/*************************************************************************
+Releases transaction locks, and releases possible other transactions waiting
+because of these locks. */
+UNIV_INTERN
+void
+lock_release_off_kernel(
+/*====================*/
+	trx_t*	trx);	/* in: transaction */
+/*************************************************************************
+Cancels a waiting lock request and releases possible other transactions
+waiting behind it. */
+UNIV_INTERN
+void
+lock_cancel_waiting_and_release(
+/*============================*/
+	lock_t*	lock);	/* in: waiting lock request */
+
+/*************************************************************************
+Removes locks on a table to be dropped or truncated.
+If remove_also_table_sx_locks is TRUE then table-level S and X locks are
+also removed in addition to other table-level and record-level locks.
+No lock, that is going to be removed, is allowed to be a wait lock. */
+UNIV_INTERN
+void
+lock_remove_all_on_table(
+/*=====================*/
+	dict_table_t*	table,			/* in: table to be dropped
+						or truncated */
+	ibool		remove_also_table_sx_locks);/* in: also removes
+						table S and X locks */
+
+/*************************************************************************
+Calculates the fold value of a page file address: used in inserting or
+searching for a lock in the hash table. */
+UNIV_INLINE
+ulint
+lock_rec_fold(
+/*==========*/
+			/* out: folded value */
+	ulint	space,	/* in: space */
+	ulint	page_no)/* in: page number */
+	__attribute__((const));
+/*************************************************************************
+Calculates the hash value of a page file address: used in inserting or
+searching for a lock in the hash table. */
+UNIV_INLINE
+ulint
+lock_rec_hash(
+/*==========*/
+			/* out: hashed value */
+	ulint	space,	/* in: space */
+	ulint	page_no);/* in: page number */
+
+/**************************************************************************
+Looks for a set bit in a record lock bitmap. Returns ULINT_UNDEFINED,
+if none found. */
+UNIV_INTERN
+ulint
+lock_rec_find_set_bit(
+/*==================*/
+				/* out: bit index == heap number of
+				the record, or ULINT_UNDEFINED if none found */
+	const lock_t*	lock);	/* in: record lock with at least one bit set */
+
+/*************************************************************************
+Gets the source table of an ALTER TABLE transaction.  The table must be
+covered by an IX or IS table lock. */
+UNIV_INTERN
+dict_table_t*
+lock_get_src_table(
+/*===============*/
+				/* out: the source table of transaction,
+				if it is covered by an IX or IS table lock;
+				dest if there is no source table, and
+				NULL if the transaction is locking more than
+				two tables or an inconsistency is found */
+	trx_t*		trx,	/* in: transaction */
+	dict_table_t*	dest,	/* in: destination of ALTER TABLE */
+	enum lock_mode*	mode);	/* out: lock mode of the source table */
+/*************************************************************************
+Determine if the given table is exclusively "owned" by the given
+transaction, i.e., transaction holds LOCK_IX and possibly LOCK_AUTO_INC
+on the table. */
+UNIV_INTERN
+ibool
+lock_is_table_exclusive(
+/*====================*/
+				/* out: TRUE if table is only locked by trx,
+				with LOCK_IX, and possibly LOCK_AUTO_INC */
+	dict_table_t*	table,	/* in: table */
+	trx_t*		trx);	/* in: transaction */
+/*************************************************************************
+Checks if a lock request lock1 has to wait for request lock2. */
+UNIV_INTERN
+ibool
+lock_has_to_wait(
+/*=============*/
+				/* out: TRUE if lock1 has to wait for
+				lock2 to be removed */
+	const lock_t*	lock1,	/* in: waiting lock */
+	const lock_t*	lock2);	/* in: another lock; NOTE that it is
+				assumed that this has a lock bit set
+				on the same record as in lock1 if the
+				locks are record locks */
+/*************************************************************************
+Checks that a transaction id is sensible, i.e., not in the future. */
+UNIV_INTERN
+ibool
+lock_check_trx_id_sanity(
+/*=====================*/
+					/* out: TRUE if ok */
+	dulint		trx_id,		/* in: trx id */
+	const rec_t*	rec,		/* in: user record */
+	dict_index_t*	index,		/* in: clustered index */
+	const ulint*	offsets,	/* in: rec_get_offsets(rec, index) */
+	ibool		has_kernel_mutex);/* in: TRUE if the caller owns the
+					kernel mutex */
+/*************************************************************************
+Prints info of a table lock. */
+UNIV_INTERN
+void
+lock_table_print(
+/*=============*/
+	FILE*		file,	/* in: file where to print */
+	const lock_t*	lock);	/* in: table type lock */
+/*************************************************************************
+Prints info of a record lock. */
+UNIV_INTERN
+void
+lock_rec_print(
+/*===========*/
+	FILE*		file,	/* in: file where to print */
+	const lock_t*	lock);	/* in: record type lock */
+/*************************************************************************
+Prints info of locks for all transactions. */
+UNIV_INTERN
+void
+lock_print_info_summary(
+/*====================*/
+	FILE*	file);	/* in: file where to print */
+/*************************************************************************
+Prints info of locks for each transaction. */
+UNIV_INTERN
+void
+lock_print_info_all_transactions(
+/*=============================*/
+	FILE*	file);	/* in: file where to print */
+/*************************************************************************
+Return approximate number or record locks (bits set in the bitmap) for
+this transaction. Since delete-marked records may be removed, the
+record count will not be precise. */
+UNIV_INTERN
+ulint
+lock_number_of_rows_locked(
+/*=======================*/
+	trx_t*	trx);	/* in: transaction */
+/***********************************************************************
+Release all the transaction's autoinc locks. */
+UNIV_INTERN
+void
+lock_release_autoinc_locks(
+/*=======================*/
+	trx_t*		trx);		/* in/out: transaction */
+
+/***********************************************************************
+Gets the type of a lock. Non-inline version for using outside of the
+lock module. */
+UNIV_INTERN
+ulint
+lock_get_type(
+/*==========*/
+				/* out: LOCK_TABLE or LOCK_REC */
+	const lock_t*	lock);	/* in: lock */
+
+/***********************************************************************
+Gets the id of the transaction owning a lock. */
+UNIV_INTERN
+ullint
+lock_get_trx_id(
+/*============*/
+				/* out: transaction id */
+	const lock_t*	lock);	/* in: lock */
+
+/***********************************************************************
+Gets the mode of a lock in a human readable string.
+The string should not be free()'d or modified. */
+
+const char*
+lock_get_mode_str(
+/*==============*/
+				/* out: lock mode */
+	const lock_t*	lock);	/* in: lock */
+
+/***********************************************************************
+Gets the type of a lock in a human readable string.
+The string should not be free()'d or modified. */
+
+const char*
+lock_get_type_str(
+/*==============*/
+				/* out: lock type */
+	const lock_t*	lock);	/* in: lock */
+
+/***********************************************************************
+Gets the id of the table on which the lock is. */
+UNIV_INTERN
+ullint
+lock_get_table_id(
+/*==============*/
+				/* out: id of the table */
+	const lock_t*	lock);	/* in: lock */
+
+/***********************************************************************
+Gets the name of the table on which the lock is.
+The string should not be free()'d or modified. */
+
+const char*
+lock_get_table_name(
+/*================*/
+				/* out: name of the table */
+	const lock_t*	lock);	/* in: lock */
+
+/***********************************************************************
+For a record lock, gets the index on which the lock is. */
+
+const dict_index_t*
+lock_rec_get_index(
+/*===============*/
+				/* out: index */
+	const lock_t*	lock);	/* in: lock */
+
+/***********************************************************************
+For a record lock, gets the name of the index on which the lock is.
+The string should not be free()'d or modified. */
+
+const char*
+lock_rec_get_index_name(
+/*====================*/
+				/* out: name of the index */
+	const lock_t*	lock);	/* in: lock */
+
+/***********************************************************************
+For a record lock, gets the tablespace number on which the lock is. */
+UNIV_INTERN
+ulint
+lock_rec_get_space_id(
+/*==================*/
+				/* out: tablespace number */
+	const lock_t*	lock);	/* in: lock */
+
+/***********************************************************************
+For a record lock, gets the page number on which the lock is. */
+UNIV_INTERN
+ulint
+lock_rec_get_page_no(
+/*=================*/
+				/* out: page number */
+	const lock_t*	lock);	/* in: lock */
+
+/* Lock modes and types */
+#define LOCK_MODE_MASK	0xFUL	/* mask used to extract mode from the
+				type_mode field in a lock */
+/* Lock types */
+#define LOCK_TABLE	16	/* these type values should be so high that */
+#define	LOCK_REC	32	/* they can be ORed to the lock mode */
+#define LOCK_TYPE_MASK	0xF0UL	/* mask used to extract lock type from the
+				type_mode field in a lock */
+/* Waiting lock flag */
+#define LOCK_WAIT	256	/* this wait bit should be so high that
+				it can be ORed to the lock mode and type;
+				when this bit is set, it means that the
+				lock has not yet been granted, it is just
+				waiting for its turn in the wait queue */
+/* Precise modes */
+#define LOCK_ORDINARY	0	/* this flag denotes an ordinary next-key lock
+				in contrast to LOCK_GAP or LOCK_REC_NOT_GAP */
+#define LOCK_GAP	512	/* this gap bit should be so high that
+				it can be ORed to the other flags;
+				when this bit is set, it means that the
+				lock holds only on the gap before the record;
+				for instance, an x-lock on the gap does not
+				give permission to modify the record on which
+				the bit is set; locks of this type are created
+				when records are removed from the index chain
+				of records */
+#define LOCK_REC_NOT_GAP 1024	/* this bit means that the lock is only on
+				the index record and does NOT block inserts
+				to the gap before the index record; this is
+				used in the case when we retrieve a record
+				with a unique key, and is also used in
+				locking plain SELECTs (not part of UPDATE
+				or DELETE) when the user has set the READ
+				COMMITTED isolation level */
+#define LOCK_INSERT_INTENTION 2048 /* this bit is set when we place a waiting
+				gap type record lock request in order to let
+				an insert of an index record to wait until
+				there are no conflicting locks by other
+				transactions on the gap; note that this flag
+				remains set when the waiting lock is granted,
+				or if the lock is inherited to a neighboring
+				record */
+
+/* When lock bits are reset, the following flags are available: */
+#define LOCK_RELEASE_WAIT	1
+#define LOCK_NOT_RELEASE_WAIT	2
+
+/* Lock operation struct */
+typedef struct lock_op_struct	lock_op_t;
+struct lock_op_struct{
+	dict_table_t*	table;	/* table to be locked */
+	enum lock_mode	mode;	/* lock mode */
+};
+
+#define LOCK_OP_START		1
+#define LOCK_OP_COMPLETE	2
+
+/* The lock system struct */
+struct lock_sys_struct{
+	hash_table_t*	rec_hash;	/* hash table of the record locks */
+};
+
+/* The lock system */
+extern lock_sys_t*	lock_sys;
+
+
+#ifndef UNIV_NONINL
+#include "lock0lock.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/lock0lock.ic b/storage/xtradb/include/lock0lock.ic
new file mode 100644
index 00000000000..f978cc70678
--- /dev/null
+++ b/storage/xtradb/include/lock0lock.ic
@@ -0,0 +1,123 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The transaction lock system
+
+Created 5/7/1996 Heikki Tuuri
+*******************************************************/
+
+#include "sync0sync.h"
+#include "srv0srv.h"
+#include "dict0dict.h"
+#include "row0row.h"
+#include "trx0sys.h"
+#include "trx0trx.h"
+#include "buf0buf.h"
+#include "page0page.h"
+#include "page0cur.h"
+#include "row0vers.h"
+#include "que0que.h"
+#include "btr0cur.h"
+#include "read0read.h"
+#include "log0recv.h"
+
+/*************************************************************************
+Calculates the fold value of a page file address: used in inserting or
+searching for a lock in the hash table. */
+UNIV_INLINE
+ulint
+lock_rec_fold(
+/*==========*/
+			/* out: folded value */
+	ulint	space,	/* in: space */
+	ulint	page_no)/* in: page number */
+{
+	return(ut_fold_ulint_pair(space, page_no));
+}
+
+/*************************************************************************
+Calculates the hash value of a page file address: used in inserting or
+searching for a lock in the hash table. */
+UNIV_INLINE
+ulint
+lock_rec_hash(
+/*==========*/
+			/* out: hashed value */
+	ulint	space,	/* in: space */
+	ulint	page_no)/* in: page number */
+{
+	return(hash_calc_hash(lock_rec_fold(space, page_no),
+			      lock_sys->rec_hash));
+}
+
+/*************************************************************************
+Checks if some transaction has an implicit x-lock on a record in a clustered
+index. */
+UNIV_INLINE
+trx_t*
+lock_clust_rec_some_has_impl(
+/*=========================*/
+				/* out: transaction which has the x-lock, or
+				NULL */
+	const rec_t*	rec,	/* in: user record */
+	dict_index_t*	index,	/* in: clustered index */
+	const ulint*	offsets)/* in: rec_get_offsets(rec, index) */
+{
+	dulint	trx_id;
+
+	ut_ad(mutex_own(&kernel_mutex));
+	ut_ad(dict_index_is_clust(index));
+	ut_ad(page_rec_is_user_rec(rec));
+
+	trx_id = row_get_rec_trx_id(rec, index, offsets);
+
+	if (trx_is_active(trx_id)) {
+		/* The modifying or inserting transaction is active */
+
+		return(trx_get_on_id(trx_id));
+	}
+
+	return(NULL);
+}
+
+/*************************************************************************
+Gets the heap_no of the smallest user record on a page. */
+UNIV_INLINE
+ulint
+lock_get_min_heap_no(
+/*=================*/
+					/* out: heap_no of smallest
+					user record, or
+					PAGE_HEAP_NO_SUPREMUM */
+	const buf_block_t*	block)	/* in: buffer block */
+{
+	const page_t*	page	= block->frame;
+
+	if (page_is_comp(page)) {
+		return(rec_get_heap_no_new(
+			       page
+			       + rec_get_next_offs(page + PAGE_NEW_INFIMUM,
+						   TRUE)));
+	} else {
+		return(rec_get_heap_no_old(
+			       page
+			       + rec_get_next_offs(page + PAGE_OLD_INFIMUM,
+						   FALSE)));
+	}
+}
diff --git a/storage/xtradb/include/lock0priv.h b/storage/xtradb/include/lock0priv.h
new file mode 100644
index 00000000000..0a0d41e6aaa
--- /dev/null
+++ b/storage/xtradb/include/lock0priv.h
@@ -0,0 +1,106 @@
+/*****************************************************************************
+
+Copyright (c) 2007, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Lock module internal structures and methods.
+
+Created July 12, 2007 Vasil Dimov
+*******************************************************/
+
+#ifndef lock0priv_h
+#define lock0priv_h
+
+#ifndef LOCK_MODULE_IMPLEMENTATION
+/* If you need to access members of the structures defined in this
+file, please write appropriate functions that retrieve them and put
+those functions in lock/ */
+#error Do not include lock0priv.h outside of the lock/ module
+#endif
+
+#include "univ.i"
+#include "dict0types.h"
+#include "hash0hash.h"
+#include "trx0types.h"
+#include "ut0lst.h"
+
+/* A table lock */
+typedef struct lock_table_struct	lock_table_t;
+struct lock_table_struct {
+	dict_table_t*	table;		/* database table in dictionary
+					cache */
+	UT_LIST_NODE_T(lock_t)
+			locks;		/* list of locks on the same
+					table */
+};
+
+/* Record lock for a page */
+typedef struct lock_rec_struct		lock_rec_t;
+struct lock_rec_struct {
+	ulint	space;			/* space id */
+	ulint	page_no;		/* page number */
+	ulint	n_bits;			/* number of bits in the lock
+					bitmap; NOTE: the lock bitmap is
+					placed immediately after the
+					lock struct */
+};
+
+/* Lock struct */
+struct lock_struct {
+	trx_t*		trx;		/* transaction owning the
+					lock */
+	UT_LIST_NODE_T(lock_t)
+			trx_locks;	/* list of the locks of the
+					transaction */
+	ulint		type_mode;	/* lock type, mode, LOCK_GAP or
+					LOCK_REC_NOT_GAP,
+					LOCK_INSERT_INTENTION,
+					wait flag, ORed */
+	hash_node_t	hash;		/* hash chain node for a record
+					lock */
+	dict_index_t*	index;		/* index for a record lock */
+	union {
+		lock_table_t	tab_lock;/* table lock */
+		lock_rec_t	rec_lock;/* record lock */
+	} un_member;
+};
+
+/*************************************************************************
+Gets the type of a lock. */
+UNIV_INLINE
+ulint
+lock_get_type_low(
+/*==============*/
+				/* out: LOCK_TABLE or LOCK_REC */
+	const lock_t*	lock);	/* in: lock */
+
+/*************************************************************************
+Gets the previous record lock set on a record. */
+
+const lock_t*
+lock_rec_get_prev(
+/*==============*/
+				/* out: previous lock on the same
+				record, NULL if none exists */
+	const lock_t*	in_lock,/* in: record lock */
+	ulint		heap_no);/* in: heap number of the record */
+
+#ifndef UNIV_NONINL
+#include "lock0priv.ic"
+#endif
+
+#endif /* lock0priv_h */
diff --git a/storage/xtradb/include/lock0priv.ic b/storage/xtradb/include/lock0priv.ic
new file mode 100644
index 00000000000..ae633a4fc61
--- /dev/null
+++ b/storage/xtradb/include/lock0priv.ic
@@ -0,0 +1,48 @@
+/*****************************************************************************
+
+Copyright (c) 2007, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Lock module internal inline methods.
+
+Created July 16, 2007 Vasil Dimov
+*******************************************************/
+
+/* This file contains only methods which are used in
+lock/lock0* files, other than lock/lock0lock.c.
+I.e. lock/lock0lock.c contains more internal inline
+methods but they are used only in that file. */
+
+#ifndef LOCK_MODULE_IMPLEMENTATION
+#error Do not include lock0priv.ic outside of the lock/ module
+#endif
+
+/*************************************************************************
+Gets the type of a lock. */
+UNIV_INLINE
+ulint
+lock_get_type_low(
+/*==============*/
+				/* out: LOCK_TABLE or LOCK_REC */
+	const lock_t*	lock)	/* in: lock */
+{
+	ut_ad(lock);
+
+	return(lock->type_mode & LOCK_TYPE_MASK);
+}
+
+/* vim: set filetype=c: */
diff --git a/storage/xtradb/include/lock0types.h b/storage/xtradb/include/lock0types.h
new file mode 100644
index 00000000000..52631b56532
--- /dev/null
+++ b/storage/xtradb/include/lock0types.h
@@ -0,0 +1,44 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The transaction lock system global types
+
+Created 5/7/1996 Heikki Tuuri
+*******************************************************/
+
+#ifndef lock0types_h
+#define lock0types_h
+
+#define lock_t ib_lock_t
+typedef struct lock_struct	lock_t;
+typedef struct lock_sys_struct	lock_sys_t;
+
+/* Basic lock modes */
+enum lock_mode {
+	LOCK_IS = 0,	/* intention shared */
+	LOCK_IX,	/* intention exclusive */
+	LOCK_S,		/* shared */
+	LOCK_X,		/* exclusive */
+	LOCK_AUTO_INC,	/* locks the auto-inc counter of a table
+			in an exclusive mode */
+	LOCK_NONE,	/* this is used elsewhere to note consistent read */
+	LOCK_NUM = LOCK_NONE/* number of lock modes */
+};
+
+#endif
diff --git a/storage/xtradb/include/log0log.h b/storage/xtradb/include/log0log.h
new file mode 100644
index 00000000000..51b57ae929c
--- /dev/null
+++ b/storage/xtradb/include/log0log.h
@@ -0,0 +1,893 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Database log
+
+Created 12/9/1995 Heikki Tuuri
+*******************************************************/
+
+#ifndef log0log_h
+#define log0log_h
+
+#include "univ.i"
+#include "ut0byte.h"
+#include "sync0sync.h"
+#include "sync0rw.h"
+
+typedef struct log_struct	log_t;
+typedef struct log_group_struct	log_group_t;
+
+#ifdef UNIV_DEBUG
+extern	ibool	log_do_write;
+extern	ibool	log_debug_writes;
+#else /* UNIV_DEBUG */
+# define log_do_write TRUE
+#endif /* UNIV_DEBUG */
+
+/* Wait modes for log_write_up_to */
+#define LOG_NO_WAIT		91
+#define LOG_WAIT_ONE_GROUP	92
+#define	LOG_WAIT_ALL_GROUPS	93
+#define LOG_MAX_N_GROUPS	32
+
+/********************************************************************
+Sets the global variable log_fsp_current_free_limit. Also makes a checkpoint,
+so that we know that the limit has been written to a log checkpoint field
+on disk. */
+UNIV_INTERN
+void
+log_fsp_current_free_limit_set_and_checkpoint(
+/*==========================================*/
+	ulint	limit);	/* in: limit to set */
+/***********************************************************************
+Calculates where in log files we find a specified lsn. */
+UNIV_INTERN
+ulint
+log_calc_where_lsn_is(
+/*==================*/
+						/* out: log file number */
+	ib_int64_t*	log_file_offset,	/* out: offset in that file
+						(including the header) */
+	ib_uint64_t	first_header_lsn,	/* in: first log file start
+						lsn */
+	ib_uint64_t	lsn,			/* in: lsn whose position to
+						determine */
+	ulint		n_log_files,		/* in: total number of log
+						files */
+	ib_int64_t	log_file_size);		/* in: log file size
+						(including the header) */
+/****************************************************************
+Writes to the log the string given. The log must be released with
+log_release. */
+UNIV_INLINE
+ib_uint64_t
+log_reserve_and_write_fast(
+/*=======================*/
+				/* out: end lsn of the log record,
+				zero if did not succeed */
+	byte*		str,	/* in: string */
+	ulint		len,	/* in: string length */
+	ib_uint64_t*	start_lsn,/* out: start lsn of the log record */
+	ibool*		success);/* out: TRUE if success */
+/***************************************************************************
+Releases the log mutex. */
+UNIV_INLINE
+void
+log_release(void);
+/*=============*/
+/***************************************************************************
+Checks if there is need for a log buffer flush or a new checkpoint, and does
+this if yes. Any database operation should call this when it has modified
+more than about 4 pages. NOTE that this function may only be called when the
+OS thread owns no synchronization objects except the dictionary mutex. */
+UNIV_INLINE
+void
+log_free_check(void);
+/*================*/
+/****************************************************************
+Opens the log for log_write_low. The log must be closed with log_close and
+released with log_release. */
+UNIV_INTERN
+ib_uint64_t
+log_reserve_and_open(
+/*=================*/
+			/* out: start lsn of the log record */
+	ulint	len);	/* in: length of data to be catenated */
+/****************************************************************
+Writes to the log the string given. It is assumed that the caller holds the
+log mutex. */
+UNIV_INTERN
+void
+log_write_low(
+/*==========*/
+	byte*	str,		/* in: string */
+	ulint	str_len);	/* in: string length */
+/****************************************************************
+Closes the log. */
+UNIV_INTERN
+ib_uint64_t
+log_close(void);
+/*===========*/
+			/* out: lsn */
+/****************************************************************
+Gets the current lsn. */
+UNIV_INLINE
+ib_uint64_t
+log_get_lsn(void);
+/*=============*/
+			/* out: current lsn */
+/**********************************************************
+Initializes the log. */
+UNIV_INTERN
+void
+log_init(void);
+/*==========*/
+/**********************************************************************
+Inits a log group to the log system. */
+UNIV_INTERN
+void
+log_group_init(
+/*===========*/
+	ulint	id,			/* in: group id */
+	ulint	n_files,		/* in: number of log files */
+	ulint	file_size,		/* in: log file size in bytes */
+	ulint	space_id,		/* in: space id of the file space
+					which contains the log files of this
+					group */
+	ulint	archive_space_id);	/* in: space id of the file space
+					which contains some archived log
+					files for this group; currently, only
+					for the first log group this is
+					used */
+/**********************************************************
+Completes an i/o to a log file. */
+UNIV_INTERN
+void
+log_io_complete(
+/*============*/
+	log_group_t*	group);	/* in: log group */
+/**********************************************************
+This function is called, e.g., when a transaction wants to commit. It checks
+that the log has been written to the log file up to the last log entry written
+by the transaction. If there is a flush running, it waits and checks if the
+flush flushed enough. If not, starts a new flush. */
+UNIV_INTERN
+void
+log_write_up_to(
+/*============*/
+	ib_uint64_t	lsn,	/* in: log sequence number up to which
+				the log should be written,
+				IB_ULONGLONG_MAX if not specified */
+	ulint		wait,	/* in: LOG_NO_WAIT, LOG_WAIT_ONE_GROUP,
+				or LOG_WAIT_ALL_GROUPS */
+	ibool		flush_to_disk);
+				/* in: TRUE if we want the written log
+				also to be flushed to disk */
+/********************************************************************
+Does a syncronous flush of the log buffer to disk. */
+UNIV_INTERN
+void
+log_buffer_flush_to_disk(void);
+/*==========================*/
+/********************************************************************
+Advances the smallest lsn for which there are unflushed dirty blocks in the
+buffer pool and also may make a new checkpoint. NOTE: this function may only
+be called if the calling thread owns no synchronization objects! */
+UNIV_INTERN
+ibool
+log_preflush_pool_modified_pages(
+/*=============================*/
+					/* out: FALSE if there was a
+					flush batch of the same type
+					running, which means that we
+					could not start this flush
+					batch */
+	ib_uint64_t	new_oldest,	/* in: try to advance
+					oldest_modified_lsn at least
+					to this lsn */
+	ibool		sync);		/* in: TRUE if synchronous
+					operation is desired */
+/**********************************************************
+Makes a checkpoint. Note that this function does not flush dirty
+blocks from the buffer pool: it only checks what is lsn of the oldest
+modification in the pool, and writes information about the lsn in
+log files. Use log_make_checkpoint_at to flush also the pool. */
+UNIV_INTERN
+ibool
+log_checkpoint(
+/*===========*/
+				/* out: TRUE if success, FALSE if a checkpoint
+				write was already running */
+	ibool	sync,		/* in: TRUE if synchronous operation is
+				desired */
+	ibool	write_always);	/* in: the function normally checks if the
+				the new checkpoint would have a greater
+				lsn than the previous one: if not, then no
+				physical write is done; by setting this
+				parameter TRUE, a physical write will always be
+				made to log files */
+/********************************************************************
+Makes a checkpoint at a given lsn or later. */
+UNIV_INTERN
+void
+log_make_checkpoint_at(
+/*===================*/
+	ib_uint64_t	lsn,		/* in: make a checkpoint at this or a
+					later lsn, if IB_ULONGLONG_MAX, makes
+					a checkpoint at the latest lsn */
+	ibool		write_always);	/* in: the function normally checks if
+					the the new checkpoint would have a
+					greater lsn than the previous one: if
+					not, then no physical write is done;
+					by setting this parameter TRUE, a
+					physical write will always be made to
+					log files */
+/********************************************************************
+Makes a checkpoint at the latest lsn and writes it to first page of each
+data file in the database, so that we know that the file spaces contain
+all modifications up to that lsn. This can only be called at database
+shutdown. This function also writes all log in log files to the log archive. */
+UNIV_INTERN
+void
+logs_empty_and_mark_files_at_shutdown(void);
+/*=======================================*/
+/**********************************************************
+Reads a checkpoint info from a log group header to log_sys->checkpoint_buf. */
+UNIV_INTERN
+void
+log_group_read_checkpoint_info(
+/*===========================*/
+	log_group_t*	group,	/* in: log group */
+	ulint		field);	/* in: LOG_CHECKPOINT_1 or LOG_CHECKPOINT_2 */
+/***********************************************************************
+Gets info from a checkpoint about a log group. */
+UNIV_INTERN
+void
+log_checkpoint_get_nth_group_info(
+/*==============================*/
+	byte*	buf,	/* in: buffer containing checkpoint info */
+	ulint	n,	/* in: nth slot */
+	ulint*	file_no,/* out: archived file number */
+	ulint*	offset);/* out: archived file offset */
+/**********************************************************
+Writes checkpoint info to groups. */
+UNIV_INTERN
+void
+log_groups_write_checkpoint_info(void);
+/*==================================*/
+#ifdef UNIV_HOTBACKUP
+/**********************************************************
+Writes info to a buffer of a log group when log files are created in
+backup restoration. */
+UNIV_INTERN
+void
+log_reset_first_header_and_checkpoint(
+/*==================================*/
+	byte*		hdr_buf,/* in: buffer which will be written to the
+				start of the first log file */
+	ib_uint64_t	start);	/* in: lsn of the start of the first log file;
+				we pretend that there is a checkpoint at
+				start + LOG_BLOCK_HDR_SIZE */
+#endif /* UNIV_HOTBACKUP */
+/************************************************************************
+Starts an archiving operation. */
+UNIV_INTERN
+ibool
+log_archive_do(
+/*===========*/
+			/* out: TRUE if succeed, FALSE if an archiving
+			operation was already running */
+	ibool	sync,	/* in: TRUE if synchronous operation is desired */
+	ulint*	n_bytes);/* out: archive log buffer size, 0 if nothing to
+			archive */
+/********************************************************************
+Writes the log contents to the archive up to the lsn when this function was
+called, and stops the archiving. When archiving is started again, the archived
+log file numbers start from a number one higher, so that the archiving will
+not write again to the archived log files which exist when this function
+returns. */
+UNIV_INTERN
+ulint
+log_archive_stop(void);
+/*==================*/
+				/* out: DB_SUCCESS or DB_ERROR */
+/********************************************************************
+Starts again archiving which has been stopped. */
+UNIV_INTERN
+ulint
+log_archive_start(void);
+/*===================*/
+			/* out: DB_SUCCESS or DB_ERROR */
+/********************************************************************
+Stop archiving the log so that a gap may occur in the archived log files. */
+UNIV_INTERN
+ulint
+log_archive_noarchivelog(void);
+/*==========================*/
+			/* out: DB_SUCCESS or DB_ERROR */
+/********************************************************************
+Start archiving the log so that a gap may occur in the archived log files. */
+UNIV_INTERN
+ulint
+log_archive_archivelog(void);
+/*========================*/
+			/* out: DB_SUCCESS or DB_ERROR */
+/**********************************************************
+Generates an archived log file name. */
+UNIV_INTERN
+void
+log_archived_file_name_gen(
+/*=======================*/
+	char*	buf,	/* in: buffer where to write */
+	ulint	id,	/* in: group id */
+	ulint	file_no);/* in: file number */
+/************************************************************************
+Checks that there is enough free space in the log to start a new query step.
+Flushes the log buffer or makes a new checkpoint if necessary. NOTE: this
+function may only be called if the calling thread owns no synchronization
+objects! */
+UNIV_INTERN
+void
+log_check_margins(void);
+/*===================*/
+/**********************************************************
+Reads a specified log segment to a buffer. */
+UNIV_INTERN
+void
+log_group_read_log_seg(
+/*===================*/
+	ulint		type,		/* in: LOG_ARCHIVE or LOG_RECOVER */
+	byte*		buf,		/* in: buffer where to read */
+	log_group_t*	group,		/* in: log group */
+	ib_uint64_t	start_lsn,	/* in: read area start */
+	ib_uint64_t	end_lsn);	/* in: read area end */
+/**********************************************************
+Writes a buffer to a log file group. */
+UNIV_INTERN
+void
+log_group_write_buf(
+/*================*/
+	log_group_t*	group,		/* in: log group */
+	byte*		buf,		/* in: buffer */
+	ulint		len,		/* in: buffer len; must be divisible
+					by OS_FILE_LOG_BLOCK_SIZE */
+	ib_uint64_t	start_lsn,	/* in: start lsn of the buffer; must
+					be divisible by
+					OS_FILE_LOG_BLOCK_SIZE */
+	ulint		new_data_offset);/* in: start offset of new data in
+					buf: this parameter is used to decide
+					if we have to write a new log file
+					header */
+/************************************************************
+Sets the field values in group to correspond to a given lsn. For this function
+to work, the values must already be correctly initialized to correspond to
+some lsn, for instance, a checkpoint lsn. */
+UNIV_INTERN
+void
+log_group_set_fields(
+/*=================*/
+	log_group_t*	group,	/* in: group */
+	ib_uint64_t	lsn);	/* in: lsn for which the values should be
+				set */
+/**********************************************************
+Calculates the data capacity of a log group, when the log file headers are not
+included. */
+UNIV_INTERN
+ulint
+log_group_get_capacity(
+/*===================*/
+				/* out: capacity in bytes */
+	log_group_t*	group);	/* in: log group */
+/****************************************************************
+Gets a log block flush bit. */
+UNIV_INLINE
+ibool
+log_block_get_flush_bit(
+/*====================*/
+				/* out: TRUE if this block was the first
+				to be written in a log flush */
+	byte*	log_block);	/* in: log block */
+/****************************************************************
+Gets a log block number stored in the header. */
+UNIV_INLINE
+ulint
+log_block_get_hdr_no(
+/*=================*/
+				/* out: log block number stored in the block
+				header */
+	byte*	log_block);	/* in: log block */
+/****************************************************************
+Gets a log block data length. */
+UNIV_INLINE
+ulint
+log_block_get_data_len(
+/*===================*/
+				/* out: log block data length measured as a
+				byte offset from the block start */
+	byte*	log_block);	/* in: log block */
+/****************************************************************
+Sets the log block data length. */
+UNIV_INLINE
+void
+log_block_set_data_len(
+/*===================*/
+	byte*	log_block,	/* in: log block */
+	ulint	len);		/* in: data length */
+/****************************************************************
+Calculates the checksum for a log block. */
+UNIV_INLINE
+ulint
+log_block_calc_checksum(
+/*====================*/
+				/* out: checksum */
+	const byte*	block);	/* in: log block */
+/****************************************************************
+Gets a log block checksum field value. */
+UNIV_INLINE
+ulint
+log_block_get_checksum(
+/*===================*/
+					/* out: checksum */
+	const byte*	log_block);	/* in: log block */
+/****************************************************************
+Sets a log block checksum field value. */
+UNIV_INLINE
+void
+log_block_set_checksum(
+/*===================*/
+	byte*	log_block,	/* in: log block */
+	ulint	checksum);	/* in: checksum */
+/****************************************************************
+Gets a log block first mtr log record group offset. */
+UNIV_INLINE
+ulint
+log_block_get_first_rec_group(
+/*==========================*/
+				/* out: first mtr log record group byte offset
+				from the block start, 0 if none */
+	byte*	log_block);	/* in: log block */
+/****************************************************************
+Sets the log block first mtr log record group offset. */
+UNIV_INLINE
+void
+log_block_set_first_rec_group(
+/*==========================*/
+	byte*	log_block,	/* in: log block */
+	ulint	offset);	/* in: offset, 0 if none */
+/****************************************************************
+Gets a log block checkpoint number field (4 lowest bytes). */
+UNIV_INLINE
+ulint
+log_block_get_checkpoint_no(
+/*========================*/
+				/* out: checkpoint no (4 lowest bytes) */
+	byte*	log_block);	/* in: log block */
+/****************************************************************
+Initializes a log block in the log buffer. */
+UNIV_INLINE
+void
+log_block_init(
+/*===========*/
+	byte*		log_block,	/* in: pointer to the log buffer */
+	ib_uint64_t	lsn);		/* in: lsn within the log block */
+/****************************************************************
+Initializes a log block in the log buffer in the old, < 3.23.52 format, where
+there was no checksum yet. */
+UNIV_INLINE
+void
+log_block_init_in_old_format(
+/*=========================*/
+	byte*		log_block,	/* in: pointer to the log buffer */
+	ib_uint64_t	lsn);		/* in: lsn within the log block */
+/****************************************************************
+Converts a lsn to a log block number. */
+UNIV_INLINE
+ulint
+log_block_convert_lsn_to_no(
+/*========================*/
+				/* out: log block number,
+				it is > 0 and <= 1G */
+	ib_uint64_t	lsn);	/* in: lsn of a byte within the block */
+/**********************************************************
+Prints info of the log. */
+UNIV_INTERN
+void
+log_print(
+/*======*/
+	FILE*	file);	/* in: file where to print */
+/**********************************************************
+Peeks the current lsn. */
+UNIV_INTERN
+ibool
+log_peek_lsn(
+/*=========*/
+				/* out: TRUE if success, FALSE if
+				could not get the log system mutex */
+	ib_uint64_t*	lsn);	/* out: if returns TRUE, current lsn is here */
+/**************************************************************************
+Refreshes the statistics used to print per-second averages. */
+UNIV_INTERN
+void
+log_refresh_stats(void);
+/*===================*/
+
+extern log_t*	log_sys;
+
+/* Values used as flags */
+#define LOG_FLUSH	7652559
+#define LOG_CHECKPOINT	78656949
+#ifdef UNIV_LOG_ARCHIVE
+# define LOG_ARCHIVE	11122331
+#endif /* UNIV_LOG_ARCHIVE */
+#define LOG_RECOVER	98887331
+
+/* The counting of lsn's starts from this value: this must be non-zero */
+#define LOG_START_LSN	((ib_uint64_t) (16 * OS_FILE_LOG_BLOCK_SIZE))
+
+#define LOG_BUFFER_SIZE		(srv_log_buffer_size * UNIV_PAGE_SIZE)
+#define LOG_ARCHIVE_BUF_SIZE	(srv_log_buffer_size * UNIV_PAGE_SIZE / 4)
+
+/* Offsets of a log block header */
+#define	LOG_BLOCK_HDR_NO	0	/* block number which must be > 0 and
+					is allowed to wrap around at 2G; the
+					highest bit is set to 1 if this is the
+					first log block in a log flush write
+					segment */
+#define LOG_BLOCK_FLUSH_BIT_MASK 0x80000000UL
+					/* mask used to get the highest bit in
+					the preceding field */
+#define	LOG_BLOCK_HDR_DATA_LEN	4	/* number of bytes of log written to
+					this block */
+#define	LOG_BLOCK_FIRST_REC_GROUP 6	/* offset of the first start of an
+					mtr log record group in this log block,
+					0 if none; if the value is the same
+					as LOG_BLOCK_HDR_DATA_LEN, it means
+					that the first rec group has not yet
+					been catenated to this log block, but
+					if it will, it will start at this
+					offset; an archive recovery can
+					start parsing the log records starting
+					from this offset in this log block,
+					if value not 0 */
+#define LOG_BLOCK_CHECKPOINT_NO	8	/* 4 lower bytes of the value of
+					log_sys->next_checkpoint_no when the
+					log block was last written to: if the
+					block has not yet been written full,
+					this value is only updated before a
+					log buffer flush */
+#define LOG_BLOCK_HDR_SIZE	12	/* size of the log block header in
+					bytes */
+
+/* Offsets of a log block trailer from the end of the block */
+#define	LOG_BLOCK_CHECKSUM	4	/* 4 byte checksum of the log block
+					contents; in InnoDB versions
+					< 3.23.52 this did not contain the
+					checksum but the same value as
+					.._HDR_NO */
+#define	LOG_BLOCK_TRL_SIZE	4	/* trailer size in bytes */
+
+/* Offsets for a checkpoint field */
+#define LOG_CHECKPOINT_NO		0
+#define LOG_CHECKPOINT_LSN		8
+#define LOG_CHECKPOINT_OFFSET		16
+#define LOG_CHECKPOINT_LOG_BUF_SIZE	20
+#define	LOG_CHECKPOINT_ARCHIVED_LSN	24
+#define	LOG_CHECKPOINT_GROUP_ARRAY	32
+
+/* For each value < LOG_MAX_N_GROUPS the following 8 bytes: */
+
+#define LOG_CHECKPOINT_ARCHIVED_FILE_NO	0
+#define LOG_CHECKPOINT_ARCHIVED_OFFSET	4
+
+#define	LOG_CHECKPOINT_ARRAY_END	(LOG_CHECKPOINT_GROUP_ARRAY\
+							+ LOG_MAX_N_GROUPS * 8)
+#define LOG_CHECKPOINT_CHECKSUM_1	LOG_CHECKPOINT_ARRAY_END
+#define LOG_CHECKPOINT_CHECKSUM_2	(4 + LOG_CHECKPOINT_ARRAY_END)
+#define LOG_CHECKPOINT_FSP_FREE_LIMIT	(8 + LOG_CHECKPOINT_ARRAY_END)
+					/* current fsp free limit in
+					tablespace 0, in units of one
+					megabyte; this information is only used
+					by ibbackup to decide if it can
+					truncate unused ends of
+					non-auto-extending data files in space
+					0 */
+#define LOG_CHECKPOINT_FSP_MAGIC_N	(12 + LOG_CHECKPOINT_ARRAY_END)
+					/* this magic number tells if the
+					checkpoint contains the above field:
+					the field was added to
+					InnoDB-3.23.50 */
+#define LOG_CHECKPOINT_SIZE		(16 + LOG_CHECKPOINT_ARRAY_END)
+
+#define LOG_CHECKPOINT_FSP_MAGIC_N_VAL	1441231243
+
+/* Offsets of a log file header */
+#define LOG_GROUP_ID		0	/* log group number */
+#define LOG_FILE_START_LSN	4	/* lsn of the start of data in this
+					log file */
+#define LOG_FILE_NO		12	/* 4-byte archived log file number;
+					this field is only defined in an
+					archived log file */
+#define LOG_FILE_WAS_CREATED_BY_HOT_BACKUP 16
+					/* a 32-byte field which contains
+					the string 'ibbackup' and the
+					creation time if the log file was
+					created by ibbackup --restore;
+					when mysqld is first time started
+					on the restored database, it can
+					print helpful info for the user */
+#define	LOG_FILE_ARCH_COMPLETED	OS_FILE_LOG_BLOCK_SIZE
+					/* this 4-byte field is TRUE when
+					the writing of an archived log file
+					has been completed; this field is
+					only defined in an archived log file */
+#define LOG_FILE_END_LSN	(OS_FILE_LOG_BLOCK_SIZE + 4)
+					/* lsn where the archived log file
+					at least extends: actually the
+					archived log file may extend to a
+					later lsn, as long as it is within the
+					same log block as this lsn; this field
+					is defined only when an archived log
+					file has been completely written */
+#define LOG_CHECKPOINT_1	OS_FILE_LOG_BLOCK_SIZE
+					/* first checkpoint field in the log
+					header; we write alternately to the
+					checkpoint fields when we make new
+					checkpoints; this field is only defined
+					in the first log file of a log group */
+#define LOG_CHECKPOINT_2	(3 * OS_FILE_LOG_BLOCK_SIZE)
+					/* second checkpoint field in the log
+					header */
+#define LOG_FILE_HDR_SIZE	(4 * OS_FILE_LOG_BLOCK_SIZE)
+
+#define LOG_GROUP_OK		301
+#define LOG_GROUP_CORRUPTED	302
+
+/* Log group consists of a number of log files, each of the same size; a log
+group is implemented as a space in the sense of the module fil0fil. */
+
+struct log_group_struct{
+	/* The following fields are protected by log_sys->mutex */
+	ulint		id;		/* log group id */
+	ulint		n_files;	/* number of files in the group */
+	ulint		file_size;	/* individual log file size in bytes,
+					including the log file header */
+	ulint		space_id;	/* file space which implements the log
+					group */
+	ulint		state;		/* LOG_GROUP_OK or
+					LOG_GROUP_CORRUPTED */
+	ib_uint64_t	lsn;		/* lsn used to fix coordinates within
+					the log group */
+	ulint		lsn_offset;	/* the offset of the above lsn */
+	ulint		n_pending_writes;/* number of currently pending flush
+					writes for this log group */
+	byte**		file_header_bufs;/* buffers for each file header in the
+					group */
+	/*-----------------------------*/
+	byte**		archive_file_header_bufs;/* buffers for each file
+					header in the group */
+	ulint		archive_space_id;/* file space which implements the log
+					group archive */
+	ulint		archived_file_no;/* file number corresponding to
+					log_sys->archived_lsn */
+	ulint		archived_offset;/* file offset corresponding to
+					log_sys->archived_lsn, 0 if we have
+					not yet written to the archive file
+					number archived_file_no */
+	ulint		next_archived_file_no;/* during an archive write,
+					until the write is completed, we
+					store the next value for
+					archived_file_no here: the write
+					completion function then sets the new
+					value to ..._file_no */
+	ulint		next_archived_offset; /* like the preceding field */
+	/*-----------------------------*/
+	ib_uint64_t	scanned_lsn;	/* used only in recovery: recovery scan
+					succeeded up to this lsn in this log
+					group */
+	byte*		checkpoint_buf;	/* checkpoint header is written from
+					this buffer to the group */
+	UT_LIST_NODE_T(log_group_t)
+			log_groups;	/* list of log groups */
+};
+
+struct log_struct{
+	byte		pad[64];	/* padding to prevent other memory
+					update hotspots from residing on the
+					same memory cache line */
+	ib_uint64_t	lsn;		/* log sequence number */
+	ulint		buf_free;	/* first free offset within the log
+					buffer */
+	mutex_t		mutex;		/* mutex protecting the log */
+	byte*		buf;		/* log buffer */
+	ulint		buf_size;	/* log buffer size in bytes */
+	ulint		max_buf_free;	/* recommended maximum value of
+					buf_free, after which the buffer is
+					flushed */
+	ulint		old_buf_free;	/* value of buf free when log was
+					last time opened; only in the debug
+					version */
+	ib_uint64_t	old_lsn;	/* value of lsn when log was last time
+					opened; only in the debug version */
+	ibool		check_flush_or_checkpoint;
+					/* this is set to TRUE when there may
+					be need to flush the log buffer, or
+					preflush buffer pool pages, or make
+					a checkpoint; this MUST be TRUE when
+					lsn - last_checkpoint_lsn >
+					max_checkpoint_age; this flag is
+					peeked at by log_free_check(), which
+					does not reserve the log mutex */
+	UT_LIST_BASE_NODE_T(log_group_t)
+			log_groups;	/* log groups */
+
+	/* The fields involved in the log buffer flush */
+
+	ulint		buf_next_to_write;/* first offset in the log buffer
+					where the byte content may not exist
+					written to file, e.g., the start
+					offset of a log record catenated
+					later; this is advanced when a flush
+					operation is completed to all the log
+					groups */
+	ib_uint64_t	written_to_some_lsn;
+					/* first log sequence number not yet
+					written to any log group; for this to
+					be advanced, it is enough that the
+					write i/o has been completed for any
+					one log group */
+	ib_uint64_t	written_to_all_lsn;
+					/* first log sequence number not yet
+					written to some log group; for this to
+					be advanced, it is enough that the
+					write i/o has been completed for all
+					log groups */
+	ib_uint64_t	write_lsn;	/* end lsn for the current running
+					write */
+	ulint		write_end_offset;/* the data in buffer has been written
+					up to this offset when the current
+					write ends: this field will then
+					be copied to buf_next_to_write */
+	ib_uint64_t	current_flush_lsn;/* end lsn for the current running
+					write + flush operation */
+	ib_uint64_t	flushed_to_disk_lsn;
+					/* how far we have written the log
+					AND flushed to disk */
+	ulint		n_pending_writes;/* number of currently pending flushes
+					or writes */
+	/* NOTE on the 'flush' in names of the fields below: starting from
+	4.0.14, we separate the write of the log file and the actual fsync()
+	or other method to flush it to disk. The names below shhould really
+	be 'flush_or_write'! */
+	os_event_t	no_flush_event;	/* this event is in the reset state
+					when a flush or a write is running;
+					a thread should wait for this without
+					owning the log mutex, but NOTE that
+					to set or reset this event, the
+					thread MUST own the log mutex! */
+	ibool		one_flushed;	/* during a flush, this is first FALSE
+					and becomes TRUE when one log group
+					has been written or flushed */
+	os_event_t	one_flushed_event;/* this event is reset when the
+					flush or write has not yet completed
+					for any log group; e.g., this means
+					that a transaction has been committed
+					when this is set; a thread should wait
+					for this without owning the log mutex,
+					but NOTE that to set or reset this
+					event, the thread MUST own the log
+					mutex! */
+	ulint		n_log_ios;	/* number of log i/os initiated thus
+					far */
+	ulint		n_log_ios_old;	/* number of log i/o's at the
+					previous printout */
+	time_t		last_printout_time;/* when log_print was last time
+					called */
+
+	/* Fields involved in checkpoints */
+	ulint		log_group_capacity; /* capacity of the log group; if
+					the checkpoint age exceeds this, it is
+					a serious error because it is possible
+					we will then overwrite log and spoil
+					crash recovery */
+	ulint		max_modified_age_async;
+					/* when this recommended value for lsn
+					- buf_pool_get_oldest_modification()
+					is exceeded, we start an asynchronous
+					preflush of pool pages */
+	ulint		max_modified_age_sync;
+					/* when this recommended value for lsn
+					- buf_pool_get_oldest_modification()
+					is exceeded, we start a synchronous
+					preflush of pool pages */
+	ulint		adm_checkpoint_interval;
+					/* administrator-specified checkpoint
+					interval in terms of log growth in
+					bytes; the interval actually used by
+					the database can be smaller */
+	ulint		max_checkpoint_age_async;
+					/* when this checkpoint age is exceeded
+					we start an asynchronous writing of a
+					new checkpoint */
+	ulint		max_checkpoint_age;
+					/* this is the maximum allowed value
+					for lsn - last_checkpoint_lsn when a
+					new query step is started */
+	ib_uint64_t	next_checkpoint_no;
+					/* next checkpoint number */
+	ib_uint64_t	last_checkpoint_lsn;
+					/* latest checkpoint lsn */
+	ib_uint64_t	next_checkpoint_lsn;
+					/* next checkpoint lsn */
+	ulint		n_pending_checkpoint_writes;
+					/* number of currently pending
+					checkpoint writes */
+	rw_lock_t	checkpoint_lock;/* this latch is x-locked when a
+					checkpoint write is running; a thread
+					should wait for this without owning
+					the log mutex */
+	byte*		checkpoint_buf;	/* checkpoint header is read to this
+					buffer */
+#ifdef UNIV_LOG_ARCHIVE
+	/* Fields involved in archiving */
+	ulint		archiving_state;/* LOG_ARCH_ON, LOG_ARCH_STOPPING
+					LOG_ARCH_STOPPED, LOG_ARCH_OFF */
+	ib_uint64_t	archived_lsn;	/* archiving has advanced to this
+					lsn */
+	ulint		max_archived_lsn_age_async;
+					/* recommended maximum age of
+					archived_lsn, before we start
+					asynchronous copying to the archive */
+	ulint		max_archived_lsn_age;
+					/* maximum allowed age for
+					archived_lsn */
+	ib_uint64_t	next_archived_lsn;/* during an archive write,
+					until the write is completed, we
+					store the next value for
+					archived_lsn here: the write
+					completion function then sets the new
+					value to archived_lsn */
+	ulint		archiving_phase;/* LOG_ARCHIVE_READ or
+					LOG_ARCHIVE_WRITE */
+	ulint		n_pending_archive_ios;
+					/* number of currently pending reads
+					or writes in archiving */
+	rw_lock_t	archive_lock;	/* this latch is x-locked when an
+					archive write is running; a thread
+					should wait for this without owning
+					the log mutex */
+	ulint		archive_buf_size;/* size of archive_buf */
+	byte*		archive_buf;	/* log segment is written to the
+					archive from this buffer */
+	os_event_t	archiving_on;	/* if archiving has been stopped,
+					a thread can wait for this event to
+					become signaled */
+#endif /* UNIV_LOG_ARCHIVE */
+};
+
+#define LOG_ARCH_ON		71
+#define LOG_ARCH_STOPPING	72
+#define LOG_ARCH_STOPPING2	73
+#define LOG_ARCH_STOPPED	74
+#define LOG_ARCH_OFF		75
+
+#ifndef UNIV_NONINL
+#include "log0log.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/log0log.ic b/storage/xtradb/include/log0log.ic
new file mode 100644
index 00000000000..85eebda4942
--- /dev/null
+++ b/storage/xtradb/include/log0log.ic
@@ -0,0 +1,407 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Database log
+
+Created 12/9/1995 Heikki Tuuri
+*******************************************************/
+
+#include "os0file.h"
+#include "mach0data.h"
+#include "mtr0mtr.h"
+
+/**********************************************************
+Checks by parsing that the catenated log segment for a single mtr is
+consistent. */
+UNIV_INTERN
+ibool
+log_check_log_recs(
+/*===============*/
+	byte*		buf,		/* in: pointer to the start of
+					the log segment in the
+					log_sys->buf log buffer */
+	ulint		len,		/* in: segment length in bytes */
+	ib_uint64_t	buf_start_lsn);	/* in: buffer start lsn */
+
+/****************************************************************
+Gets a log block flush bit. */
+UNIV_INLINE
+ibool
+log_block_get_flush_bit(
+/*====================*/
+				/* out: TRUE if this block was the first
+				to be written in a log flush */
+	byte*	log_block)	/* in: log block */
+{
+	if (LOG_BLOCK_FLUSH_BIT_MASK
+	    & mach_read_from_4(log_block + LOG_BLOCK_HDR_NO)) {
+
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
+
+/****************************************************************
+Sets the log block flush bit. */
+UNIV_INLINE
+void
+log_block_set_flush_bit(
+/*====================*/
+	byte*	log_block,	/* in: log block */
+	ibool	val)		/* in: value to set */
+{
+	ulint	field;
+
+	field = mach_read_from_4(log_block + LOG_BLOCK_HDR_NO);
+
+	if (val) {
+		field = field | LOG_BLOCK_FLUSH_BIT_MASK;
+	} else {
+		field = field & ~LOG_BLOCK_FLUSH_BIT_MASK;
+	}
+
+	mach_write_to_4(log_block + LOG_BLOCK_HDR_NO, field);
+}
+
+/****************************************************************
+Gets a log block number stored in the header. */
+UNIV_INLINE
+ulint
+log_block_get_hdr_no(
+/*=================*/
+				/* out: log block number stored in the block
+				header */
+	byte*	log_block)	/* in: log block */
+{
+	return(~LOG_BLOCK_FLUSH_BIT_MASK
+	       & mach_read_from_4(log_block + LOG_BLOCK_HDR_NO));
+}
+
+/****************************************************************
+Sets the log block number stored in the header; NOTE that this must be set
+before the flush bit! */
+UNIV_INLINE
+void
+log_block_set_hdr_no(
+/*=================*/
+	byte*	log_block,	/* in: log block */
+	ulint	n)		/* in: log block number: must be > 0 and
+				< LOG_BLOCK_FLUSH_BIT_MASK */
+{
+	ut_ad(n > 0);
+	ut_ad(n < LOG_BLOCK_FLUSH_BIT_MASK);
+
+	mach_write_to_4(log_block + LOG_BLOCK_HDR_NO, n);
+}
+
+/****************************************************************
+Gets a log block data length. */
+UNIV_INLINE
+ulint
+log_block_get_data_len(
+/*===================*/
+				/* out: log block data length measured as a
+				byte offset from the block start */
+	byte*	log_block)	/* in: log block */
+{
+	return(mach_read_from_2(log_block + LOG_BLOCK_HDR_DATA_LEN));
+}
+
+/****************************************************************
+Sets the log block data length. */
+UNIV_INLINE
+void
+log_block_set_data_len(
+/*===================*/
+	byte*	log_block,	/* in: log block */
+	ulint	len)		/* in: data length */
+{
+	mach_write_to_2(log_block + LOG_BLOCK_HDR_DATA_LEN, len);
+}
+
+/****************************************************************
+Gets a log block first mtr log record group offset. */
+UNIV_INLINE
+ulint
+log_block_get_first_rec_group(
+/*==========================*/
+				/* out: first mtr log record group byte offset
+				from the block start, 0 if none */
+	byte*	log_block)	/* in: log block */
+{
+	return(mach_read_from_2(log_block + LOG_BLOCK_FIRST_REC_GROUP));
+}
+
+/****************************************************************
+Sets the log block first mtr log record group offset. */
+UNIV_INLINE
+void
+log_block_set_first_rec_group(
+/*==========================*/
+	byte*	log_block,	/* in: log block */
+	ulint	offset)		/* in: offset, 0 if none */
+{
+	mach_write_to_2(log_block + LOG_BLOCK_FIRST_REC_GROUP, offset);
+}
+
+/****************************************************************
+Gets a log block checkpoint number field (4 lowest bytes). */
+UNIV_INLINE
+ulint
+log_block_get_checkpoint_no(
+/*========================*/
+				/* out: checkpoint no (4 lowest bytes) */
+	byte*	log_block)	/* in: log block */
+{
+	return(mach_read_from_4(log_block + LOG_BLOCK_CHECKPOINT_NO));
+}
+
+/****************************************************************
+Sets a log block checkpoint number field (4 lowest bytes). */
+UNIV_INLINE
+void
+log_block_set_checkpoint_no(
+/*========================*/
+	byte*		log_block,	/* in: log block */
+	ib_uint64_t	no)		/* in: checkpoint no */
+{
+	mach_write_to_4(log_block + LOG_BLOCK_CHECKPOINT_NO, (ulint) no);
+}
+
+/****************************************************************
+Converts a lsn to a log block number. */
+UNIV_INLINE
+ulint
+log_block_convert_lsn_to_no(
+/*========================*/
+				/* out: log block number,
+				it is > 0 and <= 1G */
+	ib_uint64_t	lsn)	/* in: lsn of a byte within the block */
+{
+	return(((ulint) (lsn / OS_FILE_LOG_BLOCK_SIZE) & 0x3FFFFFFFUL) + 1);
+}
+
+/****************************************************************
+Calculates the checksum for a log block. */
+UNIV_INLINE
+ulint
+log_block_calc_checksum(
+/*====================*/
+				/* out: checksum */
+	const byte*	block)	/* in: log block */
+{
+	ulint	sum;
+	ulint	sh;
+	ulint	i;
+
+	sum = 1;
+	sh = 0;
+
+	for (i = 0; i < OS_FILE_LOG_BLOCK_SIZE - LOG_BLOCK_TRL_SIZE; i++) {
+		ulint	b = (ulint) block[i];
+		sum &= 0x7FFFFFFFUL;
+		sum += b;
+		sum += b << sh;
+		sh++;
+		if (sh > 24) {
+			sh = 0;
+		}
+	}
+
+	return(sum);
+}
+
+/****************************************************************
+Gets a log block checksum field value. */
+UNIV_INLINE
+ulint
+log_block_get_checksum(
+/*===================*/
+					/* out: checksum */
+	const byte*	log_block)	/* in: log block */
+{
+	return(mach_read_from_4(log_block + OS_FILE_LOG_BLOCK_SIZE
+				- LOG_BLOCK_CHECKSUM));
+}
+
+/****************************************************************
+Sets a log block checksum field value. */
+UNIV_INLINE
+void
+log_block_set_checksum(
+/*===================*/
+	byte*	log_block,	/* in: log block */
+	ulint	checksum)	/* in: checksum */
+{
+	mach_write_to_4(log_block + OS_FILE_LOG_BLOCK_SIZE
+			- LOG_BLOCK_CHECKSUM,
+			checksum);
+}
+
+/****************************************************************
+Initializes a log block in the log buffer. */
+UNIV_INLINE
+void
+log_block_init(
+/*===========*/
+	byte*		log_block,	/* in: pointer to the log buffer */
+	ib_uint64_t	lsn)		/* in: lsn within the log block */
+{
+	ulint	no;
+
+	ut_ad(mutex_own(&(log_sys->mutex)));
+
+	no = log_block_convert_lsn_to_no(lsn);
+
+	log_block_set_hdr_no(log_block, no);
+
+	log_block_set_data_len(log_block, LOG_BLOCK_HDR_SIZE);
+	log_block_set_first_rec_group(log_block, 0);
+}
+
+/****************************************************************
+Initializes a log block in the log buffer in the old format, where there
+was no checksum yet. */
+UNIV_INLINE
+void
+log_block_init_in_old_format(
+/*=========================*/
+	byte*		log_block,	/* in: pointer to the log buffer */
+	ib_uint64_t	lsn)		/* in: lsn within the log block */
+{
+	ulint	no;
+
+	ut_ad(mutex_own(&(log_sys->mutex)));
+
+	no = log_block_convert_lsn_to_no(lsn);
+
+	log_block_set_hdr_no(log_block, no);
+	mach_write_to_4(log_block + OS_FILE_LOG_BLOCK_SIZE
+			- LOG_BLOCK_CHECKSUM, no);
+	log_block_set_data_len(log_block, LOG_BLOCK_HDR_SIZE);
+	log_block_set_first_rec_group(log_block, 0);
+}
+
+/****************************************************************
+Writes to the log the string given. The log must be released with
+log_release. */
+UNIV_INLINE
+ib_uint64_t
+log_reserve_and_write_fast(
+/*=======================*/
+				/* out: end lsn of the log record,
+				zero if did not succeed */
+	byte*		str,	/* in: string */
+	ulint		len,	/* in: string length */
+	ib_uint64_t*	start_lsn,/* out: start lsn of the log record */
+	ibool*		success)/* out: TRUE if success */
+{
+	log_t*		log	= log_sys;
+	ulint		data_len;
+	ib_uint64_t	lsn;
+
+	*success = TRUE;
+
+	mutex_enter(&(log->mutex));
+
+	data_len = len + log->buf_free % OS_FILE_LOG_BLOCK_SIZE;
+
+	if (data_len >= OS_FILE_LOG_BLOCK_SIZE - LOG_BLOCK_TRL_SIZE) {
+
+		/* The string does not fit within the current log block
+		or the log block would become full */
+
+		*success = FALSE;
+
+		mutex_exit(&(log->mutex));
+
+		return(0);
+	}
+
+	*start_lsn = log->lsn;
+
+	ut_memcpy(log->buf + log->buf_free, str, len);
+
+	log_block_set_data_len((byte*) ut_align_down(log->buf + log->buf_free,
+						     OS_FILE_LOG_BLOCK_SIZE),
+			       data_len);
+#ifdef UNIV_LOG_DEBUG
+	log->old_buf_free = log->buf_free;
+	log->old_lsn = log->lsn;
+#endif
+	log->buf_free += len;
+
+	ut_ad(log->buf_free <= log->buf_size);
+
+	lsn = log->lsn += len;
+
+#ifdef UNIV_LOG_DEBUG
+	log_check_log_recs(log->buf + log->old_buf_free,
+			   log->buf_free - log->old_buf_free, log->old_lsn);
+#endif
+	return(lsn);
+}
+
+/***************************************************************************
+Releases the log mutex. */
+UNIV_INLINE
+void
+log_release(void)
+/*=============*/
+{
+	mutex_exit(&(log_sys->mutex));
+}
+
+/****************************************************************
+Gets the current lsn. */
+UNIV_INLINE
+ib_uint64_t
+log_get_lsn(void)
+/*=============*/
+			/* out: current lsn */
+{
+	ib_uint64_t	lsn;
+
+	mutex_enter(&(log_sys->mutex));
+
+	lsn = log_sys->lsn;
+
+	mutex_exit(&(log_sys->mutex));
+
+	return(lsn);
+}
+
+/***************************************************************************
+Checks if there is need for a log buffer flush or a new checkpoint, and does
+this if yes. Any database operation should call this when it has modified
+more than about 4 pages. NOTE that this function may only be called when the
+OS thread owns no synchronization objects except the dictionary mutex. */
+UNIV_INLINE
+void
+log_free_check(void)
+/*================*/
+{
+	/* ut_ad(sync_thread_levels_empty()); */
+
+	if (log_sys->check_flush_or_checkpoint) {
+
+		log_check_margins();
+	}
+}
diff --git a/storage/xtradb/include/log0recv.h b/storage/xtradb/include/log0recv.h
new file mode 100644
index 00000000000..e3fe9ed330a
--- /dev/null
+++ b/storage/xtradb/include/log0recv.h
@@ -0,0 +1,392 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Recovery
+
+Created 9/20/1997 Heikki Tuuri
+*******************************************************/
+
+#ifndef log0recv_h
+#define log0recv_h
+
+#include "univ.i"
+#include "ut0byte.h"
+#include "buf0types.h"
+#include "hash0hash.h"
+#include "log0log.h"
+
+#ifdef UNIV_HOTBACKUP
+extern ibool	recv_replay_file_ops;
+
+/***********************************************************************
+Reads the checkpoint info needed in hot backup. */
+UNIV_INTERN
+ibool
+recv_read_cp_info_for_backup(
+/*=========================*/
+				/* out: TRUE if success */
+	byte*		hdr,	/* in: buffer containing the log group
+				header */
+	ib_uint64_t*	lsn,	/* out: checkpoint lsn */
+	ulint*		offset,	/* out: checkpoint offset in the log group */
+	ulint*		fsp_limit,/* out: fsp limit of space 0,
+				1000000000 if the database is running
+				with < version 3.23.50 of InnoDB */
+	ib_uint64_t*	cp_no,	/* out: checkpoint number */
+	ib_uint64_t*	first_header_lsn);
+				/* out: lsn of of the start of the
+				first log file */
+/***********************************************************************
+Scans the log segment and n_bytes_scanned is set to the length of valid
+log scanned. */
+UNIV_INTERN
+void
+recv_scan_log_seg_for_backup(
+/*=========================*/
+	byte*		buf,		/* in: buffer containing log data */
+	ulint		buf_len,	/* in: data length in that buffer */
+	ib_uint64_t*	scanned_lsn,	/* in/out: lsn of buffer start,
+					we return scanned lsn */
+	ulint*		scanned_checkpoint_no,
+					/* in/out: 4 lowest bytes of the
+					highest scanned checkpoint number so
+					far */
+	ulint*		n_bytes_scanned);/* out: how much we were able to
+					scan, smaller than buf_len if log
+					data ended here */
+#endif /* UNIV_HOTBACKUP */
+/***********************************************************************
+Returns TRUE if recovery is currently running. */
+UNIV_INLINE
+ibool
+recv_recovery_is_on(void);
+/*=====================*/
+/***********************************************************************
+Returns TRUE if recovery from backup is currently running. */
+UNIV_INLINE
+ibool
+recv_recovery_from_backup_is_on(void);
+/*=================================*/
+/****************************************************************************
+Applies the hashed log records to the page, if the page lsn is less than the
+lsn of a log record. This can be called when a buffer page has just been
+read in, or also for a page already in the buffer pool. */
+UNIV_INTERN
+void
+recv_recover_page(
+/*==============*/
+	ibool		recover_backup,
+				/* in: TRUE if we are recovering a backup
+				page: then we do not acquire any latches
+				since the page was read in outside the
+				buffer pool */
+	ibool		just_read_in,
+				/* in: TRUE if the i/o-handler calls this for
+				a freshly read page */
+	buf_block_t*	block);	/* in: buffer block */
+/************************************************************
+Recovers from a checkpoint. When this function returns, the database is able
+to start processing of new user transactions, but the function
+recv_recovery_from_checkpoint_finish should be called later to complete
+the recovery and free the resources used in it. */
+UNIV_INTERN
+ulint
+recv_recovery_from_checkpoint_start_func(
+/*=====================================*/
+					/* out: error code or DB_SUCCESS */
+#ifdef UNIV_LOG_ARCHIVE
+	ulint		type,		/* in: LOG_CHECKPOINT or LOG_ARCHIVE */
+	ib_uint64_t	limit_lsn,	/* in: recover up to this lsn
+					if possible */
+#endif /* UNIV_LOG_ARCHIVE */
+	ib_uint64_t	min_flushed_lsn,/* in: min flushed lsn from
+					data files */
+	ib_uint64_t	max_flushed_lsn);/* in: max flushed lsn from
+					 data files */
+#ifdef UNIV_LOG_ARCHIVE
+# define recv_recovery_from_checkpoint_start(type,lim,min,max)		\
+	recv_recovery_from_checkpoint_start_func(type,lim,min,max)
+#else /* UNIV_LOG_ARCHIVE */
+# define recv_recovery_from_checkpoint_start(type,lim,min,max)		\
+	recv_recovery_from_checkpoint_start_func(min,max)
+#endif /* UNIV_LOG_ARCHIVE */
+/************************************************************
+Completes recovery from a checkpoint. */
+UNIV_INTERN
+void
+recv_recovery_from_checkpoint_finish(void);
+/*======================================*/
+/***********************************************************
+Scans log from a buffer and stores new log data to the parsing buffer. Parses
+and hashes the log records if new data found. */
+UNIV_INTERN
+ibool
+recv_scan_log_recs(
+/*===============*/
+					/* out: TRUE if limit_lsn has been
+					reached, or not able to scan any more
+					in this log group */
+	ibool		apply_automatically,/* in: TRUE if we want this
+					function to apply log records
+					automatically when the hash table
+					becomes full; in the hot backup tool
+					the tool does the applying, not this
+					function */
+	ulint		available_memory,/* in: we let the hash table of recs
+					to grow to this size, at the maximum */
+	ibool		store_to_hash,	/* in: TRUE if the records should be
+					stored to the hash table; this is set
+					to FALSE if just debug checking is
+					needed */
+	byte*		buf,		/* in: buffer containing a log segment
+					or garbage */
+	ulint		len,		/* in: buffer length */
+	ib_uint64_t	start_lsn,	/* in: buffer start lsn */
+	ib_uint64_t*	contiguous_lsn,	/* in/out: it is known that all log
+					groups contain contiguous log data up
+					to this lsn */
+	ib_uint64_t*	group_scanned_lsn);/* out: scanning succeeded up to
+					this lsn */
+/**********************************************************
+Resets the logs. The contents of log files will be lost! */
+UNIV_INTERN
+void
+recv_reset_logs(
+/*============*/
+	ib_uint64_t	lsn,		/* in: reset to this lsn
+					rounded up to be divisible by
+					OS_FILE_LOG_BLOCK_SIZE, after
+					which we add
+					LOG_BLOCK_HDR_SIZE */
+#ifdef UNIV_LOG_ARCHIVE
+	ulint		arch_log_no,	/* in: next archived log file number */
+#endif /* UNIV_LOG_ARCHIVE */
+	ibool		new_logs_created);/* in: TRUE if resetting logs
+					is done at the log creation;
+					FALSE if it is done after
+					archive recovery */
+#ifdef UNIV_HOTBACKUP
+/**********************************************************
+Creates new log files after a backup has been restored. */
+UNIV_INTERN
+void
+recv_reset_log_files_for_backup(
+/*============================*/
+	const char*	log_dir,	/* in: log file directory path */
+	ulint		n_log_files,	/* in: number of log files */
+	ulint		log_file_size,	/* in: log file size */
+	ib_uint64_t	lsn);		/* in: new start lsn, must be
+					divisible by OS_FILE_LOG_BLOCK_SIZE */
+#endif /* UNIV_HOTBACKUP */
+/************************************************************
+Creates the recovery system. */
+UNIV_INTERN
+void
+recv_sys_create(void);
+/*=================*/
+/************************************************************
+Inits the recovery system for a recovery operation. */
+UNIV_INTERN
+void
+recv_sys_init(
+/*==========*/
+	ibool	recover_from_backup,	/* in: TRUE if this is called
+					to recover from a hot backup */
+	ulint	available_memory);	/* in: available memory in bytes */
+/***********************************************************************
+Empties the hash table of stored log records, applying them to appropriate
+pages. */
+UNIV_INTERN
+void
+recv_apply_hashed_log_recs(
+/*=======================*/
+	ibool	allow_ibuf);	/* in: if TRUE, also ibuf operations are
+				allowed during the application; if FALSE,
+				no ibuf operations are allowed, and after
+				the application all file pages are flushed to
+				disk and invalidated in buffer pool: this
+				alternative means that no new log records
+				can be generated during the application */
+#ifdef UNIV_HOTBACKUP
+/***********************************************************************
+Applies log records in the hash table to a backup. */
+UNIV_INTERN
+void
+recv_apply_log_recs_for_backup(void);
+/*================================*/
+#endif
+#ifdef UNIV_LOG_ARCHIVE
+/************************************************************
+Recovers from archived log files, and also from log files, if they exist. */
+UNIV_INTERN
+ulint
+recv_recovery_from_archive_start(
+/*=============================*/
+					/* out: error code or DB_SUCCESS */
+	ib_uint64_t	min_flushed_lsn,/* in: min flushed lsn field from the
+					data files */
+	ib_uint64_t	limit_lsn,	/* in: recover up to this lsn if
+					possible */
+	ulint		first_log_no);	/* in: number of the first archived
+					log file to use in the recovery; the
+					file will be searched from
+					INNOBASE_LOG_ARCH_DIR specified in
+					server config file */
+/************************************************************
+Completes recovery from archive. */
+UNIV_INTERN
+void
+recv_recovery_from_archive_finish(void);
+/*===================================*/
+#endif /* UNIV_LOG_ARCHIVE */
+
+/* Block of log record data */
+typedef struct recv_data_struct	recv_data_t;
+struct recv_data_struct{
+	recv_data_t*	next;	/* pointer to the next block or NULL */
+				/* the log record data is stored physically
+				immediately after this struct, max amount
+				RECV_DATA_BLOCK_SIZE bytes of it */
+};
+
+/* Stored log record struct */
+typedef struct recv_struct	recv_t;
+struct recv_struct{
+	byte		type;	/* log record type */
+	ulint		len;	/* log record body length in bytes */
+	recv_data_t*	data;	/* chain of blocks containing the log record
+				body */
+	ib_uint64_t	start_lsn;/* start lsn of the log segment written by
+				the mtr which generated this log record: NOTE
+				that this is not necessarily the start lsn of
+				this log record */
+	ib_uint64_t	end_lsn;/* end lsn of the log segment written by
+				the mtr which generated this log record: NOTE
+				that this is not necessarily the end lsn of
+				this log record */
+	UT_LIST_NODE_T(recv_t)
+			rec_list;/* list of log records for this page */
+};
+
+/* Hashed page file address struct */
+typedef struct recv_addr_struct	recv_addr_t;
+struct recv_addr_struct{
+	ulint		state;	/* RECV_NOT_PROCESSED, RECV_BEING_PROCESSED,
+				or RECV_PROCESSED */
+	ulint		space;	/* space id */
+	ulint		page_no;/* page number */
+	UT_LIST_BASE_NODE_T(recv_t)
+			rec_list;/* list of log records for this page */
+	hash_node_t	addr_hash;
+};
+
+/* Recovery system data structure */
+typedef struct recv_sys_struct	recv_sys_t;
+struct recv_sys_struct{
+	mutex_t		mutex;	/* mutex protecting the fields apply_log_recs,
+				n_addrs, and the state field in each recv_addr
+				struct */
+	ibool		apply_log_recs;
+				/* this is TRUE when log rec application to
+				pages is allowed; this flag tells the
+				i/o-handler if it should do log record
+				application */
+	ibool		apply_batch_on;
+				/* this is TRUE when a log rec application
+				batch is running */
+	ib_uint64_t	lsn;	/* log sequence number */
+	ulint		last_log_buf_size;
+				/* size of the log buffer when the database
+				last time wrote to the log */
+	byte*		last_block;
+				/* possible incomplete last recovered log
+				block */
+	byte*		last_block_buf_start;
+				/* the nonaligned start address of the
+				preceding buffer */
+	byte*		buf;	/* buffer for parsing log records */
+	ulint		len;	/* amount of data in buf */
+	ib_uint64_t	parse_start_lsn;
+				/* this is the lsn from which we were able to
+				start parsing log records and adding them to
+				the hash table; zero if a suitable
+				start point not found yet */
+	ib_uint64_t	scanned_lsn;
+				/* the log data has been scanned up to this
+				lsn */
+	ulint		scanned_checkpoint_no;
+				/* the log data has been scanned up to this
+				checkpoint number (lowest 4 bytes) */
+	ulint		recovered_offset;
+				/* start offset of non-parsed log records in
+				buf */
+	ib_uint64_t	recovered_lsn;
+				/* the log records have been parsed up to
+				this lsn */
+	ib_uint64_t	limit_lsn;/* recovery should be made at most up to this
+				lsn */
+	ibool		found_corrupt_log;
+				/* this is set to TRUE if we during log
+				scan find a corrupt log block, or a corrupt
+				log record, or there is a log parsing
+				buffer overflow */
+#ifdef UNIV_LOG_ARCHIVE
+	log_group_t*	archive_group;
+				/* in archive recovery: the log group whose
+				archive is read */
+#endif /* !UNIV_LOG_ARCHIVE */
+	mem_heap_t*	heap;	/* memory heap of log records and file
+				addresses*/
+	hash_table_t*	addr_hash;/* hash table of file addresses of pages */
+	ulint		n_addrs;/* number of not processed hashed file
+				addresses in the hash table */
+};
+
+extern recv_sys_t*	recv_sys;
+extern ibool		recv_recovery_on;
+extern ibool		recv_no_ibuf_operations;
+extern ibool		recv_needed_recovery;
+
+extern ibool		recv_lsn_checks_on;
+#ifdef UNIV_HOTBACKUP
+extern ibool		recv_is_making_a_backup;
+#endif /* UNIV_HOTBACKUP */
+extern ulint		recv_max_parsed_page_no;
+
+/* Size of the parsing buffer; it must accommodate RECV_SCAN_SIZE many
+times! */
+#define RECV_PARSING_BUF_SIZE	(2 * 1024 * 1024)
+
+/* Size of block reads when the log groups are scanned forward to do a
+roll-forward */
+#define RECV_SCAN_SIZE		(4 * UNIV_PAGE_SIZE)
+
+/* States of recv_addr_struct */
+#define RECV_NOT_PROCESSED	71
+#define RECV_BEING_READ		72
+#define RECV_BEING_PROCESSED	73
+#define RECV_PROCESSED		74
+
+extern ulint	recv_n_pool_free_frames;
+
+#ifndef UNIV_NONINL
+#include "log0recv.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/log0recv.ic b/storage/xtradb/include/log0recv.ic
new file mode 100644
index 00000000000..e114bede38f
--- /dev/null
+++ b/storage/xtradb/include/log0recv.ic
@@ -0,0 +1,48 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Recovery
+
+Created 9/20/1997 Heikki Tuuri
+*******************************************************/
+
+#include "univ.i"
+
+extern ibool	recv_recovery_from_backup_on;
+
+/***********************************************************************
+Returns TRUE if recovery is currently running. */
+UNIV_INLINE
+ibool
+recv_recovery_is_on(void)
+/*=====================*/
+{
+	return(UNIV_UNLIKELY(recv_recovery_on));
+}
+
+/***********************************************************************
+Returns TRUE if recovery from backup is currently running. */
+UNIV_INLINE
+ibool
+recv_recovery_from_backup_is_on(void)
+/*=================================*/
+{
+	return(recv_recovery_from_backup_on);
+}
+
diff --git a/storage/xtradb/include/mach0data.h b/storage/xtradb/include/mach0data.h
new file mode 100644
index 00000000000..78b48af0836
--- /dev/null
+++ b/storage/xtradb/include/mach0data.h
@@ -0,0 +1,398 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/**********************************************************************
+Utilities for converting data from the database file
+to the machine format.
+
+Created 11/28/1995 Heikki Tuuri
+***********************************************************************/
+
+#ifndef mach0data_h
+#define mach0data_h
+
+#include "univ.i"
+#include "ut0byte.h"
+
+/* The data and all fields are always stored in a database file
+in the same format: ascii, big-endian, ... .
+All data in the files MUST be accessed using the functions in this
+module. */
+
+/***********************************************************
+The following function is used to store data in one byte. */
+UNIV_INLINE
+void
+mach_write_to_1(
+/*============*/
+	byte*	b,	/* in: pointer to byte where to store */
+	ulint	n);	 /* in: ulint integer to be stored, >= 0, < 256 */
+/************************************************************
+The following function is used to fetch data from one byte. */
+UNIV_INLINE
+ulint
+mach_read_from_1(
+/*=============*/
+				/* out: ulint integer, >= 0, < 256 */
+	const byte*	b)	/* in: pointer to byte */
+	__attribute__((nonnull, pure));
+/***********************************************************
+The following function is used to store data in two consecutive
+bytes. We store the most significant byte to the lower address. */
+UNIV_INLINE
+void
+mach_write_to_2(
+/*============*/
+	byte*	b,	/* in: pointer to two bytes where to store */
+	ulint	n);	 /* in: ulint integer to be stored, >= 0, < 64k */
+/************************************************************
+The following function is used to fetch data from two consecutive
+bytes. The most significant byte is at the lowest address. */
+UNIV_INLINE
+ulint
+mach_read_from_2(
+/*=============*/
+				/* out: ulint integer, >= 0, < 64k */
+	const byte*	b)	/* in: pointer to two bytes */
+	__attribute__((nonnull, pure));
+
+/************************************************************
+The following function is used to convert a 16-bit data item
+to the canonical format, for fast bytewise equality test
+against memory. */
+UNIV_INLINE
+uint16
+mach_encode_2(
+/*==========*/
+			/* out: 16-bit integer in canonical format */
+	ulint	n)	/* in: integer in machine-dependent format */
+	__attribute__((const));
+/************************************************************
+The following function is used to convert a 16-bit data item
+from the canonical format, for fast bytewise equality test
+against memory. */
+UNIV_INLINE
+ulint
+mach_decode_2(
+/*==========*/
+			/* out: integer in machine-dependent format */
+	uint16	n)	/* in: 16-bit integer in canonical format */
+	__attribute__((const));
+/***********************************************************
+The following function is used to store data in 3 consecutive
+bytes. We store the most significant byte to the lowest address. */
+UNIV_INLINE
+void
+mach_write_to_3(
+/*============*/
+	byte*	b,	/* in: pointer to 3 bytes where to store */
+	ulint	n);	 /* in: ulint integer to be stored */
+/************************************************************
+The following function is used to fetch data from 3 consecutive
+bytes. The most significant byte is at the lowest address. */
+UNIV_INLINE
+ulint
+mach_read_from_3(
+/*=============*/
+				/* out: ulint integer */
+	const byte*	b)	/* in: pointer to 3 bytes */
+	__attribute__((nonnull, pure));
+/***********************************************************
+The following function is used to store data in four consecutive
+bytes. We store the most significant byte to the lowest address. */
+UNIV_INLINE
+void
+mach_write_to_4(
+/*============*/
+	byte*	b,	/* in: pointer to four bytes where to store */
+	ulint	n);	 /* in: ulint integer to be stored */
+/************************************************************
+The following function is used to fetch data from 4 consecutive
+bytes. The most significant byte is at the lowest address. */
+UNIV_INLINE
+ulint
+mach_read_from_4(
+/*=============*/
+				/* out: ulint integer */
+	const byte*	b)	/* in: pointer to four bytes */
+	__attribute__((nonnull, pure));
+/*************************************************************
+Writes a ulint in a compressed form (1..5 bytes). */
+UNIV_INLINE
+ulint
+mach_write_compressed(
+/*==================*/
+			/* out: stored size in bytes */
+	byte*	b,	/* in: pointer to memory where to store */
+	ulint	n);	/* in: ulint integer to be stored */
+/*************************************************************
+Returns the size of an ulint when written in the compressed form. */
+UNIV_INLINE
+ulint
+mach_get_compressed_size(
+/*=====================*/
+			/* out: compressed size in bytes */
+	ulint	n)	/* in: ulint integer to be stored */
+	__attribute__((const));
+/*************************************************************
+Reads a ulint in a compressed form. */
+UNIV_INLINE
+ulint
+mach_read_compressed(
+/*=================*/
+				/* out: read integer */
+	const byte*	b)	/* in: pointer to memory from where to read */
+	__attribute__((nonnull, pure));
+/***********************************************************
+The following function is used to store data in 6 consecutive
+bytes. We store the most significant byte to the lowest address. */
+UNIV_INLINE
+void
+mach_write_to_6(
+/*============*/
+	byte*	b,	/* in: pointer to 6 bytes where to store */
+	dulint	n);	 /* in: dulint integer to be stored */
+/************************************************************
+The following function is used to fetch data from 6 consecutive
+bytes. The most significant byte is at the lowest address. */
+UNIV_INLINE
+dulint
+mach_read_from_6(
+/*=============*/
+				/* out: dulint integer */
+	const byte*	b)	/* in: pointer to 6 bytes */
+	__attribute__((nonnull, pure));
+/***********************************************************
+The following function is used to store data in 7 consecutive
+bytes. We store the most significant byte to the lowest address. */
+UNIV_INLINE
+void
+mach_write_to_7(
+/*============*/
+	byte*	b,	/* in: pointer to 7 bytes where to store */
+	dulint	n);	 /* in: dulint integer to be stored */
+/************************************************************
+The following function is used to fetch data from 7 consecutive
+bytes. The most significant byte is at the lowest address. */
+UNIV_INLINE
+dulint
+mach_read_from_7(
+/*=============*/
+				/* out: dulint integer */
+	const byte*	b)	/* in: pointer to 7 bytes */
+	__attribute__((nonnull, pure));
+/***********************************************************
+The following function is used to store data in 8 consecutive
+bytes. We store the most significant byte to the lowest address. */
+UNIV_INLINE
+void
+mach_write_to_8(
+/*============*/
+	byte*	b,	/* in: pointer to 8 bytes where to store */
+	dulint	n);	/* in: dulint integer to be stored */
+/***********************************************************
+The following function is used to store data in 8 consecutive
+bytes. We store the most significant byte to the lowest address. */
+UNIV_INLINE
+void
+mach_write_ull(
+/*===========*/
+	byte*		b,	/* in: pointer to 8 bytes where to store */
+	ib_uint64_t	n);	/* in: 64-bit integer to be stored */
+/************************************************************
+The following function is used to fetch data from 8 consecutive
+bytes. The most significant byte is at the lowest address. */
+UNIV_INLINE
+dulint
+mach_read_from_8(
+/*=============*/
+				/* out: dulint integer */
+	const byte*	b)	/* in: pointer to 8 bytes */
+	__attribute__((nonnull, pure));
+/************************************************************
+The following function is used to fetch data from 8 consecutive
+bytes. The most significant byte is at the lowest address. */
+UNIV_INLINE
+ib_uint64_t
+mach_read_ull(
+/*==========*/
+				/* out: 64-bit integer */
+	const byte*	b)	/* in: pointer to 8 bytes */
+	__attribute__((nonnull, pure));
+/*************************************************************
+Writes a dulint in a compressed form (5..9 bytes). */
+UNIV_INLINE
+ulint
+mach_dulint_write_compressed(
+/*=========================*/
+			/* out: size in bytes */
+	byte*	b,	/* in: pointer to memory where to store */
+	dulint	n);	/* in: dulint integer to be stored */
+/*************************************************************
+Returns the size of a dulint when written in the compressed form. */
+UNIV_INLINE
+ulint
+mach_dulint_get_compressed_size(
+/*============================*/
+			/* out: compressed size in bytes */
+	dulint	 n);	/* in: dulint integer to be stored */
+/*************************************************************
+Reads a dulint in a compressed form. */
+UNIV_INLINE
+dulint
+mach_dulint_read_compressed(
+/*========================*/
+				/* out: read dulint */
+	const byte*	b)	/* in: pointer to memory from where to read */
+	__attribute__((nonnull, pure));
+/*************************************************************
+Writes a dulint in a compressed form (1..11 bytes). */
+UNIV_INLINE
+ulint
+mach_dulint_write_much_compressed(
+/*==============================*/
+			/* out: size in bytes */
+	byte*	b,	/* in: pointer to memory where to store */
+	dulint	n);	/* in: dulint integer to be stored */
+/*************************************************************
+Returns the size of a dulint when written in the compressed form. */
+UNIV_INLINE
+ulint
+mach_dulint_get_much_compressed_size(
+/*=================================*/
+			/* out: compressed size in bytes */
+	dulint	 n)	 /* in: dulint integer to be stored */
+	__attribute__((const));
+/*************************************************************
+Reads a dulint in a compressed form. */
+UNIV_INLINE
+dulint
+mach_dulint_read_much_compressed(
+/*=============================*/
+				/* out: read dulint */
+	const byte*	b)	/* in: pointer to memory from where to read */
+	__attribute__((nonnull, pure));
+/*************************************************************
+Reads a ulint in a compressed form if the log record fully contains it. */
+UNIV_INTERN
+byte*
+mach_parse_compressed(
+/*==================*/
+			/* out: pointer to end of the stored field, NULL if
+			not complete */
+	byte*	ptr,	/* in: pointer to buffer from where to read */
+	byte*	end_ptr,/* in: pointer to end of the buffer */
+	ulint*	val);	/* out: read value */
+/*************************************************************
+Reads a dulint in a compressed form if the log record fully contains it. */
+UNIV_INTERN
+byte*
+mach_dulint_parse_compressed(
+/*=========================*/
+			/* out: pointer to end of the stored field, NULL if
+			not complete */
+	byte*	ptr,	/* in: pointer to buffer from where to read */
+	byte*	end_ptr,/* in: pointer to end of the buffer */
+	dulint*	val);	/* out: read value */
+/*************************************************************
+Reads a double. It is stored in a little-endian format. */
+UNIV_INLINE
+double
+mach_double_read(
+/*=============*/
+				/* out: double read */
+	const byte*	b)	/* in: pointer to memory from where to read */
+	__attribute__((nonnull, pure));
+/*************************************************************
+Writes a double. It is stored in a little-endian format. */
+UNIV_INLINE
+void
+mach_double_write(
+/*==============*/
+	byte*	b,	/* in: pointer to memory where to write */
+	double	d);	/* in: double */
+/*************************************************************
+Reads a float. It is stored in a little-endian format. */
+UNIV_INLINE
+float
+mach_float_read(
+/*============*/
+				/* out: float read */
+	const byte*	b)	/* in: pointer to memory from where to read */
+	__attribute__((nonnull, pure));
+/*************************************************************
+Writes a float. It is stored in a little-endian format. */
+UNIV_INLINE
+void
+mach_float_write(
+/*=============*/
+	byte*	b,	/* in: pointer to memory where to write */
+	float	d);	/* in: float */
+/*************************************************************
+Reads a ulint stored in the little-endian format. */
+UNIV_INLINE
+ulint
+mach_read_from_n_little_endian(
+/*===========================*/
+					/* out: unsigned long int */
+	const byte*	buf,		/* in: from where to read */
+	ulint		buf_size)	/* in: from how many bytes to read */
+	__attribute__((nonnull, pure));
+/*************************************************************
+Writes a ulint in the little-endian format. */
+UNIV_INLINE
+void
+mach_write_to_n_little_endian(
+/*==========================*/
+	byte*	dest,		/* in: where to write */
+	ulint	dest_size,	/* in: into how many bytes to write */
+	ulint	n);		/* in: unsigned long int to write */
+/*************************************************************
+Reads a ulint stored in the little-endian format. */
+UNIV_INLINE
+ulint
+mach_read_from_2_little_endian(
+/*===========================*/
+					/* out: unsigned long int */
+	const byte*	buf)		/* in: from where to read */
+	__attribute__((nonnull, pure));
+/*************************************************************
+Writes a ulint in the little-endian format. */
+UNIV_INLINE
+void
+mach_write_to_2_little_endian(
+/*==========================*/
+	byte*	dest,		/* in: where to write */
+	ulint	n);		/* in: unsigned long int to write */
+
+/*************************************************************
+Convert integral type from storage byte order (big endian) to
+host byte order. */
+UNIV_INLINE
+ullint
+mach_read_int_type(
+/*===============*/
+					/* out: integer value */
+	const byte*	src,		/* in: where to read from */
+	ulint		len,		/* in: length of src */
+	ibool		unsigned_type);	/* in: signed or unsigned flag */
+#ifndef UNIV_NONINL
+#include "mach0data.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/mach0data.ic b/storage/xtradb/include/mach0data.ic
new file mode 100644
index 00000000000..5dda9aece2f
--- /dev/null
+++ b/storage/xtradb/include/mach0data.ic
@@ -0,0 +1,784 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/**********************************************************************
+Utilities for converting data from the database file
+to the machine format.
+
+Created 11/28/1995 Heikki Tuuri
+***********************************************************************/
+
+#include "ut0mem.h"
+
+/***********************************************************
+The following function is used to store data in one byte. */
+UNIV_INLINE
+void
+mach_write_to_1(
+/*============*/
+	byte*	b,	/* in: pointer to byte where to store */
+	ulint	n)	/* in: ulint integer to be stored, >= 0, < 256 */
+{
+	ut_ad(b);
+	ut_ad(n <= 0xFFUL);
+
+	b[0] = (byte)n;
+}
+
+/************************************************************
+The following function is used to fetch data from one byte. */
+UNIV_INLINE
+ulint
+mach_read_from_1(
+/*=============*/
+				/* out: ulint integer, >= 0, < 256 */
+	const byte*	b)	/* in: pointer to byte */
+{
+	ut_ad(b);
+	return((ulint)(b[0]));
+}
+
+/***********************************************************
+The following function is used to store data in two consecutive
+bytes. We store the most significant byte to the lowest address. */
+UNIV_INLINE
+void
+mach_write_to_2(
+/*============*/
+	byte*	b,	/* in: pointer to two bytes where to store */
+	ulint	n)	/* in: ulint integer to be stored */
+{
+	ut_ad(b);
+	ut_ad(n <= 0xFFFFUL);
+
+	b[0] = (byte)(n >> 8);
+	b[1] = (byte)(n);
+}
+
+/************************************************************
+The following function is used to fetch data from 2 consecutive
+bytes. The most significant byte is at the lowest address. */
+UNIV_INLINE
+ulint
+mach_read_from_2(
+/*=============*/
+				/* out: ulint integer */
+	const byte*	b)	/* in: pointer to 2 bytes */
+{
+	ut_ad(b);
+	return( ((ulint)(b[0]) << 8)
+		+ (ulint)(b[1])
+		);
+}
+
+/************************************************************
+The following function is used to convert a 16-bit data item
+to the canonical format, for fast bytewise equality test
+against memory. */
+UNIV_INLINE
+uint16
+mach_encode_2(
+/*==========*/
+			/* out: 16-bit integer in canonical format */
+	ulint	n)	/* in: integer in machine-dependent format */
+{
+	uint16	ret;
+	ut_ad(2 == sizeof ret);
+	mach_write_to_2((byte*) &ret, n);
+	return(ret);
+}
+/************************************************************
+The following function is used to convert a 16-bit data item
+from the canonical format, for fast bytewise equality test
+against memory. */
+UNIV_INLINE
+ulint
+mach_decode_2(
+/*==========*/
+			/* out: integer in machine-dependent format */
+	uint16	n)	/* in: 16-bit integer in canonical format */
+{
+	ut_ad(2 == sizeof n);
+	return(mach_read_from_2((const byte*) &n));
+}
+
+/***********************************************************
+The following function is used to store data in 3 consecutive
+bytes. We store the most significant byte to the lowest address. */
+UNIV_INLINE
+void
+mach_write_to_3(
+/*============*/
+	byte*	b,	/* in: pointer to 3 bytes where to store */
+	ulint	n)	/* in: ulint integer to be stored */
+{
+	ut_ad(b);
+	ut_ad(n <= 0xFFFFFFUL);
+
+	b[0] = (byte)(n >> 16);
+	b[1] = (byte)(n >> 8);
+	b[2] = (byte)(n);
+}
+
+/************************************************************
+The following function is used to fetch data from 3 consecutive
+bytes. The most significant byte is at the lowest address. */
+UNIV_INLINE
+ulint
+mach_read_from_3(
+/*=============*/
+				/* out: ulint integer */
+	const byte*	b)	/* in: pointer to 3 bytes */
+{
+	ut_ad(b);
+	return( ((ulint)(b[0]) << 16)
+		+ ((ulint)(b[1]) << 8)
+		+ (ulint)(b[2])
+		);
+}
+
+/***********************************************************
+The following function is used to store data in four consecutive
+bytes. We store the most significant byte to the lowest address. */
+UNIV_INLINE
+void
+mach_write_to_4(
+/*============*/
+	byte*	b,	/* in: pointer to four bytes where to store */
+	ulint	n)	/* in: ulint integer to be stored */
+{
+	ut_ad(b);
+
+	b[0] = (byte)(n >> 24);
+	b[1] = (byte)(n >> 16);
+	b[2] = (byte)(n >> 8);
+	b[3] = (byte)n;
+}
+
+/************************************************************
+The following function is used to fetch data from 4 consecutive
+bytes. The most significant byte is at the lowest address. */
+UNIV_INLINE
+ulint
+mach_read_from_4(
+/*=============*/
+				/* out: ulint integer */
+	const byte*	b)	/* in: pointer to four bytes */
+{
+	ut_ad(b);
+	return( ((ulint)(b[0]) << 24)
+		+ ((ulint)(b[1]) << 16)
+		+ ((ulint)(b[2]) << 8)
+		+ (ulint)(b[3])
+		);
+}
+
+/*************************************************************
+Writes a ulint in a compressed form where the first byte codes the
+length of the stored ulint. We look at the most significant bits of
+the byte. If the most significant bit is zero, it means 1-byte storage,
+else if the 2nd bit is 0, it means 2-byte storage, else if 3rd is 0,
+it means 3-byte storage, else if 4th is 0, it means 4-byte storage,
+else the storage is 5-byte. */
+UNIV_INLINE
+ulint
+mach_write_compressed(
+/*==================*/
+			/* out: compressed size in bytes */
+	byte*	b,	/* in: pointer to memory where to store */
+	ulint	n)	/* in: ulint integer (< 2^32) to be stored */
+{
+	ut_ad(b);
+
+	if (n < 0x80UL) {
+		mach_write_to_1(b, n);
+		return(1);
+	} else if (n < 0x4000UL) {
+		mach_write_to_2(b, n | 0x8000UL);
+		return(2);
+	} else if (n < 0x200000UL) {
+		mach_write_to_3(b, n | 0xC00000UL);
+		return(3);
+	} else if (n < 0x10000000UL) {
+		mach_write_to_4(b, n | 0xE0000000UL);
+		return(4);
+	} else {
+		mach_write_to_1(b, 0xF0UL);
+		mach_write_to_4(b + 1, n);
+		return(5);
+	}
+}
+
+/*************************************************************
+Returns the size of a ulint when written in the compressed form. */
+UNIV_INLINE
+ulint
+mach_get_compressed_size(
+/*=====================*/
+			/* out: compressed size in bytes */
+	ulint	n)	/* in: ulint integer (< 2^32) to be stored */
+{
+	if (n < 0x80UL) {
+		return(1);
+	} else if (n < 0x4000UL) {
+		return(2);
+	} else if (n < 0x200000UL) {
+		return(3);
+	} else if (n < 0x10000000UL) {
+		return(4);
+	} else {
+		return(5);
+	}
+}
+
+/*************************************************************
+Reads a ulint in a compressed form. */
+UNIV_INLINE
+ulint
+mach_read_compressed(
+/*=================*/
+				/* out: read integer (< 2^32) */
+	const byte*	b)	/* in: pointer to memory from where to read */
+{
+	ulint	flag;
+
+	ut_ad(b);
+
+	flag = mach_read_from_1(b);
+
+	if (flag < 0x80UL) {
+		return(flag);
+	} else if (flag < 0xC0UL) {
+		return(mach_read_from_2(b) & 0x7FFFUL);
+	} else if (flag < 0xE0UL) {
+		return(mach_read_from_3(b) & 0x3FFFFFUL);
+	} else if (flag < 0xF0UL) {
+		return(mach_read_from_4(b) & 0x1FFFFFFFUL);
+	} else {
+		ut_ad(flag == 0xF0UL);
+		return(mach_read_from_4(b + 1));
+	}
+}
+
+/***********************************************************
+The following function is used to store data in 8 consecutive
+bytes. We store the most significant byte to the lowest address. */
+UNIV_INLINE
+void
+mach_write_to_8(
+/*============*/
+	byte*	b,	/* in: pointer to 8 bytes where to store */
+	dulint	n)	/* in: dulint integer to be stored */
+{
+	ut_ad(b);
+
+	mach_write_to_4(b, ut_dulint_get_high(n));
+	mach_write_to_4(b + 4, ut_dulint_get_low(n));
+}
+
+/***********************************************************
+The following function is used to store data in 8 consecutive
+bytes. We store the most significant byte to the lowest address. */
+UNIV_INLINE
+void
+mach_write_ull(
+/*===========*/
+	byte*		b,	/* in: pointer to 8 bytes where to store */
+	ib_uint64_t	n)	/* in: 64-bit integer to be stored */
+{
+	ut_ad(b);
+
+	mach_write_to_4(b, (ulint) (n >> 32));
+	mach_write_to_4(b + 4, (ulint) n);
+}
+
+/************************************************************
+The following function is used to fetch data from 8 consecutive
+bytes. The most significant byte is at the lowest address. */
+UNIV_INLINE
+dulint
+mach_read_from_8(
+/*=============*/
+				/* out: dulint integer */
+	const byte*	b)	/* in: pointer to 8 bytes */
+{
+	ulint	high;
+	ulint	low;
+
+	ut_ad(b);
+
+	high = mach_read_from_4(b);
+	low = mach_read_from_4(b + 4);
+
+	return(ut_dulint_create(high, low));
+}
+
+/************************************************************
+The following function is used to fetch data from 8 consecutive
+bytes. The most significant byte is at the lowest address. */
+UNIV_INLINE
+ib_uint64_t
+mach_read_ull(
+/*==========*/
+				/* out: 64-bit integer */
+	const byte*	b)	/* in: pointer to 8 bytes */
+{
+	ib_uint64_t	ull;
+
+	ull = ((ib_uint64_t) mach_read_from_4(b)) << 32;
+	ull |= (ib_uint64_t) mach_read_from_4(b + 4);
+
+	return(ull);
+}
+
+/***********************************************************
+The following function is used to store data in 7 consecutive
+bytes. We store the most significant byte to the lowest address. */
+UNIV_INLINE
+void
+mach_write_to_7(
+/*============*/
+	byte*	b,	/* in: pointer to 7 bytes where to store */
+	dulint	n)	/* in: dulint integer to be stored */
+{
+	ut_ad(b);
+
+	mach_write_to_3(b, ut_dulint_get_high(n));
+	mach_write_to_4(b + 3, ut_dulint_get_low(n));
+}
+
+/************************************************************
+The following function is used to fetch data from 7 consecutive
+bytes. The most significant byte is at the lowest address. */
+UNIV_INLINE
+dulint
+mach_read_from_7(
+/*=============*/
+				/* out: dulint integer */
+	const byte*	b)	/* in: pointer to 7 bytes */
+{
+	ulint	high;
+	ulint	low;
+
+	ut_ad(b);
+
+	high = mach_read_from_3(b);
+	low = mach_read_from_4(b + 3);
+
+	return(ut_dulint_create(high, low));
+}
+
+/***********************************************************
+The following function is used to store data in 6 consecutive
+bytes. We store the most significant byte to the lowest address. */
+UNIV_INLINE
+void
+mach_write_to_6(
+/*============*/
+	byte*	b,	/* in: pointer to 6 bytes where to store */
+	dulint	n)	/* in: dulint integer to be stored */
+{
+	ut_ad(b);
+
+	mach_write_to_2(b, ut_dulint_get_high(n));
+	mach_write_to_4(b + 2, ut_dulint_get_low(n));
+}
+
+/************************************************************
+The following function is used to fetch data from 6 consecutive
+bytes. The most significant byte is at the lowest address. */
+UNIV_INLINE
+dulint
+mach_read_from_6(
+/*=============*/
+				/* out: dulint integer */
+	const byte*	b)	/* in: pointer to 6 bytes */
+{
+	ulint	high;
+	ulint	low;
+
+	ut_ad(b);
+
+	high = mach_read_from_2(b);
+	low = mach_read_from_4(b + 2);
+
+	return(ut_dulint_create(high, low));
+}
+
+/*************************************************************
+Writes a dulint in a compressed form (5..9 bytes). */
+UNIV_INLINE
+ulint
+mach_dulint_write_compressed(
+/*=========================*/
+			/* out: size in bytes */
+	byte*	b,	/* in: pointer to memory where to store */
+	dulint	n)	/* in: dulint integer to be stored */
+{
+	ulint	size;
+
+	ut_ad(b);
+
+	size = mach_write_compressed(b, ut_dulint_get_high(n));
+	mach_write_to_4(b + size, ut_dulint_get_low(n));
+
+	return(size + 4);
+}
+
+/*************************************************************
+Returns the size of a dulint when written in the compressed form. */
+UNIV_INLINE
+ulint
+mach_dulint_get_compressed_size(
+/*============================*/
+			/* out: compressed size in bytes */
+	dulint	 n)	/* in: dulint integer to be stored */
+{
+	return(4 + mach_get_compressed_size(ut_dulint_get_high(n)));
+}
+
+/*************************************************************
+Reads a dulint in a compressed form. */
+UNIV_INLINE
+dulint
+mach_dulint_read_compressed(
+/*========================*/
+				/* out: read dulint */
+	const byte*	b)	/* in: pointer to memory from where to read */
+{
+	ulint	high;
+	ulint	low;
+	ulint	size;
+
+	ut_ad(b);
+
+	high = mach_read_compressed(b);
+
+	size = mach_get_compressed_size(high);
+
+	low = mach_read_from_4(b + size);
+
+	return(ut_dulint_create(high, low));
+}
+
+/*************************************************************
+Writes a dulint in a compressed form (1..11 bytes). */
+UNIV_INLINE
+ulint
+mach_dulint_write_much_compressed(
+/*==============================*/
+			/* out: size in bytes */
+	byte*	b,	/* in: pointer to memory where to store */
+	dulint	n)	/* in: dulint integer to be stored */
+{
+	ulint	size;
+
+	ut_ad(b);
+
+	if (ut_dulint_get_high(n) == 0) {
+		return(mach_write_compressed(b, ut_dulint_get_low(n)));
+	}
+
+	*b = (byte)0xFF;
+	size = 1 + mach_write_compressed(b + 1, ut_dulint_get_high(n));
+
+	size += mach_write_compressed(b + size, ut_dulint_get_low(n));
+
+	return(size);
+}
+
+/*************************************************************
+Returns the size of a dulint when written in the compressed form. */
+UNIV_INLINE
+ulint
+mach_dulint_get_much_compressed_size(
+/*=================================*/
+			/* out: compressed size in bytes */
+	dulint	 n)	/* in: dulint integer to be stored */
+{
+	if (0 == ut_dulint_get_high(n)) {
+		return(mach_get_compressed_size(ut_dulint_get_low(n)));
+	}
+
+	return(1 + mach_get_compressed_size(ut_dulint_get_high(n))
+	       + mach_get_compressed_size(ut_dulint_get_low(n)));
+}
+
+/*************************************************************
+Reads a dulint in a compressed form. */
+UNIV_INLINE
+dulint
+mach_dulint_read_much_compressed(
+/*=============================*/
+				/* out: read dulint */
+	const byte*	b)	/* in: pointer to memory from where to read */
+{
+	ulint	high;
+	ulint	low;
+	ulint	size;
+
+	ut_ad(b);
+
+	if (*b != (byte)0xFF) {
+		high = 0;
+		size = 0;
+	} else {
+		high = mach_read_compressed(b + 1);
+
+		size = 1 + mach_get_compressed_size(high);
+	}
+
+	low = mach_read_compressed(b + size);
+
+	return(ut_dulint_create(high, low));
+}
+
+/*************************************************************
+Reads a double. It is stored in a little-endian format. */
+UNIV_INLINE
+double
+mach_double_read(
+/*=============*/
+				/* out: double read */
+	const byte*	b)	/* in: pointer to memory from where to read */
+{
+	double	d;
+	ulint	i;
+	byte*	ptr;
+
+	ptr = (byte*)&d;
+
+	for (i = 0; i < sizeof(double); i++) {
+#ifdef WORDS_BIGENDIAN
+		ptr[sizeof(double) - i - 1] = b[i];
+#else
+		ptr[i] = b[i];
+#endif
+	}
+
+	return(d);
+}
+
+/*************************************************************
+Writes a double. It is stored in a little-endian format. */
+UNIV_INLINE
+void
+mach_double_write(
+/*==============*/
+	byte*	b,	/* in: pointer to memory where to write */
+	double	d)	/* in: double */
+{
+	ulint	i;
+	byte*	ptr;
+
+	ptr = (byte*)&d;
+
+	for (i = 0; i < sizeof(double); i++) {
+#ifdef WORDS_BIGENDIAN
+		b[i] = ptr[sizeof(double) - i - 1];
+#else
+		b[i] = ptr[i];
+#endif
+	}
+}
+
+/*************************************************************
+Reads a float. It is stored in a little-endian format. */
+UNIV_INLINE
+float
+mach_float_read(
+/*============*/
+				/* out: float read */
+	const byte*	b)	/* in: pointer to memory from where to read */
+{
+	float	d;
+	ulint	i;
+	byte*	ptr;
+
+	ptr = (byte*)&d;
+
+	for (i = 0; i < sizeof(float); i++) {
+#ifdef WORDS_BIGENDIAN
+		ptr[sizeof(float) - i - 1] = b[i];
+#else
+		ptr[i] = b[i];
+#endif
+	}
+
+	return(d);
+}
+
+/*************************************************************
+Writes a float. It is stored in a little-endian format. */
+UNIV_INLINE
+void
+mach_float_write(
+/*=============*/
+	byte*	b,	/* in: pointer to memory where to write */
+	float	d)	/* in: float */
+{
+	ulint	i;
+	byte*	ptr;
+
+	ptr = (byte*)&d;
+
+	for (i = 0; i < sizeof(float); i++) {
+#ifdef WORDS_BIGENDIAN
+		b[i] = ptr[sizeof(float) - i - 1];
+#else
+		b[i] = ptr[i];
+#endif
+	}
+}
+
+/*************************************************************
+Reads a ulint stored in the little-endian format. */
+UNIV_INLINE
+ulint
+mach_read_from_n_little_endian(
+/*===========================*/
+					/* out: unsigned long int */
+	const byte*	buf,		/* in: from where to read */
+	ulint		buf_size)	/* in: from how many bytes to read */
+{
+	ulint	n	= 0;
+	const byte*	ptr;
+
+	ut_ad(buf_size <= sizeof(ulint));
+	ut_ad(buf_size > 0);
+
+	ptr = buf + buf_size;
+
+	for (;;) {
+		ptr--;
+
+		n = n << 8;
+
+		n += (ulint)(*ptr);
+
+		if (ptr == buf) {
+			break;
+		}
+	}
+
+	return(n);
+}
+
+/*************************************************************
+Writes a ulint in the little-endian format. */
+UNIV_INLINE
+void
+mach_write_to_n_little_endian(
+/*==========================*/
+	byte*	dest,		/* in: where to write */
+	ulint	dest_size,	/* in: into how many bytes to write */
+	ulint	n)		/* in: unsigned long int to write */
+{
+	byte*	end;
+
+	ut_ad(dest_size <= sizeof(ulint));
+	ut_ad(dest_size > 0);
+
+	end = dest + dest_size;
+
+	for (;;) {
+		*dest = (byte)(n & 0xFF);
+
+		n = n >> 8;
+
+		dest++;
+
+		if (dest == end) {
+			break;
+		}
+	}
+
+	ut_ad(n == 0);
+}
+
+/*************************************************************
+Reads a ulint stored in the little-endian format. */
+UNIV_INLINE
+ulint
+mach_read_from_2_little_endian(
+/*===========================*/
+					/* out: unsigned long int */
+	const byte*	buf)		/* in: from where to read */
+{
+	return((ulint)(*buf) + ((ulint)(*(buf + 1))) * 256);
+}
+
+/*************************************************************
+Writes a ulint in the little-endian format. */
+UNIV_INLINE
+void
+mach_write_to_2_little_endian(
+/*==========================*/
+	byte*	dest,		/* in: where to write */
+	ulint	n)		/* in: unsigned long int to write */
+{
+	ut_ad(n < 256 * 256);
+
+	*dest = (byte)(n & 0xFFUL);
+
+	n = n >> 8;
+	dest++;
+
+	*dest = (byte)(n & 0xFFUL);
+}
+
+/*************************************************************
+Convert integral type from storage byte order (big endian) to
+host byte order. */
+UNIV_INLINE
+ullint
+mach_read_int_type(
+/*===============*/
+					/* out: integer value */
+	const byte*	src,		/* in: where to read from */
+	ulint		len,		/* in: length of src */
+	ibool		unsigned_type)	/* in: signed or unsigned flag */
+{
+	/* XXX this can be optimized on big-endian machines */
+
+	ullint	ret;
+	uint	i;
+
+	if (unsigned_type || (src[0] & 0x80)) {
+
+		ret = 0x0000000000000000ULL;
+	} else {
+
+		ret = 0xFFFFFFFFFFFFFF00ULL;
+	}
+
+	if (unsigned_type) {
+
+		ret |= src[0];
+	} else {
+
+		ret |= src[0] ^ 0x80;
+	}
+
+	for (i = 1; i < len; i++) {
+		ret <<= 8;
+		ret |= src[i];
+	}
+
+	return(ret);
+}
diff --git a/storage/xtradb/include/mem0dbg.h b/storage/xtradb/include/mem0dbg.h
new file mode 100644
index 00000000000..0568a595d06
--- /dev/null
+++ b/storage/xtradb/include/mem0dbg.h
@@ -0,0 +1,142 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The memory management: the debug code. This is not a compilation module,
+but is included in mem0mem.* !
+
+Created 6/9/1994 Heikki Tuuri
+*******************************************************/
+
+/* In the debug version each allocated field is surrounded with
+check fields whose sizes are given below */
+
+#ifdef UNIV_MEM_DEBUG
+#define MEM_FIELD_HEADER_SIZE	ut_calc_align(2 * sizeof(ulint),\
+						UNIV_MEM_ALIGNMENT)
+#define MEM_FIELD_TRAILER_SIZE	sizeof(ulint)
+#else
+#define MEM_FIELD_HEADER_SIZE	0
+#endif
+
+
+/* Space needed when allocating for a user a field of
+length N. The space is allocated only in multiples of
+UNIV_MEM_ALIGNMENT. In the debug version there are also
+check fields at the both ends of the field. */
+#ifdef UNIV_MEM_DEBUG
+#define MEM_SPACE_NEEDED(N) ut_calc_align((N) + MEM_FIELD_HEADER_SIZE\
+		 + MEM_FIELD_TRAILER_SIZE, UNIV_MEM_ALIGNMENT)
+#else
+#define MEM_SPACE_NEEDED(N) ut_calc_align((N), UNIV_MEM_ALIGNMENT)
+#endif
+
+#if defined UNIV_MEM_DEBUG || defined UNIV_DEBUG
+/*******************************************************************
+Checks a memory heap for consistency and prints the contents if requested.
+Outputs the sum of sizes of buffers given to the user (only in
+the debug version), the physical size of the heap and the number of
+blocks in the heap. In case of error returns 0 as sizes and number
+of blocks. */
+UNIV_INTERN
+void
+mem_heap_validate_or_print(
+/*=======================*/
+	mem_heap_t*	heap,	/* in: memory heap */
+	byte*		top,	/* in: calculate and validate only until
+				this top pointer in the heap is reached,
+				if this pointer is NULL, ignored */
+	ibool		 print,	 /* in: if TRUE, prints the contents
+				of the heap; works only in
+				the debug version */
+	ibool*		 error,	 /* out: TRUE if error */
+	ulint*		us_size,/* out: allocated memory
+				(for the user) in the heap,
+				if a NULL pointer is passed as this
+				argument, it is ignored; in the
+				non-debug version this is always -1 */
+	ulint*		ph_size,/* out: physical size of the heap,
+				if a NULL pointer is passed as this
+				argument, it is ignored */
+	ulint*		n_blocks); /* out: number of blocks in the heap,
+				if a NULL pointer is passed as this
+				argument, it is ignored */
+/******************************************************************
+Validates the contents of a memory heap. */
+UNIV_INTERN
+ibool
+mem_heap_validate(
+/*==============*/
+				/* out: TRUE if ok */
+	mem_heap_t*   heap);	/* in: memory heap */
+#endif /* UNIV_MEM_DEBUG || UNIV_DEBUG */
+#ifdef UNIV_DEBUG
+/******************************************************************
+Checks that an object is a memory heap (or a block of it) */
+UNIV_INTERN
+ibool
+mem_heap_check(
+/*===========*/
+				/* out: TRUE if ok */
+	mem_heap_t*   heap);	/* in: memory heap */
+#endif /* UNIV_DEBUG */
+#ifdef UNIV_MEM_DEBUG
+/*********************************************************************
+TRUE if no memory is currently allocated. */
+UNIV_INTERN
+ibool
+mem_all_freed(void);
+/*===============*/
+			/* out: TRUE if no heaps exist */
+/*********************************************************************
+Validates the dynamic memory */
+UNIV_INTERN
+ibool
+mem_validate_no_assert(void);
+/*=========================*/
+			/* out: TRUE if error */
+/****************************************************************
+Validates the dynamic memory */
+UNIV_INTERN
+ibool
+mem_validate(void);
+/*===============*/
+			/* out: TRUE if ok */
+#endif /* UNIV_MEM_DEBUG */
+/****************************************************************
+Tries to find neigboring memory allocation blocks and dumps to stderr
+the neighborhood of a given pointer. */
+UNIV_INTERN
+void
+mem_analyze_corruption(
+/*===================*/
+	void*	ptr);	/* in: pointer to place of possible corruption */
+/*********************************************************************
+Prints information of dynamic memory usage and currently allocated memory
+heaps or buffers. Can only be used in the debug version. */
+UNIV_INTERN
+void
+mem_print_info(void);
+/*================*/
+/*********************************************************************
+Prints information of dynamic memory usage and currently allocated memory
+heaps or buffers since the last ..._print_info or..._print_new_info. */
+UNIV_INTERN
+void
+mem_print_new_info(void);
+/*====================*/
diff --git a/storage/xtradb/include/mem0dbg.ic b/storage/xtradb/include/mem0dbg.ic
new file mode 100644
index 00000000000..bf695fee785
--- /dev/null
+++ b/storage/xtradb/include/mem0dbg.ic
@@ -0,0 +1,109 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/************************************************************************
+The memory management: the debug code. This is not an independent
+compilation module but is included in mem0mem.*.
+
+Created 6/8/1994 Heikki Tuuri
+*************************************************************************/
+
+#ifdef UNIV_MEM_DEBUG
+extern mutex_t	mem_hash_mutex;
+extern ulint	mem_current_allocated_memory;
+
+/**********************************************************************
+Initializes an allocated memory field in the debug version. */
+UNIV_INTERN
+void
+mem_field_init(
+/*===========*/
+	byte*	buf,	/* in: memory field */
+	ulint	n);	/* in: how many bytes the user requested */
+/**********************************************************************
+Erases an allocated memory field in the debug version. */
+UNIV_INTERN
+void
+mem_field_erase(
+/*============*/
+	byte*	buf,	/* in: memory field */
+	ulint	n);	/* in: how many bytes the user requested */
+/*******************************************************************
+Initializes a buffer to a random combination of hex BA and BE.
+Used to initialize allocated memory. */
+UNIV_INTERN
+void
+mem_init_buf(
+/*=========*/
+	byte*	buf,	/* in: pointer to buffer */
+	ulint	 n);	 /* in: length of buffer */
+/*******************************************************************
+Initializes a buffer to a random combination of hex DE and AD.
+Used to erase freed memory.*/
+UNIV_INTERN
+void
+mem_erase_buf(
+/*==========*/
+	byte*	buf,	/* in: pointer to buffer */
+	ulint	 n);	 /* in: length of buffer */
+/*******************************************************************
+Inserts a created memory heap to the hash table of
+current allocated memory heaps.
+Initializes the hash table when first called. */
+UNIV_INTERN
+void
+mem_hash_insert(
+/*============*/
+	mem_heap_t*	heap,	   /* in: the created heap */
+	const char*	file_name, /* in: file name of creation */
+	ulint		line);	   /* in: line where created */
+/*******************************************************************
+Removes a memory heap (which is going to be freed by the caller)
+from the list of live memory heaps. Returns the size of the heap
+in terms of how much memory in bytes was allocated for the user of
+the heap (not the total space occupied by the heap).
+Also validates the heap.
+NOTE: This function does not free the storage occupied by the
+heap itself, only the node in the list of heaps. */
+UNIV_INTERN
+void
+mem_hash_remove(
+/*============*/
+	mem_heap_t*	heap,	   /* in: the heap to be freed */
+	const char*	file_name, /* in: file name of freeing */
+	ulint		line);	   /* in: line where freed */
+
+
+void
+mem_field_header_set_len(byte* field, ulint len);
+
+ulint
+mem_field_header_get_len(byte* field);
+
+void
+mem_field_header_set_check(byte* field, ulint check);
+
+ulint
+mem_field_header_get_check(byte* field);
+
+void
+mem_field_trailer_set_check(byte* field, ulint check);
+
+ulint
+mem_field_trailer_get_check(byte* field);
+#endif /* UNIV_MEM_DEBUG */
diff --git a/storage/xtradb/include/mem0mem.h b/storage/xtradb/include/mem0mem.h
new file mode 100644
index 00000000000..c20e7815001
--- /dev/null
+++ b/storage/xtradb/include/mem0mem.h
@@ -0,0 +1,406 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The memory management
+
+Created 6/9/1994 Heikki Tuuri
+*******************************************************/
+
+#ifndef mem0mem_h
+#define mem0mem_h
+
+#include "univ.i"
+#include "ut0mem.h"
+#include "ut0byte.h"
+#include "ut0ut.h"
+#include "ut0rnd.h"
+#include "sync0sync.h"
+#include "ut0lst.h"
+#include "mach0data.h"
+
+/* -------------------- MEMORY HEAPS ----------------------------- */
+
+/* The info structure stored at the beginning of a heap block */
+typedef struct mem_block_info_struct mem_block_info_t;
+
+/* A block of a memory heap consists of the info structure
+followed by an area of memory */
+typedef mem_block_info_t	mem_block_t;
+
+/* A memory heap is a nonempty linear list of memory blocks */
+typedef mem_block_t	mem_heap_t;
+
+/* Types of allocation for memory heaps: DYNAMIC means allocation from the
+dynamic memory pool of the C compiler, BUFFER means allocation from the
+buffer pool; the latter method is used for very big heaps */
+
+#define MEM_HEAP_DYNAMIC	0	/* the most common type */
+#define MEM_HEAP_BUFFER		1
+#define MEM_HEAP_BTR_SEARCH	2	/* this flag can optionally be
+					ORed to MEM_HEAP_BUFFER, in which
+					case heap->free_block is used in
+					some cases for memory allocations,
+					and if it's NULL, the memory
+					allocation functions can return
+					NULL. */
+
+/* The following start size is used for the first block in the memory heap if
+the size is not specified, i.e., 0 is given as the parameter in the call of
+create. The standard size is the maximum (payload) size of the blocks used for
+allocations of small buffers. */
+
+#define MEM_BLOCK_START_SIZE		64
+#define MEM_BLOCK_STANDARD_SIZE		\
+	(UNIV_PAGE_SIZE >= 16384 ? 8000 : MEM_MAX_ALLOC_IN_BUF)
+
+/* If a memory heap is allowed to grow into the buffer pool, the following
+is the maximum size for a single allocated buffer: */
+#define MEM_MAX_ALLOC_IN_BUF		(UNIV_PAGE_SIZE - 200)
+
+/**********************************************************************
+Initializes the memory system. */
+UNIV_INTERN
+void
+mem_init(
+/*=====*/
+	ulint	size);	/* in: common pool size in bytes */
+/******************************************************************
+Use this macro instead of the corresponding function! Macro for memory
+heap creation. */
+
+#define mem_heap_create(N)	mem_heap_create_func(\
+		(N), MEM_HEAP_DYNAMIC, __FILE__, __LINE__)
+/******************************************************************
+Use this macro instead of the corresponding function! Macro for memory
+heap creation. */
+
+#define mem_heap_create_in_buffer(N)	mem_heap_create_func(\
+		(N), MEM_HEAP_BUFFER, __FILE__, __LINE__)
+/******************************************************************
+Use this macro instead of the corresponding function! Macro for memory
+heap creation. */
+
+#define mem_heap_create_in_btr_search(N)	mem_heap_create_func(\
+		(N), MEM_HEAP_BTR_SEARCH | MEM_HEAP_BUFFER,\
+		__FILE__, __LINE__)
+
+/******************************************************************
+Use this macro instead of the corresponding function! Macro for memory
+heap freeing. */
+
+#define mem_heap_free(heap) mem_heap_free_func(\
+					  (heap), __FILE__, __LINE__)
+/*********************************************************************
+NOTE: Use the corresponding macros instead of this function. Creates a
+memory heap. For debugging purposes, takes also the file name and line as
+arguments. */
+UNIV_INLINE
+mem_heap_t*
+mem_heap_create_func(
+/*=================*/
+					/* out, own: memory heap, NULL if
+					did not succeed (only possible for
+					MEM_HEAP_BTR_SEARCH type heaps)*/
+	ulint		n,		/* in: desired start block size,
+					this means that a single user buffer
+					of size n will fit in the block,
+					0 creates a default size block */
+	ulint		type,		/* in: heap type */
+	const char*	file_name,	/* in: file name where created */
+	ulint		line);		/* in: line where created */
+/*********************************************************************
+NOTE: Use the corresponding macro instead of this function. Frees the space
+occupied by a memory heap. In the debug version erases the heap memory
+blocks. */
+UNIV_INLINE
+void
+mem_heap_free_func(
+/*===============*/
+	mem_heap_t*	heap,		/* in, own: heap to be freed */
+	const char*	file_name,	/* in: file name where freed */
+	ulint		line);		/* in: line where freed */
+/*******************************************************************
+Allocates and zero-fills n bytes of memory from a memory heap. */
+UNIV_INLINE
+void*
+mem_heap_zalloc(
+/*============*/
+				/* out: allocated, zero-filled storage */
+	mem_heap_t*	heap,	/* in: memory heap */
+	ulint		n);	/* in: number of bytes; if the heap is allowed
+				to grow into the buffer pool, this must be
+				<= MEM_MAX_ALLOC_IN_BUF */
+/*******************************************************************
+Allocates n bytes of memory from a memory heap. */
+UNIV_INLINE
+void*
+mem_heap_alloc(
+/*===========*/
+				/* out: allocated storage, NULL if did not
+				succeed (only possible for
+				MEM_HEAP_BTR_SEARCH type heaps) */
+	mem_heap_t*	heap,	/* in: memory heap */
+	ulint		n);	/* in: number of bytes; if the heap is allowed
+				to grow into the buffer pool, this must be
+				<= MEM_MAX_ALLOC_IN_BUF */
+/*********************************************************************
+Returns a pointer to the heap top. */
+UNIV_INLINE
+byte*
+mem_heap_get_heap_top(
+/*==================*/
+				/* out: pointer to the heap top */
+	mem_heap_t*	heap);	/* in: memory heap */
+/*********************************************************************
+Frees the space in a memory heap exceeding the pointer given. The
+pointer must have been acquired from mem_heap_get_heap_top. The first
+memory block of the heap is not freed. */
+UNIV_INLINE
+void
+mem_heap_free_heap_top(
+/*===================*/
+	mem_heap_t*	heap,	/* in: heap from which to free */
+	byte*		old_top);/* in: pointer to old top of heap */
+/*********************************************************************
+Empties a memory heap. The first memory block of the heap is not freed. */
+UNIV_INLINE
+void
+mem_heap_empty(
+/*===========*/
+	mem_heap_t*	heap);	/* in: heap to empty */
+/*********************************************************************
+Returns a pointer to the topmost element in a memory heap.
+The size of the element must be given. */
+UNIV_INLINE
+void*
+mem_heap_get_top(
+/*=============*/
+				/* out: pointer to the topmost element */
+	mem_heap_t*	heap,	/* in: memory heap */
+	ulint		n);	/* in: size of the topmost element */
+/*********************************************************************
+Frees the topmost element in a memory heap.
+The size of the element must be given. */
+UNIV_INLINE
+void
+mem_heap_free_top(
+/*==============*/
+	mem_heap_t*	heap,	/* in: memory heap */
+	ulint		n);	/* in: size of the topmost element */
+/*********************************************************************
+Returns the space in bytes occupied by a memory heap. */
+UNIV_INLINE
+ulint
+mem_heap_get_size(
+/*==============*/
+	mem_heap_t*	heap);		/* in: heap */
+/******************************************************************
+Use this macro instead of the corresponding function!
+Macro for memory buffer allocation */
+
+#define mem_zalloc(N)	memset(mem_alloc(N), 0, (N));
+
+#define mem_alloc(N)	mem_alloc_func((N), NULL, __FILE__, __LINE__)
+#define mem_alloc2(N,S)	mem_alloc_func((N), (S), __FILE__, __LINE__)
+/*******************************************************************
+NOTE: Use the corresponding macro instead of this function.
+Allocates a single buffer of memory from the dynamic memory of
+the C compiler. Is like malloc of C. The buffer must be freed
+with mem_free. */
+UNIV_INLINE
+void*
+mem_alloc_func(
+/*===========*/
+					/* out, own: free storage */
+	ulint		n,		/* in: requested size in bytes */
+	ulint*		size,		/* out: allocated size in bytes,
+					or NULL */
+	const char*	file_name,	/* in: file name where created */
+	ulint		line);		/* in: line where created */
+
+/******************************************************************
+Use this macro instead of the corresponding function!
+Macro for memory buffer freeing */
+
+#define mem_free(PTR)	mem_free_func((PTR), __FILE__, __LINE__)
+/*******************************************************************
+NOTE: Use the corresponding macro instead of this function.
+Frees a single buffer of storage from
+the dynamic memory of C compiler. Similar to free of C. */
+UNIV_INLINE
+void
+mem_free_func(
+/*==========*/
+	void*		ptr,		/* in, own: buffer to be freed */
+	const char*	file_name,	/* in: file name where created */
+	ulint		line		/* in: line where created */
+);
+
+/**************************************************************************
+Duplicates a NUL-terminated string. */
+UNIV_INLINE
+char*
+mem_strdup(
+/*=======*/
+				/* out, own: a copy of the string,
+				must be deallocated with mem_free */
+	const char*	str);	/* in: string to be copied */
+/**************************************************************************
+Makes a NUL-terminated copy of a nonterminated string. */
+UNIV_INLINE
+char*
+mem_strdupl(
+/*========*/
+				/* out, own: a copy of the string,
+				must be deallocated with mem_free */
+	const char*	str,	/* in: string to be copied */
+	ulint		len);	/* in: length of str, in bytes */
+
+/**************************************************************************
+Duplicates a NUL-terminated string, allocated from a memory heap. */
+UNIV_INTERN
+char*
+mem_heap_strdup(
+/*============*/
+				/* out, own: a copy of the string */
+	mem_heap_t*	heap,	/* in: memory heap where string is allocated */
+	const char*	str);	/* in: string to be copied */
+/**************************************************************************
+Makes a NUL-terminated copy of a nonterminated string,
+allocated from a memory heap. */
+UNIV_INLINE
+char*
+mem_heap_strdupl(
+/*=============*/
+				/* out, own: a copy of the string */
+	mem_heap_t*	heap,	/* in: memory heap where string is allocated */
+	const char*	str,	/* in: string to be copied */
+	ulint		len);	/* in: length of str, in bytes */
+
+/**************************************************************************
+Concatenate two strings and return the result, using a memory heap. */
+UNIV_INTERN
+char*
+mem_heap_strcat(
+/*============*/
+				/* out, own: the result */
+	mem_heap_t*	heap,	/* in: memory heap where string is allocated */
+	const char*	s1,	/* in: string 1 */
+	const char*	s2);	/* in: string 2 */
+
+/**************************************************************************
+Duplicate a block of data, allocated from a memory heap. */
+UNIV_INTERN
+void*
+mem_heap_dup(
+/*=========*/
+				/* out, own: a copy of the data */
+	mem_heap_t*	heap,	/* in: memory heap where copy is allocated */
+	const void*	data,	/* in: data to be copied */
+	ulint		len);	/* in: length of data, in bytes */
+
+/**************************************************************************
+Concatenate two memory blocks and return the result, using a memory heap. */
+UNIV_INTERN
+void*
+mem_heap_cat(
+/*=========*/
+				/* out, own: the result */
+	mem_heap_t*	heap,	/* in: memory heap where result is allocated */
+	const void*	b1,	/* in: block 1 */
+	ulint		len1,	/* in: length of b1, in bytes */
+	const void*	b2,	/* in: block 2 */
+	ulint		len2);	/* in: length of b2, in bytes */
+
+/********************************************************************
+A simple (s)printf replacement that dynamically allocates the space for the
+formatted string from the given heap. This supports a very limited set of
+the printf syntax: types 's' and 'u' and length modifier 'l' (which is
+required for the 'u' type). */
+UNIV_INTERN
+char*
+mem_heap_printf(
+/*============*/
+				/* out: heap-allocated formatted string */
+	mem_heap_t*	heap,	/* in: memory heap */
+	const char*	format,	/* in: format string */
+	...) __attribute__ ((format (printf, 2, 3)));
+
+#ifdef MEM_PERIODIC_CHECK
+/**********************************************************************
+Goes through the list of all allocated mem blocks, checks their magic
+numbers, and reports possible corruption. */
+UNIV_INTERN
+void
+mem_validate_all_blocks(void);
+/*=========================*/
+#endif
+
+/*#######################################################################*/
+
+/* The info header of a block in a memory heap */
+
+struct mem_block_info_struct {
+	ulint	magic_n;/* magic number for debugging */
+	char	file_name[8];/* file name where the mem heap was created */
+	ulint	line;	/* line number where the mem heap was created */
+	UT_LIST_BASE_NODE_T(mem_block_t) base; /* In the first block in the
+			the list this is the base node of the list of blocks;
+			in subsequent blocks this is undefined */
+	UT_LIST_NODE_T(mem_block_t) list; /* This contains pointers to next
+			and prev in the list. The first block allocated
+			to the heap is also the first block in this list,
+			though it also contains the base node of the list. */
+	ulint	len;	/* physical length of this block in bytes */
+	ulint	type;	/* type of heap: MEM_HEAP_DYNAMIC, or
+			MEM_HEAP_BUF possibly ORed to MEM_HEAP_BTR_SEARCH */
+	ulint	free;	/* offset in bytes of the first free position for
+			user data in the block */
+	ulint	start;	/* the value of the struct field 'free' at the
+			creation of the block */
+	void*	free_block;
+			/* if the MEM_HEAP_BTR_SEARCH bit is set in type,
+			and this is the heap root, this can contain an
+			allocated buffer frame, which can be appended as a
+			free block to the heap, if we need more space;
+			otherwise, this is NULL */
+	void*	buf_block;
+			/* if this block has been allocated from the buffer
+			pool, this contains the buf_block_t handle;
+			otherwise, this is NULL */
+#ifdef MEM_PERIODIC_CHECK
+	UT_LIST_NODE_T(mem_block_t) mem_block_list;
+			/* List of all mem blocks allocated; protected
+			by the mem_comm_pool mutex */
+#endif
+};
+
+#define MEM_BLOCK_MAGIC_N	764741555
+#define MEM_FREED_BLOCK_MAGIC_N	547711122
+
+/* Header size for a memory heap block */
+#define MEM_BLOCK_HEADER_SIZE	ut_calc_align(sizeof(mem_block_info_t),\
+							UNIV_MEM_ALIGNMENT)
+#include "mem0dbg.h"
+
+#ifndef UNIV_NONINL
+#include "mem0mem.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/mem0mem.ic b/storage/xtradb/include/mem0mem.ic
new file mode 100644
index 00000000000..04b4234904a
--- /dev/null
+++ b/storage/xtradb/include/mem0mem.ic
@@ -0,0 +1,646 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/************************************************************************
+The memory management
+
+Created 6/8/1994 Heikki Tuuri
+*************************************************************************/
+
+#include "mem0dbg.ic"
+
+#include "mem0pool.h"
+
+/*******************************************************************
+Creates a memory heap block where data can be allocated. */
+UNIV_INTERN
+mem_block_t*
+mem_heap_create_block(
+/*==================*/
+				/* out, own: memory heap block, NULL if
+				did not succeed (only possible for
+				MEM_HEAP_BTR_SEARCH type heaps) */
+	mem_heap_t*	heap,	/* in: memory heap or NULL if first block
+				should be created */
+	ulint		n,	/* in: number of bytes needed for user data */
+	ulint		type,	/* in: type of heap: MEM_HEAP_DYNAMIC or
+				MEM_HEAP_BUFFER */
+	const char*	file_name,/* in: file name where created */
+	ulint		line);	/* in: line where created */
+/**********************************************************************
+Frees a block from a memory heap. */
+UNIV_INTERN
+void
+mem_heap_block_free(
+/*================*/
+	mem_heap_t*	heap,	/* in: heap */
+	mem_block_t*	block);	/* in: block to free */
+/**********************************************************************
+Frees the free_block field from a memory heap. */
+UNIV_INTERN
+void
+mem_heap_free_block_free(
+/*=====================*/
+	mem_heap_t*	heap);	/* in: heap */
+/*******************************************************************
+Adds a new block to a memory heap. */
+UNIV_INTERN
+mem_block_t*
+mem_heap_add_block(
+/*===============*/
+				/* out: created block, NULL if did not
+				succeed (only possible for
+				MEM_HEAP_BTR_SEARCH type heaps)*/
+	mem_heap_t*	heap,	/* in: memory heap */
+	ulint		n);	/* in: number of bytes user needs */
+
+UNIV_INLINE
+void
+mem_block_set_len(mem_block_t* block, ulint len)
+{
+	ut_ad(len > 0);
+
+	block->len = len;
+}
+
+UNIV_INLINE
+ulint
+mem_block_get_len(mem_block_t* block)
+{
+	return(block->len);
+}
+
+UNIV_INLINE
+void
+mem_block_set_type(mem_block_t* block, ulint type)
+{
+	ut_ad((type == MEM_HEAP_DYNAMIC) || (type == MEM_HEAP_BUFFER)
+	      || (type == MEM_HEAP_BUFFER + MEM_HEAP_BTR_SEARCH));
+
+	block->type = type;
+}
+
+UNIV_INLINE
+ulint
+mem_block_get_type(mem_block_t* block)
+{
+	return(block->type);
+}
+
+UNIV_INLINE
+void
+mem_block_set_free(mem_block_t* block, ulint free)
+{
+	ut_ad(free > 0);
+	ut_ad(free <= mem_block_get_len(block));
+
+	block->free = free;
+}
+
+UNIV_INLINE
+ulint
+mem_block_get_free(mem_block_t* block)
+{
+	return(block->free);
+}
+
+UNIV_INLINE
+void
+mem_block_set_start(mem_block_t* block, ulint start)
+{
+	ut_ad(start > 0);
+
+	block->start = start;
+}
+
+UNIV_INLINE
+ulint
+mem_block_get_start(mem_block_t* block)
+{
+	return(block->start);
+}
+
+/*******************************************************************
+Allocates and zero-fills n bytes of memory from a memory heap. */
+UNIV_INLINE
+void*
+mem_heap_zalloc(
+/*============*/
+				/* out: allocated, zero-filled storage */
+	mem_heap_t*	heap,	/* in: memory heap */
+	ulint		n)	/* in: number of bytes; if the heap is allowed
+				to grow into the buffer pool, this must be
+				<= MEM_MAX_ALLOC_IN_BUF */
+{
+	ut_ad(heap);
+	ut_ad(!(heap->type & MEM_HEAP_BTR_SEARCH));
+	return(memset(mem_heap_alloc(heap, n), 0, n));
+}
+
+/*******************************************************************
+Allocates n bytes of memory from a memory heap. */
+UNIV_INLINE
+void*
+mem_heap_alloc(
+/*===========*/
+				/* out: allocated storage, NULL if did not
+				succeed (only possible for
+				MEM_HEAP_BTR_SEARCH type heaps) */
+	mem_heap_t*	heap,	/* in: memory heap */
+	ulint		n)	/* in: number of bytes; if the heap is allowed
+				to grow into the buffer pool, this must be
+				<= MEM_MAX_ALLOC_IN_BUF */
+{
+	mem_block_t*	block;
+	void*		buf;
+	ulint		free;
+
+	ut_ad(mem_heap_check(heap));
+
+	block = UT_LIST_GET_LAST(heap->base);
+
+	ut_ad(!(block->type & MEM_HEAP_BUFFER) || (n <= MEM_MAX_ALLOC_IN_BUF));
+
+	/* Check if there is enough space in block. If not, create a new
+	block to the heap */
+
+	if (mem_block_get_len(block)
+	    < mem_block_get_free(block) + MEM_SPACE_NEEDED(n)) {
+
+		block = mem_heap_add_block(heap, n);
+
+		if (block == NULL) {
+
+			return(NULL);
+		}
+	}
+
+	free = mem_block_get_free(block);
+
+	buf = (byte*)block + free;
+
+	mem_block_set_free(block, free + MEM_SPACE_NEEDED(n));
+
+#ifdef UNIV_MEM_DEBUG
+	UNIV_MEM_ALLOC(buf,
+		       n + MEM_FIELD_HEADER_SIZE + MEM_FIELD_TRAILER_SIZE);
+
+	/* In the debug version write debugging info to the field */
+	mem_field_init((byte*)buf, n);
+
+	/* Advance buf to point at the storage which will be given to the
+	caller */
+	buf = (byte*)buf + MEM_FIELD_HEADER_SIZE;
+
+#endif
+#ifdef UNIV_SET_MEM_TO_ZERO
+	UNIV_MEM_ALLOC(buf, n);
+	memset(buf, '\0', n);
+#endif
+	UNIV_MEM_ALLOC(buf, n);
+	return(buf);
+}
+
+/*********************************************************************
+Returns a pointer to the heap top. */
+UNIV_INLINE
+byte*
+mem_heap_get_heap_top(
+/*==================*/
+				/* out: pointer to the heap top */
+	mem_heap_t*	heap)	/* in: memory heap */
+{
+	mem_block_t*	block;
+	byte*		buf;
+
+	ut_ad(mem_heap_check(heap));
+
+	block = UT_LIST_GET_LAST(heap->base);
+
+	buf = (byte*)block + mem_block_get_free(block);
+
+	return(buf);
+}
+
+/*********************************************************************
+Frees the space in a memory heap exceeding the pointer given. The
+pointer must have been acquired from mem_heap_get_heap_top. The first
+memory block of the heap is not freed. */
+UNIV_INLINE
+void
+mem_heap_free_heap_top(
+/*===================*/
+	mem_heap_t*	heap,	/* in: heap from which to free */
+	byte*		old_top)/* in: pointer to old top of heap */
+{
+	mem_block_t*	block;
+	mem_block_t*	prev_block;
+#ifdef UNIV_MEM_DEBUG
+	ibool		error;
+	ulint		total_size;
+	ulint		size;
+#endif
+
+	ut_ad(mem_heap_check(heap));
+
+#ifdef UNIV_MEM_DEBUG
+
+	/* Validate the heap and get its total allocated size */
+	mem_heap_validate_or_print(heap, NULL, FALSE, &error, &total_size,
+				   NULL, NULL);
+	ut_a(!error);
+
+	/* Get the size below top pointer */
+	mem_heap_validate_or_print(heap, old_top, FALSE, &error, &size, NULL,
+				   NULL);
+	ut_a(!error);
+
+#endif
+
+	block = UT_LIST_GET_LAST(heap->base);
+
+	while (block != NULL) {
+		if (((byte*)block + mem_block_get_free(block) >= old_top)
+		    && ((byte*)block <= old_top)) {
+			/* Found the right block */
+
+			break;
+		}
+
+		/* Store prev_block value before freeing the current block
+		(the current block will be erased in freeing) */
+
+		prev_block = UT_LIST_GET_PREV(list, block);
+
+		mem_heap_block_free(heap, block);
+
+		block = prev_block;
+	}
+
+	ut_ad(block);
+
+	/* Set the free field of block */
+	mem_block_set_free(block, old_top - (byte*)block);
+
+#ifdef UNIV_MEM_DEBUG
+	ut_ad(mem_block_get_start(block) <= mem_block_get_free(block));
+
+	/* In the debug version erase block from top up */
+	mem_erase_buf(old_top, (byte*)block + block->len - old_top);
+
+	/* Update allocated memory count */
+	mutex_enter(&mem_hash_mutex);
+	mem_current_allocated_memory -= (total_size - size);
+	mutex_exit(&mem_hash_mutex);
+#else /* UNIV_MEM_DEBUG */
+	UNIV_MEM_ASSERT_W(old_top, (byte*)block + block->len - old_top);
+#endif /* UNIV_MEM_DEBUG */
+	UNIV_MEM_ALLOC(old_top, (byte*)block + block->len - old_top);
+
+	/* If free == start, we may free the block if it is not the first
+	one */
+
+	if ((heap != block) && (mem_block_get_free(block)
+				== mem_block_get_start(block))) {
+		mem_heap_block_free(heap, block);
+	}
+}
+
+/*********************************************************************
+Empties a memory heap. The first memory block of the heap is not freed. */
+UNIV_INLINE
+void
+mem_heap_empty(
+/*===========*/
+	mem_heap_t*	heap)	/* in: heap to empty */
+{
+	mem_heap_free_heap_top(heap, (byte*)heap + mem_block_get_start(heap));
+
+	if (heap->free_block) {
+		mem_heap_free_block_free(heap);
+	}
+}
+
+/*********************************************************************
+Returns a pointer to the topmost element in a memory heap. The size of the
+element must be given. */
+UNIV_INLINE
+void*
+mem_heap_get_top(
+/*=============*/
+				/* out: pointer to the topmost element */
+	mem_heap_t*	heap,	/* in: memory heap */
+	ulint		n)	/* in: size of the topmost element */
+{
+	mem_block_t*	block;
+	void*		buf;
+
+	ut_ad(mem_heap_check(heap));
+
+	block = UT_LIST_GET_LAST(heap->base);
+
+	buf = (byte*)block + mem_block_get_free(block) - MEM_SPACE_NEEDED(n);
+
+#ifdef UNIV_MEM_DEBUG
+	ut_ad(mem_block_get_start(block) <=(ulint)((byte*)buf - (byte*)block));
+
+	/* In the debug version, advance buf to point at the storage which
+	was given to the caller in the allocation*/
+
+	buf = (byte*)buf + MEM_FIELD_HEADER_SIZE;
+
+	/* Check that the field lengths agree */
+	ut_ad(n == (ulint)mem_field_header_get_len(buf));
+#endif
+
+	return(buf);
+}
+
+/*********************************************************************
+Frees the topmost element in a memory heap. The size of the element must be
+given. */
+UNIV_INLINE
+void
+mem_heap_free_top(
+/*==============*/
+	mem_heap_t*	heap,	/* in: memory heap */
+	ulint		n)	/* in: size of the topmost element */
+{
+	mem_block_t*	block;
+
+	ut_ad(mem_heap_check(heap));
+
+	block = UT_LIST_GET_LAST(heap->base);
+
+	/* Subtract the free field of block */
+	mem_block_set_free(block, mem_block_get_free(block)
+			   - MEM_SPACE_NEEDED(n));
+	UNIV_MEM_ASSERT_W((byte*) block + mem_block_get_free(block), n);
+#ifdef UNIV_MEM_DEBUG
+
+	ut_ad(mem_block_get_start(block) <= mem_block_get_free(block));
+
+	/* In the debug version check the consistency, and erase field */
+	mem_field_erase((byte*)block + mem_block_get_free(block), n);
+#endif
+
+	/* If free == start, we may free the block if it is not the first
+	one */
+
+	if ((heap != block) && (mem_block_get_free(block)
+				== mem_block_get_start(block))) {
+		mem_heap_block_free(heap, block);
+	} else {
+		/* Avoid a bogus UNIV_MEM_ASSERT_W() warning in a
+		subsequent invocation of mem_heap_free_top().
+		Originally, this was UNIV_MEM_FREE(), to catch writes
+		to freed memory. */
+		UNIV_MEM_ALLOC((byte*) block + mem_block_get_free(block), n);
+	}
+}
+
+/*********************************************************************
+NOTE: Use the corresponding macros instead of this function. Creates a
+memory heap. For debugging purposes, takes also the file name and line as
+argument. */
+UNIV_INLINE
+mem_heap_t*
+mem_heap_create_func(
+/*=================*/
+					/* out, own: memory heap, NULL if
+					did not succeed (only possible for
+					MEM_HEAP_BTR_SEARCH type heaps)*/
+	ulint		n,		/* in: desired start block size,
+					this means that a single user buffer
+					of size n will fit in the block,
+					0 creates a default size block */
+	ulint		type,		/* in: heap type */
+	const char*	file_name,	/* in: file name where created */
+	ulint		line)		/* in: line where created */
+{
+	mem_block_t*   block;
+
+	if (!n) {
+		n = MEM_BLOCK_START_SIZE;
+	}
+
+	block = mem_heap_create_block(NULL, n, type, file_name, line);
+
+	if (block == NULL) {
+
+		return(NULL);
+	}
+
+	UT_LIST_INIT(block->base);
+
+	/* Add the created block itself as the first block in the list */
+	UT_LIST_ADD_FIRST(list, block->base, block);
+
+#ifdef UNIV_MEM_DEBUG
+
+	mem_hash_insert(block, file_name, line);
+
+#endif
+
+	return(block);
+}
+
+/*********************************************************************
+NOTE: Use the corresponding macro instead of this function. Frees the space
+occupied by a memory heap. In the debug version erases the heap memory
+blocks. */
+UNIV_INLINE
+void
+mem_heap_free_func(
+/*===============*/
+	mem_heap_t*	heap,		/* in, own: heap to be freed */
+	const char*	file_name __attribute__((unused)),
+					/* in: file name where freed */
+	ulint		line  __attribute__((unused)))
+{
+	mem_block_t*	block;
+	mem_block_t*	prev_block;
+
+	ut_ad(mem_heap_check(heap));
+
+	block = UT_LIST_GET_LAST(heap->base);
+
+#ifdef UNIV_MEM_DEBUG
+
+	/* In the debug version remove the heap from the hash table of heaps
+	and check its consistency */
+
+	mem_hash_remove(heap, file_name, line);
+
+#endif
+
+	if (heap->free_block) {
+		mem_heap_free_block_free(heap);
+	}
+
+	while (block != NULL) {
+		/* Store the contents of info before freeing current block
+		(it is erased in freeing) */
+
+		prev_block = UT_LIST_GET_PREV(list, block);
+
+		mem_heap_block_free(heap, block);
+
+		block = prev_block;
+	}
+}
+
+/*******************************************************************
+NOTE: Use the corresponding macro instead of this function.
+Allocates a single buffer of memory from the dynamic memory of
+the C compiler. Is like malloc of C. The buffer must be freed
+with mem_free. */
+UNIV_INLINE
+void*
+mem_alloc_func(
+/*===========*/
+					/* out, own: free storage */
+	ulint		n,		/* in: desired number of bytes */
+	ulint*		size,		/* out: allocated size in bytes,
+					or NULL */
+	const char*	file_name,	/* in: file name where created */
+	ulint		line)		/* in: line where created */
+{
+	mem_heap_t*	heap;
+	void*		buf;
+
+	heap = mem_heap_create_func(n, MEM_HEAP_DYNAMIC, file_name, line);
+
+	/* Note that as we created the first block in the heap big enough
+	for the buffer requested by the caller, the buffer will be in the
+	first block and thus we can calculate the pointer to the heap from
+	the pointer to the buffer when we free the memory buffer. */
+
+	if (UNIV_LIKELY_NULL(size)) {
+		/* Adjust the allocation to the actual size of the
+		memory block. */
+		ulint	m = mem_block_get_len(heap)
+			- mem_block_get_free(heap);
+#ifdef UNIV_MEM_DEBUG
+		m -= MEM_FIELD_HEADER_SIZE + MEM_FIELD_TRAILER_SIZE;
+#endif /* UNIV_MEM_DEBUG */
+		ut_ad(m >= n);
+		*size = n = m;
+	}
+
+	buf = mem_heap_alloc(heap, n);
+
+	ut_a((byte*)heap == (byte*)buf - MEM_BLOCK_HEADER_SIZE
+	     - MEM_FIELD_HEADER_SIZE);
+	return(buf);
+}
+
+/*******************************************************************
+NOTE: Use the corresponding macro instead of this function. Frees a single
+buffer of storage from the dynamic memory of the C compiler. Similar to the
+free of C. */
+UNIV_INLINE
+void
+mem_free_func(
+/*==========*/
+	void*		ptr,		/* in, own: buffer to be freed */
+	const char*	file_name,	/* in: file name where created */
+	ulint		line		/* in: line where created */
+	)
+{
+	mem_heap_t*   heap;
+
+	heap = (mem_heap_t*)((byte*)ptr - MEM_BLOCK_HEADER_SIZE
+			     - MEM_FIELD_HEADER_SIZE);
+	mem_heap_free_func(heap, file_name, line);
+}
+
+/*********************************************************************
+Returns the space in bytes occupied by a memory heap. */
+UNIV_INLINE
+ulint
+mem_heap_get_size(
+/*==============*/
+	mem_heap_t*	heap)	/* in: heap */
+{
+	mem_block_t*	block;
+	ulint		size	= 0;
+
+	ut_ad(mem_heap_check(heap));
+
+	block = heap;
+
+	while (block != NULL) {
+
+		size += mem_block_get_len(block);
+		block = UT_LIST_GET_NEXT(list, block);
+	}
+
+	if (heap->free_block) {
+		size += UNIV_PAGE_SIZE;
+	}
+
+	return(size);
+}
+
+/**************************************************************************
+Duplicates a NUL-terminated string. */
+UNIV_INLINE
+char*
+mem_strdup(
+/*=======*/
+				/* out, own: a copy of the string,
+				must be deallocated with mem_free */
+	const char*	str)	/* in: string to be copied */
+{
+	ulint	len = strlen(str) + 1;
+	return((char*) memcpy(mem_alloc(len), str, len));
+}
+
+/**************************************************************************
+Makes a NUL-terminated copy of a nonterminated string. */
+UNIV_INLINE
+char*
+mem_strdupl(
+/*========*/
+				/* out, own: a copy of the string,
+				must be deallocated with mem_free */
+	const char*	str,	/* in: string to be copied */
+	ulint		len)	/* in: length of str, in bytes */
+{
+	char*	s = (char*) mem_alloc(len + 1);
+	s[len] = 0;
+	return((char*) memcpy(s, str, len));
+}
+
+/**************************************************************************
+Makes a NUL-terminated copy of a nonterminated string,
+allocated from a memory heap. */
+UNIV_INLINE
+char*
+mem_heap_strdupl(
+/*=============*/
+				/* out, own: a copy of the string */
+	mem_heap_t*	heap,	/* in: memory heap where string is allocated */
+	const char*	str,	/* in: string to be copied */
+	ulint		len)	/* in: length of str, in bytes */
+{
+	char*	s = (char*) mem_heap_alloc(heap, len + 1);
+	s[len] = 0;
+	return((char*) memcpy(s, str, len));
+}
diff --git a/storage/xtradb/include/mem0pool.h b/storage/xtradb/include/mem0pool.h
new file mode 100644
index 00000000000..7e51b07bfe0
--- /dev/null
+++ b/storage/xtradb/include/mem0pool.h
@@ -0,0 +1,126 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The lowest-level memory management
+
+Created 6/9/1994 Heikki Tuuri
+*******************************************************/
+
+#ifndef mem0pool_h
+#define mem0pool_h
+
+#include "univ.i"
+#include "os0file.h"
+#include "ut0lst.h"
+
+typedef struct mem_area_struct	mem_area_t;
+typedef struct mem_pool_struct	mem_pool_t;
+
+/* The common memory pool */
+extern mem_pool_t*	mem_comm_pool;
+
+/* Memory area header */
+
+struct mem_area_struct{
+	ulint		size_and_free;	/* memory area size is obtained by
+					anding with ~MEM_AREA_FREE; area in
+					a free list if ANDing with
+					MEM_AREA_FREE results in nonzero */
+	UT_LIST_NODE_T(mem_area_t)
+			free_list;	/* free list node */
+};
+
+/* Each memory area takes this many extra bytes for control information */
+#define MEM_AREA_EXTRA_SIZE	(ut_calc_align(sizeof(struct mem_area_struct),\
+			UNIV_MEM_ALIGNMENT))
+
+/************************************************************************
+Creates a memory pool. */
+UNIV_INTERN
+mem_pool_t*
+mem_pool_create(
+/*============*/
+			/* out: memory pool */
+	ulint	size);	/* in: pool size in bytes */
+/************************************************************************
+Allocates memory from a pool. NOTE: This low-level function should only be
+used in mem0mem.*! */
+UNIV_INTERN
+void*
+mem_area_alloc(
+/*===========*/
+				/* out, own: allocated memory buffer */
+	ulint*		psize,	/* in: requested size in bytes; for optimum
+				space usage, the size should be a power of 2
+				minus MEM_AREA_EXTRA_SIZE;
+				out: allocated size in bytes (greater than
+				or equal to the requested size) */
+	mem_pool_t*	pool);	/* in: memory pool */
+/************************************************************************
+Frees memory to a pool. */
+UNIV_INTERN
+void
+mem_area_free(
+/*==========*/
+	void*		ptr,	/* in, own: pointer to allocated memory
+				buffer */
+	mem_pool_t*	pool);	/* in: memory pool */
+/************************************************************************
+Returns the amount of reserved memory. */
+UNIV_INTERN
+ulint
+mem_pool_get_reserved(
+/*==================*/
+				/* out: reserved mmeory in bytes */
+	mem_pool_t*	pool);	/* in: memory pool */
+/************************************************************************
+Reserves the mem pool mutex. */
+UNIV_INTERN
+void
+mem_pool_mutex_enter(void);
+/*======================*/
+/************************************************************************
+Releases the mem pool mutex. */
+UNIV_INTERN
+void
+mem_pool_mutex_exit(void);
+/*=====================*/
+/************************************************************************
+Validates a memory pool. */
+UNIV_INTERN
+ibool
+mem_pool_validate(
+/*==============*/
+				/* out: TRUE if ok */
+	mem_pool_t*	pool);	/* in: memory pool */
+/************************************************************************
+Prints info of a memory pool. */
+UNIV_INTERN
+void
+mem_pool_print_info(
+/*================*/
+	FILE*		outfile,/* in: output file to write to */
+	mem_pool_t*	pool);	/* in: memory pool */
+
+
+#ifndef UNIV_NONINL
+#include "mem0pool.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/mem0pool.ic b/storage/xtradb/include/mem0pool.ic
new file mode 100644
index 00000000000..4cc65e754ce
--- /dev/null
+++ b/storage/xtradb/include/mem0pool.ic
@@ -0,0 +1,23 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/************************************************************************
+The lowest-level memory management
+
+Created 6/8/1994 Heikki Tuuri
+*************************************************************************/
diff --git a/storage/xtradb/include/mtr0log.h b/storage/xtradb/include/mtr0log.h
new file mode 100644
index 00000000000..44374cdf1a4
--- /dev/null
+++ b/storage/xtradb/include/mtr0log.h
@@ -0,0 +1,247 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Mini-transaction logging routines
+
+Created 12/7/1995 Heikki Tuuri
+*******************************************************/
+
+#ifndef mtr0log_h
+#define mtr0log_h
+
+#include "univ.i"
+#include "mtr0mtr.h"
+#include "dict0types.h"
+
+/************************************************************
+Writes 1 - 4 bytes to a file page buffered in the buffer pool.
+Writes the corresponding log record to the mini-transaction log. */
+UNIV_INTERN
+void
+mlog_write_ulint(
+/*=============*/
+	byte*	ptr,	/* in: pointer where to write */
+	ulint	val,	/* in: value to write */
+	byte	type,	/* in: MLOG_1BYTE, MLOG_2BYTES, MLOG_4BYTES */
+	mtr_t*	mtr);	/* in: mini-transaction handle */
+/************************************************************
+Writes 8 bytes to a file page buffered in the buffer pool.
+Writes the corresponding log record to the mini-transaction log. */
+UNIV_INTERN
+void
+mlog_write_dulint(
+/*==============*/
+	byte*	ptr,	/* in: pointer where to write */
+	dulint	val,	/* in: value to write */
+	mtr_t*	mtr);	/* in: mini-transaction handle */
+/************************************************************
+Writes a string to a file page buffered in the buffer pool. Writes the
+corresponding log record to the mini-transaction log. */
+UNIV_INTERN
+void
+mlog_write_string(
+/*==============*/
+	byte*		ptr,	/* in: pointer where to write */
+	const byte*	str,	/* in: string to write */
+	ulint		len,	/* in: string length */
+	mtr_t*		mtr);	/* in: mini-transaction handle */
+/************************************************************
+Logs a write of a string to a file page buffered in the buffer pool.
+Writes the corresponding log record to the mini-transaction log. */
+UNIV_INTERN
+void
+mlog_log_string(
+/*============*/
+	byte*	ptr,	/* in: pointer written to */
+	ulint	len,	/* in: string length */
+	mtr_t*	mtr);	/* in: mini-transaction handle */
+/************************************************************
+Writes initial part of a log record consisting of one-byte item
+type and four-byte space and page numbers. */
+UNIV_INTERN
+void
+mlog_write_initial_log_record(
+/*==========================*/
+	const byte*	ptr,	/* in: pointer to (inside) a buffer
+				frame holding the file page where
+				modification is made */
+	byte		type,	/* in: log item type: MLOG_1BYTE, ... */
+	mtr_t*		mtr);	/* in: mini-transaction handle */
+/************************************************************
+Writes a log record about an .ibd file create/delete/rename. */
+UNIV_INLINE
+byte*
+mlog_write_initial_log_record_for_file_op(
+/*======================================*/
+			/* out: new value of log_ptr */
+	ulint	type,	/* in: MLOG_FILE_CREATE, MLOG_FILE_DELETE, or
+			MLOG_FILE_RENAME */
+	ulint	space_id,/* in: space id, if applicable */
+	ulint	page_no,/* in: page number (not relevant currently) */
+	byte*	log_ptr,/* in: pointer to mtr log which has been opened */
+	mtr_t*	mtr);	/* in: mtr */
+/************************************************************
+Catenates 1 - 4 bytes to the mtr log. */
+UNIV_INLINE
+void
+mlog_catenate_ulint(
+/*================*/
+	mtr_t*	mtr,	/* in: mtr */
+	ulint	val,	/* in: value to write */
+	ulint	type);	/* in: MLOG_1BYTE, MLOG_2BYTES, MLOG_4BYTES */
+/************************************************************
+Catenates n bytes to the mtr log. */
+UNIV_INTERN
+void
+mlog_catenate_string(
+/*=================*/
+	mtr_t*		mtr,	/* in: mtr */
+	const byte*	str,	/* in: string to write */
+	ulint		len);	/* in: string length */
+/************************************************************
+Catenates a compressed ulint to mlog. */
+UNIV_INLINE
+void
+mlog_catenate_ulint_compressed(
+/*===========================*/
+	mtr_t*	mtr,	/* in: mtr */
+	ulint	val);	/* in: value to write */
+/************************************************************
+Catenates a compressed dulint to mlog. */
+UNIV_INLINE
+void
+mlog_catenate_dulint_compressed(
+/*============================*/
+	mtr_t*	mtr,	/* in: mtr */
+	dulint	val);	/* in: value to write */
+/************************************************************
+Opens a buffer to mlog. It must be closed with mlog_close. */
+UNIV_INLINE
+byte*
+mlog_open(
+/*======*/
+			/* out: buffer, NULL if log mode MTR_LOG_NONE */
+	mtr_t*	mtr,	/* in: mtr */
+	ulint	size);	/* in: buffer size in bytes; MUST be
+			smaller than DYN_ARRAY_DATA_SIZE! */
+/************************************************************
+Closes a buffer opened to mlog. */
+UNIV_INLINE
+void
+mlog_close(
+/*=======*/
+	mtr_t*	mtr,	/* in: mtr */
+	byte*	ptr);	/* in: buffer space from ptr up was not used */
+/************************************************************
+Writes the initial part of a log record (3..11 bytes).
+If the implementation of this function is changed, all
+size parameters to mlog_open() should be adjusted accordingly! */
+UNIV_INLINE
+byte*
+mlog_write_initial_log_record_fast(
+/*===============================*/
+				/* out: new value of log_ptr */
+	const byte*	ptr,	/* in: pointer to (inside) a buffer
+				frame holding the file page where
+				modification is made */
+	byte		type,	/* in: log item type: MLOG_1BYTE, ... */
+	byte*		log_ptr,/* in: pointer to mtr log which has
+				been opened */
+	mtr_t*		mtr);	/* in: mtr */
+/************************************************************
+Parses an initial log record written by mlog_write_initial_log_record. */
+UNIV_INTERN
+byte*
+mlog_parse_initial_log_record(
+/*==========================*/
+			/* out: parsed record end, NULL if not a complete
+			record */
+	byte*	ptr,	/* in: buffer */
+	byte*	end_ptr,/* in: buffer end */
+	byte*	type,	/* out: log record type: MLOG_1BYTE, ... */
+	ulint*	space,	/* out: space id */
+	ulint*	page_no);/* out: page number */
+/************************************************************
+Parses a log record written by mlog_write_ulint or mlog_write_dulint. */
+UNIV_INTERN
+byte*
+mlog_parse_nbytes(
+/*==============*/
+			/* out: parsed record end, NULL if not a complete
+			record */
+	ulint	type,	/* in: log record type: MLOG_1BYTE, ... */
+	byte*	ptr,	/* in: buffer */
+	byte*	end_ptr,/* in: buffer end */
+	byte*	page,	/* in: page where to apply the log record, or NULL */
+	void*	page_zip);/* in/out: compressed page, or NULL */
+/************************************************************
+Parses a log record written by mlog_write_string. */
+UNIV_INTERN
+byte*
+mlog_parse_string(
+/*==============*/
+			/* out: parsed record end, NULL if not a complete
+			record */
+	byte*	ptr,	/* in: buffer */
+	byte*	end_ptr,/* in: buffer end */
+	byte*	page,	/* in: page where to apply the log record, or NULL */
+	void*	page_zip);/* in/out: compressed page, or NULL */
+
+
+/************************************************************
+Opens a buffer for mlog, writes the initial log record and,
+if needed, the field lengths of an index.  Reserves space
+for further log entries.  The log entry must be closed with
+mtr_close(). */
+UNIV_INTERN
+byte*
+mlog_open_and_write_index(
+/*======================*/
+				/* out: buffer, NULL if log mode
+				MTR_LOG_NONE */
+	mtr_t*		mtr,	/* in: mtr */
+	byte*		rec,	/* in: index record or page */
+	dict_index_t*	index,	/* in: record descriptor */
+	byte		type,	/* in: log item type */
+	ulint		size);	/* in: requested buffer size in bytes
+				(if 0, calls mlog_close() and returns NULL) */
+
+/************************************************************
+Parses a log record written by mlog_open_and_write_index. */
+UNIV_INTERN
+byte*
+mlog_parse_index(
+/*=============*/
+				/* out: parsed record end,
+				NULL if not a complete record */
+	byte*		ptr,	/* in: buffer */
+	byte*		end_ptr,/* in: buffer end */
+				/* out: new value of log_ptr */
+	ibool		comp,	/* in: TRUE=compact record format */
+	dict_index_t**	index);	/* out, own: dummy index */
+
+/* Insert, update, and maybe other functions may use this value to define an
+extra mlog buffer size for variable size data */
+#define MLOG_BUF_MARGIN	256
+
+#ifndef UNIV_NONINL
+#include "mtr0log.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/mtr0log.ic b/storage/xtradb/include/mtr0log.ic
new file mode 100644
index 00000000000..5f05befb9cc
--- /dev/null
+++ b/storage/xtradb/include/mtr0log.ic
@@ -0,0 +1,247 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Mini-transaction logging routines
+
+Created 12/7/1995 Heikki Tuuri
+*******************************************************/
+
+#include "mach0data.h"
+#include "ut0lst.h"
+#include "buf0buf.h"
+
+/************************************************************
+Opens a buffer to mlog. It must be closed with mlog_close. */
+UNIV_INLINE
+byte*
+mlog_open(
+/*======*/
+			/* out: buffer, NULL if log mode MTR_LOG_NONE */
+	mtr_t*	mtr,	/* in: mtr */
+	ulint	size)	/* in: buffer size in bytes; MUST be
+			smaller than DYN_ARRAY_DATA_SIZE! */
+{
+	dyn_array_t*	mlog;
+
+	mtr->modifications = TRUE;
+
+	if (mtr_get_log_mode(mtr) == MTR_LOG_NONE) {
+
+		return(NULL);
+	}
+
+	mlog = &(mtr->log);
+
+	return(dyn_array_open(mlog, size));
+}
+
+/************************************************************
+Closes a buffer opened to mlog. */
+UNIV_INLINE
+void
+mlog_close(
+/*=======*/
+	mtr_t*	mtr,	/* in: mtr */
+	byte*	ptr)	/* in: buffer space from ptr up was not used */
+{
+	dyn_array_t*	mlog;
+
+	ut_ad(mtr_get_log_mode(mtr) != MTR_LOG_NONE);
+
+	mlog = &(mtr->log);
+
+	dyn_array_close(mlog, ptr);
+}
+
+/************************************************************
+Catenates 1 - 4 bytes to the mtr log. The value is not compressed. */
+UNIV_INLINE
+void
+mlog_catenate_ulint(
+/*================*/
+	mtr_t*	mtr,	/* in: mtr */
+	ulint	val,	/* in: value to write */
+	ulint	type)	/* in: MLOG_1BYTE, MLOG_2BYTES, MLOG_4BYTES */
+{
+	dyn_array_t*	mlog;
+	byte*		ptr;
+
+	if (mtr_get_log_mode(mtr) == MTR_LOG_NONE) {
+
+		return;
+	}
+
+	mlog = &(mtr->log);
+
+#if MLOG_1BYTE != 1
+# error "MLOG_1BYTE != 1"
+#endif
+#if MLOG_2BYTES != 2
+# error "MLOG_2BYTES != 2"
+#endif
+#if MLOG_4BYTES != 4
+# error "MLOG_4BYTES != 4"
+#endif
+#if MLOG_8BYTES != 8
+# error "MLOG_8BYTES != 8"
+#endif
+	ptr = (byte*) dyn_array_push(mlog, type);
+
+	if (type == MLOG_4BYTES) {
+		mach_write_to_4(ptr, val);
+	} else if (type == MLOG_2BYTES) {
+		mach_write_to_2(ptr, val);
+	} else {
+		ut_ad(type == MLOG_1BYTE);
+		mach_write_to_1(ptr, val);
+	}
+}
+
+/************************************************************
+Catenates a compressed ulint to mlog. */
+UNIV_INLINE
+void
+mlog_catenate_ulint_compressed(
+/*===========================*/
+	mtr_t*	mtr,	/* in: mtr */
+	ulint	val)	/* in: value to write */
+{
+	byte*	log_ptr;
+
+	log_ptr = mlog_open(mtr, 10);
+
+	/* If no logging is requested, we may return now */
+	if (log_ptr == NULL) {
+
+		return;
+	}
+
+	log_ptr += mach_write_compressed(log_ptr, val);
+
+	mlog_close(mtr, log_ptr);
+}
+
+/************************************************************
+Catenates a compressed dulint to mlog. */
+UNIV_INLINE
+void
+mlog_catenate_dulint_compressed(
+/*============================*/
+	mtr_t*	mtr,	/* in: mtr */
+	dulint	val)	/* in: value to write */
+{
+	byte*	log_ptr;
+
+	log_ptr = mlog_open(mtr, 15);
+
+	/* If no logging is requested, we may return now */
+	if (log_ptr == NULL) {
+
+		return;
+	}
+
+	log_ptr += mach_dulint_write_compressed(log_ptr, val);
+
+	mlog_close(mtr, log_ptr);
+}
+
+/************************************************************
+Writes the initial part of a log record (3..11 bytes).
+If the implementation of this function is changed, all
+size parameters to mlog_open() should be adjusted accordingly! */
+UNIV_INLINE
+byte*
+mlog_write_initial_log_record_fast(
+/*===============================*/
+				/* out: new value of log_ptr */
+	const byte*	ptr,	/* in: pointer to (inside) a buffer
+				frame holding the file page where
+				modification is made */
+	byte		type,	/* in: log item type: MLOG_1BYTE, ... */
+	byte*		log_ptr,/* in: pointer to mtr log which has
+				been opened */
+	mtr_t*		mtr)	/* in: mtr */
+{
+#ifdef UNIV_DEBUG
+	buf_block_t*	block;
+#endif
+	const byte*	page;
+	ulint		space;
+	ulint		offset;
+
+	ut_ad(mtr_memo_contains_page(mtr, ptr, MTR_MEMO_PAGE_X_FIX));
+	ut_ad(type <= MLOG_BIGGEST_TYPE);
+	ut_ad(ptr && log_ptr);
+
+	page = (const byte*) ut_align_down(ptr, UNIV_PAGE_SIZE);
+	space = mach_read_from_4(page + FIL_PAGE_ARCH_LOG_NO_OR_SPACE_ID);
+	offset = mach_read_from_4(page + FIL_PAGE_OFFSET);
+
+	mach_write_to_1(log_ptr, type);
+	log_ptr++;
+	log_ptr += mach_write_compressed(log_ptr, space);
+	log_ptr += mach_write_compressed(log_ptr, offset);
+
+	mtr->n_log_recs++;
+
+#ifdef UNIV_LOG_DEBUG
+	fprintf(stderr,
+		"Adding to mtr log record type %lu space %lu page no %lu\n",
+		(ulong) type, space, offset);
+#endif
+
+#ifdef UNIV_DEBUG
+	/* We now assume that all x-latched pages have been modified! */
+	block = (buf_block_t*) buf_block_align(ptr);
+
+	if (!mtr_memo_contains(mtr, block, MTR_MEMO_MODIFY)) {
+
+		mtr_memo_push(mtr, block, MTR_MEMO_MODIFY);
+	}
+#endif
+	return(log_ptr);
+}
+
+/************************************************************
+Writes a log record about an .ibd file create/delete/rename. */
+UNIV_INLINE
+byte*
+mlog_write_initial_log_record_for_file_op(
+/*======================================*/
+			/* out: new value of log_ptr */
+	ulint	type,	/* in: MLOG_FILE_CREATE, MLOG_FILE_DELETE, or
+			MLOG_FILE_RENAME */
+	ulint	space_id,/* in: space id, if applicable */
+	ulint	page_no,/* in: page number (not relevant currently) */
+	byte*	log_ptr,/* in: pointer to mtr log which has been opened */
+	mtr_t*	mtr)	/* in: mtr */
+{
+	ut_ad(log_ptr);
+
+	mach_write_to_1(log_ptr, type);
+	log_ptr++;
+
+	/* We write dummy space id and page number */
+	log_ptr += mach_write_compressed(log_ptr, space_id);
+	log_ptr += mach_write_compressed(log_ptr, page_no);
+
+	mtr->n_log_recs++;
+
+	return(log_ptr);
+}
diff --git a/storage/xtradb/include/mtr0mtr.h b/storage/xtradb/include/mtr0mtr.h
new file mode 100644
index 00000000000..a29f6c73141
--- /dev/null
+++ b/storage/xtradb/include/mtr0mtr.h
@@ -0,0 +1,380 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Mini-transaction buffer
+
+Created 11/26/1995 Heikki Tuuri
+*******************************************************/
+
+#ifndef mtr0mtr_h
+#define mtr0mtr_h
+
+#include "univ.i"
+#include "mem0mem.h"
+#include "dyn0dyn.h"
+#include "buf0types.h"
+#include "sync0rw.h"
+#include "ut0byte.h"
+#include "mtr0types.h"
+#include "page0types.h"
+
+/* Logging modes for a mini-transaction */
+#define MTR_LOG_ALL		21	/* default mode: log all operations
+					modifying disk-based data */
+#define	MTR_LOG_NONE		22	/* log no operations */
+/*#define	MTR_LOG_SPACE	23 */	/* log only operations modifying
+					file space page allocation data
+					(operations in fsp0fsp.* ) */
+#define	MTR_LOG_SHORT_INSERTS	24	/* inserts are logged in a shorter
+					form */
+
+/* Types for the mlock objects to store in the mtr memo; NOTE that the
+first 3 values must be RW_S_LATCH, RW_X_LATCH, RW_NO_LATCH */
+#define	MTR_MEMO_PAGE_S_FIX	RW_S_LATCH
+#define	MTR_MEMO_PAGE_X_FIX	RW_X_LATCH
+#define	MTR_MEMO_BUF_FIX	RW_NO_LATCH
+#define MTR_MEMO_MODIFY		54
+#define	MTR_MEMO_S_LOCK		55
+#define	MTR_MEMO_X_LOCK		56
+
+/* Log item types: we have made them to be of the type 'byte'
+for the compiler to warn if val and type parameters are switched
+in a call to mlog_write_ulint. NOTE! For 1 - 8 bytes, the
+flag value must give the length also! */
+#define	MLOG_SINGLE_REC_FLAG	128		/* if the mtr contains only
+						one log record for one page,
+						i.e., write_initial_log_record
+						has been called only once,
+						this flag is ORed to the type
+						of that first log record */
+#define	MLOG_1BYTE		(1)		/* one byte is written */
+#define	MLOG_2BYTES		(2)		/* 2 bytes ... */
+#define	MLOG_4BYTES		(4)		/* 4 bytes ... */
+#define	MLOG_8BYTES		(8)		/* 8 bytes ... */
+#define	MLOG_REC_INSERT		((byte)9)	/* record insert */
+#define	MLOG_REC_CLUST_DELETE_MARK ((byte)10)	/* mark clustered index record
+						deleted */
+#define	MLOG_REC_SEC_DELETE_MARK ((byte)11)	/* mark secondary index record
+						deleted */
+#define MLOG_REC_UPDATE_IN_PLACE ((byte)13)	/* update of a record,
+						preserves record field sizes */
+#define MLOG_REC_DELETE		((byte)14)	/* delete a record from a
+						page */
+#define	MLOG_LIST_END_DELETE	((byte)15)	/* delete record list end on
+						index page */
+#define	MLOG_LIST_START_DELETE	((byte)16)	/* delete record list start on
+						index page */
+#define	MLOG_LIST_END_COPY_CREATED ((byte)17)	/* copy record list end to a
+						new created index page */
+#define	MLOG_PAGE_REORGANIZE	((byte)18)	/* reorganize an index page */
+#define MLOG_PAGE_CREATE	((byte)19)	/* create an index page */
+#define	MLOG_UNDO_INSERT	((byte)20)	/* insert entry in an undo
+						log */
+#define MLOG_UNDO_ERASE_END	((byte)21)	/* erase an undo log
+						page end */
+#define	MLOG_UNDO_INIT		((byte)22)	/* initialize a page in an
+						undo log */
+#define MLOG_UNDO_HDR_DISCARD	((byte)23)	/* discard an update undo log
+						header */
+#define	MLOG_UNDO_HDR_REUSE	((byte)24)	/* reuse an insert undo log
+						header */
+#define MLOG_UNDO_HDR_CREATE	((byte)25)	/* create an undo log header */
+#define MLOG_REC_MIN_MARK	((byte)26)	/* mark an index record as the
+						predefined minimum record */
+#define MLOG_IBUF_BITMAP_INIT	((byte)27)	/* initialize an ibuf bitmap
+						page */
+/*#define	MLOG_FULL_PAGE	((byte)28)	full contents of a page */
+#define MLOG_INIT_FILE_PAGE	((byte)29)	/* this means that a file page
+						is taken into use and the prior
+						contents of the page should be
+						ignored: in recovery we must
+						not trust the lsn values stored
+						to the file page */
+#define MLOG_WRITE_STRING	((byte)30)	/* write a string to a page */
+#define	MLOG_MULTI_REC_END	((byte)31)	/* if a single mtr writes
+						log records for several pages,
+						this log record ends the
+						sequence of these records */
+#define MLOG_DUMMY_RECORD	((byte)32)	/* dummy log record used to
+						pad a log block full */
+#define MLOG_FILE_CREATE	((byte)33)	/* log record about an .ibd
+						file creation */
+#define MLOG_FILE_RENAME	((byte)34)	/* log record about an .ibd
+						file rename */
+#define MLOG_FILE_DELETE	((byte)35)	/* log record about an .ibd
+						file deletion */
+#define MLOG_COMP_REC_MIN_MARK	((byte)36)	/* mark a compact index record
+						as the predefined minimum
+						record */
+#define MLOG_COMP_PAGE_CREATE	((byte)37)	/* create a compact
+						index page */
+#define MLOG_COMP_REC_INSERT	((byte)38)	/* compact record insert */
+#define MLOG_COMP_REC_CLUST_DELETE_MARK ((byte)39)
+						/* mark compact clustered index
+						record deleted */
+#define MLOG_COMP_REC_SEC_DELETE_MARK ((byte)40)/* mark compact secondary index
+						record deleted; this log
+						record type is redundant, as
+						MLOG_REC_SEC_DELETE_MARK is
+						independent of the record
+						format. */
+#define MLOG_COMP_REC_UPDATE_IN_PLACE ((byte)41)/* update of a compact record,
+						preserves record field sizes */
+#define MLOG_COMP_REC_DELETE	((byte)42)	/* delete a compact record
+						from a page */
+#define MLOG_COMP_LIST_END_DELETE ((byte)43)	/* delete compact record list
+						end on index page */
+#define MLOG_COMP_LIST_START_DELETE ((byte)44)	/* delete compact record list
+						start on index page */
+#define MLOG_COMP_LIST_END_COPY_CREATED ((byte)45)
+						/* copy compact record list end
+						to a new created index page */
+#define MLOG_COMP_PAGE_REORGANIZE ((byte)46)	/* reorganize an index page */
+#define MLOG_FILE_CREATE2	((byte)47)	/* log record about creating
+						an .ibd file, with format */
+#define MLOG_ZIP_WRITE_NODE_PTR	((byte)48)	/* write the node pointer of
+						a record on a compressed
+						non-leaf B-tree page */
+#define MLOG_ZIP_WRITE_BLOB_PTR	((byte)49)	/* write the BLOB pointer
+						of an externally stored column
+						on a compressed page */
+#define MLOG_ZIP_WRITE_HEADER	((byte)50)	/* write to compressed page
+						header */
+#define MLOG_ZIP_PAGE_COMPRESS	((byte)51)	/* compress an index page */
+#define MLOG_BIGGEST_TYPE	((byte)51)	/* biggest value (used in
+						asserts) */
+
+/*******************************************************************
+Starts a mini-transaction and creates a mini-transaction handle
+and buffer in the memory buffer given by the caller. */
+UNIV_INLINE
+mtr_t*
+mtr_start(
+/*======*/
+			/* out: mtr buffer which also acts as
+			the mtr handle */
+	mtr_t*	mtr);	/* in: memory buffer for the mtr buffer */
+/*******************************************************************
+Commits a mini-transaction. */
+UNIV_INTERN
+void
+mtr_commit(
+/*=======*/
+	mtr_t*	mtr);	/* in: mini-transaction */
+/**************************************************************
+Sets and returns a savepoint in mtr. */
+UNIV_INLINE
+ulint
+mtr_set_savepoint(
+/*==============*/
+			/* out: savepoint */
+	mtr_t*	mtr);	/* in: mtr */
+/**************************************************************
+Releases the latches stored in an mtr memo down to a savepoint.
+NOTE! The mtr must not have made changes to buffer pages after the
+savepoint, as these can be handled only by mtr_commit. */
+UNIV_INTERN
+void
+mtr_rollback_to_savepoint(
+/*======================*/
+	mtr_t*	mtr,		/* in: mtr */
+	ulint	savepoint);	/* in: savepoint */
+/**************************************************************
+Releases the (index tree) s-latch stored in an mtr memo after a
+savepoint. */
+UNIV_INLINE
+void
+mtr_release_s_latch_at_savepoint(
+/*=============================*/
+	mtr_t*		mtr,		/* in: mtr */
+	ulint		savepoint,	/* in: savepoint */
+	rw_lock_t*	lock);		/* in: latch to release */
+/*******************************************************************
+Gets the logging mode of a mini-transaction. */
+UNIV_INLINE
+ulint
+mtr_get_log_mode(
+/*=============*/
+			/* out: logging mode: MTR_LOG_NONE, ... */
+	mtr_t*	mtr);	/* in: mtr */
+/*******************************************************************
+Changes the logging mode of a mini-transaction. */
+UNIV_INLINE
+ulint
+mtr_set_log_mode(
+/*=============*/
+			/* out: old mode */
+	mtr_t*	mtr,	/* in: mtr */
+	ulint	mode);	/* in: logging mode: MTR_LOG_NONE, ... */
+/************************************************************
+Reads 1 - 4 bytes from a file page buffered in the buffer pool. */
+UNIV_INTERN
+ulint
+mtr_read_ulint(
+/*===========*/
+				/* out: value read */
+	const byte*	ptr,	/* in: pointer from where to read */
+	ulint		type,	/* in: MLOG_1BYTE, MLOG_2BYTES, MLOG_4BYTES */
+	mtr_t*		mtr);	/* in: mini-transaction handle */
+/************************************************************
+Reads 8 bytes from a file page buffered in the buffer pool. */
+UNIV_INTERN
+dulint
+mtr_read_dulint(
+/*============*/
+				/* out: value read */
+	const byte*	ptr,	/* in: pointer from where to read */
+	mtr_t*		mtr);	/* in: mini-transaction handle */
+/*************************************************************************
+This macro locks an rw-lock in s-mode. */
+#define mtr_s_lock(B, MTR)	mtr_s_lock_func((B), __FILE__, __LINE__,\
+						(MTR))
+/*************************************************************************
+This macro locks an rw-lock in x-mode. */
+#define mtr_x_lock(B, MTR)	mtr_x_lock_func((B), __FILE__, __LINE__,\
+						(MTR))
+/*************************************************************************
+NOTE! Use the macro above!
+Locks a lock in s-mode. */
+UNIV_INLINE
+void
+mtr_s_lock_func(
+/*============*/
+	rw_lock_t*	lock,	/* in: rw-lock */
+	const char*	file,	/* in: file name */
+	ulint		line,	/* in: line number */
+	mtr_t*		mtr);	/* in: mtr */
+/*************************************************************************
+NOTE! Use the macro above!
+Locks a lock in x-mode. */
+UNIV_INLINE
+void
+mtr_x_lock_func(
+/*============*/
+	rw_lock_t*	lock,	/* in: rw-lock */
+	const char*	file,	/* in: file name */
+	ulint		line,	/* in: line number */
+	mtr_t*		mtr);	/* in: mtr */
+
+/*******************************************************
+Releases an object in the memo stack. */
+UNIV_INTERN
+void
+mtr_memo_release(
+/*=============*/
+	mtr_t*	mtr,	/* in: mtr */
+	void*	object,	/* in: object */
+	ulint	type);	/* in: object type: MTR_MEMO_S_LOCK, ... */
+#ifdef UNIV_DEBUG
+/**************************************************************
+Checks if memo contains the given item. */
+UNIV_INLINE
+ibool
+mtr_memo_contains(
+/*==============*/
+				/* out: TRUE if contains */
+	mtr_t*		mtr,	/* in: mtr */
+	const void*	object,	/* in: object to search */
+	ulint		type);	/* in: type of object */
+
+/**************************************************************
+Checks if memo contains the given page. */
+UNIV_INTERN
+ibool
+mtr_memo_contains_page(
+/*===================*/
+				/* out: TRUE if contains */
+	mtr_t*		mtr,	/* in: mtr */
+	const byte*	ptr,	/* in: pointer to buffer frame */
+	ulint		type);	/* in: type of object */
+/*************************************************************
+Prints info of an mtr handle. */
+UNIV_INTERN
+void
+mtr_print(
+/*======*/
+	mtr_t*	mtr);	/* in: mtr */
+#endif /* UNIV_DEBUG */
+/*######################################################################*/
+
+#define	MTR_BUF_MEMO_SIZE	200	/* number of slots in memo */
+
+/*******************************************************************
+Returns the log object of a mini-transaction buffer. */
+UNIV_INLINE
+dyn_array_t*
+mtr_get_log(
+/*========*/
+			/* out: log */
+	mtr_t*	mtr);	/* in: mini-transaction */
+/*******************************************************
+Pushes an object to an mtr memo stack. */
+UNIV_INLINE
+void
+mtr_memo_push(
+/*==========*/
+	mtr_t*	mtr,	/* in: mtr */
+	void*	object,	/* in: object */
+	ulint	type);	/* in: object type: MTR_MEMO_S_LOCK, ... */
+
+
+/* Type definition of a mini-transaction memo stack slot. */
+typedef	struct mtr_memo_slot_struct	mtr_memo_slot_t;
+struct mtr_memo_slot_struct{
+	ulint	type;	/* type of the stored object (MTR_MEMO_S_LOCK, ...) */
+	void*	object;	/* pointer to the object */
+};
+
+/* Mini-transaction handle and buffer */
+struct mtr_struct{
+#ifdef UNIV_DEBUG
+	ulint		state;	/* MTR_ACTIVE, MTR_COMMITTING, MTR_COMMITTED */
+#endif
+	dyn_array_t	memo;	/* memo stack for locks etc. */
+	dyn_array_t	log;	/* mini-transaction log */
+	ibool		modifications;
+				/* TRUE if the mtr made modifications to
+				buffer pool pages */
+	ulint		n_log_recs;
+				/* count of how many page initial log records
+				have been written to the mtr log */
+	ulint		log_mode; /* specifies which operations should be
+				logged; default value MTR_LOG_ALL */
+	ib_uint64_t	start_lsn;/* start lsn of the possible log entry for
+				this mtr */
+	ib_uint64_t	end_lsn;/* end lsn of the possible log entry for
+				this mtr */
+#ifdef UNIV_DEBUG
+	ulint		magic_n;
+#endif /* UNIV_DEBUG */
+};
+
+#ifdef UNIV_DEBUG
+# define MTR_MAGIC_N		54551
+#endif /* UNIV_DEBUG */
+
+#define MTR_ACTIVE		12231
+#define MTR_COMMITTING		56456
+#define MTR_COMMITTED		34676
+
+#ifndef UNIV_NONINL
+#include "mtr0mtr.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/mtr0mtr.ic b/storage/xtradb/include/mtr0mtr.ic
new file mode 100644
index 00000000000..7d6d99917b7
--- /dev/null
+++ b/storage/xtradb/include/mtr0mtr.ic
@@ -0,0 +1,266 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Mini-transaction buffer
+
+Created 11/26/1995 Heikki Tuuri
+*******************************************************/
+
+#include "sync0sync.h"
+#include "sync0rw.h"
+#include "mach0data.h"
+
+/*******************************************************************
+Starts a mini-transaction and creates a mini-transaction handle
+and a buffer in the memory buffer given by the caller. */
+UNIV_INLINE
+mtr_t*
+mtr_start(
+/*======*/
+			/* out: mtr buffer which also acts as
+			the mtr handle */
+	mtr_t*	mtr)	/* in: memory buffer for the mtr buffer */
+{
+	dyn_array_create(&(mtr->memo));
+	dyn_array_create(&(mtr->log));
+
+	mtr->log_mode = MTR_LOG_ALL;
+	mtr->modifications = FALSE;
+	mtr->n_log_recs = 0;
+
+	ut_d(mtr->state = MTR_ACTIVE);
+	ut_d(mtr->magic_n = MTR_MAGIC_N);
+
+	return(mtr);
+}
+
+/*******************************************************
+Pushes an object to an mtr memo stack. */
+UNIV_INLINE
+void
+mtr_memo_push(
+/*==========*/
+	mtr_t*	mtr,	/* in: mtr */
+	void*	object,	/* in: object */
+	ulint	type)	/* in: object type: MTR_MEMO_S_LOCK, ... */
+{
+	dyn_array_t*		memo;
+	mtr_memo_slot_t*	slot;
+
+	ut_ad(object);
+	ut_ad(type >= MTR_MEMO_PAGE_S_FIX);
+	ut_ad(type <= MTR_MEMO_X_LOCK);
+	ut_ad(mtr);
+	ut_ad(mtr->magic_n == MTR_MAGIC_N);
+
+	memo = &(mtr->memo);
+
+	slot = (mtr_memo_slot_t*) dyn_array_push(memo, sizeof *slot);
+
+	slot->object = object;
+	slot->type = type;
+}
+
+/**************************************************************
+Sets and returns a savepoint in mtr. */
+UNIV_INLINE
+ulint
+mtr_set_savepoint(
+/*==============*/
+			/* out: savepoint */
+	mtr_t*	mtr)	/* in: mtr */
+{
+	dyn_array_t*	memo;
+
+	ut_ad(mtr);
+	ut_ad(mtr->magic_n == MTR_MAGIC_N);
+
+	memo = &(mtr->memo);
+
+	return(dyn_array_get_data_size(memo));
+}
+
+/**************************************************************
+Releases the (index tree) s-latch stored in an mtr memo after a
+savepoint. */
+UNIV_INLINE
+void
+mtr_release_s_latch_at_savepoint(
+/*=============================*/
+	mtr_t*		mtr,		/* in: mtr */
+	ulint		savepoint,	/* in: savepoint */
+	rw_lock_t*	lock)		/* in: latch to release */
+{
+	mtr_memo_slot_t* slot;
+	dyn_array_t*	memo;
+
+	ut_ad(mtr);
+	ut_ad(mtr->magic_n == MTR_MAGIC_N);
+	ut_ad(mtr->state == MTR_ACTIVE);
+
+	memo = &(mtr->memo);
+
+	ut_ad(dyn_array_get_data_size(memo) > savepoint);
+
+	slot = (mtr_memo_slot_t*) dyn_array_get_element(memo, savepoint);
+
+	ut_ad(slot->object == lock);
+	ut_ad(slot->type == MTR_MEMO_S_LOCK);
+
+	rw_lock_s_unlock(lock);
+
+	slot->object = NULL;
+}
+
+#ifdef UNIV_DEBUG
+/**************************************************************
+Checks if memo contains the given item. */
+UNIV_INLINE
+ibool
+mtr_memo_contains(
+/*==============*/
+				/* out: TRUE if contains */
+	mtr_t*		mtr,	/* in: mtr */
+	const void*	object,	/* in: object to search */
+	ulint		type)	/* in: type of object */
+{
+	mtr_memo_slot_t* slot;
+	dyn_array_t*	memo;
+	ulint		offset;
+
+	ut_ad(mtr);
+	ut_ad(mtr->magic_n == MTR_MAGIC_N);
+
+	memo = &(mtr->memo);
+
+	offset = dyn_array_get_data_size(memo);
+
+	while (offset > 0) {
+		offset -= sizeof(mtr_memo_slot_t);
+
+		slot = dyn_array_get_element(memo, offset);
+
+		if ((object == slot->object) && (type == slot->type)) {
+
+			return(TRUE);
+		}
+	}
+
+	return(FALSE);
+}
+#endif /* UNIV_DEBUG */
+
+/*******************************************************************
+Returns the log object of a mini-transaction buffer. */
+UNIV_INLINE
+dyn_array_t*
+mtr_get_log(
+/*========*/
+			/* out: log */
+	mtr_t*	mtr)	/* in: mini-transaction */
+{
+	ut_ad(mtr);
+	ut_ad(mtr->magic_n == MTR_MAGIC_N);
+
+	return(&(mtr->log));
+}
+
+/*******************************************************************
+Gets the logging mode of a mini-transaction. */
+UNIV_INLINE
+ulint
+mtr_get_log_mode(
+/*=============*/
+			/* out: logging mode: MTR_LOG_NONE, ... */
+	mtr_t*	mtr)	/* in: mtr */
+{
+	ut_ad(mtr);
+	ut_ad(mtr->log_mode >= MTR_LOG_ALL);
+	ut_ad(mtr->log_mode <= MTR_LOG_SHORT_INSERTS);
+
+	return(mtr->log_mode);
+}
+
+/*******************************************************************
+Changes the logging mode of a mini-transaction. */
+UNIV_INLINE
+ulint
+mtr_set_log_mode(
+/*=============*/
+			/* out: old mode */
+	mtr_t*	mtr,	/* in: mtr */
+	ulint	mode)	/* in: logging mode: MTR_LOG_NONE, ... */
+{
+	ulint	old_mode;
+
+	ut_ad(mtr);
+	ut_ad(mode >= MTR_LOG_ALL);
+	ut_ad(mode <= MTR_LOG_SHORT_INSERTS);
+
+	old_mode = mtr->log_mode;
+
+	if ((mode == MTR_LOG_SHORT_INSERTS) && (old_mode == MTR_LOG_NONE)) {
+		/* Do nothing */
+	} else {
+		mtr->log_mode = mode;
+	}
+
+	ut_ad(old_mode >= MTR_LOG_ALL);
+	ut_ad(old_mode <= MTR_LOG_SHORT_INSERTS);
+
+	return(old_mode);
+}
+
+/*************************************************************************
+Locks a lock in s-mode. */
+UNIV_INLINE
+void
+mtr_s_lock_func(
+/*============*/
+	rw_lock_t*	lock,	/* in: rw-lock */
+	const char*	file,	/* in: file name */
+	ulint		line,	/* in: line number */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	ut_ad(mtr);
+	ut_ad(lock);
+
+	rw_lock_s_lock_func(lock, 0, file, line);
+
+	mtr_memo_push(mtr, lock, MTR_MEMO_S_LOCK);
+}
+
+/*************************************************************************
+Locks a lock in x-mode. */
+UNIV_INLINE
+void
+mtr_x_lock_func(
+/*============*/
+	rw_lock_t*	lock,	/* in: rw-lock */
+	const char*	file,	/* in: file name */
+	ulint		line,	/* in: line number */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	ut_ad(mtr);
+	ut_ad(lock);
+
+	rw_lock_x_lock_func(lock, 0, file, line);
+
+	mtr_memo_push(mtr, lock, MTR_MEMO_X_LOCK);
+}
diff --git a/storage/xtradb/include/mtr0types.h b/storage/xtradb/include/mtr0types.h
new file mode 100644
index 00000000000..23634c98827
--- /dev/null
+++ b/storage/xtradb/include/mtr0types.h
@@ -0,0 +1,30 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Mini-transaction buffer global types
+
+Created 11/26/1995 Heikki Tuuri
+*******************************************************/
+
+#ifndef mtr0types_h
+#define mtr0types_h
+
+typedef struct mtr_struct	mtr_t;
+
+#endif
diff --git a/storage/xtradb/include/mysql_addons.h b/storage/xtradb/include/mysql_addons.h
new file mode 100644
index 00000000000..2e8c87f5962
--- /dev/null
+++ b/storage/xtradb/include/mysql_addons.h
@@ -0,0 +1,32 @@
+/*****************************************************************************
+
+Copyright (c) 2007, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+This file contains functions that need to be added to
+MySQL code but have not been added yet.
+
+Whenever you add a function here submit a MySQL bug
+report (feature request) with the implementation. Then
+write the bug number in the comment before the
+function in this file.
+
+When MySQL commits the function it can be deleted from
+here. In a perfect world this file exists but is empty.
+
+Created November 07, 2007 Vasil Dimov
+*******************************************************/
diff --git a/storage/xtradb/include/os0file.h b/storage/xtradb/include/os0file.h
new file mode 100644
index 00000000000..26897226ff4
--- /dev/null
+++ b/storage/xtradb/include/os0file.h
@@ -0,0 +1,758 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The interface to the operating system file io
+
+Created 10/21/1995 Heikki Tuuri
+*******************************************************/
+
+#ifndef os0file_h
+#define os0file_h
+
+#include "univ.i"
+
+#ifndef __WIN__
+#include <dirent.h>
+#include <sys/stat.h>
+#include <time.h>
+#endif
+
+typedef	struct fil_node_struct	fil_node_t;
+
+#ifdef UNIV_DO_FLUSH
+extern ibool	os_do_not_call_flush_at_each_write;
+#endif /* UNIV_DO_FLUSH */
+extern ibool	os_has_said_disk_full;
+extern ibool	os_aio_print_debug;
+
+extern ulint	os_file_n_pending_preads;
+extern ulint	os_file_n_pending_pwrites;
+
+extern ulint	os_n_pending_reads;
+extern ulint	os_n_pending_writes;
+
+#ifdef __WIN__
+
+/* We define always WIN_ASYNC_IO, and check at run-time whether
+   the OS actually supports it: Win 95 does not, NT does. */
+#define WIN_ASYNC_IO
+
+#define UNIV_NON_BUFFERED_IO
+
+#endif
+
+#ifdef __WIN__
+#define os_file_t	HANDLE
+#define OS_FILE_FROM_FD(fd) (HANDLE) _get_osfhandle(fd)
+#else
+typedef int	os_file_t;
+#define OS_FILE_FROM_FD(fd) fd
+#endif
+
+extern ulint	os_innodb_umask;
+
+/* If this flag is TRUE, then we will use the native aio of the
+OS (provided we compiled Innobase with it in), otherwise we will
+use simulated aio we build below with threads */
+
+extern ibool	os_aio_use_native_aio;
+
+#define OS_FILE_SECTOR_SIZE		512
+
+/* The next value should be smaller or equal to the smallest sector size used
+on any disk. A log block is required to be a portion of disk which is written
+so that if the start and the end of a block get written to disk, then the
+whole block gets written. This should be true even in most cases of a crash:
+if this fails for a log block, then it is equivalent to a media failure in the
+log. */
+
+#define OS_FILE_LOG_BLOCK_SIZE		512
+
+/* Options for file_create */
+#define	OS_FILE_OPEN			51
+#define	OS_FILE_CREATE			52
+#define OS_FILE_OVERWRITE		53
+#define OS_FILE_OPEN_RAW		54
+#define	OS_FILE_CREATE_PATH		55
+#define	OS_FILE_OPEN_RETRY		56	/* for os_file_create() on
+						the first ibdata file */
+
+#define OS_FILE_READ_ONLY		333
+#define	OS_FILE_READ_WRITE		444
+#define	OS_FILE_READ_ALLOW_DELETE	555	/* for ibbackup */
+
+/* Options for file_create */
+#define	OS_FILE_AIO			61
+#define	OS_FILE_NORMAL			62
+
+/* Types for file create */
+#define	OS_DATA_FILE			100
+#define OS_LOG_FILE			101
+
+/* Error codes from os_file_get_last_error */
+#define	OS_FILE_NOT_FOUND		71
+#define	OS_FILE_DISK_FULL		72
+#define	OS_FILE_ALREADY_EXISTS		73
+#define	OS_FILE_PATH_ERROR		74
+#define	OS_FILE_AIO_RESOURCES_RESERVED	75	/* wait for OS aio resources
+						to become available again */
+#define	OS_FILE_SHARING_VIOLATION	76
+#define	OS_FILE_ERROR_NOT_SPECIFIED	77
+
+/* Types for aio operations */
+#define OS_FILE_READ	10
+#define OS_FILE_WRITE	11
+
+#define OS_FILE_LOG	256	/* This can be ORed to type */
+
+#define OS_AIO_N_PENDING_IOS_PER_THREAD 32	/* Win NT does not allow more
+						than 64 */
+
+/* Modes for aio operations */
+#define OS_AIO_NORMAL	21	/* Normal asynchronous i/o not for ibuf
+				pages or ibuf bitmap pages */
+#define OS_AIO_IBUF	22	/* Asynchronous i/o for ibuf pages or ibuf
+				bitmap pages */
+#define OS_AIO_LOG	23	/* Asynchronous i/o for the log */
+#define OS_AIO_SYNC	24	/* Asynchronous i/o where the calling thread
+				will itself wait for the i/o to complete,
+				doing also the job of the i/o-handler thread;
+				can be used for any pages, ibuf or non-ibuf.
+				This is used to save CPU time, as we can do
+				with fewer thread switches. Plain synchronous
+				i/o is not as good, because it must serialize
+				the file seek and read or write, causing a
+				bottleneck for parallelism. */
+
+#define OS_AIO_SIMULATED_WAKE_LATER	512 /* This can be ORed to mode
+				in the call of os_aio(...),
+				if the caller wants to post several i/o
+				requests in a batch, and only after that
+				wake the i/o-handler thread; this has
+				effect only in simulated aio */
+#define OS_WIN31	1
+#define OS_WIN95	2
+#define OS_WINNT	3
+#define OS_WIN2000	4
+
+extern ulint	os_n_file_reads;
+extern ulint	os_n_file_writes;
+extern ulint	os_n_fsyncs;
+
+/* File types for directory entry data type */
+
+enum os_file_type_enum{
+	OS_FILE_TYPE_UNKNOWN = 0,
+	OS_FILE_TYPE_FILE,			/* regular file */
+	OS_FILE_TYPE_DIR,			/* directory */
+	OS_FILE_TYPE_LINK			/* symbolic link */
+};
+typedef enum os_file_type_enum	  os_file_type_t;
+
+/* Maximum path string length in bytes when referring to tables with in the
+'./databasename/tablename.ibd' path format; we can allocate at least 2 buffers
+of this size from the thread stack; that is why this should not be made much
+bigger than 4000 bytes */
+#define OS_FILE_MAX_PATH	4000
+
+/* Struct used in fetching information of a file in a directory */
+struct os_file_stat_struct{
+	char		name[OS_FILE_MAX_PATH];	/* path to a file */
+	os_file_type_t	type;			/* file type */
+	ib_int64_t	size;			/* file size */
+	time_t		ctime;			/* creation time */
+	time_t		mtime;			/* modification time */
+	time_t		atime;			/* access time */
+};
+typedef struct os_file_stat_struct	os_file_stat_t;
+
+#ifdef __WIN__
+typedef HANDLE	os_file_dir_t;	/* directory stream */
+#else
+typedef DIR*	os_file_dir_t;	/* directory stream */
+#endif
+
+/***************************************************************************
+Gets the operating system version. Currently works only on Windows. */
+UNIV_INTERN
+ulint
+os_get_os_version(void);
+/*===================*/
+		  /* out: OS_WIN95, OS_WIN31, OS_WINNT, or OS_WIN2000 */
+/********************************************************************
+Creates the seek mutexes used in positioned reads and writes. */
+UNIV_INTERN
+void
+os_io_init_simple(void);
+/*===================*/
+/***************************************************************************
+Creates a temporary file.  This function is like tmpfile(3), but
+the temporary file is created in the MySQL temporary directory.
+On Netware, this function is like tmpfile(3), because the C run-time
+library of Netware does not expose the delete-on-close flag. */
+
+FILE*
+os_file_create_tmpfile(void);
+/*========================*/
+			/* out: temporary file handle, or NULL on error */
+/***************************************************************************
+The os_file_opendir() function opens a directory stream corresponding to the
+directory named by the dirname argument. The directory stream is positioned
+at the first entry. In both Unix and Windows we automatically skip the '.'
+and '..' items at the start of the directory listing. */
+UNIV_INTERN
+os_file_dir_t
+os_file_opendir(
+/*============*/
+					/* out: directory stream, NULL if
+					error */
+	const char*	dirname,	/* in: directory name; it must not
+					contain a trailing '\' or '/' */
+	ibool		error_is_fatal);/* in: TRUE if we should treat an
+					error as a fatal error; if we try to
+					open symlinks then we do not wish a
+					fatal error if it happens not to be
+					a directory */
+/***************************************************************************
+Closes a directory stream. */
+UNIV_INTERN
+int
+os_file_closedir(
+/*=============*/
+				/* out: 0 if success, -1 if failure */
+	os_file_dir_t	dir);	/* in: directory stream */
+/***************************************************************************
+This function returns information of the next file in the directory. We jump
+over the '.' and '..' entries in the directory. */
+UNIV_INTERN
+int
+os_file_readdir_next_file(
+/*======================*/
+				/* out: 0 if ok, -1 if error, 1 if at the end
+				of the directory */
+	const char*	dirname,/* in: directory name or path */
+	os_file_dir_t	dir,	/* in: directory stream */
+	os_file_stat_t*	info);	/* in/out: buffer where the info is returned */
+/*********************************************************************
+This function attempts to create a directory named pathname. The new directory
+gets default permissions. On Unix, the permissions are (0770 & ~umask). If the
+directory exists already, nothing is done and the call succeeds, unless the
+fail_if_exists arguments is true. */
+UNIV_INTERN
+ibool
+os_file_create_directory(
+/*=====================*/
+					/* out: TRUE if call succeeds,
+					FALSE on error */
+	const char*	pathname,	/* in: directory name as
+					null-terminated string */
+	ibool		fail_if_exists);/* in: if TRUE, pre-existing directory
+					is treated as an error. */
+/********************************************************************
+A simple function to open or create a file. */
+UNIV_INTERN
+os_file_t
+os_file_create_simple(
+/*==================*/
+				/* out, own: handle to the file, not defined
+				if error, error number can be retrieved with
+				os_file_get_last_error */
+	const char*	name,	/* in: name of the file or path as a
+				null-terminated string */
+	ulint		create_mode,/* in: OS_FILE_OPEN if an existing file is
+				opened (if does not exist, error), or
+				OS_FILE_CREATE if a new file is created
+				(if exists, error), or
+				OS_FILE_CREATE_PATH if new file
+				(if exists, error) and subdirectories along
+				its path are created (if needed)*/
+	ulint		access_type,/* in: OS_FILE_READ_ONLY or
+				OS_FILE_READ_WRITE */
+	ibool*		success);/* out: TRUE if succeed, FALSE if error */
+/********************************************************************
+A simple function to open or create a file. */
+UNIV_INTERN
+os_file_t
+os_file_create_simple_no_error_handling(
+/*====================================*/
+				/* out, own: handle to the file, not defined
+				if error, error number can be retrieved with
+				os_file_get_last_error */
+	const char*	name,	/* in: name of the file or path as a
+				null-terminated string */
+	ulint		create_mode,/* in: OS_FILE_OPEN if an existing file
+				is opened (if does not exist, error), or
+				OS_FILE_CREATE if a new file is created
+				(if exists, error) */
+	ulint		access_type,/* in: OS_FILE_READ_ONLY,
+				OS_FILE_READ_WRITE, or
+				OS_FILE_READ_ALLOW_DELETE; the last option is
+				used by a backup program reading the file */
+	ibool*		success);/* out: TRUE if succeed, FALSE if error */
+/********************************************************************
+Tries to disable OS caching on an opened file descriptor. */
+UNIV_INTERN
+void
+os_file_set_nocache(
+/*================*/
+	int		fd,		/* in: file descriptor to alter */
+	const char*	file_name,	/* in: file name, used in the
+					diagnostic message */
+	const char*	operation_name);/* in: "open" or "create"; used in the
+					diagnostic message */
+/********************************************************************
+Opens an existing file or creates a new. */
+UNIV_INTERN
+os_file_t
+os_file_create(
+/*===========*/
+				/* out, own: handle to the file, not defined
+				if error, error number can be retrieved with
+				os_file_get_last_error */
+	const char*	name,	/* in: name of the file or path as a
+				null-terminated string */
+	ulint		create_mode,/* in: OS_FILE_OPEN if an existing file
+				is opened (if does not exist, error), or
+				OS_FILE_CREATE if a new file is created
+				(if exists, error),
+				OS_FILE_OVERWRITE if a new file is created
+				or an old overwritten;
+				OS_FILE_OPEN_RAW, if a raw device or disk
+				partition should be opened */
+	ulint		purpose,/* in: OS_FILE_AIO, if asynchronous,
+				non-buffered i/o is desired,
+				OS_FILE_NORMAL, if any normal file;
+				NOTE that it also depends on type, os_aio_..
+				and srv_.. variables whether we really use
+				async i/o or unbuffered i/o: look in the
+				function source code for the exact rules */
+	ulint		type,	/* in: OS_DATA_FILE or OS_LOG_FILE */
+	ibool*		success);/* out: TRUE if succeed, FALSE if error */
+/***************************************************************************
+Deletes a file. The file has to be closed before calling this. */
+UNIV_INTERN
+ibool
+os_file_delete(
+/*===========*/
+				/* out: TRUE if success */
+	const char*	name);	/* in: file path as a null-terminated string */
+
+/***************************************************************************
+Deletes a file if it exists. The file has to be closed before calling this. */
+UNIV_INTERN
+ibool
+os_file_delete_if_exists(
+/*=====================*/
+				/* out: TRUE if success */
+	const char*	name);	/* in: file path as a null-terminated string */
+/***************************************************************************
+Renames a file (can also move it to another directory). It is safest that the
+file is closed before calling this function. */
+UNIV_INTERN
+ibool
+os_file_rename(
+/*===========*/
+					/* out: TRUE if success */
+	const char*	oldpath,	/* in: old file path as a
+					null-terminated string */
+	const char*	newpath);	/* in: new file path */
+/***************************************************************************
+Closes a file handle. In case of error, error number can be retrieved with
+os_file_get_last_error. */
+UNIV_INTERN
+ibool
+os_file_close(
+/*==========*/
+				/* out: TRUE if success */
+	os_file_t	file);	/* in, own: handle to a file */
+/***************************************************************************
+Closes a file handle. */
+UNIV_INTERN
+ibool
+os_file_close_no_error_handling(
+/*============================*/
+				/* out: TRUE if success */
+	os_file_t	file);	/* in, own: handle to a file */
+/***************************************************************************
+Gets a file size. */
+UNIV_INTERN
+ibool
+os_file_get_size(
+/*=============*/
+				/* out: TRUE if success */
+	os_file_t	file,	/* in: handle to a file */
+	ulint*		size,	/* out: least significant 32 bits of file
+				size */
+	ulint*		size_high);/* out: most significant 32 bits of size */
+/***************************************************************************
+Gets file size as a 64-bit integer ib_int64_t. */
+UNIV_INTERN
+ib_int64_t
+os_file_get_size_as_iblonglong(
+/*===========================*/
+				/* out: size in bytes, -1 if error */
+	os_file_t	file);	/* in: handle to a file */
+/***************************************************************************
+Write the specified number of zeros to a newly created file. */
+UNIV_INTERN
+ibool
+os_file_set_size(
+/*=============*/
+				/* out: TRUE if success */
+	const char*	name,	/* in: name of the file or path as a
+				null-terminated string */
+	os_file_t	file,	/* in: handle to a file */
+	ulint		size,	/* in: least significant 32 bits of file
+				size */
+	ulint		size_high);/* in: most significant 32 bits of size */
+/***************************************************************************
+Truncates a file at its current position. */
+UNIV_INTERN
+ibool
+os_file_set_eof(
+/*============*/
+				/* out: TRUE if success */
+	FILE*		file);	/* in: file to be truncated */
+/***************************************************************************
+Flushes the write buffers of a given file to the disk. */
+UNIV_INTERN
+ibool
+os_file_flush(
+/*==========*/
+				/* out: TRUE if success */
+	os_file_t	file);	/* in, own: handle to a file */
+/***************************************************************************
+Retrieves the last error number if an error occurs in a file io function.
+The number should be retrieved before any other OS calls (because they may
+overwrite the error number). If the number is not known to this program,
+the OS error number + 100 is returned. */
+UNIV_INTERN
+ulint
+os_file_get_last_error(
+/*===================*/
+					/* out: error number, or OS error
+					number + 100 */
+	ibool	report_all_errors);	/* in: TRUE if we want an error message
+					printed of all errors */
+/***********************************************************************
+Requests a synchronous read operation. */
+UNIV_INTERN
+ibool
+os_file_read(
+/*=========*/
+				/* out: TRUE if request was
+				successful, FALSE if fail */
+	os_file_t	file,	/* in: handle to a file */
+	void*		buf,	/* in: buffer where to read */
+	ulint		offset,	/* in: least significant 32 bits of file
+				offset where to read */
+	ulint		offset_high,/* in: most significant 32 bits of
+				offset */
+	ulint		n);	/* in: number of bytes to read */
+/***********************************************************************
+Rewind file to its start, read at most size - 1 bytes from it to str, and
+NUL-terminate str. All errors are silently ignored. This function is
+mostly meant to be used with temporary files. */
+UNIV_INTERN
+void
+os_file_read_string(
+/*================*/
+	FILE*	file,	/* in: file to read from */
+	char*	str,	/* in: buffer where to read */
+	ulint	size);	/* in: size of buffer */
+/***********************************************************************
+Requests a synchronous positioned read operation. This function does not do
+any error handling. In case of error it returns FALSE. */
+UNIV_INTERN
+ibool
+os_file_read_no_error_handling(
+/*===========================*/
+				/* out: TRUE if request was
+				successful, FALSE if fail */
+	os_file_t	file,	/* in: handle to a file */
+	void*		buf,	/* in: buffer where to read */
+	ulint		offset,	/* in: least significant 32 bits of file
+				offset where to read */
+	ulint		offset_high,/* in: most significant 32 bits of
+				offset */
+	ulint		n);	/* in: number of bytes to read */
+
+/***********************************************************************
+Requests a synchronous write operation. */
+UNIV_INTERN
+ibool
+os_file_write(
+/*==========*/
+				/* out: TRUE if request was
+				successful, FALSE if fail */
+	const char*	name,	/* in: name of the file or path as a
+				null-terminated string */
+	os_file_t	file,	/* in: handle to a file */
+	const void*	buf,	/* in: buffer from which to write */
+	ulint		offset,	/* in: least significant 32 bits of file
+				offset where to write */
+	ulint		offset_high,/* in: most significant 32 bits of
+				offset */
+	ulint		n);	/* in: number of bytes to write */
+/***********************************************************************
+Check the existence and type of the given file. */
+UNIV_INTERN
+ibool
+os_file_status(
+/*===========*/
+				/* out: TRUE if call succeeded */
+	const char*	path,	/* in:	pathname of the file */
+	ibool*		exists,	/* out: TRUE if file exists */
+	os_file_type_t* type);	/* out: type of the file (if it exists) */
+/********************************************************************
+The function os_file_dirname returns a directory component of a
+null-terminated pathname string.  In the usual case, dirname returns
+the string up to, but not including, the final '/', and basename
+is the component following the final '/'.  Trailing '/' charac­
+ters are not counted as part of the pathname.
+
+If path does not contain a slash, dirname returns the string ".".
+
+Concatenating the string returned by dirname, a "/", and the basename
+yields a complete pathname.
+
+The return value is  a copy of the directory component of the pathname.
+The copy is allocated from heap. It is the caller responsibility
+to free it after it is no longer needed.
+
+The following list of examples (taken from SUSv2) shows the strings
+returned by dirname and basename for different paths:
+
+       path	      dirname	     basename
+       "/usr/lib"     "/usr"	     "lib"
+       "/usr/"	      "/"	     "usr"
+       "usr"	      "."	     "usr"
+       "/"	      "/"	     "/"
+       "."	      "."	     "."
+       ".."	      "."	     ".."
+*/
+UNIV_INTERN
+char*
+os_file_dirname(
+/*============*/
+				/* out, own: directory component of the
+				pathname */
+	const char*	path);	/* in: pathname */
+/********************************************************************
+Creates all missing subdirectories along the given path. */
+UNIV_INTERN
+ibool
+os_file_create_subdirs_if_needed(
+/*=============================*/
+				/* out: TRUE if call succeeded
+				   FALSE otherwise */
+	const char*	path);	/* in: path name */
+/****************************************************************************
+Initializes the asynchronous io system. Creates separate aio array for
+non-ibuf read and write, a third aio array for the ibuf i/o, with just one
+segment, two aio arrays for log reads and writes with one segment, and a
+synchronous aio array of the specified size. The combined number of segments
+in the three first aio arrays is the parameter n_segments given to the
+function. The caller must create an i/o handler thread for each segment in
+the four first arrays, but not for the sync aio array. */
+UNIV_INTERN
+void
+os_aio_init(
+/*========*/
+	ulint	n,		/* in: maximum number of pending aio operations
+				allowed; n must be divisible by n_segments */
+//	ulint	n_segments,	/* in: combined number of segments in the four
+//				first aio arrays; must be >= 4 */
+	ulint	n_read_threads,  /* n_segments == 2 + n_read_threads + n_write_threads */
+	ulint	n_write_threads, /**/
+	ulint	n_slots_sync);	/* in: number of slots in the sync aio array */
+/***********************************************************************
+Requests an asynchronous i/o operation. */
+UNIV_INTERN
+ibool
+os_aio(
+/*===*/
+				/* out: TRUE if request was queued
+				successfully, FALSE if fail */
+	ulint		type,	/* in: OS_FILE_READ or OS_FILE_WRITE */
+	ulint		mode,	/* in: OS_AIO_NORMAL, ..., possibly ORed
+				to OS_AIO_SIMULATED_WAKE_LATER: the
+				last flag advises this function not to wake
+				i/o-handler threads, but the caller will
+				do the waking explicitly later, in this
+				way the caller can post several requests in
+				a batch; NOTE that the batch must not be
+				so big that it exhausts the slots in aio
+				arrays! NOTE that a simulated batch
+				may introduce hidden chances of deadlocks,
+				because i/os are not actually handled until
+				all have been posted: use with great
+				caution! */
+	const char*	name,	/* in: name of the file or path as a
+				null-terminated string */
+	os_file_t	file,	/* in: handle to a file */
+	void*		buf,	/* in: buffer where to read or from which
+				to write */
+	ulint		offset,	/* in: least significant 32 bits of file
+				offset where to read or write */
+	ulint		offset_high, /* in: most significant 32 bits of
+				offset */
+	ulint		n,	/* in: number of bytes to read or write */
+	fil_node_t*	message1,/* in: messages for the aio handler (these
+				can be used to identify a completed aio
+				operation); if mode is OS_AIO_SYNC, these
+				are ignored */
+	void*		message2);
+/****************************************************************************
+Wakes up all async i/o threads so that they know to exit themselves in
+shutdown. */
+UNIV_INTERN
+void
+os_aio_wake_all_threads_at_shutdown(void);
+/*=====================================*/
+/****************************************************************************
+Waits until there are no pending writes in os_aio_write_array. There can
+be other, synchronous, pending writes. */
+UNIV_INTERN
+void
+os_aio_wait_until_no_pending_writes(void);
+/*=====================================*/
+/**************************************************************************
+Wakes up simulated aio i/o-handler threads if they have something to do. */
+UNIV_INTERN
+void
+os_aio_simulated_wake_handler_threads(void);
+/*=======================================*/
+/**************************************************************************
+This function can be called if one wants to post a batch of reads and
+prefers an i/o-handler thread to handle them all at once later. You must
+call os_aio_simulated_wake_handler_threads later to ensure the threads
+are not left sleeping! */
+UNIV_INTERN
+void
+os_aio_simulated_put_read_threads_to_sleep(void);
+/*============================================*/
+
+#ifdef WIN_ASYNC_IO
+/**************************************************************************
+This function is only used in Windows asynchronous i/o.
+Waits for an aio operation to complete. This function is used to wait the
+for completed requests. The aio array of pending requests is divided
+into segments. The thread specifies which segment or slot it wants to wait
+for. NOTE: this function will also take care of freeing the aio slot,
+therefore no other thread is allowed to do the freeing! */
+UNIV_INTERN
+ibool
+os_aio_windows_handle(
+/*==================*/
+				/* out: TRUE if the aio operation succeeded */
+	ulint	segment,	/* in: the number of the segment in the aio
+				arrays to wait for; segment 0 is the ibuf
+				i/o thread, segment 1 the log i/o thread,
+				then follow the non-ibuf read threads, and as
+				the last are the non-ibuf write threads; if
+				this is ULINT_UNDEFINED, then it means that
+				sync aio is used, and this parameter is
+				ignored */
+	ulint	pos,		/* this parameter is used only in sync aio:
+				wait for the aio slot at this position */
+	fil_node_t**message1,	/* out: the messages passed with the aio
+				request; note that also in the case where
+				the aio operation failed, these output
+				parameters are valid and can be used to
+				restart the operation, for example */
+	void**	message2,
+	ulint*	type);		/* out: OS_FILE_WRITE or ..._READ */
+#endif
+
+/**************************************************************************
+Does simulated aio. This function should be called by an i/o-handler
+thread. */
+UNIV_INTERN
+ibool
+os_aio_simulated_handle(
+/*====================*/
+				/* out: TRUE if the aio operation succeeded */
+	ulint	segment,	/* in: the number of the segment in the aio
+				arrays to wait for; segment 0 is the ibuf
+				i/o thread, segment 1 the log i/o thread,
+				then follow the non-ibuf read threads, and as
+				the last are the non-ibuf write threads */
+	fil_node_t**message1,	/* out: the messages passed with the aio
+				request; note that also in the case where
+				the aio operation failed, these output
+				parameters are valid and can be used to
+				restart the operation, for example */
+	void**	message2,
+	ulint*	type);		/* out: OS_FILE_WRITE or ..._READ */
+/**************************************************************************
+Validates the consistency of the aio system. */
+UNIV_INTERN
+ibool
+os_aio_validate(void);
+/*=================*/
+				/* out: TRUE if ok */
+/**************************************************************************
+Prints info of the aio arrays. */
+UNIV_INTERN
+void
+os_aio_print(
+/*=========*/
+	FILE*	file);	/* in: file where to print */
+/**************************************************************************
+Refreshes the statistics used to print per-second averages. */
+UNIV_INTERN
+void
+os_aio_refresh_stats(void);
+/*======================*/
+
+#ifdef UNIV_DEBUG
+/**************************************************************************
+Checks that all slots in the system have been freed, that is, there are
+no pending io operations. */
+UNIV_INTERN
+ibool
+os_aio_all_slots_free(void);
+/*=======================*/
+#endif /* UNIV_DEBUG */
+
+/***********************************************************************
+This function returns information about the specified file */
+UNIV_INTERN
+ibool
+os_file_get_status(
+/*===============*/
+					/* out: TRUE if stat
+					information found */
+	const char*	path,		/* in:	pathname of the file */
+	os_file_stat_t* stat_info);	/* information of a file in a
+					directory */
+
+#if !defined(UNIV_HOTBACKUP) && !defined(__NETWARE__)
+/*************************************************************************
+Creates a temporary file that will be deleted on close.
+This function is defined in ha_innodb.cc. */
+UNIV_INTERN
+int
+innobase_mysql_tmpfile(void);
+/*========================*/
+			/* out: temporary file descriptor, or < 0 on error */
+#endif /* !UNIV_HOTBACKUP && !__NETWARE__ */
+
+#endif
diff --git a/storage/xtradb/include/os0proc.h b/storage/xtradb/include/os0proc.h
new file mode 100644
index 00000000000..19b0b112638
--- /dev/null
+++ b/storage/xtradb/include/os0proc.h
@@ -0,0 +1,84 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The interface to the operating system
+process control primitives
+
+Created 9/30/1995 Heikki Tuuri
+*******************************************************/
+
+#ifndef os0proc_h
+#define os0proc_h
+
+#include "univ.i"
+
+#ifdef UNIV_LINUX
+#include <sys/ipc.h>
+#include <sys/shm.h>
+#endif
+
+typedef void*			os_process_t;
+typedef unsigned long int	os_process_id_t;
+
+extern ibool os_use_large_pages;
+/* Large page size. This may be a boot-time option on some platforms */
+extern ulint os_large_page_size;
+
+/********************************************************************
+Converts the current process id to a number. It is not guaranteed that the
+number is unique. In Linux returns the 'process number' of the current
+thread. That number is the same as one sees in 'top', for example. In Linux
+the thread id is not the same as one sees in 'top'. */
+UNIV_INTERN
+ulint
+os_proc_get_number(void);
+/*====================*/
+/********************************************************************
+Allocates large pages memory. */
+UNIV_INTERN
+void*
+os_mem_alloc_large(
+/*===============*/
+					/* out: allocated memory */
+	ulint*	n);			/* in/out: number of bytes */
+/********************************************************************
+Frees large pages memory. */
+UNIV_INTERN
+void
+os_mem_free_large(
+/*==============*/
+	void	*ptr,			/* in: pointer returned by
+					os_mem_alloc_large() */
+	ulint	size);			/* in: size returned by
+					os_mem_alloc_large() */
+/********************************************************************
+Sets the priority boost for threads released from waiting within the current
+process. */
+UNIV_INTERN
+void
+os_process_set_priority_boost(
+/*==========================*/
+	ibool	do_boost);	/* in: TRUE if priority boost should be done,
+				FALSE if not */
+
+#ifndef UNIV_NONINL
+#include "os0proc.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/os0proc.ic b/storage/xtradb/include/os0proc.ic
new file mode 100644
index 00000000000..9f1fb01866d
--- /dev/null
+++ b/storage/xtradb/include/os0proc.ic
@@ -0,0 +1,26 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The interface to the operating system
+process control primitives
+
+Created 9/30/1995 Heikki Tuuri
+*******************************************************/
+
+
diff --git a/storage/xtradb/include/os0sync.h b/storage/xtradb/include/os0sync.h
new file mode 100644
index 00000000000..7e058266762
--- /dev/null
+++ b/storage/xtradb/include/os0sync.h
@@ -0,0 +1,309 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+Copyright (c) 2008, Google Inc.
+
+Portions of this file contain modifications contributed and copyrighted by
+Google, Inc. Those modifications are gratefully acknowledged and are described
+briefly in the InnoDB documentation. The contributions by Google are
+incorporated with their permission, and subject to the conditions contained in
+the file COPYING.Google.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The interface to the operating system
+synchronization primitives.
+
+Created 9/6/1995 Heikki Tuuri
+*******************************************************/
+
+#ifndef os0sync_h
+#define os0sync_h
+
+#include "univ.i"
+#include "ut0lst.h"
+
+#ifdef __WIN__
+
+#define os_fast_mutex_t CRITICAL_SECTION
+
+typedef HANDLE		os_native_event_t;
+
+typedef struct os_event_struct	os_event_struct_t;
+typedef os_event_struct_t*	os_event_t;
+
+struct os_event_struct {
+	os_native_event_t		  handle;
+					/* Windows event */
+	UT_LIST_NODE_T(os_event_struct_t) os_event_list;
+					/* list of all created events */
+};
+#else
+typedef pthread_mutex_t	os_fast_mutex_t;
+
+typedef struct os_event_struct	os_event_struct_t;
+typedef os_event_struct_t*	os_event_t;
+
+struct os_event_struct {
+	os_fast_mutex_t	os_mutex;	/* this mutex protects the next
+					fields */
+	ibool		is_set;		/* this is TRUE when the event is
+					in the signaled state, i.e., a thread
+					does not stop if it tries to wait for
+					this event */
+	ib_int64_t	signal_count;	/* this is incremented each time
+					the event becomes signaled */
+	pthread_cond_t	cond_var;	/* condition variable is used in
+					waiting for the event */
+	UT_LIST_NODE_T(os_event_struct_t) os_event_list;
+					/* list of all created events */
+};
+#endif
+
+typedef struct os_mutex_struct	os_mutex_str_t;
+typedef os_mutex_str_t*		os_mutex_t;
+
+#define OS_SYNC_INFINITE_TIME	((ulint)(-1))
+
+#define OS_SYNC_TIME_EXCEEDED	1
+
+/* Mutex protecting counts and the event and OS 'slow' mutex lists */
+extern os_mutex_t	os_sync_mutex;
+
+/* This is incremented by 1 in os_thread_create and decremented by 1 in
+os_thread_exit */
+extern ulint		os_thread_count;
+
+extern ulint		os_event_count;
+extern ulint		os_mutex_count;
+extern ulint		os_fast_mutex_count;
+
+/*************************************************************
+Initializes global event and OS 'slow' mutex lists. */
+UNIV_INTERN
+void
+os_sync_init(void);
+/*==============*/
+/*************************************************************
+Frees created events and OS 'slow' mutexes. */
+UNIV_INTERN
+void
+os_sync_free(void);
+/*==============*/
+/*************************************************************
+Creates an event semaphore, i.e., a semaphore which may just have two states:
+signaled and nonsignaled. The created event is manual reset: it must be reset
+explicitly by calling sync_os_reset_event. */
+UNIV_INTERN
+os_event_t
+os_event_create(
+/*============*/
+				/* out: the event handle */
+	const char*	name);	/* in: the name of the event, if NULL
+				the event is created without a name */
+#ifdef __WIN__
+/*************************************************************
+Creates an auto-reset event semaphore, i.e., an event which is automatically
+reset when a single thread is released. Works only in Windows. */
+UNIV_INTERN
+os_event_t
+os_event_create_auto(
+/*=================*/
+				/* out: the event handle */
+	const char*	name);	/* in: the name of the event, if NULL
+				the event is created without a name */
+#endif
+/**************************************************************
+Sets an event semaphore to the signaled state: lets waiting threads
+proceed. */
+UNIV_INTERN
+void
+os_event_set(
+/*=========*/
+	os_event_t	event);	/* in: event to set */
+/**************************************************************
+Resets an event semaphore to the nonsignaled state. Waiting threads will
+stop to wait for the event.
+The return value should be passed to os_even_wait_low() if it is desired
+that this thread should not wait in case of an intervening call to
+os_event_set() between this os_event_reset() and the
+os_event_wait_low() call. See comments for os_event_wait_low(). */
+UNIV_INTERN
+ib_int64_t
+os_event_reset(
+/*===========*/
+	os_event_t	event);	/* in: event to reset */
+/**************************************************************
+Frees an event object. */
+UNIV_INTERN
+void
+os_event_free(
+/*==========*/
+	os_event_t	event);	/* in: event to free */
+
+/**************************************************************
+Waits for an event object until it is in the signaled state. If
+srv_shutdown_state == SRV_SHUTDOWN_EXIT_THREADS this also exits the
+waiting thread when the event becomes signaled (or immediately if the
+event is already in the signaled state).
+
+Typically, if the event has been signalled after the os_event_reset()
+we'll return immediately because event->is_set == TRUE.
+There are, however, situations (e.g.: sync_array code) where we may
+lose this information. For example:
+
+thread A calls os_event_reset()
+thread B calls os_event_set()   [event->is_set == TRUE]
+thread C calls os_event_reset() [event->is_set == FALSE]
+thread A calls os_event_wait()  [infinite wait!]
+thread C calls os_event_wait()  [infinite wait!]
+
+Where such a scenario is possible, to avoid infinite wait, the
+value returned by os_event_reset() should be passed in as
+reset_sig_count. */
+UNIV_INTERN
+void
+os_event_wait_low(
+/*==============*/
+	os_event_t	event,		/* in: event to wait */
+	ib_int64_t	reset_sig_count);/* in: zero or the value
+					returned by previous call of
+					os_event_reset(). */
+
+#define os_event_wait(event) os_event_wait_low(event, 0)
+
+/**************************************************************
+Waits for an event object until it is in the signaled state or
+a timeout is exceeded. In Unix the timeout is always infinite. */
+UNIV_INTERN
+ulint
+os_event_wait_time(
+/*===============*/
+				/* out: 0 if success,
+				OS_SYNC_TIME_EXCEEDED if timeout
+				was exceeded */
+	os_event_t	event,	/* in: event to wait */
+	ulint		time);	/* in: timeout in microseconds, or
+				OS_SYNC_INFINITE_TIME */
+#ifdef __WIN__
+/**************************************************************
+Waits for any event in an OS native event array. Returns if even a single
+one is signaled or becomes signaled. */
+UNIV_INTERN
+ulint
+os_event_wait_multiple(
+/*===================*/
+					/* out: index of the event
+					which was signaled */
+	ulint			n,	/* in: number of events in the
+					array */
+	os_native_event_t*	native_event_array);
+					/* in: pointer to an array of event
+					handles */
+#endif
+/*************************************************************
+Creates an operating system mutex semaphore. Because these are slow, the
+mutex semaphore of InnoDB itself (mutex_t) should be used where possible. */
+UNIV_INTERN
+os_mutex_t
+os_mutex_create(
+/*============*/
+				/* out: the mutex handle */
+	const char*	name);	/* in: the name of the mutex, if NULL
+				the mutex is created without a name */
+/**************************************************************
+Acquires ownership of a mutex semaphore. */
+UNIV_INTERN
+void
+os_mutex_enter(
+/*===========*/
+	os_mutex_t	mutex);	/* in: mutex to acquire */
+/**************************************************************
+Releases ownership of a mutex. */
+UNIV_INTERN
+void
+os_mutex_exit(
+/*==========*/
+	os_mutex_t	mutex);	/* in: mutex to release */
+/**************************************************************
+Frees an mutex object. */
+UNIV_INTERN
+void
+os_mutex_free(
+/*==========*/
+	os_mutex_t	mutex);	/* in: mutex to free */
+/**************************************************************
+Acquires ownership of a fast mutex. Currently in Windows this is the same
+as os_fast_mutex_lock! */
+UNIV_INLINE
+ulint
+os_fast_mutex_trylock(
+/*==================*/
+						/* out: 0 if success, != 0 if
+						was reserved by another
+						thread */
+	os_fast_mutex_t*	fast_mutex);	/* in: mutex to acquire */
+/**************************************************************
+Releases ownership of a fast mutex. */
+UNIV_INTERN
+void
+os_fast_mutex_unlock(
+/*=================*/
+	os_fast_mutex_t*	fast_mutex);	/* in: mutex to release */
+/*************************************************************
+Initializes an operating system fast mutex semaphore. */
+UNIV_INTERN
+void
+os_fast_mutex_init(
+/*===============*/
+	os_fast_mutex_t*	fast_mutex);	/* in: fast mutex */
+/**************************************************************
+Acquires ownership of a fast mutex. */
+UNIV_INTERN
+void
+os_fast_mutex_lock(
+/*===============*/
+	os_fast_mutex_t*	fast_mutex);	/* in: mutex to acquire */
+/**************************************************************
+Frees an mutex object. */
+UNIV_INTERN
+void
+os_fast_mutex_free(
+/*===============*/
+	os_fast_mutex_t*	fast_mutex);	/* in: mutex to free */
+
+#ifdef HAVE_GCC_ATOMIC_BUILTINS
+/**************************************************************
+Atomic compare-and-swap for InnoDB. Currently requires GCC atomic builtins.
+Returns true if swapped, ptr is pointer to target, old_val is value to
+compare to, new_val is the value to swap in. */
+#define os_compare_and_swap(ptr, old_val, new_val) \
+	__sync_bool_compare_and_swap(ptr, old_val, new_val)
+
+/**************************************************************
+Atomic increment for InnoDB. Currently requires GCC atomic builtins.
+Returns the resulting value, ptr is pointer to target, amount is the
+amount of increment. */
+#define os_atomic_increment(ptr, amount) \
+	__sync_add_and_fetch(ptr, amount)
+
+#endif /* HAVE_GCC_ATOMIC_BUILTINS */
+
+#ifndef UNIV_NONINL
+#include "os0sync.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/os0sync.ic b/storage/xtradb/include/os0sync.ic
new file mode 100644
index 00000000000..5c03d184c7c
--- /dev/null
+++ b/storage/xtradb/include/os0sync.ic
@@ -0,0 +1,62 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The interface to the operating system synchronization primitives.
+
+Created 9/6/1995 Heikki Tuuri
+*******************************************************/
+
+#ifdef __WIN__
+#include <winbase.h>
+#endif
+
+/**************************************************************
+Acquires ownership of a fast mutex. Currently in Windows this is the same
+as os_fast_mutex_lock! */
+UNIV_INLINE
+ulint
+os_fast_mutex_trylock(
+/*==================*/
+						/* out: 0 if success, != 0 if
+						was reserved by another
+						thread */
+	os_fast_mutex_t*	fast_mutex)	/* in: mutex to acquire */
+{
+#ifdef __WIN__
+	EnterCriticalSection(fast_mutex);
+
+	return(0);
+#else
+#if defined(UNIV_HOTBACKUP) && defined(UNIV_HPUX10)
+	/* Since the hot backup version is standalone, MySQL does not redefine
+	pthread_mutex_trylock for HP-UX-10.20, and consequently we must invert
+	the return value here */
+
+	return((ulint) (1 - pthread_mutex_trylock(fast_mutex)));
+#else
+	/* NOTE that the MySQL my_pthread.h redefines pthread_mutex_trylock
+	so that it returns 0 on success. In the operating system
+	libraries, HP-UX-10.20 follows the old Posix 1003.4a Draft 4 and
+	returns 1 on success (but MySQL remaps that to 0), while Linux,
+	FreeBSD, Solaris, AIX, Tru64 Unix, HP-UX-11.0 return 0 on success. */
+
+	return((ulint) pthread_mutex_trylock(fast_mutex));
+#endif
+#endif
+}
diff --git a/storage/xtradb/include/os0thread.h b/storage/xtradb/include/os0thread.h
new file mode 100644
index 00000000000..863596bfa84
--- /dev/null
+++ b/storage/xtradb/include/os0thread.h
@@ -0,0 +1,158 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The interface to the operating system
+process and thread control primitives
+
+Created 9/8/1995 Heikki Tuuri
+*******************************************************/
+
+#ifndef os0thread_h
+#define os0thread_h
+
+#include "univ.i"
+
+/* Maximum number of threads which can be created in the program;
+this is also the size of the wait slot array for MySQL threads which
+can wait inside InnoDB */
+
+#define	OS_THREAD_MAX_N		srv_max_n_threads
+
+
+/* Possible fixed priorities for threads */
+#define OS_THREAD_PRIORITY_NONE		100
+#define OS_THREAD_PRIORITY_BACKGROUND	1
+#define OS_THREAD_PRIORITY_NORMAL	2
+#define OS_THREAD_PRIORITY_ABOVE_NORMAL	3
+
+#ifdef __WIN__
+typedef void*			os_thread_t;
+typedef ulint			os_thread_id_t;	/* In Windows the thread id
+						is an unsigned long int */
+#else
+typedef pthread_t		os_thread_t;
+typedef os_thread_t		os_thread_id_t;	/* In Unix we use the thread
+						handle itself as the id of
+						the thread */
+#endif
+
+/* Define a function pointer type to use in a typecast */
+typedef void* (*os_posix_f_t) (void*);
+
+/*******************************************************************
+Compares two thread ids for equality. */
+UNIV_INTERN
+ibool
+os_thread_eq(
+/*=========*/
+				/* out: TRUE if equal */
+	os_thread_id_t	a,	/* in: OS thread or thread id */
+	os_thread_id_t	b);	/* in: OS thread or thread id */
+/********************************************************************
+Converts an OS thread id to a ulint. It is NOT guaranteed that the ulint is
+unique for the thread though! */
+UNIV_INTERN
+ulint
+os_thread_pf(
+/*=========*/
+				/* out: unsigned long int */
+	os_thread_id_t	a);	/* in: thread or thread id */
+/********************************************************************
+Creates a new thread of execution. The execution starts from
+the function given. The start function takes a void* parameter
+and returns a ulint.
+NOTE: We count the number of threads in os_thread_exit(). A created
+thread should always use that to exit and not use return() to exit. */
+UNIV_INTERN
+os_thread_t
+os_thread_create(
+/*=============*/
+						/* out: handle to the thread */
+#ifndef __WIN__
+		 os_posix_f_t		 start_f,
+#else
+	ulint (*start_f)(void*),		/* in: pointer to function
+						from which to start */
+#endif
+	void*			arg,		/* in: argument to start
+						function */
+	os_thread_id_t*		thread_id);	/* out: id of the created
+						thread, or NULL */
+
+/*********************************************************************
+Exits the current thread. */
+UNIV_INTERN
+void
+os_thread_exit(
+/*===========*/
+	void*	exit_value);	/* in: exit value; in Windows this void*
+				is cast as a DWORD */
+/*********************************************************************
+Returns the thread identifier of current thread. */
+UNIV_INTERN
+os_thread_id_t
+os_thread_get_curr_id(void);
+/*========================*/
+/*********************************************************************
+Returns handle to the current thread. */
+UNIV_INTERN
+os_thread_t
+os_thread_get_curr(void);
+/*====================*/
+/*********************************************************************
+Advises the os to give up remainder of the thread's time slice. */
+UNIV_INTERN
+void
+os_thread_yield(void);
+/*=================*/
+/*********************************************************************
+The thread sleeps at least the time given in microseconds. */
+UNIV_INTERN
+void
+os_thread_sleep(
+/*============*/
+	ulint	tm);	/* in: time in microseconds */
+/**********************************************************************
+Gets a thread priority. */
+UNIV_INTERN
+ulint
+os_thread_get_priority(
+/*===================*/
+				/* out: priority */
+	os_thread_t	handle);/* in: OS handle to the thread */
+/**********************************************************************
+Sets a thread priority. */
+UNIV_INTERN
+void
+os_thread_set_priority(
+/*===================*/
+	os_thread_t	handle,	/* in: OS handle to the thread */
+	ulint		pri);	/* in: priority: one of OS_PRIORITY_... */
+/**********************************************************************
+Gets the last operating system error code for the calling thread. */
+UNIV_INTERN
+ulint
+os_thread_get_last_error(void);
+/*==========================*/
+
+#ifndef UNIV_NONINL
+#include "os0thread.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/os0thread.ic b/storage/xtradb/include/os0thread.ic
new file mode 100644
index 00000000000..a86b203809c
--- /dev/null
+++ b/storage/xtradb/include/os0thread.ic
@@ -0,0 +1,24 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The interface to the operating system
+process and thread control primitives
+
+Created 9/8/1995 Heikki Tuuri
+*******************************************************/
diff --git a/storage/xtradb/include/page0cur.h b/storage/xtradb/include/page0cur.h
new file mode 100644
index 00000000000..960ecdddf4e
--- /dev/null
+++ b/storage/xtradb/include/page0cur.h
@@ -0,0 +1,346 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/************************************************************************
+The page cursor
+
+Created 10/4/1994 Heikki Tuuri
+*************************************************************************/
+
+#ifndef page0cur_h
+#define page0cur_h
+
+#include "univ.i"
+
+#include "buf0types.h"
+#include "page0page.h"
+#include "rem0rec.h"
+#include "data0data.h"
+#include "mtr0mtr.h"
+
+
+#define PAGE_CUR_ADAPT
+
+/* Page cursor search modes; the values must be in this order! */
+
+#define	PAGE_CUR_UNSUPP	0
+#define	PAGE_CUR_G	1
+#define	PAGE_CUR_GE	2
+#define	PAGE_CUR_L	3
+#define	PAGE_CUR_LE	4
+/*#define PAGE_CUR_LE_OR_EXTENDS 5*/ /* This is a search mode used in
+				 "column LIKE 'abc%' ORDER BY column DESC";
+				 we have to find strings which are <= 'abc' or
+				 which extend it */
+#ifdef UNIV_SEARCH_DEBUG
+# define PAGE_CUR_DBG	6	/* As PAGE_CUR_LE, but skips search shortcut */
+#endif /* UNIV_SEARCH_DEBUG */
+
+#ifdef UNIV_DEBUG
+/*************************************************************
+Gets pointer to the page frame where the cursor is positioned. */
+UNIV_INLINE
+page_t*
+page_cur_get_page(
+/*==============*/
+				/* out: page */
+	page_cur_t*	cur);	/* in: page cursor */
+/*************************************************************
+Gets pointer to the buffer block where the cursor is positioned. */
+UNIV_INLINE
+buf_block_t*
+page_cur_get_block(
+/*===============*/
+				/* out: page */
+	page_cur_t*	cur);	/* in: page cursor */
+/*************************************************************
+Gets pointer to the page frame where the cursor is positioned. */
+UNIV_INLINE
+page_zip_des_t*
+page_cur_get_page_zip(
+/*==================*/
+				/* out: page */
+	page_cur_t*	cur);	/* in: page cursor */
+/*************************************************************
+Gets the record where the cursor is positioned. */
+UNIV_INLINE
+rec_t*
+page_cur_get_rec(
+/*=============*/
+				/* out: record */
+	page_cur_t*	cur);	/* in: page cursor */
+#else /* UNIV_DEBUG */
+# define page_cur_get_page(cur)		page_align((cur)->rec)
+# define page_cur_get_block(cur)	(cur)->block
+# define page_cur_get_page_zip(cur)	buf_block_get_page_zip((cur)->block)
+# define page_cur_get_rec(cur)		(cur)->rec
+#endif /* UNIV_DEBUG */
+/*************************************************************
+Sets the cursor object to point before the first user record
+on the page. */
+UNIV_INLINE
+void
+page_cur_set_before_first(
+/*======================*/
+	const buf_block_t*	block,	/* in: index page */
+	page_cur_t*		cur);	/* in: cursor */
+/*************************************************************
+Sets the cursor object to point after the last user record on
+the page. */
+UNIV_INLINE
+void
+page_cur_set_after_last(
+/*====================*/
+	const buf_block_t*	block,	/* in: index page */
+	page_cur_t*		cur);	/* in: cursor */
+/*************************************************************
+Returns TRUE if the cursor is before first user record on page. */
+UNIV_INLINE
+ibool
+page_cur_is_before_first(
+/*=====================*/
+					/* out: TRUE if at start */
+	const page_cur_t*	cur);	/* in: cursor */
+/*************************************************************
+Returns TRUE if the cursor is after last user record. */
+UNIV_INLINE
+ibool
+page_cur_is_after_last(
+/*===================*/
+					/* out: TRUE if at end */
+	const page_cur_t*	cur);	/* in: cursor */
+/**************************************************************
+Positions the cursor on the given record. */
+UNIV_INLINE
+void
+page_cur_position(
+/*==============*/
+	const rec_t*		rec,	/* in: record on a page */
+	const buf_block_t*	block,	/* in: buffer block containing
+					the record */
+	page_cur_t*		cur);	/* out: page cursor */
+/**************************************************************
+Invalidates a page cursor by setting the record pointer NULL. */
+UNIV_INLINE
+void
+page_cur_invalidate(
+/*================*/
+	page_cur_t*	cur);	/* out: page cursor */
+/**************************************************************
+Moves the cursor to the next record on page. */
+UNIV_INLINE
+void
+page_cur_move_to_next(
+/*==================*/
+	page_cur_t*	cur);	/* in/out: cursor; must not be after last */
+/**************************************************************
+Moves the cursor to the previous record on page. */
+UNIV_INLINE
+void
+page_cur_move_to_prev(
+/*==================*/
+	page_cur_t*	cur);	/* in/out: cursor; not before first */
+/***************************************************************
+Inserts a record next to page cursor. Returns pointer to inserted record if
+succeed, i.e., enough space available, NULL otherwise. The cursor stays at
+the same logical position, but the physical position may change if it is
+pointing to a compressed page that was reorganized. */
+UNIV_INLINE
+rec_t*
+page_cur_tuple_insert(
+/*==================*/
+				/* out: pointer to record if succeed, NULL
+				otherwise */
+	page_cur_t*	cursor,	/* in/out: a page cursor */
+	const dtuple_t*	tuple,	/* in: pointer to a data tuple */
+	dict_index_t*	index,	/* in: record descriptor */
+	ulint		n_ext,	/* in: number of externally stored columns */
+	mtr_t*		mtr);	/* in: mini-transaction handle, or NULL */
+/***************************************************************
+Inserts a record next to page cursor. Returns pointer to inserted record if
+succeed, i.e., enough space available, NULL otherwise. The cursor stays at
+the same logical position, but the physical position may change if it is
+pointing to a compressed page that was reorganized. */
+UNIV_INLINE
+rec_t*
+page_cur_rec_insert(
+/*================*/
+				/* out: pointer to record if succeed, NULL
+				otherwise */
+	page_cur_t*	cursor,	/* in/out: a page cursor */
+	const rec_t*	rec,	/* in: record to insert */
+	dict_index_t*	index,	/* in: record descriptor */
+	ulint*		offsets,/* in/out: rec_get_offsets(rec, index) */
+	mtr_t*		mtr);	/* in: mini-transaction handle, or NULL */
+/***************************************************************
+Inserts a record next to page cursor on an uncompressed page.
+Returns pointer to inserted record if succeed, i.e., enough
+space available, NULL otherwise. The cursor stays at the same position. */
+UNIV_INTERN
+rec_t*
+page_cur_insert_rec_low(
+/*====================*/
+				/* out: pointer to record if succeed, NULL
+				otherwise */
+	rec_t*		current_rec,/* in: pointer to current record after
+				which the new record is inserted */
+	dict_index_t*	index,	/* in: record descriptor */
+	const rec_t*	rec,	/* in: pointer to a physical record */
+	ulint*		offsets,/* in/out: rec_get_offsets(rec, index) */
+	mtr_t*		mtr);	/* in: mini-transaction handle, or NULL */
+/***************************************************************
+Inserts a record next to page cursor on a compressed and uncompressed
+page. Returns pointer to inserted record if succeed, i.e.,
+enough space available, NULL otherwise.
+The cursor stays at the same position. */
+UNIV_INTERN
+rec_t*
+page_cur_insert_rec_zip(
+/*====================*/
+				/* out: pointer to record if succeed, NULL
+				otherwise */
+	rec_t**		current_rec,/* in/out: pointer to current record after
+				which the new record is inserted */
+	buf_block_t*	block,	/* in: buffer block of *current_rec */
+	dict_index_t*	index,	/* in: record descriptor */
+	const rec_t*	rec,	/* in: pointer to a physical record */
+	ulint*		offsets,/* in/out: rec_get_offsets(rec, index) */
+	mtr_t*		mtr);	/* in: mini-transaction handle, or NULL */
+/*****************************************************************
+Copies records from page to a newly created page, from a given record onward,
+including that record. Infimum and supremum records are not copied. */
+UNIV_INTERN
+void
+page_copy_rec_list_end_to_created_page(
+/*===================================*/
+	page_t*		new_page,	/* in/out: index page to copy to */
+	rec_t*		rec,		/* in: first record to copy */
+	dict_index_t*	index,		/* in: record descriptor */
+	mtr_t*		mtr);		/* in: mtr */
+/***************************************************************
+Deletes a record at the page cursor. The cursor is moved to the
+next record after the deleted one. */
+UNIV_INTERN
+void
+page_cur_delete_rec(
+/*================*/
+	page_cur_t*	cursor,	/* in/out: a page cursor */
+	dict_index_t*	index,	/* in: record descriptor */
+	const ulint*	offsets,/* in: rec_get_offsets(cursor->rec, index) */
+	mtr_t*		mtr);	/* in: mini-transaction handle */
+/********************************************************************
+Searches the right position for a page cursor. */
+UNIV_INLINE
+ulint
+page_cur_search(
+/*============*/
+					/* out: number of matched
+					fields on the left */
+	const buf_block_t*	block,	/* in: buffer block */
+	const dict_index_t*	index,	/* in: record descriptor */
+	const dtuple_t*		tuple,	/* in: data tuple */
+	ulint			mode,	/* in: PAGE_CUR_L,
+					PAGE_CUR_LE, PAGE_CUR_G, or
+					PAGE_CUR_GE */
+	page_cur_t*		cursor);/* out: page cursor */
+/********************************************************************
+Searches the right position for a page cursor. */
+UNIV_INTERN
+void
+page_cur_search_with_match(
+/*=======================*/
+	const buf_block_t*	block,	/* in: buffer block */
+	const dict_index_t*	index,	/* in: record descriptor */
+	const dtuple_t*		tuple,	/* in: data tuple */
+	ulint			mode,	/* in: PAGE_CUR_L,
+					PAGE_CUR_LE, PAGE_CUR_G, or
+					PAGE_CUR_GE */
+	ulint*			iup_matched_fields,
+					/* in/out: already matched
+					fields in upper limit record */
+	ulint*			iup_matched_bytes,
+					/* in/out: already matched
+					bytes in a field not yet
+					completely matched */
+	ulint*			ilow_matched_fields,
+					/* in/out: already matched
+					fields in lower limit record */
+	ulint*			ilow_matched_bytes,
+					/* in/out: already matched
+					bytes in a field not yet
+					completely matched */
+	page_cur_t*		cursor);/* out: page cursor */
+/***************************************************************
+Positions a page cursor on a randomly chosen user record on a page. If there
+are no user records, sets the cursor on the infimum record. */
+UNIV_INTERN
+void
+page_cur_open_on_rnd_user_rec(
+/*==========================*/
+	buf_block_t*	block,	/* in: page */
+	page_cur_t*	cursor);/* out: page cursor */
+/***************************************************************
+Parses a log record of a record insert on a page. */
+UNIV_INTERN
+byte*
+page_cur_parse_insert_rec(
+/*======================*/
+				/* out: end of log record or NULL */
+	ibool		is_short,/* in: TRUE if short inserts */
+	byte*		ptr,	/* in: buffer */
+	byte*		end_ptr,/* in: buffer end */
+	buf_block_t*	block,	/* in: page or NULL */
+	dict_index_t*	index,	/* in: record descriptor */
+	mtr_t*		mtr);	/* in: mtr or NULL */
+/**************************************************************
+Parses a log record of copying a record list end to a new created page. */
+UNIV_INTERN
+byte*
+page_parse_copy_rec_list_to_created_page(
+/*=====================================*/
+				/* out: end of log record or NULL */
+	byte*		ptr,	/* in: buffer */
+	byte*		end_ptr,/* in: buffer end */
+	buf_block_t*	block,	/* in: page or NULL */
+	dict_index_t*	index,	/* in: record descriptor */
+	mtr_t*		mtr);	/* in: mtr or NULL */
+/***************************************************************
+Parses log record of a record delete on a page. */
+UNIV_INTERN
+byte*
+page_cur_parse_delete_rec(
+/*======================*/
+				/* out: pointer to record end or NULL */
+	byte*		ptr,	/* in: buffer */
+	byte*		end_ptr,/* in: buffer end */
+	buf_block_t*	block,	/* in: page or NULL */
+	dict_index_t*	index,	/* in: record descriptor */
+	mtr_t*		mtr);	/* in: mtr or NULL */
+
+/* Index page cursor */
+
+struct page_cur_struct{
+	byte*		rec;	/* pointer to a record on page */
+	buf_block_t*	block;	/* pointer to the block containing rec */
+};
+
+#ifndef UNIV_NONINL
+#include "page0cur.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/page0cur.ic b/storage/xtradb/include/page0cur.ic
new file mode 100644
index 00000000000..9cf10ea5e3f
--- /dev/null
+++ b/storage/xtradb/include/page0cur.ic
@@ -0,0 +1,300 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/************************************************************************
+The page cursor
+
+Created 10/4/1994 Heikki Tuuri
+*************************************************************************/
+
+#include "page0page.h"
+#include "buf0types.h"
+
+#ifdef UNIV_DEBUG
+/*************************************************************
+Gets pointer to the page frame where the cursor is positioned. */
+UNIV_INLINE
+page_t*
+page_cur_get_page(
+/*==============*/
+				/* out: page */
+	page_cur_t*	cur)	/* in: page cursor */
+{
+	ut_ad(cur);
+	ut_ad(page_align(cur->rec) == cur->block->frame);
+
+	return(page_align(cur->rec));
+}
+
+/*************************************************************
+Gets pointer to the buffer block where the cursor is positioned. */
+UNIV_INLINE
+buf_block_t*
+page_cur_get_block(
+/*===============*/
+				/* out: page */
+	page_cur_t*	cur)	/* in: page cursor */
+{
+	ut_ad(cur);
+	ut_ad(page_align(cur->rec) == cur->block->frame);
+	return(cur->block);
+}
+
+/*************************************************************
+Gets pointer to the page frame where the cursor is positioned. */
+UNIV_INLINE
+page_zip_des_t*
+page_cur_get_page_zip(
+/*==================*/
+				/* out: page */
+	page_cur_t*	cur)	/* in: page cursor */
+{
+	return(buf_block_get_page_zip(page_cur_get_block(cur)));
+}
+
+/*************************************************************
+Gets the record where the cursor is positioned. */
+UNIV_INLINE
+rec_t*
+page_cur_get_rec(
+/*=============*/
+				/* out: record */
+	page_cur_t*	cur)	/* in: page cursor */
+{
+	ut_ad(cur);
+	ut_ad(page_align(cur->rec) == cur->block->frame);
+
+	return(cur->rec);
+}
+#endif /* UNIV_DEBUG */
+
+/*************************************************************
+Sets the cursor object to point before the first user record
+on the page. */
+UNIV_INLINE
+void
+page_cur_set_before_first(
+/*======================*/
+	const buf_block_t*	block,	/* in: index page */
+	page_cur_t*		cur)	/* in: cursor */
+{
+	cur->block = (buf_block_t*) block;
+	cur->rec = page_get_infimum_rec(buf_block_get_frame(cur->block));
+}
+
+/*************************************************************
+Sets the cursor object to point after the last user record on
+the page. */
+UNIV_INLINE
+void
+page_cur_set_after_last(
+/*====================*/
+	const buf_block_t*	block,	/* in: index page */
+	page_cur_t*		cur)	/* in: cursor */
+{
+	cur->block = (buf_block_t*) block;
+	cur->rec = page_get_supremum_rec(buf_block_get_frame(cur->block));
+}
+
+/*************************************************************
+Returns TRUE if the cursor is before first user record on page. */
+UNIV_INLINE
+ibool
+page_cur_is_before_first(
+/*=====================*/
+					/* out: TRUE if at start */
+	const page_cur_t*	cur)	/* in: cursor */
+{
+	ut_ad(cur);
+	ut_ad(page_align(cur->rec) == cur->block->frame);
+	return(page_rec_is_infimum(cur->rec));
+}
+
+/*************************************************************
+Returns TRUE if the cursor is after last user record. */
+UNIV_INLINE
+ibool
+page_cur_is_after_last(
+/*===================*/
+					/* out: TRUE if at end */
+	const page_cur_t*	cur)	/* in: cursor */
+{
+	ut_ad(cur);
+	ut_ad(page_align(cur->rec) == cur->block->frame);
+	return(page_rec_is_supremum(cur->rec));
+}
+
+/**************************************************************
+Positions the cursor on the given record. */
+UNIV_INLINE
+void
+page_cur_position(
+/*==============*/
+	const rec_t*		rec,	/* in: record on a page */
+	const buf_block_t*	block,	/* in: buffer block containing
+					the record */
+	page_cur_t*		cur)	/* out: page cursor */
+{
+	ut_ad(rec && block && cur);
+	ut_ad(page_align(rec) == block->frame);
+
+	cur->rec = (rec_t*) rec;
+	cur->block = (buf_block_t*) block;
+}
+
+/**************************************************************
+Invalidates a page cursor by setting the record pointer NULL. */
+UNIV_INLINE
+void
+page_cur_invalidate(
+/*================*/
+	page_cur_t*	cur)	/* out: page cursor */
+{
+	ut_ad(cur);
+
+	cur->rec = NULL;
+	cur->block = NULL;
+}
+
+/**************************************************************
+Moves the cursor to the next record on page. */
+UNIV_INLINE
+void
+page_cur_move_to_next(
+/*==================*/
+	page_cur_t*	cur)	/* in/out: cursor; must not be after last */
+{
+	ut_ad(!page_cur_is_after_last(cur));
+
+	cur->rec = page_rec_get_next(cur->rec);
+}
+
+/**************************************************************
+Moves the cursor to the previous record on page. */
+UNIV_INLINE
+void
+page_cur_move_to_prev(
+/*==================*/
+	page_cur_t*	cur)	/* in/out: page cursor, not before first */
+{
+	ut_ad(!page_cur_is_before_first(cur));
+
+	cur->rec = page_rec_get_prev(cur->rec);
+}
+
+/********************************************************************
+Searches the right position for a page cursor. */
+UNIV_INLINE
+ulint
+page_cur_search(
+/*============*/
+					/* out: number of matched
+					fields on the left */
+	const buf_block_t*	block,	/* in: buffer block */
+	const dict_index_t*	index,	/* in: record descriptor */
+	const dtuple_t*		tuple,	/* in: data tuple */
+	ulint			mode,	/* in: PAGE_CUR_L,
+					PAGE_CUR_LE, PAGE_CUR_G, or
+					PAGE_CUR_GE */
+	page_cur_t*		cursor)	/* out: page cursor */
+{
+	ulint		low_matched_fields = 0;
+	ulint		low_matched_bytes = 0;
+	ulint		up_matched_fields = 0;
+	ulint		up_matched_bytes = 0;
+
+	ut_ad(dtuple_check_typed(tuple));
+
+	page_cur_search_with_match(block, index, tuple, mode,
+				   &up_matched_fields,
+				   &up_matched_bytes,
+				   &low_matched_fields,
+				   &low_matched_bytes,
+				   cursor);
+	return(low_matched_fields);
+}
+
+/***************************************************************
+Inserts a record next to page cursor. Returns pointer to inserted record if
+succeed, i.e., enough space available, NULL otherwise. The cursor stays at
+the same logical position, but the physical position may change if it is
+pointing to a compressed page that was reorganized. */
+UNIV_INLINE
+rec_t*
+page_cur_tuple_insert(
+/*==================*/
+				/* out: pointer to record if succeed, NULL
+				otherwise */
+	page_cur_t*	cursor,	/* in/out: a page cursor */
+	const dtuple_t*	tuple,	/* in: pointer to a data tuple */
+	dict_index_t*	index,	/* in: record descriptor */
+	ulint		n_ext,	/* in: number of externally stored columns */
+	mtr_t*		mtr)	/* in: mini-transaction handle, or NULL */
+{
+	mem_heap_t*	heap;
+	ulint*		offsets;
+	ulint		size
+		= rec_get_converted_size(index, tuple, n_ext);
+	rec_t*		rec;
+
+	heap = mem_heap_create(size
+			       + (4 + REC_OFFS_HEADER_SIZE
+				  + dtuple_get_n_fields(tuple))
+			       * sizeof *offsets);
+	rec = rec_convert_dtuple_to_rec((byte*) mem_heap_alloc(heap, size),
+					index, tuple, n_ext);
+	offsets = rec_get_offsets(rec, index, NULL, ULINT_UNDEFINED, &heap);
+
+	if (buf_block_get_page_zip(cursor->block)) {
+		rec = page_cur_insert_rec_zip(&cursor->rec, cursor->block,
+					      index, rec, offsets, mtr);
+	} else {
+		rec = page_cur_insert_rec_low(cursor->rec,
+					      index, rec, offsets, mtr);
+	}
+
+	mem_heap_free(heap);
+	return(rec);
+}
+
+/***************************************************************
+Inserts a record next to page cursor. Returns pointer to inserted record if
+succeed, i.e., enough space available, NULL otherwise. The cursor stays at
+the same logical position, but the physical position may change if it is
+pointing to a compressed page that was reorganized. */
+UNIV_INLINE
+rec_t*
+page_cur_rec_insert(
+/*================*/
+				/* out: pointer to record if succeed, NULL
+				otherwise */
+	page_cur_t*	cursor,	/* in/out: a page cursor */
+	const rec_t*	rec,	/* in: record to insert */
+	dict_index_t*	index,	/* in: record descriptor */
+	ulint*		offsets,/* in/out: rec_get_offsets(rec, index) */
+	mtr_t*		mtr)	/* in: mini-transaction handle, or NULL */
+{
+	if (buf_block_get_page_zip(cursor->block)) {
+		return(page_cur_insert_rec_zip(&cursor->rec, cursor->block,
+					       index, rec, offsets, mtr));
+	} else {
+		return(page_cur_insert_rec_low(cursor->rec,
+					       index, rec, offsets, mtr));
+	}
+}
+
diff --git a/storage/xtradb/include/page0page.h b/storage/xtradb/include/page0page.h
new file mode 100644
index 00000000000..e3de6901ee1
--- /dev/null
+++ b/storage/xtradb/include/page0page.h
@@ -0,0 +1,1019 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Index page routines
+
+Created 2/2/1994 Heikki Tuuri
+*******************************************************/
+
+#ifndef page0page_h
+#define page0page_h
+
+#include "univ.i"
+
+#include "page0types.h"
+#include "fil0fil.h"
+#include "buf0buf.h"
+#include "data0data.h"
+#include "dict0dict.h"
+#include "rem0rec.h"
+#include "fsp0fsp.h"
+#include "mtr0mtr.h"
+
+#ifdef UNIV_MATERIALIZE
+#undef UNIV_INLINE
+#define UNIV_INLINE
+#endif
+
+/*			PAGE HEADER
+			===========
+
+Index page header starts at the first offset left free by the FIL-module */
+
+typedef	byte		page_header_t;
+
+#define	PAGE_HEADER	FSEG_PAGE_DATA	/* index page header starts at this
+				offset */
+/*-----------------------------*/
+#define PAGE_N_DIR_SLOTS 0	/* number of slots in page directory */
+#define	PAGE_HEAP_TOP	 2	/* pointer to record heap top */
+#define	PAGE_N_HEAP	 4	/* number of records in the heap,
+				bit 15=flag: new-style compact page format */
+#define	PAGE_FREE	 6	/* pointer to start of page free record list */
+#define	PAGE_GARBAGE	 8	/* number of bytes in deleted records */
+#define	PAGE_LAST_INSERT 10	/* pointer to the last inserted record, or
+				NULL if this info has been reset by a delete,
+				for example */
+#define	PAGE_DIRECTION	 12	/* last insert direction: PAGE_LEFT, ... */
+#define	PAGE_N_DIRECTION 14	/* number of consecutive inserts to the same
+				direction */
+#define	PAGE_N_RECS	 16	/* number of user records on the page */
+#define PAGE_MAX_TRX_ID	 18	/* highest id of a trx which may have modified
+				a record on the page; a dulint; defined only
+				in secondary indexes; specifically, not in an
+				ibuf tree; NOTE: this may be modified only
+				when the thread has an x-latch to the page,
+				and ALSO an x-latch to btr_search_latch
+				if there is a hash index to the page! */
+#define PAGE_HEADER_PRIV_END 26	/* end of private data structure of the page
+				header which are set in a page create */
+/*----*/
+#define	PAGE_LEVEL	 26	/* level of the node in an index tree; the
+				leaf level is the level 0 */
+#define	PAGE_INDEX_ID	 28	/* index id where the page belongs */
+#define PAGE_BTR_SEG_LEAF 36	/* file segment header for the leaf pages in
+				a B-tree: defined only on the root page of a
+				B-tree, but not in the root of an ibuf tree */
+#define PAGE_BTR_IBUF_FREE_LIST	PAGE_BTR_SEG_LEAF
+#define PAGE_BTR_IBUF_FREE_LIST_NODE PAGE_BTR_SEG_LEAF
+				/* in the place of PAGE_BTR_SEG_LEAF and _TOP
+				there is a free list base node if the page is
+				the root page of an ibuf tree, and at the same
+				place is the free list node if the page is in
+				a free list */
+#define PAGE_BTR_SEG_TOP (36 + FSEG_HEADER_SIZE)
+				/* file segment header for the non-leaf pages
+				in a B-tree: defined only on the root page of
+				a B-tree, but not in the root of an ibuf
+				tree */
+/*----*/
+#define PAGE_DATA	(PAGE_HEADER + 36 + 2 * FSEG_HEADER_SIZE)
+				/* start of data on the page */
+
+#define PAGE_OLD_INFIMUM	(PAGE_DATA + 1 + REC_N_OLD_EXTRA_BYTES)
+				/* offset of the page infimum record on an
+				old-style page */
+#define PAGE_OLD_SUPREMUM	(PAGE_DATA + 2 + 2 * REC_N_OLD_EXTRA_BYTES + 8)
+				/* offset of the page supremum record on an
+				old-style page */
+#define PAGE_OLD_SUPREMUM_END (PAGE_OLD_SUPREMUM + 9)
+				/* offset of the page supremum record end on
+				an old-style page */
+#define PAGE_NEW_INFIMUM	(PAGE_DATA + REC_N_NEW_EXTRA_BYTES)
+				/* offset of the page infimum record on a
+				new-style compact page */
+#define PAGE_NEW_SUPREMUM	(PAGE_DATA + 2 * REC_N_NEW_EXTRA_BYTES + 8)
+				/* offset of the page supremum record on a
+				new-style compact page */
+#define PAGE_NEW_SUPREMUM_END (PAGE_NEW_SUPREMUM + 8)
+				/* offset of the page supremum record end on
+				a new-style compact page */
+/*-----------------------------*/
+
+/* Heap numbers */
+#define PAGE_HEAP_NO_INFIMUM	0	/* page infimum */
+#define PAGE_HEAP_NO_SUPREMUM	1	/* page supremum */
+#define PAGE_HEAP_NO_USER_LOW	2	/* first user record in
+					creation (insertion) order,
+					not necessarily collation order;
+					this record may have been deleted */
+
+/* Directions of cursor movement */
+#define	PAGE_LEFT		1
+#define	PAGE_RIGHT		2
+#define	PAGE_SAME_REC		3
+#define	PAGE_SAME_PAGE		4
+#define	PAGE_NO_DIRECTION	5
+
+/*			PAGE DIRECTORY
+			==============
+*/
+
+typedef	byte			page_dir_slot_t;
+typedef page_dir_slot_t		page_dir_t;
+
+/* Offset of the directory start down from the page end. We call the
+slot with the highest file address directory start, as it points to
+the first record in the list of records. */
+#define	PAGE_DIR		FIL_PAGE_DATA_END
+
+/* We define a slot in the page directory as two bytes */
+#define	PAGE_DIR_SLOT_SIZE	2
+
+/* The offset of the physically lower end of the directory, counted from
+page end, when the page is empty */
+#define PAGE_EMPTY_DIR_START	(PAGE_DIR + 2 * PAGE_DIR_SLOT_SIZE)
+
+/* The maximum and minimum number of records owned by a directory slot. The
+number may drop below the minimum in the first and the last slot in the
+directory. */
+#define PAGE_DIR_SLOT_MAX_N_OWNED	8
+#define	PAGE_DIR_SLOT_MIN_N_OWNED	4
+
+/****************************************************************
+Gets the start of a page. */
+UNIV_INLINE
+page_t*
+page_align(
+/*=======*/
+				/* out: start of the page */
+	const void*	ptr)	/* in: pointer to page frame */
+		__attribute__((const));
+/****************************************************************
+Gets the offset within a page. */
+UNIV_INLINE
+ulint
+page_offset(
+/*========*/
+				/* out: offset from the start of the page */
+	const void*	ptr)	/* in: pointer to page frame */
+		__attribute__((const));
+/*****************************************************************
+Returns the max trx id field value. */
+UNIV_INLINE
+dulint
+page_get_max_trx_id(
+/*================*/
+	const page_t*	page);	/* in: page */
+/*****************************************************************
+Sets the max trx id field value. */
+UNIV_INTERN
+void
+page_set_max_trx_id(
+/*================*/
+	buf_block_t*	block,	/* in/out: page */
+	page_zip_des_t*	page_zip,/* in/out: compressed page, or NULL */
+	dulint		trx_id);/* in: transaction id */
+/*****************************************************************
+Sets the max trx id field value if trx_id is bigger than the previous
+value. */
+UNIV_INLINE
+void
+page_update_max_trx_id(
+/*===================*/
+	buf_block_t*	block,	/* in/out: page */
+	page_zip_des_t*	page_zip,/* in/out: compressed page whose
+				uncompressed part will be updated, or NULL */
+	dulint		trx_id);/* in: transaction id */
+/*****************************************************************
+Reads the given header field. */
+UNIV_INLINE
+ulint
+page_header_get_field(
+/*==================*/
+	const page_t*	page,	/* in: page */
+	ulint		field);	/* in: PAGE_N_DIR_SLOTS, ... */
+/*****************************************************************
+Sets the given header field. */
+UNIV_INLINE
+void
+page_header_set_field(
+/*==================*/
+	page_t*		page,	/* in/out: page */
+	page_zip_des_t*	page_zip,/* in/out: compressed page whose
+				uncompressed part will be updated, or NULL */
+	ulint		field,	/* in: PAGE_N_DIR_SLOTS, ... */
+	ulint		val);	/* in: value */
+/*****************************************************************
+Returns the offset stored in the given header field. */
+UNIV_INLINE
+ulint
+page_header_get_offs(
+/*=================*/
+				/* out: offset from the start of the page,
+				or 0 */
+	const page_t*	page,	/* in: page */
+	ulint		field)	/* in: PAGE_FREE, ... */
+	__attribute__((nonnull, pure));
+
+/*****************************************************************
+Returns the pointer stored in the given header field, or NULL. */
+#define page_header_get_ptr(page, field)			\
+	(page_header_get_offs(page, field)			\
+	 ? page + page_header_get_offs(page, field) : NULL)
+/*****************************************************************
+Sets the pointer stored in the given header field. */
+UNIV_INLINE
+void
+page_header_set_ptr(
+/*================*/
+	page_t*		page,	/* in/out: page */
+	page_zip_des_t*	page_zip,/* in/out: compressed page whose
+				uncompressed part will be updated, or NULL */
+	ulint		field,	/* in/out: PAGE_FREE, ... */
+	const byte*	ptr);	/* in: pointer or NULL*/
+/*****************************************************************
+Resets the last insert info field in the page header. Writes to mlog
+about this operation. */
+UNIV_INLINE
+void
+page_header_reset_last_insert(
+/*==========================*/
+	page_t*		page,	/* in: page */
+	page_zip_des_t*	page_zip,/* in/out: compressed page whose
+				uncompressed part will be updated, or NULL */
+	mtr_t*		mtr);	/* in: mtr */
+/****************************************************************
+Gets the offset of the first record on the page. */
+UNIV_INLINE
+ulint
+page_get_infimum_offset(
+/*====================*/
+				/* out: offset of the first record
+				in record list, relative from page */
+	const page_t*	page);	/* in: page which must have record(s) */
+/****************************************************************
+Gets the offset of the last record on the page. */
+UNIV_INLINE
+ulint
+page_get_supremum_offset(
+/*=====================*/
+				/* out: offset of the last record in
+				record list, relative from page */
+	const page_t*	page);	/* in: page which must have record(s) */
+#define page_get_infimum_rec(page) ((page) + page_get_infimum_offset(page))
+#define page_get_supremum_rec(page) ((page) + page_get_supremum_offset(page))
+/****************************************************************
+Returns the middle record of record list. If there are an even number
+of records in the list, returns the first record of upper half-list. */
+UNIV_INTERN
+rec_t*
+page_get_middle_rec(
+/*================*/
+			/* out: middle record */
+	page_t*	page);	/* in: page */
+/*****************************************************************
+Compares a data tuple to a physical record. Differs from the function
+cmp_dtuple_rec_with_match in the way that the record must reside on an
+index page, and also page infimum and supremum records can be given in
+the parameter rec. These are considered as the negative infinity and
+the positive infinity in the alphabetical order. */
+UNIV_INLINE
+int
+page_cmp_dtuple_rec_with_match(
+/*===========================*/
+				/* out: 1, 0, -1, if dtuple is greater, equal,
+				less than rec, respectively, when only the
+				common first fields are compared */
+	const dtuple_t*	dtuple,	/* in: data tuple */
+	const rec_t*	rec,	/* in: physical record on a page; may also
+				be page infimum or supremum, in which case
+				matched-parameter values below are not
+				affected */
+	const ulint*	offsets,/* in: array returned by rec_get_offsets() */
+	ulint*		matched_fields, /* in/out: number of already completely
+				matched fields; when function returns
+				contains the value for current comparison */
+	ulint*		matched_bytes); /* in/out: number of already matched
+				bytes within the first field not completely
+				matched; when function returns contains the
+				value for current comparison */
+/*****************************************************************
+Gets the page number. */
+UNIV_INLINE
+ulint
+page_get_page_no(
+/*=============*/
+				/* out: page number */
+	const page_t*	page);	/* in: page */
+/*****************************************************************
+Gets the tablespace identifier. */
+UNIV_INLINE
+ulint
+page_get_space_id(
+/*==============*/
+				/* out: space id */
+	const page_t*	page);	/* in: page */
+/*****************************************************************
+Gets the number of user records on page (the infimum and supremum records
+are not user records). */
+UNIV_INLINE
+ulint
+page_get_n_recs(
+/*============*/
+				/* out: number of user records */
+	const page_t*	page);	/* in: index page */
+/*******************************************************************
+Returns the number of records before the given record in chain.
+The number includes infimum and supremum records. */
+UNIV_INTERN
+ulint
+page_rec_get_n_recs_before(
+/*=======================*/
+				/* out: number of records */
+	const rec_t*	rec);	/* in: the physical record */
+/*****************************************************************
+Gets the number of records in the heap. */
+UNIV_INLINE
+ulint
+page_dir_get_n_heap(
+/*================*/
+				/* out: number of user records */
+	const page_t*	page);	/* in: index page */
+/*****************************************************************
+Sets the number of records in the heap. */
+UNIV_INLINE
+void
+page_dir_set_n_heap(
+/*================*/
+	page_t*		page,	/* in/out: index page */
+	page_zip_des_t*	page_zip,/* in/out: compressed page whose
+				uncompressed part will be updated, or NULL.
+				Note that the size of the dense page directory
+				in the compressed page trailer is
+				n_heap * PAGE_ZIP_DIR_SLOT_SIZE. */
+	ulint		n_heap);/* in: number of records */
+/*****************************************************************
+Gets the number of dir slots in directory. */
+UNIV_INLINE
+ulint
+page_dir_get_n_slots(
+/*=================*/
+				/* out: number of slots */
+	const page_t*	page);	/* in: index page */
+/*****************************************************************
+Sets the number of dir slots in directory. */
+UNIV_INLINE
+void
+page_dir_set_n_slots(
+/*=================*/
+	page_t*		page,	/* in/out: page */
+	page_zip_des_t*	page_zip,/* in/out: compressed page whose
+				uncompressed part will be updated, or NULL */
+	ulint		n_slots);/* in: number of slots */
+#ifdef UNIV_DEBUG
+/*****************************************************************
+Gets pointer to nth directory slot. */
+UNIV_INLINE
+page_dir_slot_t*
+page_dir_get_nth_slot(
+/*==================*/
+				/* out: pointer to dir slot */
+	const page_t*	page,	/* in: index page */
+	ulint		n);	/* in: position */
+#else /* UNIV_DEBUG */
+# define page_dir_get_nth_slot(page, n)		\
+	((page) + UNIV_PAGE_SIZE - PAGE_DIR	\
+	 - (n + 1) * PAGE_DIR_SLOT_SIZE)
+#endif /* UNIV_DEBUG */
+/******************************************************************
+Used to check the consistency of a record on a page. */
+UNIV_INLINE
+ibool
+page_rec_check(
+/*===========*/
+				/* out: TRUE if succeed */
+	const rec_t*	rec);	/* in: record */
+/*******************************************************************
+Gets the record pointed to by a directory slot. */
+UNIV_INLINE
+const rec_t*
+page_dir_slot_get_rec(
+/*==================*/
+					/* out: pointer to record */
+	const page_dir_slot_t*	slot);	/* in: directory slot */
+/*******************************************************************
+This is used to set the record offset in a directory slot. */
+UNIV_INLINE
+void
+page_dir_slot_set_rec(
+/*==================*/
+	page_dir_slot_t* slot,	/* in: directory slot */
+	rec_t*		 rec);	/* in: record on the page */
+/*******************************************************************
+Gets the number of records owned by a directory slot. */
+UNIV_INLINE
+ulint
+page_dir_slot_get_n_owned(
+/*======================*/
+					/* out: number of records */
+	const page_dir_slot_t*	slot);	/* in: page directory slot */
+/*******************************************************************
+This is used to set the owned records field of a directory slot. */
+UNIV_INLINE
+void
+page_dir_slot_set_n_owned(
+/*======================*/
+	page_dir_slot_t*slot,	/* in/out: directory slot */
+	page_zip_des_t*	page_zip,/* in/out: compressed page, or NULL */
+	ulint		n);	/* in: number of records owned by the slot */
+/****************************************************************
+Calculates the space reserved for directory slots of a given
+number of records. The exact value is a fraction number
+n * PAGE_DIR_SLOT_SIZE / PAGE_DIR_SLOT_MIN_N_OWNED, and it is
+rounded upwards to an integer. */
+UNIV_INLINE
+ulint
+page_dir_calc_reserved_space(
+/*=========================*/
+	ulint	n_recs);	/* in: number of records */
+/*******************************************************************
+Looks for the directory slot which owns the given record. */
+UNIV_INTERN
+ulint
+page_dir_find_owner_slot(
+/*=====================*/
+				/* out: the directory slot number */
+	const rec_t*	rec);	/* in: the physical record */
+/****************************************************************
+Determine whether the page is in new-style compact format. */
+UNIV_INLINE
+ulint
+page_is_comp(
+/*=========*/
+				/* out: nonzero if the page is in compact
+				format, zero if it is in old-style format */
+	const page_t*	page);	/* in: index page */
+/****************************************************************
+TRUE if the record is on a page in compact format. */
+UNIV_INLINE
+ulint
+page_rec_is_comp(
+/*=============*/
+				/* out: nonzero if in compact format */
+	const rec_t*	rec);	/* in: record */
+/*******************************************************************
+Returns the heap number of a record. */
+UNIV_INLINE
+ulint
+page_rec_get_heap_no(
+/*=================*/
+				/* out: heap number */
+	const rec_t*	rec);	/* in: the physical record */
+/****************************************************************
+Determine whether the page is a B-tree leaf. */
+UNIV_INLINE
+ibool
+page_is_leaf(
+/*=========*/
+				/* out: TRUE if the page is a B-tree leaf */
+	const page_t*	page)	/* in: page */
+	__attribute__((nonnull, pure));
+/****************************************************************
+Gets the pointer to the next record on the page. */
+UNIV_INLINE
+const rec_t*
+page_rec_get_next_low(
+/*==================*/
+				/* out: pointer to next record */
+	const rec_t*	rec,	/* in: pointer to record */
+	ulint		comp);	/* in: nonzero=compact page layout */
+/****************************************************************
+Gets the pointer to the next record on the page. */
+UNIV_INLINE
+rec_t*
+page_rec_get_next(
+/*==============*/
+			/* out: pointer to next record */
+	rec_t*	rec);	/* in: pointer to record */
+/****************************************************************
+Gets the pointer to the next record on the page. */
+UNIV_INLINE
+const rec_t*
+page_rec_get_next_const(
+/*====================*/
+				/* out: pointer to next record */
+	const rec_t*	rec);	/* in: pointer to record */
+/****************************************************************
+Sets the pointer to the next record on the page. */
+UNIV_INLINE
+void
+page_rec_set_next(
+/*==============*/
+	rec_t*	rec,	/* in: pointer to record,
+			must not be page supremum */
+	rec_t*	next);	/* in: pointer to next record,
+			must not be page infimum */
+/****************************************************************
+Gets the pointer to the previous record. */
+UNIV_INLINE
+const rec_t*
+page_rec_get_prev_const(
+/*====================*/
+				/* out: pointer to previous record */
+	const rec_t*	rec);	/* in: pointer to record, must not be page
+				infimum */
+/****************************************************************
+Gets the pointer to the previous record. */
+UNIV_INLINE
+rec_t*
+page_rec_get_prev(
+/*==============*/
+				/* out: pointer to previous record */
+	rec_t*		rec);	/* in: pointer to record,
+				must not be page infimum */
+/****************************************************************
+TRUE if the record is a user record on the page. */
+UNIV_INLINE
+ibool
+page_rec_is_user_rec_low(
+/*=====================*/
+			/* out: TRUE if a user record */
+	ulint	offset)	/* in: record offset on page */
+	__attribute__((const));
+/****************************************************************
+TRUE if the record is the supremum record on a page. */
+UNIV_INLINE
+ibool
+page_rec_is_supremum_low(
+/*=====================*/
+			/* out: TRUE if the supremum record */
+	ulint	offset)	/* in: record offset on page */
+	__attribute__((const));
+/****************************************************************
+TRUE if the record is the infimum record on a page. */
+UNIV_INLINE
+ibool
+page_rec_is_infimum_low(
+/*====================*/
+			/* out: TRUE if the infimum record */
+	ulint	offset)	/* in: record offset on page */
+	__attribute__((const));
+
+/****************************************************************
+TRUE if the record is a user record on the page. */
+UNIV_INLINE
+ibool
+page_rec_is_user_rec(
+/*=================*/
+				/* out: TRUE if a user record */
+	const rec_t*	rec)	/* in: record */
+	__attribute__((const));
+/****************************************************************
+TRUE if the record is the supremum record on a page. */
+UNIV_INLINE
+ibool
+page_rec_is_supremum(
+/*=================*/
+				/* out: TRUE if the supremum record */
+	const rec_t*	rec)	/* in: record */
+	__attribute__((const));
+
+/****************************************************************
+TRUE if the record is the infimum record on a page. */
+UNIV_INLINE
+ibool
+page_rec_is_infimum(
+/*================*/
+				/* out: TRUE if the infimum record */
+	const rec_t*	rec)	/* in: record */
+	__attribute__((const));
+/*******************************************************************
+Looks for the record which owns the given record. */
+UNIV_INLINE
+rec_t*
+page_rec_find_owner_rec(
+/*====================*/
+			/* out: the owner record */
+	rec_t*	rec);	/* in: the physical record */
+/***************************************************************************
+This is a low-level operation which is used in a database index creation
+to update the page number of a created B-tree to a data dictionary
+record. */
+UNIV_INTERN
+void
+page_rec_write_index_page_no(
+/*=========================*/
+	rec_t*	rec,	/* in: record to update */
+	ulint	i,	/* in: index of the field to update */
+	ulint	page_no,/* in: value to write */
+	mtr_t*	mtr);	/* in: mtr */
+/****************************************************************
+Returns the maximum combined size of records which can be inserted on top
+of record heap. */
+UNIV_INLINE
+ulint
+page_get_max_insert_size(
+/*=====================*/
+				/* out: maximum combined size for
+				inserted records */
+	const page_t*	page,	/* in: index page */
+	ulint		n_recs);/* in: number of records */
+/****************************************************************
+Returns the maximum combined size of records which can be inserted on top
+of record heap if page is first reorganized. */
+UNIV_INLINE
+ulint
+page_get_max_insert_size_after_reorganize(
+/*======================================*/
+				/* out: maximum combined size for
+				inserted records */
+	const page_t*	page,	/* in: index page */
+	ulint		n_recs);/* in: number of records */
+/*****************************************************************
+Calculates free space if a page is emptied. */
+UNIV_INLINE
+ulint
+page_get_free_space_of_empty(
+/*=========================*/
+			/* out: free space */
+	ulint	comp)	/* in: nonzero=compact page format */
+		__attribute__((const));
+/**************************************************************
+Returns the base extra size of a physical record.  This is the
+size of the fixed header, independent of the record size. */
+UNIV_INLINE
+ulint
+page_rec_get_base_extra_size(
+/*=========================*/
+				/* out: REC_N_NEW_EXTRA_BYTES
+				or REC_N_OLD_EXTRA_BYTES */
+	const rec_t*	rec);	/* in: physical record */
+/****************************************************************
+Returns the sum of the sizes of the records in the record list
+excluding the infimum and supremum records. */
+UNIV_INLINE
+ulint
+page_get_data_size(
+/*===============*/
+				/* out: data in bytes */
+	const page_t*	page);	/* in: index page */
+/****************************************************************
+Allocates a block of memory from the head of the free list
+of an index page. */
+UNIV_INLINE
+void
+page_mem_alloc_free(
+/*================*/
+	page_t*		page,	/* in/out: index page */
+	page_zip_des_t*	page_zip,/* in/out: compressed page with enough
+				space available for inserting the record,
+				or NULL */
+	rec_t*		next_rec,/* in: pointer to the new head of the
+				free record list */
+	ulint		need);	/* in: number of bytes allocated */
+/****************************************************************
+Allocates a block of memory from the heap of an index page. */
+UNIV_INTERN
+byte*
+page_mem_alloc_heap(
+/*================*/
+				/* out: pointer to start of allocated
+				buffer, or NULL if allocation fails */
+	page_t*		page,	/* in/out: index page */
+	page_zip_des_t*	page_zip,/* in/out: compressed page with enough
+				space available for inserting the record,
+				or NULL */
+	ulint		need,	/* in: total number of bytes needed */
+	ulint*		heap_no);/* out: this contains the heap number
+				of the allocated record
+				if allocation succeeds */
+/****************************************************************
+Puts a record to free list. */
+UNIV_INLINE
+void
+page_mem_free(
+/*==========*/
+	page_t*		page,	/* in/out: index page */
+	page_zip_des_t*	page_zip,/* in/out: compressed page, or NULL */
+	rec_t*		rec,	/* in: pointer to the (origin of) record */
+	dict_index_t*	index,	/* in: index of rec */
+	const ulint*	offsets);/* in: array returned by rec_get_offsets() */
+/**************************************************************
+Create an uncompressed B-tree index page. */
+UNIV_INTERN
+page_t*
+page_create(
+/*========*/
+					/* out: pointer to the page */
+	buf_block_t*	block,		/* in: a buffer block where the
+					page is created */
+	mtr_t*		mtr,		/* in: mini-transaction handle */
+	ulint		comp);		/* in: nonzero=compact page format */
+/**************************************************************
+Create a compressed B-tree index page. */
+UNIV_INTERN
+page_t*
+page_create_zip(
+/*============*/
+					/* out: pointer to the page */
+	buf_block_t*	block,		/* in/out: a buffer frame where the
+					page is created */
+	dict_index_t*	index,		/* in: the index of the page */
+	ulint		level,		/* in: the B-tree level of the page */
+	mtr_t*		mtr);		/* in: mini-transaction handle */
+
+/*****************************************************************
+Differs from page_copy_rec_list_end, because this function does not
+touch the lock table and max trx id on page or compress the page. */
+UNIV_INTERN
+void
+page_copy_rec_list_end_no_locks(
+/*============================*/
+	buf_block_t*	new_block,	/* in: index page to copy to */
+	buf_block_t*	block,		/* in: index page of rec */
+	rec_t*		rec,		/* in: record on page */
+	dict_index_t*	index,		/* in: record descriptor */
+	mtr_t*		mtr);		/* in: mtr */
+/*****************************************************************
+Copies records from page to new_page, from the given record onward,
+including that record. Infimum and supremum records are not copied.
+The records are copied to the start of the record list on new_page. */
+UNIV_INTERN
+rec_t*
+page_copy_rec_list_end(
+/*===================*/
+					/* out: pointer to the original
+					successor of the infimum record
+					on new_page, or NULL on zip overflow
+					(new_block will be decompressed) */
+	buf_block_t*	new_block,	/* in/out: index page to copy to */
+	buf_block_t*	block,		/* in: index page containing rec */
+	rec_t*		rec,		/* in: record on page */
+	dict_index_t*	index,		/* in: record descriptor */
+	mtr_t*		mtr)		/* in: mtr */
+	__attribute__((nonnull));
+/*****************************************************************
+Copies records from page to new_page, up to the given record, NOT
+including that record. Infimum and supremum records are not copied.
+The records are copied to the end of the record list on new_page. */
+UNIV_INTERN
+rec_t*
+page_copy_rec_list_start(
+/*=====================*/
+					/* out: pointer to the original
+					predecessor of the supremum record
+					on new_page, or NULL on zip overflow
+					(new_block will be decompressed) */
+	buf_block_t*	new_block,	/* in/out: index page to copy to */
+	buf_block_t*	block,		/* in: index page containing rec */
+	rec_t*		rec,		/* in: record on page */
+	dict_index_t*	index,		/* in: record descriptor */
+	mtr_t*		mtr)		/* in: mtr */
+	__attribute__((nonnull));
+/*****************************************************************
+Deletes records from a page from a given record onward, including that record.
+The infimum and supremum records are not deleted. */
+UNIV_INTERN
+void
+page_delete_rec_list_end(
+/*=====================*/
+	rec_t*		rec,	/* in: pointer to record on page */
+	buf_block_t*	block,	/* in: buffer block of the page */
+	dict_index_t*	index,	/* in: record descriptor */
+	ulint		n_recs,	/* in: number of records to delete,
+				or ULINT_UNDEFINED if not known */
+	ulint		size,	/* in: the sum of the sizes of the
+				records in the end of the chain to
+				delete, or ULINT_UNDEFINED if not known */
+	mtr_t*		mtr)	/* in: mtr */
+	__attribute__((nonnull));
+/*****************************************************************
+Deletes records from page, up to the given record, NOT including
+that record. Infimum and supremum records are not deleted. */
+UNIV_INTERN
+void
+page_delete_rec_list_start(
+/*=======================*/
+	rec_t*		rec,	/* in: record on page */
+	buf_block_t*	block,	/* in: buffer block of the page */
+	dict_index_t*	index,	/* in: record descriptor */
+	mtr_t*		mtr)	/* in: mtr */
+	__attribute__((nonnull));
+/*****************************************************************
+Moves record list end to another page. Moved records include
+split_rec. */
+UNIV_INTERN
+ibool
+page_move_rec_list_end(
+/*===================*/
+					/* out: TRUE on success; FALSE on
+					compression failure
+					(new_block will be decompressed) */
+	buf_block_t*	new_block,	/* in/out: index page where to move */
+	buf_block_t*	block,		/* in: index page from where to move */
+	rec_t*		split_rec,	/* in: first record to move */
+	dict_index_t*	index,		/* in: record descriptor */
+	mtr_t*		mtr)		/* in: mtr */
+	__attribute__((nonnull(1, 2, 4, 5)));
+/*****************************************************************
+Moves record list start to another page. Moved records do not include
+split_rec. */
+UNIV_INTERN
+ibool
+page_move_rec_list_start(
+/*=====================*/
+					/* out: TRUE on success; FALSE on
+					compression failure */
+	buf_block_t*	new_block,	/* in/out: index page where to move */
+	buf_block_t*	block,		/* in/out: page containing split_rec */
+	rec_t*		split_rec,	/* in: first record not to move */
+	dict_index_t*	index,		/* in: record descriptor */
+	mtr_t*		mtr)		/* in: mtr */
+	__attribute__((nonnull(1, 2, 4, 5)));
+/********************************************************************
+Splits a directory slot which owns too many records. */
+UNIV_INTERN
+void
+page_dir_split_slot(
+/*================*/
+	page_t*		page,	/* in: index page */
+	page_zip_des_t*	page_zip,/* in/out: compressed page whose
+				uncompressed part will be written, or NULL */
+	ulint		slot_no)/* in: the directory slot */
+	__attribute__((nonnull(1)));
+/*****************************************************************
+Tries to balance the given directory slot with too few records
+with the upper neighbor, so that there are at least the minimum number
+of records owned by the slot; this may result in the merging of
+two slots. */
+UNIV_INTERN
+void
+page_dir_balance_slot(
+/*==================*/
+	page_t*		page,	/* in/out: index page */
+	page_zip_des_t*	page_zip,/* in/out: compressed page, or NULL */
+	ulint		slot_no)/* in: the directory slot */
+	__attribute__((nonnull(1)));
+/**************************************************************
+Parses a log record of a record list end or start deletion. */
+UNIV_INTERN
+byte*
+page_parse_delete_rec_list(
+/*=======================*/
+				/* out: end of log record or NULL */
+	byte		type,	/* in: MLOG_LIST_END_DELETE,
+				MLOG_LIST_START_DELETE,
+				MLOG_COMP_LIST_END_DELETE or
+				MLOG_COMP_LIST_START_DELETE */
+	byte*		ptr,	/* in: buffer */
+	byte*		end_ptr,/* in: buffer end */
+	buf_block_t*	block,	/* in/out: buffer block or NULL */
+	dict_index_t*	index,	/* in: record descriptor */
+	mtr_t*		mtr);	/* in: mtr or NULL */
+/***************************************************************
+Parses a redo log record of creating a page. */
+UNIV_INTERN
+byte*
+page_parse_create(
+/*==============*/
+				/* out: end of log record or NULL */
+	byte*		ptr,	/* in: buffer */
+	byte*		end_ptr,/* in: buffer end */
+	ulint		comp,	/* in: nonzero=compact page format */
+	buf_block_t*	block,	/* in: block or NULL */
+	mtr_t*		mtr);	/* in: mtr or NULL */
+/****************************************************************
+Prints record contents including the data relevant only in
+the index page context. */
+UNIV_INTERN
+void
+page_rec_print(
+/*===========*/
+	const rec_t*	rec,	/* in: physical record */
+	const ulint*	offsets);/* in: record descriptor */
+/*******************************************************************
+This is used to print the contents of the directory for
+debugging purposes. */
+UNIV_INTERN
+void
+page_dir_print(
+/*===========*/
+	page_t*	page,	/* in: index page */
+	ulint	pr_n);	/* in: print n first and n last entries */
+/*******************************************************************
+This is used to print the contents of the page record list for
+debugging purposes. */
+UNIV_INTERN
+void
+page_print_list(
+/*============*/
+	buf_block_t*	block,	/* in: index page */
+	dict_index_t*	index,	/* in: dictionary index of the page */
+	ulint		pr_n);	/* in: print n first and n last entries */
+/*******************************************************************
+Prints the info in a page header. */
+UNIV_INTERN
+void
+page_header_print(
+/*==============*/
+	const page_t*	page);
+/*******************************************************************
+This is used to print the contents of the page for
+debugging purposes. */
+UNIV_INTERN
+void
+page_print(
+/*=======*/
+	buf_block_t*	block,	/* in: index page */
+	dict_index_t*	index,	/* in: dictionary index of the page */
+	ulint		dn,	/* in: print dn first and last entries
+				in directory */
+	ulint		rn);	/* in: print rn first and last records
+				in directory */
+/*******************************************************************
+The following is used to validate a record on a page. This function
+differs from rec_validate as it can also check the n_owned field and
+the heap_no field. */
+UNIV_INTERN
+ibool
+page_rec_validate(
+/*==============*/
+				/* out: TRUE if ok */
+	rec_t*		rec,	/* in: physical record */
+	const ulint*	offsets);/* in: array returned by rec_get_offsets() */
+/*******************************************************************
+Checks that the first directory slot points to the infimum record and
+the last to the supremum. This function is intended to track if the
+bug fixed in 4.0.14 has caused corruption to users' databases. */
+UNIV_INTERN
+void
+page_check_dir(
+/*===========*/
+	const page_t*	page);	/* in: index page */
+/*******************************************************************
+This function checks the consistency of an index page when we do not
+know the index. This is also resilient so that this should never crash
+even if the page is total garbage. */
+UNIV_INTERN
+ibool
+page_simple_validate_old(
+/*=====================*/
+			/* out: TRUE if ok */
+	page_t*	page);	/* in: old-style index page */
+/*******************************************************************
+This function checks the consistency of an index page when we do not
+know the index. This is also resilient so that this should never crash
+even if the page is total garbage. */
+UNIV_INTERN
+ibool
+page_simple_validate_new(
+/*=====================*/
+			/* out: TRUE if ok */
+	page_t*	block);	/* in: new-style index page */
+/*******************************************************************
+This function checks the consistency of an index page. */
+UNIV_INTERN
+ibool
+page_validate(
+/*==========*/
+				/* out: TRUE if ok */
+	page_t*		page,	/* in: index page */
+	dict_index_t*	index);	/* in: data dictionary index containing
+				the page record type definition */
+/*******************************************************************
+Looks in the page record list for a record with the given heap number. */
+
+const rec_t*
+page_find_rec_with_heap_no(
+/*=======================*/
+				/* out: record, NULL if not found */
+	const page_t*	page,	/* in: index page */
+	ulint		heap_no);/* in: heap number */
+
+#ifdef UNIV_MATERIALIZE
+#undef UNIV_INLINE
+#define UNIV_INLINE  UNIV_INLINE_ORIGINAL
+#endif
+
+#ifndef UNIV_NONINL
+#include "page0page.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/page0page.ic b/storage/xtradb/include/page0page.ic
new file mode 100644
index 00000000000..df0f6f8b360
--- /dev/null
+++ b/storage/xtradb/include/page0page.ic
@@ -0,0 +1,1060 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Index page routines
+
+Created 2/2/1994 Heikki Tuuri
+*******************************************************/
+
+#include "mach0data.h"
+#include "rem0cmp.h"
+#include "mtr0log.h"
+#include "page0zip.h"
+
+#ifdef UNIV_MATERIALIZE
+#undef UNIV_INLINE
+#define UNIV_INLINE
+#endif
+
+/****************************************************************
+Gets the start of a page. */
+UNIV_INLINE
+page_t*
+page_align(
+/*=======*/
+				/* out: start of the page */
+	const void*	ptr)	/* in: pointer to page frame */
+{
+	return((page_t*) ut_align_down(ptr, UNIV_PAGE_SIZE));
+}
+/****************************************************************
+Gets the offset within a page. */
+UNIV_INLINE
+ulint
+page_offset(
+/*========*/
+				/* out: offset from the start of the page */
+	const void*	ptr)	/* in: pointer to page frame */
+{
+	return(ut_align_offset(ptr, UNIV_PAGE_SIZE));
+}
+/*****************************************************************
+Returns the max trx id field value. */
+UNIV_INLINE
+dulint
+page_get_max_trx_id(
+/*================*/
+	const page_t*	page)	/* in: page */
+{
+	ut_ad(page);
+
+	return(mach_read_from_8(page + PAGE_HEADER + PAGE_MAX_TRX_ID));
+}
+
+/*****************************************************************
+Sets the max trx id field value if trx_id is bigger than the previous
+value. */
+UNIV_INLINE
+void
+page_update_max_trx_id(
+/*===================*/
+	buf_block_t*	block,	/* in/out: page */
+	page_zip_des_t*	page_zip,/* in/out: compressed page whose
+				uncompressed part will be updated, or NULL */
+	dulint		trx_id)	/* in: transaction id */
+{
+	ut_ad(block);
+
+	if (ut_dulint_cmp(page_get_max_trx_id(buf_block_get_frame(block)),
+			  trx_id) < 0) {
+
+		page_set_max_trx_id(block, page_zip, trx_id);
+	}
+}
+
+/*****************************************************************
+Reads the given header field. */
+UNIV_INLINE
+ulint
+page_header_get_field(
+/*==================*/
+	const page_t*	page,	/* in: page */
+	ulint		field)	/* in: PAGE_LEVEL, ... */
+{
+	ut_ad(page);
+	ut_ad(field <= PAGE_INDEX_ID);
+
+	return(mach_read_from_2(page + PAGE_HEADER + field));
+}
+
+/*****************************************************************
+Sets the given header field. */
+UNIV_INLINE
+void
+page_header_set_field(
+/*==================*/
+	page_t*		page,	/* in/out: page */
+	page_zip_des_t*	page_zip,/* in/out: compressed page whose
+				uncompressed part will be updated, or NULL */
+	ulint		field,	/* in: PAGE_N_DIR_SLOTS, ... */
+	ulint		val)	/* in: value */
+{
+	ut_ad(page);
+	ut_ad(field <= PAGE_N_RECS);
+	ut_ad(field == PAGE_N_HEAP || val < UNIV_PAGE_SIZE);
+	ut_ad(field != PAGE_N_HEAP || (val & 0x7fff) < UNIV_PAGE_SIZE);
+
+	mach_write_to_2(page + PAGE_HEADER + field, val);
+	if (UNIV_LIKELY_NULL(page_zip)) {
+		page_zip_write_header(page_zip,
+				      page + PAGE_HEADER + field, 2, NULL);
+	}
+}
+
+/*****************************************************************
+Returns the offset stored in the given header field. */
+UNIV_INLINE
+ulint
+page_header_get_offs(
+/*=================*/
+				/* out: offset from the start of the page,
+				or 0 */
+	const page_t*	page,	/* in: page */
+	ulint		field)	/* in: PAGE_FREE, ... */
+{
+	ulint	offs;
+
+	ut_ad(page);
+	ut_ad((field == PAGE_FREE)
+	      || (field == PAGE_LAST_INSERT)
+	      || (field == PAGE_HEAP_TOP));
+
+	offs = page_header_get_field(page, field);
+
+	ut_ad((field != PAGE_HEAP_TOP) || offs);
+
+	return(offs);
+}
+
+/*****************************************************************
+Sets the pointer stored in the given header field. */
+UNIV_INLINE
+void
+page_header_set_ptr(
+/*================*/
+	page_t*		page,	/* in: page */
+	page_zip_des_t*	page_zip,/* in/out: compressed page whose
+				uncompressed part will be updated, or NULL */
+	ulint		field,	/* in: PAGE_FREE, ... */
+	const byte*	ptr)	/* in: pointer or NULL*/
+{
+	ulint	offs;
+
+	ut_ad(page);
+	ut_ad((field == PAGE_FREE)
+	      || (field == PAGE_LAST_INSERT)
+	      || (field == PAGE_HEAP_TOP));
+
+	if (ptr == NULL) {
+		offs = 0;
+	} else {
+		offs = ptr - page;
+	}
+
+	ut_ad((field != PAGE_HEAP_TOP) || offs);
+
+	page_header_set_field(page, page_zip, field, offs);
+}
+
+/*****************************************************************
+Resets the last insert info field in the page header. Writes to mlog
+about this operation. */
+UNIV_INLINE
+void
+page_header_reset_last_insert(
+/*==========================*/
+	page_t*		page,	/* in/out: page */
+	page_zip_des_t*	page_zip,/* in/out: compressed page whose
+				uncompressed part will be updated, or NULL */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	ut_ad(page && mtr);
+
+	if (UNIV_LIKELY_NULL(page_zip)) {
+		mach_write_to_2(page + (PAGE_HEADER + PAGE_LAST_INSERT), 0);
+		page_zip_write_header(page_zip,
+				      page + (PAGE_HEADER + PAGE_LAST_INSERT),
+				      2, mtr);
+	} else {
+		mlog_write_ulint(page + (PAGE_HEADER + PAGE_LAST_INSERT), 0,
+				 MLOG_2BYTES, mtr);
+	}
+}
+
+/****************************************************************
+Determine whether the page is in new-style compact format. */
+UNIV_INLINE
+ulint
+page_is_comp(
+/*=========*/
+				/* out: nonzero if the page is in compact
+				format, zero if it is in old-style format */
+	const page_t*	page)	/* in: index page */
+{
+	return(UNIV_EXPECT(page_header_get_field(page, PAGE_N_HEAP) & 0x8000,
+			   0x8000));
+}
+
+/****************************************************************
+TRUE if the record is on a page in compact format. */
+UNIV_INLINE
+ulint
+page_rec_is_comp(
+/*=============*/
+				/* out: nonzero if in compact format */
+	const rec_t*	rec)	/* in: record */
+{
+	return(page_is_comp(page_align(rec)));
+}
+
+/*******************************************************************
+Returns the heap number of a record. */
+UNIV_INLINE
+ulint
+page_rec_get_heap_no(
+/*=================*/
+				/* out: heap number */
+	const rec_t*	rec)	/* in: the physical record */
+{
+	if (page_rec_is_comp(rec)) {
+		return(rec_get_heap_no_new(rec));
+	} else {
+		return(rec_get_heap_no_old(rec));
+	}
+}
+
+/****************************************************************
+Determine whether the page is a B-tree leaf. */
+UNIV_INLINE
+ibool
+page_is_leaf(
+/*=========*/
+				/* out: TRUE if the page is a B-tree leaf */
+	const page_t*	page)	/* in: page */
+{
+	return(!*(const uint16*) (page + (PAGE_HEADER + PAGE_LEVEL)));
+}
+
+/****************************************************************
+Gets the offset of the first record on the page. */
+UNIV_INLINE
+ulint
+page_get_infimum_offset(
+/*====================*/
+				/* out: offset of the first record
+				in record list, relative from page */
+	const page_t*	page)	/* in: page which must have record(s) */
+{
+	ut_ad(page);
+	ut_ad(!page_offset(page));
+
+	if (page_is_comp(page)) {
+		return(PAGE_NEW_INFIMUM);
+	} else {
+		return(PAGE_OLD_INFIMUM);
+	}
+}
+
+/****************************************************************
+Gets the offset of the last record on the page. */
+UNIV_INLINE
+ulint
+page_get_supremum_offset(
+/*=====================*/
+				/* out: offset of the last record in
+				record list, relative from page */
+	const page_t*	page)	/* in: page which must have record(s) */
+{
+	ut_ad(page);
+	ut_ad(!page_offset(page));
+
+	if (page_is_comp(page)) {
+		return(PAGE_NEW_SUPREMUM);
+	} else {
+		return(PAGE_OLD_SUPREMUM);
+	}
+}
+
+/****************************************************************
+TRUE if the record is a user record on the page. */
+UNIV_INLINE
+ibool
+page_rec_is_user_rec_low(
+/*=====================*/
+			/* out: TRUE if a user record */
+	ulint	offset)	/* in: record offset on page */
+{
+	ut_ad(offset >= PAGE_NEW_INFIMUM);
+#if PAGE_OLD_INFIMUM < PAGE_NEW_INFIMUM
+# error "PAGE_OLD_INFIMUM < PAGE_NEW_INFIMUM"
+#endif
+#if PAGE_OLD_SUPREMUM < PAGE_NEW_SUPREMUM
+# error "PAGE_OLD_SUPREMUM < PAGE_NEW_SUPREMUM"
+#endif
+#if PAGE_NEW_INFIMUM > PAGE_OLD_SUPREMUM
+# error "PAGE_NEW_INFIMUM > PAGE_OLD_SUPREMUM"
+#endif
+#if PAGE_OLD_INFIMUM > PAGE_NEW_SUPREMUM
+# error "PAGE_OLD_INFIMUM > PAGE_NEW_SUPREMUM"
+#endif
+#if PAGE_NEW_SUPREMUM > PAGE_OLD_SUPREMUM_END
+# error "PAGE_NEW_SUPREMUM > PAGE_OLD_SUPREMUM_END"
+#endif
+#if PAGE_OLD_SUPREMUM > PAGE_NEW_SUPREMUM_END
+# error "PAGE_OLD_SUPREMUM > PAGE_NEW_SUPREMUM_END"
+#endif
+	ut_ad(offset <= UNIV_PAGE_SIZE - PAGE_EMPTY_DIR_START);
+
+	return(UNIV_LIKELY(offset != PAGE_NEW_SUPREMUM)
+	       && UNIV_LIKELY(offset != PAGE_NEW_INFIMUM)
+	       && UNIV_LIKELY(offset != PAGE_OLD_INFIMUM)
+	       && UNIV_LIKELY(offset != PAGE_OLD_SUPREMUM));
+}
+
+/****************************************************************
+TRUE if the record is the supremum record on a page. */
+UNIV_INLINE
+ibool
+page_rec_is_supremum_low(
+/*=====================*/
+			/* out: TRUE if the supremum record */
+	ulint	offset)	/* in: record offset on page */
+{
+	ut_ad(offset >= PAGE_NEW_INFIMUM);
+	ut_ad(offset <= UNIV_PAGE_SIZE - PAGE_EMPTY_DIR_START);
+
+	return(UNIV_UNLIKELY(offset == PAGE_NEW_SUPREMUM)
+	       || UNIV_UNLIKELY(offset == PAGE_OLD_SUPREMUM));
+}
+
+/****************************************************************
+TRUE if the record is the infimum record on a page. */
+UNIV_INLINE
+ibool
+page_rec_is_infimum_low(
+/*====================*/
+			/* out: TRUE if the infimum record */
+	ulint	offset)	/* in: record offset on page */
+{
+	ut_ad(offset >= PAGE_NEW_INFIMUM);
+	ut_ad(offset <= UNIV_PAGE_SIZE - PAGE_EMPTY_DIR_START);
+
+	return(UNIV_UNLIKELY(offset == PAGE_NEW_INFIMUM)
+	       || UNIV_UNLIKELY(offset == PAGE_OLD_INFIMUM));
+}
+
+/****************************************************************
+TRUE if the record is a user record on the page. */
+UNIV_INLINE
+ibool
+page_rec_is_user_rec(
+/*=================*/
+				/* out: TRUE if a user record */
+	const rec_t*	rec)	/* in: record */
+{
+	return(page_rec_is_user_rec_low(page_offset(rec)));
+}
+
+/****************************************************************
+TRUE if the record is the supremum record on a page. */
+UNIV_INLINE
+ibool
+page_rec_is_supremum(
+/*=================*/
+				/* out: TRUE if the supremum record */
+	const rec_t*	rec)	/* in: record */
+{
+	return(page_rec_is_supremum_low(page_offset(rec)));
+}
+
+/****************************************************************
+TRUE if the record is the infimum record on a page. */
+UNIV_INLINE
+ibool
+page_rec_is_infimum(
+/*================*/
+				/* out: TRUE if the infimum record */
+	const rec_t*	rec)	/* in: record */
+{
+	return(page_rec_is_infimum_low(page_offset(rec)));
+}
+
+/*****************************************************************
+Compares a data tuple to a physical record. Differs from the function
+cmp_dtuple_rec_with_match in the way that the record must reside on an
+index page, and also page infimum and supremum records can be given in
+the parameter rec. These are considered as the negative infinity and
+the positive infinity in the alphabetical order. */
+UNIV_INLINE
+int
+page_cmp_dtuple_rec_with_match(
+/*===========================*/
+				/* out: 1, 0, -1, if dtuple is greater, equal,
+				less than rec, respectively, when only the
+				common first fields are compared */
+	const dtuple_t*	dtuple,	/* in: data tuple */
+	const rec_t*	rec,	/* in: physical record on a page; may also
+				be page infimum or supremum, in which case
+				matched-parameter values below are not
+				affected */
+	const ulint*	offsets,/* in: array returned by rec_get_offsets() */
+	ulint*		matched_fields, /* in/out: number of already completely
+				matched fields; when function returns
+				contains the value for current comparison */
+	ulint*		matched_bytes) /* in/out: number of already matched
+				bytes within the first field not completely
+				matched; when function returns contains the
+				value for current comparison */
+{
+	ulint	rec_offset;
+
+	ut_ad(dtuple_check_typed(dtuple));
+	ut_ad(rec_offs_validate(rec, NULL, offsets));
+	ut_ad(!rec_offs_comp(offsets) == !page_rec_is_comp(rec));
+
+	rec_offset = page_offset(rec);
+
+	if (UNIV_UNLIKELY(rec_offset == PAGE_NEW_INFIMUM)
+	    || UNIV_UNLIKELY(rec_offset == PAGE_OLD_INFIMUM)) {
+		return(1);
+	}
+	if (UNIV_UNLIKELY(rec_offset == PAGE_NEW_SUPREMUM)
+	    || UNIV_UNLIKELY(rec_offset == PAGE_OLD_SUPREMUM)) {
+		return(-1);
+	}
+
+	return(cmp_dtuple_rec_with_match(dtuple, rec, offsets,
+					 matched_fields,
+					 matched_bytes));
+}
+
+/*****************************************************************
+Gets the page number. */
+UNIV_INLINE
+ulint
+page_get_page_no(
+/*=============*/
+				/* out: page number */
+	const page_t*	page)	/* in: page */
+{
+	ut_ad(page == page_align((page_t*) page));
+	return(mach_read_from_4(page + FIL_PAGE_OFFSET));
+}
+
+/*****************************************************************
+Gets the tablespace identifier. */
+UNIV_INLINE
+ulint
+page_get_space_id(
+/*==============*/
+				/* out: space id */
+	const page_t*	page)	/* in: page */
+{
+	ut_ad(page == page_align((page_t*) page));
+	return(mach_read_from_4(page + FIL_PAGE_ARCH_LOG_NO_OR_SPACE_ID));
+}
+
+/*****************************************************************
+Gets the number of user records on page (infimum and supremum records
+are not user records). */
+UNIV_INLINE
+ulint
+page_get_n_recs(
+/*============*/
+				/* out: number of user records */
+	const page_t*	page)	/* in: index page */
+{
+	return(page_header_get_field(page, PAGE_N_RECS));
+}
+
+/*****************************************************************
+Gets the number of dir slots in directory. */
+UNIV_INLINE
+ulint
+page_dir_get_n_slots(
+/*=================*/
+				/* out: number of slots */
+	const page_t*	page)	/* in: index page */
+{
+	return(page_header_get_field(page, PAGE_N_DIR_SLOTS));
+}
+/*****************************************************************
+Sets the number of dir slots in directory. */
+UNIV_INLINE
+void
+page_dir_set_n_slots(
+/*=================*/
+	page_t*		page,	/* in/out: page */
+	page_zip_des_t*	page_zip,/* in/out: compressed page whose
+				uncompressed part will be updated, or NULL */
+	ulint		n_slots)/* in: number of slots */
+{
+	page_header_set_field(page, page_zip, PAGE_N_DIR_SLOTS, n_slots);
+}
+
+/*****************************************************************
+Gets the number of records in the heap. */
+UNIV_INLINE
+ulint
+page_dir_get_n_heap(
+/*================*/
+				/* out: number of user records */
+	const page_t*	page)	/* in: index page */
+{
+	return(page_header_get_field(page, PAGE_N_HEAP) & 0x7fff);
+}
+
+/*****************************************************************
+Sets the number of records in the heap. */
+UNIV_INLINE
+void
+page_dir_set_n_heap(
+/*================*/
+	page_t*		page,	/* in/out: index page */
+	page_zip_des_t*	page_zip,/* in/out: compressed page whose
+				uncompressed part will be updated, or NULL.
+				Note that the size of the dense page directory
+				in the compressed page trailer is
+				n_heap * PAGE_ZIP_DIR_SLOT_SIZE. */
+	ulint		n_heap)	/* in: number of records */
+{
+	ut_ad(n_heap < 0x8000);
+	ut_ad(!page_zip || n_heap
+	      == (page_header_get_field(page, PAGE_N_HEAP) & 0x7fff) + 1);
+
+	page_header_set_field(page, page_zip, PAGE_N_HEAP, n_heap
+			      | (0x8000
+				 & page_header_get_field(page, PAGE_N_HEAP)));
+}
+
+#ifdef UNIV_DEBUG
+/*****************************************************************
+Gets pointer to nth directory slot. */
+UNIV_INLINE
+page_dir_slot_t*
+page_dir_get_nth_slot(
+/*==================*/
+				/* out: pointer to dir slot */
+	const page_t*	page,	/* in: index page */
+	ulint		n)	/* in: position */
+{
+	ut_ad(page_dir_get_n_slots(page) > n);
+
+	return((page_dir_slot_t*)
+	       page + UNIV_PAGE_SIZE - PAGE_DIR
+	       - (n + 1) * PAGE_DIR_SLOT_SIZE);
+}
+#endif /* UNIV_DEBUG */
+
+/******************************************************************
+Used to check the consistency of a record on a page. */
+UNIV_INLINE
+ibool
+page_rec_check(
+/*===========*/
+				/* out: TRUE if succeed */
+	const rec_t*	rec)	/* in: record */
+{
+	const page_t*	page = page_align(rec);
+
+	ut_a(rec);
+
+	ut_a(page_offset(rec) <= page_header_get_field(page, PAGE_HEAP_TOP));
+	ut_a(page_offset(rec) >= PAGE_DATA);
+
+	return(TRUE);
+}
+
+/*******************************************************************
+Gets the record pointed to by a directory slot. */
+UNIV_INLINE
+const rec_t*
+page_dir_slot_get_rec(
+/*==================*/
+					/* out: pointer to record */
+	const page_dir_slot_t*	slot)	/* in: directory slot */
+{
+	return(page_align(slot) + mach_read_from_2(slot));
+}
+
+/*******************************************************************
+This is used to set the record offset in a directory slot. */
+UNIV_INLINE
+void
+page_dir_slot_set_rec(
+/*==================*/
+	page_dir_slot_t* slot,	/* in: directory slot */
+	rec_t*		 rec)	/* in: record on the page */
+{
+	ut_ad(page_rec_check(rec));
+
+	mach_write_to_2(slot, page_offset(rec));
+}
+
+/*******************************************************************
+Gets the number of records owned by a directory slot. */
+UNIV_INLINE
+ulint
+page_dir_slot_get_n_owned(
+/*======================*/
+					/* out: number of records */
+	const page_dir_slot_t*	slot)	/* in: page directory slot */
+{
+	const rec_t*	rec	= page_dir_slot_get_rec(slot);
+	if (page_rec_is_comp(slot)) {
+		return(rec_get_n_owned_new(rec));
+	} else {
+		return(rec_get_n_owned_old(rec));
+	}
+}
+
+/*******************************************************************
+This is used to set the owned records field of a directory slot. */
+UNIV_INLINE
+void
+page_dir_slot_set_n_owned(
+/*======================*/
+	page_dir_slot_t*slot,	/* in/out: directory slot */
+	page_zip_des_t*	page_zip,/* in/out: compressed page, or NULL */
+	ulint		n)	/* in: number of records owned by the slot */
+{
+	rec_t*	rec	= (rec_t*) page_dir_slot_get_rec(slot);
+	if (page_rec_is_comp(slot)) {
+		rec_set_n_owned_new(rec, page_zip, n);
+	} else {
+		ut_ad(!page_zip);
+		rec_set_n_owned_old(rec, n);
+	}
+}
+
+/****************************************************************
+Calculates the space reserved for directory slots of a given number of
+records. The exact value is a fraction number n * PAGE_DIR_SLOT_SIZE /
+PAGE_DIR_SLOT_MIN_N_OWNED, and it is rounded upwards to an integer. */
+UNIV_INLINE
+ulint
+page_dir_calc_reserved_space(
+/*=========================*/
+	ulint	n_recs)		/* in: number of records */
+{
+	return((PAGE_DIR_SLOT_SIZE * n_recs + PAGE_DIR_SLOT_MIN_N_OWNED - 1)
+	       / PAGE_DIR_SLOT_MIN_N_OWNED);
+}
+
+/****************************************************************
+Gets the pointer to the next record on the page. */
+UNIV_INLINE
+const rec_t*
+page_rec_get_next_low(
+/*==================*/
+				/* out: pointer to next record */
+	const rec_t*	rec,	/* in: pointer to record */
+	ulint		comp)	/* in: nonzero=compact page layout */
+{
+	ulint		offs;
+	const page_t*	page;
+
+	ut_ad(page_rec_check(rec));
+
+	page = page_align(rec);
+
+	offs = rec_get_next_offs(rec, comp);
+
+	if (UNIV_UNLIKELY(offs >= UNIV_PAGE_SIZE)) {
+		fprintf(stderr,
+			"InnoDB: Next record offset is nonsensical %lu"
+			" in record at offset %lu\n"
+			"InnoDB: rec address %p, space id %lu, page %lu\n",
+			(ulong)offs, (ulong) page_offset(rec),
+			(void*) rec,
+			(ulong) page_get_space_id(page),
+			(ulong) page_get_page_no(page));
+		buf_page_print(page, 0);
+
+		ut_error;
+	}
+
+	if (UNIV_UNLIKELY(offs == 0)) {
+
+		return(NULL);
+	}
+
+	return(page + offs);
+}
+
+/****************************************************************
+Gets the pointer to the next record on the page. */
+UNIV_INLINE
+rec_t*
+page_rec_get_next(
+/*==============*/
+			/* out: pointer to next record */
+	rec_t*	rec)	/* in: pointer to record */
+{
+	return((rec_t*) page_rec_get_next_low(rec, page_rec_is_comp(rec)));
+}
+
+/****************************************************************
+Gets the pointer to the next record on the page. */
+UNIV_INLINE
+const rec_t*
+page_rec_get_next_const(
+/*====================*/
+				/* out: pointer to next record */
+	const rec_t*	rec)	/* in: pointer to record */
+{
+	return(page_rec_get_next_low(rec, page_rec_is_comp(rec)));
+}
+
+/****************************************************************
+Sets the pointer to the next record on the page. */
+UNIV_INLINE
+void
+page_rec_set_next(
+/*==============*/
+	rec_t*	rec,		/* in: pointer to record,
+				must not be page supremum */
+	rec_t*	next)		/* in: pointer to next record,
+				must not be page infimum */
+{
+	ulint	offs;
+
+	ut_ad(page_rec_check(rec));
+	ut_ad(!page_rec_is_supremum(rec));
+	ut_ad(rec != next);
+
+	ut_ad(!next || !page_rec_is_infimum(next));
+	ut_ad(!next || page_align(rec) == page_align(next));
+
+	if (UNIV_LIKELY(next != NULL)) {
+		offs = page_offset(next);
+	} else {
+		offs = 0;
+	}
+
+	if (page_rec_is_comp(rec)) {
+		rec_set_next_offs_new(rec, offs);
+	} else {
+		rec_set_next_offs_old(rec, offs);
+	}
+}
+
+/****************************************************************
+Gets the pointer to the previous record. */
+UNIV_INLINE
+const rec_t*
+page_rec_get_prev_const(
+/*====================*/
+				/* out: pointer to previous record */
+	const rec_t*	rec)	/* in: pointer to record, must not be page
+				infimum */
+{
+	const page_dir_slot_t*	slot;
+	ulint			slot_no;
+	const rec_t*		rec2;
+	const rec_t*		prev_rec = NULL;
+	const page_t*		page;
+
+	ut_ad(page_rec_check(rec));
+
+	page = page_align(rec);
+
+	ut_ad(!page_rec_is_infimum(rec));
+
+	slot_no = page_dir_find_owner_slot(rec);
+
+	ut_a(slot_no != 0);
+
+	slot = page_dir_get_nth_slot(page, slot_no - 1);
+
+	rec2 = page_dir_slot_get_rec(slot);
+
+	if (page_is_comp(page)) {
+		while (rec != rec2) {
+			prev_rec = rec2;
+			rec2 = page_rec_get_next_low(rec2, TRUE);
+		}
+	} else {
+		while (rec != rec2) {
+			prev_rec = rec2;
+			rec2 = page_rec_get_next_low(rec2, FALSE);
+		}
+	}
+
+	ut_a(prev_rec);
+
+	return(prev_rec);
+}
+
+/****************************************************************
+Gets the pointer to the previous record. */
+UNIV_INLINE
+rec_t*
+page_rec_get_prev(
+/*==============*/
+			/* out: pointer to previous record */
+	rec_t*	rec)	/* in: pointer to record, must not be page
+			infimum */
+{
+	return((rec_t*) page_rec_get_prev_const(rec));
+}
+
+/*******************************************************************
+Looks for the record which owns the given record. */
+UNIV_INLINE
+rec_t*
+page_rec_find_owner_rec(
+/*====================*/
+			/* out: the owner record */
+	rec_t*	rec)	/* in: the physical record */
+{
+	ut_ad(page_rec_check(rec));
+
+	if (page_rec_is_comp(rec)) {
+		while (rec_get_n_owned_new(rec) == 0) {
+			rec = page_rec_get_next(rec);
+		}
+	} else {
+		while (rec_get_n_owned_old(rec) == 0) {
+			rec = page_rec_get_next(rec);
+		}
+	}
+
+	return(rec);
+}
+
+/**************************************************************
+Returns the base extra size of a physical record.  This is the
+size of the fixed header, independent of the record size. */
+UNIV_INLINE
+ulint
+page_rec_get_base_extra_size(
+/*=========================*/
+				/* out: REC_N_NEW_EXTRA_BYTES
+				or REC_N_OLD_EXTRA_BYTES */
+	const rec_t*	rec)	/* in: physical record */
+{
+#if REC_N_NEW_EXTRA_BYTES + 1 != REC_N_OLD_EXTRA_BYTES
+# error "REC_N_NEW_EXTRA_BYTES + 1 != REC_N_OLD_EXTRA_BYTES"
+#endif
+	return(REC_N_NEW_EXTRA_BYTES + (ulint) !page_rec_is_comp(rec));
+}
+
+/****************************************************************
+Returns the sum of the sizes of the records in the record list, excluding
+the infimum and supremum records. */
+UNIV_INLINE
+ulint
+page_get_data_size(
+/*===============*/
+				/* out: data in bytes */
+	const page_t*	page)	/* in: index page */
+{
+	ulint	ret;
+
+	ret = (ulint)(page_header_get_field(page, PAGE_HEAP_TOP)
+		      - (page_is_comp(page)
+			 ? PAGE_NEW_SUPREMUM_END
+			 : PAGE_OLD_SUPREMUM_END)
+		      - page_header_get_field(page, PAGE_GARBAGE));
+
+	ut_ad(ret < UNIV_PAGE_SIZE);
+
+	return(ret);
+}
+
+
+/****************************************************************
+Allocates a block of memory from the free list of an index page. */
+UNIV_INTERN
+void
+page_mem_alloc_free(
+/*================*/
+	page_t*		page,	/* in/out: index page */
+	page_zip_des_t*	page_zip,/* in/out: compressed page with enough
+				space available for inserting the record,
+				or NULL */
+	rec_t*		next_rec,/* in: pointer to the new head of the
+				free record list */
+	ulint		need)	/* in: number of bytes allocated */
+{
+	ulint		garbage;
+
+#ifdef UNIV_DEBUG
+	const rec_t*	old_rec	= page_header_get_ptr(page, PAGE_FREE);
+	ulint		next_offs;
+
+	ut_ad(old_rec);
+	next_offs = rec_get_next_offs(old_rec, page_is_comp(page));
+	ut_ad(next_rec == (next_offs ? page + next_offs : NULL));
+#endif
+
+	page_header_set_ptr(page, page_zip, PAGE_FREE, next_rec);
+
+	garbage = page_header_get_field(page, PAGE_GARBAGE);
+	ut_ad(garbage >= need);
+
+	page_header_set_field(page, page_zip, PAGE_GARBAGE, garbage - need);
+}
+
+/*****************************************************************
+Calculates free space if a page is emptied. */
+UNIV_INLINE
+ulint
+page_get_free_space_of_empty(
+/*=========================*/
+				/* out: free space */
+	ulint	comp)		/* in: nonzero=compact page layout */
+{
+	if (UNIV_LIKELY(comp)) {
+		return((ulint)(UNIV_PAGE_SIZE
+			       - PAGE_NEW_SUPREMUM_END
+			       - PAGE_DIR
+			       - 2 * PAGE_DIR_SLOT_SIZE));
+	}
+
+	return((ulint)(UNIV_PAGE_SIZE
+		       - PAGE_OLD_SUPREMUM_END
+		       - PAGE_DIR
+		       - 2 * PAGE_DIR_SLOT_SIZE));
+}
+
+/****************************************************************
+Each user record on a page, and also the deleted user records in the heap
+takes its size plus the fraction of the dir cell size /
+PAGE_DIR_SLOT_MIN_N_OWNED bytes for it. If the sum of these exceeds the
+value of page_get_free_space_of_empty, the insert is impossible, otherwise
+it is allowed. This function returns the maximum combined size of records
+which can be inserted on top of the record heap. */
+UNIV_INLINE
+ulint
+page_get_max_insert_size(
+/*=====================*/
+				/* out: maximum combined size for
+				inserted records */
+	const page_t*	page,	/* in: index page */
+	ulint		n_recs)	/* in: number of records */
+{
+	ulint	occupied;
+	ulint	free_space;
+
+	if (page_is_comp(page)) {
+		occupied = page_header_get_field(page, PAGE_HEAP_TOP)
+			- PAGE_NEW_SUPREMUM_END
+			+ page_dir_calc_reserved_space(
+				n_recs + page_dir_get_n_heap(page) - 2);
+
+		free_space = page_get_free_space_of_empty(TRUE);
+	} else {
+		occupied = page_header_get_field(page, PAGE_HEAP_TOP)
+			- PAGE_OLD_SUPREMUM_END
+			+ page_dir_calc_reserved_space(
+				n_recs + page_dir_get_n_heap(page) - 2);
+
+		free_space = page_get_free_space_of_empty(FALSE);
+	}
+
+	/* Above the 'n_recs +' part reserves directory space for the new
+	inserted records; the '- 2' excludes page infimum and supremum
+	records */
+
+	if (occupied > free_space) {
+
+		return(0);
+	}
+
+	return(free_space - occupied);
+}
+
+/****************************************************************
+Returns the maximum combined size of records which can be inserted on top
+of the record heap if a page is first reorganized. */
+UNIV_INLINE
+ulint
+page_get_max_insert_size_after_reorganize(
+/*======================================*/
+				/* out: maximum combined size for
+				inserted records */
+	const page_t*	page,	/* in: index page */
+	ulint		n_recs)	/* in: number of records */
+{
+	ulint	occupied;
+	ulint	free_space;
+
+	occupied = page_get_data_size(page)
+		+ page_dir_calc_reserved_space(n_recs + page_get_n_recs(page));
+
+	free_space = page_get_free_space_of_empty(page_is_comp(page));
+
+	if (occupied > free_space) {
+
+		return(0);
+	}
+
+	return(free_space - occupied);
+}
+
+/****************************************************************
+Puts a record to free list. */
+UNIV_INLINE
+void
+page_mem_free(
+/*==========*/
+	page_t*		page,	/* in/out: index page */
+	page_zip_des_t*	page_zip,/* in/out: compressed page, or NULL */
+	rec_t*		rec,	/* in: pointer to the (origin of) record */
+	dict_index_t*	index,	/* in: index of rec */
+	const ulint*	offsets)/* in: array returned by rec_get_offsets() */
+{
+	rec_t*		free;
+	ulint		garbage;
+
+	ut_ad(rec_offs_validate(rec, index, offsets));
+	free = page_header_get_ptr(page, PAGE_FREE);
+
+	page_rec_set_next(rec, free);
+	page_header_set_ptr(page, page_zip, PAGE_FREE, rec);
+
+	garbage = page_header_get_field(page, PAGE_GARBAGE);
+
+	page_header_set_field(page, page_zip, PAGE_GARBAGE,
+			      garbage + rec_offs_size(offsets));
+
+	if (UNIV_LIKELY_NULL(page_zip)) {
+		page_zip_dir_delete(page_zip, rec, index, offsets, free);
+	} else {
+		page_header_set_field(page, page_zip, PAGE_N_RECS,
+				      page_get_n_recs(page) - 1);
+	}
+}
+
+#ifdef UNIV_MATERIALIZE
+#undef UNIV_INLINE
+#define UNIV_INLINE	UNIV_INLINE_ORIGINAL
+#endif
diff --git a/storage/xtradb/include/page0types.h b/storage/xtradb/include/page0types.h
new file mode 100644
index 00000000000..06af7a63d58
--- /dev/null
+++ b/storage/xtradb/include/page0types.h
@@ -0,0 +1,142 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Index page routines
+
+Created 2/2/1994 Heikki Tuuri
+*******************************************************/
+
+#ifndef page0types_h
+#define page0types_h
+
+#include "univ.i"
+#include "dict0types.h"
+#include "mtr0types.h"
+
+/* Type of the index page */
+/* The following define eliminates a name collision on HP-UX */
+#define page_t	   ib_page_t
+typedef	byte		page_t;
+typedef struct page_search_struct	page_search_t;
+typedef struct page_cur_struct	page_cur_t;
+
+typedef byte				page_zip_t;
+typedef struct page_zip_des_struct	page_zip_des_t;
+
+/* The following definitions would better belong to page0zip.h,
+but we cannot include page0zip.h from rem0rec.ic, because
+page0*.h includes rem0rec.h and may include rem0rec.ic. */
+
+#define PAGE_ZIP_SSIZE_BITS 3
+
+#define PAGE_ZIP_MIN_SIZE_SHIFT	10	/* log2 of smallest compressed size */
+#define PAGE_ZIP_MIN_SIZE	(1 << PAGE_ZIP_MIN_SIZE_SHIFT)
+
+#define PAGE_ZIP_NUM_SSIZE (UNIV_PAGE_SIZE_SHIFT - PAGE_ZIP_MIN_SIZE_SHIFT + 2)
+#if PAGE_ZIP_NUM_SSIZE > (1 << PAGE_ZIP_SSIZE_BITS)
+# error "PAGE_ZIP_NUM_SSIZE > (1 << PAGE_ZIP_SSIZE_BITS)"
+#endif
+
+/* Compressed page descriptor */
+struct page_zip_des_struct
+{
+	page_zip_t*	data;		/* compressed page data */
+
+#ifdef UNIV_DEBUG
+	unsigned	m_start:16;	/* start offset of modification log */
+#endif /* UNIV_DEBUG */
+	unsigned	m_end:16;	/* end offset of modification log */
+	unsigned	m_nonempty:1;	/* TRUE if the modification log
+					is not empty */
+	unsigned	n_blobs:12;	/* number of externally stored
+					columns on the page; the maximum
+					is 744 on a 16 KiB page */
+	unsigned	ssize:PAGE_ZIP_SSIZE_BITS;
+					/* 0 or compressed page size;
+					the size in bytes is
+					PAGE_ZIP_MIN_SIZE << (ssize - 1). */
+};
+
+/** Compression statistics for a given page size */
+struct page_zip_stat_struct {
+	/** Number of page compressions */
+	ulint		compressed;
+	/** Number of successful page compressions */
+	ulint		compressed_ok;
+	/** Number of page decompressions */
+	ulint		decompressed;
+	/** Duration of page compressions in microseconds */
+	ib_uint64_t	compressed_usec;
+	/** Duration of page decompressions in microseconds */
+	ib_uint64_t	decompressed_usec;
+};
+
+typedef struct page_zip_stat_struct page_zip_stat_t;
+
+/** Statistics on compression, indexed by page_zip_des_t::ssize - 1 */
+extern page_zip_stat_t page_zip_stat[PAGE_ZIP_NUM_SSIZE - 1];
+
+/**************************************************************************
+Write the "deleted" flag of a record on a compressed page.  The flag must
+already have been written on the uncompressed page. */
+UNIV_INTERN
+void
+page_zip_rec_set_deleted(
+/*=====================*/
+	page_zip_des_t*	page_zip,/* in/out: compressed page */
+	const byte*	rec,	/* in: record on the uncompressed page */
+	ulint		flag)	/* in: the deleted flag (nonzero=TRUE) */
+	__attribute__((nonnull));
+
+/**************************************************************************
+Write the "owned" flag of a record on a compressed page.  The n_owned field
+must already have been written on the uncompressed page. */
+UNIV_INTERN
+void
+page_zip_rec_set_owned(
+/*===================*/
+	page_zip_des_t*	page_zip,/* in/out: compressed page */
+	const byte*	rec,	/* in: record on the uncompressed page */
+	ulint		flag)	/* in: the owned flag (nonzero=TRUE) */
+	__attribute__((nonnull));
+
+/**************************************************************************
+Shift the dense page directory when a record is deleted. */
+UNIV_INTERN
+void
+page_zip_dir_delete(
+/*================*/
+	page_zip_des_t*	page_zip,/* in/out: compressed page */
+	byte*		rec,	/* in: deleted record */
+	dict_index_t*	index,	/* in: index of rec */
+	const ulint*	offsets,/* in: rec_get_offsets(rec) */
+	const byte*	free)	/* in: previous start of the free list */
+	__attribute__((nonnull(1,2,3,4)));
+
+/**************************************************************************
+Add a slot to the dense page directory. */
+UNIV_INTERN
+void
+page_zip_dir_add_slot(
+/*==================*/
+	page_zip_des_t*	page_zip,	/* in/out: compressed page */
+	ulint		is_clustered)	/* in: nonzero for clustered index,
+					zero for others */
+	__attribute__((nonnull));
+#endif
diff --git a/storage/xtradb/include/page0zip.h b/storage/xtradb/include/page0zip.h
new file mode 100644
index 00000000000..0183e013d05
--- /dev/null
+++ b/storage/xtradb/include/page0zip.h
@@ -0,0 +1,455 @@
+/*****************************************************************************
+
+Copyright (c) 2005, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Compressed page interface
+
+Created June 2005 by Marko Makela
+*******************************************************/
+
+#ifndef page0zip_h
+#define page0zip_h
+
+#ifdef UNIV_MATERIALIZE
+# undef UNIV_INLINE
+# define UNIV_INLINE
+#endif
+
+#include "mtr0types.h"
+#include "page0types.h"
+#include "buf0types.h"
+#include "dict0types.h"
+#include "mem0mem.h"
+
+/**************************************************************************
+Determine the size of a compressed page in bytes. */
+UNIV_INLINE
+ulint
+page_zip_get_size(
+/*==============*/
+						/* out: size in bytes */
+	const page_zip_des_t*	page_zip)	/* in: compressed page */
+	__attribute__((nonnull, pure));
+/**************************************************************************
+Set the size of a compressed page in bytes. */
+UNIV_INLINE
+void
+page_zip_set_size(
+/*==============*/
+	page_zip_des_t*	page_zip,	/* in/out: compressed page */
+	ulint		size);		/* in: size in bytes */
+
+/**************************************************************************
+Determine if a record is so big that it needs to be stored externally. */
+UNIV_INLINE
+ibool
+page_zip_rec_needs_ext(
+/*===================*/
+				/* out: FALSE if the entire record
+				can be stored locally on the page */
+	ulint	rec_size,	/* in: length of the record in bytes */
+	ulint	comp,		/* in: nonzero=compact format */
+	ulint	n_fields,	/* in: number of fields in the record;
+				ignored if zip_size == 0 */
+	ulint	zip_size)	/* in: compressed page size in bytes, or 0 */
+	__attribute__((const));
+
+/**************************************************************************
+Determine the guaranteed free space on an empty page. */
+UNIV_INTERN
+ulint
+page_zip_empty_size(
+/*================*/
+				/* out: minimum payload size on the page */
+	ulint	n_fields,	/* in: number of columns in the index */
+	ulint	zip_size)	/* in: compressed page size in bytes */
+	__attribute__((const));
+
+/**************************************************************************
+Initialize a compressed page descriptor. */
+UNIV_INLINE
+void
+page_zip_des_init(
+/*==============*/
+	page_zip_des_t*	page_zip);	/* in/out: compressed page
+					descriptor */
+
+/**************************************************************************
+Configure the zlib allocator to use the given memory heap. */
+UNIV_INTERN
+void
+page_zip_set_alloc(
+/*===============*/
+	void*		stream,		/* in/out: zlib stream */
+	mem_heap_t*	heap);		/* in: memory heap to use */
+
+/**************************************************************************
+Compress a page. */
+UNIV_INTERN
+ibool
+page_zip_compress(
+/*==============*/
+				/* out: TRUE on success, FALSE on failure;
+				page_zip will be left intact on failure. */
+	page_zip_des_t*	page_zip,/* in: size; out: data, n_blobs,
+				m_start, m_end, m_nonempty */
+	const page_t*	page,	/* in: uncompressed page */
+	dict_index_t*	index,	/* in: index of the B-tree node */
+	mtr_t*		mtr)	/* in: mini-transaction, or NULL */
+	__attribute__((nonnull(1,2,3)));
+
+/**************************************************************************
+Decompress a page.  This function should tolerate errors on the compressed
+page.  Instead of letting assertions fail, it will return FALSE if an
+inconsistency is detected. */
+UNIV_INTERN
+ibool
+page_zip_decompress(
+/*================*/
+				/* out: TRUE on success, FALSE on failure */
+	page_zip_des_t*	page_zip,/* in: data, ssize;
+				out: m_start, m_end, m_nonempty, n_blobs */
+	page_t*		page)	/* out: uncompressed page, may be trashed */
+	__attribute__((nonnull));
+
+#ifdef UNIV_DEBUG
+/**************************************************************************
+Validate a compressed page descriptor. */
+UNIV_INLINE
+ibool
+page_zip_simple_validate(
+/*=====================*/
+						/* out: TRUE if ok */
+	const page_zip_des_t*	page_zip);	/* in: compressed page
+						descriptor */
+#endif /* UNIV_DEBUG */
+
+#ifdef UNIV_ZIP_DEBUG
+/**************************************************************************
+Check that the compressed and decompressed pages match. */
+UNIV_INTERN
+ibool
+page_zip_validate_low(
+/*==================*/
+					/* out: TRUE if valid, FALSE if not */
+	const page_zip_des_t*	page_zip,/* in: compressed page */
+	const page_t*		page,	/* in: uncompressed page */
+	ibool			sloppy)	/* in: FALSE=strict,
+					TRUE=ignore the MIN_REC_FLAG */
+	__attribute__((nonnull));
+/**************************************************************************
+Check that the compressed and decompressed pages match. */
+UNIV_INTERN
+ibool
+page_zip_validate(
+/*==============*/
+	const page_zip_des_t*	page_zip,/* in: compressed page */
+	const page_t*		page)	/* in: uncompressed page */
+	__attribute__((nonnull));
+#endif /* UNIV_ZIP_DEBUG */
+
+/**************************************************************************
+Determine how big record can be inserted without recompressing the page. */
+UNIV_INLINE
+lint
+page_zip_max_ins_size(
+/*==================*/
+					/* out: a positive number
+					indicating the maximum size of
+					a record whose insertion is
+					guaranteed to succeed, or
+					zero or negative */
+	const page_zip_des_t*	page_zip,/* in: compressed page */
+	ibool			is_clust)/* in: TRUE if clustered index */
+	__attribute__((nonnull, pure));
+
+/**************************************************************************
+Determine if enough space is available in the modification log. */
+UNIV_INLINE
+ibool
+page_zip_available(
+/*===============*/
+					/* out: TRUE if page_zip_write_rec()
+					will succeed */
+	const page_zip_des_t*	page_zip,/* in: compressed page */
+	ibool			is_clust,/* in: TRUE if clustered index */
+	ulint			length,	/* in: combined size of the record */
+	ulint			create)	/* in: nonzero=add the record to
+					the heap */
+	__attribute__((nonnull, pure));
+
+/**************************************************************************
+Write data to the uncompressed header portion of a page.  The data must
+already have been written to the uncompressed page. */
+UNIV_INLINE
+void
+page_zip_write_header(
+/*==================*/
+	page_zip_des_t*	page_zip,/* in/out: compressed page */
+	const byte*	str,	/* in: address on the uncompressed page */
+	ulint		length,	/* in: length of the data */
+	mtr_t*		mtr)	/* in: mini-transaction, or NULL */
+	__attribute__((nonnull(1,2)));
+
+/**************************************************************************
+Write an entire record on the compressed page.  The data must already
+have been written to the uncompressed page. */
+UNIV_INTERN
+void
+page_zip_write_rec(
+/*===============*/
+	page_zip_des_t*	page_zip,/* in/out: compressed page */
+	const byte*	rec,	/* in: record being written */
+	dict_index_t*	index,	/* in: the index the record belongs to */
+	const ulint*	offsets,/* in: rec_get_offsets(rec, index) */
+	ulint		create)	/* in: nonzero=insert, zero=update */
+	__attribute__((nonnull));
+
+/***************************************************************
+Parses a log record of writing a BLOB pointer of a record. */
+UNIV_INTERN
+byte*
+page_zip_parse_write_blob_ptr(
+/*==========================*/
+				/* out: end of log record or NULL */
+	byte*		ptr,	/* in: redo log buffer */
+	byte*		end_ptr,/* in: redo log buffer end */
+	page_t*		page,	/* in/out: uncompressed page */
+	page_zip_des_t*	page_zip);/* in/out: compressed page */
+
+/**************************************************************************
+Write a BLOB pointer of a record on the leaf page of a clustered index.
+The information must already have been updated on the uncompressed page. */
+UNIV_INTERN
+void
+page_zip_write_blob_ptr(
+/*====================*/
+	page_zip_des_t*	page_zip,/* in/out: compressed page */
+	const byte*	rec,	/* in/out: record whose data is being
+				written */
+	dict_index_t*	index,	/* in: index of the page */
+	const ulint*	offsets,/* in: rec_get_offsets(rec, index) */
+	ulint		n,	/* in: column index */
+	mtr_t*		mtr)	/* in: mini-transaction handle,
+				or NULL if no logging is needed */
+	__attribute__((nonnull(1,2,3,4)));
+
+/***************************************************************
+Parses a log record of writing the node pointer of a record. */
+UNIV_INTERN
+byte*
+page_zip_parse_write_node_ptr(
+/*==========================*/
+				/* out: end of log record or NULL */
+	byte*		ptr,	/* in: redo log buffer */
+	byte*		end_ptr,/* in: redo log buffer end */
+	page_t*		page,	/* in/out: uncompressed page */
+	page_zip_des_t*	page_zip);/* in/out: compressed page */
+
+/**************************************************************************
+Write the node pointer of a record on a non-leaf compressed page. */
+UNIV_INTERN
+void
+page_zip_write_node_ptr(
+/*====================*/
+	page_zip_des_t*	page_zip,/* in/out: compressed page */
+	byte*		rec,	/* in/out: record */
+	ulint		size,	/* in: data size of rec */
+	ulint		ptr,	/* in: node pointer */
+	mtr_t*		mtr)	/* in: mini-transaction, or NULL */
+	__attribute__((nonnull(1,2)));
+
+/**************************************************************************
+Write the trx_id and roll_ptr of a record on a B-tree leaf node page. */
+UNIV_INTERN
+void
+page_zip_write_trx_id_and_roll_ptr(
+/*===============================*/
+	page_zip_des_t*	page_zip,/* in/out: compressed page */
+	byte*		rec,	/* in/out: record */
+	const ulint*	offsets,/* in: rec_get_offsets(rec, index) */
+	ulint		trx_id_col,/* in: column number of TRX_ID in rec */
+	dulint		trx_id,	/* in: transaction identifier */
+	dulint		roll_ptr)/* in: roll_ptr */
+	__attribute__((nonnull));
+
+/**************************************************************************
+Write the "deleted" flag of a record on a compressed page.  The flag must
+already have been written on the uncompressed page. */
+UNIV_INTERN
+void
+page_zip_rec_set_deleted(
+/*=====================*/
+	page_zip_des_t*	page_zip,/* in/out: compressed page */
+	const byte*	rec,	/* in: record on the uncompressed page */
+	ulint		flag)	/* in: the deleted flag (nonzero=TRUE) */
+	__attribute__((nonnull));
+
+/**************************************************************************
+Write the "owned" flag of a record on a compressed page.  The n_owned field
+must already have been written on the uncompressed page. */
+UNIV_INTERN
+void
+page_zip_rec_set_owned(
+/*===================*/
+	page_zip_des_t*	page_zip,/* in/out: compressed page */
+	const byte*	rec,	/* in: record on the uncompressed page */
+	ulint		flag)	/* in: the owned flag (nonzero=TRUE) */
+	__attribute__((nonnull));
+
+/**************************************************************************
+Insert a record to the dense page directory. */
+UNIV_INTERN
+void
+page_zip_dir_insert(
+/*================*/
+	page_zip_des_t*	page_zip,/* in/out: compressed page */
+	const byte*	prev_rec,/* in: record after which to insert */
+	const byte*	free_rec,/* in: record from which rec was
+				allocated, or NULL */
+	byte*		rec);	/* in: record to insert */
+
+/**************************************************************************
+Shift the dense page directory and the array of BLOB pointers
+when a record is deleted. */
+UNIV_INTERN
+void
+page_zip_dir_delete(
+/*================*/
+	page_zip_des_t*	page_zip,/* in/out: compressed page */
+	byte*		rec,	/* in: deleted record */
+	dict_index_t*	index,	/* in: index of rec */
+	const ulint*	offsets,/* in: rec_get_offsets(rec) */
+	const byte*	free)	/* in: previous start of the free list */
+	__attribute__((nonnull(1,2,3,4)));
+
+/**************************************************************************
+Add a slot to the dense page directory. */
+UNIV_INTERN
+void
+page_zip_dir_add_slot(
+/*==================*/
+	page_zip_des_t*	page_zip,	/* in/out: compressed page */
+	ulint		is_clustered)	/* in: nonzero for clustered index,
+					zero for others */
+	__attribute__((nonnull));
+
+/***************************************************************
+Parses a log record of writing to the header of a page. */
+UNIV_INTERN
+byte*
+page_zip_parse_write_header(
+/*========================*/
+				/* out: end of log record or NULL */
+	byte*		ptr,	/* in: redo log buffer */
+	byte*		end_ptr,/* in: redo log buffer end */
+	page_t*		page,	/* in/out: uncompressed page */
+	page_zip_des_t*	page_zip);/* in/out: compressed page */
+
+/**************************************************************************
+Write data to the uncompressed header portion of a page.  The data must
+already have been written to the uncompressed page.
+However, the data portion of the uncompressed page may differ from
+the compressed page when a record is being inserted in
+page_cur_insert_rec_low(). */
+UNIV_INLINE
+void
+page_zip_write_header(
+/*==================*/
+	page_zip_des_t*	page_zip,/* in/out: compressed page */
+	const byte*	str,	/* in: address on the uncompressed page */
+	ulint		length,	/* in: length of the data */
+	mtr_t*		mtr)	/* in: mini-transaction, or NULL */
+	__attribute__((nonnull(1,2)));
+
+/**************************************************************************
+Reorganize and compress a page.  This is a low-level operation for
+compressed pages, to be used when page_zip_compress() fails.
+On success, a redo log entry MLOG_ZIP_PAGE_COMPRESS will be written.
+The function btr_page_reorganize() should be preferred whenever possible.
+IMPORTANT: if page_zip_reorganize() is invoked on a leaf page of a
+non-clustered index, the caller must update the insert buffer free
+bits in the same mini-transaction in such a way that the modification
+will be redo-logged. */
+UNIV_INTERN
+ibool
+page_zip_reorganize(
+/*================*/
+				/* out: TRUE on success, FALSE on failure;
+				page and page_zip will be left intact
+				on failure. */
+	buf_block_t*	block,	/* in/out: page with compressed page;
+				on the compressed page, in: size;
+				out: data, n_blobs,
+				m_start, m_end, m_nonempty */
+	dict_index_t*	index,	/* in: index of the B-tree node */
+	mtr_t*		mtr)	/* in: mini-transaction */
+	__attribute__((nonnull));
+/**************************************************************************
+Copy the records of a page byte for byte.  Do not copy the page header
+or trailer, except those B-tree header fields that are directly
+related to the storage of records.  Also copy PAGE_MAX_TRX_ID.
+NOTE: The caller must update the lock table and the adaptive hash index. */
+UNIV_INTERN
+void
+page_zip_copy_recs(
+/*===============*/
+	page_zip_des_t*		page_zip,	/* out: copy of src_zip
+						(n_blobs, m_start, m_end,
+						m_nonempty, data[0..size-1]) */
+	page_t*			page,		/* out: copy of src */
+	const page_zip_des_t*	src_zip,	/* in: compressed page */
+	const page_t*		src,		/* in: page */
+	dict_index_t*		index,		/* in: index of the B-tree */
+	mtr_t*			mtr)		/* in: mini-transaction */
+	__attribute__((nonnull(1,2,3,4)));
+
+/**************************************************************************
+Parses a log record of compressing an index page. */
+UNIV_INTERN
+byte*
+page_zip_parse_compress(
+/*====================*/
+				/* out: end of log record or NULL */
+	byte*		ptr,	/* in: buffer */
+	byte*		end_ptr,/* in: buffer end */
+	page_t*		page,	/* out: uncompressed page */
+	page_zip_des_t*	page_zip)/* out: compressed page */
+	__attribute__((nonnull(1,2)));
+
+/**************************************************************************
+Calculate the compressed page checksum. */
+UNIV_INTERN
+ulint
+page_zip_calc_checksum(
+/*===================*/
+                                /* out: page checksum */
+        const void*     data,   /* in: compressed page */
+        ulint           size)   /* in: size of compressed page */
+	__attribute__((nonnull));
+
+#ifdef UNIV_MATERIALIZE
+# undef UNIV_INLINE
+# define UNIV_INLINE	UNIV_INLINE_ORIGINAL
+#endif
+
+#ifndef UNIV_NONINL
+# include "page0zip.ic"
+#endif
+
+#endif /* page0zip_h */
diff --git a/storage/xtradb/include/page0zip.ic b/storage/xtradb/include/page0zip.ic
new file mode 100644
index 00000000000..3db5f025c31
--- /dev/null
+++ b/storage/xtradb/include/page0zip.ic
@@ -0,0 +1,398 @@
+/*****************************************************************************
+
+Copyright (c) 2005, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Compressed page interface
+
+Created June 2005 by Marko Makela
+*******************************************************/
+
+#ifdef UNIV_MATERIALIZE
+# undef UNIV_INLINE
+# define UNIV_INLINE
+#endif
+
+#include "page0zip.h"
+#include "page0page.h"
+
+/* The format of compressed pages is as follows.
+
+The header and trailer of the uncompressed pages, excluding the page
+directory in the trailer, are copied as is to the header and trailer
+of the compressed page.
+
+At the end of the compressed page, there is a dense page directory
+pointing to every user record contained on the page, including deleted
+records on the free list.  The dense directory is indexed in the
+collation order, i.e., in the order in which the record list is
+linked on the uncompressed page.  The infimum and supremum records are
+excluded.  The two most significant bits of the entries are allocated
+for the delete-mark and an n_owned flag indicating the last record in
+a chain of records pointed to from the sparse page directory on the
+uncompressed page.
+
+The data between PAGE_ZIP_START and the last page directory entry will
+be written in compressed format, starting at offset PAGE_DATA.
+Infimum and supremum records are not stored.  We exclude the
+REC_N_NEW_EXTRA_BYTES in every record header.  These can be recovered
+from the dense page directory stored at the end of the compressed
+page.
+
+The fields node_ptr (in non-leaf B-tree nodes; level>0), trx_id and
+roll_ptr (in leaf B-tree nodes; level=0), and BLOB pointers of
+externally stored columns are stored separately, in ascending order of
+heap_no and column index, starting backwards from the dense page
+directory.
+
+The compressed data stream may be followed by a modification log
+covering the compressed portion of the page, as follows.
+
+MODIFICATION LOG ENTRY FORMAT
+- write record:
+  - (heap_no - 1) << 1 (1..2 bytes)
+  - extra bytes backwards
+  - data bytes
+- clear record:
+  - (heap_no - 1) << 1 | 1 (1..2 bytes)
+
+The integer values are stored in a variable-length format:
+- 0xxxxxxx: 0..127
+- 1xxxxxxx xxxxxxxx: 0..32767
+
+The end of the modification log is marked by a 0 byte.
+
+In summary, the compressed page looks like this:
+
+(1) Uncompressed page header (PAGE_DATA bytes)
+(2) Compressed index information
+(3) Compressed page data
+(4) Page modification log (page_zip->m_start..page_zip->m_end)
+(5) Empty zero-filled space
+(6) BLOB pointers (on leaf pages)
+  - BTR_EXTERN_FIELD_REF_SIZE for each externally stored column
+  - in descending collation order
+(7) Uncompressed columns of user records, n_dense * uncompressed_size bytes,
+  - indexed by heap_no
+  - DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN for leaf pages of clustered indexes
+  - REC_NODE_PTR_SIZE for non-leaf pages
+  - 0 otherwise
+(8) dense page directory, stored backwards
+  - n_dense = n_heap - 2
+  - existing records in ascending collation order
+  - deleted records (free list) in link order
+*/
+
+/* Start offset of the area that will be compressed */
+#define PAGE_ZIP_START		PAGE_NEW_SUPREMUM_END
+/* Size of an compressed page directory entry */
+#define PAGE_ZIP_DIR_SLOT_SIZE	2
+/* Mask of record offsets */
+#define PAGE_ZIP_DIR_SLOT_MASK	0x3fff
+/* 'owned' flag */
+#define PAGE_ZIP_DIR_SLOT_OWNED	0x4000
+/* 'deleted' flag */
+#define PAGE_ZIP_DIR_SLOT_DEL	0x8000
+
+/**************************************************************************
+Determine the size of a compressed page in bytes. */
+UNIV_INLINE
+ulint
+page_zip_get_size(
+/*==============*/
+						/* out: size in bytes */
+	const page_zip_des_t*	page_zip)	/* in: compressed page */
+{
+	ulint	size;
+
+	if (UNIV_UNLIKELY(!page_zip->ssize)) {
+		return(0);
+	}
+
+	size = (PAGE_ZIP_MIN_SIZE >> 1) << page_zip->ssize;
+
+	ut_ad(size >= PAGE_ZIP_MIN_SIZE);
+	ut_ad(size <= UNIV_PAGE_SIZE);
+
+	return(size);
+}
+/**************************************************************************
+Set the size of a compressed page in bytes. */
+UNIV_INLINE
+void
+page_zip_set_size(
+/*==============*/
+	page_zip_des_t*	page_zip,	/* in/out: compressed page */
+	ulint		size)		/* in: size in bytes */
+{
+	if (size) {
+		int	ssize;
+
+		ut_ad(ut_is_2pow(size));
+
+		for (ssize = 1; size > (ulint) (512 << ssize); ssize++) {
+		}
+
+		page_zip->ssize = ssize;
+	} else {
+		page_zip->ssize = 0;
+	}
+
+	ut_ad(page_zip_get_size(page_zip) == size);
+}
+
+/**************************************************************************
+Determine if a record is so big that it needs to be stored externally. */
+UNIV_INLINE
+ibool
+page_zip_rec_needs_ext(
+/*===================*/
+				/* out: FALSE if the entire record
+				can be stored locally on the page */
+	ulint	rec_size,	/* in: length of the record in bytes */
+	ulint	comp,		/* in: nonzero=compact format */
+	ulint	n_fields,	/* in: number of fields in the record;
+				ignored if zip_size == 0 */
+	ulint	zip_size)	/* in: compressed page size in bytes, or 0 */
+{
+	ut_ad(rec_size > comp ? REC_N_NEW_EXTRA_BYTES : REC_N_OLD_EXTRA_BYTES);
+	ut_ad(ut_is_2pow(zip_size));
+	ut_ad(comp || !zip_size);
+
+#if UNIV_PAGE_SIZE > REC_MAX_DATA_SIZE
+	if (UNIV_UNLIKELY(rec_size >= REC_MAX_DATA_SIZE)) {
+		return(TRUE);
+	}
+#endif
+
+	if (UNIV_UNLIKELY(zip_size)) {
+		ut_ad(comp);
+		/* On a compressed page, there is a two-byte entry in
+		the dense page directory for every record.  But there
+		is no record header.  There should be enough room for
+		one record on an empty leaf page.  Subtract 1 byte for
+		the encoded heap number.  Check also the available space
+		on the uncompressed page. */
+		return(rec_size - (REC_N_NEW_EXTRA_BYTES - 2)
+		       >= (page_zip_empty_size(n_fields, zip_size) - 1)
+		       || rec_size >= page_get_free_space_of_empty(TRUE) / 2);
+	}
+
+	return(rec_size >= page_get_free_space_of_empty(comp) / 2);
+}
+
+#ifdef UNIV_DEBUG
+/**************************************************************************
+Validate a compressed page descriptor. */
+UNIV_INLINE
+ibool
+page_zip_simple_validate(
+/*=====================*/
+					/* out: TRUE if ok */
+	const page_zip_des_t*	page_zip)/* in: compressed page descriptor */
+{
+	ut_ad(page_zip);
+	ut_ad(page_zip->data);
+	ut_ad(page_zip->ssize < PAGE_ZIP_NUM_SSIZE);
+	ut_ad(page_zip_get_size(page_zip)
+	      > PAGE_DATA + PAGE_ZIP_DIR_SLOT_SIZE);
+	ut_ad(page_zip->m_start <= page_zip->m_end);
+	ut_ad(page_zip->m_end < page_zip_get_size(page_zip));
+	ut_ad(page_zip->n_blobs
+	      < page_zip_get_size(page_zip) / BTR_EXTERN_FIELD_REF_SIZE);
+	return(TRUE);
+}
+#endif /* UNIV_DEBUG */
+
+/**************************************************************************
+Determine if the length of the page trailer. */
+UNIV_INLINE
+ibool
+page_zip_get_trailer_len(
+/*=====================*/
+					/* out: length of the page trailer,
+					in bytes, not including the terminating
+					zero byte of the modification log */
+	const page_zip_des_t*	page_zip,/* in: compressed page */
+	ibool			is_clust,/* in: TRUE if clustered index */
+	ulint*			entry_size)/* out: size of the uncompressed
+					portion of a user record */
+{
+	ulint	uncompressed_size;
+
+	ut_ad(page_zip_simple_validate(page_zip));
+	UNIV_MEM_ASSERT_RW(page_zip->data, page_zip_get_size(page_zip));
+
+	if (UNIV_UNLIKELY(!page_is_leaf(page_zip->data))) {
+		uncompressed_size = PAGE_ZIP_DIR_SLOT_SIZE
+			+ REC_NODE_PTR_SIZE;
+		ut_ad(!page_zip->n_blobs);
+	} else if (UNIV_UNLIKELY(is_clust)) {
+		uncompressed_size = PAGE_ZIP_DIR_SLOT_SIZE
+			+ DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN;
+	} else {
+		uncompressed_size = PAGE_ZIP_DIR_SLOT_SIZE;
+		ut_ad(!page_zip->n_blobs);
+	}
+
+	if (entry_size) {
+		*entry_size = uncompressed_size;
+	}
+
+	return((page_dir_get_n_heap(page_zip->data) - 2)
+	       * uncompressed_size
+	       + page_zip->n_blobs * BTR_EXTERN_FIELD_REF_SIZE);
+}
+
+/**************************************************************************
+Determine how big record can be inserted without recompressing the page. */
+UNIV_INLINE
+lint
+page_zip_max_ins_size(
+/*==================*/
+					/* out: a positive number
+					indicating the maximum size of
+					a record whose insertion is
+					guaranteed to succeed, or
+					zero or negative */
+	const page_zip_des_t*	page_zip,/* in: compressed page */
+	ibool			is_clust)/* in: TRUE if clustered index */
+{
+	ulint	uncompressed_size;
+	ulint	trailer_len;
+
+	trailer_len = page_zip_get_trailer_len(page_zip, is_clust,
+					       &uncompressed_size);
+
+	/* When a record is created, a pointer may be added to
+	the dense directory.
+	Likewise, space for the columns that will not be
+	compressed will be allocated from the page trailer.
+	Also the BLOB pointers will be allocated from there, but
+	we may as well count them in the length of the record. */
+
+	trailer_len += uncompressed_size;
+
+	return((lint) page_zip_get_size(page_zip)
+	       - trailer_len - page_zip->m_end
+	       - (REC_N_NEW_EXTRA_BYTES - 2));
+}
+
+/**************************************************************************
+Determine if enough space is available in the modification log. */
+UNIV_INLINE
+ibool
+page_zip_available(
+/*===============*/
+					/* out: TRUE if enough space
+					is available */
+	const page_zip_des_t*	page_zip,/* in: compressed page */
+	ibool			is_clust,/* in: TRUE if clustered index */
+	ulint			length,	/* in: combined size of the record */
+	ulint			create)	/* in: nonzero=add the record to
+					the heap */
+{
+	ulint	uncompressed_size;
+	ulint	trailer_len;
+
+	ut_ad(length > REC_N_NEW_EXTRA_BYTES);
+
+	trailer_len = page_zip_get_trailer_len(page_zip, is_clust,
+					       &uncompressed_size);
+
+	/* Subtract the fixed extra bytes and add the maximum
+	space needed for identifying the record (encoded heap_no). */
+	length -= REC_N_NEW_EXTRA_BYTES - 2;
+
+	if (UNIV_UNLIKELY(create)) {
+		/* When a record is created, a pointer may be added to
+		the dense directory.
+		Likewise, space for the columns that will not be
+		compressed will be allocated from the page trailer.
+		Also the BLOB pointers will be allocated from there, but
+		we may as well count them in the length of the record. */
+
+		trailer_len += uncompressed_size;
+	}
+
+	return(UNIV_LIKELY(length
+			   + trailer_len
+			   + page_zip->m_end
+			   < page_zip_get_size(page_zip)));
+}
+
+/**************************************************************************
+Initialize a compressed page descriptor. */
+UNIV_INLINE
+void
+page_zip_des_init(
+/*==============*/
+	page_zip_des_t*	page_zip)	/* in/out: compressed page
+					descriptor */
+{
+	memset(page_zip, 0, sizeof *page_zip);
+}
+
+/**************************************************************************
+Write a log record of writing to the uncompressed header portion of a page. */
+UNIV_INTERN
+void
+page_zip_write_header_log(
+/*======================*/
+	const byte*	data,/* in: data on the uncompressed page */
+	ulint		length,	/* in: length of the data */
+	mtr_t*		mtr);	/* in: mini-transaction */
+
+/**************************************************************************
+Write data to the uncompressed header portion of a page.  The data must
+already have been written to the uncompressed page.
+However, the data portion of the uncompressed page may differ from
+the compressed page when a record is being inserted in
+page_cur_insert_rec_zip(). */
+UNIV_INLINE
+void
+page_zip_write_header(
+/*==================*/
+	page_zip_des_t*	page_zip,/* in/out: compressed page */
+	const byte*	str,	/* in: address on the uncompressed page */
+	ulint		length,	/* in: length of the data */
+	mtr_t*		mtr)	/* in: mini-transaction, or NULL */
+{
+	ulint	pos;
+
+	ut_ad(buf_frame_get_page_zip(str) == page_zip);
+	ut_ad(page_zip_simple_validate(page_zip));
+	UNIV_MEM_ASSERT_RW(page_zip->data, page_zip_get_size(page_zip));
+
+	pos = page_offset(str);
+
+	ut_ad(pos < PAGE_DATA);
+
+	memcpy(page_zip->data + pos, str, length);
+
+	/* The following would fail in page_cur_insert_rec_zip(). */
+	/* ut_ad(page_zip_validate(page_zip, str - pos)); */
+
+	if (UNIV_LIKELY_NULL(mtr)) {
+		page_zip_write_header_log(str, length, mtr);
+	}
+}
+
+#ifdef UNIV_MATERIALIZE
+# undef UNIV_INLINE
+# define UNIV_INLINE	UNIV_INLINE_ORIGINAL
+#endif
diff --git a/storage/xtradb/include/pars0grm.h b/storage/xtradb/include/pars0grm.h
new file mode 100644
index 00000000000..3de233eed3a
--- /dev/null
+++ b/storage/xtradb/include/pars0grm.h
@@ -0,0 +1,236 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+Copyright (c) 1984, 1989, 1990, 2000, 2001, 2002, 2003, 2004 Free Software
+Foundation, Inc.
+
+As a special exception, when this file is copied by Bison into a
+Bison output file, you may use that output file without restriction.
+This special exception was added by the Free Software Foundation
+in version 1.24 of Bison.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/* A Bison parser, made by GNU Bison 1.875d.  */
+
+/* Tokens.  */
+#ifndef YYTOKENTYPE
+# define YYTOKENTYPE
+   /* Put the tokens into the symbol table, so that GDB and other debuggers
+      know about them.  */
+   enum yytokentype {
+     PARS_INT_LIT = 258,
+     PARS_FLOAT_LIT = 259,
+     PARS_STR_LIT = 260,
+     PARS_FIXBINARY_LIT = 261,
+     PARS_BLOB_LIT = 262,
+     PARS_NULL_LIT = 263,
+     PARS_ID_TOKEN = 264,
+     PARS_AND_TOKEN = 265,
+     PARS_OR_TOKEN = 266,
+     PARS_NOT_TOKEN = 267,
+     PARS_GE_TOKEN = 268,
+     PARS_LE_TOKEN = 269,
+     PARS_NE_TOKEN = 270,
+     PARS_PROCEDURE_TOKEN = 271,
+     PARS_IN_TOKEN = 272,
+     PARS_OUT_TOKEN = 273,
+     PARS_BINARY_TOKEN = 274,
+     PARS_BLOB_TOKEN = 275,
+     PARS_INT_TOKEN = 276,
+     PARS_INTEGER_TOKEN = 277,
+     PARS_FLOAT_TOKEN = 278,
+     PARS_CHAR_TOKEN = 279,
+     PARS_IS_TOKEN = 280,
+     PARS_BEGIN_TOKEN = 281,
+     PARS_END_TOKEN = 282,
+     PARS_IF_TOKEN = 283,
+     PARS_THEN_TOKEN = 284,
+     PARS_ELSE_TOKEN = 285,
+     PARS_ELSIF_TOKEN = 286,
+     PARS_LOOP_TOKEN = 287,
+     PARS_WHILE_TOKEN = 288,
+     PARS_RETURN_TOKEN = 289,
+     PARS_SELECT_TOKEN = 290,
+     PARS_SUM_TOKEN = 291,
+     PARS_COUNT_TOKEN = 292,
+     PARS_DISTINCT_TOKEN = 293,
+     PARS_FROM_TOKEN = 294,
+     PARS_WHERE_TOKEN = 295,
+     PARS_FOR_TOKEN = 296,
+     PARS_DDOT_TOKEN = 297,
+     PARS_READ_TOKEN = 298,
+     PARS_ORDER_TOKEN = 299,
+     PARS_BY_TOKEN = 300,
+     PARS_ASC_TOKEN = 301,
+     PARS_DESC_TOKEN = 302,
+     PARS_INSERT_TOKEN = 303,
+     PARS_INTO_TOKEN = 304,
+     PARS_VALUES_TOKEN = 305,
+     PARS_UPDATE_TOKEN = 306,
+     PARS_SET_TOKEN = 307,
+     PARS_DELETE_TOKEN = 308,
+     PARS_CURRENT_TOKEN = 309,
+     PARS_OF_TOKEN = 310,
+     PARS_CREATE_TOKEN = 311,
+     PARS_TABLE_TOKEN = 312,
+     PARS_INDEX_TOKEN = 313,
+     PARS_UNIQUE_TOKEN = 314,
+     PARS_CLUSTERED_TOKEN = 315,
+     PARS_DOES_NOT_FIT_IN_MEM_TOKEN = 316,
+     PARS_ON_TOKEN = 317,
+     PARS_ASSIGN_TOKEN = 318,
+     PARS_DECLARE_TOKEN = 319,
+     PARS_CURSOR_TOKEN = 320,
+     PARS_SQL_TOKEN = 321,
+     PARS_OPEN_TOKEN = 322,
+     PARS_FETCH_TOKEN = 323,
+     PARS_CLOSE_TOKEN = 324,
+     PARS_NOTFOUND_TOKEN = 325,
+     PARS_TO_CHAR_TOKEN = 326,
+     PARS_TO_NUMBER_TOKEN = 327,
+     PARS_TO_BINARY_TOKEN = 328,
+     PARS_BINARY_TO_NUMBER_TOKEN = 329,
+     PARS_SUBSTR_TOKEN = 330,
+     PARS_REPLSTR_TOKEN = 331,
+     PARS_CONCAT_TOKEN = 332,
+     PARS_INSTR_TOKEN = 333,
+     PARS_LENGTH_TOKEN = 334,
+     PARS_SYSDATE_TOKEN = 335,
+     PARS_PRINTF_TOKEN = 336,
+     PARS_ASSERT_TOKEN = 337,
+     PARS_RND_TOKEN = 338,
+     PARS_RND_STR_TOKEN = 339,
+     PARS_ROW_PRINTF_TOKEN = 340,
+     PARS_COMMIT_TOKEN = 341,
+     PARS_ROLLBACK_TOKEN = 342,
+     PARS_WORK_TOKEN = 343,
+     PARS_UNSIGNED_TOKEN = 344,
+     PARS_EXIT_TOKEN = 345,
+     PARS_FUNCTION_TOKEN = 346,
+     PARS_LOCK_TOKEN = 347,
+     PARS_SHARE_TOKEN = 348,
+     PARS_MODE_TOKEN = 349,
+     NEG = 350
+   };
+#endif
+#define PARS_INT_LIT 258
+#define PARS_FLOAT_LIT 259
+#define PARS_STR_LIT 260
+#define PARS_FIXBINARY_LIT 261
+#define PARS_BLOB_LIT 262
+#define PARS_NULL_LIT 263
+#define PARS_ID_TOKEN 264
+#define PARS_AND_TOKEN 265
+#define PARS_OR_TOKEN 266
+#define PARS_NOT_TOKEN 267
+#define PARS_GE_TOKEN 268
+#define PARS_LE_TOKEN 269
+#define PARS_NE_TOKEN 270
+#define PARS_PROCEDURE_TOKEN 271
+#define PARS_IN_TOKEN 272
+#define PARS_OUT_TOKEN 273
+#define PARS_BINARY_TOKEN 274
+#define PARS_BLOB_TOKEN 275
+#define PARS_INT_TOKEN 276
+#define PARS_INTEGER_TOKEN 277
+#define PARS_FLOAT_TOKEN 278
+#define PARS_CHAR_TOKEN 279
+#define PARS_IS_TOKEN 280
+#define PARS_BEGIN_TOKEN 281
+#define PARS_END_TOKEN 282
+#define PARS_IF_TOKEN 283
+#define PARS_THEN_TOKEN 284
+#define PARS_ELSE_TOKEN 285
+#define PARS_ELSIF_TOKEN 286
+#define PARS_LOOP_TOKEN 287
+#define PARS_WHILE_TOKEN 288
+#define PARS_RETURN_TOKEN 289
+#define PARS_SELECT_TOKEN 290
+#define PARS_SUM_TOKEN 291
+#define PARS_COUNT_TOKEN 292
+#define PARS_DISTINCT_TOKEN 293
+#define PARS_FROM_TOKEN 294
+#define PARS_WHERE_TOKEN 295
+#define PARS_FOR_TOKEN 296
+#define PARS_DDOT_TOKEN 297
+#define PARS_READ_TOKEN 298
+#define PARS_ORDER_TOKEN 299
+#define PARS_BY_TOKEN 300
+#define PARS_ASC_TOKEN 301
+#define PARS_DESC_TOKEN 302
+#define PARS_INSERT_TOKEN 303
+#define PARS_INTO_TOKEN 304
+#define PARS_VALUES_TOKEN 305
+#define PARS_UPDATE_TOKEN 306
+#define PARS_SET_TOKEN 307
+#define PARS_DELETE_TOKEN 308
+#define PARS_CURRENT_TOKEN 309
+#define PARS_OF_TOKEN 310
+#define PARS_CREATE_TOKEN 311
+#define PARS_TABLE_TOKEN 312
+#define PARS_INDEX_TOKEN 313
+#define PARS_UNIQUE_TOKEN 314
+#define PARS_CLUSTERED_TOKEN 315
+#define PARS_DOES_NOT_FIT_IN_MEM_TOKEN 316
+#define PARS_ON_TOKEN 317
+#define PARS_ASSIGN_TOKEN 318
+#define PARS_DECLARE_TOKEN 319
+#define PARS_CURSOR_TOKEN 320
+#define PARS_SQL_TOKEN 321
+#define PARS_OPEN_TOKEN 322
+#define PARS_FETCH_TOKEN 323
+#define PARS_CLOSE_TOKEN 324
+#define PARS_NOTFOUND_TOKEN 325
+#define PARS_TO_CHAR_TOKEN 326
+#define PARS_TO_NUMBER_TOKEN 327
+#define PARS_TO_BINARY_TOKEN 328
+#define PARS_BINARY_TO_NUMBER_TOKEN 329
+#define PARS_SUBSTR_TOKEN 330
+#define PARS_REPLSTR_TOKEN 331
+#define PARS_CONCAT_TOKEN 332
+#define PARS_INSTR_TOKEN 333
+#define PARS_LENGTH_TOKEN 334
+#define PARS_SYSDATE_TOKEN 335
+#define PARS_PRINTF_TOKEN 336
+#define PARS_ASSERT_TOKEN 337
+#define PARS_RND_TOKEN 338
+#define PARS_RND_STR_TOKEN 339
+#define PARS_ROW_PRINTF_TOKEN 340
+#define PARS_COMMIT_TOKEN 341
+#define PARS_ROLLBACK_TOKEN 342
+#define PARS_WORK_TOKEN 343
+#define PARS_UNSIGNED_TOKEN 344
+#define PARS_EXIT_TOKEN 345
+#define PARS_FUNCTION_TOKEN 346
+#define PARS_LOCK_TOKEN 347
+#define PARS_SHARE_TOKEN 348
+#define PARS_MODE_TOKEN 349
+#define NEG 350
+
+
+
+
+#if ! defined (YYSTYPE) && ! defined (YYSTYPE_IS_DECLARED)
+typedef int YYSTYPE;
+# define yystype YYSTYPE /* obsolescent; will be withdrawn */
+# define YYSTYPE_IS_DECLARED 1
+# define YYSTYPE_IS_TRIVIAL 1
+#endif
+
+extern YYSTYPE yylval;
+
+
+
diff --git a/storage/xtradb/include/pars0opt.h b/storage/xtradb/include/pars0opt.h
new file mode 100644
index 00000000000..02524e9d893
--- /dev/null
+++ b/storage/xtradb/include/pars0opt.h
@@ -0,0 +1,74 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Simple SQL optimizer
+
+Created 12/21/1997 Heikki Tuuri
+*******************************************************/
+
+#ifndef pars0opt_h
+#define pars0opt_h
+
+#include "univ.i"
+#include "que0types.h"
+#include "usr0types.h"
+#include "pars0sym.h"
+#include "dict0types.h"
+#include "row0sel.h"
+
+/***********************************************************************
+Optimizes a select. Decides which indexes to tables to use. The tables
+are accessed in the order that they were written to the FROM part in the
+select statement. */
+UNIV_INTERN
+void
+opt_search_plan(
+/*============*/
+	sel_node_t*	sel_node);	/* in: parsed select node */
+/***********************************************************************
+Looks for occurrences of the columns of the table in the query subgraph and
+adds them to the list of columns if an occurrence of the same column does not
+already exist in the list. If the column is already in the list, puts a value
+indirection to point to the occurrence in the column list, except if the
+column occurrence we are looking at is in the column list, in which case
+nothing is done. */
+UNIV_INTERN
+void
+opt_find_all_cols(
+/*==============*/
+	ibool		copy_val,	/* in: if TRUE, new found columns are
+					added as columns to copy */
+	dict_index_t*	index,		/* in: index to use */
+	sym_node_list_t* col_list,	/* in: base node of a list where
+					to add new found columns */
+	plan_t*		plan,		/* in: plan or NULL */
+	que_node_t*	exp);		/* in: expression or condition */
+/************************************************************************
+Prints info of a query plan. */
+UNIV_INTERN
+void
+opt_print_query_plan(
+/*=================*/
+	sel_node_t*	sel_node);	/* in: select node */
+
+#ifndef UNIV_NONINL
+#include "pars0opt.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/pars0opt.ic b/storage/xtradb/include/pars0opt.ic
new file mode 100644
index 00000000000..35653453b30
--- /dev/null
+++ b/storage/xtradb/include/pars0opt.ic
@@ -0,0 +1,23 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Simple SQL optimizer
+
+Created 12/21/1997 Heikki Tuuri
+*******************************************************/
diff --git a/storage/xtradb/include/pars0pars.h b/storage/xtradb/include/pars0pars.h
new file mode 100644
index 00000000000..e5693ee5575
--- /dev/null
+++ b/storage/xtradb/include/pars0pars.h
@@ -0,0 +1,747 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+SQL parser
+
+Created 11/19/1996 Heikki Tuuri
+*******************************************************/
+
+#ifndef pars0pars_h
+#define pars0pars_h
+
+#include "univ.i"
+#include "que0types.h"
+#include "usr0types.h"
+#include "pars0types.h"
+#include "row0types.h"
+#include "trx0types.h"
+#include "ut0vec.h"
+
+/* Type of the user functions. The first argument is always InnoDB-supplied
+and varies in type, while 'user_arg' is a user-supplied argument. The
+meaning of the return type also varies. See the individual use cases, e.g.
+the FETCH statement, for details on them. */
+typedef void* (*pars_user_func_cb_t)(void* arg, void* user_arg);
+
+extern int	yydebug;
+
+/* If the following is set TRUE, the lexer will print the SQL string
+as it tokenizes it */
+
+#ifdef UNIV_SQL_DEBUG
+extern ibool	pars_print_lexed;
+#endif /* UNIV_SQL_DEBUG */
+
+/* Global variable used while parsing a single procedure or query : the code is
+NOT re-entrant */
+extern sym_tab_t*	pars_sym_tab_global;
+
+extern pars_res_word_t	pars_to_char_token;
+extern pars_res_word_t	pars_to_number_token;
+extern pars_res_word_t	pars_to_binary_token;
+extern pars_res_word_t	pars_binary_to_number_token;
+extern pars_res_word_t	pars_substr_token;
+extern pars_res_word_t	pars_replstr_token;
+extern pars_res_word_t	pars_concat_token;
+extern pars_res_word_t	pars_length_token;
+extern pars_res_word_t	pars_instr_token;
+extern pars_res_word_t	pars_sysdate_token;
+extern pars_res_word_t	pars_printf_token;
+extern pars_res_word_t	pars_assert_token;
+extern pars_res_word_t	pars_rnd_token;
+extern pars_res_word_t	pars_rnd_str_token;
+extern pars_res_word_t	pars_count_token;
+extern pars_res_word_t	pars_sum_token;
+extern pars_res_word_t	pars_distinct_token;
+extern pars_res_word_t	pars_binary_token;
+extern pars_res_word_t	pars_blob_token;
+extern pars_res_word_t	pars_int_token;
+extern pars_res_word_t	pars_char_token;
+extern pars_res_word_t	pars_float_token;
+extern pars_res_word_t	pars_update_token;
+extern pars_res_word_t	pars_asc_token;
+extern pars_res_word_t	pars_desc_token;
+extern pars_res_word_t	pars_open_token;
+extern pars_res_word_t	pars_close_token;
+extern pars_res_word_t	pars_share_token;
+extern pars_res_word_t	pars_unique_token;
+extern pars_res_word_t	pars_clustered_token;
+
+extern ulint		pars_star_denoter;
+
+/* Procedure parameter types */
+#define PARS_INPUT	0
+#define PARS_OUTPUT	1
+#define PARS_NOT_PARAM	2
+
+int
+yyparse(void);
+
+/*****************************************************************
+Parses an SQL string returning the query graph. */
+UNIV_INTERN
+que_t*
+pars_sql(
+/*=====*/
+				/* out, own: the query graph */
+	pars_info_t*	info,	/* in: extra information, or NULL */
+	const char*	str);	/* in: SQL string */
+/*****************************************************************
+Retrieves characters to the lexical analyzer. */
+UNIV_INTERN
+void
+pars_get_lex_chars(
+/*===============*/
+	char*	buf,		/* in/out: buffer where to copy */
+	int*	result,		/* out: number of characters copied or EOF */
+	int	max_size);	/* in: maximum number of characters which fit
+				in the buffer */
+/*****************************************************************
+Called by yyparse on error. */
+UNIV_INTERN
+void
+yyerror(
+/*====*/
+	const char*	s);	/* in: error message string */
+/*************************************************************************
+Parses a variable declaration. */
+UNIV_INTERN
+sym_node_t*
+pars_variable_declaration(
+/*======================*/
+				/* out, own: symbol table node of type
+				SYM_VAR */
+	sym_node_t*	node,	/* in: symbol table node allocated for the
+				id of the variable */
+	pars_res_word_t* type);	/* in: pointer to a type token */
+/*************************************************************************
+Parses a function expression. */
+UNIV_INTERN
+func_node_t*
+pars_func(
+/*======*/
+				/* out, own: function node in a query tree */
+	que_node_t*	res_word,/* in: function name reserved word */
+	que_node_t*	arg);	/* in: first argument in the argument list */
+/*************************************************************************
+Parses an operator expression. */
+UNIV_INTERN
+func_node_t*
+pars_op(
+/*====*/
+				/* out, own: function node in a query tree */
+	int		func,	/* in: operator token code */
+	que_node_t*	arg1,	/* in: first argument */
+	que_node_t*	arg2);	/* in: second argument or NULL for an unary
+				operator */
+/*************************************************************************
+Parses an ORDER BY clause. Order by a single column only is supported. */
+UNIV_INTERN
+order_node_t*
+pars_order_by(
+/*==========*/
+				/* out, own: order-by node in a query tree */
+	sym_node_t*	column,	/* in: column name */
+	pars_res_word_t* asc);	/* in: &pars_asc_token or pars_desc_token */
+/*************************************************************************
+Parses a select list; creates a query graph node for the whole SELECT
+statement. */
+UNIV_INTERN
+sel_node_t*
+pars_select_list(
+/*=============*/
+					/* out, own: select node in a query
+					tree */
+	que_node_t*	select_list,	/* in: select list */
+	sym_node_t*	into_list);	/* in: variables list or NULL */
+/*************************************************************************
+Parses a cursor declaration. */
+UNIV_INTERN
+que_node_t*
+pars_cursor_declaration(
+/*====================*/
+					/* out: sym_node */
+	sym_node_t*	sym_node,	/* in: cursor id node in the symbol
+					table */
+	sel_node_t*	select_node);	/* in: select node */
+/*************************************************************************
+Parses a function declaration. */
+UNIV_INTERN
+que_node_t*
+pars_function_declaration(
+/*======================*/
+					/* out: sym_node */
+	sym_node_t*	sym_node);	/* in: function id node in the symbol
+					table */
+/*************************************************************************
+Parses a select statement. */
+UNIV_INTERN
+sel_node_t*
+pars_select_statement(
+/*==================*/
+					/* out, own: select node in a query
+					tree */
+	sel_node_t*	select_node,	/* in: select node already containing
+					the select list */
+	sym_node_t*	table_list,	/* in: table list */
+	que_node_t*	search_cond,	/* in: search condition or NULL */
+	pars_res_word_t* for_update,	/* in: NULL or &pars_update_token */
+	pars_res_word_t* consistent_read,/* in: NULL or
+						&pars_consistent_token */
+	order_node_t*	order_by);	/* in: NULL or an order-by node */
+/*************************************************************************
+Parses a column assignment in an update. */
+UNIV_INTERN
+col_assign_node_t*
+pars_column_assignment(
+/*===================*/
+				/* out: column assignment node */
+	sym_node_t*	column,	/* in: column to assign */
+	que_node_t*	exp);	/* in: value to assign */
+/*************************************************************************
+Parses a delete or update statement start. */
+UNIV_INTERN
+upd_node_t*
+pars_update_statement_start(
+/*========================*/
+					/* out, own: update node in a query
+					tree */
+	ibool		is_delete,	/* in: TRUE if delete */
+	sym_node_t*	table_sym,	/* in: table name node */
+	col_assign_node_t* col_assign_list);/* in: column assignment list, NULL
+					if delete */
+/*************************************************************************
+Parses an update or delete statement. */
+UNIV_INTERN
+upd_node_t*
+pars_update_statement(
+/*==================*/
+					/* out, own: update node in a query
+					tree */
+	upd_node_t*	node,		/* in: update node */
+	sym_node_t*	cursor_sym,	/* in: pointer to a cursor entry in
+					the symbol table or NULL */
+	que_node_t*	search_cond);	/* in: search condition or NULL */
+/*************************************************************************
+Parses an insert statement. */
+UNIV_INTERN
+ins_node_t*
+pars_insert_statement(
+/*==================*/
+					/* out, own: update node in a query
+					tree */
+	sym_node_t*	table_sym,	/* in: table name node */
+	que_node_t*	values_list,	/* in: value expression list or NULL */
+	sel_node_t*	select);	/* in: select condition or NULL */
+/*************************************************************************
+Parses a procedure parameter declaration. */
+UNIV_INTERN
+sym_node_t*
+pars_parameter_declaration(
+/*=======================*/
+				/* out, own: symbol table node of type
+				SYM_VAR */
+	sym_node_t*	node,	/* in: symbol table node allocated for the
+				id of the parameter */
+	ulint		param_type,
+				/* in: PARS_INPUT or PARS_OUTPUT */
+	pars_res_word_t* type);	/* in: pointer to a type token */
+/*************************************************************************
+Parses an elsif element. */
+UNIV_INTERN
+elsif_node_t*
+pars_elsif_element(
+/*===============*/
+					/* out: elsif node */
+	que_node_t*	cond,		/* in: if-condition */
+	que_node_t*	stat_list);	/* in: statement list */
+/*************************************************************************
+Parses an if-statement. */
+UNIV_INTERN
+if_node_t*
+pars_if_statement(
+/*==============*/
+					/* out: if-statement node */
+	que_node_t*	cond,		/* in: if-condition */
+	que_node_t*	stat_list,	/* in: statement list */
+	que_node_t*	else_part);	/* in: else-part statement list */
+/*************************************************************************
+Parses a for-loop-statement. */
+UNIV_INTERN
+for_node_t*
+pars_for_statement(
+/*===============*/
+					/* out: for-statement node */
+	sym_node_t*	loop_var,	/* in: loop variable */
+	que_node_t*	loop_start_limit,/* in: loop start expression */
+	que_node_t*	loop_end_limit,	/* in: loop end expression */
+	que_node_t*	stat_list);	/* in: statement list */
+/*************************************************************************
+Parses a while-statement. */
+UNIV_INTERN
+while_node_t*
+pars_while_statement(
+/*=================*/
+					/* out: while-statement node */
+	que_node_t*	cond,		/* in: while-condition */
+	que_node_t*	stat_list);	/* in: statement list */
+/*************************************************************************
+Parses an exit statement. */
+UNIV_INTERN
+exit_node_t*
+pars_exit_statement(void);
+/*=====================*/
+					/* out: exit statement node */
+/*************************************************************************
+Parses a return-statement. */
+UNIV_INTERN
+return_node_t*
+pars_return_statement(void);
+/*=======================*/
+					/* out: return-statement node */
+/*************************************************************************
+Parses a procedure call. */
+UNIV_INTERN
+func_node_t*
+pars_procedure_call(
+/*================*/
+				/* out: function node */
+	que_node_t*	res_word,/* in: procedure name reserved word */
+	que_node_t*	args);	/* in: argument list */
+/*************************************************************************
+Parses an assignment statement. */
+UNIV_INTERN
+assign_node_t*
+pars_assignment_statement(
+/*======================*/
+				/* out: assignment statement node */
+	sym_node_t*	var,	/* in: variable to assign */
+	que_node_t*	val);	/* in: value to assign */
+/*************************************************************************
+Parses a fetch statement. into_list or user_func (but not both) must be
+non-NULL. */
+UNIV_INTERN
+fetch_node_t*
+pars_fetch_statement(
+/*=================*/
+					/* out: fetch statement node */
+	sym_node_t*	cursor,		/* in: cursor node */
+	sym_node_t*	into_list,	/* in: variables to set, or NULL */
+	sym_node_t*	user_func);	/* in: user function name, or NULL */
+/*************************************************************************
+Parses an open or close cursor statement. */
+UNIV_INTERN
+open_node_t*
+pars_open_statement(
+/*================*/
+				/* out: fetch statement node */
+	ulint		type,	/* in: ROW_SEL_OPEN_CURSOR
+				or ROW_SEL_CLOSE_CURSOR */
+	sym_node_t*	cursor);	/* in: cursor node */
+/*************************************************************************
+Parses a row_printf-statement. */
+UNIV_INTERN
+row_printf_node_t*
+pars_row_printf_statement(
+/*======================*/
+					/* out: row_printf-statement node */
+	sel_node_t*	sel_node);	/* in: select node */
+/*************************************************************************
+Parses a commit statement. */
+UNIV_INTERN
+commit_node_t*
+pars_commit_statement(void);
+/*=======================*/
+/*************************************************************************
+Parses a rollback statement. */
+UNIV_INTERN
+roll_node_t*
+pars_rollback_statement(void);
+/*=========================*/
+/*************************************************************************
+Parses a column definition at a table creation. */
+UNIV_INTERN
+sym_node_t*
+pars_column_def(
+/*============*/
+						/* out: column sym table
+						node */
+	sym_node_t*		sym_node,	/* in: column node in the
+						symbol table */
+	pars_res_word_t*	type,		/* in: data type */
+	sym_node_t*		len,		/* in: length of column, or
+						NULL */
+	void*			is_unsigned,	/* in: if not NULL, column
+						is of type UNSIGNED. */
+	void*			is_not_null);	/* in: if not NULL, column
+						is of type NOT NULL. */
+/*************************************************************************
+Parses a table creation operation. */
+UNIV_INTERN
+tab_node_t*
+pars_create_table(
+/*==============*/
+					/* out: table create subgraph */
+	sym_node_t*	table_sym,	/* in: table name node in the symbol
+					table */
+	sym_node_t*	column_defs,	/* in: list of column names */
+	void*		not_fit_in_memory);/* in: a non-NULL pointer means that
+					this is a table which in simulations
+					should be simulated as not fitting
+					in memory; thread is put to sleep
+					to simulate disk accesses; NOTE that
+					this flag is not stored to the data
+					dictionary on disk, and the database
+					will forget about non-NULL value if
+					it has to reload the table definition
+					from disk */
+/*************************************************************************
+Parses an index creation operation. */
+UNIV_INTERN
+ind_node_t*
+pars_create_index(
+/*==============*/
+					/* out: index create subgraph */
+	pars_res_word_t* unique_def,	/* in: not NULL if a unique index */
+	pars_res_word_t* clustered_def,	/* in: not NULL if a clustered index */
+	sym_node_t*	index_sym,	/* in: index name node in the symbol
+					table */
+	sym_node_t*	table_sym,	/* in: table name node in the symbol
+					table */
+	sym_node_t*	column_list);	/* in: list of column names */
+/*************************************************************************
+Parses a procedure definition. */
+UNIV_INTERN
+que_fork_t*
+pars_procedure_definition(
+/*======================*/
+					/* out: query fork node */
+	sym_node_t*	sym_node,	/* in: procedure id node in the symbol
+					table */
+	sym_node_t*	param_list,	/* in: parameter declaration list */
+	que_node_t*	stat_list);	/* in: statement list */
+
+/*****************************************************************
+Parses a stored procedure call, when this is not within another stored
+procedure, that is, the client issues a procedure call directly.
+In MySQL/InnoDB, stored InnoDB procedures are invoked via the
+parsed procedure tree, not via InnoDB SQL, so this function is not used. */
+UNIV_INTERN
+que_fork_t*
+pars_stored_procedure_call(
+/*=======================*/
+					/* out: query graph */
+	sym_node_t*	sym_node);	/* in: stored procedure name */
+/**********************************************************************
+Completes a query graph by adding query thread and fork nodes
+above it and prepares the graph for running. The fork created is of
+type QUE_FORK_MYSQL_INTERFACE. */
+UNIV_INTERN
+que_thr_t*
+pars_complete_graph_for_exec(
+/*=========================*/
+				/* out: query thread node to run */
+	que_node_t*	node,	/* in: root node for an incomplete
+				query graph */
+	trx_t*		trx,	/* in: transaction handle */
+	mem_heap_t*	heap);	/* in: memory heap from which allocated */
+
+/********************************************************************
+Create parser info struct.*/
+UNIV_INTERN
+pars_info_t*
+pars_info_create(void);
+/*==================*/
+		/* out, own: info struct */
+
+/********************************************************************
+Free info struct and everything it contains.*/
+UNIV_INTERN
+void
+pars_info_free(
+/*===========*/
+	pars_info_t*	info);	/* in: info struct */
+
+/********************************************************************
+Add bound literal. */
+UNIV_INTERN
+void
+pars_info_add_literal(
+/*==================*/
+	pars_info_t*	info,		/* in: info struct */
+	const char*	name,		/* in: name */
+	const void*	address,	/* in: address */
+	ulint		length,		/* in: length of data */
+	ulint		type,		/* in: type, e.g. DATA_FIXBINARY */
+	ulint		prtype);	/* in: precise type, e.g.
+					DATA_UNSIGNED */
+
+/********************************************************************
+Equivalent to pars_info_add_literal(info, name, str, strlen(str),
+DATA_VARCHAR, DATA_ENGLISH). */
+UNIV_INTERN
+void
+pars_info_add_str_literal(
+/*======================*/
+	pars_info_t*	info,		/* in: info struct */
+	const char*	name,		/* in: name */
+	const char*	str);		/* in: string */
+
+/********************************************************************
+Equivalent to:
+
+char buf[4];
+mach_write_to_4(buf, val);
+pars_info_add_literal(info, name, buf, 4, DATA_INT, 0);
+
+except that the buffer is dynamically allocated from the info struct's
+heap. */
+UNIV_INTERN
+void
+pars_info_add_int4_literal(
+/*=======================*/
+	pars_info_t*	info,		/* in: info struct */
+	const char*	name,		/* in: name */
+	lint		val);		/* in: value */
+
+/********************************************************************
+Equivalent to:
+
+char buf[8];
+mach_write_to_8(buf, val);
+pars_info_add_literal(info, name, buf, 8, DATA_BINARY, 0);
+
+except that the buffer is dynamically allocated from the info struct's
+heap. */
+UNIV_INTERN
+void
+pars_info_add_dulint_literal(
+/*=========================*/
+	pars_info_t*	info,		/* in: info struct */
+	const char*	name,		/* in: name */
+	dulint		val);		/* in: value */
+/********************************************************************
+Add user function. */
+UNIV_INTERN
+void
+pars_info_add_function(
+/*===================*/
+	pars_info_t*		info,	/* in: info struct */
+	const char*		name,	/* in: function name */
+	pars_user_func_cb_t	func,	/* in: function address */
+	void*			arg);	/* in: user-supplied argument */
+
+/********************************************************************
+Add bound id. */
+UNIV_INTERN
+void
+pars_info_add_id(
+/*=============*/
+	pars_info_t*	info,		/* in: info struct */
+	const char*	name,		/* in: name */
+	const char*	id);		/* in: id */
+
+/********************************************************************
+Get user function with the given name.*/
+UNIV_INTERN
+pars_user_func_t*
+pars_info_get_user_func(
+/*====================*/
+					/* out: user func, or NULL if not
+					found */
+	pars_info_t*		info,	/* in: info struct */
+	const char*		name);	/* in: function name to find*/
+
+/********************************************************************
+Get bound literal with the given name.*/
+UNIV_INTERN
+pars_bound_lit_t*
+pars_info_get_bound_lit(
+/*====================*/
+					/* out: bound literal, or NULL if
+					not found */
+	pars_info_t*		info,	/* in: info struct */
+	const char*		name);	/* in: bound literal name to find */
+
+/********************************************************************
+Get bound id with the given name.*/
+UNIV_INTERN
+pars_bound_id_t*
+pars_info_get_bound_id(
+/*===================*/
+					/* out: bound id, or NULL if not
+					found */
+	pars_info_t*		info,	/* in: info struct */
+	const char*		name);	/* in: bound id name to find */
+
+
+/* Extra information supplied for pars_sql(). */
+struct pars_info_struct {
+	mem_heap_t*	heap;		/* our own memory heap */
+
+	ib_vector_t*	funcs;		/* user functions, or NUll
+					(pars_user_func_t*) */
+	ib_vector_t*	bound_lits;	/* bound literals, or NULL
+					(pars_bound_lit_t*) */
+	ib_vector_t*	bound_ids;	/* bound ids, or NULL
+					(pars_bound_id_t*) */
+
+	ibool		graph_owns_us;	/* if TRUE (which is the default),
+					que_graph_free() will free us */
+};
+
+/* User-supplied function and argument. */
+struct pars_user_func_struct {
+	const char*		name;		/* function name */
+	pars_user_func_cb_t	func;		/* function address */
+	void*			arg;		/* user-supplied argument */
+};
+
+/* Bound literal. */
+struct pars_bound_lit_struct {
+	const char*	name;		/* name */
+	const void*	address;	/* address */
+	ulint		length;		/* length of data */
+	ulint		type;		/* type, e.g. DATA_FIXBINARY */
+	ulint		prtype;		/* precise type, e.g. DATA_UNSIGNED */
+};
+
+/* Bound id. */
+struct pars_bound_id_struct {
+	const char*	name;		/* name */
+	const char*	id;		/* id */
+};
+
+/* Struct used to denote a reserved word in a parsing tree */
+struct pars_res_word_struct{
+	int	code;	/* the token code for the reserved word from
+			pars0grm.h */
+};
+
+/* A predefined function or operator node in a parsing tree; this construct
+is also used for some non-functions like the assignment ':=' */
+struct func_node_struct{
+	que_common_t	common;	/* type: QUE_NODE_FUNC */
+	int		func;	/* token code of the function name */
+	ulint		class;	/* class of the function */
+	que_node_t*	args;	/* argument(s) of the function */
+	UT_LIST_NODE_T(func_node_t) cond_list;
+				/* list of comparison conditions; defined
+				only for comparison operator nodes except,
+				presently, for OPT_SCROLL_TYPE ones */
+	UT_LIST_NODE_T(func_node_t) func_node_list;
+				/* list of function nodes in a parsed
+				query graph */
+};
+
+/* An order-by node in a select */
+struct order_node_struct{
+	que_common_t	common;	/* type: QUE_NODE_ORDER */
+	sym_node_t*	column;	/* order-by column */
+	ibool		asc;	/* TRUE if ascending, FALSE if descending */
+};
+
+/* Procedure definition node */
+struct proc_node_struct{
+	que_common_t	common;		/* type: QUE_NODE_PROC */
+	sym_node_t*	proc_id;	/* procedure name symbol in the symbol
+					table of this same procedure */
+	sym_node_t*	param_list;	/* input and output parameters */
+	que_node_t*	stat_list;	/* statement list */
+	sym_tab_t*	sym_tab;	/* symbol table of this procedure */
+};
+
+/* elsif-element node */
+struct elsif_node_struct{
+	que_common_t	common;		/* type: QUE_NODE_ELSIF */
+	que_node_t*	cond;		/* if condition */
+	que_node_t*	stat_list;	/* statement list */
+};
+
+/* if-statement node */
+struct if_node_struct{
+	que_common_t	common;		/* type: QUE_NODE_IF */
+	que_node_t*	cond;		/* if condition */
+	que_node_t*	stat_list;	/* statement list */
+	que_node_t*	else_part;	/* else-part statement list */
+	elsif_node_t*	elsif_list;	/* elsif element list */
+};
+
+/* while-statement node */
+struct while_node_struct{
+	que_common_t	common;		/* type: QUE_NODE_WHILE */
+	que_node_t*	cond;		/* while condition */
+	que_node_t*	stat_list;	/* statement list */
+};
+
+/* for-loop-statement node */
+struct for_node_struct{
+	que_common_t	common;		/* type: QUE_NODE_FOR */
+	sym_node_t*	loop_var;	/* loop variable: this is the
+					dereferenced symbol from the
+					variable declarations, not the
+					symbol occurrence in the for loop
+					definition */
+	que_node_t*	loop_start_limit;/* initial value of loop variable */
+	que_node_t*	loop_end_limit;	/* end value of loop variable */
+	int		loop_end_value;	/* evaluated value for the end value:
+					it is calculated only when the loop
+					is entered, and will not change within
+					the loop */
+	que_node_t*	stat_list;	/* statement list */
+};
+
+/* exit statement node */
+struct exit_node_struct{
+	que_common_t	common;		/* type: QUE_NODE_EXIT */
+};
+
+/* return-statement node */
+struct return_node_struct{
+	que_common_t	common;		/* type: QUE_NODE_RETURN */
+};
+
+/* Assignment statement node */
+struct assign_node_struct{
+	que_common_t	common;		/* type: QUE_NODE_ASSIGNMENT */
+	sym_node_t*	var;		/* variable to set */
+	que_node_t*	val;		/* value to assign */
+};
+
+/* Column assignment node */
+struct col_assign_node_struct{
+	que_common_t	common;		/* type: QUE_NODE_COL_ASSIGN */
+	sym_node_t*	col;		/* column to set */
+	que_node_t*	val;		/* value to assign */
+};
+
+/* Classes of functions */
+#define PARS_FUNC_ARITH		1	/* +, -, *, / */
+#define	PARS_FUNC_LOGICAL	2
+#define PARS_FUNC_CMP		3
+#define	PARS_FUNC_PREDEFINED	4	/* TO_NUMBER, SUBSTR, ... */
+#define	PARS_FUNC_AGGREGATE	5	/* COUNT, DISTINCT, SUM */
+#define	PARS_FUNC_OTHER		6	/* these are not real functions,
+					e.g., := */
+
+#ifndef UNIV_NONINL
+#include "pars0pars.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/pars0pars.ic b/storage/xtradb/include/pars0pars.ic
new file mode 100644
index 00000000000..3a55ad86f48
--- /dev/null
+++ b/storage/xtradb/include/pars0pars.ic
@@ -0,0 +1,23 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+SQL parser
+
+Created 11/19/1996 Heikki Tuuri
+*******************************************************/
diff --git a/storage/xtradb/include/pars0sym.h b/storage/xtradb/include/pars0sym.h
new file mode 100644
index 00000000000..69227a2917e
--- /dev/null
+++ b/storage/xtradb/include/pars0sym.h
@@ -0,0 +1,239 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+SQL parser symbol table
+
+Created 12/15/1997 Heikki Tuuri
+*******************************************************/
+
+#ifndef pars0sym_h
+#define pars0sym_h
+
+#include "univ.i"
+#include "que0types.h"
+#include "usr0types.h"
+#include "dict0types.h"
+#include "pars0types.h"
+#include "row0types.h"
+
+/**********************************************************************
+Creates a symbol table for a single stored procedure or query. */
+UNIV_INTERN
+sym_tab_t*
+sym_tab_create(
+/*===========*/
+				/* out, own: symbol table */
+	mem_heap_t*	heap);	/* in: memory heap where to create */
+/**********************************************************************
+Frees the memory allocated dynamically AFTER parsing phase for variables
+etc. in the symbol table. Does not free the mem heap where the table was
+originally created. Frees also SQL explicit cursor definitions. */
+UNIV_INTERN
+void
+sym_tab_free_private(
+/*=================*/
+	sym_tab_t*	sym_tab);	/* in, own: symbol table */
+/**********************************************************************
+Adds an integer literal to a symbol table. */
+UNIV_INTERN
+sym_node_t*
+sym_tab_add_int_lit(
+/*================*/
+					/* out: symbol table node */
+	sym_tab_t*	sym_tab,	/* in: symbol table */
+	ulint		val);		/* in: integer value */
+/**********************************************************************
+Adds an string literal to a symbol table. */
+UNIV_INTERN
+sym_node_t*
+sym_tab_add_str_lit(
+/*================*/
+					/* out: symbol table node */
+	sym_tab_t*	sym_tab,	/* in: symbol table */
+	byte*		str,		/* in: string with no quotes around
+					it */
+	ulint		len);		/* in: string length */
+/**********************************************************************
+Add a bound literal to a symbol table. */
+UNIV_INTERN
+sym_node_t*
+sym_tab_add_bound_lit(
+/*==================*/
+					/* out: symbol table node */
+	sym_tab_t*	sym_tab,	/* in: symbol table */
+	const char*	name,		/* in: name of bound literal */
+	ulint*		lit_type);	/* out: type of literal (PARS_*_LIT) */
+/**********************************************************************
+Adds an SQL null literal to a symbol table. */
+UNIV_INTERN
+sym_node_t*
+sym_tab_add_null_lit(
+/*=================*/
+					/* out: symbol table node */
+	sym_tab_t*	sym_tab);	/* in: symbol table */
+/**********************************************************************
+Adds an identifier to a symbol table. */
+UNIV_INTERN
+sym_node_t*
+sym_tab_add_id(
+/*===========*/
+					/* out: symbol table node */
+	sym_tab_t*	sym_tab,	/* in: symbol table */
+	byte*		name,		/* in: identifier name */
+	ulint		len);		/* in: identifier length */
+
+/**********************************************************************
+Add a bound identifier to a symbol table. */
+UNIV_INTERN
+sym_node_t*
+sym_tab_add_bound_id(
+/*===========*/
+					/* out: symbol table node */
+	sym_tab_t*	sym_tab,	/* in: symbol table */
+	const char*	name);		/* in: name of bound id */
+
+#define	SYM_CLUST_FIELD_NO	0
+#define	SYM_SEC_FIELD_NO	1
+
+struct sym_node_struct{
+	que_common_t			common;		/* node type:
+							QUE_NODE_SYMBOL */
+	/* NOTE: if the data field in 'common.val' is not NULL and the symbol
+	table node is not for a temporary column, the memory for the value has
+	been allocated from dynamic memory and it should be freed when the
+	symbol table is discarded */
+
+	/* 'alias' and 'indirection' are almost the same, but not quite.
+	'alias' always points to the primary instance of the variable, while
+	'indirection' does the same only if we should use the primary
+	instance's values for the node's data. This is usually the case, but
+	when initializing a cursor (e.g., "DECLARE CURSOR c IS SELECT * FROM
+	t WHERE id = x;"), we copy the values from the primary instance to
+	the cursor's instance so that they are fixed for the duration of the
+	cursor, and set 'indirection' to NULL. If we did not, the value of
+	'x' could change between fetches and things would break horribly.
+
+	TODO: It would be cleaner to make 'indirection' a boolean field and
+	always use 'alias' to refer to the primary node. */
+
+	sym_node_t*			indirection;	/* pointer to
+							another symbol table
+							node which contains
+							the value for this
+							node, NULL otherwise */
+	sym_node_t*			alias;		/* pointer to
+							another symbol table
+							node for which this
+							node is an alias,
+							NULL otherwise */
+	UT_LIST_NODE_T(sym_node_t)	col_var_list;	/* list of table
+							columns or a list of
+							input variables for an
+							explicit cursor */
+	ibool				copy_val;	/* TRUE if a column
+							and its value should
+							be copied to dynamic
+							memory when fetched */
+	ulint				field_nos[2];	/* if a column, in
+							the position
+							SYM_CLUST_FIELD_NO is
+							the field number in the
+							clustered index; in
+							the position
+							SYM_SEC_FIELD_NO
+							the field number in the
+							non-clustered index to
+							use first; if not found
+							from the index, then
+							ULINT_UNDEFINED */
+	ibool				resolved;	/* TRUE if the
+							meaning of a variable
+							or a column has been
+							resolved; for literals
+							this is always TRUE */
+	ulint				token_type;	/* SYM_VAR, SYM_COLUMN,
+							SYM_IMPLICIT_VAR,
+							SYM_LIT, SYM_TABLE,
+							SYM_CURSOR, ... */
+	const char*			name;		/* name of an id */
+	ulint				name_len;	/* id name length */
+	dict_table_t*			table;		/* table definition
+							if a table id or a
+							column id */
+	ulint				col_no;		/* column number if a
+							column */
+	sel_buf_t*			prefetch_buf;	/* NULL, or a buffer
+							for cached column
+							values for prefetched
+							rows */
+	sel_node_t*			cursor_def;	/* cursor definition
+							select node if a
+							named cursor */
+	ulint				param_type;	/* PARS_INPUT,
+							PARS_OUTPUT, or
+							PARS_NOT_PARAM if not a
+							procedure parameter */
+	sym_tab_t*			sym_table;	/* back pointer to
+							the symbol table */
+	UT_LIST_NODE_T(sym_node_t)	sym_list;	/* list of symbol
+							nodes */
+};
+
+struct sym_tab_struct{
+	que_t*			query_graph;
+					/* query graph generated by the
+					parser */
+	const char*		sql_string;
+					/* SQL string to parse */
+	size_t			string_len;
+					/* SQL string length */
+	int			next_char_pos;
+					/* position of the next character in
+					sql_string to give to the lexical
+					analyzer */
+	pars_info_t*		info;	/* extra information, or NULL */
+	sym_node_list_t		sym_list;
+					/* list of symbol nodes in the symbol
+					table */
+	UT_LIST_BASE_NODE_T(func_node_t)
+				func_node_list;
+					/* list of function nodes in the
+					parsed query graph */
+	mem_heap_t*		heap;	/* memory heap from which we can
+					allocate space */
+};
+
+/* Types of a symbol table entry */
+#define	SYM_VAR			91	/* declared parameter or local
+					variable of a procedure */
+#define SYM_IMPLICIT_VAR	92	/* storage for a intermediate result
+					of a calculation */
+#define SYM_LIT			93	/* literal */
+#define SYM_TABLE		94	/* database table name */
+#define SYM_COLUMN		95	/* database table name */
+#define SYM_CURSOR		96	/* named cursor */
+#define SYM_PROCEDURE_NAME	97	/* stored procedure name */
+#define SYM_INDEX		98	/* database index name */
+#define SYM_FUNCTION		99	/* user function name */
+
+#ifndef UNIV_NONINL
+#include "pars0sym.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/pars0sym.ic b/storage/xtradb/include/pars0sym.ic
new file mode 100644
index 00000000000..235d6819ae9
--- /dev/null
+++ b/storage/xtradb/include/pars0sym.ic
@@ -0,0 +1,23 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+SQL parser symbol table
+
+Created 12/15/1997 Heikki Tuuri
+*******************************************************/
diff --git a/storage/xtradb/include/pars0types.h b/storage/xtradb/include/pars0types.h
new file mode 100644
index 00000000000..e0902d0611a
--- /dev/null
+++ b/storage/xtradb/include/pars0types.h
@@ -0,0 +1,49 @@
+/*****************************************************************************
+
+Copyright (c) 1998, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+SQL parser global types
+
+Created 1/11/1998 Heikki Tuuri
+*******************************************************/
+
+#ifndef pars0types_h
+#define pars0types_h
+
+typedef struct pars_info_struct		pars_info_t;
+typedef struct pars_user_func_struct	pars_user_func_t;
+typedef struct pars_bound_lit_struct	pars_bound_lit_t;
+typedef struct pars_bound_id_struct	pars_bound_id_t;
+typedef struct sym_node_struct		sym_node_t;
+typedef struct sym_tab_struct		sym_tab_t;
+typedef struct pars_res_word_struct	pars_res_word_t;
+typedef struct func_node_struct		func_node_t;
+typedef struct order_node_struct	order_node_t;
+typedef struct proc_node_struct		proc_node_t;
+typedef struct elsif_node_struct	elsif_node_t;
+typedef struct if_node_struct		if_node_t;
+typedef struct while_node_struct	while_node_t;
+typedef struct for_node_struct		for_node_t;
+typedef struct exit_node_struct		exit_node_t;
+typedef struct return_node_struct	return_node_t;
+typedef struct assign_node_struct	assign_node_t;
+typedef struct col_assign_node_struct	col_assign_node_t;
+
+typedef UT_LIST_BASE_NODE_T(sym_node_t)	sym_node_list_t;
+
+#endif
diff --git a/storage/xtradb/include/que0que.h b/storage/xtradb/include/que0que.h
new file mode 100644
index 00000000000..a534cb7e464
--- /dev/null
+++ b/storage/xtradb/include/que0que.h
@@ -0,0 +1,526 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Query graph
+
+Created 5/27/1996 Heikki Tuuri
+*******************************************************/
+
+#ifndef que0que_h
+#define que0que_h
+
+#include "univ.i"
+#include "data0data.h"
+#include "dict0types.h"
+#include "trx0trx.h"
+#include "srv0srv.h"
+#include "usr0types.h"
+#include "que0types.h"
+#include "row0types.h"
+#include "pars0types.h"
+
+/* If the following flag is set TRUE, the module will print trace info
+of SQL execution in the UNIV_SQL_DEBUG version */
+extern ibool	que_trace_on;
+
+/***************************************************************************
+Adds a query graph to the session's list of graphs. */
+UNIV_INTERN
+void
+que_graph_publish(
+/*==============*/
+	que_t*	graph,	/* in: graph */
+	sess_t*	sess);	/* in: session */
+/***************************************************************************
+Creates a query graph fork node. */
+UNIV_INTERN
+que_fork_t*
+que_fork_create(
+/*============*/
+					/* out, own: fork node */
+	que_t*		graph,		/* in: graph, if NULL then this
+					fork node is assumed to be the
+					graph root */
+	que_node_t*	parent,		/* in: parent node */
+	ulint		fork_type,	/* in: fork type */
+	mem_heap_t*	heap);		/* in: memory heap where created */
+/***************************************************************************
+Gets the first thr in a fork. */
+UNIV_INLINE
+que_thr_t*
+que_fork_get_first_thr(
+/*===================*/
+	que_fork_t*	fork);	/* in: query fork */
+/***************************************************************************
+Gets the child node of the first thr in a fork. */
+UNIV_INLINE
+que_node_t*
+que_fork_get_child(
+/*===============*/
+	que_fork_t*	fork);	/* in: query fork */
+/***************************************************************************
+Sets the parent of a graph node. */
+UNIV_INLINE
+void
+que_node_set_parent(
+/*================*/
+	que_node_t*	node,	/* in: graph node */
+	que_node_t*	parent);/* in: parent */
+/***************************************************************************
+Creates a query graph thread node. */
+UNIV_INTERN
+que_thr_t*
+que_thr_create(
+/*===========*/
+				/* out, own: query thread node */
+	que_fork_t*	parent,	/* in: parent node, i.e., a fork node */
+	mem_heap_t*	heap);	/* in: memory heap where created */
+/**************************************************************************
+Checks if the query graph is in a state where it should be freed, and
+frees it in that case. If the session is in a state where it should be
+closed, also this is done. */
+UNIV_INTERN
+ibool
+que_graph_try_free(
+/*===============*/
+			/* out: TRUE if freed */
+	que_t*	graph);	/* in: query graph */
+/**************************************************************************
+Frees a query graph, but not the heap where it was created. Does not free
+explicit cursor declarations, they are freed in que_graph_free. */
+UNIV_INTERN
+void
+que_graph_free_recursive(
+/*=====================*/
+	que_node_t*	node);	/* in: query graph node */
+/**************************************************************************
+Frees a query graph. */
+UNIV_INTERN
+void
+que_graph_free(
+/*===========*/
+	que_t*	graph);	/* in: query graph; we assume that the memory
+			heap where this graph was created is private
+			to this graph: if not, then use
+			que_graph_free_recursive and free the heap
+			afterwards! */
+/**************************************************************************
+Stops a query thread if graph or trx is in a state requiring it. The
+conditions are tested in the order (1) graph, (2) trx. The kernel mutex has
+to be reserved. */
+UNIV_INTERN
+ibool
+que_thr_stop(
+/*=========*/
+				/* out: TRUE if stopped */
+	que_thr_t*	thr);	/* in: query thread */
+/**************************************************************************
+Moves a thread from another state to the QUE_THR_RUNNING state. Increments
+the n_active_thrs counters of the query graph and transaction. */
+UNIV_INTERN
+void
+que_thr_move_to_run_state_for_mysql(
+/*================================*/
+	que_thr_t*	thr,	/* in: an query thread */
+	trx_t*		trx);	/* in: transaction */
+/**************************************************************************
+A patch for MySQL used to 'stop' a dummy query thread used in MySQL
+select, when there is no error or lock wait. */
+UNIV_INTERN
+void
+que_thr_stop_for_mysql_no_error(
+/*============================*/
+	que_thr_t*	thr,	/* in: query thread */
+	trx_t*		trx);	/* in: transaction */
+/**************************************************************************
+A patch for MySQL used to 'stop' a dummy query thread used in MySQL. The
+query thread is stopped and made inactive, except in the case where
+it was put to the lock wait state in lock0lock.c, but the lock has already
+been granted or the transaction chosen as a victim in deadlock resolution. */
+UNIV_INTERN
+void
+que_thr_stop_for_mysql(
+/*===================*/
+	que_thr_t*	thr);	/* in: query thread */
+/**************************************************************************
+Run a query thread. Handles lock waits. */
+UNIV_INTERN
+void
+que_run_threads(
+/*============*/
+	que_thr_t*	thr);	/* in: query thread */
+/**************************************************************************
+After signal handling is finished, returns control to a query graph error
+handling routine. (Currently, just returns the control to the root of the
+graph so that the graph can communicate an error message to the client.) */
+UNIV_INTERN
+void
+que_fork_error_handle(
+/*==================*/
+	trx_t*	trx,	/* in: trx */
+	que_t*	fork);	/* in: query graph which was run before signal
+			handling started, NULL not allowed */
+/**************************************************************************
+Moves a suspended query thread to the QUE_THR_RUNNING state and releases
+a single worker thread to execute it. This function should be used to end
+the wait state of a query thread waiting for a lock or a stored procedure
+completion. */
+UNIV_INTERN
+void
+que_thr_end_wait(
+/*=============*/
+	que_thr_t*	thr,		/* in: query thread in the
+					QUE_THR_LOCK_WAIT,
+					or QUE_THR_PROCEDURE_WAIT, or
+					QUE_THR_SIG_REPLY_WAIT state */
+	que_thr_t**	next_thr);	/* in/out: next query thread to run;
+					if the value which is passed in is
+					a pointer to a NULL pointer, then the
+					calling function can start running
+					a new query thread */
+/**************************************************************************
+Same as que_thr_end_wait, but no parameter next_thr available. */
+UNIV_INTERN
+void
+que_thr_end_wait_no_next_thr(
+/*=========================*/
+	que_thr_t*	thr);		/* in: query thread in the
+					QUE_THR_LOCK_WAIT,
+					or QUE_THR_PROCEDURE_WAIT, or
+					QUE_THR_SIG_REPLY_WAIT state */
+/**************************************************************************
+Starts execution of a command in a query fork. Picks a query thread which
+is not in the QUE_THR_RUNNING state and moves it to that state. If none
+can be chosen, a situation which may arise in parallelized fetches, NULL
+is returned. */
+UNIV_INTERN
+que_thr_t*
+que_fork_start_command(
+/*===================*/
+				/* out: a query thread of the graph moved to
+				QUE_THR_RUNNING state, or NULL; the query
+				thread should be executed by que_run_threads
+				by the caller */
+	que_fork_t*	fork);	/* in: a query fork */
+/***************************************************************************
+Gets the trx of a query thread. */
+UNIV_INLINE
+trx_t*
+thr_get_trx(
+/*========*/
+	que_thr_t*	thr);	/* in: query thread */
+/***************************************************************************
+Gets the type of a graph node. */
+UNIV_INLINE
+ulint
+que_node_get_type(
+/*==============*/
+	que_node_t*	node);	/* in: graph node */
+/***************************************************************************
+Gets pointer to the value data type field of a graph node. */
+UNIV_INLINE
+dtype_t*
+que_node_get_data_type(
+/*===================*/
+	que_node_t*	node);	/* in: graph node */
+/***************************************************************************
+Gets pointer to the value dfield of a graph node. */
+UNIV_INLINE
+dfield_t*
+que_node_get_val(
+/*=============*/
+	que_node_t*	node);	/* in: graph node */
+/***************************************************************************
+Gets the value buffer size of a graph node. */
+UNIV_INLINE
+ulint
+que_node_get_val_buf_size(
+/*======================*/
+				/* out: val buffer size, not defined if
+				val.data == NULL in node */
+	que_node_t*	node);	/* in: graph node */
+/***************************************************************************
+Sets the value buffer size of a graph node. */
+UNIV_INLINE
+void
+que_node_set_val_buf_size(
+/*======================*/
+	que_node_t*	node,	/* in: graph node */
+	ulint		size);	/* in: size */
+/*************************************************************************
+Gets the next list node in a list of query graph nodes. */
+UNIV_INLINE
+que_node_t*
+que_node_get_next(
+/*==============*/
+	que_node_t*	node);	/* in: node in a list */
+/*************************************************************************
+Gets the parent node of a query graph node. */
+UNIV_INLINE
+que_node_t*
+que_node_get_parent(
+/*================*/
+				/* out: parent node or NULL */
+	que_node_t*	node);	/* in: node */
+/********************************************************************
+Get the first containing loop node (e.g. while_node_t or for_node_t) for the
+given node, or NULL if the node is not within a loop. */
+UNIV_INTERN
+que_node_t*
+que_node_get_containing_loop_node(
+/*==============================*/
+				/* out: containing loop node, or NULL. */
+	que_node_t*	node);	/* in: node */
+/*************************************************************************
+Catenates a query graph node to a list of them, possible empty list. */
+UNIV_INLINE
+que_node_t*
+que_node_list_add_last(
+/*===================*/
+					/* out: one-way list of nodes */
+	que_node_t*	node_list,	/* in: node list, or NULL */
+	que_node_t*	node);		/* in: node */
+/*************************************************************************
+Gets a query graph node list length. */
+UNIV_INLINE
+ulint
+que_node_list_get_len(
+/*==================*/
+					/* out: length, for NULL list 0 */
+	que_node_t*	node_list);	/* in: node list, or NULL */
+/**************************************************************************
+Checks if graph, trx, or session is in a state where the query thread should
+be stopped. */
+UNIV_INLINE
+ibool
+que_thr_peek_stop(
+/*==============*/
+				/* out: TRUE if should be stopped; NOTE that
+				if the peek is made without reserving the
+				kernel mutex, then another peek with the
+				mutex reserved is necessary before deciding
+				the actual stopping */
+	que_thr_t*	thr);	/* in: query thread */
+/***************************************************************************
+Returns TRUE if the query graph is for a SELECT statement. */
+UNIV_INLINE
+ibool
+que_graph_is_select(
+/*================*/
+					/* out: TRUE if a select */
+	que_t*		graph);		/* in: graph */
+/**************************************************************************
+Prints info of an SQL query graph node. */
+UNIV_INTERN
+void
+que_node_print_info(
+/*================*/
+	que_node_t*	node);	/* in: query graph node */
+/*************************************************************************
+Evaluate the given SQL */
+UNIV_INTERN
+ulint
+que_eval_sql(
+/*=========*/
+				/* out: error code or DB_SUCCESS */
+	pars_info_t*	info,	/* in: info struct, or NULL */
+	const char*	sql,	/* in: SQL string */
+	ibool		reserve_dict_mutex,
+				/* in: if TRUE, acquire/release
+				dict_sys->mutex around call to pars_sql. */
+	trx_t*		trx);	/* in: trx */
+
+/* Query graph query thread node: the fields are protected by the kernel
+mutex with the exceptions named below */
+
+struct que_thr_struct{
+	que_common_t	common;		/* type: QUE_NODE_THR */
+	ulint		magic_n;	/* magic number to catch memory
+					corruption */
+	que_node_t*	child;		/* graph child node */
+	que_t*		graph;		/* graph where this node belongs */
+	ibool		is_active;	/* TRUE if the thread has been set
+					to the run state in
+					que_thr_move_to_run_state, but not
+					deactivated in
+					que_thr_dec_reference_count */
+	ulint		state;		/* state of the query thread */
+	UT_LIST_NODE_T(que_thr_t)
+			thrs;		/* list of thread nodes of the fork
+					node */
+	UT_LIST_NODE_T(que_thr_t)
+			trx_thrs;	/* lists of threads in wait list of
+					the trx */
+	UT_LIST_NODE_T(que_thr_t)
+			queue;		/* list of runnable thread nodes in
+					the server task queue */
+	/*------------------------------*/
+	/* The following fields are private to the OS thread executing the
+	query thread, and are not protected by the kernel mutex: */
+
+	que_node_t*	run_node;	/* pointer to the node where the
+					subgraph down from this node is
+					currently executed */
+	que_node_t*	prev_node;	/* pointer to the node from which
+					the control came */
+	ulint		resource;	/* resource usage of the query thread
+					thus far */
+	ulint		lock_state;	/* lock state of thread (table or
+					row) */
+};
+
+#define QUE_THR_MAGIC_N		8476583
+#define QUE_THR_MAGIC_FREED	123461526
+
+/* Query graph fork node: its fields are protected by the kernel mutex */
+struct que_fork_struct{
+	que_common_t	common;		/* type: QUE_NODE_FORK */
+	que_t*		graph;		/* query graph of this node */
+	ulint		fork_type;	/* fork type */
+	ulint		n_active_thrs;	/* if this is the root of a graph, the
+					number query threads that have been
+					started in que_thr_move_to_run_state
+					but for which que_thr_dec_refer_count
+					has not yet been called */
+	trx_t*		trx;		/* transaction: this is set only in
+					the root node */
+	ulint		state;		/* state of the fork node */
+	que_thr_t*	caller;		/* pointer to a possible calling query
+					thread */
+	UT_LIST_BASE_NODE_T(que_thr_t)
+			thrs;		/* list of query threads */
+	/*------------------------------*/
+	/* The fields in this section are defined only in the root node */
+	sym_tab_t*	sym_tab;	/* symbol table of the query,
+					generated by the parser, or NULL
+					if the graph was created 'by hand' */
+	pars_info_t*	info;		/* in: info struct, or NULL */
+	/* The following cur_... fields are relevant only in a select graph */
+
+	ulint		cur_end;	/* QUE_CUR_NOT_DEFINED, QUE_CUR_START,
+					QUE_CUR_END */
+	ulint		cur_pos;	/* if there are n rows in the result
+					set, values 0 and n + 1 mean before
+					first row, or after last row, depending
+					on cur_end; values 1...n mean a row
+					index */
+	ibool		cur_on_row;	/* TRUE if cursor is on a row, i.e.,
+					it is not before the first row or
+					after the last row */
+	dulint		n_inserts;	/* number of rows inserted */
+	dulint		n_updates;	/* number of rows updated */
+	dulint		n_deletes;	/* number of rows deleted */
+	sel_node_t*	last_sel_node;	/* last executed select node, or NULL
+					if none */
+	UT_LIST_NODE_T(que_fork_t)
+			graphs;		/* list of query graphs of a session
+					or a stored procedure */
+	/*------------------------------*/
+	mem_heap_t*	heap;		/* memory heap where the fork was
+					created */
+
+};
+
+/* Query fork (or graph) types */
+#define QUE_FORK_SELECT_NON_SCROLL	1	/* forward-only cursor */
+#define QUE_FORK_SELECT_SCROLL		2	/* scrollable cursor */
+#define QUE_FORK_INSERT			3
+#define QUE_FORK_UPDATE			4
+#define QUE_FORK_ROLLBACK		5
+			/* This is really the undo graph used in rollback,
+			no signal-sending roll_node in this graph */
+#define QUE_FORK_PURGE			6
+#define	QUE_FORK_EXECUTE		7
+#define QUE_FORK_PROCEDURE		8
+#define QUE_FORK_PROCEDURE_CALL		9
+#define QUE_FORK_MYSQL_INTERFACE	10
+#define	QUE_FORK_RECOVERY		11
+
+/* Query fork (or graph) states */
+#define QUE_FORK_ACTIVE		1
+#define QUE_FORK_COMMAND_WAIT	2
+#define QUE_FORK_INVALID	3
+#define QUE_FORK_BEING_FREED	4
+
+/* Flag which is ORed to control structure statement node types */
+#define QUE_NODE_CONTROL_STAT	1024
+
+/* Query graph node types */
+#define	QUE_NODE_LOCK		1
+#define	QUE_NODE_INSERT		2
+#define QUE_NODE_UPDATE		4
+#define	QUE_NODE_CURSOR		5
+#define	QUE_NODE_SELECT		6
+#define	QUE_NODE_AGGREGATE	7
+#define QUE_NODE_FORK		8
+#define QUE_NODE_THR		9
+#define QUE_NODE_UNDO		10
+#define QUE_NODE_COMMIT		11
+#define QUE_NODE_ROLLBACK	12
+#define QUE_NODE_PURGE		13
+#define QUE_NODE_CREATE_TABLE	14
+#define QUE_NODE_CREATE_INDEX	15
+#define QUE_NODE_SYMBOL		16
+#define QUE_NODE_RES_WORD	17
+#define QUE_NODE_FUNC		18
+#define QUE_NODE_ORDER		19
+#define QUE_NODE_PROC		(20 + QUE_NODE_CONTROL_STAT)
+#define QUE_NODE_IF		(21 + QUE_NODE_CONTROL_STAT)
+#define QUE_NODE_WHILE		(22 + QUE_NODE_CONTROL_STAT)
+#define QUE_NODE_ASSIGNMENT	23
+#define QUE_NODE_FETCH		24
+#define QUE_NODE_OPEN		25
+#define QUE_NODE_COL_ASSIGNMENT	26
+#define QUE_NODE_FOR		(27 + QUE_NODE_CONTROL_STAT)
+#define QUE_NODE_RETURN		28
+#define QUE_NODE_ROW_PRINTF	29
+#define QUE_NODE_ELSIF		30
+#define QUE_NODE_CALL		31
+#define QUE_NODE_EXIT		32
+
+/* Query thread states */
+#define QUE_THR_RUNNING		1
+#define QUE_THR_PROCEDURE_WAIT	2
+#define	QUE_THR_COMPLETED	3	/* in selects this means that the
+					thread is at the end of its result set
+					(or start, in case of a scroll cursor);
+					in other statements, this means the
+					thread has done its task */
+#define QUE_THR_COMMAND_WAIT	4
+#define QUE_THR_LOCK_WAIT	5
+#define QUE_THR_SIG_REPLY_WAIT	6
+#define QUE_THR_SUSPENDED	7
+#define QUE_THR_ERROR		8
+
+/* Query thread lock states */
+#define QUE_THR_LOCK_NOLOCK	0
+#define QUE_THR_LOCK_ROW	1
+#define QUE_THR_LOCK_TABLE	2
+
+/* From where the cursor position is counted */
+#define QUE_CUR_NOT_DEFINED	1
+#define QUE_CUR_START		2
+#define	QUE_CUR_END		3
+
+
+#ifndef UNIV_NONINL
+#include "que0que.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/que0que.ic b/storage/xtradb/include/que0que.ic
new file mode 100644
index 00000000000..e9a6b00b9ab
--- /dev/null
+++ b/storage/xtradb/include/que0que.ic
@@ -0,0 +1,275 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Query graph
+
+Created 5/27/1996 Heikki Tuuri
+*******************************************************/
+
+#include "usr0sess.h"
+
+/***************************************************************************
+Gets the trx of a query thread. */
+UNIV_INLINE
+trx_t*
+thr_get_trx(
+/*========*/
+	que_thr_t*	thr)	/* in: query thread */
+{
+	ut_ad(thr);
+
+	return(thr->graph->trx);
+}
+
+/***************************************************************************
+Gets the first thr in a fork. */
+UNIV_INLINE
+que_thr_t*
+que_fork_get_first_thr(
+/*===================*/
+	que_fork_t*	fork)	/* in: query fork */
+{
+	return(UT_LIST_GET_FIRST(fork->thrs));
+}
+
+/***************************************************************************
+Gets the child node of the first thr in a fork. */
+UNIV_INLINE
+que_node_t*
+que_fork_get_child(
+/*===============*/
+	que_fork_t*	fork)	/* in: query fork */
+{
+	que_thr_t*	thr;
+
+	thr = UT_LIST_GET_FIRST(fork->thrs);
+
+	return(thr->child);
+}
+
+/***************************************************************************
+Gets the type of a graph node. */
+UNIV_INLINE
+ulint
+que_node_get_type(
+/*==============*/
+	que_node_t*	node)	/* in: graph node */
+{
+	ut_ad(node);
+
+	return(((que_common_t*)node)->type);
+}
+
+/***************************************************************************
+Gets pointer to the value dfield of a graph node. */
+UNIV_INLINE
+dfield_t*
+que_node_get_val(
+/*=============*/
+	que_node_t*	node)	/* in: graph node */
+{
+	ut_ad(node);
+
+	return(&(((que_common_t*)node)->val));
+}
+
+/***************************************************************************
+Gets the value buffer size of a graph node. */
+UNIV_INLINE
+ulint
+que_node_get_val_buf_size(
+/*======================*/
+				/* out: val buffer size, not defined if
+				val.data == NULL in node */
+	que_node_t*	node)	/* in: graph node */
+{
+	ut_ad(node);
+
+	return(((que_common_t*)node)->val_buf_size);
+}
+
+/***************************************************************************
+Sets the value buffer size of a graph node. */
+UNIV_INLINE
+void
+que_node_set_val_buf_size(
+/*======================*/
+	que_node_t*	node,	/* in: graph node */
+	ulint		size)	/* in: size */
+{
+	ut_ad(node);
+
+	((que_common_t*)node)->val_buf_size = size;
+}
+
+/***************************************************************************
+Sets the parent of a graph node. */
+UNIV_INLINE
+void
+que_node_set_parent(
+/*================*/
+	que_node_t*	node,	/* in: graph node */
+	que_node_t*	parent)	/* in: parent */
+{
+	ut_ad(node);
+
+	((que_common_t*)node)->parent = parent;
+}
+
+/***************************************************************************
+Gets pointer to the value data type field of a graph node. */
+UNIV_INLINE
+dtype_t*
+que_node_get_data_type(
+/*===================*/
+	que_node_t*	node)	/* in: graph node */
+{
+	ut_ad(node);
+
+	return(dfield_get_type(&((que_common_t*) node)->val));
+}
+
+/*************************************************************************
+Catenates a query graph node to a list of them, possible empty list. */
+UNIV_INLINE
+que_node_t*
+que_node_list_add_last(
+/*===================*/
+					/* out: one-way list of nodes */
+	que_node_t*	node_list,	/* in: node list, or NULL */
+	que_node_t*	node)		/* in: node */
+{
+	que_common_t*	cnode;
+	que_common_t*	cnode2;
+
+	cnode = (que_common_t*) node;
+
+	cnode->brother = NULL;
+
+	if (node_list == NULL) {
+
+		return(node);
+	}
+
+	cnode2 = (que_common_t*) node_list;
+
+	while (cnode2->brother != NULL) {
+		cnode2 = (que_common_t*) cnode2->brother;
+	}
+
+	cnode2->brother = node;
+
+	return(node_list);
+}
+
+/*************************************************************************
+Gets the next list node in a list of query graph nodes. */
+UNIV_INLINE
+que_node_t*
+que_node_get_next(
+/*==============*/
+				/* out: next node in a list of nodes */
+	que_node_t*	node)	/* in: node in a list */
+{
+	return(((que_common_t*)node)->brother);
+}
+
+/*************************************************************************
+Gets a query graph node list length. */
+UNIV_INLINE
+ulint
+que_node_list_get_len(
+/*==================*/
+					/* out: length, for NULL list 0 */
+	que_node_t*	node_list)	/* in: node list, or NULL */
+{
+	const que_common_t*	cnode;
+	ulint			len;
+
+	cnode = (const que_common_t*) node_list;
+	len = 0;
+
+	while (cnode != NULL) {
+		len++;
+		cnode = (const que_common_t*) cnode->brother;
+	}
+
+	return(len);
+}
+
+/*************************************************************************
+Gets the parent node of a query graph node. */
+UNIV_INLINE
+que_node_t*
+que_node_get_parent(
+/*================*/
+				/* out: parent node or NULL */
+	que_node_t*	node)	/* in: node */
+{
+	return(((que_common_t*)node)->parent);
+}
+
+/**************************************************************************
+Checks if graph, trx, or session is in a state where the query thread should
+be stopped. */
+UNIV_INLINE
+ibool
+que_thr_peek_stop(
+/*==============*/
+				/* out: TRUE if should be stopped; NOTE that
+				if the peek is made without reserving the
+				kernel mutex, then another peek with the
+				mutex reserved is necessary before deciding
+				the actual stopping */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	trx_t*	trx;
+	que_t*	graph;
+
+	graph = thr->graph;
+	trx = graph->trx;
+
+	if (graph->state != QUE_FORK_ACTIVE
+	    || trx->que_state == TRX_QUE_LOCK_WAIT
+	    || (UT_LIST_GET_LEN(trx->signals) > 0
+		&& trx->que_state == TRX_QUE_RUNNING)) {
+
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
+
+/***************************************************************************
+Returns TRUE if the query graph is for a SELECT statement. */
+UNIV_INLINE
+ibool
+que_graph_is_select(
+/*================*/
+					/* out: TRUE if a select */
+	que_t*		graph)		/* in: graph */
+{
+	if (graph->fork_type == QUE_FORK_SELECT_SCROLL
+	    || graph->fork_type == QUE_FORK_SELECT_NON_SCROLL) {
+
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
diff --git a/storage/xtradb/include/que0types.h b/storage/xtradb/include/que0types.h
new file mode 100644
index 00000000000..1d3217fb491
--- /dev/null
+++ b/storage/xtradb/include/que0types.h
@@ -0,0 +1,59 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Query graph global types
+
+Created 5/27/1996 Heikki Tuuri
+*******************************************************/
+
+#ifndef que0types_h
+#define que0types_h
+
+#include "data0data.h"
+#include "dict0types.h"
+
+/* Pseudotype for all graph nodes */
+typedef void	que_node_t;
+
+typedef struct que_fork_struct	que_fork_t;
+
+/* Query graph root is a fork node */
+typedef	que_fork_t	que_t;
+
+typedef struct que_thr_struct		que_thr_t;
+typedef struct que_common_struct	que_common_t;
+
+/* Common struct at the beginning of each query graph node; the name of this
+substruct must be 'common' */
+
+struct que_common_struct{
+	ulint		type;	/* query node type */
+	que_node_t*	parent;	/* back pointer to parent node, or NULL */
+	que_node_t*	brother;/* pointer to a possible brother node */
+	dfield_t	val;	/* evaluated value for an expression */
+	ulint		val_buf_size;
+				/* buffer size for the evaluated value data,
+				if the buffer has been allocated dynamically:
+				if this field is != 0, and the node is a
+				symbol node or a function node, then we
+				have to free the data field in val
+				explicitly */
+};
+
+#endif
diff --git a/storage/xtradb/include/read0read.h b/storage/xtradb/include/read0read.h
new file mode 100644
index 00000000000..7ea8bdaf8dd
--- /dev/null
+++ b/storage/xtradb/include/read0read.h
@@ -0,0 +1,181 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Cursor read
+
+Created 2/16/1997 Heikki Tuuri
+*******************************************************/
+
+#ifndef read0read_h
+#define read0read_h
+
+#include "univ.i"
+
+
+#include "ut0byte.h"
+#include "ut0lst.h"
+#include "trx0trx.h"
+#include "read0types.h"
+
+/*************************************************************************
+Opens a read view where exactly the transactions serialized before this
+point in time are seen in the view. */
+UNIV_INTERN
+read_view_t*
+read_view_open_now(
+/*===============*/
+					/* out, own: read view struct */
+	dulint		cr_trx_id,	/* in: trx_id of creating
+					transaction, or (0, 0) used in
+					purge */
+	mem_heap_t*	heap);		/* in: memory heap from which
+					allocated */
+/*************************************************************************
+Makes a copy of the oldest existing read view, or opens a new. The view
+must be closed with ..._close. */
+UNIV_INTERN
+read_view_t*
+read_view_oldest_copy_or_open_new(
+/*==============================*/
+					/* out, own: read view struct */
+	dulint		cr_trx_id,	/* in: trx_id of creating
+					transaction, or (0, 0) used in
+					purge */
+	mem_heap_t*	heap);		/* in: memory heap from which
+					allocated */
+/*************************************************************************
+Closes a read view. */
+UNIV_INTERN
+void
+read_view_close(
+/*============*/
+	read_view_t*	view);	/* in: read view */
+/*************************************************************************
+Closes a consistent read view for MySQL. This function is called at an SQL
+statement end if the trx isolation level is <= TRX_ISO_READ_COMMITTED. */
+UNIV_INTERN
+void
+read_view_close_for_mysql(
+/*======================*/
+	trx_t*	trx);	/* in: trx which has a read view */
+/*************************************************************************
+Checks if a read view sees the specified transaction. */
+UNIV_INLINE
+ibool
+read_view_sees_trx_id(
+/*==================*/
+				/* out: TRUE if sees */
+	read_view_t*	view,	/* in: read view */
+	dulint		trx_id);/* in: trx id */
+/*************************************************************************
+Prints a read view to stderr. */
+UNIV_INTERN
+void
+read_view_print(
+/*============*/
+	read_view_t*	view);	/* in: read view */
+/*************************************************************************
+Create a consistent cursor view for mysql to be used in cursors. In this
+consistent read view modifications done by the creating transaction or future
+transactions are not visible. */
+UNIV_INTERN
+cursor_view_t*
+read_cursor_view_create_for_mysql(
+/*==============================*/
+	trx_t*		cr_trx);/* in: trx where cursor view is created */
+/*************************************************************************
+Close a given consistent cursor view for mysql and restore global read view
+back to a transaction read view. */
+UNIV_INTERN
+void
+read_cursor_view_close_for_mysql(
+/*=============================*/
+	trx_t*		trx,		/* in: trx */
+	cursor_view_t*	curview);	/* in: cursor view to be closed */
+/*************************************************************************
+This function sets a given consistent cursor view to a transaction
+read view if given consistent cursor view is not NULL. Otherwise, function
+restores a global read view to a transaction read view. */
+UNIV_INTERN
+void
+read_cursor_set_for_mysql(
+/*======================*/
+	trx_t*		trx,	/* in: transaction where cursor is set */
+	cursor_view_t*	curview);/* in: consistent cursor view to be set */
+
+/* Read view lists the trx ids of those transactions for which a consistent
+read should not see the modifications to the database. */
+
+struct read_view_struct{
+	ulint	type;		/* VIEW_NORMAL, VIEW_HIGH_GRANULARITY */
+	dulint	undo_no;	/* (0, 0) or if type is VIEW_HIGH_GRANULARITY
+				transaction undo_no when this high-granularity
+				consistent read view was created */
+	dulint	low_limit_no;	/* The view does not need to see the undo
+				logs for transactions whose transaction number
+				is strictly smaller (<) than this value: they
+				can be removed in purge if not needed by other
+				views */
+	dulint	low_limit_id;	/* The read should not see any transaction
+				with trx id >= this value */
+	dulint	up_limit_id;	/* The read should see all trx ids which
+				are strictly smaller (<) than this value */
+	ulint	n_trx_ids;	/* Number of cells in the trx_ids array */
+	dulint*	trx_ids;	/* Additional trx ids which the read should
+				not see: typically, these are the active
+				transactions at the time when the read is
+				serialized, except the reading transaction
+				itself; the trx ids in this array are in a
+				descending order */
+	dulint	creator_trx_id;	/* trx id of creating transaction, or
+				(0, 0) used in purge */
+	UT_LIST_NODE_T(read_view_t) view_list;
+				/* List of read views in trx_sys */
+};
+
+/* Read view types */
+#define VIEW_NORMAL		1	/* Normal consistent read view
+					where transaction does not see changes
+					made by active transactions except
+					creating transaction. */
+#define VIEW_HIGH_GRANULARITY	2	/* High-granularity read view where
+					transaction does not see changes
+					made by active transactions and own
+					changes after a point in time when this
+					read view was created. */
+
+/* Implement InnoDB framework to support consistent read views in
+cursors. This struct holds both heap where consistent read view
+is allocated and pointer to a read view. */
+
+struct cursor_view_struct{
+	mem_heap_t*	heap;
+				/* Memory heap for the cursor view */
+	read_view_t*	read_view;
+				/* Consistent read view of the cursor*/
+	ulint		n_mysql_tables_in_use;
+				/* number of Innobase tables used in the
+				processing of this cursor */
+};
+
+#ifndef UNIV_NONINL
+#include "read0read.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/read0read.ic b/storage/xtradb/include/read0read.ic
new file mode 100644
index 00000000000..9fc6af04e88
--- /dev/null
+++ b/storage/xtradb/include/read0read.ic
@@ -0,0 +1,97 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Cursor read
+
+Created 2/16/1997 Heikki Tuuri
+*******************************************************/
+
+/*************************************************************************
+Gets the nth trx id in a read view. */
+UNIV_INLINE
+dulint
+read_view_get_nth_trx_id(
+/*=====================*/
+				/* out: trx id */
+	read_view_t*	view,	/* in: read view */
+	ulint		n)	/* in: position */
+{
+	ut_ad(n < view->n_trx_ids);
+
+	return(*(view->trx_ids + n));
+}
+
+/*************************************************************************
+Sets the nth trx id in a read view. */
+UNIV_INLINE
+void
+read_view_set_nth_trx_id(
+/*=====================*/
+	read_view_t*	view,	/* in: read view */
+	ulint		n,	/* in: position */
+	dulint		trx_id)	/* in: trx id to set */
+{
+	ut_ad(n < view->n_trx_ids);
+
+	*(view->trx_ids + n) = trx_id;
+}
+
+/*************************************************************************
+Checks if a read view sees the specified transaction. */
+UNIV_INLINE
+ibool
+read_view_sees_trx_id(
+/*==================*/
+				/* out: TRUE if sees */
+	read_view_t*	view,	/* in: read view */
+	dulint		trx_id)	/* in: trx id */
+{
+	ulint	n_ids;
+	int	cmp;
+	ulint	i;
+
+	if (ut_dulint_cmp(trx_id, view->up_limit_id) < 0) {
+
+		return(TRUE);
+	}
+
+	if (ut_dulint_cmp(trx_id, view->low_limit_id) >= 0) {
+
+		return(FALSE);
+	}
+
+	/* We go through the trx ids in the array smallest first: this order
+	may save CPU time, because if there was a very long running
+	transaction in the trx id array, its trx id is looked at first, and
+	the first two comparisons may well decide the visibility of trx_id. */
+
+	n_ids = view->n_trx_ids;
+
+	for (i = 0; i < n_ids; i++) {
+
+		cmp = ut_dulint_cmp(
+			trx_id,
+			read_view_get_nth_trx_id(view, n_ids - i - 1));
+		if (cmp <= 0) {
+			return(cmp < 0);
+		}
+	}
+
+	return(TRUE);
+}
diff --git a/storage/xtradb/include/read0types.h b/storage/xtradb/include/read0types.h
new file mode 100644
index 00000000000..44849cbb498
--- /dev/null
+++ b/storage/xtradb/include/read0types.h
@@ -0,0 +1,31 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Cursor read
+
+Created 2/16/1997 Heikki Tuuri
+*******************************************************/
+
+#ifndef read0types_h
+#define read0types_h
+
+typedef struct read_view_struct	read_view_t;
+typedef struct cursor_view_struct	cursor_view_t;
+
+#endif
diff --git a/storage/xtradb/include/rem0cmp.h b/storage/xtradb/include/rem0cmp.h
new file mode 100644
index 00000000000..f32bae73a13
--- /dev/null
+++ b/storage/xtradb/include/rem0cmp.h
@@ -0,0 +1,205 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/***********************************************************************
+Comparison services for records
+
+Created 7/1/1994 Heikki Tuuri
+************************************************************************/
+
+#ifndef rem0cmp_h
+#define rem0cmp_h
+
+#include "univ.i"
+#include "data0data.h"
+#include "data0type.h"
+#include "dict0dict.h"
+#include "rem0rec.h"
+
+/*****************************************************************
+Returns TRUE if two columns are equal for comparison purposes. */
+UNIV_INTERN
+ibool
+cmp_cols_are_equal(
+/*===============*/
+					/* out: TRUE if the columns are
+					considered equal in comparisons */
+	const dict_col_t*	col1,	/* in: column 1 */
+	const dict_col_t*	col2,	/* in: column 2 */
+	ibool			check_charsets);
+					/* in: whether to check charsets */
+/*****************************************************************
+This function is used to compare two data fields for which we know the
+data type. */
+UNIV_INLINE
+int
+cmp_data_data(
+/*==========*/
+				/* out: 1, 0, -1, if data1 is greater, equal,
+				less than data2, respectively */
+	ulint		mtype,	/* in: main type */
+	ulint		prtype,	/* in: precise type */
+	const byte*	data1,	/* in: data field (== a pointer to a memory
+				buffer) */
+	ulint		len1,	/* in: data field length or UNIV_SQL_NULL */
+	const byte*	data2,	/* in: data field (== a pointer to a memory
+				buffer) */
+	ulint		len2);	/* in: data field length or UNIV_SQL_NULL */
+/*****************************************************************
+This function is used to compare two data fields for which we know the
+data type. */
+UNIV_INTERN
+int
+cmp_data_data_slow(
+/*===============*/
+				/* out: 1, 0, -1, if data1 is greater, equal,
+				less than data2, respectively */
+	ulint		mtype,	/* in: main type */
+	ulint		prtype,	/* in: precise type */
+	const byte*	data1,	/* in: data field (== a pointer to a memory
+				buffer) */
+	ulint		len1,	/* in: data field length or UNIV_SQL_NULL */
+	const byte*	data2,	/* in: data field (== a pointer to a memory
+				buffer) */
+	ulint		len2);	/* in: data field length or UNIV_SQL_NULL */
+/*****************************************************************
+This function is used to compare two dfields where at least the first
+has its data type field set. */
+UNIV_INLINE
+int
+cmp_dfield_dfield(
+/*==============*/
+				/* out: 1, 0, -1, if dfield1 is greater, equal,
+				less than dfield2, respectively */
+	const dfield_t*	dfield1,/* in: data field; must have type field set */
+	const dfield_t*	dfield2);/* in: data field */
+/*****************************************************************
+This function is used to compare a data tuple to a physical record.
+Only dtuple->n_fields_cmp first fields are taken into account for
+the the data tuple! If we denote by n = n_fields_cmp, then rec must
+have either m >= n fields, or it must differ from dtuple in some of
+the m fields rec has. If rec has an externally stored field we do not
+compare it but return with value 0 if such a comparison should be
+made. */
+UNIV_INTERN
+int
+cmp_dtuple_rec_with_match(
+/*======================*/
+				/* out: 1, 0, -1, if dtuple is greater, equal,
+				less than rec, respectively, when only the
+				common first fields are compared, or
+				until the first externally stored field in
+				rec */
+	const dtuple_t*	dtuple,	/* in: data tuple */
+	const rec_t*	rec,	/* in: physical record which differs from
+				dtuple in some of the common fields, or which
+				has an equal number or more fields than
+				dtuple */
+	const ulint*	offsets,/* in: array returned by rec_get_offsets() */
+	ulint*		matched_fields, /* in/out: number of already completely
+				matched fields; when function returns,
+				contains the value for current comparison */
+	ulint*		matched_bytes); /* in/out: number of already matched
+				bytes within the first field not completely
+				matched; when function returns, contains the
+				value for current comparison */
+/******************************************************************
+Compares a data tuple to a physical record. */
+UNIV_INTERN
+int
+cmp_dtuple_rec(
+/*===========*/
+				/* out: 1, 0, -1, if dtuple is greater, equal,
+				less than rec, respectively; see the comments
+				for cmp_dtuple_rec_with_match */
+	const dtuple_t*	dtuple,	/* in: data tuple */
+	const rec_t*	rec,	/* in: physical record */
+	const ulint*	offsets);/* in: array returned by rec_get_offsets() */
+/******************************************************************
+Checks if a dtuple is a prefix of a record. The last field in dtuple
+is allowed to be a prefix of the corresponding field in the record. */
+UNIV_INTERN
+ibool
+cmp_dtuple_is_prefix_of_rec(
+/*========================*/
+				/* out: TRUE if prefix */
+	const dtuple_t*	dtuple,	/* in: data tuple */
+	const rec_t*	rec,	/* in: physical record */
+	const ulint*	offsets);/* in: array returned by rec_get_offsets() */
+#ifndef UNIV_HOTBACKUP
+/*****************************************************************
+Compare two physical records that contain the same number of columns,
+none of which are stored externally. */
+UNIV_INTERN
+int
+cmp_rec_rec_simple(
+/*===============*/
+					/* out: 1, 0 , -1 if rec1 is greater,
+					equal, less, respectively, than rec2 */
+	const rec_t*		rec1,	/* in: physical record */
+	const rec_t*		rec2,	/* in: physical record */
+	const ulint*		offsets1,/* in: rec_get_offsets(rec1, index) */
+	const ulint*		offsets2,/* in: rec_get_offsets(rec2, index) */
+	const dict_index_t*	index);	/* in: data dictionary index */
+#endif /* !UNIV_HOTBACKUP */
+/*****************************************************************
+This function is used to compare two physical records. Only the common
+first fields are compared, and if an externally stored field is
+encountered, then 0 is returned. */
+UNIV_INTERN
+int
+cmp_rec_rec_with_match(
+/*===================*/
+				/* out: 1, 0 , -1 if rec1 is greater, equal,
+				less, respectively, than rec2; only the common
+				first fields are compared */
+	const rec_t*	rec1,	/* in: physical record */
+	const rec_t*	rec2,	/* in: physical record */
+	const ulint*	offsets1,/* in: rec_get_offsets(rec1, index) */
+	const ulint*	offsets2,/* in: rec_get_offsets(rec2, index) */
+	dict_index_t*	index,	/* in: data dictionary index */
+	ulint*		matched_fields, /* in/out: number of already completely
+				matched fields; when the function returns,
+				contains the value the for current
+				comparison */
+	ulint*		matched_bytes);/* in/out: number of already matched
+				bytes within the first field not completely
+				matched; when the function returns, contains
+				the value for the current comparison */
+/*****************************************************************
+This function is used to compare two physical records. Only the common
+first fields are compared. */
+UNIV_INLINE
+int
+cmp_rec_rec(
+/*========*/
+				/* out: 1, 0 , -1 if rec1 is greater, equal,
+				less, respectively, than rec2; only the common
+				first fields are compared */
+	const rec_t*	rec1,	/* in: physical record */
+	const rec_t*	rec2,	/* in: physical record */
+	const ulint*	offsets1,/* in: rec_get_offsets(rec1, index) */
+	const ulint*	offsets2,/* in: rec_get_offsets(rec2, index) */
+	dict_index_t*	index);	/* in: data dictionary index */
+
+
+#ifndef UNIV_NONINL
+#include "rem0cmp.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/rem0cmp.ic b/storage/xtradb/include/rem0cmp.ic
new file mode 100644
index 00000000000..6c58d9e5a25
--- /dev/null
+++ b/storage/xtradb/include/rem0cmp.ic
@@ -0,0 +1,92 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/***********************************************************************
+Comparison services for records
+
+Created 7/1/1994 Heikki Tuuri
+************************************************************************/
+
+/*****************************************************************
+This function is used to compare two data fields for which we know the
+data type. */
+UNIV_INLINE
+int
+cmp_data_data(
+/*==========*/
+				/* out: 1, 0, -1, if data1 is greater, equal,
+				less than data2, respectively */
+	ulint		mtype,	/* in: main type */
+	ulint		prtype,	/* in: precise type */
+	const byte*	data1,	/* in: data field (== a pointer to a memory
+				buffer) */
+	ulint		len1,	/* in: data field length or UNIV_SQL_NULL */
+	const byte*	data2,	/* in: data field (== a pointer to a memory
+				buffer) */
+	ulint		len2)	/* in: data field length or UNIV_SQL_NULL */
+{
+	return(cmp_data_data_slow(mtype, prtype, data1, len1, data2, len2));
+}
+
+/*****************************************************************
+This function is used to compare two dfields where at least the first
+has its data type field set. */
+UNIV_INLINE
+int
+cmp_dfield_dfield(
+/*==============*/
+				/* out: 1, 0, -1, if dfield1 is greater, equal,
+				less than dfield2, respectively */
+	const dfield_t*	dfield1,/* in: data field; must have type field set */
+	const dfield_t*	dfield2)/* in: data field */
+{
+	const dtype_t*	type;
+
+	ut_ad(dfield_check_typed(dfield1));
+
+	type = dfield_get_type(dfield1);
+
+	return(cmp_data_data(type->mtype, type->prtype,
+			     (const byte*) dfield_get_data(dfield1),
+			     dfield_get_len(dfield1),
+			     (const byte*) dfield_get_data(dfield2),
+			     dfield_get_len(dfield2)));
+}
+
+/*****************************************************************
+This function is used to compare two physical records. Only the common
+first fields are compared. */
+UNIV_INLINE
+int
+cmp_rec_rec(
+/*========*/
+				/* out: 1, 0 , -1 if rec1 is greater, equal,
+				less, respectively, than rec2; only the common
+				first fields are compared */
+	const rec_t*	rec1,	/* in: physical record */
+	const rec_t*	rec2,	/* in: physical record */
+	const ulint*	offsets1,/* in: rec_get_offsets(rec1, index) */
+	const ulint*	offsets2,/* in: rec_get_offsets(rec2, index) */
+	dict_index_t*	index)	/* in: data dictionary index */
+{
+	ulint	match_f		= 0;
+	ulint	match_b		= 0;
+
+	return(cmp_rec_rec_with_match(rec1, rec2, offsets1, offsets2, index,
+				      &match_f, &match_b));
+}
diff --git a/storage/xtradb/include/rem0rec.h b/storage/xtradb/include/rem0rec.h
new file mode 100644
index 00000000000..cb72a5fa25b
--- /dev/null
+++ b/storage/xtradb/include/rem0rec.h
@@ -0,0 +1,822 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/************************************************************************
+Record manager
+
+Created 5/30/1994 Heikki Tuuri
+*************************************************************************/
+
+#ifndef rem0rec_h
+#define rem0rec_h
+
+#include "univ.i"
+#include "data0data.h"
+#include "rem0types.h"
+#include "mtr0types.h"
+#include "page0types.h"
+
+/* Info bit denoting the predefined minimum record: this bit is set
+if and only if the record is the first user record on a non-leaf
+B-tree page that is the leftmost page on its level
+(PAGE_LEVEL is nonzero and FIL_PAGE_PREV is FIL_NULL). */
+#define REC_INFO_MIN_REC_FLAG	0x10UL
+/* The deleted flag in info bits */
+#define REC_INFO_DELETED_FLAG	0x20UL	/* when bit is set to 1, it means the
+					record has been delete marked */
+
+/* Number of extra bytes in an old-style record,
+in addition to the data and the offsets */
+#define REC_N_OLD_EXTRA_BYTES	6
+/* Number of extra bytes in a new-style record,
+in addition to the data and the offsets */
+#define REC_N_NEW_EXTRA_BYTES	5
+
+/* Record status values */
+#define REC_STATUS_ORDINARY	0
+#define REC_STATUS_NODE_PTR	1
+#define REC_STATUS_INFIMUM	2
+#define REC_STATUS_SUPREMUM	3
+
+/* The following four constants are needed in page0zip.c in order to
+efficiently compress and decompress pages. */
+
+/* The offset of heap_no in a compact record */
+#define REC_NEW_HEAP_NO		4
+/* The shift of heap_no in a compact record.
+The status is stored in the low-order bits. */
+#define	REC_HEAP_NO_SHIFT	3
+
+/* Length of a B-tree node pointer, in bytes */
+#define REC_NODE_PTR_SIZE	4
+
+#ifdef UNIV_DEBUG
+/* Length of the rec_get_offsets() header */
+# define REC_OFFS_HEADER_SIZE	4
+#else /* UNIV_DEBUG */
+/* Length of the rec_get_offsets() header */
+# define REC_OFFS_HEADER_SIZE	2
+#endif /* UNIV_DEBUG */
+
+/* Number of elements that should be initially allocated for the
+offsets[] array, first passed to rec_get_offsets() */
+#define REC_OFFS_NORMAL_SIZE	100
+#define REC_OFFS_SMALL_SIZE	10
+
+/**********************************************************
+The following function is used to get the pointer of the next chained record
+on the same page. */
+UNIV_INLINE
+const rec_t*
+rec_get_next_ptr_const(
+/*===================*/
+				/* out: pointer to the next chained record, or
+				NULL if none */
+	const rec_t*	rec,	/* in: physical record */
+	ulint		comp);	/* in: nonzero=compact page format */
+/**********************************************************
+The following function is used to get the pointer of the next chained record
+on the same page. */
+UNIV_INLINE
+rec_t*
+rec_get_next_ptr(
+/*=============*/
+			/* out: pointer to the next chained record, or
+			NULL if none */
+	rec_t*	rec,	/* in: physical record */
+	ulint	comp);	/* in: nonzero=compact page format */
+/**********************************************************
+The following function is used to get the offset of the
+next chained record on the same page. */
+UNIV_INLINE
+ulint
+rec_get_next_offs(
+/*==============*/
+				/* out: the page offset of the next
+				chained record, or 0 if none */
+	const rec_t*	rec,	/* in: physical record */
+	ulint		comp);	/* in: nonzero=compact page format */
+/**********************************************************
+The following function is used to set the next record offset field
+of an old-style record. */
+UNIV_INLINE
+void
+rec_set_next_offs_old(
+/*==================*/
+	rec_t*	rec,	/* in: old-style physical record */
+	ulint	next);	/* in: offset of the next record */
+/**********************************************************
+The following function is used to set the next record offset field
+of a new-style record. */
+UNIV_INLINE
+void
+rec_set_next_offs_new(
+/*==================*/
+	rec_t*	rec,	/* in/out: new-style physical record */
+	ulint	next);	/* in: offset of the next record */
+/**********************************************************
+The following function is used to get the number of fields
+in an old-style record. */
+UNIV_INLINE
+ulint
+rec_get_n_fields_old(
+/*=================*/
+				/* out: number of data fields */
+	const rec_t*	rec);	/* in: physical record */
+/**********************************************************
+The following function is used to get the number of fields
+in a record. */
+UNIV_INLINE
+ulint
+rec_get_n_fields(
+/*=============*/
+					/* out: number of data fields */
+	const rec_t*		rec,	/* in: physical record */
+	const dict_index_t*	index);	/* in: record descriptor */
+/**********************************************************
+The following function is used to get the number of records owned by the
+previous directory record. */
+UNIV_INLINE
+ulint
+rec_get_n_owned_old(
+/*================*/
+				/* out: number of owned records */
+	const rec_t*	rec);	/* in: old-style physical record */
+/**********************************************************
+The following function is used to set the number of owned records. */
+UNIV_INLINE
+void
+rec_set_n_owned_old(
+/*================*/
+				/* out: TRUE on success */
+	rec_t*	rec,		/* in: old-style physical record */
+	ulint	n_owned);	/* in: the number of owned */
+/**********************************************************
+The following function is used to get the number of records owned by the
+previous directory record. */
+UNIV_INLINE
+ulint
+rec_get_n_owned_new(
+/*================*/
+				/* out: number of owned records */
+	const rec_t*	rec);	/* in: new-style physical record */
+/**********************************************************
+The following function is used to set the number of owned records. */
+UNIV_INLINE
+void
+rec_set_n_owned_new(
+/*================*/
+	rec_t*		rec,	/* in/out: new-style physical record */
+	page_zip_des_t*	page_zip,/* in/out: compressed page, or NULL */
+	ulint		n_owned);/* in: the number of owned */
+/**********************************************************
+The following function is used to retrieve the info bits of
+a record. */
+UNIV_INLINE
+ulint
+rec_get_info_bits(
+/*==============*/
+				/* out: info bits */
+	const rec_t*	rec,	/* in: physical record */
+	ulint		comp);	/* in: nonzero=compact page format */
+/**********************************************************
+The following function is used to set the info bits of a record. */
+UNIV_INLINE
+void
+rec_set_info_bits_old(
+/*==================*/
+	rec_t*	rec,	/* in: old-style physical record */
+	ulint	bits);	/* in: info bits */
+/**********************************************************
+The following function is used to set the info bits of a record. */
+UNIV_INLINE
+void
+rec_set_info_bits_new(
+/*==================*/
+	rec_t*	rec,	/* in/out: new-style physical record */
+	ulint	bits);	/* in: info bits */
+/**********************************************************
+The following function retrieves the status bits of a new-style record. */
+UNIV_INLINE
+ulint
+rec_get_status(
+/*===========*/
+				/* out: status bits */
+	const rec_t*	rec);	/* in: physical record */
+
+/**********************************************************
+The following function is used to set the status bits of a new-style record. */
+UNIV_INLINE
+void
+rec_set_status(
+/*===========*/
+	rec_t*	rec,	/* in/out: physical record */
+	ulint	bits);	/* in: info bits */
+
+/**********************************************************
+The following function is used to retrieve the info and status
+bits of a record.  (Only compact records have status bits.) */
+UNIV_INLINE
+ulint
+rec_get_info_and_status_bits(
+/*=========================*/
+				/* out: info bits */
+	const rec_t*	rec,	/* in: physical record */
+	ulint		comp);	/* in: nonzero=compact page format */
+/**********************************************************
+The following function is used to set the info and status
+bits of a record.  (Only compact records have status bits.) */
+UNIV_INLINE
+void
+rec_set_info_and_status_bits(
+/*=========================*/
+	rec_t*	rec,	/* in/out: compact physical record */
+	ulint	bits);	/* in: info bits */
+
+/**********************************************************
+The following function tells if record is delete marked. */
+UNIV_INLINE
+ulint
+rec_get_deleted_flag(
+/*=================*/
+				/* out: nonzero if delete marked */
+	const rec_t*	rec,	/* in: physical record */
+	ulint		comp);	/* in: nonzero=compact page format */
+/**********************************************************
+The following function is used to set the deleted bit. */
+UNIV_INLINE
+void
+rec_set_deleted_flag_old(
+/*=====================*/
+	rec_t*	rec,	/* in: old-style physical record */
+	ulint	flag);	/* in: nonzero if delete marked */
+/**********************************************************
+The following function is used to set the deleted bit. */
+UNIV_INLINE
+void
+rec_set_deleted_flag_new(
+/*=====================*/
+	rec_t*		rec,	/* in/out: new-style physical record */
+	page_zip_des_t*	page_zip,/* in/out: compressed page, or NULL */
+	ulint		flag);	/* in: nonzero if delete marked */
+/**********************************************************
+The following function tells if a new-style record is a node pointer. */
+UNIV_INLINE
+ibool
+rec_get_node_ptr_flag(
+/*==================*/
+				/* out: TRUE if node pointer */
+	const rec_t*	rec);	/* in: physical record */
+/**********************************************************
+The following function is used to get the order number
+of an old-style record in the heap of the index page. */
+UNIV_INLINE
+ulint
+rec_get_heap_no_old(
+/*================*/
+				/* out: heap order number */
+	const rec_t*	rec);	/* in: physical record */
+/**********************************************************
+The following function is used to set the heap number
+field in an old-style record. */
+UNIV_INLINE
+void
+rec_set_heap_no_old(
+/*================*/
+	rec_t*	rec,	/* in: physical record */
+	ulint	heap_no);/* in: the heap number */
+/**********************************************************
+The following function is used to get the order number
+of a new-style record in the heap of the index page. */
+UNIV_INLINE
+ulint
+rec_get_heap_no_new(
+/*================*/
+				/* out: heap order number */
+	const rec_t*	rec);	/* in: physical record */
+/**********************************************************
+The following function is used to set the heap number
+field in a new-style record. */
+UNIV_INLINE
+void
+rec_set_heap_no_new(
+/*================*/
+	rec_t*	rec,	/* in/out: physical record */
+	ulint	heap_no);/* in: the heap number */
+/**********************************************************
+The following function is used to test whether the data offsets
+in the record are stored in one-byte or two-byte format. */
+UNIV_INLINE
+ibool
+rec_get_1byte_offs_flag(
+/*====================*/
+				/* out: TRUE if 1-byte form */
+	const rec_t*	rec);	/* in: physical record */
+
+/**********************************************************
+Determine how many of the first n columns in a compact
+physical record are stored externally. */
+UNIV_INTERN
+ulint
+rec_get_n_extern_new(
+/*=================*/
+				/* out: number of externally stored columns */
+	const rec_t*	rec,	/* in: compact physical record */
+	dict_index_t*	index,	/* in: record descriptor */
+	ulint		n);	/* in: number of columns to scan */
+
+/**********************************************************
+The following function determines the offsets to each field
+in the record.	It can reuse a previously allocated array. */
+UNIV_INTERN
+ulint*
+rec_get_offsets_func(
+/*=================*/
+					/* out: the new offsets */
+	const rec_t*		rec,	/* in: physical record */
+	const dict_index_t*	index,	/* in: record descriptor */
+	ulint*			offsets,/* in/out: array consisting of
+					offsets[0] allocated elements,
+					or an array from rec_get_offsets(),
+					or NULL */
+	ulint			n_fields,/* in: maximum number of
+					initialized fields
+					 (ULINT_UNDEFINED if all fields) */
+	mem_heap_t**		heap,	/* in/out: memory heap */
+	const char*		file,	/* in: file name where called */
+	ulint			line);	/* in: line number where called */
+
+#define rec_get_offsets(rec,index,offsets,n,heap)	\
+	rec_get_offsets_func(rec,index,offsets,n,heap,__FILE__,__LINE__)
+
+/**********************************************************
+Determine the offset to each field in a leaf-page record
+in ROW_FORMAT=COMPACT.  This is a special case of
+rec_init_offsets() and rec_get_offsets_func(). */
+UNIV_INTERN
+void
+rec_init_offsets_comp_ordinary(
+/*===========================*/
+	const rec_t*		rec,	/* in: physical record in
+					ROW_FORMAT=COMPACT */
+	ulint			extra,	/* in: number of bytes to reserve
+					between the record header and
+					the data payload
+					(usually REC_N_NEW_EXTRA_BYTES) */
+	const dict_index_t*	index,	/* in: record descriptor */
+	ulint*			offsets);/* in/out: array of offsets;
+					in: n=rec_offs_n_fields(offsets) */
+
+/**********************************************************
+The following function determines the offsets to each field
+in the record.  It can reuse a previously allocated array. */
+UNIV_INTERN
+void
+rec_get_offsets_reverse(
+/*====================*/
+	const byte*		extra,	/* in: the extra bytes of a
+					compact record in reverse order,
+					excluding the fixed-size
+					REC_N_NEW_EXTRA_BYTES */
+	const dict_index_t*	index,	/* in: record descriptor */
+	ulint			node_ptr,/* in: nonzero=node pointer,
+					0=leaf node */
+	ulint*			offsets);/* in/out: array consisting of
+					offsets[0] allocated elements */
+
+/****************************************************************
+Validates offsets returned by rec_get_offsets(). */
+UNIV_INLINE
+ibool
+rec_offs_validate(
+/*==============*/
+					/* out: TRUE if valid */
+	const rec_t*		rec,	/* in: record or NULL */
+	const dict_index_t*	index,	/* in: record descriptor or NULL */
+	const ulint*		offsets);/* in: array returned by
+					rec_get_offsets() */
+#ifdef UNIV_DEBUG
+/****************************************************************
+Updates debug data in offsets, in order to avoid bogus
+rec_offs_validate() failures. */
+UNIV_INLINE
+void
+rec_offs_make_valid(
+/*================*/
+	const rec_t*		rec,	/* in: record */
+	const dict_index_t*	index,	/* in: record descriptor */
+	ulint*			offsets);/* in: array returned by
+					rec_get_offsets() */
+#else
+# define rec_offs_make_valid(rec, index, offsets) ((void) 0)
+#endif /* UNIV_DEBUG */
+
+/****************************************************************
+The following function is used to get the offset to the nth
+data field in an old-style record. */
+UNIV_INTERN
+ulint
+rec_get_nth_field_offs_old(
+/*=======================*/
+				/* out: offset to the field */
+	const rec_t*	rec,	/* in: record */
+	ulint		n,	/* in: index of the field */
+	ulint*		len);	/* out: length of the field; UNIV_SQL_NULL
+				if SQL null */
+#define rec_get_nth_field_old(rec, n, len) \
+((rec) + rec_get_nth_field_offs_old(rec, n, len))
+/****************************************************************
+Gets the physical size of an old-style field.
+Also an SQL null may have a field of size > 0,
+if the data type is of a fixed size. */
+UNIV_INLINE
+ulint
+rec_get_nth_field_size(
+/*===================*/
+				/* out: field size in bytes */
+	const rec_t*	rec,	/* in: record */
+	ulint		n);	/* in: index of the field */
+/****************************************************************
+The following function is used to get an offset to the nth
+data field in a record. */
+UNIV_INLINE
+ulint
+rec_get_nth_field_offs(
+/*===================*/
+				/* out: offset from the origin of rec */
+	const ulint*	offsets,/* in: array returned by rec_get_offsets() */
+	ulint		n,	/* in: index of the field */
+	ulint*		len);	/* out: length of the field; UNIV_SQL_NULL
+				if SQL null */
+#define rec_get_nth_field(rec, offsets, n, len) \
+((rec) + rec_get_nth_field_offs(offsets, n, len))
+/**********************************************************
+Determine if the offsets are for a record in the new
+compact format. */
+UNIV_INLINE
+ulint
+rec_offs_comp(
+/*==========*/
+				/* out: nonzero if compact format */
+	const ulint*	offsets);/* in: array returned by rec_get_offsets() */
+/**********************************************************
+Determine if the offsets are for a record containing
+externally stored columns. */
+UNIV_INLINE
+ulint
+rec_offs_any_extern(
+/*================*/
+				/* out: nonzero if externally stored */
+	const ulint*	offsets);/* in: array returned by rec_get_offsets() */
+/**********************************************************
+Returns nonzero if the extern bit is set in nth field of rec. */
+UNIV_INLINE
+ulint
+rec_offs_nth_extern(
+/*================*/
+				/* out: nonzero if externally stored */
+	const ulint*	offsets,/* in: array returned by rec_get_offsets() */
+	ulint		n);	/* in: nth field */
+/**********************************************************
+Returns nonzero if the SQL NULL bit is set in nth field of rec. */
+UNIV_INLINE
+ulint
+rec_offs_nth_sql_null(
+/*==================*/
+				/* out: nonzero if SQL NULL */
+	const ulint*	offsets,/* in: array returned by rec_get_offsets() */
+	ulint		n);	/* in: nth field */
+/**********************************************************
+Gets the physical size of a field. */
+UNIV_INLINE
+ulint
+rec_offs_nth_size(
+/*==============*/
+				/* out: length of field */
+	const ulint*	offsets,/* in: array returned by rec_get_offsets() */
+	ulint		n);	/* in: nth field */
+
+/**********************************************************
+Returns the number of extern bits set in a record. */
+UNIV_INLINE
+ulint
+rec_offs_n_extern(
+/*==============*/
+				/* out: number of externally stored fields */
+	const ulint*	offsets);/* in: array returned by rec_get_offsets() */
+/***************************************************************
+This is used to modify the value of an already existing field in a record.
+The previous value must have exactly the same size as the new value. If len
+is UNIV_SQL_NULL then the field is treated as an SQL null.
+For records in ROW_FORMAT=COMPACT (new-style records), len must not be
+UNIV_SQL_NULL unless the field already is SQL null. */
+UNIV_INLINE
+void
+rec_set_nth_field(
+/*==============*/
+	rec_t*		rec,	/* in: record */
+	const ulint*	offsets,/* in: array returned by rec_get_offsets() */
+	ulint		n,	/* in: index number of the field */
+	const void*	data,	/* in: pointer to the data if not SQL null */
+	ulint		len);	/* in: length of the data or UNIV_SQL_NULL */
+/**************************************************************
+The following function returns the data size of an old-style physical
+record, that is the sum of field lengths. SQL null fields
+are counted as length 0 fields. The value returned by the function
+is the distance from record origin to record end in bytes. */
+UNIV_INLINE
+ulint
+rec_get_data_size_old(
+/*==================*/
+				/* out: size */
+	const rec_t*	rec);	/* in: physical record */
+/**************************************************************
+The following function returns the number of allocated elements
+for an array of offsets. */
+UNIV_INLINE
+ulint
+rec_offs_get_n_alloc(
+/*=================*/
+				/* out: number of elements */
+	const ulint*	offsets);/* in: array for rec_get_offsets() */
+/**************************************************************
+The following function sets the number of allocated elements
+for an array of offsets. */
+UNIV_INLINE
+void
+rec_offs_set_n_alloc(
+/*=================*/
+	ulint*	offsets,	/* out: array for rec_get_offsets(),
+				must be allocated */
+	ulint	n_alloc);	/* in: number of elements */
+#define rec_offs_init(offsets) \
+	rec_offs_set_n_alloc(offsets, (sizeof offsets) / sizeof *offsets)
+/**************************************************************
+The following function returns the number of fields in a record. */
+UNIV_INLINE
+ulint
+rec_offs_n_fields(
+/*==============*/
+				/* out: number of fields */
+	const ulint*	offsets);/* in: array returned by rec_get_offsets() */
+/**************************************************************
+The following function returns the data size of a physical
+record, that is the sum of field lengths. SQL null fields
+are counted as length 0 fields. The value returned by the function
+is the distance from record origin to record end in bytes. */
+UNIV_INLINE
+ulint
+rec_offs_data_size(
+/*===============*/
+				/* out: size */
+	const ulint*	offsets);/* in: array returned by rec_get_offsets() */
+/**************************************************************
+Returns the total size of record minus data size of record.
+The value returned by the function is the distance from record
+start to record origin in bytes. */
+UNIV_INLINE
+ulint
+rec_offs_extra_size(
+/*================*/
+				/* out: size */
+	const ulint*	offsets);/* in: array returned by rec_get_offsets() */
+/**************************************************************
+Returns the total size of a physical record.  */
+UNIV_INLINE
+ulint
+rec_offs_size(
+/*==========*/
+				/* out: size */
+	const ulint*	offsets);/* in: array returned by rec_get_offsets() */
+/**************************************************************
+Returns a pointer to the start of the record. */
+UNIV_INLINE
+byte*
+rec_get_start(
+/*==========*/
+				/* out: pointer to start */
+	rec_t*		rec,	/* in: pointer to record */
+	const ulint*	offsets);/* in: array returned by rec_get_offsets() */
+/**************************************************************
+Returns a pointer to the end of the record. */
+UNIV_INLINE
+byte*
+rec_get_end(
+/*========*/
+				/* out: pointer to end */
+	rec_t*		rec,	/* in: pointer to record */
+	const ulint*	offsets);/* in: array returned by rec_get_offsets() */
+/*******************************************************************
+Copies a physical record to a buffer. */
+UNIV_INLINE
+rec_t*
+rec_copy(
+/*=====*/
+				/* out: pointer to the origin of the copy */
+	void*		buf,	/* in: buffer */
+	const rec_t*	rec,	/* in: physical record */
+	const ulint*	offsets);/* in: array returned by rec_get_offsets() */
+/******************************************************************
+Copies the first n fields of a physical record to a new physical record in
+a buffer. */
+UNIV_INTERN
+rec_t*
+rec_copy_prefix_to_buf(
+/*===================*/
+						/* out, own: copied record */
+	const rec_t*		rec,		/* in: physical record */
+	const dict_index_t*	index,		/* in: record descriptor */
+	ulint			n_fields,	/* in: number of fields
+						to copy */
+	byte**			buf,		/* in/out: memory buffer
+						for the copied prefix,
+						or NULL */
+	ulint*			buf_size);	/* in/out: buffer size */
+/****************************************************************
+Folds a prefix of a physical record to a ulint. */
+UNIV_INLINE
+ulint
+rec_fold(
+/*=====*/
+					/* out: the folded value */
+	const rec_t*	rec,		/* in: the physical record */
+	const ulint*	offsets,	/* in: array returned by
+					rec_get_offsets() */
+	ulint		n_fields,	/* in: number of complete
+					fields to fold */
+	ulint		n_bytes,	/* in: number of bytes to fold
+					in an incomplete last field */
+	dulint		tree_id)	/* in: index tree id */
+	__attribute__((pure));
+/*************************************************************
+Builds a ROW_FORMAT=COMPACT record out of a data tuple. */
+UNIV_INTERN
+void
+rec_convert_dtuple_to_rec_comp(
+/*===========================*/
+	rec_t*			rec,	/* in: origin of record */
+	ulint			extra,	/* in: number of bytes to
+					reserve between the record
+					header and the data payload
+					(normally REC_N_NEW_EXTRA_BYTES) */
+	const dict_index_t*	index,	/* in: record descriptor */
+	ulint			status,	/* in: status bits of the record */
+	const dfield_t*		fields,	/* in: array of data fields */
+	ulint			n_fields);/* in: number of data fields */
+/*************************************************************
+Builds a physical record out of a data tuple and
+stores it into the given buffer. */
+UNIV_INTERN
+rec_t*
+rec_convert_dtuple_to_rec(
+/*======================*/
+					/* out: pointer to the origin
+					of physical record */
+	byte*			buf,	/* in: start address of the
+					physical record */
+	const dict_index_t*	index,	/* in: record descriptor */
+	const dtuple_t*		dtuple,	/* in: data tuple */
+	ulint			n_ext);	/* in: number of
+					externally stored columns */
+/**************************************************************
+Returns the extra size of an old-style physical record if we know its
+data size and number of fields. */
+UNIV_INLINE
+ulint
+rec_get_converted_extra_size(
+/*=========================*/
+				/* out: extra size */
+	ulint	data_size,	/* in: data size */
+	ulint	n_fields,	/* in: number of fields */
+	ulint	n_ext)		/* in: number of externally stored columns */
+		__attribute__((const));
+/**************************************************************
+Determines the size of a data tuple prefix in ROW_FORMAT=COMPACT. */
+UNIV_INTERN
+ulint
+rec_get_converted_size_comp_prefix(
+/*===============================*/
+					/* out: total size */
+	const dict_index_t*	index,	/* in: record descriptor;
+					dict_table_is_comp() is
+					assumed to hold, even if
+					it does not */
+	const dfield_t*		fields,	/* in: array of data fields */
+	ulint			n_fields,/* in: number of data fields */
+	ulint*			extra);	/* out: extra size */
+/**************************************************************
+Determines the size of a data tuple in ROW_FORMAT=COMPACT. */
+UNIV_INTERN
+ulint
+rec_get_converted_size_comp(
+/*========================*/
+					/* out: total size */
+	const dict_index_t*	index,	/* in: record descriptor;
+					dict_table_is_comp() is
+					assumed to hold, even if
+					it does not */
+	ulint			status,	/* in: status bits of the record */
+	const dfield_t*		fields,	/* in: array of data fields */
+	ulint			n_fields,/* in: number of data fields */
+	ulint*			extra);	/* out: extra size */
+/**************************************************************
+The following function returns the size of a data tuple when converted to
+a physical record. */
+UNIV_INLINE
+ulint
+rec_get_converted_size(
+/*===================*/
+				/* out: size */
+	dict_index_t*	index,	/* in: record descriptor */
+	const dtuple_t*	dtuple,	/* in: data tuple */
+	ulint		n_ext);	/* in: number of externally stored columns */
+/******************************************************************
+Copies the first n fields of a physical record to a data tuple.
+The fields are copied to the memory heap. */
+UNIV_INTERN
+void
+rec_copy_prefix_to_dtuple(
+/*======================*/
+	dtuple_t*		tuple,		/* out: data tuple */
+	const rec_t*		rec,		/* in: physical record */
+	const dict_index_t*	index,		/* in: record descriptor */
+	ulint			n_fields,	/* in: number of fields
+						to copy */
+	mem_heap_t*		heap);		/* in: memory heap */
+/*******************************************************************
+Validates the consistency of a physical record. */
+UNIV_INTERN
+ibool
+rec_validate(
+/*=========*/
+				/* out: TRUE if ok */
+	const rec_t*	rec,	/* in: physical record */
+	const ulint*	offsets);/* in: array returned by rec_get_offsets() */
+/*******************************************************************
+Prints an old-style physical record. */
+UNIV_INTERN
+void
+rec_print_old(
+/*==========*/
+	FILE*		file,	/* in: file where to print */
+	const rec_t*	rec);	/* in: physical record */
+/*******************************************************************
+Prints a physical record in ROW_FORMAT=COMPACT.  Ignores the
+record header. */
+UNIV_INTERN
+void
+rec_print_comp(
+/*===========*/
+	FILE*		file,	/* in: file where to print */
+	const rec_t*	rec,	/* in: physical record */
+	const ulint*	offsets);/* in: array returned by rec_get_offsets() */
+/*******************************************************************
+Prints a physical record. */
+UNIV_INTERN
+void
+rec_print_new(
+/*==========*/
+	FILE*		file,	/* in: file where to print */
+	const rec_t*	rec,	/* in: physical record */
+	const ulint*	offsets);/* in: array returned by rec_get_offsets() */
+/*******************************************************************
+Prints a physical record. */
+UNIV_INTERN
+void
+rec_print(
+/*======*/
+	FILE*		file,	/* in: file where to print */
+	const rec_t*	rec,	/* in: physical record */
+	dict_index_t*	index);	/* in: record descriptor */
+
+#define REC_INFO_BITS		6	/* This is single byte bit-field */
+
+/* Maximum lengths for the data in a physical record if the offsets
+are given in one byte (resp. two byte) format. */
+#define REC_1BYTE_OFFS_LIMIT	0x7FUL
+#define REC_2BYTE_OFFS_LIMIT	0x7FFFUL
+
+/* The data size of record must be smaller than this because we reserve
+two upmost bits in a two byte offset for special purposes */
+#define REC_MAX_DATA_SIZE	(16 * 1024)
+
+#ifndef UNIV_NONINL
+#include "rem0rec.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/rem0rec.ic b/storage/xtradb/include/rem0rec.ic
new file mode 100644
index 00000000000..0b2b9f4a685
--- /dev/null
+++ b/storage/xtradb/include/rem0rec.ic
@@ -0,0 +1,1652 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/************************************************************************
+Record manager
+
+Created 5/30/1994 Heikki Tuuri
+*************************************************************************/
+
+#include "mach0data.h"
+#include "ut0byte.h"
+#include "dict0dict.h"
+
+/* Compact flag ORed to the extra size returned by rec_get_offsets() */
+#define REC_OFFS_COMPACT	((ulint) 1 << 31)
+/* SQL NULL flag in offsets returned by rec_get_offsets() */
+#define REC_OFFS_SQL_NULL	((ulint) 1 << 31)
+/* External flag in offsets returned by rec_get_offsets() */
+#define REC_OFFS_EXTERNAL	((ulint) 1 << 30)
+/* Mask for offsets returned by rec_get_offsets() */
+#define REC_OFFS_MASK		(REC_OFFS_EXTERNAL - 1)
+
+/* Offsets of the bit-fields in an old-style record. NOTE! In the table the
+most significant bytes and bits are written below less significant.
+
+	(1) byte offset		(2) bit usage within byte
+	downward from
+	origin ->	1	8 bits pointer to next record
+			2	8 bits pointer to next record
+			3	1 bit short flag
+				7 bits number of fields
+			4	3 bits number of fields
+				5 bits heap number
+			5	8 bits heap number
+			6	4 bits n_owned
+				4 bits info bits
+*/
+
+/* Offsets of the bit-fields in a new-style record. NOTE! In the table the
+most significant bytes and bits are written below less significant.
+
+	(1) byte offset		(2) bit usage within byte
+	downward from
+	origin ->	1	8 bits relative offset of next record
+			2	8 bits relative offset of next record
+				  the relative offset is an unsigned 16-bit
+				  integer:
+				  (offset_of_next_record
+				   - offset_of_this_record) mod 64Ki,
+				  where mod is the modulo as a non-negative
+				  number;
+				  we can calculate the the offset of the next
+				  record with the formula:
+				  relative_offset + offset_of_this_record
+				  mod UNIV_PAGE_SIZE
+			3	3 bits status:
+					000=conventional record
+					001=node pointer record (inside B-tree)
+					010=infimum record
+					011=supremum record
+					1xx=reserved
+				5 bits heap number
+			4	8 bits heap number
+			5	4 bits n_owned
+				4 bits info bits
+*/
+
+/* We list the byte offsets from the origin of the record, the mask,
+and the shift needed to obtain each bit-field of the record. */
+
+#define REC_NEXT		2
+#define REC_NEXT_MASK		0xFFFFUL
+#define REC_NEXT_SHIFT		0
+
+#define REC_OLD_SHORT		3	/* This is single byte bit-field */
+#define REC_OLD_SHORT_MASK	0x1UL
+#define REC_OLD_SHORT_SHIFT	0
+
+#define REC_OLD_N_FIELDS	4
+#define REC_OLD_N_FIELDS_MASK	0x7FEUL
+#define REC_OLD_N_FIELDS_SHIFT	1
+
+#define REC_NEW_STATUS		3	/* This is single byte bit-field */
+#define REC_NEW_STATUS_MASK	0x7UL
+#define REC_NEW_STATUS_SHIFT	0
+
+#define REC_OLD_HEAP_NO		5
+#define REC_HEAP_NO_MASK	0xFFF8UL
+#if 0 /* defined in rem0rec.h for use of page0zip.c */
+#define REC_NEW_HEAP_NO		4
+#define	REC_HEAP_NO_SHIFT	3
+#endif
+
+#define REC_OLD_N_OWNED		6	/* This is single byte bit-field */
+#define REC_NEW_N_OWNED		5	/* This is single byte bit-field */
+#define	REC_N_OWNED_MASK	0xFUL
+#define REC_N_OWNED_SHIFT	0
+
+#define REC_OLD_INFO_BITS	6	/* This is single byte bit-field */
+#define REC_NEW_INFO_BITS	5	/* This is single byte bit-field */
+#define	REC_INFO_BITS_MASK	0xF0UL
+#define REC_INFO_BITS_SHIFT	0
+
+/* The following masks are used to filter the SQL null bit from
+one-byte and two-byte offsets */
+
+#define REC_1BYTE_SQL_NULL_MASK	0x80UL
+#define REC_2BYTE_SQL_NULL_MASK	0x8000UL
+
+/* In a 2-byte offset the second most significant bit denotes
+a field stored to another page: */
+
+#define REC_2BYTE_EXTERN_MASK	0x4000UL
+
+#if REC_OLD_SHORT_MASK << (8 * (REC_OLD_SHORT - 3)) \
+		^ REC_OLD_N_FIELDS_MASK << (8 * (REC_OLD_N_FIELDS - 4)) \
+		^ REC_HEAP_NO_MASK << (8 * (REC_OLD_HEAP_NO - 4)) \
+		^ REC_N_OWNED_MASK << (8 * (REC_OLD_N_OWNED - 3)) \
+		^ REC_INFO_BITS_MASK << (8 * (REC_OLD_INFO_BITS - 3)) \
+		^ 0xFFFFFFFFUL
+# error "sum of old-style masks != 0xFFFFFFFFUL"
+#endif
+#if REC_NEW_STATUS_MASK << (8 * (REC_NEW_STATUS - 3)) \
+		^ REC_HEAP_NO_MASK << (8 * (REC_NEW_HEAP_NO - 4)) \
+		^ REC_N_OWNED_MASK << (8 * (REC_NEW_N_OWNED - 3)) \
+		^ REC_INFO_BITS_MASK << (8 * (REC_NEW_INFO_BITS - 3)) \
+		^ 0xFFFFFFUL
+# error "sum of new-style masks != 0xFFFFFFUL"
+#endif
+
+/***************************************************************
+Sets the value of the ith field SQL null bit of an old-style record. */
+UNIV_INTERN
+void
+rec_set_nth_field_null_bit(
+/*=======================*/
+	rec_t*	rec,	/* in: record */
+	ulint	i,	/* in: ith field */
+	ibool	val);	/* in: value to set */
+/***************************************************************
+Sets an old-style record field to SQL null.
+The physical size of the field is not changed. */
+UNIV_INTERN
+void
+rec_set_nth_field_sql_null(
+/*=======================*/
+	rec_t*	rec,	/* in: record */
+	ulint	n);	/* in: index of the field */
+
+/**********************************************************
+Gets a bit field from within 1 byte. */
+UNIV_INLINE
+ulint
+rec_get_bit_field_1(
+/*================*/
+	const rec_t*	rec,	/* in: pointer to record origin */
+	ulint		offs,	/* in: offset from the origin down */
+	ulint		mask,	/* in: mask used to filter bits */
+	ulint		shift)	/* in: shift right applied after masking */
+{
+	ut_ad(rec);
+
+	return((mach_read_from_1(rec - offs) & mask) >> shift);
+}
+
+/**********************************************************
+Sets a bit field within 1 byte. */
+UNIV_INLINE
+void
+rec_set_bit_field_1(
+/*================*/
+	rec_t*	rec,	/* in: pointer to record origin */
+	ulint	val,	/* in: value to set */
+	ulint	offs,	/* in: offset from the origin down */
+	ulint	mask,	/* in: mask used to filter bits */
+	ulint	shift)	/* in: shift right applied after masking */
+{
+	ut_ad(rec);
+	ut_ad(offs <= REC_N_OLD_EXTRA_BYTES);
+	ut_ad(mask);
+	ut_ad(mask <= 0xFFUL);
+	ut_ad(((mask >> shift) << shift) == mask);
+	ut_ad(((val << shift) & mask) == (val << shift));
+
+	mach_write_to_1(rec - offs,
+			(mach_read_from_1(rec - offs) & ~mask)
+			| (val << shift));
+}
+
+/**********************************************************
+Gets a bit field from within 2 bytes. */
+UNIV_INLINE
+ulint
+rec_get_bit_field_2(
+/*================*/
+	const rec_t*	rec,	/* in: pointer to record origin */
+	ulint		offs,	/* in: offset from the origin down */
+	ulint		mask,	/* in: mask used to filter bits */
+	ulint		shift)	/* in: shift right applied after masking */
+{
+	ut_ad(rec);
+
+	return((mach_read_from_2(rec - offs) & mask) >> shift);
+}
+
+/**********************************************************
+Sets a bit field within 2 bytes. */
+UNIV_INLINE
+void
+rec_set_bit_field_2(
+/*================*/
+	rec_t*	rec,	/* in: pointer to record origin */
+	ulint	val,	/* in: value to set */
+	ulint	offs,	/* in: offset from the origin down */
+	ulint	mask,	/* in: mask used to filter bits */
+	ulint	shift)	/* in: shift right applied after masking */
+{
+	ut_ad(rec);
+	ut_ad(offs <= REC_N_OLD_EXTRA_BYTES);
+	ut_ad(mask > 0xFFUL);
+	ut_ad(mask <= 0xFFFFUL);
+	ut_ad((mask >> shift) & 1);
+	ut_ad(0 == ((mask >> shift) & ((mask >> shift) + 1)));
+	ut_ad(((mask >> shift) << shift) == mask);
+	ut_ad(((val << shift) & mask) == (val << shift));
+
+	mach_write_to_2(rec - offs,
+			(mach_read_from_2(rec - offs) & ~mask)
+			| (val << shift));
+}
+
+/**********************************************************
+The following function is used to get the pointer of the next chained record
+on the same page. */
+UNIV_INLINE
+const rec_t*
+rec_get_next_ptr_const(
+/*===================*/
+				/* out: pointer to the next chained record, or
+				NULL if none */
+	const rec_t*	rec,	/* in: physical record */
+	ulint		comp)	/* in: nonzero=compact page format */
+{
+	ulint	field_value;
+
+	ut_ad(REC_NEXT_MASK == 0xFFFFUL);
+	ut_ad(REC_NEXT_SHIFT == 0);
+
+	field_value = mach_read_from_2(rec - REC_NEXT);
+
+	if (UNIV_UNLIKELY(field_value == 0)) {
+
+		return(NULL);
+	}
+
+	if (UNIV_EXPECT(comp, REC_OFFS_COMPACT)) {
+#if UNIV_PAGE_SIZE <= 32768
+		/* Note that for 64 KiB pages, field_value can 'wrap around'
+		and the debug assertion is not valid */
+
+		/* In the following assertion, field_value is interpreted
+		as signed 16-bit integer in 2's complement arithmetics.
+		If all platforms defined int16_t in the standard headers,
+		the expression could be written simpler as
+		(int16_t) field_value + ut_align_offset(...) < UNIV_PAGE_SIZE
+		*/
+		ut_ad((field_value >= 32768
+		       ? field_value - 65536
+		       : field_value)
+		      + ut_align_offset(rec, UNIV_PAGE_SIZE)
+		      < UNIV_PAGE_SIZE);
+#endif
+		/* There must be at least REC_N_NEW_EXTRA_BYTES + 1
+		between each record. */
+		ut_ad((field_value > REC_N_NEW_EXTRA_BYTES
+		       && field_value < 32768)
+		      || field_value < (uint16) -REC_N_NEW_EXTRA_BYTES);
+
+		return((byte*) ut_align_down(rec, UNIV_PAGE_SIZE)
+		       + ut_align_offset(rec + field_value, UNIV_PAGE_SIZE));
+	} else {
+		ut_ad(field_value < UNIV_PAGE_SIZE);
+
+		return((byte*) ut_align_down(rec, UNIV_PAGE_SIZE)
+		       + field_value);
+	}
+}
+
+/**********************************************************
+The following function is used to get the pointer of the next chained record
+on the same page. */
+UNIV_INLINE
+rec_t*
+rec_get_next_ptr(
+/*=============*/
+			/* out: pointer to the next chained record, or
+			NULL if none */
+	rec_t*	rec,	/* in: physical record */
+	ulint	comp)	/* in: nonzero=compact page format */
+{
+	return((rec_t*) rec_get_next_ptr_const(rec, comp));
+}
+
+/**********************************************************
+The following function is used to get the offset of the next chained record
+on the same page. */
+UNIV_INLINE
+ulint
+rec_get_next_offs(
+/*==============*/
+				/* out: the page offset of the next
+				chained record, or 0 if none */
+	const rec_t*	rec,	/* in: physical record */
+	ulint		comp)	/* in: nonzero=compact page format */
+{
+	ulint	field_value;
+#if REC_NEXT_MASK != 0xFFFFUL
+# error "REC_NEXT_MASK != 0xFFFFUL"
+#endif
+#if REC_NEXT_SHIFT
+# error "REC_NEXT_SHIFT != 0"
+#endif
+
+	field_value = mach_read_from_2(rec - REC_NEXT);
+
+	if (UNIV_EXPECT(comp, REC_OFFS_COMPACT)) {
+#if UNIV_PAGE_SIZE <= 32768
+		/* Note that for 64 KiB pages, field_value can 'wrap around'
+		and the debug assertion is not valid */
+
+		/* In the following assertion, field_value is interpreted
+		as signed 16-bit integer in 2's complement arithmetics.
+		If all platforms defined int16_t in the standard headers,
+		the expression could be written simpler as
+		(int16_t) field_value + ut_align_offset(...) < UNIV_PAGE_SIZE
+		*/
+		ut_ad((field_value >= 32768
+		       ? field_value - 65536
+		       : field_value)
+		      + ut_align_offset(rec, UNIV_PAGE_SIZE)
+		      < UNIV_PAGE_SIZE);
+#endif
+		if (UNIV_UNLIKELY(field_value == 0)) {
+
+			return(0);
+		}
+
+		/* There must be at least REC_N_NEW_EXTRA_BYTES + 1
+		between each record. */
+		ut_ad((field_value > REC_N_NEW_EXTRA_BYTES
+		       && field_value < 32768)
+		      || field_value < (uint16) -REC_N_NEW_EXTRA_BYTES);
+
+		return(ut_align_offset(rec + field_value, UNIV_PAGE_SIZE));
+	} else {
+		ut_ad(field_value < UNIV_PAGE_SIZE);
+
+		return(field_value);
+	}
+}
+
+/**********************************************************
+The following function is used to set the next record offset field
+of an old-style record. */
+UNIV_INLINE
+void
+rec_set_next_offs_old(
+/*==================*/
+	rec_t*	rec,	/* in: old-style physical record */
+	ulint	next)	/* in: offset of the next record */
+{
+	ut_ad(rec);
+	ut_ad(UNIV_PAGE_SIZE > next);
+#if REC_NEXT_MASK != 0xFFFFUL
+# error "REC_NEXT_MASK != 0xFFFFUL"
+#endif
+#if REC_NEXT_SHIFT
+# error "REC_NEXT_SHIFT != 0"
+#endif
+
+	mach_write_to_2(rec - REC_NEXT, next);
+}
+
+/**********************************************************
+The following function is used to set the next record offset field
+of a new-style record. */
+UNIV_INLINE
+void
+rec_set_next_offs_new(
+/*==================*/
+	rec_t*	rec,	/* in/out: new-style physical record */
+	ulint	next)	/* in: offset of the next record */
+{
+	ulint	field_value;
+
+	ut_ad(rec);
+	ut_ad(UNIV_PAGE_SIZE > next);
+
+	if (UNIV_UNLIKELY(!next)) {
+		field_value = 0;
+	} else {
+		/* The following two statements calculate
+		next - offset_of_rec mod 64Ki, where mod is the modulo
+		as a non-negative number */
+
+		field_value = (ulint)
+			((lint) next 
+			 - (lint) ut_align_offset(rec, UNIV_PAGE_SIZE));
+		field_value &= REC_NEXT_MASK;
+	}
+
+	mach_write_to_2(rec - REC_NEXT, field_value);
+}
+
+/**********************************************************
+The following function is used to get the number of fields
+in an old-style record. */
+UNIV_INLINE
+ulint
+rec_get_n_fields_old(
+/*=================*/
+				/* out: number of data fields */
+	const rec_t*	rec)	/* in: physical record */
+{
+	ulint	ret;
+
+	ut_ad(rec);
+
+	ret = rec_get_bit_field_2(rec, REC_OLD_N_FIELDS,
+				  REC_OLD_N_FIELDS_MASK,
+				  REC_OLD_N_FIELDS_SHIFT);
+	ut_ad(ret <= REC_MAX_N_FIELDS);
+	ut_ad(ret > 0);
+
+	return(ret);
+}
+
+/**********************************************************
+The following function is used to set the number of fields
+in an old-style record. */
+UNIV_INLINE
+void
+rec_set_n_fields_old(
+/*=================*/
+	rec_t*	rec,		/* in: physical record */
+	ulint	n_fields)	/* in: the number of fields */
+{
+	ut_ad(rec);
+	ut_ad(n_fields <= REC_MAX_N_FIELDS);
+	ut_ad(n_fields > 0);
+
+	rec_set_bit_field_2(rec, n_fields, REC_OLD_N_FIELDS,
+			    REC_OLD_N_FIELDS_MASK, REC_OLD_N_FIELDS_SHIFT);
+}
+
+/**********************************************************
+The following function retrieves the status bits of a new-style record. */
+UNIV_INLINE
+ulint
+rec_get_status(
+/*===========*/
+				/* out: status bits */
+	const rec_t*	rec)	/* in: physical record */
+{
+	ulint	ret;
+
+	ut_ad(rec);
+
+	ret = rec_get_bit_field_1(rec, REC_NEW_STATUS,
+				  REC_NEW_STATUS_MASK, REC_NEW_STATUS_SHIFT);
+	ut_ad((ret & ~REC_NEW_STATUS_MASK) == 0);
+
+	return(ret);
+}
+
+/**********************************************************
+The following function is used to get the number of fields
+in a record. */
+UNIV_INLINE
+ulint
+rec_get_n_fields(
+/*=============*/
+					/* out: number of data fields */
+	const rec_t*		rec,	/* in: physical record */
+	const dict_index_t*	index)	/* in: record descriptor */
+{
+	ut_ad(rec);
+	ut_ad(index);
+
+	if (!dict_table_is_comp(index->table)) {
+		return(rec_get_n_fields_old(rec));
+	}
+
+	switch (rec_get_status(rec)) {
+	case REC_STATUS_ORDINARY:
+		return(dict_index_get_n_fields(index));
+	case REC_STATUS_NODE_PTR:
+		return(dict_index_get_n_unique_in_tree(index) + 1);
+	case REC_STATUS_INFIMUM:
+	case REC_STATUS_SUPREMUM:
+		return(1);
+	default:
+		ut_error;
+		return(ULINT_UNDEFINED);
+	}
+}
+
+/**********************************************************
+The following function is used to get the number of records owned by the
+previous directory record. */
+UNIV_INLINE
+ulint
+rec_get_n_owned_old(
+/*================*/
+				/* out: number of owned records */
+	const rec_t*	rec)	/* in: old-style physical record */
+{
+	return(rec_get_bit_field_1(rec, REC_OLD_N_OWNED,
+				   REC_N_OWNED_MASK, REC_N_OWNED_SHIFT));
+}
+
+/**********************************************************
+The following function is used to set the number of owned records. */
+UNIV_INLINE
+void
+rec_set_n_owned_old(
+/*================*/
+				/* out: TRUE on success */
+	rec_t*	rec,		/* in: old-style physical record */
+	ulint	n_owned)	/* in: the number of owned */
+{
+	rec_set_bit_field_1(rec, n_owned, REC_OLD_N_OWNED,
+			    REC_N_OWNED_MASK, REC_N_OWNED_SHIFT);
+}
+
+/**********************************************************
+The following function is used to get the number of records owned by the
+previous directory record. */
+UNIV_INLINE
+ulint
+rec_get_n_owned_new(
+/*================*/
+				/* out: number of owned records */
+	const rec_t*	rec)	/* in: new-style physical record */
+{
+	return(rec_get_bit_field_1(rec, REC_NEW_N_OWNED,
+				   REC_N_OWNED_MASK, REC_N_OWNED_SHIFT));
+}
+
+/**********************************************************
+The following function is used to set the number of owned records. */
+UNIV_INLINE
+void
+rec_set_n_owned_new(
+/*================*/
+	rec_t*		rec,	/* in/out: new-style physical record */
+	page_zip_des_t*	page_zip,/* in/out: compressed page, or NULL */
+	ulint		n_owned)/* in: the number of owned */
+{
+	rec_set_bit_field_1(rec, n_owned, REC_NEW_N_OWNED,
+			    REC_N_OWNED_MASK, REC_N_OWNED_SHIFT);
+	if (UNIV_LIKELY_NULL(page_zip)
+	    && UNIV_LIKELY(rec_get_status(rec)
+			   != REC_STATUS_SUPREMUM)) {
+		page_zip_rec_set_owned(page_zip, rec, n_owned);
+	}
+}
+
+/**********************************************************
+The following function is used to retrieve the info bits of a record. */
+UNIV_INLINE
+ulint
+rec_get_info_bits(
+/*==============*/
+				/* out: info bits */
+	const rec_t*	rec,	/* in: physical record */
+	ulint		comp)	/* in: nonzero=compact page format */
+{
+	return(rec_get_bit_field_1(
+		       rec, comp ? REC_NEW_INFO_BITS : REC_OLD_INFO_BITS,
+		       REC_INFO_BITS_MASK, REC_INFO_BITS_SHIFT));
+}
+
+/**********************************************************
+The following function is used to set the info bits of a record. */
+UNIV_INLINE
+void
+rec_set_info_bits_old(
+/*==================*/
+	rec_t*	rec,	/* in: old-style physical record */
+	ulint	bits)	/* in: info bits */
+{
+	rec_set_bit_field_1(rec, bits, REC_OLD_INFO_BITS,
+			    REC_INFO_BITS_MASK, REC_INFO_BITS_SHIFT);
+}
+/**********************************************************
+The following function is used to set the info bits of a record. */
+UNIV_INLINE
+void
+rec_set_info_bits_new(
+/*==================*/
+	rec_t*	rec,	/* in/out: new-style physical record */
+	ulint	bits)	/* in: info bits */
+{
+	rec_set_bit_field_1(rec, bits, REC_NEW_INFO_BITS,
+			    REC_INFO_BITS_MASK, REC_INFO_BITS_SHIFT);
+}
+
+/**********************************************************
+The following function is used to set the status bits of a new-style record. */
+UNIV_INLINE
+void
+rec_set_status(
+/*===========*/
+	rec_t*	rec,	/* in/out: physical record */
+	ulint	bits)	/* in: info bits */
+{
+	rec_set_bit_field_1(rec, bits, REC_NEW_STATUS,
+			    REC_NEW_STATUS_MASK, REC_NEW_STATUS_SHIFT);
+}
+
+/**********************************************************
+The following function is used to retrieve the info and status
+bits of a record.  (Only compact records have status bits.) */
+UNIV_INLINE
+ulint
+rec_get_info_and_status_bits(
+/*=========================*/
+				/* out: info bits */
+	const rec_t*	rec,	/* in: physical record */
+	ulint		comp)	/* in: nonzero=compact page format */
+{
+	ulint	bits;
+#if (REC_NEW_STATUS_MASK >> REC_NEW_STATUS_SHIFT) \
+& (REC_INFO_BITS_MASK >> REC_INFO_BITS_SHIFT)
+# error "REC_NEW_STATUS_MASK and REC_INFO_BITS_MASK overlap"
+#endif
+	if (UNIV_EXPECT(comp, REC_OFFS_COMPACT)) {
+		bits = rec_get_info_bits(rec, TRUE) | rec_get_status(rec);
+	} else {
+		bits = rec_get_info_bits(rec, FALSE);
+		ut_ad(!(bits & ~(REC_INFO_BITS_MASK >> REC_INFO_BITS_SHIFT)));
+	}
+	return(bits);
+}
+/**********************************************************
+The following function is used to set the info and status
+bits of a record.  (Only compact records have status bits.) */
+UNIV_INLINE
+void
+rec_set_info_and_status_bits(
+/*=========================*/
+	rec_t*	rec,	/* in/out: physical record */
+	ulint	bits)	/* in: info bits */
+{
+#if (REC_NEW_STATUS_MASK >> REC_NEW_STATUS_SHIFT) \
+& (REC_INFO_BITS_MASK >> REC_INFO_BITS_SHIFT)
+# error "REC_NEW_STATUS_MASK and REC_INFO_BITS_MASK overlap"
+#endif
+	rec_set_status(rec, bits & REC_NEW_STATUS_MASK);
+	rec_set_info_bits_new(rec, bits & ~REC_NEW_STATUS_MASK);
+}
+
+/**********************************************************
+The following function tells if record is delete marked. */
+UNIV_INLINE
+ulint
+rec_get_deleted_flag(
+/*=================*/
+				/* out: nonzero if delete marked */
+	const rec_t*	rec,	/* in: physical record */
+	ulint		comp)	/* in: nonzero=compact page format */
+{
+	if (UNIV_EXPECT(comp, REC_OFFS_COMPACT)) {
+		return(UNIV_UNLIKELY(
+			       rec_get_bit_field_1(rec, REC_NEW_INFO_BITS,
+						   REC_INFO_DELETED_FLAG,
+						   REC_INFO_BITS_SHIFT)));
+	} else {
+		return(UNIV_UNLIKELY(
+			       rec_get_bit_field_1(rec, REC_OLD_INFO_BITS,
+						   REC_INFO_DELETED_FLAG,
+						   REC_INFO_BITS_SHIFT)));
+	}
+}
+
+/**********************************************************
+The following function is used to set the deleted bit. */
+UNIV_INLINE
+void
+rec_set_deleted_flag_old(
+/*=====================*/
+	rec_t*	rec,	/* in: old-style physical record */
+	ulint	flag)	/* in: nonzero if delete marked */
+{
+	ulint	val;
+
+	val = rec_get_info_bits(rec, FALSE);
+
+	if (flag) {
+		val |= REC_INFO_DELETED_FLAG;
+	} else {
+		val &= ~REC_INFO_DELETED_FLAG;
+	}
+
+	rec_set_info_bits_old(rec, val);
+}
+
+/**********************************************************
+The following function is used to set the deleted bit. */
+UNIV_INLINE
+void
+rec_set_deleted_flag_new(
+/*=====================*/
+	rec_t*		rec,	/* in/out: new-style physical record */
+	page_zip_des_t*	page_zip,/* in/out: compressed page, or NULL */
+	ulint		flag)	/* in: nonzero if delete marked */
+{
+	ulint	val;
+
+	val = rec_get_info_bits(rec, TRUE);
+
+	if (flag) {
+		val |= REC_INFO_DELETED_FLAG;
+	} else {
+		val &= ~REC_INFO_DELETED_FLAG;
+	}
+
+	rec_set_info_bits_new(rec, val);
+
+	if (UNIV_LIKELY_NULL(page_zip)) {
+		page_zip_rec_set_deleted(page_zip, rec, flag);
+	}
+}
+
+/**********************************************************
+The following function tells if a new-style record is a node pointer. */
+UNIV_INLINE
+ibool
+rec_get_node_ptr_flag(
+/*==================*/
+				/* out: TRUE if node pointer */
+	const rec_t*	rec)	/* in: physical record */
+{
+	return(REC_STATUS_NODE_PTR == rec_get_status(rec));
+}
+
+/**********************************************************
+The following function is used to get the order number
+of an old-style record in the heap of the index page. */
+UNIV_INLINE
+ulint
+rec_get_heap_no_old(
+/*================*/
+				/* out: heap order number */
+	const rec_t*	rec)	/* in: physical record */
+{
+	return(rec_get_bit_field_2(rec, REC_OLD_HEAP_NO,
+				   REC_HEAP_NO_MASK, REC_HEAP_NO_SHIFT));
+}
+
+/**********************************************************
+The following function is used to set the heap number
+field in an old-style record. */
+UNIV_INLINE
+void
+rec_set_heap_no_old(
+/*================*/
+	rec_t*	rec,	/* in: physical record */
+	ulint	heap_no)/* in: the heap number */
+{
+	rec_set_bit_field_2(rec, heap_no, REC_OLD_HEAP_NO,
+			    REC_HEAP_NO_MASK, REC_HEAP_NO_SHIFT);
+}
+
+/**********************************************************
+The following function is used to get the order number
+of a new-style record in the heap of the index page. */
+UNIV_INLINE
+ulint
+rec_get_heap_no_new(
+/*================*/
+				/* out: heap order number */
+	const rec_t*	rec)	/* in: physical record */
+{
+	return(rec_get_bit_field_2(rec, REC_NEW_HEAP_NO,
+				   REC_HEAP_NO_MASK, REC_HEAP_NO_SHIFT));
+}
+
+/**********************************************************
+The following function is used to set the heap number
+field in a new-style record. */
+UNIV_INLINE
+void
+rec_set_heap_no_new(
+/*================*/
+	rec_t*	rec,	/* in/out: physical record */
+	ulint	heap_no)/* in: the heap number */
+{
+	rec_set_bit_field_2(rec, heap_no, REC_NEW_HEAP_NO,
+			    REC_HEAP_NO_MASK, REC_HEAP_NO_SHIFT);
+}
+
+/**********************************************************
+The following function is used to test whether the data offsets in the record
+are stored in one-byte or two-byte format. */
+UNIV_INLINE
+ibool
+rec_get_1byte_offs_flag(
+/*====================*/
+				/* out: TRUE if 1-byte form */
+	const rec_t*	rec)	/* in: physical record */
+{
+#if TRUE != 1
+#error "TRUE != 1"
+#endif
+
+	return(rec_get_bit_field_1(rec, REC_OLD_SHORT, REC_OLD_SHORT_MASK,
+				   REC_OLD_SHORT_SHIFT));
+}
+
+/**********************************************************
+The following function is used to set the 1-byte offsets flag. */
+UNIV_INLINE
+void
+rec_set_1byte_offs_flag(
+/*====================*/
+	rec_t*	rec,	/* in: physical record */
+	ibool	flag)	/* in: TRUE if 1byte form */
+{
+#if TRUE != 1
+#error "TRUE != 1"
+#endif
+	ut_ad(flag <= TRUE);
+
+	rec_set_bit_field_1(rec, flag, REC_OLD_SHORT, REC_OLD_SHORT_MASK,
+			    REC_OLD_SHORT_SHIFT);
+}
+
+/**********************************************************
+Returns the offset of nth field end if the record is stored in the 1-byte
+offsets form. If the field is SQL null, the flag is ORed in the returned
+value. */
+UNIV_INLINE
+ulint
+rec_1_get_field_end_info(
+/*=====================*/
+				/* out: offset of the start of the
+				field, SQL null flag ORed */
+	const rec_t*	rec,	/* in: record */
+	ulint		n)	/* in: field index */
+{
+	ut_ad(rec_get_1byte_offs_flag(rec));
+	ut_ad(n < rec_get_n_fields_old(rec));
+
+	return(mach_read_from_1(rec - (REC_N_OLD_EXTRA_BYTES + n + 1)));
+}
+
+/**********************************************************
+Returns the offset of nth field end if the record is stored in the 2-byte
+offsets form. If the field is SQL null, the flag is ORed in the returned
+value. */
+UNIV_INLINE
+ulint
+rec_2_get_field_end_info(
+/*=====================*/
+				/* out: offset of the start of the
+				field, SQL null flag and extern
+				storage flag ORed */
+	const rec_t*	rec,	/* in: record */
+	ulint		n)	/* in: field index */
+{
+	ut_ad(!rec_get_1byte_offs_flag(rec));
+	ut_ad(n < rec_get_n_fields_old(rec));
+
+	return(mach_read_from_2(rec - (REC_N_OLD_EXTRA_BYTES + 2 * n + 2)));
+}
+
+/* Get the base address of offsets.  The extra_size is stored at
+this position, and following positions hold the end offsets of
+the fields. */
+#define rec_offs_base(offsets) (offsets + REC_OFFS_HEADER_SIZE)
+
+/**************************************************************
+The following function returns the number of allocated elements
+for an array of offsets. */
+UNIV_INLINE
+ulint
+rec_offs_get_n_alloc(
+/*=================*/
+				/* out: number of elements */
+	const ulint*	offsets)/* in: array for rec_get_offsets() */
+{
+	ulint	n_alloc;
+	ut_ad(offsets);
+	n_alloc = offsets[0];
+	ut_ad(n_alloc > REC_OFFS_HEADER_SIZE);
+	UNIV_MEM_ASSERT_W(offsets, n_alloc * sizeof *offsets);
+	return(n_alloc);
+}
+
+/**************************************************************
+The following function sets the number of allocated elements
+for an array of offsets. */
+UNIV_INLINE
+void
+rec_offs_set_n_alloc(
+/*=================*/
+	ulint*	offsets,	/* out: array for rec_get_offsets(),
+				must be allocated */
+	ulint	n_alloc)	/* in: number of elements */
+{
+	ut_ad(offsets);
+	ut_ad(n_alloc > REC_OFFS_HEADER_SIZE);
+	UNIV_MEM_ASSERT_AND_ALLOC(offsets, n_alloc * sizeof *offsets);
+	offsets[0] = n_alloc;
+}
+
+/**************************************************************
+The following function returns the number of fields in a record. */
+UNIV_INLINE
+ulint
+rec_offs_n_fields(
+/*==============*/
+				/* out: number of fields */
+	const ulint*	offsets)/* in: array returned by rec_get_offsets() */
+{
+	ulint	n_fields;
+	ut_ad(offsets);
+	n_fields = offsets[1];
+	ut_ad(n_fields > 0);
+	ut_ad(n_fields <= REC_MAX_N_FIELDS);
+	ut_ad(n_fields + REC_OFFS_HEADER_SIZE
+	      <= rec_offs_get_n_alloc(offsets));
+	return(n_fields);
+}
+
+/****************************************************************
+Validates offsets returned by rec_get_offsets(). */
+UNIV_INLINE
+ibool
+rec_offs_validate(
+/*==============*/
+					/* out: TRUE if valid */
+	const rec_t*		rec,	/* in: record or NULL */
+	const dict_index_t*	index,	/* in: record descriptor or NULL */
+	const ulint*		offsets)/* in: array returned by
+					rec_get_offsets() */
+{
+	ulint	i	= rec_offs_n_fields(offsets);
+	ulint	last	= ULINT_MAX;
+	ulint	comp	= *rec_offs_base(offsets) & REC_OFFS_COMPACT;
+
+	if (rec) {
+		ut_ad((ulint) rec == offsets[2]);
+		if (!comp) {
+			ut_a(rec_get_n_fields_old(rec) >= i);
+		}
+	}
+	if (index) {
+		ulint max_n_fields;
+		ut_ad((ulint) index == offsets[3]);
+		max_n_fields = ut_max(
+			dict_index_get_n_fields(index),
+			dict_index_get_n_unique_in_tree(index) + 1);
+		if (comp && rec) {
+			switch (rec_get_status(rec)) {
+			case REC_STATUS_ORDINARY:
+				break;
+			case REC_STATUS_NODE_PTR:
+				max_n_fields = dict_index_get_n_unique_in_tree(
+					index) + 1;
+				break;
+			case REC_STATUS_INFIMUM:
+			case REC_STATUS_SUPREMUM:
+				max_n_fields = 1;
+				break;
+			default:
+				ut_error;
+			}
+		}
+		/* index->n_def == 0 for dummy indexes if !comp */
+		ut_a(!comp || index->n_def);
+		ut_a(!index->n_def || i <= max_n_fields);
+	}
+	while (i--) {
+		ulint	curr = rec_offs_base(offsets)[1 + i] & REC_OFFS_MASK;
+		ut_a(curr <= last);
+		last = curr;
+	}
+	return(TRUE);
+}
+#ifdef UNIV_DEBUG
+/****************************************************************
+Updates debug data in offsets, in order to avoid bogus
+rec_offs_validate() failures. */
+UNIV_INLINE
+void
+rec_offs_make_valid(
+/*================*/
+	const rec_t*		rec,	/* in: record */
+	const dict_index_t*	index,	/* in: record descriptor */
+	ulint*			offsets)/* in: array returned by
+					rec_get_offsets() */
+{
+	ut_ad(rec);
+	ut_ad(index);
+	ut_ad(offsets);
+	ut_ad(rec_get_n_fields(rec, index) >= rec_offs_n_fields(offsets));
+	offsets[2] = (ulint) rec;
+	offsets[3] = (ulint) index;
+}
+#endif /* UNIV_DEBUG */
+
+/****************************************************************
+The following function is used to get an offset to the nth
+data field in a record. */
+UNIV_INLINE
+ulint
+rec_get_nth_field_offs(
+/*===================*/
+				/* out: offset from the origin of rec */
+	const ulint*	offsets,/* in: array returned by rec_get_offsets() */
+	ulint		n,	/* in: index of the field */
+	ulint*		len)	/* out: length of the field; UNIV_SQL_NULL
+				if SQL null */
+{
+	ulint	offs;
+	ulint	length;
+	ut_ad(n < rec_offs_n_fields(offsets));
+	ut_ad(len);
+
+	if (UNIV_UNLIKELY(n == 0)) {
+		offs = 0;
+	} else {
+		offs = rec_offs_base(offsets)[n] & REC_OFFS_MASK;
+	}
+
+	length = rec_offs_base(offsets)[1 + n];
+
+	if (length & REC_OFFS_SQL_NULL) {
+		length = UNIV_SQL_NULL;
+	} else {
+		length &= REC_OFFS_MASK;
+		length -= offs;
+	}
+
+	*len = length;
+	return(offs);
+}
+
+/**********************************************************
+Determine if the offsets are for a record in the new
+compact format. */
+UNIV_INLINE
+ulint
+rec_offs_comp(
+/*==========*/
+				/* out: nonzero if compact format */
+	const ulint*	offsets)/* in: array returned by rec_get_offsets() */
+{
+	ut_ad(rec_offs_validate(NULL, NULL, offsets));
+	return(*rec_offs_base(offsets) & REC_OFFS_COMPACT);
+}
+
+/**********************************************************
+Determine if the offsets are for a record containing
+externally stored columns. */
+UNIV_INLINE
+ulint
+rec_offs_any_extern(
+/*================*/
+				/* out: nonzero if externally stored */
+	const ulint*	offsets)/* in: array returned by rec_get_offsets() */
+{
+	ut_ad(rec_offs_validate(NULL, NULL, offsets));
+	return(UNIV_UNLIKELY(*rec_offs_base(offsets) & REC_OFFS_EXTERNAL));
+}
+
+/**********************************************************
+Returns nonzero if the extern bit is set in nth field of rec. */
+UNIV_INLINE
+ulint
+rec_offs_nth_extern(
+/*================*/
+				/* out: nonzero if externally stored */
+	const ulint*	offsets,/* in: array returned by rec_get_offsets() */
+	ulint		n)	/* in: nth field */
+{
+	ut_ad(rec_offs_validate(NULL, NULL, offsets));
+	ut_ad(n < rec_offs_n_fields(offsets));
+	return(UNIV_UNLIKELY(rec_offs_base(offsets)[1 + n]
+			     & REC_OFFS_EXTERNAL));
+}
+
+/**********************************************************
+Returns nonzero if the SQL NULL bit is set in nth field of rec. */
+UNIV_INLINE
+ulint
+rec_offs_nth_sql_null(
+/*==================*/
+				/* out: nonzero if SQL NULL */
+	const ulint*	offsets,/* in: array returned by rec_get_offsets() */
+	ulint		n)	/* in: nth field */
+{
+	ut_ad(rec_offs_validate(NULL, NULL, offsets));
+	ut_ad(n < rec_offs_n_fields(offsets));
+	return(UNIV_UNLIKELY(rec_offs_base(offsets)[1 + n]
+			     & REC_OFFS_SQL_NULL));
+}
+
+/**********************************************************
+Gets the physical size of a field. */
+UNIV_INLINE
+ulint
+rec_offs_nth_size(
+/*==============*/
+				/* out: length of field */
+	const ulint*	offsets,/* in: array returned by rec_get_offsets() */
+	ulint		n)	/* in: nth field */
+{
+	ut_ad(rec_offs_validate(NULL, NULL, offsets));
+	ut_ad(n < rec_offs_n_fields(offsets));
+	if (!n) {
+		return(rec_offs_base(offsets)[1 + n] & REC_OFFS_MASK);
+	}
+	return((rec_offs_base(offsets)[1 + n] - rec_offs_base(offsets)[n])
+	       & REC_OFFS_MASK);
+}
+
+/**********************************************************
+Returns the number of extern bits set in a record. */
+UNIV_INLINE
+ulint
+rec_offs_n_extern(
+/*==============*/
+				/* out: number of externally stored fields */
+	const ulint*	offsets)/* in: array returned by rec_get_offsets() */
+{
+	ulint	n = 0;
+
+	if (rec_offs_any_extern(offsets)) {
+		ulint	i;
+
+		for (i = rec_offs_n_fields(offsets); i--; ) {
+			if (rec_offs_nth_extern(offsets, i)) {
+				n++;
+			}
+		}
+	}
+
+	return(n);
+}
+
+/**********************************************************
+Returns the offset of n - 1th field end if the record is stored in the 1-byte
+offsets form. If the field is SQL null, the flag is ORed in the returned
+value. This function and the 2-byte counterpart are defined here because the
+C-compiler was not able to sum negative and positive constant offsets, and
+warned of constant arithmetic overflow within the compiler. */
+UNIV_INLINE
+ulint
+rec_1_get_prev_field_end_info(
+/*==========================*/
+				/* out: offset of the start of the
+				PREVIOUS field, SQL null flag ORed */
+	const rec_t*	rec,	/* in: record */
+	ulint		n)	/* in: field index */
+{
+	ut_ad(rec_get_1byte_offs_flag(rec));
+	ut_ad(n <= rec_get_n_fields_old(rec));
+
+	return(mach_read_from_1(rec - (REC_N_OLD_EXTRA_BYTES + n)));
+}
+
+/**********************************************************
+Returns the offset of n - 1th field end if the record is stored in the 2-byte
+offsets form. If the field is SQL null, the flag is ORed in the returned
+value. */
+UNIV_INLINE
+ulint
+rec_2_get_prev_field_end_info(
+/*==========================*/
+				/* out: offset of the start of the
+				PREVIOUS field, SQL null flag ORed */
+	const rec_t*	rec,	/* in: record */
+	ulint		n)	/* in: field index */
+{
+	ut_ad(!rec_get_1byte_offs_flag(rec));
+	ut_ad(n <= rec_get_n_fields_old(rec));
+
+	return(mach_read_from_2(rec - (REC_N_OLD_EXTRA_BYTES + 2 * n)));
+}
+
+/**********************************************************
+Sets the field end info for the nth field if the record is stored in the
+1-byte format. */
+UNIV_INLINE
+void
+rec_1_set_field_end_info(
+/*=====================*/
+	rec_t*	rec,	/* in: record */
+	ulint	n,	/* in: field index */
+	ulint	info)	/* in: value to set */
+{
+	ut_ad(rec_get_1byte_offs_flag(rec));
+	ut_ad(n < rec_get_n_fields_old(rec));
+
+	mach_write_to_1(rec - (REC_N_OLD_EXTRA_BYTES + n + 1), info);
+}
+
+/**********************************************************
+Sets the field end info for the nth field if the record is stored in the
+2-byte format. */
+UNIV_INLINE
+void
+rec_2_set_field_end_info(
+/*=====================*/
+	rec_t*	rec,	/* in: record */
+	ulint	n,	/* in: field index */
+	ulint	info)	/* in: value to set */
+{
+	ut_ad(!rec_get_1byte_offs_flag(rec));
+	ut_ad(n < rec_get_n_fields_old(rec));
+
+	mach_write_to_2(rec - (REC_N_OLD_EXTRA_BYTES + 2 * n + 2), info);
+}
+
+/**********************************************************
+Returns the offset of nth field start if the record is stored in the 1-byte
+offsets form. */
+UNIV_INLINE
+ulint
+rec_1_get_field_start_offs(
+/*=======================*/
+				/* out: offset of the start of the field */
+	const rec_t*	rec,	/* in: record */
+	ulint		n)	/* in: field index */
+{
+	ut_ad(rec_get_1byte_offs_flag(rec));
+	ut_ad(n <= rec_get_n_fields_old(rec));
+
+	if (n == 0) {
+
+		return(0);
+	}
+
+	return(rec_1_get_prev_field_end_info(rec, n)
+	       & ~REC_1BYTE_SQL_NULL_MASK);
+}
+
+/**********************************************************
+Returns the offset of nth field start if the record is stored in the 2-byte
+offsets form. */
+UNIV_INLINE
+ulint
+rec_2_get_field_start_offs(
+/*=======================*/
+				/* out: offset of the start of the field */
+	const rec_t*	rec,	/* in: record */
+	ulint		n)	/* in: field index */
+{
+	ut_ad(!rec_get_1byte_offs_flag(rec));
+	ut_ad(n <= rec_get_n_fields_old(rec));
+
+	if (n == 0) {
+
+		return(0);
+	}
+
+	return(rec_2_get_prev_field_end_info(rec, n)
+	       & ~(REC_2BYTE_SQL_NULL_MASK | REC_2BYTE_EXTERN_MASK));
+}
+
+/**********************************************************
+The following function is used to read the offset of the start of a data field
+in the record. The start of an SQL null field is the end offset of the
+previous non-null field, or 0, if none exists. If n is the number of the last
+field + 1, then the end offset of the last field is returned. */
+UNIV_INLINE
+ulint
+rec_get_field_start_offs(
+/*=====================*/
+				/* out: offset of the start of the field */
+	const rec_t*	rec,	/* in: record */
+	ulint		n)	/* in: field index */
+{
+	ut_ad(rec);
+	ut_ad(n <= rec_get_n_fields_old(rec));
+
+	if (n == 0) {
+
+		return(0);
+	}
+
+	if (rec_get_1byte_offs_flag(rec)) {
+
+		return(rec_1_get_field_start_offs(rec, n));
+	}
+
+	return(rec_2_get_field_start_offs(rec, n));
+}
+
+/****************************************************************
+Gets the physical size of an old-style field.
+Also an SQL null may have a field of size > 0,
+if the data type is of a fixed size. */
+UNIV_INLINE
+ulint
+rec_get_nth_field_size(
+/*===================*/
+				/* out: field size in bytes */
+	const rec_t*	rec,	/* in: record */
+	ulint		n)	/* in: index of the field */
+{
+	ulint	os;
+	ulint	next_os;
+
+	os = rec_get_field_start_offs(rec, n);
+	next_os = rec_get_field_start_offs(rec, n + 1);
+
+	ut_ad(next_os - os < UNIV_PAGE_SIZE);
+
+	return(next_os - os);
+}
+
+/***************************************************************
+This is used to modify the value of an already existing field in a record.
+The previous value must have exactly the same size as the new value. If len
+is UNIV_SQL_NULL then the field is treated as an SQL null.
+For records in ROW_FORMAT=COMPACT (new-style records), len must not be
+UNIV_SQL_NULL unless the field already is SQL null. */
+UNIV_INLINE
+void
+rec_set_nth_field(
+/*==============*/
+	rec_t*		rec,	/* in: record */
+	const ulint*	offsets,/* in: array returned by rec_get_offsets() */
+	ulint		n,	/* in: index number of the field */
+	const void*	data,	/* in: pointer to the data
+				if not SQL null */
+	ulint		len)	/* in: length of the data or UNIV_SQL_NULL */
+{
+	byte*	data2;
+	ulint	len2;
+
+	ut_ad(rec);
+	ut_ad(rec_offs_validate(rec, NULL, offsets));
+
+	if (UNIV_UNLIKELY(len == UNIV_SQL_NULL)) {
+		if (!rec_offs_nth_sql_null(offsets, n)) {
+			ut_a(!rec_offs_comp(offsets));
+			rec_set_nth_field_sql_null(rec, n);
+		}
+
+		return;
+	}
+
+	data2 = rec_get_nth_field(rec, offsets, n, &len2);
+	if (len2 == UNIV_SQL_NULL) {
+		ut_ad(!rec_offs_comp(offsets));
+		rec_set_nth_field_null_bit(rec, n, FALSE);
+		ut_ad(len == rec_get_nth_field_size(rec, n));
+	} else {
+		ut_ad(len2 == len);
+	}
+
+	ut_memcpy(data2, data, len);
+}
+
+/**************************************************************
+The following function returns the data size of an old-style physical
+record, that is the sum of field lengths. SQL null fields
+are counted as length 0 fields. The value returned by the function
+is the distance from record origin to record end in bytes. */
+UNIV_INLINE
+ulint
+rec_get_data_size_old(
+/*==================*/
+				/* out: size */
+	const rec_t*	rec)	/* in: physical record */
+{
+	ut_ad(rec);
+
+	return(rec_get_field_start_offs(rec, rec_get_n_fields_old(rec)));
+}
+
+/**************************************************************
+The following function sets the number of fields in offsets. */
+UNIV_INLINE
+void
+rec_offs_set_n_fields(
+/*==================*/
+	ulint*	offsets,	/* in/out: array returned by
+				rec_get_offsets() */
+	ulint	n_fields)	/* in: number of fields */
+{
+	ut_ad(offsets);
+	ut_ad(n_fields > 0);
+	ut_ad(n_fields <= REC_MAX_N_FIELDS);
+	ut_ad(n_fields + REC_OFFS_HEADER_SIZE
+	      <= rec_offs_get_n_alloc(offsets));
+	offsets[1] = n_fields;
+}
+
+/**************************************************************
+The following function returns the data size of a physical
+record, that is the sum of field lengths. SQL null fields
+are counted as length 0 fields. The value returned by the function
+is the distance from record origin to record end in bytes. */
+UNIV_INLINE
+ulint
+rec_offs_data_size(
+/*===============*/
+				/* out: size */
+	const ulint*	offsets)/* in: array returned by rec_get_offsets() */
+{
+	ulint	size;
+
+	ut_ad(rec_offs_validate(NULL, NULL, offsets));
+	size = rec_offs_base(offsets)[rec_offs_n_fields(offsets)]
+		& REC_OFFS_MASK;
+	ut_ad(size < UNIV_PAGE_SIZE);
+	return(size);
+}
+
+/**************************************************************
+Returns the total size of record minus data size of record. The value
+returned by the function is the distance from record start to record origin
+in bytes. */
+UNIV_INLINE
+ulint
+rec_offs_extra_size(
+/*================*/
+				/* out: size */
+	const ulint*	offsets)/* in: array returned by rec_get_offsets() */
+{
+	ulint	size;
+	ut_ad(rec_offs_validate(NULL, NULL, offsets));
+	size = *rec_offs_base(offsets) & ~(REC_OFFS_COMPACT | REC_OFFS_EXTERNAL);
+	ut_ad(size < UNIV_PAGE_SIZE);
+	return(size);
+}
+
+/**************************************************************
+Returns the total size of a physical record.  */
+UNIV_INLINE
+ulint
+rec_offs_size(
+/*==========*/
+				/* out: size */
+	const ulint*	offsets)/* in: array returned by rec_get_offsets() */
+{
+	return(rec_offs_data_size(offsets) + rec_offs_extra_size(offsets));
+}
+
+/**************************************************************
+Returns a pointer to the end of the record. */
+UNIV_INLINE
+byte*
+rec_get_end(
+/*========*/
+				/* out: pointer to end */
+	rec_t*		rec,	/* in: pointer to record */
+	const ulint*	offsets)/* in: array returned by rec_get_offsets() */
+{
+	ut_ad(rec_offs_validate(rec, NULL, offsets));
+	return(rec + rec_offs_data_size(offsets));
+}
+
+/**************************************************************
+Returns a pointer to the start of the record. */
+UNIV_INLINE
+byte*
+rec_get_start(
+/*==========*/
+				/* out: pointer to start */
+	rec_t*		rec,	/* in: pointer to record */
+	const ulint*	offsets)/* in: array returned by rec_get_offsets() */
+{
+	ut_ad(rec_offs_validate(rec, NULL, offsets));
+	return(rec - rec_offs_extra_size(offsets));
+}
+
+/*******************************************************************
+Copies a physical record to a buffer. */
+UNIV_INLINE
+rec_t*
+rec_copy(
+/*=====*/
+				/* out: pointer to the origin of the copy */
+	void*		buf,	/* in: buffer */
+	const rec_t*	rec,	/* in: physical record */
+	const ulint*	offsets)/* in: array returned by rec_get_offsets() */
+{
+	ulint	extra_len;
+	ulint	data_len;
+
+	ut_ad(rec && buf);
+	ut_ad(rec_offs_validate((rec_t*) rec, NULL, offsets));
+	ut_ad(rec_validate(rec, offsets));
+
+	extra_len = rec_offs_extra_size(offsets);
+	data_len = rec_offs_data_size(offsets);
+
+	ut_memcpy(buf, rec - extra_len, extra_len + data_len);
+
+	return((byte*)buf + extra_len);
+}
+
+/**************************************************************
+Returns the extra size of an old-style physical record if we know its
+data size and number of fields. */
+UNIV_INLINE
+ulint
+rec_get_converted_extra_size(
+/*=========================*/
+				/* out: extra size */
+	ulint	data_size,	/* in: data size */
+	ulint	n_fields,	/* in: number of fields */
+	ulint	n_ext)		/* in: number of externally stored columns */
+{
+	if (!n_ext && data_size <= REC_1BYTE_OFFS_LIMIT) {
+
+		return(REC_N_OLD_EXTRA_BYTES + n_fields);
+	}
+
+	return(REC_N_OLD_EXTRA_BYTES + 2 * n_fields);
+}
+
+/**************************************************************
+The following function returns the size of a data tuple when converted to
+a physical record. */
+UNIV_INLINE
+ulint
+rec_get_converted_size(
+/*===================*/
+				/* out: size */
+	dict_index_t*	index,	/* in: record descriptor */
+	const dtuple_t*	dtuple,	/* in: data tuple */
+	ulint		n_ext)	/* in: number of externally stored columns */
+{
+	ulint	data_size;
+	ulint	extra_size;
+
+	ut_ad(index);
+	ut_ad(dtuple);
+	ut_ad(dtuple_check_typed(dtuple));
+
+	ut_ad(index->type & DICT_UNIVERSAL
+	      || dtuple_get_n_fields(dtuple)
+	      == (((dtuple_get_info_bits(dtuple) & REC_NEW_STATUS_MASK)
+		   == REC_STATUS_NODE_PTR)
+		  ? dict_index_get_n_unique_in_tree(index) + 1
+		  : dict_index_get_n_fields(index)));
+
+	if (dict_table_is_comp(index->table)) {
+		return(rec_get_converted_size_comp(index,
+						   dtuple_get_info_bits(dtuple)
+						   & REC_NEW_STATUS_MASK,
+						   dtuple->fields,
+						   dtuple->n_fields, NULL));
+	}
+
+	data_size = dtuple_get_data_size(dtuple);
+
+	extra_size = rec_get_converted_extra_size(
+		data_size, dtuple_get_n_fields(dtuple), n_ext);
+
+	return(data_size + extra_size);
+}
+
+/****************************************************************
+Folds a prefix of a physical record to a ulint. Folds only existing fields,
+that is, checks that we do not run out of the record. */
+UNIV_INLINE
+ulint
+rec_fold(
+/*=====*/
+					/* out: the folded value */
+	const rec_t*	rec,		/* in: the physical record */
+	const ulint*	offsets,	/* in: array returned by
+					rec_get_offsets() */
+	ulint		n_fields,	/* in: number of complete
+					fields to fold */
+	ulint		n_bytes,	/* in: number of bytes to fold
+					in an incomplete last field */
+	dulint		tree_id)	/* in: index tree id */
+{
+	ulint		i;
+	const byte*	data;
+	ulint		len;
+	ulint		fold;
+	ulint		n_fields_rec;
+
+	ut_ad(rec_offs_validate(rec, NULL, offsets));
+	ut_ad(rec_validate(rec, offsets));
+	ut_ad(n_fields + n_bytes > 0);
+
+	n_fields_rec = rec_offs_n_fields(offsets);
+	ut_ad(n_fields <= n_fields_rec);
+	ut_ad(n_fields < n_fields_rec || n_bytes == 0);
+
+	if (n_fields > n_fields_rec) {
+		n_fields = n_fields_rec;
+	}
+
+	if (n_fields == n_fields_rec) {
+		n_bytes = 0;
+	}
+
+	fold = ut_fold_dulint(tree_id);
+
+	for (i = 0; i < n_fields; i++) {
+		data = rec_get_nth_field(rec, offsets, i, &len);
+
+		if (len != UNIV_SQL_NULL) {
+			fold = ut_fold_ulint_pair(fold,
+						  ut_fold_binary(data, len));
+		}
+	}
+
+	if (n_bytes > 0) {
+		data = rec_get_nth_field(rec, offsets, i, &len);
+
+		if (len != UNIV_SQL_NULL) {
+			if (len > n_bytes) {
+				len = n_bytes;
+			}
+
+			fold = ut_fold_ulint_pair(fold,
+						  ut_fold_binary(data, len));
+		}
+	}
+
+	return(fold);
+}
diff --git a/storage/xtradb/include/rem0types.h b/storage/xtradb/include/rem0types.h
new file mode 100644
index 00000000000..d0b11b92495
--- /dev/null
+++ b/storage/xtradb/include/rem0types.h
@@ -0,0 +1,45 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/************************************************************************
+Record manager global types
+
+Created 5/30/1994 Heikki Tuuri
+*************************************************************************/
+
+#ifndef rem0types_h
+#define rem0types_h
+
+/* We define the physical record simply as an array of bytes */
+typedef byte	rec_t;
+
+/* Maximum values for various fields (for non-blob tuples) */
+#define REC_MAX_N_FIELDS	(1024 - 1)
+#define REC_MAX_HEAP_NO		(2 * 8192 - 1)
+#define REC_MAX_N_OWNED		(16 - 1)
+
+/* REC_MAX_INDEX_COL_LEN is measured in bytes and is the maximum
+indexed column length (or indexed prefix length). It is set to 3*256,
+so that one can create a column prefix index on 256 characters of a
+TEXT or VARCHAR column also in the UTF-8 charset. In that charset,
+a character may take at most 3 bytes.
+This constant MUST NOT BE CHANGED, or the compatibility of InnoDB data
+files would be at risk! */
+#define REC_MAX_INDEX_COL_LEN	768
+
+#endif
diff --git a/storage/xtradb/include/row0ext.h b/storage/xtradb/include/row0ext.h
new file mode 100644
index 00000000000..08ebafa4d98
--- /dev/null
+++ b/storage/xtradb/include/row0ext.h
@@ -0,0 +1,98 @@
+/*****************************************************************************
+
+Copyright (c) 2006, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Caching of externally stored column prefixes
+
+Created September 2006 Marko Makela
+*******************************************************/
+
+#ifndef row0ext_h
+#define row0ext_h
+
+#include "univ.i"
+#include "row0types.h"
+#include "data0types.h"
+#include "mem0mem.h"
+
+/************************************************************************
+Creates a cache of column prefixes of externally stored columns. */
+UNIV_INTERN
+row_ext_t*
+row_ext_create(
+/*===========*/
+				/* out,own: column prefix cache */
+	ulint		n_ext,	/* in: number of externally stored columns */
+	const ulint*	ext,	/* in: col_no's of externally stored columns
+				in the InnoDB table object, as reported by
+				dict_col_get_no(); NOT relative to the records
+				in the clustered index */
+	const dtuple_t*	tuple,	/* in: data tuple containing the field
+				references of the externally stored
+				columns; must be indexed by col_no;
+				the clustered index record must be
+				covered by a lock or a page latch
+				to prevent deletion (rollback or purge). */
+	ulint		zip_size,/* compressed page size in bytes, or 0 */
+	mem_heap_t*	heap);	/* in: heap where created */
+
+/************************************************************************
+Looks up a column prefix of an externally stored column. */
+UNIV_INLINE
+const byte*
+row_ext_lookup_ith(
+/*===============*/
+					/* out: column prefix, or NULL if
+					the column is not stored externally,
+					or pointer to field_ref_zero
+					if the BLOB pointer is unset */
+	const row_ext_t*	ext,	/* in/out: column prefix cache */
+	ulint			i,	/* in: index of ext->ext[] */
+	ulint*			len);	/* out: length of prefix, in bytes,
+					at most REC_MAX_INDEX_COL_LEN */
+/************************************************************************
+Looks up a column prefix of an externally stored column. */
+UNIV_INLINE
+const byte*
+row_ext_lookup(
+/*===========*/
+					/* out: column prefix, or NULL if
+					the column is not stored externally,
+					or pointer to field_ref_zero
+					if the BLOB pointer is unset */
+	const row_ext_t*	ext,	/* in: column prefix cache */
+	ulint			col,	/* in: column number in the InnoDB
+					table object, as reported by
+					dict_col_get_no(); NOT relative to the
+					records in the clustered index */
+	ulint*			len);	/* out: length of prefix, in bytes,
+					at most REC_MAX_INDEX_COL_LEN */
+
+/* Prefixes of externally stored columns */
+struct row_ext_struct{
+	ulint		n_ext;	/* number of externally stored columns */
+	const ulint*	ext;	/* col_no's of externally stored columns */
+	byte*		buf;	/* backing store of the column prefix cache */
+	ulint		len[1];	/* prefix lengths; 0 if not cached */
+};
+
+#ifndef UNIV_NONINL
+#include "row0ext.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/row0ext.ic b/storage/xtradb/include/row0ext.ic
new file mode 100644
index 00000000000..e56fc175764
--- /dev/null
+++ b/storage/xtradb/include/row0ext.ic
@@ -0,0 +1,87 @@
+/*****************************************************************************
+
+Copyright (c) 2006, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Caching of externally stored column prefixes
+
+Created September 2006 Marko Makela
+*******************************************************/
+
+#include "rem0types.h"
+#include "btr0types.h"
+
+/************************************************************************
+Looks up a column prefix of an externally stored column. */
+UNIV_INLINE
+const byte*
+row_ext_lookup_ith(
+/*===============*/
+					/* out: column prefix, or NULL if
+					the column is not stored externally,
+					or pointer to field_ref_zero
+					if the BLOB pointer is unset */
+	const row_ext_t*	ext,	/* in/out: column prefix cache */
+	ulint			i,	/* in: index of ext->ext[] */
+	ulint*			len)	/* out: length of prefix, in bytes,
+					at most REC_MAX_INDEX_COL_LEN */
+{
+	ut_ad(ext);
+	ut_ad(len);
+	ut_ad(i < ext->n_ext);
+
+	*len = ext->len[i];
+
+	if (UNIV_UNLIKELY(*len == 0)) {
+		/* The BLOB could not be fetched to the cache. */
+		return(field_ref_zero);
+	} else {
+		return(ext->buf + i * REC_MAX_INDEX_COL_LEN);
+	}
+}
+
+/************************************************************************
+Looks up a column prefix of an externally stored column. */
+UNIV_INLINE
+const byte*
+row_ext_lookup(
+/*===========*/
+					/* out: column prefix, or NULL if
+					the column is not stored externally,
+					or pointer to field_ref_zero
+					if the BLOB pointer is unset */
+	const row_ext_t*	ext,	/* in: column prefix cache */
+	ulint			col,	/* in: column number in the InnoDB
+					table object, as reported by
+					dict_col_get_no(); NOT relative to the
+					records in the clustered index */
+	ulint*			len)	/* out: length of prefix, in bytes,
+					at most REC_MAX_INDEX_COL_LEN */
+{
+	ulint	i;
+
+	ut_ad(ext);
+	ut_ad(len);
+
+	for (i = 0; i < ext->n_ext; i++) {
+		if (col == ext->ext[i]) {
+			return(row_ext_lookup_ith(ext, i, len));
+		}
+	}
+
+	return(NULL);
+}
diff --git a/storage/xtradb/include/row0ins.h b/storage/xtradb/include/row0ins.h
new file mode 100644
index 00000000000..6aa83bed0f6
--- /dev/null
+++ b/storage/xtradb/include/row0ins.h
@@ -0,0 +1,157 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Insert into a table
+
+Created 4/20/1996 Heikki Tuuri
+*******************************************************/
+
+#ifndef row0ins_h
+#define row0ins_h
+
+#include "univ.i"
+#include "data0data.h"
+#include "que0types.h"
+#include "dict0types.h"
+#include "trx0types.h"
+#include "row0types.h"
+
+/*******************************************************************
+Checks if foreign key constraint fails for an index entry. Sets shared locks
+which lock either the success or the failure of the constraint. NOTE that
+the caller must have a shared latch on dict_foreign_key_check_lock. */
+UNIV_INTERN
+ulint
+row_ins_check_foreign_constraint(
+/*=============================*/
+				/* out: DB_SUCCESS, DB_LOCK_WAIT,
+				DB_NO_REFERENCED_ROW,
+				or DB_ROW_IS_REFERENCED */
+	ibool		check_ref,/* in: TRUE If we want to check that
+				the referenced table is ok, FALSE if we
+				want to to check the foreign key table */
+	dict_foreign_t*	foreign,/* in: foreign constraint; NOTE that the
+				tables mentioned in it must be in the
+				dictionary cache if they exist at all */
+	dict_table_t*	table,	/* in: if check_ref is TRUE, then the foreign
+				table, else the referenced table */
+	dtuple_t*	entry,	/* in: index entry for index */
+	que_thr_t*	thr);	/* in: query thread */
+/*************************************************************************
+Creates an insert node struct. */
+UNIV_INTERN
+ins_node_t*
+ins_node_create(
+/*============*/
+					/* out, own: insert node struct */
+	ulint		ins_type,	/* in: INS_VALUES, ... */
+	dict_table_t*	table,		/* in: table where to insert */
+	mem_heap_t*	heap);		/* in: mem heap where created */
+/*************************************************************************
+Sets a new row to insert for an INS_DIRECT node. This function is only used
+if we have constructed the row separately, which is a rare case; this
+function is quite slow. */
+UNIV_INTERN
+void
+ins_node_set_new_row(
+/*=================*/
+	ins_node_t*	node,	/* in: insert node */
+	dtuple_t*	row);	/* in: new row (or first row) for the node */
+/*******************************************************************
+Inserts an index entry to index. Tries first optimistic, then pessimistic
+descent down the tree. If the entry matches enough to a delete marked record,
+performs the insert by updating or delete unmarking the delete marked
+record. */
+UNIV_INTERN
+ulint
+row_ins_index_entry(
+/*================*/
+				/* out: DB_SUCCESS, DB_LOCK_WAIT,
+				DB_DUPLICATE_KEY, or some other error code */
+	dict_index_t*	index,	/* in: index */
+	dtuple_t*	entry,	/* in: index entry to insert */
+	ulint		n_ext,	/* in: number of externally stored columns */
+	ibool		foreign,/* in: TRUE=check foreign key constraints */
+	que_thr_t*	thr);	/* in: query thread */
+/***************************************************************
+Inserts a row to a table. This is a high-level function used in
+SQL execution graphs. */
+UNIV_INTERN
+que_thr_t*
+row_ins_step(
+/*=========*/
+				/* out: query thread to run next or NULL */
+	que_thr_t*	thr);	/* in: query thread */
+/***************************************************************
+Creates an entry template for each index of a table. */
+UNIV_INTERN
+void
+ins_node_create_entry_list(
+/*=======================*/
+	ins_node_t*	node);	/* in: row insert node */
+
+/* Insert node structure */
+
+struct ins_node_struct{
+	que_common_t	common;	/* node type: QUE_NODE_INSERT */
+	ulint		ins_type;/* INS_VALUES, INS_SEARCHED, or INS_DIRECT */
+	dtuple_t*	row;	/* row to insert */
+	dict_table_t*	table;	/* table where to insert */
+	sel_node_t*	select;	/* select in searched insert */
+	que_node_t*	values_list;/* list of expressions to evaluate and
+				insert in an INS_VALUES insert */
+	ulint		state;	/* node execution state */
+	dict_index_t*	index;	/* NULL, or the next index where the index
+				entry should be inserted */
+	dtuple_t*	entry;	/* NULL, or entry to insert in the index;
+				after a successful insert of the entry,
+				this should be reset to NULL */
+	UT_LIST_BASE_NODE_T(dtuple_t)
+			entry_list;/* list of entries, one for each index */
+	byte*		row_id_buf;/* buffer for the row id sys field in row */
+	dulint		trx_id;	/* trx id or the last trx which executed the
+				node */
+	byte*		trx_id_buf;/* buffer for the trx id sys field in row */
+	mem_heap_t*	entry_sys_heap;
+				/* memory heap used as auxiliary storage;
+				entry_list and sys fields are stored here;
+				if this is NULL, entry list should be created
+				and buffers for sys fields in row allocated */
+	ulint		magic_n;
+};
+
+#define	INS_NODE_MAGIC_N	15849075
+
+/* Insert node types */
+#define INS_SEARCHED	0	/* INSERT INTO ... SELECT ... */
+#define INS_VALUES	1	/* INSERT INTO ... VALUES ... */
+#define INS_DIRECT	2	/* this is for internal use in dict0crea:
+				insert the row directly */
+
+/* Node execution states */
+#define	INS_NODE_SET_IX_LOCK	1	/* we should set an IX lock on table */
+#define INS_NODE_ALLOC_ROW_ID	2	/* row id should be allocated */
+#define	INS_NODE_INSERT_ENTRIES 3	/* index entries should be built and
+					inserted */
+
+#ifndef UNIV_NONINL
+#include "row0ins.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/row0ins.ic b/storage/xtradb/include/row0ins.ic
new file mode 100644
index 00000000000..b7aeaf97834
--- /dev/null
+++ b/storage/xtradb/include/row0ins.ic
@@ -0,0 +1,25 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Insert into a table
+
+Created 4/20/1996 Heikki Tuuri
+*******************************************************/
+
+
diff --git a/storage/xtradb/include/row0merge.h b/storage/xtradb/include/row0merge.h
new file mode 100644
index 00000000000..9975497cbeb
--- /dev/null
+++ b/storage/xtradb/include/row0merge.h
@@ -0,0 +1,198 @@
+/*****************************************************************************
+
+Copyright (c) 2005, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Index build routines using a merge sort
+
+Created 13/06/2005 Jan Lindstrom
+*******************************************************/
+
+#ifndef row0merge_h
+#define row0merge_h
+
+#include "univ.i"
+#include "data0data.h"
+#include "dict0types.h"
+#include "trx0types.h"
+#include "que0types.h"
+#include "mtr0mtr.h"
+#include "rem0types.h"
+#include "rem0rec.h"
+#include "read0types.h"
+#include "btr0types.h"
+#include "row0mysql.h"
+#include "lock0types.h"
+
+/* This structure holds index field definitions */
+
+struct merge_index_field_struct {
+	ulint		prefix_len;	/* Prefix len */
+	const char*	field_name;	/* Field name */
+};
+
+typedef struct merge_index_field_struct merge_index_field_t;
+
+/* This structure holds index definitions */
+
+struct merge_index_def_struct {
+	const char*		name;		/* Index name */
+	ulint			ind_type;	/* 0, DICT_UNIQUE,
+						or DICT_CLUSTERED */
+	ulint			n_fields;	/* Number of fields in index */
+	merge_index_field_t*	fields;		/* Field definitions */
+};
+
+typedef struct merge_index_def_struct merge_index_def_t;
+
+/*************************************************************************
+Sets an exclusive lock on a table, for the duration of creating indexes. */
+UNIV_INTERN
+ulint
+row_merge_lock_table(
+/*=================*/
+					/* out: error code or DB_SUCCESS */
+	trx_t*		trx,		/* in/out: transaction */
+	dict_table_t*	table,		/* in: table to lock */
+	enum lock_mode	mode);		/* in: LOCK_X or LOCK_S */
+/*************************************************************************
+Drop an index from the InnoDB system tables.  The data dictionary must
+have been locked exclusively by the caller, because the transaction
+will not be committed. */
+UNIV_INTERN
+void
+row_merge_drop_index(
+/*=================*/
+	dict_index_t*	index,	/* in: index to be removed */
+	dict_table_t*	table,	/* in: table */
+	trx_t*		trx);	/* in: transaction handle */
+/*************************************************************************
+Drop those indexes which were created before an error occurred when
+building an index.  The data dictionary must have been locked
+exclusively by the caller, because the transaction will not be
+committed. */
+UNIV_INTERN
+void
+row_merge_drop_indexes(
+/*===================*/
+	trx_t*		trx,		/* in: transaction */
+	dict_table_t*	table,		/* in: table containing the indexes */
+	dict_index_t**	index,		/* in: indexes to drop */
+	ulint		num_created);	/* in: number of elements in index[] */
+/*************************************************************************
+Drop all partially created indexes during crash recovery. */
+UNIV_INTERN
+void
+row_merge_drop_temp_indexes(void);
+/*=============================*/
+/*************************************************************************
+Rename the tables in the data dictionary.  The data dictionary must
+have been locked exclusively by the caller, because the transaction
+will not be committed. */
+UNIV_INTERN
+ulint
+row_merge_rename_tables(
+/*====================*/
+					/* out: error code or DB_SUCCESS */
+	dict_table_t*	old_table,	/* in/out: old table, renamed to
+					tmp_name */
+	dict_table_t*	new_table,	/* in/out: new table, renamed to
+					old_table->name */
+	const char*	tmp_name,	/* in: new name for old_table */
+	trx_t*		trx);		/* in: transaction handle */
+
+/*************************************************************************
+Create a temporary table for creating a primary key, using the definition
+of an existing table. */
+UNIV_INTERN
+dict_table_t*
+row_merge_create_temporary_table(
+/*=============================*/
+						/* out: table,
+						or NULL on error */
+	const char*		table_name,	/* in: new table name */
+	const merge_index_def_t*index_def,	/* in: the index definition
+						of the primary key */
+	const dict_table_t*	table,		/* in: old table definition */
+	trx_t*			trx);		/* in/out: transaction
+						(sets error_state) */
+/*************************************************************************
+Rename the temporary indexes in the dictionary to permanent ones.  The
+data dictionary must have been locked exclusively by the caller,
+because the transaction will not be committed. */
+UNIV_INTERN
+ulint
+row_merge_rename_indexes(
+/*=====================*/
+					/* out: DB_SUCCESS if all OK */
+	trx_t*		trx,		/* in/out: transaction */
+	dict_table_t*	table);		/* in/out: table with new indexes */
+/*************************************************************************
+Create the index and load in to the dictionary. */
+UNIV_INTERN
+dict_index_t*
+row_merge_create_index(
+/*===================*/
+					/* out: index, or NULL on error */
+	trx_t*		trx,		/* in/out: trx (sets error_state) */
+	dict_table_t*	table,		/* in: the index is on this table */
+	const merge_index_def_t*	/* in: the index definition */
+			index_def);
+#ifdef ROW_MERGE_IS_INDEX_USABLE
+/*************************************************************************
+Check if a transaction can use an index. */
+UNIV_INTERN
+ibool
+row_merge_is_index_usable(
+/*======================*/
+					/* out: TRUE if index can be used by
+					the transaction else FALSE*/
+	const trx_t*		trx,	/* in: transaction */
+	const dict_index_t*	index);	/* in: index to check */
+#endif /* ROW_MERGE_IS_INDEX_USABLE */
+/*************************************************************************
+If there are views that refer to the old table name then we "attach" to
+the new instance of the table else we drop it immediately. */
+UNIV_INTERN
+ulint
+row_merge_drop_table(
+/*=================*/
+					/* out: DB_SUCCESS or error code */
+	trx_t*		trx,		/* in: transaction */
+	dict_table_t*	table);		/* in: table instance to drop */
+
+/*************************************************************************
+Build indexes on a table by reading a clustered index,
+creating a temporary file containing index entries, merge sorting
+these index entries and inserting sorted index entries to indexes. */
+UNIV_INTERN
+ulint
+row_merge_build_indexes(
+/*====================*/
+					/* out: DB_SUCCESS or error code */
+	trx_t*		trx,		/* in: transaction */
+	dict_table_t*	old_table,	/* in: table where rows are
+					read from */
+	dict_table_t*	new_table,	/* in: table where indexes are
+					created; identical to old_table
+					unless creating a PRIMARY KEY */
+	dict_index_t**	indexes,	/* in: indexes to be created */
+	ulint		n_indexes,	/* in: size of indexes[] */
+	TABLE*		table);		/* in/out: MySQL table, for
+					reporting erroneous key value
+					if applicable */
+#endif /* row0merge.h */
diff --git a/storage/xtradb/include/row0mysql.h b/storage/xtradb/include/row0mysql.h
new file mode 100644
index 00000000000..c1e11124a5d
--- /dev/null
+++ b/storage/xtradb/include/row0mysql.h
@@ -0,0 +1,769 @@
+/*****************************************************************************
+
+Copyright (c) 2000, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Interface between Innobase row operations and MySQL.
+Contains also create table and other data dictionary operations.
+
+Created 9/17/2000 Heikki Tuuri
+*******************************************************/
+
+#ifndef row0mysql_h
+#define row0mysql_h
+
+#include "univ.i"
+#include "data0data.h"
+#include "que0types.h"
+#include "dict0types.h"
+#include "trx0types.h"
+#include "row0types.h"
+#include "btr0pcur.h"
+#include "trx0types.h"
+
+extern ibool row_rollback_on_timeout;
+
+typedef struct row_prebuilt_struct row_prebuilt_t;
+
+/***********************************************************************
+Frees the blob heap in prebuilt when no longer needed. */
+UNIV_INTERN
+void
+row_mysql_prebuilt_free_blob_heap(
+/*==============================*/
+	row_prebuilt_t*	prebuilt);	/* in: prebuilt struct of a
+					ha_innobase:: table handle */
+/***********************************************************************
+Stores a >= 5.0.3 format true VARCHAR length to dest, in the MySQL row
+format. */
+UNIV_INTERN
+byte*
+row_mysql_store_true_var_len(
+/*=========================*/
+			/* out: pointer to the data, we skip the 1 or 2 bytes
+			at the start that are used to store the len */
+	byte*	dest,	/* in: where to store */
+	ulint	len,	/* in: length, must fit in two bytes */
+	ulint	lenlen);/* in: storage length of len: either 1 or 2 bytes */
+/***********************************************************************
+Reads a >= 5.0.3 format true VARCHAR length, in the MySQL row format, and
+returns a pointer to the data. */
+
+const byte*
+row_mysql_read_true_varchar(
+/*========================*/
+				/* out: pointer to the data, we skip
+				the 1 or 2 bytes at the start that are
+				used to store the len */
+	ulint*		len,	/* out: variable-length field length */
+	const byte*	field,	/* in: field in the MySQL format */
+	ulint		lenlen);/* in: storage length of len: either 1
+				or 2 bytes */
+/***********************************************************************
+Stores a reference to a BLOB in the MySQL format. */
+UNIV_INTERN
+void
+row_mysql_store_blob_ref(
+/*=====================*/
+	byte*		dest,	/* in: where to store */
+	ulint		col_len,/* in: dest buffer size: determines into
+				how many bytes the BLOB length is stored,
+				the space for the length may vary from 1
+				to 4 bytes */
+	const void*	data,	/* in: BLOB data; if the value to store
+				is SQL NULL this should be NULL pointer */
+	ulint		len);	/* in: BLOB length; if the value to store
+				is SQL NULL this should be 0; remember
+				also to set the NULL bit in the MySQL record
+				header! */
+/***********************************************************************
+Reads a reference to a BLOB in the MySQL format. */
+
+const byte*
+row_mysql_read_blob_ref(
+/*====================*/
+					/* out: pointer to BLOB data */
+	ulint*		len,		/* out: BLOB length */
+	const byte*	ref,		/* in: BLOB reference in the
+					MySQL format */
+	ulint		col_len);	/* in: BLOB reference length
+					(not BLOB length) */
+/******************************************************************
+Stores a non-SQL-NULL field given in the MySQL format in the InnoDB format.
+The counterpart of this function is row_sel_field_store_in_mysql_format() in
+row0sel.c. */
+UNIV_INTERN
+byte*
+row_mysql_store_col_in_innobase_format(
+/*===================================*/
+					/* out: up to which byte we used
+					buf in the conversion */
+	dfield_t*	dfield,		/* in/out: dfield where dtype
+					information must be already set when
+					this function is called! */
+	byte*		buf,		/* in/out: buffer for a converted
+					integer value; this must be at least
+					col_len long then! */
+	ibool		row_format_col,	/* TRUE if the mysql_data is from
+					a MySQL row, FALSE if from a MySQL
+					key value;
+					in MySQL, a true VARCHAR storage
+					format differs in a row and in a
+					key value: in a key value the length
+					is always stored in 2 bytes! */
+	const byte*	mysql_data,	/* in: MySQL column value, not
+					SQL NULL; NOTE that dfield may also
+					get a pointer to mysql_data,
+					therefore do not discard this as long
+					as dfield is used! */
+	ulint		col_len,	/* in: MySQL column length; NOTE that
+					this is the storage length of the
+					column in the MySQL format row, not
+					necessarily the length of the actual
+					payload data; if the column is a true
+					VARCHAR then this is irrelevant */
+	ulint		comp);		/* in: nonzero=compact format */
+/********************************************************************
+Handles user errors and lock waits detected by the database engine. */
+UNIV_INTERN
+ibool
+row_mysql_handle_errors(
+/*====================*/
+				/* out: TRUE if it was a lock wait and
+				we should continue running the query thread */
+	ulint*		new_err,/* out: possible new error encountered in
+				rollback, or the old error which was
+				during the function entry */
+	trx_t*		trx,	/* in: transaction */
+	que_thr_t*	thr,	/* in: query thread */
+	trx_savept_t*	savept);/* in: savepoint */
+/************************************************************************
+Create a prebuilt struct for a MySQL table handle. */
+UNIV_INTERN
+row_prebuilt_t*
+row_create_prebuilt(
+/*================*/
+				/* out, own: a prebuilt struct */
+	dict_table_t*	table);	/* in: Innobase table handle */
+/************************************************************************
+Free a prebuilt struct for a MySQL table handle. */
+UNIV_INTERN
+void
+row_prebuilt_free(
+/*==============*/
+	row_prebuilt_t*	prebuilt,	/* in, own: prebuilt struct */
+	ibool		dict_locked);	/* in: TRUE=data dictionary locked */
+/*************************************************************************
+Updates the transaction pointers in query graphs stored in the prebuilt
+struct. */
+UNIV_INTERN
+void
+row_update_prebuilt_trx(
+/*====================*/
+					/* out: prebuilt dtuple */
+	row_prebuilt_t*	prebuilt,	/* in: prebuilt struct in MySQL
+					handle */
+	trx_t*		trx);		/* in: transaction handle */
+/*************************************************************************
+Unlocks AUTO_INC type locks that were possibly reserved by a trx. */
+UNIV_INTERN
+void
+row_unlock_table_autoinc_for_mysql(
+/*===============================*/
+	trx_t*	trx);			/* in/out: transaction */
+/*************************************************************************
+Sets an AUTO_INC type lock on the table mentioned in prebuilt. The
+AUTO_INC lock gives exclusive access to the auto-inc counter of the
+table. The lock is reserved only for the duration of an SQL statement.
+It is not compatible with another AUTO_INC or exclusive lock on the
+table. */
+UNIV_INTERN
+int
+row_lock_table_autoinc_for_mysql(
+/*=============================*/
+					/* out: error code or DB_SUCCESS */
+	row_prebuilt_t*	prebuilt);	/* in: prebuilt struct in the MySQL
+					table handle */
+/*************************************************************************
+Sets a table lock on the table mentioned in prebuilt. */
+UNIV_INTERN
+int
+row_lock_table_for_mysql(
+/*=====================*/
+					/* out: error code or DB_SUCCESS */
+	row_prebuilt_t*	prebuilt,	/* in: prebuilt struct in the MySQL
+					table handle */
+	dict_table_t*	table,		/* in: table to lock, or NULL
+					if prebuilt->table should be
+					locked as
+					prebuilt->select_lock_type */
+	ulint		mode);		/* in: lock mode of table
+					(ignored if table==NULL) */
+
+/*************************************************************************
+Does an insert for MySQL. */
+UNIV_INTERN
+int
+row_insert_for_mysql(
+/*=================*/
+					/* out: error code or DB_SUCCESS */
+	byte*		mysql_rec,	/* in: row in the MySQL format */
+	row_prebuilt_t*	prebuilt);	/* in: prebuilt struct in MySQL
+					handle */
+/*************************************************************************
+Builds a dummy query graph used in selects. */
+UNIV_INTERN
+void
+row_prebuild_sel_graph(
+/*===================*/
+	row_prebuilt_t*	prebuilt);	/* in: prebuilt struct in MySQL
+					handle */
+/*************************************************************************
+Gets pointer to a prebuilt update vector used in updates. If the update
+graph has not yet been built in the prebuilt struct, then this function
+first builds it. */
+UNIV_INTERN
+upd_t*
+row_get_prebuilt_update_vector(
+/*===========================*/
+					/* out: prebuilt update vector */
+	row_prebuilt_t*	prebuilt);	/* in: prebuilt struct in MySQL
+					handle */
+/*************************************************************************
+Checks if a table is such that we automatically created a clustered
+index on it (on row id). */
+UNIV_INTERN
+ibool
+row_table_got_default_clust_index(
+/*==============================*/
+	const dict_table_t*	table);
+/*************************************************************************
+Calculates the key number used inside MySQL for an Innobase index. We have
+to take into account if we generated a default clustered index for the table */
+UNIV_INTERN
+ulint
+row_get_mysql_key_number_for_index(
+/*===============================*/
+	const dict_index_t*	index);
+/*************************************************************************
+Does an update or delete of a row for MySQL. */
+UNIV_INTERN
+int
+row_update_for_mysql(
+/*=================*/
+					/* out: error code or DB_SUCCESS */
+	byte*		mysql_rec,	/* in: the row to be updated, in
+					the MySQL format */
+	row_prebuilt_t*	prebuilt);	/* in: prebuilt struct in MySQL
+					handle */
+/*************************************************************************
+This can only be used when srv_locks_unsafe_for_binlog is TRUE or
+session is using a READ COMMITTED isolation level. Before
+calling this function we must use trx_reset_new_rec_lock_info() and
+trx_register_new_rec_lock() to store the information which new record locks
+really were set. This function removes a newly set lock under prebuilt->pcur,
+and also under prebuilt->clust_pcur. Currently, this is only used and tested
+in the case of an UPDATE or a DELETE statement, where the row lock is of the
+LOCK_X type.
+Thus, this implements a 'mini-rollback' that releases the latest record
+locks we set. */
+UNIV_INTERN
+int
+row_unlock_for_mysql(
+/*=================*/
+					/* out: error code or DB_SUCCESS */
+	row_prebuilt_t*	prebuilt,	/* in: prebuilt struct in MySQL
+					handle */
+	ibool		has_latches_on_recs);/* TRUE if called so that we have
+					the latches on the records under pcur
+					and clust_pcur, and we do not need to
+					reposition the cursors. */
+/*************************************************************************
+Creates an query graph node of 'update' type to be used in the MySQL
+interface. */
+UNIV_INTERN
+upd_node_t*
+row_create_update_node_for_mysql(
+/*=============================*/
+				/* out, own: update node */
+	dict_table_t*	table,	/* in: table to update */
+	mem_heap_t*	heap);	/* in: mem heap from which allocated */
+/**************************************************************************
+Does a cascaded delete or set null in a foreign key operation. */
+UNIV_INTERN
+ulint
+row_update_cascade_for_mysql(
+/*=========================*/
+				/* out: error code or DB_SUCCESS */
+	que_thr_t*	thr,	/* in: query thread */
+	upd_node_t*	node,	/* in: update node used in the cascade
+				or set null operation */
+	dict_table_t*	table);	/* in: table where we do the operation */
+/*************************************************************************
+Locks the data dictionary exclusively for performing a table create or other
+data dictionary modification operation. */
+UNIV_INTERN
+void
+row_mysql_lock_data_dictionary_func(
+/*================================*/
+	trx_t*		trx,	/* in/out: transaction */
+	const char*	file,	/* in: file name */
+	ulint		line);	/* in: line number */
+#define row_mysql_lock_data_dictionary(trx)				\
+	row_mysql_lock_data_dictionary_func(trx, __FILE__, __LINE__)
+/*************************************************************************
+Unlocks the data dictionary exclusive lock. */
+UNIV_INTERN
+void
+row_mysql_unlock_data_dictionary(
+/*=============================*/
+	trx_t*	trx);	/* in/out: transaction */
+/*************************************************************************
+Locks the data dictionary in shared mode from modifications, for performing
+foreign key check, rollback, or other operation invisible to MySQL. */
+UNIV_INTERN
+void
+row_mysql_freeze_data_dictionary_func(
+/*==================================*/
+	trx_t*		trx,	/* in/out: transaction */
+	const char*	file,	/* in: file name */
+	ulint		line);	/* in: line number */
+#define row_mysql_freeze_data_dictionary(trx)				\
+	row_mysql_freeze_data_dictionary_func(trx, __FILE__, __LINE__)
+/*************************************************************************
+Unlocks the data dictionary shared lock. */
+UNIV_INTERN
+void
+row_mysql_unfreeze_data_dictionary(
+/*===============================*/
+	trx_t*	trx);	/* in/out: transaction */
+#ifndef UNIV_HOTBACKUP
+/*************************************************************************
+Creates a table for MySQL. If the name of the table ends in
+one of "innodb_monitor", "innodb_lock_monitor", "innodb_tablespace_monitor",
+"innodb_table_monitor", then this will also start the printing of monitor
+output by the master thread. If the table name ends in "innodb_mem_validate",
+InnoDB will try to invoke mem_validate(). */
+UNIV_INTERN
+int
+row_create_table_for_mysql(
+/*=======================*/
+					/* out: error code or DB_SUCCESS */
+	dict_table_t*	table,		/* in, own: table definition
+					(will be freed) */
+	trx_t*		trx);		/* in: transaction handle */
+/*************************************************************************
+Does an index creation operation for MySQL. TODO: currently failure
+to create an index results in dropping the whole table! This is no problem
+currently as all indexes must be created at the same time as the table. */
+UNIV_INTERN
+int
+row_create_index_for_mysql(
+/*=======================*/
+					/* out: error number or DB_SUCCESS */
+	dict_index_t*	index,		/* in, own: index definition
+					(will be freed) */
+	trx_t*		trx,		/* in: transaction handle */
+	const ulint*	field_lengths); /* in: if not NULL, must contain
+					dict_index_get_n_fields(index)
+					actual field lengths for the
+					index columns, which are
+					then checked for not being too
+					large. */
+/*************************************************************************
+Scans a table create SQL string and adds to the data dictionary
+the foreign key constraints declared in the string. This function
+should be called after the indexes for a table have been created.
+Each foreign key constraint must be accompanied with indexes in
+bot participating tables. The indexes are allowed to contain more
+fields than mentioned in the constraint. */
+UNIV_INTERN
+int
+row_table_add_foreign_constraints(
+/*==============================*/
+					/* out: error code or DB_SUCCESS */
+	trx_t*		trx,		/* in: transaction */
+	const char*	sql_string,	/* in: table create statement where
+					foreign keys are declared like:
+				FOREIGN KEY (a, b) REFERENCES table2(c, d),
+					table2 can be written also with the
+					database name before it: test.table2 */
+	const char*	name,		/* in: table full name in the
+					normalized form
+					database_name/table_name */
+	ibool		reject_fks);	/* in: if TRUE, fail with error
+					code DB_CANNOT_ADD_CONSTRAINT if
+					any foreign keys are found. */
+
+/*************************************************************************
+The master thread in srv0srv.c calls this regularly to drop tables which
+we must drop in background after queries to them have ended. Such lazy
+dropping of tables is needed in ALTER TABLE on Unix. */
+UNIV_INTERN
+ulint
+row_drop_tables_for_mysql_in_background(void);
+/*=========================================*/
+					/* out: how many tables dropped
+					+ remaining tables in list */
+/*************************************************************************
+Get the background drop list length. NOTE: the caller must own the kernel
+mutex! */
+UNIV_INTERN
+ulint
+row_get_background_drop_list_len_low(void);
+/*======================================*/
+					/* out: how many tables in list */
+/*************************************************************************
+Truncates a table for MySQL. */
+UNIV_INTERN
+int
+row_truncate_table_for_mysql(
+/*=========================*/
+				/* out: error code or DB_SUCCESS */
+	dict_table_t*	table,	/* in: table handle */
+	trx_t*		trx);	/* in: transaction handle */
+/*************************************************************************
+Drops a table for MySQL.  If the name of the dropped table ends in
+one of "innodb_monitor", "innodb_lock_monitor", "innodb_tablespace_monitor",
+"innodb_table_monitor", then this will also stop the printing of monitor
+output by the master thread.  If the data dictionary was not already locked
+by the transaction, the transaction will be committed.  Otherwise, the
+data dictionary will remain locked. */
+UNIV_INTERN
+int
+row_drop_table_for_mysql(
+/*=====================*/
+				/* out: error code or DB_SUCCESS */
+	const char*	name,	/* in: table name */
+	trx_t*		trx,	/* in: transaction handle */
+	ibool		drop_db);/* in: TRUE=dropping whole database */
+
+/*************************************************************************
+Discards the tablespace of a table which stored in an .ibd file. Discarding
+means that this function deletes the .ibd file and assigns a new table id for
+the table. Also the flag table->ibd_file_missing is set TRUE. */
+UNIV_INTERN
+int
+row_discard_tablespace_for_mysql(
+/*=============================*/
+				/* out: error code or DB_SUCCESS */
+	const char*	name,	/* in: table name */
+	trx_t*		trx);	/* in: transaction handle */
+/*********************************************************************
+Imports a tablespace. The space id in the .ibd file must match the space id
+of the table in the data dictionary. */
+UNIV_INTERN
+int
+row_import_tablespace_for_mysql(
+/*============================*/
+				/* out: error code or DB_SUCCESS */
+	const char*	name,	/* in: table name */
+	trx_t*		trx);	/* in: transaction handle */
+/*************************************************************************
+Drops a database for MySQL. */
+UNIV_INTERN
+int
+row_drop_database_for_mysql(
+/*========================*/
+				/* out: error code or DB_SUCCESS */
+	const char*	name,	/* in: database name which ends to '/' */
+	trx_t*		trx);	/* in: transaction handle */
+/*************************************************************************
+Renames a table for MySQL. */
+UNIV_INTERN
+ulint
+row_rename_table_for_mysql(
+/*=======================*/
+					/* out: error code or DB_SUCCESS */
+	const char*	old_name,	/* in: old table name */
+	const char*	new_name,	/* in: new table name */
+	trx_t*		trx,		/* in: transaction handle */
+	ibool		commit);	/* in: if TRUE then commit trx */
+/*************************************************************************
+Checks a table for corruption. */
+UNIV_INTERN
+ulint
+row_check_table_for_mysql(
+/*======================*/
+					/* out: DB_ERROR or DB_SUCCESS */
+	row_prebuilt_t*	prebuilt);	/* in: prebuilt struct in MySQL
+					handle */
+#endif /* !UNIV_HOTBACKUP */
+
+/*************************************************************************
+Determines if a table is a magic monitor table. */
+UNIV_INTERN
+ibool
+row_is_magic_monitor_table(
+/*=======================*/
+					/* out: TRUE if monitor table */
+	const char*	table_name);	/* in: name of the table, in the
+					form database/table_name */
+
+/* A struct describing a place for an individual column in the MySQL
+row format which is presented to the table handler in ha_innobase.
+This template struct is used to speed up row transformations between
+Innobase and MySQL. */
+
+typedef struct mysql_row_templ_struct mysql_row_templ_t;
+struct mysql_row_templ_struct {
+	ulint	col_no;			/* column number of the column */
+	ulint	rec_field_no;		/* field number of the column in an
+					Innobase record in the current index;
+					not defined if template_type is
+					ROW_MYSQL_WHOLE_ROW */
+	ulint	mysql_col_offset;	/* offset of the column in the MySQL
+					row format */
+	ulint	mysql_col_len;		/* length of the column in the MySQL
+					row format */
+	ulint	mysql_null_byte_offset;	/* MySQL NULL bit byte offset in a
+					MySQL record */
+	ulint	mysql_null_bit_mask;	/* bit mask to get the NULL bit,
+					zero if column cannot be NULL */
+	ulint	type;			/* column type in Innobase mtype
+					numbers DATA_CHAR... */
+	ulint	mysql_type;		/* MySQL type code; this is always
+					< 256 */
+	ulint	mysql_length_bytes;	/* if mysql_type
+					== DATA_MYSQL_TRUE_VARCHAR, this tells
+					whether we should use 1 or 2 bytes to
+					store the MySQL true VARCHAR data
+					length at the start of row in the MySQL
+					format (NOTE that the MySQL key value
+					format always uses 2 bytes for the data
+					len) */
+	ulint	charset;		/* MySQL charset-collation code
+					of the column, or zero */
+	ulint	mbminlen;		/* minimum length of a char, in bytes,
+					or zero if not a char type */
+	ulint	mbmaxlen;		/* maximum length of a char, in bytes,
+					or zero if not a char type */
+	ulint	is_unsigned;		/* if a column type is an integer
+					type and this field is != 0, then
+					it is an unsigned integer type */
+};
+
+#define MYSQL_FETCH_CACHE_SIZE		8
+/* After fetching this many rows, we start caching them in fetch_cache */
+#define MYSQL_FETCH_CACHE_THRESHOLD	4
+
+#define ROW_PREBUILT_ALLOCATED	78540783
+#define ROW_PREBUILT_FREED	26423527
+
+/* A struct for (sometimes lazily) prebuilt structures in an Innobase table
+handle used within MySQL; these are used to save CPU time. */
+
+struct row_prebuilt_struct {
+	ulint		magic_n;	/* this magic number is set to
+					ROW_PREBUILT_ALLOCATED when created,
+					or ROW_PREBUILT_FREED when the
+					struct has been freed */
+	dict_table_t*	table;		/* Innobase table handle */
+	trx_t*		trx;		/* current transaction handle */
+	ibool		sql_stat_start;	/* TRUE when we start processing of
+					an SQL statement: we may have to set
+					an intention lock on the table,
+					create a consistent read view etc. */
+	ibool		mysql_has_locked; /* this is set TRUE when MySQL
+					calls external_lock on this handle
+					with a lock flag, and set FALSE when
+					with the F_UNLOCK flag */
+	ibool		clust_index_was_generated;
+					/* if the user did not define a
+					primary key in MySQL, then Innobase
+					automatically generated a clustered
+					index where the ordering column is
+					the row id: in this case this flag
+					is set to TRUE */
+	dict_index_t*	index;		/* current index for a search, if
+					any */
+	ulint		read_just_key;	/* set to 1 when MySQL calls
+					ha_innobase::extra with the
+					argument HA_EXTRA_KEYREAD; it is enough
+					to read just columns defined in
+					the index (i.e., no read of the
+					clustered index record necessary) */
+	ibool		used_in_HANDLER;/* TRUE if we have been using this
+					handle in a MySQL HANDLER low level
+					index cursor command: then we must
+					store the pcur position even in a
+					unique search from a clustered index,
+					because HANDLER allows NEXT and PREV
+					in such a situation */
+	ulint		template_type;	/* ROW_MYSQL_WHOLE_ROW,
+					ROW_MYSQL_REC_FIELDS,
+					ROW_MYSQL_DUMMY_TEMPLATE, or
+					ROW_MYSQL_NO_TEMPLATE */
+	ulint		n_template;	/* number of elements in the
+					template */
+	ulint		null_bitmap_len;/* number of bytes in the SQL NULL
+					bitmap at the start of a row in the
+					MySQL format */
+	ibool		need_to_access_clustered; /* if we are fetching
+					columns through a secondary index
+					and at least one column is not in
+					the secondary index, then this is
+					set to TRUE */
+	ibool		templ_contains_blob;/* TRUE if the template contains
+					BLOB column(s) */
+	mysql_row_templ_t* mysql_template;/* template used to transform
+					rows fast between MySQL and Innobase
+					formats; memory for this template
+					is not allocated from 'heap' */
+	mem_heap_t*	heap;		/* memory heap from which
+					these auxiliary structures are
+					allocated when needed */
+	ins_node_t*	ins_node;	/* Innobase SQL insert node
+					used to perform inserts
+					to the table */
+	byte*		ins_upd_rec_buff;/* buffer for storing data converted
+					to the Innobase format from the MySQL
+					format */
+	const byte*	default_rec;	/* the default values of all columns
+					(a "default row") in MySQL format */
+	ulint		hint_need_to_fetch_extra_cols;
+					/* normally this is set to 0; if this
+					is set to ROW_RETRIEVE_PRIMARY_KEY,
+					then we should at least retrieve all
+					columns in the primary key; if this
+					is set to ROW_RETRIEVE_ALL_COLS, then
+					we must retrieve all columns in the
+					key (if read_just_key == 1), or all
+					columns in the table */
+	upd_node_t*	upd_node;	/* Innobase SQL update node used
+					to perform updates and deletes */
+	que_fork_t*	ins_graph;	/* Innobase SQL query graph used
+					in inserts */
+	que_fork_t*	upd_graph;	/* Innobase SQL query graph used
+					in updates or deletes */
+	btr_pcur_t*	pcur;		/* persistent cursor used in selects
+					and updates */
+	btr_pcur_t*	clust_pcur;	/* persistent cursor used in
+					some selects and updates */
+	que_fork_t*	sel_graph;	/* dummy query graph used in
+					selects */
+	dtuple_t*	search_tuple;	/* prebuilt dtuple used in selects */
+	byte		row_id[DATA_ROW_ID_LEN];
+					/* if the clustered index was
+					generated, the row id of the
+					last row fetched is stored
+					here */
+	dtuple_t*	clust_ref;	/* prebuilt dtuple used in
+					sel/upd/del */
+	ulint		select_lock_type;/* LOCK_NONE, LOCK_S, or LOCK_X */
+	ulint		stored_select_lock_type;/* this field is used to
+					remember the original select_lock_type
+					that was decided in ha_innodb.cc,
+					::store_lock(), ::external_lock(),
+					etc. */
+	ulint		row_read_type;	/* ROW_READ_WITH_LOCKS if row locks
+					should be the obtained for records
+					under an UPDATE or DELETE cursor.
+					If innodb_locks_unsafe_for_binlog
+					is TRUE, this can be set to
+					ROW_READ_TRY_SEMI_CONSISTENT, so that
+					if the row under an UPDATE or DELETE
+					cursor was locked by another
+					transaction, InnoDB will resort
+					to reading the last committed value
+					('semi-consistent read').  Then,
+					this field will be set to
+					ROW_READ_DID_SEMI_CONSISTENT to
+					indicate that.	If the row does not
+					match the WHERE condition, MySQL will
+					invoke handler::unlock_row() to
+					clear the flag back to
+					ROW_READ_TRY_SEMI_CONSISTENT and
+					to simply skip the row.	 If
+					the row matches, the next call to
+					row_search_for_mysql() will lock
+					the row.
+					This eliminates lock waits in some
+					cases; note that this breaks
+					serializability. */
+	ulint		mysql_prefix_len;/* byte offset of the end of
+					the last requested column */
+	ulint		mysql_row_len;	/* length in bytes of a row in the
+					MySQL format */
+	ulint		n_rows_fetched;	/* number of rows fetched after
+					positioning the current cursor */
+	ulint		fetch_direction;/* ROW_SEL_NEXT or ROW_SEL_PREV */
+	byte*		fetch_cache[MYSQL_FETCH_CACHE_SIZE];
+					/* a cache for fetched rows if we
+					fetch many rows from the same cursor:
+					it saves CPU time to fetch them in a
+					batch; we reserve mysql_row_len
+					bytes for each such row; these
+					pointers point 4 bytes past the
+					allocated mem buf start, because
+					there is a 4 byte magic number at the
+					start and at the end */
+	ibool		keep_other_fields_on_keyread; /* when using fetch
+					cache with HA_EXTRA_KEYREAD, don't
+					overwrite other fields in mysql row
+					row buffer.*/
+	ulint		fetch_cache_first;/* position of the first not yet
+					fetched row in fetch_cache */
+	ulint		n_fetch_cached;	/* number of not yet fetched rows
+					in fetch_cache */
+	mem_heap_t*	blob_heap;	/* in SELECTS BLOB fields are copied
+					to this heap */
+	mem_heap_t*	old_vers_heap;	/* memory heap where a previous
+					version is built in consistent read */
+	/*----------------------*/
+	ulonglong	autoinc_last_value;/* last value of AUTO-INC interval */
+	ulonglong	autoinc_increment;/* The increment step of the auto 
+					increment column. Value must be
+					greater than or equal to 1. Required to
+					calculate the next value */
+	ulonglong	autoinc_offset; /* The offset passed to
+					get_auto_increment() by MySQL. Required
+					to calculate the next value */
+	ulint		autoinc_error;	/* The actual error code encountered
+					while trying to init or read the
+					autoinc value from the table. We
+					store it here so that we can return
+					it to MySQL */
+	/*----------------------*/
+	UT_LIST_NODE_T(row_prebuilt_t)	prebuilts;
+					/* list node of table->prebuilts */
+	ulint		magic_n2;	/* this should be the same as
+					magic_n */
+};
+
+#define ROW_PREBUILT_FETCH_MAGIC_N	465765687
+
+#define ROW_MYSQL_WHOLE_ROW	0
+#define ROW_MYSQL_REC_FIELDS	1
+#define ROW_MYSQL_NO_TEMPLATE	2
+#define ROW_MYSQL_DUMMY_TEMPLATE 3	/* dummy template used in
+					row_scan_and_check_index */
+
+/* Values for hint_need_to_fetch_extra_cols */
+#define ROW_RETRIEVE_PRIMARY_KEY	1
+#define ROW_RETRIEVE_ALL_COLS		2
+
+/* Values for row_read_type */
+#define ROW_READ_WITH_LOCKS		0
+#define ROW_READ_TRY_SEMI_CONSISTENT	1
+#define ROW_READ_DID_SEMI_CONSISTENT	2
+
+#ifndef UNIV_NONINL
+#include "row0mysql.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/row0mysql.ic b/storage/xtradb/include/row0mysql.ic
new file mode 100644
index 00000000000..5260ae17924
--- /dev/null
+++ b/storage/xtradb/include/row0mysql.ic
@@ -0,0 +1,23 @@
+/*****************************************************************************
+
+Copyright (c) 2001, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+MySQL interface for Innobase
+
+Created 1/23/2001 Heikki Tuuri
+*******************************************************/
diff --git a/storage/xtradb/include/row0purge.h b/storage/xtradb/include/row0purge.h
new file mode 100644
index 00000000000..fbc12f8d389
--- /dev/null
+++ b/storage/xtradb/include/row0purge.h
@@ -0,0 +1,95 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Purge obsolete records
+
+Created 3/14/1997 Heikki Tuuri
+*******************************************************/
+
+#ifndef row0purge_h
+#define row0purge_h
+
+#include "univ.i"
+#include "data0data.h"
+#include "btr0types.h"
+#include "btr0pcur.h"
+#include "dict0types.h"
+#include "trx0types.h"
+#include "que0types.h"
+#include "row0types.h"
+
+/************************************************************************
+Creates a purge node to a query graph. */
+UNIV_INTERN
+purge_node_t*
+row_purge_node_create(
+/*==================*/
+				/* out, own: purge node */
+	que_thr_t*	parent,	/* in: parent node, i.e., a thr node */
+	mem_heap_t*	heap);	/* in: memory heap where created */
+/***************************************************************
+Does the purge operation for a single undo log record. This is a high-level
+function used in an SQL execution graph. */
+UNIV_INTERN
+que_thr_t*
+row_purge_step(
+/*===========*/
+				/* out: query thread to run next or NULL */
+	que_thr_t*	thr);	/* in: query thread */
+
+/* Purge node structure */
+
+struct purge_node_struct{
+	que_common_t	common;	/* node type: QUE_NODE_PURGE */
+	/*----------------------*/
+	/* Local storage for this graph node */
+	dulint		roll_ptr;/* roll pointer to undo log record */
+	trx_undo_rec_t*	undo_rec;/* undo log record */
+	trx_undo_inf_t*	reservation;/* reservation for the undo log record in
+				the purge array */
+	dulint		undo_no;/* undo number of the record */
+	ulint		rec_type;/* undo log record type: TRX_UNDO_INSERT_REC,
+				... */
+	btr_pcur_t	pcur;	/* persistent cursor used in searching the
+				clustered index record */
+	ibool		found_clust;/* TRUE if the clustered index record
+				determined by ref was found in the clustered
+				index, and we were able to position pcur on
+				it */
+	dict_table_t*	table;	/* table where purge is done */
+	ulint		cmpl_info;/* compiler analysis info of an update */
+	upd_t*		update;	/* update vector for a clustered index
+				record */
+	dtuple_t*	ref;	/* NULL, or row reference to the next row to
+				handle */
+	dtuple_t*	row;	/* NULL, or a copy (also fields copied to
+				heap) of the indexed fields of the row to
+				handle */
+	dict_index_t*	index;	/* NULL, or the next index whose record should
+				be handled */
+	mem_heap_t*	heap;	/* memory heap used as auxiliary storage for
+				row; this must be emptied after a successful
+				purge of a row */
+};
+
+#ifndef UNIV_NONINL
+#include "row0purge.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/row0purge.ic b/storage/xtradb/include/row0purge.ic
new file mode 100644
index 00000000000..5fc665e9d20
--- /dev/null
+++ b/storage/xtradb/include/row0purge.ic
@@ -0,0 +1,24 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+
+/******************************************************
+Purge obsolete records
+
+Created 3/14/1997 Heikki Tuuri
+*******************************************************/
diff --git a/storage/xtradb/include/row0row.h b/storage/xtradb/include/row0row.h
new file mode 100644
index 00000000000..26c4b5e4e71
--- /dev/null
+++ b/storage/xtradb/include/row0row.h
@@ -0,0 +1,331 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+General row routines
+
+Created 4/20/1996 Heikki Tuuri
+*******************************************************/
+
+#ifndef row0row_h
+#define row0row_h
+
+#include "univ.i"
+#include "data0data.h"
+#include "dict0types.h"
+#include "trx0types.h"
+#include "que0types.h"
+#include "mtr0mtr.h"
+#include "rem0types.h"
+#include "read0types.h"
+#include "row0types.h"
+#include "btr0types.h"
+
+/*************************************************************************
+Gets the offset of the trx id field, in bytes relative to the origin of
+a clustered index record. */
+UNIV_INTERN
+ulint
+row_get_trx_id_offset(
+/*==================*/
+				/* out: offset of DATA_TRX_ID */
+	const rec_t*	rec,	/* in: record */
+	dict_index_t*	index,	/* in: clustered index */
+	const ulint*	offsets);/* in: rec_get_offsets(rec, index) */
+/*************************************************************************
+Reads the trx id field from a clustered index record. */
+UNIV_INLINE
+dulint
+row_get_rec_trx_id(
+/*===============*/
+				/* out: value of the field */
+	const rec_t*	rec,	/* in: record */
+	dict_index_t*	index,	/* in: clustered index */
+	const ulint*	offsets);/* in: rec_get_offsets(rec, index) */
+/*************************************************************************
+Reads the roll pointer field from a clustered index record. */
+UNIV_INLINE
+dulint
+row_get_rec_roll_ptr(
+/*=================*/
+				/* out: value of the field */
+	const rec_t*	rec,	/* in: record */
+	dict_index_t*	index,	/* in: clustered index */
+	const ulint*	offsets);/* in: rec_get_offsets(rec, index) */
+/*********************************************************************
+When an insert or purge to a table is performed, this function builds
+the entry to be inserted into or purged from an index on the table. */
+UNIV_INTERN
+dtuple_t*
+row_build_index_entry(
+/*==================*/
+				/* out: index entry which should be
+				inserted or purged, or NULL if the
+				externally stored columns in the
+				clustered index record are unavailable
+				and ext != NULL */
+	const dtuple_t*	row,	/* in: row which should be
+				inserted or purged */
+	row_ext_t*	ext,	/* in: externally stored column prefixes,
+				or NULL */
+	dict_index_t*	index,	/* in: index on the table */
+	mem_heap_t*	heap);	/* in: memory heap from which the memory for
+				the index entry is allocated */
+/***********************************************************************
+An inverse function to row_build_index_entry. Builds a row from a
+record in a clustered index. */
+UNIV_INTERN
+dtuple_t*
+row_build(
+/*======*/
+					/* out, own: row built;
+					see the NOTE below! */
+	ulint			type,	/* in: ROW_COPY_POINTERS or
+					ROW_COPY_DATA; the latter
+					copies also the data fields to
+					heap while the first only
+					places pointers to data fields
+					on the index page, and thus is
+					more efficient */
+	const dict_index_t*	index,	/* in: clustered index */
+	const rec_t*		rec,	/* in: record in the clustered
+					index; NOTE: in the case
+					ROW_COPY_POINTERS the data
+					fields in the row will point
+					directly into this record,
+					therefore, the buffer page of
+					this record must be at least
+					s-latched and the latch held
+					as long as the row dtuple is used! */
+	const ulint*		offsets,/* in: rec_get_offsets(rec,index)
+					or NULL, in which case this function
+					will invoke rec_get_offsets() */
+	const dict_table_t*	col_table,
+					/* in: table, to check which
+					externally stored columns
+					occur in the ordering columns
+					of an index, or NULL if
+					index->table should be
+					consulted instead; the user
+					columns in this table should be
+					the same columns as in index->table */
+	row_ext_t**		ext,	/* out, own: cache of
+					externally stored column
+					prefixes, or NULL */
+	mem_heap_t*		heap);	/* in: memory heap from which
+					the memory needed is allocated */
+/***********************************************************************
+Converts an index record to a typed data tuple. */
+UNIV_INTERN
+dtuple_t*
+row_rec_to_index_entry_low(
+/*=======================*/
+					/* out: index entry built; does not
+					set info_bits, and the data fields in
+					the entry will point directly to rec */
+	const rec_t*		rec,	/* in: record in the index */
+	const dict_index_t*	index,	/* in: index */
+	const ulint*		offsets,/* in: rec_get_offsets(rec, index) */
+	ulint*			n_ext,	/* out: number of externally
+					stored columns */
+	mem_heap_t*		heap);	/* in: memory heap from which
+					the memory needed is allocated */
+/***********************************************************************
+Converts an index record to a typed data tuple. NOTE that externally
+stored (often big) fields are NOT copied to heap. */
+UNIV_INTERN
+dtuple_t*
+row_rec_to_index_entry(
+/*===================*/
+					/* out, own: index entry
+					built; see the NOTE below! */
+	ulint			type,	/* in: ROW_COPY_DATA, or
+					ROW_COPY_POINTERS: the former
+					copies also the data fields to
+					heap as the latter only places
+					pointers to data fields on the
+					index page */
+	const rec_t*		rec,	/* in: record in the index;
+					NOTE: in the case
+					ROW_COPY_POINTERS the data
+					fields in the row will point
+					directly into this record,
+					therefore, the buffer page of
+					this record must be at least
+					s-latched and the latch held
+					as long as the dtuple is used! */
+	const dict_index_t*	index,	/* in: index */
+	ulint*			offsets,/* in/out: rec_get_offsets(rec) */
+	ulint*			n_ext,	/* out: number of externally
+					stored columns */
+	mem_heap_t*		heap);	/* in: memory heap from which
+					the memory needed is allocated */
+/***********************************************************************
+Builds from a secondary index record a row reference with which we can
+search the clustered index record. */
+UNIV_INTERN
+dtuple_t*
+row_build_row_ref(
+/*==============*/
+				/* out, own: row reference built; see the
+				NOTE below! */
+	ulint		type,	/* in: ROW_COPY_DATA, or ROW_COPY_POINTERS:
+				the former copies also the data fields to
+				heap, whereas the latter only places pointers
+				to data fields on the index page */
+	dict_index_t*	index,	/* in: secondary index */
+	const rec_t*	rec,	/* in: record in the index;
+				NOTE: in the case ROW_COPY_POINTERS
+				the data fields in the row will point
+				directly into this record, therefore,
+				the buffer page of this record must be
+				at least s-latched and the latch held
+				as long as the row reference is used! */
+	mem_heap_t*	heap);	/* in: memory heap from which the memory
+				needed is allocated */
+/***********************************************************************
+Builds from a secondary index record a row reference with which we can
+search the clustered index record. */
+UNIV_INTERN
+void
+row_build_row_ref_in_tuple(
+/*=======================*/
+	dtuple_t*		ref,	/* in/out: row reference built;
+					see the NOTE below! */
+	const rec_t*		rec,	/* in: record in the index;
+					NOTE: the data fields in ref
+					will point directly into this
+					record, therefore, the buffer
+					page of this record must be at
+					least s-latched and the latch
+					held as long as the row
+					reference is used! */
+	const dict_index_t*	index,	/* in: secondary index */
+	ulint*			offsets,/* in: rec_get_offsets(rec, index)
+					or NULL */
+	trx_t*			trx);	/* in: transaction */
+/***********************************************************************
+From a row build a row reference with which we can search the clustered
+index record. */
+UNIV_INTERN
+void
+row_build_row_ref_from_row(
+/*=======================*/
+	dtuple_t*		ref,	/* in/out: row reference built;
+					see the NOTE below!
+					ref must have the right number
+					of fields! */
+	const dict_table_t*	table,	/* in: table */
+	const dtuple_t*		row);	/* in: row
+					NOTE: the data fields in ref will point
+					directly into data of this row */
+/***********************************************************************
+Builds from a secondary index record a row reference with which we can
+search the clustered index record. */
+UNIV_INLINE
+void
+row_build_row_ref_fast(
+/*===================*/
+	dtuple_t*	ref,	/* in/out: typed data tuple where the
+				reference is built */
+	const ulint*	map,	/* in: array of field numbers in rec
+				telling how ref should be built from
+				the fields of rec */
+	const rec_t*	rec,	/* in: record in the index; must be
+				preserved while ref is used, as we do
+				not copy field values to heap */
+	const ulint*	offsets);/* in: array returned by rec_get_offsets() */
+/*******************************************************************
+Searches the clustered index record for a row, if we have the row
+reference. */
+UNIV_INTERN
+ibool
+row_search_on_row_ref(
+/*==================*/
+					/* out: TRUE if found */
+	btr_pcur_t*		pcur,	/* out: persistent cursor, which must
+					be closed by the caller */
+	ulint			mode,	/* in: BTR_MODIFY_LEAF, ... */
+	const dict_table_t*	table,	/* in: table */
+	const dtuple_t*		ref,	/* in: row reference */
+	mtr_t*			mtr);	/* in/out: mtr */
+/*************************************************************************
+Fetches the clustered index record for a secondary index record. The latches
+on the secondary index record are preserved. */
+UNIV_INTERN
+rec_t*
+row_get_clust_rec(
+/*==============*/
+				/* out: record or NULL, if no record found */
+	ulint		mode,	/* in: BTR_MODIFY_LEAF, ... */
+	const rec_t*	rec,	/* in: record in a secondary index */
+	dict_index_t*	index,	/* in: secondary index */
+	dict_index_t**	clust_index,/* out: clustered index */
+	mtr_t*		mtr);	/* in: mtr */
+/*******************************************************************
+Searches an index record. */
+UNIV_INTERN
+ibool
+row_search_index_entry(
+/*===================*/
+				/* out: TRUE if found */
+	dict_index_t*	index,	/* in: index */
+	const dtuple_t*	entry,	/* in: index entry */
+	ulint		mode,	/* in: BTR_MODIFY_LEAF, ... */
+	btr_pcur_t*	pcur,	/* in/out: persistent cursor, which must
+				be closed by the caller */
+	mtr_t*		mtr);	/* in: mtr */
+
+
+#define ROW_COPY_DATA		1
+#define ROW_COPY_POINTERS	2
+
+/* The allowed latching order of index records is the following:
+(1) a secondary index record ->
+(2) the clustered index record ->
+(3) rollback segment data for the clustered index record.
+
+No new latches may be obtained while the kernel mutex is reserved.
+However, the kernel mutex can be reserved while latches are owned. */
+
+/***********************************************************************
+Formats the raw data in "data" (in InnoDB on-disk format) using
+"dict_field" and writes the result to "buf".
+Not more than "buf_size" bytes are written to "buf".
+The result is always '\0'-terminated (provided buf_size > 0) and the
+number of bytes that were written to "buf" is returned (including the
+terminating '\0'). */
+UNIV_INTERN
+ulint
+row_raw_format(
+/*===========*/
+						/* out: number of bytes
+						that were written */
+	const char*		data,		/* in: raw data */
+	ulint			data_len,	/* in: raw data length
+						in bytes */
+	const dict_field_t*	dict_field,	/* in: index field */
+	char*			buf,		/* out: output buffer */
+	ulint			buf_size);	/* in: output buffer size
+						in bytes */
+
+#ifndef UNIV_NONINL
+#include "row0row.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/row0row.ic b/storage/xtradb/include/row0row.ic
new file mode 100644
index 00000000000..9947dd43257
--- /dev/null
+++ b/storage/xtradb/include/row0row.ic
@@ -0,0 +1,119 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+General row routines
+
+Created 4/20/1996 Heikki Tuuri
+*******************************************************/
+
+#include "dict0dict.h"
+#include "rem0rec.h"
+#include "trx0undo.h"
+
+/*************************************************************************
+Reads the trx id field from a clustered index record. */
+UNIV_INLINE
+dulint
+row_get_rec_trx_id(
+/*===============*/
+				/* out: value of the field */
+	const rec_t*	rec,	/* in: record */
+	dict_index_t*	index,	/* in: clustered index */
+	const ulint*	offsets)/* in: rec_get_offsets(rec, index) */
+{
+	ulint	offset;
+
+	ut_ad(dict_index_is_clust(index));
+	ut_ad(rec_offs_validate(rec, index, offsets));
+
+	offset = index->trx_id_offset;
+
+	if (!offset) {
+		offset = row_get_trx_id_offset(rec, index, offsets);
+	}
+
+	return(trx_read_trx_id(rec + offset));
+}
+
+/*************************************************************************
+Reads the roll pointer field from a clustered index record. */
+UNIV_INLINE
+dulint
+row_get_rec_roll_ptr(
+/*=================*/
+				/* out: value of the field */
+	const rec_t*	rec,	/* in: record */
+	dict_index_t*	index,	/* in: clustered index */
+	const ulint*	offsets)/* in: rec_get_offsets(rec, index) */
+{
+	ulint	offset;
+
+	ut_ad(dict_index_is_clust(index));
+	ut_ad(rec_offs_validate(rec, index, offsets));
+
+	offset = index->trx_id_offset;
+
+	if (!offset) {
+		offset = row_get_trx_id_offset(rec, index, offsets);
+	}
+
+	return(trx_read_roll_ptr(rec + offset + DATA_TRX_ID_LEN));
+}
+
+/***********************************************************************
+Builds from a secondary index record a row reference with which we can
+search the clustered index record. */
+UNIV_INLINE
+void
+row_build_row_ref_fast(
+/*===================*/
+	dtuple_t*	ref,	/* in/out: typed data tuple where the
+				reference is built */
+	const ulint*	map,	/* in: array of field numbers in rec
+				telling how ref should be built from
+				the fields of rec */
+	const rec_t*	rec,	/* in: record in the index; must be
+				preserved while ref is used, as we do
+				not copy field values to heap */
+	const ulint*	offsets)/* in: array returned by rec_get_offsets() */
+{
+	dfield_t*	dfield;
+	const byte*	field;
+	ulint		len;
+	ulint		ref_len;
+	ulint		field_no;
+	ulint		i;
+
+	ut_ad(rec_offs_validate(rec, NULL, offsets));
+	ut_ad(!rec_offs_any_extern(offsets));
+	ref_len = dtuple_get_n_fields(ref);
+
+	for (i = 0; i < ref_len; i++) {
+		dfield = dtuple_get_nth_field(ref, i);
+
+		field_no = *(map + i);
+
+		if (field_no != ULINT_UNDEFINED) {
+
+			field = rec_get_nth_field(rec, offsets,
+						  field_no, &len);
+			dfield_set_data(dfield, field, len);
+		}
+	}
+}
diff --git a/storage/xtradb/include/row0sel.h b/storage/xtradb/include/row0sel.h
new file mode 100644
index 00000000000..2f8574d0691
--- /dev/null
+++ b/storage/xtradb/include/row0sel.h
@@ -0,0 +1,401 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Select
+
+Created 12/19/1997 Heikki Tuuri
+*******************************************************/
+
+#ifndef row0sel_h
+#define row0sel_h
+
+#include "univ.i"
+#include "data0data.h"
+#include "que0types.h"
+#include "dict0types.h"
+#include "trx0types.h"
+#include "row0types.h"
+#include "que0types.h"
+#include "pars0sym.h"
+#include "btr0pcur.h"
+#include "read0read.h"
+#include "row0mysql.h"
+
+/*************************************************************************
+Creates a select node struct. */
+UNIV_INTERN
+sel_node_t*
+sel_node_create(
+/*============*/
+				/* out, own: select node struct */
+	mem_heap_t*	heap);	/* in: memory heap where created */
+/*************************************************************************
+Frees the memory private to a select node when a query graph is freed,
+does not free the heap where the node was originally created. */
+UNIV_INTERN
+void
+sel_node_free_private(
+/*==================*/
+	sel_node_t*	node);	/* in: select node struct */
+/*************************************************************************
+Frees a prefetch buffer for a column, including the dynamically allocated
+memory for data stored there. */
+UNIV_INTERN
+void
+sel_col_prefetch_buf_free(
+/*======================*/
+	sel_buf_t*	prefetch_buf);	/* in, own: prefetch buffer */
+/*************************************************************************
+Gets the plan node for the nth table in a join. */
+UNIV_INLINE
+plan_t*
+sel_node_get_nth_plan(
+/*==================*/
+	sel_node_t*	node,
+	ulint		i);
+/**************************************************************************
+Performs a select step. This is a high-level function used in SQL execution
+graphs. */
+UNIV_INTERN
+que_thr_t*
+row_sel_step(
+/*=========*/
+				/* out: query thread to run next or NULL */
+	que_thr_t*	thr);	/* in: query thread */
+/**************************************************************************
+Performs an execution step of an open or close cursor statement node. */
+UNIV_INLINE
+que_thr_t*
+open_step(
+/*======*/
+				/* out: query thread to run next or NULL */
+	que_thr_t*	thr);	/* in: query thread */
+/**************************************************************************
+Performs a fetch for a cursor. */
+UNIV_INTERN
+que_thr_t*
+fetch_step(
+/*=======*/
+				/* out: query thread to run next or NULL */
+	que_thr_t*	thr);	/* in: query thread */
+/********************************************************************
+Sample callback function for fetch that prints each row.*/
+UNIV_INTERN
+void*
+row_fetch_print(
+/*============*/
+				/* out: always returns non-NULL */
+	void*	row,		/* in:  sel_node_t* */
+	void*	user_arg);	/* in:  not used */
+/********************************************************************
+Callback function for fetch that stores an unsigned 4 byte integer to the
+location pointed. The column's type must be DATA_INT, DATA_UNSIGNED, length
+= 4. */
+UNIV_INTERN
+void*
+row_fetch_store_uint4(
+/*==================*/
+				/* out: always returns NULL */
+	void*	row,		/* in:  sel_node_t* */
+	void*	user_arg);	/* in:  data pointer */
+/***************************************************************
+Prints a row in a select result. */
+UNIV_INTERN
+que_thr_t*
+row_printf_step(
+/*============*/
+				/* out: query thread to run next or NULL */
+	que_thr_t*	thr);	/* in: query thread */
+/********************************************************************
+Converts a key value stored in MySQL format to an Innobase dtuple. The last
+field of the key value may be just a prefix of a fixed length field: hence
+the parameter key_len. But currently we do not allow search keys where the
+last field is only a prefix of the full key field len and print a warning if
+such appears. */
+UNIV_INTERN
+void
+row_sel_convert_mysql_key_to_innobase(
+/*==================================*/
+	dtuple_t*	tuple,		/* in/out: tuple where to build;
+					NOTE: we assume that the type info
+					in the tuple is already according
+					to index! */
+	byte*		buf,		/* in: buffer to use in field
+					conversions */
+	ulint		buf_len,	/* in: buffer length */
+	dict_index_t*	index,		/* in: index of the key value */
+	const byte*	key_ptr,	/* in: MySQL key value */
+	ulint		key_len,	/* in: MySQL key value length */
+	trx_t*		trx);		/* in: transaction */
+/************************************************************************
+Searches for rows in the database. This is used in the interface to
+MySQL. This function opens a cursor, and also implements fetch next
+and fetch prev. NOTE that if we do a search with a full key value
+from a unique index (ROW_SEL_EXACT), then we will not store the cursor
+position and fetch next or fetch prev must not be tried to the cursor! */
+UNIV_INTERN
+ulint
+row_search_for_mysql(
+/*=================*/
+					/* out: DB_SUCCESS,
+					DB_RECORD_NOT_FOUND,
+					DB_END_OF_INDEX, DB_DEADLOCK,
+					DB_LOCK_TABLE_FULL,
+					or DB_TOO_BIG_RECORD */
+	byte*		buf,		/* in/out: buffer for the fetched
+					row in the MySQL format */
+	ulint		mode,		/* in: search mode PAGE_CUR_L, ... */
+	row_prebuilt_t*	prebuilt,	/* in: prebuilt struct for the
+					table handle; this contains the info
+					of search_tuple, index; if search
+					tuple contains 0 fields then we
+					position the cursor at the start or
+					the end of the index, depending on
+					'mode' */
+	ulint		match_mode,	/* in: 0 or ROW_SEL_EXACT or
+					ROW_SEL_EXACT_PREFIX */
+	ulint		direction);	/* in: 0 or ROW_SEL_NEXT or
+					ROW_SEL_PREV; NOTE: if this is != 0,
+					then prebuilt must have a pcur
+					with stored position! In opening of a
+					cursor 'direction' should be 0. */
+/***********************************************************************
+Checks if MySQL at the moment is allowed for this table to retrieve a
+consistent read result, or store it to the query cache. */
+UNIV_INTERN
+ibool
+row_search_check_if_query_cache_permitted(
+/*======================================*/
+					/* out: TRUE if storing or retrieving
+					from the query cache is permitted */
+	trx_t*		trx,		/* in: transaction object */
+	const char*	norm_name);	/* in: concatenation of database name,
+					'/' char, table name */
+/***********************************************************************
+Read the max AUTOINC value from an index. */
+UNIV_INTERN
+ulint
+row_search_max_autoinc(
+/*===================*/
+					/* out: DB_SUCCESS if all OK else
+					error code */
+	dict_index_t*	index,		/* in: index to search */
+	const char*	col_name,	/* in: autoinc column name */
+	ib_uint64_t*	value);		/* out: AUTOINC value read */
+
+/* A structure for caching column values for prefetched rows */
+struct sel_buf_struct{
+	byte*		data;	/* data, or NULL; if not NULL, this field
+				has allocated memory which must be explicitly
+				freed; can be != NULL even when len is
+				UNIV_SQL_NULL */
+	ulint		len;	/* data length or UNIV_SQL_NULL */
+	ulint		val_buf_size;
+				/* size of memory buffer allocated for data:
+				this can be more than len; this is defined
+				when data != NULL */
+};
+
+struct plan_struct{
+	dict_table_t*	table;		/* table struct in the dictionary
+					cache */
+	dict_index_t*	index;		/* table index used in the search */
+	btr_pcur_t	pcur;		/* persistent cursor used to search
+					the index */
+	ibool		asc;		/* TRUE if cursor traveling upwards */
+	ibool		pcur_is_open;	/* TRUE if pcur has been positioned
+					and we can try to fetch new rows */
+	ibool		cursor_at_end;	/* TRUE if the cursor is open but
+					we know that there are no more
+					qualifying rows left to retrieve from
+					the index tree; NOTE though, that
+					there may still be unprocessed rows in
+					the prefetch stack; always FALSE when
+					pcur_is_open is FALSE */
+	ibool		stored_cursor_rec_processed;
+					/* TRUE if the pcur position has been
+					stored and the record it is positioned
+					on has already been processed */
+	que_node_t**	tuple_exps;	/* array of expressions which are used
+					to calculate the field values in the
+					search tuple: there is one expression
+					for each field in the search tuple */
+	dtuple_t*	tuple;		/* search tuple */
+	ulint		mode;		/* search mode: PAGE_CUR_G, ... */
+	ulint		n_exact_match;	/* number of first fields in the search
+					tuple which must be exactly matched */
+	ibool		unique_search;	/* TRUE if we are searching an
+					index record with a unique key */
+	ulint		n_rows_fetched;	/* number of rows fetched using pcur
+					after it was opened */
+	ulint		n_rows_prefetched;/* number of prefetched rows cached
+					for fetch: fetching several rows in
+					the same mtr saves CPU time */
+	ulint		first_prefetched;/* index of the first cached row in
+					select buffer arrays for each column */
+	ibool		no_prefetch;	/* no prefetch for this table */
+	sym_node_list_t	columns;	/* symbol table nodes for the columns
+					to retrieve from the table */
+	UT_LIST_BASE_NODE_T(func_node_t)
+			end_conds;	/* conditions which determine the
+					fetch limit of the index segment we
+					have to look at: when one of these
+					fails, the result set has been
+					exhausted for the cursor in this
+					index; these conditions are normalized
+					so that in a comparison the column
+					for this table is the first argument */
+	UT_LIST_BASE_NODE_T(func_node_t)
+			other_conds;	/* the rest of search conditions we can
+					test at this table in a join */
+	ibool		must_get_clust;	/* TRUE if index is a non-clustered
+					index and we must also fetch the
+					clustered index record; this is the
+					case if the non-clustered record does
+					not contain all the needed columns, or
+					if this is a single-table explicit
+					cursor, or a searched update or
+					delete */
+	ulint*		clust_map;	/* map telling how clust_ref is built
+					from the fields of a non-clustered
+					record */
+	dtuple_t*	clust_ref;	/* the reference to the clustered
+					index entry is built here if index is
+					a non-clustered index */
+	btr_pcur_t	clust_pcur;	/* if index is non-clustered, we use
+					this pcur to search the clustered
+					index */
+	mem_heap_t*	old_vers_heap;	/* memory heap used in building an old
+					version of a row, or NULL */
+};
+
+struct sel_node_struct{
+	que_common_t	common;		/* node type: QUE_NODE_SELECT */
+	ulint		state;		/* node state */
+	que_node_t*	select_list;	/* select list */
+	sym_node_t*	into_list;	/* variables list or NULL */
+	sym_node_t*	table_list;	/* table list */
+	ibool		asc;		/* TRUE if the rows should be fetched
+					in an ascending order */
+	ibool		set_x_locks;	/* TRUE if the cursor is for update or
+					delete, which means that a row x-lock
+					should be placed on the cursor row */
+	ulint		row_lock_mode;	/* LOCK_X or LOCK_S */
+	ulint		n_tables;	/* number of tables */
+	ulint		fetch_table;	/* number of the next table to access
+					in the join */
+	plan_t*		plans;		/* array of n_tables many plan nodes
+					containing the search plan and the
+					search data structures */
+	que_node_t*	search_cond;	/* search condition */
+	read_view_t*	read_view;	/* if the query is a non-locking
+					consistent read, its read view is
+					placed here, otherwise NULL */
+	ibool		consistent_read;/* TRUE if the select is a consistent,
+					non-locking read */
+	order_node_t*	order_by;	/* order by column definition, or
+					NULL */
+	ibool		is_aggregate;	/* TRUE if the select list consists of
+					aggregate functions */
+	ibool		aggregate_already_fetched;
+					/* TRUE if the aggregate row has
+					already been fetched for the current
+					cursor */
+	ibool		can_get_updated;/* this is TRUE if the select
+					is in a single-table explicit
+					cursor which can get updated
+					within the stored procedure,
+					or in a searched update or
+					delete; NOTE that to determine
+					of an explicit cursor if it
+					can get updated, the parser
+					checks from a stored procedure
+					if it contains positioned
+					update or delete statements */
+	sym_node_t*	explicit_cursor;/* not NULL if an explicit cursor */
+	UT_LIST_BASE_NODE_T(sym_node_t)
+			copy_variables; /* variables whose values we have to
+					copy when an explicit cursor is opened,
+					so that they do not change between
+					fetches */
+};
+
+/* Select node states */
+#define	SEL_NODE_CLOSED		0	/* it is a declared cursor which is not
+					currently open */
+#define SEL_NODE_OPEN		1	/* intention locks not yet set on
+					tables */
+#define SEL_NODE_FETCH		2	/* intention locks have been set */
+#define SEL_NODE_NO_MORE_ROWS	3	/* cursor has reached the result set
+					end */
+
+/* Fetch statement node */
+struct fetch_node_struct{
+	que_common_t	common;		/* type: QUE_NODE_FETCH */
+	sel_node_t*	cursor_def;	/* cursor definition */
+	sym_node_t*	into_list;	/* variables to set */
+
+	pars_user_func_t*
+			func;		/* User callback function or NULL.
+					The first argument to the function
+					is a sel_node_t*, containing the
+					results of the SELECT operation for
+					one row. If the function returns
+					NULL, it is not interested in
+					further rows and the cursor is
+					modified so (cursor % NOTFOUND) is
+					true. If it returns not-NULL,
+					continue normally. See
+					row_fetch_print() for an example
+					(and a useful debugging tool). */
+};
+
+/* Open or close cursor statement node */
+struct open_node_struct{
+	que_common_t	common;		/* type: QUE_NODE_OPEN */
+	ulint		op_type;	/* ROW_SEL_OPEN_CURSOR or
+					ROW_SEL_CLOSE_CURSOR */
+	sel_node_t*	cursor_def;	/* cursor definition */
+};
+
+/* Row printf statement node */
+struct row_printf_node_struct{
+	que_common_t	common;		/* type: QUE_NODE_ROW_PRINTF */
+	sel_node_t*	sel_node;	/* select */
+};
+
+#define ROW_SEL_OPEN_CURSOR	0
+#define ROW_SEL_CLOSE_CURSOR	1
+
+/* Flags for the MySQL interface */
+#define ROW_SEL_NEXT		1
+#define ROW_SEL_PREV		2
+
+#define ROW_SEL_EXACT		1	/* search using a complete key value */
+#define ROW_SEL_EXACT_PREFIX	2	/* search using a key prefix which
+					must match to rows: the prefix may
+					contain an incomplete field (the
+					last field in prefix may be just
+					a prefix of a fixed length column) */
+
+#ifndef UNIV_NONINL
+#include "row0sel.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/row0sel.ic b/storage/xtradb/include/row0sel.ic
new file mode 100644
index 00000000000..a21181e3237
--- /dev/null
+++ b/storage/xtradb/include/row0sel.ic
@@ -0,0 +1,104 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Select
+
+Created 12/19/1997 Heikki Tuuri
+*******************************************************/
+
+#include "que0que.h"
+
+/*************************************************************************
+Gets the plan node for the nth table in a join. */
+UNIV_INLINE
+plan_t*
+sel_node_get_nth_plan(
+/*==================*/
+				/* out: plan node */
+	sel_node_t*	node,	/* in: select node */
+	ulint		i)	/* in: get ith plan node */
+{
+	ut_ad(i < node->n_tables);
+
+	return(node->plans + i);
+}
+
+/*************************************************************************
+Resets the cursor defined by sel_node to the SEL_NODE_OPEN state, which means
+that it will start fetching from the start of the result set again, regardless
+of where it was before, and it will set intention locks on the tables. */
+UNIV_INLINE
+void
+sel_node_reset_cursor(
+/*==================*/
+	sel_node_t*	node)	/* in: select node */
+{
+	node->state = SEL_NODE_OPEN;
+}
+
+/**************************************************************************
+Performs an execution step of an open or close cursor statement node. */
+UNIV_INLINE
+que_thr_t*
+open_step(
+/*======*/
+				/* out: query thread to run next or NULL */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	sel_node_t*	sel_node;
+	open_node_t*	node;
+	ulint		err;
+
+	ut_ad(thr);
+
+	node = (open_node_t*) thr->run_node;
+	ut_ad(que_node_get_type(node) == QUE_NODE_OPEN);
+
+	sel_node = node->cursor_def;
+
+	err = DB_SUCCESS;
+
+	if (node->op_type == ROW_SEL_OPEN_CURSOR) {
+
+		/*		if (sel_node->state == SEL_NODE_CLOSED) { */
+
+		sel_node_reset_cursor(sel_node);
+		/*		} else {
+		err = DB_ERROR;
+		} */
+	} else {
+		if (sel_node->state != SEL_NODE_CLOSED) {
+
+			sel_node->state = SEL_NODE_CLOSED;
+		} else {
+			err = DB_ERROR;
+		}
+	}
+
+	if (UNIV_EXPECT(err, DB_SUCCESS) != DB_SUCCESS) {
+		/* SQL error detected */
+		fprintf(stderr, "SQL error %lu\n", (ulong) err);
+
+		ut_error;
+	}
+
+	thr->run_node = que_node_get_parent(node);
+
+	return(thr);
+}
diff --git a/storage/xtradb/include/row0types.h b/storage/xtradb/include/row0types.h
new file mode 100644
index 00000000000..f0af7c2bf53
--- /dev/null
+++ b/storage/xtradb/include/row0types.h
@@ -0,0 +1,58 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Row operation global types
+
+Created 12/27/1996 Heikki Tuuri
+*******************************************************/
+
+#ifndef row0types_h
+#define row0types_h
+
+typedef struct plan_struct plan_t;
+
+typedef	struct upd_struct upd_t;
+
+typedef struct upd_field_struct upd_field_t;
+
+typedef	struct upd_node_struct upd_node_t;
+
+typedef	struct del_node_struct del_node_t;
+
+typedef	struct ins_node_struct ins_node_t;
+
+typedef struct sel_node_struct	sel_node_t;
+
+typedef struct open_node_struct	open_node_t;
+
+typedef struct fetch_node_struct fetch_node_t;
+
+typedef struct row_printf_node_struct	row_printf_node_t;
+typedef struct sel_buf_struct	sel_buf_t;
+
+typedef	struct undo_node_struct undo_node_t;
+
+typedef	struct purge_node_struct purge_node_t;
+
+typedef struct row_ext_struct row_ext_t;
+
+/* MySQL data types */
+typedef struct st_table TABLE;
+
+#endif
diff --git a/storage/xtradb/include/row0uins.h b/storage/xtradb/include/row0uins.h
new file mode 100644
index 00000000000..16bbbbd0d12
--- /dev/null
+++ b/storage/xtradb/include/row0uins.h
@@ -0,0 +1,53 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Fresh insert undo
+
+Created 2/25/1997 Heikki Tuuri
+*******************************************************/
+
+#ifndef row0uins_h
+#define row0uins_h
+
+#include "univ.i"
+#include "data0data.h"
+#include "dict0types.h"
+#include "trx0types.h"
+#include "que0types.h"
+#include "row0types.h"
+#include "mtr0mtr.h"
+
+/***************************************************************
+Undoes a fresh insert of a row to a table. A fresh insert means that
+the same clustered index unique key did not have any record, even delete
+marked, at the time of the insert.  InnoDB is eager in a rollback:
+if it figures out that an index record will be removed in the purge
+anyway, it will remove it in the rollback. */
+UNIV_INTERN
+ulint
+row_undo_ins(
+/*=========*/
+				/* out: DB_SUCCESS */
+	undo_node_t*	node);	/* in: row undo node */
+
+#ifndef UNIV_NONINL
+#include "row0uins.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/row0uins.ic b/storage/xtradb/include/row0uins.ic
new file mode 100644
index 00000000000..75bef8431eb
--- /dev/null
+++ b/storage/xtradb/include/row0uins.ic
@@ -0,0 +1,24 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Fresh insert undo
+
+Created 2/25/1997 Heikki Tuuri
+*******************************************************/
+
diff --git a/storage/xtradb/include/row0umod.h b/storage/xtradb/include/row0umod.h
new file mode 100644
index 00000000000..3a4e8c2f9a3
--- /dev/null
+++ b/storage/xtradb/include/row0umod.h
@@ -0,0 +1,51 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Undo modify of a row
+
+Created 2/27/1997 Heikki Tuuri
+*******************************************************/
+
+#ifndef row0umod_h
+#define row0umod_h
+
+#include "univ.i"
+#include "data0data.h"
+#include "dict0types.h"
+#include "trx0types.h"
+#include "que0types.h"
+#include "row0types.h"
+#include "mtr0mtr.h"
+
+/***************************************************************
+Undoes a modify operation on a row of a table. */
+UNIV_INTERN
+ulint
+row_undo_mod(
+/*=========*/
+				/* out: DB_SUCCESS or error code */
+	undo_node_t*	node,	/* in: row undo node */
+	que_thr_t*	thr);	/* in: query thread */
+
+
+#ifndef UNIV_NONINL
+#include "row0umod.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/row0umod.ic b/storage/xtradb/include/row0umod.ic
new file mode 100644
index 00000000000..7ac7bc2fea7
--- /dev/null
+++ b/storage/xtradb/include/row0umod.ic
@@ -0,0 +1,23 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Undo modify of a row
+
+Created 2/27/1997 Heikki Tuuri
+*******************************************************/
diff --git a/storage/xtradb/include/row0undo.h b/storage/xtradb/include/row0undo.h
new file mode 100644
index 00000000000..a17cfb1babd
--- /dev/null
+++ b/storage/xtradb/include/row0undo.h
@@ -0,0 +1,136 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Row undo
+
+Created 1/8/1997 Heikki Tuuri
+*******************************************************/
+
+#ifndef row0undo_h
+#define row0undo_h
+
+#include "univ.i"
+#include "mtr0mtr.h"
+#include "trx0sys.h"
+#include "btr0types.h"
+#include "btr0pcur.h"
+#include "dict0types.h"
+#include "trx0types.h"
+#include "que0types.h"
+#include "row0types.h"
+
+/************************************************************************
+Creates a row undo node to a query graph. */
+UNIV_INTERN
+undo_node_t*
+row_undo_node_create(
+/*=================*/
+				/* out, own: undo node */
+	trx_t*		trx,	/* in: transaction */
+	que_thr_t*	parent,	/* in: parent node, i.e., a thr node */
+	mem_heap_t*	heap);	/* in: memory heap where created */
+/***************************************************************
+Looks for the clustered index record when node has the row reference.
+The pcur in node is used in the search. If found, stores the row to node,
+and stores the position of pcur, and detaches it. The pcur must be closed
+by the caller in any case. */
+UNIV_INTERN
+ibool
+row_undo_search_clust_to_pcur(
+/*==========================*/
+				/* out: TRUE if found; NOTE the node->pcur
+				must be closed by the caller, regardless of
+				the return value */
+	undo_node_t*	node);	/* in: row undo node */
+/***************************************************************
+Undoes a row operation in a table. This is a high-level function used
+in SQL execution graphs. */
+UNIV_INTERN
+que_thr_t*
+row_undo_step(
+/*==========*/
+				/* out: query thread to run next or NULL */
+	que_thr_t*	thr);	/* in: query thread */
+
+/* A single query thread will try to perform the undo for all successive
+versions of a clustered index record, if the transaction has modified it
+several times during the execution which is rolled back. It may happen
+that the task is transferred to another query thread, if the other thread
+is assigned to handle an undo log record in the chain of different versions
+of the record, and the other thread happens to get the x-latch to the
+clustered index record at the right time.
+	If a query thread notices that the clustered index record it is looking
+for is missing, or the roll ptr field in the record doed not point to the
+undo log record the thread was assigned to handle, then it gives up the undo
+task for that undo log record, and fetches the next. This situation can occur
+just in the case where the transaction modified the same record several times
+and another thread is currently doing the undo for successive versions of
+that index record. */
+
+/* Undo node structure */
+
+struct undo_node_struct{
+	que_common_t	common;	/* node type: QUE_NODE_UNDO */
+	ulint		state;	/* node execution state */
+	trx_t*		trx;	/* trx for which undo is done */
+	dulint		roll_ptr;/* roll pointer to undo log record */
+	trx_undo_rec_t*	undo_rec;/* undo log record */
+	dulint		undo_no;/* undo number of the record */
+	ulint		rec_type;/* undo log record type: TRX_UNDO_INSERT_REC,
+				... */
+	dulint		new_roll_ptr; /* roll ptr to restore to clustered index
+				record */
+	dulint		new_trx_id; /* trx id to restore to clustered index
+				record */
+	btr_pcur_t	pcur;	/* persistent cursor used in searching the
+				clustered index record */
+	dict_table_t*	table;	/* table where undo is done */
+	ulint		cmpl_info;/* compiler analysis of an update */
+	upd_t*		update;	/* update vector for a clustered index
+				record */
+	dtuple_t*	ref;	/* row reference to the next row to handle */
+	dtuple_t*	row;	/* a copy (also fields copied to heap) of the
+				row to handle */
+	row_ext_t*	ext;	/* NULL, or prefixes of the externally
+				stored columns of the row */
+	dtuple_t*	undo_row;/* NULL, or the row after undo */
+	row_ext_t*	undo_ext;/* NULL, or prefixes of the externally
+				stored columns of undo_row */
+	dict_index_t*	index;	/* the next index whose record should be
+				handled */
+	mem_heap_t*	heap;	/* memory heap used as auxiliary storage for
+				row; this must be emptied after undo is tried
+				on a row */
+};
+
+/* Execution states for an undo node */
+#define	UNDO_NODE_FETCH_NEXT	1	/* we should fetch the next undo log
+					record */
+#define	UNDO_NODE_PREV_VERS	2	/* the roll ptr to previous version of
+					a row is stored in node, and undo
+					should be done based on it */
+#define UNDO_NODE_INSERT	3
+#define UNDO_NODE_MODIFY	4
+
+
+#ifndef UNIV_NONINL
+#include "row0undo.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/row0undo.ic b/storage/xtradb/include/row0undo.ic
new file mode 100644
index 00000000000..921e3633b10
--- /dev/null
+++ b/storage/xtradb/include/row0undo.ic
@@ -0,0 +1,23 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Row undo
+
+Created 1/8/1997 Heikki Tuuri
+*******************************************************/
diff --git a/storage/xtradb/include/row0upd.h b/storage/xtradb/include/row0upd.h
new file mode 100644
index 00000000000..71aa20d158c
--- /dev/null
+++ b/storage/xtradb/include/row0upd.h
@@ -0,0 +1,475 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Update of a row
+
+Created 12/27/1996 Heikki Tuuri
+*******************************************************/
+
+#ifndef row0upd_h
+#define row0upd_h
+
+#include "univ.i"
+#include "data0data.h"
+#include "btr0types.h"
+#include "btr0pcur.h"
+#include "dict0types.h"
+#include "trx0types.h"
+#include "que0types.h"
+#include "row0types.h"
+#include "pars0types.h"
+
+/*************************************************************************
+Creates an update vector object. */
+UNIV_INLINE
+upd_t*
+upd_create(
+/*=======*/
+				/* out, own: update vector object */
+	ulint		n,	/* in: number of fields */
+	mem_heap_t*	heap);	/* in: heap from which memory allocated */
+/*************************************************************************
+Returns the number of fields in the update vector == number of columns
+to be updated by an update vector. */
+UNIV_INLINE
+ulint
+upd_get_n_fields(
+/*=============*/
+					/* out: number of fields */
+	const upd_t*	update);	/* in: update vector */
+#ifdef UNIV_DEBUG
+/*************************************************************************
+Returns the nth field of an update vector. */
+UNIV_INLINE
+upd_field_t*
+upd_get_nth_field(
+/*==============*/
+				/* out: update vector field */
+	const upd_t*	update,	/* in: update vector */
+	ulint		n);	/* in: field position in update vector */
+#else
+# define upd_get_nth_field(update, n) ((update)->fields + (n))
+#endif
+/*************************************************************************
+Sets an index field number to be updated by an update vector field. */
+UNIV_INLINE
+void
+upd_field_set_field_no(
+/*===================*/
+	upd_field_t*	upd_field,	/* in: update vector field */
+	ulint		field_no,	/* in: field number in a clustered
+					index */
+	dict_index_t*	index,		/* in: index */
+	trx_t*		trx);		/* in: transaction */
+/*************************************************************************
+Returns a field of an update vector by field_no. */
+UNIV_INLINE
+const upd_field_t*
+upd_get_field_by_field_no(
+/*======================*/
+				/* out: update vector field, or NULL */
+	const upd_t*	update,	/* in: update vector */
+	ulint		no)	/* in: field_no */
+	__attribute__((nonnull, pure));
+/*************************************************************************
+Writes into the redo log the values of trx id and roll ptr and enough info
+to determine their positions within a clustered index record. */
+UNIV_INTERN
+byte*
+row_upd_write_sys_vals_to_log(
+/*==========================*/
+				/* out: new pointer to mlog */
+	dict_index_t*	index,	/* in: clustered index */
+	trx_t*		trx,	/* in: transaction */
+	dulint		roll_ptr,/* in: roll ptr of the undo log record */
+	byte*		log_ptr,/* pointer to a buffer of size > 20 opened
+				in mlog */
+	mtr_t*		mtr);	/* in: mtr */
+/*************************************************************************
+Updates the trx id and roll ptr field in a clustered index record when
+a row is updated or marked deleted. */
+UNIV_INLINE
+void
+row_upd_rec_sys_fields(
+/*===================*/
+	rec_t*		rec,	/* in/out: record */
+	page_zip_des_t*	page_zip,/* in/out: compressed page whose
+				uncompressed part will be updated, or NULL */
+	dict_index_t*	index,	/* in: clustered index */
+	const ulint*	offsets,/* in: rec_get_offsets(rec, index) */
+	trx_t*		trx,	/* in: transaction */
+	dulint		roll_ptr);/* in: roll ptr of the undo log record */
+/*************************************************************************
+Sets the trx id or roll ptr field of a clustered index entry. */
+UNIV_INTERN
+void
+row_upd_index_entry_sys_field(
+/*==========================*/
+	const dtuple_t*	entry,	/* in: index entry, where the memory buffers
+				for sys fields are already allocated:
+				the function just copies the new values to
+				them */
+	dict_index_t*	index,	/* in: clustered index */
+	ulint		type,	/* in: DATA_TRX_ID or DATA_ROLL_PTR */
+	dulint		val);	/* in: value to write */
+/*************************************************************************
+Creates an update node for a query graph. */
+UNIV_INTERN
+upd_node_t*
+upd_node_create(
+/*============*/
+				/* out, own: update node */
+	mem_heap_t*	heap);	/* in: mem heap where created */
+/***************************************************************
+Writes to the redo log the new values of the fields occurring in the index. */
+UNIV_INTERN
+void
+row_upd_index_write_log(
+/*====================*/
+	const upd_t*	update,	/* in: update vector */
+	byte*		log_ptr,/* in: pointer to mlog buffer: must
+				contain at least MLOG_BUF_MARGIN bytes
+				of free space; the buffer is closed
+				within this function */
+	mtr_t*		mtr);	/* in: mtr into whose log to write */
+/***************************************************************
+Returns TRUE if row update changes size of some field in index or if some
+field to be updated is stored externally in rec or update. */
+UNIV_INTERN
+ibool
+row_upd_changes_field_size_or_external(
+/*===================================*/
+				/* out: TRUE if the update changes the size of
+				some field in index or the field is external
+				in rec or update */
+	dict_index_t*	index,	/* in: index */
+	const ulint*	offsets,/* in: rec_get_offsets(rec, index) */
+	const upd_t*	update);/* in: update vector */
+/***************************************************************
+Replaces the new column values stored in the update vector to the record
+given. No field size changes are allowed. */
+UNIV_INTERN
+void
+row_upd_rec_in_place(
+/*=================*/
+	rec_t*		rec,	/* in/out: record where replaced */
+	dict_index_t*	index,	/* in: the index the record belongs to */
+	const ulint*	offsets,/* in: array returned by rec_get_offsets() */
+	const upd_t*	update,	/* in: update vector */
+	page_zip_des_t*	page_zip);/* in: compressed page with enough space
+				available, or NULL */
+/*******************************************************************
+Builds an update vector from those fields which in a secondary index entry
+differ from a record that has the equal ordering fields. NOTE: we compare
+the fields as binary strings! */
+UNIV_INTERN
+upd_t*
+row_upd_build_sec_rec_difference_binary(
+/*====================================*/
+				/* out, own: update vector of differing
+				fields */
+	dict_index_t*	index,	/* in: index */
+	const dtuple_t*	entry,	/* in: entry to insert */
+	const rec_t*	rec,	/* in: secondary index record */
+	trx_t*		trx,	/* in: transaction */
+	mem_heap_t*	heap);	/* in: memory heap from which allocated */
+/*******************************************************************
+Builds an update vector from those fields, excluding the roll ptr and
+trx id fields, which in an index entry differ from a record that has
+the equal ordering fields. NOTE: we compare the fields as binary strings! */
+UNIV_INTERN
+upd_t*
+row_upd_build_difference_binary(
+/*============================*/
+				/* out, own: update vector of differing
+				fields, excluding roll ptr and trx id */
+	dict_index_t*	index,	/* in: clustered index */
+	const dtuple_t*	entry,	/* in: entry to insert */
+	const rec_t*	rec,	/* in: clustered index record */
+	trx_t*		trx,	/* in: transaction */
+	mem_heap_t*	heap);	/* in: memory heap from which allocated */
+/***************************************************************
+Replaces the new column values stored in the update vector to the index entry
+given. */
+UNIV_INTERN
+void
+row_upd_index_replace_new_col_vals_index_pos(
+/*=========================================*/
+	dtuple_t*	entry,	/* in/out: index entry where replaced;
+				the clustered index record must be
+				covered by a lock or a page latch to
+				prevent deletion (rollback or purge) */
+	dict_index_t*	index,	/* in: index; NOTE that this may also be a
+				non-clustered index */
+	const upd_t*	update,	/* in: an update vector built for the index so
+				that the field number in an upd_field is the
+				index position */
+	ibool		order_only,
+				/* in: if TRUE, limit the replacement to
+				ordering fields of index; note that this
+				does not work for non-clustered indexes. */
+	mem_heap_t*	heap)	/* in: memory heap for allocating and
+				copying the new values */
+	__attribute__((nonnull));
+/***************************************************************
+Replaces the new column values stored in the update vector to the index entry
+given. */
+UNIV_INTERN
+void
+row_upd_index_replace_new_col_vals(
+/*===============================*/
+	dtuple_t*	entry,	/* in/out: index entry where replaced;
+				the clustered index record must be
+				covered by a lock or a page latch to
+				prevent deletion (rollback or purge) */
+	dict_index_t*	index,	/* in: index; NOTE that this may also be a
+				non-clustered index */
+	const upd_t*	update,	/* in: an update vector built for the
+				CLUSTERED index so that the field number in
+				an upd_field is the clustered index position */
+	mem_heap_t*	heap)	/* in: memory heap for allocating and
+				copying the new values */
+	__attribute__((nonnull));
+/***************************************************************
+Replaces the new column values stored in the update vector. */
+UNIV_INTERN
+void
+row_upd_replace(
+/*============*/
+	dtuple_t*		row,	/* in/out: row where replaced,
+					indexed by col_no;
+					the clustered index record must be
+					covered by a lock or a page latch to
+					prevent deletion (rollback or purge) */
+	row_ext_t**		ext,	/* out, own: NULL, or externally
+					stored column prefixes */
+	const dict_index_t*	index,	/* in: clustered index */
+	const upd_t*		update,	/* in: an update vector built for the
+					clustered index */
+	mem_heap_t*		heap);	/* in: memory heap */
+/***************************************************************
+Checks if an update vector changes an ordering field of an index record.
+This function is fast if the update vector is short or the number of ordering
+fields in the index is small. Otherwise, this can be quadratic.
+NOTE: we compare the fields as binary strings! */
+UNIV_INTERN
+ibool
+row_upd_changes_ord_field_binary(
+/*=============================*/
+				/* out: TRUE if update vector changes
+				an ordering field in the index record;
+				NOTE: the fields are compared as binary
+				strings */
+	const dtuple_t*	row,	/* in: old value of row, or NULL if the
+				row and the data values in update are not
+				known when this function is called, e.g., at
+				compile time */
+	dict_index_t*	index,	/* in: index of the record */
+	const upd_t*	update);/* in: update vector for the row; NOTE: the
+				field numbers in this MUST be clustered index
+				positions! */
+/***************************************************************
+Checks if an update vector changes an ordering field of an index record.
+This function is fast if the update vector is short or the number of ordering
+fields in the index is small. Otherwise, this can be quadratic.
+NOTE: we compare the fields as binary strings! */
+UNIV_INTERN
+ibool
+row_upd_changes_some_index_ord_field_binary(
+/*========================================*/
+					/* out: TRUE if update vector
+					may change an ordering field
+					in an index record */
+	const dict_table_t*	table,	/* in: table */
+	const upd_t*		update);/* in: update vector for the row */
+/***************************************************************
+Updates a row in a table. This is a high-level function used
+in SQL execution graphs. */
+UNIV_INTERN
+que_thr_t*
+row_upd_step(
+/*=========*/
+				/* out: query thread to run next or NULL */
+	que_thr_t*	thr);	/* in: query thread */
+/*************************************************************************
+Parses the log data of system field values. */
+UNIV_INTERN
+byte*
+row_upd_parse_sys_vals(
+/*===================*/
+			/* out: log data end or NULL */
+	byte*	ptr,	/* in: buffer */
+	byte*	end_ptr,/* in: buffer end */
+	ulint*	pos,	/* out: TRX_ID position in record */
+	dulint*	trx_id,	/* out: trx id */
+	dulint*	roll_ptr);/* out: roll ptr */
+/*************************************************************************
+Updates the trx id and roll ptr field in a clustered index record in database
+recovery. */
+UNIV_INTERN
+void
+row_upd_rec_sys_fields_in_recovery(
+/*===============================*/
+	rec_t*		rec,	/* in/out: record */
+	page_zip_des_t*	page_zip,/* in/out: compressed page, or NULL */
+	const ulint*	offsets,/* in: array returned by rec_get_offsets() */
+	ulint		pos,	/* in: TRX_ID position in rec */
+	dulint		trx_id,	/* in: transaction id */
+	dulint		roll_ptr);/* in: roll ptr of the undo log record */
+/*************************************************************************
+Parses the log data written by row_upd_index_write_log. */
+UNIV_INTERN
+byte*
+row_upd_index_parse(
+/*================*/
+				/* out: log data end or NULL */
+	byte*		ptr,	/* in: buffer */
+	byte*		end_ptr,/* in: buffer end */
+	mem_heap_t*	heap,	/* in: memory heap where update vector is
+				built */
+	upd_t**		update_out);/* out: update vector */
+
+
+/* Update vector field */
+struct upd_field_struct{
+	unsigned	field_no:16;	/* field number in an index, usually
+					the clustered index, but in updating
+					a secondary index record in btr0cur.c
+					this is the position in the secondary
+					index */
+	unsigned	orig_len:16;	/* original length of the locally
+					stored part of an externally stored
+					column, or 0 */
+	que_node_t*	exp;		/* expression for calculating a new
+					value: it refers to column values and
+					constants in the symbol table of the
+					query graph */
+	dfield_t	new_val;	/* new value for the column */
+};
+
+/* Update vector structure */
+struct upd_struct{
+	ulint		info_bits;	/* new value of info bits to record;
+					default is 0 */
+	ulint		n_fields;	/* number of update fields */
+	upd_field_t*	fields;		/* array of update fields */
+};
+
+/* Update node structure which also implements the delete operation
+of a row */
+
+struct upd_node_struct{
+	que_common_t	common;	/* node type: QUE_NODE_UPDATE */
+	ibool		is_delete;/* TRUE if delete, FALSE if update */
+	ibool		searched_update;
+				/* TRUE if searched update, FALSE if
+				positioned */
+	ibool		in_mysql_interface;
+				/* TRUE if the update node was created
+				for the MySQL interface */
+	dict_foreign_t*	foreign;/* NULL or pointer to a foreign key
+				constraint if this update node is used in
+				doing an ON DELETE or ON UPDATE operation */
+	upd_node_t*	cascade_node;/* NULL or an update node template which
+				is used to implement ON DELETE/UPDATE CASCADE
+				or ... SET NULL for foreign keys */
+	mem_heap_t*	cascade_heap;/* NULL or a mem heap where the cascade
+				node is created */
+	sel_node_t*	select;	/* query graph subtree implementing a base
+				table cursor: the rows returned will be
+				updated */
+	btr_pcur_t*	pcur;	/* persistent cursor placed on the clustered
+				index record which should be updated or
+				deleted; the cursor is stored in the graph
+				of 'select' field above, except in the case
+				of the MySQL interface */
+	dict_table_t*	table;	/* table where updated */
+	upd_t*		update;	/* update vector for the row */
+	ulint		update_n_fields;
+				/* when this struct is used to implement
+				a cascade operation for foreign keys, we store
+				here the size of the buffer allocated for use
+				as the update vector */
+	sym_node_list_t	columns;/* symbol table nodes for the columns
+				to retrieve from the table */
+	ibool		has_clust_rec_x_lock;
+				/* TRUE if the select which retrieves the
+				records to update already sets an x-lock on
+				the clustered record; note that it must always
+				set at least an s-lock */
+	ulint		cmpl_info;/* information extracted during query
+				compilation; speeds up execution:
+				UPD_NODE_NO_ORD_CHANGE and
+				UPD_NODE_NO_SIZE_CHANGE, ORed */
+	/*----------------------*/
+	/* Local storage for this graph node */
+	ulint		state;	/* node execution state */
+	dict_index_t*	index;	/* NULL, or the next index whose record should
+				be updated */
+	dtuple_t*	row;	/* NULL, or a copy (also fields copied to
+				heap) of the row to update; this must be reset
+				to NULL after a successful update */
+	row_ext_t*	ext;	/* NULL, or prefixes of the externally
+				stored columns in the old row */
+	dtuple_t*	upd_row;/* NULL, or a copy of the updated row */
+	row_ext_t*	upd_ext;/* NULL, or prefixes of the externally
+				stored columns in upd_row */
+	mem_heap_t*	heap;	/* memory heap used as auxiliary storage;
+				this must be emptied after a successful
+				update */
+	/*----------------------*/
+	sym_node_t*	table_sym;/* table node in symbol table */
+	que_node_t*	col_assign_list;
+				/* column assignment list */
+	ulint		magic_n;
+};
+
+#define	UPD_NODE_MAGIC_N	1579975
+
+/* Node execution states */
+#define UPD_NODE_SET_IX_LOCK	   1	/* execution came to the node from
+					a node above and if the field
+					has_clust_rec_x_lock is FALSE, we
+					should set an intention x-lock on
+					the table */
+#define UPD_NODE_UPDATE_CLUSTERED  2	/* clustered index record should be
+					updated */
+#define UPD_NODE_INSERT_CLUSTERED  3	/* clustered index record should be
+					inserted, old record is already delete
+					marked */
+#define UPD_NODE_UPDATE_ALL_SEC	   4	/* an ordering field of the clustered
+					index record was changed, or this is
+					a delete operation: should update
+					all the secondary index records */
+#define	UPD_NODE_UPDATE_SOME_SEC   5	/* secondary index entries should be
+					looked at and updated if an ordering
+					field changed */
+
+/* Compilation info flags: these must fit within 3 bits; see trx0rec.h */
+#define UPD_NODE_NO_ORD_CHANGE	1	/* no secondary index record will be
+					changed in the update and no ordering
+					field of the clustered index */
+#define UPD_NODE_NO_SIZE_CHANGE	2	/* no record field size will be
+					changed in the update */
+
+#ifndef UNIV_NONINL
+#include "row0upd.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/row0upd.ic b/storage/xtradb/include/row0upd.ic
new file mode 100644
index 00000000000..a0c23aa6b07
--- /dev/null
+++ b/storage/xtradb/include/row0upd.ic
@@ -0,0 +1,179 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Update of a row
+
+Created 12/27/1996 Heikki Tuuri
+*******************************************************/
+
+#include "mtr0log.h"
+#include "trx0trx.h"
+#include "trx0undo.h"
+#include "row0row.h"
+#include "btr0sea.h"
+#include "page0zip.h"
+
+/*************************************************************************
+Creates an update vector object. */
+UNIV_INLINE
+upd_t*
+upd_create(
+/*=======*/
+				/* out, own: update vector object */
+	ulint		n,	/* in: number of fields */
+	mem_heap_t*	heap)	/* in: heap from which memory allocated */
+{
+	upd_t*	update;
+
+	update = (upd_t*) mem_heap_alloc(heap, sizeof(upd_t));
+
+	update->info_bits = 0;
+	update->n_fields = n;
+	update->fields = (upd_field_t*)
+		mem_heap_alloc(heap, sizeof(upd_field_t) * n);
+
+	return(update);
+}
+
+/*************************************************************************
+Returns the number of fields in the update vector == number of columns
+to be updated by an update vector. */
+UNIV_INLINE
+ulint
+upd_get_n_fields(
+/*=============*/
+				/* out: number of fields */
+	const upd_t*	update)	/* in: update vector */
+{
+	ut_ad(update);
+
+	return(update->n_fields);
+}
+
+#ifdef UNIV_DEBUG
+/*************************************************************************
+Returns the nth field of an update vector. */
+UNIV_INLINE
+upd_field_t*
+upd_get_nth_field(
+/*==============*/
+				/* out: update vector field */
+	const upd_t*	update,	/* in: update vector */
+	ulint		n)	/* in: field position in update vector */
+{
+	ut_ad(update);
+	ut_ad(n < update->n_fields);
+
+	return((upd_field_t*) update->fields + n);
+}
+#endif /* UNIV_DEBUG */
+
+/*************************************************************************
+Sets an index field number to be updated by an update vector field. */
+UNIV_INLINE
+void
+upd_field_set_field_no(
+/*===================*/
+	upd_field_t*	upd_field,	/* in: update vector field */
+	ulint		field_no,	/* in: field number in a clustered
+					index */
+	dict_index_t*	index,		/* in: index */
+	trx_t*		trx)		/* in: transaction */
+{
+	upd_field->field_no = field_no;
+	upd_field->orig_len = 0;
+
+	if (UNIV_UNLIKELY(field_no >= dict_index_get_n_fields(index))) {
+		fprintf(stderr,
+			"InnoDB: Error: trying to access field %lu in ",
+			(ulong) field_no);
+		dict_index_name_print(stderr, trx, index);
+		fprintf(stderr, "\n"
+			"InnoDB: but index only has %lu fields\n",
+			(ulong) dict_index_get_n_fields(index));
+	}
+
+	dict_col_copy_type(dict_index_get_nth_col(index, field_no),
+			   dfield_get_type(&upd_field->new_val));
+}
+
+/*************************************************************************
+Returns a field of an update vector by field_no. */
+UNIV_INLINE
+const upd_field_t*
+upd_get_field_by_field_no(
+/*======================*/
+				/* out: update vector field, or NULL */
+	const upd_t*	update,	/* in: update vector */
+	ulint		no)	/* in: field_no */
+{
+	ulint	i;
+	for (i = 0; i < upd_get_n_fields(update); i++) {
+		const upd_field_t*	uf = upd_get_nth_field(update, i);
+
+		if (uf->field_no == no) {
+
+			return(uf);
+		}
+	}
+
+	return(NULL);
+}
+
+/*************************************************************************
+Updates the trx id and roll ptr field in a clustered index record when
+a row is updated or marked deleted. */
+UNIV_INLINE
+void
+row_upd_rec_sys_fields(
+/*===================*/
+	rec_t*		rec,	/* in/out: record */
+	page_zip_des_t*	page_zip,/* in/out: compressed page whose
+				uncompressed part will be updated, or NULL */
+	dict_index_t*	index,	/* in: clustered index */
+	const ulint*	offsets,/* in: rec_get_offsets(rec, index) */
+	trx_t*		trx,	/* in: transaction */
+	dulint		roll_ptr)/* in: roll ptr of the undo log record */
+{
+	ut_ad(dict_index_is_clust(index));
+	ut_ad(rec_offs_validate(rec, index, offsets));
+#ifdef UNIV_SYNC_DEBUG
+	if (!rw_lock_own(&btr_search_latch, RW_LOCK_EX)) {
+		ut_ad(!buf_block_align(rec)->is_hashed);
+	}
+#endif /* UNIV_SYNC_DEBUG */
+
+	if (UNIV_LIKELY_NULL(page_zip)) {
+		ulint	pos = dict_index_get_sys_col_pos(index, DATA_TRX_ID);
+		page_zip_write_trx_id_and_roll_ptr(page_zip, rec, offsets,
+						   pos, trx->id, roll_ptr);
+	} else {
+		ulint	offset = index->trx_id_offset;
+
+		if (!offset) {
+			offset = row_get_trx_id_offset(rec, index, offsets);
+		}
+
+#if DATA_TRX_ID + 1 != DATA_ROLL_PTR
+# error "DATA_TRX_ID + 1 != DATA_ROLL_PTR"
+#endif
+		trx_write_trx_id(rec + offset, trx->id);
+		trx_write_roll_ptr(rec + offset + DATA_TRX_ID_LEN, roll_ptr);
+	}
+}
diff --git a/storage/xtradb/include/row0vers.h b/storage/xtradb/include/row0vers.h
new file mode 100644
index 00000000000..0feae77e8b5
--- /dev/null
+++ b/storage/xtradb/include/row0vers.h
@@ -0,0 +1,142 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Row versions
+
+Created 2/6/1997 Heikki Tuuri
+*******************************************************/
+
+#ifndef row0vers_h
+#define row0vers_h
+
+#include "univ.i"
+#include "data0data.h"
+#include "dict0types.h"
+#include "trx0types.h"
+#include "que0types.h"
+#include "rem0types.h"
+#include "mtr0mtr.h"
+#include "read0types.h"
+
+/*********************************************************************
+Finds out if an active transaction has inserted or modified a secondary
+index record. NOTE: the kernel mutex is temporarily released in this
+function! */
+UNIV_INTERN
+trx_t*
+row_vers_impl_x_locked_off_kernel(
+/*==============================*/
+				/* out: NULL if committed, else the active
+				transaction; NOTE that the kernel mutex is
+				temporarily released! */
+	const rec_t*	rec,	/* in: record in a secondary index */
+	dict_index_t*	index,	/* in: the secondary index */
+	const ulint*	offsets);/* in: rec_get_offsets(rec, index) */
+/*********************************************************************
+Finds out if we must preserve a delete marked earlier version of a clustered
+index record, because it is >= the purge view. */
+UNIV_INTERN
+ibool
+row_vers_must_preserve_del_marked(
+/*==============================*/
+			/* out: TRUE if earlier version should be preserved */
+	dulint	trx_id,	/* in: transaction id in the version */
+	mtr_t*	mtr);	/* in: mtr holding the latch on the clustered index
+			record; it will also hold the latch on purge_view */
+/*********************************************************************
+Finds out if a version of the record, where the version >= the current
+purge view, should have ientry as its secondary index entry. We check
+if there is any not delete marked version of the record where the trx
+id >= purge view, and the secondary index entry == ientry; exactly in
+this case we return TRUE. */
+UNIV_INTERN
+ibool
+row_vers_old_has_index_entry(
+/*=========================*/
+				/* out: TRUE if earlier version should have */
+	ibool		also_curr,/* in: TRUE if also rec is included in the
+				versions to search; otherwise only versions
+				prior to it are searched */
+	const rec_t*	rec,	/* in: record in the clustered index; the
+				caller must have a latch on the page */
+	mtr_t*		mtr,	/* in: mtr holding the latch on rec; it will
+				also hold the latch on purge_view */
+	dict_index_t*	index,	/* in: the secondary index */
+	const dtuple_t*	ientry);/* in: the secondary index entry */
+/*********************************************************************
+Constructs the version of a clustered index record which a consistent
+read should see. We assume that the trx id stored in rec is such that
+the consistent read should not see rec in its present version. */
+UNIV_INTERN
+ulint
+row_vers_build_for_consistent_read(
+/*===============================*/
+				/* out: DB_SUCCESS or DB_MISSING_HISTORY */
+	const rec_t*	rec,	/* in: record in a clustered index; the
+				caller must have a latch on the page; this
+				latch locks the top of the stack of versions
+				of this records */
+	mtr_t*		mtr,	/* in: mtr holding the latch on rec; it will
+				also hold the latch on purge_view */
+	dict_index_t*	index,	/* in: the clustered index */
+	ulint**		offsets,/* in/out: offsets returned by
+				rec_get_offsets(rec, index) */
+	read_view_t*	view,	/* in: the consistent read view */
+	mem_heap_t**	offset_heap,/* in/out: memory heap from which
+				the offsets are allocated */
+	mem_heap_t*	in_heap,/* in: memory heap from which the memory for
+				*old_vers is allocated; memory for possible
+				intermediate versions is allocated and freed
+				locally within the function */
+	rec_t**		old_vers);/* out, own: old version, or NULL if the
+				record does not exist in the view, that is,
+				it was freshly inserted afterwards */
+
+/*********************************************************************
+Constructs the last committed version of a clustered index record,
+which should be seen by a semi-consistent read. */
+UNIV_INTERN
+ulint
+row_vers_build_for_semi_consistent_read(
+/*====================================*/
+				/* out: DB_SUCCESS or DB_MISSING_HISTORY */
+	const rec_t*	rec,	/* in: record in a clustered index; the
+				caller must have a latch on the page; this
+				latch locks the top of the stack of versions
+				of this records */
+	mtr_t*		mtr,	/* in: mtr holding the latch on rec */
+	dict_index_t*	index,	/* in: the clustered index */
+	ulint**		offsets,/* in/out: offsets returned by
+				rec_get_offsets(rec, index) */
+	mem_heap_t**	offset_heap,/* in/out: memory heap from which
+				the offsets are allocated */
+	mem_heap_t*	in_heap,/* in: memory heap from which the memory for
+				*old_vers is allocated; memory for possible
+				intermediate versions is allocated and freed
+				locally within the function */
+	const rec_t**	old_vers);/* out: rec, old version, or NULL if the
+				record does not exist in the view, that is,
+				it was freshly inserted afterwards */
+
+
+#ifndef UNIV_NONINL
+#include "row0vers.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/row0vers.ic b/storage/xtradb/include/row0vers.ic
new file mode 100644
index 00000000000..aac95ea6593
--- /dev/null
+++ b/storage/xtradb/include/row0vers.ic
@@ -0,0 +1,29 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Row versions
+
+Created 2/6/1997 Heikki Tuuri
+*******************************************************/
+
+#include "row0row.h"
+#include "dict0dict.h"
+#include "read0read.h"
+#include "page0page.h"
+#include "log0recv.h"
diff --git a/storage/xtradb/include/srv0que.h b/storage/xtradb/include/srv0que.h
new file mode 100644
index 00000000000..88db1a013f6
--- /dev/null
+++ b/storage/xtradb/include/srv0que.h
@@ -0,0 +1,68 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Server query execution
+
+Created 6/5/1996 Heikki Tuuri
+*******************************************************/
+
+#ifndef srv0que_h
+#define srv0que_h
+
+#include "univ.i"
+#include "que0types.h"
+
+/**************************************************************************
+Checks if there is work to do in the server task queue. If there is, the
+thread starts processing a task. Before leaving, it again checks the task
+queue and picks a new task if any exists. This is called by a SRV_WORKER
+thread. */
+UNIV_INTERN
+void
+srv_que_task_queue_check(void);
+/*==========================*/
+/**************************************************************************
+Performs round-robin on the server tasks. This is called by a SRV_WORKER
+thread every second or so. */
+UNIV_INTERN
+que_thr_t*
+srv_que_round_robin(
+/*================*/
+				/* out: the new (may be == thr) query thread
+				to run */
+	que_thr_t*	thr);	/* in: query thread */
+/**************************************************************************
+Enqueues a task to server task queue and releases a worker thread, if
+there exists one suspended. */
+UNIV_INTERN
+void
+srv_que_task_enqueue(
+/*=================*/
+	que_thr_t*	thr);	/* in: query thread */
+/**************************************************************************
+Enqueues a task to server task queue and releases a worker thread, if
+there exists one suspended. */
+UNIV_INTERN
+void
+srv_que_task_enqueue_low(
+/*=====================*/
+	que_thr_t*	thr);	/* in: query thread */
+
+#endif
+
diff --git a/storage/xtradb/include/srv0srv.h b/storage/xtradb/include/srv0srv.h
new file mode 100644
index 00000000000..cb78d66da1b
--- /dev/null
+++ b/storage/xtradb/include/srv0srv.h
@@ -0,0 +1,608 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+Copyright (c) 2008, Google Inc.
+
+Portions of this file contain modifications contributed and copyrighted by
+Google, Inc. Those modifications are gratefully acknowledged and are described
+briefly in the InnoDB documentation. The contributions by Google are
+incorporated with their permission, and subject to the conditions contained in
+the file COPYING.Google.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The server main program
+
+Created 10/10/1995 Heikki Tuuri
+*******************************************************/
+
+#ifndef srv0srv_h
+#define srv0srv_h
+
+#include "univ.i"
+#include "sync0sync.h"
+#include "os0sync.h"
+#include "que0types.h"
+#include "trx0types.h"
+
+extern const char*	srv_main_thread_op_info;
+
+/* Prefix used by MySQL to indicate pre-5.1 table name encoding */
+extern const char	srv_mysql50_table_name_prefix[9];
+
+/* When this event is set the lock timeout and InnoDB monitor
+thread starts running */
+extern os_event_t	srv_lock_timeout_thread_event;
+
+/* If the last data file is auto-extended, we add this many pages to it
+at a time */
+#define SRV_AUTO_EXTEND_INCREMENT	\
+	(srv_auto_extend_increment * ((1024 * 1024) / UNIV_PAGE_SIZE))
+
+/* This is set to TRUE if the MySQL user has set it in MySQL */
+extern ibool	srv_lower_case_table_names;
+
+/* Mutex for locking srv_monitor_file */
+extern mutex_t	srv_monitor_file_mutex;
+/* Temporary file for innodb monitor output */
+extern FILE*	srv_monitor_file;
+/* Mutex for locking srv_dict_tmpfile.
+This mutex has a very high rank; threads reserving it should not
+be holding any InnoDB latches. */
+extern mutex_t	srv_dict_tmpfile_mutex;
+/* Temporary file for output from the data dictionary */
+extern FILE*	srv_dict_tmpfile;
+/* Mutex for locking srv_misc_tmpfile.
+This mutex has a very low rank; threads reserving it should not
+acquire any further latches or sleep before releasing this one. */
+extern mutex_t	srv_misc_tmpfile_mutex;
+/* Temporary file for miscellanous diagnostic output */
+extern FILE*	srv_misc_tmpfile;
+
+/* Server parameters which are read from the initfile */
+
+extern char*	srv_data_home;
+#ifdef UNIV_LOG_ARCHIVE
+extern char*	srv_arch_dir;
+#endif /* UNIV_LOG_ARCHIVE */
+
+/* store to its own file each table created by an user; data
+dictionary tables are in the system tablespace 0 */
+extern my_bool	srv_file_per_table;
+/* The file format to use on new *.ibd files. */
+extern ulint	srv_file_format;
+/* Whether to check file format during startup.*/
+extern ulint	srv_check_file_format_at_startup;
+/* Place locks to records only i.e. do not use next-key locking except
+on duplicate key checking and foreign key checking */
+extern ibool	srv_locks_unsafe_for_binlog;
+
+extern ulint	srv_n_data_files;
+extern char**	srv_data_file_names;
+extern ulint*	srv_data_file_sizes;
+extern ulint*	srv_data_file_is_raw_partition;
+
+extern ibool	srv_extra_undoslots;
+
+extern ibool	srv_auto_extend_last_data_file;
+extern ulint	srv_last_file_size_max;
+extern ulong	srv_auto_extend_increment;
+
+extern ibool	srv_created_new_raw;
+
+#define SRV_NEW_RAW	1
+#define SRV_OLD_RAW	2
+
+extern char**	srv_log_group_home_dirs;
+
+extern ulint	srv_n_log_groups;
+extern ulint	srv_n_log_files;
+extern ulint	srv_log_file_size;
+extern ulint	srv_log_buffer_size;
+extern ulong	srv_flush_log_at_trx_commit;
+
+extern ulint    srv_show_locks_held;
+extern ulint    srv_show_verbose_locks;
+
+/* The sort order table of the MySQL latin1_swedish_ci character set
+collation */
+extern const byte*	srv_latin1_ordering;
+extern my_bool	srv_use_sys_malloc;
+extern ulint	srv_buf_pool_size;	/* requested size in bytes */
+extern ulint	srv_buf_pool_old_size;	/* previously requested size */
+extern ulint	srv_buf_pool_curr_size;	/* current size in bytes */
+extern ulint	srv_mem_pool_size;
+extern ulint	srv_lock_table_size;
+
+extern ulint	srv_n_file_io_threads;
+extern ulint	srv_n_read_io_threads;
+extern ulint	srv_n_write_io_threads;
+
+#ifdef UNIV_LOG_ARCHIVE
+extern ibool	srv_log_archive_on;
+extern ibool	srv_archive_recovery;
+extern dulint	srv_archive_recovery_limit_lsn;
+#endif /* UNIV_LOG_ARCHIVE */
+
+extern char*	srv_file_flush_method_str;
+extern ulint	srv_unix_file_flush_method;
+extern ulint	srv_win_file_flush_method;
+
+extern ulint	srv_max_n_open_files;
+
+extern ulint	srv_max_dirty_pages_pct;
+
+extern ulint	srv_force_recovery;
+extern ulong	srv_thread_concurrency;
+extern ulong	srv_commit_concurrency;
+
+extern ulint	srv_max_n_threads;
+
+extern lint	srv_conc_n_threads;
+
+extern ulint	srv_fast_shutdown;	 /* If this is 1, do not do a
+					 purge and index buffer merge.
+					 If this 2, do not even flush the
+					 buffer pool to data files at the
+					 shutdown: we effectively 'crash'
+					 InnoDB (but lose no committed
+					 transactions). */
+extern ibool	srv_innodb_status;
+
+extern unsigned long long	srv_stats_sample_pages;
+
+extern ibool	srv_use_doublewrite_buf;
+extern ibool	srv_use_checksums;
+
+extern ibool	srv_set_thread_priorities;
+extern int	srv_query_thread_priority;
+
+extern ulong	srv_max_buf_pool_modified_pct;
+extern ulong	srv_max_purge_lag;
+
+extern ulong	srv_replication_delay;
+
+extern ulint	srv_io_capacity;
+extern long long	srv_ibuf_max_size;
+extern ulint	srv_ibuf_active_contract;
+extern ulint	srv_ibuf_accel_rate;
+extern ulint	srv_flush_neighbor_pages;
+extern ulint	srv_enable_unsafe_group_commit;
+extern ulint	srv_read_ahead;
+extern ulint	srv_adaptive_checkpoint;
+
+extern ulint	srv_extra_rsegments;
+
+/*-------------------------------------------*/
+
+extern ulint	srv_n_rows_inserted;
+extern ulint	srv_n_rows_updated;
+extern ulint	srv_n_rows_deleted;
+extern ulint	srv_n_rows_read;
+
+extern ibool	srv_print_innodb_monitor;
+extern ibool	srv_print_innodb_lock_monitor;
+extern ibool	srv_print_innodb_tablespace_monitor;
+extern ibool	srv_print_verbose_log;
+extern ibool	srv_print_innodb_table_monitor;
+
+extern ibool	srv_lock_timeout_and_monitor_active;
+extern ibool	srv_error_monitor_active;
+
+extern ulong	srv_n_spin_wait_rounds;
+extern ulong	srv_n_free_tickets_to_enter;
+extern ulong	srv_thread_sleep_delay;
+extern ulint	srv_spin_wait_delay;
+extern ibool	srv_priority_boost;
+
+extern	ulint	srv_mem_pool_size;
+extern	ulint	srv_lock_table_size;
+
+#ifdef UNIV_DEBUG
+extern	ibool	srv_print_thread_releases;
+extern	ibool	srv_print_lock_waits;
+extern	ibool	srv_print_buf_io;
+extern	ibool	srv_print_log_io;
+extern	ibool	srv_print_latch_waits;
+#else /* UNIV_DEBUG */
+# define srv_print_thread_releases	FALSE
+# define srv_print_lock_waits		FALSE
+# define srv_print_buf_io		FALSE
+# define srv_print_log_io		FALSE
+# define srv_print_latch_waits		FALSE
+#endif /* UNIV_DEBUG */
+
+extern ulint	srv_activity_count;
+extern ulint	srv_fatal_semaphore_wait_threshold;
+extern ulint	srv_dml_needed_delay;
+
+extern mutex_t*	kernel_mutex_temp;/* mutex protecting the server, trx structs,
+				query threads, and lock table: we allocate
+				it from dynamic memory to get it to the
+				same DRAM page as other hotspot semaphores */
+#define kernel_mutex (*kernel_mutex_temp)
+
+#define SRV_MAX_N_IO_THREADS	100
+
+/* Array of English strings describing the current state of an
+i/o handler thread */
+extern const char* srv_io_thread_op_info[];
+extern const char* srv_io_thread_function[];
+
+/* the number of the log write requests done */
+extern ulint srv_log_write_requests;
+
+/* the number of physical writes to the log performed */
+extern ulint srv_log_writes;
+
+/* amount of data written to the log files in bytes */
+extern ulint srv_os_log_written;
+
+/* amount of writes being done to the log files */
+extern ulint srv_os_log_pending_writes;
+
+/* we increase this counter, when there we don't have enough space in the
+log buffer and have to flush it */
+extern ulint srv_log_waits;
+
+/* variable that counts amount of data read in total (in bytes) */
+extern ulint srv_data_read;
+
+/* here we count the amount of data written in total (in bytes) */
+extern ulint srv_data_written;
+
+/* this variable counts the amount of times, when the doublewrite buffer
+was flushed */
+extern ulint srv_dblwr_writes;
+
+/* here we store the number of pages that have been flushed to the
+doublewrite buffer */
+extern ulint srv_dblwr_pages_written;
+
+/* in this variable we store the number of write requests issued */
+extern ulint srv_buf_pool_write_requests;
+
+/* here we store the number of times when we had to wait for a free page
+in the buffer pool. It happens when the buffer pool is full and we need
+to make a flush, in order to be able to read or create a page. */
+extern ulint srv_buf_pool_wait_free;
+
+/* variable to count the number of pages that were written from the
+buffer pool to disk */
+extern ulint srv_buf_pool_flushed;
+
+/* variable to count the number of buffer pool reads that led to the
+reading of a disk page */
+extern ulint srv_buf_pool_reads;
+
+/* variable to count the number of sequential read-aheads were done */
+extern ulint srv_read_ahead_seq;
+
+/* variable to count the number of random read-aheads were done */
+extern ulint srv_read_ahead_rnd;
+
+/* In this structure we store status variables to be passed to MySQL */
+typedef struct export_var_struct export_struc;
+
+extern export_struc export_vars;
+
+typedef struct srv_sys_struct	srv_sys_t;
+
+/* The server system */
+extern srv_sys_t*	srv_sys;
+
+/* Alternatives for the file flush option in Unix; see the InnoDB manual
+about what these mean */
+#define SRV_UNIX_FSYNC		1	/* This is the default */
+#define SRV_UNIX_O_DSYNC	2
+#define SRV_UNIX_LITTLESYNC	3
+#define SRV_UNIX_NOSYNC		4
+#define SRV_UNIX_O_DIRECT	5
+
+/* Alternatives for file i/o in Windows */
+#define SRV_WIN_IO_NORMAL		1
+#define SRV_WIN_IO_UNBUFFERED		2	/* This is the default */
+
+/* Alternatives for srv_force_recovery. Non-zero values are intended
+to help the user get a damaged database up so that he can dump intact
+tables and rows with SELECT INTO OUTFILE. The database must not otherwise
+be used with these options! A bigger number below means that all precautions
+of lower numbers are included. */
+
+#define SRV_FORCE_IGNORE_CORRUPT 1	/* let the server run even if it
+					detects a corrupt page */
+#define SRV_FORCE_NO_BACKGROUND	2	/* prevent the main thread from
+					running: if a crash would occur
+					in purge, this prevents it */
+#define SRV_FORCE_NO_TRX_UNDO	3	/* do not run trx rollback after
+					recovery */
+#define SRV_FORCE_NO_IBUF_MERGE	4	/* prevent also ibuf operations:
+					if they would cause a crash, better
+					not do them */
+#define	SRV_FORCE_NO_UNDO_LOG_SCAN 5	/* do not look at undo logs when
+					starting the database: InnoDB will
+					treat even incomplete transactions
+					as committed */
+#define SRV_FORCE_NO_LOG_REDO	6	/* do not do the log roll-forward
+					in connection with recovery */
+
+/** Types of threads existing in the system. */
+enum srv_thread_type {
+	SRV_COM = 1,	/**< threads serving communication and queries */
+	SRV_CONSOLE,	/**< thread serving console */
+	SRV_WORKER,	/**< threads serving parallelized queries and
+			queries released from lock wait */
+#if 0
+	/* Utility threads */
+	SRV_BUFFER,	/**< thread flushing dirty buffer blocks */
+	SRV_RECOVERY,	/**< threads finishing a recovery */
+	SRV_INSERT,	/**< thread flushing the insert buffer to disk */
+#endif
+	SRV_MASTER	/**< the master thread, (whose type number must
+			be biggest) */
+};
+
+/*************************************************************************
+Boots Innobase server. */
+UNIV_INTERN
+ulint
+srv_boot(void);
+/*==========*/
+			/* out: DB_SUCCESS or error code */
+/*************************************************************************
+Initializes the server. */
+UNIV_INTERN
+void
+srv_init(void);
+/*==========*/
+/*************************************************************************
+Frees the OS fast mutex created in srv_boot(). */
+UNIV_INTERN
+void
+srv_free(void);
+/*==========*/
+/*************************************************************************
+Initializes the synchronization primitives, memory system, and the thread
+local storage. */
+UNIV_INTERN
+void
+srv_general_init(void);
+/*==================*/
+/*************************************************************************
+Gets the number of threads in the system. */
+UNIV_INTERN
+ulint
+srv_get_n_threads(void);
+/*===================*/
+/*************************************************************************
+Returns the calling thread type. */
+
+enum srv_thread_type
+srv_get_thread_type(void);
+/*=====================*/
+			/* out: SRV_COM, ... */
+/*************************************************************************
+Sets the info describing an i/o thread current state. */
+UNIV_INTERN
+void
+srv_set_io_thread_op_info(
+/*======================*/
+	ulint		i,	/* in: the 'segment' of the i/o thread */
+	const char*	str);	/* in: constant char string describing the
+				state */
+/*************************************************************************
+Releases threads of the type given from suspension in the thread table.
+NOTE! The server mutex has to be reserved by the caller! */
+UNIV_INTERN
+ulint
+srv_release_threads(
+/*================*/
+					/* out: number of threads
+					released: this may be < n if
+					not enough threads were
+					suspended at the moment */
+	enum srv_thread_type	type,	/* in: thread type */
+	ulint			n);	/* in: number of threads to release */
+/*************************************************************************
+The master thread controlling the server. */
+UNIV_INTERN
+os_thread_ret_t
+srv_master_thread(
+/*==============*/
+			/* out: a dummy parameter */
+	void*	arg);	/* in: a dummy parameter required by
+			os_thread_create */
+/***********************************************************************
+Tells the Innobase server that there has been activity in the database
+and wakes up the master thread if it is suspended (not sleeping). Used
+in the MySQL interface. Note that there is a small chance that the master
+thread stays suspended (we do not protect our operation with the kernel
+mutex, for performace reasons). */
+UNIV_INTERN
+void
+srv_active_wake_master_thread(void);
+/*===============================*/
+/***********************************************************************
+Wakes up the master thread if it is suspended or being suspended. */
+UNIV_INTERN
+void
+srv_wake_master_thread(void);
+/*========================*/
+/*************************************************************************
+Puts an OS thread to wait if there are too many concurrent threads
+(>= srv_thread_concurrency) inside InnoDB. The threads wait in a FIFO queue. */
+UNIV_INTERN
+void
+srv_conc_enter_innodb(
+/*==================*/
+	trx_t*	trx);	/* in: transaction object associated with the
+			thread */
+/*************************************************************************
+This lets a thread enter InnoDB regardless of the number of threads inside
+InnoDB. This must be called when a thread ends a lock wait. */
+UNIV_INTERN
+void
+srv_conc_force_enter_innodb(
+/*========================*/
+	trx_t*	trx);	/* in: transaction object associated with the
+			thread */
+/*************************************************************************
+This must be called when a thread exits InnoDB in a lock wait or at the
+end of an SQL statement. */
+UNIV_INTERN
+void
+srv_conc_force_exit_innodb(
+/*=======================*/
+	trx_t*	trx);	/* in: transaction object associated with the
+			thread */
+/*************************************************************************
+This must be called when a thread exits InnoDB. */
+UNIV_INTERN
+void
+srv_conc_exit_innodb(
+/*=================*/
+	trx_t*	trx);	/* in: transaction object associated with the
+			thread */
+/*******************************************************************
+Puts a MySQL OS thread to wait for a lock to be released. If an error
+occurs during the wait trx->error_state associated with thr is
+!= DB_SUCCESS when we return. DB_LOCK_WAIT_TIMEOUT and DB_DEADLOCK
+are possible errors. DB_DEADLOCK is returned if selective deadlock
+resolution chose this transaction as a victim. */
+UNIV_INTERN
+void
+srv_suspend_mysql_thread(
+/*=====================*/
+	que_thr_t*	thr);	/* in: query thread associated with the MySQL
+				OS thread */
+/************************************************************************
+Releases a MySQL OS thread waiting for a lock to be released, if the
+thread is already suspended. */
+UNIV_INTERN
+void
+srv_release_mysql_thread_if_suspended(
+/*==================================*/
+	que_thr_t*	thr);	/* in: query thread associated with the
+				MySQL OS thread	 */
+/*************************************************************************
+A thread which wakes up threads whose lock wait may have lasted too long.
+This also prints the info output by various InnoDB monitors. */
+UNIV_INTERN
+os_thread_ret_t
+srv_lock_timeout_and_monitor_thread(
+/*================================*/
+			/* out: a dummy parameter */
+	void*	arg);	/* in: a dummy parameter required by
+			os_thread_create */
+/*************************************************************************
+A thread which prints warnings about semaphore waits which have lasted
+too long. These can be used to track bugs which cause hangs. */
+UNIV_INTERN
+os_thread_ret_t
+srv_error_monitor_thread(
+/*=====================*/
+			/* out: a dummy parameter */
+	void*	arg);	/* in: a dummy parameter required by
+			os_thread_create */
+/**********************************************************************
+Outputs to a file the output of the InnoDB Monitor. */
+UNIV_INTERN
+void
+srv_printf_innodb_monitor(
+/*======================*/
+	FILE*	file,		/* in: output stream */
+	ulint*	trx_start,	/* out: file position of the start of
+				the list of active transactions */
+	ulint*	trx_end);	/* out: file position of the end of
+				the list of active transactions */
+
+/**********************************************************************
+Function to pass InnoDB status variables to MySQL */
+UNIV_INTERN
+void
+srv_export_innodb_status(void);
+/*=====================*/
+
+/* Thread slot in the thread table */
+typedef struct srv_slot_struct	srv_slot_t;
+
+/* Thread table is an array of slots */
+typedef srv_slot_t	srv_table_t;
+
+/* In this structure we store status variables to be passed to MySQL */
+struct export_var_struct{
+	ulint innodb_data_pending_reads;
+	ulint innodb_data_pending_writes;
+	ulint innodb_data_pending_fsyncs;
+	ulint innodb_data_fsyncs;
+	ulint innodb_data_read;
+	ulint innodb_data_writes;
+	ulint innodb_data_written;
+	ulint innodb_data_reads;
+	ulint innodb_buffer_pool_pages_total;
+	ulint innodb_buffer_pool_pages_data;
+	ulint innodb_buffer_pool_pages_dirty;
+	ulint innodb_buffer_pool_pages_misc;
+	ulint innodb_buffer_pool_pages_free;
+#ifdef UNIV_DEBUG
+	ulint innodb_buffer_pool_pages_latched;
+#endif /* UNIV_DEBUG */
+	ulint innodb_buffer_pool_read_requests;
+	ulint innodb_buffer_pool_reads;
+	ulint innodb_buffer_pool_wait_free;
+	ulint innodb_buffer_pool_pages_flushed;
+	ulint innodb_buffer_pool_write_requests;
+	ulint innodb_buffer_pool_read_ahead_seq;
+	ulint innodb_buffer_pool_read_ahead_rnd;
+	ulint innodb_dblwr_pages_written;
+	ulint innodb_dblwr_writes;
+	ibool innodb_have_atomic_builtins;
+	ulint innodb_log_waits;
+	ulint innodb_log_write_requests;
+	ulint innodb_log_writes;
+	ulint innodb_os_log_written;
+	ulint innodb_os_log_fsyncs;
+	ulint innodb_os_log_pending_writes;
+	ulint innodb_os_log_pending_fsyncs;
+	ulint innodb_page_size;
+	ulint innodb_pages_created;
+	ulint innodb_pages_read;
+	ulint innodb_pages_written;
+	ulint innodb_row_lock_waits;
+	ulint innodb_row_lock_current_waits;
+	ib_int64_t innodb_row_lock_time;
+	ulint innodb_row_lock_time_avg;
+	ulint innodb_row_lock_time_max;
+	ulint innodb_rows_read;
+	ulint innodb_rows_inserted;
+	ulint innodb_rows_updated;
+	ulint innodb_rows_deleted;
+};
+
+/* The server system struct */
+struct srv_sys_struct{
+	srv_table_t*	threads;	/* server thread table */
+	UT_LIST_BASE_NODE_T(que_thr_t)
+			tasks;		/* task queue */
+	dict_index_t*	dummy_ind1;	/* dummy index for old-style
+					supremum and infimum records */
+	dict_index_t*	dummy_ind2;	/* dummy index for new-style
+					supremum and infimum records */
+};
+
+extern ulint	srv_n_threads_active[];
+
+#endif
diff --git a/storage/xtradb/include/srv0srv.ic b/storage/xtradb/include/srv0srv.ic
new file mode 100644
index 00000000000..93d675f1dca
--- /dev/null
+++ b/storage/xtradb/include/srv0srv.ic
@@ -0,0 +1,23 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Server main program
+
+Created 10/4/1995 Heikki Tuuri
+*******************************************************/
diff --git a/storage/xtradb/include/srv0start.h b/storage/xtradb/include/srv0start.h
new file mode 100644
index 00000000000..15fa3b8f95f
--- /dev/null
+++ b/storage/xtradb/include/srv0start.h
@@ -0,0 +1,118 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Starts the Innobase database server
+
+Created 10/10/1995 Heikki Tuuri
+*******************************************************/
+
+#ifndef srv0start_h
+#define srv0start_h
+
+#include "univ.i"
+#include "ut0byte.h"
+
+/*************************************************************************
+Normalizes a directory path for Windows: converts slashes to backslashes. */
+UNIV_INTERN
+void
+srv_normalize_path_for_win(
+/*=======================*/
+	char*	str);	/* in/out: null-terminated character string */
+/*************************************************************************
+Reads the data files and their sizes from a character string given in
+the .cnf file. */
+UNIV_INTERN
+ibool
+srv_parse_data_file_paths_and_sizes(
+/*================================*/
+			/* out: TRUE if ok, FALSE on parse error */
+	char*	str);	/* in/out: the data file path string */
+/*************************************************************************
+Reads log group home directories from a character string given in
+the .cnf file. */
+UNIV_INTERN
+ibool
+srv_parse_log_group_home_dirs(
+/*==========================*/
+			/* out: TRUE if ok, FALSE on parse error */
+	char*	str);	/* in/out: character string */
+/*************************************************************************
+Frees the memory allocated by srv_parse_data_file_paths_and_sizes()
+and srv_parse_log_group_home_dirs(). */
+UNIV_INTERN
+void
+srv_free_paths_and_sizes(void);
+/*==========================*/
+/*************************************************************************
+Adds a slash or a backslash to the end of a string if it is missing
+and the string is not empty. */
+UNIV_INTERN
+char*
+srv_add_path_separator_if_needed(
+/*=============================*/
+			/* out: string which has the separator if the
+			string is not empty */
+	char*	str);	/* in: null-terminated character string */
+/********************************************************************
+Starts Innobase and creates a new database if database files
+are not found and the user wants. */
+UNIV_INTERN
+int
+innobase_start_or_create_for_mysql(void);
+/*====================================*/
+				/* out: DB_SUCCESS or error code */
+/********************************************************************
+Shuts down the Innobase database. */
+UNIV_INTERN
+int
+innobase_shutdown_for_mysql(void);
+/*=============================*/
+				/* out: DB_SUCCESS or error code */
+extern	ib_uint64_t	srv_shutdown_lsn;
+extern	ib_uint64_t	srv_start_lsn;
+
+#ifdef __NETWARE__
+void set_panic_flag_for_netware(void);
+#endif
+
+#ifdef HAVE_DARWIN_THREADS
+extern	ibool	srv_have_fullfsync;
+#endif
+
+extern	ibool	srv_is_being_started;
+extern	ibool	srv_was_started;
+extern	ibool	srv_startup_is_before_trx_rollback_phase;
+extern	ibool	srv_is_being_shut_down;
+
+extern	ibool	srv_start_raw_disk_in_use;
+
+/* At a shutdown the value first climbs from 0 to SRV_SHUTDOWN_CLEANUP
+and then to SRV_SHUTDOWN_LAST_PHASE, and so on */
+
+extern	ulint	srv_shutdown_state;
+
+#define SRV_SHUTDOWN_CLEANUP	   1
+#define SRV_SHUTDOWN_LAST_PHASE	   2
+#define SRV_SHUTDOWN_EXIT_THREADS  3
+
+/* Log 'spaces' have id's >= this */
+#define SRV_LOG_SPACE_FIRST_ID		0xFFFFFFF0UL
+
+#endif
diff --git a/storage/xtradb/include/sync0arr.h b/storage/xtradb/include/sync0arr.h
new file mode 100644
index 00000000000..cc01c9ac5c8
--- /dev/null
+++ b/storage/xtradb/include/sync0arr.h
@@ -0,0 +1,138 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The wait array used in synchronization primitives
+
+Created 9/5/1995 Heikki Tuuri
+*******************************************************/
+
+#ifndef sync0arr_h
+#define sync0arr_h
+
+#include "univ.i"
+#include "ut0lst.h"
+#include "ut0mem.h"
+#include "os0thread.h"
+
+typedef struct sync_cell_struct		sync_cell_t;
+typedef struct sync_array_struct	sync_array_t;
+
+#define SYNC_ARRAY_OS_MUTEX	1
+#define SYNC_ARRAY_MUTEX	2
+
+/***********************************************************************
+Creates a synchronization wait array. It is protected by a mutex
+which is automatically reserved when the functions operating on it
+are called. */
+UNIV_INTERN
+sync_array_t*
+sync_array_create(
+/*==============*/
+				/* out, own: created wait array */
+	ulint	n_cells,	/* in: number of cells in the array
+				to create */
+	ulint	protection);	/* in: either SYNC_ARRAY_OS_MUTEX or
+				SYNC_ARRAY_MUTEX: determines the type
+				of mutex protecting the data structure */
+/**********************************************************************
+Frees the resources in a wait array. */
+UNIV_INTERN
+void
+sync_array_free(
+/*============*/
+	sync_array_t*	arr);	/* in, own: sync wait array */
+/**********************************************************************
+Reserves a wait array cell for waiting for an object.
+The event of the cell is reset to nonsignalled state. */
+UNIV_INTERN
+void
+sync_array_reserve_cell(
+/*====================*/
+	sync_array_t*	arr,	/* in: wait array */
+	void*		object, /* in: pointer to the object to wait for */
+	ulint		type,	/* in: lock request type */
+	const char*	file,	/* in: file where requested */
+	ulint		line,	/* in: line where requested */
+	ulint*		index); /* out: index of the reserved cell */
+/**********************************************************************
+This function should be called when a thread starts to wait on
+a wait array cell. In the debug version this function checks
+if the wait for a semaphore will result in a deadlock, in which
+case prints info and asserts. */
+UNIV_INTERN
+void
+sync_array_wait_event(
+/*==================*/
+	sync_array_t*	arr,	/* in: wait array */
+	ulint		index);	 /* in: index of the reserved cell */
+/**********************************************************************
+Frees the cell. NOTE! sync_array_wait_event frees the cell
+automatically! */
+UNIV_INTERN
+void
+sync_array_free_cell(
+/*=================*/
+	sync_array_t*	arr,	/* in: wait array */
+	ulint		index);	/* in: index of the cell in array */
+/**************************************************************************
+Note that one of the wait objects was signalled. */
+UNIV_INTERN
+void
+sync_array_object_signalled(
+/*========================*/
+	sync_array_t*	arr);	/* in: wait array */
+/**************************************************************************
+If the wakeup algorithm does not work perfectly at semaphore relases,
+this function will do the waking (see the comment in mutex_exit). This
+function should be called about every 1 second in the server. */
+UNIV_INTERN
+void
+sync_arr_wake_threads_if_sema_free(void);
+/*====================================*/
+/**************************************************************************
+Prints warnings of long semaphore waits to stderr. */
+UNIV_INTERN
+ibool
+sync_array_print_long_waits(void);
+/*=============================*/
+			/* out: TRUE if fatal semaphore wait threshold
+			was exceeded */
+/************************************************************************
+Validates the integrity of the wait array. Checks
+that the number of reserved cells equals the count variable. */
+UNIV_INTERN
+void
+sync_array_validate(
+/*================*/
+	sync_array_t*	arr);	/* in: sync wait array */
+/**************************************************************************
+Prints info of the wait array. */
+UNIV_INTERN
+void
+sync_array_print_info(
+/*==================*/
+	FILE*		file,	/* in: file where to print */
+	sync_array_t*	arr);	/* in: wait array */
+
+
+#ifndef UNIV_NONINL
+#include "sync0arr.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/sync0arr.ic b/storage/xtradb/include/sync0arr.ic
new file mode 100644
index 00000000000..09a562a4723
--- /dev/null
+++ b/storage/xtradb/include/sync0arr.ic
@@ -0,0 +1,26 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The wait array for synchronization primitives
+
+Inline code
+
+Created 9/5/1995 Heikki Tuuri
+*******************************************************/
+
diff --git a/storage/xtradb/include/sync0rw.h b/storage/xtradb/include/sync0rw.h
new file mode 100644
index 00000000000..e3fe0dc9ccc
--- /dev/null
+++ b/storage/xtradb/include/sync0rw.h
@@ -0,0 +1,567 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+Copyright (c) 2008, Google Inc.
+
+Portions of this file contain modifications contributed and copyrighted by
+Google, Inc. Those modifications are gratefully acknowledged and are described
+briefly in the InnoDB documentation. The contributions by Google are
+incorporated with their permission, and subject to the conditions contained in
+the file COPYING.Google.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The read-write lock (for threads, not for database transactions)
+
+Created 9/11/1995 Heikki Tuuri
+*******************************************************/
+
+#ifndef sync0rw_h
+#define sync0rw_h
+
+#include "univ.i"
+#include "ut0lst.h"
+#include "sync0sync.h"
+#include "os0sync.h"
+
+/* The following undef is to prevent a name conflict with a macro
+in MySQL: */
+#undef rw_lock_t
+
+/* Latch types; these are used also in btr0btr.h: keep the numerical values
+smaller than 30 and the order of the numerical values like below! */
+#define RW_S_LATCH	1
+#define	RW_X_LATCH	2
+#define	RW_NO_LATCH	3
+
+/* We decrement lock_word by this amount for each x_lock. It is also the
+start value for the lock_word, meaning that it limits the maximum number
+of concurrent read locks before the rw_lock breaks. The current value of
+0x00100000 allows 1,048,575 concurrent readers and 2047 recursive writers.*/
+#define X_LOCK_DECR		0x00100000
+
+typedef struct rw_lock_struct		rw_lock_t;
+#ifdef UNIV_SYNC_DEBUG
+typedef struct rw_lock_debug_struct	rw_lock_debug_t;
+#endif /* UNIV_SYNC_DEBUG */
+
+typedef UT_LIST_BASE_NODE_T(rw_lock_t)	rw_lock_list_t;
+
+extern rw_lock_list_t	rw_lock_list;
+extern mutex_t		rw_lock_list_mutex;
+
+#ifdef UNIV_SYNC_DEBUG
+/* The global mutex which protects debug info lists of all rw-locks.
+To modify the debug info list of an rw-lock, this mutex has to be
+
+acquired in addition to the mutex protecting the lock. */
+extern mutex_t		rw_lock_debug_mutex;
+extern os_event_t	rw_lock_debug_event;	/* If deadlock detection does
+					not get immediately the mutex it
+					may wait for this event */
+extern ibool		rw_lock_debug_waiters;	/* This is set to TRUE, if
+					there may be waiters for the event */
+#endif /* UNIV_SYNC_DEBUG */
+
+extern	ib_int64_t	rw_s_spin_wait_count;
+extern	ib_int64_t	rw_s_spin_round_count;
+extern	ib_int64_t	rw_s_exit_count;
+extern	ib_int64_t	rw_s_os_wait_count;
+extern	ib_int64_t	rw_x_spin_wait_count;
+extern	ib_int64_t	rw_x_spin_round_count;
+extern	ib_int64_t	rw_x_os_wait_count;
+extern	ib_int64_t	rw_x_exit_count;
+
+/**********************************************************************
+Creates, or rather, initializes an rw-lock object in a specified memory
+location (which must be appropriately aligned). The rw-lock is initialized
+to the non-locked state. Explicit freeing of the rw-lock with rw_lock_free
+is necessary only if the memory block containing it is freed. */
+#ifdef UNIV_DEBUG
+# ifdef UNIV_SYNC_DEBUG
+#  define rw_lock_create(L, level) 					\
+	rw_lock_create_func((L), (level), #L, __FILE__, __LINE__)
+# else /* UNIV_SYNC_DEBUG */
+#  define rw_lock_create(L, level) 					\
+	rw_lock_create_func((L), #L, __FILE__, __LINE__)
+# endif /* UNIV_SYNC_DEBUG */
+#else /* UNIV_DEBUG */
+# define rw_lock_create(L, level) 					\
+	rw_lock_create_func((L), __FILE__, __LINE__)
+#endif /* UNIV_DEBUG */
+
+/**********************************************************************
+Creates, or rather, initializes an rw-lock object in a specified memory
+location (which must be appropriately aligned). The rw-lock is initialized
+to the non-locked state. Explicit freeing of the rw-lock with rw_lock_free
+is necessary only if the memory block containing it is freed. */
+UNIV_INTERN
+void
+rw_lock_create_func(
+/*================*/
+	rw_lock_t*	lock,		/* in: pointer to memory */
+#ifdef UNIV_DEBUG
+# ifdef UNIV_SYNC_DEBUG
+	ulint		level,		/* in: level */
+# endif /* UNIV_SYNC_DEBUG */
+	const char*	cmutex_name, 	/* in: mutex name */
+#endif /* UNIV_DEBUG */
+	const char*	cfile_name,	/* in: file name where created */
+	ulint 		cline);		/* in: file line where created */
+/**********************************************************************
+Calling this function is obligatory only if the memory buffer containing
+the rw-lock is freed. Removes an rw-lock object from the global list. The
+rw-lock is checked to be in the non-locked state. */
+UNIV_INTERN
+void
+rw_lock_free(
+/*=========*/
+	rw_lock_t*	lock);	/* in: rw-lock */
+#ifdef UNIV_DEBUG
+/**********************************************************************
+Checks that the rw-lock has been initialized and that there are no
+simultaneous shared and exclusive locks. */
+UNIV_INTERN
+ibool
+rw_lock_validate(
+/*=============*/
+	rw_lock_t*	lock);
+#endif /* UNIV_DEBUG */
+/******************************************************************
+NOTE! The following macros should be used in rw s-locking, not the
+corresponding function. */
+
+#define rw_lock_s_lock(M)	rw_lock_s_lock_func(\
+		(M), 0, __FILE__, __LINE__)
+/******************************************************************
+NOTE! The following macros should be used in rw s-locking, not the
+corresponding function. */
+
+#define rw_lock_s_lock_gen(M, P)	rw_lock_s_lock_func(\
+		(M), (P), __FILE__, __LINE__)
+/******************************************************************
+NOTE! The following macros should be used in rw s-locking, not the
+corresponding function. */
+
+#define rw_lock_s_lock_nowait(M, F, L)    rw_lock_s_lock_low(\
+					  (M), 0, (F), (L))
+/**********************************************************************
+Low-level function which tries to lock an rw-lock in s-mode. Performs no
+spinning. */
+UNIV_INLINE
+ibool
+rw_lock_s_lock_low(
+/*===============*/
+				/* out: TRUE if success */
+	rw_lock_t*	lock,	/* in: pointer to rw-lock */
+	ulint		pass __attribute__((unused)),
+				/* in: pass value; != 0, if the lock will be
+				passed to another thread to unlock */
+	const char*	file_name, /* in: file name where lock requested */
+	ulint		line);	/* in: line where requested */
+/**********************************************************************
+NOTE! Use the corresponding macro, not directly this function, except if
+you supply the file name and line number. Lock an rw-lock in shared mode
+for the current thread. If the rw-lock is locked in exclusive mode, or
+there is an exclusive lock request waiting, the function spins a preset
+time (controlled by SYNC_SPIN_ROUNDS), waiting for the lock, before
+suspending the thread. */
+UNIV_INLINE
+void
+rw_lock_s_lock_func(
+/*================*/
+	rw_lock_t*	lock,	/* in: pointer to rw-lock */
+	ulint		pass,	/* in: pass value; != 0, if the lock will
+				be passed to another thread to unlock */
+	const char*	file_name,/* in: file name where lock requested */
+	ulint		line);	/* in: line where requested */
+/**********************************************************************
+NOTE! Use the corresponding macro, not directly this function! Lock an
+rw-lock in exclusive mode for the current thread if the lock can be
+obtained immediately. */
+UNIV_INLINE
+ibool
+rw_lock_x_lock_func_nowait(
+/*=======================*/
+				/* out: TRUE if success */
+	rw_lock_t*	lock,	/* in: pointer to rw-lock */
+	const char*	file_name,/* in: file name where lock requested */
+	ulint		line);	/* in: line where requested */
+/**********************************************************************
+Releases a shared mode lock. */
+UNIV_INLINE
+void
+rw_lock_s_unlock_func(
+/*==================*/
+	rw_lock_t*	lock	/* in: rw-lock */
+#ifdef UNIV_SYNC_DEBUG
+	,ulint		pass	/* in: pass value; != 0, if the lock may have
+				been passed to another thread to unlock */
+#endif
+	);
+/***********************************************************************
+Releases a shared mode lock. */
+
+#ifdef UNIV_SYNC_DEBUG
+#define rw_lock_s_unlock(L)	rw_lock_s_unlock_func(L, 0)
+#else
+#define rw_lock_s_unlock(L)	rw_lock_s_unlock_func(L)
+#endif
+/***********************************************************************
+Releases a shared mode lock. */
+
+#ifdef UNIV_SYNC_DEBUG
+#define rw_lock_s_unlock_gen(L, P)	rw_lock_s_unlock_func(L, P)
+#else
+#define rw_lock_s_unlock_gen(L, P)	rw_lock_s_unlock_func(L)
+#endif
+/******************************************************************
+NOTE! The following macro should be used in rw x-locking, not the
+corresponding function. */
+
+#define rw_lock_x_lock(M)	rw_lock_x_lock_func(\
+		(M), 0, __FILE__, __LINE__)
+/******************************************************************
+NOTE! The following macro should be used in rw x-locking, not the
+corresponding function. */
+
+#define rw_lock_x_lock_gen(M, P)	rw_lock_x_lock_func(\
+		(M), (P), __FILE__, __LINE__)
+/******************************************************************
+NOTE! The following macros should be used in rw x-locking, not the
+corresponding function. */
+
+#define rw_lock_x_lock_nowait(M)	rw_lock_x_lock_func_nowait(\
+		(M), __FILE__, __LINE__)
+/**********************************************************************
+NOTE! Use the corresponding macro, not directly this function! Lock an
+rw-lock in exclusive mode for the current thread. If the rw-lock is locked
+in shared or exclusive mode, or there is an exclusive lock request waiting,
+the function spins a preset time (controlled by SYNC_SPIN_ROUNDS), waiting
+for the lock, before suspending the thread. If the same thread has an x-lock
+on the rw-lock, locking succeed, with the following exception: if pass != 0,
+only a single x-lock may be taken on the lock. NOTE: If the same thread has
+an s-lock, locking does not succeed! */
+UNIV_INTERN
+void
+rw_lock_x_lock_func(
+/*================*/
+	rw_lock_t*	lock,	/* in: pointer to rw-lock */
+	ulint		pass,	/* in: pass value; != 0, if the lock will
+				be passed to another thread to unlock */
+	const char*	file_name,/* in: file name where lock requested */
+	ulint		line);	/* in: line where requested */
+/**********************************************************************
+Releases an exclusive mode lock. */
+UNIV_INLINE
+void
+rw_lock_x_unlock_func(
+/*==================*/
+	rw_lock_t*	lock	/* in: rw-lock */
+#ifdef UNIV_SYNC_DEBUG
+	,ulint		pass	/* in: pass value; != 0, if the lock may have
+				been passed to another thread to unlock */
+#endif
+	);
+/***********************************************************************
+Releases an exclusive mode lock. */
+
+#ifdef UNIV_SYNC_DEBUG
+#define rw_lock_x_unlock(L)	rw_lock_x_unlock_func(L, 0)
+#else
+#define rw_lock_x_unlock(L)	rw_lock_x_unlock_func(L)
+#endif
+/***********************************************************************
+Releases an exclusive mode lock. */
+
+#ifdef UNIV_SYNC_DEBUG
+#define rw_lock_x_unlock_gen(L, P)	rw_lock_x_unlock_func(L, P)
+#else
+#define rw_lock_x_unlock_gen(L, P)	rw_lock_x_unlock_func(L)
+#endif
+/**********************************************************************
+Low-level function which locks an rw-lock in s-mode when we know that it
+is possible and none else is currently accessing the rw-lock structure.
+Then we can do the locking without reserving the mutex. */
+UNIV_INLINE
+void
+rw_lock_s_lock_direct(
+/*==================*/
+	rw_lock_t*	lock,		/* in: pointer to rw-lock */
+	const char*	file_name,	/* in: file name where requested */
+	ulint		line		/* in: line where lock requested */
+);
+/**********************************************************************
+Low-level function which locks an rw-lock in x-mode when we know that it
+is not locked and none else is currently accessing the rw-lock structure.
+Then we can do the locking without reserving the mutex. */
+UNIV_INLINE
+void
+rw_lock_x_lock_direct(
+/*==================*/
+	rw_lock_t*	lock,		/* in: pointer to rw-lock */
+	const char*	file_name,	/* in: file name where requested */
+	ulint		line		/* in: line where lock requested */
+);
+/**********************************************************************
+This function is used in the insert buffer to move the ownership of an
+x-latch on a buffer frame to the current thread. The x-latch was set by
+the buffer read operation and it protected the buffer frame while the
+read was done. The ownership is moved because we want that the current
+thread is able to acquire a second x-latch which is stored in an mtr.
+This, in turn, is needed to pass the debug checks of index page
+operations. */
+UNIV_INTERN
+void
+rw_lock_x_lock_move_ownership(
+/*==========================*/
+	rw_lock_t*	lock);	/* in: lock which was x-locked in the
+				buffer read */
+/**********************************************************************
+Releases a shared mode lock when we know there are no waiters and none
+else will access the lock during the time this function is executed. */
+UNIV_INLINE
+void
+rw_lock_s_unlock_direct(
+/*====================*/
+	rw_lock_t*	lock);	/* in: rw-lock */
+/**********************************************************************
+Releases an exclusive mode lock when we know there are no waiters, and
+none else will access the lock durint the time this function is executed. */
+UNIV_INLINE
+void
+rw_lock_x_unlock_direct(
+/*====================*/
+	rw_lock_t*	lock);	/* in: rw-lock */
+/**********************************************************************
+Returns the value of writer_count for the lock. Does not reserve the lock
+mutex, so the caller must be sure it is not changed during the call. */
+UNIV_INLINE
+ulint
+rw_lock_get_x_lock_count(
+/*=====================*/
+				/* out: value of writer_count */
+	rw_lock_t*	lock);	/* in: rw-lock */
+/************************************************************************
+Accessor functions for rw lock. */
+UNIV_INLINE
+ulint
+rw_lock_get_waiters(
+/*================*/
+	rw_lock_t*	lock);
+UNIV_INLINE
+ulint
+rw_lock_get_writer(
+/*===============*/
+	rw_lock_t*	lock);
+UNIV_INLINE
+ulint
+rw_lock_get_reader_count(
+/*=====================*/
+	rw_lock_t*	lock);
+/**********************************************************************
+Decrements lock_word the specified amount if it is greater than 0.
+This is used by both s_lock and x_lock operations. */
+UNIV_INLINE
+ibool
+rw_lock_lock_word_decr(
+/*===================*/
+					/* out: TRUE if decr occurs */
+	rw_lock_t*	lock,		/* in: rw-lock */
+	ulint		amount);	/* in: amount to decrement */
+/**********************************************************************
+Increments lock_word the specified amount and returns new value. */
+UNIV_INLINE
+lint
+rw_lock_lock_word_incr(
+/*===================*/
+					/* out: TRUE if decr occurs */
+	rw_lock_t*	lock,
+	ulint		amount);	/* in: rw-lock */
+/**********************************************************************
+This function sets the lock->writer_thread and lock->recursive fields.
+For platforms where we are using atomic builtins instead of lock->mutex
+it sets the lock->writer_thread field using atomics to ensure memory
+ordering. Note that it is assumed that the caller of this function
+effectively owns the lock i.e.: nobody else is allowed to modify
+lock->writer_thread at this point in time.
+The protocol is that lock->writer_thread MUST be updated BEFORE the
+lock->recursive flag is set. */
+UNIV_INLINE
+void
+rw_lock_set_writer_id_and_recursion_flag(
+/*=====================================*/
+	rw_lock_t*	lock,		/* in/out: lock to work on */
+	ibool		recursive);	/* in: TRUE if recursion
+					allowed */
+#ifdef UNIV_SYNC_DEBUG
+/**********************************************************************
+Checks if the thread has locked the rw-lock in the specified mode, with
+the pass value == 0. */
+UNIV_INTERN
+ibool
+rw_lock_own(
+/*========*/
+	rw_lock_t*	lock,		/* in: rw-lock */
+	ulint		lock_type);	/* in: lock type: RW_LOCK_SHARED,
+					RW_LOCK_EX */
+#endif /* UNIV_SYNC_DEBUG */
+/**********************************************************************
+Checks if somebody has locked the rw-lock in the specified mode. */
+UNIV_INTERN
+ibool
+rw_lock_is_locked(
+/*==============*/
+	rw_lock_t*	lock,		/* in: rw-lock */
+	ulint		lock_type);	/* in: lock type: RW_LOCK_SHARED,
+					RW_LOCK_EX */
+#ifdef UNIV_SYNC_DEBUG
+/*******************************************************************
+Prints debug info of an rw-lock. */
+UNIV_INTERN
+void
+rw_lock_print(
+/*==========*/
+	rw_lock_t*	lock);	/* in: rw-lock */
+/*******************************************************************
+Prints debug info of currently locked rw-locks. */
+UNIV_INTERN
+void
+rw_lock_list_print_info(
+/*====================*/
+	FILE*	file);		/* in: file where to print */
+/*******************************************************************
+Returns the number of currently locked rw-locks.
+Works only in the debug version. */
+UNIV_INTERN
+ulint
+rw_lock_n_locked(void);
+/*==================*/
+
+/*#####################################################################*/
+
+/**********************************************************************
+Acquires the debug mutex. We cannot use the mutex defined in sync0sync,
+because the debug mutex is also acquired in sync0arr while holding the OS
+mutex protecting the sync array, and the ordinary mutex_enter might
+recursively call routines in sync0arr, leading to a deadlock on the OS
+mutex. */
+UNIV_INTERN
+void
+rw_lock_debug_mutex_enter(void);
+/*==========================*/
+/**********************************************************************
+Releases the debug mutex. */
+UNIV_INTERN
+void
+rw_lock_debug_mutex_exit(void);
+/*==========================*/
+/*************************************************************************
+Prints info of a debug struct. */
+UNIV_INTERN
+void
+rw_lock_debug_print(
+/*================*/
+	rw_lock_debug_t*	info);	/* in: debug struct */
+#endif /* UNIV_SYNC_DEBUG */
+
+/* NOTE! The structure appears here only for the compiler to know its size.
+Do not use its fields directly! The structure used in the spin lock
+implementation of a read-write lock. Several threads may have a shared lock
+simultaneously in this lock, but only one writer may have an exclusive lock,
+in which case no shared locks are allowed. To prevent starving of a writer
+blocked by readers, a writer may queue for x-lock by decrementing lock_word:
+no new readers will be let in while the thread waits for readers to exit. */
+
+struct rw_lock_struct {
+	volatile lint	lock_word;
+				/* Holds the state of the lock. */
+	volatile ulint	waiters;/* 1: there are waiters */
+	volatile ibool	recursive;/* Default value FALSE which means the lock
+				is non-recursive. The value is typically set
+				to TRUE making normal rw_locks recursive. In
+				case of asynchronous IO, when a non-zero
+				value of 'pass' is passed then we keep the
+				lock non-recursive.
+				This flag also tells us about the state of
+				writer_thread field. If this flag is set
+				then writer_thread MUST contain the thread
+				id of the current x-holder or wait-x thread.
+				This flag must be reset in x_unlock
+				functions before incrementing the lock_word */
+	volatile os_thread_id_t	writer_thread;
+				/* Thread id of writer thread. Is only
+				guaranteed to have sane and non-stale
+				value iff recursive flag is set. */
+	os_event_t	event;	/* Used by sync0arr.c for thread queueing */
+	os_event_t	wait_ex_event;
+				/* Event for next-writer to wait on. A thread
+				must decrement lock_word before waiting. */
+#ifndef INNODB_RW_LOCKS_USE_ATOMICS
+	mutex_t	mutex;		/* The mutex protecting rw_lock_struct */
+#endif /* INNODB_RW_LOCKS_USE_ATOMICS */
+
+	UT_LIST_NODE_T(rw_lock_t) list;
+				/* All allocated rw locks are put into a
+				list */
+#ifdef UNIV_SYNC_DEBUG
+	UT_LIST_BASE_NODE_T(rw_lock_debug_t) debug_list;
+				/* In the debug version: pointer to the debug
+				info list of the lock */
+	ulint	level;		/* Level in the global latching order. */
+#endif /* UNIV_SYNC_DEBUG */
+	ulint count_os_wait;	/* Count of os_waits. May not be accurate */
+	const char*	cfile_name;/* File name where lock created */
+        /* last s-lock file/line is not guaranteed to be correct */
+	const char*	last_s_file_name;/* File name where last s-locked */
+	const char*	last_x_file_name;/* File name where last x-locked */
+	ibool		writer_is_wait_ex;
+				/* This is TRUE if the writer field is
+				RW_LOCK_WAIT_EX; this field is located far
+				from the memory update hotspot fields which
+				are at the start of this struct, thus we can
+				peek this field without causing much memory
+				bus traffic */
+	unsigned	cline:14;	/* Line where created */
+	unsigned	last_s_line:14;	/* Line number where last time s-locked */
+	unsigned	last_x_line:14;	/* Line number where last time x-locked */
+	ulint	magic_n;
+};
+
+#define	RW_LOCK_MAGIC_N	22643
+
+#ifdef UNIV_SYNC_DEBUG
+/* The structure for storing debug info of an rw-lock */
+struct	rw_lock_debug_struct {
+
+	os_thread_id_t thread_id;  /* The thread id of the thread which
+				locked the rw-lock */
+	ulint	pass;		/* Pass value given in the lock operation */
+	ulint	lock_type;	/* Type of the lock: RW_LOCK_EX,
+				RW_LOCK_SHARED, RW_LOCK_WAIT_EX */
+	const char*	file_name;/* File name where the lock was obtained */
+	ulint	line;		/* Line where the rw-lock was locked */
+	UT_LIST_NODE_T(rw_lock_debug_t) list;
+				/* Debug structs are linked in a two-way
+				list */
+};
+#endif /* UNIV_SYNC_DEBUG */
+
+#ifndef UNIV_NONINL
+#include "sync0rw.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/sync0rw.ic b/storage/xtradb/include/sync0rw.ic
new file mode 100644
index 00000000000..9e7e4dc9bd8
--- /dev/null
+++ b/storage/xtradb/include/sync0rw.ic
@@ -0,0 +1,635 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+Copyright (c) 2008, Google Inc.
+
+Portions of this file contain modifications contributed and copyrighted by
+Google, Inc. Those modifications are gratefully acknowledged and are described
+briefly in the InnoDB documentation. The contributions by Google are
+incorporated with their permission, and subject to the conditions contained in
+the file COPYING.Google.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The read-write lock (for threads)
+
+Created 9/11/1995 Heikki Tuuri
+*******************************************************/
+
+/**********************************************************************
+Lock an rw-lock in shared mode for the current thread. If the rw-lock is
+locked in exclusive mode, or there is an exclusive lock request waiting,
+the function spins a preset time (controlled by SYNC_SPIN_ROUNDS),
+waiting for the lock before suspending the thread. */
+UNIV_INTERN
+void
+rw_lock_s_lock_spin(
+/*================*/
+	rw_lock_t*	lock,	/* in: pointer to rw-lock */
+	ulint		pass,	/* in: pass value; != 0, if the lock will
+				be passed to another thread to unlock */
+	const char*	file_name,/* in: file name where lock requested */
+	ulint		line);	/* in: line where requested */
+#ifdef UNIV_SYNC_DEBUG
+/**********************************************************************
+Inserts the debug information for an rw-lock. */
+UNIV_INTERN
+void
+rw_lock_add_debug_info(
+/*===================*/
+	rw_lock_t*	lock,		/* in: rw-lock */
+	ulint		pass,		/* in: pass value */
+	ulint		lock_type,	/* in: lock type */
+	const char*	file_name,	/* in: file where requested */
+	ulint		line);		/* in: line where requested */
+/**********************************************************************
+Removes a debug information struct for an rw-lock. */
+UNIV_INTERN
+void
+rw_lock_remove_debug_info(
+/*======================*/
+	rw_lock_t*	lock,		/* in: rw-lock */
+	ulint		pass,		/* in: pass value */
+	ulint		lock_type);	/* in: lock type */
+#endif /* UNIV_SYNC_DEBUG */
+
+/************************************************************************
+Accessor functions for rw lock. */
+UNIV_INLINE
+ulint
+rw_lock_get_waiters(
+/*================*/
+				/* out: 1 if waiters, 0 otherwise */
+	rw_lock_t*	lock)	/* in: rw-lock */
+{
+	return(lock->waiters);
+}
+
+/************************************************************************
+Sets lock->waiters to 1. It is not an error if lock->waiters is already
+1. On platforms where ATOMIC builtins are used this function enforces a
+memory barrier. */
+UNIV_INLINE
+void
+rw_lock_set_waiter_flag(
+/*====================*/
+	rw_lock_t*	lock)	/* in: rw-lock */
+{
+#ifdef INNODB_RW_LOCKS_USE_ATOMICS
+	os_compare_and_swap(&lock->waiters, 0, 1);
+#else /* INNODB_RW_LOCKS_USE_ATOMICS */
+	lock->waiters = 1;
+#endif /* INNODB_RW_LOCKS_USE_ATOMICS */
+}
+
+/************************************************************************
+Resets lock->waiters to 0. It is not an error if lock->waiters is already
+0. On platforms where ATOMIC builtins are used this function enforces a
+memory barrier. */
+UNIV_INLINE
+void
+rw_lock_reset_waiter_flag(
+/*======================*/
+	rw_lock_t*	lock)	/* in: rw-lock */
+{
+#ifdef INNODB_RW_LOCKS_USE_ATOMICS
+	os_compare_and_swap(&lock->waiters, 1, 0);
+#else /* INNODB_RW_LOCKS_USE_ATOMICS */
+	lock->waiters = 0;
+#endif /* INNODB_RW_LOCKS_USE_ATOMICS */
+}
+
+/**********************************************************************
+Returns the write-status of the lock - this function made more sense
+with the old rw_lock implementation. */
+UNIV_INLINE
+ulint
+rw_lock_get_writer(
+/*===============*/
+	rw_lock_t*	lock)
+{
+	lint lock_word = lock->lock_word;
+	if(lock_word > 0) {
+		/* return NOT_LOCKED in s-lock state, like the writer
+		member of the old lock implementation. */
+		return(RW_LOCK_NOT_LOCKED);
+	} else if (((-lock_word) % X_LOCK_DECR) == 0) {
+		return(RW_LOCK_EX);
+	} else {
+                ut_ad(lock_word > -X_LOCK_DECR);
+		return(RW_LOCK_WAIT_EX);
+	}
+}
+
+/**********************************************************************
+Returns number of readers. */
+UNIV_INLINE
+ulint
+rw_lock_get_reader_count(
+/*=====================*/
+	rw_lock_t*	lock)
+{
+	lint lock_word = lock->lock_word;
+	if(lock_word > 0) {
+		/* s-locked, no x-waiters */
+		return(X_LOCK_DECR - lock_word);
+	} else if (lock_word < 0 && lock_word > -X_LOCK_DECR) {
+		/* s-locked, with x-waiters */
+		return((ulint)(-lock_word));
+	}
+	return(0);
+}
+
+#ifndef INNODB_RW_LOCKS_USE_ATOMICS
+UNIV_INLINE
+mutex_t*
+rw_lock_get_mutex(
+/*==============*/
+	rw_lock_t*	lock)
+{
+	return(&(lock->mutex));
+}
+#endif
+
+/**********************************************************************
+Returns the value of writer_count for the lock. Does not reserve the lock
+mutex, so the caller must be sure it is not changed during the call. */
+UNIV_INLINE
+ulint
+rw_lock_get_x_lock_count(
+/*=====================*/
+				/* out: value of writer_count */
+	rw_lock_t*	lock)	/* in: rw-lock */
+{
+	lint lock_copy = lock->lock_word;
+	/* If there is a reader, lock_word is not divisible by X_LOCK_DECR */
+	if(lock_copy > 0 || (-lock_copy) % X_LOCK_DECR != 0) {
+		return(0);
+	}
+	return(((-lock_copy) / X_LOCK_DECR) + 1);
+}
+
+/**********************************************************************
+Two different implementations for decrementing the lock_word of a rw_lock:
+one for systems supporting atomic operations, one for others. This does
+does not support recusive x-locks: they should be handled by the caller and
+need not be atomic since they are performed by the current lock holder.
+Returns true if the decrement was made, false if not. */
+UNIV_INLINE
+ibool
+rw_lock_lock_word_decr(
+/*===================*/
+				/* out: TRUE if decr occurs */
+	rw_lock_t*	lock,	/* in: rw-lock */
+	ulint		amount)	/* in: amount of decrement */
+{
+
+#ifdef INNODB_RW_LOCKS_USE_ATOMICS
+
+        lint local_lock_word = lock->lock_word;
+	while (local_lock_word > 0) {
+		if(os_compare_and_swap(&(lock->lock_word),
+                                       local_lock_word,
+                                       local_lock_word - amount)) {
+			return(TRUE);
+		}
+		local_lock_word = lock->lock_word;
+	}
+	return(FALSE);
+
+#else /* INNODB_RW_LOCKS_USE_ATOMICS */
+
+	ibool success = FALSE;
+	mutex_enter(&(lock->mutex));
+	if(lock->lock_word > 0) {
+		lock->lock_word -= amount;
+		success = TRUE;
+	}
+	mutex_exit(&(lock->mutex));
+	return(success);
+
+#endif /* INNODB_RW_LOCKS_USE_ATOMICS */
+}
+
+/**********************************************************************
+Two different implementations for incrementing the lock_word of a rw_lock:
+one for systems supporting atomic operations, one for others.
+Returns the value of lock_word after increment. */
+UNIV_INLINE
+lint
+rw_lock_lock_word_incr(
+/*===================*/
+				/* out: lock->lock_word after increment */
+	rw_lock_t*	lock,	/* in: rw-lock */
+	ulint		amount)	/* in: amount of increment */
+{
+
+#ifdef INNODB_RW_LOCKS_USE_ATOMICS
+
+	return(os_atomic_increment(&(lock->lock_word), amount));
+
+#else /* INNODB_RW_LOCKS_USE_ATOMICS */
+
+	lint local_lock_word;
+
+	mutex_enter(&(lock->mutex));
+
+	lock->lock_word += amount;
+	local_lock_word = lock->lock_word;
+
+	mutex_exit(&(lock->mutex));
+
+        return(local_lock_word);
+
+#endif /* INNODB_RW_LOCKS_USE_ATOMICS */
+}
+
+/**********************************************************************
+This function sets the lock->writer_thread and lock->recursive fields.
+For platforms where we are using atomic builtins instead of lock->mutex
+it sets the lock->writer_thread field using atomics to ensure memory
+ordering. Note that it is assumed that the caller of this function
+effectively owns the lock i.e.: nobody else is allowed to modify
+lock->writer_thread at this point in time.
+The protocol is that lock->writer_thread MUST be updated BEFORE the
+lock->recursive flag is set. */
+UNIV_INLINE
+void
+rw_lock_set_writer_id_and_recursion_flag(
+/*=====================================*/
+	rw_lock_t*	lock,		/* in/out: lock to work on */
+	ibool		recursive)	/* in: TRUE if recursion
+					allowed */
+{
+	os_thread_id_t	curr_thread	= os_thread_get_curr_id();
+
+#ifdef INNODB_RW_LOCKS_USE_ATOMICS
+	os_thread_id_t	local_thread;
+	ibool		success;
+
+	/* Prevent Valgrind warnings about writer_thread being
+	uninitialized.  It does not matter if writer_thread is
+	uninitialized, because we are comparing writer_thread against
+	itself, and the operation should always succeed. */
+	UNIV_MEM_VALID(&lock->writer_thread, sizeof lock->writer_thread);
+
+	local_thread = lock->writer_thread;
+	success = os_compare_and_swap(&lock->writer_thread,
+				      local_thread, curr_thread);
+	ut_a(success);
+	lock->recursive = recursive;
+
+#else /* INNODB_RW_LOCKS_USE_ATOMICS */
+
+	mutex_enter(&lock->mutex);
+	lock->writer_thread = curr_thread;
+	lock->recursive = recursive;
+	mutex_exit(&lock->mutex);
+
+#endif /* INNODB_RW_LOCKS_USE_ATOMICS */
+}
+
+/**********************************************************************
+Low-level function which tries to lock an rw-lock in s-mode. Performs no
+spinning. */
+UNIV_INLINE
+ibool
+rw_lock_s_lock_low(
+/*===============*/
+				/* out: TRUE if success */
+	rw_lock_t*	lock,	/* in: pointer to rw-lock */
+	ulint		pass __attribute__((unused)),
+				/* in: pass value; != 0, if the lock will be
+				passed to another thread to unlock */
+	const char*	file_name, /* in: file name where lock requested */
+	ulint		line)	/* in: line where requested */
+{
+	/* TODO: study performance of UNIV_LIKELY branch prediction hints. */
+	if (!rw_lock_lock_word_decr(lock, 1)) {
+		/* Locking did not succeed */
+		return(FALSE);
+	}
+
+#ifdef UNIV_SYNC_DEBUG
+	rw_lock_add_debug_info(lock, pass, RW_LOCK_SHARED, file_name, line);
+#endif
+	/* These debugging values are not set safely: they may be incorrect
+        or even refer to a line that is invalid for the file name. */
+	lock->last_s_file_name = file_name;
+	lock->last_s_line = line;
+
+	return(TRUE);	/* locking succeeded */
+}
+
+/**********************************************************************
+Low-level function which locks an rw-lock in s-mode when we know that it
+is possible and none else is currently accessing the rw-lock structure.
+Then we can do the locking without reserving the mutex. */
+UNIV_INLINE
+void
+rw_lock_s_lock_direct(
+/*==================*/
+	rw_lock_t*	lock,		/* in: pointer to rw-lock */
+	const char*	file_name,	/* in: file name where requested */
+	ulint		line)		/* in: line where lock requested */
+{
+	ut_ad(lock->lock_word == X_LOCK_DECR);
+
+	/* Indicate there is a new reader by decrementing lock_word */
+	lock->lock_word--;
+
+	lock->last_s_file_name = file_name;
+	lock->last_s_line = line;
+
+#ifdef UNIV_SYNC_DEBUG
+	rw_lock_add_debug_info(lock, 0, RW_LOCK_SHARED, file_name, line);
+#endif
+}
+
+/**********************************************************************
+Low-level function which locks an rw-lock in x-mode when we know that it
+is not locked and none else is currently accessing the rw-lock structure.
+Then we can do the locking without reserving the mutex. */
+UNIV_INLINE
+void
+rw_lock_x_lock_direct(
+/*==================*/
+	rw_lock_t*	lock,		/* in: pointer to rw-lock */
+	const char*	file_name,	/* in: file name where requested */
+	ulint		line)		/* in: line where lock requested */
+{
+	ut_ad(rw_lock_validate(lock));
+	ut_ad(lock->lock_word == X_LOCK_DECR);
+
+	lock->lock_word -= X_LOCK_DECR;
+	lock->writer_thread = os_thread_get_curr_id();
+	lock->recursive = TRUE;
+
+	lock->last_x_file_name = file_name;
+	lock->last_x_line = line;
+
+#ifdef UNIV_SYNC_DEBUG
+	rw_lock_add_debug_info(lock, 0, RW_LOCK_EX, file_name, line);
+#endif
+}
+
+/**********************************************************************
+NOTE! Use the corresponding macro, not directly this function! Lock an
+rw-lock in shared mode for the current thread. If the rw-lock is locked
+in exclusive mode, or there is an exclusive lock request waiting, the
+function spins a preset time (controlled by SYNC_SPIN_ROUNDS), waiting for
+the lock, before suspending the thread. */
+UNIV_INLINE
+void
+rw_lock_s_lock_func(
+/*================*/
+	rw_lock_t*	lock,	/* in: pointer to rw-lock */
+	ulint		pass,	/* in: pass value; != 0, if the lock will
+				be passed to another thread to unlock */
+	const char*	file_name,/* in: file name where lock requested */
+	ulint		line)	/* in: line where requested */
+{
+	/* NOTE: As we do not know the thread ids for threads which have
+	s-locked a latch, and s-lockers will be served only after waiting
+	x-lock requests have been fulfilled, then if this thread already
+	owns an s-lock here, it may end up in a deadlock with another thread
+	which requests an x-lock here. Therefore, we will forbid recursive
+	s-locking of a latch: the following assert will warn the programmer
+	of the possibility of this kind of a deadlock. If we want to implement
+	safe recursive s-locking, we should keep in a list the thread ids of
+	the threads which have s-locked a latch. This would use some CPU
+	time. */
+
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(!rw_lock_own(lock, RW_LOCK_SHARED)); /* see NOTE above */
+#endif /* UNIV_SYNC_DEBUG */
+
+	/* TODO: study performance of UNIV_LIKELY branch prediction hints. */
+	if (rw_lock_s_lock_low(lock, pass, file_name, line)) {
+
+		return; /* Success */
+	} else {
+		/* Did not succeed, try spin wait */
+
+		rw_lock_s_lock_spin(lock, pass, file_name, line);
+
+		return;
+	}
+}
+
+/**********************************************************************
+NOTE! Use the corresponding macro, not directly this function! Lock an
+rw-lock in exclusive mode for the current thread if the lock can be
+obtained immediately. */
+UNIV_INLINE
+ibool
+rw_lock_x_lock_func_nowait(
+/*=======================*/
+				/* out: TRUE if success */
+	rw_lock_t*	lock,	/* in: pointer to rw-lock */
+	const char*	file_name,/* in: file name where lock requested */
+	ulint		line)	/* in: line where requested */
+{
+	os_thread_id_t	curr_thread	= os_thread_get_curr_id();
+
+	ibool success;
+
+#ifdef INNODB_RW_LOCKS_USE_ATOMICS
+	success = os_compare_and_swap(&(lock->lock_word), X_LOCK_DECR, 0);
+#else
+
+	success = FALSE;
+	mutex_enter(&(lock->mutex));
+	if (lock->lock_word == X_LOCK_DECR) {
+		lock->lock_word = 0;
+		success = TRUE;
+	}
+	mutex_exit(&(lock->mutex));
+
+#endif
+	if (success) {
+		rw_lock_set_writer_id_and_recursion_flag(lock, TRUE);
+
+	} else if (lock->recursive
+		   && os_thread_eq(lock->writer_thread, curr_thread)) {
+		/* Relock: this lock_word modification is safe since no other
+		threads can modify (lock, unlock, or reserve) lock_word while
+		there is an exclusive writer and this is the writer thread. */
+		lock->lock_word -= X_LOCK_DECR;
+
+		ut_ad(((-lock->lock_word) % X_LOCK_DECR) == 0);
+
+	} else {
+		/* Failure */
+		return(FALSE);
+	}
+#ifdef UNIV_SYNC_DEBUG
+	rw_lock_add_debug_info(lock, 0, RW_LOCK_EX, file_name, line);
+#endif
+
+	lock->last_x_file_name = file_name;
+	lock->last_x_line = line;
+
+	ut_ad(rw_lock_validate(lock));
+
+	return(TRUE);
+}
+
+/**********************************************************************
+Releases a shared mode lock. */
+UNIV_INLINE
+void
+rw_lock_s_unlock_func(
+/*==================*/
+	rw_lock_t*	lock	/* in: rw-lock */
+#ifdef UNIV_SYNC_DEBUG
+	,ulint		pass	/* in: pass value; != 0, if the lock may have
+				been passed to another thread to unlock */
+#endif
+	)
+{
+	ut_ad((lock->lock_word % X_LOCK_DECR) != 0);
+
+#ifdef UNIV_SYNC_DEBUG
+	rw_lock_remove_debug_info(lock, pass, RW_LOCK_SHARED);
+#endif
+
+	/* Increment lock_word to indicate 1 less reader */
+	if (rw_lock_lock_word_incr(lock, 1) == 0) {
+
+		/* wait_ex waiter exists. It may not be asleep, but we signal
+                anyway. We do not wake other waiters, because they can't
+                exist without wait_ex waiter and wait_ex waiter goes first.*/
+		os_event_set(lock->wait_ex_event);
+		sync_array_object_signalled(sync_primary_wait_array);
+
+	}
+
+	ut_ad(rw_lock_validate(lock));
+
+#ifdef UNIV_SYNC_PERF_STAT
+	rw_s_exit_count++;
+#endif
+}
+
+/**********************************************************************
+Releases a shared mode lock when we know there are no waiters and none
+else will access the lock during the time this function is executed. */
+UNIV_INLINE
+void
+rw_lock_s_unlock_direct(
+/*====================*/
+	rw_lock_t*	lock)	/* in: rw-lock */
+{
+	ut_ad(lock->lock_word < X_LOCK_DECR);
+
+#ifdef UNIV_SYNC_DEBUG
+	rw_lock_remove_debug_info(lock, 0, RW_LOCK_SHARED);
+#endif
+
+	/* Decrease reader count by incrementing lock_word */
+	lock->lock_word++;
+
+	ut_ad(!lock->waiters);
+	ut_ad(rw_lock_validate(lock));
+#ifdef UNIV_SYNC_PERF_STAT
+	rw_s_exit_count++;
+#endif
+}
+
+/**********************************************************************
+Releases an exclusive mode lock. */
+UNIV_INLINE
+void
+rw_lock_x_unlock_func(
+/*==================*/
+	rw_lock_t*	lock	/* in: rw-lock */
+#ifdef UNIV_SYNC_DEBUG
+	,ulint		pass	/* in: pass value; != 0, if the lock may have
+				been passed to another thread to unlock */
+#endif
+	)
+{
+	ut_ad((lock->lock_word % X_LOCK_DECR) == 0);
+
+	/* lock->recursive flag also indicates if lock->writer_thread is
+	valid or stale. If we are the last of the recursive callers
+	then we must unset lock->recursive flag to indicate that the
+	lock->writer_thread is now stale.
+	Note that since we still hold the x-lock we can safely read the
+	lock_word. */
+	if (lock->lock_word == 0) {
+		/* Last caller in a possible recursive chain. */
+		lock->recursive = FALSE;
+		UNIV_MEM_INVALID(&lock->writer_thread,
+				 sizeof lock->writer_thread);
+	}
+
+#ifdef UNIV_SYNC_DEBUG
+	rw_lock_remove_debug_info(lock, pass, RW_LOCK_EX);
+#endif
+
+	if (rw_lock_lock_word_incr(lock, X_LOCK_DECR) == X_LOCK_DECR) {
+		/* Lock is now free. May have to signal read/write waiters.
+                We do not need to signal wait_ex waiters, since they cannot
+                exist when there is a writer. */
+		if (lock->waiters) {
+			rw_lock_reset_waiter_flag(lock);
+			os_event_set(lock->event);
+			sync_array_object_signalled(sync_primary_wait_array);
+		}
+	}
+
+	ut_ad(rw_lock_validate(lock));
+
+#ifdef UNIV_SYNC_PERF_STAT
+	rw_x_exit_count++;
+#endif
+}
+
+/**********************************************************************
+Releases an exclusive mode lock when we know there are no waiters, and
+none else will access the lock during the time this function is executed. */
+UNIV_INLINE
+void
+rw_lock_x_unlock_direct(
+/*====================*/
+	rw_lock_t*	lock)	/* in: rw-lock */
+{
+	/* Reset the exclusive lock if this thread no longer has an x-mode
+	lock */
+
+	ut_ad((lock->lock_word % X_LOCK_DECR) == 0);
+
+#ifdef UNIV_SYNC_DEBUG
+	rw_lock_remove_debug_info(lock, 0, RW_LOCK_EX);
+#endif
+
+	if (lock->lock_word == 0) {
+		lock->recursive = FALSE;
+		UNIV_MEM_INVALID(&lock->writer_thread,
+				 sizeof lock->writer_thread);
+	}
+
+	lock->lock_word += X_LOCK_DECR;
+
+	ut_ad(!lock->waiters);
+	ut_ad(rw_lock_validate(lock));
+
+#ifdef UNIV_SYNC_PERF_STAT
+	rw_x_exit_count++;
+#endif
+}
diff --git a/storage/xtradb/include/sync0sync.h b/storage/xtradb/include/sync0sync.h
new file mode 100644
index 00000000000..ea4abddbbf4
--- /dev/null
+++ b/storage/xtradb/include/sync0sync.h
@@ -0,0 +1,569 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+Copyright (c) 2008, Google Inc.
+
+Portions of this file contain modifications contributed and copyrighted by
+Google, Inc. Those modifications are gratefully acknowledged and are described
+briefly in the InnoDB documentation. The contributions by Google are
+incorporated with their permission, and subject to the conditions contained in
+the file COPYING.Google.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Mutex, the basic synchronization primitive
+
+Created 9/5/1995 Heikki Tuuri
+*******************************************************/
+
+#ifndef sync0sync_h
+#define sync0sync_h
+
+#include "univ.i"
+#include "sync0types.h"
+#include "ut0lst.h"
+#include "ut0mem.h"
+#include "os0thread.h"
+#include "os0sync.h"
+#include "sync0arr.h"
+
+#ifndef UNIV_HOTBACKUP
+extern my_bool	timed_mutexes;
+#endif /* UNIV_HOTBACKUP */
+
+/**********************************************************************
+Initializes the synchronization data structures. */
+UNIV_INTERN
+void
+sync_init(void);
+/*===========*/
+/**********************************************************************
+Frees the resources in synchronization data structures. */
+UNIV_INTERN
+void
+sync_close(void);
+/*===========*/
+/**********************************************************************
+Creates, or rather, initializes a mutex object to a specified memory
+location (which must be appropriately aligned). The mutex is initialized
+in the reset state. Explicit freeing of the mutex with mutex_free is
+necessary only if the memory block containing it is freed. */
+
+#ifdef UNIV_DEBUG
+# ifdef UNIV_SYNC_DEBUG
+#  define mutex_create(M, level)					\
+	mutex_create_func((M), #M, (level), __FILE__, __LINE__)
+# else
+#  define mutex_create(M, level)					\
+	mutex_create_func((M), #M, __FILE__, __LINE__)
+# endif
+#else
+# define mutex_create(M, level)					\
+	mutex_create_func((M), __FILE__, __LINE__)
+#endif
+
+/**********************************************************************
+Creates, or rather, initializes a mutex object in a specified memory
+location (which must be appropriately aligned). The mutex is initialized
+in the reset state. Explicit freeing of the mutex with mutex_free is
+necessary only if the memory block containing it is freed. */
+UNIV_INTERN
+void
+mutex_create_func(
+/*==============*/
+	mutex_t*	mutex,		/* in: pointer to memory */
+#ifdef UNIV_DEBUG
+	const char*	cmutex_name,	/* in: mutex name */
+# ifdef UNIV_SYNC_DEBUG
+	ulint		level,		/* in: level */
+# endif /* UNIV_SYNC_DEBUG */
+#endif /* UNIV_DEBUG */
+	const char*	cfile_name,	/* in: file name where created */
+	ulint		cline);		/* in: file line where created */
+
+#undef mutex_free			/* Fix for MacOS X */
+
+/**********************************************************************
+Calling this function is obligatory only if the memory buffer containing
+the mutex is freed. Removes a mutex object from the mutex list. The mutex
+is checked to be in the reset state. */
+UNIV_INTERN
+void
+mutex_free(
+/*=======*/
+	mutex_t*	mutex);	/* in: mutex */
+/******************************************************************
+NOTE! The following macro should be used in mutex locking, not the
+corresponding function. */
+
+#define mutex_enter(M)	  mutex_enter_func((M), __FILE__, __LINE__)
+/******************************************************************
+NOTE! The following macro should be used in mutex locking, not the
+corresponding function. */
+
+/* NOTE! currently same as mutex_enter! */
+
+#define mutex_enter_fast(M)	mutex_enter_func((M), __FILE__, __LINE__)
+/**********************************************************************
+NOTE! Use the corresponding macro in the header file, not this function
+directly. Locks a mutex for the current thread. If the mutex is reserved
+the function spins a preset time (controlled by SYNC_SPIN_ROUNDS) waiting
+for the mutex before suspending the thread. */
+UNIV_INLINE
+void
+mutex_enter_func(
+/*=============*/
+	mutex_t*	mutex,		/* in: pointer to mutex */
+	const char*	file_name,	/* in: file name where locked */
+	ulint		line);		/* in: line where locked */
+/******************************************************************
+NOTE! The following macro should be used in mutex locking, not the
+corresponding function. */
+
+#define mutex_enter_nowait(M)	\
+	mutex_enter_nowait_func((M), __FILE__, __LINE__)
+/************************************************************************
+NOTE! Use the corresponding macro in the header file, not this function
+directly. Tries to lock the mutex for the current thread. If the lock is not
+acquired immediately, returns with return value 1. */
+UNIV_INTERN
+ulint
+mutex_enter_nowait_func(
+/*====================*/
+					/* out: 0 if succeed, 1 if not */
+	mutex_t*	mutex,		/* in: pointer to mutex */
+	const char*	file_name,	/* in: file name where mutex
+					requested */
+	ulint		line);		/* in: line where requested */
+/**********************************************************************
+Unlocks a mutex owned by the current thread. */
+UNIV_INLINE
+void
+mutex_exit(
+/*=======*/
+	mutex_t*	mutex);	/* in: pointer to mutex */
+/**********************************************************************
+Returns TRUE if no mutex or rw-lock is currently locked.
+Works only in the debug version. */
+UNIV_INTERN
+ibool
+sync_all_freed(void);
+/*================*/
+/*#####################################################################
+FUNCTION PROTOTYPES FOR DEBUGGING */
+/***********************************************************************
+Prints wait info of the sync system. */
+UNIV_INTERN
+void
+sync_print_wait_info(
+/*=================*/
+	FILE*	file);		/* in: file where to print */
+/***********************************************************************
+Prints info of the sync system. */
+UNIV_INTERN
+void
+sync_print(
+/*=======*/
+	FILE*	file);		/* in: file where to print */
+#ifdef UNIV_DEBUG
+/**********************************************************************
+Checks that the mutex has been initialized. */
+UNIV_INTERN
+ibool
+mutex_validate(
+/*===========*/
+	const mutex_t*	mutex);
+/**********************************************************************
+Checks that the current thread owns the mutex. Works only
+in the debug version. */
+UNIV_INTERN
+ibool
+mutex_own(
+/*======*/
+				/* out: TRUE if owns */
+	const mutex_t*	mutex);	/* in: mutex */
+#endif /* UNIV_DEBUG */
+#ifdef UNIV_SYNC_DEBUG
+/**********************************************************************
+Adds a latch and its level in the thread level array. Allocates the memory
+for the array if called first time for this OS thread. Makes the checks
+against other latch levels stored in the array for this thread. */
+UNIV_INTERN
+void
+sync_thread_add_level(
+/*==================*/
+	void*	latch,	/* in: pointer to a mutex or an rw-lock */
+	ulint	level);	/* in: level in the latching order; if
+			SYNC_LEVEL_VARYING, nothing is done */
+/**********************************************************************
+Removes a latch from the thread level array if it is found there. */
+UNIV_INTERN
+ibool
+sync_thread_reset_level(
+/*====================*/
+			/* out: TRUE if found from the array; it is no error
+			if the latch is not found, as we presently are not
+			able to determine the level for every latch
+			reservation the program does */
+	void*	latch);	/* in: pointer to a mutex or an rw-lock */
+/**********************************************************************
+Checks that the level array for the current thread is empty. */
+UNIV_INTERN
+ibool
+sync_thread_levels_empty(void);
+/*==========================*/
+			/* out: TRUE if empty */
+/**********************************************************************
+Checks that the level array for the current thread is empty. */
+UNIV_INTERN
+ibool
+sync_thread_levels_empty_gen(
+/*=========================*/
+					/* out: TRUE if empty except the
+					exceptions specified below */
+	ibool	dict_mutex_allowed);	/* in: TRUE if dictionary mutex is
+					allowed to be owned by the thread,
+					also purge_is_running mutex is
+					allowed */
+/**********************************************************************
+Gets the debug information for a reserved mutex. */
+UNIV_INTERN
+void
+mutex_get_debug_info(
+/*=================*/
+	mutex_t*	mutex,		/* in: mutex */
+	const char**	file_name,	/* out: file where requested */
+	ulint*		line,		/* out: line where requested */
+	os_thread_id_t* thread_id);	/* out: id of the thread which owns
+					the mutex */
+/**********************************************************************
+Counts currently reserved mutexes. Works only in the debug version. */
+UNIV_INTERN
+ulint
+mutex_n_reserved(void);
+/*==================*/
+#endif /* UNIV_SYNC_DEBUG */
+/**********************************************************************
+NOT to be used outside this module except in debugging! Gets the value
+of the lock word. */
+UNIV_INLINE
+byte
+mutex_get_lock_word(
+/*================*/
+	const mutex_t*	mutex);	/* in: mutex */
+#ifdef UNIV_SYNC_DEBUG
+/**********************************************************************
+NOT to be used outside this module except in debugging! Gets the waiters
+field in a mutex. */
+UNIV_INLINE
+ulint
+mutex_get_waiters(
+/*==============*/
+				/* out: value to set */
+	const mutex_t*	mutex);	/* in: mutex */
+#endif /* UNIV_SYNC_DEBUG */
+
+/*
+		LATCHING ORDER WITHIN THE DATABASE
+		==================================
+
+The mutex or latch in the central memory object, for instance, a rollback
+segment object, must be acquired before acquiring the latch or latches to
+the corresponding file data structure. In the latching order below, these
+file page object latches are placed immediately below the corresponding
+central memory object latch or mutex.
+
+Synchronization object			Notes
+----------------------			-----
+
+Dictionary mutex			If we have a pointer to a dictionary
+|					object, e.g., a table, it can be
+|					accessed without reserving the
+|					dictionary mutex. We must have a
+|					reservation, a memoryfix, to the
+|					appropriate table object in this case,
+|					and the table must be explicitly
+|					released later.
+V
+Dictionary header
+|
+V
+Secondary index tree latch		The tree latch protects also all
+|					the B-tree non-leaf pages. These
+V					can be read with the page only
+Secondary index non-leaf		bufferfixed to save CPU time,
+|					no s-latch is needed on the page.
+|					Modification of a page requires an
+|					x-latch on the page, however. If a
+|					thread owns an x-latch to the tree,
+|					it is allowed to latch non-leaf pages
+|					even after it has acquired the fsp
+|					latch.
+V
+Secondary index leaf			The latch on the secondary index leaf
+|					can be kept while accessing the
+|					clustered index, to save CPU time.
+V
+Clustered index tree latch		To increase concurrency, the tree
+|					latch is usually released when the
+|					leaf page latch has been acquired.
+V
+Clustered index non-leaf
+|
+V
+Clustered index leaf
+|
+V
+Transaction system header
+|
+V
+Transaction undo mutex			The undo log entry must be written
+|					before any index page is modified.
+|					Transaction undo mutex is for the undo
+|					logs the analogue of the tree latch
+|					for a B-tree. If a thread has the
+|					trx undo mutex reserved, it is allowed
+|					to latch the undo log pages in any
+|					order, and also after it has acquired
+|					the fsp latch.
+V
+Rollback segment mutex			The rollback segment mutex must be
+|					reserved, if, e.g., a new page must
+|					be added to an undo log. The rollback
+|					segment and the undo logs in its
+|					history list can be seen as an
+|					analogue of a B-tree, and the latches
+|					reserved similarly, using a version of
+|					lock-coupling. If an undo log must be
+|					extended by a page when inserting an
+|					undo log record, this corresponds to
+|					a pessimistic insert in a B-tree.
+V
+Rollback segment header
+|
+V
+Purge system latch
+|
+V
+Undo log pages				If a thread owns the trx undo mutex,
+|					or for a log in the history list, the
+|					rseg mutex, it is allowed to latch
+|					undo log pages in any order, and even
+|					after it has acquired the fsp latch.
+|					If a thread does not have the
+|					appropriate mutex, it is allowed to
+|					latch only a single undo log page in
+|					a mini-transaction.
+V
+File space management latch		If a mini-transaction must allocate
+|					several file pages, it can do that,
+|					because it keeps the x-latch to the
+|					file space management in its memo.
+V
+File system pages
+|
+V
+Kernel mutex				If a kernel operation needs a file
+|					page allocation, it must reserve the
+|					fsp x-latch before acquiring the kernel
+|					mutex.
+V
+Search system mutex
+|
+V
+Buffer pool mutex
+|
+V
+Log mutex
+|
+Any other latch
+|
+V
+Memory pool mutex */
+
+/* Latching order levels */
+
+/* User transaction locks are higher than any of the latch levels below:
+no latches are allowed when a thread goes to wait for a normal table
+or row lock! */
+#define SYNC_USER_TRX_LOCK	9999
+#define SYNC_NO_ORDER_CHECK	3000	/* this can be used to suppress
+					latching order checking */
+#define	SYNC_LEVEL_VARYING	2000	/* Level is varying. Only used with
+					buffer pool page locks, which do not
+					have a fixed level, but instead have
+					their level set after the page is
+					locked; see e.g.
+					ibuf_bitmap_get_map_page(). */
+#define SYNC_TRX_I_S_RWLOCK	1910	/* Used for
+					trx_i_s_cache_t::rw_lock */
+#define SYNC_TRX_I_S_LAST_READ	1900	/* Used for
+					trx_i_s_cache_t::last_read_mutex */
+#define SYNC_FILE_FORMAT_TAG	1200	/* Used to serialize access to the
+					file format tag */
+#define	SYNC_DICT_OPERATION	1001	/* table create, drop, etc. reserve
+					this in X-mode, implicit or backround
+					operations purge, rollback, foreign
+					key checks reserve this in S-mode */
+#define SYNC_DICT		1000
+#define SYNC_DICT_AUTOINC_MUTEX	999
+#define SYNC_DICT_HEADER	995
+#define SYNC_IBUF_HEADER	914
+#define SYNC_IBUF_PESS_INSERT_MUTEX 912
+#define SYNC_IBUF_MUTEX		910	/* ibuf mutex is really below
+					SYNC_FSP_PAGE: we assign a value this
+					high only to make the program to pass
+					the debug checks */
+/*-------------------------------*/
+#define	SYNC_INDEX_TREE		900
+#define SYNC_TREE_NODE_NEW	892
+#define SYNC_TREE_NODE_FROM_HASH 891
+#define SYNC_TREE_NODE		890
+#define	SYNC_PURGE_SYS		810
+#define	SYNC_PURGE_LATCH	800
+#define	SYNC_TRX_UNDO		700
+#define SYNC_RSEG		600
+#define SYNC_RSEG_HEADER_NEW	591
+#define SYNC_RSEG_HEADER	590
+#define SYNC_TRX_UNDO_PAGE	570
+#define SYNC_EXTERN_STORAGE	500
+#define	SYNC_FSP		400
+#define	SYNC_FSP_PAGE		395
+/*------------------------------------- Insert buffer headers */
+/*------------------------------------- ibuf_mutex */
+/*------------------------------------- Insert buffer tree */
+#define	SYNC_IBUF_BITMAP_MUTEX	351
+#define	SYNC_IBUF_BITMAP	350
+/*------------------------------------- MySQL query cache mutex */
+/*------------------------------------- MySQL binlog mutex */
+/*-------------------------------*/
+#define	SYNC_KERNEL		300
+#define SYNC_REC_LOCK		299
+#define	SYNC_TRX_LOCK_HEAP	298
+#define SYNC_TRX_SYS_HEADER	290
+#define SYNC_LOG		170
+#define SYNC_RECV		168
+#define	SYNC_WORK_QUEUE		162
+#define	SYNC_SEARCH_SYS_CONF	161	/* for assigning btr_search_enabled */
+#define	SYNC_SEARCH_SYS		160	/* NOTE that if we have a memory
+					heap that can be extended to the
+					buffer pool, its logical level is
+					SYNC_SEARCH_SYS, as memory allocation
+					can call routines there! Otherwise
+					the level is SYNC_MEM_HASH. */
+#define	SYNC_BUF_POOL		150
+#define	SYNC_BUF_BLOCK		149
+#define SYNC_DOUBLEWRITE	140
+#define	SYNC_ANY_LATCH		135
+#define SYNC_THR_LOCAL		133
+#define	SYNC_MEM_HASH		131
+#define	SYNC_MEM_POOL		130
+
+/* Codes used to designate lock operations */
+#define RW_LOCK_NOT_LOCKED	350
+#define RW_LOCK_EX		351
+#define RW_LOCK_EXCLUSIVE	351
+#define RW_LOCK_SHARED		352
+#define RW_LOCK_WAIT_EX		353
+#define SYNC_MUTEX		354
+
+/* NOTE! The structure appears here only for the compiler to know its size.
+Do not use its fields directly! The structure used in the spin lock
+implementation of a mutual exclusion semaphore. */
+
+struct mutex_struct {
+	os_event_t	event;	/* Used by sync0arr.c for the wait queue */
+	byte	lock_word;	/* This byte is the target of the atomic
+				test-and-set instruction in Win32 and
+				x86 32/64 with GCC 4.1.0 or later version */
+#if defined(_WIN32) && defined(UNIV_CAN_USE_X86_ASSEMBLER)
+#elif defined(HAVE_GCC_ATOMIC_BUILTINS)
+#else
+	os_fast_mutex_t
+		os_fast_mutex;	/* In other systems we use this OS mutex
+				in place of lock_word */
+#endif
+	ulint	waiters;	/* This ulint is set to 1 if there are (or
+				may be) threads waiting in the global wait
+				array for this mutex to be released.
+				Otherwise, this is 0. */
+	UT_LIST_NODE_T(mutex_t)	list; /* All allocated mutexes are put into
+				a list.	Pointers to the next and prev. */
+#ifdef UNIV_SYNC_DEBUG
+	const char*	file_name;	/* File where the mutex was locked */
+	ulint	line;		/* Line where the mutex was locked */
+	ulint	level;		/* Level in the global latching order */
+#endif /* UNIV_SYNC_DEBUG */
+	const char*	cfile_name;/* File name where mutex created */
+	ulint		cline;	/* Line where created */
+#ifdef UNIV_DEBUG
+	os_thread_id_t thread_id; /* The thread id of the thread
+				which locked the mutex. */
+	ulint		magic_n;
+# define MUTEX_MAGIC_N	(ulint)979585
+#endif /* UNIV_DEBUG */
+#ifndef UNIV_HOTBACKUP
+	ulong		count_os_wait; /* count of os_wait */
+# ifdef UNIV_DEBUG
+	ulong		count_using; /* count of times mutex used */
+	ulong		count_spin_loop; /* count of spin loops */
+	ulong		count_spin_rounds; /* count of spin rounds */
+	ulong		count_os_yield; /* count of os_wait */
+	ulonglong	lspent_time; /* mutex os_wait timer msec */
+	ulonglong	lmax_spent_time; /* mutex os_wait timer msec */
+	const char*	cmutex_name;/* mutex name */
+	ulint		mutex_type;/* 0 - usual mutex 1 - rw_lock mutex	 */
+# endif /* UNIV_DEBUG */
+#endif /* !UNIV_HOTBACKUP */
+};
+
+/* The global array of wait cells for implementation of the databases own
+mutexes and read-write locks. Appears here for debugging purposes only! */
+
+extern sync_array_t*	sync_primary_wait_array;
+
+/* Constant determining how long spin wait is continued before suspending
+the thread. A value 600 rounds on a 1995 100 MHz Pentium seems to correspond
+to 20 microseconds. */
+
+#define	SYNC_SPIN_ROUNDS	srv_n_spin_wait_rounds
+
+/* The number of system calls made in this module. Intended for performance
+monitoring. */
+
+extern	ib_int64_t	mutex_exit_count;
+
+#ifdef UNIV_SYNC_DEBUG
+/* Latching order checks start when this is set TRUE */
+extern ibool	sync_order_checks_on;
+#endif /* UNIV_SYNC_DEBUG */
+
+/* This variable is set to TRUE when sync_init is called */
+extern ibool	sync_initialized;
+
+/* Global list of database mutexes (not OS mutexes) created. */
+typedef UT_LIST_BASE_NODE_T(mutex_t)  ut_list_base_node_t;
+extern ut_list_base_node_t  mutex_list;
+
+/* Mutex protecting the mutex_list variable */
+extern mutex_t mutex_list_mutex;
+
+
+#ifndef UNIV_NONINL
+#include "sync0sync.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/sync0sync.ic b/storage/xtradb/include/sync0sync.ic
new file mode 100644
index 00000000000..c43121ebd0b
--- /dev/null
+++ b/storage/xtradb/include/sync0sync.ic
@@ -0,0 +1,270 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+Copyright (c) 2008, Google Inc.
+
+Portions of this file contain modifications contributed and copyrighted by
+Google, Inc. Those modifications are gratefully acknowledged and are described
+briefly in the InnoDB documentation. The contributions by Google are
+incorporated with their permission, and subject to the conditions contained in
+the file COPYING.Google.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Mutex, the basic synchronization primitive
+
+Created 9/5/1995 Heikki Tuuri
+*******************************************************/
+
+/**********************************************************************
+Sets the waiters field in a mutex. */
+UNIV_INTERN
+void
+mutex_set_waiters(
+/*==============*/
+	mutex_t*	mutex,	/* in: mutex */
+	ulint		n);	/* in: value to set */
+/**********************************************************************
+Reserves a mutex for the current thread. If the mutex is reserved, the
+function spins a preset time (controlled by SYNC_SPIN_ROUNDS) waiting
+for the mutex before suspending the thread. */
+UNIV_INTERN
+void
+mutex_spin_wait(
+/*============*/
+	mutex_t*	mutex,		/* in: pointer to mutex */
+	const char*	file_name,	/* in: file name where mutex
+					requested */
+	ulint		line);		/* in: line where requested */
+#ifdef UNIV_SYNC_DEBUG
+/**********************************************************************
+Sets the debug information for a reserved mutex. */
+UNIV_INTERN
+void
+mutex_set_debug_info(
+/*=================*/
+	mutex_t*	mutex,		/* in: mutex */
+	const char*	file_name,	/* in: file where requested */
+	ulint		line);		/* in: line where requested */
+#endif /* UNIV_SYNC_DEBUG */
+/**********************************************************************
+Releases the threads waiting in the primary wait array for this mutex. */
+UNIV_INTERN
+void
+mutex_signal_object(
+/*================*/
+	mutex_t*	mutex);	/* in: mutex */
+
+/**********************************************************************
+Performs an atomic test-and-set instruction to the lock_word field of a
+mutex. */
+UNIV_INLINE
+byte
+mutex_test_and_set(
+/*===============*/
+				/* out: the previous value of lock_word: 0 or
+				1 */
+	mutex_t*	mutex)	/* in: mutex */
+{
+#if defined(_WIN32) && defined(UNIV_CAN_USE_X86_ASSEMBLER)
+	byte	res;
+	byte*	lw;		/* assembler code is used to ensure that
+				lock_word is loaded from memory */
+	ut_ad(mutex);
+	ut_ad(sizeof(byte) == 1);
+
+	lw = &(mutex->lock_word);
+
+	__asm	MOV	ECX, lw
+		__asm	MOV	EDX, 1
+		__asm	XCHG	DL, BYTE PTR [ECX]
+		__asm	MOV	res, DL
+
+		/* The fence below would prevent this thread from
+		reading the data structure protected by the mutex
+		before the test-and-set operation is committed, but
+		the fence is apparently not needed:
+
+		In a posting to comp.arch newsgroup (August 10, 1997)
+		Andy Glew said that in P6 a LOCKed instruction like
+		XCHG establishes a fence with respect to memory reads
+		and writes and thus an explicit fence is not
+		needed. In P5 he seemed to agree with a previous
+		newsgroup poster that LOCKed instructions serialize
+		all instruction execution, and, consequently, also
+		memory operations. This is confirmed in Intel Software
+		Dev. Manual, Vol. 3. */
+
+		/* mutex_fence(); */
+
+		return(res);
+#elif defined(HAVE_GCC_ATOMIC_BUILTINS)
+	return __sync_lock_test_and_set(&(mutex->lock_word), 1);
+#else
+	ibool	ret;
+
+	ret = os_fast_mutex_trylock(&(mutex->os_fast_mutex));
+
+	if (ret == 0) {
+		/* We check that os_fast_mutex_trylock does not leak
+		and allow race conditions */
+		ut_a(mutex->lock_word == 0);
+
+		mutex->lock_word = 1;
+	}
+
+	return((byte)ret);
+#endif
+}
+
+/**********************************************************************
+Performs a reset instruction to the lock_word field of a mutex. This
+instruction also serializes memory operations to the program order. */
+UNIV_INLINE
+void
+mutex_reset_lock_word(
+/*==================*/
+	mutex_t*	mutex)	/* in: mutex */
+{
+#if defined(_WIN32) && defined(UNIV_CAN_USE_X86_ASSEMBLER)
+	byte*	lw;		/* assembler code is used to ensure that
+				lock_word is loaded from memory */
+	ut_ad(mutex);
+
+	lw = &(mutex->lock_word);
+
+	__asm	MOV	EDX, 0
+		__asm	MOV	ECX, lw
+		__asm	XCHG	DL, BYTE PTR [ECX]
+#elif defined(HAVE_GCC_ATOMIC_BUILTINS)
+	/* In theory __sync_lock_release should be used to release the lock.
+	Unfortunately, it does not work properly alone. The workaround is
+	that more conservative __sync_lock_test_and_set is used instead. */
+	__sync_lock_test_and_set(&(mutex->lock_word), 0);
+#else
+	mutex->lock_word = 0;
+
+	os_fast_mutex_unlock(&(mutex->os_fast_mutex));
+#endif
+}
+
+/**********************************************************************
+Gets the value of the lock word. */
+UNIV_INLINE
+byte
+mutex_get_lock_word(
+/*================*/
+	const mutex_t*	mutex)	/* in: mutex */
+{
+	const volatile byte*	ptr;	/* declared volatile to ensure that
+					lock_word is loaded from memory */
+	ut_ad(mutex);
+
+	ptr = &(mutex->lock_word);
+
+	return(*ptr);
+}
+
+/**********************************************************************
+Gets the waiters field in a mutex. */
+UNIV_INLINE
+ulint
+mutex_get_waiters(
+/*==============*/
+				/* out: value to set */
+	const mutex_t*	mutex)	/* in: mutex */
+{
+	const volatile ulint*	ptr;	/* declared volatile to ensure that
+					the value is read from memory */
+	ut_ad(mutex);
+
+	ptr = &(mutex->waiters);
+
+	return(*ptr);		/* Here we assume that the read of a single
+				word from memory is atomic */
+}
+
+/**********************************************************************
+Unlocks a mutex owned by the current thread. */
+UNIV_INLINE
+void
+mutex_exit(
+/*=======*/
+	mutex_t*	mutex)	/* in: pointer to mutex */
+{
+	ut_ad(mutex_own(mutex));
+
+	ut_d(mutex->thread_id = (os_thread_id_t) ULINT_UNDEFINED);
+
+#ifdef UNIV_SYNC_DEBUG
+	sync_thread_reset_level(mutex);
+#endif
+	mutex_reset_lock_word(mutex);
+
+	/* A problem: we assume that mutex_reset_lock word
+	is a memory barrier, that is when we read the waiters
+	field next, the read must be serialized in memory
+	after the reset. A speculative processor might
+	perform the read first, which could leave a waiting
+	thread hanging indefinitely.
+
+	Our current solution call every second
+	sync_arr_wake_threads_if_sema_free()
+	to wake up possible hanging threads if
+	they are missed in mutex_signal_object. */
+
+	if (mutex_get_waiters(mutex) != 0) {
+
+		mutex_signal_object(mutex);
+	}
+
+#ifdef UNIV_SYNC_PERF_STAT
+	mutex_exit_count++;
+#endif
+}
+
+/**********************************************************************
+Locks a mutex for the current thread. If the mutex is reserved, the function
+spins a preset time (controlled by SYNC_SPIN_ROUNDS), waiting for the mutex
+before suspending the thread. */
+UNIV_INLINE
+void
+mutex_enter_func(
+/*=============*/
+	mutex_t*	mutex,		/* in: pointer to mutex */
+	const char*	file_name,	/* in: file name where locked */
+	ulint		line)		/* in: line where locked */
+{
+	ut_ad(mutex_validate(mutex));
+	ut_ad(!mutex_own(mutex));
+
+	/* Note that we do not peek at the value of lock_word before trying
+	the atomic test_and_set; we could peek, and possibly save time. */
+
+#if defined UNIV_DEBUG && !defined UNIV_HOTBACKUP
+	mutex->count_using++;
+#endif /* UNIV_DEBUG && !UNIV_HOTBACKUP */
+
+	if (!mutex_test_and_set(mutex)) {
+		ut_d(mutex->thread_id = os_thread_get_curr_id());
+#ifdef UNIV_SYNC_DEBUG
+		mutex_set_debug_info(mutex, file_name, line);
+#endif
+		return;	/* Succeeded! */
+	}
+
+	mutex_spin_wait(mutex, file_name, line);
+}
diff --git a/storage/xtradb/include/sync0types.h b/storage/xtradb/include/sync0types.h
new file mode 100644
index 00000000000..3c1021b1a30
--- /dev/null
+++ b/storage/xtradb/include/sync0types.h
@@ -0,0 +1,31 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Global types for sync
+
+Created 9/5/1995 Heikki Tuuri
+*******************************************************/
+
+#ifndef sync0types_h
+#define sync0types_h
+
+#define mutex_t ib_mutex_t
+typedef struct mutex_struct		mutex_t;
+
+#endif
diff --git a/storage/xtradb/include/thr0loc.h b/storage/xtradb/include/thr0loc.h
new file mode 100644
index 00000000000..96ec13cc8e4
--- /dev/null
+++ b/storage/xtradb/include/thr0loc.h
@@ -0,0 +1,94 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The thread local storage
+
+Created 10/5/1995 Heikki Tuuri
+*******************************************************/
+
+/* This module implements storage private to each thread,
+a capability useful in some situations like storing the
+OS handle to the current thread, or its priority. */
+
+#ifndef thr0loc_h
+#define thr0loc_h
+
+#include "univ.i"
+#include "os0thread.h"
+
+/********************************************************************
+Initializes the thread local storage module. */
+UNIV_INTERN
+void
+thr_local_init(void);
+/*================*/
+/***********************************************************************
+Creates a local storage struct for the calling new thread. */
+UNIV_INTERN
+void
+thr_local_create(void);
+/*==================*/
+/***********************************************************************
+Frees the local storage struct for the specified thread. */
+UNIV_INTERN
+void
+thr_local_free(
+/*===========*/
+	os_thread_id_t	id);	/* in: thread id */
+/***********************************************************************
+Gets the slot number in the thread table of a thread. */
+UNIV_INTERN
+ulint
+thr_local_get_slot_no(
+/*==================*/
+				/* out: slot number */
+	os_thread_id_t	id);	/* in: thread id of the thread */
+/***********************************************************************
+Sets in the local storage the slot number in the thread table of a thread. */
+UNIV_INTERN
+void
+thr_local_set_slot_no(
+/*==================*/
+	os_thread_id_t	id,	/* in: thread id of the thread */
+	ulint		slot_no);/* in: slot number */
+/***********************************************************************
+Returns pointer to the 'in_ibuf' field within the current thread local
+storage. */
+UNIV_INTERN
+ibool*
+thr_local_get_in_ibuf_field(void);
+/*=============================*/
+			/* out: pointer to the in_ibuf field */
+
+/*************************************************************************
+Return local hash table informations. */
+
+ulint
+thr_local_hash_cells(void);
+/*=======================*/
+
+ulint
+thr_local_hash_nodes(void);
+/*=======================*/
+
+#ifndef UNIV_NONINL
+#include "thr0loc.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/thr0loc.ic b/storage/xtradb/include/thr0loc.ic
new file mode 100644
index 00000000000..6de183fd857
--- /dev/null
+++ b/storage/xtradb/include/thr0loc.ic
@@ -0,0 +1,23 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Thread local storage
+
+Created 10/4/1995 Heikki Tuuri
+*******************************************************/
diff --git a/storage/xtradb/include/trx0i_s.h b/storage/xtradb/include/trx0i_s.h
new file mode 100644
index 00000000000..cf2865af127
--- /dev/null
+++ b/storage/xtradb/include/trx0i_s.h
@@ -0,0 +1,212 @@
+/*****************************************************************************
+
+Copyright (c) 2007, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+INFORMATION SCHEMA innodb_trx, innodb_locks and
+innodb_lock_waits tables cache structures and public
+functions.
+
+Created July 17, 2007 Vasil Dimov
+*******************************************************/
+
+#ifndef trx0i_s_h
+#define trx0i_s_h
+
+#include "univ.i"
+#include "trx0types.h"
+#include "ut0ut.h"
+
+/* the maximum amount of memory that can be consumed by innodb_trx,
+innodb_locks and innodb_lock_waits information schema tables. */
+#define TRX_I_S_MEM_LIMIT		16777216 /* 16 MiB */
+
+/* the maximum length of a string that can be stored in
+i_s_locks_row_t::lock_data */
+#define TRX_I_S_LOCK_DATA_MAX_LEN	8192
+
+/* the maximum length of a string that can be stored in
+i_s_trx_row_t::trx_query */
+#define TRX_I_S_TRX_QUERY_MAX_LEN	1024
+
+typedef struct i_s_locks_row_struct	i_s_locks_row_t;
+typedef struct i_s_hash_chain_struct	i_s_hash_chain_t;
+
+/* Objects of this type are added to the hash table
+trx_i_s_cache_t::locks_hash */
+struct i_s_hash_chain_struct {
+	i_s_locks_row_t*	value;
+	i_s_hash_chain_t*	next;
+};
+
+/* This structure represents INFORMATION_SCHEMA.innodb_locks row */
+struct i_s_locks_row_struct {
+	ullint		lock_trx_id;
+	const char*	lock_mode;
+	const char*	lock_type;
+	const char*	lock_table;
+	const char*	lock_index;
+	ulint		lock_space;
+	ulint		lock_page;
+	ulint		lock_rec;
+	const char*	lock_data;
+
+	/* The following are auxiliary and not included in the table */
+	ullint		lock_table_id;
+	i_s_hash_chain_t hash_chain; /* this object is added to the hash
+				    table
+				    trx_i_s_cache_t::locks_hash */
+};
+
+/* This structure represents INFORMATION_SCHEMA.innodb_trx row */
+typedef struct i_s_trx_row_struct {
+	ullint			trx_id;
+	const char*		trx_state;
+	ib_time_t		trx_started;
+	const i_s_locks_row_t*	requested_lock_row;
+	ib_time_t		trx_wait_started;
+	ullint			trx_weight;
+	ulint			trx_mysql_thread_id;
+	const char*		trx_query;
+} i_s_trx_row_t;
+
+/* This structure represents INFORMATION_SCHEMA.innodb_lock_waits row */
+typedef struct i_s_lock_waits_row_struct {
+	const i_s_locks_row_t*	requested_lock_row;
+	const i_s_locks_row_t*	blocking_lock_row;
+} i_s_lock_waits_row_t;
+
+/* This type is opaque and is defined in trx/trx0i_s.c */
+typedef struct trx_i_s_cache_struct	trx_i_s_cache_t;
+
+/* Auxiliary enum used by functions that need to select one of the
+INFORMATION_SCHEMA tables */
+enum i_s_table {
+	I_S_INNODB_TRX,
+	I_S_INNODB_LOCKS,
+	I_S_INNODB_LOCK_WAITS
+};
+
+/* This is the intermediate buffer where data needed to fill the
+INFORMATION SCHEMA tables is fetched and later retrieved by the C++
+code in handler/i_s.cc. */
+extern trx_i_s_cache_t*	trx_i_s_cache;
+
+/***********************************************************************
+Initialize INFORMATION SCHEMA trx related cache. */
+UNIV_INTERN
+void
+trx_i_s_cache_init(
+/*===============*/
+	trx_i_s_cache_t*	cache);	/* out: cache to init */
+
+/***********************************************************************
+Issue a shared/read lock on the tables cache. */
+UNIV_INTERN
+void
+trx_i_s_cache_start_read(
+/*=====================*/
+	trx_i_s_cache_t*	cache);	/* in: cache */
+
+/***********************************************************************
+Release a shared/read lock on the tables cache. */
+UNIV_INTERN
+void
+trx_i_s_cache_end_read(
+/*===================*/
+	trx_i_s_cache_t*	cache);	/* in: cache */
+
+/***********************************************************************
+Issue an exclusive/write lock on the tables cache. */
+UNIV_INTERN
+void
+trx_i_s_cache_start_write(
+/*======================*/
+	trx_i_s_cache_t*	cache);	/* in: cache */
+
+/***********************************************************************
+Release an exclusive/write lock on the tables cache. */
+UNIV_INTERN
+void
+trx_i_s_cache_end_write(
+/*====================*/
+	trx_i_s_cache_t*	cache);	/* in: cache */
+
+
+/***********************************************************************
+Retrieves the number of used rows in the cache for a given
+INFORMATION SCHEMA table. */
+UNIV_INTERN
+ulint
+trx_i_s_cache_get_rows_used(
+/*========================*/
+					/* out: number of rows */
+	trx_i_s_cache_t*	cache,	/* in: cache */
+	enum i_s_table		table);	/* in: which table */
+
+/***********************************************************************
+Retrieves the nth row in the cache for a given INFORMATION SCHEMA
+table. */
+UNIV_INTERN
+void*
+trx_i_s_cache_get_nth_row(
+/*======================*/
+					/* out: row */
+	trx_i_s_cache_t*	cache,	/* in: cache */
+	enum i_s_table		table,	/* in: which table */
+	ulint			n);	/* in: row number */
+
+/***********************************************************************
+Update the transactions cache if it has not been read for some time. */
+UNIV_INTERN
+int
+trx_i_s_possibly_fetch_data_into_cache(
+/*===================================*/
+					/* out: 0 - fetched, 1 - not */
+	trx_i_s_cache_t*	cache);	/* in/out: cache */
+
+/***********************************************************************
+Returns TRUE if the data in the cache is truncated due to the memory
+limit posed by TRX_I_S_MEM_LIMIT. */
+UNIV_INTERN
+ibool
+trx_i_s_cache_is_truncated(
+/*=======================*/
+					/* out: TRUE if truncated */
+	trx_i_s_cache_t*	cache);	/* in: cache */
+
+/* The maximum length of a resulting lock_id_size in
+trx_i_s_create_lock_id(), not including the terminating '\0'.
+":%lu:%lu:%lu" -> 63 chars */
+#define TRX_I_S_LOCK_ID_MAX_LEN	(TRX_ID_MAX_LEN + 63)
+
+/***********************************************************************
+Crafts a lock id string from a i_s_locks_row_t object. Returns its
+second argument. This function aborts if there is not enough space in
+lock_id. Be sure to provide at least TRX_I_S_LOCK_ID_MAX_LEN + 1 if you
+want to be 100% sure that it will not abort. */
+UNIV_INTERN
+char*
+trx_i_s_create_lock_id(
+/*===================*/
+					/* out: resulting lock id */
+	const i_s_locks_row_t*	row,	/* in: innodb_locks row */
+	char*			lock_id,/* out: resulting lock_id */
+	ulint			lock_id_size);/* in: size of the lock id
+					buffer */
+
+#endif /* trx0i_s_h */
diff --git a/storage/xtradb/include/trx0purge.h b/storage/xtradb/include/trx0purge.h
new file mode 100644
index 00000000000..4921b860485
--- /dev/null
+++ b/storage/xtradb/include/trx0purge.h
@@ -0,0 +1,185 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Purge old versions
+
+Created 3/26/1996 Heikki Tuuri
+*******************************************************/
+
+#ifndef trx0purge_h
+#define trx0purge_h
+
+#include "univ.i"
+#include "trx0types.h"
+#include "mtr0mtr.h"
+#include "trx0sys.h"
+#include "que0types.h"
+#include "page0page.h"
+#include "usr0sess.h"
+#include "fil0fil.h"
+
+/* The global data structure coordinating a purge */
+extern trx_purge_t*	purge_sys;
+
+/* A dummy undo record used as a return value when we have a whole undo log
+which needs no purge */
+extern trx_undo_rec_t	trx_purge_dummy_rec;
+
+/************************************************************************
+Calculates the file address of an undo log header when we have the file
+address of its history list node. */
+UNIV_INLINE
+fil_addr_t
+trx_purge_get_log_from_hist(
+/*========================*/
+					/* out: file address of the log */
+	fil_addr_t	node_addr);	/* in: file address of the history
+					list node of the log */
+/*********************************************************************
+Checks if trx_id is >= purge_view: then it is guaranteed that its update
+undo log still exists in the system. */
+UNIV_INTERN
+ibool
+trx_purge_update_undo_must_exist(
+/*=============================*/
+			/* out: TRUE if is sure that it is preserved, also
+			if the function returns FALSE, it is possible that
+			the undo log still exists in the system */
+	dulint	trx_id);/* in: transaction id */
+/************************************************************************
+Creates the global purge system control structure and inits the history
+mutex. */
+UNIV_INTERN
+void
+trx_purge_sys_create(void);
+/*======================*/
+/************************************************************************
+Adds the update undo log as the first log in the history list. Removes the
+update undo log segment from the rseg slot if it is too big for reuse. */
+UNIV_INTERN
+void
+trx_purge_add_update_undo_to_history(
+/*=================================*/
+	trx_t*	trx,		/* in: transaction */
+	page_t*	undo_page,	/* in: update undo log header page,
+				x-latched */
+	mtr_t*	mtr);		/* in: mtr */
+/************************************************************************
+Fetches the next undo log record from the history list to purge. It must be
+released with the corresponding release function. */
+UNIV_INTERN
+trx_undo_rec_t*
+trx_purge_fetch_next_rec(
+/*=====================*/
+				/* out: copy of an undo log record, or
+				pointer to the dummy undo log record
+				&trx_purge_dummy_rec if the whole undo log
+				can skipped in purge; NULL if none left */
+	dulint*		roll_ptr,/* out: roll pointer to undo record */
+	trx_undo_inf_t** cell,	/* out: storage cell for the record in the
+				purge array */
+	mem_heap_t*	heap);	/* in: memory heap where copied */
+/***********************************************************************
+Releases a reserved purge undo record. */
+UNIV_INTERN
+void
+trx_purge_rec_release(
+/*==================*/
+	trx_undo_inf_t*	cell);	/* in: storage cell */
+/***********************************************************************
+This function runs a purge batch. */
+UNIV_INTERN
+ulint
+trx_purge(void);
+/*===========*/
+				/* out: number of undo log pages handled in
+				the batch */
+/**********************************************************************
+Prints information of the purge system to stderr. */
+UNIV_INTERN
+void
+trx_purge_sys_print(void);
+/*======================*/
+
+/* The control structure used in the purge operation */
+struct trx_purge_struct{
+	ulint		state;		/* Purge system state */
+	sess_t*		sess;		/* System session running the purge
+					query */
+	trx_t*		trx;		/* System transaction running the purge
+					query: this trx is not in the trx list
+					of the trx system and it never ends */
+	que_t*		query;		/* The query graph which will do the
+					parallelized purge operation */
+	rw_lock_t	latch;		/* The latch protecting the purge view.
+					A purge operation must acquire an
+					x-latch here for the instant at which
+					it changes the purge view: an undo
+					log operation can prevent this by
+					obtaining an s-latch here. */
+	read_view_t*	view;		/* The purge will not remove undo logs
+					which are >= this view (purge view) */
+	mutex_t		mutex;		/* Mutex protecting the fields below */
+	ulint		n_pages_handled;/* Approximate number of undo log
+					pages processed in purge */
+	ulint		handle_limit;	/* Target of how many pages to get
+					processed in the current purge */
+	/*------------------------------*/
+	/* The following two fields form the 'purge pointer' which advances
+	during a purge, and which is used in history list truncation */
+
+	dulint		purge_trx_no;	/* Purge has advanced past all
+					transactions whose number is less
+					than this */
+	dulint		purge_undo_no;	/* Purge has advanced past all records
+					whose undo number is less than this */
+	/*-----------------------------*/
+	ibool		next_stored;	/* TRUE if the info of the next record
+					to purge is stored below: if yes, then
+					the transaction number and the undo
+					number of the record are stored in
+					purge_trx_no and purge_undo_no above */
+	trx_rseg_t*	rseg;		/* Rollback segment for the next undo
+					record to purge */
+	ulint		page_no;	/* Page number for the next undo
+					record to purge, page number of the
+					log header, if dummy record */
+	ulint		offset;		/* Page offset for the next undo
+					record to purge, 0 if the dummy
+					record */
+	ulint		hdr_page_no;	/* Header page of the undo log where
+					the next record to purge belongs */
+	ulint		hdr_offset;	/* Header byte offset on the page */
+	/*-----------------------------*/
+	trx_undo_arr_t*	arr;		/* Array of transaction numbers and
+					undo numbers of the undo records
+					currently under processing in purge */
+	mem_heap_t*	heap;		/* Temporary storage used during a
+					purge: can be emptied after purge
+					completes */
+};
+
+#define TRX_PURGE_ON		1	/* purge operation is running */
+#define TRX_STOP_PURGE		2	/* purge operation is stopped, or
+					it should be stopped */
+#ifndef UNIV_NONINL
+#include "trx0purge.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/trx0purge.ic b/storage/xtradb/include/trx0purge.ic
new file mode 100644
index 00000000000..2c1d2ac75af
--- /dev/null
+++ b/storage/xtradb/include/trx0purge.ic
@@ -0,0 +1,42 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Purge old versions
+
+Created 3/26/1996 Heikki Tuuri
+*******************************************************/
+
+#include "trx0undo.h"
+
+/************************************************************************
+Calculates the file address of an undo log header when we have the file
+address of its history list node. */
+UNIV_INLINE
+fil_addr_t
+trx_purge_get_log_from_hist(
+/*========================*/
+					/* out: file address of the log */
+	fil_addr_t	node_addr)	/* in: file address of the history
+					list node of the log */
+{
+	node_addr.boffset -= TRX_UNDO_HISTORY_NODE;
+
+	return(node_addr);
+}
+
diff --git a/storage/xtradb/include/trx0rec.h b/storage/xtradb/include/trx0rec.h
new file mode 100644
index 00000000000..444d39e39db
--- /dev/null
+++ b/storage/xtradb/include/trx0rec.h
@@ -0,0 +1,333 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Transaction undo log record
+
+Created 3/26/1996 Heikki Tuuri
+*******************************************************/
+
+#ifndef trx0rec_h
+#define trx0rec_h
+
+#include "univ.i"
+#include "trx0types.h"
+#include "row0types.h"
+#include "mtr0mtr.h"
+#include "dict0types.h"
+#include "que0types.h"
+#include "data0data.h"
+#include "rem0types.h"
+
+/***************************************************************************
+Copies the undo record to the heap. */
+UNIV_INLINE
+trx_undo_rec_t*
+trx_undo_rec_copy(
+/*==============*/
+					/* out, own: copy of undo log record */
+	trx_undo_rec_t*	undo_rec,	/* in: undo log record */
+	mem_heap_t*	heap);		/* in: heap where copied */
+/**************************************************************************
+Reads the undo log record type. */
+UNIV_INLINE
+ulint
+trx_undo_rec_get_type(
+/*==================*/
+					/* out: record type */
+	trx_undo_rec_t*	undo_rec);	/* in: undo log record */
+/**************************************************************************
+Reads from an undo log record the record compiler info. */
+UNIV_INLINE
+ulint
+trx_undo_rec_get_cmpl_info(
+/*=======================*/
+					/* out: compiler info */
+	trx_undo_rec_t*	undo_rec);	/* in: undo log record */
+/**************************************************************************
+Returns TRUE if an undo log record contains an extern storage field. */
+UNIV_INLINE
+ibool
+trx_undo_rec_get_extern_storage(
+/*============================*/
+					/* out: TRUE if extern */
+	trx_undo_rec_t*	undo_rec);	/* in: undo log record */
+/**************************************************************************
+Reads the undo log record number. */
+UNIV_INLINE
+dulint
+trx_undo_rec_get_undo_no(
+/*=====================*/
+					/* out: undo no */
+	trx_undo_rec_t*	undo_rec);	/* in: undo log record */
+/**************************************************************************
+ * Returns the start of the undo record data area. */
+
+UNIV_INLINE
+byte*
+trx_undo_rec_get_ptr(
+/*==================*/
+					/* out: compiler info */
+	trx_undo_rec_t* undo_rec,       /* in: undo log record */
+	dulint          undo_no);        /* in: undo no read from node */
+
+/**************************************************************************
+Reads from an undo log record the general parameters. */
+UNIV_INTERN
+byte*
+trx_undo_rec_get_pars(
+/*==================*/
+					/* out: remaining part of undo log
+					record after reading these values */
+	trx_undo_rec_t*	undo_rec,	/* in: undo log record */
+	ulint*		type,		/* out: undo record type:
+					TRX_UNDO_INSERT_REC, ... */
+	ulint*		cmpl_info,	/* out: compiler info, relevant only
+					for update type records */
+	ibool*		updated_extern,	/* out: TRUE if we updated an
+					externally stored fild */
+	dulint*		undo_no,	/* out: undo log record number */
+	dulint*		table_id);	/* out: table id */
+/***********************************************************************
+Builds a row reference from an undo log record. */
+UNIV_INTERN
+byte*
+trx_undo_rec_get_row_ref(
+/*=====================*/
+				/* out: pointer to remaining part of undo
+				record */
+	byte*		ptr,	/* in: remaining part of a copy of an undo log
+				record, at the start of the row reference;
+				NOTE that this copy of the undo log record must
+				be preserved as long as the row reference is
+				used, as we do NOT copy the data in the
+				record! */
+	dict_index_t*	index,	/* in: clustered index */
+	dtuple_t**	ref,	/* out, own: row reference */
+	mem_heap_t*	heap);	/* in: memory heap from which the memory
+				needed is allocated */
+/***********************************************************************
+Skips a row reference from an undo log record. */
+UNIV_INTERN
+byte*
+trx_undo_rec_skip_row_ref(
+/*======================*/
+				/* out: pointer to remaining part of undo
+				record */
+	byte*		ptr,	/* in: remaining part in update undo log
+				record, at the start of the row reference */
+	dict_index_t*	index);	/* in: clustered index */
+/**************************************************************************
+Reads from an undo log update record the system field values of the old
+version. */
+UNIV_INTERN
+byte*
+trx_undo_update_rec_get_sys_cols(
+/*=============================*/
+				/* out: remaining part of undo log
+				record after reading these values */
+	byte*	ptr,		/* in: remaining part of undo log
+				record after reading general
+				parameters */
+	dulint*	trx_id,		/* out: trx id */
+	dulint*	roll_ptr,	/* out: roll ptr */
+	ulint*	info_bits);	/* out: info bits state */
+/***********************************************************************
+Builds an update vector based on a remaining part of an undo log record. */
+UNIV_INTERN
+byte*
+trx_undo_update_rec_get_update(
+/*===========================*/
+				/* out: remaining part of the record,
+				NULL if an error detected, which means that
+				the record is corrupted */
+	byte*		ptr,	/* in: remaining part in update undo log
+				record, after reading the row reference
+				NOTE that this copy of the undo log record must
+				be preserved as long as the update vector is
+				used, as we do NOT copy the data in the
+				record! */
+	dict_index_t*	index,	/* in: clustered index */
+	ulint		type,	/* in: TRX_UNDO_UPD_EXIST_REC,
+				TRX_UNDO_UPD_DEL_REC, or
+				TRX_UNDO_DEL_MARK_REC; in the last case,
+				only trx id and roll ptr fields are added to
+				the update vector */
+	dulint		trx_id,	/* in: transaction id from this undorecord */
+	dulint		roll_ptr,/* in: roll pointer from this undo record */
+	ulint		info_bits,/* in: info bits from this undo record */
+	trx_t*		trx,	/* in: transaction */
+	mem_heap_t*	heap,	/* in: memory heap from which the memory
+				needed is allocated */
+	upd_t**		upd);	/* out, own: update vector */
+/***********************************************************************
+Builds a partial row from an update undo log record. It contains the
+columns which occur as ordering in any index of the table. */
+UNIV_INTERN
+byte*
+trx_undo_rec_get_partial_row(
+/*=========================*/
+				/* out: pointer to remaining part of undo
+				record */
+	byte*		ptr,	/* in: remaining part in update undo log
+				record of a suitable type, at the start of
+				the stored index columns;
+				NOTE that this copy of the undo log record must
+				be preserved as long as the partial row is
+				used, as we do NOT copy the data in the
+				record! */
+	dict_index_t*	index,	/* in: clustered index */
+	dtuple_t**	row,	/* out, own: partial row */
+	ibool		ignore_prefix, /* in: flag to indicate if we
+				expect blob prefixes in undo. Used
+				only in the assertion. */
+	mem_heap_t*	heap);	/* in: memory heap from which the memory
+				needed is allocated */
+/***************************************************************************
+Writes information to an undo log about an insert, update, or a delete marking
+of a clustered index record. This information is used in a rollback of the
+transaction and in consistent reads that must look to the history of this
+transaction. */
+UNIV_INTERN
+ulint
+trx_undo_report_row_operation(
+/*==========================*/
+					/* out: DB_SUCCESS or error code */
+	ulint		flags,		/* in: if BTR_NO_UNDO_LOG_FLAG bit is
+					set, does nothing */
+	ulint		op_type,	/* in: TRX_UNDO_INSERT_OP or
+					TRX_UNDO_MODIFY_OP */
+	que_thr_t*	thr,		/* in: query thread */
+	dict_index_t*	index,		/* in: clustered index */
+	const dtuple_t*	clust_entry,	/* in: in the case of an insert,
+					index entry to insert into the
+					clustered index, otherwise NULL */
+	const upd_t*	update,		/* in: in the case of an update,
+					the update vector, otherwise NULL */
+	ulint		cmpl_info,	/* in: compiler info on secondary
+					index updates */
+	const rec_t*	rec,		/* in: case of an update or delete
+					marking, the record in the clustered
+					index, otherwise NULL */
+	dulint*		roll_ptr);	/* out: rollback pointer to the
+					inserted undo log record,
+					ut_dulint_zero if BTR_NO_UNDO_LOG
+					flag was specified */
+/**********************************************************************
+Copies an undo record to heap. This function can be called if we know that
+the undo log record exists. */
+UNIV_INTERN
+trx_undo_rec_t*
+trx_undo_get_undo_rec_low(
+/*======================*/
+					/* out, own: copy of the record */
+	dulint		roll_ptr,	/* in: roll pointer to record */
+	mem_heap_t*	heap);		/* in: memory heap where copied */
+/**********************************************************************
+Copies an undo record to heap. */
+UNIV_INTERN
+ulint
+trx_undo_get_undo_rec(
+/*==================*/
+					/* out: DB_SUCCESS, or
+					DB_MISSING_HISTORY if the undo log
+					has been truncated and we cannot
+					fetch the old version; NOTE: the
+					caller must have latches on the
+					clustered index page and purge_view */
+	dulint		roll_ptr,	/* in: roll pointer to record */
+	dulint		trx_id,		/* in: id of the trx that generated
+					the roll pointer: it points to an
+					undo log of this transaction */
+	trx_undo_rec_t** undo_rec,	/* out, own: copy of the record */
+	mem_heap_t*	heap);		/* in: memory heap where copied */
+/***********************************************************************
+Build a previous version of a clustered index record. This function checks
+that the caller has a latch on the index page of the clustered index record
+and an s-latch on the purge_view. This guarantees that the stack of versions
+is locked. */
+UNIV_INTERN
+ulint
+trx_undo_prev_version_build(
+/*========================*/
+				/* out: DB_SUCCESS, or DB_MISSING_HISTORY if
+				the previous version is not >= purge_view,
+				which means that it may have been removed,
+				DB_ERROR if corrupted record */
+	const rec_t*	index_rec,/* in: clustered index record in the
+				index tree */
+	mtr_t*		index_mtr,/* in: mtr which contains the latch to
+				index_rec page and purge_view */
+	const rec_t*	rec,	/* in: version of a clustered index record */
+	dict_index_t*	index,	/* in: clustered index */
+	ulint*		offsets,/* in: rec_get_offsets(rec, index) */
+	mem_heap_t*	heap,	/* in: memory heap from which the memory
+				needed is allocated */
+	rec_t**		old_vers);/* out, own: previous version, or NULL if
+				rec is the first inserted version, or if
+				history data has been deleted */
+/***************************************************************
+Parses a redo log record of adding an undo log record. */
+UNIV_INTERN
+byte*
+trx_undo_parse_add_undo_rec(
+/*========================*/
+			/* out: end of log record or NULL */
+	byte*	ptr,	/* in: buffer */
+	byte*	end_ptr,/* in: buffer end */
+	page_t*	page);	/* in: page or NULL */
+/***************************************************************
+Parses a redo log record of erasing of an undo page end. */
+UNIV_INTERN
+byte*
+trx_undo_parse_erase_page_end(
+/*==========================*/
+			/* out: end of log record or NULL */
+	byte*	ptr,	/* in: buffer */
+	byte*	end_ptr,/* in: buffer end */
+	page_t*	page,	/* in: page or NULL */
+	mtr_t*	mtr);	/* in: mtr or NULL */
+
+/* Types of an undo log record: these have to be smaller than 16, as the
+compilation info multiplied by 16 is ORed to this value in an undo log
+record */
+
+#define	TRX_UNDO_INSERT_REC	11	/* fresh insert into clustered index */
+#define	TRX_UNDO_UPD_EXIST_REC	12	/* update of a non-delete-marked
+					record */
+#define	TRX_UNDO_UPD_DEL_REC	13	/* update of a delete marked record to
+					a not delete marked record; also the
+					fields of the record can change */
+#define	TRX_UNDO_DEL_MARK_REC	14	/* delete marking of a record; fields
+					do not change */
+#define	TRX_UNDO_CMPL_INFO_MULT	16	/* compilation info is multiplied by
+					this and ORed to the type above */
+#define	TRX_UNDO_UPD_EXTERN	128	/* This bit can be ORed to type_cmpl
+					to denote that we updated external
+					storage fields: used by purge to
+					free the external storage */
+
+/* Operation type flags used in trx_undo_report_row_operation */
+#define	TRX_UNDO_INSERT_OP		1
+#define	TRX_UNDO_MODIFY_OP		2
+
+#ifndef UNIV_NONINL
+#include "trx0rec.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/trx0rec.ic b/storage/xtradb/include/trx0rec.ic
new file mode 100644
index 00000000000..bfd74eb9dfb
--- /dev/null
+++ b/storage/xtradb/include/trx0rec.ic
@@ -0,0 +1,116 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Transaction undo log record
+
+Created 3/26/1996 Heikki Tuuri
+*******************************************************/
+
+/**************************************************************************
+Reads from an undo log record the record type. */
+UNIV_INLINE
+ulint
+trx_undo_rec_get_type(
+/*==================*/
+					/* out: record type */
+	trx_undo_rec_t*	undo_rec)	/* in: undo log record */
+{
+	return(mach_read_from_1(undo_rec + 2) & (TRX_UNDO_CMPL_INFO_MULT - 1));
+}
+
+/**************************************************************************
+Reads from an undo log record the record compiler info. */
+UNIV_INLINE
+ulint
+trx_undo_rec_get_cmpl_info(
+/*=======================*/
+					/* out: compiler info */
+	trx_undo_rec_t*	undo_rec)	/* in: undo log record */
+{
+	return(mach_read_from_1(undo_rec + 2) / TRX_UNDO_CMPL_INFO_MULT);
+}
+
+/**************************************************************************
+Returns TRUE if an undo log record contains an extern storage field. */
+UNIV_INLINE
+ibool
+trx_undo_rec_get_extern_storage(
+/*============================*/
+					/* out: TRUE if extern */
+	trx_undo_rec_t*	undo_rec)	/* in: undo log record */
+{
+	if (mach_read_from_1(undo_rec + 2) & TRX_UNDO_UPD_EXTERN) {
+
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
+
+/**************************************************************************
+Reads the undo log record number. */
+UNIV_INLINE
+dulint
+trx_undo_rec_get_undo_no(
+/*=====================*/
+					/* out: undo no */
+	trx_undo_rec_t*	undo_rec)	/* in: undo log record */
+{
+	byte*	ptr;
+
+	ptr = undo_rec + 3;
+
+	return(mach_dulint_read_much_compressed(ptr));
+}
+
+/**************************************************************************
+Returns the start of the undo record data area. */
+UNIV_INLINE
+byte*
+trx_undo_rec_get_ptr(
+/*=================*/
+					/* out: compiler info */
+	trx_undo_rec_t*	undo_rec,	/* in: undo log record */
+	dulint		undo_no)	/* in: undo no read from node */
+{
+	return (((byte*) undo_rec) + 3
+		+ mach_dulint_get_much_compressed_size(undo_no));
+}
+
+/***************************************************************************
+Copies the undo record to the heap. */
+UNIV_INLINE
+trx_undo_rec_t*
+trx_undo_rec_copy(
+/*==============*/
+					/* out, own: copy of undo log record */
+	trx_undo_rec_t*	undo_rec,	/* in: undo log record */
+	mem_heap_t*	heap)		/* in: heap where copied */
+{
+	ulint		len;
+	trx_undo_rec_t*	rec_copy;
+
+	len = mach_read_from_2(undo_rec)
+		- ut_align_offset(undo_rec, UNIV_PAGE_SIZE);
+	rec_copy = mem_heap_alloc(heap, len);
+
+	ut_memcpy(rec_copy, undo_rec, len);
+
+	return(rec_copy);
+}
diff --git a/storage/xtradb/include/trx0roll.h b/storage/xtradb/include/trx0roll.h
new file mode 100644
index 00000000000..3318a5985d7
--- /dev/null
+++ b/storage/xtradb/include/trx0roll.h
@@ -0,0 +1,340 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Transaction rollback
+
+Created 3/26/1996 Heikki Tuuri
+*******************************************************/
+
+#ifndef trx0roll_h
+#define trx0roll_h
+
+#include "univ.i"
+#include "trx0trx.h"
+#include "trx0types.h"
+#include "mtr0mtr.h"
+#include "trx0sys.h"
+
+#define trx_roll_free_all_savepoints(s) trx_roll_savepoints_free((s), NULL)
+
+/***********************************************************************
+Determines if this transaction is rolling back an incomplete transaction
+in crash recovery. */
+UNIV_INTERN
+ibool
+trx_is_recv(
+/*========*/
+				/* out: TRUE if trx is an incomplete
+				transaction that is being rolled back
+				in crash recovery */
+	const trx_t*	trx);	/* in: transaction */
+/***********************************************************************
+Returns a transaction savepoint taken at this point in time. */
+UNIV_INTERN
+trx_savept_t
+trx_savept_take(
+/*============*/
+			/* out: savepoint */
+	trx_t*	trx);	/* in: transaction */
+/***********************************************************************
+Creates an undo number array. */
+UNIV_INTERN
+trx_undo_arr_t*
+trx_undo_arr_create(void);
+/*=====================*/
+/***********************************************************************
+Frees an undo number array. */
+UNIV_INTERN
+void
+trx_undo_arr_free(
+/*==============*/
+	trx_undo_arr_t*	arr);	/* in: undo number array */
+/***********************************************************************
+Returns pointer to nth element in an undo number array. */
+UNIV_INLINE
+trx_undo_inf_t*
+trx_undo_arr_get_nth_info(
+/*======================*/
+				/* out: pointer to the nth element */
+	trx_undo_arr_t*	arr,	/* in: undo number array */
+	ulint		n);	/* in: position */
+/***************************************************************************
+Tries truncate the undo logs. */
+UNIV_INTERN
+void
+trx_roll_try_truncate(
+/*==================*/
+	trx_t*	trx);	/* in: transaction */
+/************************************************************************
+Pops the topmost record when the two undo logs of a transaction are seen
+as a single stack of records ordered by their undo numbers. Inserts the
+undo number of the popped undo record to the array of currently processed
+undo numbers in the transaction. When the query thread finishes processing
+of this undo record, it must be released with trx_undo_rec_release. */
+UNIV_INTERN
+trx_undo_rec_t*
+trx_roll_pop_top_rec_of_trx(
+/*========================*/
+				/* out: undo log record copied to heap, NULL
+				if none left, or if the undo number of the
+				top record would be less than the limit */
+	trx_t*		trx,	/* in: transaction */
+	dulint		limit,	/* in: least undo number we need */
+	dulint*		roll_ptr,/* out: roll pointer to undo record */
+	mem_heap_t*	heap);	/* in: memory heap where copied */
+/************************************************************************
+Reserves an undo log record for a query thread to undo. This should be
+called if the query thread gets the undo log record not using the pop
+function above. */
+UNIV_INTERN
+ibool
+trx_undo_rec_reserve(
+/*=================*/
+			/* out: TRUE if succeeded */
+	trx_t*	trx,	/* in: transaction */
+	dulint	undo_no);/* in: undo number of the record */
+/***********************************************************************
+Releases a reserved undo record. */
+UNIV_INTERN
+void
+trx_undo_rec_release(
+/*=================*/
+	trx_t*	trx,	/* in: transaction */
+	dulint	undo_no);/* in: undo number */
+/*************************************************************************
+Starts a rollback operation. */
+UNIV_INTERN
+void
+trx_rollback(
+/*=========*/
+	trx_t*		trx,	/* in: transaction */
+	trx_sig_t*	sig,	/* in: signal starting the rollback */
+	que_thr_t**	next_thr);/* in/out: next query thread to run;
+				if the value which is passed in is
+				a pointer to a NULL pointer, then the
+				calling function can start running
+				a new query thread */
+/***********************************************************************
+Rollback or clean up any incomplete transactions which were
+encountered in crash recovery.  If the transaction already was
+committed, then we clean up a possible insert undo log. If the
+transaction was not yet committed, then we roll it back.
+Note: this is done in a background thread. */
+UNIV_INTERN
+os_thread_ret_t
+trx_rollback_or_clean_all_recovered(
+/*================================*/
+			/* out: a dummy parameter */
+	void*	arg __attribute__((unused)));
+			/* in: a dummy parameter required by
+			os_thread_create */
+/********************************************************************
+Finishes a transaction rollback. */
+UNIV_INTERN
+void
+trx_finish_rollback_off_kernel(
+/*===========================*/
+	que_t*		graph,	/* in: undo graph which can now be freed */
+	trx_t*		trx,	/* in: transaction */
+	que_thr_t**	next_thr);/* in/out: next query thread to run;
+				if the value which is passed in is
+				a pointer to a NULL pointer, then the
+				calling function can start running
+				a new query thread; if this parameter is
+				NULL, it is ignored */
+/********************************************************************
+Builds an undo 'query' graph for a transaction. The actual rollback is
+performed by executing this query graph like a query subprocedure call.
+The reply about the completion of the rollback will be sent by this
+graph. */
+UNIV_INTERN
+que_t*
+trx_roll_graph_build(
+/*=================*/
+			/* out, own: the query graph */
+	trx_t*	trx);	/* in: trx handle */
+/*************************************************************************
+Creates a rollback command node struct. */
+UNIV_INTERN
+roll_node_t*
+roll_node_create(
+/*=============*/
+				/* out, own: rollback node struct */
+	mem_heap_t*	heap);	/* in: mem heap where created */
+/***************************************************************
+Performs an execution step for a rollback command node in a query graph. */
+UNIV_INTERN
+que_thr_t*
+trx_rollback_step(
+/*==============*/
+				/* out: query thread to run next, or NULL */
+	que_thr_t*	thr);	/* in: query thread */
+/***********************************************************************
+Rollback a transaction used in MySQL. */
+UNIV_INTERN
+int
+trx_rollback_for_mysql(
+/*===================*/
+			/* out: error code or DB_SUCCESS */
+	trx_t*	trx);	/* in: transaction handle */
+/***********************************************************************
+Rollback the latest SQL statement for MySQL. */
+UNIV_INTERN
+int
+trx_rollback_last_sql_stat_for_mysql(
+/*=================================*/
+			/* out: error code or DB_SUCCESS */
+	trx_t*	trx);	/* in: transaction handle */
+/***********************************************************************
+Rollback a transaction used in MySQL. */
+UNIV_INTERN
+int
+trx_general_rollback_for_mysql(
+/*===========================*/
+				/* out: error code or DB_SUCCESS */
+	trx_t*		trx,	/* in: transaction handle */
+	ibool		partial,/* in: TRUE if partial rollback requested */
+	trx_savept_t*	savept);/* in: pointer to savepoint undo number, if
+				partial rollback requested */
+/***********************************************************************
+Rolls back a transaction back to a named savepoint. Modifications after the
+savepoint are undone but InnoDB does NOT release the corresponding locks
+which are stored in memory. If a lock is 'implicit', that is, a new inserted
+row holds a lock where the lock information is carried by the trx id stored in
+the row, these locks are naturally released in the rollback. Savepoints which
+were set after this savepoint are deleted. */
+UNIV_INTERN
+ulint
+trx_rollback_to_savepoint_for_mysql(
+/*================================*/
+						/* out: if no savepoint
+						of the name found then
+						DB_NO_SAVEPOINT,
+						otherwise DB_SUCCESS */
+	trx_t*		trx,			/* in: transaction handle */
+	const char*	savepoint_name,		/* in: savepoint name */
+	ib_int64_t*	mysql_binlog_cache_pos);/* out: the MySQL binlog cache
+						position corresponding to this
+						savepoint; MySQL needs this
+						information to remove the
+						binlog entries of the queries
+						executed after the savepoint */
+/***********************************************************************
+Creates a named savepoint. If the transaction is not yet started, starts it.
+If there is already a savepoint of the same name, this call erases that old
+savepoint and replaces it with a new. Savepoints are deleted in a transaction
+commit or rollback. */
+UNIV_INTERN
+ulint
+trx_savepoint_for_mysql(
+/*====================*/
+						/* out: always DB_SUCCESS */
+	trx_t*		trx,			/* in: transaction handle */
+	const char*	savepoint_name,		/* in: savepoint name */
+	ib_int64_t	binlog_cache_pos);	/* in: MySQL binlog cache
+						position corresponding to this
+						connection at the time of the
+						savepoint */
+
+/***********************************************************************
+Releases a named savepoint. Savepoints which
+were set after this savepoint are deleted. */
+UNIV_INTERN
+ulint
+trx_release_savepoint_for_mysql(
+/*============================*/
+						/* out: if no savepoint
+						of the name found then
+						DB_NO_SAVEPOINT,
+						otherwise DB_SUCCESS */
+	trx_t*		trx,			/* in: transaction handle */
+	const char*	savepoint_name);	/* in: savepoint name */
+
+/***********************************************************************
+Frees a single savepoint struct. */
+UNIV_INTERN
+void
+trx_roll_savepoint_free(
+/*=====================*/
+	trx_t*			trx,	/* in: transaction handle */
+	trx_named_savept_t*	savep);	/* in: savepoint to free */
+
+/***********************************************************************
+Frees savepoint structs starting from savep, if savep == NULL then
+free all savepoints. */
+
+void
+trx_roll_savepoints_free(
+/*=====================*/
+	trx_t*			trx,	/* in: transaction handle */
+	trx_named_savept_t*	savep);	/* in: free all savepoints > this one;
+					if this is NULL, free all savepoints
+					of trx */
+
+/* A cell in the array used during a rollback and a purge */
+struct	trx_undo_inf_struct{
+	dulint	trx_no;		/* transaction number: not defined during
+				a rollback */
+	dulint	undo_no;	/* undo number of an undo record */
+	ibool	in_use;		/* TRUE if the cell is in use */
+};
+
+/* During a rollback and a purge, undo numbers of undo records currently being
+processed are stored in this array */
+
+struct trx_undo_arr_struct{
+	ulint		n_cells;	/* number of cells in the array */
+	ulint		n_used;		/* number of cells currently in use */
+	trx_undo_inf_t*	infos;		/* the array of undo infos */
+	mem_heap_t*	heap;		/* memory heap from which allocated */
+};
+
+/* Rollback command node in a query graph */
+struct roll_node_struct{
+	que_common_t	common;	/* node type: QUE_NODE_ROLLBACK */
+	ulint		state;	/* node execution state */
+	ibool		partial;/* TRUE if we want a partial rollback */
+	trx_savept_t	savept;	/* savepoint to which to roll back, in the
+				case of a partial rollback */
+};
+
+/* A savepoint set with SQL's "SAVEPOINT savepoint_id" command */
+struct trx_named_savept_struct{
+	char*		name;		/* savepoint name */
+	trx_savept_t	savept;		/* the undo number corresponding to
+					the savepoint */
+	ib_int64_t	mysql_binlog_cache_pos;
+					/* the MySQL binlog cache position
+					corresponding to this savepoint, not
+					defined if the MySQL binlogging is not
+					enabled */
+	UT_LIST_NODE_T(trx_named_savept_t)
+			trx_savepoints;	/* the list of savepoints of a
+					transaction */
+};
+
+/* Rollback node states */
+#define ROLL_NODE_SEND	1
+#define ROLL_NODE_WAIT	2
+
+#ifndef UNIV_NONINL
+#include "trx0roll.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/trx0roll.ic b/storage/xtradb/include/trx0roll.ic
new file mode 100644
index 00000000000..513b8b44847
--- /dev/null
+++ b/storage/xtradb/include/trx0roll.ic
@@ -0,0 +1,39 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Transaction rollback
+
+Created 3/26/1996 Heikki Tuuri
+*******************************************************/
+
+/***********************************************************************
+Returns pointer to nth element in an undo number array. */
+UNIV_INLINE
+trx_undo_inf_t*
+trx_undo_arr_get_nth_info(
+/*======================*/
+				/* out: pointer to the nth element */
+	trx_undo_arr_t*	arr,	/* in: undo number array */
+	ulint		n)	/* in: position */
+{
+	ut_ad(arr);
+	ut_ad(n < arr->n_cells);
+
+	return(arr->infos + n);
+}
diff --git a/storage/xtradb/include/trx0rseg.h b/storage/xtradb/include/trx0rseg.h
new file mode 100644
index 00000000000..f3aa736f788
--- /dev/null
+++ b/storage/xtradb/include/trx0rseg.h
@@ -0,0 +1,220 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Rollback segment
+
+Created 3/26/1996 Heikki Tuuri
+*******************************************************/
+
+#ifndef trx0rseg_h
+#define trx0rseg_h
+
+#include "univ.i"
+#include "trx0types.h"
+#include "trx0sys.h"
+
+/**********************************************************************
+Gets a rollback segment header. */
+UNIV_INLINE
+trx_rsegf_t*
+trx_rsegf_get(
+/*==========*/
+				/* out: rollback segment header, page
+				x-latched */
+	ulint	space,		/* in: space where placed */
+	ulint	zip_size,	/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	ulint	page_no,	/* in: page number of the header */
+	mtr_t*	mtr);		/* in: mtr */
+/**********************************************************************
+Gets a newly created rollback segment header. */
+UNIV_INLINE
+trx_rsegf_t*
+trx_rsegf_get_new(
+/*==============*/
+				/* out: rollback segment header, page
+				x-latched */
+	ulint	space,		/* in: space where placed */
+	ulint	zip_size,	/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	ulint	page_no,	/* in: page number of the header */
+	mtr_t*	mtr);		/* in: mtr */
+/*******************************************************************
+Gets the file page number of the nth undo log slot. */
+UNIV_INLINE
+ulint
+trx_rsegf_get_nth_undo(
+/*===================*/
+				/* out: page number of the undo log segment */
+	trx_rsegf_t*	rsegf,	/* in: rollback segment header */
+	ulint		n,	/* in: index of slot */
+	mtr_t*		mtr);	/* in: mtr */
+/*******************************************************************
+Sets the file page number of the nth undo log slot. */
+UNIV_INLINE
+void
+trx_rsegf_set_nth_undo(
+/*===================*/
+	trx_rsegf_t*	rsegf,	/* in: rollback segment header */
+	ulint		n,	/* in: index of slot */
+	ulint		page_no,/* in: page number of the undo log segment */
+	mtr_t*		mtr);	/* in: mtr */
+/********************************************************************
+Looks for a free slot for an undo log segment. */
+UNIV_INLINE
+ulint
+trx_rsegf_undo_find_free(
+/*=====================*/
+				/* out: slot index or ULINT_UNDEFINED if not
+				found */
+	trx_rsegf_t*	rsegf,	/* in: rollback segment header */
+	mtr_t*		mtr);	/* in: mtr */
+/**********************************************************************
+Looks for a rollback segment, based on the rollback segment id. */
+UNIV_INTERN
+trx_rseg_t*
+trx_rseg_get_on_id(
+/*===============*/
+			/* out: rollback segment */
+	ulint	id);	/* in: rollback segment id */
+/********************************************************************
+Creates a rollback segment header. This function is called only when
+a new rollback segment is created in the database. */
+UNIV_INTERN
+ulint
+trx_rseg_header_create(
+/*===================*/
+				/* out: page number of the created segment,
+				FIL_NULL if fail */
+	ulint	space,		/* in: space id */
+	ulint	zip_size,	/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	ulint	max_size,	/* in: max size in pages */
+	ulint*	slot_no,	/* out: rseg id == slot number in trx sys */
+	mtr_t*	mtr);		/* in: mtr */
+/*************************************************************************
+Creates the memory copies for rollback segments and initializes the
+rseg list and array in trx_sys at a database startup. */
+UNIV_INTERN
+void
+trx_rseg_list_and_array_init(
+/*=========================*/
+	trx_sysf_t*	sys_header,	/* in: trx system header */
+	mtr_t*		mtr);		/* in: mtr */
+/********************************************************************
+Creates a new rollback segment to the database. */
+UNIV_INTERN
+trx_rseg_t*
+trx_rseg_create(
+/*============*/
+				/* out: the created segment object, NULL if
+				fail */
+	ulint	space,		/* in: space id */
+	ulint	max_size,	/* in: max size in pages */
+	ulint*	id,		/* out: rseg id */
+	mtr_t*	mtr);		/* in: mtr */
+
+
+/* Real max value may be 4076 in usual. But reserve 4 slot for safety or etc... */
+#define TRX_RSEG_N_EXTRA_SLOTS	(((UNIV_PAGE_SIZE - (FIL_PAGE_DATA + FIL_PAGE_DATA_END + TRX_RSEG_UNDO_SLOTS)) / TRX_RSEG_SLOT_SIZE) - 4)
+
+/* Number of undo log slots in a rollback segment file copy */
+#define TRX_RSEG_N_SLOTS	(srv_extra_undoslots ? TRX_RSEG_N_EXTRA_SLOTS : (UNIV_PAGE_SIZE / 16))
+
+/* Maximum number of transactions supported by a single rollback segment */
+#define TRX_RSEG_MAX_N_TRXS	(TRX_RSEG_N_SLOTS / 2)
+
+/* The rollback segment memory object */
+struct trx_rseg_struct{
+	/*--------------------------------------------------------*/
+	ulint		id;	/* rollback segment id == the index of
+				its slot in the trx system file copy */
+	mutex_t		mutex;	/* mutex protecting the fields in this
+				struct except id; NOTE that the latching
+				order must always be kernel mutex ->
+				rseg mutex */
+	ulint		space;	/* space where the rollback segment is
+				header is placed */
+	ulint		zip_size;/* in: compressed page size of space
+				in bytes, or 0 for uncompressed spaces */
+	ulint		page_no;/* page number of the rollback segment
+				header */
+	ulint		max_size;/* maximum allowed size in pages */
+	ulint		curr_size;/* current size in pages */
+	/*--------------------------------------------------------*/
+	/* Fields for update undo logs */
+	UT_LIST_BASE_NODE_T(trx_undo_t) update_undo_list;
+					/* List of update undo logs */
+	UT_LIST_BASE_NODE_T(trx_undo_t) update_undo_cached;
+					/* List of update undo log segments
+					cached for fast reuse */
+	/*--------------------------------------------------------*/
+	/* Fields for insert undo logs */
+	UT_LIST_BASE_NODE_T(trx_undo_t) insert_undo_list;
+					/* List of insert undo logs */
+	UT_LIST_BASE_NODE_T(trx_undo_t) insert_undo_cached;
+					/* List of insert undo log segments
+					cached for fast reuse */
+	/*--------------------------------------------------------*/
+	ulint		last_page_no;	/* Page number of the last not yet
+					purged log header in the history list;
+					FIL_NULL if all list purged */
+	ulint		last_offset;	/* Byte offset of the last not yet
+					purged log header */
+	dulint		last_trx_no;	/* Transaction number of the last not
+					yet purged log */
+	ibool		last_del_marks;	/* TRUE if the last not yet purged log
+					needs purging */
+	/*--------------------------------------------------------*/
+	UT_LIST_NODE_T(trx_rseg_t) rseg_list;
+					/* the list of the rollback segment
+					memory objects */
+};
+
+/* Undo log segment slot in a rollback segment header */
+/*-------------------------------------------------------------*/
+#define	TRX_RSEG_SLOT_PAGE_NO	0	/* Page number of the header page of
+					an undo log segment */
+/*-------------------------------------------------------------*/
+/* Slot size */
+#define TRX_RSEG_SLOT_SIZE	4
+
+/* The offset of the rollback segment header on its page */
+#define	TRX_RSEG		FSEG_PAGE_DATA
+
+/* Transaction rollback segment header */
+/*-------------------------------------------------------------*/
+#define	TRX_RSEG_MAX_SIZE	0	/* Maximum allowed size for rollback
+					segment in pages */
+#define	TRX_RSEG_HISTORY_SIZE	4	/* Number of file pages occupied
+					by the logs in the history list */
+#define	TRX_RSEG_HISTORY	8	/* The update undo logs for committed
+					transactions */
+#define	TRX_RSEG_FSEG_HEADER	(8 + FLST_BASE_NODE_SIZE)
+					/* Header for the file segment where
+					this page is placed */
+#define TRX_RSEG_UNDO_SLOTS	(8 + FLST_BASE_NODE_SIZE + FSEG_HEADER_SIZE)
+					/* Undo log segment slots */
+/*-------------------------------------------------------------*/
+
+#ifndef UNIV_NONINL
+#include "trx0rseg.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/trx0rseg.ic b/storage/xtradb/include/trx0rseg.ic
new file mode 100644
index 00000000000..e665a40fa8b
--- /dev/null
+++ b/storage/xtradb/include/trx0rseg.ic
@@ -0,0 +1,146 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Rollback segment
+
+Created 3/26/1996 Heikki Tuuri
+*******************************************************/
+
+#include "srv0srv.h"
+
+/**********************************************************************
+Gets a rollback segment header. */
+UNIV_INLINE
+trx_rsegf_t*
+trx_rsegf_get(
+/*==========*/
+				/* out: rollback segment header, page
+				x-latched */
+	ulint	space,		/* in: space where placed */
+	ulint	zip_size,	/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	ulint	page_no,	/* in: page number of the header */
+	mtr_t*	mtr)		/* in: mtr */
+{
+	buf_block_t*	block;
+	trx_rsegf_t*	header;
+
+	block = buf_page_get(space, zip_size, page_no, RW_X_LATCH, mtr);
+	buf_block_dbg_add_level(block, SYNC_RSEG_HEADER);
+
+	header = TRX_RSEG + buf_block_get_frame(block);
+
+	return(header);
+}
+
+/**********************************************************************
+Gets a newly created rollback segment header. */
+UNIV_INLINE
+trx_rsegf_t*
+trx_rsegf_get_new(
+/*==============*/
+				/* out: rollback segment header, page
+				x-latched */
+	ulint	space,		/* in: space where placed */
+	ulint	zip_size,	/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	ulint	page_no,	/* in: page number of the header */
+	mtr_t*	mtr)		/* in: mtr */
+{
+	buf_block_t*	block;
+	trx_rsegf_t*	header;
+
+	block = buf_page_get(space, zip_size, page_no, RW_X_LATCH, mtr);
+	buf_block_dbg_add_level(block, SYNC_RSEG_HEADER_NEW);
+
+	header = TRX_RSEG + buf_block_get_frame(block);
+
+	return(header);
+}
+
+/*******************************************************************
+Gets the file page number of the nth undo log slot. */
+UNIV_INLINE
+ulint
+trx_rsegf_get_nth_undo(
+/*===================*/
+				/* out: page number of the undo log segment */
+	trx_rsegf_t*	rsegf,	/* in: rollback segment header */
+	ulint		n,	/* in: index of slot */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	if (UNIV_UNLIKELY(n >= TRX_RSEG_N_SLOTS)) {
+		fprintf(stderr,
+			"InnoDB: Error: trying to get slot %lu of rseg\n",
+			(ulong) n);
+		ut_error;
+	}
+
+	return(mtr_read_ulint(rsegf + TRX_RSEG_UNDO_SLOTS
+			      + n * TRX_RSEG_SLOT_SIZE, MLOG_4BYTES, mtr));
+}
+
+/*******************************************************************
+Sets the file page number of the nth undo log slot. */
+UNIV_INLINE
+void
+trx_rsegf_set_nth_undo(
+/*===================*/
+	trx_rsegf_t*	rsegf,	/* in: rollback segment header */
+	ulint		n,	/* in: index of slot */
+	ulint		page_no,/* in: page number of the undo log segment */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	if (UNIV_UNLIKELY(n >= TRX_RSEG_N_SLOTS)) {
+		fprintf(stderr,
+			"InnoDB: Error: trying to set slot %lu of rseg\n",
+			(ulong) n);
+		ut_error;
+	}
+
+	mlog_write_ulint(rsegf + TRX_RSEG_UNDO_SLOTS + n * TRX_RSEG_SLOT_SIZE,
+			 page_no, MLOG_4BYTES, mtr);
+}
+
+/********************************************************************
+Looks for a free slot for an undo log segment. */
+UNIV_INLINE
+ulint
+trx_rsegf_undo_find_free(
+/*=====================*/
+				/* out: slot index or ULINT_UNDEFINED if not
+				found */
+	trx_rsegf_t*	rsegf,	/* in: rollback segment header */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	ulint		i;
+	ulint		page_no;
+
+	for (i = 0; i < TRX_RSEG_N_SLOTS; i++) {
+
+		page_no = trx_rsegf_get_nth_undo(rsegf, i, mtr);
+
+		if (page_no == FIL_NULL) {
+
+			return(i);
+		}
+	}
+
+	return(ULINT_UNDEFINED);
+}
diff --git a/storage/xtradb/include/trx0sys.h b/storage/xtradb/include/trx0sys.h
new file mode 100644
index 00000000000..f7e7e082278
--- /dev/null
+++ b/storage/xtradb/include/trx0sys.h
@@ -0,0 +1,560 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Transaction system
+
+Created 3/26/1996 Heikki Tuuri
+*******************************************************/
+
+#ifndef trx0sys_h
+#define trx0sys_h
+
+#include "univ.i"
+
+#include "trx0types.h"
+#include "mtr0mtr.h"
+#include "mtr0log.h"
+#include "ut0byte.h"
+#include "mem0mem.h"
+#include "sync0sync.h"
+#include "ut0lst.h"
+#include "buf0buf.h"
+#include "fil0fil.h"
+#include "fut0lst.h"
+#include "fsp0fsp.h"
+#include "read0types.h"
+#include "page0types.h"
+
+/* In a MySQL replication slave, in crash recovery we store the master log
+file name and position here. We have successfully got the updates to InnoDB
+up to this position. If .._pos is -1, it means no crash recovery was needed,
+or there was no master log position info inside InnoDB. */
+
+extern char		trx_sys_mysql_master_log_name[];
+extern ib_int64_t	trx_sys_mysql_master_log_pos;
+
+extern char		trx_sys_mysql_relay_log_name[];
+extern ib_int64_t	trx_sys_mysql_relay_log_pos;
+
+/* If this MySQL server uses binary logging, after InnoDB has been inited
+and if it has done a crash recovery, we store the binlog file name and position
+here. If .._pos is -1, it means there was no binlog position info inside
+InnoDB. */
+
+extern char		trx_sys_mysql_bin_log_name[];
+extern ib_int64_t	trx_sys_mysql_bin_log_pos;
+
+/* The transaction system */
+extern trx_sys_t*	trx_sys;
+
+/* Doublewrite system */
+extern trx_doublewrite_t*	trx_doublewrite;
+extern ibool			trx_doublewrite_must_reset_space_ids;
+extern ibool			trx_sys_multiple_tablespace_format;
+
+/********************************************************************
+Creates the doublewrite buffer to a new InnoDB installation. The header of the
+doublewrite buffer is placed on the trx system header page. */
+UNIV_INTERN
+void
+trx_sys_create_doublewrite_buf(void);
+/*================================*/
+/********************************************************************
+At a database startup initializes the doublewrite buffer memory structure if
+we already have a doublewrite buffer created in the data files. If we are
+upgrading to an InnoDB version which supports multiple tablespaces, then this
+function performs the necessary update operations. If we are in a crash
+recovery, this function uses a possible doublewrite buffer to restore
+half-written pages in the data files. */
+UNIV_INTERN
+void
+trx_sys_doublewrite_init_or_restore_pages(
+/*======================================*/
+	ibool	restore_corrupt_pages);
+/********************************************************************
+Marks the trx sys header when we have successfully upgraded to the >= 4.1.x
+multiple tablespace format. */
+UNIV_INTERN
+void
+trx_sys_mark_upgraded_to_multiple_tablespaces(void);
+/*===============================================*/
+/********************************************************************
+Determines if a page number is located inside the doublewrite buffer. */
+UNIV_INTERN
+ibool
+trx_doublewrite_page_inside(
+/*========================*/
+				/* out: TRUE if the location is inside
+				the two blocks of the doublewrite buffer */
+	ulint	page_no);	/* in: page number */
+/*******************************************************************
+Checks if a page address is the trx sys header page. */
+UNIV_INLINE
+ibool
+trx_sys_hdr_page(
+/*=============*/
+			/* out: TRUE if trx sys header page */
+	ulint	space,	/* in: space */
+	ulint	page_no);/* in: page number */
+/*********************************************************************
+Creates and initializes the central memory structures for the transaction
+system. This is called when the database is started. */
+UNIV_INTERN
+void
+trx_sys_init_at_db_start(void);
+/*==========================*/
+/*********************************************************************
+Creates and initializes the transaction system at the database creation. */
+UNIV_INTERN
+void
+trx_sys_create(void);
+/*================*/
+/*********************************************************************
+Create extra rollback segments when create_new_db */
+UNIV_INTERN
+void
+trx_sys_create_extra_rseg(
+/*======================*/
+	ulint	num);	/* in: number of extra user rollback segments */
+/********************************************************************
+Looks for a free slot for a rollback segment in the trx system file copy. */
+UNIV_INTERN
+ulint
+trx_sysf_rseg_find_free(
+/*====================*/
+					/* out: slot index or ULINT_UNDEFINED
+					if not found */
+	mtr_t*		mtr);		/* in: mtr */
+/*******************************************************************
+Gets the pointer in the nth slot of the rseg array. */
+UNIV_INLINE
+trx_rseg_t*
+trx_sys_get_nth_rseg(
+/*=================*/
+				/* out: pointer to rseg object, NULL if slot
+				not in use */
+	trx_sys_t*	sys,	/* in: trx system */
+	ulint		n);	/* in: index of slot */
+/*******************************************************************
+Sets the pointer in the nth slot of the rseg array. */
+UNIV_INLINE
+void
+trx_sys_set_nth_rseg(
+/*=================*/
+	trx_sys_t*	sys,	/* in: trx system */
+	ulint		n,	/* in: index of slot */
+	trx_rseg_t*	rseg);	/* in: pointer to rseg object, NULL if slot
+				not in use */
+/**************************************************************************
+Gets a pointer to the transaction system file copy and x-locks its page. */
+UNIV_INLINE
+trx_sysf_t*
+trx_sysf_get(
+/*=========*/
+			/* out: pointer to system file copy, page x-locked */
+	mtr_t*	mtr);	/* in: mtr */
+/*********************************************************************
+Gets the space of the nth rollback segment slot in the trx system
+file copy. */
+UNIV_INLINE
+ulint
+trx_sysf_rseg_get_space(
+/*====================*/
+					/* out: space id */
+	trx_sysf_t*	sys_header,	/* in: trx sys file copy */
+	ulint		i,		/* in: slot index == rseg id */
+	mtr_t*		mtr);		/* in: mtr */
+/*********************************************************************
+Gets the page number of the nth rollback segment slot in the trx system
+file copy. */
+UNIV_INLINE
+ulint
+trx_sysf_rseg_get_page_no(
+/*======================*/
+					/* out: page number, FIL_NULL
+					if slot unused */
+	trx_sysf_t*	sys_header,	/* in: trx sys file copy */
+	ulint		i,		/* in: slot index == rseg id */
+	mtr_t*		mtr);		/* in: mtr */
+/*********************************************************************
+Sets the space id of the nth rollback segment slot in the trx system
+file copy. */
+UNIV_INLINE
+void
+trx_sysf_rseg_set_space(
+/*====================*/
+	trx_sysf_t*	sys_header,	/* in: trx sys file copy */
+	ulint		i,		/* in: slot index == rseg id */
+	ulint		space,		/* in: space id */
+	mtr_t*		mtr);		/* in: mtr */
+/*********************************************************************
+Sets the page number of the nth rollback segment slot in the trx system
+file copy. */
+UNIV_INLINE
+void
+trx_sysf_rseg_set_page_no(
+/*======================*/
+	trx_sysf_t*	sys_header,	/* in: trx sys file copy */
+	ulint		i,		/* in: slot index == rseg id */
+	ulint		page_no,	/* in: page number, FIL_NULL if
+					the slot is reset to unused */
+	mtr_t*		mtr);		/* in: mtr */
+/*********************************************************************
+Allocates a new transaction id. */
+UNIV_INLINE
+dulint
+trx_sys_get_new_trx_id(void);
+/*========================*/
+			/* out: new, allocated trx id */
+/*********************************************************************
+Allocates a new transaction number. */
+UNIV_INLINE
+dulint
+trx_sys_get_new_trx_no(void);
+/*========================*/
+			/* out: new, allocated trx number */
+/*********************************************************************
+Writes a trx id to an index page. In case that the id size changes in
+some future version, this function should be used instead of
+mach_write_... */
+UNIV_INLINE
+void
+trx_write_trx_id(
+/*=============*/
+	byte*	ptr,	/* in: pointer to memory where written */
+	dulint	id);	/* in: id */
+/*********************************************************************
+Reads a trx id from an index page. In case that the id size changes in
+some future version, this function should be used instead of
+mach_read_... */
+UNIV_INLINE
+dulint
+trx_read_trx_id(
+/*============*/
+				/* out: id */
+	const byte*	ptr);	/* in: pointer to memory from where to read */
+/********************************************************************
+Looks for the trx handle with the given id in trx_list. */
+UNIV_INLINE
+trx_t*
+trx_get_on_id(
+/*==========*/
+			/* out: the trx handle or NULL if not found */
+	dulint	trx_id);	/* in: trx id to search for */
+/********************************************************************
+Returns the minumum trx id in trx list. This is the smallest id for which
+the trx can possibly be active. (But, you must look at the trx->conc_state to
+find out if the minimum trx id transaction itself is active, or already
+committed.) */
+UNIV_INLINE
+dulint
+trx_list_get_min_trx_id(void);
+/*=========================*/
+			/* out: the minimum trx id, or trx_sys->max_trx_id
+			if the trx list is empty */
+/********************************************************************
+Checks if a transaction with the given id is active. */
+UNIV_INLINE
+ibool
+trx_is_active(
+/*==========*/
+			/* out: TRUE if active */
+	dulint	trx_id);/* in: trx id of the transaction */
+/********************************************************************
+Checks that trx is in the trx list. */
+UNIV_INTERN
+ibool
+trx_in_trx_list(
+/*============*/
+			/* out: TRUE if is in */
+	trx_t*	in_trx);/* in: trx */
+/*********************************************************************
+Updates the offset information about the end of the MySQL binlog entry
+which corresponds to the transaction just being committed. In a MySQL
+replication slave updates the latest master binlog position up to which
+replication has proceeded. */
+UNIV_INTERN
+void
+trx_sys_update_mysql_binlog_offset(
+/*===============================*/
+	const char*	file_name_in,/* in: MySQL log file name */
+	ib_int64_t	offset,	/* in: position in that log file */
+	ulint		field,	/* in: offset of the MySQL log info field in
+				the trx sys header */
+	mtr_t*		mtr);	/* in: mtr */
+/*********************************************************************
+Prints to stderr the MySQL binlog offset info in the trx system header if
+the magic number shows it valid. */
+UNIV_INTERN
+void
+trx_sys_print_mysql_binlog_offset(void);
+/*===================================*/
+#ifdef UNIV_HOTBACKUP
+/*********************************************************************
+Prints to stderr the MySQL binlog info in the system header if the
+magic number shows it valid. */
+UNIV_INTERN
+void
+trx_sys_print_mysql_binlog_offset_from_page(
+/*========================================*/
+	const byte*	page);	/* in: buffer containing the trx
+				system header page, i.e., page number
+				TRX_SYS_PAGE_NO in the tablespace */
+#endif /* UNIV_HOTBACKUP */
+/*********************************************************************
+Prints to stderr the MySQL master log offset info in the trx system header if
+the magic number shows it valid. */
+UNIV_INTERN
+void
+trx_sys_print_mysql_master_log_pos(void);
+/*====================================*/
+/*********************************************************************
+Initializes the tablespace tag system. */
+UNIV_INTERN
+void
+trx_sys_file_format_init(void);
+/*==========================*/
+/*********************************************************************
+Closes the tablespace tag system. */
+UNIV_INTERN
+void
+trx_sys_file_format_close(void);
+/*===========================*/
+/************************************************************************
+Tags the system table space with minimum format id if it has not been
+tagged yet.
+WARNING: This function is only called during the startup and AFTER the
+redo log application during recovery has finished. */
+UNIV_INTERN
+void
+trx_sys_file_format_tag_init(void);
+/*==============================*/
+/*********************************************************************
+Get the name representation of the file format from its id. */
+UNIV_INTERN
+const char*
+trx_sys_file_format_id_to_name(
+/*===========================*/
+					/* out: pointer to the name */
+	const ulint	id);		/* in: id of the file format */
+/*********************************************************************
+Set the file format id unconditionally except if it's already the
+same value. */
+UNIV_INTERN
+ibool
+trx_sys_file_format_max_set(
+/*========================*/
+					/* out: TRUE if value updated */
+	ulint		format_id,	/* in: file format id */
+	const char**	name);		/* out: max file format name or
+					NULL if not needed. */
+/*********************************************************************
+Get the name representation of the file format from its id. */
+UNIV_INTERN
+const char*
+trx_sys_file_format_max_get(void);
+/*=============================*/
+				/* out: pointer to the max format name */
+/*********************************************************************
+Check for the max file format tag stored on disk. */
+UNIV_INTERN
+ulint
+trx_sys_file_format_max_check(
+/*==========================*/
+					/* out: DB_SUCCESS or error code */
+	ulint		max_format_id);	/* in: the max format id to check */
+/************************************************************************
+Update the file format tag in the system tablespace only if the given
+format id is greater than the known max id. */
+UNIV_INTERN
+ibool
+trx_sys_file_format_max_upgrade(
+/*============================*/
+					/* out: TRUE if format_id was
+					bigger than the known max id */
+	const char**	name,		/* out: max file format name */
+	ulint		format_id);	/* in: file format identifier */
+/* The automatically created system rollback segment has this id */
+#define TRX_SYS_SYSTEM_RSEG_ID	0
+
+/* Space id and page no where the trx system file copy resides */
+#define	TRX_SYS_SPACE	0	/* the SYSTEM tablespace */
+#define	TRX_SYS_PAGE_NO	FSP_TRX_SYS_PAGE_NO
+
+/* The offset of the transaction system header on the page */
+#define	TRX_SYS		FSEG_PAGE_DATA
+
+/* Transaction system header */
+/*-------------------------------------------------------------*/
+#define	TRX_SYS_TRX_ID_STORE	0	/* the maximum trx id or trx number
+					modulo TRX_SYS_TRX_ID_UPDATE_MARGIN
+					written to a file page by any
+					transaction; the assignment of
+					transaction ids continues from this
+					number rounded up by .._MARGIN plus
+					.._MARGIN when the database is
+					started */
+#define TRX_SYS_FSEG_HEADER	8	/* segment header for the tablespace
+					segment the trx system is created
+					into */
+#define	TRX_SYS_RSEGS		(8 + FSEG_HEADER_SIZE)
+					/* the start of the array of rollback
+					segment specification slots */
+/*-------------------------------------------------------------*/
+
+/* Max number of rollback segments: the number of segment specification slots
+in the transaction system array; rollback segment id must fit in one byte,
+therefore 256; each slot is currently 8 bytes in size */
+#define	TRX_SYS_N_RSEGS		256
+
+#define TRX_SYS_MYSQL_LOG_NAME_LEN	512
+#define TRX_SYS_MYSQL_MASTER_LOG_NAME_LEN	480	/* (500 - 12) is dead line. */
+#define TRX_SYS_MYSQL_LOG_MAGIC_N	873422344
+
+#if UNIV_PAGE_SIZE < 4096
+# error "UNIV_PAGE_SIZE < 4096"
+#endif
+/* The offset of the MySQL replication info in the trx system header;
+this contains the same fields as TRX_SYS_MYSQL_LOG_INFO below */
+#define TRX_SYS_MYSQL_MASTER_LOG_INFO	(UNIV_PAGE_SIZE - 2000)
+#define TRX_SYS_MYSQL_RELAY_LOG_INFO	(UNIV_PAGE_SIZE - 1500)
+
+/* The offset of the MySQL binlog offset info in the trx system header */
+#define TRX_SYS_MYSQL_LOG_INFO		(UNIV_PAGE_SIZE - 1000)
+#define	TRX_SYS_MYSQL_LOG_MAGIC_N_FLD	0	/* magic number which shows
+						if we have valid data in the
+						MySQL binlog info; the value
+						is ..._MAGIC_N if yes */
+#define TRX_SYS_MYSQL_LOG_OFFSET_HIGH	4	/* high 4 bytes of the offset
+						within that file */
+#define TRX_SYS_MYSQL_LOG_OFFSET_LOW	8	/* low 4 bytes of the offset
+						within that file */
+#define TRX_SYS_MYSQL_LOG_NAME		12	/* MySQL log file name */
+
+/* The offset of the doublewrite buffer header on the trx system header page */
+#define TRX_SYS_DOUBLEWRITE		(UNIV_PAGE_SIZE - 200)
+/*-------------------------------------------------------------*/
+#define TRX_SYS_DOUBLEWRITE_FSEG	0	/* fseg header of the fseg
+						containing the doublewrite
+						buffer */
+#define TRX_SYS_DOUBLEWRITE_MAGIC	FSEG_HEADER_SIZE
+						/* 4-byte magic number which
+						shows if we already have
+						created the doublewrite
+						buffer */
+#define TRX_SYS_DOUBLEWRITE_BLOCK1	(4 + FSEG_HEADER_SIZE)
+						/* page number of the
+						first page in the first
+						sequence of 64
+						(= FSP_EXTENT_SIZE) consecutive
+						pages in the doublewrite
+						buffer */
+#define TRX_SYS_DOUBLEWRITE_BLOCK2	(8 + FSEG_HEADER_SIZE)
+						/* page number of the
+						first page in the second
+						sequence of 64 consecutive
+						pages in the doublewrite
+						buffer */
+#define TRX_SYS_DOUBLEWRITE_REPEAT	12	/* we repeat the above 3
+						numbers so that if the trx
+						sys header is half-written
+						to disk, we still may be able
+						to recover the information */
+#define TRX_SYS_DOUBLEWRITE_SPACE_ID_STORED (24 + FSEG_HEADER_SIZE)
+						/* If this is not yet set to
+						.._N, we must reset the
+						doublewrite buffer, because
+						starting from 4.1.x the space
+						id of a data page is stored to
+					FIL_PAGE_ARCH_LOG_NO_OR_SPACE_NO */
+/*-------------------------------------------------------------*/
+#define TRX_SYS_DOUBLEWRITE_MAGIC_N	536853855
+#define TRX_SYS_DOUBLEWRITE_SPACE_ID_STORED_N 1783657386
+
+
+#define TRX_SYS_DOUBLEWRITE_BLOCK_SIZE	FSP_EXTENT_SIZE
+
+/* The offset of the file format tag on the trx system header page */
+#define TRX_SYS_FILE_FORMAT_TAG		(UNIV_PAGE_SIZE - 16)
+
+/* We use these random constants to reduce the probability of reading
+garbage (from previous versions) that maps to an actual format id. We
+use these as bit masks at the time of  reading and writing from/to disk. */
+#define TRX_SYS_FILE_FORMAT_TAG_MAGIC_N_LOW	3645922177UL
+#define TRX_SYS_FILE_FORMAT_TAG_MAGIC_N_HIGH	2745987765UL
+
+/* Doublewrite control struct */
+struct trx_doublewrite_struct{
+	mutex_t	mutex;		/* mutex protecting the first_free field and
+				write_buf */
+	ulint	block1;		/* the page number of the first
+				doublewrite block (64 pages) */
+	ulint	block2;		/* page number of the second block */
+	ulint	first_free;	/* first free position in write_buf measured
+				in units of UNIV_PAGE_SIZE */
+	byte*	write_buf;	/* write buffer used in writing to the
+				doublewrite buffer, aligned to an
+				address divisible by UNIV_PAGE_SIZE
+				(which is required by Windows aio) */
+	byte*	write_buf_unaligned; /* pointer to write_buf, but unaligned */
+	buf_page_t**
+		buf_block_arr;	/* array to store pointers to the buffer
+				blocks which have been cached to write_buf */
+};
+
+/* The transaction system central memory data structure; protected by the
+kernel mutex */
+struct trx_sys_struct{
+	dulint		max_trx_id;	/* The smallest number not yet
+					assigned as a transaction id or
+					transaction number */
+	UT_LIST_BASE_NODE_T(trx_t) trx_list;
+					/* List of active and committed in
+					memory transactions, sorted on trx id,
+					biggest first */
+	UT_LIST_BASE_NODE_T(trx_t) mysql_trx_list;
+					/* List of transactions created
+					for MySQL */
+	UT_LIST_BASE_NODE_T(trx_rseg_t) rseg_list;
+					/* List of rollback segment objects */
+	trx_rseg_t*	latest_rseg;	/* Latest rollback segment in the
+					round-robin assignment of rollback
+					segments to transactions */
+	trx_rseg_t*	rseg_array[TRX_SYS_N_RSEGS];
+					/* Pointer array to rollback segments;
+					NULL if slot not in use */
+	ulint		rseg_history_len;/* Length of the TRX_RSEG_HISTORY
+					list (update undo logs for committed
+					transactions), protected by
+					rseg->mutex */
+	UT_LIST_BASE_NODE_T(read_view_t) view_list;
+					/* List of read views sorted on trx no,
+					biggest first */
+};
+
+/* When a trx id which is zero modulo this number (which must be a power of
+two) is assigned, the field TRX_SYS_TRX_ID_STORE on the transaction system
+page is updated */
+#define TRX_SYS_TRX_ID_WRITE_MARGIN	256
+
+#ifndef UNIV_NONINL
+#include "trx0sys.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/trx0sys.ic b/storage/xtradb/include/trx0sys.ic
new file mode 100644
index 00000000000..4437133188f
--- /dev/null
+++ b/storage/xtradb/include/trx0sys.ic
@@ -0,0 +1,383 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Transaction system
+
+Created 3/26/1996 Heikki Tuuri
+*******************************************************/
+
+#include "srv0srv.h"
+#include "trx0trx.h"
+
+/* The typedef for rseg slot in the file copy */
+typedef byte	trx_sysf_rseg_t;
+
+/* Rollback segment specification slot offsets */
+/*-------------------------------------------------------------*/
+#define	TRX_SYS_RSEG_SPACE	0	/* space where the the segment
+					header is placed; starting with
+					MySQL/InnoDB 5.1.7, this is
+					UNIV_UNDEFINED if the slot is unused */
+#define	TRX_SYS_RSEG_PAGE_NO	4	/*  page number where the the segment
+					header is placed; this is FIL_NULL
+					if the slot is unused */
+/*-------------------------------------------------------------*/
+/* Size of a rollback segment specification slot */
+#define TRX_SYS_RSEG_SLOT_SIZE	8
+
+/*********************************************************************
+Writes the value of max_trx_id to the file based trx system header. */
+UNIV_INTERN
+void
+trx_sys_flush_max_trx_id(void);
+/*==========================*/
+
+/*******************************************************************
+Checks if a page address is the trx sys header page. */
+UNIV_INLINE
+ibool
+trx_sys_hdr_page(
+/*=============*/
+			/* out: TRUE if trx sys header page */
+	ulint	space,	/* in: space */
+	ulint	page_no)/* in: page number */
+{
+	if ((space == TRX_SYS_SPACE) && (page_no == TRX_SYS_PAGE_NO)) {
+
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
+
+/*******************************************************************
+Gets the pointer in the nth slot of the rseg array. */
+UNIV_INLINE
+trx_rseg_t*
+trx_sys_get_nth_rseg(
+/*=================*/
+				/* out: pointer to rseg object, NULL if slot
+				not in use */
+	trx_sys_t*	sys,	/* in: trx system */
+	ulint		n)	/* in: index of slot */
+{
+	ut_ad(mutex_own(&(kernel_mutex)));
+	ut_ad(n < TRX_SYS_N_RSEGS);
+
+	return(sys->rseg_array[n]);
+}
+
+/*******************************************************************
+Sets the pointer in the nth slot of the rseg array. */
+UNIV_INLINE
+void
+trx_sys_set_nth_rseg(
+/*=================*/
+	trx_sys_t*	sys,	/* in: trx system */
+	ulint		n,	/* in: index of slot */
+	trx_rseg_t*	rseg)	/* in: pointer to rseg object, NULL if slot
+				not in use */
+{
+	ut_ad(n < TRX_SYS_N_RSEGS);
+
+	sys->rseg_array[n] = rseg;
+}
+
+/**************************************************************************
+Gets a pointer to the transaction system header and x-latches its page. */
+UNIV_INLINE
+trx_sysf_t*
+trx_sysf_get(
+/*=========*/
+			/* out: pointer to system header, page x-latched. */
+	mtr_t*	mtr)	/* in: mtr */
+{
+	buf_block_t*	block;
+	trx_sysf_t*	header;
+
+	ut_ad(mtr);
+
+	block = buf_page_get(TRX_SYS_SPACE, 0, TRX_SYS_PAGE_NO,
+			     RW_X_LATCH, mtr);
+	buf_block_dbg_add_level(block, SYNC_TRX_SYS_HEADER);
+
+	header = TRX_SYS + buf_block_get_frame(block);
+
+	return(header);
+}
+
+/*********************************************************************
+Gets the space of the nth rollback segment slot in the trx system
+file copy. */
+UNIV_INLINE
+ulint
+trx_sysf_rseg_get_space(
+/*====================*/
+					/* out: space id */
+	trx_sysf_t*	sys_header,	/* in: trx sys header */
+	ulint		i,		/* in: slot index == rseg id */
+	mtr_t*		mtr)		/* in: mtr */
+{
+	ut_ad(mutex_own(&(kernel_mutex)));
+	ut_ad(sys_header);
+	ut_ad(i < TRX_SYS_N_RSEGS);
+
+	return(mtr_read_ulint(sys_header + TRX_SYS_RSEGS
+			      + i * TRX_SYS_RSEG_SLOT_SIZE
+			      + TRX_SYS_RSEG_SPACE, MLOG_4BYTES, mtr));
+}
+
+/*********************************************************************
+Gets the page number of the nth rollback segment slot in the trx system
+header. */
+UNIV_INLINE
+ulint
+trx_sysf_rseg_get_page_no(
+/*======================*/
+					/* out: page number, FIL_NULL
+					if slot unused */
+	trx_sysf_t*	sys_header,	/* in: trx system header */
+	ulint		i,		/* in: slot index == rseg id */
+	mtr_t*		mtr)		/* in: mtr */
+{
+	ut_ad(sys_header);
+	ut_ad(mutex_own(&(kernel_mutex)));
+	ut_ad(i < TRX_SYS_N_RSEGS);
+
+	return(mtr_read_ulint(sys_header + TRX_SYS_RSEGS
+			      + i * TRX_SYS_RSEG_SLOT_SIZE
+			      + TRX_SYS_RSEG_PAGE_NO, MLOG_4BYTES, mtr));
+}
+
+/*********************************************************************
+Sets the space id of the nth rollback segment slot in the trx system
+file copy. */
+UNIV_INLINE
+void
+trx_sysf_rseg_set_space(
+/*====================*/
+	trx_sysf_t*	sys_header,	/* in: trx sys file copy */
+	ulint		i,		/* in: slot index == rseg id */
+	ulint		space,		/* in: space id */
+	mtr_t*		mtr)		/* in: mtr */
+{
+	ut_ad(mutex_own(&(kernel_mutex)));
+	ut_ad(sys_header);
+	ut_ad(i < TRX_SYS_N_RSEGS);
+
+	mlog_write_ulint(sys_header + TRX_SYS_RSEGS
+			 + i * TRX_SYS_RSEG_SLOT_SIZE
+			 + TRX_SYS_RSEG_SPACE,
+			 space,
+			 MLOG_4BYTES, mtr);
+}
+
+/*********************************************************************
+Sets the page number of the nth rollback segment slot in the trx system
+header. */
+UNIV_INLINE
+void
+trx_sysf_rseg_set_page_no(
+/*======================*/
+	trx_sysf_t*	sys_header,	/* in: trx sys header */
+	ulint		i,		/* in: slot index == rseg id */
+	ulint		page_no,	/* in: page number, FIL_NULL if the
+					slot is reset to unused */
+	mtr_t*		mtr)		/* in: mtr */
+{
+	ut_ad(mutex_own(&(kernel_mutex)));
+	ut_ad(sys_header);
+	ut_ad(i < TRX_SYS_N_RSEGS);
+
+	mlog_write_ulint(sys_header + TRX_SYS_RSEGS
+			 + i * TRX_SYS_RSEG_SLOT_SIZE
+			 + TRX_SYS_RSEG_PAGE_NO,
+			 page_no,
+			 MLOG_4BYTES, mtr);
+}
+
+/*********************************************************************
+Writes a trx id to an index page. In case that the id size changes in
+some future version, this function should be used instead of
+mach_write_... */
+UNIV_INLINE
+void
+trx_write_trx_id(
+/*=============*/
+	byte*	ptr,	/* in: pointer to memory where written */
+	dulint	id)	/* in: id */
+{
+#if DATA_TRX_ID_LEN != 6
+# error "DATA_TRX_ID_LEN != 6"
+#endif
+	mach_write_to_6(ptr, id);
+}
+
+/*********************************************************************
+Reads a trx id from an index page. In case that the id size changes in
+some future version, this function should be used instead of
+mach_read_... */
+UNIV_INLINE
+dulint
+trx_read_trx_id(
+/*============*/
+				/* out: id */
+	const byte*	ptr)	/* in: pointer to memory from where to read */
+{
+#if DATA_TRX_ID_LEN != 6
+# error "DATA_TRX_ID_LEN != 6"
+#endif
+	return(mach_read_from_6(ptr));
+}
+
+/********************************************************************
+Looks for the trx handle with the given id in trx_list. */
+UNIV_INLINE
+trx_t*
+trx_get_on_id(
+/*==========*/
+			/* out: the trx handle or NULL if not found */
+	dulint	trx_id)	/* in: trx id to search for */
+{
+	trx_t*	trx;
+
+	ut_ad(mutex_own(&(kernel_mutex)));
+
+	trx = UT_LIST_GET_FIRST(trx_sys->trx_list);
+
+	while (trx != NULL) {
+		if (0 == ut_dulint_cmp(trx_id, trx->id)) {
+
+			return(trx);
+		}
+
+		trx = UT_LIST_GET_NEXT(trx_list, trx);
+	}
+
+	return(NULL);
+}
+
+/********************************************************************
+Returns the minumum trx id in trx list. This is the smallest id for which
+the trx can possibly be active. (But, you must look at the trx->conc_state to
+find out if the minimum trx id transaction itself is active, or already
+committed.) */
+UNIV_INLINE
+dulint
+trx_list_get_min_trx_id(void)
+/*=========================*/
+			/* out: the minimum trx id, or trx_sys->max_trx_id
+			if the trx list is empty */
+{
+	trx_t*	trx;
+
+	ut_ad(mutex_own(&(kernel_mutex)));
+
+	trx = UT_LIST_GET_LAST(trx_sys->trx_list);
+
+	if (trx == NULL) {
+
+		return(trx_sys->max_trx_id);
+	}
+
+	return(trx->id);
+}
+
+/********************************************************************
+Checks if a transaction with the given id is active. */
+UNIV_INLINE
+ibool
+trx_is_active(
+/*==========*/
+			/* out: TRUE if active */
+	dulint	trx_id)	/* in: trx id of the transaction */
+{
+	trx_t*	trx;
+
+	ut_ad(mutex_own(&(kernel_mutex)));
+
+	if (ut_dulint_cmp(trx_id, trx_list_get_min_trx_id()) < 0) {
+
+		return(FALSE);
+	}
+
+	if (ut_dulint_cmp(trx_id, trx_sys->max_trx_id) >= 0) {
+
+		/* There must be corruption: we return TRUE because this
+		function is only called by lock_clust_rec_some_has_impl()
+		and row_vers_impl_x_locked_off_kernel() and they have
+		diagnostic prints in this case */
+
+		return(TRUE);
+	}
+
+	trx = trx_get_on_id(trx_id);
+	if (trx && (trx->conc_state == TRX_ACTIVE
+		    || trx->conc_state == TRX_PREPARED)) {
+
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
+
+/*********************************************************************
+Allocates a new transaction id. */
+UNIV_INLINE
+dulint
+trx_sys_get_new_trx_id(void)
+/*========================*/
+			/* out: new, allocated trx id */
+{
+	dulint	id;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	/* VERY important: after the database is started, max_trx_id value is
+	divisible by TRX_SYS_TRX_ID_WRITE_MARGIN, and the following if
+	will evaluate to TRUE when this function is first time called,
+	and the value for trx id will be written to disk-based header!
+	Thus trx id values will not overlap when the database is
+	repeatedly started! */
+
+	if (ut_dulint_get_low(trx_sys->max_trx_id)
+	    % TRX_SYS_TRX_ID_WRITE_MARGIN == 0) {
+
+		trx_sys_flush_max_trx_id();
+	}
+
+	id = trx_sys->max_trx_id;
+
+	UT_DULINT_INC(trx_sys->max_trx_id);
+
+	return(id);
+}
+
+/*********************************************************************
+Allocates a new transaction number. */
+UNIV_INLINE
+dulint
+trx_sys_get_new_trx_no(void)
+/*========================*/
+			/* out: new, allocated trx number */
+{
+	ut_ad(mutex_own(&kernel_mutex));
+
+	return(trx_sys_get_new_trx_id());
+}
diff --git a/storage/xtradb/include/trx0trx.h b/storage/xtradb/include/trx0trx.h
new file mode 100644
index 00000000000..e7f7539f9ee
--- /dev/null
+++ b/storage/xtradb/include/trx0trx.h
@@ -0,0 +1,860 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The transaction
+
+Created 3/26/1996 Heikki Tuuri
+*******************************************************/
+
+#ifndef trx0trx_h
+#define trx0trx_h
+
+#include "univ.i"
+#include "trx0types.h"
+#include "lock0types.h"
+#include "usr0types.h"
+#include "que0types.h"
+#include "mem0mem.h"
+#include "read0types.h"
+#include "dict0types.h"
+#include "trx0xa.h"
+#include "ut0vec.h"
+
+/* Dummy session used currently in MySQL interface */
+extern sess_t*	trx_dummy_sess;
+
+/* Number of transactions currently allocated for MySQL: protected by
+the kernel mutex */
+extern ulint	trx_n_mysql_transactions;
+
+/*****************************************************************
+Resets the new record lock info in a transaction struct. */
+UNIV_INLINE
+void
+trx_reset_new_rec_lock_info(
+/*========================*/
+	trx_t*	trx);	/* in: transaction struct */
+/*****************************************************************
+Registers that we have set a new record lock on an index. We only have space
+to store 2 indexes! If this is called to store more than 2 indexes after
+trx_reset_new_rec_lock_info(), then this function does nothing. */
+UNIV_INLINE
+void
+trx_register_new_rec_lock(
+/*======================*/
+	trx_t*		trx,	/* in: transaction struct */
+	dict_index_t*	index);	/* in: trx sets a new record lock on this
+				index */
+/*****************************************************************
+Checks if trx has set a new record lock on an index. */
+UNIV_INLINE
+ibool
+trx_new_rec_locks_contain(
+/*======================*/
+				/* out: TRUE if trx has set a new record lock
+				on index */
+	trx_t*		trx,	/* in: transaction struct */
+	dict_index_t*	index);	/* in: index */
+/************************************************************************
+Releases the search latch if trx has reserved it. */
+UNIV_INTERN
+void
+trx_search_latch_release_if_reserved(
+/*=================================*/
+	trx_t*	   trx); /* in: transaction */
+/**********************************************************************
+Set detailed error message for the transaction. */
+UNIV_INTERN
+void
+trx_set_detailed_error(
+/*===================*/
+	trx_t*		trx,	/* in: transaction struct */
+	const char*	msg);	/* in: detailed error message */
+/*****************************************************************
+Set detailed error message for the transaction from a file. Note that the
+file is rewinded before reading from it. */
+UNIV_INTERN
+void
+trx_set_detailed_error_from_file(
+/*=============================*/
+	trx_t*	trx,	/* in: transaction struct */
+	FILE*	file);	/* in: file to read message from */
+/********************************************************************
+Retrieves the error_info field from a trx. */
+UNIV_INLINE
+const dict_index_t*
+trx_get_error_info(
+/*===============*/
+				/* out: the error info */
+	const trx_t*	trx);	/* in: trx object */
+/********************************************************************
+Creates and initializes a transaction object. */
+UNIV_INTERN
+trx_t*
+trx_create(
+/*=======*/
+			/* out, own: the transaction */
+	sess_t*	sess)	/* in: session */
+	__attribute__((nonnull));
+/************************************************************************
+Creates a transaction object for MySQL. */
+UNIV_INTERN
+trx_t*
+trx_allocate_for_mysql(void);
+/*========================*/
+				/* out, own: transaction object */
+/************************************************************************
+Creates a transaction object for background operations by the master thread. */
+UNIV_INTERN
+trx_t*
+trx_allocate_for_background(void);
+/*=============================*/
+				/* out, own: transaction object */
+/************************************************************************
+Frees a transaction object. */
+UNIV_INTERN
+void
+trx_free(
+/*=====*/
+	trx_t*	trx);	/* in, own: trx object */
+/************************************************************************
+Frees a transaction object for MySQL. */
+UNIV_INTERN
+void
+trx_free_for_mysql(
+/*===============*/
+	trx_t*	trx);	/* in, own: trx object */
+/************************************************************************
+Frees a transaction object of a background operation of the master thread. */
+UNIV_INTERN
+void
+trx_free_for_background(
+/*====================*/
+	trx_t*	trx);	/* in, own: trx object */
+/********************************************************************
+Creates trx objects for transactions and initializes the trx list of
+trx_sys at database start. Rollback segment and undo log lists must
+already exist when this function is called, because the lists of
+transactions to be rolled back or cleaned up are built based on the
+undo log lists. */
+UNIV_INTERN
+void
+trx_lists_init_at_db_start(void);
+/*============================*/
+/********************************************************************
+Starts a new transaction. */
+UNIV_INTERN
+ibool
+trx_start(
+/*======*/
+			/* out: TRUE if success, FALSE if the rollback
+			segment could not support this many transactions */
+	trx_t*	trx,	/* in: transaction */
+	ulint	rseg_id);/* in: rollback segment id; if ULINT_UNDEFINED
+			is passed, the system chooses the rollback segment
+			automatically in a round-robin fashion */
+/********************************************************************
+Starts a new transaction. */
+UNIV_INTERN
+ibool
+trx_start_low(
+/*==========*/
+			/* out: TRUE */
+	trx_t*	trx,	/* in: transaction */
+	ulint	rseg_id);/* in: rollback segment id; if ULINT_UNDEFINED
+			is passed, the system chooses the rollback segment
+			automatically in a round-robin fashion */
+/*****************************************************************
+Starts the transaction if it is not yet started. */
+UNIV_INLINE
+void
+trx_start_if_not_started(
+/*=====================*/
+	trx_t*	trx);	/* in: transaction */
+/*****************************************************************
+Starts the transaction if it is not yet started. Assumes we have reserved
+the kernel mutex! */
+UNIV_INLINE
+void
+trx_start_if_not_started_low(
+/*=========================*/
+	trx_t*	trx);	/* in: transaction */
+/********************************************************************
+Commits a transaction. */
+UNIV_INTERN
+void
+trx_commit_off_kernel(
+/*==================*/
+	trx_t*	trx);	/* in: transaction */
+/********************************************************************
+Cleans up a transaction at database startup. The cleanup is needed if
+the transaction already got to the middle of a commit when the database
+crashed, andf we cannot roll it back. */
+UNIV_INTERN
+void
+trx_cleanup_at_db_startup(
+/*======================*/
+	trx_t*	trx);	/* in: transaction */
+/**************************************************************************
+Does the transaction commit for MySQL. */
+UNIV_INTERN
+ulint
+trx_commit_for_mysql(
+/*=================*/
+			/* out: DB_SUCCESS or error number */
+	trx_t*	trx);	/* in: trx handle */
+/**************************************************************************
+Does the transaction prepare for MySQL. */
+UNIV_INTERN
+ulint
+trx_prepare_for_mysql(
+/*==================*/
+			/* out: 0 or error number */
+	trx_t*	trx);	/* in: trx handle */
+/**************************************************************************
+This function is used to find number of prepared transactions and
+their transaction objects for a recovery. */
+UNIV_INTERN
+int
+trx_recover_for_mysql(
+/*==================*/
+				/* out: number of prepared transactions */
+	XID*	xid_list,	/* in/out: prepared transactions */
+	ulint	len);		/* in: number of slots in xid_list */
+/***********************************************************************
+This function is used to find one X/Open XA distributed transaction
+which is in the prepared state */
+UNIV_INTERN
+trx_t *
+trx_get_trx_by_xid(
+/*===============*/
+			/* out: trx or NULL */
+	XID*	xid);	/*  in: X/Open XA transaction identification */
+/**************************************************************************
+If required, flushes the log to disk if we called trx_commit_for_mysql()
+with trx->flush_log_later == TRUE. */
+UNIV_INTERN
+ulint
+trx_commit_complete_for_mysql(
+/*==========================*/
+			/* out: 0 or error number */
+	trx_t*	trx);	/* in: trx handle */
+/**************************************************************************
+Marks the latest SQL statement ended. */
+UNIV_INTERN
+void
+trx_mark_sql_stat_end(
+/*==================*/
+	trx_t*	trx);	/* in: trx handle */
+/************************************************************************
+Assigns a read view for a consistent read query. All the consistent reads
+within the same transaction will get the same read view, which is created
+when this function is first called for a new started transaction. */
+UNIV_INTERN
+read_view_t*
+trx_assign_read_view(
+/*=================*/
+			/* out: consistent read view */
+	trx_t*	trx);	/* in: active transaction */
+/***************************************************************
+The transaction must be in the TRX_QUE_LOCK_WAIT state. Puts it to
+the TRX_QUE_RUNNING state and releases query threads which were
+waiting for a lock in the wait_thrs list. */
+UNIV_INTERN
+void
+trx_end_lock_wait(
+/*==============*/
+	trx_t*	trx);	/* in: transaction */
+/********************************************************************
+Sends a signal to a trx object. */
+UNIV_INTERN
+void
+trx_sig_send(
+/*=========*/
+	trx_t*		trx,		/* in: trx handle */
+	ulint		type,		/* in: signal type */
+	ulint		sender,		/* in: TRX_SIG_SELF or
+					TRX_SIG_OTHER_SESS */
+	que_thr_t*	receiver_thr,	/* in: query thread which wants the
+					reply, or NULL; if type is
+					TRX_SIG_END_WAIT, this must be NULL */
+	trx_savept_t*	savept,		/* in: possible rollback savepoint, or
+					NULL */
+	que_thr_t**	next_thr);	/* in/out: next query thread to run;
+					if the value which is passed in is
+					a pointer to a NULL pointer, then the
+					calling function can start running
+					a new query thread; if the parameter
+					is NULL, it is ignored */
+/********************************************************************
+Send the reply message when a signal in the queue of the trx has
+been handled. */
+UNIV_INTERN
+void
+trx_sig_reply(
+/*==========*/
+	trx_sig_t*	sig,		/* in: signal */
+	que_thr_t**	next_thr);	/* in/out: next query thread to run;
+					if the value which is passed in is
+					a pointer to a NULL pointer, then the
+					calling function can start running
+					a new query thread */
+/********************************************************************
+Removes the signal object from a trx signal queue. */
+UNIV_INTERN
+void
+trx_sig_remove(
+/*===========*/
+	trx_t*		trx,	/* in: trx handle */
+	trx_sig_t*	sig);	/* in, own: signal */
+/********************************************************************
+Starts handling of a trx signal. */
+UNIV_INTERN
+void
+trx_sig_start_handle(
+/*=================*/
+	trx_t*		trx,		/* in: trx handle */
+	que_thr_t**	next_thr);	/* in/out: next query thread to run;
+					if the value which is passed in is
+					a pointer to a NULL pointer, then the
+					calling function can start running
+					a new query thread */
+/********************************************************************
+Ends signal handling. If the session is in the error state, and
+trx->graph_before_signal_handling != NULL, returns control to the error
+handling routine of the graph (currently only returns the control to the
+graph root which then sends an error message to the client). */
+UNIV_INTERN
+void
+trx_end_signal_handling(
+/*====================*/
+	trx_t*	trx);	/* in: trx */
+/*************************************************************************
+Creates a commit command node struct. */
+UNIV_INTERN
+commit_node_t*
+commit_node_create(
+/*===============*/
+				/* out, own: commit node struct */
+	mem_heap_t*	heap);	/* in: mem heap where created */
+/***************************************************************
+Performs an execution step for a commit type node in a query graph. */
+UNIV_INTERN
+que_thr_t*
+trx_commit_step(
+/*============*/
+				/* out: query thread to run next, or NULL */
+	que_thr_t*	thr);	/* in: query thread */
+
+/**************************************************************************
+Prints info about a transaction to the given file. The caller must own the
+kernel mutex and must have called
+innobase_mysql_prepare_print_arbitrary_thd(), unless he knows that MySQL
+or InnoDB cannot meanwhile change the info printed here. */
+UNIV_INTERN
+void
+trx_print(
+/*======*/
+	FILE*	f,		/* in: output stream */
+	trx_t*	trx,		/* in: transaction */
+	ulint	max_query_len);	/* in: max query length to print, or 0 to
+				   use the default max length */
+
+/** Type of data dictionary operation */
+enum trx_dict_op {
+	/** The transaction is not modifying the data dictionary. */
+	TRX_DICT_OP_NONE = 0,
+	/** The transaction is creating a table or an index, or
+	dropping a table.  The table must be dropped in crash
+	recovery.  This and TRX_DICT_OP_NONE are the only possible
+	operation modes in crash recovery. */
+	TRX_DICT_OP_TABLE = 1,
+	/** The transaction is creating or dropping an index in an
+	existing table.  In crash recovery, the the data dictionary
+	must be locked, but the table must not be dropped. */
+	TRX_DICT_OP_INDEX = 2
+};
+
+/**************************************************************************
+Determine if a transaction is a dictionary operation. */
+UNIV_INLINE
+enum trx_dict_op
+trx_get_dict_operation(
+/*===================*/
+				/* out: dictionary operation mode */
+	const trx_t*	trx)	/* in: transaction */
+	__attribute__((pure));
+/**************************************************************************
+Flag a transaction a dictionary operation. */
+UNIV_INLINE
+void
+trx_set_dict_operation(
+/*===================*/
+	trx_t*			trx,	/* in/out: transaction */
+	enum trx_dict_op	op);	/* in: operation, not
+					TRX_DICT_OP_NONE */
+
+#ifndef UNIV_HOTBACKUP
+/**************************************************************************
+Determines if the currently running transaction has been interrupted. */
+UNIV_INTERN
+ibool
+trx_is_interrupted(
+/*===============*/
+			/* out: TRUE if interrupted */
+	trx_t*	trx);	/* in: transaction */
+#else /* !UNIV_HOTBACKUP */
+#define trx_is_interrupted(trx) FALSE
+#endif /* !UNIV_HOTBACKUP */
+
+/***********************************************************************
+Calculates the "weight" of a transaction. The weight of one transaction
+is estimated as the number of altered rows + the number of locked rows.
+*/
+
+#define TRX_WEIGHT(t)	\
+	ut_dulint_add((t)->undo_no, UT_LIST_GET_LEN((t)->trx_locks))
+
+/***********************************************************************
+Compares the "weight" (or size) of two transactions. Transactions that
+have edited non-transactional tables are considered heavier than ones
+that have not. */
+UNIV_INTERN
+int
+trx_weight_cmp(
+/*===========*/
+				/* out: <0, 0 or >0; similar to strcmp(3) */
+	const trx_t*	a,	/* in: the first transaction to be compared */
+	const trx_t*	b);	/* in: the second transaction to be compared */
+
+/***********************************************************************
+Retrieves transacion's id, represented as unsigned long long. */
+UNIV_INLINE
+ullint
+trx_get_id(
+/*=======*/
+				/* out: transaction's id */
+	const trx_t*	trx);	/* in: transaction */
+
+/* Maximum length of a string that can be returned by
+trx_get_que_state_str(). */
+#define TRX_QUE_STATE_STR_MAX_LEN	12 /* "ROLLING BACK" */
+
+/***********************************************************************
+Retrieves transaction's que state in a human readable string. The string
+should not be free()'d or modified. */
+UNIV_INLINE
+const char*
+trx_get_que_state_str(
+/*==================*/
+				/* out: string in the data segment */
+	const trx_t*	trx);	/* in: transaction */
+
+/* Signal to a transaction */
+struct trx_sig_struct{
+	unsigned	type:3;		/* signal type */
+	unsigned	sender:1;	/* TRX_SIG_SELF or
+					TRX_SIG_OTHER_SESS */
+	que_thr_t*	receiver;	/* non-NULL if the sender of the signal
+					wants reply after the operation induced
+					by the signal is completed */
+	trx_savept_t	savept;		/* possible rollback savepoint */
+	UT_LIST_NODE_T(trx_sig_t)
+			signals;	/* queue of pending signals to the
+					transaction */
+	UT_LIST_NODE_T(trx_sig_t)
+			reply_signals;	/* list of signals for which the sender
+					transaction is waiting a reply */
+};
+
+#define TRX_MAGIC_N	91118598
+
+/* The transaction handle; every session has a trx object which is freed only
+when the session is freed; in addition there may be session-less transactions
+rolling back after a database recovery */
+
+struct trx_struct{
+	ulint		magic_n;
+	/* All the next fields are protected by the kernel mutex, except the
+	undo logs which are protected by undo_mutex */
+	const char*	op_info;	/* English text describing the
+					current operation, or an empty
+					string */
+	unsigned	is_purge:1;	/* 0=user transaction, 1=purge */
+	unsigned	is_recovered:1;	/* 0=normal transaction,
+					1=recovered, must be rolled back */
+	unsigned	conc_state:2;	/* state of the trx from the point
+					of view of concurrency control:
+					TRX_ACTIVE, TRX_COMMITTED_IN_MEMORY,
+					... */
+	unsigned	que_state:2;	/* valid when conc_state == TRX_ACTIVE:
+					TRX_QUE_RUNNING, TRX_QUE_LOCK_WAIT,
+					... */
+	unsigned	isolation_level:2;/* TRX_ISO_REPEATABLE_READ, ... */
+	unsigned	check_foreigns:1;/* normally TRUE, but if the user
+					wants to suppress foreign key checks,
+					(in table imports, for example) we
+					set this FALSE */
+	unsigned	check_unique_secondary:1;
+					/* normally TRUE, but if the user
+					wants to speed up inserts by
+					suppressing unique key checks
+					for secondary indexes when we decide
+					if we can use the insert buffer for
+					them, we set this FALSE */
+	unsigned	support_xa:1;	/* normally we do the XA two-phase
+					commit steps, but by setting this to
+					FALSE, one can save CPU time and about
+					150 bytes in the undo log size as then
+					we skip XA steps */
+	unsigned	flush_log_later:1;/* when we commit the transaction
+					in MySQL's binlog write, we will
+					flush the log to disk later in
+					a separate call */
+	unsigned	must_flush_log_later:1;/* this flag is set to TRUE in
+					trx_commit_off_kernel() if
+					flush_log_later was TRUE, and there
+					were modifications by the transaction;
+					in that case we must flush the log
+					in trx_commit_complete_for_mysql() */
+	unsigned	dict_operation:2;/**< @see enum trx_dict_op */
+	unsigned	duplicates:2;	/* TRX_DUP_IGNORE | TRX_DUP_REPLACE */
+	unsigned	active_trans:2;	/* 1 - if a transaction in MySQL
+					is active. 2 - if prepare_commit_mutex
+					was taken */
+	unsigned	has_search_latch:1;
+					/* TRUE if this trx has latched the
+					search system latch in S-mode */
+	unsigned	declared_to_be_inside_innodb:1;
+					/* this is TRUE if we have declared
+					this transaction in
+					srv_conc_enter_innodb to be inside the
+					InnoDB engine */
+	unsigned	handling_signals:1;/* this is TRUE as long as the trx
+					is handling signals */
+	unsigned	dict_operation_lock_mode:2;
+					/* 0, RW_S_LATCH, or RW_X_LATCH:
+					the latch mode trx currently holds
+					on dict_operation_lock */
+	time_t		start_time;	/* time the trx object was created
+					or the state last time became
+					TRX_ACTIVE */
+	dulint		id;		/* transaction id */
+	XID		xid;		/* X/Open XA transaction
+					identification to identify a
+					transaction branch */
+	dulint		no;		/* transaction serialization number ==
+					max trx id when the transaction is
+					moved to COMMITTED_IN_MEMORY state */
+	ib_uint64_t	commit_lsn;	/* lsn at the time of the commit */
+	dulint		table_id;	/* Table to drop iff dict_operation
+					is TRUE, or ut_dulint_zero. */
+	/*------------------------------*/
+	void*		mysql_thd;	/* MySQL thread handle corresponding
+					to this trx, or NULL */
+	char**		mysql_query_str;/* pointer to the field in mysqld_thd
+					which contains the pointer to the
+					current SQL query string */
+	const char*	mysql_log_file_name;
+					/* if MySQL binlog is used, this field
+					contains a pointer to the latest file
+					name; this is NULL if binlog is not
+					used */
+	ib_int64_t	mysql_log_offset;/* if MySQL binlog is used, this field
+					contains the end offset of the binlog
+					entry */
+	const char*	mysql_master_log_file_name;
+					/* if the database server is a MySQL
+					replication slave, we have here the
+					master binlog name up to which
+					replication has processed; otherwise
+					this is a pointer to a null
+					character */
+	ib_int64_t	mysql_master_log_pos;
+					/* if the database server is a MySQL
+					replication slave, this is the
+					position in the log file up to which
+					replication has processed */
+	const char*	mysql_relay_log_file_name;
+	ib_int64_t	mysql_relay_log_pos;
+
+	os_thread_id_t	mysql_thread_id;/* id of the MySQL thread associated
+					with this transaction object */
+	ulint		mysql_process_no;/* since in Linux, 'top' reports
+					process id's and not thread id's, we
+					store the process number too */
+	/*------------------------------*/
+	ulint		n_mysql_tables_in_use; /* number of Innobase tables
+					used in the processing of the current
+					SQL statement in MySQL */
+	ulint		mysql_n_tables_locked;
+					/* how many tables the current SQL
+					statement uses, except those
+					in consistent read */
+	ulint		search_latch_timeout;
+					/* If we notice that someone is
+					waiting for our S-lock on the search
+					latch to be released, we wait in
+					row0sel.c for BTR_SEA_TIMEOUT new
+					searches until we try to keep
+					the search latch again over
+					calls from MySQL; this is intended
+					to reduce contention on the search
+					latch */
+	/*------------------------------*/
+	ulint		n_tickets_to_enter_innodb;
+					/* this can be > 0 only when
+					declared_to_... is TRUE; when we come
+					to srv_conc_innodb_enter, if the value
+					here is > 0, we decrement this by 1 */
+	/*------------------------------*/
+	dict_index_t*	new_rec_locks[2];/* these are normally NULL; if
+					srv_locks_unsafe_for_binlog is TRUE
+					or session is using READ COMMITTED
+					isolation level,
+					in a cursor search, if we set a new
+					record lock on an index, this is set
+					to point to the index; this is
+					used in releasing the locks under the
+					cursors if we are performing an UPDATE
+					and we determine after retrieving
+					the row that it does not need to be
+					locked; thus, these can be used to
+					implement a 'mini-rollback' that
+					releases the latest record locks */
+	UT_LIST_NODE_T(trx_t)
+			trx_list;	/* list of transactions */
+	UT_LIST_NODE_T(trx_t)
+			mysql_trx_list;	/* list of transactions created for
+					MySQL */
+	/*------------------------------*/
+	ulint		error_state;	/* 0 if no error, otherwise error
+					number; NOTE That ONLY the thread
+					doing the transaction is allowed to
+					set this field: this is NOT protected
+					by the kernel mutex */
+	const dict_index_t*error_info;	/* if the error number indicates a
+					duplicate key error, a pointer to
+					the problematic index is stored here */
+	ulint		error_key_num;	/* if the index creation fails to a
+					duplicate key error, a mysql key
+					number of that index is stored here */
+	sess_t*		sess;		/* session of the trx, NULL if none */
+	que_t*		graph;		/* query currently run in the session,
+					or NULL if none; NOTE that the query
+					belongs to the session, and it can
+					survive over a transaction commit, if
+					it is a stored procedure with a COMMIT
+					WORK statement, for instance */
+	ulint		n_active_thrs;	/* number of active query threads */
+	que_t*		graph_before_signal_handling;
+					/* value of graph when signal handling
+					for this trx started: this is used to
+					return control to the original query
+					graph for error processing */
+	trx_sig_t	sig;		/* one signal object can be allocated
+					in this space, avoiding mem_alloc */
+	UT_LIST_BASE_NODE_T(trx_sig_t)
+			signals;	/* queue of processed or pending
+					signals to the trx */
+	UT_LIST_BASE_NODE_T(trx_sig_t)
+			reply_signals;	/* list of signals sent by the query
+					threads of this trx for which a thread
+					is waiting for a reply; if this trx is
+					killed, the reply requests in the list
+					must be canceled */
+	/*------------------------------*/
+	lock_t*		wait_lock;	/* if trx execution state is
+					TRX_QUE_LOCK_WAIT, this points to
+					the lock request, otherwise this is
+					NULL */
+	ibool		was_chosen_as_deadlock_victim;
+					/* when the transaction decides to wait
+					for a lock, it sets this to FALSE;
+					if another transaction chooses this
+					transaction as a victim in deadlock
+					resolution, it sets this to TRUE */
+	time_t		wait_started;	/* lock wait started at this time */
+	UT_LIST_BASE_NODE_T(que_thr_t)
+			wait_thrs;	/* query threads belonging to this
+					trx that are in the QUE_THR_LOCK_WAIT
+					state */
+	ulint		deadlock_mark;	/* a mark field used in deadlock
+					checking algorithm.  This must be
+					in its own machine word, because
+					it can be changed by other
+					threads while holding kernel_mutex. */
+	/*------------------------------*/
+	mem_heap_t*	lock_heap;	/* memory heap for the locks of the
+					transaction */
+	UT_LIST_BASE_NODE_T(lock_t)
+			trx_locks;	/* locks reserved by the transaction */
+	/*------------------------------*/
+	mem_heap_t*	global_read_view_heap;
+					/* memory heap for the global read
+					view */
+	read_view_t*	global_read_view;
+					/* consistent read view associated
+					to a transaction or NULL */
+	read_view_t*	read_view;	/* consistent read view used in the
+					transaction or NULL, this read view
+					if defined can be normal read view
+					associated to a transaction (i.e.
+					same as global_read_view) or read view
+					associated to a cursor */
+	/*------------------------------*/
+	UT_LIST_BASE_NODE_T(trx_named_savept_t)
+			trx_savepoints;	/* savepoints set with SAVEPOINT ...,
+					oldest first */
+	/*------------------------------*/
+	mutex_t		undo_mutex;	/* mutex protecting the fields in this
+					section (down to undo_no_arr), EXCEPT
+					last_sql_stat_start, which can be
+					accessed only when we know that there
+					cannot be any activity in the undo
+					logs! */
+	dulint		undo_no;	/* next undo log record number to
+					assign; since the undo log is
+					private for a transaction, this
+					is a simple ascending sequence
+					with no gaps; thus it represents
+					the number of modified/inserted
+					rows in a transaction */
+	trx_savept_t	last_sql_stat_start;
+					/* undo_no when the last sql statement
+					was started: in case of an error, trx
+					is rolled back down to this undo
+					number; see note at undo_mutex! */
+	trx_rseg_t*	rseg;		/* rollback segment assigned to the
+					transaction, or NULL if not assigned
+					yet */
+	trx_undo_t*	insert_undo;	/* pointer to the insert undo log, or
+					NULL if no inserts performed yet */
+	trx_undo_t*	update_undo;	/* pointer to the update undo log, or
+					NULL if no update performed yet */
+	dulint		roll_limit;	/* least undo number to undo during
+					a rollback */
+	ulint		pages_undone;	/* number of undo log pages undone
+					since the last undo log truncation */
+	trx_undo_arr_t*	undo_no_arr;	/* array of undo numbers of undo log
+					records which are currently processed
+					by a rollback operation */
+	/*------------------------------*/
+	ulint		n_autoinc_rows;	/* no. of AUTO-INC rows required for
+					an SQL statement. This is useful for
+					multi-row INSERTs */
+	ib_vector_t*    autoinc_locks;  /* AUTOINC locks held by this
+					transaction. Note that these are
+					also in the lock list trx_locks. This
+					vector needs to be freed explicitly
+					when the trx_t instance is desrtoyed */
+	/*------------------------------*/
+	char detailed_error[256];	/* detailed error message for last
+					error, or empty. */
+};
+
+#define TRX_MAX_N_THREADS	32	/* maximum number of
+					concurrent threads running a
+					single operation of a
+					transaction, e.g., a parallel
+					query */
+/* Transaction concurrency states (trx->conc_state) */
+#define	TRX_NOT_STARTED		0
+#define	TRX_ACTIVE		1
+#define	TRX_COMMITTED_IN_MEMORY	2
+#define	TRX_PREPARED		3	/* Support for 2PC/XA */
+
+/* Transaction execution states when trx->conc_state == TRX_ACTIVE */
+#define TRX_QUE_RUNNING		0	/* transaction is running */
+#define TRX_QUE_LOCK_WAIT	1	/* transaction is waiting for a lock */
+#define TRX_QUE_ROLLING_BACK	2	/* transaction is rolling back */
+#define TRX_QUE_COMMITTING	3	/* transaction is committing */
+
+/* Transaction isolation levels (trx->isolation_level) */
+#define TRX_ISO_READ_UNCOMMITTED	0	/* dirty read: non-locking
+						SELECTs are performed so that
+						we do not look at a possible
+						earlier version of a record;
+						thus they are not 'consistent'
+						reads under this isolation
+						level; otherwise like level
+						2 */
+
+#define TRX_ISO_READ_COMMITTED		1	/* somewhat Oracle-like
+						isolation, except that in
+						range UPDATE and DELETE we
+						must block phantom rows
+						with next-key locks;
+						SELECT ... FOR UPDATE and ...
+						LOCK IN SHARE MODE only lock
+						the index records, NOT the
+						gaps before them, and thus
+						allow free inserting;
+						each consistent read reads its
+						own snapshot */
+
+#define TRX_ISO_REPEATABLE_READ		2	/* this is the default;
+						all consistent reads in the
+						same trx read the same
+						snapshot;
+						full next-key locking used
+						in locking reads to block
+						insertions into gaps */
+
+#define TRX_ISO_SERIALIZABLE		3	/* all plain SELECTs are
+						converted to LOCK IN SHARE
+						MODE reads */
+
+/* Treatment of duplicate values (trx->duplicates; for example, in inserts).
+Multiple flags can be combined with bitwise OR. */
+#define TRX_DUP_IGNORE	1	/* duplicate rows are to be updated */
+#define TRX_DUP_REPLACE	2	/* duplicate rows are to be replaced */
+
+
+/* Types of a trx signal */
+#define TRX_SIG_NO_SIGNAL		0
+#define TRX_SIG_TOTAL_ROLLBACK		1
+#define TRX_SIG_ROLLBACK_TO_SAVEPT	2
+#define TRX_SIG_COMMIT			3
+#define	TRX_SIG_ERROR_OCCURRED		4
+#define TRX_SIG_BREAK_EXECUTION		5
+
+/* Sender types of a signal */
+#define TRX_SIG_SELF		0	/* sent by the session itself, or
+					by an error occurring within this
+					session */
+#define TRX_SIG_OTHER_SESS	1	/* sent by another session (which
+					must hold rights to this) */
+
+/* Commit command node in a query graph */
+struct commit_node_struct{
+	que_common_t	common;	/* node type: QUE_NODE_COMMIT */
+	ulint		state;	/* node execution state */
+};
+
+/* Commit node states */
+#define COMMIT_NODE_SEND	1
+#define COMMIT_NODE_WAIT	2
+
+
+#ifndef UNIV_NONINL
+#include "trx0trx.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/trx0trx.ic b/storage/xtradb/include/trx0trx.ic
new file mode 100644
index 00000000000..6da89f002fe
--- /dev/null
+++ b/storage/xtradb/include/trx0trx.ic
@@ -0,0 +1,221 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The transaction
+
+Created 3/26/1996 Heikki Tuuri
+*******************************************************/
+
+/*****************************************************************
+Starts the transaction if it is not yet started. */
+UNIV_INLINE
+void
+trx_start_if_not_started(
+/*=====================*/
+	trx_t*	trx)	/* in: transaction */
+{
+	ut_ad(trx->conc_state != TRX_COMMITTED_IN_MEMORY);
+
+	if (trx->conc_state == TRX_NOT_STARTED) {
+
+		trx_start(trx, ULINT_UNDEFINED);
+	}
+}
+
+/*****************************************************************
+Starts the transaction if it is not yet started. Assumes we have reserved
+the kernel mutex! */
+UNIV_INLINE
+void
+trx_start_if_not_started_low(
+/*=========================*/
+	trx_t*	trx)	/* in: transaction */
+{
+	ut_ad(trx->conc_state != TRX_COMMITTED_IN_MEMORY);
+
+	if (trx->conc_state == TRX_NOT_STARTED) {
+
+		trx_start_low(trx, ULINT_UNDEFINED);
+	}
+}
+
+/*****************************************************************
+Resets the new record lock info in a transaction struct. */
+UNIV_INLINE
+void
+trx_reset_new_rec_lock_info(
+/*========================*/
+	trx_t*	trx)	/* in: transaction struct */
+{
+	trx->new_rec_locks[0] = NULL;
+	trx->new_rec_locks[1] = NULL;
+}
+
+/*****************************************************************
+Registers that we have set a new record lock on an index. We only have space
+to store 2 indexes! If this is called to store more than 2 indexes after
+trx_reset_new_rec_lock_info(), then this function does nothing. */
+UNIV_INLINE
+void
+trx_register_new_rec_lock(
+/*======================*/
+	trx_t*		trx,	/* in: transaction struct */
+	dict_index_t*	index)	/* in: trx sets a new record lock on this
+				index */
+{
+	if (trx->new_rec_locks[0] == NULL) {
+		trx->new_rec_locks[0] = index;
+
+		return;
+	}
+
+	if (trx->new_rec_locks[0] == index) {
+
+		return;
+	}
+
+	if (trx->new_rec_locks[1] != NULL) {
+
+		return;
+	}
+
+	trx->new_rec_locks[1] = index;
+}
+
+/*****************************************************************
+Checks if trx has set a new record lock on an index. */
+UNIV_INLINE
+ibool
+trx_new_rec_locks_contain(
+/*======================*/
+				/* out: TRUE if trx has set a new record lock
+				on index */
+	trx_t*		trx,	/* in: transaction struct */
+	dict_index_t*	index)	/* in: index */
+{
+	return(trx->new_rec_locks[0] == index
+	       || trx->new_rec_locks[1] == index);
+}
+
+/********************************************************************
+Retrieves the error_info field from a trx. */
+UNIV_INLINE
+const dict_index_t*
+trx_get_error_info(
+/*===============*/
+				/* out: the error info */
+	const trx_t*	trx)	/* in: trx object */
+{
+	return(trx->error_info);
+}
+
+/***********************************************************************
+Retrieves transacion's id, represented as unsigned long long. */
+UNIV_INLINE
+ullint
+trx_get_id(
+/*=======*/
+				/* out: transaction's id */
+	const trx_t*	trx)	/* in: transaction */
+{
+	return((ullint)ut_conv_dulint_to_longlong(trx->id));
+}
+
+/***********************************************************************
+Retrieves transaction's que state in a human readable string. The string
+should not be free()'d or modified. */
+UNIV_INLINE
+const char*
+trx_get_que_state_str(
+/*==================*/
+				/* out: string in the data segment */
+	const trx_t*	trx)	/* in: transaction */
+{
+	/* be sure to adjust TRX_QUE_STATE_STR_MAX_LEN if you change this */
+	switch (trx->que_state) {
+	case TRX_QUE_RUNNING:
+		return("RUNNING");
+	case TRX_QUE_LOCK_WAIT:
+		return("LOCK WAIT");
+	case TRX_QUE_ROLLING_BACK:
+		return("ROLLING BACK");
+	case TRX_QUE_COMMITTING:
+		return("COMMITTING");
+	default:
+		return("UNKNOWN");
+	}
+}
+
+/**************************************************************************
+Determine if a transaction is a dictionary operation. */
+UNIV_INLINE
+enum trx_dict_op
+trx_get_dict_operation(
+/*===================*/
+				/* out: dictionary operation mode */
+	const trx_t*	trx)	/* in: transaction */
+{
+	enum trx_dict_op op = (enum trx_dict_op) trx->dict_operation;
+
+#ifdef UNIV_DEBUG
+	switch (op) {
+	case TRX_DICT_OP_NONE:
+	case TRX_DICT_OP_TABLE:
+	case TRX_DICT_OP_INDEX:
+		return(op);
+	}
+	ut_error;
+#endif /* UNIV_DEBUG */
+	return((enum trx_dict_op) UNIV_EXPECT(op, TRX_DICT_OP_NONE));
+}
+/**************************************************************************
+Flag a transaction a dictionary operation. */
+UNIV_INLINE
+void
+trx_set_dict_operation(
+/*===================*/
+	trx_t*			trx,	/* in/out: transaction */
+	enum trx_dict_op	op)	/* in: operation, not
+					TRX_DICT_OP_NONE */
+{
+#ifdef UNIV_DEBUG
+	enum trx_dict_op	old_op = trx_get_dict_operation(trx);
+
+	switch (op) {
+	case TRX_DICT_OP_NONE:
+		ut_error;
+		break;
+	case TRX_DICT_OP_TABLE:
+		switch (old_op) {
+		case TRX_DICT_OP_NONE:
+		case TRX_DICT_OP_INDEX:
+		case TRX_DICT_OP_TABLE:
+			goto ok;
+		}
+		ut_error;
+		break;
+	case TRX_DICT_OP_INDEX:
+		ut_ad(old_op == TRX_DICT_OP_NONE);
+		break;
+	}
+ok:
+#endif /* UNIV_DEBUG */
+
+	trx->dict_operation = op;
+}
diff --git a/storage/xtradb/include/trx0types.h b/storage/xtradb/include/trx0types.h
new file mode 100644
index 00000000000..896f4e8c0a2
--- /dev/null
+++ b/storage/xtradb/include/trx0types.h
@@ -0,0 +1,78 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Transaction system global type definitions
+
+Created 3/26/1996 Heikki Tuuri
+*******************************************************/
+
+#ifndef trx0types_h
+#define trx0types_h
+
+#include "ut0byte.h"
+
+/* prepare trx_t::id for being printed via printf(3) */
+#define TRX_ID_PREP_PRINTF(id)	(ullint) ut_conv_dulint_to_longlong(id)
+
+/* printf(3) format used for printing TRX_ID_PRINTF_PREP() */
+#define TRX_ID_FMT		"%llX"
+
+/* maximum length that a formatted trx_t::id could take, not including
+the terminating '\0'. */
+#define TRX_ID_MAX_LEN		17
+
+/* Memory objects */
+typedef struct trx_struct	trx_t;
+typedef struct trx_sys_struct	trx_sys_t;
+typedef struct trx_doublewrite_struct	trx_doublewrite_t;
+typedef struct trx_sig_struct	trx_sig_t;
+typedef struct trx_rseg_struct	trx_rseg_t;
+typedef struct trx_undo_struct	trx_undo_t;
+typedef struct trx_undo_arr_struct trx_undo_arr_t;
+typedef struct trx_undo_inf_struct trx_undo_inf_t;
+typedef struct trx_purge_struct	trx_purge_t;
+typedef struct roll_node_struct	roll_node_t;
+typedef struct commit_node_struct commit_node_t;
+typedef struct trx_named_savept_struct trx_named_savept_t;
+
+/* Rollback contexts */
+enum trx_rb_ctx {
+	RB_NONE = 0,	/* no rollback */
+	RB_NORMAL,	/* normal rollback */
+	RB_RECOVERY,	/* rolling back an incomplete transaction,
+			in crash recovery */
+};
+
+/* Transaction savepoint */
+typedef struct trx_savept_struct trx_savept_t;
+struct trx_savept_struct{
+	dulint	least_undo_no;	/* least undo number to undo */
+};
+
+/* File objects */
+typedef byte	trx_sysf_t;
+typedef byte	trx_rsegf_t;
+typedef byte	trx_usegf_t;
+typedef byte	trx_ulogf_t;
+typedef byte	trx_upagef_t;
+
+/* Undo log record */
+typedef	byte	trx_undo_rec_t;
+
+#endif
diff --git a/storage/xtradb/include/trx0undo.h b/storage/xtradb/include/trx0undo.h
new file mode 100644
index 00000000000..7f7408931da
--- /dev/null
+++ b/storage/xtradb/include/trx0undo.h
@@ -0,0 +1,526 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Transaction undo log
+
+Created 3/26/1996 Heikki Tuuri
+*******************************************************/
+
+#ifndef trx0undo_h
+#define trx0undo_h
+
+#include "univ.i"
+#include "trx0types.h"
+#include "mtr0mtr.h"
+#include "trx0sys.h"
+#include "page0types.h"
+#include "trx0xa.h"
+
+/***************************************************************************
+Builds a roll pointer dulint. */
+UNIV_INLINE
+dulint
+trx_undo_build_roll_ptr(
+/*====================*/
+				/* out: roll pointer */
+	ibool	is_insert,	/* in: TRUE if insert undo log */
+	ulint	rseg_id,	/* in: rollback segment id */
+	ulint	page_no,	/* in: page number */
+	ulint	offset);	/* in: offset of the undo entry within page */
+/***************************************************************************
+Decodes a roll pointer dulint. */
+UNIV_INLINE
+void
+trx_undo_decode_roll_ptr(
+/*=====================*/
+	dulint	roll_ptr,	/* in: roll pointer */
+	ibool*	is_insert,	/* out: TRUE if insert undo log */
+	ulint*	rseg_id,	/* out: rollback segment id */
+	ulint*	page_no,	/* out: page number */
+	ulint*	offset);	/* out: offset of the undo entry within page */
+/***************************************************************************
+Returns TRUE if the roll pointer is of the insert type. */
+UNIV_INLINE
+ibool
+trx_undo_roll_ptr_is_insert(
+/*========================*/
+				/* out: TRUE if insert undo log */
+	dulint	roll_ptr);	/* in: roll pointer */
+/*********************************************************************
+Writes a roll ptr to an index page. In case that the size changes in
+some future version, this function should be used instead of
+mach_write_... */
+UNIV_INLINE
+void
+trx_write_roll_ptr(
+/*===============*/
+	byte*	ptr,		/* in: pointer to memory where written */
+	dulint	roll_ptr);	/* in: roll ptr */
+/*********************************************************************
+Reads a roll ptr from an index page. In case that the roll ptr size
+changes in some future version, this function should be used instead of
+mach_read_... */
+UNIV_INLINE
+dulint
+trx_read_roll_ptr(
+/*==============*/
+				/* out: roll ptr */
+	const byte*	ptr);	/* in: pointer to memory from where to read */
+/**********************************************************************
+Gets an undo log page and x-latches it. */
+UNIV_INLINE
+page_t*
+trx_undo_page_get(
+/*==============*/
+				/* out: pointer to page x-latched */
+	ulint	space,		/* in: space where placed */
+	ulint	zip_size,	/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	ulint	page_no,	/* in: page number */
+	mtr_t*	mtr);		/* in: mtr */
+/**********************************************************************
+Gets an undo log page and s-latches it. */
+UNIV_INLINE
+page_t*
+trx_undo_page_get_s_latched(
+/*========================*/
+				/* out: pointer to page s-latched */
+	ulint	space,		/* in: space where placed */
+	ulint	zip_size,	/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	ulint	page_no,	/* in: page number */
+	mtr_t*	mtr);		/* in: mtr */
+/**********************************************************************
+Returns the previous undo record on the page in the specified log, or
+NULL if none exists. */
+UNIV_INLINE
+trx_undo_rec_t*
+trx_undo_page_get_prev_rec(
+/*=======================*/
+				/* out: pointer to record, NULL if none */
+	trx_undo_rec_t*	rec,	/* in: undo log record */
+	ulint		page_no,/* in: undo log header page number */
+	ulint		offset);/* in: undo log header offset on page */
+/**********************************************************************
+Returns the next undo log record on the page in the specified log, or
+NULL if none exists. */
+UNIV_INLINE
+trx_undo_rec_t*
+trx_undo_page_get_next_rec(
+/*=======================*/
+				/* out: pointer to record, NULL if none */
+	trx_undo_rec_t*	rec,	/* in: undo log record */
+	ulint		page_no,/* in: undo log header page number */
+	ulint		offset);/* in: undo log header offset on page */
+/**********************************************************************
+Returns the last undo record on the page in the specified undo log, or
+NULL if none exists. */
+UNIV_INLINE
+trx_undo_rec_t*
+trx_undo_page_get_last_rec(
+/*=======================*/
+			/* out: pointer to record, NULL if none */
+	page_t*	undo_page,/* in: undo log page */
+	ulint	page_no,/* in: undo log header page number */
+	ulint	offset);	/* in: undo log header offset on page */
+/**********************************************************************
+Returns the first undo record on the page in the specified undo log, or
+NULL if none exists. */
+UNIV_INLINE
+trx_undo_rec_t*
+trx_undo_page_get_first_rec(
+/*========================*/
+			/* out: pointer to record, NULL if none */
+	page_t*	undo_page,/* in: undo log page */
+	ulint	page_no,/* in: undo log header page number */
+	ulint	offset);/* in: undo log header offset on page */
+/***************************************************************************
+Gets the previous record in an undo log. */
+UNIV_INTERN
+trx_undo_rec_t*
+trx_undo_get_prev_rec(
+/*==================*/
+				/* out: undo log record, the page s-latched,
+				NULL if none */
+	trx_undo_rec_t*	rec,	/* in: undo record */
+	ulint		page_no,/* in: undo log header page number */
+	ulint		offset,	/* in: undo log header offset on page */
+	mtr_t*		mtr);	/* in: mtr */
+/***************************************************************************
+Gets the next record in an undo log. */
+UNIV_INTERN
+trx_undo_rec_t*
+trx_undo_get_next_rec(
+/*==================*/
+				/* out: undo log record, the page s-latched,
+				NULL if none */
+	trx_undo_rec_t*	rec,	/* in: undo record */
+	ulint		page_no,/* in: undo log header page number */
+	ulint		offset,	/* in: undo log header offset on page */
+	mtr_t*		mtr);	/* in: mtr */
+/***************************************************************************
+Gets the first record in an undo log. */
+UNIV_INTERN
+trx_undo_rec_t*
+trx_undo_get_first_rec(
+/*===================*/
+			/* out: undo log record, the page latched, NULL if
+			none */
+	ulint	space,	/* in: undo log header space */
+	ulint	zip_size,/* in: compressed page size in bytes
+			or 0 for uncompressed pages */
+	ulint	page_no,/* in: undo log header page number */
+	ulint	offset,	/* in: undo log header offset on page */
+	ulint	mode,	/* in: latching mode: RW_S_LATCH or RW_X_LATCH */
+	mtr_t*	mtr);	/* in: mtr */
+/************************************************************************
+Tries to add a page to the undo log segment where the undo log is placed. */
+UNIV_INTERN
+ulint
+trx_undo_add_page(
+/*==============*/
+				/* out: page number if success, else
+				FIL_NULL */
+	trx_t*		trx,	/* in: transaction */
+	trx_undo_t*	undo,	/* in: undo log memory object */
+	mtr_t*		mtr);	/* in: mtr which does not have a latch to any
+				undo log page; the caller must have reserved
+				the rollback segment mutex */
+/***************************************************************************
+Truncates an undo log from the end. This function is used during a rollback
+to free space from an undo log. */
+UNIV_INTERN
+void
+trx_undo_truncate_end(
+/*==================*/
+	trx_t*		trx,	/* in: transaction whose undo log it is */
+	trx_undo_t*	undo,	/* in: undo log */
+	dulint		limit);	/* in: all undo records with undo number
+				>= this value should be truncated */
+/***************************************************************************
+Truncates an undo log from the start. This function is used during a purge
+operation. */
+UNIV_INTERN
+void
+trx_undo_truncate_start(
+/*====================*/
+	trx_rseg_t* rseg,	/* in: rollback segment */
+	ulint	space,		/* in: space id of the log */
+	ulint	hdr_page_no,	/* in: header page number */
+	ulint	hdr_offset,	/* in: header offset on the page */
+	dulint	limit);		/* in: all undo pages with undo numbers <
+				this value should be truncated; NOTE that
+				the function only frees whole pages; the
+				header page is not freed, but emptied, if
+				all the records there are < limit */
+/************************************************************************
+Initializes the undo log lists for a rollback segment memory copy.
+This function is only called when the database is started or a new
+rollback segment created. */
+UNIV_INTERN
+ulint
+trx_undo_lists_init(
+/*================*/
+				/* out: the combined size of undo log segments
+				in pages */
+	trx_rseg_t*	rseg);	/* in: rollback segment memory object */
+/**************************************************************************
+Assigns an undo log for a transaction. A new undo log is created or a cached
+undo log reused. */
+UNIV_INTERN
+ulint
+trx_undo_assign_undo(
+/*=================*/
+				/* out: DB_SUCCESS if undo log assign
+				successful, possible error codes are:
+				DB_TOO_MANY_CONCURRENT_TRXS
+				DB_OUT_OF_FILE_SPACE DB_OUT_OF_MEMORY*/
+	trx_t*		trx,	/* in: transaction */
+	ulint		type);	/* in: TRX_UNDO_INSERT or TRX_UNDO_UPDATE */
+/**********************************************************************
+Sets the state of the undo log segment at a transaction finish. */
+UNIV_INTERN
+page_t*
+trx_undo_set_state_at_finish(
+/*=========================*/
+				/* out: undo log segment header page,
+				x-latched */
+	trx_rseg_t*	rseg,	/* in: rollback segment memory object */
+	trx_t*		trx,	/* in: transaction */
+	trx_undo_t*	undo,	/* in: undo log memory copy */
+	mtr_t*		mtr);	/* in: mtr */
+/**********************************************************************
+Sets the state of the undo log segment at a transaction prepare. */
+UNIV_INTERN
+page_t*
+trx_undo_set_state_at_prepare(
+/*==========================*/
+				/* out: undo log segment header page,
+				x-latched */
+	trx_t*		trx,	/* in: transaction */
+	trx_undo_t*	undo,	/* in: undo log memory copy */
+	mtr_t*		mtr);	/* in: mtr */
+
+/**************************************************************************
+Adds the update undo log header as the first in the history list, and
+frees the memory object, or puts it to the list of cached update undo log
+segments. */
+UNIV_INTERN
+void
+trx_undo_update_cleanup(
+/*====================*/
+	trx_t*	trx,		/* in: trx owning the update undo log */
+	page_t*	undo_page,	/* in: update undo log header page,
+				x-latched */
+	mtr_t*	mtr);		/* in: mtr */
+/**********************************************************************
+Frees or caches an insert undo log after a transaction commit or rollback.
+Knowledge of inserts is not needed after a commit or rollback, therefore
+the data can be discarded. */
+UNIV_INTERN
+void
+trx_undo_insert_cleanup(
+/*====================*/
+	trx_t*	trx);	/* in: transaction handle */
+/***************************************************************
+Parses the redo log entry of an undo log page initialization. */
+UNIV_INTERN
+byte*
+trx_undo_parse_page_init(
+/*=====================*/
+			/* out: end of log record or NULL */
+	byte*	ptr,	/* in: buffer */
+	byte*	end_ptr,/* in: buffer end */
+	page_t*	page,	/* in: page or NULL */
+	mtr_t*	mtr);	/* in: mtr or NULL */
+/***************************************************************
+Parses the redo log entry of an undo log page header create or reuse. */
+UNIV_INTERN
+byte*
+trx_undo_parse_page_header(
+/*=======================*/
+			/* out: end of log record or NULL */
+	ulint	type,	/* in: MLOG_UNDO_HDR_CREATE or MLOG_UNDO_HDR_REUSE */
+	byte*	ptr,	/* in: buffer */
+	byte*	end_ptr,/* in: buffer end */
+	page_t*	page,	/* in: page or NULL */
+	mtr_t*	mtr);	/* in: mtr or NULL */
+/***************************************************************
+Parses the redo log entry of an undo log page header discard. */
+UNIV_INTERN
+byte*
+trx_undo_parse_discard_latest(
+/*==========================*/
+			/* out: end of log record or NULL */
+	byte*	ptr,	/* in: buffer */
+	byte*	end_ptr,/* in: buffer end */
+	page_t*	page,	/* in: page or NULL */
+	mtr_t*	mtr);	/* in: mtr or NULL */
+
+/* Types of an undo log segment */
+#define	TRX_UNDO_INSERT		1	/* contains undo entries for inserts */
+#define	TRX_UNDO_UPDATE		2	/* contains undo entries for updates
+					and delete markings: in short,
+					modifys (the name 'UPDATE' is a
+					historical relic) */
+/* States of an undo log segment */
+#define TRX_UNDO_ACTIVE		1	/* contains an undo log of an active
+					transaction */
+#define	TRX_UNDO_CACHED		2	/* cached for quick reuse */
+#define	TRX_UNDO_TO_FREE	3	/* insert undo segment can be freed */
+#define	TRX_UNDO_TO_PURGE	4	/* update undo segment will not be
+					reused: it can be freed in purge when
+					all undo data in it is removed */
+#define	TRX_UNDO_PREPARED	5	/* contains an undo log of an
+					prepared transaction */
+
+/* Transaction undo log memory object; this is protected by the undo_mutex
+in the corresponding transaction object */
+
+struct trx_undo_struct{
+	/*-----------------------------*/
+	ulint		id;		/* undo log slot number within the
+					rollback segment */
+	ulint		type;		/* TRX_UNDO_INSERT or
+					TRX_UNDO_UPDATE */
+	ulint		state;		/* state of the corresponding undo log
+					segment */
+	ibool		del_marks;	/* relevant only in an update undo log:
+					this is TRUE if the transaction may
+					have delete marked records, because of
+					a delete of a row or an update of an
+					indexed field; purge is then
+					necessary; also TRUE if the transaction
+					has updated an externally stored
+					field */
+	dulint		trx_id;		/* id of the trx assigned to the undo
+					log */
+	XID		xid;		/* X/Open XA transaction
+					identification */
+	ibool		dict_operation;	/* TRUE if a dict operation trx */
+	dulint		table_id;	/* if a dict operation, then the table
+					id */
+	trx_rseg_t*	rseg;		/* rseg where the undo log belongs */
+	/*-----------------------------*/
+	ulint		space;		/* space id where the undo log
+					placed */
+	ulint		zip_size;	/* in: compressed page size of space
+					in bytes, or 0 for uncompressed */
+	ulint		hdr_page_no;	/* page number of the header page in
+					the undo log */
+	ulint		hdr_offset;	/* header offset of the undo log on the
+					page */
+	ulint		last_page_no;	/* page number of the last page in the
+					undo log; this may differ from
+					top_page_no during a rollback */
+	ulint		size;		/* current size in pages */
+	/*-----------------------------*/
+	ulint		empty;		/* TRUE if the stack of undo log
+					records is currently empty */
+	ulint		top_page_no;	/* page number where the latest undo
+					log record was catenated; during
+					rollback the page from which the latest
+					undo record was chosen */
+	ulint		top_offset;	/* offset of the latest undo record,
+					i.e., the topmost element in the undo
+					log if we think of it as a stack */
+	dulint		top_undo_no;	/* undo number of the latest record */
+	buf_block_t*	guess_block;	/* guess for the buffer block where
+					the top page might reside */
+	/*-----------------------------*/
+	UT_LIST_NODE_T(trx_undo_t) undo_list;
+					/* undo log objects in the rollback
+					segment are chained into lists */
+};
+
+/* The offset of the undo log page header on pages of the undo log */
+#define	TRX_UNDO_PAGE_HDR	FSEG_PAGE_DATA
+/*-------------------------------------------------------------*/
+/* Transaction undo log page header offsets */
+#define	TRX_UNDO_PAGE_TYPE	0	/* TRX_UNDO_INSERT or
+					TRX_UNDO_UPDATE */
+#define	TRX_UNDO_PAGE_START	2	/* Byte offset where the undo log
+					records for the LATEST transaction
+					start on this page (remember that
+					in an update undo log, the first page
+					can contain several undo logs) */
+#define	TRX_UNDO_PAGE_FREE	4	/* On each page of the undo log this
+					field contains the byte offset of the
+					first free byte on the page */
+#define TRX_UNDO_PAGE_NODE	6	/* The file list node in the chain
+					of undo log pages */
+/*-------------------------------------------------------------*/
+#define TRX_UNDO_PAGE_HDR_SIZE	(6 + FLST_NODE_SIZE)
+
+/* An update undo segment with just one page can be reused if it has
+< this number bytes used; we must leave space at least for one new undo
+log header on the page */
+
+#define TRX_UNDO_PAGE_REUSE_LIMIT	(3 * UNIV_PAGE_SIZE / 4)
+
+/* An update undo log segment may contain several undo logs on its first page
+if the undo logs took so little space that the segment could be cached and
+reused. All the undo log headers are then on the first page, and the last one
+owns the undo log records on subsequent pages if the segment is bigger than
+one page. If an undo log is stored in a segment, then on the first page it is
+allowed to have zero undo records, but if the segment extends to several
+pages, then all the rest of the pages must contain at least one undo log
+record. */
+
+/* The offset of the undo log segment header on the first page of the undo
+log segment */
+
+#define	TRX_UNDO_SEG_HDR	(TRX_UNDO_PAGE_HDR + TRX_UNDO_PAGE_HDR_SIZE)
+/*-------------------------------------------------------------*/
+#define	TRX_UNDO_STATE		0	/* TRX_UNDO_ACTIVE, ... */
+#define	TRX_UNDO_LAST_LOG	2	/* Offset of the last undo log header
+					on the segment header page, 0 if
+					none */
+#define	TRX_UNDO_FSEG_HEADER	4	/* Header for the file segment which
+					the undo log segment occupies */
+#define	TRX_UNDO_PAGE_LIST	(4 + FSEG_HEADER_SIZE)
+					/* Base node for the list of pages in
+					the undo log segment; defined only on
+					the undo log segment's first page */
+/*-------------------------------------------------------------*/
+/* Size of the undo log segment header */
+#define TRX_UNDO_SEG_HDR_SIZE	(4 + FSEG_HEADER_SIZE + FLST_BASE_NODE_SIZE)
+
+
+/* The undo log header. There can be several undo log headers on the first
+page of an update undo log segment. */
+/*-------------------------------------------------------------*/
+#define	TRX_UNDO_TRX_ID		0	/* Transaction id */
+#define	TRX_UNDO_TRX_NO		8	/* Transaction number of the
+					transaction; defined only if the log
+					is in a history list */
+#define TRX_UNDO_DEL_MARKS	16	/* Defined only in an update undo
+					log: TRUE if the transaction may have
+					done delete markings of records, and
+					thus purge is necessary */
+#define	TRX_UNDO_LOG_START	18	/* Offset of the first undo log record
+					of this log on the header page; purge
+					may remove undo log record from the
+					log start, and therefore this is not
+					necessarily the same as this log
+					header end offset */
+#define	TRX_UNDO_XID_EXISTS	20	/* TRUE if undo log header includes
+					X/Open XA transaction identification
+					XID */
+#define	TRX_UNDO_DICT_TRANS	21	/* TRUE if the transaction is a table
+					create, index create, or drop
+					transaction: in recovery
+					the transaction cannot be rolled back
+					in the usual way: a 'rollback' rather
+					means dropping the created or dropped
+					table, if it still exists */
+#define TRX_UNDO_TABLE_ID	22	/* Id of the table if the preceding
+					field is TRUE */
+#define	TRX_UNDO_NEXT_LOG	30	/* Offset of the next undo log header
+					on this page, 0 if none */
+#define	TRX_UNDO_PREV_LOG	32	/* Offset of the previous undo log
+					header on this page, 0 if none */
+#define TRX_UNDO_HISTORY_NODE	34	/* If the log is put to the history
+					list, the file list node is here */
+/*-------------------------------------------------------------*/
+#define TRX_UNDO_LOG_OLD_HDR_SIZE (34 + FLST_NODE_SIZE)
+
+/* Note: the writing of the undo log old header is coded by a log record
+MLOG_UNDO_HDR_CREATE or MLOG_UNDO_HDR_REUSE. The appending of an XID to the
+header is logged separately. In this sense, the XID is not really a member
+of the undo log header. TODO: do not append the XID to the log header if XA
+is not needed by the user. The XID wastes about 150 bytes of space in every
+undo log. In the history list we may have millions of undo logs, which means
+quite a large overhead. */
+
+/* X/Open XA Transaction Identification (XID) */
+
+#define	TRX_UNDO_XA_FORMAT	(TRX_UNDO_LOG_OLD_HDR_SIZE)
+#define	TRX_UNDO_XA_TRID_LEN	(TRX_UNDO_XA_FORMAT + 4)
+#define	TRX_UNDO_XA_BQUAL_LEN	(TRX_UNDO_XA_TRID_LEN + 4)
+#define	TRX_UNDO_XA_XID		(TRX_UNDO_XA_BQUAL_LEN + 4)
+/*--------------------------------------------------------------*/
+#define TRX_UNDO_LOG_XA_HDR_SIZE (TRX_UNDO_XA_XID + XIDDATASIZE)
+				/* Total size of the header with the XA XID */
+
+#ifndef UNIV_NONINL
+#include "trx0undo.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/trx0undo.ic b/storage/xtradb/include/trx0undo.ic
new file mode 100644
index 00000000000..0bd8b79414b
--- /dev/null
+++ b/storage/xtradb/include/trx0undo.ic
@@ -0,0 +1,344 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Transaction undo log
+
+Created 3/26/1996 Heikki Tuuri
+*******************************************************/
+
+#include "data0type.h"
+#include "page0page.h"
+
+/***************************************************************************
+Builds a roll pointer dulint. */
+UNIV_INLINE
+dulint
+trx_undo_build_roll_ptr(
+/*====================*/
+				/* out: roll pointer */
+	ibool	is_insert,	/* in: TRUE if insert undo log */
+	ulint	rseg_id,	/* in: rollback segment id */
+	ulint	page_no,	/* in: page number */
+	ulint	offset)		/* in: offset of the undo entry within page */
+{
+#if DATA_ROLL_PTR_LEN != 7
+# error "DATA_ROLL_PTR_LEN != 7"
+#endif
+	ut_ad(rseg_id < 128);
+
+	return(ut_dulint_create(is_insert * 128 * 256 * 256
+				+ rseg_id * 256 * 256
+				+ (page_no / 256) / 256,
+				(page_no % (256 * 256)) * 256 * 256
+				+ offset));
+}
+
+/***************************************************************************
+Decodes a roll pointer dulint. */
+UNIV_INLINE
+void
+trx_undo_decode_roll_ptr(
+/*=====================*/
+	dulint	roll_ptr,	/* in: roll pointer */
+	ibool*	is_insert,	/* out: TRUE if insert undo log */
+	ulint*	rseg_id,	/* out: rollback segment id */
+	ulint*	page_no,	/* out: page number */
+	ulint*	offset)		/* out: offset of the undo entry within page */
+{
+	ulint	low;
+	ulint	high;
+#if DATA_ROLL_PTR_LEN != 7
+# error "DATA_ROLL_PTR_LEN != 7"
+#endif
+#if TRUE != 1
+# error "TRUE != 1"
+#endif
+	high = ut_dulint_get_high(roll_ptr);
+	low = ut_dulint_get_low(roll_ptr);
+
+	*offset = low % (256 * 256);
+
+	*is_insert = high / (256 * 256 * 128);	/* TRUE == 1 */
+	*rseg_id = (high / (256 * 256)) % 128;
+
+	*page_no = (high % (256 * 256)) * 256 * 256
+		+ (low / 256) / 256;
+}
+
+/***************************************************************************
+Returns TRUE if the roll pointer is of the insert type. */
+UNIV_INLINE
+ibool
+trx_undo_roll_ptr_is_insert(
+/*========================*/
+				/* out: TRUE if insert undo log */
+	dulint	roll_ptr)	/* in: roll pointer */
+{
+	ulint	high;
+#if DATA_ROLL_PTR_LEN != 7
+# error "DATA_ROLL_PTR_LEN != 7"
+#endif
+#if TRUE != 1
+# error "TRUE != 1"
+#endif
+	high = ut_dulint_get_high(roll_ptr);
+
+	return(high / (256 * 256 * 128));
+}
+
+/*********************************************************************
+Writes a roll ptr to an index page. In case that the size changes in
+some future version, this function should be used instead of
+mach_write_... */
+UNIV_INLINE
+void
+trx_write_roll_ptr(
+/*===============*/
+	byte*	ptr,		/* in: pointer to memory where written */
+	dulint	roll_ptr)	/* in: roll ptr */
+{
+#if DATA_ROLL_PTR_LEN != 7
+# error "DATA_ROLL_PTR_LEN != 7"
+#endif
+	mach_write_to_7(ptr, roll_ptr);
+}
+
+/*********************************************************************
+Reads a roll ptr from an index page. In case that the roll ptr size
+changes in some future version, this function should be used instead of
+mach_read_... */
+UNIV_INLINE
+dulint
+trx_read_roll_ptr(
+/*==============*/
+				/* out: roll ptr */
+	const byte*	ptr)	/* in: pointer to memory from where to read */
+{
+#if DATA_ROLL_PTR_LEN != 7
+# error "DATA_ROLL_PTR_LEN != 7"
+#endif
+	return(mach_read_from_7(ptr));
+}
+
+/**********************************************************************
+Gets an undo log page and x-latches it. */
+UNIV_INLINE
+page_t*
+trx_undo_page_get(
+/*==============*/
+				/* out: pointer to page x-latched */
+	ulint	space,		/* in: space where placed */
+	ulint	zip_size,	/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	ulint	page_no,	/* in: page number */
+	mtr_t*	mtr)		/* in: mtr */
+{
+	buf_block_t*	block = buf_page_get(space, zip_size, page_no,
+					     RW_X_LATCH, mtr);
+	buf_block_dbg_add_level(block, SYNC_TRX_UNDO_PAGE);
+
+	return(buf_block_get_frame(block));
+}
+
+/**********************************************************************
+Gets an undo log page and s-latches it. */
+UNIV_INLINE
+page_t*
+trx_undo_page_get_s_latched(
+/*========================*/
+				/* out: pointer to page s-latched */
+	ulint	space,		/* in: space where placed */
+	ulint	zip_size,	/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	ulint	page_no,	/* in: page number */
+	mtr_t*	mtr)		/* in: mtr */
+{
+	buf_block_t*	block = buf_page_get(space, zip_size, page_no,
+					     RW_S_LATCH, mtr);
+	buf_block_dbg_add_level(block, SYNC_TRX_UNDO_PAGE);
+
+	return(buf_block_get_frame(block));
+}
+
+/**********************************************************************
+Returns the start offset of the undo log records of the specified undo
+log on the page. */
+UNIV_INLINE
+ulint
+trx_undo_page_get_start(
+/*====================*/
+			/* out: start offset */
+	page_t*	undo_page,/* in: undo log page */
+	ulint	page_no,/* in: undo log header page number */
+	ulint	offset)	/* in: undo log header offset on page */
+{
+	ulint	start;
+
+	if (page_no == page_get_page_no(undo_page)) {
+
+		start = mach_read_from_2(offset + undo_page
+					 + TRX_UNDO_LOG_START);
+	} else {
+		start = TRX_UNDO_PAGE_HDR + TRX_UNDO_PAGE_HDR_SIZE;
+	}
+
+	return(start);
+}
+
+/**********************************************************************
+Returns the end offset of the undo log records of the specified undo
+log on the page. */
+UNIV_INLINE
+ulint
+trx_undo_page_get_end(
+/*==================*/
+			/* out: end offset */
+	page_t*	undo_page,/* in: undo log page */
+	ulint	page_no,/* in: undo log header page number */
+	ulint	offset)	/* in: undo log header offset on page */
+{
+	trx_ulogf_t*	log_hdr;
+	ulint		end;
+
+	if (page_no == page_get_page_no(undo_page)) {
+
+		log_hdr = undo_page + offset;
+
+		end = mach_read_from_2(log_hdr + TRX_UNDO_NEXT_LOG);
+
+		if (end == 0) {
+			end = mach_read_from_2(undo_page + TRX_UNDO_PAGE_HDR
+					       + TRX_UNDO_PAGE_FREE);
+		}
+	} else {
+		end = mach_read_from_2(undo_page + TRX_UNDO_PAGE_HDR
+				       + TRX_UNDO_PAGE_FREE);
+	}
+
+	return(end);
+}
+
+/**********************************************************************
+Returns the previous undo record on the page in the specified log, or
+NULL if none exists. */
+UNIV_INLINE
+trx_undo_rec_t*
+trx_undo_page_get_prev_rec(
+/*=======================*/
+				/* out: pointer to record, NULL if none */
+	trx_undo_rec_t*	rec,	/* in: undo log record */
+	ulint		page_no,/* in: undo log header page number */
+	ulint		offset)	/* in: undo log header offset on page */
+{
+	page_t*	undo_page;
+	ulint	start;
+
+	undo_page = (page_t*) ut_align_down(rec, UNIV_PAGE_SIZE);
+
+	start = trx_undo_page_get_start(undo_page, page_no, offset);
+
+	if (start + undo_page == rec) {
+
+		return(NULL);
+	}
+
+	return(undo_page + mach_read_from_2(rec - 2));
+}
+
+/**********************************************************************
+Returns the next undo log record on the page in the specified log, or
+NULL if none exists. */
+UNIV_INLINE
+trx_undo_rec_t*
+trx_undo_page_get_next_rec(
+/*=======================*/
+				/* out: pointer to record, NULL if none */
+	trx_undo_rec_t*	rec,	/* in: undo log record */
+	ulint		page_no,/* in: undo log header page number */
+	ulint		offset)	/* in: undo log header offset on page */
+{
+	page_t*	undo_page;
+	ulint	end;
+	ulint	next;
+
+	undo_page = (page_t*) ut_align_down(rec, UNIV_PAGE_SIZE);
+
+	end = trx_undo_page_get_end(undo_page, page_no, offset);
+
+	next = mach_read_from_2(rec);
+
+	if (next == end) {
+
+		return(NULL);
+	}
+
+	return(undo_page + next);
+}
+
+/**********************************************************************
+Returns the last undo record on the page in the specified undo log, or
+NULL if none exists. */
+UNIV_INLINE
+trx_undo_rec_t*
+trx_undo_page_get_last_rec(
+/*=======================*/
+			/* out: pointer to record, NULL if none */
+	page_t*	undo_page,/* in: undo log page */
+	ulint	page_no,/* in: undo log header page number */
+	ulint	offset)	/* in: undo log header offset on page */
+{
+	ulint	start;
+	ulint	end;
+
+	start = trx_undo_page_get_start(undo_page, page_no, offset);
+	end = trx_undo_page_get_end(undo_page, page_no, offset);
+
+	if (start == end) {
+
+		return(NULL);
+	}
+
+	return(undo_page + mach_read_from_2(undo_page + end - 2));
+}
+
+/**********************************************************************
+Returns the first undo record on the page in the specified undo log, or
+NULL if none exists. */
+UNIV_INLINE
+trx_undo_rec_t*
+trx_undo_page_get_first_rec(
+/*========================*/
+			/* out: pointer to record, NULL if none */
+	page_t*	undo_page,/* in: undo log page */
+	ulint	page_no,/* in: undo log header page number */
+	ulint	offset)	/* in: undo log header offset on page */
+{
+	ulint	start;
+	ulint	end;
+
+	start = trx_undo_page_get_start(undo_page, page_no, offset);
+	end = trx_undo_page_get_end(undo_page, page_no, offset);
+
+	if (start == end) {
+
+		return(NULL);
+	}
+
+	return(undo_page + start);
+}
diff --git a/storage/xtradb/include/trx0xa.h b/storage/xtradb/include/trx0xa.h
new file mode 100644
index 00000000000..0e040b8d8e5
--- /dev/null
+++ b/storage/xtradb/include/trx0xa.h
@@ -0,0 +1,61 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/*
+ * Start of xa.h header
+ *
+ * Define a symbol to prevent multiple inclusions of this header file
+ */
+#ifndef	XA_H
+#define	XA_H
+
+/*
+ * Transaction branch identification: XID and NULLXID:
+ */
+#ifndef XIDDATASIZE
+
+#define	XIDDATASIZE	128		/* size in bytes */
+#define	MAXGTRIDSIZE	 64		/* maximum size in bytes of gtrid */
+#define	MAXBQUALSIZE	 64		/* maximum size in bytes of bqual */
+
+struct xid_t {
+	long formatID;			/* format identifier; -1
+					means that the XID is null */
+	long gtrid_length;		/* value from 1 through 64 */
+	long bqual_length;		/* value from 1 through 64 */
+	char data[XIDDATASIZE];
+};
+typedef	struct xid_t XID;
+#endif
+#define	XA_OK		0		/* normal execution */
+#define	XAER_ASYNC	-2		/* asynchronous operation already
+					outstanding */
+#define	XAER_RMERR	-3		/* a resource manager error occurred in
+					 the transaction branch */
+#define	XAER_NOTA	-4		/* the XID is not valid */
+#define	XAER_INVAL	-5		/* invalid arguments were given */
+#define	XAER_PROTO	-6		/* routine invoked in an improper
+					context */
+#define	XAER_RMFAIL	-7		/* resource manager unavailable */
+#define	XAER_DUPID	-8		/* the XID already exists */
+#define	XAER_OUTSIDE	-9		/* resource manager doing work outside
+					transaction */
+#endif /* ifndef XA_H */
+/*
+ * End of xa.h header
+ */
diff --git a/storage/xtradb/include/univ.i b/storage/xtradb/include/univ.i
new file mode 100644
index 00000000000..3b01401a0b9
--- /dev/null
+++ b/storage/xtradb/include/univ.i
@@ -0,0 +1,466 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+Copyright (c) 2008, Google Inc.
+
+Portions of this file contain modifications contributed and copyrighted by
+Google, Inc. Those modifications are gratefully acknowledged and are described
+briefly in the InnoDB documentation. The contributions by Google are
+incorporated with their permission, and subject to the conditions contained in
+the file COPYING.Google.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/***************************************************************************
+Version control for database, common definitions, and include files
+
+Created 1/20/1994 Heikki Tuuri
+****************************************************************************/
+
+#ifndef univ_i
+#define univ_i
+
+#define INNODB_VERSION_MAJOR	1
+#define INNODB_VERSION_MINOR	0
+#define INNODB_VERSION_BUGFIX	3
+#define PERCONA_INNODB_VERSION	5a
+
+/* The following is the InnoDB version as shown in
+SELECT plugin_version FROM information_schema.plugins;
+calculated in in make_version_string() in sql/sql_show.cc like this:
+"version >> 8" . "version & 0xff"
+because the version is shown with only one dot, we skip the last
+component, i.e. we show M.N.P as M.N */
+#define INNODB_VERSION_SHORT	\
+	(INNODB_VERSION_MAJOR << 8 | INNODB_VERSION_MINOR)
+
+/* auxiliary macros to help creating the version as string */
+#define __INNODB_VERSION(a, b, c, d)	(#a "." #b "." #c "-" #d)
+#define _INNODB_VERSION(a, b, c, d)	__INNODB_VERSION(a, b, c, d)
+
+#define INNODB_VERSION_STR			\
+	_INNODB_VERSION(INNODB_VERSION_MAJOR,	\
+			INNODB_VERSION_MINOR,	\
+			INNODB_VERSION_BUGFIX,	\
+			PERCONA_INNODB_VERSION)
+
+#ifdef MYSQL_DYNAMIC_PLUGIN
+/* In the dynamic plugin, redefine some externally visible symbols
+in order not to conflict with the symbols of a builtin InnoDB. */
+
+/* Rename all C++ classes that contain virtual functions, because we
+have not figured out how to apply the visibility=hidden attribute to
+the virtual method table (vtable) in GCC 3. */
+# define ha_innobase ha_innodb
+#endif /* MYSQL_DYNAMIC_PLUGIN */
+
+#if (defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)) && !defined(MYSQL_SERVER) && !defined(__WIN__)
+# undef __WIN__
+# define __WIN__
+
+# include <windows.h>
+
+# if !defined(WIN64) && !defined(_WIN64)
+#  define UNIV_CAN_USE_X86_ASSEMBLER
+# endif
+
+# ifdef _NT_
+#  define __NT__
+# endif
+
+#else
+/* The defines used with MySQL */
+
+/* Include two header files from MySQL to make the Unix flavor used
+in compiling more Posix-compatible. These headers also define __WIN__
+if we are compiling on Windows. */
+
+# include <my_global.h>
+# include <my_pthread.h>
+
+/* Include <sys/stat.h> to get S_I... macros defined for os0file.c */
+# include <sys/stat.h>
+# if !defined(__NETWARE__) && !defined(__WIN__) 
+#  include <sys/mman.h> /* mmap() for os0proc.c */
+# endif
+
+# undef PACKAGE
+# undef VERSION
+
+/* Include the header file generated by GNU autoconf */
+# ifndef __WIN__
+#  include "config.h"
+# endif
+
+# ifdef HAVE_SCHED_H
+#  include <sched.h>
+# endif
+
+/* When compiling for Itanium IA64, undefine the flag below to prevent use
+of the 32-bit x86 assembler in mutex operations. */
+
+# if defined(__WIN__) && !defined(WIN64) && !defined(_WIN64)
+#  define UNIV_CAN_USE_X86_ASSEMBLER
+# endif
+
+/* For InnoDB rw_locks to work with atomics we need the thread_id
+to be no more than machine word wide. The following enables using
+atomics for InnoDB rw_locks where these conditions are met. */
+#ifdef HAVE_GCC_ATOMIC_BUILTINS
+/* if HAVE_ATOMIC_PTHREAD_T is defined at this point that means that
+the code from plug.in has defined it and we do not need to include
+ut0auxconf.h which would either define HAVE_ATOMIC_PTHREAD_T or will
+be empty */
+# ifndef HAVE_ATOMIC_PTHREAD_T
+#  include "ut0auxconf.h"
+# endif /* HAVE_ATOMIC_PTHREAD_T */
+/* now HAVE_ATOMIC_PTHREAD_T is eventually defined either by plug.in or
+from Makefile.in->ut0auxconf.h */
+# ifdef HAVE_ATOMIC_PTHREAD_T
+#  define INNODB_RW_LOCKS_USE_ATOMICS
+# endif /* HAVE_ATOMIC_PTHREAD_T */
+#endif /* HAVE_GCC_ATOMIC_BUILTINS */
+
+/* We only try to do explicit inlining of functions with gcc and
+Microsoft Visual C++ */
+
+# if !defined(__GNUC__)
+#  undef  UNIV_MUST_NOT_INLINE			/* Remove compiler warning */
+#  define UNIV_MUST_NOT_INLINE
+# endif
+
+# ifdef HAVE_PREAD
+#  define HAVE_PWRITE
+# endif
+
+#endif /* #if (defined(WIN32) || ... */
+
+/*			DEBUG VERSION CONTROL
+			===================== */
+
+/* The following flag will make InnoDB to initialize
+all memory it allocates to zero. It hides Purify
+warnings about reading unallocated memory unless
+memory is read outside the allocated blocks. */
+/*
+#define UNIV_INIT_MEM_TO_ZERO
+*/
+
+/* When this macro is defined then additional test functions will be
+compiled. These functions live at the end of each relevant source file
+and have "test_" prefix. These functions are not called from anywhere in
+the code, they can be called from gdb after
+innobase_start_or_create_for_mysql() has executed using the call
+command. Not tested on Windows. */
+/*
+#define UNIV_COMPILE_TEST_FUNCS
+*/
+
+#if 0
+#define UNIV_DEBUG_VALGRIND			/* Enable extra
+						Valgrind instrumentation */
+#define UNIV_DEBUG_PRINT			/* Enable the compilation of
+						some debug print functions */
+#define UNIV_AHI_DEBUG				/* Enable adaptive hash index
+						debugging without UNIV_DEBUG */
+#define UNIV_BUF_DEBUG				/* Enable buffer pool
+						debugging without UNIV_DEBUG */
+#define UNIV_DEBUG				/* Enable ut_ad() assertions
+						and disable UNIV_INLINE */
+#define UNIV_DEBUG_FILE_ACCESSES		/* Debug .ibd file access
+						(field file_page_was_freed
+						in buf_page_t) */
+#define UNIV_LRU_DEBUG				/* debug the buffer pool LRU */
+#define UNIV_HASH_DEBUG				/* debug HASH_ macros */
+#define UNIV_LIST_DEBUG				/* debug UT_LIST_ macros */
+#define UNIV_MEM_DEBUG				/* detect memory leaks etc */
+#define UNIV_IBUF_DEBUG				/* debug the insert buffer */
+#define UNIV_IBUF_COUNT_DEBUG			/* debug the insert buffer;
+this limits the database to IBUF_COUNT_N_SPACES and IBUF_COUNT_N_PAGES,
+and the insert buffer must be empty when the database is started */
+#define UNIV_SYNC_DEBUG				/* debug mutex and latch
+operations (very slow); also UNIV_DEBUG must be defined */
+#define UNIV_SEARCH_DEBUG			/* debug B-tree comparisons */
+#define UNIV_SYNC_PERF_STAT			/* operation counts for
+						rw-locks and mutexes */
+#define UNIV_SEARCH_PERF_STAT			/* statistics for the
+						adaptive hash index */
+#define UNIV_SRV_PRINT_LATCH_WAITS		/* enable diagnostic output
+						in sync0sync.c */
+#define UNIV_BTR_PRINT				/* enable functions for
+						printing B-trees */
+#define UNIV_ZIP_DEBUG				/* extensive consistency checks
+						for compressed pages */
+#define UNIV_ZIP_COPY				/* call page_zip_copy_recs()
+						more often */
+#endif
+
+#define UNIV_BTR_DEBUG				/* check B-tree links */
+#define UNIV_LIGHT_MEM_DEBUG			/* light memory debugging */
+
+#ifdef HAVE_valgrind
+/* The following sets all new allocated memory to zero before use:
+this can be used to eliminate unnecessary Purify warnings, but note that
+it also masks many bugs Purify could detect. For detailed Purify analysis it
+is best to remove the define below and look through the warnings one
+by one. */
+#define UNIV_SET_MEM_TO_ZERO
+#endif
+
+/*
+#define UNIV_SQL_DEBUG
+#define UNIV_LOG_DEBUG
+*/
+			/* the above option prevents forcing of log to disk
+			at a buffer page write: it should be tested with this
+			option off; also some ibuf tests are suppressed */
+/*
+#define UNIV_BASIC_LOG_DEBUG
+*/
+			/* the above option enables basic recovery debugging:
+			new allocated file pages are reset */
+
+/* Linkage specifier for non-static InnoDB symbols (variables and functions)
+that are only referenced from within InnoDB, not from MySQL */
+#ifdef __WIN__
+# define UNIV_INTERN
+#else
+# define UNIV_INTERN __attribute__((visibility ("hidden")))
+#endif
+
+#if (!defined(UNIV_DEBUG) && !defined(UNIV_MUST_NOT_INLINE))
+/* Definition for inline version */
+
+#ifdef __WIN__
+#define UNIV_INLINE	__inline
+#else
+#define UNIV_INLINE static __inline__
+#endif
+
+#else
+/* If we want to compile a noninlined version we use the following macro
+definitions: */
+
+#define UNIV_NONINL
+#define UNIV_INLINE	UNIV_INTERN
+
+#endif	/* UNIV_DEBUG */
+
+#ifdef _WIN32
+#define UNIV_WORD_SIZE		4
+#elif defined(_WIN64)
+#define UNIV_WORD_SIZE		8
+#else
+/* MySQL config.h generated by GNU autoconf will define SIZEOF_LONG in Posix */
+#define UNIV_WORD_SIZE		SIZEOF_LONG
+#endif
+
+/* The following alignment is used in memory allocations in memory heap
+management to ensure correct alignment for doubles etc. */
+#define UNIV_MEM_ALIGNMENT      8
+
+/* The following alignment is used in aligning lints etc. */
+#define UNIV_WORD_ALIGNMENT	UNIV_WORD_SIZE
+
+/*
+			DATABASE VERSION CONTROL
+			========================
+*/
+
+/* The 2-logarithm of UNIV_PAGE_SIZE: */
+#define UNIV_PAGE_SIZE_SHIFT	14
+/* The universal page size of the database */
+#define UNIV_PAGE_SIZE		(1 << UNIV_PAGE_SIZE_SHIFT)
+
+/* Maximum number of parallel threads in a parallelized operation */
+#define UNIV_MAX_PARALLELISM	32
+
+/*
+			UNIVERSAL TYPE DEFINITIONS
+			==========================
+*/
+
+/* Note that inside MySQL 'byte' is defined as char on Linux! */
+#define byte			unsigned char
+
+/* Define an unsigned integer type that is exactly 32 bits. */
+
+#if SIZEOF_INT == 4
+typedef unsigned int		ib_uint32_t;
+#elif SIZEOF_LONG == 4
+typedef unsigned long		ib_uint32_t;
+#else
+#error "Neither int or long is 4 bytes"
+#endif
+
+/* Another basic type we use is unsigned long integer which should be equal to
+the word size of the machine, that is on a 32-bit platform 32 bits, and on a
+64-bit platform 64 bits. We also give the printf format for the type as a
+macro ULINTPF. */
+
+#ifdef _WIN64
+typedef unsigned __int64	ulint;
+#define ULINTPF			"%I64u"
+typedef __int64			lint;
+#else
+typedef unsigned long int	ulint;
+#define ULINTPF			"%lu"
+typedef long int		lint;
+#endif
+
+#ifdef __WIN__
+typedef __int64			ib_int64_t;
+typedef unsigned __int64	ib_uint64_t;
+#else
+/* Note: longlong and ulonglong come from MySQL headers. */
+typedef longlong		ib_int64_t;
+typedef ulonglong		ib_uint64_t;
+#endif
+
+typedef unsigned long long int	ullint;
+
+#ifndef __WIN__
+#if SIZEOF_LONG != SIZEOF_VOIDP
+#error "Error: InnoDB's ulint must be of the same size as void*"
+#endif
+#endif
+
+/* The 'undefined' value for a ulint */
+#define ULINT_UNDEFINED		((ulint)(-1))
+
+/* The undefined 32-bit unsigned integer */
+#define	ULINT32_UNDEFINED	0xFFFFFFFF
+
+/* Maximum value for a ulint */
+#define ULINT_MAX		((ulint)(-2))
+
+/* Maximum value for ib_uint64_t */
+#define IB_ULONGLONG_MAX	((ib_uint64_t) (~0ULL))
+
+/* This 'ibool' type is used within Innobase. Remember that different included
+headers may define 'bool' differently. Do not assume that 'bool' is a ulint! */
+#define ibool			ulint
+
+#ifndef TRUE
+
+#define TRUE    1
+#define FALSE   0
+
+#endif
+
+/* The following number as the length of a logical field means that the field
+has the SQL NULL as its value. NOTE that because we assume that the length
+of a field is a 32-bit integer when we store it, for example, to an undo log
+on disk, we must have also this number fit in 32 bits, also in 64-bit
+computers! */
+
+#define UNIV_SQL_NULL ULINT32_UNDEFINED
+
+/* Lengths which are not UNIV_SQL_NULL, but bigger than the following
+number indicate that a field contains a reference to an externally
+stored part of the field in the tablespace. The length field then
+contains the sum of the following flag and the locally stored len. */
+
+#define UNIV_EXTERN_STORAGE_FIELD (UNIV_SQL_NULL - UNIV_PAGE_SIZE)
+
+/* Some macros to improve branch prediction and reduce cache misses */
+#if defined(__GNUC__) && (__GNUC__ > 2) && ! defined(__INTEL_COMPILER)
+/* Tell the compiler that 'expr' probably evaluates to 'constant'. */
+# define UNIV_EXPECT(expr,constant) __builtin_expect(expr, constant)
+/* Tell the compiler that a pointer is likely to be NULL */
+# define UNIV_LIKELY_NULL(ptr) __builtin_expect((ulint) ptr, 0)
+/* Minimize cache-miss latency by moving data at addr into a cache before
+it is read. */
+# define UNIV_PREFETCH_R(addr) __builtin_prefetch(addr, 0, 3)
+/* Minimize cache-miss latency by moving data at addr into a cache before
+it is read or written. */
+# define UNIV_PREFETCH_RW(addr) __builtin_prefetch(addr, 1, 3)
+#else
+/* Dummy versions of the macros */
+# define UNIV_EXPECT(expr,value) (expr)
+# define UNIV_LIKELY_NULL(expr) (expr)
+# define UNIV_PREFETCH_R(addr) ((void) 0)
+# define UNIV_PREFETCH_RW(addr) ((void) 0)
+#endif
+/* Tell the compiler that cond is likely to hold */
+#define UNIV_LIKELY(cond) UNIV_EXPECT(cond, TRUE)
+/* Tell the compiler that cond is unlikely to hold */
+#define UNIV_UNLIKELY(cond) UNIV_EXPECT(cond, FALSE)
+
+/* Compile-time constant of the given array's size. */
+#define UT_ARR_SIZE(a) (sizeof(a) / sizeof((a)[0]))
+
+/* The return type from a thread's start function differs between Unix and
+Windows, so define a typedef for it and a macro to use at the end of such
+functions. */
+
+#ifdef __WIN__
+typedef ulint os_thread_ret_t;
+#define OS_THREAD_DUMMY_RETURN return(0)
+#else
+typedef void* os_thread_ret_t;
+#define OS_THREAD_DUMMY_RETURN return(NULL)
+#endif
+
+#include <stdio.h>
+#include "ut0dbg.h"
+#include "ut0ut.h"
+#include "db0err.h"
+#ifdef UNIV_DEBUG_VALGRIND
+# include <valgrind/memcheck.h>
+# define UNIV_MEM_VALID(addr, size) VALGRIND_MAKE_MEM_DEFINED(addr, size)
+# define UNIV_MEM_INVALID(addr, size) VALGRIND_MAKE_MEM_UNDEFINED(addr, size)
+# define UNIV_MEM_FREE(addr, size) VALGRIND_MAKE_MEM_NOACCESS(addr, size)
+# define UNIV_MEM_ALLOC(addr, size) VALGRIND_MAKE_MEM_UNDEFINED(addr, size)
+# define UNIV_MEM_DESC(addr, size, b) VALGRIND_CREATE_BLOCK(addr, size, b)
+# define UNIV_MEM_UNDESC(b) VALGRIND_DISCARD(b)
+# define UNIV_MEM_ASSERT_RW(addr, size) do {				\
+	const void* _p = (const void*) (ulint)				\
+		VALGRIND_CHECK_MEM_IS_DEFINED(addr, size);		\
+	if (UNIV_LIKELY_NULL(_p))					\
+		fprintf(stderr, "%s:%d: %p[%u] undefined at %ld\n",	\
+			__FILE__, __LINE__,				\
+			(const void*) (addr), (unsigned) (size), (long)	\
+			(((const char*) _p) - ((const char*) (addr))));	\
+	} while (0)
+# define UNIV_MEM_ASSERT_W(addr, size) do {				\
+	const void* _p = (const void*) (ulint)				\
+		VALGRIND_CHECK_MEM_IS_ADDRESSABLE(addr, size);		\
+	if (UNIV_LIKELY_NULL(_p))					\
+		fprintf(stderr, "%s:%d: %p[%u] unwritable at %ld\n",	\
+			__FILE__, __LINE__,				\
+			(const void*) (addr), (unsigned) (size), (long)	\
+			(((const char*) _p) - ((const char*) (addr))));	\
+	} while (0)
+#else
+# define UNIV_MEM_VALID(addr, size) do {} while(0)
+# define UNIV_MEM_INVALID(addr, size) do {} while(0)
+# define UNIV_MEM_FREE(addr, size) do {} while(0)
+# define UNIV_MEM_ALLOC(addr, size) do {} while(0)
+# define UNIV_MEM_DESC(addr, size, b) do {} while(0)
+# define UNIV_MEM_UNDESC(b) do {} while(0)
+# define UNIV_MEM_ASSERT_RW(addr, size) do {} while(0)
+# define UNIV_MEM_ASSERT_W(addr, size) do {} while(0)
+#endif
+#define UNIV_MEM_ASSERT_AND_FREE(addr, size) do {	\
+	UNIV_MEM_ASSERT_W(addr, size);			\
+	UNIV_MEM_FREE(addr, size);			\
+} while (0)
+#define UNIV_MEM_ASSERT_AND_ALLOC(addr, size) do {	\
+	UNIV_MEM_ASSERT_W(addr, size);			\
+	UNIV_MEM_ALLOC(addr, size);			\
+} while (0)
+
+#endif
diff --git a/storage/xtradb/include/usr0sess.h b/storage/xtradb/include/usr0sess.h
new file mode 100644
index 00000000000..08c6c70066f
--- /dev/null
+++ b/storage/xtradb/include/usr0sess.h
@@ -0,0 +1,77 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Sessions
+
+Created 6/25/1996 Heikki Tuuri
+*******************************************************/
+
+#ifndef usr0sess_h
+#define usr0sess_h
+
+#include "univ.i"
+#include "ut0byte.h"
+#include "trx0types.h"
+#include "srv0srv.h"
+#include "trx0types.h"
+#include "usr0types.h"
+#include "que0types.h"
+#include "data0data.h"
+#include "rem0rec.h"
+
+/*************************************************************************
+Opens a session. */
+UNIV_INTERN
+sess_t*
+sess_open(void);
+/*============*/
+					/* out, own: session object */
+/*************************************************************************
+Closes a session, freeing the memory occupied by it, if it is in a state
+where it should be closed. */
+UNIV_INTERN
+ibool
+sess_try_close(
+/*===========*/
+				/* out: TRUE if closed */
+	sess_t*		sess);	/* in, own: session object */
+
+/* The session handle. All fields are protected by the kernel mutex */
+struct sess_struct{
+	ulint		state;		/* state of the session */
+	trx_t*		trx;		/* transaction object permanently
+					assigned for the session: the
+					transaction instance designated by the
+					trx id changes, but the memory
+					structure is preserved */
+	UT_LIST_BASE_NODE_T(que_t)
+			graphs;		/* query graphs belonging to this
+					session */
+};
+
+/* Session states */
+#define SESS_ACTIVE		1
+#define SESS_ERROR		2	/* session contains an error message
+					which has not yet been communicated
+					to the client */
+#ifndef UNIV_NONINL
+#include "usr0sess.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/usr0sess.ic b/storage/xtradb/include/usr0sess.ic
new file mode 100644
index 00000000000..5eefed382da
--- /dev/null
+++ b/storage/xtradb/include/usr0sess.ic
@@ -0,0 +1,23 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Sessions
+
+Created 6/25/1996 Heikki Tuuri
+*******************************************************/
diff --git a/storage/xtradb/include/usr0types.h b/storage/xtradb/include/usr0types.h
new file mode 100644
index 00000000000..7f7d12f7bf5
--- /dev/null
+++ b/storage/xtradb/include/usr0types.h
@@ -0,0 +1,30 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Users and sessions global types
+
+Created 6/25/1996 Heikki Tuuri
+*******************************************************/
+
+#ifndef usr0types_h
+#define usr0types_h
+
+typedef struct sess_struct	sess_t;
+
+#endif
diff --git a/storage/xtradb/include/ut0auxconf.h b/storage/xtradb/include/ut0auxconf.h
new file mode 100644
index 00000000000..6362b7ca412
--- /dev/null
+++ b/storage/xtradb/include/ut0auxconf.h
@@ -0,0 +1,14 @@
+/* Do not remove this file even though it is empty.
+This file is included in univ.i and will cause compilation failure
+if not present.
+A custom check has been added in the generated
+storage/innobase/Makefile.in that is shipped with with the InnoDB Plugin
+source archive. This check tries to compile a test program and if
+successful then adds "#define HAVE_ATOMIC_PTHREAD_T" to this file.
+This is a hack that has been developed in order to check for pthread_t
+atomicity without the need to regenerate the ./configure script that is
+distributed in the MySQL 5.1 official source archives.
+If by any chance Makefile.in and ./configure are regenerated and thus
+the hack from Makefile.in wiped away then the "real" check from plug.in
+will take over.
+*/
diff --git a/storage/xtradb/include/ut0byte.h b/storage/xtradb/include/ut0byte.h
new file mode 100644
index 00000000000..24aac1678b3
--- /dev/null
+++ b/storage/xtradb/include/ut0byte.h
@@ -0,0 +1,268 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/**********************************************************************
+Utilities for byte operations
+
+Created 1/20/1994 Heikki Tuuri
+***********************************************************************/
+
+#ifndef ut0byte_h
+#define ut0byte_h
+
+
+#include "univ.i"
+
+/* Type definition for a 64-bit unsigned integer, which works also
+in 32-bit machines. NOTE! Access the fields only with the accessor
+functions. This definition appears here only for the compiler to
+know the size of a dulint. */
+
+typedef	struct dulint_struct	dulint;
+struct dulint_struct{
+	ulint	high;	/* most significant 32 bits */
+	ulint	low;	/* least significant 32 bits */
+};
+
+/* Zero value for a dulint */
+extern const dulint	ut_dulint_zero;
+
+/* Maximum value for a dulint */
+extern const dulint	ut_dulint_max;
+
+/***********************************************************
+Creates a 64-bit dulint out of two ulints. */
+UNIV_INLINE
+dulint
+ut_dulint_create(
+/*=============*/
+			/* out: created dulint */
+	ulint	high,	/* in: high-order 32 bits */
+	ulint	low);	/* in: low-order 32 bits */
+/***********************************************************
+Gets the high-order 32 bits of a dulint. */
+UNIV_INLINE
+ulint
+ut_dulint_get_high(
+/*===============*/
+			/* out: 32 bits in ulint */
+	dulint	d);	/* in: dulint */
+/***********************************************************
+Gets the low-order 32 bits of a dulint. */
+UNIV_INLINE
+ulint
+ut_dulint_get_low(
+/*==============*/
+			/* out: 32 bits in ulint */
+	dulint	d);	/* in: dulint */
+/***********************************************************
+Converts a dulint (a struct of 2 ulints) to ib_int64_t, which is a 64-bit
+integer type. */
+UNIV_INLINE
+ib_int64_t
+ut_conv_dulint_to_longlong(
+/*=======================*/
+			/* out: value in ib_int64_t type */
+	dulint	d);	/* in: dulint */
+/***********************************************************
+Tests if a dulint is zero. */
+UNIV_INLINE
+ibool
+ut_dulint_is_zero(
+/*==============*/
+			/* out: TRUE if zero */
+	dulint	a);	/* in: dulint */
+/***********************************************************
+Compares two dulints. */
+UNIV_INLINE
+int
+ut_dulint_cmp(
+/*==========*/
+			/* out: -1 if a < b, 0 if a == b,
+			1 if a > b */
+	dulint	a,	/* in: dulint */
+	dulint	b);	/* in: dulint */
+/***********************************************************
+Calculates the max of two dulints. */
+UNIV_INLINE
+dulint
+ut_dulint_get_max(
+/*==============*/
+			/* out: max(a, b) */
+	dulint	a,	/* in: dulint */
+	dulint	b);	/* in: dulint */
+/***********************************************************
+Calculates the min of two dulints. */
+UNIV_INLINE
+dulint
+ut_dulint_get_min(
+/*==============*/
+			/* out: min(a, b) */
+	dulint	a,	/* in: dulint */
+	dulint	b);	/* in: dulint */
+/***********************************************************
+Adds a ulint to a dulint. */
+UNIV_INLINE
+dulint
+ut_dulint_add(
+/*==========*/
+			/* out: sum a + b */
+	dulint	a,	/* in: dulint */
+	ulint	b);	/* in: ulint */
+/***********************************************************
+Subtracts a ulint from a dulint. */
+UNIV_INLINE
+dulint
+ut_dulint_subtract(
+/*===============*/
+			/* out: a - b */
+	dulint	a,	/* in: dulint */
+	ulint	b);	/* in: ulint, b <= a */
+/***********************************************************
+Subtracts a dulint from another. NOTE that the difference must be positive
+and smaller that 4G. */
+UNIV_INLINE
+ulint
+ut_dulint_minus(
+/*============*/
+			/* out: a - b */
+	dulint	a,	/* in: dulint; NOTE a must be >= b and at most
+			2 to power 32 - 1 greater */
+	dulint	b);	/* in: dulint */
+/************************************************************
+Rounds a dulint downward to a multiple of a power of 2. */
+UNIV_INLINE
+dulint
+ut_dulint_align_down(
+/*=================*/
+				/* out: rounded value */
+	dulint	 n,		/* in: number to be rounded */
+	ulint	 align_no);	/* in: align by this number which must be a
+				power of 2 */
+/************************************************************
+Rounds a dulint upward to a multiple of a power of 2. */
+UNIV_INLINE
+dulint
+ut_dulint_align_up(
+/*===============*/
+				/* out: rounded value */
+	dulint	 n,		/* in: number to be rounded */
+	ulint	 align_no);	/* in: align by this number which must be a
+				power of 2 */
+/************************************************************
+Rounds a dulint downward to a multiple of a power of 2. */
+UNIV_INLINE
+ib_uint64_t
+ut_uint64_align_down(
+/*=================*/
+					/* out: rounded value */
+	ib_uint64_t	 n,		/* in: number to be rounded */
+	ulint		 align_no);	/* in: align by this number
+					which must be a power of 2 */
+/************************************************************
+Rounds ib_uint64_t upward to a multiple of a power of 2. */
+UNIV_INLINE
+ib_uint64_t
+ut_uint64_align_up(
+/*===============*/
+					/* out: rounded value */
+	ib_uint64_t	 n,		/* in: number to be rounded */
+	ulint		 align_no);	/* in: align by this number
+					which must be a power of 2 */
+/***********************************************************
+Increments a dulint variable by 1. */
+#define UT_DULINT_INC(D)\
+{\
+	if ((D).low == 0xFFFFFFFFUL) {\
+		(D).high = (D).high + 1;\
+		(D).low = 0;\
+	} else {\
+		(D).low = (D).low + 1;\
+	}\
+}
+/***********************************************************
+Tests if two dulints are equal. */
+#define UT_DULINT_EQ(D1, D2)	(((D1).low == (D2).low)\
+						&& ((D1).high == (D2).high))
+#ifdef notdefined
+/****************************************************************
+Sort function for dulint arrays. */
+UNIV_INTERN
+void
+ut_dulint_sort(dulint* arr, dulint* aux_arr, ulint low, ulint high);
+/*===============================================================*/
+#endif /* notdefined */
+
+/*************************************************************
+The following function rounds up a pointer to the nearest aligned address. */
+UNIV_INLINE
+void*
+ut_align(
+/*=====*/
+				/* out: aligned pointer */
+	void*	ptr,		/* in: pointer */
+	ulint	align_no);	/* in: align by this number */
+/*************************************************************
+The following function rounds down a pointer to the nearest
+aligned address. */
+UNIV_INLINE
+void*
+ut_align_down(
+/*==========*/
+					/* out: aligned pointer */
+	const void*	ptr,		/* in: pointer */
+	ulint		align_no)	/* in: align by this number */
+		__attribute__((const));
+/*************************************************************
+The following function computes the offset of a pointer from the nearest
+aligned address. */
+UNIV_INLINE
+ulint
+ut_align_offset(
+/*============*/
+					/* out: distance from aligned
+					pointer */
+	const void*	ptr,		/* in: pointer */
+	ulint		align_no)	/* in: align by this number */
+			__attribute__((const));
+/*********************************************************************
+Gets the nth bit of a ulint. */
+UNIV_INLINE
+ibool
+ut_bit_get_nth(
+/*===========*/
+			/* out: TRUE if nth bit is 1; 0th bit is defined to
+			be the least significant */
+	ulint	a,	/* in: ulint */
+	ulint	n);	/* in: nth bit requested */
+/*********************************************************************
+Sets the nth bit of a ulint. */
+UNIV_INLINE
+ulint
+ut_bit_set_nth(
+/*===========*/
+			/* out: the ulint with the bit set as requested */
+	ulint	a,	/* in: ulint */
+	ulint	n,	/* in: nth bit requested */
+	ibool	val);	/* in: value for the bit to set */
+
+#ifndef UNIV_NONINL
+#include "ut0byte.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/ut0byte.ic b/storage/xtradb/include/ut0byte.ic
new file mode 100644
index 00000000000..021a3a15009
--- /dev/null
+++ b/storage/xtradb/include/ut0byte.ic
@@ -0,0 +1,413 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************************
+Utilities for byte operations
+
+Created 5/30/1994 Heikki Tuuri
+*******************************************************************/
+
+/***********************************************************
+Creates a 64-bit dulint out of two ulints. */
+UNIV_INLINE
+dulint
+ut_dulint_create(
+/*=============*/
+			/* out: created dulint */
+	ulint	high,	/* in: high-order 32 bits */
+	ulint	low)	/* in: low-order 32 bits */
+{
+	dulint	res;
+
+	ut_ad(high <= 0xFFFFFFFF);
+	ut_ad(low <= 0xFFFFFFFF);
+
+	res.high = high;
+	res.low	 = low;
+
+	return(res);
+}
+
+/***********************************************************
+Gets the high-order 32 bits of a dulint. */
+UNIV_INLINE
+ulint
+ut_dulint_get_high(
+/*===============*/
+			/* out: 32 bits in ulint */
+	dulint	d)	/* in: dulint */
+{
+	return(d.high);
+}
+
+/***********************************************************
+Gets the low-order 32 bits of a dulint. */
+UNIV_INLINE
+ulint
+ut_dulint_get_low(
+/*==============*/
+			/* out: 32 bits in ulint */
+	dulint	d)	/* in: dulint */
+{
+	return(d.low);
+}
+
+/***********************************************************
+Converts a dulint (a struct of 2 ulints) to ib_int64_t, which is a 64-bit
+integer type. */
+UNIV_INLINE
+ib_int64_t
+ut_conv_dulint_to_longlong(
+/*=======================*/
+			/* out: value in ib_int64_t type */
+	dulint	d)	/* in: dulint */
+{
+	return((ib_int64_t)d.low
+	       + (((ib_int64_t)d.high) << 32));
+}
+
+/***********************************************************
+Tests if a dulint is zero. */
+UNIV_INLINE
+ibool
+ut_dulint_is_zero(
+/*==============*/
+			/* out: TRUE if zero */
+	dulint	a)	/* in: dulint */
+{
+	if ((a.low == 0) && (a.high == 0)) {
+
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
+
+/***********************************************************
+Compares two dulints. */
+UNIV_INLINE
+int
+ut_dulint_cmp(
+/*==========*/
+			/* out: -1 if a < b, 0 if a == b,
+			1 if a > b */
+	dulint	a,	/* in: dulint */
+	dulint	b)	/* in: dulint */
+{
+	if (a.high > b.high) {
+		return(1);
+	} else if (a.high < b.high) {
+		return(-1);
+	} else if (a.low > b.low) {
+		return(1);
+	} else if (a.low < b.low) {
+		return(-1);
+	} else {
+		return(0);
+	}
+}
+
+/***********************************************************
+Calculates the max of two dulints. */
+UNIV_INLINE
+dulint
+ut_dulint_get_max(
+/*==============*/
+			/* out: max(a, b) */
+	dulint	a,	/* in: dulint */
+	dulint	b)	/* in: dulint */
+{
+	if (ut_dulint_cmp(a, b) > 0) {
+
+		return(a);
+	}
+
+	return(b);
+}
+
+/***********************************************************
+Calculates the min of two dulints. */
+UNIV_INLINE
+dulint
+ut_dulint_get_min(
+/*==============*/
+			/* out: min(a, b) */
+	dulint	a,	/* in: dulint */
+	dulint	b)	/* in: dulint */
+{
+	if (ut_dulint_cmp(a, b) > 0) {
+
+		return(b);
+	}
+
+	return(a);
+}
+
+/***********************************************************
+Adds a ulint to a dulint. */
+UNIV_INLINE
+dulint
+ut_dulint_add(
+/*==========*/
+			/* out: sum a + b */
+	dulint	a,	/* in: dulint */
+	ulint	b)	/* in: ulint */
+{
+	if (0xFFFFFFFFUL - b >= a.low) {
+		a.low += b;
+
+		return(a);
+	}
+
+	a.low = a.low - (0xFFFFFFFFUL - b) - 1;
+
+	a.high++;
+
+	return(a);
+}
+
+/***********************************************************
+Subtracts a ulint from a dulint. */
+UNIV_INLINE
+dulint
+ut_dulint_subtract(
+/*===============*/
+			/* out: a - b */
+	dulint	a,	/* in: dulint */
+	ulint	b)	/* in: ulint, b <= a */
+{
+	if (a.low >= b) {
+		a.low -= b;
+
+		return(a);
+	}
+
+	b -= a.low + 1;
+
+	a.low = 0xFFFFFFFFUL - b;
+
+	ut_ad(a.high > 0);
+
+	a.high--;
+
+	return(a);
+}
+
+/***********************************************************
+Subtracts a dulint from another. NOTE that the difference must be positive
+and smaller that 4G. */
+UNIV_INLINE
+ulint
+ut_dulint_minus(
+/*============*/
+			/* out: a - b */
+	dulint	a,	/* in: dulint; NOTE a must be >= b and at most
+			2 to power 32 - 1 greater */
+	dulint	b)	/* in: dulint */
+{
+	ulint	diff;
+
+	if (a.high == b.high) {
+		ut_ad(a.low >= b.low);
+
+		return(a.low - b.low);
+	}
+
+	ut_ad(a.high == b.high + 1);
+
+	diff = (ulint)(0xFFFFFFFFUL - b.low);
+	diff += 1 + a.low;
+
+	ut_ad(diff > a.low);
+
+	return(diff);
+}
+
+/************************************************************
+Rounds a dulint downward to a multiple of a power of 2. */
+UNIV_INLINE
+dulint
+ut_dulint_align_down(
+/*=================*/
+				/* out: rounded value */
+	dulint	 n,		/* in: number to be rounded */
+	ulint	 align_no)	/* in: align by this number which must be a
+				power of 2 */
+{
+	ulint	low, high;
+
+	ut_ad(align_no > 0);
+	ut_ad(((align_no - 1) & align_no) == 0);
+
+	low = ut_dulint_get_low(n);
+	high = ut_dulint_get_high(n);
+
+	low = low & ~(align_no - 1);
+
+	return(ut_dulint_create(high, low));
+}
+
+/************************************************************
+Rounds a dulint upward to a multiple of a power of 2. */
+UNIV_INLINE
+dulint
+ut_dulint_align_up(
+/*===============*/
+				/* out: rounded value */
+	dulint	 n,		/* in: number to be rounded */
+	ulint	 align_no)	/* in: align by this number which must be a
+				power of 2 */
+{
+	return(ut_dulint_align_down(ut_dulint_add(n, align_no - 1), align_no));
+}
+
+/************************************************************
+Rounds ib_uint64_t downward to a multiple of a power of 2. */
+UNIV_INLINE
+ib_uint64_t
+ut_uint64_align_down(
+/*=================*/
+					/* out: rounded value */
+	ib_uint64_t	 n,		/* in: number to be rounded */
+	ulint		 align_no)	/* in: align by this number
+					which must be a power of 2 */
+{
+	ut_ad(align_no > 0);
+	ut_ad(ut_is_2pow(align_no));
+
+	return(n & ~((ib_uint64_t) align_no - 1));
+}
+
+/************************************************************
+Rounds ib_uint64_t upward to a multiple of a power of 2. */
+UNIV_INLINE
+ib_uint64_t
+ut_uint64_align_up(
+/*===============*/
+					/* out: rounded value */
+	ib_uint64_t	 n,		/* in: number to be rounded */
+	ulint		 align_no)	/* in: align by this number
+					which must be a power of 2 */
+{
+	ib_uint64_t	align_1 = (ib_uint64_t) align_no - 1;
+
+	ut_ad(align_no > 0);
+	ut_ad(ut_is_2pow(align_no));
+
+	return((n + align_1) & ~align_1);
+}
+
+/*************************************************************
+The following function rounds up a pointer to the nearest aligned address. */
+UNIV_INLINE
+void*
+ut_align(
+/*=====*/
+				/* out: aligned pointer */
+	void*	ptr,		/* in: pointer */
+	ulint	align_no)	/* in: align by this number */
+{
+	ut_ad(align_no > 0);
+	ut_ad(((align_no - 1) & align_no) == 0);
+	ut_ad(ptr);
+
+	ut_ad(sizeof(void*) == sizeof(ulint));
+
+	return((void*)((((ulint)ptr) + align_no - 1) & ~(align_no - 1)));
+}
+
+/*************************************************************
+The following function rounds down a pointer to the nearest
+aligned address. */
+UNIV_INLINE
+void*
+ut_align_down(
+/*==========*/
+					/* out: aligned pointer */
+	const void*	ptr,		/* in: pointer */
+	ulint		align_no)	/* in: align by this number */
+{
+	ut_ad(align_no > 0);
+	ut_ad(((align_no - 1) & align_no) == 0);
+	ut_ad(ptr);
+
+	ut_ad(sizeof(void*) == sizeof(ulint));
+
+	return((void*)((((ulint)ptr)) & ~(align_no - 1)));
+}
+
+/*************************************************************
+The following function computes the offset of a pointer from the nearest
+aligned address. */
+UNIV_INLINE
+ulint
+ut_align_offset(
+/*============*/
+					/* out: distance from
+					aligned pointer */
+	const void*	ptr,		/* in: pointer */
+	ulint		align_no)	/* in: align by this number */
+{
+	ut_ad(align_no > 0);
+	ut_ad(((align_no - 1) & align_no) == 0);
+	ut_ad(ptr);
+
+	ut_ad(sizeof(void*) == sizeof(ulint));
+
+	return(((ulint)ptr) & (align_no - 1));
+}
+
+/*********************************************************************
+Gets the nth bit of a ulint. */
+UNIV_INLINE
+ibool
+ut_bit_get_nth(
+/*===========*/
+			/* out: TRUE if nth bit is 1; 0th bit is defined to
+			be the least significant */
+	ulint	a,	/* in: ulint */
+	ulint	n)	/* in: nth bit requested */
+{
+	ut_ad(n < 8 * sizeof(ulint));
+#if TRUE != 1
+# error "TRUE != 1"
+#endif
+	return(1 & (a >> n));
+}
+
+/*********************************************************************
+Sets the nth bit of a ulint. */
+UNIV_INLINE
+ulint
+ut_bit_set_nth(
+/*===========*/
+			/* out: the ulint with the bit set as requested */
+	ulint	a,	/* in: ulint */
+	ulint	n,	/* in: nth bit requested */
+	ibool	val)	/* in: value for the bit to set */
+{
+	ut_ad(n < 8 * sizeof(ulint));
+#if TRUE != 1
+# error "TRUE != 1"
+#endif
+	if (val) {
+		return(((ulint) 1 << n) | a);
+	} else {
+		return(~((ulint) 1 << n) & a);
+	}
+}
diff --git a/storage/xtradb/include/ut0dbg.h b/storage/xtradb/include/ut0dbg.h
new file mode 100644
index 00000000000..a206789fd4c
--- /dev/null
+++ b/storage/xtradb/include/ut0dbg.h
@@ -0,0 +1,160 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/*********************************************************************
+Debug utilities for Innobase
+
+Created 1/30/1994 Heikki Tuuri
+**********************************************************************/
+
+#ifndef ut0dbg_h
+#define ut0dbg_h
+
+#include "univ.i"
+#include <stdlib.h>
+#include "os0thread.h"
+
+#if defined(__GNUC__) && (__GNUC__ > 2)
+# define UT_DBG_FAIL(EXPR) UNIV_UNLIKELY(!((ulint)(EXPR)))
+#else
+extern ulint	ut_dbg_zero; /* This is used to eliminate
+				compiler warnings */
+# define UT_DBG_FAIL(EXPR) !((ulint)(EXPR) + ut_dbg_zero)
+#endif
+
+/*****************************************************************
+Report a failed assertion. */
+UNIV_INTERN
+void
+ut_dbg_assertion_failed(
+/*====================*/
+	const char* expr,	/* in: the failed assertion */
+	const char* file,	/* in: source file containing the assertion */
+	ulint line);		/* in: line number of the assertion */
+
+#ifdef __NETWARE__
+/* Flag for ignoring further assertion failures.
+On NetWare, have a graceful exit rather than a segfault to avoid abends. */
+extern ibool	panic_shutdown;
+/* Abort the execution. */
+void ut_dbg_panic(void);
+# define UT_DBG_PANIC ut_dbg_panic()
+/* Stop threads in ut_a(). */
+# define UT_DBG_STOP	do {} while (0)	/* We do not do this on NetWare */
+#else /* __NETWARE__ */
+# if defined(__WIN__) || defined(__INTEL_COMPILER)
+#  undef UT_DBG_USE_ABORT
+# elif defined(__GNUC__) && (__GNUC__ > 2)
+#  define UT_DBG_USE_ABORT
+# endif
+
+# ifndef UT_DBG_USE_ABORT
+/* A null pointer that will be dereferenced to trigger a memory trap */
+extern ulint*	ut_dbg_null_ptr;
+# endif
+
+# if defined(UNIV_SYNC_DEBUG) || !defined(UT_DBG_USE_ABORT)
+/* Flag for indicating that all threads should stop.  This will be set
+by ut_dbg_assertion_failed(). */
+extern ibool	ut_dbg_stop_threads;
+
+/*****************************************************************
+Stop a thread after assertion failure. */
+UNIV_INTERN
+void
+ut_dbg_stop_thread(
+/*===============*/
+	const char*	file,
+	ulint		line);
+# endif
+
+# ifdef UT_DBG_USE_ABORT
+/* Abort the execution. */
+#  define UT_DBG_PANIC abort()
+/* Stop threads (null operation) */
+#  define UT_DBG_STOP do {} while (0)
+# else /* UT_DBG_USE_ABORT */
+/* Abort the execution. */
+#  define UT_DBG_PANIC					\
+	if (*(ut_dbg_null_ptr)) ut_dbg_null_ptr = NULL
+/* Stop threads in ut_a(). */
+#  define UT_DBG_STOP do						\
+	if (UNIV_UNLIKELY(ut_dbg_stop_threads)) {		\
+		ut_dbg_stop_thread(__FILE__, (ulint) __LINE__);	\
+	} while (0)
+# endif /* UT_DBG_USE_ABORT */
+#endif /* __NETWARE__ */
+
+/* Abort execution if EXPR does not evaluate to nonzero. */
+#define ut_a(EXPR) do {						\
+	if (UT_DBG_FAIL(EXPR)) {				\
+		ut_dbg_assertion_failed(#EXPR,			\
+				__FILE__, (ulint) __LINE__);	\
+		UT_DBG_PANIC;					\
+	}							\
+	UT_DBG_STOP;						\
+} while (0)
+
+/* Abort execution. */
+#define ut_error do {						\
+	ut_dbg_assertion_failed(0, __FILE__, (ulint) __LINE__);	\
+	UT_DBG_PANIC;						\
+} while (0)
+
+#ifdef UNIV_DEBUG
+#define ut_ad(EXPR)	ut_a(EXPR)
+#define ut_d(EXPR)	do {EXPR;} while (0)
+#else
+#define ut_ad(EXPR)
+#define ut_d(EXPR)
+#endif
+
+#define UT_NOT_USED(A)	A = A
+
+#ifdef UNIV_COMPILE_TEST_FUNCS
+
+#include <sys/types.h>
+#include <sys/time.h>
+#include <sys/resource.h>
+
+/* structure used for recording usage statistics */
+typedef struct speedo_struct {
+	struct rusage	ru;
+	struct timeval	tv;
+} speedo_t;
+
+/***********************************************************************
+Resets a speedo (records the current time in it). */
+UNIV_INTERN
+void
+speedo_reset(
+/*=========*/
+	speedo_t*	speedo);	/* out: speedo */
+
+/***********************************************************************
+Shows the time elapsed and usage statistics since the last reset of a
+speedo. */
+UNIV_INTERN
+void
+speedo_show(
+/*========*/
+	const speedo_t*	speedo);	/* in: speedo */
+
+#endif /* UNIV_COMPILE_TEST_FUNCS */
+
+#endif
diff --git a/storage/xtradb/include/ut0list.h b/storage/xtradb/include/ut0list.h
new file mode 100644
index 00000000000..034aa400af9
--- /dev/null
+++ b/storage/xtradb/include/ut0list.h
@@ -0,0 +1,165 @@
+/*****************************************************************************
+
+Copyright (c) 2006, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/***********************************************************************
+A double-linked list. This differs from the one in ut0lst.h in that in this
+one, each list node contains a pointer to the data, whereas the one in
+ut0lst.h uses a strategy where the list pointers are embedded in the data
+items themselves.
+
+Use this one when you need to store arbitrary data in the list where you
+can't embed the list pointers in the data, if a data item needs to be
+stored in multiple lists, etc.
+
+Note about the memory management: ib_list_t is a fixed-size struct whose
+allocation/deallocation is done through ib_list_create/ib_list_free, but the
+memory for the list nodes is allocated through a user-given memory heap,
+which can either be the same for all nodes or vary per node. Most users will
+probably want to create a memory heap to store the item-specific data, and
+pass in this same heap to the list node creation functions, thus
+automatically freeing the list node when the item's heap is freed.
+
+************************************************************************/
+
+#ifndef IB_LIST_H
+#define IB_LIST_H
+
+#include "mem0mem.h"
+
+typedef struct ib_list_struct ib_list_t;
+typedef struct ib_list_node_struct ib_list_node_t;
+typedef struct ib_list_helper_struct ib_list_helper_t;
+
+/********************************************************************
+Create a new list using mem_alloc. Lists created with this function must be
+freed with ib_list_free. */
+UNIV_INTERN
+ib_list_t*
+ib_list_create(void);
+/*=================*/
+			/* out: list */
+
+
+/********************************************************************
+Create a new list using the given heap. ib_list_free MUST NOT BE CALLED for
+lists created with this function. */
+UNIV_INTERN
+ib_list_t*
+ib_list_create_heap(
+/*================*/
+				/* out: list */
+	mem_heap_t*	heap);	/* in: memory heap to use */
+
+/********************************************************************
+Free a list. */
+UNIV_INTERN
+void
+ib_list_free(
+/*=========*/
+	ib_list_t*	list);	/* in: list */
+
+/********************************************************************
+Add the data to the start of the list. */
+UNIV_INTERN
+ib_list_node_t*
+ib_list_add_first(
+/*==============*/
+				/* out: new list node*/
+	ib_list_t*	list,	/* in: list */
+	void*		data,	/* in: data */
+	mem_heap_t*	heap);	/* in: memory heap to use */
+
+/********************************************************************
+Add the data to the end of the list. */
+UNIV_INTERN
+ib_list_node_t*
+ib_list_add_last(
+/*=============*/
+				/* out: new list node*/
+	ib_list_t*	list,	/* in: list */
+	void*		data,	/* in: data */
+	mem_heap_t*	heap);	/* in: memory heap to use */
+
+/********************************************************************
+Add the data after the indicated node. */
+UNIV_INTERN
+ib_list_node_t*
+ib_list_add_after(
+/*==============*/
+					/* out: new list node*/
+	ib_list_t*	list,		/* in: list */
+	ib_list_node_t*	prev_node,	/* in: node preceding new node (can
+					be NULL) */
+	void*		data,		/* in: data */
+	mem_heap_t*	heap);		/* in: memory heap to use */
+
+/********************************************************************
+Remove the node from the list. */
+UNIV_INTERN
+void
+ib_list_remove(
+/*===========*/
+	ib_list_t*	list,	/* in: list */
+	ib_list_node_t*	node);	/* in: node to remove */
+
+/********************************************************************
+Get the first node in the list. */
+UNIV_INLINE
+ib_list_node_t*
+ib_list_get_first(
+/*==============*/
+				/* out: first node, or NULL */
+	ib_list_t*	list);	/* in: list */
+
+/********************************************************************
+Get the last node in the list. */
+UNIV_INLINE
+ib_list_node_t*
+ib_list_get_last(
+/*=============*/
+				/* out: last node, or NULL */
+	ib_list_t*	list);	/* in: list */
+
+/* List. */
+struct ib_list_struct {
+	ib_list_node_t*		first;		/* first node */
+	ib_list_node_t*		last;		/* last node */
+	ibool			is_heap_list;	/* TRUE if this list was
+						allocated through a heap */
+};
+
+/* A list node. */
+struct ib_list_node_struct {
+	ib_list_node_t*		prev;		/* previous node */
+	ib_list_node_t*		next;		/* next node */
+	void*			data;		/* user data */
+};
+
+/* Quite often, the only additional piece of data you need is the per-item
+memory heap, so we have this generic struct available to use in those
+cases. */
+struct ib_list_helper_struct {
+	mem_heap_t*	heap;		/* memory heap */
+	void*		data;		/* user data */
+};
+
+#ifndef UNIV_NONINL
+#include "ut0list.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/ut0list.ic b/storage/xtradb/include/ut0list.ic
new file mode 100644
index 00000000000..c79a0cf18dc
--- /dev/null
+++ b/storage/xtradb/include/ut0list.ic
@@ -0,0 +1,41 @@
+/*****************************************************************************
+
+Copyright (c) 2006, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/********************************************************************
+Get the first node in the list. */
+UNIV_INLINE
+ib_list_node_t*
+ib_list_get_first(
+/*==============*/
+				/* out: first node, or NULL */
+	ib_list_t*	list)	/* in: list */
+{
+	return(list->first);
+}
+
+/********************************************************************
+Get the last node in the list. */
+UNIV_INLINE
+ib_list_node_t*
+ib_list_get_last(
+/*=============*/
+				/* out: last node, or NULL */
+	ib_list_t*	list)	/* in: list */
+{
+	return(list->last);
+}
diff --git a/storage/xtradb/include/ut0lst.h b/storage/xtradb/include/ut0lst.h
new file mode 100644
index 00000000000..46ee23a2538
--- /dev/null
+++ b/storage/xtradb/include/ut0lst.h
@@ -0,0 +1,243 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/**********************************************************************
+List utilities
+
+Created 9/10/1995 Heikki Tuuri
+***********************************************************************/
+
+#ifndef ut0lst_h
+#define ut0lst_h
+
+#include "univ.i"
+
+/* This module implements the two-way linear list which should be used
+if a list is used in the database. Note that a single struct may belong
+to two or more lists, provided that the list are given different names.
+An example of the usage of the lists can be found in fil0fil.c. */
+
+/***********************************************************************
+This macro expands to the unnamed type definition of a struct which acts
+as the two-way list base node. The base node contains pointers
+to both ends of the list and a count of nodes in the list (excluding
+the base node from the count). TYPE should be the list node type name. */
+
+#define UT_LIST_BASE_NODE_T(TYPE)\
+struct {\
+	ulint	count;	/* count of nodes in list */\
+	TYPE *	start;	/* pointer to list start, NULL if empty */\
+	TYPE *	end;	/* pointer to list end, NULL if empty */\
+}\
+
+/***********************************************************************
+This macro expands to the unnamed type definition of a struct which
+should be embedded in the nodes of the list, the node type must be a struct.
+This struct contains the pointers to next and previous nodes in the list.
+The name of the field in the node struct should be the name given
+to the list. TYPE should be the list node type name. Example of usage:
+
+typedef struct LRU_node_struct	LRU_node_t;
+struct LRU_node_struct {
+	UT_LIST_NODE_T(LRU_node_t)	LRU_list;
+	...
+}
+The example implements an LRU list of name LRU_list. Its nodes are of type
+LRU_node_t.
+*/
+
+#define UT_LIST_NODE_T(TYPE)\
+struct {\
+	TYPE *	prev;	/* pointer to the previous node,\
+			NULL if start of list */\
+	TYPE *	next;	/* pointer to next node, NULL if end of list */\
+}\
+
+/***********************************************************************
+Initializes the base node of a two-way list. */
+
+#define UT_LIST_INIT(BASE)\
+{\
+	(BASE).count = 0;\
+	(BASE).start = NULL;\
+	(BASE).end   = NULL;\
+}\
+
+/***********************************************************************
+Adds the node as the first element in a two-way linked list.
+BASE has to be the base node (not a pointer to it). N has to be
+the pointer to the node to be added to the list. NAME is the list name. */
+
+#define UT_LIST_ADD_FIRST(NAME, BASE, N)\
+{\
+	ut_ad(N);\
+	((BASE).count)++;\
+	((N)->NAME).next = (BASE).start;\
+	((N)->NAME).prev = NULL;\
+	if (UNIV_LIKELY((BASE).start != NULL)) {\
+		ut_ad((BASE).start != (N));\
+		(((BASE).start)->NAME).prev = (N);\
+	}\
+	(BASE).start = (N);\
+	if (UNIV_UNLIKELY((BASE).end == NULL)) {\
+		(BASE).end = (N);\
+	}\
+}\
+
+/***********************************************************************
+Adds the node as the last element in a two-way linked list.
+BASE has to be the base node (not a pointer to it). N has to be
+the pointer to the node to be added to the list. NAME is the list name. */
+
+#define UT_LIST_ADD_LAST(NAME, BASE, N)\
+{\
+	ut_ad(N);\
+	((BASE).count)++;\
+	((N)->NAME).prev = (BASE).end;\
+	((N)->NAME).next = NULL;\
+	if ((BASE).end != NULL) {\
+		ut_ad((BASE).end != (N));\
+		(((BASE).end)->NAME).next = (N);\
+	}\
+	(BASE).end = (N);\
+	if ((BASE).start == NULL) {\
+		(BASE).start = (N);\
+	}\
+}\
+
+/***********************************************************************
+Inserts a NODE2 after NODE1 in a list.
+BASE has to be the base node (not a pointer to it). NAME is the list
+name, NODE1 and NODE2 are pointers to nodes. */
+
+#define UT_LIST_INSERT_AFTER(NAME, BASE, NODE1, NODE2)\
+{\
+	ut_ad(NODE1);\
+	ut_ad(NODE2);\
+	ut_ad((NODE1) != (NODE2));\
+	((BASE).count)++;\
+	((NODE2)->NAME).prev = (NODE1);\
+	((NODE2)->NAME).next = ((NODE1)->NAME).next;\
+	if (((NODE1)->NAME).next != NULL) {\
+		((((NODE1)->NAME).next)->NAME).prev = (NODE2);\
+	}\
+	((NODE1)->NAME).next = (NODE2);\
+	if ((BASE).end == (NODE1)) {\
+		(BASE).end = (NODE2);\
+	}\
+}\
+
+/* Invalidate the pointers in a list node. */
+#ifdef UNIV_LIST_DEBUG
+# define UT_LIST_REMOVE_CLEAR(NAME, N)		\
+((N)->NAME.prev = (N)->NAME.next = (void*) -1)
+#else
+# define UT_LIST_REMOVE_CLEAR(NAME, N) while (0)
+#endif
+
+/***********************************************************************
+Removes a node from a two-way linked list. BASE has to be the base node
+(not a pointer to it). N has to be the pointer to the node to be removed
+from the list. NAME is the list name. */
+
+#define UT_LIST_REMOVE(NAME, BASE, N)					\
+do {									\
+	ut_ad(N);							\
+	ut_a((BASE).count > 0);						\
+	((BASE).count)--;						\
+	if (((N)->NAME).next != NULL) {					\
+		((((N)->NAME).next)->NAME).prev = ((N)->NAME).prev;	\
+	} else {							\
+		(BASE).end = ((N)->NAME).prev;				\
+	}								\
+	if (((N)->NAME).prev != NULL) {					\
+		((((N)->NAME).prev)->NAME).next = ((N)->NAME).next;	\
+	} else {							\
+		(BASE).start = ((N)->NAME).next;			\
+	}								\
+	UT_LIST_REMOVE_CLEAR(NAME, N);					\
+} while (0)
+
+/************************************************************************
+Gets the next node in a two-way list. NAME is the name of the list
+and N is pointer to a node. */
+
+#define UT_LIST_GET_NEXT(NAME, N)\
+	(((N)->NAME).next)
+
+/************************************************************************
+Gets the previous node in a two-way list. NAME is the name of the list
+and N is pointer to a node. */
+
+#define UT_LIST_GET_PREV(NAME, N)\
+	(((N)->NAME).prev)
+
+/************************************************************************
+Alternative macro to get the number of nodes in a two-way list, i.e.,
+its length. BASE is the base node (not a pointer to it). */
+
+#define UT_LIST_GET_LEN(BASE)\
+	(BASE).count
+
+/************************************************************************
+Gets the first node in a two-way list, or returns NULL,
+if the list is empty. BASE is the base node (not a pointer to it). */
+
+#define UT_LIST_GET_FIRST(BASE)\
+	(BASE).start
+
+/************************************************************************
+Gets the last node in a two-way list, or returns NULL,
+if the list is empty. BASE is the base node (not a pointer to it). */
+
+#define UT_LIST_GET_LAST(BASE)\
+	(BASE).end
+
+/************************************************************************
+Checks the consistency of a two-way list. NAME is the name of the list,
+TYPE is the node type, and BASE is the base node (not a pointer to it). */
+
+#define UT_LIST_VALIDATE(NAME, TYPE, BASE)\
+{\
+	ulint	ut_list_i_313;\
+	TYPE *	ut_list_node_313;\
+\
+	ut_list_node_313 = (BASE).start;\
+\
+	for (ut_list_i_313 = 0; ut_list_i_313 < (BASE).count;\
+						ut_list_i_313++) {\
+		ut_a(ut_list_node_313);\
+		ut_list_node_313 = (ut_list_node_313->NAME).next;\
+	}\
+\
+	ut_a(ut_list_node_313 == NULL);\
+\
+	ut_list_node_313 = (BASE).end;\
+\
+	for (ut_list_i_313 = 0; ut_list_i_313 < (BASE).count;\
+						ut_list_i_313++) {\
+		ut_a(ut_list_node_313);\
+		ut_list_node_313 = (ut_list_node_313->NAME).prev;\
+	}\
+\
+	ut_a(ut_list_node_313 == NULL);\
+}\
+
+
+#endif
+
diff --git a/storage/xtradb/include/ut0mem.h b/storage/xtradb/include/ut0mem.h
new file mode 100644
index 00000000000..f8dec99ed4a
--- /dev/null
+++ b/storage/xtradb/include/ut0mem.h
@@ -0,0 +1,271 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/***********************************************************************
+Memory primitives
+
+Created 5/30/1994 Heikki Tuuri
+************************************************************************/
+
+#ifndef ut0mem_h
+#define ut0mem_h
+
+#include "univ.i"
+#include "os0sync.h"
+#include <string.h>
+
+/* The total amount of memory currently allocated from the operating
+system with os_mem_alloc_large() or malloc().  Does not count malloc()
+if srv_use_sys_malloc is set.  Protected by ut_list_mutex. */
+extern ulint		ut_total_allocated_memory;
+
+/* Mutex protecting ut_total_allocated_memory and ut_mem_block_list */
+extern os_fast_mutex_t	ut_list_mutex;
+
+UNIV_INLINE
+void*
+ut_memcpy(void* dest, const void* sour, ulint n);
+
+UNIV_INLINE
+void*
+ut_memmove(void* dest, const void* sour, ulint n);
+
+UNIV_INLINE
+int
+ut_memcmp(const void* str1, const void* str2, ulint n);
+
+/**************************************************************************
+Initializes the mem block list at database startup. */
+UNIV_INTERN
+void
+ut_mem_init(void);
+/*=============*/
+
+/**************************************************************************
+Allocates memory. Sets it also to zero if UNIV_SET_MEM_TO_ZERO is
+defined and set_to_zero is TRUE. */
+UNIV_INTERN
+void*
+ut_malloc_low(
+/*==========*/
+					/* out, own: allocated memory */
+	ulint	n,			/* in: number of bytes to allocate */
+	ibool	set_to_zero,		/* in: TRUE if allocated memory
+					should be set to zero if
+					UNIV_SET_MEM_TO_ZERO is defined */
+	ibool	assert_on_error);	/* in: if TRUE, we crash mysqld if
+					the memory cannot be allocated */
+/**************************************************************************
+Allocates memory. Sets it also to zero if UNIV_SET_MEM_TO_ZERO is
+defined. */
+UNIV_INTERN
+void*
+ut_malloc(
+/*======*/
+			/* out, own: allocated memory */
+	ulint	n);	/* in: number of bytes to allocate */
+/**************************************************************************
+Tests if malloc of n bytes would succeed. ut_malloc() asserts if memory runs
+out. It cannot be used if we want to return an error message. Prints to
+stderr a message if fails. */
+UNIV_INTERN
+ibool
+ut_test_malloc(
+/*===========*/
+			/* out: TRUE if succeeded */
+	ulint	n);	/* in: try to allocate this many bytes */
+/**************************************************************************
+Frees a memory block allocated with ut_malloc. */
+UNIV_INTERN
+void
+ut_free(
+/*====*/
+	void* ptr);  /* in, own: memory block */
+/**************************************************************************
+Implements realloc. This is needed by /pars/lexyy.c. Otherwise, you should not
+use this function because the allocation functions in mem0mem.h are the
+recommended ones in InnoDB.
+
+man realloc in Linux, 2004:
+
+       realloc()  changes the size of the memory block pointed to
+       by ptr to size bytes.  The contents will be  unchanged  to
+       the minimum of the old and new sizes; newly allocated mem­
+       ory will be uninitialized.  If ptr is NULL,  the	 call  is
+       equivalent  to malloc(size); if size is equal to zero, the
+       call is equivalent to free(ptr).	 Unless ptr is	NULL,  it
+       must  have  been	 returned by an earlier call to malloc(),
+       calloc() or realloc().
+
+RETURN VALUE
+       realloc() returns a pointer to the newly allocated memory,
+       which is suitably aligned for any kind of variable and may
+       be different from ptr, or NULL if the  request  fails.  If
+       size  was equal to 0, either NULL or a pointer suitable to
+       be passed to free() is returned.	 If realloc()  fails  the
+       original	 block	is  left  untouched  - it is not freed or
+       moved. */
+UNIV_INTERN
+void*
+ut_realloc(
+/*=======*/
+			/* out, own: pointer to new mem block or NULL */
+	void*	ptr,	/* in: pointer to old block or NULL */
+	ulint	size);	/* in: desired size */
+/**************************************************************************
+Frees in shutdown all allocated memory not freed yet. */
+UNIV_INTERN
+void
+ut_free_all_mem(void);
+/*=================*/
+
+UNIV_INLINE
+char*
+ut_strcpy(char* dest, const char* sour);
+
+UNIV_INLINE
+ulint
+ut_strlen(const char* str);
+
+UNIV_INLINE
+int
+ut_strcmp(const char* str1, const char* str2);
+
+/**************************************************************************
+Copies up to size - 1 characters from the NUL-terminated string src to
+dst, NUL-terminating the result. Returns strlen(src), so truncation
+occurred if the return value >= size. */
+UNIV_INTERN
+ulint
+ut_strlcpy(
+/*=======*/
+				/* out: strlen(src) */
+	char*		dst,	/* in: destination buffer */
+	const char*	src,	/* in: source buffer */
+	ulint		size);	/* in: size of destination buffer */
+
+/**************************************************************************
+Like ut_strlcpy, but if src doesn't fit in dst completely, copies the last
+(size - 1) bytes of src, not the first. */
+UNIV_INTERN
+ulint
+ut_strlcpy_rev(
+/*===========*/
+				/* out: strlen(src) */
+	char*		dst,	/* in: destination buffer */
+	const char*	src,	/* in: source buffer */
+	ulint		size);	/* in: size of destination buffer */
+
+/**************************************************************************
+Compute strlen(ut_strcpyq(str, q)). */
+UNIV_INLINE
+ulint
+ut_strlenq(
+/*=======*/
+				/* out: length of the string when quoted */
+	const char*	str,	/* in: null-terminated string */
+	char		q);	/* in: the quote character */
+
+/**************************************************************************
+Make a quoted copy of a NUL-terminated string.	Leading and trailing
+quotes will not be included; only embedded quotes will be escaped.
+See also ut_strlenq() and ut_memcpyq(). */
+UNIV_INTERN
+char*
+ut_strcpyq(
+/*=======*/
+				/* out: pointer to end of dest */
+	char*		dest,	/* in: output buffer */
+	char		q,	/* in: the quote character */
+	const char*	src);	/* in: null-terminated string */
+
+/**************************************************************************
+Make a quoted copy of a fixed-length string.  Leading and trailing
+quotes will not be included; only embedded quotes will be escaped.
+See also ut_strlenq() and ut_strcpyq(). */
+UNIV_INTERN
+char*
+ut_memcpyq(
+/*=======*/
+				/* out: pointer to end of dest */
+	char*		dest,	/* in: output buffer */
+	char		q,	/* in: the quote character */
+	const char*	src,	/* in: string to be quoted */
+	ulint		len);	/* in: length of src */
+
+/**************************************************************************
+Return the number of times s2 occurs in s1. Overlapping instances of s2
+are only counted once. */
+UNIV_INTERN
+ulint
+ut_strcount(
+/*========*/
+				/* out: the number of times s2 occurs in s1 */
+	const char*	s1,	/* in: string to search in */
+	const char*	s2);	/* in: string to search for */
+
+/**************************************************************************
+Replace every occurrence of s1 in str with s2. Overlapping instances of s1
+are only replaced once. */
+UNIV_INTERN
+char*
+ut_strreplace(
+/*==========*/
+				/* out, own: modified string, must be
+				freed with mem_free() */
+	const char*	str,	/* in: string to operate on */
+	const char*	s1,	/* in: string to replace */
+	const char*	s2);	/* in: string to replace s1 with */
+
+/**************************************************************************
+Converts a raw binary data to a '\0'-terminated hex string. The output is
+truncated if there is not enough space in "hex", make sure "hex_size" is at
+least (2 * raw_size + 1) if you do not want this to happen. Returns the
+actual number of characters written to "hex" (including the '\0'). */
+UNIV_INLINE
+ulint
+ut_raw_to_hex(
+/*==========*/
+					/* out: number of chars written */
+	const void*	raw,		/* in: raw data */
+	ulint		raw_size,	/* in: "raw" length in bytes */
+	char*		hex,		/* out: hex string */
+	ulint		hex_size);	/* in: "hex" size in bytes */
+
+/***********************************************************************
+Adds single quotes to the start and end of string and escapes any quotes
+by doubling them. Returns the number of bytes that were written to "buf"
+(including the terminating '\0'). If buf_size is too small then the
+trailing bytes from "str" are discarded. */
+UNIV_INLINE
+ulint
+ut_str_sql_format(
+/*==============*/
+					/* out: number of bytes
+					that were written */
+	const char*	str,		/* in: string */
+	ulint		str_len,	/* in: string length in bytes */
+	char*		buf,		/* out: output buffer */
+	ulint		buf_size);	/* in: output buffer size
+					in bytes */
+
+#ifndef UNIV_NONINL
+#include "ut0mem.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/ut0mem.ic b/storage/xtradb/include/ut0mem.ic
new file mode 100644
index 00000000000..5078c721706
--- /dev/null
+++ b/storage/xtradb/include/ut0mem.ic
@@ -0,0 +1,308 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/***********************************************************************
+Memory primitives
+
+Created 5/30/1994 Heikki Tuuri
+************************************************************************/
+
+#include "ut0byte.h"
+#include "mach0data.h"
+
+UNIV_INLINE
+void*
+ut_memcpy(void* dest, const void* sour, ulint n)
+{
+	return(memcpy(dest, sour, n));
+}
+
+UNIV_INLINE
+void*
+ut_memmove(void* dest, const void* sour, ulint n)
+{
+	return(memmove(dest, sour, n));
+}
+
+UNIV_INLINE
+int
+ut_memcmp(const void* str1, const void* str2, ulint n)
+{
+	return(memcmp(str1, str2, n));
+}
+
+UNIV_INLINE
+char*
+ut_strcpy(char* dest, const char* sour)
+{
+	return(strcpy(dest, sour));
+}
+
+UNIV_INLINE
+ulint
+ut_strlen(const char* str)
+{
+	return(strlen(str));
+}
+
+UNIV_INLINE
+int
+ut_strcmp(const char* str1, const char* str2)
+{
+	return(strcmp(str1, str2));
+}
+
+/**************************************************************************
+Compute strlen(ut_strcpyq(str, q)). */
+UNIV_INLINE
+ulint
+ut_strlenq(
+/*=======*/
+				/* out: length of the string when quoted */
+	const char*	str,	/* in: null-terminated string */
+	char		q)	/* in: the quote character */
+{
+	ulint len;
+
+	for (len = 0; *str; len++, str++) {
+		if (*str == q) {
+			len++;
+		}
+	}
+
+	return(len);
+}
+
+/**************************************************************************
+Converts a raw binary data to a '\0'-terminated hex string. The output is
+truncated if there is not enough space in "hex", make sure "hex_size" is at
+least (2 * raw_size + 1) if you do not want this to happen. Returns the
+actual number of characters written to "hex" (including the '\0'). */
+UNIV_INLINE
+ulint
+ut_raw_to_hex(
+/*==========*/
+					/* out: number of chars written */
+	const void*	raw,		/* in: raw data */
+	ulint		raw_size,	/* in: "raw" length in bytes */
+	char*		hex,		/* out: hex string */
+	ulint		hex_size)	/* in: "hex" size in bytes */
+{
+
+#ifdef WORDS_BIGENDIAN
+
+#define MK_UINT16(a, b) (((uint16) (a)) << 8 | (uint16) (b))
+
+#define UINT16_GET_A(u)	((unsigned char) ((u) >> 8))
+#define UINT16_GET_B(u)	((unsigned char) ((u) & 0xFF))
+
+#else /* WORDS_BIGENDIAN */
+
+#define MK_UINT16(a, b) (((uint16) (b)) << 8 | (uint16) (a))
+
+#define UINT16_GET_A(u)	((unsigned char) ((u) & 0xFF))
+#define UINT16_GET_B(u)	((unsigned char) ((u) >> 8))
+
+#endif /* WORDS_BIGENDIAN */
+
+#define MK_ALL_UINT16_WITH_A(a)	\
+	MK_UINT16(a, '0'),	\
+	MK_UINT16(a, '1'),	\
+	MK_UINT16(a, '2'),	\
+	MK_UINT16(a, '3'),	\
+	MK_UINT16(a, '4'),	\
+	MK_UINT16(a, '5'),	\
+	MK_UINT16(a, '6'),	\
+	MK_UINT16(a, '7'),	\
+	MK_UINT16(a, '8'),	\
+	MK_UINT16(a, '9'),	\
+	MK_UINT16(a, 'A'),	\
+	MK_UINT16(a, 'B'),	\
+	MK_UINT16(a, 'C'),	\
+	MK_UINT16(a, 'D'),	\
+	MK_UINT16(a, 'E'),	\
+	MK_UINT16(a, 'F')
+
+	static const uint16	hex_map[256] = {
+		MK_ALL_UINT16_WITH_A('0'),
+		MK_ALL_UINT16_WITH_A('1'),
+		MK_ALL_UINT16_WITH_A('2'),
+		MK_ALL_UINT16_WITH_A('3'),
+		MK_ALL_UINT16_WITH_A('4'),
+		MK_ALL_UINT16_WITH_A('5'),
+		MK_ALL_UINT16_WITH_A('6'),
+		MK_ALL_UINT16_WITH_A('7'),
+		MK_ALL_UINT16_WITH_A('8'),
+		MK_ALL_UINT16_WITH_A('9'),
+		MK_ALL_UINT16_WITH_A('A'),
+		MK_ALL_UINT16_WITH_A('B'),
+		MK_ALL_UINT16_WITH_A('C'),
+		MK_ALL_UINT16_WITH_A('D'),
+		MK_ALL_UINT16_WITH_A('E'),
+		MK_ALL_UINT16_WITH_A('F')
+	};
+	const unsigned char*	rawc;
+	ulint			read_bytes;
+	ulint			write_bytes;
+	ulint			i;
+
+	rawc = (const unsigned char*) raw;
+
+	if (hex_size == 0) {
+
+		return(0);
+	}
+
+	if (hex_size <= 2 * raw_size) {
+
+		read_bytes = hex_size / 2;
+		write_bytes = hex_size;
+	} else {
+
+		read_bytes = raw_size;
+		write_bytes = 2 * raw_size + 1;
+	}
+
+#define LOOP_READ_BYTES(ASSIGN)			\
+	for (i = 0; i < read_bytes; i++) {	\
+		ASSIGN;				\
+		hex += 2;			\
+		rawc++;				\
+	}
+
+	if (ut_align_offset(hex, 2) == 0) {
+
+		LOOP_READ_BYTES(
+			*(uint16*) hex = hex_map[*rawc]
+		);
+	} else {
+
+		LOOP_READ_BYTES(
+			*hex       = UINT16_GET_A(hex_map[*rawc]);
+			*(hex + 1) = UINT16_GET_B(hex_map[*rawc])
+		);
+	}
+
+	if (hex_size <= 2 * raw_size && hex_size % 2 == 0) {
+
+		hex--;
+	}
+
+	*hex = '\0';
+
+	return(write_bytes);
+}
+
+/***********************************************************************
+Adds single quotes to the start and end of string and escapes any quotes
+by doubling them. Returns the number of bytes that were written to "buf"
+(including the terminating '\0'). If buf_size is too small then the
+trailing bytes from "str" are discarded. */
+UNIV_INLINE
+ulint
+ut_str_sql_format(
+/*==============*/
+					/* out: number of bytes
+					that were written */
+	const char*	str,		/* in: string */
+	ulint		str_len,	/* in: string length in bytes */
+	char*		buf,		/* out: output buffer */
+	ulint		buf_size)	/* in: output buffer size
+					in bytes */
+{
+	ulint	str_i;
+	ulint	buf_i;
+
+	buf_i = 0;
+
+	switch (buf_size) {
+	case 3:
+
+		if (str_len == 0) {
+
+			buf[buf_i] = '\'';
+			buf_i++;
+			buf[buf_i] = '\'';
+			buf_i++;
+		}
+		/* FALLTHROUGH */
+	case 2:
+	case 1:
+
+		buf[buf_i] = '\0';
+		buf_i++;
+		/* FALLTHROUGH */
+	case 0:
+
+		return(buf_i);
+	}
+
+	/* buf_size >= 4 */
+
+	buf[0] = '\'';
+	buf_i = 1;
+
+	for (str_i = 0; str_i < str_len; str_i++) {
+
+		char	ch;
+
+		if (buf_size - buf_i == 2) {
+
+			break;
+		}
+
+		ch = str[str_i];
+
+		switch (ch) {
+		case '\0':
+
+			if (UNIV_UNLIKELY(buf_size - buf_i < 4)) {
+
+				goto func_exit;
+			}
+			buf[buf_i] = '\\';
+			buf_i++;
+			buf[buf_i] = '0';
+			buf_i++;
+			break;
+		case '\'':
+		case '\\':
+
+			if (UNIV_UNLIKELY(buf_size - buf_i < 4)) {
+
+				goto func_exit;
+			}
+			buf[buf_i] = ch;
+			buf_i++;
+			/* FALLTHROUGH */
+		default:
+
+			buf[buf_i] = ch;
+			buf_i++;
+		}
+	}
+
+func_exit:
+
+	buf[buf_i] = '\'';
+	buf_i++;
+	buf[buf_i] = '\0';
+	buf_i++;
+
+	return(buf_i);
+}
diff --git a/storage/xtradb/include/ut0rnd.h b/storage/xtradb/include/ut0rnd.h
new file mode 100644
index 00000000000..b9e23d7cd14
--- /dev/null
+++ b/storage/xtradb/include/ut0rnd.h
@@ -0,0 +1,142 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/**********************************************************************
+Random numbers and hashing
+
+Created 1/20/1994 Heikki Tuuri
+***********************************************************************/
+
+#ifndef ut0rnd_h
+#define ut0rnd_h
+
+#include "univ.i"
+
+#include "ut0byte.h"
+
+/* The 'character code' for end of field or string (used
+in folding records */
+#define UT_END_OF_FIELD		257
+
+/************************************************************
+This is used to set the random number seed. */
+UNIV_INLINE
+void
+ut_rnd_set_seed(
+/*============*/
+	ulint	 seed);		 /* in: seed */
+/************************************************************
+The following function generates a series of 'random' ulint integers. */
+UNIV_INLINE
+ulint
+ut_rnd_gen_next_ulint(
+/*==================*/
+			/* out: the next 'random' number */
+	ulint	rnd);	/* in: the previous random number value */
+/*************************************************************
+The following function generates 'random' ulint integers which
+enumerate the value space (let there be N of them) of ulint integers
+in a pseudo-random fashion. Note that the same integer is repeated
+always after N calls to the generator. */
+UNIV_INLINE
+ulint
+ut_rnd_gen_ulint(void);
+/*==================*/
+			/* out: the 'random' number */
+/************************************************************
+Generates a random integer from a given interval. */
+UNIV_INLINE
+ulint
+ut_rnd_interval(
+/*============*/
+			/* out: the 'random' number */
+	ulint	low,	/* in: low limit; can generate also this value */
+	ulint	high);	/* in: high limit; can generate also this value */
+/*************************************************************
+Generates a random iboolean value. */
+UNIV_INLINE
+ibool
+ut_rnd_gen_ibool(void);
+/*=================*/
+			/* out: the random value */
+/***********************************************************
+The following function generates a hash value for a ulint integer
+to a hash table of size table_size, which should be a prime or some
+random number to work reliably. */
+UNIV_INLINE
+ulint
+ut_hash_ulint(
+/*==========*/
+				/* out: hash value */
+	ulint	 key,		/* in: value to be hashed */
+	ulint	 table_size);	/* in: hash table size */
+/*****************************************************************
+Folds a pair of ulints. */
+UNIV_INLINE
+ulint
+ut_fold_ulint_pair(
+/*===============*/
+			/* out: folded value */
+	ulint	n1,	/* in: ulint */
+	ulint	n2)	/* in: ulint */
+	__attribute__((const));
+/*****************************************************************
+Folds a dulint. */
+UNIV_INLINE
+ulint
+ut_fold_dulint(
+/*===========*/
+			/* out: folded value */
+	dulint	d)	/* in: dulint */
+	__attribute__((const));
+/*****************************************************************
+Folds a character string ending in the null character. */
+UNIV_INLINE
+ulint
+ut_fold_string(
+/*===========*/
+				/* out: folded value */
+	const char*	str)	/* in: null-terminated string */
+	__attribute__((pure));
+/*****************************************************************
+Folds a binary string. */
+UNIV_INLINE
+ulint
+ut_fold_binary(
+/*===========*/
+				/* out: folded value */
+	const byte*	str,	/* in: string of bytes */
+	ulint		len)	/* in: length */
+	__attribute__((pure));
+/***************************************************************
+Looks for a prime number slightly greater than the given argument.
+The prime is chosen so that it is not near any power of 2. */
+UNIV_INTERN
+ulint
+ut_find_prime(
+/*==========*/
+			/* out: prime */
+	ulint	n)	/* in: positive number > 100 */
+	__attribute__((const));
+
+
+#ifndef UNIV_NONINL
+#include "ut0rnd.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/ut0rnd.ic b/storage/xtradb/include/ut0rnd.ic
new file mode 100644
index 00000000000..d72100d16a1
--- /dev/null
+++ b/storage/xtradb/include/ut0rnd.ic
@@ -0,0 +1,228 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************************
+Random numbers and hashing
+
+Created 5/30/1994 Heikki Tuuri
+*******************************************************************/
+
+#define UT_HASH_RANDOM_MASK	1463735687
+#define UT_HASH_RANDOM_MASK2	1653893711
+#define UT_RND1			151117737
+#define UT_RND2			119785373
+#define UT_RND3			 85689495
+#define UT_RND4			 76595339
+#define UT_SUM_RND2		 98781234
+#define UT_SUM_RND3		126792457
+#define UT_SUM_RND4		 63498502
+#define UT_XOR_RND1		187678878
+#define UT_XOR_RND2		143537923
+
+extern	ulint	 ut_rnd_ulint_counter;
+
+/************************************************************
+This is used to set the random number seed. */
+UNIV_INLINE
+void
+ut_rnd_set_seed(
+/*============*/
+	ulint	 seed)		 /* in: seed */
+{
+	ut_rnd_ulint_counter = seed;
+}
+
+/************************************************************
+The following function generates a series of 'random' ulint integers. */
+UNIV_INLINE
+ulint
+ut_rnd_gen_next_ulint(
+/*==================*/
+			/* out: the next 'random' number */
+	ulint	rnd)	/* in: the previous random number value */
+{
+	ulint	n_bits;
+
+	n_bits = 8 * sizeof(ulint);
+
+	rnd = UT_RND2 * rnd + UT_SUM_RND3;
+	rnd = UT_XOR_RND1 ^ rnd;
+	rnd = (rnd << 20) + (rnd >> (n_bits - 20));
+	rnd = UT_RND3 * rnd + UT_SUM_RND4;
+	rnd = UT_XOR_RND2 ^ rnd;
+	rnd = (rnd << 20) + (rnd >> (n_bits - 20));
+	rnd = UT_RND1 * rnd + UT_SUM_RND2;
+
+	return(rnd);
+}
+
+/************************************************************
+The following function generates 'random' ulint integers which
+enumerate the value space of ulint integers in a pseudo random
+fashion. Note that the same integer is repeated always after
+2 to power 32 calls to the generator (if ulint is 32-bit). */
+UNIV_INLINE
+ulint
+ut_rnd_gen_ulint(void)
+/*==================*/
+			/* out: the 'random' number */
+{
+	ulint	rnd;
+	ulint	n_bits;
+
+	n_bits = 8 * sizeof(ulint);
+
+	ut_rnd_ulint_counter = UT_RND1 * ut_rnd_ulint_counter + UT_RND2;
+
+	rnd = ut_rnd_gen_next_ulint(ut_rnd_ulint_counter);
+
+	return(rnd);
+}
+
+/************************************************************
+Generates a random integer from a given interval. */
+UNIV_INLINE
+ulint
+ut_rnd_interval(
+/*============*/
+			/* out: the 'random' number */
+	ulint	low,	/* in: low limit; can generate also this value */
+	ulint	high)	/* in: high limit; can generate also this value */
+{
+	ulint	rnd;
+
+	ut_ad(high >= low);
+
+	if (low == high) {
+
+		return(low);
+	}
+
+	rnd = ut_rnd_gen_ulint();
+
+	return(low + (rnd % (high - low + 1)));
+}
+
+/*************************************************************
+Generates a random iboolean value. */
+UNIV_INLINE
+ibool
+ut_rnd_gen_ibool(void)
+/*=================*/
+			/* out: the random value */
+{
+	ulint	 x;
+
+	x = ut_rnd_gen_ulint();
+
+	if (((x >> 20) + (x >> 15)) & 1) {
+
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
+
+/***********************************************************
+The following function generates a hash value for a ulint integer
+to a hash table of size table_size, which should be a prime
+or some random number for the hash table to work reliably. */
+UNIV_INLINE
+ulint
+ut_hash_ulint(
+/*==========*/
+				/* out: hash value */
+	ulint	 key,		/* in: value to be hashed */
+	ulint	 table_size)	/* in: hash table size */
+{
+	key = key ^ UT_HASH_RANDOM_MASK2;
+
+	return(key % table_size);
+}
+
+/*****************************************************************
+Folds a pair of ulints. */
+UNIV_INLINE
+ulint
+ut_fold_ulint_pair(
+/*===============*/
+			/* out: folded value */
+	ulint	n1,	/* in: ulint */
+	ulint	n2)	/* in: ulint */
+{
+	return(((((n1 ^ n2 ^ UT_HASH_RANDOM_MASK2) << 8) + n1)
+		^ UT_HASH_RANDOM_MASK) + n2);
+}
+
+/*****************************************************************
+Folds a dulint. */
+UNIV_INLINE
+ulint
+ut_fold_dulint(
+/*===========*/
+			/* out: folded value */
+	dulint	d)	/* in: dulint */
+{
+	return(ut_fold_ulint_pair(ut_dulint_get_low(d),
+				  ut_dulint_get_high(d)));
+}
+
+/*****************************************************************
+Folds a character string ending in the null character. */
+UNIV_INLINE
+ulint
+ut_fold_string(
+/*===========*/
+				/* out: folded value */
+	const char*	str)	/* in: null-terminated string */
+{
+	ulint	fold = 0;
+
+	ut_ad(str);
+
+	while (*str != '\0') {
+		fold = ut_fold_ulint_pair(fold, (ulint)(*str));
+		str++;
+	}
+
+	return(fold);
+}
+
+/*****************************************************************
+Folds a binary string. */
+UNIV_INLINE
+ulint
+ut_fold_binary(
+/*===========*/
+				/* out: folded value */
+	const byte*	str,	/* in: string of bytes */
+	ulint		len)	/* in: length */
+{
+	const byte*	str_end	= str + len;
+	ulint		fold = 0;
+
+	ut_ad(str || !len);
+
+	while (str < str_end) {
+		fold = ut_fold_ulint_pair(fold, (ulint)(*str));
+
+		str++;
+	}
+
+	return(fold);
+}
diff --git a/storage/xtradb/include/ut0sort.h b/storage/xtradb/include/ut0sort.h
new file mode 100644
index 00000000000..5fd5db54832
--- /dev/null
+++ b/storage/xtradb/include/ut0sort.h
@@ -0,0 +1,105 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/**********************************************************************
+Sort utility
+
+Created 11/9/1995 Heikki Tuuri
+***********************************************************************/
+
+#ifndef ut0sort_h
+#define ut0sort_h
+
+#include "univ.i"
+
+/* This module gives a macro definition of the body of
+a standard sort function for an array of elements of any
+type. The comparison function is given as a parameter to
+the macro. The sort algorithm is mergesort which has logarithmic
+worst case.
+*/
+
+/***********************************************************************
+This macro expands to the body of a standard sort function.
+The sort function uses mergesort and must be defined separately
+for each type of array.
+Also the comparison function has to be defined individually
+for each array cell type. SORT_FUN is the sort function name.
+The function takes the array to be sorted (ARR),
+the array of auxiliary space (AUX_ARR) of same size,
+and the low (LOW), inclusive, and high (HIGH), noninclusive,
+limits for the sort interval as arguments.
+CMP_FUN is the comparison function name. It takes as arguments
+two elements from the array and returns 1, if the first is bigger,
+0 if equal, and -1 if the second bigger. */
+
+#define UT_SORT_FUNCTION_BODY(SORT_FUN, ARR, AUX_ARR, LOW, HIGH, CMP_FUN)\
+{\
+	ulint		ut_sort_mid77;\
+	ulint		ut_sort_i77;\
+	ulint		ut_sort_low77;\
+	ulint		ut_sort_high77;\
+\
+	ut_ad((LOW) < (HIGH));\
+	ut_ad(ARR);\
+	ut_ad(AUX_ARR);\
+\
+	if ((LOW) == (HIGH) - 1) {\
+		return;\
+	} else if ((LOW) == (HIGH) - 2) {\
+		if (CMP_FUN((ARR)[LOW], (ARR)[(HIGH) - 1]) > 0) {\
+			(AUX_ARR)[LOW] = (ARR)[LOW];\
+			(ARR)[LOW] = (ARR)[(HIGH) - 1];\
+			(ARR)[(HIGH) - 1] = (AUX_ARR)[LOW];\
+		}\
+		return;\
+	}\
+\
+	ut_sort_mid77 = ((LOW) + (HIGH)) / 2;\
+\
+	SORT_FUN((ARR), (AUX_ARR), (LOW), ut_sort_mid77);\
+	SORT_FUN((ARR), (AUX_ARR), ut_sort_mid77, (HIGH));\
+\
+	ut_sort_low77 = (LOW);\
+	ut_sort_high77 = ut_sort_mid77;\
+\
+	for (ut_sort_i77 = (LOW); ut_sort_i77 < (HIGH); ut_sort_i77++) {\
+\
+		if (ut_sort_low77 >= ut_sort_mid77) {\
+			(AUX_ARR)[ut_sort_i77] = (ARR)[ut_sort_high77];\
+			ut_sort_high77++;\
+		} else if (ut_sort_high77 >= (HIGH)) {\
+			(AUX_ARR)[ut_sort_i77] = (ARR)[ut_sort_low77];\
+			ut_sort_low77++;\
+		} else if (CMP_FUN((ARR)[ut_sort_low77],\
+				   (ARR)[ut_sort_high77]) > 0) {\
+			(AUX_ARR)[ut_sort_i77] = (ARR)[ut_sort_high77];\
+			ut_sort_high77++;\
+		} else {\
+			(AUX_ARR)[ut_sort_i77] = (ARR)[ut_sort_low77];\
+			ut_sort_low77++;\
+		}\
+	}\
+\
+	memcpy((void*) ((ARR) + (LOW)), (AUX_ARR) + (LOW),\
+	       ((HIGH) - (LOW)) * sizeof *(ARR));\
+}\
+
+
+#endif
+
diff --git a/storage/xtradb/include/ut0ut.h b/storage/xtradb/include/ut0ut.h
new file mode 100644
index 00000000000..3ca14acd2ef
--- /dev/null
+++ b/storage/xtradb/include/ut0ut.h
@@ -0,0 +1,328 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/**********************************************************************
+Various utilities
+
+Created 1/20/1994 Heikki Tuuri
+***********************************************************************/
+
+#ifndef ut0ut_h
+#define ut0ut_h
+
+#include "univ.i"
+#include <time.h>
+#ifndef MYSQL_SERVER
+#include <ctype.h>
+#endif
+
+#define	TEMP_INDEX_PREFIX	'\377'	/* Index name prefix in fast index
+					creation */
+
+typedef time_t	ib_time_t;
+
+/*************************************************************************
+Delays execution for at most max_wait_us microseconds or returns earlier
+if cond becomes true; cond is evaluated every 2 ms. */
+
+#define UT_WAIT_FOR(cond, max_wait_us)				\
+do {								\
+	ullint	start_us;					\
+	start_us = ut_time_us(NULL);				\
+	while (!(cond) 						\
+	       && ut_time_us(NULL) - start_us < (max_wait_us)) {\
+								\
+		os_thread_sleep(2000 /* 2 ms */);		\
+	}							\
+} while (0)
+
+/************************************************************
+Gets the high 32 bits in a ulint. That is makes a shift >> 32,
+but since there seem to be compiler bugs in both gcc and Visual C++,
+we do this by a special conversion. */
+UNIV_INTERN
+ulint
+ut_get_high32(
+/*==========*/
+			/* out: a >> 32 */
+	ulint	a);	/* in: ulint */
+/**********************************************************
+Calculates the minimum of two ulints. */
+UNIV_INLINE
+ulint
+ut_min(
+/*===*/
+			/* out: minimum */
+	ulint	 n1,	/* in: first number */
+	ulint	 n2);	/* in: second number */
+/**********************************************************
+Calculates the maximum of two ulints. */
+UNIV_INLINE
+ulint
+ut_max(
+/*===*/
+			/* out: maximum */
+	ulint	 n1,	/* in: first number */
+	ulint	 n2);	/* in: second number */
+/********************************************************************
+Calculates minimum of two ulint-pairs. */
+UNIV_INLINE
+void
+ut_pair_min(
+/*========*/
+	ulint*	a,	/* out: more significant part of minimum */
+	ulint*	b,	/* out: less significant part of minimum */
+	ulint	a1,	/* in: more significant part of first pair */
+	ulint	b1,	/* in: less significant part of first pair */
+	ulint	a2,	/* in: more significant part of second pair */
+	ulint	b2);	/* in: less significant part of second pair */
+/**********************************************************
+Compares two ulints. */
+UNIV_INLINE
+int
+ut_ulint_cmp(
+/*=========*/
+			/* out: 1 if a > b, 0 if a == b, -1 if a < b */
+	ulint	a,	/* in: ulint */
+	ulint	b);	/* in: ulint */
+/***********************************************************
+Compares two pairs of ulints. */
+UNIV_INLINE
+int
+ut_pair_cmp(
+/*========*/
+			/* out: -1 if a < b, 0 if a == b,
+			1 if a > b */
+	ulint	a1,	/* in: more significant part of first pair */
+	ulint	a2,	/* in: less significant part of first pair */
+	ulint	b1,	/* in: more significant part of second pair */
+	ulint	b2);	/* in: less significant part of second pair */
+/*****************************************************************
+Determines if a number is zero or a power of two. */
+#define ut_is_2pow(n) UNIV_LIKELY(!((n) & ((n) - 1)))
+/*****************************************************************
+Calculates fast the remainder of n/m when m is a power of two. */
+#define ut_2pow_remainder(n, m) ((n) & ((m) - 1))
+/*****************************************************************
+Calculates the biggest multiple of m that is not bigger than n
+when m is a power of two.  In other words, rounds n down to m * k. */
+#define ut_2pow_round(n, m) ((n) & ~((m) - 1))
+#define ut_calc_align_down(n, m) ut_2pow_round(n, m)
+/************************************************************
+Calculates the smallest multiple of m that is not smaller than n
+when m is a power of two.  In other words, rounds n up to m * k. */
+#define ut_calc_align(n, m) (((n) + ((m) - 1)) & ~((m) - 1))
+/*****************************************************************
+Calculates fast the 2-logarithm of a number, rounded upward to an
+integer. */
+UNIV_INLINE
+ulint
+ut_2_log(
+/*=====*/
+			/* out: logarithm in the base 2, rounded upward */
+	ulint	n);	/* in: number */
+/*****************************************************************
+Calculates 2 to power n. */
+UNIV_INLINE
+ulint
+ut_2_exp(
+/*=====*/
+			/* out: 2 to power n */
+	ulint	n);	/* in: number */
+/*****************************************************************
+Calculates fast the number rounded up to the nearest power of 2. */
+UNIV_INTERN
+ulint
+ut_2_power_up(
+/*==========*/
+			/* out: first power of 2 which is >= n */
+	ulint	n)	/* in: number != 0 */
+	__attribute__((const));
+
+/* Determine how many bytes (groups of 8 bits) are needed to
+store the given number of bits. */
+#define UT_BITS_IN_BYTES(b) (((b) + 7) / 8)
+
+/**************************************************************
+Returns system time. We do not specify the format of the time returned:
+the only way to manipulate it is to use the function ut_difftime. */
+UNIV_INTERN
+ib_time_t
+ut_time(void);
+/*=========*/
+/**************************************************************
+Returns system time.
+Upon successful completion, the value 0 is returned; otherwise the
+value -1 is returned and the global variable errno is set to indicate the
+error. */
+UNIV_INTERN
+int
+ut_usectime(
+/*========*/
+			/* out: 0 on success, -1 otherwise */
+	ulint*	sec,	/* out: seconds since the Epoch */
+	ulint*	ms);	/* out: microseconds since the Epoch+*sec */
+
+/**************************************************************
+Returns the number of microseconds since epoch. Similar to
+time(3), the return value is also stored in *tloc, provided
+that tloc is non-NULL. */
+UNIV_INTERN
+ullint
+ut_time_us(
+/*=======*/
+			/* out: us since epoch */
+	ullint*	tloc);	/* out: us since epoch, if non-NULL */
+
+/**************************************************************
+Returns the difference of two times in seconds. */
+UNIV_INTERN
+double
+ut_difftime(
+/*========*/
+				/* out: time2 - time1 expressed in seconds */
+	ib_time_t	time2,	/* in: time */
+	ib_time_t	time1);	/* in: time */
+/**************************************************************
+Prints a timestamp to a file. */
+UNIV_INTERN
+void
+ut_print_timestamp(
+/*===============*/
+	FILE*  file); /* in: file where to print */
+/**************************************************************
+Sprintfs a timestamp to a buffer, 13..14 chars plus terminating NUL. */
+UNIV_INTERN
+void
+ut_sprintf_timestamp(
+/*=================*/
+	char*	buf); /* in: buffer where to sprintf */
+#ifdef UNIV_HOTBACKUP
+/**************************************************************
+Sprintfs a timestamp to a buffer with no spaces and with ':' characters
+replaced by '_'. */
+UNIV_INTERN
+void
+ut_sprintf_timestamp_without_extra_chars(
+/*=====================================*/
+	char*	buf); /* in: buffer where to sprintf */
+/**************************************************************
+Returns current year, month, day. */
+UNIV_INTERN
+void
+ut_get_year_month_day(
+/*==================*/
+	ulint*	year,	/* out: current year */
+	ulint*	month,	/* out: month */
+	ulint*	day);	/* out: day */
+#endif /* UNIV_HOTBACKUP */
+/*****************************************************************
+Runs an idle loop on CPU. The argument gives the desired delay
+in microseconds on 100 MHz Pentium + Visual C++. */
+UNIV_INTERN
+ulint
+ut_delay(
+/*=====*/
+			/* out: dummy value */
+	ulint	delay);	/* in: delay in microseconds on 100 MHz Pentium */
+/*****************************************************************
+Prints the contents of a memory buffer in hex and ascii. */
+UNIV_INTERN
+void
+ut_print_buf(
+/*=========*/
+	FILE*		file,	/* in: file where to print */
+	const void*	buf,	/* in: memory buffer */
+	ulint		len);	/* in: length of the buffer */
+
+/**************************************************************************
+Outputs a NUL-terminated file name, quoted with apostrophes. */
+UNIV_INTERN
+void
+ut_print_filename(
+/*==============*/
+	FILE*		f,	/* in: output stream */
+	const char*	name);	/* in: name to print */
+
+/* Forward declaration of transaction handle */
+struct trx_struct;
+
+/**************************************************************************
+Outputs a fixed-length string, quoted as an SQL identifier.
+If the string contains a slash '/', the string will be
+output as two identifiers separated by a period (.),
+as in SQL database_name.identifier. */
+UNIV_INTERN
+void
+ut_print_name(
+/*==========*/
+	FILE*		f,	/* in: output stream */
+	struct trx_struct*trx,	/* in: transaction */
+	ibool		table_id,/* in: TRUE=print a table name,
+				FALSE=print other identifier */
+	const char*	name);	/* in: name to print */
+
+/**************************************************************************
+Outputs a fixed-length string, quoted as an SQL identifier.
+If the string contains a slash '/', the string will be
+output as two identifiers separated by a period (.),
+as in SQL database_name.identifier. */
+UNIV_INTERN
+void
+ut_print_namel(
+/*===========*/
+	FILE*		f,	/* in: output stream */
+	struct trx_struct*trx,	/* in: transaction (NULL=no quotes) */
+	ibool		table_id,/* in: TRUE=print a table name,
+				FALSE=print other identifier */
+	const char*	name,	/* in: name to print */
+	ulint		namelen);/* in: length of name */
+
+/**************************************************************************
+Catenate files. */
+UNIV_INTERN
+void
+ut_copy_file(
+/*=========*/
+	FILE*	dest,	/* in: output file */
+	FILE*	src);	/* in: input file to be appended to output */
+
+/**************************************************************************
+snprintf(). */
+
+#ifdef __WIN__
+int
+ut_snprintf(
+				/* out: number of characters that would
+				have been printed if the size were
+				unlimited, not including the terminating
+				'\0'. */
+	char*		str,	/* out: string */
+	size_t		size,	/* in: str size */
+	const char*	fmt,	/* in: format */
+	...);			/* in: format values */
+#else
+#define ut_snprintf	snprintf
+#endif /* __WIN__ */
+
+#ifndef UNIV_NONINL
+#include "ut0ut.ic"
+#endif
+
+#endif
+
diff --git a/storage/xtradb/include/ut0ut.ic b/storage/xtradb/include/ut0ut.ic
new file mode 100644
index 00000000000..e4e0a2acce6
--- /dev/null
+++ b/storage/xtradb/include/ut0ut.ic
@@ -0,0 +1,161 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************************
+Various utilities
+
+Created 5/30/1994 Heikki Tuuri
+*******************************************************************/
+
+/**********************************************************
+Calculates the minimum of two ulints. */
+UNIV_INLINE
+ulint
+ut_min(
+/*===*/
+			/* out: minimum */
+	ulint	 n1,	/* in: first number */
+	ulint	 n2)	/* in: second number */
+{
+	return((n1 <= n2) ? n1 : n2);
+}
+
+/**********************************************************
+Calculates the maximum of two ulints. */
+UNIV_INLINE
+ulint
+ut_max(
+/*===*/
+			/* out: maximum */
+	ulint	 n1,	/* in: first number */
+	ulint	 n2)	/* in: second number */
+{
+	return((n1 <= n2) ? n2 : n1);
+}
+
+/********************************************************************
+Calculates minimum of two ulint-pairs. */
+UNIV_INLINE
+void
+ut_pair_min(
+/*========*/
+	ulint*	a,	/* out: more significant part of minimum */
+	ulint*	b,	/* out: less significant part of minimum */
+	ulint	a1,	/* in: more significant part of first pair */
+	ulint	b1,	/* in: less significant part of first pair */
+	ulint	a2,	/* in: more significant part of second pair */
+	ulint	b2)	/* in: less significant part of second pair */
+{
+	if (a1 == a2) {
+		*a = a1;
+		*b = ut_min(b1, b2);
+	} else if (a1 < a2) {
+		*a = a1;
+		*b = b1;
+	} else {
+		*a = a2;
+		*b = b2;
+	}
+}
+
+/**********************************************************
+Compares two ulints. */
+UNIV_INLINE
+int
+ut_ulint_cmp(
+/*=========*/
+			/* out: 1 if a > b, 0 if a == b, -1 if a < b */
+	ulint	a,	/* in: ulint */
+	ulint	b)	/* in: ulint */
+{
+	if (a < b) {
+		return(-1);
+	} else if (a == b) {
+		return(0);
+	} else {
+		return(1);
+	}
+}
+
+/***********************************************************
+Compares two pairs of ulints. */
+UNIV_INLINE
+int
+ut_pair_cmp(
+/*========*/
+			/* out: -1 if a < b, 0 if a == b, 1 if a > b */
+	ulint	a1,	/* in: more significant part of first pair */
+	ulint	a2,	/* in: less significant part of first pair */
+	ulint	b1,	/* in: more significant part of second pair */
+	ulint	b2)	/* in: less significant part of second pair */
+{
+	if (a1 > b1) {
+		return(1);
+	} else if (a1 < b1) {
+		return(-1);
+	} else if (a2 > b2) {
+		return(1);
+	} else if (a2 < b2) {
+		return(-1);
+	} else {
+		return(0);
+	}
+}
+
+/*****************************************************************
+Calculates fast the 2-logarithm of a number, rounded upward to an
+integer. */
+UNIV_INLINE
+ulint
+ut_2_log(
+/*=====*/
+			/* out: logarithm in the base 2, rounded upward */
+	ulint	n)	/* in: number != 0 */
+{
+	ulint	res;
+
+	res = 0;
+
+	ut_ad(n > 0);
+
+	n = n - 1;
+
+	for (;;) {
+		n = n / 2;
+
+		if (n == 0) {
+			break;
+		}
+
+		res++;
+	}
+
+	return(res + 1);
+}
+
+/*****************************************************************
+Calculates 2 to power n. */
+UNIV_INLINE
+ulint
+ut_2_exp(
+/*=====*/
+			/* out: 2 to power n */
+	ulint	n)	/* in: number */
+{
+	return((ulint) 1 << n);
+}
diff --git a/storage/xtradb/include/ut0vec.h b/storage/xtradb/include/ut0vec.h
new file mode 100644
index 00000000000..aeb7e168dc6
--- /dev/null
+++ b/storage/xtradb/include/ut0vec.h
@@ -0,0 +1,117 @@
+/*****************************************************************************
+
+Copyright (c) 2006, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+#ifndef IB_VECTOR_H
+#define IB_VECTOR_H
+
+#include "univ.i"
+#include "mem0mem.h"
+
+typedef struct ib_vector_struct ib_vector_t;
+
+/* An automatically resizing vector datatype with the following properties:
+
+ -Contains void* items.
+
+ -The items are owned by the caller.
+
+ -All memory allocation is done through a heap owned by the caller, who is
+ responsible for freeing it when done with the vector.
+
+ -When the vector is resized, the old memory area is left allocated since it
+ uses the same heap as the new memory area, so this is best used for
+ relatively small or short-lived uses.
+*/
+
+/********************************************************************
+Create a new vector with the given initial size. */
+UNIV_INTERN
+ib_vector_t*
+ib_vector_create(
+/*=============*/
+				/* out: vector */
+	mem_heap_t*	heap,	/* in: heap */
+	ulint		size);	/* in: initial size */
+
+/********************************************************************
+Push a new element to the vector, increasing its size if necessary. */
+UNIV_INTERN
+void
+ib_vector_push(
+/*===========*/
+	ib_vector_t*	vec,	/* in: vector */
+	void*		elem);	/* in: data element */
+
+/********************************************************************
+Get the number of elements in the vector. */
+UNIV_INLINE
+ulint
+ib_vector_size(
+/*===========*/
+				/* out: number of elements in vector */
+	const ib_vector_t*	vec);	/* in: vector */
+
+/********************************************************************
+Test whether a vector is empty or not. */
+UNIV_INLINE
+ibool
+ib_vector_is_empty(
+/*===============*/
+				/* out: TRUE if empty */
+	const ib_vector_t*	vec);	/* in: vector */
+
+/********************************************************************
+Get the n'th element. */
+UNIV_INLINE
+void*
+ib_vector_get(
+/*==========*/
+				/* out: n'th element */
+	ib_vector_t*	vec,	/* in: vector */
+	ulint		n);	/* in: element index to get */
+
+/********************************************************************
+Remove the last element from the vector. */
+UNIV_INLINE
+void*
+ib_vector_pop(
+/*==========*/
+	ib_vector_t*	vec);	/* in: vector */
+
+/********************************************************************
+Free the underlying heap of the vector. Note that vec is invalid
+after this call. */
+UNIV_INLINE
+void
+ib_vector_free(
+/*===========*/
+	ib_vector_t*	vec);	/* in,own: vector */
+
+/* See comment at beginning of file. */
+struct ib_vector_struct {
+	mem_heap_t*	heap;	/* heap */
+	void**		data;	/* data elements */
+	ulint		used;	/* number of elements currently used */
+	ulint		total;	/* number of elements allocated */
+};
+
+#ifndef UNIV_NONINL
+#include "ut0vec.ic"
+#endif
+
+#endif
diff --git a/storage/xtradb/include/ut0vec.ic b/storage/xtradb/include/ut0vec.ic
new file mode 100644
index 00000000000..b0e853717e3
--- /dev/null
+++ b/storage/xtradb/include/ut0vec.ic
@@ -0,0 +1,88 @@
+/*****************************************************************************
+
+Copyright (c) 2006, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/********************************************************************
+Get number of elements in vector. */
+UNIV_INLINE
+ulint
+ib_vector_size(
+/*===========*/
+				/* out: number of elements in vector */
+	const ib_vector_t*	vec)	/* in: vector */
+{
+	return(vec->used);
+}
+
+/********************************************************************
+Get n'th element. */
+UNIV_INLINE
+void*
+ib_vector_get(
+/*==========*/
+				/* out: n'th element */
+	ib_vector_t*	vec,	/* in: vector */
+	ulint		n)	/* in: element index to get */
+{
+	ut_a(n < vec->used);
+
+	return(vec->data[n]);
+}
+
+/********************************************************************
+Remove the last element from the vector. */
+UNIV_INLINE
+void*
+ib_vector_pop(
+/*==========*/
+				/* out: last vector element */
+	ib_vector_t*    vec)    /* in/out: vector */
+{
+	void*           elem;
+
+	ut_a(vec->used > 0);
+	--vec->used;
+	elem = vec->data[vec->used];
+
+	ut_d(vec->data[vec->used] = NULL);
+	UNIV_MEM_INVALID(&vec->data[vec->used], sizeof(*vec->data));
+
+	return(elem);
+}
+
+/********************************************************************
+Free the underlying heap of the vector. Note that vec is invalid
+after this call. */
+UNIV_INLINE
+void
+ib_vector_free(
+/*===========*/
+	ib_vector_t*    vec)    /* in, own: vector */
+{
+	mem_heap_free(vec->heap);
+}
+
+/********************************************************************
+Test whether a vector is empty or not. */
+UNIV_INLINE
+ibool
+ib_vector_is_empty(
+/*===============*/			/* out: TRUE if empty else FALSE */
+	const ib_vector_t*	vec)	/* in vector to test */
+{
+	return(ib_vector_size(vec) == 0);
+}
diff --git a/storage/xtradb/include/ut0wqueue.h b/storage/xtradb/include/ut0wqueue.h
new file mode 100644
index 00000000000..6bb80dad532
--- /dev/null
+++ b/storage/xtradb/include/ut0wqueue.h
@@ -0,0 +1,77 @@
+/*****************************************************************************
+
+Copyright (c) 2006, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/***********************************************************************
+A Work queue. Threads can add work items to the queue and other threads can
+wait for work items to be available and take them off the queue for
+processing.
+************************************************************************/
+
+#ifndef IB_WORK_QUEUE_H
+#define IB_WORK_QUEUE_H
+
+#include "ut0list.h"
+#include "mem0mem.h"
+#include "os0sync.h"
+#include "sync0types.h"
+
+typedef struct ib_wqueue_struct ib_wqueue_t;
+
+/********************************************************************
+Create a new work queue. */
+UNIV_INTERN
+ib_wqueue_t*
+ib_wqueue_create(void);
+/*===================*/
+			/* out: work queue */
+
+/********************************************************************
+Free a work queue. */
+UNIV_INTERN
+void
+ib_wqueue_free(
+/*===========*/
+	ib_wqueue_t*	wq);	/* in: work queue */
+
+/********************************************************************
+Add a work item to the queue. */
+UNIV_INTERN
+void
+ib_wqueue_add(
+/*==========*/
+	ib_wqueue_t*	wq,	/* in: work queue */
+	void*		item,	/* in: work item */
+	mem_heap_t*	heap);	/* in: memory heap to use for allocating the
+				list node */
+
+/********************************************************************
+Wait for a work item to appear in the queue. */
+UNIV_INTERN
+void*
+ib_wqueue_wait(
+				/* out: work item */
+	ib_wqueue_t*	wq);	/* in: work queue */
+
+/* Work queue. */
+struct ib_wqueue_struct {
+	mutex_t		mutex;	/* mutex protecting everything */
+	ib_list_t*	items;	/* work item list */
+	os_event_t	event;	/* event we use to signal additions to list */
+};
+
+#endif
diff --git a/storage/xtradb/lock/lock0iter.c b/storage/xtradb/lock/lock0iter.c
new file mode 100644
index 00000000000..e7a128d0db3
--- /dev/null
+++ b/storage/xtradb/lock/lock0iter.c
@@ -0,0 +1,113 @@
+/*****************************************************************************
+
+Copyright (c) 2007, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Lock queue iterator. Can iterate over table and record
+lock queues.
+
+Created July 16, 2007 Vasil Dimov
+*******************************************************/
+
+#define LOCK_MODULE_IMPLEMENTATION
+
+#include "univ.i"
+#include "lock0iter.h"
+#include "lock0lock.h"
+#include "lock0priv.h"
+#include "ut0dbg.h"
+#include "ut0lst.h"
+#ifdef UNIV_DEBUG
+# include "srv0srv.h" /* kernel_mutex */
+#endif /* UNIV_DEBUG */
+
+/***********************************************************************
+Initialize lock queue iterator so that it starts to iterate from
+"lock". bit_no specifies the record number within the heap where the
+record is stored. It can be undefined (ULINT_UNDEFINED) in two cases:
+1. If the lock is a table lock, thus we have a table lock queue;
+2. If the lock is a record lock and it is a wait lock. In this case
+   bit_no is calculated in this function by using
+   lock_rec_find_set_bit(). There is exactly one bit set in the bitmap
+   of a wait lock. */
+UNIV_INTERN
+void
+lock_queue_iterator_reset(
+/*======================*/
+	lock_queue_iterator_t*	iter,	/* out: iterator */
+	const lock_t*		lock,	/* in: lock to start from */
+	ulint			bit_no)	/* in: record number in the
+					heap */
+{
+	ut_ad(mutex_own(&kernel_mutex));
+
+	iter->current_lock = lock;
+
+	if (bit_no != ULINT_UNDEFINED) {
+
+		iter->bit_no = bit_no;
+	} else {
+
+		switch (lock_get_type_low(lock)) {
+		case LOCK_TABLE:
+			iter->bit_no = ULINT_UNDEFINED;
+			break;
+		case LOCK_REC:
+			iter->bit_no = lock_rec_find_set_bit(lock);
+			ut_a(iter->bit_no != ULINT_UNDEFINED);
+			break;
+		default:
+			ut_error;
+		}
+	}
+}
+
+/***********************************************************************
+Gets the previous lock in the lock queue, returns NULL if there are no
+more locks (i.e. the current lock is the first one). The iterator is
+receded (if not-NULL is returned). */
+UNIV_INTERN
+const lock_t*
+lock_queue_iterator_get_prev(
+/*=========================*/
+					/* out: previous lock or NULL */
+	lock_queue_iterator_t*	iter)	/* in/out: iterator */
+{
+	const lock_t*	prev_lock;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	switch (lock_get_type_low(iter->current_lock)) {
+	case LOCK_REC:
+		prev_lock = lock_rec_get_prev(
+			iter->current_lock, iter->bit_no);
+		break;
+	case LOCK_TABLE:
+		prev_lock = UT_LIST_GET_PREV(
+			un_member.tab_lock.locks, iter->current_lock);
+		break;
+	default:
+		ut_error;
+	}
+
+	if (prev_lock != NULL) {
+
+		iter->current_lock = prev_lock;
+	}
+
+	return(prev_lock);
+}
diff --git a/storage/xtradb/lock/lock0lock.c b/storage/xtradb/lock/lock0lock.c
new file mode 100644
index 00000000000..3730c66313d
--- /dev/null
+++ b/storage/xtradb/lock/lock0lock.c
@@ -0,0 +1,5680 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The transaction lock system
+
+Created 5/7/1996 Heikki Tuuri
+*******************************************************/
+
+#define LOCK_MODULE_IMPLEMENTATION
+
+#include "lock0lock.h"
+#include "lock0priv.h"
+
+#ifdef UNIV_NONINL
+#include "lock0lock.ic"
+#include "lock0priv.ic"
+#endif
+
+#include "ha_prototypes.h"
+#include "usr0sess.h"
+#include "trx0purge.h"
+#include "dict0mem.h"
+#include "trx0sys.h"
+
+/* Restricts the length of search we will do in the waits-for
+graph of transactions */
+#define LOCK_MAX_N_STEPS_IN_DEADLOCK_CHECK 1000000
+
+/* Restricts the recursion depth of the search we will do in the waits-for
+graph of transactions */
+#define LOCK_MAX_DEPTH_IN_DEADLOCK_CHECK 200
+
+/* When releasing transaction locks, this specifies how often we release
+the kernel mutex for a moment to give also others access to it */
+
+#define LOCK_RELEASE_KERNEL_INTERVAL	1000
+
+/* Safety margin when creating a new record lock: this many extra records
+can be inserted to the page without need to create a lock with a bigger
+bitmap */
+
+#define LOCK_PAGE_BITMAP_MARGIN		64
+
+/* An explicit record lock affects both the record and the gap before it.
+An implicit x-lock does not affect the gap, it only locks the index
+record from read or update.
+
+If a transaction has modified or inserted an index record, then
+it owns an implicit x-lock on the record. On a secondary index record,
+a transaction has an implicit x-lock also if it has modified the
+clustered index record, the max trx id of the page where the secondary
+index record resides is >= trx id of the transaction (or database recovery
+is running), and there are no explicit non-gap lock requests on the
+secondary index record.
+
+This complicated definition for a secondary index comes from the
+implementation: we want to be able to determine if a secondary index
+record has an implicit x-lock, just by looking at the present clustered
+index record, not at the historical versions of the record. The
+complicated definition can be explained to the user so that there is
+nondeterminism in the access path when a query is answered: we may,
+or may not, access the clustered index record and thus may, or may not,
+bump into an x-lock set there.
+
+Different transaction can have conflicting locks set on the gap at the
+same time. The locks on the gap are purely inhibitive: an insert cannot
+be made, or a select cursor may have to wait if a different transaction
+has a conflicting lock on the gap. An x-lock on the gap does not give
+the right to insert into the gap.
+
+An explicit lock can be placed on a user record or the supremum record of
+a page. The locks on the supremum record are always thought to be of the gap
+type, though the gap bit is not set. When we perform an update of a record
+where the size of the record changes, we may temporarily store its explicit
+locks on the infimum record of the page, though the infimum otherwise never
+carries locks.
+
+A waiting record lock can also be of the gap type. A waiting lock request
+can be granted when there is no conflicting mode lock request by another
+transaction ahead of it in the explicit lock queue.
+
+In version 4.0.5 we added yet another explicit lock type: LOCK_REC_NOT_GAP.
+It only locks the record it is placed on, not the gap before the record.
+This lock type is necessary to emulate an Oracle-like READ COMMITTED isolation
+level.
+
+-------------------------------------------------------------------------
+RULE 1: If there is an implicit x-lock on a record, and there are non-gap
+-------
+lock requests waiting in the queue, then the transaction holding the implicit
+x-lock also has an explicit non-gap record x-lock. Therefore, as locks are
+released, we can grant locks to waiting lock requests purely by looking at
+the explicit lock requests in the queue.
+
+RULE 3: Different transactions cannot have conflicting granted non-gap locks
+-------
+on a record at the same time. However, they can have conflicting granted gap
+locks.
+RULE 4: If a there is a waiting lock request in a queue, no lock request,
+-------
+gap or not, can be inserted ahead of it in the queue. In record deletes
+and page splits new gap type locks can be created by the database manager
+for a transaction, and without rule 4, the waits-for graph of transactions
+might become cyclic without the database noticing it, as the deadlock check
+is only performed when a transaction itself requests a lock!
+-------------------------------------------------------------------------
+
+An insert is allowed to a gap if there are no explicit lock requests by
+other transactions on the next record. It does not matter if these lock
+requests are granted or waiting, gap bit set or not, with the exception
+that a gap type request set by another transaction to wait for
+its turn to do an insert is ignored. On the other hand, an
+implicit x-lock by another transaction does not prevent an insert, which
+allows for more concurrency when using an Oracle-style sequence number
+generator for the primary key with many transactions doing inserts
+concurrently.
+
+A modify of a record is allowed if the transaction has an x-lock on the
+record, or if other transactions do not have any non-gap lock requests on the
+record.
+
+A read of a single user record with a cursor is allowed if the transaction
+has a non-gap explicit, or an implicit lock on the record, or if the other
+transactions have no x-lock requests on the record. At a page supremum a
+read is always allowed.
+
+In summary, an implicit lock is seen as a granted x-lock only on the
+record, not on the gap. An explicit lock with no gap bit set is a lock
+both on the record and the gap. If the gap bit is set, the lock is only
+on the gap. Different transaction cannot own conflicting locks on the
+record at the same time, but they may own conflicting locks on the gap.
+Granted locks on a record give an access right to the record, but gap type
+locks just inhibit operations.
+
+NOTE: Finding out if some transaction has an implicit x-lock on a secondary
+index record can be cumbersome. We may have to look at previous versions of
+the corresponding clustered index record to find out if a delete marked
+secondary index record was delete marked by an active transaction, not by
+a committed one.
+
+FACT A: If a transaction has inserted a row, it can delete it any time
+without need to wait for locks.
+
+PROOF: The transaction has an implicit x-lock on every index record inserted
+for the row, and can thus modify each record without the need to wait. Q.E.D.
+
+FACT B: If a transaction has read some result set with a cursor, it can read
+it again, and retrieves the same result set, if it has not modified the
+result set in the meantime. Hence, there is no phantom problem. If the
+biggest record, in the alphabetical order, touched by the cursor is removed,
+a lock wait may occur, otherwise not.
+
+PROOF: When a read cursor proceeds, it sets an s-lock on each user record
+it passes, and a gap type s-lock on each page supremum. The cursor must
+wait until it has these locks granted. Then no other transaction can
+have a granted x-lock on any of the user records, and therefore cannot
+modify the user records. Neither can any other transaction insert into
+the gaps which were passed over by the cursor. Page splits and merges,
+and removal of obsolete versions of records do not affect this, because
+when a user record or a page supremum is removed, the next record inherits
+its locks as gap type locks, and therefore blocks inserts to the same gap.
+Also, if a page supremum is inserted, it inherits its locks from the successor
+record. When the cursor is positioned again at the start of the result set,
+the records it will touch on its course are either records it touched
+during the last pass or new inserted page supremums. It can immediately
+access all these records, and when it arrives at the biggest record, it
+notices that the result set is complete. If the biggest record was removed,
+lock wait can occur because the next record only inherits a gap type lock,
+and a wait may be needed. Q.E.D. */
+
+/* If an index record should be changed or a new inserted, we must check
+the lock on the record or the next. When a read cursor starts reading,
+we will set a record level s-lock on each record it passes, except on the
+initial record on which the cursor is positioned before we start to fetch
+records. Our index tree search has the convention that the B-tree
+cursor is positioned BEFORE the first possibly matching record in
+the search. Optimizations are possible here: if the record is searched
+on an equality condition to a unique key, we could actually set a special
+lock on the record, a lock which would not prevent any insert before
+this record. In the next key locking an x-lock set on a record also
+prevents inserts just before that record.
+	There are special infimum and supremum records on each page.
+A supremum record can be locked by a read cursor. This records cannot be
+updated but the lock prevents insert of a user record to the end of
+the page.
+	Next key locks will prevent the phantom problem where new rows
+could appear to SELECT result sets after the select operation has been
+performed. Prevention of phantoms ensures the serilizability of
+transactions.
+	What should we check if an insert of a new record is wanted?
+Only the lock on the next record on the same page, because also the
+supremum record can carry a lock. An s-lock prevents insertion, but
+what about an x-lock? If it was set by a searched update, then there
+is implicitly an s-lock, too, and the insert should be prevented.
+What if our transaction owns an x-lock to the next record, but there is
+a waiting s-lock request on the next record? If this s-lock was placed
+by a read cursor moving in the ascending order in the index, we cannot
+do the insert immediately, because when we finally commit our transaction,
+the read cursor should see also the new inserted record. So we should
+move the read cursor backward from the the next record for it to pass over
+the new inserted record. This move backward may be too cumbersome to
+implement. If we in this situation just enqueue a second x-lock request
+for our transaction on the next record, then the deadlock mechanism
+notices a deadlock between our transaction and the s-lock request
+transaction. This seems to be an ok solution.
+	We could have the convention that granted explicit record locks,
+lock the corresponding records from changing, and also lock the gaps
+before them from inserting. A waiting explicit lock request locks the gap
+before from inserting. Implicit record x-locks, which we derive from the
+transaction id in the clustered index record, only lock the record itself
+from modification, not the gap before it from inserting.
+	How should we store update locks? If the search is done by a unique
+key, we could just modify the record trx id. Otherwise, we could put a record
+x-lock on the record. If the update changes ordering fields of the
+clustered index record, the inserted new record needs no record lock in
+lock table, the trx id is enough. The same holds for a secondary index
+record. Searched delete is similar to update.
+
+PROBLEM:
+What about waiting lock requests? If a transaction is waiting to make an
+update to a record which another modified, how does the other transaction
+know to send the end-lock-wait signal to the waiting transaction? If we have
+the convention that a transaction may wait for just one lock at a time, how
+do we preserve it if lock wait ends?
+
+PROBLEM:
+Checking the trx id label of a secondary index record. In the case of a
+modification, not an insert, is this necessary? A secondary index record
+is modified only by setting or resetting its deleted flag. A secondary index
+record contains fields to uniquely determine the corresponding clustered
+index record. A secondary index record is therefore only modified if we
+also modify the clustered index record, and the trx id checking is done
+on the clustered index record, before we come to modify the secondary index
+record. So, in the case of delete marking or unmarking a secondary index
+record, we do not have to care about trx ids, only the locks in the lock
+table must be checked. In the case of a select from a secondary index, the
+trx id is relevant, and in this case we may have to search the clustered
+index record.
+
+PROBLEM: How to update record locks when page is split or merged, or
+--------------------------------------------------------------------
+a record is deleted or updated?
+If the size of fields in a record changes, we perform the update by
+a delete followed by an insert. How can we retain the locks set or
+waiting on the record? Because a record lock is indexed in the bitmap
+by the heap number of the record, when we remove the record from the
+record list, it is possible still to keep the lock bits. If the page
+is reorganized, we could make a table of old and new heap numbers,
+and permute the bitmaps in the locks accordingly. We can add to the
+table a row telling where the updated record ended. If the update does
+not require a reorganization of the page, we can simply move the lock
+bits for the updated record to the position determined by its new heap
+number (we may have to allocate a new lock, if we run out of the bitmap
+in the old one).
+	A more complicated case is the one where the reinsertion of the
+updated record is done pessimistically, because the structure of the
+tree may change.
+
+PROBLEM: If a supremum record is removed in a page merge, or a record
+---------------------------------------------------------------------
+removed in a purge, what to do to the waiting lock requests? In a split to
+the right, we just move the lock requests to the new supremum. If a record
+is removed, we could move the waiting lock request to its inheritor, the
+next record in the index. But, the next record may already have lock
+requests on its own queue. A new deadlock check should be made then. Maybe
+it is easier just to release the waiting transactions. They can then enqueue
+new lock requests on appropriate records.
+
+PROBLEM: When a record is inserted, what locks should it inherit from the
+-------------------------------------------------------------------------
+upper neighbor? An insert of a new supremum record in a page split is
+always possible, but an insert of a new user record requires that the upper
+neighbor does not have any lock requests by other transactions, granted or
+waiting, in its lock queue. Solution: We can copy the locks as gap type
+locks, so that also the waiting locks are transformed to granted gap type
+locks on the inserted record. */
+
+/* LOCK COMPATIBILITY MATRIX
+ *    IS IX S  X  AI
+ * IS +	 +  +  -  +
+ * IX +	 +  -  -  +
+ * S  +	 -  +  -  -
+ * X  -	 -  -  -  -
+ * AI +	 +  -  -  -
+ *
+ * Note that for rows, InnoDB only acquires S or X locks.
+ * For tables, InnoDB normally acquires IS or IX locks.
+ * S or X table locks are only acquired for LOCK TABLES.
+ * Auto-increment (AI) locks are needed because of
+ * statement-level MySQL binlog.
+ * See also lock_mode_compatible().
+ */
+#define LK(a,b) (1 << ((a) * LOCK_NUM + (b)))
+#define LKS(a,b) LK(a,b) | LK(b,a)
+
+/* Define the lock compatibility matrix in a ulint.  The first line below
+defines the diagonal entries.  The following lines define the compatibility
+for LOCK_IX, LOCK_S, and LOCK_AUTO_INC using LKS(), since the matrix
+is symmetric. */
+#define LOCK_MODE_COMPATIBILITY 0					\
+ | LK(LOCK_IS, LOCK_IS) | LK(LOCK_IX, LOCK_IX) | LK(LOCK_S, LOCK_S)	\
+ | LKS(LOCK_IX, LOCK_IS) | LKS(LOCK_IS, LOCK_AUTO_INC)			\
+ | LKS(LOCK_S, LOCK_IS)							\
+ | LKS(LOCK_AUTO_INC, LOCK_IS) | LKS(LOCK_AUTO_INC, LOCK_IX)
+
+/* STRONGER-OR-EQUAL RELATION (mode1=row, mode2=column)
+ *    IS IX S  X  AI
+ * IS +  -  -  -  -
+ * IX +  +  -  -  -
+ * S  +  -  +  -  -
+ * X  +  +  +  +  +
+ * AI -  -  -  -  +
+ * See lock_mode_stronger_or_eq().
+ */
+
+/* Define the stronger-or-equal lock relation in a ulint.  This relation
+contains all pairs LK(mode1, mode2) where mode1 is stronger than or
+equal to mode2. */
+#define LOCK_MODE_STRONGER_OR_EQ 0					\
+ | LK(LOCK_IS, LOCK_IS)							\
+ | LK(LOCK_IX, LOCK_IS) | LK(LOCK_IX, LOCK_IX)				\
+ | LK(LOCK_S, LOCK_IS) | LK(LOCK_S, LOCK_S)				\
+ | LK(LOCK_AUTO_INC, LOCK_AUTO_INC)					\
+ | LK(LOCK_X, LOCK_IS) | LK(LOCK_X, LOCK_IX) | LK(LOCK_X, LOCK_S)	\
+ | LK(LOCK_X, LOCK_AUTO_INC) | LK(LOCK_X, LOCK_X)
+
+#ifdef UNIV_DEBUG
+UNIV_INTERN ibool	lock_print_waits	= FALSE;
+
+/*************************************************************************
+Validates the lock system. */
+static
+ibool
+lock_validate(void);
+/*===============*/
+			/* out: TRUE if ok */
+
+/*************************************************************************
+Validates the record lock queues on a page. */
+static
+ibool
+lock_rec_validate_page(
+/*===================*/
+			/* out: TRUE if ok */
+	ulint	space,	/* in: space id */
+	ulint	page_no);/* in: page number */
+
+/* Define the following in order to enable lock_rec_validate_page() checks. */
+# undef UNIV_DEBUG_LOCK_VALIDATE
+#endif /* UNIV_DEBUG */
+
+/* The lock system */
+UNIV_INTERN lock_sys_t*	lock_sys	= NULL;
+
+/* We store info on the latest deadlock error to this buffer. InnoDB
+Monitor will then fetch it and print */
+UNIV_INTERN ibool	lock_deadlock_found = FALSE;
+UNIV_INTERN FILE*	lock_latest_err_file;
+
+/* Flags for recursive deadlock search */
+#define LOCK_VICTIM_IS_START	1
+#define LOCK_VICTIM_IS_OTHER	2
+
+/************************************************************************
+Checks if a lock request results in a deadlock. */
+static
+ibool
+lock_deadlock_occurs(
+/*=================*/
+			/* out: TRUE if a deadlock was detected and we
+			chose trx as a victim; FALSE if no deadlock, or
+			there was a deadlock, but we chose other
+			transaction(s) as victim(s) */
+	lock_t*	lock,	/* in: lock the transaction is requesting */
+	trx_t*	trx);	/* in: transaction */
+/************************************************************************
+Looks recursively for a deadlock. */
+static
+ulint
+lock_deadlock_recursive(
+/*====================*/
+				/* out: 0 if no deadlock found,
+				LOCK_VICTIM_IS_START if there was a deadlock
+				and we chose 'start' as the victim,
+				LOCK_VICTIM_IS_OTHER if a deadlock
+				was found and we chose some other trx as a
+				victim: we must do the search again in this
+				last case because there may be another
+				deadlock! */
+	trx_t*	start,		/* in: recursion starting point */
+	trx_t*	trx,		/* in: a transaction waiting for a lock */
+	lock_t*	wait_lock,	/* in: the lock trx is waiting to be granted */
+	ulint*	cost,		/* in/out: number of calculation steps thus
+				far: if this exceeds LOCK_MAX_N_STEPS_...
+				we return LOCK_VICTIM_IS_START */
+	ulint	depth);		/* in: recursion depth: if this exceeds
+				LOCK_MAX_DEPTH_IN_DEADLOCK_CHECK, we
+				return LOCK_VICTIM_IS_START */
+
+/*************************************************************************
+Gets the nth bit of a record lock. */
+UNIV_INLINE
+ibool
+lock_rec_get_nth_bit(
+/*=================*/
+				/* out: TRUE if bit set */
+	const lock_t*	lock,	/* in: record lock */
+	ulint		i)	/* in: index of the bit */
+{
+	ulint	byte_index;
+	ulint	bit_index;
+
+	ut_ad(lock);
+	ut_ad(lock_get_type_low(lock) == LOCK_REC);
+
+	if (i >= lock->un_member.rec_lock.n_bits) {
+
+		return(FALSE);
+	}
+
+	byte_index = i / 8;
+	bit_index = i % 8;
+
+	return(1 & ((const byte*) &lock[1])[byte_index] >> bit_index);
+}
+
+/*************************************************************************/
+
+#define lock_mutex_enter_kernel()	mutex_enter(&kernel_mutex)
+#define lock_mutex_exit_kernel()	mutex_exit(&kernel_mutex)
+
+/*************************************************************************
+Checks that a transaction id is sensible, i.e., not in the future. */
+UNIV_INTERN
+ibool
+lock_check_trx_id_sanity(
+/*=====================*/
+					/* out: TRUE if ok */
+	dulint		trx_id,		/* in: trx id */
+	const rec_t*	rec,		/* in: user record */
+	dict_index_t*	index,		/* in: index */
+	const ulint*	offsets,	/* in: rec_get_offsets(rec, index) */
+	ibool		has_kernel_mutex)/* in: TRUE if the caller owns the
+					kernel mutex */
+{
+	ibool	is_ok		= TRUE;
+
+	ut_ad(rec_offs_validate(rec, index, offsets));
+
+	if (!has_kernel_mutex) {
+		mutex_enter(&kernel_mutex);
+	}
+
+	/* A sanity check: the trx_id in rec must be smaller than the global
+	trx id counter */
+
+	if (ut_dulint_cmp(trx_id, trx_sys->max_trx_id) >= 0) {
+		ut_print_timestamp(stderr);
+		fputs("  InnoDB: Error: transaction id associated"
+		      " with record\n",
+		      stderr);
+		rec_print_new(stderr, rec, offsets);
+		fputs("InnoDB: in ", stderr);
+		dict_index_name_print(stderr, NULL, index);
+		fprintf(stderr, "\n"
+			"InnoDB: is " TRX_ID_FMT " which is higher than the"
+			" global trx id counter " TRX_ID_FMT "!\n"
+			"InnoDB: The table is corrupt. You have to do"
+			" dump + drop + reimport.\n",
+			TRX_ID_PREP_PRINTF(trx_id),
+			TRX_ID_PREP_PRINTF(trx_sys->max_trx_id));
+
+		is_ok = FALSE;
+	}
+
+	if (!has_kernel_mutex) {
+		mutex_exit(&kernel_mutex);
+	}
+
+	return(is_ok);
+}
+
+/*************************************************************************
+Checks that a record is seen in a consistent read. */
+UNIV_INTERN
+ibool
+lock_clust_rec_cons_read_sees(
+/*==========================*/
+				/* out: TRUE if sees, or FALSE if an earlier
+				version of the record should be retrieved */
+	const rec_t*	rec,	/* in: user record which should be read or
+				passed over by a read cursor */
+	dict_index_t*	index,	/* in: clustered index */
+	const ulint*	offsets,/* in: rec_get_offsets(rec, index) */
+	read_view_t*	view)	/* in: consistent read view */
+{
+	dulint	trx_id;
+
+	ut_ad(dict_index_is_clust(index));
+	ut_ad(page_rec_is_user_rec(rec));
+	ut_ad(rec_offs_validate(rec, index, offsets));
+
+	/* NOTE that we call this function while holding the search
+	system latch. To obey the latching order we must NOT reserve the
+	kernel mutex here! */
+
+	trx_id = row_get_rec_trx_id(rec, index, offsets);
+
+	return(read_view_sees_trx_id(view, trx_id));
+}
+
+/*************************************************************************
+Checks that a non-clustered index record is seen in a consistent read. */
+UNIV_INTERN
+ulint
+lock_sec_rec_cons_read_sees(
+/*========================*/
+					/* out: TRUE if certainly
+					sees, or FALSE if an earlier
+					version of the clustered index
+					record might be needed: NOTE
+					that a non-clustered index
+					page contains so little
+					information on its
+					modifications that also in the
+					case FALSE, the present
+					version of rec may be the
+					right, but we must check this
+					from the clustered index
+					record */
+	const rec_t*		rec,	/* in: user record which
+					should be read or passed over
+					by a read cursor */
+	const read_view_t*	view)	/* in: consistent read view */
+{
+	dulint	max_trx_id;
+
+	ut_ad(page_rec_is_user_rec(rec));
+
+	/* NOTE that we might call this function while holding the search
+	system latch. To obey the latching order we must NOT reserve the
+	kernel mutex here! */
+
+	if (recv_recovery_is_on()) {
+
+		return(FALSE);
+	}
+
+	max_trx_id = page_get_max_trx_id(page_align(rec));
+
+	return(ut_dulint_cmp(max_trx_id, view->up_limit_id) < 0);
+}
+
+/*************************************************************************
+Creates the lock system at database start. */
+UNIV_INTERN
+void
+lock_sys_create(
+/*============*/
+	ulint	n_cells)	/* in: number of slots in lock hash table */
+{
+	lock_sys = mem_alloc(sizeof(lock_sys_t));
+
+	lock_sys->rec_hash = hash_create(n_cells);
+
+	/* hash_create_mutexes(lock_sys->rec_hash, 2, SYNC_REC_LOCK); */
+
+	lock_latest_err_file = os_file_create_tmpfile();
+	ut_a(lock_latest_err_file);
+}
+
+/*************************************************************************
+Gets the size of a lock struct. */
+UNIV_INTERN
+ulint
+lock_get_size(void)
+/*===============*/
+			/* out: size in bytes */
+{
+	return((ulint)sizeof(lock_t));
+}
+
+/*************************************************************************
+Gets the mode of a lock. */
+UNIV_INLINE
+enum lock_mode
+lock_get_mode(
+/*==========*/
+				/* out: mode */
+	const lock_t*	lock)	/* in: lock */
+{
+	ut_ad(lock);
+
+	return(lock->type_mode & LOCK_MODE_MASK);
+}
+
+/*************************************************************************
+Gets the wait flag of a lock. */
+UNIV_INLINE
+ibool
+lock_get_wait(
+/*==========*/
+				/* out: TRUE if waiting */
+	const lock_t*	lock)	/* in: lock */
+{
+	ut_ad(lock);
+
+	if (UNIV_UNLIKELY(lock->type_mode & LOCK_WAIT)) {
+
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
+
+/*************************************************************************
+Gets the source table of an ALTER TABLE transaction.  The table must be
+covered by an IX or IS table lock. */
+UNIV_INTERN
+dict_table_t*
+lock_get_src_table(
+/*===============*/
+				/* out: the source table of transaction,
+				if it is covered by an IX or IS table lock;
+				dest if there is no source table, and
+				NULL if the transaction is locking more than
+				two tables or an inconsistency is found */
+	trx_t*		trx,	/* in: transaction */
+	dict_table_t*	dest,	/* in: destination of ALTER TABLE */
+	enum lock_mode*	mode)	/* out: lock mode of the source table */
+{
+	dict_table_t*	src;
+	lock_t*		lock;
+
+	src = NULL;
+	*mode = LOCK_NONE;
+
+	for (lock = UT_LIST_GET_FIRST(trx->trx_locks);
+	     lock;
+	     lock = UT_LIST_GET_NEXT(trx_locks, lock)) {
+		lock_table_t*	tab_lock;
+		enum lock_mode	lock_mode;
+		if (!(lock_get_type_low(lock) & LOCK_TABLE)) {
+			/* We are only interested in table locks. */
+			continue;
+		}
+		tab_lock = &lock->un_member.tab_lock;
+		if (dest == tab_lock->table) {
+			/* We are not interested in the destination table. */
+			continue;
+		} else if (!src) {
+			/* This presumably is the source table. */
+			src = tab_lock->table;
+			if (UT_LIST_GET_LEN(src->locks) != 1
+			    || UT_LIST_GET_FIRST(src->locks) != lock) {
+				/* We only support the case when
+				there is only one lock on this table. */
+				return(NULL);
+			}
+		} else if (src != tab_lock->table) {
+			/* The transaction is locking more than
+			two tables (src and dest): abort */
+			return(NULL);
+		}
+
+		/* Check that the source table is locked by
+		LOCK_IX or LOCK_IS. */
+		lock_mode = lock_get_mode(lock);
+		if (lock_mode == LOCK_IX || lock_mode == LOCK_IS) {
+			if (*mode != LOCK_NONE && *mode != lock_mode) {
+				/* There are multiple locks on src. */
+				return(NULL);
+			}
+			*mode = lock_mode;
+		}
+	}
+
+	if (!src) {
+		/* No source table lock found: flag the situation to caller */
+		src = dest;
+	}
+
+	return(src);
+}
+
+/*************************************************************************
+Determine if the given table is exclusively "owned" by the given
+transaction, i.e., transaction holds LOCK_IX and possibly LOCK_AUTO_INC
+on the table. */
+UNIV_INTERN
+ibool
+lock_is_table_exclusive(
+/*====================*/
+				/* out: TRUE if table is only locked by trx,
+				with LOCK_IX, and possibly LOCK_AUTO_INC */
+	dict_table_t*	table,	/* in: table */
+	trx_t*		trx)	/* in: transaction */
+{
+	const lock_t*	lock;
+	ibool		ok	= FALSE;
+
+	ut_ad(table);
+	ut_ad(trx);
+
+	lock_mutex_enter_kernel();
+
+	for (lock = UT_LIST_GET_FIRST(table->locks);
+	     lock;
+	     lock = UT_LIST_GET_NEXT(locks, &lock->un_member.tab_lock)) {
+		if (lock->trx != trx) {
+			/* A lock on the table is held
+			by some other transaction. */
+			goto not_ok;
+		}
+
+		if (!(lock_get_type_low(lock) & LOCK_TABLE)) {
+			/* We are interested in table locks only. */
+			continue;
+		}
+
+		switch (lock_get_mode(lock)) {
+		case LOCK_IX:
+			ok = TRUE;
+			break;
+		case LOCK_AUTO_INC:
+			/* It is allowed for trx to hold an
+			auto_increment lock. */
+			break;
+		default:
+not_ok:
+			/* Other table locks than LOCK_IX are not allowed. */
+			ok = FALSE;
+			goto func_exit;
+		}
+	}
+
+func_exit:
+	lock_mutex_exit_kernel();
+
+	return(ok);
+}
+
+/*************************************************************************
+Sets the wait flag of a lock and the back pointer in trx to lock. */
+UNIV_INLINE
+void
+lock_set_lock_and_trx_wait(
+/*=======================*/
+	lock_t*	lock,	/* in: lock */
+	trx_t*	trx)	/* in: trx */
+{
+	ut_ad(lock);
+	ut_ad(trx->wait_lock == NULL);
+
+	trx->wait_lock = lock;
+	lock->type_mode |= LOCK_WAIT;
+}
+
+/**************************************************************************
+The back pointer to a waiting lock request in the transaction is set to NULL
+and the wait bit in lock type_mode is reset. */
+UNIV_INLINE
+void
+lock_reset_lock_and_trx_wait(
+/*=========================*/
+	lock_t*	lock)	/* in: record lock */
+{
+	ut_ad((lock->trx)->wait_lock == lock);
+	ut_ad(lock_get_wait(lock));
+
+	/* Reset the back pointer in trx to this waiting lock request */
+
+	(lock->trx)->wait_lock = NULL;
+	lock->type_mode &= ~LOCK_WAIT;
+}
+
+/*************************************************************************
+Gets the gap flag of a record lock. */
+UNIV_INLINE
+ibool
+lock_rec_get_gap(
+/*=============*/
+				/* out: TRUE if gap flag set */
+	const lock_t*	lock)	/* in: record lock */
+{
+	ut_ad(lock);
+	ut_ad(lock_get_type_low(lock) == LOCK_REC);
+
+	if (lock->type_mode & LOCK_GAP) {
+
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
+
+/*************************************************************************
+Gets the LOCK_REC_NOT_GAP flag of a record lock. */
+UNIV_INLINE
+ibool
+lock_rec_get_rec_not_gap(
+/*=====================*/
+				/* out: TRUE if LOCK_REC_NOT_GAP flag set */
+	const lock_t*	lock)	/* in: record lock */
+{
+	ut_ad(lock);
+	ut_ad(lock_get_type_low(lock) == LOCK_REC);
+
+	if (lock->type_mode & LOCK_REC_NOT_GAP) {
+
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
+
+/*************************************************************************
+Gets the waiting insert flag of a record lock. */
+UNIV_INLINE
+ibool
+lock_rec_get_insert_intention(
+/*==========================*/
+				/* out: TRUE if gap flag set */
+	const lock_t*	lock)	/* in: record lock */
+{
+	ut_ad(lock);
+	ut_ad(lock_get_type_low(lock) == LOCK_REC);
+
+	if (lock->type_mode & LOCK_INSERT_INTENTION) {
+
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
+
+/*************************************************************************
+Calculates if lock mode 1 is stronger or equal to lock mode 2. */
+UNIV_INLINE
+ulint
+lock_mode_stronger_or_eq(
+/*=====================*/
+				/* out: nonzero
+				if mode1 stronger or equal to mode2 */
+	enum lock_mode	mode1,	/* in: lock mode */
+	enum lock_mode	mode2)	/* in: lock mode */
+{
+	ut_ad(mode1 == LOCK_X || mode1 == LOCK_S || mode1 == LOCK_IX
+	      || mode1 == LOCK_IS || mode1 == LOCK_AUTO_INC);
+	ut_ad(mode2 == LOCK_X || mode2 == LOCK_S || mode2 == LOCK_IX
+	      || mode2 == LOCK_IS || mode2 == LOCK_AUTO_INC);
+
+	return((LOCK_MODE_STRONGER_OR_EQ) & LK(mode1, mode2));
+}
+
+/*************************************************************************
+Calculates if lock mode 1 is compatible with lock mode 2. */
+UNIV_INLINE
+ulint
+lock_mode_compatible(
+/*=================*/
+			/* out: nonzero if mode1 compatible with mode2 */
+	enum lock_mode	mode1,	/* in: lock mode */
+	enum lock_mode	mode2)	/* in: lock mode */
+{
+	ut_ad(mode1 == LOCK_X || mode1 == LOCK_S || mode1 == LOCK_IX
+	      || mode1 == LOCK_IS || mode1 == LOCK_AUTO_INC);
+	ut_ad(mode2 == LOCK_X || mode2 == LOCK_S || mode2 == LOCK_IX
+	      || mode2 == LOCK_IS || mode2 == LOCK_AUTO_INC);
+
+	return((LOCK_MODE_COMPATIBILITY) & LK(mode1, mode2));
+}
+
+/*************************************************************************
+Checks if a lock request for a new lock has to wait for request lock2. */
+UNIV_INLINE
+ibool
+lock_rec_has_to_wait(
+/*=================*/
+				/* out: TRUE if new lock has to wait
+				for lock2 to be removed */
+	const trx_t*	trx,	/* in: trx of new lock */
+	ulint		type_mode,/* in: precise mode of the new lock
+				to set: LOCK_S or LOCK_X, possibly
+				ORed to LOCK_GAP or LOCK_REC_NOT_GAP,
+				LOCK_INSERT_INTENTION */
+	const lock_t*	lock2,	/* in: another record lock; NOTE that
+				it is assumed that this has a lock bit
+				set on the same record as in the new
+				lock we are setting */
+	ibool lock_is_on_supremum)  /* in: TRUE if we are setting the
+				lock on the 'supremum' record of an
+				index page: we know then that the lock
+				request is really for a 'gap' type lock */
+{
+	ut_ad(trx && lock2);
+	ut_ad(lock_get_type_low(lock2) == LOCK_REC);
+
+	if (trx != lock2->trx
+	    && !lock_mode_compatible(LOCK_MODE_MASK & type_mode,
+				     lock_get_mode(lock2))) {
+
+		/* We have somewhat complex rules when gap type record locks
+		cause waits */
+
+		if ((lock_is_on_supremum || (type_mode & LOCK_GAP))
+		    && !(type_mode & LOCK_INSERT_INTENTION)) {
+
+			/* Gap type locks without LOCK_INSERT_INTENTION flag
+			do not need to wait for anything. This is because
+			different users can have conflicting lock types
+			on gaps. */
+
+			return(FALSE);
+		}
+
+		if (!(type_mode & LOCK_INSERT_INTENTION)
+		    && lock_rec_get_gap(lock2)) {
+
+			/* Record lock (LOCK_ORDINARY or LOCK_REC_NOT_GAP
+			does not need to wait for a gap type lock */
+
+			return(FALSE);
+		}
+
+		if ((type_mode & LOCK_GAP)
+		    && lock_rec_get_rec_not_gap(lock2)) {
+
+			/* Lock on gap does not need to wait for
+			a LOCK_REC_NOT_GAP type lock */
+
+			return(FALSE);
+		}
+
+		if (lock_rec_get_insert_intention(lock2)) {
+
+			/* No lock request needs to wait for an insert
+			intention lock to be removed. This is ok since our
+			rules allow conflicting locks on gaps. This eliminates
+			a spurious deadlock caused by a next-key lock waiting
+			for an insert intention lock; when the insert
+			intention lock was granted, the insert deadlocked on
+			the waiting next-key lock.
+
+			Also, insert intention locks do not disturb each
+			other. */
+
+			return(FALSE);
+		}
+
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
+
+/*************************************************************************
+Checks if a lock request lock1 has to wait for request lock2. */
+UNIV_INTERN
+ibool
+lock_has_to_wait(
+/*=============*/
+				/* out: TRUE if lock1 has to wait for
+				lock2 to be removed */
+	const lock_t*	lock1,	/* in: waiting lock */
+	const lock_t*	lock2)	/* in: another lock; NOTE that it is
+				assumed that this has a lock bit set
+				on the same record as in lock1 if the
+				locks are record locks */
+{
+	ut_ad(lock1 && lock2);
+
+	if (lock1->trx != lock2->trx
+	    && !lock_mode_compatible(lock_get_mode(lock1),
+				     lock_get_mode(lock2))) {
+		if (lock_get_type_low(lock1) == LOCK_REC) {
+			ut_ad(lock_get_type_low(lock2) == LOCK_REC);
+
+			/* If this lock request is for a supremum record
+			then the second bit on the lock bitmap is set */
+
+			return(lock_rec_has_to_wait(lock1->trx,
+						    lock1->type_mode, lock2,
+						    lock_rec_get_nth_bit(
+							    lock1, 1)));
+		}
+
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
+
+/*============== RECORD LOCK BASIC FUNCTIONS ============================*/
+
+/*************************************************************************
+Gets the number of bits in a record lock bitmap. */
+UNIV_INLINE
+ulint
+lock_rec_get_n_bits(
+/*================*/
+				/* out: number of bits */
+	const lock_t*	lock)	/* in: record lock */
+{
+	return(lock->un_member.rec_lock.n_bits);
+}
+
+/**************************************************************************
+Sets the nth bit of a record lock to TRUE. */
+UNIV_INLINE
+void
+lock_rec_set_nth_bit(
+/*=================*/
+	lock_t*	lock,	/* in: record lock */
+	ulint	i)	/* in: index of the bit */
+{
+	ulint	byte_index;
+	ulint	bit_index;
+
+	ut_ad(lock);
+	ut_ad(lock_get_type_low(lock) == LOCK_REC);
+	ut_ad(i < lock->un_member.rec_lock.n_bits);
+
+	byte_index = i / 8;
+	bit_index = i % 8;
+
+	((byte*) &lock[1])[byte_index] |= 1 << bit_index;
+}
+
+/**************************************************************************
+Looks for a set bit in a record lock bitmap. Returns ULINT_UNDEFINED,
+if none found. */
+UNIV_INTERN
+ulint
+lock_rec_find_set_bit(
+/*==================*/
+				/* out: bit index == heap number of
+				the record, or ULINT_UNDEFINED if none found */
+	const lock_t*	lock)	/* in: record lock with at least one bit set */
+{
+	ulint	i;
+
+	for (i = 0; i < lock_rec_get_n_bits(lock); i++) {
+
+		if (lock_rec_get_nth_bit(lock, i)) {
+
+			return(i);
+		}
+	}
+
+	return(ULINT_UNDEFINED);
+}
+
+/**************************************************************************
+Resets the nth bit of a record lock. */
+UNIV_INLINE
+void
+lock_rec_reset_nth_bit(
+/*===================*/
+	lock_t*	lock,	/* in: record lock */
+	ulint	i)	/* in: index of the bit which must be set to TRUE
+			when this function is called */
+{
+	ulint	byte_index;
+	ulint	bit_index;
+
+	ut_ad(lock);
+	ut_ad(lock_get_type_low(lock) == LOCK_REC);
+	ut_ad(i < lock->un_member.rec_lock.n_bits);
+
+	byte_index = i / 8;
+	bit_index = i % 8;
+
+	((byte*) &lock[1])[byte_index] &= ~(1 << bit_index);
+}
+
+/*************************************************************************
+Gets the first or next record lock on a page. */
+UNIV_INLINE
+lock_t*
+lock_rec_get_next_on_page(
+/*======================*/
+			/* out: next lock, NULL if none exists */
+	lock_t*	lock)	/* in: a record lock */
+{
+	ulint	space;
+	ulint	page_no;
+
+	ut_ad(mutex_own(&kernel_mutex));
+	ut_ad(lock_get_type_low(lock) == LOCK_REC);
+
+	space = lock->un_member.rec_lock.space;
+	page_no = lock->un_member.rec_lock.page_no;
+
+	for (;;) {
+		lock = HASH_GET_NEXT(hash, lock);
+
+		if (!lock) {
+
+			break;
+		}
+
+		if ((lock->un_member.rec_lock.space == space)
+		    && (lock->un_member.rec_lock.page_no == page_no)) {
+
+			break;
+		}
+	}
+
+	return(lock);
+}
+
+/*************************************************************************
+Gets the first record lock on a page, where the page is identified by its
+file address. */
+UNIV_INLINE
+lock_t*
+lock_rec_get_first_on_page_addr(
+/*============================*/
+			/* out: first lock, NULL if none exists */
+	ulint	space,	/* in: space */
+	ulint	page_no)/* in: page number */
+{
+	lock_t*	lock;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	lock = HASH_GET_FIRST(lock_sys->rec_hash,
+			      lock_rec_hash(space, page_no));
+	while (lock) {
+		if ((lock->un_member.rec_lock.space == space)
+		    && (lock->un_member.rec_lock.page_no == page_no)) {
+
+			break;
+		}
+
+		lock = HASH_GET_NEXT(hash, lock);
+	}
+
+	return(lock);
+}
+
+/*************************************************************************
+Returns TRUE if there are explicit record locks on a page. */
+UNIV_INTERN
+ibool
+lock_rec_expl_exist_on_page(
+/*========================*/
+			/* out: TRUE if there are explicit record locks on
+			the page */
+	ulint	space,	/* in: space id */
+	ulint	page_no)/* in: page number */
+{
+	ibool	ret;
+
+	mutex_enter(&kernel_mutex);
+
+	if (lock_rec_get_first_on_page_addr(space, page_no)) {
+		ret = TRUE;
+	} else {
+		ret = FALSE;
+	}
+
+	mutex_exit(&kernel_mutex);
+
+	return(ret);
+}
+
+/*************************************************************************
+Gets the first record lock on a page, where the page is identified by a
+pointer to it. */
+UNIV_INLINE
+lock_t*
+lock_rec_get_first_on_page(
+/*=======================*/
+					/* out: first lock, NULL if
+					none exists */
+	const buf_block_t*	block)	/* in: buffer block */
+{
+	ulint	hash;
+	lock_t*	lock;
+	ulint	space	= buf_block_get_space(block);
+	ulint	page_no	= buf_block_get_page_no(block);
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	hash = buf_block_get_lock_hash_val(block);
+
+	lock = HASH_GET_FIRST(lock_sys->rec_hash, hash);
+
+	while (lock) {
+		if ((lock->un_member.rec_lock.space == space)
+		    && (lock->un_member.rec_lock.page_no == page_no)) {
+
+			break;
+		}
+
+		lock = HASH_GET_NEXT(hash, lock);
+	}
+
+	return(lock);
+}
+
+/*************************************************************************
+Gets the next explicit lock request on a record. */
+UNIV_INLINE
+lock_t*
+lock_rec_get_next(
+/*==============*/
+			/* out: next lock, NULL if none exists */
+	ulint	heap_no,/* in: heap number of the record */
+	lock_t*	lock)	/* in: lock */
+{
+	ut_ad(mutex_own(&kernel_mutex));
+
+	do {
+		ut_ad(lock_get_type_low(lock) == LOCK_REC);
+		lock = lock_rec_get_next_on_page(lock);
+	} while (lock && !lock_rec_get_nth_bit(lock, heap_no));
+
+	return(lock);
+}
+
+/*************************************************************************
+Gets the first explicit lock request on a record. */
+UNIV_INLINE
+lock_t*
+lock_rec_get_first(
+/*===============*/
+					/* out: first lock, NULL if
+					none exists */
+	const buf_block_t*	block,	/* in: block containing the record */
+	ulint			heap_no)/* in: heap number of the record */
+{
+	lock_t*	lock;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	for (lock = lock_rec_get_first_on_page(block); lock;
+	     lock = lock_rec_get_next_on_page(lock)) {
+		if (lock_rec_get_nth_bit(lock, heap_no)) {
+			break;
+		}
+	}
+
+	return(lock);
+}
+
+/*************************************************************************
+Resets the record lock bitmap to zero. NOTE: does not touch the wait_lock
+pointer in the transaction! This function is used in lock object creation
+and resetting. */
+static
+void
+lock_rec_bitmap_reset(
+/*==================*/
+	lock_t*	lock)	/* in: record lock */
+{
+	ulint	n_bytes;
+
+	ut_ad(lock_get_type_low(lock) == LOCK_REC);
+
+	/* Reset to zero the bitmap which resides immediately after the lock
+	struct */
+
+	n_bytes = lock_rec_get_n_bits(lock) / 8;
+
+	ut_ad((lock_rec_get_n_bits(lock) % 8) == 0);
+
+	memset(&lock[1], 0, n_bytes);
+}
+
+/*************************************************************************
+Copies a record lock to heap. */
+static
+lock_t*
+lock_rec_copy(
+/*==========*/
+				/* out: copy of lock */
+	const lock_t*	lock,	/* in: record lock */
+	mem_heap_t*	heap)	/* in: memory heap */
+{
+	ulint	size;
+
+	ut_ad(lock_get_type_low(lock) == LOCK_REC);
+
+	size = sizeof(lock_t) + lock_rec_get_n_bits(lock) / 8;
+
+	return(mem_heap_dup(heap, lock, size));
+}
+
+/*************************************************************************
+Gets the previous record lock set on a record. */
+UNIV_INTERN
+const lock_t*
+lock_rec_get_prev(
+/*==============*/
+				/* out: previous lock on the same
+				record, NULL if none exists */
+	const lock_t*	in_lock,/* in: record lock */
+	ulint		heap_no)/* in: heap number of the record */
+{
+	lock_t*	lock;
+	ulint	space;
+	ulint	page_no;
+	lock_t*	found_lock	= NULL;
+
+	ut_ad(mutex_own(&kernel_mutex));
+	ut_ad(lock_get_type_low(in_lock) == LOCK_REC);
+
+	space = in_lock->un_member.rec_lock.space;
+	page_no = in_lock->un_member.rec_lock.page_no;
+
+	lock = lock_rec_get_first_on_page_addr(space, page_no);
+
+	for (;;) {
+		ut_ad(lock);
+
+		if (lock == in_lock) {
+
+			return(found_lock);
+		}
+
+		if (lock_rec_get_nth_bit(lock, heap_no)) {
+
+			found_lock = lock;
+		}
+
+		lock = lock_rec_get_next_on_page(lock);
+	}
+}
+
+/*============= FUNCTIONS FOR ANALYZING TABLE LOCK QUEUE ================*/
+
+/*************************************************************************
+Checks if a transaction has the specified table lock, or stronger. */
+UNIV_INLINE
+lock_t*
+lock_table_has(
+/*===========*/
+				/* out: lock or NULL */
+	trx_t*		trx,	/* in: transaction */
+	dict_table_t*	table,	/* in: table */
+	enum lock_mode	mode)	/* in: lock mode */
+{
+	lock_t*	lock;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	/* Look for stronger locks the same trx already has on the table */
+
+	lock = UT_LIST_GET_LAST(table->locks);
+
+	while (lock != NULL) {
+
+		if (lock->trx == trx
+		    && lock_mode_stronger_or_eq(lock_get_mode(lock), mode)) {
+
+			/* The same trx already has locked the table in
+			a mode stronger or equal to the mode given */
+
+			ut_ad(!lock_get_wait(lock));
+
+			return(lock);
+		}
+
+		lock = UT_LIST_GET_PREV(un_member.tab_lock.locks, lock);
+	}
+
+	return(NULL);
+}
+
+/*============= FUNCTIONS FOR ANALYZING RECORD LOCK QUEUE ================*/
+
+/*************************************************************************
+Checks if a transaction has a GRANTED explicit lock on rec stronger or equal
+to precise_mode. */
+UNIV_INLINE
+lock_t*
+lock_rec_has_expl(
+/*==============*/
+					/* out: lock or NULL */
+	ulint			precise_mode,/* in: LOCK_S or LOCK_X
+					possibly ORed to LOCK_GAP or
+					LOCK_REC_NOT_GAP, for a
+					supremum record we regard this
+					always a gap type request */
+	const buf_block_t*	block,	/* in: buffer block containing
+					the record */
+	ulint			heap_no,/* in: heap number of the record */
+	trx_t*			trx)	/* in: transaction */
+{
+	lock_t*	lock;
+
+	ut_ad(mutex_own(&kernel_mutex));
+	ut_ad((precise_mode & LOCK_MODE_MASK) == LOCK_S
+	      || (precise_mode & LOCK_MODE_MASK) == LOCK_X);
+	ut_ad(!(precise_mode & LOCK_INSERT_INTENTION));
+
+	lock = lock_rec_get_first(block, heap_no);
+
+	while (lock) {
+		if (lock->trx == trx
+		    && lock_mode_stronger_or_eq(lock_get_mode(lock),
+						precise_mode & LOCK_MODE_MASK)
+		    && !lock_get_wait(lock)
+		    && (!lock_rec_get_rec_not_gap(lock)
+			|| (precise_mode & LOCK_REC_NOT_GAP)
+			|| heap_no == PAGE_HEAP_NO_SUPREMUM)
+		    && (!lock_rec_get_gap(lock)
+			|| (precise_mode & LOCK_GAP)
+			|| heap_no == PAGE_HEAP_NO_SUPREMUM)
+		    && (!lock_rec_get_insert_intention(lock))) {
+
+			return(lock);
+		}
+
+		lock = lock_rec_get_next(heap_no, lock);
+	}
+
+	return(NULL);
+}
+
+#ifdef UNIV_DEBUG
+# ifndef UNIV_HOTBACKUP
+/*************************************************************************
+Checks if some other transaction has a lock request in the queue. */
+static
+lock_t*
+lock_rec_other_has_expl_req(
+/*========================*/
+					/* out: lock or NULL */
+	enum lock_mode		mode,	/* in: LOCK_S or LOCK_X */
+	ulint			gap,	/* in: LOCK_GAP if also gap
+					locks are taken into account,
+					or 0 if not */
+	ulint			wait,	/* in: LOCK_WAIT if also
+					waiting locks are taken into
+					account, or 0 if not */
+	const buf_block_t*	block,	/* in: buffer block containing
+					the record */
+	ulint			heap_no,/* in: heap number of the record */
+	const trx_t*		trx)	/* in: transaction, or NULL if
+					requests by all transactions
+					are taken into account */
+{
+	lock_t*	lock;
+
+	ut_ad(mutex_own(&kernel_mutex));
+	ut_ad(mode == LOCK_X || mode == LOCK_S);
+	ut_ad(gap == 0 || gap == LOCK_GAP);
+	ut_ad(wait == 0 || wait == LOCK_WAIT);
+
+	lock = lock_rec_get_first(block, heap_no);
+
+	while (lock) {
+		if (lock->trx != trx
+		    && (gap
+			|| !(lock_rec_get_gap(lock)
+			     || heap_no == PAGE_HEAP_NO_SUPREMUM))
+		    && (wait || !lock_get_wait(lock))
+		    && lock_mode_stronger_or_eq(lock_get_mode(lock), mode)) {
+
+			return(lock);
+		}
+
+		lock = lock_rec_get_next(heap_no, lock);
+	}
+
+	return(NULL);
+}
+# endif /* !UNIV_HOTBACKUP */
+#endif /* UNIV_DEBUG */
+
+/*************************************************************************
+Checks if some other transaction has a conflicting explicit lock request
+in the queue, so that we have to wait. */
+static
+lock_t*
+lock_rec_other_has_conflicting(
+/*===========================*/
+					/* out: lock or NULL */
+	enum lock_mode		mode,	/* in: LOCK_S or LOCK_X,
+					possibly ORed to LOCK_GAP or
+					LOC_REC_NOT_GAP,
+					LOCK_INSERT_INTENTION */
+	const buf_block_t*	block,	/* in: buffer block containing
+					the record */
+	ulint			heap_no,/* in: heap number of the record */
+	trx_t*			trx)	/* in: our transaction */
+{
+	lock_t*	lock;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	lock = lock_rec_get_first(block, heap_no);
+
+	if (UNIV_LIKELY_NULL(lock)) {
+		if (UNIV_UNLIKELY(heap_no == PAGE_HEAP_NO_SUPREMUM)) {
+
+			do {
+				if (lock_rec_has_to_wait(trx, mode, lock,
+							 TRUE)) {
+					return(lock);
+				}
+
+				lock = lock_rec_get_next(heap_no, lock);
+			} while (lock);
+		} else {
+
+			do {
+				if (lock_rec_has_to_wait(trx, mode, lock,
+							 FALSE)) {
+					return(lock);
+				}
+
+				lock = lock_rec_get_next(heap_no, lock);
+			} while (lock);
+		}
+	}
+
+	return(NULL);
+}
+
+/*************************************************************************
+Looks for a suitable type record lock struct by the same trx on the same page.
+This can be used to save space when a new record lock should be set on a page:
+no new struct is needed, if a suitable old is found. */
+UNIV_INLINE
+lock_t*
+lock_rec_find_similar_on_page(
+/*==========================*/
+					/* out: lock or NULL */
+	ulint		type_mode,	/* in: lock type_mode field */
+	ulint		heap_no,	/* in: heap number of the record */
+	lock_t*		lock,		/* in: lock_rec_get_first_on_page() */
+	const trx_t*	trx)		/* in: transaction */
+{
+	ut_ad(mutex_own(&kernel_mutex));
+
+	while (lock != NULL) {
+		if (lock->trx == trx
+		    && lock->type_mode == type_mode
+		    && lock_rec_get_n_bits(lock) > heap_no) {
+
+			return(lock);
+		}
+
+		lock = lock_rec_get_next_on_page(lock);
+	}
+
+	return(NULL);
+}
+
+/*************************************************************************
+Checks if some transaction has an implicit x-lock on a record in a secondary
+index. */
+static
+trx_t*
+lock_sec_rec_some_has_impl_off_kernel(
+/*==================================*/
+				/* out: transaction which has the x-lock, or
+				NULL */
+	const rec_t*	rec,	/* in: user record */
+	dict_index_t*	index,	/* in: secondary index */
+	const ulint*	offsets)/* in: rec_get_offsets(rec, index) */
+{
+	const page_t*	page = page_align(rec);
+
+	ut_ad(mutex_own(&kernel_mutex));
+	ut_ad(!dict_index_is_clust(index));
+	ut_ad(page_rec_is_user_rec(rec));
+	ut_ad(rec_offs_validate(rec, index, offsets));
+
+	/* Some transaction may have an implicit x-lock on the record only
+	if the max trx id for the page >= min trx id for the trx list, or
+	database recovery is running. We do not write the changes of a page
+	max trx id to the log, and therefore during recovery, this value
+	for a page may be incorrect. */
+
+	if (!(ut_dulint_cmp(page_get_max_trx_id(page),
+			    trx_list_get_min_trx_id()) >= 0)
+	    && !recv_recovery_is_on()) {
+
+		return(NULL);
+	}
+
+	/* Ok, in this case it is possible that some transaction has an
+	implicit x-lock. We have to look in the clustered index. */
+
+	if (!lock_check_trx_id_sanity(page_get_max_trx_id(page),
+				      rec, index, offsets, TRUE)) {
+		buf_page_print(page, 0);
+
+		/* The page is corrupt: try to avoid a crash by returning
+		NULL */
+		return(NULL);
+	}
+
+	return(row_vers_impl_x_locked_off_kernel(rec, index, offsets));
+}
+
+/*************************************************************************
+Return approximate number or record locks (bits set in the bitmap) for
+this transaction. Since delete-marked records may be removed, the
+record count will not be precise. */
+UNIV_INTERN
+ulint
+lock_number_of_rows_locked(
+/*=======================*/
+	trx_t*	trx)	/* in: transaction */
+{
+	lock_t*	lock;
+	ulint   n_records = 0;
+	ulint	n_bits;
+	ulint	n_bit;
+
+	lock = UT_LIST_GET_FIRST(trx->trx_locks);
+
+	while (lock) {
+		if (lock_get_type_low(lock) == LOCK_REC) {
+			n_bits = lock_rec_get_n_bits(lock);
+
+			for (n_bit = 0; n_bit < n_bits; n_bit++) {
+				if (lock_rec_get_nth_bit(lock, n_bit)) {
+					n_records++;
+				}
+			}
+		}
+
+		lock = UT_LIST_GET_NEXT(trx_locks, lock);
+	}
+
+	return (n_records);
+}
+
+/*============== RECORD LOCK CREATION AND QUEUE MANAGEMENT =============*/
+
+/*************************************************************************
+Creates a new record lock and inserts it to the lock queue. Does NOT check
+for deadlocks or lock compatibility! */
+static
+lock_t*
+lock_rec_create(
+/*============*/
+					/* out: created lock */
+	ulint			type_mode,/* in: lock mode and wait
+					flag, type is ignored and
+					replaced by LOCK_REC */
+	const buf_block_t*	block,	/* in: buffer block containing
+					the record */
+	ulint			heap_no,/* in: heap number of the record */
+	dict_index_t*		index,	/* in: index of record */
+	trx_t*			trx)	/* in: transaction */
+{
+	lock_t*		lock;
+	ulint		page_no;
+	ulint		space;
+	ulint		n_bits;
+	ulint		n_bytes;
+	const page_t*	page;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	space = buf_block_get_space(block);
+	page_no	= buf_block_get_page_no(block);
+	page = block->frame;
+
+	ut_ad(!!page_is_comp(page) == dict_table_is_comp(index->table));
+
+	/* If rec is the supremum record, then we reset the gap and
+	LOCK_REC_NOT_GAP bits, as all locks on the supremum are
+	automatically of the gap type */
+
+	if (UNIV_UNLIKELY(heap_no == PAGE_HEAP_NO_SUPREMUM)) {
+		ut_ad(!(type_mode & LOCK_REC_NOT_GAP));
+
+		type_mode = type_mode & ~(LOCK_GAP | LOCK_REC_NOT_GAP);
+	}
+
+	/* Make lock bitmap bigger by a safety margin */
+	n_bits = page_dir_get_n_heap(page) + LOCK_PAGE_BITMAP_MARGIN;
+	n_bytes = 1 + n_bits / 8;
+
+	lock = mem_heap_alloc(trx->lock_heap, sizeof(lock_t) + n_bytes);
+
+	UT_LIST_ADD_LAST(trx_locks, trx->trx_locks, lock);
+
+	lock->trx = trx;
+
+	lock->type_mode = (type_mode & ~LOCK_TYPE_MASK) | LOCK_REC;
+	lock->index = index;
+
+	lock->un_member.rec_lock.space = space;
+	lock->un_member.rec_lock.page_no = page_no;
+	lock->un_member.rec_lock.n_bits = n_bytes * 8;
+
+	/* Reset to zero the bitmap which resides immediately after the
+	lock struct */
+
+	lock_rec_bitmap_reset(lock);
+
+	/* Set the bit corresponding to rec */
+	lock_rec_set_nth_bit(lock, heap_no);
+
+	HASH_INSERT(lock_t, hash, lock_sys->rec_hash,
+		    lock_rec_fold(space, page_no), lock);
+	if (UNIV_UNLIKELY(type_mode & LOCK_WAIT)) {
+
+		lock_set_lock_and_trx_wait(lock, trx);
+	}
+
+	return(lock);
+}
+
+/*************************************************************************
+Enqueues a waiting request for a lock which cannot be granted immediately.
+Checks for deadlocks. */
+static
+ulint
+lock_rec_enqueue_waiting(
+/*=====================*/
+					/* out: DB_LOCK_WAIT,
+					DB_DEADLOCK, or
+					DB_QUE_THR_SUSPENDED, or
+					DB_SUCCESS; DB_SUCCESS means
+					that there was a deadlock, but
+					another transaction was chosen
+					as a victim, and we got the
+					lock immediately: no need to
+					wait then */
+	ulint			type_mode,/* in: lock mode this
+					transaction is requesting:
+					LOCK_S or LOCK_X, possibly
+					ORed with LOCK_GAP or
+					LOCK_REC_NOT_GAP, ORed with
+					LOCK_INSERT_INTENTION if this
+					waiting lock request is set
+					when performing an insert of
+					an index record */
+	const buf_block_t*	block,	/* in: buffer block containing
+					the record */
+	ulint			heap_no,/* in: heap number of the record */
+	dict_index_t*		index,	/* in: index of record */
+	que_thr_t*		thr)	/* in: query thread */
+{
+	lock_t*	lock;
+	trx_t*	trx;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	/* Test if there already is some other reason to suspend thread:
+	we do not enqueue a lock request if the query thread should be
+	stopped anyway */
+
+	if (UNIV_UNLIKELY(que_thr_stop(thr))) {
+
+		ut_error;
+
+		return(DB_QUE_THR_SUSPENDED);
+	}
+
+	trx = thr_get_trx(thr);
+
+	switch (trx_get_dict_operation(trx)) {
+	case TRX_DICT_OP_NONE:
+		break;
+	case TRX_DICT_OP_TABLE:
+	case TRX_DICT_OP_INDEX:
+		ut_print_timestamp(stderr);
+		fputs("  InnoDB: Error: a record lock wait happens"
+		      " in a dictionary operation!\n"
+		      "InnoDB: ", stderr);
+		dict_index_name_print(stderr, trx, index);
+		fputs(".\n"
+		      "InnoDB: Submit a detailed bug report"
+		      " to http://bugs.mysql.com\n",
+		      stderr);
+	}
+
+	/* Enqueue the lock request that will wait to be granted */
+	lock = lock_rec_create(type_mode | LOCK_WAIT,
+			       block, heap_no, index, trx);
+
+	/* Check if a deadlock occurs: if yes, remove the lock request and
+	return an error code */
+
+	if (UNIV_UNLIKELY(lock_deadlock_occurs(lock, trx))) {
+
+		lock_reset_lock_and_trx_wait(lock);
+		lock_rec_reset_nth_bit(lock, heap_no);
+
+		return(DB_DEADLOCK);
+	}
+
+	/* If there was a deadlock but we chose another transaction as a
+	victim, it is possible that we already have the lock now granted! */
+
+	if (trx->wait_lock == NULL) {
+
+		return(DB_SUCCESS);
+	}
+
+	trx->que_state = TRX_QUE_LOCK_WAIT;
+	trx->was_chosen_as_deadlock_victim = FALSE;
+	trx->wait_started = time(NULL);
+
+	ut_a(que_thr_stop(thr));
+
+#ifdef UNIV_DEBUG
+	if (lock_print_waits) {
+		fprintf(stderr, "Lock wait for trx %lu in index ",
+			(ulong) ut_dulint_get_low(trx->id));
+		ut_print_name(stderr, trx, FALSE, index->name);
+	}
+#endif /* UNIV_DEBUG */
+
+	return(DB_LOCK_WAIT);
+}
+
+/*************************************************************************
+Adds a record lock request in the record queue. The request is normally
+added as the last in the queue, but if there are no waiting lock requests
+on the record, and the request to be added is not a waiting request, we
+can reuse a suitable record lock object already existing on the same page,
+just setting the appropriate bit in its bitmap. This is a low-level function
+which does NOT check for deadlocks or lock compatibility! */
+static
+lock_t*
+lock_rec_add_to_queue(
+/*==================*/
+					/* out: lock where the bit was set */
+	ulint			type_mode,/* in: lock mode, wait, gap
+					etc. flags; type is ignored
+					and replaced by LOCK_REC */
+	const buf_block_t*	block,	/* in: buffer block containing
+					the record */
+	ulint			heap_no,/* in: heap number of the record */
+	dict_index_t*		index,	/* in: index of record */
+	trx_t*			trx)	/* in: transaction */
+{
+	lock_t*	lock;
+
+	ut_ad(mutex_own(&kernel_mutex));
+#ifdef UNIV_DEBUG
+	switch (type_mode & LOCK_MODE_MASK) {
+	case LOCK_X:
+	case LOCK_S:
+		break;
+	default:
+		ut_error;
+	}
+
+	if (!(type_mode & (LOCK_WAIT | LOCK_GAP))) {
+		enum lock_mode	mode = (type_mode & LOCK_MODE_MASK) == LOCK_S
+			? LOCK_X
+			: LOCK_S;
+		lock_t*		other_lock
+			= lock_rec_other_has_expl_req(mode, 0, LOCK_WAIT,
+						      block, heap_no, trx);
+		ut_a(!other_lock);
+	}
+#endif /* UNIV_DEBUG */
+
+	type_mode |= LOCK_REC;
+
+	/* If rec is the supremum record, then we can reset the gap bit, as
+	all locks on the supremum are automatically of the gap type, and we
+	try to avoid unnecessary memory consumption of a new record lock
+	struct for a gap type lock */
+
+	if (UNIV_UNLIKELY(heap_no == PAGE_HEAP_NO_SUPREMUM)) {
+		ut_ad(!(type_mode & LOCK_REC_NOT_GAP));
+
+		/* There should never be LOCK_REC_NOT_GAP on a supremum
+		record, but let us play safe */
+
+		type_mode = type_mode & ~(LOCK_GAP | LOCK_REC_NOT_GAP);
+	}
+
+	/* Look for a waiting lock request on the same record or on a gap */
+
+	lock = lock_rec_get_first_on_page(block);
+
+	while (lock != NULL) {
+		if (lock_get_wait(lock)
+		    && (lock_rec_get_nth_bit(lock, heap_no))) {
+
+			goto somebody_waits;
+		}
+
+		lock = lock_rec_get_next_on_page(lock);
+	}
+
+	if (UNIV_LIKELY(!(type_mode & LOCK_WAIT))) {
+
+		/* Look for a similar record lock on the same page:
+		if one is found and there are no waiting lock requests,
+		we can just set the bit */
+
+		lock = lock_rec_find_similar_on_page(
+			type_mode, heap_no,
+			lock_rec_get_first_on_page(block), trx);
+
+		if (lock) {
+
+			lock_rec_set_nth_bit(lock, heap_no);
+
+			return(lock);
+		}
+	}
+
+somebody_waits:
+	return(lock_rec_create(type_mode, block, heap_no, index, trx));
+}
+
+/*************************************************************************
+This is a fast routine for locking a record in the most common cases:
+there are no explicit locks on the page, or there is just one lock, owned
+by this transaction, and of the right type_mode. This is a low-level function
+which does NOT look at implicit locks! Checks lock compatibility within
+explicit locks. This function sets a normal next-key lock, or in the case of
+a page supremum record, a gap type lock. */
+UNIV_INLINE
+ibool
+lock_rec_lock_fast(
+/*===============*/
+					/* out: TRUE if locking succeeded */
+	ibool			impl,	/* in: if TRUE, no lock is set
+					if no wait is necessary: we
+					assume that the caller will
+					set an implicit lock */
+	ulint			mode,	/* in: lock mode: LOCK_X or
+					LOCK_S possibly ORed to either
+					LOCK_GAP or LOCK_REC_NOT_GAP */
+	const buf_block_t*	block,	/* in: buffer block containing
+					the record */
+	ulint			heap_no,/* in: heap number of record */
+	dict_index_t*		index,	/* in: index of record */
+	que_thr_t*		thr)	/* in: query thread */
+{
+	lock_t*	lock;
+	trx_t*	trx;
+
+	ut_ad(mutex_own(&kernel_mutex));
+	ut_ad((LOCK_MODE_MASK & mode) != LOCK_S
+	      || lock_table_has(thr_get_trx(thr), index->table, LOCK_IS));
+	ut_ad((LOCK_MODE_MASK & mode) != LOCK_X
+	      || lock_table_has(thr_get_trx(thr), index->table, LOCK_IX));
+	ut_ad((LOCK_MODE_MASK & mode) == LOCK_S
+	      || (LOCK_MODE_MASK & mode) == LOCK_X);
+	ut_ad(mode - (LOCK_MODE_MASK & mode) == LOCK_GAP
+	      || mode - (LOCK_MODE_MASK & mode) == 0
+	      || mode - (LOCK_MODE_MASK & mode) == LOCK_REC_NOT_GAP);
+
+	lock = lock_rec_get_first_on_page(block);
+
+	trx = thr_get_trx(thr);
+
+	if (lock == NULL) {
+		if (!impl) {
+			lock_rec_create(mode, block, heap_no, index, trx);
+
+			if (srv_locks_unsafe_for_binlog
+			    || trx->isolation_level
+			    == TRX_ISO_READ_COMMITTED) {
+				trx_register_new_rec_lock(trx, index);
+			}
+		}
+
+		return(TRUE);
+	}
+
+	if (lock_rec_get_next_on_page(lock)) {
+
+		return(FALSE);
+	}
+
+	if (lock->trx != trx
+	    || lock->type_mode != (mode | LOCK_REC)
+	    || lock_rec_get_n_bits(lock) <= heap_no) {
+
+		return(FALSE);
+	}
+
+	if (!impl) {
+		/* If the nth bit of the record lock is already set then we
+		do not set a new lock bit, otherwise we do set */
+
+		if (!lock_rec_get_nth_bit(lock, heap_no)) {
+			lock_rec_set_nth_bit(lock, heap_no);
+			if (srv_locks_unsafe_for_binlog
+			    || trx->isolation_level
+			    == TRX_ISO_READ_COMMITTED) {
+				trx_register_new_rec_lock(trx, index);
+			}
+		}
+	}
+
+	return(TRUE);
+}
+
+/*************************************************************************
+This is the general, and slower, routine for locking a record. This is a
+low-level function which does NOT look at implicit locks! Checks lock
+compatibility within explicit locks. This function sets a normal next-key
+lock, or in the case of a page supremum record, a gap type lock. */
+static
+ulint
+lock_rec_lock_slow(
+/*===============*/
+					/* out: DB_SUCCESS,
+					DB_LOCK_WAIT, or error code */
+	ibool			impl,	/* in: if TRUE, no lock is set
+					if no wait is necessary: we
+					assume that the caller will
+					set an implicit lock */
+	ulint			mode,	/* in: lock mode: LOCK_X or
+					LOCK_S possibly ORed to either
+					LOCK_GAP or LOCK_REC_NOT_GAP */
+	const buf_block_t*	block,	/* in: buffer block containing
+					the record */
+	ulint			heap_no,/* in: heap number of record */
+	dict_index_t*		index,	/* in: index of record */
+	que_thr_t*		thr)	/* in: query thread */
+{
+	trx_t*	trx;
+	ulint	err;
+
+	ut_ad(mutex_own(&kernel_mutex));
+	ut_ad((LOCK_MODE_MASK & mode) != LOCK_S
+	      || lock_table_has(thr_get_trx(thr), index->table, LOCK_IS));
+	ut_ad((LOCK_MODE_MASK & mode) != LOCK_X
+	      || lock_table_has(thr_get_trx(thr), index->table, LOCK_IX));
+	ut_ad((LOCK_MODE_MASK & mode) == LOCK_S
+	      || (LOCK_MODE_MASK & mode) == LOCK_X);
+	ut_ad(mode - (LOCK_MODE_MASK & mode) == LOCK_GAP
+	      || mode - (LOCK_MODE_MASK & mode) == 0
+	      || mode - (LOCK_MODE_MASK & mode) == LOCK_REC_NOT_GAP);
+
+	trx = thr_get_trx(thr);
+
+	if (lock_rec_has_expl(mode, block, heap_no, trx)) {
+		/* The trx already has a strong enough lock on rec: do
+		nothing */
+
+		err = DB_SUCCESS;
+	} else if (lock_rec_other_has_conflicting(mode, block, heap_no, trx)) {
+
+		/* If another transaction has a non-gap conflicting request in
+		the queue, as this transaction does not have a lock strong
+		enough already granted on the record, we have to wait. */
+
+		err = lock_rec_enqueue_waiting(mode, block, heap_no,
+					       index, thr);
+
+		if (srv_locks_unsafe_for_binlog
+		    || trx->isolation_level == TRX_ISO_READ_COMMITTED) {
+			trx_register_new_rec_lock(trx, index);
+		}
+	} else {
+		if (!impl) {
+			/* Set the requested lock on the record */
+
+			lock_rec_add_to_queue(LOCK_REC | mode, block,
+					      heap_no, index, trx);
+			if (srv_locks_unsafe_for_binlog
+			    || trx->isolation_level
+			    == TRX_ISO_READ_COMMITTED) {
+				trx_register_new_rec_lock(trx, index);
+			}
+		}
+
+		err = DB_SUCCESS;
+	}
+
+	return(err);
+}
+
+/*************************************************************************
+Tries to lock the specified record in the mode requested. If not immediately
+possible, enqueues a waiting lock request. This is a low-level function
+which does NOT look at implicit locks! Checks lock compatibility within
+explicit locks. This function sets a normal next-key lock, or in the case
+of a page supremum record, a gap type lock. */
+static
+ulint
+lock_rec_lock(
+/*==========*/
+					/* out: DB_SUCCESS,
+					DB_LOCK_WAIT, or error code */
+	ibool			impl,	/* in: if TRUE, no lock is set
+					if no wait is necessary: we
+					assume that the caller will
+					set an implicit lock */
+	ulint			mode,	/* in: lock mode: LOCK_X or
+					LOCK_S possibly ORed to either
+					LOCK_GAP or LOCK_REC_NOT_GAP */
+	const buf_block_t*	block,	/* in: buffer block containing
+					the record */
+	ulint			heap_no,/* in: heap number of record */
+	dict_index_t*		index,	/* in: index of record */
+	que_thr_t*		thr)	/* in: query thread */
+{
+	ulint	err;
+
+	ut_ad(mutex_own(&kernel_mutex));
+	ut_ad((LOCK_MODE_MASK & mode) != LOCK_S
+	      || lock_table_has(thr_get_trx(thr), index->table, LOCK_IS));
+	ut_ad((LOCK_MODE_MASK & mode) != LOCK_X
+	      || lock_table_has(thr_get_trx(thr), index->table, LOCK_IX));
+	ut_ad((LOCK_MODE_MASK & mode) == LOCK_S
+	      || (LOCK_MODE_MASK & mode) == LOCK_X);
+	ut_ad(mode - (LOCK_MODE_MASK & mode) == LOCK_GAP
+	      || mode - (LOCK_MODE_MASK & mode) == LOCK_REC_NOT_GAP
+	      || mode - (LOCK_MODE_MASK & mode) == 0);
+
+	if (lock_rec_lock_fast(impl, mode, block, heap_no, index, thr)) {
+
+		/* We try a simplified and faster subroutine for the most
+		common cases */
+
+		err = DB_SUCCESS;
+	} else {
+		err = lock_rec_lock_slow(impl, mode, block,
+					 heap_no, index, thr);
+	}
+
+	return(err);
+}
+
+/*************************************************************************
+Checks if a waiting record lock request still has to wait in a queue. */
+static
+ibool
+lock_rec_has_to_wait_in_queue(
+/*==========================*/
+				/* out: TRUE if still has to wait */
+	lock_t*	wait_lock)	/* in: waiting record lock */
+{
+	lock_t*	lock;
+	ulint	space;
+	ulint	page_no;
+	ulint	heap_no;
+
+	ut_ad(mutex_own(&kernel_mutex));
+	ut_ad(lock_get_wait(wait_lock));
+	ut_ad(lock_get_type_low(wait_lock) == LOCK_REC);
+
+	space = wait_lock->un_member.rec_lock.space;
+	page_no = wait_lock->un_member.rec_lock.page_no;
+	heap_no = lock_rec_find_set_bit(wait_lock);
+
+	lock = lock_rec_get_first_on_page_addr(space, page_no);
+
+	while (lock != wait_lock) {
+
+		if (lock_rec_get_nth_bit(lock, heap_no)
+		    && lock_has_to_wait(wait_lock, lock)) {
+
+			return(TRUE);
+		}
+
+		lock = lock_rec_get_next_on_page(lock);
+	}
+
+	return(FALSE);
+}
+
+/*****************************************************************
+Grants a lock to a waiting lock request and releases the waiting
+transaction. */
+static
+void
+lock_grant(
+/*=======*/
+	lock_t*	lock)	/* in/out: waiting lock request */
+{
+	ut_ad(mutex_own(&kernel_mutex));
+
+	lock_reset_lock_and_trx_wait(lock);
+
+	if (lock_get_mode(lock) == LOCK_AUTO_INC) {
+		trx_t*		trx = lock->trx;
+		dict_table_t*	table = lock->un_member.tab_lock.table;
+
+		if (table->autoinc_trx == trx) {
+			fprintf(stderr,
+				"InnoDB: Error: trx already had"
+				" an AUTO-INC lock!\n");
+		} else {
+			table->autoinc_trx = trx;
+
+			ib_vector_push(trx->autoinc_locks, lock);
+		}
+	}
+
+#ifdef UNIV_DEBUG
+	if (lock_print_waits) {
+		fprintf(stderr, "Lock wait for trx %lu ends\n",
+			(ulong) ut_dulint_get_low(lock->trx->id));
+	}
+#endif /* UNIV_DEBUG */
+
+	/* If we are resolving a deadlock by choosing another transaction
+	as a victim, then our original transaction may not be in the
+	TRX_QUE_LOCK_WAIT state, and there is no need to end the lock wait
+	for it */
+
+	if (lock->trx->que_state == TRX_QUE_LOCK_WAIT) {
+		trx_end_lock_wait(lock->trx);
+	}
+}
+
+/*****************************************************************
+Cancels a waiting record lock request and releases the waiting transaction
+that requested it. NOTE: does NOT check if waiting lock requests behind this
+one can now be granted! */
+static
+void
+lock_rec_cancel(
+/*============*/
+	lock_t*	lock)	/* in: waiting record lock request */
+{
+	ut_ad(mutex_own(&kernel_mutex));
+	ut_ad(lock_get_type_low(lock) == LOCK_REC);
+
+	/* Reset the bit (there can be only one set bit) in the lock bitmap */
+	lock_rec_reset_nth_bit(lock, lock_rec_find_set_bit(lock));
+
+	/* Reset the wait flag and the back pointer to lock in trx */
+
+	lock_reset_lock_and_trx_wait(lock);
+
+	/* The following function releases the trx from lock wait */
+
+	trx_end_lock_wait(lock->trx);
+}
+
+/*****************************************************************
+Removes a record lock request, waiting or granted, from the queue and
+grants locks to other transactions in the queue if they now are entitled
+to a lock. NOTE: all record locks contained in in_lock are removed. */
+static
+void
+lock_rec_dequeue_from_page(
+/*=======================*/
+	lock_t*	in_lock)/* in: record lock object: all record locks which
+			are contained in this lock object are removed;
+			transactions waiting behind will get their lock
+			requests granted, if they are now qualified to it */
+{
+	ulint	space;
+	ulint	page_no;
+	lock_t*	lock;
+	trx_t*	trx;
+
+	ut_ad(mutex_own(&kernel_mutex));
+	ut_ad(lock_get_type_low(in_lock) == LOCK_REC);
+
+	trx = in_lock->trx;
+
+	space = in_lock->un_member.rec_lock.space;
+	page_no = in_lock->un_member.rec_lock.page_no;
+
+	HASH_DELETE(lock_t, hash, lock_sys->rec_hash,
+		    lock_rec_fold(space, page_no), in_lock);
+
+	UT_LIST_REMOVE(trx_locks, trx->trx_locks, in_lock);
+
+	/* Check if waiting locks in the queue can now be granted: grant
+	locks if there are no conflicting locks ahead. */
+
+	lock = lock_rec_get_first_on_page_addr(space, page_no);
+
+	while (lock != NULL) {
+		if (lock_get_wait(lock)
+		    && !lock_rec_has_to_wait_in_queue(lock)) {
+
+			/* Grant the lock */
+			lock_grant(lock);
+		}
+
+		lock = lock_rec_get_next_on_page(lock);
+	}
+}
+
+/*****************************************************************
+Removes a record lock request, waiting or granted, from the queue. */
+static
+void
+lock_rec_discard(
+/*=============*/
+	lock_t*	in_lock)/* in: record lock object: all record locks which
+			are contained in this lock object are removed */
+{
+	ulint	space;
+	ulint	page_no;
+	trx_t*	trx;
+
+	ut_ad(mutex_own(&kernel_mutex));
+	ut_ad(lock_get_type_low(in_lock) == LOCK_REC);
+
+	trx = in_lock->trx;
+
+	space = in_lock->un_member.rec_lock.space;
+	page_no = in_lock->un_member.rec_lock.page_no;
+
+	HASH_DELETE(lock_t, hash, lock_sys->rec_hash,
+		    lock_rec_fold(space, page_no), in_lock);
+
+	UT_LIST_REMOVE(trx_locks, trx->trx_locks, in_lock);
+}
+
+/*****************************************************************
+Removes record lock objects set on an index page which is discarded. This
+function does not move locks, or check for waiting locks, therefore the
+lock bitmaps must already be reset when this function is called. */
+static
+void
+lock_rec_free_all_from_discard_page(
+/*================================*/
+	const buf_block_t*	block)	/* in: page to be discarded */
+{
+	ulint	space;
+	ulint	page_no;
+	lock_t*	lock;
+	lock_t*	next_lock;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	space = buf_block_get_space(block);
+	page_no = buf_block_get_page_no(block);
+
+	lock = lock_rec_get_first_on_page_addr(space, page_no);
+
+	while (lock != NULL) {
+		ut_ad(lock_rec_find_set_bit(lock) == ULINT_UNDEFINED);
+		ut_ad(!lock_get_wait(lock));
+
+		next_lock = lock_rec_get_next_on_page(lock);
+
+		lock_rec_discard(lock);
+
+		lock = next_lock;
+	}
+}
+
+/*============= RECORD LOCK MOVING AND INHERITING ===================*/
+
+/*****************************************************************
+Resets the lock bits for a single record. Releases transactions waiting for
+lock requests here. */
+static
+void
+lock_rec_reset_and_release_wait(
+/*============================*/
+	const buf_block_t*	block,	/* in: buffer block containing
+					the record */
+	ulint			heap_no)/* in: heap number of record */
+{
+	lock_t*	lock;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	lock = lock_rec_get_first(block, heap_no);
+
+	while (lock != NULL) {
+		if (lock_get_wait(lock)) {
+			lock_rec_cancel(lock);
+		} else {
+			lock_rec_reset_nth_bit(lock, heap_no);
+		}
+
+		lock = lock_rec_get_next(heap_no, lock);
+	}
+}
+
+/*****************************************************************
+Makes a record to inherit the locks (except LOCK_INSERT_INTENTION type)
+of another record as gap type locks, but does not reset the lock bits of
+the other record. Also waiting lock requests on rec are inherited as
+GRANTED gap locks. */
+static
+void
+lock_rec_inherit_to_gap(
+/*====================*/
+	const buf_block_t*	heir_block,	/* in: block containing the
+						record which inherits */
+	const buf_block_t*	block,		/* in: block containing the
+						record from which inherited;
+						does NOT reset the locks on
+						this record */
+	ulint			heir_heap_no,	/* in: heap_no of the
+						inheriting record */
+	ulint			heap_no)	/* in: heap_no of the
+						donating record */
+{
+	lock_t*	lock;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	lock = lock_rec_get_first(block, heap_no);
+
+	/* If srv_locks_unsafe_for_binlog is TRUE or session is using
+	READ COMMITTED isolation level, we do not want locks set
+	by an UPDATE or a DELETE to be inherited as gap type locks. But we
+	DO want S-locks set by a consistency constraint to be inherited also
+	then. */
+
+	while (lock != NULL) {
+		if (!lock_rec_get_insert_intention(lock)
+		    && !((srv_locks_unsafe_for_binlog
+			  || lock->trx->isolation_level
+			  == TRX_ISO_READ_COMMITTED)
+			 && lock_get_mode(lock) == LOCK_X)) {
+
+			lock_rec_add_to_queue(LOCK_REC | LOCK_GAP
+					      | lock_get_mode(lock),
+					      heir_block, heir_heap_no,
+					      lock->index, lock->trx);
+		}
+
+		lock = lock_rec_get_next(heap_no, lock);
+	}
+}
+
+/*****************************************************************
+Makes a record to inherit the gap locks (except LOCK_INSERT_INTENTION type)
+of another record as gap type locks, but does not reset the lock bits of the
+other record. Also waiting lock requests are inherited as GRANTED gap locks. */
+static
+void
+lock_rec_inherit_to_gap_if_gap_lock(
+/*================================*/
+	const buf_block_t*	block,		/* in: buffer block */
+	ulint			heir_heap_no,	/* in: heap_no of
+						record which inherits */
+	ulint			heap_no)	/* in: heap_no of record
+						from which inherited;
+						does NOT reset the locks
+						on this record */
+{
+	lock_t*	lock;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	lock = lock_rec_get_first(block, heap_no);
+
+	while (lock != NULL) {
+		if (!lock_rec_get_insert_intention(lock)
+		    && (heap_no == PAGE_HEAP_NO_SUPREMUM
+			|| !lock_rec_get_rec_not_gap(lock))) {
+
+			lock_rec_add_to_queue(LOCK_REC | LOCK_GAP
+					      | lock_get_mode(lock),
+					      block, heir_heap_no,
+					      lock->index, lock->trx);
+		}
+
+		lock = lock_rec_get_next(heap_no, lock);
+	}
+}
+
+/*****************************************************************
+Moves the locks of a record to another record and resets the lock bits of
+the donating record. */
+static
+void
+lock_rec_move(
+/*==========*/
+	const buf_block_t*	receiver,	/* in: buffer block containing
+						the receiving record */
+	const buf_block_t*	donator,	/* in: buffer block containing
+						the donating record */
+	ulint			receiver_heap_no,/* in: heap_no of the record
+						which gets the locks; there
+						must be no lock requests
+						on it! */
+	ulint			donator_heap_no)/* in: heap_no of the record
+						which gives the locks */
+{
+	lock_t*	lock;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	lock = lock_rec_get_first(donator, donator_heap_no);
+
+	ut_ad(lock_rec_get_first(receiver, receiver_heap_no) == NULL);
+
+	while (lock != NULL) {
+		const ulint	type_mode = lock->type_mode;
+
+		lock_rec_reset_nth_bit(lock, donator_heap_no);
+
+		if (UNIV_UNLIKELY(type_mode & LOCK_WAIT)) {
+			lock_reset_lock_and_trx_wait(lock);
+		}
+
+		/* Note that we FIRST reset the bit, and then set the lock:
+		the function works also if donator == receiver */
+
+		lock_rec_add_to_queue(type_mode, receiver, receiver_heap_no,
+				      lock->index, lock->trx);
+		lock = lock_rec_get_next(donator_heap_no, lock);
+	}
+
+	ut_ad(lock_rec_get_first(donator, donator_heap_no) == NULL);
+}
+
+/*****************************************************************
+Updates the lock table when we have reorganized a page. NOTE: we copy
+also the locks set on the infimum of the page; the infimum may carry
+locks if an update of a record is occurring on the page, and its locks
+were temporarily stored on the infimum. */
+UNIV_INTERN
+void
+lock_move_reorganize_page(
+/*======================*/
+	const buf_block_t*	block,	/* in: old index page, now
+					reorganized */
+	const buf_block_t*	oblock)	/* in: copy of the old, not
+					reorganized page */
+{
+	lock_t*		lock;
+	UT_LIST_BASE_NODE_T(lock_t)	old_locks;
+	mem_heap_t*	heap		= NULL;
+	ulint		comp;
+
+	lock_mutex_enter_kernel();
+
+	lock = lock_rec_get_first_on_page(block);
+
+	if (lock == NULL) {
+		lock_mutex_exit_kernel();
+
+		return;
+	}
+
+	heap = mem_heap_create(256);
+
+	/* Copy first all the locks on the page to heap and reset the
+	bitmaps in the original locks; chain the copies of the locks
+	using the trx_locks field in them. */
+
+	UT_LIST_INIT(old_locks);
+
+	do {
+		/* Make a copy of the lock */
+		lock_t*	old_lock = lock_rec_copy(lock, heap);
+
+		UT_LIST_ADD_LAST(trx_locks, old_locks, old_lock);
+
+		/* Reset bitmap of lock */
+		lock_rec_bitmap_reset(lock);
+
+		if (lock_get_wait(lock)) {
+			lock_reset_lock_and_trx_wait(lock);
+		}
+
+		lock = lock_rec_get_next_on_page(lock);
+	} while (lock != NULL);
+
+	comp = page_is_comp(block->frame);
+	ut_ad(comp == page_is_comp(oblock->frame));
+
+	for (lock = UT_LIST_GET_FIRST(old_locks); lock;
+	     lock = UT_LIST_GET_NEXT(trx_locks, lock)) {
+		/* NOTE: we copy also the locks set on the infimum and
+		supremum of the page; the infimum may carry locks if an
+		update of a record is occurring on the page, and its locks
+		were temporarily stored on the infimum */
+		page_cur_t	cur1;
+		page_cur_t	cur2;
+
+		page_cur_set_before_first(block, &cur1);
+		page_cur_set_before_first(oblock, &cur2);
+
+		/* Set locks according to old locks */
+		for (;;) {
+			ulint	old_heap_no;
+			ulint	new_heap_no;
+
+			ut_ad(comp || !memcmp(page_cur_get_rec(&cur1),
+					      page_cur_get_rec(&cur2),
+					      rec_get_data_size_old(
+						      page_cur_get_rec(
+							      &cur2))));
+			if (UNIV_LIKELY(comp)) {
+				old_heap_no = rec_get_heap_no_new(
+					page_cur_get_rec(&cur2));
+				new_heap_no = rec_get_heap_no_new(
+					page_cur_get_rec(&cur1));
+			} else {
+				old_heap_no = rec_get_heap_no_old(
+					page_cur_get_rec(&cur2));
+				new_heap_no = rec_get_heap_no_old(
+					page_cur_get_rec(&cur1));
+			}
+
+			if (lock_rec_get_nth_bit(lock, old_heap_no)) {
+
+				/* Clear the bit in old_lock. */
+				ut_d(lock_rec_reset_nth_bit(lock,
+							    old_heap_no));
+
+				/* NOTE that the old lock bitmap could be too
+				small for the new heap number! */
+
+				lock_rec_add_to_queue(lock->type_mode, block,
+						      new_heap_no,
+						      lock->index, lock->trx);
+
+				/* if (new_heap_no == PAGE_HEAP_NO_SUPREMUM
+				&& lock_get_wait(lock)) {
+				fprintf(stderr,
+				"---\n--\n!!!Lock reorg: supr type %lu\n",
+				lock->type_mode);
+				} */
+			}
+
+			if (UNIV_UNLIKELY
+			    (new_heap_no == PAGE_HEAP_NO_SUPREMUM)) {
+
+				ut_ad(old_heap_no == PAGE_HEAP_NO_SUPREMUM);
+				break;
+			}
+
+			page_cur_move_to_next(&cur1);
+			page_cur_move_to_next(&cur2);
+		}
+
+#ifdef UNIV_DEBUG
+		{
+			ulint	i = lock_rec_find_set_bit(lock);
+
+			/* Check that all locks were moved. */
+			if (UNIV_UNLIKELY(i != ULINT_UNDEFINED)) {
+				fprintf(stderr,
+					"lock_move_reorganize_page():"
+					" %lu not moved in %p\n",
+					(ulong) i, (void*) lock);
+				ut_error;
+			}
+		}
+#endif /* UNIV_DEBUG */
+	}
+
+	lock_mutex_exit_kernel();
+
+	mem_heap_free(heap);
+
+#ifdef UNIV_DEBUG_LOCK_VALIDATE
+	ut_ad(lock_rec_validate_page(buf_block_get_space(block),
+				     buf_block_get_page_no(block)));
+#endif
+}
+
+/*****************************************************************
+Moves the explicit locks on user records to another page if a record
+list end is moved to another page. */
+UNIV_INTERN
+void
+lock_move_rec_list_end(
+/*===================*/
+	const buf_block_t*	new_block,	/* in: index page to move to */
+	const buf_block_t*	block,		/* in: index page */
+	const rec_t*		rec)		/* in: record on page: this
+						is the first record moved */
+{
+	lock_t*		lock;
+	const ulint	comp	= page_rec_is_comp(rec);
+
+	lock_mutex_enter_kernel();
+
+	/* Note: when we move locks from record to record, waiting locks
+	and possible granted gap type locks behind them are enqueued in
+	the original order, because new elements are inserted to a hash
+	table to the end of the hash chain, and lock_rec_add_to_queue
+	does not reuse locks if there are waiters in the queue. */
+
+	for (lock = lock_rec_get_first_on_page(block); lock;
+	     lock = lock_rec_get_next_on_page(lock)) {
+		page_cur_t	cur1;
+		page_cur_t	cur2;
+		const ulint	type_mode = lock->type_mode;
+
+		page_cur_position(rec, block, &cur1);
+
+		if (page_cur_is_before_first(&cur1)) {
+			page_cur_move_to_next(&cur1);
+		}
+
+		page_cur_set_before_first(new_block, &cur2);
+		page_cur_move_to_next(&cur2);
+
+		/* Copy lock requests on user records to new page and
+		reset the lock bits on the old */
+
+		while (!page_cur_is_after_last(&cur1)) {
+			ulint	heap_no;
+
+			if (comp) {
+				heap_no = rec_get_heap_no_new(
+					page_cur_get_rec(&cur1));
+			} else {
+				heap_no = rec_get_heap_no_old(
+					page_cur_get_rec(&cur1));
+				ut_ad(!memcmp(page_cur_get_rec(&cur1),
+					 page_cur_get_rec(&cur2),
+					 rec_get_data_size_old(
+						 page_cur_get_rec(&cur2))));
+			}
+
+			if (lock_rec_get_nth_bit(lock, heap_no)) {
+				lock_rec_reset_nth_bit(lock, heap_no);
+
+				if (UNIV_UNLIKELY(type_mode & LOCK_WAIT)) {
+					lock_reset_lock_and_trx_wait(lock);
+				}
+
+				if (comp) {
+					heap_no = rec_get_heap_no_new(
+						page_cur_get_rec(&cur2));
+				} else {
+					heap_no = rec_get_heap_no_old(
+						page_cur_get_rec(&cur2));
+				}
+
+				lock_rec_add_to_queue(type_mode,
+						      new_block, heap_no,
+						      lock->index, lock->trx);
+			}
+
+			page_cur_move_to_next(&cur1);
+			page_cur_move_to_next(&cur2);
+		}
+	}
+
+	lock_mutex_exit_kernel();
+
+#ifdef UNIV_DEBUG_LOCK_VALIDATE
+	ut_ad(lock_rec_validate_page(buf_block_get_space(block),
+				     buf_block_get_page_no(block)));
+	ut_ad(lock_rec_validate_page(buf_block_get_space(new_block),
+				     buf_block_get_page_no(new_block)));
+#endif
+}
+
+/*****************************************************************
+Moves the explicit locks on user records to another page if a record
+list start is moved to another page. */
+UNIV_INTERN
+void
+lock_move_rec_list_start(
+/*=====================*/
+	const buf_block_t*	new_block,	/* in: index page to move to */
+	const buf_block_t*	block,		/* in: index page */
+	const rec_t*		rec,		/* in: record on page:
+						this is the first
+						record NOT copied */
+	const rec_t*		old_end)	/* in: old
+						previous-to-last
+						record on new_page
+						before the records
+						were copied */
+{
+	lock_t*		lock;
+	const ulint	comp	= page_rec_is_comp(rec);
+
+	ut_ad(block->frame == page_align(rec));
+	ut_ad(new_block->frame == page_align(old_end));
+
+	lock_mutex_enter_kernel();
+
+	for (lock = lock_rec_get_first_on_page(block); lock;
+	     lock = lock_rec_get_next_on_page(lock)) {
+		page_cur_t	cur1;
+		page_cur_t	cur2;
+		const ulint	type_mode = lock->type_mode;
+
+		page_cur_set_before_first(block, &cur1);
+		page_cur_move_to_next(&cur1);
+
+		page_cur_position(old_end, new_block, &cur2);
+		page_cur_move_to_next(&cur2);
+
+		/* Copy lock requests on user records to new page and
+		reset the lock bits on the old */
+
+		while (page_cur_get_rec(&cur1) != rec) {
+			ulint	heap_no;
+
+			if (comp) {
+				heap_no = rec_get_heap_no_new(
+					page_cur_get_rec(&cur1));
+			} else {
+				heap_no = rec_get_heap_no_old(
+					page_cur_get_rec(&cur1));
+				ut_ad(!memcmp(page_cur_get_rec(&cur1),
+					      page_cur_get_rec(&cur2),
+					      rec_get_data_size_old(
+						      page_cur_get_rec(
+							      &cur2))));
+			}
+
+			if (lock_rec_get_nth_bit(lock, heap_no)) {
+				lock_rec_reset_nth_bit(lock, heap_no);
+
+				if (UNIV_UNLIKELY(type_mode & LOCK_WAIT)) {
+					lock_reset_lock_and_trx_wait(lock);
+				}
+
+				if (comp) {
+					heap_no = rec_get_heap_no_new(
+						page_cur_get_rec(&cur2));
+				} else {
+					heap_no = rec_get_heap_no_old(
+						page_cur_get_rec(&cur2));
+				}
+
+				lock_rec_add_to_queue(type_mode,
+						      new_block, heap_no,
+						      lock->index, lock->trx);
+			}
+
+			page_cur_move_to_next(&cur1);
+			page_cur_move_to_next(&cur2);
+		}
+
+#ifdef UNIV_DEBUG
+		if (page_rec_is_supremum(rec)) {
+			ulint	i;
+
+			for (i = PAGE_HEAP_NO_USER_LOW;
+			     i < lock_rec_get_n_bits(lock); i++) {
+				if (UNIV_UNLIKELY
+				    (lock_rec_get_nth_bit(lock, i))) {
+
+					fprintf(stderr,
+						"lock_move_rec_list_start():"
+						" %lu not moved in %p\n",
+						(ulong) i, (void*) lock);
+					ut_error;
+				}
+			}
+		}
+#endif /* UNIV_DEBUG */
+	}
+
+	lock_mutex_exit_kernel();
+
+#ifdef UNIV_DEBUG_LOCK_VALIDATE
+	ut_ad(lock_rec_validate_page(buf_block_get_space(block),
+				     buf_block_get_page_no(block)));
+#endif
+}
+
+/*****************************************************************
+Updates the lock table when a page is split to the right. */
+UNIV_INTERN
+void
+lock_update_split_right(
+/*====================*/
+	const buf_block_t*	right_block,	/* in: right page */
+	const buf_block_t*	left_block)	/* in: left page */
+{
+	ulint	heap_no = lock_get_min_heap_no(right_block);
+
+	lock_mutex_enter_kernel();
+
+	/* Move the locks on the supremum of the left page to the supremum
+	of the right page */
+
+	lock_rec_move(right_block, left_block,
+		      PAGE_HEAP_NO_SUPREMUM, PAGE_HEAP_NO_SUPREMUM);
+
+	/* Inherit the locks to the supremum of left page from the successor
+	of the infimum on right page */
+
+	lock_rec_inherit_to_gap(left_block, right_block,
+				PAGE_HEAP_NO_SUPREMUM, heap_no);
+
+	lock_mutex_exit_kernel();
+}
+
+/*****************************************************************
+Updates the lock table when a page is merged to the right. */
+UNIV_INTERN
+void
+lock_update_merge_right(
+/*====================*/
+	const buf_block_t*	right_block,	/* in: right page to
+						which merged */
+	const rec_t*		orig_succ,	/* in: original
+						successor of infimum
+						on the right page
+						before merge */
+	const buf_block_t*	left_block)	/* in: merged index
+						page which will be
+						discarded */
+{
+	lock_mutex_enter_kernel();
+
+	/* Inherit the locks from the supremum of the left page to the
+	original successor of infimum on the right page, to which the left
+	page was merged */
+
+	lock_rec_inherit_to_gap(right_block, left_block,
+				page_rec_get_heap_no(orig_succ),
+				PAGE_HEAP_NO_SUPREMUM);
+
+	/* Reset the locks on the supremum of the left page, releasing
+	waiting transactions */
+
+	lock_rec_reset_and_release_wait(left_block,
+					PAGE_HEAP_NO_SUPREMUM);
+
+	lock_rec_free_all_from_discard_page(left_block);
+
+	lock_mutex_exit_kernel();
+}
+
+/*****************************************************************
+Updates the lock table when the root page is copied to another in
+btr_root_raise_and_insert. Note that we leave lock structs on the
+root page, even though they do not make sense on other than leaf
+pages: the reason is that in a pessimistic update the infimum record
+of the root page will act as a dummy carrier of the locks of the record
+to be updated. */
+UNIV_INTERN
+void
+lock_update_root_raise(
+/*===================*/
+	const buf_block_t*	block,	/* in: index page to which copied */
+	const buf_block_t*	root)	/* in: root page */
+{
+	lock_mutex_enter_kernel();
+
+	/* Move the locks on the supremum of the root to the supremum
+	of block */
+
+	lock_rec_move(block, root,
+		      PAGE_HEAP_NO_SUPREMUM, PAGE_HEAP_NO_SUPREMUM);
+	lock_mutex_exit_kernel();
+}
+
+/*****************************************************************
+Updates the lock table when a page is copied to another and the original page
+is removed from the chain of leaf pages, except if page is the root! */
+UNIV_INTERN
+void
+lock_update_copy_and_discard(
+/*=========================*/
+	const buf_block_t*	new_block,	/* in: index page to
+						which copied */
+	const buf_block_t*	block)		/* in: index page;
+						NOT the root! */
+{
+	lock_mutex_enter_kernel();
+
+	/* Move the locks on the supremum of the old page to the supremum
+	of new_page */
+
+	lock_rec_move(new_block, block,
+		      PAGE_HEAP_NO_SUPREMUM, PAGE_HEAP_NO_SUPREMUM);
+	lock_rec_free_all_from_discard_page(block);
+
+	lock_mutex_exit_kernel();
+}
+
+/*****************************************************************
+Updates the lock table when a page is split to the left. */
+UNIV_INTERN
+void
+lock_update_split_left(
+/*===================*/
+	const buf_block_t*	right_block,	/* in: right page */
+	const buf_block_t*	left_block)	/* in: left page */
+{
+	ulint	heap_no = lock_get_min_heap_no(right_block);
+
+	lock_mutex_enter_kernel();
+
+	/* Inherit the locks to the supremum of the left page from the
+	successor of the infimum on the right page */
+
+	lock_rec_inherit_to_gap(left_block, right_block,
+				PAGE_HEAP_NO_SUPREMUM, heap_no);
+
+	lock_mutex_exit_kernel();
+}
+
+/*****************************************************************
+Updates the lock table when a page is merged to the left. */
+UNIV_INTERN
+void
+lock_update_merge_left(
+/*===================*/
+	const buf_block_t*	left_block,	/* in: left page to
+						which merged */
+	const rec_t*		orig_pred,	/* in: original predecessor
+						of supremum on the left page
+						before merge */
+	const buf_block_t*	right_block)	/* in: merged index page
+						which will be discarded */
+{
+	const rec_t*	left_next_rec;
+
+	ut_ad(left_block->frame == page_align(orig_pred));
+
+	lock_mutex_enter_kernel();
+
+	left_next_rec = page_rec_get_next_const(orig_pred);
+
+	if (!page_rec_is_supremum(left_next_rec)) {
+
+		/* Inherit the locks on the supremum of the left page to the
+		first record which was moved from the right page */
+
+		lock_rec_inherit_to_gap(left_block, left_block,
+					page_rec_get_heap_no(left_next_rec),
+					PAGE_HEAP_NO_SUPREMUM);
+
+		/* Reset the locks on the supremum of the left page,
+		releasing waiting transactions */
+
+		lock_rec_reset_and_release_wait(left_block,
+						PAGE_HEAP_NO_SUPREMUM);
+	}
+
+	/* Move the locks from the supremum of right page to the supremum
+	of the left page */
+
+	lock_rec_move(left_block, right_block,
+		      PAGE_HEAP_NO_SUPREMUM, PAGE_HEAP_NO_SUPREMUM);
+
+	lock_rec_free_all_from_discard_page(right_block);
+
+	lock_mutex_exit_kernel();
+}
+
+/*****************************************************************
+Resets the original locks on heir and replaces them with gap type locks
+inherited from rec. */
+UNIV_INTERN
+void
+lock_rec_reset_and_inherit_gap_locks(
+/*=================================*/
+	const buf_block_t*	heir_block,	/* in: block containing the
+						record which inherits */
+	const buf_block_t*	block,		/* in: block containing the
+						record from which inherited;
+						does NOT reset the locks on
+						this record */
+	ulint			heir_heap_no,	/* in: heap_no of the
+						inheriting record */
+	ulint			heap_no)	/* in: heap_no of the
+						donating record */
+{
+	mutex_enter(&kernel_mutex);
+
+	lock_rec_reset_and_release_wait(heir_block, heir_heap_no);
+
+	lock_rec_inherit_to_gap(heir_block, block, heir_heap_no, heap_no);
+
+	mutex_exit(&kernel_mutex);
+}
+
+/*****************************************************************
+Updates the lock table when a page is discarded. */
+UNIV_INTERN
+void
+lock_update_discard(
+/*================*/
+	const buf_block_t*	heir_block,	/* in: index page
+						which will inherit the locks */
+	ulint			heir_heap_no,	/* in: heap_no of the record
+						which will inherit the locks */
+	const buf_block_t*	block)		/* in: index page
+						which will be discarded */
+{
+	const page_t*	page = block->frame;
+	const rec_t*	rec;
+	ulint		heap_no;
+
+	lock_mutex_enter_kernel();
+
+	if (!lock_rec_get_first_on_page(block)) {
+		/* No locks exist on page, nothing to do */
+
+		lock_mutex_exit_kernel();
+
+		return;
+	}
+
+	/* Inherit all the locks on the page to the record and reset all
+	the locks on the page */
+
+	if (page_is_comp(page)) {
+		rec = page + PAGE_NEW_INFIMUM;
+
+		do {
+			heap_no = rec_get_heap_no_new(rec);
+
+			lock_rec_inherit_to_gap(heir_block, block,
+						heir_heap_no, heap_no);
+
+			lock_rec_reset_and_release_wait(block, heap_no);
+
+			rec = page + rec_get_next_offs(rec, TRUE);
+		} while (heap_no != PAGE_HEAP_NO_SUPREMUM);
+	} else {
+		rec = page + PAGE_OLD_INFIMUM;
+
+		do {
+			heap_no = rec_get_heap_no_old(rec);
+
+			lock_rec_inherit_to_gap(heir_block, block,
+						heir_heap_no, heap_no);
+
+			lock_rec_reset_and_release_wait(block, heap_no);
+
+			rec = page + rec_get_next_offs(rec, FALSE);
+		} while (heap_no != PAGE_HEAP_NO_SUPREMUM);
+	}
+
+	lock_rec_free_all_from_discard_page(block);
+
+	lock_mutex_exit_kernel();
+}
+
+/*****************************************************************
+Updates the lock table when a new user record is inserted. */
+UNIV_INTERN
+void
+lock_update_insert(
+/*===============*/
+	const buf_block_t*	block,	/* in: buffer block containing rec */
+	const rec_t*		rec)	/* in: the inserted record */
+{
+	ulint	receiver_heap_no;
+	ulint	donator_heap_no;
+
+	ut_ad(block->frame == page_align(rec));
+
+	/* Inherit the gap-locking locks for rec, in gap mode, from the next
+	record */
+
+	if (page_rec_is_comp(rec)) {
+		receiver_heap_no = rec_get_heap_no_new(rec);
+		donator_heap_no = rec_get_heap_no_new(
+			page_rec_get_next_low(rec, TRUE));
+	} else {
+		receiver_heap_no = rec_get_heap_no_old(rec);
+		donator_heap_no = rec_get_heap_no_old(
+			page_rec_get_next_low(rec, FALSE));
+	}
+
+	lock_mutex_enter_kernel();
+	lock_rec_inherit_to_gap_if_gap_lock(block,
+					    receiver_heap_no, donator_heap_no);
+	lock_mutex_exit_kernel();
+}
+
+/*****************************************************************
+Updates the lock table when a record is removed. */
+UNIV_INTERN
+void
+lock_update_delete(
+/*===============*/
+	const buf_block_t*	block,	/* in: buffer block containing rec */
+	const rec_t*		rec)	/* in: the record to be removed */
+{
+	const page_t*	page = block->frame;
+	ulint		heap_no;
+	ulint		next_heap_no;
+
+	ut_ad(page == page_align(rec));
+
+	if (page_is_comp(page)) {
+		heap_no = rec_get_heap_no_new(rec);
+		next_heap_no = rec_get_heap_no_new(page
+						   + rec_get_next_offs(rec,
+								       TRUE));
+	} else {
+		heap_no = rec_get_heap_no_old(rec);
+		next_heap_no = rec_get_heap_no_old(page
+						   + rec_get_next_offs(rec,
+								       FALSE));
+	}
+
+	lock_mutex_enter_kernel();
+
+	/* Let the next record inherit the locks from rec, in gap mode */
+
+	lock_rec_inherit_to_gap(block, block, next_heap_no, heap_no);
+
+	/* Reset the lock bits on rec and release waiting transactions */
+
+	lock_rec_reset_and_release_wait(block, heap_no);
+
+	lock_mutex_exit_kernel();
+}
+
+/*************************************************************************
+Stores on the page infimum record the explicit locks of another record.
+This function is used to store the lock state of a record when it is
+updated and the size of the record changes in the update. The record
+is moved in such an update, perhaps to another page. The infimum record
+acts as a dummy carrier record, taking care of lock releases while the
+actual record is being moved. */
+UNIV_INTERN
+void
+lock_rec_store_on_page_infimum(
+/*===========================*/
+	const buf_block_t*	block,	/* in: buffer block containing rec */
+	const rec_t*		rec)	/* in: record whose lock state
+					is stored on the infimum
+					record of the same page; lock
+					bits are reset on the
+					record */
+{
+	ulint	heap_no = page_rec_get_heap_no(rec);
+
+	ut_ad(block->frame == page_align(rec));
+
+	lock_mutex_enter_kernel();
+
+	lock_rec_move(block, block, PAGE_HEAP_NO_INFIMUM, heap_no);
+
+	lock_mutex_exit_kernel();
+}
+
+/*************************************************************************
+Restores the state of explicit lock requests on a single record, where the
+state was stored on the infimum of the page. */
+UNIV_INTERN
+void
+lock_rec_restore_from_page_infimum(
+/*===============================*/
+	const buf_block_t*	block,	/* in: buffer block containing rec */
+	const rec_t*		rec,	/* in: record whose lock state
+					is restored */
+	const buf_block_t*	donator)/* in: page (rec is not
+					necessarily on this page)
+					whose infimum stored the lock
+					state; lock bits are reset on
+					the infimum */
+{
+	ulint	heap_no = page_rec_get_heap_no(rec);
+
+	lock_mutex_enter_kernel();
+
+	lock_rec_move(block, donator, heap_no, PAGE_HEAP_NO_INFIMUM);
+
+	lock_mutex_exit_kernel();
+}
+
+/*=========== DEADLOCK CHECKING ======================================*/
+
+/************************************************************************
+Checks if a lock request results in a deadlock. */
+static
+ibool
+lock_deadlock_occurs(
+/*=================*/
+			/* out: TRUE if a deadlock was detected and we
+			chose trx as a victim; FALSE if no deadlock, or
+			there was a deadlock, but we chose other
+			transaction(s) as victim(s) */
+	lock_t*	lock,	/* in: lock the transaction is requesting */
+	trx_t*	trx)	/* in: transaction */
+{
+	dict_table_t*	table;
+	dict_index_t*	index;
+	trx_t*		mark_trx;
+	ulint		ret;
+	ulint		cost	= 0;
+
+	ut_ad(trx);
+	ut_ad(lock);
+	ut_ad(mutex_own(&kernel_mutex));
+retry:
+	/* We check that adding this trx to the waits-for graph
+	does not produce a cycle. First mark all active transactions
+	with 0: */
+
+	mark_trx = UT_LIST_GET_FIRST(trx_sys->trx_list);
+
+	while (mark_trx) {
+		mark_trx->deadlock_mark = 0;
+		mark_trx = UT_LIST_GET_NEXT(trx_list, mark_trx);
+	}
+
+	ret = lock_deadlock_recursive(trx, trx, lock, &cost, 0);
+
+	if (ret == LOCK_VICTIM_IS_OTHER) {
+		/* We chose some other trx as a victim: retry if there still
+		is a deadlock */
+
+		goto retry;
+	}
+
+	if (UNIV_UNLIKELY(ret == LOCK_VICTIM_IS_START)) {
+		if (lock_get_type_low(lock) & LOCK_TABLE) {
+			table = lock->un_member.tab_lock.table;
+			index = NULL;
+		} else {
+			index = lock->index;
+			table = index->table;
+		}
+
+		lock_deadlock_found = TRUE;
+
+		fputs("*** WE ROLL BACK TRANSACTION (2)\n",
+		      lock_latest_err_file);
+
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
+
+/************************************************************************
+Looks recursively for a deadlock. */
+static
+ulint
+lock_deadlock_recursive(
+/*====================*/
+				/* out: 0 if no deadlock found,
+				LOCK_VICTIM_IS_START if there was a deadlock
+				and we chose 'start' as the victim,
+				LOCK_VICTIM_IS_OTHER if a deadlock
+				was found and we chose some other trx as a
+				victim: we must do the search again in this
+				last case because there may be another
+				deadlock! */
+	trx_t*	start,		/* in: recursion starting point */
+	trx_t*	trx,		/* in: a transaction waiting for a lock */
+	lock_t*	wait_lock,	/* in: the lock trx is waiting to be granted */
+	ulint*	cost,		/* in/out: number of calculation steps thus
+				far: if this exceeds LOCK_MAX_N_STEPS_...
+				we return LOCK_VICTIM_IS_START */
+	ulint	depth)		/* in: recursion depth: if this exceeds
+				LOCK_MAX_DEPTH_IN_DEADLOCK_CHECK, we
+				return LOCK_VICTIM_IS_START */
+{
+	lock_t*	lock;
+	ulint	bit_no		= ULINT_UNDEFINED;
+	trx_t*	lock_trx;
+	ulint	ret;
+
+	ut_a(trx);
+	ut_a(start);
+	ut_a(wait_lock);
+	ut_ad(mutex_own(&kernel_mutex));
+
+	if (trx->deadlock_mark == 1) {
+		/* We have already exhaustively searched the subtree starting
+		from this trx */
+
+		return(0);
+	}
+
+	*cost = *cost + 1;
+
+	lock = wait_lock;
+
+	if (lock_get_type_low(wait_lock) == LOCK_REC) {
+
+		bit_no = lock_rec_find_set_bit(wait_lock);
+
+		ut_a(bit_no != ULINT_UNDEFINED);
+	}
+
+	/* Look at the locks ahead of wait_lock in the lock queue */
+
+	for (;;) {
+		if (lock_get_type_low(lock) & LOCK_TABLE) {
+
+			lock = UT_LIST_GET_PREV(un_member.tab_lock.locks,
+						lock);
+		} else {
+			ut_ad(lock_get_type_low(lock) == LOCK_REC);
+			ut_a(bit_no != ULINT_UNDEFINED);
+
+			lock = (lock_t*) lock_rec_get_prev(lock, bit_no);
+		}
+
+		if (lock == NULL) {
+			/* We can mark this subtree as searched */
+			trx->deadlock_mark = 1;
+
+			return(FALSE);
+		}
+
+		if (lock_has_to_wait(wait_lock, lock)) {
+
+			ibool	too_far
+				= depth > LOCK_MAX_DEPTH_IN_DEADLOCK_CHECK
+				|| *cost > LOCK_MAX_N_STEPS_IN_DEADLOCK_CHECK;
+
+			lock_trx = lock->trx;
+
+			if (lock_trx == start || too_far) {
+
+				/* We came back to the recursion starting
+				point: a deadlock detected; or we have
+				searched the waits-for graph too long */
+
+				FILE*	ef = lock_latest_err_file;
+
+				rewind(ef);
+				ut_print_timestamp(ef);
+
+				fputs("\n*** (1) TRANSACTION:\n", ef);
+
+				trx_print(ef, wait_lock->trx, 3000);
+
+				fputs("*** (1) WAITING FOR THIS LOCK"
+				      " TO BE GRANTED:\n", ef);
+
+				if (lock_get_type_low(wait_lock) == LOCK_REC) {
+					lock_rec_print(ef, wait_lock);
+				} else {
+					lock_table_print(ef, wait_lock);
+				}
+
+				fputs("*** (2) TRANSACTION:\n", ef);
+
+				trx_print(ef, lock->trx, 3000);
+
+				fputs("*** (2) HOLDS THE LOCK(S):\n", ef);
+
+				if (lock_get_type_low(lock) == LOCK_REC) {
+					lock_rec_print(ef, lock);
+				} else {
+					lock_table_print(ef, lock);
+				}
+
+				fputs("*** (2) WAITING FOR THIS LOCK"
+				      " TO BE GRANTED:\n", ef);
+
+				if (lock_get_type_low(start->wait_lock)
+				    == LOCK_REC) {
+					lock_rec_print(ef, start->wait_lock);
+				} else {
+					lock_table_print(ef, start->wait_lock);
+				}
+#ifdef UNIV_DEBUG
+				if (lock_print_waits) {
+					fputs("Deadlock detected"
+					      " or too long search\n",
+					      stderr);
+				}
+#endif /* UNIV_DEBUG */
+				if (too_far) {
+
+					fputs("TOO DEEP OR LONG SEARCH"
+					      " IN THE LOCK TABLE"
+					      " WAITS-FOR GRAPH\n", ef);
+
+					return(LOCK_VICTIM_IS_START);
+				}
+
+				if (trx_weight_cmp(wait_lock->trx,
+						   start) >= 0) {
+					/* Our recursion starting point
+					transaction is 'smaller', let us
+					choose 'start' as the victim and roll
+					back it */
+
+					return(LOCK_VICTIM_IS_START);
+				}
+
+				lock_deadlock_found = TRUE;
+
+				/* Let us choose the transaction of wait_lock
+				as a victim to try to avoid deadlocking our
+				recursion starting point transaction */
+
+				fputs("*** WE ROLL BACK TRANSACTION (1)\n",
+				      ef);
+
+				wait_lock->trx->was_chosen_as_deadlock_victim
+					= TRUE;
+
+				lock_cancel_waiting_and_release(wait_lock);
+
+				/* Since trx and wait_lock are no longer
+				in the waits-for graph, we can return FALSE;
+				note that our selective algorithm can choose
+				several transactions as victims, but still
+				we may end up rolling back also the recursion
+				starting point transaction! */
+
+				return(LOCK_VICTIM_IS_OTHER);
+			}
+
+			if (lock_trx->que_state == TRX_QUE_LOCK_WAIT) {
+
+				/* Another trx ahead has requested lock	in an
+				incompatible mode, and is itself waiting for
+				a lock */
+
+				ret = lock_deadlock_recursive(
+					start, lock_trx,
+					lock_trx->wait_lock, cost, depth + 1);
+				if (ret != 0) {
+
+					return(ret);
+				}
+			}
+		}
+	}/* end of the 'for (;;)'-loop */
+}
+
+/*========================= TABLE LOCKS ==============================*/
+
+/*************************************************************************
+Creates a table lock object and adds it as the last in the lock queue
+of the table. Does NOT check for deadlocks or lock compatibility. */
+UNIV_INLINE
+lock_t*
+lock_table_create(
+/*==============*/
+				/* out, own: new lock object */
+	dict_table_t*	table,	/* in: database table in dictionary cache */
+	ulint		type_mode,/* in: lock mode possibly ORed with
+				LOCK_WAIT */
+	trx_t*		trx)	/* in: trx */
+{
+	lock_t*	lock;
+
+	ut_ad(table && trx);
+	ut_ad(mutex_own(&kernel_mutex));
+
+	if ((type_mode & LOCK_MODE_MASK) == LOCK_AUTO_INC) {
+		++table->n_waiting_or_granted_auto_inc_locks;
+	}
+
+	/* For AUTOINC locking we reuse the lock instance only if
+	there is no wait involved else we allocate the waiting lock
+	from the transaction lock heap. */
+	if (type_mode == LOCK_AUTO_INC) {
+
+		lock = table->autoinc_lock;
+
+		table->autoinc_trx = trx;
+
+		ib_vector_push(trx->autoinc_locks, lock);
+	} else {
+		lock = mem_heap_alloc(trx->lock_heap, sizeof(lock_t));
+	}
+
+	UT_LIST_ADD_LAST(trx_locks, trx->trx_locks, lock);
+
+	lock->type_mode = type_mode | LOCK_TABLE;
+	lock->trx = trx;
+
+	lock->un_member.tab_lock.table = table;
+
+	UT_LIST_ADD_LAST(un_member.tab_lock.locks, table->locks, lock);
+
+	if (UNIV_UNLIKELY(type_mode & LOCK_WAIT)) {
+
+		lock_set_lock_and_trx_wait(lock, trx);
+	}
+
+	return(lock);
+}
+
+/*****************************************************************
+Removes a table lock request from the queue and the trx list of locks;
+this is a low-level function which does NOT check if waiting requests
+can now be granted. */
+UNIV_INLINE
+void
+lock_table_remove_low(
+/*==================*/
+	lock_t*	lock)	/* in: table lock */
+{
+	trx_t*		trx;
+	dict_table_t*	table;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	trx = lock->trx;
+	table = lock->un_member.tab_lock.table;
+
+	/* Remove the table from the transaction's AUTOINC vector, if
+	the lock that is being release is an AUTOINC lock. */
+	if (lock_get_mode(lock) == LOCK_AUTO_INC) {
+
+		/* The table's AUTOINC lock can get transferred to
+		another transaction before we get here. */
+		if (table->autoinc_trx == trx) {
+			table->autoinc_trx = NULL;
+		}
+
+		/* The locks must be freed in the reverse order from
+		the one in which they were acquired. This is to avoid
+		traversing the AUTOINC lock vector unnecessarily. 
+
+		We only store locks that were granted in the
+		trx->autoinc_locks vector (see lock_table_create()
+		and lock_grant()). Therefore it can be empty and we
+		need to check for that. */
+
+		if (!ib_vector_is_empty(trx->autoinc_locks)) {
+			lock_t*	autoinc_lock;
+
+			autoinc_lock = ib_vector_pop(trx->autoinc_locks);
+			ut_a(autoinc_lock == lock);
+		}
+
+		ut_a(table->n_waiting_or_granted_auto_inc_locks > 0);
+		--table->n_waiting_or_granted_auto_inc_locks;
+	}
+
+	UT_LIST_REMOVE(trx_locks, trx->trx_locks, lock);
+	UT_LIST_REMOVE(un_member.tab_lock.locks, table->locks, lock);
+}
+
+/*************************************************************************
+Enqueues a waiting request for a table lock which cannot be granted
+immediately. Checks for deadlocks. */
+static
+ulint
+lock_table_enqueue_waiting(
+/*=======================*/
+				/* out: DB_LOCK_WAIT, DB_DEADLOCK, or
+				DB_QUE_THR_SUSPENDED, or DB_SUCCESS;
+				DB_SUCCESS means that there was a deadlock,
+				but another transaction was chosen as a
+				victim, and we got the lock immediately:
+				no need to wait then */
+	ulint		mode,	/* in: lock mode this transaction is
+				requesting */
+	dict_table_t*	table,	/* in: table */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	lock_t*	lock;
+	trx_t*	trx;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	/* Test if there already is some other reason to suspend thread:
+	we do not enqueue a lock request if the query thread should be
+	stopped anyway */
+
+	if (que_thr_stop(thr)) {
+		ut_error;
+
+		return(DB_QUE_THR_SUSPENDED);
+	}
+
+	trx = thr_get_trx(thr);
+
+	switch (trx_get_dict_operation(trx)) {
+	case TRX_DICT_OP_NONE:
+		break;
+	case TRX_DICT_OP_TABLE:
+	case TRX_DICT_OP_INDEX:
+		ut_print_timestamp(stderr);
+		fputs("  InnoDB: Error: a table lock wait happens"
+		      " in a dictionary operation!\n"
+		      "InnoDB: Table name ", stderr);
+		ut_print_name(stderr, trx, TRUE, table->name);
+		fputs(".\n"
+		      "InnoDB: Submit a detailed bug report"
+		      " to http://bugs.mysql.com\n",
+		      stderr);
+	}
+
+	/* Enqueue the lock request that will wait to be granted */
+
+	lock = lock_table_create(table, mode | LOCK_WAIT, trx);
+
+	/* Check if a deadlock occurs: if yes, remove the lock request and
+	return an error code */
+
+	if (lock_deadlock_occurs(lock, trx)) {
+
+		lock_reset_lock_and_trx_wait(lock);
+		lock_table_remove_low(lock);
+
+		return(DB_DEADLOCK);
+	}
+
+	if (trx->wait_lock == NULL) {
+		/* Deadlock resolution chose another transaction as a victim,
+		and we accidentally got our lock granted! */
+
+		return(DB_SUCCESS);
+	}
+
+	trx->que_state = TRX_QUE_LOCK_WAIT;
+	trx->was_chosen_as_deadlock_victim = FALSE;
+	trx->wait_started = time(NULL);
+
+	ut_a(que_thr_stop(thr));
+
+	return(DB_LOCK_WAIT);
+}
+
+/*************************************************************************
+Checks if other transactions have an incompatible mode lock request in
+the lock queue. */
+UNIV_INLINE
+ibool
+lock_table_other_has_incompatible(
+/*==============================*/
+	trx_t*		trx,	/* in: transaction, or NULL if all
+				transactions should be included */
+	ulint		wait,	/* in: LOCK_WAIT if also waiting locks are
+				taken into account, or 0 if not */
+	dict_table_t*	table,	/* in: table */
+	enum lock_mode	mode)	/* in: lock mode */
+{
+	lock_t*	lock;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	lock = UT_LIST_GET_LAST(table->locks);
+
+	while (lock != NULL) {
+
+		if ((lock->trx != trx)
+		    && (!lock_mode_compatible(lock_get_mode(lock), mode))
+		    && (wait || !(lock_get_wait(lock)))) {
+
+			return(TRUE);
+		}
+
+		lock = UT_LIST_GET_PREV(un_member.tab_lock.locks, lock);
+	}
+
+	return(FALSE);
+}
+
+/*************************************************************************
+Locks the specified database table in the mode given. If the lock cannot
+be granted immediately, the query thread is put to wait. */
+UNIV_INTERN
+ulint
+lock_table(
+/*=======*/
+				/* out: DB_SUCCESS, DB_LOCK_WAIT,
+				DB_DEADLOCK, or DB_QUE_THR_SUSPENDED */
+	ulint		flags,	/* in: if BTR_NO_LOCKING_FLAG bit is set,
+				does nothing */
+	dict_table_t*	table,	/* in: database table in dictionary cache */
+	enum lock_mode	mode,	/* in: lock mode */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	trx_t*	trx;
+	ulint	err;
+
+	ut_ad(table && thr);
+
+	if (flags & BTR_NO_LOCKING_FLAG) {
+
+		return(DB_SUCCESS);
+	}
+
+	ut_a(flags == 0);
+
+	trx = thr_get_trx(thr);
+
+	lock_mutex_enter_kernel();
+
+	/* Look for stronger locks the same trx already has on the table */
+
+	if (lock_table_has(trx, table, mode)) {
+
+		lock_mutex_exit_kernel();
+
+		return(DB_SUCCESS);
+	}
+
+	/* We have to check if the new lock is compatible with any locks
+	other transactions have in the table lock queue. */
+
+	if (lock_table_other_has_incompatible(trx, LOCK_WAIT, table, mode)) {
+
+		/* Another trx has a request on the table in an incompatible
+		mode: this trx may have to wait */
+
+		err = lock_table_enqueue_waiting(mode | flags, table, thr);
+
+		lock_mutex_exit_kernel();
+
+		return(err);
+	}
+
+	lock_table_create(table, mode | flags, trx);
+
+	ut_a(!flags || mode == LOCK_S || mode == LOCK_X);
+
+	lock_mutex_exit_kernel();
+
+	return(DB_SUCCESS);
+}
+
+/*************************************************************************
+Checks if there are any locks set on the table. */
+UNIV_INTERN
+ibool
+lock_is_on_table(
+/*=============*/
+				/* out: TRUE if there are lock(s) */
+	dict_table_t*	table)	/* in: database table in dictionary cache */
+{
+	ibool	ret;
+
+	ut_ad(table);
+
+	lock_mutex_enter_kernel();
+
+	if (UT_LIST_GET_LAST(table->locks)) {
+		ret = TRUE;
+	} else {
+		ret = FALSE;
+	}
+
+	lock_mutex_exit_kernel();
+
+	return(ret);
+}
+
+/*************************************************************************
+Checks if a waiting table lock request still has to wait in a queue. */
+static
+ibool
+lock_table_has_to_wait_in_queue(
+/*============================*/
+				/* out: TRUE if still has to wait */
+	lock_t*	wait_lock)	/* in: waiting table lock */
+{
+	dict_table_t*	table;
+	lock_t*		lock;
+
+	ut_ad(mutex_own(&kernel_mutex));
+	ut_ad(lock_get_wait(wait_lock));
+
+	table = wait_lock->un_member.tab_lock.table;
+
+	lock = UT_LIST_GET_FIRST(table->locks);
+
+	while (lock != wait_lock) {
+
+		if (lock_has_to_wait(wait_lock, lock)) {
+
+			return(TRUE);
+		}
+
+		lock = UT_LIST_GET_NEXT(un_member.tab_lock.locks, lock);
+	}
+
+	return(FALSE);
+}
+
+/*****************************************************************
+Removes a table lock request, waiting or granted, from the queue and grants
+locks to other transactions in the queue, if they now are entitled to a
+lock. */
+static
+void
+lock_table_dequeue(
+/*===============*/
+	lock_t*	in_lock)/* in: table lock object; transactions waiting
+			behind will get their lock requests granted, if
+			they are now qualified to it */
+{
+	lock_t*	lock;
+
+	ut_ad(mutex_own(&kernel_mutex));
+	ut_a(lock_get_type_low(in_lock) == LOCK_TABLE);
+
+	lock = UT_LIST_GET_NEXT(un_member.tab_lock.locks, in_lock);
+
+	lock_table_remove_low(in_lock);
+
+	/* Check if waiting locks in the queue can now be granted: grant
+	locks if there are no conflicting locks ahead. */
+
+	while (lock != NULL) {
+
+		if (lock_get_wait(lock)
+		    && !lock_table_has_to_wait_in_queue(lock)) {
+
+			/* Grant the lock */
+			lock_grant(lock);
+		}
+
+		lock = UT_LIST_GET_NEXT(un_member.tab_lock.locks, lock);
+	}
+}
+
+/*=========================== LOCK RELEASE ==============================*/
+
+/*****************************************************************
+Removes a granted record lock of a transaction from the queue and grants
+locks to other transactions waiting in the queue if they now are entitled
+to a lock. */
+UNIV_INTERN
+void
+lock_rec_unlock(
+/*============*/
+	trx_t*			trx,	/* in: transaction that has
+					set a record lock */
+	const buf_block_t*	block,	/* in: buffer block containing rec */
+	const rec_t*		rec,	/* in: record */
+	enum lock_mode		lock_mode)/* in: LOCK_S or LOCK_X */
+{
+	lock_t*	lock;
+	lock_t*	release_lock	= NULL;
+	ulint	heap_no;
+
+	ut_ad(trx && rec);
+	ut_ad(block->frame == page_align(rec));
+
+	heap_no = page_rec_get_heap_no(rec);
+
+	mutex_enter(&kernel_mutex);
+
+	lock = lock_rec_get_first(block, heap_no);
+
+	/* Find the last lock with the same lock_mode and transaction
+	from the record. */
+
+	while (lock != NULL) {
+		if (lock->trx == trx && lock_get_mode(lock) == lock_mode) {
+			release_lock = lock;
+			ut_a(!lock_get_wait(lock));
+		}
+
+		lock = lock_rec_get_next(heap_no, lock);
+	}
+
+	/* If a record lock is found, release the record lock */
+
+	if (UNIV_LIKELY(release_lock != NULL)) {
+		lock_rec_reset_nth_bit(release_lock, heap_no);
+	} else {
+		mutex_exit(&kernel_mutex);
+		ut_print_timestamp(stderr);
+		fprintf(stderr,
+			"  InnoDB: Error: unlock row could not"
+			" find a %lu mode lock on the record\n",
+			(ulong) lock_mode);
+
+		return;
+	}
+
+	/* Check if we can now grant waiting lock requests */
+
+	lock = lock_rec_get_first(block, heap_no);
+
+	while (lock != NULL) {
+		if (lock_get_wait(lock)
+		    && !lock_rec_has_to_wait_in_queue(lock)) {
+
+			/* Grant the lock */
+			lock_grant(lock);
+		}
+
+		lock = lock_rec_get_next(heap_no, lock);
+	}
+
+	mutex_exit(&kernel_mutex);
+}
+
+/*************************************************************************
+Releases a table lock.
+Releases possible other transactions waiting for this lock. */
+UNIV_INTERN
+void
+lock_table_unlock(
+/*==============*/
+	lock_t*	lock)	/* in: lock */
+{
+	mutex_enter(&kernel_mutex);
+
+	lock_table_dequeue(lock);
+
+	mutex_exit(&kernel_mutex);
+}
+
+/*************************************************************************
+Releases transaction locks, and releases possible other transactions waiting
+because of these locks. */
+UNIV_INTERN
+void
+lock_release_off_kernel(
+/*====================*/
+	trx_t*	trx)	/* in: transaction */
+{
+	dict_table_t*	table;
+	ulint		count;
+	lock_t*		lock;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	lock = UT_LIST_GET_LAST(trx->trx_locks);
+
+	count = 0;
+
+	while (lock != NULL) {
+
+		count++;
+
+		if (lock_get_type_low(lock) == LOCK_REC) {
+
+			lock_rec_dequeue_from_page(lock);
+		} else {
+			ut_ad(lock_get_type_low(lock) & LOCK_TABLE);
+
+			if (lock_get_mode(lock) != LOCK_IS
+			    && !ut_dulint_is_zero(trx->undo_no)) {
+
+				/* The trx may have modified the table. We
+				block the use of the MySQL query cache for
+				all currently active transactions. */
+
+				table = lock->un_member.tab_lock.table;
+
+				table->query_cache_inv_trx_id
+					= trx_sys->max_trx_id;
+			}
+
+			lock_table_dequeue(lock);
+		}
+
+		if (count == LOCK_RELEASE_KERNEL_INTERVAL) {
+			/* Release the kernel mutex for a while, so that we
+			do not monopolize it */
+
+			lock_mutex_exit_kernel();
+
+			lock_mutex_enter_kernel();
+
+			count = 0;
+		}
+
+		lock = UT_LIST_GET_LAST(trx->trx_locks);
+	}
+
+	ut_a(ib_vector_size(trx->autoinc_locks) == 0);
+
+	mem_heap_empty(trx->lock_heap);
+}
+
+/*************************************************************************
+Cancels a waiting lock request and releases possible other transactions
+waiting behind it. */
+UNIV_INTERN
+void
+lock_cancel_waiting_and_release(
+/*============================*/
+	lock_t*	lock)	/* in: waiting lock request */
+{
+	ut_ad(mutex_own(&kernel_mutex));
+
+	if (lock_get_type_low(lock) == LOCK_REC) {
+
+		lock_rec_dequeue_from_page(lock);
+	} else {
+		ut_ad(lock_get_type_low(lock) & LOCK_TABLE);
+
+		if (lock->trx->autoinc_locks != NULL) {
+			/* Release the transaction's AUTOINC locks/ */
+			lock_release_autoinc_locks(lock->trx);
+		}
+
+		lock_table_dequeue(lock);
+	}
+
+	/* Reset the wait flag and the back pointer to lock in trx */
+
+	lock_reset_lock_and_trx_wait(lock);
+
+	/* The following function releases the trx from lock wait */
+
+	trx_end_lock_wait(lock->trx);
+}
+
+/* True if a lock mode is S or X */
+#define IS_LOCK_S_OR_X(lock) \
+	(lock_get_mode(lock) == LOCK_S \
+	 || lock_get_mode(lock) == LOCK_X)
+
+
+/*************************************************************************
+Removes locks of a transaction on a table to be dropped.
+If remove_also_table_sx_locks is TRUE then table-level S and X locks are
+also removed in addition to other table-level and record-level locks.
+No lock, that is going to be removed, is allowed to be a wait lock. */
+static
+void
+lock_remove_all_on_table_for_trx(
+/*=============================*/
+	dict_table_t*	table,			/* in: table to be dropped */
+	trx_t*		trx,			/* in: a transaction */
+	ibool		remove_also_table_sx_locks)/* in: also removes
+						table S and X locks */
+{
+	lock_t*	lock;
+	lock_t*	prev_lock;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	lock = UT_LIST_GET_LAST(trx->trx_locks);
+
+	while (lock != NULL) {
+		prev_lock = UT_LIST_GET_PREV(trx_locks, lock);
+
+		if (lock_get_type_low(lock) == LOCK_REC
+		    && lock->index->table == table) {
+			ut_a(!lock_get_wait(lock));
+
+			lock_rec_discard(lock);
+		} else if (lock_get_type_low(lock) & LOCK_TABLE
+			   && lock->un_member.tab_lock.table == table
+			   && (remove_also_table_sx_locks
+			       || !IS_LOCK_S_OR_X(lock))) {
+
+			ut_a(!lock_get_wait(lock));
+
+			lock_table_remove_low(lock);
+		}
+
+		lock = prev_lock;
+	}
+}
+
+/*************************************************************************
+Removes locks on a table to be dropped or truncated.
+If remove_also_table_sx_locks is TRUE then table-level S and X locks are
+also removed in addition to other table-level and record-level locks.
+No lock, that is going to be removed, is allowed to be a wait lock. */
+UNIV_INTERN
+void
+lock_remove_all_on_table(
+/*=====================*/
+	dict_table_t*	table,			/* in: table to be dropped
+						or truncated */
+	ibool		remove_also_table_sx_locks)/* in: also removes
+						table S and X locks */
+{
+	lock_t*	lock;
+	lock_t*	prev_lock;
+
+	mutex_enter(&kernel_mutex);
+
+	lock = UT_LIST_GET_FIRST(table->locks);
+
+	while (lock != NULL) {
+
+		prev_lock = UT_LIST_GET_PREV(un_member.tab_lock.locks,
+					     lock);
+
+		/* If we should remove all locks (remove_also_table_sx_locks
+		is TRUE), or if the lock is not table-level S or X lock,
+		then check we are not going to remove a wait lock. */
+		if (remove_also_table_sx_locks
+		    || !(lock_get_type(lock) == LOCK_TABLE
+			 && IS_LOCK_S_OR_X(lock))) {
+
+			ut_a(!lock_get_wait(lock));
+		}
+
+		lock_remove_all_on_table_for_trx(table, lock->trx,
+						 remove_also_table_sx_locks);
+
+		if (prev_lock == NULL) {
+			if (lock == UT_LIST_GET_FIRST(table->locks)) {
+				/* lock was not removed, pick its successor */
+				lock = UT_LIST_GET_NEXT(
+					un_member.tab_lock.locks, lock);
+			} else {
+				/* lock was removed, pick the first one */
+				lock = UT_LIST_GET_FIRST(table->locks);
+			}
+		} else if (UT_LIST_GET_NEXT(un_member.tab_lock.locks,
+					    prev_lock) != lock) {
+			/* If lock was removed by
+			lock_remove_all_on_table_for_trx() then pick the
+			successor of prev_lock ... */
+			lock = UT_LIST_GET_NEXT(
+				un_member.tab_lock.locks, prev_lock);
+		} else {
+			/* ... otherwise pick the successor of lock. */
+			lock = UT_LIST_GET_NEXT(
+				un_member.tab_lock.locks, lock);
+		}
+	}
+
+	mutex_exit(&kernel_mutex);
+}
+
+/*===================== VALIDATION AND DEBUGGING  ====================*/
+
+/*************************************************************************
+Prints info of a table lock. */
+UNIV_INTERN
+void
+lock_table_print(
+/*=============*/
+	FILE*		file,	/* in: file where to print */
+	const lock_t*	lock)	/* in: table type lock */
+{
+	ut_ad(mutex_own(&kernel_mutex));
+	ut_a(lock_get_type_low(lock) == LOCK_TABLE);
+
+	fputs("TABLE LOCK table ", file);
+	ut_print_name(file, lock->trx, TRUE,
+		      lock->un_member.tab_lock.table->name);
+	fprintf(file, " trx id " TRX_ID_FMT,
+		TRX_ID_PREP_PRINTF(lock->trx->id));
+
+	if (lock_get_mode(lock) == LOCK_S) {
+		fputs(" lock mode S", file);
+	} else if (lock_get_mode(lock) == LOCK_X) {
+		fputs(" lock mode X", file);
+	} else if (lock_get_mode(lock) == LOCK_IS) {
+		fputs(" lock mode IS", file);
+	} else if (lock_get_mode(lock) == LOCK_IX) {
+		fputs(" lock mode IX", file);
+	} else if (lock_get_mode(lock) == LOCK_AUTO_INC) {
+		fputs(" lock mode AUTO-INC", file);
+	} else {
+		fprintf(file, " unknown lock mode %lu",
+			(ulong) lock_get_mode(lock));
+	}
+
+	if (lock_get_wait(lock)) {
+		fputs(" waiting", file);
+	}
+
+	putc('\n', file);
+}
+
+/*************************************************************************
+Prints info of a record lock. */
+UNIV_INTERN
+void
+lock_rec_print(
+/*===========*/
+	FILE*		file,	/* in: file where to print */
+	const lock_t*	lock)	/* in: record type lock */
+{
+	const buf_block_t*	block;
+	ulint			space;
+	ulint			page_no;
+	ulint			i;
+	mtr_t			mtr;
+	mem_heap_t*		heap		= NULL;
+	ulint			offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint*			offsets		= offsets_;
+	rec_offs_init(offsets_);
+
+	ut_ad(mutex_own(&kernel_mutex));
+	ut_a(lock_get_type_low(lock) == LOCK_REC);
+
+	space = lock->un_member.rec_lock.space;
+	page_no = lock->un_member.rec_lock.page_no;
+
+	fprintf(file, "RECORD LOCKS space id %lu page no %lu n bits %lu ",
+		(ulong) space, (ulong) page_no,
+		(ulong) lock_rec_get_n_bits(lock));
+	dict_index_name_print(file, lock->trx, lock->index);
+	fprintf(file, " trx id " TRX_ID_FMT,
+		TRX_ID_PREP_PRINTF(lock->trx->id));
+
+	if (lock_get_mode(lock) == LOCK_S) {
+		fputs(" lock mode S", file);
+	} else if (lock_get_mode(lock) == LOCK_X) {
+		fputs(" lock_mode X", file);
+	} else {
+		ut_error;
+	}
+
+	if (lock_rec_get_gap(lock)) {
+		fputs(" locks gap before rec", file);
+	}
+
+	if (lock_rec_get_rec_not_gap(lock)) {
+		fputs(" locks rec but not gap", file);
+	}
+
+	if (lock_rec_get_insert_intention(lock)) {
+		fputs(" insert intention", file);
+	}
+
+	if (lock_get_wait(lock)) {
+		fputs(" waiting", file);
+	}
+
+	mtr_start(&mtr);
+
+	putc('\n', file);
+
+	if ( srv_show_verbose_locks ) {
+		block = buf_page_try_get(space, page_no, &mtr);
+		if (block) {
+			for (i = 0; i < lock_rec_get_n_bits(lock); i++) {
+
+				if (lock_rec_get_nth_bit(lock, i)) {
+
+					const rec_t*	rec
+						= page_find_rec_with_heap_no(
+								buf_block_get_frame(block), i);
+					offsets = rec_get_offsets(
+							rec, lock->index, offsets,
+							ULINT_UNDEFINED, &heap);
+
+					fprintf(file, "Record lock, heap no %lu ",
+							(ulong) i);
+					rec_print_new(file, rec, offsets);
+					putc('\n', file);
+				}
+			}
+		} else {
+			for (i = 0; i < lock_rec_get_n_bits(lock); i++) {
+				fprintf(file, "Record lock, heap no %lu\n", (ulong) i);
+			}
+		}
+	}
+	mtr_commit(&mtr);
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+}
+
+#ifndef UNIV_HOTBACKUP
+
+#ifdef UNIV_DEBUG
+/* Print the number of lock structs from lock_print_info_summary() only
+in non-production builds for performance reasons, see
+http://bugs.mysql.com/36942 */
+#define PRINT_NUM_OF_LOCK_STRUCTS
+#endif /* UNIV_DEBUG */
+
+#ifdef PRINT_NUM_OF_LOCK_STRUCTS
+/*************************************************************************
+Calculates the number of record lock structs in the record lock hash table. */
+static
+ulint
+lock_get_n_rec_locks(void)
+/*======================*/
+{
+	lock_t*	lock;
+	ulint	n_locks	= 0;
+	ulint	i;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	for (i = 0; i < hash_get_n_cells(lock_sys->rec_hash); i++) {
+
+		lock = HASH_GET_FIRST(lock_sys->rec_hash, i);
+
+		while (lock) {
+			n_locks++;
+
+			lock = HASH_GET_NEXT(hash, lock);
+		}
+	}
+
+	return(n_locks);
+}
+#endif /* PRINT_NUM_OF_LOCK_STRUCTS */
+
+/*************************************************************************
+Prints info of locks for all transactions. */
+UNIV_INTERN
+void
+lock_print_info_summary(
+/*====================*/
+	FILE*	file)	/* in: file where to print */
+{
+	/* We must protect the MySQL thd->query field with a MySQL mutex, and
+	because the MySQL mutex must be reserved before the kernel_mutex of
+	InnoDB, we call innobase_mysql_prepare_print_arbitrary_thd() here. */
+
+	innobase_mysql_prepare_print_arbitrary_thd();
+	lock_mutex_enter_kernel();
+
+	if (lock_deadlock_found) {
+		fputs("------------------------\n"
+		      "LATEST DETECTED DEADLOCK\n"
+		      "------------------------\n", file);
+
+		ut_copy_file(file, lock_latest_err_file);
+	}
+
+	fputs("------------\n"
+	      "TRANSACTIONS\n"
+	      "------------\n", file);
+
+	fprintf(file, "Trx id counter " TRX_ID_FMT "\n",
+		TRX_ID_PREP_PRINTF(trx_sys->max_trx_id));
+
+	fprintf(file,
+		"Purge done for trx's n:o < " TRX_ID_FMT
+		" undo n:o < " TRX_ID_FMT "\n",
+		TRX_ID_PREP_PRINTF(purge_sys->purge_trx_no),
+		TRX_ID_PREP_PRINTF(purge_sys->purge_undo_no));
+
+	fprintf(file,
+		"History list length %lu\n",
+		(ulong) trx_sys->rseg_history_len);
+
+#ifdef PRINT_NUM_OF_LOCK_STRUCTS
+	fprintf(file,
+		"Total number of lock structs in row lock hash table %lu\n",
+		(ulong) lock_get_n_rec_locks());
+#endif /* PRINT_NUM_OF_LOCK_STRUCTS */
+}
+
+/*************************************************************************
+Prints info of locks for each transaction. */
+UNIV_INTERN
+void
+lock_print_info_all_transactions(
+/*=============================*/
+	FILE*	file)	/* in: file where to print */
+{
+	lock_t*	lock;
+	ibool	load_page_first = TRUE;
+	ulint	nth_trx		= 0;
+	ulint	nth_lock	= 0;
+	ulint	i;
+	mtr_t	mtr;
+	trx_t*	trx;
+
+	fprintf(file, "LIST OF TRANSACTIONS FOR EACH SESSION:\n");
+
+	/* First print info on non-active transactions */
+
+	trx = UT_LIST_GET_FIRST(trx_sys->mysql_trx_list);
+
+	while (trx) {
+		if (trx->conc_state == TRX_NOT_STARTED) {
+			fputs("---", file);
+			trx_print(file, trx, 600);
+		}
+
+		trx = UT_LIST_GET_NEXT(mysql_trx_list, trx);
+	}
+
+loop:
+	trx = UT_LIST_GET_FIRST(trx_sys->trx_list);
+
+	i = 0;
+
+	/* Since we temporarily release the kernel mutex when
+	reading a database page in below, variable trx may be
+	obsolete now and we must loop through the trx list to
+	get probably the same trx, or some other trx. */
+
+	while (trx && (i < nth_trx)) {
+		trx = UT_LIST_GET_NEXT(trx_list, trx);
+		i++;
+	}
+
+	if (trx == NULL) {
+		lock_mutex_exit_kernel();
+		innobase_mysql_end_print_arbitrary_thd();
+
+		ut_ad(lock_validate());
+
+		return;
+	}
+
+	if (nth_lock == 0) {
+		fputs("---", file);
+		trx_print(file, trx, 600);
+
+		if (trx->read_view) {
+			fprintf(file,
+				"Trx read view will not see trx with"
+				" id >= " TRX_ID_FMT
+				", sees < " TRX_ID_FMT "\n",
+				TRX_ID_PREP_PRINTF(
+					trx->read_view->low_limit_id),
+				TRX_ID_PREP_PRINTF(
+					trx->read_view->up_limit_id));
+		}
+
+		if (trx->que_state == TRX_QUE_LOCK_WAIT) {
+			fprintf(file,
+				"------- TRX HAS BEEN WAITING %lu SEC"
+				" FOR THIS LOCK TO BE GRANTED:\n",
+				(ulong) difftime(time(NULL),
+						 trx->wait_started));
+
+			if (lock_get_type_low(trx->wait_lock) == LOCK_REC) {
+				lock_rec_print(file, trx->wait_lock);
+			} else {
+				lock_table_print(file, trx->wait_lock);
+			}
+
+			fputs("------------------\n", file);
+		}
+	}
+
+        if (!srv_print_innodb_lock_monitor && !srv_show_locks_held) {
+		nth_trx++;
+		goto loop;
+	}
+
+	i = 0;
+
+	/* Look at the note about the trx loop above why we loop here:
+	lock may be an obsolete pointer now. */
+
+	lock = UT_LIST_GET_FIRST(trx->trx_locks);
+
+	while (lock && (i < nth_lock)) {
+		lock = UT_LIST_GET_NEXT(trx_locks, lock);
+		i++;
+	}
+
+	if (lock == NULL) {
+		nth_trx++;
+		nth_lock = 0;
+
+		goto loop;
+	}
+
+	if (lock_get_type_low(lock) == LOCK_REC) {
+		if (load_page_first) {
+			ulint	space	= lock->un_member.rec_lock.space;
+			ulint	zip_size= fil_space_get_zip_size(space);
+			ulint	page_no = lock->un_member.rec_lock.page_no;
+
+			lock_mutex_exit_kernel();
+			innobase_mysql_end_print_arbitrary_thd();
+
+			mtr_start(&mtr);
+
+			buf_page_get_with_no_latch(space, zip_size,
+						   page_no, &mtr);
+
+			mtr_commit(&mtr);
+
+			load_page_first = FALSE;
+
+			innobase_mysql_prepare_print_arbitrary_thd();
+			lock_mutex_enter_kernel();
+
+			goto loop;
+		}
+
+		lock_rec_print(file, lock);
+	} else {
+		ut_ad(lock_get_type_low(lock) & LOCK_TABLE);
+
+		lock_table_print(file, lock);
+	}
+
+	load_page_first = TRUE;
+
+	nth_lock++;
+
+	if (nth_lock >= srv_show_locks_held) {
+		fputs("TOO LOCKS PRINTED FOR THIS TRX:"
+		      " SUPPRESSING FURTHER PRINTS\n",
+		      file);
+
+		nth_trx++;
+		nth_lock = 0;
+
+		goto loop;
+	}
+
+	goto loop;
+}
+
+# ifdef UNIV_DEBUG
+/*************************************************************************
+Validates the lock queue on a table. */
+static
+ibool
+lock_table_queue_validate(
+/*======================*/
+				/* out: TRUE if ok */
+	dict_table_t*	table)	/* in: table */
+{
+	lock_t*	lock;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	lock = UT_LIST_GET_FIRST(table->locks);
+
+	while (lock) {
+		ut_a(((lock->trx)->conc_state == TRX_ACTIVE)
+		     || ((lock->trx)->conc_state == TRX_PREPARED)
+		     || ((lock->trx)->conc_state == TRX_COMMITTED_IN_MEMORY));
+
+		if (!lock_get_wait(lock)) {
+
+			ut_a(!lock_table_other_has_incompatible(
+				     lock->trx, 0, table,
+				     lock_get_mode(lock)));
+		} else {
+
+			ut_a(lock_table_has_to_wait_in_queue(lock));
+		}
+
+		lock = UT_LIST_GET_NEXT(un_member.tab_lock.locks, lock);
+	}
+
+	return(TRUE);
+}
+
+/*************************************************************************
+Validates the lock queue on a single record. */
+static
+ibool
+lock_rec_queue_validate(
+/*====================*/
+					/* out: TRUE if ok */
+	const buf_block_t*	block,	/* in: buffer block containing rec */
+	const rec_t*		rec,	/* in: record to look at */
+	dict_index_t*		index,	/* in: index, or NULL if not known */
+	const ulint*		offsets)/* in: rec_get_offsets(rec, index) */
+{
+	trx_t*	impl_trx;
+	lock_t*	lock;
+	ulint	heap_no;
+
+	ut_a(rec);
+	ut_a(block->frame == page_align(rec));
+	ut_ad(rec_offs_validate(rec, index, offsets));
+	ut_ad(!page_rec_is_comp(rec) == !rec_offs_comp(offsets));
+
+	heap_no = page_rec_get_heap_no(rec);
+
+	lock_mutex_enter_kernel();
+
+	if (!page_rec_is_user_rec(rec)) {
+
+		lock = lock_rec_get_first(block, heap_no);
+
+		while (lock) {
+			switch(lock->trx->conc_state) {
+			case TRX_ACTIVE:
+			case TRX_PREPARED:
+			case TRX_COMMITTED_IN_MEMORY:
+				break;
+			default:
+				ut_error;
+			}
+
+			ut_a(trx_in_trx_list(lock->trx));
+
+			if (lock_get_wait(lock)) {
+				ut_a(lock_rec_has_to_wait_in_queue(lock));
+			}
+
+			if (index) {
+				ut_a(lock->index == index);
+			}
+
+			lock = lock_rec_get_next(heap_no, lock);
+		}
+
+		lock_mutex_exit_kernel();
+
+		return(TRUE);
+	}
+
+	if (!index);
+	else if (dict_index_is_clust(index)) {
+
+		impl_trx = lock_clust_rec_some_has_impl(rec, index, offsets);
+
+		if (impl_trx
+		    && lock_rec_other_has_expl_req(LOCK_S, 0, LOCK_WAIT,
+						   block, heap_no, impl_trx)) {
+
+			ut_a(lock_rec_has_expl(LOCK_X | LOCK_REC_NOT_GAP,
+					       block, heap_no, impl_trx));
+		}
+	} else {
+
+		/* The kernel mutex may get released temporarily in the
+		next function call: we have to release lock table mutex
+		to obey the latching order */
+
+		impl_trx = lock_sec_rec_some_has_impl_off_kernel(
+			rec, index, offsets);
+
+		if (impl_trx
+		    && lock_rec_other_has_expl_req(LOCK_S, 0, LOCK_WAIT,
+						   block, heap_no, impl_trx)) {
+
+			ut_a(lock_rec_has_expl(LOCK_X | LOCK_REC_NOT_GAP,
+					       block, heap_no, impl_trx));
+		}
+	}
+
+	lock = lock_rec_get_first(block, heap_no);
+
+	while (lock) {
+		ut_a(lock->trx->conc_state == TRX_ACTIVE
+		     || lock->trx->conc_state == TRX_PREPARED
+		     || lock->trx->conc_state == TRX_COMMITTED_IN_MEMORY);
+		ut_a(trx_in_trx_list(lock->trx));
+
+		if (index) {
+			ut_a(lock->index == index);
+		}
+
+		if (!lock_rec_get_gap(lock) && !lock_get_wait(lock)) {
+
+			enum lock_mode	mode;
+
+			if (lock_get_mode(lock) == LOCK_S) {
+				mode = LOCK_X;
+			} else {
+				mode = LOCK_S;
+			}
+			ut_a(!lock_rec_other_has_expl_req(
+				     mode, 0, 0, block, heap_no, lock->trx));
+
+		} else if (lock_get_wait(lock) && !lock_rec_get_gap(lock)) {
+
+			ut_a(lock_rec_has_to_wait_in_queue(lock));
+		}
+
+		lock = lock_rec_get_next(heap_no, lock);
+	}
+
+	lock_mutex_exit_kernel();
+
+	return(TRUE);
+}
+
+/*************************************************************************
+Validates the record lock queues on a page. */
+static
+ibool
+lock_rec_validate_page(
+/*===================*/
+			/* out: TRUE if ok */
+	ulint	space,	/* in: space id */
+	ulint	page_no)/* in: page number */
+{
+	dict_index_t*	index;
+	buf_block_t*	block;
+	const page_t*	page;
+	lock_t*		lock;
+	const rec_t*	rec;
+	ulint		nth_lock	= 0;
+	ulint		nth_bit		= 0;
+	ulint		i;
+	mtr_t		mtr;
+	mem_heap_t*	heap		= NULL;
+	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint*		offsets		= offsets_;
+	rec_offs_init(offsets_);
+
+	ut_ad(!mutex_own(&kernel_mutex));
+
+	mtr_start(&mtr);
+
+	block = buf_page_get(space, fil_space_get_zip_size(space),
+			     page_no, RW_X_LATCH, &mtr);
+	buf_block_dbg_add_level(block, SYNC_NO_ORDER_CHECK);
+
+	page = block->frame;
+
+	lock_mutex_enter_kernel();
+loop:
+	lock = lock_rec_get_first_on_page_addr(space, page_no);
+
+	if (!lock) {
+		goto function_exit;
+	}
+
+	for (i = 0; i < nth_lock; i++) {
+
+		lock = lock_rec_get_next_on_page(lock);
+
+		if (!lock) {
+			goto function_exit;
+		}
+	}
+
+	ut_a(trx_in_trx_list(lock->trx));
+	ut_a(lock->trx->conc_state == TRX_ACTIVE
+	     || lock->trx->conc_state == TRX_PREPARED
+	     || lock->trx->conc_state == TRX_COMMITTED_IN_MEMORY);
+
+	for (i = nth_bit; i < lock_rec_get_n_bits(lock); i++) {
+
+		if (i == 1 || lock_rec_get_nth_bit(lock, i)) {
+
+			index = lock->index;
+			rec = page_find_rec_with_heap_no(page, i);
+			ut_a(rec);
+			offsets = rec_get_offsets(rec, index, offsets,
+						  ULINT_UNDEFINED, &heap);
+
+			fprintf(stderr,
+				"Validating %lu %lu\n",
+				(ulong) space, (ulong) page_no);
+
+			lock_mutex_exit_kernel();
+
+			lock_rec_queue_validate(block, rec, index, offsets);
+
+			lock_mutex_enter_kernel();
+
+			nth_bit = i + 1;
+
+			goto loop;
+		}
+	}
+
+	nth_bit = 0;
+	nth_lock++;
+
+	goto loop;
+
+function_exit:
+	lock_mutex_exit_kernel();
+
+	mtr_commit(&mtr);
+
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+	return(TRUE);
+}
+
+/*************************************************************************
+Validates the lock system. */
+static
+ibool
+lock_validate(void)
+/*===============*/
+			/* out: TRUE if ok */
+{
+	lock_t*	lock;
+	trx_t*	trx;
+	dulint	limit;
+	ulint	space;
+	ulint	page_no;
+	ulint	i;
+
+	lock_mutex_enter_kernel();
+
+	trx = UT_LIST_GET_FIRST(trx_sys->trx_list);
+
+	while (trx) {
+		lock = UT_LIST_GET_FIRST(trx->trx_locks);
+
+		while (lock) {
+			if (lock_get_type_low(lock) & LOCK_TABLE) {
+
+				lock_table_queue_validate(
+					lock->un_member.tab_lock.table);
+			}
+
+			lock = UT_LIST_GET_NEXT(trx_locks, lock);
+		}
+
+		trx = UT_LIST_GET_NEXT(trx_list, trx);
+	}
+
+	for (i = 0; i < hash_get_n_cells(lock_sys->rec_hash); i++) {
+
+		limit = ut_dulint_zero;
+
+		for (;;) {
+			lock = HASH_GET_FIRST(lock_sys->rec_hash, i);
+
+			while (lock) {
+				ut_a(trx_in_trx_list(lock->trx));
+
+				space = lock->un_member.rec_lock.space;
+				page_no = lock->un_member.rec_lock.page_no;
+
+				if (ut_dulint_cmp(
+					    ut_dulint_create(space, page_no),
+					    limit) >= 0) {
+					break;
+				}
+
+				lock = HASH_GET_NEXT(hash, lock);
+			}
+
+			if (!lock) {
+
+				break;
+			}
+
+			lock_mutex_exit_kernel();
+
+			lock_rec_validate_page(space, page_no);
+
+			lock_mutex_enter_kernel();
+
+			limit = ut_dulint_create(space, page_no + 1);
+		}
+	}
+
+	lock_mutex_exit_kernel();
+
+	return(TRUE);
+}
+# endif /* UNIV_DEBUG */
+#endif /* !UNIV_HOTBACKUP */
+/*============ RECORD LOCK CHECKS FOR ROW OPERATIONS ====================*/
+
+/*************************************************************************
+Checks if locks of other transactions prevent an immediate insert of
+a record. If they do, first tests if the query thread should anyway
+be suspended for some reason; if not, then puts the transaction and
+the query thread to the lock wait state and inserts a waiting request
+for a gap x-lock to the lock queue. */
+UNIV_INTERN
+ulint
+lock_rec_insert_check_and_lock(
+/*===========================*/
+				/* out: DB_SUCCESS, DB_LOCK_WAIT,
+				DB_DEADLOCK, or DB_QUE_THR_SUSPENDED */
+	ulint		flags,	/* in: if BTR_NO_LOCKING_FLAG bit is
+				set, does nothing */
+	rec_t*		rec,	/* in: record after which to insert */
+	buf_block_t*	block,	/* in/out: buffer block of rec */
+	dict_index_t*	index,	/* in: index */
+	que_thr_t*	thr,	/* in: query thread */
+	ibool*		inherit)/* out: set to TRUE if the new
+				inserted record maybe should inherit
+				LOCK_GAP type locks from the successor
+				record */
+{
+	const rec_t*	next_rec;
+	trx_t*		trx;
+	lock_t*		lock;
+	ulint		err;
+	ulint		next_rec_heap_no;
+
+	ut_ad(block->frame == page_align(rec));
+
+	if (flags & BTR_NO_LOCKING_FLAG) {
+
+		return(DB_SUCCESS);
+	}
+
+	trx = thr_get_trx(thr);
+	next_rec = page_rec_get_next(rec);
+	next_rec_heap_no = page_rec_get_heap_no(next_rec);
+
+	lock_mutex_enter_kernel();
+
+	/* When inserting a record into an index, the table must be at
+	least IX-locked or we must be building an index, in which case
+	the table must be at least S-locked. */
+	ut_ad(lock_table_has(trx, index->table, LOCK_IX)
+	      || (*index->name == TEMP_INDEX_PREFIX
+		  && lock_table_has(trx, index->table, LOCK_S)));
+
+	lock = lock_rec_get_first(block, next_rec_heap_no);
+
+	if (UNIV_LIKELY(lock == NULL)) {
+		/* We optimize CPU time usage in the simplest case */
+
+		lock_mutex_exit_kernel();
+
+		if (!dict_index_is_clust(index)) {
+			/* Update the page max trx id field */
+			page_update_max_trx_id(block,
+					       buf_block_get_page_zip(block),
+					       trx->id);
+		}
+
+		*inherit = FALSE;
+
+		return(DB_SUCCESS);
+	}
+
+	*inherit = TRUE;
+
+	/* If another transaction has an explicit lock request which locks
+	the gap, waiting or granted, on the successor, the insert has to wait.
+
+	An exception is the case where the lock by the another transaction
+	is a gap type lock which it placed to wait for its turn to insert. We
+	do not consider that kind of a lock conflicting with our insert. This
+	eliminates an unnecessary deadlock which resulted when 2 transactions
+	had to wait for their insert. Both had waiting gap type lock requests
+	on the successor, which produced an unnecessary deadlock. */
+
+	if (lock_rec_other_has_conflicting(
+		    LOCK_X | LOCK_GAP | LOCK_INSERT_INTENTION,
+		    block, next_rec_heap_no, trx)) {
+
+		/* Note that we may get DB_SUCCESS also here! */
+		err = lock_rec_enqueue_waiting(LOCK_X | LOCK_GAP
+					       | LOCK_INSERT_INTENTION,
+					       block, next_rec_heap_no,
+					       index, thr);
+	} else {
+		err = DB_SUCCESS;
+	}
+
+	lock_mutex_exit_kernel();
+
+	if ((err == DB_SUCCESS) && !dict_index_is_clust(index)) {
+		/* Update the page max trx id field */
+		page_update_max_trx_id(block,
+				       buf_block_get_page_zip(block),
+				       trx->id);
+	}
+
+#ifdef UNIV_DEBUG
+	{
+		mem_heap_t*	heap		= NULL;
+		ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+		const ulint*	offsets;
+		rec_offs_init(offsets_);
+
+		offsets = rec_get_offsets(next_rec, index, offsets_,
+					  ULINT_UNDEFINED, &heap);
+		ut_ad(lock_rec_queue_validate(block,
+					      next_rec, index, offsets));
+		if (UNIV_LIKELY_NULL(heap)) {
+			mem_heap_free(heap);
+		}
+	}
+#endif /* UNIV_DEBUG */
+
+	return(err);
+}
+
+/*************************************************************************
+If a transaction has an implicit x-lock on a record, but no explicit x-lock
+set on the record, sets one for it. NOTE that in the case of a secondary
+index, the kernel mutex may get temporarily released. */
+static
+void
+lock_rec_convert_impl_to_expl(
+/*==========================*/
+	const buf_block_t*	block,	/* in: buffer block of rec */
+	const rec_t*		rec,	/* in: user record on page */
+	dict_index_t*		index,	/* in: index of record */
+	const ulint*		offsets)/* in: rec_get_offsets(rec, index) */
+{
+	trx_t*	impl_trx;
+
+	ut_ad(mutex_own(&kernel_mutex));
+	ut_ad(page_rec_is_user_rec(rec));
+	ut_ad(rec_offs_validate(rec, index, offsets));
+	ut_ad(!page_rec_is_comp(rec) == !rec_offs_comp(offsets));
+
+	if (dict_index_is_clust(index)) {
+		impl_trx = lock_clust_rec_some_has_impl(rec, index, offsets);
+	} else {
+		impl_trx = lock_sec_rec_some_has_impl_off_kernel(
+			rec, index, offsets);
+	}
+
+	if (impl_trx) {
+		ulint	heap_no = page_rec_get_heap_no(rec);
+
+		/* If the transaction has no explicit x-lock set on the
+		record, set one for it */
+
+		if (!lock_rec_has_expl(LOCK_X | LOCK_REC_NOT_GAP, block,
+				       heap_no, impl_trx)) {
+
+			lock_rec_add_to_queue(
+				LOCK_REC | LOCK_X | LOCK_REC_NOT_GAP,
+				block, heap_no, index, impl_trx);
+		}
+	}
+}
+
+/*************************************************************************
+Checks if locks of other transactions prevent an immediate modify (update,
+delete mark, or delete unmark) of a clustered index record. If they do,
+first tests if the query thread should anyway be suspended for some
+reason; if not, then puts the transaction and the query thread to the
+lock wait state and inserts a waiting request for a record x-lock to the
+lock queue. */
+UNIV_INTERN
+ulint
+lock_clust_rec_modify_check_and_lock(
+/*=================================*/
+					/* out: DB_SUCCESS,
+					DB_LOCK_WAIT, DB_DEADLOCK, or
+					DB_QUE_THR_SUSPENDED */
+	ulint			flags,	/* in: if BTR_NO_LOCKING_FLAG
+					bit is set, does nothing */
+	const buf_block_t*	block,	/* in: buffer block of rec */
+	const rec_t*		rec,	/* in: record which should be
+					modified */
+	dict_index_t*		index,	/* in: clustered index */
+	const ulint*		offsets,/* in: rec_get_offsets(rec, index) */
+	que_thr_t*		thr)	/* in: query thread */
+{
+	ulint	err;
+	ulint	heap_no;
+
+	ut_ad(rec_offs_validate(rec, index, offsets));
+	ut_ad(dict_index_is_clust(index));
+	ut_ad(block->frame == page_align(rec));
+
+	if (flags & BTR_NO_LOCKING_FLAG) {
+
+		return(DB_SUCCESS);
+	}
+
+	heap_no = rec_offs_comp(offsets)
+		? rec_get_heap_no_new(rec)
+		: rec_get_heap_no_old(rec);
+
+	lock_mutex_enter_kernel();
+
+	ut_ad(lock_table_has(thr_get_trx(thr), index->table, LOCK_IX));
+
+	/* If a transaction has no explicit x-lock set on the record, set one
+	for it */
+
+	lock_rec_convert_impl_to_expl(block, rec, index, offsets);
+
+	err = lock_rec_lock(TRUE, LOCK_X | LOCK_REC_NOT_GAP,
+			    block, heap_no, index, thr);
+
+	lock_mutex_exit_kernel();
+
+	ut_ad(lock_rec_queue_validate(block, rec, index, offsets));
+
+	return(err);
+}
+
+/*************************************************************************
+Checks if locks of other transactions prevent an immediate modify (delete
+mark or delete unmark) of a secondary index record. */
+UNIV_INTERN
+ulint
+lock_sec_rec_modify_check_and_lock(
+/*===============================*/
+				/* out: DB_SUCCESS, DB_LOCK_WAIT,
+				DB_DEADLOCK, or DB_QUE_THR_SUSPENDED */
+	ulint		flags,	/* in: if BTR_NO_LOCKING_FLAG
+				bit is set, does nothing */
+	buf_block_t*	block,	/* in/out: buffer block of rec */
+	rec_t*		rec,	/* in: record which should be
+				modified; NOTE: as this is a secondary
+				index, we always have to modify the
+				clustered index record first: see the
+				comment below */
+	dict_index_t*	index,	/* in: secondary index */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	ulint	err;
+	ulint	heap_no;
+
+	ut_ad(!dict_index_is_clust(index));
+	ut_ad(block->frame == page_align(rec));
+
+	if (flags & BTR_NO_LOCKING_FLAG) {
+
+		return(DB_SUCCESS);
+	}
+
+	heap_no = page_rec_get_heap_no(rec);
+
+	/* Another transaction cannot have an implicit lock on the record,
+	because when we come here, we already have modified the clustered
+	index record, and this would not have been possible if another active
+	transaction had modified this secondary index record. */
+
+	lock_mutex_enter_kernel();
+
+	ut_ad(lock_table_has(thr_get_trx(thr), index->table, LOCK_IX));
+
+	err = lock_rec_lock(TRUE, LOCK_X | LOCK_REC_NOT_GAP,
+			    block, heap_no, index, thr);
+
+	lock_mutex_exit_kernel();
+
+#ifdef UNIV_DEBUG
+	{
+		mem_heap_t*	heap		= NULL;
+		ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+		const ulint*	offsets;
+		rec_offs_init(offsets_);
+
+		offsets = rec_get_offsets(rec, index, offsets_,
+					  ULINT_UNDEFINED, &heap);
+		ut_ad(lock_rec_queue_validate(block, rec, index, offsets));
+		if (UNIV_LIKELY_NULL(heap)) {
+			mem_heap_free(heap);
+		}
+	}
+#endif /* UNIV_DEBUG */
+
+	if (err == DB_SUCCESS) {
+		/* Update the page max trx id field */
+		page_update_max_trx_id(block,
+				       buf_block_get_page_zip(block),
+				       thr_get_trx(thr)->id);
+	}
+
+	return(err);
+}
+
+/*************************************************************************
+Like the counterpart for a clustered index below, but now we read a
+secondary index record. */
+UNIV_INTERN
+ulint
+lock_sec_rec_read_check_and_lock(
+/*=============================*/
+					/* out: DB_SUCCESS,
+					DB_LOCK_WAIT, DB_DEADLOCK, or
+					DB_QUE_THR_SUSPENDED */
+	ulint			flags,	/* in: if BTR_NO_LOCKING_FLAG
+					bit is set, does nothing */
+	const buf_block_t*	block,	/* in: buffer block of rec */
+	const rec_t*		rec,	/* in: user record or page
+					supremum record which should
+					be read or passed over by a
+					read cursor */
+	dict_index_t*		index,	/* in: secondary index */
+	const ulint*		offsets,/* in: rec_get_offsets(rec, index) */
+	enum lock_mode		mode,	/* in: mode of the lock which
+					the read cursor should set on
+					records: LOCK_S or LOCK_X; the
+					latter is possible in
+					SELECT FOR UPDATE */
+	ulint			gap_mode,/* in: LOCK_ORDINARY, LOCK_GAP, or
+					LOCK_REC_NOT_GAP */
+	que_thr_t*		thr)	/* in: query thread */
+{
+	ulint	err;
+	ulint	heap_no;
+
+	ut_ad(!dict_index_is_clust(index));
+	ut_ad(block->frame == page_align(rec));
+	ut_ad(page_rec_is_user_rec(rec) || page_rec_is_supremum(rec));
+	ut_ad(rec_offs_validate(rec, index, offsets));
+	ut_ad(mode == LOCK_X || mode == LOCK_S);
+
+	if (flags & BTR_NO_LOCKING_FLAG) {
+
+		return(DB_SUCCESS);
+	}
+
+	heap_no = page_rec_get_heap_no(rec);
+
+	lock_mutex_enter_kernel();
+
+	ut_ad(mode != LOCK_X
+	      || lock_table_has(thr_get_trx(thr), index->table, LOCK_IX));
+	ut_ad(mode != LOCK_S
+	      || lock_table_has(thr_get_trx(thr), index->table, LOCK_IS));
+
+	/* Some transaction may have an implicit x-lock on the record only
+	if the max trx id for the page >= min trx id for the trx list or a
+	database recovery is running. */
+
+	if (((ut_dulint_cmp(page_get_max_trx_id(block->frame),
+			    trx_list_get_min_trx_id()) >= 0)
+	     || recv_recovery_is_on())
+	    && !page_rec_is_supremum(rec)) {
+
+		lock_rec_convert_impl_to_expl(block, rec, index, offsets);
+	}
+
+	err = lock_rec_lock(FALSE, mode | gap_mode,
+			    block, heap_no, index, thr);
+
+	lock_mutex_exit_kernel();
+
+	ut_ad(lock_rec_queue_validate(block, rec, index, offsets));
+
+	return(err);
+}
+
+/*************************************************************************
+Checks if locks of other transactions prevent an immediate read, or passing
+over by a read cursor, of a clustered index record. If they do, first tests
+if the query thread should anyway be suspended for some reason; if not, then
+puts the transaction and the query thread to the lock wait state and inserts a
+waiting request for a record lock to the lock queue. Sets the requested mode
+lock on the record. */
+UNIV_INTERN
+ulint
+lock_clust_rec_read_check_and_lock(
+/*===============================*/
+					/* out: DB_SUCCESS,
+					DB_LOCK_WAIT, DB_DEADLOCK, or
+					DB_QUE_THR_SUSPENDED */
+	ulint			flags,	/* in: if BTR_NO_LOCKING_FLAG
+					bit is set, does nothing */
+	const buf_block_t*	block,	/* in: buffer block of rec */
+	const rec_t*		rec,	/* in: user record or page
+					supremum record which should
+					be read or passed over by a
+					read cursor */
+	dict_index_t*		index,	/* in: clustered index */
+	const ulint*		offsets,/* in: rec_get_offsets(rec, index) */
+	enum lock_mode		mode,	/* in: mode of the lock which
+					the read cursor should set on
+					records: LOCK_S or LOCK_X; the
+					latter is possible in
+					SELECT FOR UPDATE */
+	ulint			gap_mode,/* in: LOCK_ORDINARY, LOCK_GAP, or
+					LOCK_REC_NOT_GAP */
+	que_thr_t*		thr)	/* in: query thread */
+{
+	ulint	err;
+	ulint	heap_no;
+
+	ut_ad(dict_index_is_clust(index));
+	ut_ad(block->frame == page_align(rec));
+	ut_ad(page_rec_is_user_rec(rec) || page_rec_is_supremum(rec));
+	ut_ad(gap_mode == LOCK_ORDINARY || gap_mode == LOCK_GAP
+	      || gap_mode == LOCK_REC_NOT_GAP);
+	ut_ad(rec_offs_validate(rec, index, offsets));
+
+	if (flags & BTR_NO_LOCKING_FLAG) {
+
+		return(DB_SUCCESS);
+	}
+
+	heap_no = page_rec_get_heap_no(rec);
+
+	lock_mutex_enter_kernel();
+
+	ut_ad(mode != LOCK_X
+	      || lock_table_has(thr_get_trx(thr), index->table, LOCK_IX));
+	ut_ad(mode != LOCK_S
+	      || lock_table_has(thr_get_trx(thr), index->table, LOCK_IS));
+
+	if (UNIV_LIKELY(heap_no != PAGE_HEAP_NO_SUPREMUM)) {
+
+		lock_rec_convert_impl_to_expl(block, rec, index, offsets);
+	}
+
+	err = lock_rec_lock(FALSE, mode | gap_mode,
+			    block, heap_no, index, thr);
+
+	lock_mutex_exit_kernel();
+
+	ut_ad(lock_rec_queue_validate(block, rec, index, offsets));
+
+	return(err);
+}
+/*************************************************************************
+Checks if locks of other transactions prevent an immediate read, or passing
+over by a read cursor, of a clustered index record. If they do, first tests
+if the query thread should anyway be suspended for some reason; if not, then
+puts the transaction and the query thread to the lock wait state and inserts a
+waiting request for a record lock to the lock queue. Sets the requested mode
+lock on the record. This is an alternative version of
+lock_clust_rec_read_check_and_lock() that does not require the parameter
+"offsets". */
+UNIV_INTERN
+ulint
+lock_clust_rec_read_check_and_lock_alt(
+/*===================================*/
+					/* out: DB_SUCCESS,
+					DB_LOCK_WAIT, DB_DEADLOCK, or
+					DB_QUE_THR_SUSPENDED */
+	ulint			flags,	/* in: if BTR_NO_LOCKING_FLAG
+					bit is set, does nothing */
+	const buf_block_t*	block,	/* in: buffer block of rec */
+	const rec_t*		rec,	/* in: user record or page
+					supremum record which should
+					be read or passed over by a
+					read cursor */
+	dict_index_t*		index,	/* in: clustered index */
+	enum lock_mode		mode,	/* in: mode of the lock which
+					the read cursor should set on
+					records: LOCK_S or LOCK_X; the
+					latter is possible in
+					SELECT FOR UPDATE */
+	ulint			gap_mode,/* in: LOCK_ORDINARY, LOCK_GAP, or
+					LOCK_REC_NOT_GAP */
+	que_thr_t*		thr)	/* in: query thread */
+{
+	mem_heap_t*	tmp_heap	= NULL;
+	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint*		offsets		= offsets_;
+	ulint		ret;
+	rec_offs_init(offsets_);
+
+	offsets = rec_get_offsets(rec, index, offsets,
+				  ULINT_UNDEFINED, &tmp_heap);
+	ret = lock_clust_rec_read_check_and_lock(flags, block, rec, index,
+						 offsets, mode, gap_mode, thr);
+	if (tmp_heap) {
+		mem_heap_free(tmp_heap);
+	}
+	return(ret);
+}
+
+/***********************************************************************
+Release the last lock from the transaction's autoinc locks. */
+UNIV_INLINE
+void
+lock_release_autoinc_last_lock(
+/*===========================*/
+	ib_vector_t*	autoinc_locks)	/* in/out: vector of AUTOINC locks */
+{
+	ulint		last;
+	lock_t*		lock;
+
+	ut_ad(mutex_own(&kernel_mutex));
+	ut_a(!ib_vector_is_empty(autoinc_locks));
+
+	/* The lock to be release must be the last lock acquired. */
+	last = ib_vector_size(autoinc_locks) - 1;
+	lock = ib_vector_get(autoinc_locks, last);
+
+	/* Should have only AUTOINC locks in the vector. */
+	ut_a(lock_get_mode(lock) == LOCK_AUTO_INC);
+	ut_a(lock_get_type(lock) == LOCK_TABLE);
+
+	ut_a(lock->un_member.tab_lock.table != NULL);
+
+	/* This will remove the lock from the trx autoinc_locks too. */
+	lock_table_dequeue(lock);
+}
+
+/***********************************************************************
+Release all the transaction's autoinc locks. */
+UNIV_INTERN
+void
+lock_release_autoinc_locks(
+/*=======================*/
+	trx_t*		trx)		/* in/out: transaction */
+{
+	ut_ad(mutex_own(&kernel_mutex));
+
+	ut_a(trx->autoinc_locks != NULL);
+
+	/* We release the locks in the reverse order. This is to
+	avoid searching the vector for the element to delete at
+	the lower level. See (lock_table_remove_low()) for details. */
+	while (!ib_vector_is_empty(trx->autoinc_locks)) {
+
+		/* lock_table_remove_low() will also remove the lock from
+		the transaction's autoinc_locks vector. */
+		lock_release_autoinc_last_lock(trx->autoinc_locks);
+	}
+
+	/* Should release all locks. */
+	ut_a(ib_vector_is_empty(trx->autoinc_locks));
+}
+
+/***********************************************************************
+Gets the type of a lock. Non-inline version for using outside of the
+lock module. */
+UNIV_INTERN
+ulint
+lock_get_type(
+/*==========*/
+				/* out: LOCK_TABLE or LOCK_REC */
+	const lock_t*	lock)	/* in: lock */
+{
+	return(lock_get_type_low(lock));
+}
+
+/***********************************************************************
+Gets the id of the transaction owning a lock. */
+UNIV_INTERN
+ullint
+lock_get_trx_id(
+/*============*/
+				/* out: transaction id */
+	const lock_t*	lock)	/* in: lock */
+{
+	return(trx_get_id(lock->trx));
+}
+
+/***********************************************************************
+Gets the mode of a lock in a human readable string.
+The string should not be free()'d or modified. */
+UNIV_INTERN
+const char*
+lock_get_mode_str(
+/*==============*/
+				/* out: lock mode */
+	const lock_t*	lock)	/* in: lock */
+{
+	ibool	is_gap_lock;
+
+	is_gap_lock = lock_get_type_low(lock) == LOCK_REC
+		&& lock_rec_get_gap(lock);
+
+	switch (lock_get_mode(lock)) {
+	case LOCK_S:
+		if (is_gap_lock) {
+			return("S,GAP");
+		} else {
+			return("S");
+		}
+	case LOCK_X:
+		if (is_gap_lock) {
+			return("X,GAP");
+		} else {
+			return("X");
+		}
+	case LOCK_IS:
+		if (is_gap_lock) {
+			return("IS,GAP");
+		} else {
+			return("IS");
+		}
+	case LOCK_IX:
+		if (is_gap_lock) {
+			return("IX,GAP");
+		} else {
+			return("IX");
+		}
+	case LOCK_AUTO_INC:
+		return("AUTO_INC");
+	default:
+		return("UNKNOWN");
+	}
+}
+
+/***********************************************************************
+Gets the type of a lock in a human readable string.
+The string should not be free()'d or modified. */
+UNIV_INTERN
+const char*
+lock_get_type_str(
+/*==============*/
+				/* out: lock type */
+	const lock_t*	lock)	/* in: lock */
+{
+	switch (lock_get_type_low(lock)) {
+	case LOCK_REC:
+		return("RECORD");
+	case LOCK_TABLE:
+		return("TABLE");
+	default:
+		return("UNKNOWN");
+	}
+}
+
+/***********************************************************************
+Gets the table on which the lock is. */
+UNIV_INLINE
+dict_table_t*
+lock_get_table(
+/*===========*/
+				/* out: table */
+	const lock_t*	lock)	/* in: lock */
+{
+	switch (lock_get_type_low(lock)) {
+	case LOCK_REC:
+		return(lock->index->table);
+	case LOCK_TABLE:
+		return(lock->un_member.tab_lock.table);
+	default:
+		ut_error;
+		return(NULL);
+	}
+}
+
+/***********************************************************************
+Gets the id of the table on which the lock is. */
+UNIV_INTERN
+ullint
+lock_get_table_id(
+/*==============*/
+				/* out: id of the table */
+	const lock_t*	lock)	/* in: lock */
+{
+	dict_table_t*	table;
+
+	table = lock_get_table(lock);
+
+	return((ullint)ut_conv_dulint_to_longlong(table->id));
+}
+
+/***********************************************************************
+Gets the name of the table on which the lock is.
+The string should not be free()'d or modified. */
+UNIV_INTERN
+const char*
+lock_get_table_name(
+/*================*/
+				/* out: name of the table */
+	const lock_t*	lock)	/* in: lock */
+{
+	dict_table_t*	table;
+
+	table = lock_get_table(lock);
+
+	return(table->name);
+}
+
+/***********************************************************************
+For a record lock, gets the index on which the lock is. */
+UNIV_INTERN
+const dict_index_t*
+lock_rec_get_index(
+/*===============*/
+				/* out: index */
+	const lock_t*	lock)	/* in: lock */
+{
+	ut_a(lock_get_type_low(lock) == LOCK_REC);
+
+	return(lock->index);
+}
+
+/***********************************************************************
+For a record lock, gets the name of the index on which the lock is.
+The string should not be free()'d or modified. */
+UNIV_INTERN
+const char*
+lock_rec_get_index_name(
+/*====================*/
+				/* out: name of the index */
+	const lock_t*	lock)	/* in: lock */
+{
+	ut_a(lock_get_type_low(lock) == LOCK_REC);
+
+	return(lock->index->name);
+}
+
+/***********************************************************************
+For a record lock, gets the tablespace number on which the lock is. */
+UNIV_INTERN
+ulint
+lock_rec_get_space_id(
+/*==================*/
+				/* out: tablespace number */
+	const lock_t*	lock)	/* in: lock */
+{
+	ut_a(lock_get_type_low(lock) == LOCK_REC);
+
+	return(lock->un_member.rec_lock.space);
+}
+
+/***********************************************************************
+For a record lock, gets the page number on which the lock is. */
+UNIV_INTERN
+ulint
+lock_rec_get_page_no(
+/*=================*/
+				/* out: page number */
+	const lock_t*	lock)	/* in: lock */
+{
+	ut_a(lock_get_type_low(lock) == LOCK_REC);
+
+	return(lock->un_member.rec_lock.page_no);
+}
diff --git a/storage/xtradb/log/log0log.c b/storage/xtradb/log/log0log.c
new file mode 100644
index 00000000000..d2ea0507705
--- /dev/null
+++ b/storage/xtradb/log/log0log.c
@@ -0,0 +1,3316 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Database log
+
+Created 12/9/1995 Heikki Tuuri
+*******************************************************/
+
+#include "log0log.h"
+
+#ifdef UNIV_NONINL
+#include "log0log.ic"
+#endif
+
+#include "mem0mem.h"
+#include "buf0buf.h"
+#include "buf0flu.h"
+#include "srv0srv.h"
+#include "log0recv.h"
+#include "fil0fil.h"
+#include "dict0boot.h"
+#include "srv0srv.h"
+#include "srv0start.h"
+#include "trx0sys.h"
+#include "trx0trx.h"
+
+/*
+General philosophy of InnoDB redo-logs:
+
+1) Every change to a contents of a data page must be done
+through mtr, which in mtr_commit() writes log records
+to the InnoDB redo log.
+
+2) Normally these changes are performed using a mlog_write_ulint()
+or similar function.
+
+3) In some page level operations only a code number of a
+c-function and its parameters are written to the log to
+reduce the size of the log.
+
+  3a) You should not add parameters to these kind of functions
+  (e.g. trx_undo_header_create(), trx_undo_insert_header_reuse())
+
+  3b) You should not add such functionality which either change
+  working when compared with the old or are dependent on data
+  outside of the page. These kind of functions should implement
+  self-contained page transformation and it should be unchanged
+  if you don't have very essential reasons to change log
+  semantics or format.
+
+*/
+
+/* Current free limit of space 0; protected by the log sys mutex; 0 means
+uninitialized */
+UNIV_INTERN ulint	log_fsp_current_free_limit		= 0;
+
+/* Global log system variable */
+UNIV_INTERN log_t*	log_sys	= NULL;
+
+#ifdef UNIV_DEBUG
+UNIV_INTERN ibool	log_do_write = TRUE;
+
+UNIV_INTERN ibool	log_debug_writes = FALSE;
+#endif /* UNIV_DEBUG */
+
+/* These control how often we print warnings if the last checkpoint is too
+old */
+UNIV_INTERN ibool	log_has_printed_chkp_warning = FALSE;
+UNIV_INTERN time_t	log_last_warning_time;
+
+#ifdef UNIV_LOG_ARCHIVE
+/* Pointer to this variable is used as the i/o-message when we do i/o to an
+archive */
+UNIV_INTERN byte	log_archive_io;
+#endif /* UNIV_LOG_ARCHIVE */
+
+/* A margin for free space in the log buffer before a log entry is catenated */
+#define LOG_BUF_WRITE_MARGIN	(4 * OS_FILE_LOG_BLOCK_SIZE)
+
+/* Margins for free space in the log buffer after a log entry is catenated */
+#define LOG_BUF_FLUSH_RATIO	2
+#define LOG_BUF_FLUSH_MARGIN	(LOG_BUF_WRITE_MARGIN + 4 * UNIV_PAGE_SIZE)
+
+/* Margin for the free space in the smallest log group, before a new query
+step which modifies the database, is started */
+
+#define LOG_CHECKPOINT_FREE_PER_THREAD	(4 * UNIV_PAGE_SIZE)
+#define LOG_CHECKPOINT_EXTRA_FREE	(8 * UNIV_PAGE_SIZE)
+
+/* This parameter controls asynchronous making of a new checkpoint; the value
+should be bigger than LOG_POOL_PREFLUSH_RATIO_SYNC */
+
+#define LOG_POOL_CHECKPOINT_RATIO_ASYNC	32
+
+/* This parameter controls synchronous preflushing of modified buffer pages */
+#define LOG_POOL_PREFLUSH_RATIO_SYNC	16
+
+/* The same ratio for asynchronous preflushing; this value should be less than
+the previous */
+#define LOG_POOL_PREFLUSH_RATIO_ASYNC	8
+
+/* Extra margin, in addition to one log file, used in archiving */
+#define LOG_ARCHIVE_EXTRA_MARGIN	(4 * UNIV_PAGE_SIZE)
+
+/* This parameter controls asynchronous writing to the archive */
+#define LOG_ARCHIVE_RATIO_ASYNC		16
+
+/* Codes used in unlocking flush latches */
+#define LOG_UNLOCK_NONE_FLUSHED_LOCK	1
+#define LOG_UNLOCK_FLUSH_LOCK		2
+
+/* States of an archiving operation */
+#define	LOG_ARCHIVE_READ	1
+#define	LOG_ARCHIVE_WRITE	2
+
+/**********************************************************
+Completes a checkpoint write i/o to a log file. */
+static
+void
+log_io_complete_checkpoint(void);
+/*============================*/
+#ifdef UNIV_LOG_ARCHIVE
+/**********************************************************
+Completes an archiving i/o. */
+static
+void
+log_io_complete_archive(void);
+/*=========================*/
+#endif /* UNIV_LOG_ARCHIVE */
+
+/********************************************************************
+Sets the global variable log_fsp_current_free_limit. Also makes a checkpoint,
+so that we know that the limit has been written to a log checkpoint field
+on disk. */
+UNIV_INTERN
+void
+log_fsp_current_free_limit_set_and_checkpoint(
+/*==========================================*/
+	ulint	limit)	/* in: limit to set */
+{
+	ibool	success;
+
+	mutex_enter(&(log_sys->mutex));
+
+	log_fsp_current_free_limit = limit;
+
+	mutex_exit(&(log_sys->mutex));
+
+	/* Try to make a synchronous checkpoint */
+
+	success = FALSE;
+
+	while (!success) {
+		success = log_checkpoint(TRUE, TRUE);
+	}
+}
+
+/********************************************************************
+Returns the oldest modified block lsn in the pool, or log_sys->lsn if none
+exists. */
+static
+ib_uint64_t
+log_buf_pool_get_oldest_modification(void)
+/*======================================*/
+{
+	ib_uint64_t	lsn;
+
+	ut_ad(mutex_own(&(log_sys->mutex)));
+
+	lsn = buf_pool_get_oldest_modification();
+
+	if (!lsn) {
+
+		lsn = log_sys->lsn;
+	}
+
+	return(lsn);
+}
+
+/****************************************************************
+Opens the log for log_write_low. The log must be closed with log_close and
+released with log_release. */
+UNIV_INTERN
+ib_uint64_t
+log_reserve_and_open(
+/*=================*/
+			/* out: start lsn of the log record */
+	ulint	len)	/* in: length of data to be catenated */
+{
+	log_t*	log			= log_sys;
+	ulint	len_upper_limit;
+#ifdef UNIV_LOG_ARCHIVE
+	ulint	archived_lsn_age;
+	ulint	dummy;
+#endif /* UNIV_LOG_ARCHIVE */
+#ifdef UNIV_DEBUG
+	ulint	count			= 0;
+#endif /* UNIV_DEBUG */
+
+	ut_a(len < log->buf_size / 2);
+loop:
+	mutex_enter(&(log->mutex));
+
+	/* Calculate an upper limit for the space the string may take in the
+	log buffer */
+
+	len_upper_limit = LOG_BUF_WRITE_MARGIN + (5 * len) / 4;
+
+	if (log->buf_free + len_upper_limit > log->buf_size) {
+
+		mutex_exit(&(log->mutex));
+
+		/* Not enough free space, do a syncronous flush of the log
+		buffer */
+
+		log_buffer_flush_to_disk();
+
+		srv_log_waits++;
+
+		ut_ad(++count < 50);
+
+		goto loop;
+	}
+
+#ifdef UNIV_LOG_ARCHIVE
+	if (log->archiving_state != LOG_ARCH_OFF) {
+
+		archived_lsn_age = log->lsn - log->archived_lsn;
+		if (archived_lsn_age + len_upper_limit
+		    > log->max_archived_lsn_age) {
+			/* Not enough free archived space in log groups: do a
+			synchronous archive write batch: */
+
+			mutex_exit(&(log->mutex));
+
+			ut_ad(len_upper_limit <= log->max_archived_lsn_age);
+
+			log_archive_do(TRUE, &dummy);
+
+			ut_ad(++count < 50);
+
+			goto loop;
+		}
+	}
+#endif /* UNIV_LOG_ARCHIVE */
+
+#ifdef UNIV_LOG_DEBUG
+	log->old_buf_free = log->buf_free;
+	log->old_lsn = log->lsn;
+#endif
+	return(log->lsn);
+}
+
+/****************************************************************
+Writes to the log the string given. It is assumed that the caller holds the
+log mutex. */
+UNIV_INTERN
+void
+log_write_low(
+/*==========*/
+	byte*	str,		/* in: string */
+	ulint	str_len)	/* in: string length */
+{
+	log_t*	log	= log_sys;
+	ulint	len;
+	ulint	data_len;
+	byte*	log_block;
+
+	ut_ad(mutex_own(&(log->mutex)));
+part_loop:
+	/* Calculate a part length */
+
+	data_len = (log->buf_free % OS_FILE_LOG_BLOCK_SIZE) + str_len;
+
+	if (data_len <= OS_FILE_LOG_BLOCK_SIZE - LOG_BLOCK_TRL_SIZE) {
+
+		/* The string fits within the current log block */
+
+		len = str_len;
+	} else {
+		data_len = OS_FILE_LOG_BLOCK_SIZE - LOG_BLOCK_TRL_SIZE;
+
+		len = OS_FILE_LOG_BLOCK_SIZE
+			- (log->buf_free % OS_FILE_LOG_BLOCK_SIZE)
+			- LOG_BLOCK_TRL_SIZE;
+	}
+
+	ut_memcpy(log->buf + log->buf_free, str, len);
+
+	str_len -= len;
+	str = str + len;
+
+	log_block = ut_align_down(log->buf + log->buf_free,
+				  OS_FILE_LOG_BLOCK_SIZE);
+	log_block_set_data_len(log_block, data_len);
+
+	if (data_len == OS_FILE_LOG_BLOCK_SIZE - LOG_BLOCK_TRL_SIZE) {
+		/* This block became full */
+		log_block_set_data_len(log_block, OS_FILE_LOG_BLOCK_SIZE);
+		log_block_set_checkpoint_no(log_block,
+					    log_sys->next_checkpoint_no);
+		len += LOG_BLOCK_HDR_SIZE + LOG_BLOCK_TRL_SIZE;
+
+		log->lsn += len;
+
+		/* Initialize the next block header */
+		log_block_init(log_block + OS_FILE_LOG_BLOCK_SIZE, log->lsn);
+	} else {
+		log->lsn += len;
+	}
+
+	log->buf_free += len;
+
+	ut_ad(log->buf_free <= log->buf_size);
+
+	if (str_len > 0) {
+		goto part_loop;
+	}
+
+	srv_log_write_requests++;
+}
+
+/****************************************************************
+Closes the log. */
+UNIV_INTERN
+ib_uint64_t
+log_close(void)
+/*===========*/
+			/* out: lsn */
+{
+	byte*		log_block;
+	ulint		first_rec_group;
+	ib_uint64_t	oldest_lsn;
+	ib_uint64_t	lsn;
+	log_t*		log	= log_sys;
+	ib_uint64_t	checkpoint_age;
+
+	ut_ad(mutex_own(&(log->mutex)));
+
+	lsn = log->lsn;
+
+	log_block = ut_align_down(log->buf + log->buf_free,
+				  OS_FILE_LOG_BLOCK_SIZE);
+	first_rec_group = log_block_get_first_rec_group(log_block);
+
+	if (first_rec_group == 0) {
+		/* We initialized a new log block which was not written
+		full by the current mtr: the next mtr log record group
+		will start within this block at the offset data_len */
+
+		log_block_set_first_rec_group(
+			log_block, log_block_get_data_len(log_block));
+	}
+
+	if (log->buf_free > log->max_buf_free) {
+
+		log->check_flush_or_checkpoint = TRUE;
+	}
+
+	checkpoint_age = lsn - log->last_checkpoint_lsn;
+
+	if (checkpoint_age >= log->log_group_capacity) {
+		/* TODO: split btr_store_big_rec_extern_fields() into small
+		steps so that we can release all latches in the middle, and
+		call log_free_check() to ensure we never write over log written
+		after the latest checkpoint. In principle, we should split all
+		big_rec operations, but other operations are smaller. */
+
+		if (!log_has_printed_chkp_warning
+		    || difftime(time(NULL), log_last_warning_time) > 15) {
+
+			log_has_printed_chkp_warning = TRUE;
+			log_last_warning_time = time(NULL);
+
+			ut_print_timestamp(stderr);
+			fprintf(stderr,
+				"  InnoDB: ERROR: the age of the last"
+				" checkpoint is %lu,\n"
+				"InnoDB: which exceeds the log group"
+				" capacity %lu.\n"
+				"InnoDB: If you are using big"
+				" BLOB or TEXT rows, you must set the\n"
+				"InnoDB: combined size of log files"
+				" at least 10 times bigger than the\n"
+				"InnoDB: largest such row.\n",
+				(ulong) checkpoint_age,
+				(ulong) log->log_group_capacity);
+		}
+	}
+
+	if (checkpoint_age <= log->max_modified_age_async) {
+
+		goto function_exit;
+	}
+
+	oldest_lsn = buf_pool_get_oldest_modification();
+
+	if (!oldest_lsn
+	    || lsn - oldest_lsn > log->max_modified_age_async
+	    || checkpoint_age > log->max_checkpoint_age_async) {
+
+		log->check_flush_or_checkpoint = TRUE;
+	}
+function_exit:
+
+#ifdef UNIV_LOG_DEBUG
+	log_check_log_recs(log->buf + log->old_buf_free,
+			   log->buf_free - log->old_buf_free, log->old_lsn);
+#endif
+
+	return(lsn);
+}
+
+#ifdef UNIV_LOG_ARCHIVE
+/**********************************************************
+Pads the current log block full with dummy log records. Used in producing
+consistent archived log files. */
+static
+void
+log_pad_current_log_block(void)
+/*===========================*/
+{
+	byte		b		= MLOG_DUMMY_RECORD;
+	ulint		pad_length;
+	ulint		i;
+	ib_uint64_t	lsn;
+
+	/* We retrieve lsn only because otherwise gcc crashed on HP-UX */
+	lsn = log_reserve_and_open(OS_FILE_LOG_BLOCK_SIZE);
+
+	pad_length = OS_FILE_LOG_BLOCK_SIZE
+		- (log_sys->buf_free % OS_FILE_LOG_BLOCK_SIZE)
+		- LOG_BLOCK_TRL_SIZE;
+
+	for (i = 0; i < pad_length; i++) {
+		log_write_low(&b, 1);
+	}
+
+	lsn = log_sys->lsn;
+
+	log_close();
+	log_release();
+
+	ut_a(lsn % OS_FILE_LOG_BLOCK_SIZE == LOG_BLOCK_HDR_SIZE);
+}
+#endif /* UNIV_LOG_ARCHIVE */
+
+/**********************************************************
+Calculates the data capacity of a log group, when the log file headers are not
+included. */
+UNIV_INTERN
+ulint
+log_group_get_capacity(
+/*===================*/
+				/* out: capacity in bytes */
+	log_group_t*	group)	/* in: log group */
+{
+	ut_ad(mutex_own(&(log_sys->mutex)));
+
+	return((group->file_size - LOG_FILE_HDR_SIZE) * group->n_files);
+}
+
+/**********************************************************
+Calculates the offset within a log group, when the log file headers are not
+included. */
+UNIV_INLINE
+ulint
+log_group_calc_size_offset(
+/*=======================*/
+				/* out: size offset (<= offset) */
+	ulint		offset,	/* in: real offset within the log group */
+	log_group_t*	group)	/* in: log group */
+{
+	ut_ad(mutex_own(&(log_sys->mutex)));
+
+	return(offset - LOG_FILE_HDR_SIZE * (1 + offset / group->file_size));
+}
+
+/**********************************************************
+Calculates the offset within a log group, when the log file headers are
+included. */
+UNIV_INLINE
+ulint
+log_group_calc_real_offset(
+/*=======================*/
+				/* out: real offset (>= offset) */
+	ulint		offset,	/* in: size offset within the log group */
+	log_group_t*	group)	/* in: log group */
+{
+	ut_ad(mutex_own(&(log_sys->mutex)));
+
+	return(offset + LOG_FILE_HDR_SIZE
+	       * (1 + offset / (group->file_size - LOG_FILE_HDR_SIZE)));
+}
+
+/**********************************************************
+Calculates the offset of an lsn within a log group. */
+static
+ulint
+log_group_calc_lsn_offset(
+/*======================*/
+				/* out: offset within the log group */
+	ib_uint64_t	lsn,	/* in: lsn, must be within 4 GB of
+				group->lsn */
+	log_group_t*	group)	/* in: log group */
+{
+	ib_uint64_t	gr_lsn;
+	ib_int64_t	gr_lsn_size_offset;
+	ib_int64_t	difference;
+	ib_int64_t	group_size;
+	ib_int64_t	offset;
+
+	ut_ad(mutex_own(&(log_sys->mutex)));
+
+	/* If total log file size is > 2 GB we can easily get overflows
+	with 32-bit integers. Use 64-bit integers instead. */
+
+	gr_lsn = group->lsn;
+
+	gr_lsn_size_offset = (ib_int64_t)
+		log_group_calc_size_offset(group->lsn_offset, group);
+
+	group_size = (ib_int64_t) log_group_get_capacity(group);
+
+	if (lsn >= gr_lsn) {
+
+		difference = (ib_int64_t) (lsn - gr_lsn);
+	} else {
+		difference = (ib_int64_t) (gr_lsn - lsn);
+
+		difference = difference % group_size;
+
+		difference = group_size - difference;
+	}
+
+	offset = (gr_lsn_size_offset + difference) % group_size;
+
+	ut_a(offset < (((ib_int64_t) 1) << 32)); /* offset must be < 4 GB */
+
+	/* fprintf(stderr,
+	"Offset is %lu gr_lsn_offset is %lu difference is %lu\n",
+	(ulint)offset,(ulint)gr_lsn_size_offset, (ulint)difference);
+	*/
+
+	return(log_group_calc_real_offset((ulint)offset, group));
+}
+
+/***********************************************************************
+Calculates where in log files we find a specified lsn. */
+UNIV_INTERN
+ulint
+log_calc_where_lsn_is(
+/*==================*/
+						/* out: log file number */
+	ib_int64_t*	log_file_offset,	/* out: offset in that file
+						(including the header) */
+	ib_uint64_t	first_header_lsn,	/* in: first log file start
+						lsn */
+	ib_uint64_t	lsn,			/* in: lsn whose position to
+						determine */
+	ulint		n_log_files,		/* in: total number of log
+						files */
+	ib_int64_t	log_file_size)		/* in: log file size
+						(including the header) */
+{
+	ib_int64_t	capacity	= log_file_size - LOG_FILE_HDR_SIZE;
+	ulint		file_no;
+	ib_int64_t	add_this_many;
+
+	if (lsn < first_header_lsn) {
+		add_this_many = 1 + (first_header_lsn - lsn)
+			/ (capacity * (ib_int64_t)n_log_files);
+		lsn += add_this_many
+			* capacity * (ib_int64_t)n_log_files;
+	}
+
+	ut_a(lsn >= first_header_lsn);
+
+	file_no = ((ulint)((lsn - first_header_lsn) / capacity))
+		% n_log_files;
+	*log_file_offset = (lsn - first_header_lsn) % capacity;
+
+	*log_file_offset = *log_file_offset + LOG_FILE_HDR_SIZE;
+
+	return(file_no);
+}
+
+/************************************************************
+Sets the field values in group to correspond to a given lsn. For this function
+to work, the values must already be correctly initialized to correspond to
+some lsn, for instance, a checkpoint lsn. */
+UNIV_INTERN
+void
+log_group_set_fields(
+/*=================*/
+	log_group_t*	group,	/* in: group */
+	ib_uint64_t	lsn)	/* in: lsn for which the values should be
+				set */
+{
+	group->lsn_offset = log_group_calc_lsn_offset(lsn, group);
+	group->lsn = lsn;
+}
+
+/*********************************************************************
+Calculates the recommended highest values for lsn - last_checkpoint_lsn,
+lsn - buf_get_oldest_modification(), and lsn - max_archive_lsn_age. */
+static
+ibool
+log_calc_max_ages(void)
+/*===================*/
+			/* out: error value FALSE if the smallest log group is
+			too small to accommodate the number of OS threads in
+			the database server */
+{
+	log_group_t*	group;
+	ulint		margin;
+	ulint		free;
+	ibool		success		= TRUE;
+	ulint		smallest_capacity;
+	ulint		archive_margin;
+	ulint		smallest_archive_margin;
+
+	ut_ad(!mutex_own(&(log_sys->mutex)));
+
+	mutex_enter(&(log_sys->mutex));
+
+	group = UT_LIST_GET_FIRST(log_sys->log_groups);
+
+	ut_ad(group);
+
+	smallest_capacity = ULINT_MAX;
+	smallest_archive_margin = ULINT_MAX;
+
+	while (group) {
+		if (log_group_get_capacity(group) < smallest_capacity) {
+
+			smallest_capacity = log_group_get_capacity(group);
+		}
+
+		archive_margin = log_group_get_capacity(group)
+			- (group->file_size - LOG_FILE_HDR_SIZE)
+			- LOG_ARCHIVE_EXTRA_MARGIN;
+
+		if (archive_margin < smallest_archive_margin) {
+
+			smallest_archive_margin = archive_margin;
+		}
+
+		group = UT_LIST_GET_NEXT(log_groups, group);
+	}
+
+	/* Add extra safety */
+	smallest_capacity = smallest_capacity - smallest_capacity / 10;
+
+	/* For each OS thread we must reserve so much free space in the
+	smallest log group that it can accommodate the log entries produced
+	by single query steps: running out of free log space is a serious
+	system error which requires rebooting the database. */
+
+	free = LOG_CHECKPOINT_FREE_PER_THREAD * (10 + srv_thread_concurrency)
+		+ LOG_CHECKPOINT_EXTRA_FREE;
+	if (free >= smallest_capacity / 2) {
+		success = FALSE;
+
+		goto failure;
+	} else {
+		margin = smallest_capacity - free;
+	}
+
+	margin = ut_min(margin, log_sys->adm_checkpoint_interval);
+
+	margin = margin - margin / 10;	/* Add still some extra safety */
+
+	log_sys->log_group_capacity = smallest_capacity;
+
+	log_sys->max_modified_age_async = margin
+		- margin / LOG_POOL_PREFLUSH_RATIO_ASYNC;
+	log_sys->max_modified_age_sync = margin
+		- margin / LOG_POOL_PREFLUSH_RATIO_SYNC;
+
+	log_sys->max_checkpoint_age_async = margin - margin
+		/ LOG_POOL_CHECKPOINT_RATIO_ASYNC;
+	log_sys->max_checkpoint_age = margin;
+
+#ifdef UNIV_LOG_ARCHIVE
+	log_sys->max_archived_lsn_age = smallest_archive_margin;
+
+	log_sys->max_archived_lsn_age_async = smallest_archive_margin
+		- smallest_archive_margin / LOG_ARCHIVE_RATIO_ASYNC;
+#endif /* UNIV_LOG_ARCHIVE */
+failure:
+	mutex_exit(&(log_sys->mutex));
+
+	if (!success) {
+		fprintf(stderr,
+			"InnoDB: Error: ib_logfiles are too small"
+			" for innodb_thread_concurrency %lu.\n"
+			"InnoDB: The combined size of ib_logfiles"
+			" should be bigger than\n"
+			"InnoDB: 200 kB * innodb_thread_concurrency.\n"
+			"InnoDB: To get mysqld to start up, set"
+			" innodb_thread_concurrency in my.cnf\n"
+			"InnoDB: to a lower value, for example, to 8."
+			" After an ERROR-FREE shutdown\n"
+			"InnoDB: of mysqld you can adjust the size of"
+			" ib_logfiles, as explained in\n"
+			"InnoDB: http://dev.mysql.com/doc/refman/5.1/en/"
+			"adding-and-removing.html\n"
+			"InnoDB: Cannot continue operation."
+			" Calling exit(1).\n",
+			(ulong)srv_thread_concurrency);
+
+		exit(1);
+	}
+
+	return(success);
+}
+
+/**********************************************************
+Initializes the log. */
+UNIV_INTERN
+void
+log_init(void)
+/*==========*/
+{
+	byte*	buf;
+
+	log_sys = mem_alloc(sizeof(log_t));
+
+	mutex_create(&log_sys->mutex, SYNC_LOG);
+
+	mutex_enter(&(log_sys->mutex));
+
+	/* Start the lsn from one log block from zero: this way every
+	log record has a start lsn != zero, a fact which we will use */
+
+	log_sys->lsn = LOG_START_LSN;
+
+	ut_a(LOG_BUFFER_SIZE >= 16 * OS_FILE_LOG_BLOCK_SIZE);
+	ut_a(LOG_BUFFER_SIZE >= 4 * UNIV_PAGE_SIZE);
+
+	buf = mem_alloc(LOG_BUFFER_SIZE + OS_FILE_LOG_BLOCK_SIZE);
+	log_sys->buf = ut_align(buf, OS_FILE_LOG_BLOCK_SIZE);
+
+	log_sys->buf_size = LOG_BUFFER_SIZE;
+
+	memset(log_sys->buf, '\0', LOG_BUFFER_SIZE);
+
+	log_sys->max_buf_free = log_sys->buf_size / LOG_BUF_FLUSH_RATIO
+		- LOG_BUF_FLUSH_MARGIN;
+	log_sys->check_flush_or_checkpoint = TRUE;
+	UT_LIST_INIT(log_sys->log_groups);
+
+	log_sys->n_log_ios = 0;
+
+	log_sys->n_log_ios_old = log_sys->n_log_ios;
+	log_sys->last_printout_time = time(NULL);
+	/*----------------------------*/
+
+	log_sys->buf_next_to_write = 0;
+
+	log_sys->write_lsn = 0;
+	log_sys->current_flush_lsn = 0;
+	log_sys->flushed_to_disk_lsn = 0;
+
+	log_sys->written_to_some_lsn = log_sys->lsn;
+	log_sys->written_to_all_lsn = log_sys->lsn;
+
+	log_sys->n_pending_writes = 0;
+
+	log_sys->no_flush_event = os_event_create(NULL);
+
+	os_event_set(log_sys->no_flush_event);
+
+	log_sys->one_flushed_event = os_event_create(NULL);
+
+	os_event_set(log_sys->one_flushed_event);
+
+	/*----------------------------*/
+	log_sys->adm_checkpoint_interval = ULINT_MAX;
+
+	log_sys->next_checkpoint_no = 0;
+	log_sys->last_checkpoint_lsn = log_sys->lsn;
+	log_sys->n_pending_checkpoint_writes = 0;
+
+	rw_lock_create(&log_sys->checkpoint_lock, SYNC_NO_ORDER_CHECK);
+
+	log_sys->checkpoint_buf
+		= ut_align(mem_alloc(2 * OS_FILE_LOG_BLOCK_SIZE),
+			   OS_FILE_LOG_BLOCK_SIZE);
+	memset(log_sys->checkpoint_buf, '\0', OS_FILE_LOG_BLOCK_SIZE);
+	/*----------------------------*/
+
+#ifdef UNIV_LOG_ARCHIVE
+	/* Under MySQL, log archiving is always off */
+	log_sys->archiving_state = LOG_ARCH_OFF;
+	log_sys->archived_lsn = log_sys->lsn;
+	log_sys->next_archived_lsn = 0;
+
+	log_sys->n_pending_archive_ios = 0;
+
+	rw_lock_create(&log_sys->archive_lock, SYNC_NO_ORDER_CHECK);
+
+	log_sys->archive_buf = NULL;
+
+	/* ut_align(
+	ut_malloc(LOG_ARCHIVE_BUF_SIZE
+	+ OS_FILE_LOG_BLOCK_SIZE),
+	OS_FILE_LOG_BLOCK_SIZE); */
+	log_sys->archive_buf_size = 0;
+
+	/* memset(log_sys->archive_buf, '\0', LOG_ARCHIVE_BUF_SIZE); */
+
+	log_sys->archiving_on = os_event_create(NULL);
+#endif /* UNIV_LOG_ARCHIVE */
+
+	/*----------------------------*/
+
+	log_block_init(log_sys->buf, log_sys->lsn);
+	log_block_set_first_rec_group(log_sys->buf, LOG_BLOCK_HDR_SIZE);
+
+	log_sys->buf_free = LOG_BLOCK_HDR_SIZE;
+	log_sys->lsn = LOG_START_LSN + LOG_BLOCK_HDR_SIZE;
+
+	mutex_exit(&(log_sys->mutex));
+
+#ifdef UNIV_LOG_DEBUG
+	recv_sys_create();
+	recv_sys_init(FALSE, buf_pool_get_curr_size());
+
+	recv_sys->parse_start_lsn = log_sys->lsn;
+	recv_sys->scanned_lsn = log_sys->lsn;
+	recv_sys->scanned_checkpoint_no = 0;
+	recv_sys->recovered_lsn = log_sys->lsn;
+	recv_sys->limit_lsn = IB_ULONGLONG_MAX;
+#endif
+}
+
+/**********************************************************************
+Inits a log group to the log system. */
+UNIV_INTERN
+void
+log_group_init(
+/*===========*/
+	ulint	id,			/* in: group id */
+	ulint	n_files,		/* in: number of log files */
+	ulint	file_size,		/* in: log file size in bytes */
+	ulint	space_id,		/* in: space id of the file space
+					which contains the log files of this
+					group */
+	ulint	archive_space_id __attribute__((unused)))
+					/* in: space id of the file space
+					which contains some archived log
+					files for this group; currently, only
+					for the first log group this is
+					used */
+{
+	ulint	i;
+
+	log_group_t*	group;
+
+	group = mem_alloc(sizeof(log_group_t));
+
+	group->id = id;
+	group->n_files = n_files;
+	group->file_size = file_size;
+	group->space_id = space_id;
+	group->state = LOG_GROUP_OK;
+	group->lsn = LOG_START_LSN;
+	group->lsn_offset = LOG_FILE_HDR_SIZE;
+	group->n_pending_writes = 0;
+
+	group->file_header_bufs = mem_alloc(sizeof(byte*) * n_files);
+#ifdef UNIV_LOG_ARCHIVE
+	group->archive_file_header_bufs = mem_alloc(sizeof(byte*) * n_files);
+#endif /* UNIV_LOG_ARCHIVE */
+
+	for (i = 0; i < n_files; i++) {
+		*(group->file_header_bufs + i) = ut_align(
+			mem_alloc(LOG_FILE_HDR_SIZE + OS_FILE_LOG_BLOCK_SIZE),
+			OS_FILE_LOG_BLOCK_SIZE);
+
+		memset(*(group->file_header_bufs + i), '\0',
+		       LOG_FILE_HDR_SIZE);
+
+#ifdef UNIV_LOG_ARCHIVE
+		*(group->archive_file_header_bufs + i) = ut_align(
+			mem_alloc(LOG_FILE_HDR_SIZE + OS_FILE_LOG_BLOCK_SIZE),
+			OS_FILE_LOG_BLOCK_SIZE);
+		memset(*(group->archive_file_header_bufs + i), '\0',
+		       LOG_FILE_HDR_SIZE);
+#endif /* UNIV_LOG_ARCHIVE */
+	}
+
+#ifdef UNIV_LOG_ARCHIVE
+	group->archive_space_id = archive_space_id;
+
+	group->archived_file_no = 0;
+	group->archived_offset = 0;
+#endif /* UNIV_LOG_ARCHIVE */
+
+	group->checkpoint_buf = ut_align(
+		mem_alloc(2 * OS_FILE_LOG_BLOCK_SIZE), OS_FILE_LOG_BLOCK_SIZE);
+
+	memset(group->checkpoint_buf, '\0', OS_FILE_LOG_BLOCK_SIZE);
+
+	UT_LIST_ADD_LAST(log_groups, log_sys->log_groups, group);
+
+	ut_a(log_calc_max_ages());
+}
+
+/**********************************************************************
+Does the unlockings needed in flush i/o completion. */
+UNIV_INLINE
+void
+log_flush_do_unlocks(
+/*=================*/
+	ulint	code)	/* in: any ORed combination of LOG_UNLOCK_FLUSH_LOCK
+			and LOG_UNLOCK_NONE_FLUSHED_LOCK */
+{
+	ut_ad(mutex_own(&(log_sys->mutex)));
+
+	/* NOTE that we must own the log mutex when doing the setting of the
+	events: this is because transactions will wait for these events to
+	be set, and at that moment the log flush they were waiting for must
+	have ended. If the log mutex were not reserved here, the i/o-thread
+	calling this function might be preempted for a while, and when it
+	resumed execution, it might be that a new flush had been started, and
+	this function would erroneously signal the NEW flush as completed.
+	Thus, the changes in the state of these events are performed
+	atomically in conjunction with the changes in the state of
+	log_sys->n_pending_writes etc. */
+
+	if (code & LOG_UNLOCK_NONE_FLUSHED_LOCK) {
+		os_event_set(log_sys->one_flushed_event);
+	}
+
+	if (code & LOG_UNLOCK_FLUSH_LOCK) {
+		os_event_set(log_sys->no_flush_event);
+	}
+}
+
+/**********************************************************************
+Checks if a flush is completed for a log group and does the completion
+routine if yes. */
+UNIV_INLINE
+ulint
+log_group_check_flush_completion(
+/*=============================*/
+				/* out: LOG_UNLOCK_NONE_FLUSHED_LOCK or 0 */
+	log_group_t*	group)	/* in: log group */
+{
+	ut_ad(mutex_own(&(log_sys->mutex)));
+
+	if (!log_sys->one_flushed && group->n_pending_writes == 0) {
+#ifdef UNIV_DEBUG
+		if (log_debug_writes) {
+			fprintf(stderr,
+				"Log flushed first to group %lu\n",
+				(ulong) group->id);
+		}
+#endif /* UNIV_DEBUG */
+		log_sys->written_to_some_lsn = log_sys->write_lsn;
+		log_sys->one_flushed = TRUE;
+
+		return(LOG_UNLOCK_NONE_FLUSHED_LOCK);
+	}
+
+#ifdef UNIV_DEBUG
+	if (log_debug_writes && (group->n_pending_writes == 0)) {
+
+		fprintf(stderr, "Log flushed to group %lu\n",
+			(ulong) group->id);
+	}
+#endif /* UNIV_DEBUG */
+	return(0);
+}
+
+/**********************************************************
+Checks if a flush is completed and does the completion routine if yes. */
+static
+ulint
+log_sys_check_flush_completion(void)
+/*================================*/
+			/* out: LOG_UNLOCK_FLUSH_LOCK or 0 */
+{
+	ulint	move_start;
+	ulint	move_end;
+
+	ut_ad(mutex_own(&(log_sys->mutex)));
+
+	if (log_sys->n_pending_writes == 0) {
+
+		log_sys->written_to_all_lsn = log_sys->write_lsn;
+		log_sys->buf_next_to_write = log_sys->write_end_offset;
+
+		if (log_sys->write_end_offset > log_sys->max_buf_free / 2) {
+			/* Move the log buffer content to the start of the
+			buffer */
+
+			move_start = ut_calc_align_down(
+				log_sys->write_end_offset,
+				OS_FILE_LOG_BLOCK_SIZE);
+			move_end = ut_calc_align(log_sys->buf_free,
+						 OS_FILE_LOG_BLOCK_SIZE);
+
+			ut_memmove(log_sys->buf, log_sys->buf + move_start,
+				   move_end - move_start);
+			log_sys->buf_free -= move_start;
+
+			log_sys->buf_next_to_write -= move_start;
+		}
+
+		return(LOG_UNLOCK_FLUSH_LOCK);
+	}
+
+	return(0);
+}
+
+/**********************************************************
+Completes an i/o to a log file. */
+UNIV_INTERN
+void
+log_io_complete(
+/*============*/
+	log_group_t*	group)	/* in: log group or a dummy pointer */
+{
+	ulint	unlock;
+
+#ifdef UNIV_LOG_ARCHIVE
+	if ((byte*)group == &log_archive_io) {
+		/* It was an archive write */
+
+		log_io_complete_archive();
+
+		return;
+	}
+#endif /* UNIV_LOG_ARCHIVE */
+
+	if ((ulint)group & 0x1UL) {
+		/* It was a checkpoint write */
+		group = (log_group_t*)((ulint)group - 1);
+
+		if (srv_unix_file_flush_method != SRV_UNIX_O_DSYNC
+		    && srv_unix_file_flush_method != SRV_UNIX_NOSYNC) {
+
+			fil_flush(group->space_id);
+		}
+
+#ifdef UNIV_DEBUG
+		if (log_debug_writes) {
+			fprintf(stderr,
+				"Checkpoint info written to group %lu\n",
+				group->id);
+		}
+#endif /* UNIV_DEBUG */
+		log_io_complete_checkpoint();
+
+		return;
+	}
+
+	ut_error;	/* We currently use synchronous writing of the
+			logs and cannot end up here! */
+
+	if (srv_unix_file_flush_method != SRV_UNIX_O_DSYNC
+	    && srv_unix_file_flush_method != SRV_UNIX_NOSYNC
+	    && srv_flush_log_at_trx_commit != 2) {
+
+		fil_flush(group->space_id);
+	}
+
+	mutex_enter(&(log_sys->mutex));
+
+	ut_a(group->n_pending_writes > 0);
+	ut_a(log_sys->n_pending_writes > 0);
+
+	group->n_pending_writes--;
+	log_sys->n_pending_writes--;
+
+	unlock = log_group_check_flush_completion(group);
+	unlock = unlock | log_sys_check_flush_completion();
+
+	log_flush_do_unlocks(unlock);
+
+	mutex_exit(&(log_sys->mutex));
+}
+
+/**********************************************************
+Writes a log file header to a log file space. */
+static
+void
+log_group_file_header_flush(
+/*========================*/
+	log_group_t*	group,		/* in: log group */
+	ulint		nth_file,	/* in: header to the nth file in the
+					log file space */
+	ib_uint64_t	start_lsn)	/* in: log file data starts at this
+					lsn */
+{
+	byte*	buf;
+	ulint	dest_offset;
+
+	ut_ad(mutex_own(&(log_sys->mutex)));
+	ut_a(nth_file < group->n_files);
+
+	buf = *(group->file_header_bufs + nth_file);
+
+	mach_write_to_4(buf + LOG_GROUP_ID, group->id);
+	mach_write_ull(buf + LOG_FILE_START_LSN, start_lsn);
+
+	/* Wipe over possible label of ibbackup --restore */
+	memcpy(buf + LOG_FILE_WAS_CREATED_BY_HOT_BACKUP, "    ", 4);
+
+	dest_offset = nth_file * group->file_size;
+
+#ifdef UNIV_DEBUG
+	if (log_debug_writes) {
+		fprintf(stderr,
+			"Writing log file header to group %lu file %lu\n",
+			(ulong) group->id, (ulong) nth_file);
+	}
+#endif /* UNIV_DEBUG */
+	if (log_do_write) {
+		log_sys->n_log_ios++;
+
+		srv_os_log_pending_writes++;
+
+		fil_io(OS_FILE_WRITE | OS_FILE_LOG, TRUE, group->space_id, 0,
+		       dest_offset / UNIV_PAGE_SIZE,
+		       dest_offset % UNIV_PAGE_SIZE,
+		       OS_FILE_LOG_BLOCK_SIZE,
+		       buf, group);
+
+		srv_os_log_pending_writes--;
+	}
+}
+
+/**********************************************************
+Stores a 4-byte checksum to the trailer checksum field of a log block
+before writing it to a log file. This checksum is used in recovery to
+check the consistency of a log block. */
+static
+void
+log_block_store_checksum(
+/*=====================*/
+	byte*	block)	/* in/out: pointer to a log block */
+{
+	log_block_set_checksum(block, log_block_calc_checksum(block));
+}
+
+/**********************************************************
+Writes a buffer to a log file group. */
+UNIV_INTERN
+void
+log_group_write_buf(
+/*================*/
+	log_group_t*	group,		/* in: log group */
+	byte*		buf,		/* in: buffer */
+	ulint		len,		/* in: buffer len; must be divisible
+					by OS_FILE_LOG_BLOCK_SIZE */
+	ib_uint64_t	start_lsn,	/* in: start lsn of the buffer; must
+					be divisible by
+					OS_FILE_LOG_BLOCK_SIZE */
+	ulint		new_data_offset)/* in: start offset of new data in
+					buf: this parameter is used to decide
+					if we have to write a new log file
+					header */
+{
+	ulint	write_len;
+	ibool	write_header;
+	ulint	next_offset;
+	ulint	i;
+
+	ut_ad(mutex_own(&(log_sys->mutex)));
+	ut_a(len % OS_FILE_LOG_BLOCK_SIZE == 0);
+	ut_a(((ulint) start_lsn) % OS_FILE_LOG_BLOCK_SIZE == 0);
+
+	if (new_data_offset == 0) {
+		write_header = TRUE;
+	} else {
+		write_header = FALSE;
+	}
+loop:
+	if (len == 0) {
+
+		return;
+	}
+
+	next_offset = log_group_calc_lsn_offset(start_lsn, group);
+
+	if ((next_offset % group->file_size == LOG_FILE_HDR_SIZE)
+	    && write_header) {
+		/* We start to write a new log file instance in the group */
+
+		log_group_file_header_flush(group,
+					    next_offset / group->file_size,
+					    start_lsn);
+		srv_os_log_written+= OS_FILE_LOG_BLOCK_SIZE;
+		srv_log_writes++;
+	}
+
+	if ((next_offset % group->file_size) + len > group->file_size) {
+
+		write_len = group->file_size
+			- (next_offset % group->file_size);
+	} else {
+		write_len = len;
+	}
+
+#ifdef UNIV_DEBUG
+	if (log_debug_writes) {
+
+		fprintf(stderr,
+			"Writing log file segment to group %lu"
+			" offset %lu len %lu\n"
+			"start lsn %llu\n"
+			"First block n:o %lu last block n:o %lu\n",
+			(ulong) group->id, (ulong) next_offset,
+			(ulong) write_len,
+			start_lsn,
+			(ulong) log_block_get_hdr_no(buf),
+			(ulong) log_block_get_hdr_no(
+				buf + write_len - OS_FILE_LOG_BLOCK_SIZE));
+		ut_a(log_block_get_hdr_no(buf)
+		     == log_block_convert_lsn_to_no(start_lsn));
+
+		for (i = 0; i < write_len / OS_FILE_LOG_BLOCK_SIZE; i++) {
+
+			ut_a(log_block_get_hdr_no(buf) + i
+			     == log_block_get_hdr_no(
+				     buf + i * OS_FILE_LOG_BLOCK_SIZE));
+		}
+	}
+#endif /* UNIV_DEBUG */
+	/* Calculate the checksums for each log block and write them to
+	the trailer fields of the log blocks */
+
+	for (i = 0; i < write_len / OS_FILE_LOG_BLOCK_SIZE; i++) {
+		log_block_store_checksum(buf + i * OS_FILE_LOG_BLOCK_SIZE);
+	}
+
+	if (log_do_write) {
+		log_sys->n_log_ios++;
+
+		srv_os_log_pending_writes++;
+
+		fil_io(OS_FILE_WRITE | OS_FILE_LOG, TRUE, group->space_id, 0,
+		       next_offset / UNIV_PAGE_SIZE,
+		       next_offset % UNIV_PAGE_SIZE, write_len, buf, group);
+
+		srv_os_log_pending_writes--;
+
+		srv_os_log_written+= write_len;
+		srv_log_writes++;
+	}
+
+	if (write_len < len) {
+		start_lsn += write_len;
+		len -= write_len;
+		buf += write_len;
+
+		write_header = TRUE;
+
+		goto loop;
+	}
+}
+
+/**********************************************************
+This function is called, e.g., when a transaction wants to commit. It checks
+that the log has been written to the log file up to the last log entry written
+by the transaction. If there is a flush running, it waits and checks if the
+flush flushed enough. If not, starts a new flush. */
+UNIV_INTERN
+void
+log_write_up_to(
+/*============*/
+	ib_uint64_t	lsn,	/* in: log sequence number up to which
+				the log should be written,
+				IB_ULONGLONG_MAX if not specified */
+	ulint		wait,	/* in: LOG_NO_WAIT, LOG_WAIT_ONE_GROUP,
+				or LOG_WAIT_ALL_GROUPS */
+	ibool		flush_to_disk)
+				/* in: TRUE if we want the written log
+				also to be flushed to disk */
+{
+	log_group_t*	group;
+	ulint		start_offset;
+	ulint		end_offset;
+	ulint		area_start;
+	ulint		area_end;
+#ifdef UNIV_DEBUG
+	ulint		loop_count	= 0;
+#endif /* UNIV_DEBUG */
+	ulint		unlock;
+
+	if (recv_no_ibuf_operations) {
+		/* Recovery is running and no operations on the log files are
+		allowed yet (the variable name .._no_ibuf_.. is misleading) */
+
+		return;
+	}
+
+loop:
+#ifdef UNIV_DEBUG
+	loop_count++;
+
+	ut_ad(loop_count < 5);
+
+# if 0
+	if (loop_count > 2) {
+		fprintf(stderr, "Log loop count %lu\n", loop_count);
+	}
+# endif
+#endif
+
+	mutex_enter(&(log_sys->mutex));
+
+	if (flush_to_disk
+	    && log_sys->flushed_to_disk_lsn >= lsn) {
+
+		mutex_exit(&(log_sys->mutex));
+
+		return;
+	}
+
+	if (!flush_to_disk
+	    && (log_sys->written_to_all_lsn >= lsn
+		|| (log_sys->written_to_some_lsn >= lsn
+		    && wait != LOG_WAIT_ALL_GROUPS))) {
+
+		mutex_exit(&(log_sys->mutex));
+
+		return;
+	}
+
+	if (log_sys->n_pending_writes > 0) {
+		/* A write (+ possibly flush to disk) is running */
+
+		if (flush_to_disk
+		    && log_sys->current_flush_lsn >= lsn) {
+			/* The write + flush will write enough: wait for it to
+			complete  */
+
+			goto do_waits;
+		}
+
+		if (!flush_to_disk
+		    && log_sys->write_lsn >= lsn) {
+			/* The write will write enough: wait for it to
+			complete  */
+
+			goto do_waits;
+		}
+
+		mutex_exit(&(log_sys->mutex));
+
+		/* Wait for the write to complete and try to start a new
+		write */
+
+		os_event_wait(log_sys->no_flush_event);
+
+		goto loop;
+	}
+
+	if (!flush_to_disk
+	    && log_sys->buf_free == log_sys->buf_next_to_write) {
+		/* Nothing to write and no flush to disk requested */
+
+		mutex_exit(&(log_sys->mutex));
+
+		return;
+	}
+
+#ifdef UNIV_DEBUG
+	if (log_debug_writes) {
+		fprintf(stderr,
+			"Writing log from %llu up to lsn %llu\n",
+			log_sys->written_to_all_lsn,
+			log_sys->lsn);
+	}
+#endif /* UNIV_DEBUG */
+	log_sys->n_pending_writes++;
+
+	group = UT_LIST_GET_FIRST(log_sys->log_groups);
+	group->n_pending_writes++;	/* We assume here that we have only
+					one log group! */
+
+	os_event_reset(log_sys->no_flush_event);
+	os_event_reset(log_sys->one_flushed_event);
+
+	start_offset = log_sys->buf_next_to_write;
+	end_offset = log_sys->buf_free;
+
+	area_start = ut_calc_align_down(start_offset, OS_FILE_LOG_BLOCK_SIZE);
+	area_end = ut_calc_align(end_offset, OS_FILE_LOG_BLOCK_SIZE);
+
+	ut_ad(area_end - area_start > 0);
+
+	log_sys->write_lsn = log_sys->lsn;
+
+	if (flush_to_disk) {
+		log_sys->current_flush_lsn = log_sys->lsn;
+	}
+
+	log_sys->one_flushed = FALSE;
+
+	log_block_set_flush_bit(log_sys->buf + area_start, TRUE);
+	log_block_set_checkpoint_no(
+		log_sys->buf + area_end - OS_FILE_LOG_BLOCK_SIZE,
+		log_sys->next_checkpoint_no);
+
+	/* Copy the last, incompletely written, log block a log block length
+	up, so that when the flush operation writes from the log buffer, the
+	segment to write will not be changed by writers to the log */
+
+	ut_memcpy(log_sys->buf + area_end,
+		  log_sys->buf + area_end - OS_FILE_LOG_BLOCK_SIZE,
+		  OS_FILE_LOG_BLOCK_SIZE);
+
+	log_sys->buf_free += OS_FILE_LOG_BLOCK_SIZE;
+	log_sys->write_end_offset = log_sys->buf_free;
+
+	group = UT_LIST_GET_FIRST(log_sys->log_groups);
+
+	/* Do the write to the log files */
+
+	while (group) {
+		log_group_write_buf(
+			group, log_sys->buf + area_start,
+			area_end - area_start,
+			ut_uint64_align_down(log_sys->written_to_all_lsn,
+					     OS_FILE_LOG_BLOCK_SIZE),
+			start_offset - area_start);
+
+		log_group_set_fields(group, log_sys->write_lsn);
+
+		group = UT_LIST_GET_NEXT(log_groups, group);
+	}
+
+	mutex_exit(&(log_sys->mutex));
+
+	if (srv_unix_file_flush_method == SRV_UNIX_O_DSYNC) {
+		/* O_DSYNC means the OS did not buffer the log file at all:
+		so we have also flushed to disk what we have written */
+
+		log_sys->flushed_to_disk_lsn = log_sys->write_lsn;
+
+	} else if (flush_to_disk) {
+
+		group = UT_LIST_GET_FIRST(log_sys->log_groups);
+
+		fil_flush(group->space_id);
+		log_sys->flushed_to_disk_lsn = log_sys->write_lsn;
+	}
+
+	mutex_enter(&(log_sys->mutex));
+
+	group = UT_LIST_GET_FIRST(log_sys->log_groups);
+
+	ut_a(group->n_pending_writes == 1);
+	ut_a(log_sys->n_pending_writes == 1);
+
+	group->n_pending_writes--;
+	log_sys->n_pending_writes--;
+
+	unlock = log_group_check_flush_completion(group);
+	unlock = unlock | log_sys_check_flush_completion();
+
+	log_flush_do_unlocks(unlock);
+
+	mutex_exit(&(log_sys->mutex));
+
+	return;
+
+do_waits:
+	mutex_exit(&(log_sys->mutex));
+
+	switch (wait) {
+	case LOG_WAIT_ONE_GROUP:
+		os_event_wait(log_sys->one_flushed_event);
+		break;
+	case LOG_WAIT_ALL_GROUPS:
+		os_event_wait(log_sys->no_flush_event);
+		break;
+#ifdef UNIV_DEBUG
+	case LOG_NO_WAIT:
+		break;
+	default:
+		ut_error;
+#endif /* UNIV_DEBUG */
+	}
+}
+
+/********************************************************************
+Does a syncronous flush of the log buffer to disk. */
+UNIV_INTERN
+void
+log_buffer_flush_to_disk(void)
+/*==========================*/
+{
+	ib_uint64_t	lsn;
+
+	mutex_enter(&(log_sys->mutex));
+
+	lsn = log_sys->lsn;
+
+	mutex_exit(&(log_sys->mutex));
+
+	log_write_up_to(lsn, LOG_WAIT_ALL_GROUPS, TRUE);
+}
+
+/********************************************************************
+Tries to establish a big enough margin of free space in the log buffer, such
+that a new log entry can be catenated without an immediate need for a flush. */
+static
+void
+log_flush_margin(void)
+/*==================*/
+{
+	log_t*		log	= log_sys;
+	ib_uint64_t	lsn	= 0;
+
+	mutex_enter(&(log->mutex));
+
+	if (log->buf_free > log->max_buf_free) {
+
+		if (log->n_pending_writes > 0) {
+			/* A flush is running: hope that it will provide enough
+			free space */
+		} else {
+			lsn = log->lsn;
+		}
+	}
+
+	mutex_exit(&(log->mutex));
+
+	if (lsn) {
+		log_write_up_to(lsn, LOG_NO_WAIT, FALSE);
+	}
+}
+
+/********************************************************************
+Advances the smallest lsn for which there are unflushed dirty blocks in the
+buffer pool. NOTE: this function may only be called if the calling thread owns
+no synchronization objects! */
+UNIV_INTERN
+ibool
+log_preflush_pool_modified_pages(
+/*=============================*/
+					/* out: FALSE if there was a
+					flush batch of the same type
+					running, which means that we
+					could not start this flush
+					batch */
+	ib_uint64_t	new_oldest,	/* in: try to advance
+					oldest_modified_lsn at least
+					to this lsn */
+	ibool		sync)		/* in: TRUE if synchronous
+					operation is desired */
+{
+	ulint	n_pages;
+
+	if (recv_recovery_on) {
+		/* If the recovery is running, we must first apply all
+		log records to their respective file pages to get the
+		right modify lsn values to these pages: otherwise, there
+		might be pages on disk which are not yet recovered to the
+		current lsn, and even after calling this function, we could
+		not know how up-to-date the disk version of the database is,
+		and we could not make a new checkpoint on the basis of the
+		info on the buffer pool only. */
+
+		recv_apply_hashed_log_recs(TRUE);
+	}
+
+	n_pages = buf_flush_batch(BUF_FLUSH_LIST, ULINT_MAX, new_oldest);
+
+	if (sync) {
+		buf_flush_wait_batch_end(BUF_FLUSH_LIST);
+	}
+
+	if (n_pages == ULINT_UNDEFINED) {
+
+		return(FALSE);
+	}
+
+	return(TRUE);
+}
+
+/**********************************************************
+Completes a checkpoint. */
+static
+void
+log_complete_checkpoint(void)
+/*=========================*/
+{
+	ut_ad(mutex_own(&(log_sys->mutex)));
+	ut_ad(log_sys->n_pending_checkpoint_writes == 0);
+
+	log_sys->next_checkpoint_no++;
+
+	log_sys->last_checkpoint_lsn = log_sys->next_checkpoint_lsn;
+
+	rw_lock_x_unlock_gen(&(log_sys->checkpoint_lock), LOG_CHECKPOINT);
+}
+
+/**********************************************************
+Completes an asynchronous checkpoint info write i/o to a log file. */
+static
+void
+log_io_complete_checkpoint(void)
+/*============================*/
+{
+	mutex_enter(&(log_sys->mutex));
+
+	ut_ad(log_sys->n_pending_checkpoint_writes > 0);
+
+	log_sys->n_pending_checkpoint_writes--;
+
+	if (log_sys->n_pending_checkpoint_writes == 0) {
+		log_complete_checkpoint();
+	}
+
+	mutex_exit(&(log_sys->mutex));
+}
+
+/***********************************************************************
+Writes info to a checkpoint about a log group. */
+static
+void
+log_checkpoint_set_nth_group_info(
+/*==============================*/
+	byte*	buf,	/* in: buffer for checkpoint info */
+	ulint	n,	/* in: nth slot */
+	ulint	file_no,/* in: archived file number */
+	ulint	offset)	/* in: archived file offset */
+{
+	ut_ad(n < LOG_MAX_N_GROUPS);
+
+	mach_write_to_4(buf + LOG_CHECKPOINT_GROUP_ARRAY
+			+ 8 * n + LOG_CHECKPOINT_ARCHIVED_FILE_NO, file_no);
+	mach_write_to_4(buf + LOG_CHECKPOINT_GROUP_ARRAY
+			+ 8 * n + LOG_CHECKPOINT_ARCHIVED_OFFSET, offset);
+}
+
+/***********************************************************************
+Gets info from a checkpoint about a log group. */
+UNIV_INTERN
+void
+log_checkpoint_get_nth_group_info(
+/*==============================*/
+	byte*	buf,	/* in: buffer containing checkpoint info */
+	ulint	n,	/* in: nth slot */
+	ulint*	file_no,/* out: archived file number */
+	ulint*	offset)	/* out: archived file offset */
+{
+	ut_ad(n < LOG_MAX_N_GROUPS);
+
+	*file_no = mach_read_from_4(buf + LOG_CHECKPOINT_GROUP_ARRAY
+				    + 8 * n + LOG_CHECKPOINT_ARCHIVED_FILE_NO);
+	*offset = mach_read_from_4(buf + LOG_CHECKPOINT_GROUP_ARRAY
+				   + 8 * n + LOG_CHECKPOINT_ARCHIVED_OFFSET);
+}
+
+/**********************************************************
+Writes the checkpoint info to a log group header. */
+static
+void
+log_group_checkpoint(
+/*=================*/
+	log_group_t*	group)	/* in: log group */
+{
+	log_group_t*	group2;
+#ifdef UNIV_LOG_ARCHIVE
+	ib_uint64_t	archived_lsn;
+	ib_uint64_t	next_archived_lsn;
+#endif /* UNIV_LOG_ARCHIVE */
+	ulint		write_offset;
+	ulint		fold;
+	byte*		buf;
+	ulint		i;
+
+	ut_ad(mutex_own(&(log_sys->mutex)));
+#if LOG_CHECKPOINT_SIZE > OS_FILE_LOG_BLOCK_SIZE
+# error "LOG_CHECKPOINT_SIZE > OS_FILE_LOG_BLOCK_SIZE"
+#endif
+
+	buf = group->checkpoint_buf;
+
+	mach_write_ull(buf + LOG_CHECKPOINT_NO, log_sys->next_checkpoint_no);
+	mach_write_ull(buf + LOG_CHECKPOINT_LSN, log_sys->next_checkpoint_lsn);
+
+	mach_write_to_4(buf + LOG_CHECKPOINT_OFFSET,
+			log_group_calc_lsn_offset(
+				log_sys->next_checkpoint_lsn, group));
+
+	mach_write_to_4(buf + LOG_CHECKPOINT_LOG_BUF_SIZE, log_sys->buf_size);
+
+#ifdef UNIV_LOG_ARCHIVE
+	if (log_sys->archiving_state == LOG_ARCH_OFF) {
+		archived_lsn = IB_ULONGLONG_MAX;
+	} else {
+		archived_lsn = log_sys->archived_lsn;
+
+		if (archived_lsn != log_sys->next_archived_lsn) {
+			next_archived_lsn = log_sys->next_archived_lsn;
+			/* For debugging only */
+		}
+	}
+
+	mach_write_ull(buf + LOG_CHECKPOINT_ARCHIVED_LSN, archived_lsn);
+#else /* UNIV_LOG_ARCHIVE */
+	mach_write_ull(buf + LOG_CHECKPOINT_ARCHIVED_LSN, IB_ULONGLONG_MAX);
+#endif /* UNIV_LOG_ARCHIVE */
+
+	for (i = 0; i < LOG_MAX_N_GROUPS; i++) {
+		log_checkpoint_set_nth_group_info(buf, i, 0, 0);
+	}
+
+	group2 = UT_LIST_GET_FIRST(log_sys->log_groups);
+
+	while (group2) {
+		log_checkpoint_set_nth_group_info(buf, group2->id,
+#ifdef UNIV_LOG_ARCHIVE
+						  group2->archived_file_no,
+						  group2->archived_offset
+#else /* UNIV_LOG_ARCHIVE */
+						  0, 0
+#endif /* UNIV_LOG_ARCHIVE */
+						  );
+
+		group2 = UT_LIST_GET_NEXT(log_groups, group2);
+	}
+
+	fold = ut_fold_binary(buf, LOG_CHECKPOINT_CHECKSUM_1);
+	mach_write_to_4(buf + LOG_CHECKPOINT_CHECKSUM_1, fold);
+
+	fold = ut_fold_binary(buf + LOG_CHECKPOINT_LSN,
+			      LOG_CHECKPOINT_CHECKSUM_2 - LOG_CHECKPOINT_LSN);
+	mach_write_to_4(buf + LOG_CHECKPOINT_CHECKSUM_2, fold);
+
+	/* Starting from InnoDB-3.23.50, we also write info on allocated
+	size in the tablespace */
+
+	mach_write_to_4(buf + LOG_CHECKPOINT_FSP_FREE_LIMIT,
+			log_fsp_current_free_limit);
+
+	mach_write_to_4(buf + LOG_CHECKPOINT_FSP_MAGIC_N,
+			LOG_CHECKPOINT_FSP_MAGIC_N_VAL);
+
+	/* We alternate the physical place of the checkpoint info in the first
+	log file */
+
+	if ((log_sys->next_checkpoint_no & 1) == 0) {
+		write_offset = LOG_CHECKPOINT_1;
+	} else {
+		write_offset = LOG_CHECKPOINT_2;
+	}
+
+	if (log_do_write) {
+		if (log_sys->n_pending_checkpoint_writes == 0) {
+
+			rw_lock_x_lock_gen(&(log_sys->checkpoint_lock),
+					   LOG_CHECKPOINT);
+		}
+
+		log_sys->n_pending_checkpoint_writes++;
+
+		log_sys->n_log_ios++;
+
+		/* We send as the last parameter the group machine address
+		added with 1, as we want to distinguish between a normal log
+		file write and a checkpoint field write */
+
+		fil_io(OS_FILE_WRITE | OS_FILE_LOG, FALSE, group->space_id, 0,
+		       write_offset / UNIV_PAGE_SIZE,
+		       write_offset % UNIV_PAGE_SIZE,
+		       OS_FILE_LOG_BLOCK_SIZE,
+		       buf, ((byte*)group + 1));
+
+		ut_ad(((ulint)group & 0x1UL) == 0);
+	}
+}
+
+#ifdef UNIV_HOTBACKUP
+/**********************************************************
+Writes info to a buffer of a log group when log files are created in
+backup restoration. */
+UNIV_INTERN
+void
+log_reset_first_header_and_checkpoint(
+/*==================================*/
+	byte*		hdr_buf,/* in: buffer which will be written to the
+				start of the first log file */
+	ib_uint64_t	start)	/* in: lsn of the start of the first log file;
+				we pretend that there is a checkpoint at
+				start + LOG_BLOCK_HDR_SIZE */
+{
+	ulint		fold;
+	byte*		buf;
+	ib_uint64_t	lsn;
+
+	mach_write_to_4(hdr_buf + LOG_GROUP_ID, 0);
+	mach_write_ull(hdr_buf + LOG_FILE_START_LSN, start);
+
+	lsn = start + LOG_BLOCK_HDR_SIZE;
+
+	/* Write the label of ibbackup --restore */
+	strcpy((char*) hdr_buf + LOG_FILE_WAS_CREATED_BY_HOT_BACKUP,
+	       "ibbackup ");
+	ut_sprintf_timestamp((char*) hdr_buf
+			     + (LOG_FILE_WAS_CREATED_BY_HOT_BACKUP
+				+ (sizeof "ibbackup ") - 1));
+	buf = hdr_buf + LOG_CHECKPOINT_1;
+
+	mach_write_ull(buf + LOG_CHECKPOINT_NO, 0);
+	mach_write_ull(buf + LOG_CHECKPOINT_LSN, lsn);
+
+	mach_write_to_4(buf + LOG_CHECKPOINT_OFFSET,
+			LOG_FILE_HDR_SIZE + LOG_BLOCK_HDR_SIZE);
+
+	mach_write_to_4(buf + LOG_CHECKPOINT_LOG_BUF_SIZE, 2 * 1024 * 1024);
+
+	mach_write_ull(buf + LOG_CHECKPOINT_ARCHIVED_LSN, IB_ULONGLONG_MAX);
+
+	fold = ut_fold_binary(buf, LOG_CHECKPOINT_CHECKSUM_1);
+	mach_write_to_4(buf + LOG_CHECKPOINT_CHECKSUM_1, fold);
+
+	fold = ut_fold_binary(buf + LOG_CHECKPOINT_LSN,
+			      LOG_CHECKPOINT_CHECKSUM_2 - LOG_CHECKPOINT_LSN);
+	mach_write_to_4(buf + LOG_CHECKPOINT_CHECKSUM_2, fold);
+
+	/* Starting from InnoDB-3.23.50, we should also write info on
+	allocated size in the tablespace, but unfortunately we do not
+	know it here */
+}
+#endif /* UNIV_HOTBACKUP */
+
+/**********************************************************
+Reads a checkpoint info from a log group header to log_sys->checkpoint_buf. */
+UNIV_INTERN
+void
+log_group_read_checkpoint_info(
+/*===========================*/
+	log_group_t*	group,	/* in: log group */
+	ulint		field)	/* in: LOG_CHECKPOINT_1 or LOG_CHECKPOINT_2 */
+{
+	ut_ad(mutex_own(&(log_sys->mutex)));
+
+	log_sys->n_log_ios++;
+
+	fil_io(OS_FILE_READ | OS_FILE_LOG, TRUE, group->space_id, 0,
+	       field / UNIV_PAGE_SIZE, field % UNIV_PAGE_SIZE,
+	       OS_FILE_LOG_BLOCK_SIZE, log_sys->checkpoint_buf, NULL);
+}
+
+/**********************************************************
+Writes checkpoint info to groups. */
+UNIV_INTERN
+void
+log_groups_write_checkpoint_info(void)
+/*==================================*/
+{
+	log_group_t*	group;
+
+	ut_ad(mutex_own(&(log_sys->mutex)));
+
+	group = UT_LIST_GET_FIRST(log_sys->log_groups);
+
+	while (group) {
+		log_group_checkpoint(group);
+
+		group = UT_LIST_GET_NEXT(log_groups, group);
+	}
+}
+
+/**********************************************************
+Makes a checkpoint. Note that this function does not flush dirty
+blocks from the buffer pool: it only checks what is lsn of the oldest
+modification in the pool, and writes information about the lsn in
+log files. Use log_make_checkpoint_at to flush also the pool. */
+UNIV_INTERN
+ibool
+log_checkpoint(
+/*===========*/
+				/* out: TRUE if success, FALSE if a checkpoint
+				write was already running */
+	ibool	sync,		/* in: TRUE if synchronous operation is
+				desired */
+	ibool	write_always)	/* in: the function normally checks if the
+				the new checkpoint would have a greater
+				lsn than the previous one: if not, then no
+				physical write is done; by setting this
+				parameter TRUE, a physical write will always be
+				made to log files */
+{
+	ib_uint64_t	oldest_lsn;
+
+	if (recv_recovery_is_on()) {
+		recv_apply_hashed_log_recs(TRUE);
+	}
+
+	if (srv_unix_file_flush_method != SRV_UNIX_NOSYNC) {
+		fil_flush_file_spaces(FIL_TABLESPACE);
+	}
+
+	mutex_enter(&(log_sys->mutex));
+
+	oldest_lsn = log_buf_pool_get_oldest_modification();
+
+	mutex_exit(&(log_sys->mutex));
+
+	/* Because log also contains headers and dummy log records,
+	if the buffer pool contains no dirty buffers, oldest_lsn
+	gets the value log_sys->lsn from the previous function,
+	and we must make sure that the log is flushed up to that
+	lsn. If there are dirty buffers in the buffer pool, then our
+	write-ahead-logging algorithm ensures that the log has been flushed
+	up to oldest_lsn. */
+
+	log_write_up_to(oldest_lsn, LOG_WAIT_ALL_GROUPS, TRUE);
+
+	mutex_enter(&(log_sys->mutex));
+
+	if (!write_always
+	    && log_sys->last_checkpoint_lsn >= oldest_lsn) {
+
+		mutex_exit(&(log_sys->mutex));
+
+		return(TRUE);
+	}
+
+	ut_ad(log_sys->written_to_all_lsn >= oldest_lsn);
+
+	if (log_sys->n_pending_checkpoint_writes > 0) {
+		/* A checkpoint write is running */
+
+		mutex_exit(&(log_sys->mutex));
+
+		if (sync) {
+			/* Wait for the checkpoint write to complete */
+			rw_lock_s_lock(&(log_sys->checkpoint_lock));
+			rw_lock_s_unlock(&(log_sys->checkpoint_lock));
+		}
+
+		return(FALSE);
+	}
+
+	log_sys->next_checkpoint_lsn = oldest_lsn;
+
+#ifdef UNIV_DEBUG
+	if (log_debug_writes) {
+		fprintf(stderr, "Making checkpoint no %lu at lsn %llu\n",
+			(ulong) log_sys->next_checkpoint_no,
+			oldest_lsn);
+	}
+#endif /* UNIV_DEBUG */
+
+	log_groups_write_checkpoint_info();
+
+	mutex_exit(&(log_sys->mutex));
+
+	if (sync) {
+		/* Wait for the checkpoint write to complete */
+		rw_lock_s_lock(&(log_sys->checkpoint_lock));
+		rw_lock_s_unlock(&(log_sys->checkpoint_lock));
+	}
+
+	return(TRUE);
+}
+
+/********************************************************************
+Makes a checkpoint at a given lsn or later. */
+UNIV_INTERN
+void
+log_make_checkpoint_at(
+/*===================*/
+	ib_uint64_t	lsn,		/* in: make a checkpoint at this or a
+					later lsn, if IB_ULONGLONG_MAX, makes
+					a checkpoint at the latest lsn */
+	ibool		write_always)	/* in: the function normally checks if
+					the the new checkpoint would have a
+					greater lsn than the previous one: if
+					not, then no physical write is done;
+					by setting this parameter TRUE, a
+					physical write will always be made to
+					log files */
+{
+	/* Preflush pages synchronously */
+
+	while (!log_preflush_pool_modified_pages(lsn, TRUE));
+
+	while (!log_checkpoint(TRUE, write_always));
+}
+
+/********************************************************************
+Tries to establish a big enough margin of free space in the log groups, such
+that a new log entry can be catenated without an immediate need for a
+checkpoint. NOTE: this function may only be called if the calling thread
+owns no synchronization objects! */
+static
+void
+log_checkpoint_margin(void)
+/*=======================*/
+{
+	log_t*		log		= log_sys;
+	ib_uint64_t	age;
+	ib_uint64_t	checkpoint_age;
+	ib_uint64_t	advance;
+	ib_uint64_t	oldest_lsn;
+	ibool		sync;
+	ibool		checkpoint_sync;
+	ibool		do_checkpoint;
+	ibool		success;
+loop:
+	sync = FALSE;
+	checkpoint_sync = FALSE;
+	do_checkpoint = FALSE;
+
+	mutex_enter(&(log->mutex));
+
+	if (log->check_flush_or_checkpoint == FALSE) {
+		mutex_exit(&(log->mutex));
+
+		return;
+	}
+
+	oldest_lsn = log_buf_pool_get_oldest_modification();
+
+	age = log->lsn - oldest_lsn;
+
+	if (age > log->max_modified_age_sync) {
+
+		/* A flush is urgent: we have to do a synchronous preflush */
+
+		sync = TRUE;
+		advance = 2 * (age - log->max_modified_age_sync);
+	} else if (age > log->max_modified_age_async) {
+
+		/* A flush is not urgent: we do an asynchronous preflush */
+		advance = age - log->max_modified_age_async;
+	} else {
+		advance = 0;
+	}
+
+	checkpoint_age = log->lsn - log->last_checkpoint_lsn;
+
+	if (checkpoint_age > log->max_checkpoint_age) {
+		/* A checkpoint is urgent: we do it synchronously */
+
+		checkpoint_sync = TRUE;
+
+		do_checkpoint = TRUE;
+
+	} else if (checkpoint_age > log->max_checkpoint_age_async) {
+		/* A checkpoint is not urgent: do it asynchronously */
+
+		do_checkpoint = TRUE;
+
+		log->check_flush_or_checkpoint = FALSE;
+	} else {
+		log->check_flush_or_checkpoint = FALSE;
+	}
+
+	mutex_exit(&(log->mutex));
+
+	if (advance) {
+		ib_uint64_t	new_oldest = oldest_lsn + advance;
+
+		success = log_preflush_pool_modified_pages(new_oldest, sync);
+
+		/* If the flush succeeded, this thread has done its part
+		and can proceed. If it did not succeed, there was another
+		thread doing a flush at the same time. If sync was FALSE,
+		the flush was not urgent, and we let this thread proceed.
+		Otherwise, we let it start from the beginning again. */
+
+		if (sync && !success) {
+			mutex_enter(&(log->mutex));
+
+			log->check_flush_or_checkpoint = TRUE;
+
+			mutex_exit(&(log->mutex));
+			goto loop;
+		}
+	}
+
+	if (do_checkpoint) {
+		log_checkpoint(checkpoint_sync, FALSE);
+
+		if (checkpoint_sync) {
+
+			goto loop;
+		}
+	}
+}
+
+/**********************************************************
+Reads a specified log segment to a buffer. */
+UNIV_INTERN
+void
+log_group_read_log_seg(
+/*===================*/
+	ulint		type,		/* in: LOG_ARCHIVE or LOG_RECOVER */
+	byte*		buf,		/* in: buffer where to read */
+	log_group_t*	group,		/* in: log group */
+	ib_uint64_t	start_lsn,	/* in: read area start */
+	ib_uint64_t	end_lsn)	/* in: read area end */
+{
+	ulint	len;
+	ulint	source_offset;
+	ibool	sync;
+
+	ut_ad(mutex_own(&(log_sys->mutex)));
+
+	sync = (type == LOG_RECOVER);
+loop:
+	source_offset = log_group_calc_lsn_offset(start_lsn, group);
+
+	len = (ulint) (end_lsn - start_lsn);
+
+	ut_ad(len != 0);
+
+	if ((source_offset % group->file_size) + len > group->file_size) {
+
+		len = group->file_size - (source_offset % group->file_size);
+	}
+
+#ifdef UNIV_LOG_ARCHIVE
+	if (type == LOG_ARCHIVE) {
+
+		log_sys->n_pending_archive_ios++;
+	}
+#endif /* UNIV_LOG_ARCHIVE */
+
+	log_sys->n_log_ios++;
+
+	fil_io(OS_FILE_READ | OS_FILE_LOG, sync, group->space_id, 0,
+	       source_offset / UNIV_PAGE_SIZE, source_offset % UNIV_PAGE_SIZE,
+	       len, buf, NULL);
+
+	start_lsn += len;
+	buf += len;
+
+	if (start_lsn != end_lsn) {
+
+		goto loop;
+	}
+}
+
+#ifdef UNIV_LOG_ARCHIVE
+/**********************************************************
+Generates an archived log file name. */
+UNIV_INTERN
+void
+log_archived_file_name_gen(
+/*=======================*/
+	char*	buf,	/* in: buffer where to write */
+	ulint	id __attribute__((unused)),
+			/* in: group id;
+			currently we only archive the first group */
+	ulint	file_no)/* in: file number */
+{
+	sprintf(buf, "%sib_arch_log_%010lu", srv_arch_dir, (ulong) file_no);
+}
+
+/**********************************************************
+Writes a log file header to a log file space. */
+static
+void
+log_group_archive_file_header_write(
+/*================================*/
+	log_group_t*	group,		/* in: log group */
+	ulint		nth_file,	/* in: header to the nth file in the
+					archive log file space */
+	ulint		file_no,	/* in: archived file number */
+	ib_uint64_t	start_lsn)	/* in: log file data starts at this
+					lsn */
+{
+	byte*	buf;
+	ulint	dest_offset;
+
+	ut_ad(mutex_own(&(log_sys->mutex)));
+
+	ut_a(nth_file < group->n_files);
+
+	buf = *(group->archive_file_header_bufs + nth_file);
+
+	mach_write_to_4(buf + LOG_GROUP_ID, group->id);
+	mach_write_ull(buf + LOG_FILE_START_LSN, start_lsn);
+	mach_write_to_4(buf + LOG_FILE_NO, file_no);
+
+	mach_write_to_4(buf + LOG_FILE_ARCH_COMPLETED, FALSE);
+
+	dest_offset = nth_file * group->file_size;
+
+	log_sys->n_log_ios++;
+
+	fil_io(OS_FILE_WRITE | OS_FILE_LOG, TRUE, group->archive_space_id,
+	       dest_offset / UNIV_PAGE_SIZE,
+	       dest_offset % UNIV_PAGE_SIZE,
+	       2 * OS_FILE_LOG_BLOCK_SIZE,
+	       buf, &log_archive_io);
+}
+
+/**********************************************************
+Writes a log file header to a completed archived log file. */
+static
+void
+log_group_archive_completed_header_write(
+/*=====================================*/
+	log_group_t*	group,		/* in: log group */
+	ulint		nth_file,	/* in: header to the nth file in the
+					archive log file space */
+	ib_uint64_t	end_lsn)	/* in: end lsn of the file */
+{
+	byte*	buf;
+	ulint	dest_offset;
+
+	ut_ad(mutex_own(&(log_sys->mutex)));
+	ut_a(nth_file < group->n_files);
+
+	buf = *(group->archive_file_header_bufs + nth_file);
+
+	mach_write_to_4(buf + LOG_FILE_ARCH_COMPLETED, TRUE);
+	mach_write_ull(buf + LOG_FILE_END_LSN, end_lsn);
+
+	dest_offset = nth_file * group->file_size + LOG_FILE_ARCH_COMPLETED;
+
+	log_sys->n_log_ios++;
+
+	fil_io(OS_FILE_WRITE | OS_FILE_LOG, TRUE, group->archive_space_id,
+	       dest_offset / UNIV_PAGE_SIZE,
+	       dest_offset % UNIV_PAGE_SIZE,
+	       OS_FILE_LOG_BLOCK_SIZE,
+	       buf + LOG_FILE_ARCH_COMPLETED,
+	       &log_archive_io);
+}
+
+/**********************************************************
+Does the archive writes for a single log group. */
+static
+void
+log_group_archive(
+/*==============*/
+	log_group_t*	group)	/* in: log group */
+{
+	os_file_t	 file_handle;
+	ib_uint64_t	start_lsn;
+	ib_uint64_t	end_lsn;
+	char		name[1024];
+	byte*		buf;
+	ulint		len;
+	ibool		ret;
+	ulint		next_offset;
+	ulint		n_files;
+	ulint		open_mode;
+
+	ut_ad(mutex_own(&(log_sys->mutex)));
+
+	start_lsn = log_sys->archived_lsn;
+
+	ut_a(start_lsn % OS_FILE_LOG_BLOCK_SIZE == 0);
+
+	end_lsn = log_sys->next_archived_lsn;
+
+	ut_a(end_lsn % OS_FILE_LOG_BLOCK_SIZE == 0);
+
+	buf = log_sys->archive_buf;
+
+	n_files = 0;
+
+	next_offset = group->archived_offset;
+loop:
+	if ((next_offset % group->file_size == 0)
+	    || (fil_space_get_size(group->archive_space_id) == 0)) {
+
+		/* Add the file to the archive file space; create or open the
+		file */
+
+		if (next_offset % group->file_size == 0) {
+			open_mode = OS_FILE_CREATE;
+		} else {
+			open_mode = OS_FILE_OPEN;
+		}
+
+		log_archived_file_name_gen(name, group->id,
+					   group->archived_file_no + n_files);
+
+		file_handle = os_file_create(name, open_mode, OS_FILE_AIO,
+					     OS_DATA_FILE, &ret);
+
+		if (!ret && (open_mode == OS_FILE_CREATE)) {
+			file_handle = os_file_create(
+				name, OS_FILE_OPEN, OS_FILE_AIO,
+				OS_DATA_FILE, &ret);
+		}
+
+		if (!ret) {
+			fprintf(stderr,
+				"InnoDB: Cannot create or open"
+				" archive log file %s.\n"
+				"InnoDB: Cannot continue operation.\n"
+				"InnoDB: Check that the log archive"
+				" directory exists,\n"
+				"InnoDB: you have access rights to it, and\n"
+				"InnoDB: there is space available.\n", name);
+			exit(1);
+		}
+
+#ifdef UNIV_DEBUG
+		if (log_debug_writes) {
+			fprintf(stderr, "Created archive file %s\n", name);
+		}
+#endif /* UNIV_DEBUG */
+
+		ret = os_file_close(file_handle);
+
+		ut_a(ret);
+
+		/* Add the archive file as a node to the space */
+
+		fil_node_create(name, group->file_size / UNIV_PAGE_SIZE,
+				group->archive_space_id, FALSE);
+
+		if (next_offset % group->file_size == 0) {
+			log_group_archive_file_header_write(
+				group, n_files,
+				group->archived_file_no + n_files,
+				start_lsn);
+
+			next_offset += LOG_FILE_HDR_SIZE;
+		}
+	}
+
+	len = end_lsn - start_lsn;
+
+	if (group->file_size < (next_offset % group->file_size) + len) {
+
+		len = group->file_size - (next_offset % group->file_size);
+	}
+
+#ifdef UNIV_DEBUG
+	if (log_debug_writes) {
+		fprintf(stderr,
+			"Archiving starting at lsn %llu, len %lu"
+			" to group %lu\n",
+			start_lsn,
+			(ulong) len, (ulong) group->id);
+	}
+#endif /* UNIV_DEBUG */
+
+	log_sys->n_pending_archive_ios++;
+
+	log_sys->n_log_ios++;
+
+	fil_io(OS_FILE_WRITE | OS_FILE_LOG, FALSE, group->archive_space_id,
+	       next_offset / UNIV_PAGE_SIZE, next_offset % UNIV_PAGE_SIZE,
+	       ut_calc_align(len, OS_FILE_LOG_BLOCK_SIZE), buf,
+	       &log_archive_io);
+
+	start_lsn += len;
+	next_offset += len;
+	buf += len;
+
+	if (next_offset % group->file_size == 0) {
+		n_files++;
+	}
+
+	if (end_lsn != start_lsn) {
+
+		goto loop;
+	}
+
+	group->next_archived_file_no = group->archived_file_no + n_files;
+	group->next_archived_offset = next_offset % group->file_size;
+
+	ut_a(group->next_archived_offset % OS_FILE_LOG_BLOCK_SIZE == 0);
+}
+
+/*********************************************************
+(Writes to the archive of each log group.) Currently, only the first
+group is archived. */
+static
+void
+log_archive_groups(void)
+/*====================*/
+{
+	log_group_t*	group;
+
+	ut_ad(mutex_own(&(log_sys->mutex)));
+
+	group = UT_LIST_GET_FIRST(log_sys->log_groups);
+
+	log_group_archive(group);
+}
+
+/*********************************************************
+Completes the archiving write phase for (each log group), currently,
+the first log group. */
+static
+void
+log_archive_write_complete_groups(void)
+/*===================================*/
+{
+	log_group_t*	group;
+	ulint		end_offset;
+	ulint		trunc_files;
+	ulint		n_files;
+	ib_uint64_t	start_lsn;
+	ib_uint64_t	end_lsn;
+	ulint		i;
+
+	ut_ad(mutex_own(&(log_sys->mutex)));
+
+	group = UT_LIST_GET_FIRST(log_sys->log_groups);
+
+	group->archived_file_no = group->next_archived_file_no;
+	group->archived_offset = group->next_archived_offset;
+
+	/* Truncate from the archive file space all but the last
+	file, or if it has been written full, all files */
+
+	n_files = (UNIV_PAGE_SIZE
+		   * fil_space_get_size(group->archive_space_id))
+		/ group->file_size;
+	ut_ad(n_files > 0);
+
+	end_offset = group->archived_offset;
+
+	if (end_offset % group->file_size == 0) {
+
+		trunc_files = n_files;
+	} else {
+		trunc_files = n_files - 1;
+	}
+
+#ifdef UNIV_DEBUG
+	if (log_debug_writes && trunc_files) {
+		fprintf(stderr,
+			"Complete file(s) archived to group %lu\n",
+			(ulong) group->id);
+	}
+#endif /* UNIV_DEBUG */
+
+	/* Calculate the archive file space start lsn */
+	start_lsn = log_sys->next_archived_lsn
+		- (end_offset - LOG_FILE_HDR_SIZE + trunc_files
+		   * (group->file_size - LOG_FILE_HDR_SIZE));
+	end_lsn = start_lsn;
+
+	for (i = 0; i < trunc_files; i++) {
+
+		end_lsn += group->file_size - LOG_FILE_HDR_SIZE;
+
+		/* Write a notice to the headers of archived log
+		files that the file write has been completed */
+
+		log_group_archive_completed_header_write(group, i, end_lsn);
+	}
+
+	fil_space_truncate_start(group->archive_space_id,
+				 trunc_files * group->file_size);
+
+#ifdef UNIV_DEBUG
+	if (log_debug_writes) {
+		fputs("Archiving writes completed\n", stderr);
+	}
+#endif /* UNIV_DEBUG */
+}
+
+/**********************************************************
+Completes an archiving i/o. */
+static
+void
+log_archive_check_completion_low(void)
+/*==================================*/
+{
+	ut_ad(mutex_own(&(log_sys->mutex)));
+
+	if (log_sys->n_pending_archive_ios == 0
+	    && log_sys->archiving_phase == LOG_ARCHIVE_READ) {
+
+#ifdef UNIV_DEBUG
+		if (log_debug_writes) {
+			fputs("Archiving read completed\n", stderr);
+		}
+#endif /* UNIV_DEBUG */
+
+		/* Archive buffer has now been read in: start archive writes */
+
+		log_sys->archiving_phase = LOG_ARCHIVE_WRITE;
+
+		log_archive_groups();
+	}
+
+	if (log_sys->n_pending_archive_ios == 0
+	    && log_sys->archiving_phase == LOG_ARCHIVE_WRITE) {
+
+		log_archive_write_complete_groups();
+
+		log_sys->archived_lsn = log_sys->next_archived_lsn;
+
+		rw_lock_x_unlock_gen(&(log_sys->archive_lock), LOG_ARCHIVE);
+	}
+}
+
+/**********************************************************
+Completes an archiving i/o. */
+static
+void
+log_io_complete_archive(void)
+/*=========================*/
+{
+	log_group_t*	group;
+
+	mutex_enter(&(log_sys->mutex));
+
+	group = UT_LIST_GET_FIRST(log_sys->log_groups);
+
+	mutex_exit(&(log_sys->mutex));
+
+	fil_flush(group->archive_space_id);
+
+	mutex_enter(&(log_sys->mutex));
+
+	ut_ad(log_sys->n_pending_archive_ios > 0);
+
+	log_sys->n_pending_archive_ios--;
+
+	log_archive_check_completion_low();
+
+	mutex_exit(&(log_sys->mutex));
+}
+
+/************************************************************************
+Starts an archiving operation. */
+UNIV_INTERN
+ibool
+log_archive_do(
+/*===========*/
+			/* out: TRUE if succeed, FALSE if an archiving
+			operation was already running */
+	ibool	sync,	/* in: TRUE if synchronous operation is desired */
+	ulint*	n_bytes)/* out: archive log buffer size, 0 if nothing to
+			archive */
+{
+	ibool		calc_new_limit;
+	ib_uint64_t	start_lsn;
+	ib_uint64_t	limit_lsn;
+
+	calc_new_limit = TRUE;
+loop:
+	mutex_enter(&(log_sys->mutex));
+
+	switch (log_sys->archiving_state) {
+	case LOG_ARCH_OFF:
+arch_none:
+		mutex_exit(&(log_sys->mutex));
+
+		*n_bytes = 0;
+
+		return(TRUE);
+	case LOG_ARCH_STOPPED:
+	case LOG_ARCH_STOPPING2:
+		mutex_exit(&(log_sys->mutex));
+
+		os_event_wait(log_sys->archiving_on);
+
+		goto loop;
+	}
+
+	start_lsn = log_sys->archived_lsn;
+
+	if (calc_new_limit) {
+		ut_a(log_sys->archive_buf_size % OS_FILE_LOG_BLOCK_SIZE == 0);
+		limit_lsn = start_lsn + log_sys->archive_buf_size;
+
+		*n_bytes = log_sys->archive_buf_size;
+
+		if (limit_lsn >= log_sys->lsn) {
+
+			limit_lsn = ut_uint64_align_down(
+				log_sys->lsn, OS_FILE_LOG_BLOCK_SIZE);
+		}
+	}
+
+	if (log_sys->archived_lsn >= limit_lsn) {
+
+		goto arch_none;
+	}
+
+	if (log_sys->written_to_all_lsn < limit_lsn) {
+
+		mutex_exit(&(log_sys->mutex));
+
+		log_write_up_to(limit_lsn, LOG_WAIT_ALL_GROUPS, TRUE);
+
+		calc_new_limit = FALSE;
+
+		goto loop;
+	}
+
+	if (log_sys->n_pending_archive_ios > 0) {
+		/* An archiving operation is running */
+
+		mutex_exit(&(log_sys->mutex));
+
+		if (sync) {
+			rw_lock_s_lock(&(log_sys->archive_lock));
+			rw_lock_s_unlock(&(log_sys->archive_lock));
+		}
+
+		*n_bytes = log_sys->archive_buf_size;
+
+		return(FALSE);
+	}
+
+	rw_lock_x_lock_gen(&(log_sys->archive_lock), LOG_ARCHIVE);
+
+	log_sys->archiving_phase = LOG_ARCHIVE_READ;
+
+	log_sys->next_archived_lsn = limit_lsn;
+
+#ifdef UNIV_DEBUG
+	if (log_debug_writes) {
+		fprintf(stderr,
+			"Archiving from lsn %llu to lsn %llu\n",
+			log_sys->archived_lsn, limit_lsn);
+	}
+#endif /* UNIV_DEBUG */
+
+	/* Read the log segment to the archive buffer */
+
+	log_group_read_log_seg(LOG_ARCHIVE, log_sys->archive_buf,
+			       UT_LIST_GET_FIRST(log_sys->log_groups),
+			       start_lsn, limit_lsn);
+
+	mutex_exit(&(log_sys->mutex));
+
+	if (sync) {
+		rw_lock_s_lock(&(log_sys->archive_lock));
+		rw_lock_s_unlock(&(log_sys->archive_lock));
+	}
+
+	*n_bytes = log_sys->archive_buf_size;
+
+	return(TRUE);
+}
+
+/********************************************************************
+Writes the log contents to the archive at least up to the lsn when this
+function was called. */
+static
+void
+log_archive_all(void)
+/*=================*/
+{
+	ib_uint64_t	present_lsn;
+	ulint		dummy;
+
+	mutex_enter(&(log_sys->mutex));
+
+	if (log_sys->archiving_state == LOG_ARCH_OFF) {
+		mutex_exit(&(log_sys->mutex));
+
+		return;
+	}
+
+	present_lsn = log_sys->lsn;
+
+	mutex_exit(&(log_sys->mutex));
+
+	log_pad_current_log_block();
+
+	for (;;) {
+		mutex_enter(&(log_sys->mutex));
+
+		if (present_lsn <= log_sys->archived_lsn) {
+
+			mutex_exit(&(log_sys->mutex));
+
+			return;
+		}
+
+		mutex_exit(&(log_sys->mutex));
+
+		log_archive_do(TRUE, &dummy);
+	}
+}
+
+/*********************************************************
+Closes the possible open archive log file (for each group) the first group,
+and if it was open, increments the group file count by 2, if desired. */
+static
+void
+log_archive_close_groups(
+/*=====================*/
+	ibool	increment_file_count)	/* in: TRUE if we want to increment
+					the file count */
+{
+	log_group_t*	group;
+	ulint		trunc_len;
+
+	ut_ad(mutex_own(&(log_sys->mutex)));
+
+	if (log_sys->archiving_state == LOG_ARCH_OFF) {
+
+		return;
+	}
+
+	group = UT_LIST_GET_FIRST(log_sys->log_groups);
+
+	trunc_len = UNIV_PAGE_SIZE
+		* fil_space_get_size(group->archive_space_id);
+	if (trunc_len > 0) {
+		ut_a(trunc_len == group->file_size);
+
+		/* Write a notice to the headers of archived log
+		files that the file write has been completed */
+
+		log_group_archive_completed_header_write(
+			group, 0, log_sys->archived_lsn);
+
+		fil_space_truncate_start(group->archive_space_id,
+					 trunc_len);
+		if (increment_file_count) {
+			group->archived_offset = 0;
+			group->archived_file_no += 2;
+		}
+
+#ifdef UNIV_DEBUG
+		if (log_debug_writes) {
+			fprintf(stderr,
+				"Incrementing arch file no to %lu"
+				" in log group %lu\n",
+				(ulong) group->archived_file_no + 2,
+				(ulong) group->id);
+		}
+#endif /* UNIV_DEBUG */
+	}
+}
+
+/********************************************************************
+Writes the log contents to the archive up to the lsn when this function was
+called, and stops the archiving. When archiving is started again, the archived
+log file numbers start from 2 higher, so that the archiving will not write
+again to the archived log files which exist when this function returns. */
+UNIV_INTERN
+ulint
+log_archive_stop(void)
+/*==================*/
+			/* out: DB_SUCCESS or DB_ERROR */
+{
+	ibool	success;
+
+	mutex_enter(&(log_sys->mutex));
+
+	if (log_sys->archiving_state != LOG_ARCH_ON) {
+
+		mutex_exit(&(log_sys->mutex));
+
+		return(DB_ERROR);
+	}
+
+	log_sys->archiving_state = LOG_ARCH_STOPPING;
+
+	mutex_exit(&(log_sys->mutex));
+
+	log_archive_all();
+
+	mutex_enter(&(log_sys->mutex));
+
+	log_sys->archiving_state = LOG_ARCH_STOPPING2;
+	os_event_reset(log_sys->archiving_on);
+
+	mutex_exit(&(log_sys->mutex));
+
+	/* Wait for a possible archiving operation to end */
+
+	rw_lock_s_lock(&(log_sys->archive_lock));
+	rw_lock_s_unlock(&(log_sys->archive_lock));
+
+	mutex_enter(&(log_sys->mutex));
+
+	/* Close all archived log files, incrementing the file count by 2,
+	if appropriate */
+
+	log_archive_close_groups(TRUE);
+
+	mutex_exit(&(log_sys->mutex));
+
+	/* Make a checkpoint, so that if recovery is needed, the file numbers
+	of new archived log files will start from the right value */
+
+	success = FALSE;
+
+	while (!success) {
+		success = log_checkpoint(TRUE, TRUE);
+	}
+
+	mutex_enter(&(log_sys->mutex));
+
+	log_sys->archiving_state = LOG_ARCH_STOPPED;
+
+	mutex_exit(&(log_sys->mutex));
+
+	return(DB_SUCCESS);
+}
+
+/********************************************************************
+Starts again archiving which has been stopped. */
+UNIV_INTERN
+ulint
+log_archive_start(void)
+/*===================*/
+			/* out: DB_SUCCESS or DB_ERROR */
+{
+	mutex_enter(&(log_sys->mutex));
+
+	if (log_sys->archiving_state != LOG_ARCH_STOPPED) {
+
+		mutex_exit(&(log_sys->mutex));
+
+		return(DB_ERROR);
+	}
+
+	log_sys->archiving_state = LOG_ARCH_ON;
+
+	os_event_set(log_sys->archiving_on);
+
+	mutex_exit(&(log_sys->mutex));
+
+	return(DB_SUCCESS);
+}
+
+/********************************************************************
+Stop archiving the log so that a gap may occur in the archived log files. */
+UNIV_INTERN
+ulint
+log_archive_noarchivelog(void)
+/*==========================*/
+			/* out: DB_SUCCESS or DB_ERROR */
+{
+loop:
+	mutex_enter(&(log_sys->mutex));
+
+	if (log_sys->archiving_state == LOG_ARCH_STOPPED
+	    || log_sys->archiving_state == LOG_ARCH_OFF) {
+
+		log_sys->archiving_state = LOG_ARCH_OFF;
+
+		os_event_set(log_sys->archiving_on);
+
+		mutex_exit(&(log_sys->mutex));
+
+		return(DB_SUCCESS);
+	}
+
+	mutex_exit(&(log_sys->mutex));
+
+	log_archive_stop();
+
+	os_thread_sleep(500000);
+
+	goto loop;
+}
+
+/********************************************************************
+Start archiving the log so that a gap may occur in the archived log files. */
+UNIV_INTERN
+ulint
+log_archive_archivelog(void)
+/*========================*/
+			/* out: DB_SUCCESS or DB_ERROR */
+{
+	mutex_enter(&(log_sys->mutex));
+
+	if (log_sys->archiving_state == LOG_ARCH_OFF) {
+
+		log_sys->archiving_state = LOG_ARCH_ON;
+
+		log_sys->archived_lsn
+			= ut_uint64_align_down(log_sys->lsn,
+					       OS_FILE_LOG_BLOCK_SIZE);
+		mutex_exit(&(log_sys->mutex));
+
+		return(DB_SUCCESS);
+	}
+
+	mutex_exit(&(log_sys->mutex));
+
+	return(DB_ERROR);
+}
+
+/********************************************************************
+Tries to establish a big enough margin of free space in the log groups, such
+that a new log entry can be catenated without an immediate need for
+archiving. */
+static
+void
+log_archive_margin(void)
+/*====================*/
+{
+	log_t*	log		= log_sys;
+	ulint	age;
+	ibool	sync;
+	ulint	dummy;
+loop:
+	mutex_enter(&(log->mutex));
+
+	if (log->archiving_state == LOG_ARCH_OFF) {
+		mutex_exit(&(log->mutex));
+
+		return;
+	}
+
+	age = log->lsn - log->archived_lsn;
+
+	if (age > log->max_archived_lsn_age) {
+
+		/* An archiving is urgent: we have to do synchronous i/o */
+
+		sync = TRUE;
+
+	} else if (age > log->max_archived_lsn_age_async) {
+
+		/* An archiving is not urgent: we do asynchronous i/o */
+
+		sync = FALSE;
+	} else {
+		/* No archiving required yet */
+
+		mutex_exit(&(log->mutex));
+
+		return;
+	}
+
+	mutex_exit(&(log->mutex));
+
+	log_archive_do(sync, &dummy);
+
+	if (sync == TRUE) {
+		/* Check again that enough was written to the archive */
+
+		goto loop;
+	}
+}
+#endif /* UNIV_LOG_ARCHIVE */
+
+/************************************************************************
+Checks that there is enough free space in the log to start a new query step.
+Flushes the log buffer or makes a new checkpoint if necessary. NOTE: this
+function may only be called if the calling thread owns no synchronization
+objects! */
+UNIV_INTERN
+void
+log_check_margins(void)
+/*===================*/
+{
+loop:
+	log_flush_margin();
+
+	log_checkpoint_margin();
+
+#ifdef UNIV_LOG_ARCHIVE
+	log_archive_margin();
+#endif /* UNIV_LOG_ARCHIVE */
+
+	mutex_enter(&(log_sys->mutex));
+
+	if (log_sys->check_flush_or_checkpoint) {
+
+		mutex_exit(&(log_sys->mutex));
+
+		goto loop;
+	}
+
+	mutex_exit(&(log_sys->mutex));
+}
+
+/********************************************************************
+Makes a checkpoint at the latest lsn and writes it to first page of each
+data file in the database, so that we know that the file spaces contain
+all modifications up to that lsn. This can only be called at database
+shutdown. This function also writes all log in log files to the log archive. */
+UNIV_INTERN
+void
+logs_empty_and_mark_files_at_shutdown(void)
+/*=======================================*/
+{
+	ib_uint64_t	lsn;
+	ulint		arch_log_no;
+
+	if (srv_print_verbose_log) {
+		ut_print_timestamp(stderr);
+		fprintf(stderr, "  InnoDB: Starting shutdown...\n");
+	}
+	/* Wait until the master thread and all other operations are idle: our
+	algorithm only works if the server is idle at shutdown */
+
+	srv_shutdown_state = SRV_SHUTDOWN_CLEANUP;
+loop:
+	os_thread_sleep(100000);
+
+	mutex_enter(&kernel_mutex);
+
+	/* We need the monitor threads to stop before we proceed with a
+	normal shutdown. In case of very fast shutdown, however, we can
+	proceed without waiting for monitor threads. */
+
+	if (srv_fast_shutdown < 2
+	   && (srv_error_monitor_active
+	      || srv_lock_timeout_and_monitor_active)) {
+
+		mutex_exit(&kernel_mutex);
+
+		goto loop;
+	}
+
+	/* Check that there are no longer transactions. We need this wait even
+	for the 'very fast' shutdown, because the InnoDB layer may have
+	committed or prepared transactions and we don't want to lose them. */
+
+	if (trx_n_mysql_transactions > 0
+	    || UT_LIST_GET_LEN(trx_sys->trx_list) > 0) {
+
+		mutex_exit(&kernel_mutex);
+
+		goto loop;
+	}
+
+	if (srv_fast_shutdown == 2) {
+		/* In this fastest shutdown we do not flush the buffer pool:
+		it is essentially a 'crash' of the InnoDB server. Make sure
+		that the log is all flushed to disk, so that we can recover
+		all committed transactions in a crash recovery. We must not
+		write the lsn stamps to the data files, since at a startup
+		InnoDB deduces from the stamps if the previous shutdown was
+		clean. */
+
+		log_buffer_flush_to_disk();
+
+		return; /* We SKIP ALL THE REST !! */
+	}
+
+	/* Check that the master thread is suspended */
+
+	if (srv_n_threads_active[SRV_MASTER] != 0) {
+
+		mutex_exit(&kernel_mutex);
+
+		goto loop;
+	}
+
+	mutex_exit(&kernel_mutex);
+
+	mutex_enter(&(log_sys->mutex));
+
+	if (log_sys->n_pending_checkpoint_writes
+#ifdef UNIV_LOG_ARCHIVE
+	    || log_sys->n_pending_archive_ios
+#endif /* UNIV_LOG_ARCHIVE */
+	    || log_sys->n_pending_writes) {
+
+		mutex_exit(&(log_sys->mutex));
+
+		goto loop;
+	}
+
+	mutex_exit(&(log_sys->mutex));
+
+	if (!buf_pool_check_no_pending_io()) {
+
+		goto loop;
+	}
+
+#ifdef UNIV_LOG_ARCHIVE
+	log_archive_all();
+#endif /* UNIV_LOG_ARCHIVE */
+
+	log_make_checkpoint_at(IB_ULONGLONG_MAX, TRUE);
+
+	mutex_enter(&(log_sys->mutex));
+
+	lsn = log_sys->lsn;
+
+	if (lsn != log_sys->last_checkpoint_lsn
+#ifdef UNIV_LOG_ARCHIVE
+	    || (srv_log_archive_on
+		&& lsn != log_sys->archived_lsn + LOG_BLOCK_HDR_SIZE)
+#endif /* UNIV_LOG_ARCHIVE */
+	    ) {
+
+		mutex_exit(&(log_sys->mutex));
+
+		goto loop;
+	}
+
+	arch_log_no = 0;
+
+#ifdef UNIV_LOG_ARCHIVE
+	UT_LIST_GET_FIRST(log_sys->log_groups)->archived_file_no;
+
+	if (0 == UT_LIST_GET_FIRST(log_sys->log_groups)->archived_offset) {
+
+		arch_log_no--;
+	}
+
+	log_archive_close_groups(TRUE);
+#endif /* UNIV_LOG_ARCHIVE */
+
+	mutex_exit(&(log_sys->mutex));
+
+	mutex_enter(&kernel_mutex);
+	/* Check that the master thread has stayed suspended */
+	if (srv_n_threads_active[SRV_MASTER] != 0) {
+		fprintf(stderr,
+			"InnoDB: Warning: the master thread woke up"
+			" during shutdown\n");
+
+		mutex_exit(&kernel_mutex);
+
+		goto loop;
+	}
+	mutex_exit(&kernel_mutex);
+
+	fil_flush_file_spaces(FIL_TABLESPACE);
+	fil_flush_file_spaces(FIL_LOG);
+
+	/* The call fil_write_flushed_lsn_to_data_files() will pass the buffer
+	pool: therefore it is essential that the buffer pool has been
+	completely flushed to disk! (We do not call fil_write... if the
+	'very fast' shutdown is enabled.) */
+
+	if (!buf_all_freed()) {
+
+		goto loop;
+	}
+
+	srv_shutdown_state = SRV_SHUTDOWN_LAST_PHASE;
+
+	/* Make some checks that the server really is quiet */
+	ut_a(srv_n_threads_active[SRV_MASTER] == 0);
+	ut_a(buf_all_freed());
+	ut_a(lsn == log_sys->lsn);
+
+	if (lsn < srv_start_lsn) {
+		fprintf(stderr,
+			"InnoDB: Error: log sequence number"
+			" at shutdown %llu\n"
+			"InnoDB: is lower than at startup %llu!\n",
+			lsn, srv_start_lsn);
+	}
+
+	srv_shutdown_lsn = lsn;
+
+	fil_write_flushed_lsn_to_data_files(lsn, arch_log_no);
+
+	fil_flush_file_spaces(FIL_TABLESPACE);
+
+	fil_close_all_files();
+
+	/* Make some checks that the server really is quiet */
+	ut_a(srv_n_threads_active[SRV_MASTER] == 0);
+	ut_a(buf_all_freed());
+	ut_a(lsn == log_sys->lsn);
+}
+
+/**********************************************************
+Checks by parsing that the catenated log segment for a single mtr is
+consistent. */
+UNIV_INTERN
+ibool
+log_check_log_recs(
+/*===============*/
+	byte*		buf,		/* in: pointer to the start of
+					the log segment in the
+					log_sys->buf log buffer */
+	ulint		len,		/* in: segment length in bytes */
+	ib_uint64_t	buf_start_lsn)	/* in: buffer start lsn */
+{
+	ib_uint64_t	contiguous_lsn;
+	ib_uint64_t	scanned_lsn;
+	byte*		start;
+	byte*		end;
+	byte*		buf1;
+	byte*		scan_buf;
+
+	ut_ad(mutex_own(&(log_sys->mutex)));
+
+	if (len == 0) {
+
+		return(TRUE);
+	}
+
+	start = ut_align_down(buf, OS_FILE_LOG_BLOCK_SIZE);
+	end = ut_align(buf + len, OS_FILE_LOG_BLOCK_SIZE);
+
+	buf1 = mem_alloc((end - start) + OS_FILE_LOG_BLOCK_SIZE);
+	scan_buf = ut_align(buf1, OS_FILE_LOG_BLOCK_SIZE);
+
+	ut_memcpy(scan_buf, start, end - start);
+
+	recv_scan_log_recs(TRUE,
+			   (buf_pool->curr_size
+			    - recv_n_pool_free_frames) * UNIV_PAGE_SIZE,
+			   FALSE, scan_buf, end - start,
+			   ut_uint64_align_down(buf_start_lsn,
+						OS_FILE_LOG_BLOCK_SIZE),
+			   &contiguous_lsn, &scanned_lsn);
+
+	ut_a(scanned_lsn == buf_start_lsn + len);
+	ut_a(recv_sys->recovered_lsn == scanned_lsn);
+
+	mem_free(buf1);
+
+	return(TRUE);
+}
+
+/**********************************************************
+Peeks the current lsn. */
+UNIV_INTERN
+ibool
+log_peek_lsn(
+/*=========*/
+				/* out: TRUE if success, FALSE if
+				could not get the log system mutex */
+	ib_uint64_t*	lsn)	/* out: if returns TRUE, current lsn is here */
+{
+	if (0 == mutex_enter_nowait(&(log_sys->mutex))) {
+		*lsn = log_sys->lsn;
+
+		mutex_exit(&(log_sys->mutex));
+
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
+
+/**********************************************************
+Prints info of the log. */
+UNIV_INTERN
+void
+log_print(
+/*======*/
+	FILE*	file)	/* in: file where to print */
+{
+	double	time_elapsed;
+	time_t	current_time;
+
+	mutex_enter(&(log_sys->mutex));
+
+	fprintf(file,
+		"Log sequence number %llu\n"
+		"Log flushed up to   %llu\n"
+		"Last checkpoint at  %llu\n",
+		log_sys->lsn,
+		log_sys->flushed_to_disk_lsn,
+		log_sys->last_checkpoint_lsn);
+
+	fprintf(file,
+		"Max checkpoint age  %lu\n"
+		"Modified age        %lu\n"
+		"Checkpoint age      %lu\n",
+			(ulong) log_sys->max_checkpoint_age,
+			(ulong) (log_sys->lsn -
+					log_buf_pool_get_oldest_modification()),
+			(ulong) (log_sys->lsn - log_sys->last_checkpoint_lsn));
+
+	current_time = time(NULL);
+
+	time_elapsed = 0.001 + difftime(current_time,
+					log_sys->last_printout_time);
+	fprintf(file,
+		"%lu pending log writes, %lu pending chkp writes\n"
+		"%lu log i/o's done, %.2f log i/o's/second\n",
+		(ulong) log_sys->n_pending_writes,
+		(ulong) log_sys->n_pending_checkpoint_writes,
+		(ulong) log_sys->n_log_ios,
+		((log_sys->n_log_ios - log_sys->n_log_ios_old)
+		 / time_elapsed));
+
+	log_sys->n_log_ios_old = log_sys->n_log_ios;
+	log_sys->last_printout_time = current_time;
+
+	mutex_exit(&(log_sys->mutex));
+}
+
+/**************************************************************************
+Refreshes the statistics used to print per-second averages. */
+UNIV_INTERN
+void
+log_refresh_stats(void)
+/*===================*/
+{
+	log_sys->n_log_ios_old = log_sys->n_log_ios;
+	log_sys->last_printout_time = time(NULL);
+}
diff --git a/storage/xtradb/log/log0recv.c b/storage/xtradb/log/log0recv.c
new file mode 100644
index 00000000000..b72dde4efcf
--- /dev/null
+++ b/storage/xtradb/log/log0recv.c
@@ -0,0 +1,3476 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Recovery
+
+Created 9/20/1997 Heikki Tuuri
+*******************************************************/
+
+#include "log0recv.h"
+
+#ifdef UNIV_NONINL
+#include "log0recv.ic"
+#endif
+
+#include "mem0mem.h"
+#include "buf0buf.h"
+#include "buf0flu.h"
+#include "buf0rea.h"
+#include "srv0srv.h"
+#include "srv0start.h"
+#include "mtr0log.h"
+#include "page0cur.h"
+#include "page0zip.h"
+#include "btr0cur.h"
+#include "ibuf0ibuf.h"
+#include "trx0undo.h"
+#include "trx0rec.h"
+#include "trx0roll.h"
+#include "row0merge.h"
+
+#ifdef UNIV_HOTBACKUP
+/* This is set to FALSE if the backup was originally taken with the
+ibbackup --include regexp option: then we do not want to create tables in
+directories which were not included */
+UNIV_INTERN ibool	recv_replay_file_ops	= TRUE;
+#endif /* UNIV_HOTBACKUP */
+
+/* Log records are stored in the hash table in chunks at most of this size;
+this must be less than UNIV_PAGE_SIZE as it is stored in the buffer pool */
+#define RECV_DATA_BLOCK_SIZE	(MEM_MAX_ALLOC_IN_BUF - sizeof(recv_data_t))
+
+/* Read-ahead area in applying log records to file pages */
+#define RECV_READ_AHEAD_AREA	32
+
+UNIV_INTERN recv_sys_t*	recv_sys = NULL;
+UNIV_INTERN ibool	recv_recovery_on = FALSE;
+UNIV_INTERN ibool	recv_recovery_from_backup_on = FALSE;
+
+UNIV_INTERN ibool	recv_needed_recovery = FALSE;
+
+UNIV_INTERN ibool	recv_lsn_checks_on = FALSE;
+
+/* There are two conditions under which we scan the logs, the first
+is normal startup and the second is when we do a recovery from an
+archive.
+This flag is set if we are doing a scan from the last checkpoint during
+startup. If we find log entries that were written after the last checkpoint
+we know that the server was not cleanly shutdown. We must then initialize
+the crash recovery environment before attempting to store these entries in
+the log hash table. */
+UNIV_INTERN ibool	recv_log_scan_is_startup_type = FALSE;
+
+/* If the following is TRUE, the buffer pool file pages must be invalidated
+after recovery and no ibuf operations are allowed; this becomes TRUE if
+the log record hash table becomes too full, and log records must be merged
+to file pages already before the recovery is finished: in this case no
+ibuf operations are allowed, as they could modify the pages read in the
+buffer pool before the pages have been recovered to the up-to-date state */
+
+/* Recovery is running and no operations on the log files are allowed
+yet: the variable name is misleading */
+
+UNIV_INTERN ibool	recv_no_ibuf_operations = FALSE;
+
+/* The following counter is used to decide when to print info on
+log scan */
+UNIV_INTERN ulint	recv_scan_print_counter	= 0;
+
+UNIV_INTERN ibool	recv_is_from_backup	= FALSE;
+#ifdef UNIV_HOTBACKUP
+UNIV_INTERN ibool	recv_is_making_a_backup = FALSE;
+#else
+# define recv_is_making_a_backup FALSE
+#endif /* UNIV_HOTBACKUP */
+
+UNIV_INTERN ulint	recv_previous_parsed_rec_type	= 999999;
+UNIV_INTERN ulint	recv_previous_parsed_rec_offset	= 0;
+UNIV_INTERN ulint	recv_previous_parsed_rec_is_multi = 0;
+
+UNIV_INTERN ulint	recv_max_parsed_page_no		= 0;
+
+/* This many frames must be left free in the buffer pool when we scan
+the log and store the scanned log records in the buffer pool: we will
+use these free frames to read in pages when we start applying the
+log records to the database. */
+
+UNIV_INTERN ulint	recv_n_pool_free_frames		= 256;
+
+/* The maximum lsn we see for a page during the recovery process. If this
+is bigger than the lsn we are able to scan up to, that is an indication that
+the recovery failed and the database may be corrupt. */
+
+UNIV_INTERN ib_uint64_t	recv_max_page_lsn;
+
+/* prototypes */
+
+/***********************************************************
+Initialize crash recovery environment. Can be called iff
+recv_needed_recovery == FALSE. */
+static
+void
+recv_init_crash_recovery(void);
+/*===========================*/
+
+/************************************************************
+Creates the recovery system. */
+UNIV_INTERN
+void
+recv_sys_create(void)
+/*=================*/
+{
+	if (recv_sys != NULL) {
+
+		return;
+	}
+
+	recv_sys = mem_alloc(sizeof(recv_sys_t));
+
+	mutex_create(&recv_sys->mutex, SYNC_RECV);
+
+	recv_sys->heap = NULL;
+	recv_sys->addr_hash = NULL;
+}
+
+/************************************************************
+Inits the recovery system for a recovery operation. */
+UNIV_INTERN
+void
+recv_sys_init(
+/*==========*/
+	ibool	recover_from_backup,	/* in: TRUE if this is called
+					to recover from a hot backup */
+	ulint	available_memory)	/* in: available memory in bytes */
+{
+	if (recv_sys->heap != NULL) {
+
+		return;
+	}
+
+	mutex_enter(&(recv_sys->mutex));
+
+	if (!recover_from_backup) {
+		recv_sys->heap = mem_heap_create_in_buffer(256);
+	} else {
+		recv_sys->heap = mem_heap_create(256);
+		recv_is_from_backup = TRUE;
+	}
+
+	recv_sys->buf = ut_malloc(RECV_PARSING_BUF_SIZE);
+	recv_sys->len = 0;
+	recv_sys->recovered_offset = 0;
+
+	recv_sys->addr_hash = hash_create(available_memory / 64);
+	recv_sys->n_addrs = 0;
+
+	recv_sys->apply_log_recs = FALSE;
+	recv_sys->apply_batch_on = FALSE;
+
+	recv_sys->last_block_buf_start = mem_alloc(2 * OS_FILE_LOG_BLOCK_SIZE);
+
+	recv_sys->last_block = ut_align(recv_sys->last_block_buf_start,
+					OS_FILE_LOG_BLOCK_SIZE);
+	recv_sys->found_corrupt_log = FALSE;
+
+	recv_max_page_lsn = 0;
+
+	mutex_exit(&(recv_sys->mutex));
+}
+
+/************************************************************
+Empties the hash table when it has been fully processed. */
+static
+void
+recv_sys_empty_hash(void)
+/*=====================*/
+{
+	ut_ad(mutex_own(&(recv_sys->mutex)));
+
+	if (recv_sys->n_addrs != 0) {
+		fprintf(stderr,
+			"InnoDB: Error: %lu pages with log records"
+			" were left unprocessed!\n"
+			"InnoDB: Maximum page number with"
+			" log records on it %lu\n",
+			(ulong) recv_sys->n_addrs,
+			(ulong) recv_max_parsed_page_no);
+		ut_error;
+	}
+
+	hash_table_free(recv_sys->addr_hash);
+	mem_heap_empty(recv_sys->heap);
+
+	recv_sys->addr_hash = hash_create(buf_pool_get_curr_size() / 256);
+}
+
+#ifndef UNIV_LOG_DEBUG
+/************************************************************
+Frees the recovery system. */
+static
+void
+recv_sys_free(void)
+/*===============*/
+{
+	mutex_enter(&(recv_sys->mutex));
+
+	hash_table_free(recv_sys->addr_hash);
+	mem_heap_free(recv_sys->heap);
+	ut_free(recv_sys->buf);
+	mem_free(recv_sys->last_block_buf_start);
+
+	recv_sys->addr_hash = NULL;
+	recv_sys->heap = NULL;
+
+	mutex_exit(&(recv_sys->mutex));
+}
+#endif /* UNIV_LOG_DEBUG */
+
+/************************************************************
+Truncates possible corrupted or extra records from a log group. */
+static
+void
+recv_truncate_group(
+/*================*/
+	log_group_t*	group,		/* in: log group */
+	ib_uint64_t	recovered_lsn,	/* in: recovery succeeded up to this
+					lsn */
+	ib_uint64_t	limit_lsn,	/* in: this was the limit for
+					recovery */
+	ib_uint64_t	checkpoint_lsn,	/* in: recovery was started from this
+					checkpoint */
+	ib_uint64_t	archived_lsn)	/* in: the log has been archived up to
+					this lsn */
+{
+	ib_uint64_t	start_lsn;
+	ib_uint64_t	end_lsn;
+	ib_uint64_t	finish_lsn1;
+	ib_uint64_t	finish_lsn2;
+	ib_uint64_t	finish_lsn;
+	ulint		len;
+	ulint		i;
+
+	if (archived_lsn == IB_ULONGLONG_MAX) {
+		/* Checkpoint was taken in the NOARCHIVELOG mode */
+		archived_lsn = checkpoint_lsn;
+	}
+
+	finish_lsn1 = ut_uint64_align_down(archived_lsn,
+					   OS_FILE_LOG_BLOCK_SIZE)
+		+ log_group_get_capacity(group);
+
+	finish_lsn2 = ut_uint64_align_up(recovered_lsn,
+					 OS_FILE_LOG_BLOCK_SIZE)
+		+ recv_sys->last_log_buf_size;
+
+	if (limit_lsn != IB_ULONGLONG_MAX) {
+		/* We do not know how far we should erase log records: erase
+		as much as possible */
+
+		finish_lsn = finish_lsn1;
+	} else {
+		/* It is enough to erase the length of the log buffer */
+		finish_lsn = finish_lsn1 < finish_lsn2
+			? finish_lsn1 : finish_lsn2;
+	}
+
+	ut_a(RECV_SCAN_SIZE <= log_sys->buf_size);
+
+	/* Write the log buffer full of zeros */
+	for (i = 0; i < RECV_SCAN_SIZE; i++) {
+
+		*(log_sys->buf + i) = '\0';
+	}
+
+	start_lsn = ut_uint64_align_down(recovered_lsn,
+					 OS_FILE_LOG_BLOCK_SIZE);
+
+	if (start_lsn != recovered_lsn) {
+		/* Copy the last incomplete log block to the log buffer and
+		edit its data length: */
+
+		ut_memcpy(log_sys->buf, recv_sys->last_block,
+			  OS_FILE_LOG_BLOCK_SIZE);
+		log_block_set_data_len(log_sys->buf,
+				       (ulint) (recovered_lsn - start_lsn));
+	}
+
+	if (start_lsn >= finish_lsn) {
+
+		return;
+	}
+
+	for (;;) {
+		end_lsn = start_lsn + RECV_SCAN_SIZE;
+
+		if (end_lsn > finish_lsn) {
+
+			end_lsn = finish_lsn;
+		}
+
+		len = (ulint) (end_lsn - start_lsn);
+
+		log_group_write_buf(group, log_sys->buf, len, start_lsn, 0);
+		if (end_lsn >= finish_lsn) {
+
+			return;
+		}
+
+		/* Write the log buffer full of zeros */
+		for (i = 0; i < RECV_SCAN_SIZE; i++) {
+
+			*(log_sys->buf + i) = '\0';
+		}
+
+		start_lsn = end_lsn;
+	}
+}
+
+/************************************************************
+Copies the log segment between group->recovered_lsn and recovered_lsn from the
+most up-to-date log group to group, so that it contains the latest log data. */
+static
+void
+recv_copy_group(
+/*============*/
+	log_group_t*	up_to_date_group,	/* in: the most up-to-date log
+						group */
+	log_group_t*	group,			/* in: copy to this log
+						group */
+	ib_uint64_t	recovered_lsn)		/* in: recovery succeeded up
+						to this lsn */
+{
+	ib_uint64_t	start_lsn;
+	ib_uint64_t	end_lsn;
+	ulint		len;
+
+	if (group->scanned_lsn >= recovered_lsn) {
+
+		return;
+	}
+
+	ut_a(RECV_SCAN_SIZE <= log_sys->buf_size);
+
+	start_lsn = ut_uint64_align_down(group->scanned_lsn,
+					 OS_FILE_LOG_BLOCK_SIZE);
+	for (;;) {
+		end_lsn = start_lsn + RECV_SCAN_SIZE;
+
+		if (end_lsn > recovered_lsn) {
+			end_lsn = ut_uint64_align_up(recovered_lsn,
+						     OS_FILE_LOG_BLOCK_SIZE);
+		}
+
+		log_group_read_log_seg(LOG_RECOVER, log_sys->buf,
+				       up_to_date_group, start_lsn, end_lsn);
+
+		len = (ulint) (end_lsn - start_lsn);
+
+		log_group_write_buf(group, log_sys->buf, len, start_lsn, 0);
+
+		if (end_lsn >= recovered_lsn) {
+
+			return;
+		}
+
+		start_lsn = end_lsn;
+	}
+}
+
+/************************************************************
+Copies a log segment from the most up-to-date log group to the other log
+groups, so that they all contain the latest log data. Also writes the info
+about the latest checkpoint to the groups, and inits the fields in the group
+memory structs to up-to-date values. */
+static
+void
+recv_synchronize_groups(
+/*====================*/
+	log_group_t*	up_to_date_group)	/* in: the most up-to-date
+						log group */
+{
+	log_group_t*	group;
+	ib_uint64_t	start_lsn;
+	ib_uint64_t	end_lsn;
+	ib_uint64_t	recovered_lsn;
+	ib_uint64_t	limit_lsn;
+
+	recovered_lsn = recv_sys->recovered_lsn;
+	limit_lsn = recv_sys->limit_lsn;
+
+	/* Read the last recovered log block to the recovery system buffer:
+	the block is always incomplete */
+
+	start_lsn = ut_uint64_align_down(recovered_lsn,
+					 OS_FILE_LOG_BLOCK_SIZE);
+	end_lsn = ut_uint64_align_up(recovered_lsn, OS_FILE_LOG_BLOCK_SIZE);
+
+	ut_a(start_lsn != end_lsn);
+
+	log_group_read_log_seg(LOG_RECOVER, recv_sys->last_block,
+			       up_to_date_group, start_lsn, end_lsn);
+
+	group = UT_LIST_GET_FIRST(log_sys->log_groups);
+
+	while (group) {
+		if (group != up_to_date_group) {
+
+			/* Copy log data if needed */
+
+			recv_copy_group(group, up_to_date_group,
+					recovered_lsn);
+		}
+
+		/* Update the fields in the group struct to correspond to
+		recovered_lsn */
+
+		log_group_set_fields(group, recovered_lsn);
+
+		group = UT_LIST_GET_NEXT(log_groups, group);
+	}
+
+	/* Copy the checkpoint info to the groups; remember that we have
+	incremented checkpoint_no by one, and the info will not be written
+	over the max checkpoint info, thus making the preservation of max
+	checkpoint info on disk certain */
+
+	log_groups_write_checkpoint_info();
+
+	mutex_exit(&(log_sys->mutex));
+
+	/* Wait for the checkpoint write to complete */
+	rw_lock_s_lock(&(log_sys->checkpoint_lock));
+	rw_lock_s_unlock(&(log_sys->checkpoint_lock));
+
+	mutex_enter(&(log_sys->mutex));
+}
+
+/***************************************************************************
+Checks the consistency of the checkpoint info */
+static
+ibool
+recv_check_cp_is_consistent(
+/*========================*/
+			/* out: TRUE if ok */
+	byte*	buf)	/* in: buffer containing checkpoint info */
+{
+	ulint	fold;
+
+	fold = ut_fold_binary(buf, LOG_CHECKPOINT_CHECKSUM_1);
+
+	if ((fold & 0xFFFFFFFFUL) != mach_read_from_4(
+		    buf + LOG_CHECKPOINT_CHECKSUM_1)) {
+		return(FALSE);
+	}
+
+	fold = ut_fold_binary(buf + LOG_CHECKPOINT_LSN,
+			      LOG_CHECKPOINT_CHECKSUM_2 - LOG_CHECKPOINT_LSN);
+
+	if ((fold & 0xFFFFFFFFUL) != mach_read_from_4(
+		    buf + LOG_CHECKPOINT_CHECKSUM_2)) {
+		return(FALSE);
+	}
+
+	return(TRUE);
+}
+
+/************************************************************
+Looks for the maximum consistent checkpoint from the log groups. */
+static
+ulint
+recv_find_max_checkpoint(
+/*=====================*/
+					/* out: error code or DB_SUCCESS */
+	log_group_t**	max_group,	/* out: max group */
+	ulint*		max_field)	/* out: LOG_CHECKPOINT_1 or
+					LOG_CHECKPOINT_2 */
+{
+	log_group_t*	group;
+	ib_uint64_t	max_no;
+	ib_uint64_t	checkpoint_no;
+	ulint		field;
+	byte*		buf;
+
+	group = UT_LIST_GET_FIRST(log_sys->log_groups);
+
+	max_no = 0;
+	*max_group = NULL;
+	*max_field = 0;
+
+	buf = log_sys->checkpoint_buf;
+
+	while (group) {
+		group->state = LOG_GROUP_CORRUPTED;
+
+		for (field = LOG_CHECKPOINT_1; field <= LOG_CHECKPOINT_2;
+		     field += LOG_CHECKPOINT_2 - LOG_CHECKPOINT_1) {
+
+			log_group_read_checkpoint_info(group, field);
+
+			if (!recv_check_cp_is_consistent(buf)) {
+#ifdef UNIV_DEBUG
+				if (log_debug_writes) {
+					fprintf(stderr,
+						"InnoDB: Checkpoint in group"
+						" %lu at %lu invalid, %lu\n",
+						(ulong) group->id,
+						(ulong) field,
+						(ulong) mach_read_from_4(
+							buf
+							+ LOG_CHECKPOINT_CHECKSUM_1));
+
+				}
+#endif /* UNIV_DEBUG */
+				goto not_consistent;
+			}
+
+			group->state = LOG_GROUP_OK;
+
+			group->lsn = mach_read_ull(
+				buf + LOG_CHECKPOINT_LSN);
+			group->lsn_offset = mach_read_from_4(
+				buf + LOG_CHECKPOINT_OFFSET);
+			checkpoint_no = mach_read_ull(
+				buf + LOG_CHECKPOINT_NO);
+
+#ifdef UNIV_DEBUG
+			if (log_debug_writes) {
+				fprintf(stderr,
+					"InnoDB: Checkpoint number %lu"
+					" found in group %lu\n",
+					(ulong) checkpoint_no,
+					(ulong) group->id);
+			}
+#endif /* UNIV_DEBUG */
+
+			if (checkpoint_no >= max_no) {
+				*max_group = group;
+				*max_field = field;
+				max_no = checkpoint_no;
+			}
+
+not_consistent:
+			;
+		}
+
+		group = UT_LIST_GET_NEXT(log_groups, group);
+	}
+
+	if (*max_group == NULL) {
+
+		fprintf(stderr,
+			"InnoDB: No valid checkpoint found.\n"
+			"InnoDB: If this error appears when you are"
+			" creating an InnoDB database,\n"
+			"InnoDB: the problem may be that during"
+			" an earlier attempt you managed\n"
+			"InnoDB: to create the InnoDB data files,"
+			" but log file creation failed.\n"
+			"InnoDB: If that is the case, please refer to\n"
+			"InnoDB: http://dev.mysql.com/doc/refman/5.1/en/"
+			"error-creating-innodb.html\n");
+		return(DB_ERROR);
+	}
+
+	return(DB_SUCCESS);
+}
+
+#ifdef UNIV_HOTBACKUP
+/***********************************************************************
+Reads the checkpoint info needed in hot backup. */
+UNIV_INTERN
+ibool
+recv_read_cp_info_for_backup(
+/*=========================*/
+				/* out: TRUE if success */
+	byte*		hdr,	/* in: buffer containing the log group
+				header */
+	ib_uint64_t*	lsn,	/* out: checkpoint lsn */
+	ulint*		offset,	/* out: checkpoint offset in the log group */
+	ulint*		fsp_limit,/* out: fsp limit of space 0,
+				1000000000 if the database is running
+				with < version 3.23.50 of InnoDB */
+	ib_uint64_t*	cp_no,	/* out: checkpoint number */
+	ib_uint64_t*	first_header_lsn)
+				/* out: lsn of of the start of the
+				first log file */
+{
+	ulint		max_cp		= 0;
+	ib_uint64_t	max_cp_no	= 0;
+	byte*		cp_buf;
+
+	cp_buf = hdr + LOG_CHECKPOINT_1;
+
+	if (recv_check_cp_is_consistent(cp_buf)) {
+		max_cp_no = mach_read_ull(cp_buf + LOG_CHECKPOINT_NO);
+		max_cp = LOG_CHECKPOINT_1;
+	}
+
+	cp_buf = hdr + LOG_CHECKPOINT_2;
+
+	if (recv_check_cp_is_consistent(cp_buf)) {
+		if (mach_read_ull(cp_buf + LOG_CHECKPOINT_NO) > max_cp_no) {
+			max_cp = LOG_CHECKPOINT_2;
+		}
+	}
+
+	if (max_cp == 0) {
+		return(FALSE);
+	}
+
+	cp_buf = hdr + max_cp;
+
+	*lsn = mach_read_ull(cp_buf + LOG_CHECKPOINT_LSN);
+	*offset = mach_read_from_4(cp_buf + LOG_CHECKPOINT_OFFSET);
+
+	/* If the user is running a pre-3.23.50 version of InnoDB, its
+	checkpoint data does not contain the fsp limit info */
+	if (mach_read_from_4(cp_buf + LOG_CHECKPOINT_FSP_MAGIC_N)
+	    == LOG_CHECKPOINT_FSP_MAGIC_N_VAL) {
+
+		*fsp_limit = mach_read_from_4(
+			cp_buf + LOG_CHECKPOINT_FSP_FREE_LIMIT);
+
+		if (*fsp_limit == 0) {
+			*fsp_limit = 1000000000;
+		}
+	} else {
+		*fsp_limit = 1000000000;
+	}
+
+	/*	fprintf(stderr, "fsp limit %lu MB\n", *fsp_limit); */
+
+	*cp_no = mach_read_ull(cp_buf + LOG_CHECKPOINT_NO);
+
+	*first_header_lsn = mach_read_ull(hdr + LOG_FILE_START_LSN);
+
+	return(TRUE);
+}
+#endif /* UNIV_HOTBACKUP */
+
+/**********************************************************
+Checks the 4-byte checksum to the trailer checksum field of a log block.
+We also accept a log block in the old format < InnoDB-3.23.52 where the
+checksum field contains the log block number. */
+static
+ibool
+log_block_checksum_is_ok_or_old_format(
+/*===================================*/
+			/* out: TRUE if ok, or if the log block may be in the
+			format of InnoDB version < 3.23.52 */
+	byte*	block)	/* in: pointer to a log block */
+{
+#ifdef UNIV_LOG_DEBUG
+	return(TRUE);
+#endif /* UNIV_LOG_DEBUG */
+	if (log_block_calc_checksum(block) == log_block_get_checksum(block)) {
+
+		return(TRUE);
+	}
+
+	if (log_block_get_hdr_no(block) == log_block_get_checksum(block)) {
+
+		/* We assume the log block is in the format of
+		InnoDB version < 3.23.52 and the block is ok */
+#if 0
+		fprintf(stderr,
+			"InnoDB: Scanned old format < InnoDB-3.23.52"
+			" log block number %lu\n",
+			log_block_get_hdr_no(block));
+#endif
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
+
+#ifdef UNIV_HOTBACKUP
+/***********************************************************************
+Scans the log segment and n_bytes_scanned is set to the length of valid
+log scanned. */
+UNIV_INTERN
+void
+recv_scan_log_seg_for_backup(
+/*=========================*/
+	byte*		buf,		/* in: buffer containing log data */
+	ulint		buf_len,	/* in: data length in that buffer */
+	ib_uint64_t*	scanned_lsn,	/* in/out: lsn of buffer start,
+					we return scanned lsn */
+	ulint*		scanned_checkpoint_no,
+					/* in/out: 4 lowest bytes of the
+					highest scanned checkpoint number so
+					far */
+	ulint*		n_bytes_scanned)/* out: how much we were able to
+					scan, smaller than buf_len if log
+					data ended here */
+{
+	ulint	data_len;
+	byte*	log_block;
+	ulint	no;
+
+	*n_bytes_scanned = 0;
+
+	for (log_block = buf; log_block < buf + buf_len;
+	     log_block += OS_FILE_LOG_BLOCK_SIZE) {
+
+		no = log_block_get_hdr_no(log_block);
+
+#if 0
+		fprintf(stderr, "Log block header no %lu\n", no);
+#endif
+
+		if (no != log_block_convert_lsn_to_no(*scanned_lsn)
+		    || !log_block_checksum_is_ok_or_old_format(log_block)) {
+#if 0
+			fprintf(stderr,
+				"Log block n:o %lu, scanned lsn n:o %lu\n",
+				no, log_block_convert_lsn_to_no(*scanned_lsn));
+#endif
+			/* Garbage or an incompletely written log block */
+
+			log_block += OS_FILE_LOG_BLOCK_SIZE;
+#if 0
+			fprintf(stderr,
+				"Next log block n:o %lu\n",
+				log_block_get_hdr_no(log_block));
+#endif
+			break;
+		}
+
+		if (*scanned_checkpoint_no > 0
+		    && log_block_get_checkpoint_no(log_block)
+		    < *scanned_checkpoint_no
+		    && *scanned_checkpoint_no
+		    - log_block_get_checkpoint_no(log_block)
+		    > 0x80000000UL) {
+
+			/* Garbage from a log buffer flush which was made
+			before the most recent database recovery */
+#if 0
+			fprintf(stderr,
+				"Scanned cp n:o %lu, block cp n:o %lu\n",
+				*scanned_checkpoint_no,
+				log_block_get_checkpoint_no(log_block));
+#endif
+			break;
+		}
+
+		data_len = log_block_get_data_len(log_block);
+
+		*scanned_checkpoint_no
+			= log_block_get_checkpoint_no(log_block);
+		*scanned_lsn += data_len;
+
+		*n_bytes_scanned += data_len;
+
+		if (data_len < OS_FILE_LOG_BLOCK_SIZE) {
+			/* Log data ends here */
+
+#if 0
+			fprintf(stderr, "Log block data len %lu\n",
+				data_len);
+#endif
+			break;
+		}
+	}
+}
+#endif /* UNIV_HOTBACKUP */
+
+/***********************************************************************
+Tries to parse a single log record body and also applies it to a page if
+specified. File ops are parsed, but not applied in this function. */
+static
+byte*
+recv_parse_or_apply_log_rec_body(
+/*=============================*/
+				/* out: log record end, NULL if not a
+				complete record */
+	byte		type,	/* in: type */
+	byte*		ptr,	/* in: pointer to a buffer */
+	byte*		end_ptr,/* in: pointer to the buffer end */
+	buf_block_t*	block,	/* in/out: buffer block or NULL; if
+				not NULL, then the log record is
+				applied to the page, and the log
+				record should be complete then */
+	mtr_t*		mtr)	/* in: mtr or NULL; should be non-NULL
+				if and only if block is non-NULL */
+{
+	dict_index_t*	index	= NULL;
+	page_t*		page;
+	page_zip_des_t*	page_zip;
+
+	ut_ad(!block == !mtr);
+
+	if (block) {
+		page = block->frame;
+		page_zip = buf_block_get_page_zip(block);
+	} else {
+		page = NULL;
+		page_zip = NULL;
+	}
+
+	switch (type) {
+	case MLOG_1BYTE: case MLOG_2BYTES: case MLOG_4BYTES: case MLOG_8BYTES:
+		ptr = mlog_parse_nbytes(type, ptr, end_ptr, page, page_zip);
+		break;
+	case MLOG_REC_INSERT: case MLOG_COMP_REC_INSERT:
+		if (NULL != (ptr = mlog_parse_index(
+				     ptr, end_ptr,
+				     type == MLOG_COMP_REC_INSERT,
+				     &index))) {
+			ut_a(!page
+			     || (ibool)!!page_is_comp(page)
+			     == dict_table_is_comp(index->table));
+			ptr = page_cur_parse_insert_rec(FALSE, ptr, end_ptr,
+							block, index, mtr);
+		}
+		break;
+	case MLOG_REC_CLUST_DELETE_MARK: case MLOG_COMP_REC_CLUST_DELETE_MARK:
+		if (NULL != (ptr = mlog_parse_index(
+				     ptr, end_ptr,
+				     type == MLOG_COMP_REC_CLUST_DELETE_MARK,
+				     &index))) {
+			ut_a(!page
+			     || (ibool)!!page_is_comp(page)
+			     == dict_table_is_comp(index->table));
+			ptr = btr_cur_parse_del_mark_set_clust_rec(
+				ptr, end_ptr, page, page_zip, index);
+		}
+		break;
+	case MLOG_COMP_REC_SEC_DELETE_MARK:
+		/* This log record type is obsolete, but we process it for
+		backward compatibility with MySQL 5.0.3 and 5.0.4. */
+		ut_a(!page || page_is_comp(page));
+		ut_a(!page_zip);
+		ptr = mlog_parse_index(ptr, end_ptr, TRUE, &index);
+		if (!ptr) {
+			break;
+		}
+		/* Fall through */
+	case MLOG_REC_SEC_DELETE_MARK:
+		ptr = btr_cur_parse_del_mark_set_sec_rec(ptr, end_ptr,
+							 page, page_zip);
+		break;
+	case MLOG_REC_UPDATE_IN_PLACE: case MLOG_COMP_REC_UPDATE_IN_PLACE:
+		if (NULL != (ptr = mlog_parse_index(
+				     ptr, end_ptr,
+				     type == MLOG_COMP_REC_UPDATE_IN_PLACE,
+				     &index))) {
+			ut_a(!page
+			     || (ibool)!!page_is_comp(page)
+			     == dict_table_is_comp(index->table));
+			ptr = btr_cur_parse_update_in_place(ptr, end_ptr, page,
+							    page_zip, index);
+		}
+		break;
+	case MLOG_LIST_END_DELETE: case MLOG_COMP_LIST_END_DELETE:
+	case MLOG_LIST_START_DELETE: case MLOG_COMP_LIST_START_DELETE:
+		if (NULL != (ptr = mlog_parse_index(
+				     ptr, end_ptr,
+				     type == MLOG_COMP_LIST_END_DELETE
+				     || type == MLOG_COMP_LIST_START_DELETE,
+				     &index))) {
+			ut_a(!page
+			     || (ibool)!!page_is_comp(page)
+			     == dict_table_is_comp(index->table));
+			ptr = page_parse_delete_rec_list(type, ptr, end_ptr,
+							 block, index, mtr);
+		}
+		break;
+	case MLOG_LIST_END_COPY_CREATED: case MLOG_COMP_LIST_END_COPY_CREATED:
+		if (NULL != (ptr = mlog_parse_index(
+				     ptr, end_ptr,
+				     type == MLOG_COMP_LIST_END_COPY_CREATED,
+				     &index))) {
+			ut_a(!page
+			     || (ibool)!!page_is_comp(page)
+			     == dict_table_is_comp(index->table));
+			ptr = page_parse_copy_rec_list_to_created_page(
+				ptr, end_ptr, block, index, mtr);
+		}
+		break;
+	case MLOG_PAGE_REORGANIZE: case MLOG_COMP_PAGE_REORGANIZE:
+		if (NULL != (ptr = mlog_parse_index(
+				     ptr, end_ptr,
+				     type == MLOG_COMP_PAGE_REORGANIZE,
+				     &index))) {
+			ut_a(!page
+			     || (ibool)!!page_is_comp(page)
+			     == dict_table_is_comp(index->table));
+			ptr = btr_parse_page_reorganize(ptr, end_ptr, index,
+							block, mtr);
+		}
+		break;
+	case MLOG_PAGE_CREATE: case MLOG_COMP_PAGE_CREATE:
+		ut_a(!page_zip);
+		ptr = page_parse_create(ptr, end_ptr,
+					type == MLOG_COMP_PAGE_CREATE,
+					block, mtr);
+		break;
+	case MLOG_UNDO_INSERT:
+		ptr = trx_undo_parse_add_undo_rec(ptr, end_ptr, page);
+		break;
+	case MLOG_UNDO_ERASE_END:
+		ptr = trx_undo_parse_erase_page_end(ptr, end_ptr, page, mtr);
+		break;
+	case MLOG_UNDO_INIT:
+		ptr = trx_undo_parse_page_init(ptr, end_ptr, page, mtr);
+		break;
+	case MLOG_UNDO_HDR_DISCARD:
+		ptr = trx_undo_parse_discard_latest(ptr, end_ptr, page, mtr);
+		break;
+	case MLOG_UNDO_HDR_CREATE:
+	case MLOG_UNDO_HDR_REUSE:
+		ptr = trx_undo_parse_page_header(type, ptr, end_ptr,
+						 page, mtr);
+		break;
+	case MLOG_REC_MIN_MARK: case MLOG_COMP_REC_MIN_MARK:
+		/* On a compressed page, MLOG_COMP_REC_MIN_MARK
+		will be followed by MLOG_COMP_REC_DELETE
+		or MLOG_ZIP_WRITE_HEADER(FIL_PAGE_PREV, FIL_NULL)
+		in the same mini-transaction. */
+		ut_a(type == MLOG_COMP_REC_MIN_MARK || !page_zip);
+		ptr = btr_parse_set_min_rec_mark(
+			ptr, end_ptr, type == MLOG_COMP_REC_MIN_MARK,
+			page, mtr);
+		break;
+	case MLOG_REC_DELETE: case MLOG_COMP_REC_DELETE:
+		if (NULL != (ptr = mlog_parse_index(
+				     ptr, end_ptr,
+				     type == MLOG_COMP_REC_DELETE,
+				     &index))) {
+			ut_a(!page
+			     || (ibool)!!page_is_comp(page)
+			     == dict_table_is_comp(index->table));
+			ptr = page_cur_parse_delete_rec(ptr, end_ptr,
+							block, index, mtr);
+		}
+		break;
+	case MLOG_IBUF_BITMAP_INIT:
+		ptr = ibuf_parse_bitmap_init(ptr, end_ptr, block, mtr);
+		break;
+	case MLOG_INIT_FILE_PAGE:
+		ptr = fsp_parse_init_file_page(ptr, end_ptr, block);
+		break;
+	case MLOG_WRITE_STRING:
+		ptr = mlog_parse_string(ptr, end_ptr, page, page_zip);
+		break;
+	case MLOG_FILE_CREATE:
+	case MLOG_FILE_RENAME:
+	case MLOG_FILE_DELETE:
+	case MLOG_FILE_CREATE2:
+		ptr = fil_op_log_parse_or_replay(ptr, end_ptr, type, 0);
+		break;
+	case MLOG_ZIP_WRITE_NODE_PTR:
+		ptr = page_zip_parse_write_node_ptr(ptr, end_ptr,
+						    page, page_zip);
+		break;
+	case MLOG_ZIP_WRITE_BLOB_PTR:
+		ptr = page_zip_parse_write_blob_ptr(ptr, end_ptr,
+						    page, page_zip);
+		break;
+	case MLOG_ZIP_WRITE_HEADER:
+		ptr = page_zip_parse_write_header(ptr, end_ptr,
+						  page, page_zip);
+		break;
+	case MLOG_ZIP_PAGE_COMPRESS:
+		ptr = page_zip_parse_compress(ptr, end_ptr,
+					      page, page_zip);
+		break;
+	default:
+		ptr = NULL;
+		recv_sys->found_corrupt_log = TRUE;
+	}
+
+	if (index) {
+		dict_table_t*	table = index->table;
+
+		dict_mem_index_free(index);
+		dict_mem_table_free(table);
+	}
+
+	return(ptr);
+}
+
+/*************************************************************************
+Calculates the fold value of a page file address: used in inserting or
+searching for a log record in the hash table. */
+UNIV_INLINE
+ulint
+recv_fold(
+/*======*/
+			/* out: folded value */
+	ulint	space,	/* in: space */
+	ulint	page_no)/* in: page number */
+{
+	return(ut_fold_ulint_pair(space, page_no));
+}
+
+/*************************************************************************
+Calculates the hash value of a page file address: used in inserting or
+searching for a log record in the hash table. */
+UNIV_INLINE
+ulint
+recv_hash(
+/*======*/
+			/* out: folded value */
+	ulint	space,	/* in: space */
+	ulint	page_no)/* in: page number */
+{
+	return(hash_calc_hash(recv_fold(space, page_no), recv_sys->addr_hash));
+}
+
+/*************************************************************************
+Gets the hashed file address struct for a page. */
+static
+recv_addr_t*
+recv_get_fil_addr_struct(
+/*=====================*/
+			/* out: file address struct, NULL if not found from
+			the hash table */
+	ulint	space,	/* in: space id */
+	ulint	page_no)/* in: page number */
+{
+	recv_addr_t*	recv_addr;
+
+	recv_addr = HASH_GET_FIRST(recv_sys->addr_hash,
+				   recv_hash(space, page_no));
+	while (recv_addr) {
+		if ((recv_addr->space == space)
+		    && (recv_addr->page_no == page_no)) {
+
+			break;
+		}
+
+		recv_addr = HASH_GET_NEXT(addr_hash, recv_addr);
+	}
+
+	return(recv_addr);
+}
+
+/***********************************************************************
+Adds a new log record to the hash table of log records. */
+static
+void
+recv_add_to_hash_table(
+/*===================*/
+	byte		type,		/* in: log record type */
+	ulint		space,		/* in: space id */
+	ulint		page_no,	/* in: page number */
+	byte*		body,		/* in: log record body */
+	byte*		rec_end,	/* in: log record end */
+	ib_uint64_t	start_lsn,	/* in: start lsn of the mtr */
+	ib_uint64_t	end_lsn)	/* in: end lsn of the mtr */
+{
+	recv_t*		recv;
+	ulint		len;
+	recv_data_t*	recv_data;
+	recv_data_t**	prev_field;
+	recv_addr_t*	recv_addr;
+
+	if (fil_tablespace_deleted_or_being_deleted_in_mem(space, -1)) {
+		/* The tablespace does not exist any more: do not store the
+		log record */
+
+		return;
+	}
+
+	len = rec_end - body;
+
+	recv = mem_heap_alloc(recv_sys->heap, sizeof(recv_t));
+	recv->type = type;
+	recv->len = rec_end - body;
+	recv->start_lsn = start_lsn;
+	recv->end_lsn = end_lsn;
+
+	recv_addr = recv_get_fil_addr_struct(space, page_no);
+
+	if (recv_addr == NULL) {
+		recv_addr = mem_heap_alloc(recv_sys->heap,
+					   sizeof(recv_addr_t));
+		recv_addr->space = space;
+		recv_addr->page_no = page_no;
+		recv_addr->state = RECV_NOT_PROCESSED;
+
+		UT_LIST_INIT(recv_addr->rec_list);
+
+		HASH_INSERT(recv_addr_t, addr_hash, recv_sys->addr_hash,
+			    recv_fold(space, page_no), recv_addr);
+		recv_sys->n_addrs++;
+#if 0
+		fprintf(stderr, "Inserting log rec for space %lu, page %lu\n",
+			space, page_no);
+#endif
+	}
+
+	UT_LIST_ADD_LAST(rec_list, recv_addr->rec_list, recv);
+
+	prev_field = &(recv->data);
+
+	/* Store the log record body in chunks of less than UNIV_PAGE_SIZE:
+	recv_sys->heap grows into the buffer pool, and bigger chunks could not
+	be allocated */
+
+	while (rec_end > body) {
+
+		len = rec_end - body;
+
+		if (len > RECV_DATA_BLOCK_SIZE) {
+			len = RECV_DATA_BLOCK_SIZE;
+		}
+
+		recv_data = mem_heap_alloc(recv_sys->heap,
+					   sizeof(recv_data_t) + len);
+		*prev_field = recv_data;
+
+		ut_memcpy(((byte*)recv_data) + sizeof(recv_data_t), body, len);
+
+		prev_field = &(recv_data->next);
+
+		body += len;
+	}
+
+	*prev_field = NULL;
+}
+
+/*************************************************************************
+Copies the log record body from recv to buf. */
+static
+void
+recv_data_copy_to_buf(
+/*==================*/
+	byte*	buf,	/* in: buffer of length at least recv->len */
+	recv_t*	recv)	/* in: log record */
+{
+	recv_data_t*	recv_data;
+	ulint		part_len;
+	ulint		len;
+
+	len = recv->len;
+	recv_data = recv->data;
+
+	while (len > 0) {
+		if (len > RECV_DATA_BLOCK_SIZE) {
+			part_len = RECV_DATA_BLOCK_SIZE;
+		} else {
+			part_len = len;
+		}
+
+		ut_memcpy(buf, ((byte*)recv_data) + sizeof(recv_data_t),
+			  part_len);
+		buf += part_len;
+		len -= part_len;
+
+		recv_data = recv_data->next;
+	}
+}
+
+/****************************************************************************
+Applies the hashed log records to the page, if the page lsn is less than the
+lsn of a log record. This can be called when a buffer page has just been
+read in, or also for a page already in the buffer pool. */
+UNIV_INTERN
+void
+recv_recover_page(
+/*==============*/
+	ibool		recover_backup,
+				/* in: TRUE if we are recovering a backup
+				page: then we do not acquire any latches
+				since the page was read in outside the
+				buffer pool */
+	ibool		just_read_in,
+				/* in: TRUE if the i/o-handler calls this for
+				a freshly read page */
+	buf_block_t*	block)	/* in: buffer block */
+{
+	page_t*		page;
+	recv_addr_t*	recv_addr;
+	recv_t*		recv;
+	byte*		buf;
+	ib_uint64_t	start_lsn;
+	ib_uint64_t	end_lsn;
+	ib_uint64_t	page_lsn;
+	ib_uint64_t	page_newest_lsn;
+	ibool		modification_to_page;
+	ibool		success;
+	mtr_t		mtr;
+
+	mutex_enter(&(recv_sys->mutex));
+
+	if (recv_sys->apply_log_recs == FALSE) {
+
+		/* Log records should not be applied now */
+
+		mutex_exit(&(recv_sys->mutex));
+
+		return;
+	}
+
+	recv_addr = recv_get_fil_addr_struct(buf_block_get_space(block),
+					     buf_block_get_page_no(block));
+
+	if ((recv_addr == NULL)
+	    || (recv_addr->state == RECV_BEING_PROCESSED)
+	    || (recv_addr->state == RECV_PROCESSED)) {
+
+		mutex_exit(&(recv_sys->mutex));
+
+		return;
+	}
+
+#if 0
+	fprintf(stderr, "Recovering space %lu, page %lu\n",
+		buf_block_get_space(block), buf_block_get_page_no(block));
+#endif
+
+	recv_addr->state = RECV_BEING_PROCESSED;
+
+	mutex_exit(&(recv_sys->mutex));
+
+	mtr_start(&mtr);
+	mtr_set_log_mode(&mtr, MTR_LOG_NONE);
+
+	page = block->frame;
+
+	if (!recover_backup) {
+		if (just_read_in) {
+			/* Move the ownership of the x-latch on the
+			page to this OS thread, so that we can acquire
+			a second x-latch on it. This is needed for the
+			operations to the page to pass the debug
+			checks. */
+
+			rw_lock_x_lock_move_ownership(&(block->lock));
+		}
+
+		success = buf_page_get_known_nowait(RW_X_LATCH, block,
+						    BUF_KEEP_OLD,
+						    __FILE__, __LINE__,
+						    &mtr);
+		ut_a(success);
+
+		buf_block_dbg_add_level(block, SYNC_NO_ORDER_CHECK);
+	}
+
+	/* Read the newest modification lsn from the page */
+	page_lsn = mach_read_ull(page + FIL_PAGE_LSN);
+
+	if (!recover_backup) {
+		/* It may be that the page has been modified in the buffer
+		pool: read the newest modification lsn there */
+
+		page_newest_lsn
+			= buf_page_get_newest_modification(&block->page);
+
+		if (page_newest_lsn) {
+
+			page_lsn = page_newest_lsn;
+		}
+	} else {
+		/* In recovery from a backup we do not really use the buffer
+		pool */
+
+		page_newest_lsn = 0;
+	}
+
+	modification_to_page = FALSE;
+	start_lsn = end_lsn = 0;
+
+	recv = UT_LIST_GET_FIRST(recv_addr->rec_list);
+
+	while (recv) {
+		end_lsn = recv->end_lsn;
+
+		if (recv->len > RECV_DATA_BLOCK_SIZE) {
+			/* We have to copy the record body to a separate
+			buffer */
+
+			buf = mem_alloc(recv->len);
+
+			recv_data_copy_to_buf(buf, recv);
+		} else {
+			buf = ((byte*)(recv->data)) + sizeof(recv_data_t);
+		}
+
+		if (recv->type == MLOG_INIT_FILE_PAGE) {
+			page_lsn = page_newest_lsn;
+
+			mach_write_ull(page + UNIV_PAGE_SIZE
+				       - FIL_PAGE_END_LSN_OLD_CHKSUM, 0);
+			mach_write_ull(page + FIL_PAGE_LSN, 0);
+		}
+
+		if (recv->start_lsn >= page_lsn) {
+
+			if (!modification_to_page) {
+
+				modification_to_page = TRUE;
+				start_lsn = recv->start_lsn;
+			}
+
+#ifdef UNIV_DEBUG
+			if (log_debug_writes) {
+				fprintf(stderr,
+					"InnoDB: Applying log rec"
+					" type %lu len %lu"
+					" to space %lu page no %lu\n",
+					(ulong) recv->type, (ulong) recv->len,
+					(ulong) recv_addr->space,
+					(ulong) recv_addr->page_no);
+			}
+#endif /* UNIV_DEBUG */
+
+			recv_parse_or_apply_log_rec_body(recv->type, buf,
+							 buf + recv->len,
+							 block, &mtr);
+			mach_write_ull(page + UNIV_PAGE_SIZE
+				       - FIL_PAGE_END_LSN_OLD_CHKSUM,
+				       recv->start_lsn + recv->len);
+			mach_write_ull(page + FIL_PAGE_LSN,
+				       recv->start_lsn + recv->len);
+		}
+
+		if (recv->len > RECV_DATA_BLOCK_SIZE) {
+			mem_free(buf);
+		}
+
+		recv = UT_LIST_GET_NEXT(rec_list, recv);
+	}
+
+#ifdef UNIV_ZIP_DEBUG
+	if (fil_page_get_type(page) == FIL_PAGE_INDEX) {
+		page_zip_des_t*	page_zip = buf_block_get_page_zip(block);
+
+		if (page_zip) {
+			ut_a(page_zip_validate_low(page_zip, page, FALSE));
+		}
+	}
+#endif /* UNIV_ZIP_DEBUG */
+
+	mutex_enter(&(recv_sys->mutex));
+
+	if (recv_max_page_lsn < page_lsn) {
+		recv_max_page_lsn = page_lsn;
+	}
+
+	recv_addr->state = RECV_PROCESSED;
+
+	ut_a(recv_sys->n_addrs);
+	recv_sys->n_addrs--;
+
+	mutex_exit(&(recv_sys->mutex));
+
+	if (!recover_backup && modification_to_page) {
+		ut_a(block);
+
+		buf_flush_recv_note_modification(block, start_lsn, end_lsn);
+	}
+
+	/* Make sure that committing mtr does not change the modification
+	lsn values of page */
+
+	mtr.modifications = FALSE;
+
+	mtr_commit(&mtr);
+}
+
+/***********************************************************************
+Reads in pages which have hashed log records, from an area around a given
+page number. */
+static
+ulint
+recv_read_in_area(
+/*==============*/
+			/* out: number of pages found */
+	ulint	space,	/* in: space */
+	ulint	zip_size,/* in: compressed page size in bytes, or 0 */
+	ulint	page_no)/* in: page number */
+{
+	recv_addr_t* recv_addr;
+	ulint	page_nos[RECV_READ_AHEAD_AREA];
+	ulint	low_limit;
+	ulint	n;
+
+	low_limit = page_no - (page_no % RECV_READ_AHEAD_AREA);
+
+	n = 0;
+
+	for (page_no = low_limit; page_no < low_limit + RECV_READ_AHEAD_AREA;
+	     page_no++) {
+		recv_addr = recv_get_fil_addr_struct(space, page_no);
+
+		if (recv_addr && !buf_page_peek(space, page_no)) {
+
+			mutex_enter(&(recv_sys->mutex));
+
+			if (recv_addr->state == RECV_NOT_PROCESSED) {
+				recv_addr->state = RECV_BEING_READ;
+
+				page_nos[n] = page_no;
+
+				n++;
+			}
+
+			mutex_exit(&(recv_sys->mutex));
+		}
+	}
+
+	buf_read_recv_pages(FALSE, space, zip_size, page_nos, n);
+	/*
+	fprintf(stderr, "Recv pages at %lu n %lu\n", page_nos[0], n);
+	*/
+	return(n);
+}
+
+/***********************************************************************
+Empties the hash table of stored log records, applying them to appropriate
+pages. */
+UNIV_INTERN
+void
+recv_apply_hashed_log_recs(
+/*=======================*/
+	ibool	allow_ibuf)	/* in: if TRUE, also ibuf operations are
+				allowed during the application; if FALSE,
+				no ibuf operations are allowed, and after
+				the application all file pages are flushed to
+				disk and invalidated in buffer pool: this
+				alternative means that no new log records
+				can be generated during the application;
+				the caller must in this case own the log
+				mutex */
+{
+	recv_addr_t* recv_addr;
+	ulint	i;
+	ulint	n_pages;
+	ibool	has_printed	= FALSE;
+	mtr_t	mtr;
+loop:
+	mutex_enter(&(recv_sys->mutex));
+
+	if (recv_sys->apply_batch_on) {
+
+		mutex_exit(&(recv_sys->mutex));
+
+		os_thread_sleep(500000);
+
+		goto loop;
+	}
+
+	ut_ad(!allow_ibuf == mutex_own(&log_sys->mutex));
+
+	if (!allow_ibuf) {
+		recv_no_ibuf_operations = TRUE;
+	}
+
+	recv_sys->apply_log_recs = TRUE;
+	recv_sys->apply_batch_on = TRUE;
+
+	for (i = 0; i < hash_get_n_cells(recv_sys->addr_hash); i++) {
+
+		recv_addr = HASH_GET_FIRST(recv_sys->addr_hash, i);
+
+		while (recv_addr) {
+			ulint	space = recv_addr->space;
+			ulint	zip_size = fil_space_get_zip_size(space);
+			ulint	page_no = recv_addr->page_no;
+
+			if (recv_addr->state == RECV_NOT_PROCESSED) {
+				if (!has_printed) {
+					ut_print_timestamp(stderr);
+					fputs("  InnoDB: Starting an"
+					      " apply batch of log records"
+					      " to the database...\n"
+					      "InnoDB: Progress in percents: ",
+					      stderr);
+					has_printed = TRUE;
+				}
+
+				mutex_exit(&(recv_sys->mutex));
+
+				if (buf_page_peek(space, page_no)) {
+					buf_block_t*	block;
+
+					mtr_start(&mtr);
+
+					block = buf_page_get(
+						space, zip_size, page_no,
+						RW_X_LATCH, &mtr);
+					buf_block_dbg_add_level(
+						block, SYNC_NO_ORDER_CHECK);
+
+					recv_recover_page(FALSE, FALSE, block);
+					mtr_commit(&mtr);
+				} else {
+					recv_read_in_area(space, zip_size,
+							  page_no);
+				}
+
+				mutex_enter(&(recv_sys->mutex));
+			}
+
+			recv_addr = HASH_GET_NEXT(addr_hash, recv_addr);
+		}
+
+		if (has_printed
+		    && (i * 100) / hash_get_n_cells(recv_sys->addr_hash)
+		    != ((i + 1) * 100)
+		    / hash_get_n_cells(recv_sys->addr_hash)) {
+
+			fprintf(stderr, "%lu ", (ulong)
+				((i * 100)
+				 / hash_get_n_cells(recv_sys->addr_hash)));
+		}
+	}
+
+	/* Wait until all the pages have been processed */
+
+	while (recv_sys->n_addrs != 0) {
+
+		mutex_exit(&(recv_sys->mutex));
+
+		os_thread_sleep(500000);
+
+		mutex_enter(&(recv_sys->mutex));
+	}
+
+	if (has_printed) {
+
+		fprintf(stderr, "\n");
+	}
+
+	if (!allow_ibuf) {
+		/* Flush all the file pages to disk and invalidate them in
+		the buffer pool */
+
+		mutex_exit(&(recv_sys->mutex));
+		mutex_exit(&(log_sys->mutex));
+
+		n_pages = buf_flush_batch(BUF_FLUSH_LIST, ULINT_MAX,
+					  IB_ULONGLONG_MAX);
+		ut_a(n_pages != ULINT_UNDEFINED);
+
+		buf_flush_wait_batch_end(BUF_FLUSH_LIST);
+
+		buf_pool_invalidate();
+
+		mutex_enter(&(log_sys->mutex));
+		mutex_enter(&(recv_sys->mutex));
+
+		recv_no_ibuf_operations = FALSE;
+	}
+
+	recv_sys->apply_log_recs = FALSE;
+	recv_sys->apply_batch_on = FALSE;
+
+	recv_sys_empty_hash();
+
+	if (has_printed) {
+		fprintf(stderr, "InnoDB: Apply batch completed\n");
+	}
+
+	mutex_exit(&(recv_sys->mutex));
+}
+
+#ifdef UNIV_HOTBACKUP
+/***********************************************************************
+Applies log records in the hash table to a backup. */
+UNIV_INTERN
+void
+recv_apply_log_recs_for_backup(void)
+/*================================*/
+{
+	recv_addr_t*	recv_addr;
+	ulint		n_hash_cells;
+	buf_block_t*	block;
+	ulint		actual_size;
+	ibool		success;
+	ulint		error;
+	ulint		i;
+
+	recv_sys->apply_log_recs = TRUE;
+	recv_sys->apply_batch_on = TRUE;
+
+	block = buf_LRU_get_free_block(UNIV_PAGE_SIZE);
+
+	fputs("InnoDB: Starting an apply batch of log records"
+	      " to the database...\n"
+	      "InnoDB: Progress in percents: ", stderr);
+
+	n_hash_cells = hash_get_n_cells(recv_sys->addr_hash);
+
+	for (i = 0; i < n_hash_cells; i++) {
+		/* The address hash table is externally chained */
+		recv_addr = hash_get_nth_cell(recv_sys->addr_hash, i)->node;
+
+		while (recv_addr != NULL) {
+
+			ulint	zip_size
+				= fil_space_get_zip_size(recv_addr->space);
+
+			if (zip_size == ULINT_UNDEFINED) {
+#if 0
+				fprintf(stderr,
+					"InnoDB: Warning: cannot apply"
+					" log record to"
+					" tablespace %lu page %lu,\n"
+					"InnoDB: because tablespace with"
+					" that id does not exist.\n",
+					recv_addr->space, recv_addr->page_no);
+#endif
+				recv_addr->state = RECV_PROCESSED;
+
+				ut_a(recv_sys->n_addrs);
+				recv_sys->n_addrs--;
+
+				goto skip_this_recv_addr;
+			}
+
+			/* We simulate a page read made by the buffer pool, to
+			make sure the recovery apparatus works ok. We must init
+			the block. */
+
+			buf_page_init_for_backup_restore(
+				recv_addr->space, recv_addr->page_no,
+				zip_size, block);
+
+			/* Extend the tablespace's last file if the page_no
+			does not fall inside its bounds; we assume the last
+			file is auto-extending, and ibbackup copied the file
+			when it still was smaller */
+
+			success = fil_extend_space_to_desired_size(
+				&actual_size,
+				recv_addr->space, recv_addr->page_no + 1);
+			if (!success) {
+				fprintf(stderr,
+					"InnoDB: Fatal error: cannot extend"
+					" tablespace %lu to hold %lu pages\n",
+					recv_addr->space, recv_addr->page_no);
+
+				exit(1);
+			}
+
+			/* Read the page from the tablespace file using the
+			fil0fil.c routines */
+
+			if (zip_size) {
+				error = fil_io(OS_FILE_READ, TRUE,
+					       recv_addr->space, zip_size,
+					       recv_addr->page_no, 0, zip_size,
+					       block->page.zip.data, NULL);
+			} else {
+				error = fil_io(OS_FILE_READ, TRUE,
+					       recv_addr->space, 0,
+					       recv_addr->page_no, 0,
+					       UNIV_PAGE_SIZE,
+					       block->frame, NULL);
+			}
+
+			if (error != DB_SUCCESS) {
+				fprintf(stderr,
+					"InnoDB: Fatal error: cannot read"
+					" from tablespace"
+					" %lu page number %lu\n",
+					(ulong) recv_addr->space,
+					(ulong) recv_addr->page_no);
+
+				exit(1);
+			}
+
+			/* Apply the log records to this page */
+			recv_recover_page(TRUE, FALSE, block);
+
+			/* Write the page back to the tablespace file using the
+			fil0fil.c routines */
+
+			buf_flush_init_for_writing(
+				block->frame, buf_block_get_page_zip(block),
+				mach_read_ull(block->frame + FIL_PAGE_LSN));
+
+			if (zip_size) {
+				error = fil_io(OS_FILE_WRITE, TRUE,
+					       recv_addr->space, zip_size,
+					       recv_addr->page_no, 0,
+					       zip_size,
+					       block->page.zip.data, NULL);
+			} else {
+				error = fil_io(OS_FILE_WRITE, TRUE,
+					       recv_addr->space, 0,
+					       recv_addr->page_no, 0,
+					       UNIV_PAGE_SIZE,
+					       block->frame, NULL);
+			}
+skip_this_recv_addr:
+			recv_addr = HASH_GET_NEXT(addr_hash, recv_addr);
+		}
+
+		if ((100 * i) / n_hash_cells
+		    != (100 * (i + 1)) / n_hash_cells) {
+			fprintf(stderr, "%lu ",
+				(ulong) ((100 * i) / n_hash_cells));
+			fflush(stderr);
+		}
+	}
+
+	buf_block_free(block);
+	recv_sys_empty_hash();
+}
+#endif /* UNIV_HOTBACKUP */
+
+/***********************************************************************
+Tries to parse a single log record and returns its length. */
+static
+ulint
+recv_parse_log_rec(
+/*===============*/
+			/* out: length of the record, or 0 if the record was
+			not complete */
+	byte*	ptr,	/* in: pointer to a buffer */
+	byte*	end_ptr,/* in: pointer to the buffer end */
+	byte*	type,	/* out: type */
+	ulint*	space,	/* out: space id */
+	ulint*	page_no,/* out: page number */
+	byte**	body)	/* out: log record body start */
+{
+	byte*	new_ptr;
+
+	*body = NULL;
+
+	if (ptr == end_ptr) {
+
+		return(0);
+	}
+
+	if (*ptr == MLOG_MULTI_REC_END) {
+
+		*type = *ptr;
+
+		return(1);
+	}
+
+	if (*ptr == MLOG_DUMMY_RECORD) {
+		*type = *ptr;
+
+		*space = ULINT_UNDEFINED - 1; /* For debugging */
+
+		return(1);
+	}
+
+	new_ptr = mlog_parse_initial_log_record(ptr, end_ptr, type, space,
+						page_no);
+	*body = new_ptr;
+
+	if (UNIV_UNLIKELY(!new_ptr)) {
+
+		return(0);
+	}
+
+	/* Check that page_no is sensible */
+
+	if (UNIV_UNLIKELY(*page_no > 0x8FFFFFFFUL)) {
+
+		recv_sys->found_corrupt_log = TRUE;
+
+		return(0);
+	}
+
+	new_ptr = recv_parse_or_apply_log_rec_body(*type, new_ptr, end_ptr,
+						   NULL, NULL);
+	if (UNIV_UNLIKELY(new_ptr == NULL)) {
+
+		return(0);
+	}
+
+	if (*page_no > recv_max_parsed_page_no) {
+		recv_max_parsed_page_no = *page_no;
+	}
+
+	return(new_ptr - ptr);
+}
+
+/***********************************************************
+Calculates the new value for lsn when more data is added to the log. */
+static
+ib_uint64_t
+recv_calc_lsn_on_data_add(
+/*======================*/
+	ib_uint64_t	lsn,	/* in: old lsn */
+	ib_uint64_t	len)	/* in: this many bytes of data is
+				added, log block headers not included */
+{
+	ulint	frag_len;
+	ulint	lsn_len;
+
+	frag_len = (((ulint) lsn) % OS_FILE_LOG_BLOCK_SIZE)
+		- LOG_BLOCK_HDR_SIZE;
+	ut_ad(frag_len < OS_FILE_LOG_BLOCK_SIZE - LOG_BLOCK_HDR_SIZE
+	      - LOG_BLOCK_TRL_SIZE);
+	lsn_len = (ulint) len;
+	lsn_len += (lsn_len + frag_len)
+		/ (OS_FILE_LOG_BLOCK_SIZE - LOG_BLOCK_HDR_SIZE
+		   - LOG_BLOCK_TRL_SIZE)
+		* (LOG_BLOCK_HDR_SIZE + LOG_BLOCK_TRL_SIZE);
+
+	return(lsn + lsn_len);
+}
+
+#ifdef UNIV_LOG_DEBUG
+/***********************************************************
+Checks that the parser recognizes incomplete initial segments of a log
+record as incomplete. */
+static
+void
+recv_check_incomplete_log_recs(
+/*===========================*/
+	byte*	ptr,	/* in: pointer to a complete log record */
+	ulint	len)	/* in: length of the log record */
+{
+	ulint	i;
+	byte	type;
+	ulint	space;
+	ulint	page_no;
+	byte*	body;
+
+	for (i = 0; i < len; i++) {
+		ut_a(0 == recv_parse_log_rec(ptr, ptr + i, &type, &space,
+					     &page_no, &body));
+	}
+}
+#endif /* UNIV_LOG_DEBUG */
+
+/***********************************************************
+Prints diagnostic info of corrupt log. */
+static
+void
+recv_report_corrupt_log(
+/*====================*/
+	byte*	ptr,	/* in: pointer to corrupt log record */
+	byte	type,	/* in: type of the record */
+	ulint	space,	/* in: space id, this may also be garbage */
+	ulint	page_no)/* in: page number, this may also be garbage */
+{
+	fprintf(stderr,
+		"InnoDB: ############### CORRUPT LOG RECORD FOUND\n"
+		"InnoDB: Log record type %lu, space id %lu, page number %lu\n"
+		"InnoDB: Log parsing proceeded successfully up to %llu\n"
+		"InnoDB: Previous log record type %lu, is multi %lu\n"
+		"InnoDB: Recv offset %lu, prev %lu\n",
+		(ulong) type, (ulong) space, (ulong) page_no,
+		recv_sys->recovered_lsn,
+		(ulong) recv_previous_parsed_rec_type,
+		(ulong) recv_previous_parsed_rec_is_multi,
+		(ulong) (ptr - recv_sys->buf),
+		(ulong) recv_previous_parsed_rec_offset);
+
+	if ((ulint)(ptr - recv_sys->buf + 100)
+	    > recv_previous_parsed_rec_offset
+	    && (ulint)(ptr - recv_sys->buf + 100
+		       - recv_previous_parsed_rec_offset)
+	    < 200000) {
+		fputs("InnoDB: Hex dump of corrupt log starting"
+		      " 100 bytes before the start\n"
+		      "InnoDB: of the previous log rec,\n"
+		      "InnoDB: and ending 100 bytes after the start"
+		      " of the corrupt rec:\n",
+		      stderr);
+
+		ut_print_buf(stderr,
+			     recv_sys->buf
+			     + recv_previous_parsed_rec_offset - 100,
+			     ptr - recv_sys->buf + 200
+			     - recv_previous_parsed_rec_offset);
+		putc('\n', stderr);
+	}
+
+	fputs("InnoDB: WARNING: the log file may have been corrupt and it\n"
+	      "InnoDB: is possible that the log scan did not proceed\n"
+	      "InnoDB: far enough in recovery! Please run CHECK TABLE\n"
+	      "InnoDB: on your InnoDB tables to check that they are ok!\n"
+	      "InnoDB: If mysqld crashes after this recovery, look at\n"
+	      "InnoDB: http://dev.mysql.com/doc/refman/5.1/en/"
+	      "forcing-recovery.html\n"
+	      "InnoDB: about forcing recovery.\n", stderr);
+
+	fflush(stderr);
+}
+
+/***********************************************************
+Parses log records from a buffer and stores them to a hash table to wait
+merging to file pages. */
+static
+ibool
+recv_parse_log_recs(
+/*================*/
+				/* out: currently always returns FALSE */
+	ibool	store_to_hash)	/* in: TRUE if the records should be stored
+				to the hash table; this is set to FALSE if just
+				debug checking is needed */
+{
+	byte*		ptr;
+	byte*		end_ptr;
+	ulint		single_rec;
+	ulint		len;
+	ulint		total_len;
+	ib_uint64_t	new_recovered_lsn;
+	ib_uint64_t	old_lsn;
+	byte		type;
+	ulint		space;
+	ulint		page_no;
+	byte*		body;
+	ulint		n_recs;
+
+	ut_ad(mutex_own(&(log_sys->mutex)));
+	ut_ad(recv_sys->parse_start_lsn != 0);
+loop:
+	ptr = recv_sys->buf + recv_sys->recovered_offset;
+
+	end_ptr = recv_sys->buf + recv_sys->len;
+
+	if (ptr == end_ptr) {
+
+		return(FALSE);
+	}
+
+	single_rec = (ulint)*ptr & MLOG_SINGLE_REC_FLAG;
+
+	if (single_rec || *ptr == MLOG_DUMMY_RECORD) {
+		/* The mtr only modified a single page, or this is a file op */
+
+		old_lsn = recv_sys->recovered_lsn;
+
+		/* Try to parse a log record, fetching its type, space id,
+		page no, and a pointer to the body of the log record */
+
+		len = recv_parse_log_rec(ptr, end_ptr, &type, &space,
+					 &page_no, &body);
+
+		if (len == 0 || recv_sys->found_corrupt_log) {
+			if (recv_sys->found_corrupt_log) {
+
+				recv_report_corrupt_log(ptr,
+							type, space, page_no);
+			}
+
+			return(FALSE);
+		}
+
+		new_recovered_lsn = recv_calc_lsn_on_data_add(old_lsn, len);
+
+		if (new_recovered_lsn > recv_sys->scanned_lsn) {
+			/* The log record filled a log block, and we require
+			that also the next log block should have been scanned
+			in */
+
+			return(FALSE);
+		}
+
+		recv_previous_parsed_rec_type = (ulint)type;
+		recv_previous_parsed_rec_offset = recv_sys->recovered_offset;
+		recv_previous_parsed_rec_is_multi = 0;
+
+		recv_sys->recovered_offset += len;
+		recv_sys->recovered_lsn = new_recovered_lsn;
+
+#ifdef UNIV_DEBUG
+		if (log_debug_writes) {
+			fprintf(stderr,
+				"InnoDB: Parsed a single log rec"
+				" type %lu len %lu space %lu page no %lu\n",
+				(ulong) type, (ulong) len, (ulong) space,
+				(ulong) page_no);
+		}
+#endif /* UNIV_DEBUG */
+
+		if (type == MLOG_DUMMY_RECORD) {
+			/* Do nothing */
+
+		} else if (!store_to_hash) {
+			/* In debug checking, update a replicate page
+			according to the log record, and check that it
+			becomes identical with the original page */
+#ifdef UNIV_LOG_DEBUG
+			recv_check_incomplete_log_recs(ptr, len);
+#endif/* UNIV_LOG_DEBUG */
+
+		} else if (type == MLOG_FILE_CREATE
+			   || type == MLOG_FILE_CREATE2
+			   || type == MLOG_FILE_RENAME
+			   || type == MLOG_FILE_DELETE) {
+			ut_a(space);
+#ifdef UNIV_HOTBACKUP
+			if (recv_replay_file_ops) {
+
+				/* In ibbackup --apply-log, replay an .ibd file
+				operation, if possible; note that
+				fil_path_to_mysql_datadir is set in ibbackup to
+				point to the datadir we should use there */
+
+				if (NULL == fil_op_log_parse_or_replay(
+					    body, end_ptr, type, space)) {
+					fprintf(stderr,
+						"InnoDB: Error: file op"
+						" log record of type %lu"
+						" space %lu not complete in\n"
+						"InnoDB: the replay phase."
+						" Path %s\n",
+						(ulint)type, space,
+						(char*)(body + 2));
+
+					ut_error;
+				}
+			}
+#endif
+			/* In normal mysqld crash recovery we do not try to
+			replay file operations */
+		} else {
+			recv_add_to_hash_table(type, space, page_no, body,
+					       ptr + len, old_lsn,
+					       recv_sys->recovered_lsn);
+		}
+	} else {
+		/* Check that all the records associated with the single mtr
+		are included within the buffer */
+
+		total_len = 0;
+		n_recs = 0;
+
+		for (;;) {
+			len = recv_parse_log_rec(ptr, end_ptr, &type, &space,
+						 &page_no, &body);
+			if (len == 0 || recv_sys->found_corrupt_log) {
+
+				if (recv_sys->found_corrupt_log) {
+
+					recv_report_corrupt_log(
+						ptr, type, space, page_no);
+				}
+
+				return(FALSE);
+			}
+
+			recv_previous_parsed_rec_type = (ulint)type;
+			recv_previous_parsed_rec_offset
+				= recv_sys->recovered_offset + total_len;
+			recv_previous_parsed_rec_is_multi = 1;
+
+			if ((!store_to_hash) && (type != MLOG_MULTI_REC_END)) {
+#ifdef UNIV_LOG_DEBUG
+				recv_check_incomplete_log_recs(ptr, len);
+#endif /* UNIV_LOG_DEBUG */
+			}
+
+#ifdef UNIV_DEBUG
+			if (log_debug_writes) {
+				fprintf(stderr,
+					"InnoDB: Parsed a multi log rec"
+					" type %lu len %lu"
+					" space %lu page no %lu\n",
+					(ulong) type, (ulong) len,
+					(ulong) space, (ulong) page_no);
+			}
+#endif /* UNIV_DEBUG */
+
+			total_len += len;
+			n_recs++;
+
+			ptr += len;
+
+			if (type == MLOG_MULTI_REC_END) {
+
+				/* Found the end mark for the records */
+
+				break;
+			}
+		}
+
+		new_recovered_lsn = recv_calc_lsn_on_data_add(
+			recv_sys->recovered_lsn, total_len);
+
+		if (new_recovered_lsn > recv_sys->scanned_lsn) {
+			/* The log record filled a log block, and we require
+			that also the next log block should have been scanned
+			in */
+
+			return(FALSE);
+		}
+
+		/* Add all the records to the hash table */
+
+		ptr = recv_sys->buf + recv_sys->recovered_offset;
+
+		for (;;) {
+			old_lsn = recv_sys->recovered_lsn;
+			len = recv_parse_log_rec(ptr, end_ptr, &type, &space,
+						 &page_no, &body);
+			if (recv_sys->found_corrupt_log) {
+
+				recv_report_corrupt_log(ptr,
+							type, space, page_no);
+			}
+
+			ut_a(len != 0);
+			ut_a(0 == ((ulint)*ptr & MLOG_SINGLE_REC_FLAG));
+
+			recv_sys->recovered_offset += len;
+			recv_sys->recovered_lsn
+				= recv_calc_lsn_on_data_add(old_lsn, len);
+			if (type == MLOG_MULTI_REC_END) {
+
+				/* Found the end mark for the records */
+
+				break;
+			}
+
+			if (store_to_hash) {
+				recv_add_to_hash_table(type, space, page_no,
+						       body, ptr + len,
+						       old_lsn,
+						       new_recovered_lsn);
+			}
+
+			ptr += len;
+		}
+	}
+
+	goto loop;
+}
+
+/***********************************************************
+Adds data from a new log block to the parsing buffer of recv_sys if
+recv_sys->parse_start_lsn is non-zero. */
+static
+ibool
+recv_sys_add_to_parsing_buf(
+/*========================*/
+					/* out: TRUE if more data added */
+	byte*		log_block,	/* in: log block */
+	ib_uint64_t	scanned_lsn)	/* in: lsn of how far we were able
+					to find data in this log block */
+{
+	ulint	more_len;
+	ulint	data_len;
+	ulint	start_offset;
+	ulint	end_offset;
+
+	ut_ad(scanned_lsn >= recv_sys->scanned_lsn);
+
+	if (!recv_sys->parse_start_lsn) {
+		/* Cannot start parsing yet because no start point for
+		it found */
+
+		return(FALSE);
+	}
+
+	data_len = log_block_get_data_len(log_block);
+
+	if (recv_sys->parse_start_lsn >= scanned_lsn) {
+
+		return(FALSE);
+
+	} else if (recv_sys->scanned_lsn >= scanned_lsn) {
+
+		return(FALSE);
+
+	} else if (recv_sys->parse_start_lsn > recv_sys->scanned_lsn) {
+		more_len = (ulint) (scanned_lsn - recv_sys->parse_start_lsn);
+	} else {
+		more_len = (ulint) (scanned_lsn - recv_sys->scanned_lsn);
+	}
+
+	if (more_len == 0) {
+
+		return(FALSE);
+	}
+
+	ut_ad(data_len >= more_len);
+
+	start_offset = data_len - more_len;
+
+	if (start_offset < LOG_BLOCK_HDR_SIZE) {
+		start_offset = LOG_BLOCK_HDR_SIZE;
+	}
+
+	end_offset = data_len;
+
+	if (end_offset > OS_FILE_LOG_BLOCK_SIZE - LOG_BLOCK_TRL_SIZE) {
+		end_offset = OS_FILE_LOG_BLOCK_SIZE - LOG_BLOCK_TRL_SIZE;
+	}
+
+	ut_ad(start_offset <= end_offset);
+
+	if (start_offset < end_offset) {
+		ut_memcpy(recv_sys->buf + recv_sys->len,
+			  log_block + start_offset, end_offset - start_offset);
+
+		recv_sys->len += end_offset - start_offset;
+
+		ut_a(recv_sys->len <= RECV_PARSING_BUF_SIZE);
+	}
+
+	return(TRUE);
+}
+
+/***********************************************************
+Moves the parsing buffer data left to the buffer start. */
+static
+void
+recv_sys_justify_left_parsing_buf(void)
+/*===================================*/
+{
+	ut_memmove(recv_sys->buf, recv_sys->buf + recv_sys->recovered_offset,
+		   recv_sys->len - recv_sys->recovered_offset);
+
+	recv_sys->len -= recv_sys->recovered_offset;
+
+	recv_sys->recovered_offset = 0;
+}
+
+/***********************************************************
+Scans log from a buffer and stores new log data to the parsing buffer. Parses
+and hashes the log records if new data found. */
+UNIV_INTERN
+ibool
+recv_scan_log_recs(
+/*===============*/
+					/* out: TRUE if limit_lsn has been
+					reached, or not able to scan any more
+					in this log group */
+	ibool		apply_automatically,/* in: TRUE if we want this
+					function to apply log records
+					automatically when the hash table
+					becomes full; in the hot backup tool
+					the tool does the applying, not this
+					function */
+	ulint		available_memory,/* in: we let the hash table of recs
+					to grow to this size, at the maximum */
+	ibool		store_to_hash,	/* in: TRUE if the records should be
+					stored to the hash table; this is set
+					to FALSE if just debug checking is
+					needed */
+	byte*		buf,		/* in: buffer containing a log segment
+					or garbage */
+	ulint		len,		/* in: buffer length */
+	ib_uint64_t	start_lsn,	/* in: buffer start lsn */
+	ib_uint64_t*	contiguous_lsn,	/* in/out: it is known that all log
+					groups contain contiguous log data up
+					to this lsn */
+	ib_uint64_t*	group_scanned_lsn)/* out: scanning succeeded up to
+					this lsn */
+{
+	byte*		log_block;
+	ulint		no;
+	ib_uint64_t	scanned_lsn;
+	ibool		finished;
+	ulint		data_len;
+	ibool		more_data;
+
+	ut_ad(start_lsn % OS_FILE_LOG_BLOCK_SIZE == 0);
+	ut_ad(len % OS_FILE_LOG_BLOCK_SIZE == 0);
+	ut_ad(len > 0);
+	ut_a(apply_automatically <= TRUE);
+	ut_a(store_to_hash <= TRUE);
+
+	finished = FALSE;
+
+	log_block = buf;
+	scanned_lsn = start_lsn;
+	more_data = FALSE;
+
+	while (log_block < buf + len && !finished) {
+
+		no = log_block_get_hdr_no(log_block);
+		/*
+		fprintf(stderr, "Log block header no %lu\n", no);
+
+		fprintf(stderr, "Scanned lsn no %lu\n",
+		log_block_convert_lsn_to_no(scanned_lsn));
+		*/
+		if (no != log_block_convert_lsn_to_no(scanned_lsn)
+		    || !log_block_checksum_is_ok_or_old_format(log_block)) {
+
+			if (no == log_block_convert_lsn_to_no(scanned_lsn)
+			    && !log_block_checksum_is_ok_or_old_format(
+				    log_block)) {
+				fprintf(stderr,
+					"InnoDB: Log block no %lu at"
+					" lsn %llu has\n"
+					"InnoDB: ok header, but checksum field"
+					" contains %lu, should be %lu\n",
+					(ulong) no,
+					scanned_lsn,
+					(ulong) log_block_get_checksum(
+						log_block),
+					(ulong) log_block_calc_checksum(
+						log_block));
+			}
+
+			/* Garbage or an incompletely written log block */
+
+			finished = TRUE;
+
+			break;
+		}
+
+		if (log_block_get_flush_bit(log_block)) {
+			/* This block was a start of a log flush operation:
+			we know that the previous flush operation must have
+			been completed for all log groups before this block
+			can have been flushed to any of the groups. Therefore,
+			we know that log data is contiguous up to scanned_lsn
+			in all non-corrupt log groups. */
+
+			if (scanned_lsn > *contiguous_lsn) {
+				*contiguous_lsn = scanned_lsn;
+			}
+		}
+
+		data_len = log_block_get_data_len(log_block);
+
+		if ((store_to_hash || (data_len == OS_FILE_LOG_BLOCK_SIZE))
+		    && scanned_lsn + data_len > recv_sys->scanned_lsn
+		    && (recv_sys->scanned_checkpoint_no > 0)
+		    && (log_block_get_checkpoint_no(log_block)
+			< recv_sys->scanned_checkpoint_no)
+		    && (recv_sys->scanned_checkpoint_no
+			- log_block_get_checkpoint_no(log_block)
+			> 0x80000000UL)) {
+
+			/* Garbage from a log buffer flush which was made
+			before the most recent database recovery */
+
+			finished = TRUE;
+#ifdef UNIV_LOG_DEBUG
+			/* This is not really an error, but currently
+			we stop here in the debug version: */
+
+			ut_error;
+#endif
+			break;
+		}
+
+		if (!recv_sys->parse_start_lsn
+		    && (log_block_get_first_rec_group(log_block) > 0)) {
+
+			/* We found a point from which to start the parsing
+			of log records */
+
+			recv_sys->parse_start_lsn = scanned_lsn
+				+ log_block_get_first_rec_group(log_block);
+			recv_sys->scanned_lsn = recv_sys->parse_start_lsn;
+			recv_sys->recovered_lsn = recv_sys->parse_start_lsn;
+		}
+
+		scanned_lsn += data_len;
+
+		if (scanned_lsn > recv_sys->scanned_lsn) {
+
+			/* We have found more entries. If this scan is
+ 			of startup type, we must initiate crash recovery
+			environment before parsing these log records. */
+
+			if (recv_log_scan_is_startup_type
+			    && !recv_needed_recovery) {
+
+				fprintf(stderr,
+					"InnoDB: Log scan progressed"
+					" past the checkpoint lsn %llu\n",
+					recv_sys->scanned_lsn);
+				recv_init_crash_recovery();
+			}
+
+			/* We were able to find more log data: add it to the
+			parsing buffer if parse_start_lsn is already
+			non-zero */
+
+			if (recv_sys->len + 4 * OS_FILE_LOG_BLOCK_SIZE
+			    >= RECV_PARSING_BUF_SIZE) {
+				fprintf(stderr,
+					"InnoDB: Error: log parsing"
+					" buffer overflow."
+					" Recovery may have failed!\n");
+
+				recv_sys->found_corrupt_log = TRUE;
+
+			} else if (!recv_sys->found_corrupt_log) {
+				more_data = recv_sys_add_to_parsing_buf(
+					log_block, scanned_lsn);
+			}
+
+			recv_sys->scanned_lsn = scanned_lsn;
+			recv_sys->scanned_checkpoint_no
+				= log_block_get_checkpoint_no(log_block);
+		}
+
+		if (data_len < OS_FILE_LOG_BLOCK_SIZE) {
+			/* Log data for this group ends here */
+
+			finished = TRUE;
+		} else {
+			log_block += OS_FILE_LOG_BLOCK_SIZE;
+		}
+	}
+
+	*group_scanned_lsn = scanned_lsn;
+
+	if (recv_needed_recovery
+	    || (recv_is_from_backup && !recv_is_making_a_backup)) {
+		recv_scan_print_counter++;
+
+		if (finished || (recv_scan_print_counter % 80 == 0)) {
+
+			fprintf(stderr,
+				"InnoDB: Doing recovery: scanned up to"
+				" log sequence number %llu\n",
+				*group_scanned_lsn);
+		}
+	}
+
+	if (more_data && !recv_sys->found_corrupt_log) {
+		/* Try to parse more log records */
+
+		recv_parse_log_recs(store_to_hash);
+
+		if (store_to_hash && mem_heap_get_size(recv_sys->heap)
+		    > available_memory
+		    && apply_automatically) {
+
+			/* Hash table of log records has grown too big:
+			empty it; FALSE means no ibuf operations
+			allowed, as we cannot add new records to the
+			log yet: they would be produced by ibuf
+			operations */
+
+			recv_apply_hashed_log_recs(FALSE);
+		}
+
+		if (recv_sys->recovered_offset > RECV_PARSING_BUF_SIZE / 4) {
+			/* Move parsing buffer data to the buffer start */
+
+			recv_sys_justify_left_parsing_buf();
+		}
+	}
+
+	return(finished);
+}
+
+/***********************************************************
+Scans log from a buffer and stores new log data to the parsing buffer. Parses
+and hashes the log records if new data found. */
+static
+void
+recv_group_scan_log_recs(
+/*=====================*/
+	log_group_t*	group,		/* in: log group */
+	ib_uint64_t*	contiguous_lsn,	/* in/out: it is known that all log
+					groups contain contiguous log data up
+					to this lsn */
+	ib_uint64_t*	group_scanned_lsn)/* out: scanning succeeded up to
+					this lsn */
+{
+	ibool		finished;
+	ib_uint64_t	start_lsn;
+	ib_uint64_t	end_lsn;
+
+	finished = FALSE;
+
+	start_lsn = *contiguous_lsn;
+
+	while (!finished) {
+		end_lsn = start_lsn + RECV_SCAN_SIZE;
+
+		log_group_read_log_seg(LOG_RECOVER, log_sys->buf,
+				       group, start_lsn, end_lsn);
+
+		finished = recv_scan_log_recs(
+			TRUE, (buf_pool->curr_size - recv_n_pool_free_frames)
+			* UNIV_PAGE_SIZE, TRUE, log_sys->buf, RECV_SCAN_SIZE,
+			start_lsn, contiguous_lsn, group_scanned_lsn);
+		start_lsn = end_lsn;
+	}
+
+#ifdef UNIV_DEBUG
+	if (log_debug_writes) {
+		fprintf(stderr,
+			"InnoDB: Scanned group %lu up to"
+			" log sequence number %llu\n",
+			(ulong) group->id,
+			*group_scanned_lsn);
+	}
+#endif /* UNIV_DEBUG */
+}
+
+/***********************************************************
+Initialize crash recovery environment. Can be called iff
+recv_needed_recovery == FALSE. */
+static
+void
+recv_init_crash_recovery(void)
+/*==========================*/
+{
+	ut_a(!recv_needed_recovery);
+
+	recv_needed_recovery = TRUE;
+
+	ut_print_timestamp(stderr);
+
+	fprintf(stderr,
+		"  InnoDB: Database was not"
+		" shut down normally!\n"
+		"InnoDB: Starting crash recovery.\n");
+
+	fprintf(stderr,
+		"InnoDB: Reading tablespace information"
+		" from the .ibd files...\n");
+
+	fil_load_single_table_tablespaces();
+
+	/* If we are using the doublewrite method, we will
+	check if there are half-written pages in data files,
+	and restore them from the doublewrite buffer if
+	possible */
+
+	if (srv_force_recovery < SRV_FORCE_NO_LOG_REDO) {
+
+		fprintf(stderr,
+			"InnoDB: Restoring possible"
+			" half-written data pages from"
+			" the doublewrite\n"
+			"InnoDB: buffer...\n");
+		trx_sys_doublewrite_init_or_restore_pages(TRUE);
+	}
+}
+
+/************************************************************
+Recovers from a checkpoint. When this function returns, the database is able
+to start processing of new user transactions, but the function
+recv_recovery_from_checkpoint_finish should be called later to complete
+the recovery and free the resources used in it. */
+UNIV_INTERN
+ulint
+recv_recovery_from_checkpoint_start_func(
+/*=====================================*/
+					/* out: error code or DB_SUCCESS */
+#ifdef UNIV_LOG_ARCHIVE
+	ulint		type,		/* in: LOG_CHECKPOINT or LOG_ARCHIVE */
+	ib_uint64_t	limit_lsn,	/* in: recover up to this lsn
+					if possible */
+#endif /* UNIV_LOG_ARCHIVE */
+	ib_uint64_t	min_flushed_lsn,/* in: min flushed lsn from
+					data files */
+	ib_uint64_t	max_flushed_lsn)/* in: max flushed lsn from
+					data files */
+{
+	log_group_t*	group;
+	log_group_t*	max_cp_group;
+	log_group_t*	up_to_date_group;
+	ulint		max_cp_field;
+	ib_uint64_t	checkpoint_lsn;
+	ib_uint64_t	checkpoint_no;
+	ib_uint64_t	old_scanned_lsn;
+	ib_uint64_t	group_scanned_lsn;
+	ib_uint64_t	contiguous_lsn;
+	ib_uint64_t	archived_lsn;
+	byte*		buf;
+	byte		log_hdr_buf[LOG_FILE_HDR_SIZE];
+	ulint		err;
+
+#ifdef UNIV_LOG_ARCHIVE
+	ut_ad(type != LOG_CHECKPOINT || limit_lsn == IB_ULONGLONG_MAX);
+# define TYPE_CHECKPOINT	(type == LOG_CHECKPOINT)
+# define LIMIT_LSN		limit_lsn
+#else /* UNIV_LOG_ARCHIVE */
+# define TYPE_CHECKPOINT	1
+# define LIMIT_LSN		IB_ULONGLONG_MAX
+#endif /* UNIV_LOG_ARCHIVE */
+
+	if (TYPE_CHECKPOINT) {
+		recv_sys_create();
+		recv_sys_init(FALSE, buf_pool_get_curr_size());
+	}
+
+	if (srv_force_recovery >= SRV_FORCE_NO_LOG_REDO) {
+		fprintf(stderr,
+			"InnoDB: The user has set SRV_FORCE_NO_LOG_REDO on\n");
+		fprintf(stderr,
+			"InnoDB: Skipping log redo\n");
+
+		return(DB_SUCCESS);
+	}
+
+	recv_recovery_on = TRUE;
+
+	recv_sys->limit_lsn = LIMIT_LSN;
+
+	mutex_enter(&(log_sys->mutex));
+
+	/* Look for the latest checkpoint from any of the log groups */
+
+	err = recv_find_max_checkpoint(&max_cp_group, &max_cp_field);
+
+	if (err != DB_SUCCESS) {
+
+		mutex_exit(&(log_sys->mutex));
+
+		return(err);
+	}
+
+	log_group_read_checkpoint_info(max_cp_group, max_cp_field);
+
+	buf = log_sys->checkpoint_buf;
+
+	checkpoint_lsn = mach_read_ull(buf + LOG_CHECKPOINT_LSN);
+	checkpoint_no = mach_read_ull(buf + LOG_CHECKPOINT_NO);
+	archived_lsn = mach_read_ull(buf + LOG_CHECKPOINT_ARCHIVED_LSN);
+
+	/* Read the first log file header to print a note if this is
+	a recovery from a restored InnoDB Hot Backup */
+
+	fil_io(OS_FILE_READ | OS_FILE_LOG, TRUE, max_cp_group->space_id, 0,
+	       0, 0, LOG_FILE_HDR_SIZE,
+	       log_hdr_buf, max_cp_group);
+
+	if (0 == ut_memcmp(log_hdr_buf + LOG_FILE_WAS_CREATED_BY_HOT_BACKUP,
+			   (byte*)"ibbackup", (sizeof "ibbackup") - 1)) {
+		/* This log file was created by ibbackup --restore: print
+		a note to the user about it */
+
+		fprintf(stderr,
+			"InnoDB: The log file was created by"
+			" ibbackup --apply-log at\n"
+			"InnoDB: %s\n",
+			log_hdr_buf + LOG_FILE_WAS_CREATED_BY_HOT_BACKUP);
+		fprintf(stderr,
+			"InnoDB: NOTE: the following crash recovery"
+			" is part of a normal restore.\n");
+
+		/* Wipe over the label now */
+
+		memset(log_hdr_buf + LOG_FILE_WAS_CREATED_BY_HOT_BACKUP,
+		       ' ', 4);
+		/* Write to the log file to wipe over the label */
+		fil_io(OS_FILE_WRITE | OS_FILE_LOG, TRUE,
+		       max_cp_group->space_id, 0,
+		       0, 0, OS_FILE_LOG_BLOCK_SIZE,
+		       log_hdr_buf, max_cp_group);
+	}
+
+#ifdef UNIV_LOG_ARCHIVE
+	group = UT_LIST_GET_FIRST(log_sys->log_groups);
+
+	while (group) {
+		log_checkpoint_get_nth_group_info(buf, group->id,
+						  &(group->archived_file_no),
+						  &(group->archived_offset));
+
+		group = UT_LIST_GET_NEXT(log_groups, group);
+	}
+#endif /* UNIV_LOG_ARCHIVE */
+
+	if (TYPE_CHECKPOINT) {
+		/* Start reading the log groups from the checkpoint lsn up. The
+		variable contiguous_lsn contains an lsn up to which the log is
+		known to be contiguously written to all log groups. */
+
+		recv_sys->parse_start_lsn = checkpoint_lsn;
+		recv_sys->scanned_lsn = checkpoint_lsn;
+		recv_sys->scanned_checkpoint_no = 0;
+		recv_sys->recovered_lsn = checkpoint_lsn;
+
+		srv_start_lsn = checkpoint_lsn;
+	}
+
+	contiguous_lsn = ut_uint64_align_down(recv_sys->scanned_lsn,
+					      OS_FILE_LOG_BLOCK_SIZE);
+	if (TYPE_CHECKPOINT) {
+		up_to_date_group = max_cp_group;
+#ifdef UNIV_LOG_ARCHIVE
+	} else {
+		ulint	capacity;
+
+		/* Try to recover the remaining part from logs: first from
+		the logs of the archived group */
+
+		group = recv_sys->archive_group;
+		capacity = log_group_get_capacity(group);
+
+		if (recv_sys->scanned_lsn > checkpoint_lsn + capacity
+		    || checkpoint_lsn > recv_sys->scanned_lsn + capacity) {
+
+			mutex_exit(&(log_sys->mutex));
+
+			/* The group does not contain enough log: probably
+			an archived log file was missing or corrupt */
+
+			return(DB_ERROR);
+		}
+
+		recv_group_scan_log_recs(group, &contiguous_lsn,
+					 &group_scanned_lsn);
+		if (recv_sys->scanned_lsn < checkpoint_lsn) {
+
+			mutex_exit(&(log_sys->mutex));
+
+			/* The group did not contain enough log: an archived
+			log file was missing or invalid, or the log group
+			was corrupt */
+
+			return(DB_ERROR);
+		}
+
+		group->scanned_lsn = group_scanned_lsn;
+		up_to_date_group = group;
+#endif /* UNIV_LOG_ARCHIVE */
+	}
+
+	ut_ad(RECV_SCAN_SIZE <= log_sys->buf_size);
+
+	group = UT_LIST_GET_FIRST(log_sys->log_groups);
+
+#ifdef UNIV_LOG_ARCHIVE
+	if ((type == LOG_ARCHIVE) && (group == recv_sys->archive_group)) {
+		group = UT_LIST_GET_NEXT(log_groups, group);
+	}
+#endif /* UNIV_LOG_ARCHIVE */
+
+	/* Set the flag to publish that we are doing startup scan. */
+	recv_log_scan_is_startup_type = TYPE_CHECKPOINT;
+	while (group) {
+		old_scanned_lsn = recv_sys->scanned_lsn;
+
+		recv_group_scan_log_recs(group, &contiguous_lsn,
+					 &group_scanned_lsn);
+		group->scanned_lsn = group_scanned_lsn;
+
+		if (old_scanned_lsn < group_scanned_lsn) {
+			/* We found a more up-to-date group */
+
+			up_to_date_group = group;
+		}
+
+#ifdef UNIV_LOG_ARCHIVE
+		if ((type == LOG_ARCHIVE)
+		    && (group == recv_sys->archive_group)) {
+			group = UT_LIST_GET_NEXT(log_groups, group);
+		}
+#endif /* UNIV_LOG_ARCHIVE */
+
+		group = UT_LIST_GET_NEXT(log_groups, group);
+	}
+
+	/* Done with startup scan. Clear the flag. */
+	recv_log_scan_is_startup_type = FALSE;
+	if (TYPE_CHECKPOINT) {
+		/* NOTE: we always do a 'recovery' at startup, but only if
+		there is something wrong we will print a message to the
+		user about recovery: */
+
+		if (checkpoint_lsn != max_flushed_lsn
+		    || checkpoint_lsn != min_flushed_lsn) {
+
+			if (checkpoint_lsn < max_flushed_lsn) {
+				fprintf(stderr,
+					"InnoDB: #########################"
+					"#################################\n"
+					"InnoDB:                          "
+					"WARNING!\n"
+					"InnoDB: The log sequence number"
+					" in ibdata files is higher\n"
+					"InnoDB: than the log sequence number"
+					" in the ib_logfiles! Are you sure\n"
+					"InnoDB: you are using the right"
+					" ib_logfiles to start up"
+					" the database?\n"
+					"InnoDB: Log sequence number in"
+					" ib_logfiles is %llu, log\n"
+					"InnoDB: sequence numbers stamped"
+					" to ibdata file headers are between\n"
+					"InnoDB: %llu and %llu.\n"
+					"InnoDB: #########################"
+					"#################################\n",
+					checkpoint_lsn,
+					min_flushed_lsn,
+					max_flushed_lsn);
+			}
+
+			if (!recv_needed_recovery) {
+				fprintf(stderr,
+					"InnoDB: The log sequence number"
+					" in ibdata files does not match\n"
+					"InnoDB: the log sequence number"
+					" in the ib_logfiles!\n");
+				recv_init_crash_recovery();
+			}
+		}
+
+		if (!recv_needed_recovery) {
+			/* Init the doublewrite buffer memory structure */
+			trx_sys_doublewrite_init_or_restore_pages(FALSE);
+		}
+	}
+
+	/* We currently have only one log group */
+	if (group_scanned_lsn < checkpoint_lsn) {
+		ut_print_timestamp(stderr);
+		fprintf(stderr,
+			"  InnoDB: ERROR: We were only able to scan the log"
+			" up to\n"
+			"InnoDB: %llu, but a checkpoint was at %llu.\n"
+			"InnoDB: It is possible that"
+			" the database is now corrupt!\n",
+			group_scanned_lsn,
+			checkpoint_lsn);
+	}
+
+	if (group_scanned_lsn < recv_max_page_lsn) {
+		ut_print_timestamp(stderr);
+		fprintf(stderr,
+			"  InnoDB: ERROR: We were only able to scan the log"
+			" up to %llu\n"
+			"InnoDB: but a database page a had an lsn %llu."
+			" It is possible that the\n"
+			"InnoDB: database is now corrupt!\n",
+			group_scanned_lsn,
+			recv_max_page_lsn);
+	}
+
+	if (recv_sys->recovered_lsn < checkpoint_lsn) {
+
+		mutex_exit(&(log_sys->mutex));
+
+		if (recv_sys->recovered_lsn >= LIMIT_LSN) {
+
+			return(DB_SUCCESS);
+		}
+
+		ut_error;
+
+		return(DB_ERROR);
+	}
+
+	/* Synchronize the uncorrupted log groups to the most up-to-date log
+	group; we also copy checkpoint info to groups */
+
+	log_sys->next_checkpoint_lsn = checkpoint_lsn;
+	log_sys->next_checkpoint_no = checkpoint_no + 1;
+
+#ifdef UNIV_LOG_ARCHIVE
+	log_sys->archived_lsn = archived_lsn;
+#endif /* UNIV_LOG_ARCHIVE */
+
+	recv_synchronize_groups(up_to_date_group);
+
+	if (!recv_needed_recovery) {
+		ut_a(checkpoint_lsn == recv_sys->recovered_lsn);
+	} else {
+		srv_start_lsn = recv_sys->recovered_lsn;
+	}
+
+	log_sys->lsn = recv_sys->recovered_lsn;
+
+	ut_memcpy(log_sys->buf, recv_sys->last_block, OS_FILE_LOG_BLOCK_SIZE);
+
+	log_sys->buf_free = (ulint) log_sys->lsn % OS_FILE_LOG_BLOCK_SIZE;
+	log_sys->buf_next_to_write = log_sys->buf_free;
+	log_sys->written_to_some_lsn = log_sys->lsn;
+	log_sys->written_to_all_lsn = log_sys->lsn;
+
+	log_sys->last_checkpoint_lsn = checkpoint_lsn;
+
+	log_sys->next_checkpoint_no = checkpoint_no + 1;
+
+#ifdef UNIV_LOG_ARCHIVE
+	if (archived_lsn == IB_ULONGLONG_MAX) {
+
+		log_sys->archiving_state = LOG_ARCH_OFF;
+	}
+#endif /* UNIV_LOG_ARCHIVE */
+
+	mutex_enter(&(recv_sys->mutex));
+
+	recv_sys->apply_log_recs = TRUE;
+
+	mutex_exit(&(recv_sys->mutex));
+
+	mutex_exit(&(log_sys->mutex));
+
+	recv_lsn_checks_on = TRUE;
+
+	/* The database is now ready to start almost normal processing of user
+	transactions: transaction rollbacks and the application of the log
+	records in the hash table can be run in background. */
+
+	return(DB_SUCCESS);
+
+#undef TYPE_CHECKPOINT
+#undef LIMIT_LSN
+}
+
+/************************************************************
+Completes recovery from a checkpoint. */
+UNIV_INTERN
+void
+recv_recovery_from_checkpoint_finish(void)
+/*======================================*/
+{
+	int		i;
+
+	/* Apply the hashed log records to the respective file pages */
+
+	if (srv_force_recovery < SRV_FORCE_NO_LOG_REDO) {
+
+		recv_apply_hashed_log_recs(TRUE);
+	}
+
+#ifdef UNIV_DEBUG
+	if (log_debug_writes) {
+		fprintf(stderr,
+			"InnoDB: Log records applied to the database\n");
+	}
+#endif /* UNIV_DEBUG */
+
+	if (recv_needed_recovery) {
+		trx_sys_print_mysql_master_log_pos();
+		trx_sys_print_mysql_binlog_offset();
+	}
+
+	if (recv_sys->found_corrupt_log) {
+
+		fprintf(stderr,
+			"InnoDB: WARNING: the log file may have been"
+			" corrupt and it\n"
+			"InnoDB: is possible that the log scan or parsing"
+			" did not proceed\n"
+			"InnoDB: far enough in recovery. Please run"
+			" CHECK TABLE\n"
+			"InnoDB: on your InnoDB tables to check that"
+			" they are ok!\n"
+			"InnoDB: It may be safest to recover your"
+			" InnoDB database from\n"
+			"InnoDB: a backup!\n");
+	}
+
+	/* Free the resources of the recovery system */
+
+	recv_recovery_on = FALSE;
+
+#ifndef UNIV_LOG_DEBUG
+	recv_sys_free();
+#endif
+
+	/* Drop partially created indexes. */
+	row_merge_drop_temp_indexes();
+
+#ifdef UNIV_SYNC_DEBUG
+	/* Wait for a while so that created threads have time to suspend
+	themselves before we switch the latching order checks on */
+	os_thread_sleep(1000000);
+
+	/* Switch latching order checks on in sync0sync.c */
+	sync_order_checks_on = TRUE;
+#endif
+	if (srv_force_recovery < SRV_FORCE_NO_TRX_UNDO) {
+		/* Rollback the uncommitted transactions which have no user
+		session */
+
+		os_thread_create(trx_rollback_or_clean_all_recovered,
+				 (void *)&i, NULL);
+	}
+}
+
+/**********************************************************
+Resets the logs. The contents of log files will be lost! */
+UNIV_INTERN
+void
+recv_reset_logs(
+/*============*/
+	ib_uint64_t	lsn,		/* in: reset to this lsn
+					rounded up to be divisible by
+					OS_FILE_LOG_BLOCK_SIZE, after
+					which we add
+					LOG_BLOCK_HDR_SIZE */
+#ifdef UNIV_LOG_ARCHIVE
+	ulint		arch_log_no,	/* in: next archived log file number */
+#endif /* UNIV_LOG_ARCHIVE */
+	ibool		new_logs_created)/* in: TRUE if resetting logs
+					is done at the log creation;
+					FALSE if it is done after
+					archive recovery */
+{
+	log_group_t*	group;
+
+	ut_ad(mutex_own(&(log_sys->mutex)));
+
+	log_sys->lsn = ut_uint64_align_up(lsn, OS_FILE_LOG_BLOCK_SIZE);
+
+	group = UT_LIST_GET_FIRST(log_sys->log_groups);
+
+	while (group) {
+		group->lsn = log_sys->lsn;
+		group->lsn_offset = LOG_FILE_HDR_SIZE;
+#ifdef UNIV_LOG_ARCHIVE
+		group->archived_file_no = arch_log_no;
+		group->archived_offset = 0;
+#endif /* UNIV_LOG_ARCHIVE */
+
+		if (!new_logs_created) {
+			recv_truncate_group(group, group->lsn, group->lsn,
+					    group->lsn, group->lsn);
+		}
+
+		group = UT_LIST_GET_NEXT(log_groups, group);
+	}
+
+	log_sys->buf_next_to_write = 0;
+	log_sys->written_to_some_lsn = log_sys->lsn;
+	log_sys->written_to_all_lsn = log_sys->lsn;
+
+	log_sys->next_checkpoint_no = 0;
+	log_sys->last_checkpoint_lsn = 0;
+
+#ifdef UNIV_LOG_ARCHIVE
+	log_sys->archived_lsn = log_sys->lsn;
+#endif /* UNIV_LOG_ARCHIVE */
+
+	log_block_init(log_sys->buf, log_sys->lsn);
+	log_block_set_first_rec_group(log_sys->buf, LOG_BLOCK_HDR_SIZE);
+
+	log_sys->buf_free = LOG_BLOCK_HDR_SIZE;
+	log_sys->lsn += LOG_BLOCK_HDR_SIZE;
+
+	mutex_exit(&(log_sys->mutex));
+
+	/* Reset the checkpoint fields in logs */
+
+	log_make_checkpoint_at(IB_ULONGLONG_MAX, TRUE);
+	log_make_checkpoint_at(IB_ULONGLONG_MAX, TRUE);
+
+	mutex_enter(&(log_sys->mutex));
+}
+
+#ifdef UNIV_HOTBACKUP
+/**********************************************************
+Creates new log files after a backup has been restored. */
+UNIV_INTERN
+void
+recv_reset_log_files_for_backup(
+/*============================*/
+	const char*	log_dir,	/* in: log file directory path */
+	ulint		n_log_files,	/* in: number of log files */
+	ulint		log_file_size,	/* in: log file size */
+	ib_uint64_t	lsn)		/* in: new start lsn, must be
+					divisible by OS_FILE_LOG_BLOCK_SIZE */
+{
+	os_file_t	log_file;
+	ibool		success;
+	byte*		buf;
+	ulint		i;
+	ulint		log_dir_len;
+	char		name[5000];
+	static const char ib_logfile_basename[] = "ib_logfile";
+
+	log_dir_len = strlen(log_dir);
+	/* full path name of ib_logfile consists of log dir path + basename
+	+ number. This must fit in the name buffer.
+	*/
+	ut_a(log_dir_len + strlen(ib_logfile_basename) + 11  < sizeof(name));
+
+	buf = ut_malloc(LOG_FILE_HDR_SIZE + OS_FILE_LOG_BLOCK_SIZE);
+	memset(buf, '\0', LOG_FILE_HDR_SIZE + OS_FILE_LOG_BLOCK_SIZE);
+
+	for (i = 0; i < n_log_files; i++) {
+
+		sprintf(name, "%s%s%lu", log_dir,
+			ib_logfile_basename, (ulong)i);
+
+		log_file = os_file_create_simple(name, OS_FILE_CREATE,
+						 OS_FILE_READ_WRITE, &success);
+		if (!success) {
+			fprintf(stderr,
+				"InnoDB: Cannot create %s. Check that"
+				" the file does not exist yet.\n", name);
+
+			exit(1);
+		}
+
+		fprintf(stderr,
+			"Setting log file size to %lu %lu\n",
+			(ulong) ut_get_high32(log_file_size),
+			(ulong) log_file_size & 0xFFFFFFFFUL);
+
+		success = os_file_set_size(name, log_file,
+					   log_file_size & 0xFFFFFFFFUL,
+					   ut_get_high32(log_file_size));
+
+		if (!success) {
+			fprintf(stderr,
+				"InnoDB: Cannot set %s size to %lu %lu\n",
+				name, (ulong) ut_get_high32(log_file_size),
+				(ulong) (log_file_size & 0xFFFFFFFFUL));
+			exit(1);
+		}
+
+		os_file_flush(log_file);
+		os_file_close(log_file);
+	}
+
+	/* We pretend there is a checkpoint at lsn + LOG_BLOCK_HDR_SIZE */
+
+	log_reset_first_header_and_checkpoint(buf, lsn);
+
+	log_block_init_in_old_format(buf + LOG_FILE_HDR_SIZE, lsn);
+	log_block_set_first_rec_group(buf + LOG_FILE_HDR_SIZE,
+				      LOG_BLOCK_HDR_SIZE);
+	sprintf(name, "%s%s%lu", log_dir, ib_logfile_basename, (ulong)0);
+
+	log_file = os_file_create_simple(name, OS_FILE_OPEN,
+					 OS_FILE_READ_WRITE, &success);
+	if (!success) {
+		fprintf(stderr, "InnoDB: Cannot open %s.\n", name);
+
+		exit(1);
+	}
+
+	os_file_write(name, log_file, buf, 0, 0,
+		      LOG_FILE_HDR_SIZE + OS_FILE_LOG_BLOCK_SIZE);
+	os_file_flush(log_file);
+	os_file_close(log_file);
+
+	ut_free(buf);
+}
+#endif /* UNIV_HOTBACKUP */
+
+#ifdef UNIV_LOG_ARCHIVE
+/**********************************************************
+Reads from the archive of a log group and performs recovery. */
+static
+ibool
+log_group_recover_from_archive_file(
+/*================================*/
+					/* out: TRUE if no more complete
+					consistent archive files */
+	log_group_t*	group)		/* in: log group */
+{
+	os_file_t	file_handle;
+	ib_uint64_t	start_lsn;
+	ib_uint64_t	file_end_lsn;
+	ib_uint64_t	dummy_lsn;
+	ib_uint64_t	scanned_lsn;
+	ulint		len;
+	ibool		ret;
+	byte*		buf;
+	ulint		read_offset;
+	ulint		file_size;
+	ulint		file_size_high;
+	int		input_char;
+	char		name[10000];
+
+	ut_a(0);
+
+try_open_again:
+	buf = log_sys->buf;
+
+	/* Add the file to the archive file space; open the file */
+
+	log_archived_file_name_gen(name, group->id, group->archived_file_no);
+
+	file_handle = os_file_create(name, OS_FILE_OPEN,
+				     OS_FILE_LOG, OS_FILE_AIO, &ret);
+
+	if (ret == FALSE) {
+ask_again:
+		fprintf(stderr,
+			"InnoDB: Do you want to copy additional"
+			" archived log files\n"
+			"InnoDB: to the directory\n");
+		fprintf(stderr,
+			"InnoDB: or were these all the files needed"
+			" in recovery?\n");
+		fprintf(stderr,
+			"InnoDB: (Y == copy more files; N == this is all)?");
+
+		input_char = getchar();
+
+		if (input_char == (int) 'N') {
+
+			return(TRUE);
+		} else if (input_char == (int) 'Y') {
+
+			goto try_open_again;
+		} else {
+			goto ask_again;
+		}
+	}
+
+	ret = os_file_get_size(file_handle, &file_size, &file_size_high);
+	ut_a(ret);
+
+	ut_a(file_size_high == 0);
+
+	fprintf(stderr, "InnoDB: Opened archived log file %s\n", name);
+
+	ret = os_file_close(file_handle);
+
+	if (file_size < LOG_FILE_HDR_SIZE) {
+		fprintf(stderr,
+			"InnoDB: Archive file header incomplete %s\n", name);
+
+		return(TRUE);
+	}
+
+	ut_a(ret);
+
+	/* Add the archive file as a node to the space */
+
+	fil_node_create(name, 1 + file_size / UNIV_PAGE_SIZE,
+			group->archive_space_id, FALSE);
+#if RECV_SCAN_SIZE < LOG_FILE_HDR_SIZE
+# error "RECV_SCAN_SIZE < LOG_FILE_HDR_SIZE"
+#endif
+
+	/* Read the archive file header */
+	fil_io(OS_FILE_READ | OS_FILE_LOG, TRUE, group->archive_space_id, 0, 0,
+	       LOG_FILE_HDR_SIZE, buf, NULL);
+
+	/* Check if the archive file header is consistent */
+
+	if (mach_read_from_4(buf + LOG_GROUP_ID) != group->id
+	    || mach_read_from_4(buf + LOG_FILE_NO)
+	    != group->archived_file_no) {
+		fprintf(stderr,
+			"InnoDB: Archive file header inconsistent %s\n", name);
+
+		return(TRUE);
+	}
+
+	if (!mach_read_from_4(buf + LOG_FILE_ARCH_COMPLETED)) {
+		fprintf(stderr,
+			"InnoDB: Archive file not completely written %s\n",
+			name);
+
+		return(TRUE);
+	}
+
+	start_lsn = mach_read_ull(buf + LOG_FILE_START_LSN);
+	file_end_lsn = mach_read_ull(buf + LOG_FILE_END_LSN);
+
+	if (!recv_sys->scanned_lsn) {
+
+		if (recv_sys->parse_start_lsn < start_lsn) {
+			fprintf(stderr,
+				"InnoDB: Archive log file %s"
+				" starts from too big a lsn\n",
+				name);
+			return(TRUE);
+		}
+
+		recv_sys->scanned_lsn = start_lsn;
+	}
+
+	if (recv_sys->scanned_lsn != start_lsn) {
+
+		fprintf(stderr,
+			"InnoDB: Archive log file %s starts from"
+			" a wrong lsn\n",
+			name);
+		return(TRUE);
+	}
+
+	read_offset = LOG_FILE_HDR_SIZE;
+
+	for (;;) {
+		len = RECV_SCAN_SIZE;
+
+		if (read_offset + len > file_size) {
+			len = ut_calc_align_down(file_size - read_offset,
+						 OS_FILE_LOG_BLOCK_SIZE);
+		}
+
+		if (len == 0) {
+
+			break;
+		}
+
+#ifdef UNIV_DEBUG
+		if (log_debug_writes) {
+			fprintf(stderr,
+				"InnoDB: Archive read starting at"
+				" lsn %llu, len %lu from file %s\n",
+				start_lsn,
+				(ulong) len, name);
+		}
+#endif /* UNIV_DEBUG */
+
+		fil_io(OS_FILE_READ | OS_FILE_LOG, TRUE,
+		       group->archive_space_id, read_offset / UNIV_PAGE_SIZE,
+		       read_offset % UNIV_PAGE_SIZE, len, buf, NULL);
+
+		ret = recv_scan_log_recs(
+			TRUE, (buf_pool->n_frames - recv_n_pool_free_frames)
+			* UNIV_PAGE_SIZE, TRUE, buf, len, start_lsn,
+			&dummy_lsn, &scanned_lsn);
+
+		if (scanned_lsn == file_end_lsn) {
+
+			return(FALSE);
+		}
+
+		if (ret) {
+			fprintf(stderr,
+				"InnoDB: Archive log file %s"
+				" does not scan right\n",
+				name);
+			return(TRUE);
+		}
+
+		read_offset += len;
+		start_lsn += len;
+
+		ut_ad(start_lsn == scanned_lsn);
+	}
+
+	return(FALSE);
+}
+
+/************************************************************
+Recovers from archived log files, and also from log files, if they exist. */
+UNIV_INTERN
+ulint
+recv_recovery_from_archive_start(
+/*=============================*/
+					/* out: error code or DB_SUCCESS */
+	ib_uint64_t	min_flushed_lsn,/* in: min flushed lsn field from the
+					data files */
+	ib_uint64_t	limit_lsn,	/* in: recover up to this lsn if
+					possible */
+	ulint		first_log_no)	/* in: number of the first archived
+					log file to use in the recovery; the
+					file will be searched from
+					INNOBASE_LOG_ARCH_DIR specified in
+					server config file */
+{
+	log_group_t*	group;
+	ulint		group_id;
+	ulint		trunc_len;
+	ibool		ret;
+	ulint		err;
+
+	ut_a(0);
+
+	recv_sys_create();
+	recv_sys_init(FALSE, buf_pool_get_curr_size());
+
+	recv_recovery_on = TRUE;
+	recv_recovery_from_backup_on = TRUE;
+
+	recv_sys->limit_lsn = limit_lsn;
+
+	group_id = 0;
+
+	group = UT_LIST_GET_FIRST(log_sys->log_groups);
+
+	while (group) {
+		if (group->id == group_id) {
+
+			break;
+		}
+
+		group = UT_LIST_GET_NEXT(log_groups, group);
+	}
+
+	if (!group) {
+		fprintf(stderr,
+			"InnoDB: There is no log group defined with id %lu!\n",
+			(ulong) group_id);
+		return(DB_ERROR);
+	}
+
+	group->archived_file_no = first_log_no;
+
+	recv_sys->parse_start_lsn = min_flushed_lsn;
+
+	recv_sys->scanned_lsn = 0;
+	recv_sys->scanned_checkpoint_no = 0;
+	recv_sys->recovered_lsn = recv_sys->parse_start_lsn;
+
+	recv_sys->archive_group = group;
+
+	ret = FALSE;
+
+	mutex_enter(&(log_sys->mutex));
+
+	while (!ret) {
+		ret = log_group_recover_from_archive_file(group);
+
+		/* Close and truncate a possible processed archive file
+		from the file space */
+
+		trunc_len = UNIV_PAGE_SIZE
+			* fil_space_get_size(group->archive_space_id);
+		if (trunc_len > 0) {
+			fil_space_truncate_start(group->archive_space_id,
+						 trunc_len);
+		}
+
+		group->archived_file_no++;
+	}
+
+	if (recv_sys->recovered_lsn < limit_lsn) {
+
+		if (!recv_sys->scanned_lsn) {
+
+			recv_sys->scanned_lsn = recv_sys->parse_start_lsn;
+		}
+
+		mutex_exit(&(log_sys->mutex));
+
+		err = recv_recovery_from_checkpoint_start(LOG_ARCHIVE,
+							  limit_lsn,
+							  IB_ULONGLONG_MAX,
+							  IB_ULONGLONG_MAX);
+		if (err != DB_SUCCESS) {
+
+			return(err);
+		}
+
+		mutex_enter(&(log_sys->mutex));
+	}
+
+	if (limit_lsn != IB_ULONGLONG_MAX) {
+
+		recv_apply_hashed_log_recs(FALSE);
+
+		recv_reset_logs(recv_sys->recovered_lsn, 0, FALSE);
+	}
+
+	mutex_exit(&(log_sys->mutex));
+
+	return(DB_SUCCESS);
+}
+
+/************************************************************
+Completes recovery from archive. */
+UNIV_INTERN
+void
+recv_recovery_from_archive_finish(void)
+/*===================================*/
+{
+	recv_recovery_from_checkpoint_finish();
+
+	recv_recovery_from_backup_on = FALSE;
+}
+#endif /* UNIV_LOG_ARCHIVE */
diff --git a/storage/xtradb/mach/mach0data.c b/storage/xtradb/mach/mach0data.c
new file mode 100644
index 00000000000..5deb475318d
--- /dev/null
+++ b/storage/xtradb/mach/mach0data.c
@@ -0,0 +1,135 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/**********************************************************************
+Utilities for converting data from the database file
+to the machine format.
+
+Created 11/28/1995 Heikki Tuuri
+***********************************************************************/
+
+#include "mach0data.h"
+
+#ifdef UNIV_NONINL
+#include "mach0data.ic"
+#endif
+
+/*************************************************************
+Reads a ulint in a compressed form if the log record fully contains it. */
+UNIV_INTERN
+byte*
+mach_parse_compressed(
+/*==================*/
+			/* out: pointer to end of the stored field, NULL if
+			not complete */
+	byte*	ptr,	/* in: pointer to buffer from where to read */
+	byte*	end_ptr,/* in: pointer to end of the buffer */
+	ulint*	val)	/* out: read value (< 2^32) */
+{
+	ulint	flag;
+
+	ut_ad(ptr && end_ptr && val);
+
+	if (ptr >= end_ptr) {
+
+		return(NULL);
+	}
+
+	flag = mach_read_from_1(ptr);
+
+	if (flag < 0x80UL) {
+		*val = flag;
+		return(ptr + 1);
+
+	} else if (flag < 0xC0UL) {
+		if (end_ptr < ptr + 2) {
+			return(NULL);
+		}
+
+		*val = mach_read_from_2(ptr) & 0x7FFFUL;
+
+		return(ptr + 2);
+
+	} else if (flag < 0xE0UL) {
+		if (end_ptr < ptr + 3) {
+			return(NULL);
+		}
+
+		*val = mach_read_from_3(ptr) & 0x3FFFFFUL;
+
+		return(ptr + 3);
+	} else if (flag < 0xF0UL) {
+		if (end_ptr < ptr + 4) {
+			return(NULL);
+		}
+
+		*val = mach_read_from_4(ptr) & 0x1FFFFFFFUL;
+
+		return(ptr + 4);
+	} else {
+		ut_ad(flag == 0xF0UL);
+
+		if (end_ptr < ptr + 5) {
+			return(NULL);
+		}
+
+		*val = mach_read_from_4(ptr + 1);
+		return(ptr + 5);
+	}
+}
+
+/*************************************************************
+Reads a dulint in a compressed form if the log record fully contains it. */
+UNIV_INTERN
+byte*
+mach_dulint_parse_compressed(
+/*=========================*/
+			/* out: pointer to end of the stored field, NULL if
+			not complete */
+	byte*	ptr,	/* in: pointer to buffer from where to read */
+	byte*	end_ptr,/* in: pointer to end of the buffer */
+	dulint*	val)	/* out: read value */
+{
+	ulint	high;
+	ulint	low;
+	ulint	size;
+
+	ut_ad(ptr && end_ptr && val);
+
+	if (end_ptr < ptr + 5) {
+
+		return(NULL);
+	}
+
+	high = mach_read_compressed(ptr);
+
+	size = mach_get_compressed_size(high);
+
+	ptr += size;
+
+	if (end_ptr < ptr + 4) {
+
+		return(NULL);
+	}
+
+	low = mach_read_from_4(ptr);
+
+	*val = ut_dulint_create(high, low);
+
+	return(ptr + 4);
+}
diff --git a/storage/xtradb/mem/mem0dbg.c b/storage/xtradb/mem/mem0dbg.c
new file mode 100644
index 00000000000..a1647462922
--- /dev/null
+++ b/storage/xtradb/mem/mem0dbg.c
@@ -0,0 +1,1014 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/************************************************************************
+The memory management: the debug code. This is not a compilation module,
+but is included in mem0mem.* !
+
+Created 6/9/1994 Heikki Tuuri
+*************************************************************************/
+
+#ifdef UNIV_MEM_DEBUG
+/* The mutex which protects in the debug version the hash table
+containing the list of live memory heaps, and also the global
+variables below. */
+UNIV_INTERN mutex_t	mem_hash_mutex;
+
+/* The following variables contain information about the
+extent of memory allocations. Only used in the debug version.
+Protected by mem_hash_mutex above. */
+
+static ulint		mem_n_created_heaps		= 0;
+static ulint		mem_n_allocations		= 0;
+static ulint		mem_total_allocated_memory	= 0;
+UNIV_INTERN ulint	mem_current_allocated_memory	= 0;
+static ulint		mem_max_allocated_memory	= 0;
+static ulint		mem_last_print_info		= 0;
+
+/* Size of the hash table for memory management tracking */
+#define	MEM_HASH_SIZE	997
+
+/* The node of the list containing currently allocated memory heaps */
+
+typedef struct mem_hash_node_struct mem_hash_node_t;
+struct mem_hash_node_struct {
+	UT_LIST_NODE_T(mem_hash_node_t)
+				list;	/* hash list node */
+	mem_heap_t*		heap;	/* memory heap */
+	const char*		file_name;/* file where heap was created*/
+	ulint			line;	/* file line of creation */
+	ulint			nth_heap;/* this is the nth heap created */
+	UT_LIST_NODE_T(mem_hash_node_t)
+				all_list;/* list of all created heaps */
+};
+
+typedef UT_LIST_BASE_NODE_T(mem_hash_node_t) mem_hash_cell_t;
+
+/* The hash table of allocated heaps */
+static mem_hash_cell_t		mem_hash_table[MEM_HASH_SIZE];
+
+/* The base node of the list of all allocated heaps */
+static mem_hash_cell_t		mem_all_list_base;
+
+static ibool	mem_hash_initialized	= FALSE;
+
+
+UNIV_INLINE
+mem_hash_cell_t*
+mem_hash_get_nth_cell(ulint i);
+
+/* Accessor function for the hash table. Returns a pointer to the
+table cell. */
+UNIV_INLINE
+mem_hash_cell_t*
+mem_hash_get_nth_cell(ulint i)
+{
+	ut_a(i < MEM_HASH_SIZE);
+
+	return(&(mem_hash_table[i]));
+}
+
+/* Accessor functions for a memory field in the debug version */
+UNIV_INTERN
+void
+mem_field_header_set_len(byte* field, ulint len)
+{
+	mach_write_to_4(field - 2 * sizeof(ulint), len);
+}
+
+UNIV_INTERN
+ulint
+mem_field_header_get_len(byte* field)
+{
+	return(mach_read_from_4(field - 2 * sizeof(ulint)));
+}
+
+UNIV_INTERN
+void
+mem_field_header_set_check(byte* field, ulint check)
+{
+	mach_write_to_4(field - sizeof(ulint), check);
+}
+
+UNIV_INTERN
+ulint
+mem_field_header_get_check(byte* field)
+{
+	return(mach_read_from_4(field - sizeof(ulint)));
+}
+
+UNIV_INTERN
+void
+mem_field_trailer_set_check(byte* field, ulint check)
+{
+	mach_write_to_4(field + mem_field_header_get_len(field), check);
+}
+
+UNIV_INTERN
+ulint
+mem_field_trailer_get_check(byte* field)
+{
+	return(mach_read_from_4(field
+				+ mem_field_header_get_len(field)));
+}
+#endif /* UNIV_MEM_DEBUG */
+
+/**********************************************************************
+Initializes the memory system. */
+UNIV_INTERN
+void
+mem_init(
+/*=====*/
+	ulint	size)	/* in: common pool size in bytes */
+{
+#ifdef UNIV_MEM_DEBUG
+
+	ulint	i;
+
+	/* Initialize the hash table */
+	ut_a(FALSE == mem_hash_initialized);
+
+	mutex_create(&mem_hash_mutex, SYNC_MEM_HASH);
+
+	for (i = 0; i < MEM_HASH_SIZE; i++) {
+		UT_LIST_INIT(*mem_hash_get_nth_cell(i));
+	}
+
+	UT_LIST_INIT(mem_all_list_base);
+
+	mem_hash_initialized = TRUE;
+#endif
+
+	if (UNIV_LIKELY(srv_use_sys_malloc)) {
+		/* When innodb_use_sys_malloc is set, the
+		mem_comm_pool won't be used for any allocations.  We
+		create a dummy mem_comm_pool, because some statistics
+		and debugging code relies on it being initialized. */
+		size = 1;
+	}
+
+	mem_comm_pool = mem_pool_create(size);
+}
+
+#ifdef UNIV_MEM_DEBUG
+/**********************************************************************
+Initializes an allocated memory field in the debug version. */
+UNIV_INTERN
+void
+mem_field_init(
+/*===========*/
+	byte*	buf,	/* in: memory field */
+	ulint	n)	/* in: how many bytes the user requested */
+{
+	ulint	rnd;
+	byte*	usr_buf;
+
+	usr_buf = buf + MEM_FIELD_HEADER_SIZE;
+
+	/* In the debug version write the length field and the
+	check fields to the start and the end of the allocated storage.
+	The field header consists of a length field and
+	a random number field, in this order. The field trailer contains
+	the same random number as a check field. */
+
+	mem_field_header_set_len(usr_buf, n);
+
+	rnd = ut_rnd_gen_ulint();
+
+	mem_field_header_set_check(usr_buf, rnd);
+	mem_field_trailer_set_check(usr_buf, rnd);
+
+	/* Update the memory allocation information */
+
+	mutex_enter(&mem_hash_mutex);
+
+	mem_total_allocated_memory += n;
+	mem_current_allocated_memory += n;
+	mem_n_allocations++;
+
+	if (mem_current_allocated_memory > mem_max_allocated_memory) {
+		mem_max_allocated_memory = mem_current_allocated_memory;
+	}
+
+	mutex_exit(&mem_hash_mutex);
+
+	/* In the debug version set the buffer to a random
+	combination of 0xBA and 0xBE */
+
+	mem_init_buf(usr_buf, n);
+}
+
+/**********************************************************************
+Erases an allocated memory field in the debug version. */
+UNIV_INTERN
+void
+mem_field_erase(
+/*============*/
+	byte*	buf,	/* in: memory field */
+	ulint	n __attribute__((unused)))
+			/* in: how many bytes the user requested */
+{
+	byte*	usr_buf;
+
+	usr_buf = buf + MEM_FIELD_HEADER_SIZE;
+
+	mutex_enter(&mem_hash_mutex);
+	mem_current_allocated_memory	-= n;
+	mutex_exit(&mem_hash_mutex);
+
+	/* Check that the field lengths agree */
+	ut_ad(n == (ulint)mem_field_header_get_len(usr_buf));
+
+	/* In the debug version, set the freed space to a random
+	combination of 0xDE and 0xAD */
+
+	mem_erase_buf(buf, MEM_SPACE_NEEDED(n));
+}
+
+/*******************************************************************
+Initializes a buffer to a random combination of hex BA and BE.
+Used to initialize allocated memory. */
+UNIV_INTERN
+void
+mem_init_buf(
+/*=========*/
+	byte*	buf,	/* in: pointer to buffer */
+	ulint	 n)	/* in: length of buffer */
+{
+	byte*	ptr;
+
+	UNIV_MEM_ASSERT_W(buf, n);
+
+	for (ptr = buf; ptr < buf + n; ptr++) {
+
+		if (ut_rnd_gen_ibool()) {
+			*ptr = 0xBA;
+		} else {
+			*ptr = 0xBE;
+		}
+	}
+
+	UNIV_MEM_INVALID(buf, n);
+}
+
+/*******************************************************************
+Initializes a buffer to a random combination of hex DE and AD.
+Used to erase freed memory.*/
+UNIV_INTERN
+void
+mem_erase_buf(
+/*==========*/
+	byte*	buf,	/* in: pointer to buffer */
+	ulint	 n)	 /* in: length of buffer */
+{
+	byte*	ptr;
+
+	UNIV_MEM_ASSERT_W(buf, n);
+
+	for (ptr = buf; ptr < buf + n; ptr++) {
+		if (ut_rnd_gen_ibool()) {
+			*ptr = 0xDE;
+		} else {
+			*ptr = 0xAD;
+		}
+	}
+
+	UNIV_MEM_FREE(buf, n);
+}
+
+/*******************************************************************
+Inserts a created memory heap to the hash table of current allocated
+memory heaps. */
+UNIV_INTERN
+void
+mem_hash_insert(
+/*============*/
+	mem_heap_t*	heap,	   /* in: the created heap */
+	const char*	file_name, /* in: file name of creation */
+	ulint		line)	   /* in: line where created */
+{
+	mem_hash_node_t*	new_node;
+	ulint			cell_no	;
+
+	ut_ad(mem_heap_check(heap));
+
+	mutex_enter(&mem_hash_mutex);
+
+	cell_no = ut_hash_ulint((ulint)heap, MEM_HASH_SIZE);
+
+	/* Allocate a new node to the list */
+	new_node = ut_malloc(sizeof(mem_hash_node_t));
+
+	new_node->heap = heap;
+	new_node->file_name = file_name;
+	new_node->line = line;
+	new_node->nth_heap = mem_n_created_heaps;
+
+	/* Insert into lists */
+	UT_LIST_ADD_FIRST(list, *mem_hash_get_nth_cell(cell_no), new_node);
+
+	UT_LIST_ADD_LAST(all_list, mem_all_list_base, new_node);
+
+	mem_n_created_heaps++;
+
+	mutex_exit(&mem_hash_mutex);
+}
+
+/*******************************************************************
+Removes a memory heap (which is going to be freed by the caller)
+from the list of live memory heaps. Returns the size of the heap
+in terms of how much memory in bytes was allocated for the user of
+the heap (not the total space occupied by the heap).
+Also validates the heap.
+NOTE: This function does not free the storage occupied by the
+heap itself, only the node in the list of heaps. */
+UNIV_INTERN
+void
+mem_hash_remove(
+/*============*/
+	mem_heap_t*	heap,	   /* in: the heap to be freed */
+	const char*	file_name, /* in: file name of freeing */
+	ulint		line)	   /* in: line where freed */
+{
+	mem_hash_node_t*	node;
+	ulint			cell_no;
+	ibool			error;
+	ulint			size;
+
+	ut_ad(mem_heap_check(heap));
+
+	mutex_enter(&mem_hash_mutex);
+
+	cell_no = ut_hash_ulint((ulint)heap, MEM_HASH_SIZE);
+
+	/* Look for the heap in the hash table list */
+	node = UT_LIST_GET_FIRST(*mem_hash_get_nth_cell(cell_no));
+
+	while (node != NULL) {
+		if (node->heap == heap) {
+
+			break;
+		}
+
+		node = UT_LIST_GET_NEXT(list, node);
+	}
+
+	if (node == NULL) {
+		fprintf(stderr,
+			"Memory heap or buffer freed in %s line %lu"
+			" did not exist.\n",
+			file_name, (ulong) line);
+		ut_error;
+	}
+
+	/* Remove from lists */
+	UT_LIST_REMOVE(list, *mem_hash_get_nth_cell(cell_no), node);
+
+	UT_LIST_REMOVE(all_list, mem_all_list_base, node);
+
+	/* Validate the heap which will be freed */
+	mem_heap_validate_or_print(node->heap, NULL, FALSE, &error, &size,
+				   NULL, NULL);
+	if (error) {
+		fprintf(stderr,
+			"Inconsistency in memory heap or"
+			" buffer n:o %lu created\n"
+			"in %s line %lu and tried to free in %s line %lu.\n"
+			"Hex dump of 400 bytes around memory heap"
+			" first block start:\n",
+			node->nth_heap, node->file_name, (ulong) node->line,
+			file_name, (ulong) line);
+		ut_print_buf(stderr, (byte*)node->heap - 200, 400);
+		fputs("\nDump of the mem heap:\n", stderr);
+		mem_heap_validate_or_print(node->heap, NULL, TRUE, &error,
+					   &size, NULL, NULL);
+		ut_error;
+	}
+
+	/* Free the memory occupied by the node struct */
+	ut_free(node);
+
+	mem_current_allocated_memory -= size;
+
+	mutex_exit(&mem_hash_mutex);
+}
+#endif /* UNIV_MEM_DEBUG */
+
+#if defined UNIV_MEM_DEBUG || defined UNIV_DEBUG
+/*******************************************************************
+Checks a memory heap for consistency and prints the contents if requested.
+Outputs the sum of sizes of buffers given to the user (only in
+the debug version), the physical size of the heap and the number of
+blocks in the heap. In case of error returns 0 as sizes and number
+of blocks. */
+UNIV_INTERN
+void
+mem_heap_validate_or_print(
+/*=======================*/
+	mem_heap_t*	heap,	/* in: memory heap */
+	byte*		top __attribute__((unused)),
+				/* in: calculate and validate only until
+				this top pointer in the heap is reached,
+				if this pointer is NULL, ignored */
+	ibool		print,	/* in: if TRUE, prints the contents
+				of the heap; works only in
+				the debug version */
+	ibool*		error,	/* out: TRUE if error */
+	ulint*		us_size,/* out: allocated memory
+				(for the user) in the heap,
+				if a NULL pointer is passed as this
+				argument, it is ignored; in the
+				non-debug version this is always -1 */
+	ulint*		ph_size,/* out: physical size of the heap,
+				if a NULL pointer is passed as this
+				argument, it is ignored */
+	ulint*		n_blocks) /* out: number of blocks in the heap,
+				if a NULL pointer is passed as this
+				argument, it is ignored */
+{
+	mem_block_t*	block;
+	ulint		total_len	= 0;
+	ulint		block_count	= 0;
+	ulint		phys_len	= 0;
+#ifdef UNIV_MEM_DEBUG
+	ulint		len;
+	byte*		field;
+	byte*		user_field;
+	ulint		check_field;
+#endif
+
+	/* Pessimistically, we set the parameters to error values */
+	if (us_size != NULL) {
+		*us_size = 0;
+	}
+	if (ph_size != NULL) {
+		*ph_size = 0;
+	}
+	if (n_blocks != NULL) {
+		*n_blocks = 0;
+	}
+	*error = TRUE;
+
+	block = heap;
+
+	if (block->magic_n != MEM_BLOCK_MAGIC_N) {
+		return;
+	}
+
+	if (print) {
+		fputs("Memory heap:", stderr);
+	}
+
+	while (block != NULL) {
+		phys_len += mem_block_get_len(block);
+
+		if ((block->type == MEM_HEAP_BUFFER)
+		    && (mem_block_get_len(block) > UNIV_PAGE_SIZE)) {
+
+			fprintf(stderr,
+				"InnoDB: Error: mem block %p"
+				" length %lu > UNIV_PAGE_SIZE\n",
+				(void*) block,
+				(ulong) mem_block_get_len(block));
+			/* error */
+
+			return;
+		}
+
+#ifdef UNIV_MEM_DEBUG
+		/* We can trace the fields of the block only in the debug
+		version */
+		if (print) {
+			fprintf(stderr, " Block %ld:", block_count);
+		}
+
+		field = (byte*)block + mem_block_get_start(block);
+
+		if (top && (field == top)) {
+
+			goto completed;
+		}
+
+		while (field < (byte*)block + mem_block_get_free(block)) {
+
+			/* Calculate the pointer to the storage
+			which was given to the user */
+
+			user_field = field + MEM_FIELD_HEADER_SIZE;
+
+			len = mem_field_header_get_len(user_field);
+
+			if (print) {
+				ut_print_buf(stderr, user_field, len);
+				putc('\n', stderr);
+			}
+
+			total_len += len;
+			check_field = mem_field_header_get_check(user_field);
+
+			if (check_field
+			    != mem_field_trailer_get_check(user_field)) {
+				/* error */
+
+				fprintf(stderr,
+					"InnoDB: Error: block %lx mem"
+					" field %lx len %lu\n"
+					"InnoDB: header check field is"
+					" %lx but trailer %lx\n",
+					(ulint)block,
+					(ulint)field, len, check_field,
+					mem_field_trailer_get_check(
+						user_field));
+
+				return;
+			}
+
+			/* Move to next field */
+			field = field + MEM_SPACE_NEEDED(len);
+
+			if (top && (field == top)) {
+
+				goto completed;
+			}
+
+		}
+
+		/* At the end check that we have arrived to the first free
+		position */
+
+		if (field != (byte*)block + mem_block_get_free(block)) {
+			/* error */
+
+			fprintf(stderr,
+				"InnoDB: Error: block %lx end of"
+				" mem fields %lx\n"
+				"InnoDB: but block free at %lx\n",
+				(ulint)block, (ulint)field,
+				(ulint)((byte*)block
+					+ mem_block_get_free(block)));
+
+			return;
+		}
+
+#endif
+
+		block = UT_LIST_GET_NEXT(list, block);
+		block_count++;
+	}
+#ifdef UNIV_MEM_DEBUG
+completed:
+#endif
+	if (us_size != NULL) {
+		*us_size = total_len;
+	}
+	if (ph_size != NULL) {
+		*ph_size = phys_len;
+	}
+	if (n_blocks != NULL) {
+		*n_blocks = block_count;
+	}
+	*error = FALSE;
+}
+
+/******************************************************************
+Prints the contents of a memory heap. */
+static
+void
+mem_heap_print(
+/*===========*/
+	mem_heap_t*	heap)	/* in: memory heap */
+{
+	ibool	error;
+	ulint	us_size;
+	ulint	phys_size;
+	ulint	n_blocks;
+
+	ut_ad(mem_heap_check(heap));
+
+	mem_heap_validate_or_print(heap, NULL, TRUE, &error,
+				   &us_size, &phys_size, &n_blocks);
+	fprintf(stderr,
+		"\nheap type: %lu; size: user size %lu;"
+		" physical size %lu; blocks %lu.\n",
+		(ulong) heap->type, (ulong) us_size,
+		(ulong) phys_size, (ulong) n_blocks);
+	ut_a(!error);
+}
+
+/******************************************************************
+Validates the contents of a memory heap. */
+UNIV_INTERN
+ibool
+mem_heap_validate(
+/*==============*/
+				/* out: TRUE if ok */
+	mem_heap_t*	heap)	/* in: memory heap */
+{
+	ibool	error;
+	ulint	us_size;
+	ulint	phys_size;
+	ulint	n_blocks;
+
+	ut_ad(mem_heap_check(heap));
+
+	mem_heap_validate_or_print(heap, NULL, FALSE, &error, &us_size,
+				   &phys_size, &n_blocks);
+	if (error) {
+		mem_heap_print(heap);
+	}
+
+	ut_a(!error);
+
+	return(TRUE);
+}
+#endif /* UNIV_MEM_DEBUG || UNIV_DEBUG */
+
+#ifdef UNIV_DEBUG
+/******************************************************************
+Checks that an object is a memory heap (or a block of it). */
+UNIV_INTERN
+ibool
+mem_heap_check(
+/*===========*/
+				/* out: TRUE if ok */
+	mem_heap_t*	heap)	/* in: memory heap */
+{
+	ut_a(heap->magic_n == MEM_BLOCK_MAGIC_N);
+
+	return(TRUE);
+}
+#endif /* UNIV_DEBUG */
+
+#ifdef UNIV_MEM_DEBUG
+/*********************************************************************
+TRUE if no memory is currently allocated. */
+UNIV_INTERN
+ibool
+mem_all_freed(void)
+/*===============*/
+			/* out: TRUE if no heaps exist */
+{
+	mem_hash_node_t*	node;
+	ulint			heap_count	= 0;
+	ulint			i;
+
+	mem_validate();
+
+	mutex_enter(&mem_hash_mutex);
+
+	for (i = 0; i < MEM_HASH_SIZE; i++) {
+
+		node = UT_LIST_GET_FIRST(*mem_hash_get_nth_cell(i));
+		while (node != NULL) {
+			heap_count++;
+			node = UT_LIST_GET_NEXT(list, node);
+		}
+	}
+
+	mutex_exit(&mem_hash_mutex);
+
+	if (heap_count == 0) {
+
+		ut_a(mem_pool_get_reserved(mem_comm_pool) == 0);
+
+		return(TRUE);
+	} else {
+		return(FALSE);
+	}
+}
+
+/*********************************************************************
+Validates the dynamic memory allocation system. */
+UNIV_INTERN
+ibool
+mem_validate_no_assert(void)
+/*========================*/
+			/* out: TRUE if error */
+{
+	mem_hash_node_t*	node;
+	ulint			n_heaps			= 0;
+	ulint			allocated_mem;
+	ulint			ph_size;
+	ulint			total_allocated_mem	= 0;
+	ibool			error			= FALSE;
+	ulint			n_blocks;
+	ulint			i;
+
+	mem_pool_validate(mem_comm_pool);
+
+	mutex_enter(&mem_hash_mutex);
+
+	for (i = 0; i < MEM_HASH_SIZE; i++) {
+
+		node = UT_LIST_GET_FIRST(*mem_hash_get_nth_cell(i));
+
+		while (node != NULL) {
+			n_heaps++;
+
+			mem_heap_validate_or_print(node->heap, NULL,
+						   FALSE, &error,
+						   &allocated_mem,
+						   &ph_size, &n_blocks);
+
+			if (error) {
+				fprintf(stderr,
+					"\nERROR!!!!!!!!!!!!!!!!!!!"
+					"!!!!!!!!!!!!!!!!!!!!!!!\n\n"
+					"Inconsistency in memory heap"
+					" or buffer created\n"
+					"in %s line %lu.\n",
+					node->file_name, node->line);
+
+				mutex_exit(&mem_hash_mutex);
+
+				return(TRUE);
+			}
+
+			total_allocated_mem += allocated_mem;
+			node = UT_LIST_GET_NEXT(list, node);
+		}
+	}
+
+	if ((n_heaps == 0) && (mem_current_allocated_memory != 0)) {
+		error = TRUE;
+	}
+
+	if (mem_total_allocated_memory < mem_current_allocated_memory) {
+		error = TRUE;
+	}
+
+	if (mem_max_allocated_memory > mem_total_allocated_memory) {
+		error = TRUE;
+	}
+
+	if (mem_n_created_heaps < n_heaps) {
+		error = TRUE;
+	}
+
+	mutex_exit(&mem_hash_mutex);
+
+	return(error);
+}
+
+/****************************************************************
+Validates the dynamic memory */
+UNIV_INTERN
+ibool
+mem_validate(void)
+/*==============*/
+			/* out: TRUE if ok */
+{
+	ut_a(!mem_validate_no_assert());
+
+	return(TRUE);
+}
+#endif /* UNIV_MEM_DEBUG */
+
+/****************************************************************
+Tries to find neigboring memory allocation blocks and dumps to stderr
+the neighborhood of a given pointer. */
+UNIV_INTERN
+void
+mem_analyze_corruption(
+/*===================*/
+	void*	ptr)	/* in: pointer to place of possible corruption */
+{
+	byte*	p;
+	ulint	i;
+	ulint	dist;
+
+	fputs("InnoDB: Apparent memory corruption: mem dump ", stderr);
+	ut_print_buf(stderr, (byte*)ptr - 250, 500);
+
+	fputs("\nInnoDB: Scanning backward trying to find"
+	      " previous allocated mem blocks\n", stderr);
+
+	p = (byte*)ptr;
+	dist = 0;
+
+	for (i = 0; i < 10; i++) {
+		for (;;) {
+			if (((ulint)p) % 4 == 0) {
+
+				if (*((ulint*)p) == MEM_BLOCK_MAGIC_N) {
+					fprintf(stderr,
+						"Mem block at - %lu,"
+						" file %s, line %lu\n",
+						(ulong) dist,
+						(p + sizeof(ulint)),
+						(ulong)
+						(*(ulint*)(p + 8
+							   + sizeof(ulint))));
+
+					break;
+				}
+
+				if (*((ulint*)p) == MEM_FREED_BLOCK_MAGIC_N) {
+					fprintf(stderr,
+						"Freed mem block at - %lu,"
+						" file %s, line %lu\n",
+						(ulong) dist,
+						(p + sizeof(ulint)),
+						(ulong)
+						(*(ulint*)(p + 8
+							   + sizeof(ulint))));
+
+					break;
+				}
+			}
+
+			p--;
+			dist++;
+		}
+
+		p--;
+		dist++;
+	}
+
+	fprintf(stderr,
+		"InnoDB: Scanning forward trying to find next"
+		" allocated mem blocks\n");
+
+	p = (byte*)ptr;
+	dist = 0;
+
+	for (i = 0; i < 10; i++) {
+		for (;;) {
+			if (((ulint)p) % 4 == 0) {
+
+				if (*((ulint*)p) == MEM_BLOCK_MAGIC_N) {
+					fprintf(stderr,
+						"Mem block at + %lu, file %s,"
+						" line %lu\n",
+						(ulong) dist,
+						(p + sizeof(ulint)),
+						(ulong)
+						(*(ulint*)(p + 8
+							   + sizeof(ulint))));
+
+					break;
+				}
+
+				if (*((ulint*)p) == MEM_FREED_BLOCK_MAGIC_N) {
+					fprintf(stderr,
+						"Freed mem block at + %lu,"
+						" file %s, line %lu\n",
+						(ulong) dist,
+						(p + sizeof(ulint)),
+						(ulong)
+						(*(ulint*)(p + 8
+							   + sizeof(ulint))));
+
+					break;
+				}
+			}
+
+			p++;
+			dist++;
+		}
+
+		p++;
+		dist++;
+	}
+}
+
+/*********************************************************************
+Prints information of dynamic memory usage and currently allocated
+memory heaps or buffers. Can only be used in the debug version. */
+static
+void
+mem_print_info_low(
+/*===============*/
+	ibool	print_all)	/* in: if TRUE, all heaps are printed,
+				else only the heaps allocated after the
+				previous call of this function */
+{
+#ifdef UNIV_MEM_DEBUG
+	mem_hash_node_t*	node;
+	ulint			n_heaps			= 0;
+	ulint			allocated_mem;
+	ulint			ph_size;
+	ulint			total_allocated_mem	= 0;
+	ibool			error;
+	ulint			n_blocks;
+#endif
+	FILE*			outfile;
+
+	/* outfile = fopen("ibdebug", "a"); */
+
+	outfile = stdout;
+
+	fprintf(outfile, "\n");
+	fprintf(outfile,
+		"________________________________________________________\n");
+	fprintf(outfile, "MEMORY ALLOCATION INFORMATION\n\n");
+
+#ifndef UNIV_MEM_DEBUG
+
+	UT_NOT_USED(print_all);
+
+	mem_pool_print_info(outfile, mem_comm_pool);
+
+	fprintf(outfile,
+		"Sorry, non-debug version cannot give more memory info\n");
+
+	/* fclose(outfile); */
+
+	return;
+#else
+	mutex_enter(&mem_hash_mutex);
+
+	fprintf(outfile, "LIST OF CREATED HEAPS AND ALLOCATED BUFFERS: \n\n");
+
+	if (!print_all) {
+		fprintf(outfile, "AFTER THE LAST PRINT INFO\n");
+	}
+
+	node = UT_LIST_GET_FIRST(mem_all_list_base);
+
+	while (node != NULL) {
+		n_heaps++;
+
+		if (!print_all && node->nth_heap < mem_last_print_info) {
+
+			goto next_heap;
+		}
+
+		mem_heap_validate_or_print(node->heap, NULL,
+					   FALSE, &error, &allocated_mem,
+					   &ph_size, &n_blocks);
+		total_allocated_mem += allocated_mem;
+
+		fprintf(outfile,
+			"%lu: file %s line %lu of size %lu phys.size %lu"
+			" with %lu blocks, type %lu\n",
+			node->nth_heap, node->file_name, node->line,
+			allocated_mem, ph_size, n_blocks,
+			(node->heap)->type);
+next_heap:
+		node = UT_LIST_GET_NEXT(all_list, node);
+	}
+
+	fprintf(outfile, "\n");
+
+	fprintf(outfile, "Current allocated memory              : %lu\n",
+		mem_current_allocated_memory);
+	fprintf(outfile, "Current allocated heaps and buffers   : %lu\n",
+		n_heaps);
+	fprintf(outfile, "Cumulative allocated memory           : %lu\n",
+		mem_total_allocated_memory);
+	fprintf(outfile, "Maximum allocated memory              : %lu\n",
+		mem_max_allocated_memory);
+	fprintf(outfile, "Cumulative created heaps and buffers  : %lu\n",
+		mem_n_created_heaps);
+	fprintf(outfile, "Cumulative number of allocations      : %lu\n",
+		mem_n_allocations);
+
+	mem_last_print_info = mem_n_created_heaps;
+
+	mutex_exit(&mem_hash_mutex);
+
+	mem_pool_print_info(outfile, mem_comm_pool);
+
+	/*	mem_validate(); */
+
+	/*	fclose(outfile); */
+#endif
+}
+
+/*********************************************************************
+Prints information of dynamic memory usage and currently allocated memory
+heaps or buffers. Can only be used in the debug version. */
+UNIV_INTERN
+void
+mem_print_info(void)
+/*================*/
+{
+	mem_print_info_low(TRUE);
+}
+
+/*********************************************************************
+Prints information of dynamic memory usage and currently allocated memory
+heaps or buffers since the last ..._print_info or..._print_new_info. */
+UNIV_INTERN
+void
+mem_print_new_info(void)
+/*====================*/
+{
+	mem_print_info_low(FALSE);
+}
diff --git a/storage/xtradb/mem/mem0mem.c b/storage/xtradb/mem/mem0mem.c
new file mode 100644
index 00000000000..b7345f5846b
--- /dev/null
+++ b/storage/xtradb/mem/mem0mem.c
@@ -0,0 +1,553 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/************************************************************************
+The memory management
+
+Created 6/9/1994 Heikki Tuuri
+*************************************************************************/
+
+#include "mem0mem.h"
+#ifdef UNIV_NONINL
+#include "mem0mem.ic"
+#endif
+
+#include "buf0buf.h"
+#include "srv0srv.h"
+#include "mem0dbg.c"
+#include <stdarg.h>
+
+/*
+			THE MEMORY MANAGEMENT
+			=====================
+
+The basic element of the memory management is called a memory
+heap. A memory heap is conceptually a
+stack from which memory can be allocated. The stack may grow infinitely.
+The top element of the stack may be freed, or
+the whole stack can be freed at one time. The advantage of the
+memory heap concept is that we can avoid using the malloc and free
+functions of C which are quite expensive, for example, on the Solaris + GCC
+system (50 MHz Sparc, 1993) the pair takes 3 microseconds,
+on Win NT + 100MHz Pentium, 2.5 microseconds.
+When we use a memory heap,
+we can allocate larger blocks of memory at a time and thus
+reduce overhead. Slightly more efficient the method is when we
+allocate the memory from the index page buffer pool, as we can
+claim a new page fast. This is called buffer allocation.
+When we allocate the memory from the dynamic memory of the
+C environment, that is called dynamic allocation.
+
+The default way of operation of the memory heap is the following.
+First, when the heap is created, an initial block of memory is
+allocated. In dynamic allocation this may be about 50 bytes.
+If more space is needed, additional blocks are allocated
+and they are put into a linked list.
+After the initial block, each allocated block is twice the size of the
+previous, until a threshold is attained, after which the sizes
+of the blocks stay the same. An exception is, of course, the case
+where the caller requests a memory buffer whose size is
+bigger than the threshold. In that case a block big enough must
+be allocated.
+
+The heap is physically arranged so that if the current block
+becomes full, a new block is allocated and always inserted in the
+chain of blocks as the last block.
+
+In the debug version of the memory management, all the allocated
+heaps are kept in a list (which is implemented as a hash table).
+Thus we can notice if the caller tries to free an already freed
+heap. In addition, each buffer given to the caller contains
+start field at the start and a trailer field at the end of the buffer.
+
+The start field has the following content:
+A. sizeof(ulint) bytes of field length (in the standard byte order)
+B. sizeof(ulint) bytes of check field (a random number)
+
+The trailer field contains:
+A. sizeof(ulint) bytes of check field (the same random number as at the start)
+
+Thus we can notice if something has been copied over the
+borders of the buffer, which is illegal.
+The memory in the buffers is initialized to a random byte sequence.
+After freeing, all the blocks in the heap are set to random bytes
+to help us discover errors which result from the use of
+buffers in an already freed heap. */
+
+#ifdef MEM_PERIODIC_CHECK
+
+ibool					mem_block_list_inited;
+/* List of all mem blocks allocated; protected by the mem_comm_pool mutex */
+UT_LIST_BASE_NODE_T(mem_block_t)	mem_block_list;
+
+#endif
+
+/**************************************************************************
+Duplicates a NUL-terminated string, allocated from a memory heap. */
+UNIV_INTERN
+char*
+mem_heap_strdup(
+/*============*/
+				/* out, own: a copy of the string */
+	mem_heap_t*	heap,	/* in: memory heap where string is allocated */
+	const char*	str)	/* in: string to be copied */
+{
+	return(mem_heap_dup(heap, str, strlen(str) + 1));
+}
+
+/**************************************************************************
+Duplicate a block of data, allocated from a memory heap. */
+UNIV_INTERN
+void*
+mem_heap_dup(
+/*=========*/
+				/* out, own: a copy of the data */
+	mem_heap_t*	heap,	/* in: memory heap where copy is allocated */
+	const void*	data,	/* in: data to be copied */
+	ulint		len)	/* in: length of data, in bytes */
+{
+	return(memcpy(mem_heap_alloc(heap, len), data, len));
+}
+
+/**************************************************************************
+Concatenate two memory blocks and return the result, using a memory heap. */
+UNIV_INTERN
+void*
+mem_heap_cat(
+/*=========*/
+				/* out, own: the result */
+	mem_heap_t*	heap,	/* in: memory heap where result is allocated */
+	const void*	b1,	/* in: block 1 */
+	ulint		len1,	/* in: length of b1, in bytes */
+	const void*	b2,	/* in: block 2 */
+	ulint		len2)	/* in: length of b2, in bytes */
+{
+	void*	res = mem_heap_alloc(heap, len1 + len2);
+
+	memcpy(res, b1, len1);
+	memcpy((char*)res + len1, b2, len2);
+
+	return(res);
+}
+
+/**************************************************************************
+Concatenate two strings and return the result, using a memory heap. */
+UNIV_INTERN
+char*
+mem_heap_strcat(
+/*============*/
+				/* out, own: the result */
+	mem_heap_t*	heap,	/* in: memory heap where string is allocated */
+	const char*	s1,	/* in: string 1 */
+	const char*	s2)	/* in: string 2 */
+{
+	char*	s;
+	ulint	s1_len = strlen(s1);
+	ulint	s2_len = strlen(s2);
+
+	s = mem_heap_alloc(heap, s1_len + s2_len + 1);
+
+	memcpy(s, s1, s1_len);
+	memcpy(s + s1_len, s2, s2_len);
+
+	s[s1_len + s2_len] = '\0';
+
+	return(s);
+}
+
+
+/********************************************************************
+Helper function for mem_heap_printf. */
+static
+ulint
+mem_heap_printf_low(
+/*================*/
+				/* out: length of formatted string,
+				including terminating NUL */
+	char*		buf,	/* in/out: buffer to store formatted string
+				in, or NULL to just calculate length */
+	const char*	format,	/* in: format string */
+	va_list		ap)	/* in: arguments */
+{
+	ulint 		len = 0;
+
+	while (*format) {
+
+		/* Does this format specifier have the 'l' length modifier. */
+		ibool	is_long = FALSE;
+
+		/* Length of one parameter. */
+		size_t	plen;
+
+		if (*format++ != '%') {
+			/* Non-format character. */
+
+			len++;
+
+			if (buf) {
+				*buf++ = *(format - 1);
+			}
+
+			continue;
+		}
+
+		if (*format == 'l') {
+			is_long = TRUE;
+			format++;
+		}
+
+		switch (*format++) {
+		case 's':
+			/* string */
+			{
+				char*	s = va_arg(ap, char*);
+
+				/* "%ls" is a non-sensical format specifier. */
+				ut_a(!is_long);
+
+				plen = strlen(s);
+				len += plen;
+
+				if (buf) {
+					memcpy(buf, s, plen);
+					buf += plen;
+				}
+			}
+
+			break;
+
+		case 'u':
+			/* unsigned int */
+			{
+				char		tmp[32];
+				unsigned long	val;
+
+				/* We only support 'long' values for now. */
+				ut_a(is_long);
+
+				val = va_arg(ap, unsigned long);
+
+				plen = sprintf(tmp, "%lu", val);
+				len += plen;
+
+				if (buf) {
+					memcpy(buf, tmp, plen);
+					buf += plen;
+				}
+			}
+
+			break;
+
+		case '%':
+
+			/* "%l%" is a non-sensical format specifier. */
+			ut_a(!is_long);
+
+			len++;
+
+			if (buf) {
+				*buf++ = '%';
+			}
+
+			break;
+
+		default:
+			ut_error;
+		}
+	}
+
+	/* For the NUL character. */
+	len++;
+
+	if (buf) {
+		*buf = '\0';
+	}
+
+	return(len);
+}
+
+/********************************************************************
+A simple (s)printf replacement that dynamically allocates the space for the
+formatted string from the given heap. This supports a very limited set of
+the printf syntax: types 's' and 'u' and length modifier 'l' (which is
+required for the 'u' type). */
+UNIV_INTERN
+char*
+mem_heap_printf(
+/*============*/
+				/* out: heap-allocated formatted string */
+	mem_heap_t*	heap,	/* in: memory heap */
+	const char*	format,	/* in: format string */
+	...)
+{
+	va_list		ap;
+	char*		str;
+	ulint 		len;
+
+	/* Calculate length of string */
+	len = 0;
+	va_start(ap, format);
+	len = mem_heap_printf_low(NULL, format, ap);
+	va_end(ap);
+
+	/* Now create it for real. */
+	str = mem_heap_alloc(heap, len);
+	va_start(ap, format);
+	mem_heap_printf_low(str, format, ap);
+	va_end(ap);
+
+	return(str);
+}
+
+/*******************************************************************
+Creates a memory heap block where data can be allocated. */
+UNIV_INTERN
+mem_block_t*
+mem_heap_create_block(
+/*==================*/
+				/* out, own: memory heap block, NULL if
+				did not succeed (only possible for
+				MEM_HEAP_BTR_SEARCH type heaps) */
+	mem_heap_t*	heap,	/* in: memory heap or NULL if first block
+				should be created */
+	ulint		n,	/* in: number of bytes needed for user data */
+	ulint		type,	/* in: type of heap: MEM_HEAP_DYNAMIC or
+				MEM_HEAP_BUFFER */
+	const char*	file_name,/* in: file name where created */
+	ulint		line)	/* in: line where created */
+{
+	buf_block_t*	buf_block = NULL;
+	mem_block_t*	block;
+	ulint		len;
+
+	ut_ad((type == MEM_HEAP_DYNAMIC) || (type == MEM_HEAP_BUFFER)
+	      || (type == MEM_HEAP_BUFFER + MEM_HEAP_BTR_SEARCH));
+
+	if (heap && heap->magic_n != MEM_BLOCK_MAGIC_N) {
+		mem_analyze_corruption(heap);
+	}
+
+	/* In dynamic allocation, calculate the size: block header + data. */
+	len = MEM_BLOCK_HEADER_SIZE + MEM_SPACE_NEEDED(n);
+
+	if (type == MEM_HEAP_DYNAMIC || len < UNIV_PAGE_SIZE / 2) {
+
+		ut_ad(type == MEM_HEAP_DYNAMIC || n <= MEM_MAX_ALLOC_IN_BUF);
+
+		block = mem_area_alloc(&len, mem_comm_pool);
+	} else {
+		len = UNIV_PAGE_SIZE;
+
+		if ((type & MEM_HEAP_BTR_SEARCH) && heap) {
+			/* We cannot allocate the block from the
+			buffer pool, but must get the free block from
+			the heap header free block field */
+
+			buf_block = heap->free_block;
+			heap->free_block = NULL;
+
+			if (UNIV_UNLIKELY(!buf_block)) {
+
+				return(NULL);
+			}
+		} else {
+			buf_block = buf_block_alloc(0);
+		}
+
+		block = (mem_block_t*) buf_block->frame;
+	}
+
+	ut_ad(block);
+	block->buf_block = buf_block;
+	block->magic_n = MEM_BLOCK_MAGIC_N;
+	ut_strlcpy_rev(block->file_name, file_name, sizeof(block->file_name));
+	block->line = line;
+
+#ifdef MEM_PERIODIC_CHECK
+	mem_pool_mutex_enter();
+
+	if (!mem_block_list_inited) {
+		mem_block_list_inited = TRUE;
+		UT_LIST_INIT(mem_block_list);
+	}
+
+	UT_LIST_ADD_LAST(mem_block_list, mem_block_list, block);
+
+	mem_pool_mutex_exit();
+#endif
+	mem_block_set_len(block, len);
+	mem_block_set_type(block, type);
+	mem_block_set_free(block, MEM_BLOCK_HEADER_SIZE);
+	mem_block_set_start(block, MEM_BLOCK_HEADER_SIZE);
+
+	block->free_block = NULL;
+
+	ut_ad((ulint)MEM_BLOCK_HEADER_SIZE < len);
+
+	return(block);
+}
+
+/*******************************************************************
+Adds a new block to a memory heap. */
+UNIV_INTERN
+mem_block_t*
+mem_heap_add_block(
+/*===============*/
+				/* out: created block, NULL if did not
+				succeed (only possible for
+				MEM_HEAP_BTR_SEARCH type heaps)*/
+	mem_heap_t*	heap,	/* in: memory heap */
+	ulint		n)	/* in: number of bytes user needs */
+{
+	mem_block_t*	block;
+	mem_block_t*	new_block;
+	ulint		new_size;
+
+	ut_ad(mem_heap_check(heap));
+
+	block = UT_LIST_GET_LAST(heap->base);
+
+	/* We have to allocate a new block. The size is always at least
+	doubled until the standard size is reached. After that the size
+	stays the same, except in cases where the caller needs more space. */
+
+	new_size = 2 * mem_block_get_len(block);
+
+	if (heap->type != MEM_HEAP_DYNAMIC) {
+		/* From the buffer pool we allocate buffer frames */
+		ut_a(n <= MEM_MAX_ALLOC_IN_BUF);
+
+		if (new_size > MEM_MAX_ALLOC_IN_BUF) {
+			new_size = MEM_MAX_ALLOC_IN_BUF;
+		}
+	} else if (new_size > MEM_BLOCK_STANDARD_SIZE) {
+
+		new_size = MEM_BLOCK_STANDARD_SIZE;
+	}
+
+	if (new_size < n) {
+		new_size = n;
+	}
+
+	new_block = mem_heap_create_block(heap, new_size, heap->type,
+					  heap->file_name, heap->line);
+	if (new_block == NULL) {
+
+		return(NULL);
+	}
+
+	/* Add the new block as the last block */
+
+	UT_LIST_INSERT_AFTER(list, heap->base, block, new_block);
+
+	return(new_block);
+}
+
+/**********************************************************************
+Frees a block from a memory heap. */
+UNIV_INTERN
+void
+mem_heap_block_free(
+/*================*/
+	mem_heap_t*	heap,	/* in: heap */
+	mem_block_t*	block)	/* in: block to free */
+{
+	ulint		type;
+	ulint		len;
+	buf_block_t*	buf_block;
+
+	if (block->magic_n != MEM_BLOCK_MAGIC_N) {
+		mem_analyze_corruption(block);
+	}
+
+	UT_LIST_REMOVE(list, heap->base, block);
+
+#ifdef MEM_PERIODIC_CHECK
+	mem_pool_mutex_enter();
+
+	UT_LIST_REMOVE(mem_block_list, mem_block_list, block);
+
+	mem_pool_mutex_exit();
+#endif
+	type = heap->type;
+	len = block->len;
+	buf_block = block->buf_block;
+	block->magic_n = MEM_FREED_BLOCK_MAGIC_N;
+
+#ifdef UNIV_MEM_DEBUG
+	/* In the debug version we set the memory to a random combination
+	of hex 0xDE and 0xAD. */
+
+	mem_erase_buf((byte*)block, len);
+#else /* UNIV_MEM_DEBUG */
+	UNIV_MEM_ASSERT_AND_FREE(block, len);
+#endif /* UNIV_MEM_DEBUG */
+
+	if (type == MEM_HEAP_DYNAMIC || len < UNIV_PAGE_SIZE / 2) {
+
+		ut_ad(!buf_block);
+		mem_area_free(block, mem_comm_pool);
+	} else {
+		ut_ad(type & MEM_HEAP_BUFFER);
+
+		buf_block_free(buf_block);
+	}
+}
+
+/**********************************************************************
+Frees the free_block field from a memory heap. */
+UNIV_INTERN
+void
+mem_heap_free_block_free(
+/*=====================*/
+	mem_heap_t*	heap)	/* in: heap */
+{
+	if (UNIV_LIKELY_NULL(heap->free_block)) {
+
+		buf_block_free(heap->free_block);
+
+		heap->free_block = NULL;
+	}
+}
+
+#ifdef MEM_PERIODIC_CHECK
+/**********************************************************************
+Goes through the list of all allocated mem blocks, checks their magic
+numbers, and reports possible corruption. */
+UNIV_INTERN
+void
+mem_validate_all_blocks(void)
+/*=========================*/
+{
+	mem_block_t*	block;
+
+	mem_pool_mutex_enter();
+
+	block = UT_LIST_GET_FIRST(mem_block_list);
+
+	while (block) {
+		if (block->magic_n != MEM_BLOCK_MAGIC_N) {
+			mem_analyze_corruption(block);
+		}
+
+		block = UT_LIST_GET_NEXT(mem_block_list, block);
+	}
+
+	mem_pool_mutex_exit();
+}
+#endif
diff --git a/storage/xtradb/mem/mem0pool.c b/storage/xtradb/mem/mem0pool.c
new file mode 100644
index 00000000000..34de6b2a706
--- /dev/null
+++ b/storage/xtradb/mem/mem0pool.c
@@ -0,0 +1,704 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/************************************************************************
+The lowest-level memory management
+
+Created 5/12/1997 Heikki Tuuri
+*************************************************************************/
+
+#include "mem0pool.h"
+#ifdef UNIV_NONINL
+#include "mem0pool.ic"
+#endif
+
+#include "srv0srv.h"
+#include "sync0sync.h"
+#include "ut0mem.h"
+#include "ut0lst.h"
+#include "ut0byte.h"
+#include "mem0mem.h"
+
+/* We would like to use also the buffer frames to allocate memory. This
+would be desirable, because then the memory consumption of the database
+would be fixed, and we might even lock the buffer pool to the main memory.
+The problem here is that the buffer management routines can themselves call
+memory allocation, while the buffer pool mutex is reserved.
+
+The main components of the memory consumption are:
+
+1. buffer pool,
+2. parsed and optimized SQL statements,
+3. data dictionary cache,
+4. log buffer,
+5. locks for each transaction,
+6. hash table for the adaptive index,
+7. state and buffers for each SQL query currently being executed,
+8. session for each user, and
+9. stack for each OS thread.
+
+Items 1 and 2 are managed by an LRU algorithm. Items 5 and 6 can potentially
+consume very much memory. Items 7 and 8 should consume quite little memory,
+and the OS should take care of item 9, which too should consume little memory.
+
+A solution to the memory management:
+
+1. the buffer pool size is set separately;
+2. log buffer size is set separately;
+3. the common pool size for all the other entries, except 8, is set separately.
+
+Problems: we may waste memory if the common pool is set too big. Another
+problem is the locks, which may take very much space in big transactions.
+Then the shared pool size should be set very big. We can allow locks to take
+space from the buffer pool, but the SQL optimizer is then unaware of the
+usable size of the buffer pool. We could also combine the objects in the
+common pool and the buffers in the buffer pool into a single LRU list and
+manage it uniformly, but this approach does not take into account the parsing
+and other costs unique to SQL statements.
+
+The locks for a transaction can be seen as a part of the state of the
+transaction. Hence, they should be stored in the common pool. We still
+have the problem of a very big update transaction, for example, which
+will set very many x-locks on rows, and the locks will consume a lot
+of memory, say, half of the buffer pool size.
+
+Another problem is what to do if we are not able to malloc a requested
+block of memory from the common pool. Then we can request memory from
+the operating system. If it does not help, a system error results.
+
+Because 5 and 6 may potentially consume very much memory, we let them grow
+into the buffer pool. We may let the locks of a transaction take frames
+from the buffer pool, when the corresponding memory heap block has grown to
+the size of a buffer frame. Similarly for the hash node cells of the locks,
+and for the adaptive index. Thus, for each individual transaction, its locks
+can occupy at most about the size of the buffer frame of memory in the common
+pool, and after that its locks will grow into the buffer pool. */
+
+/* Mask used to extract the free bit from area->size */
+#define MEM_AREA_FREE	1
+
+/* The smallest memory area total size */
+#define MEM_AREA_MIN_SIZE	(2 * MEM_AREA_EXTRA_SIZE)
+
+
+/* Data structure for a memory pool. The space is allocated using the buddy
+algorithm, where free list i contains areas of size 2 to power i. */
+struct mem_pool_struct{
+	byte*		buf;		/* memory pool */
+	ulint		size;		/* memory common pool size */
+	ulint		reserved;	/* amount of currently allocated
+					memory */
+	mutex_t		mutex;		/* mutex protecting this struct */
+	UT_LIST_BASE_NODE_T(mem_area_t)
+			free_list[64];	/* lists of free memory areas: an
+					area is put to the list whose number
+					is the 2-logarithm of the area size */
+};
+
+/* The common memory pool */
+UNIV_INTERN mem_pool_t*	mem_comm_pool	= NULL;
+
+/* We use this counter to check that the mem pool mutex does not leak;
+this is to track a strange assertion failure reported at
+mysql@lists.mysql.com */
+
+UNIV_INTERN ulint	mem_n_threads_inside		= 0;
+
+/************************************************************************
+Reserves the mem pool mutex. */
+UNIV_INTERN
+void
+mem_pool_mutex_enter(void)
+/*======================*/
+{
+	mutex_enter(&(mem_comm_pool->mutex));
+}
+
+/************************************************************************
+Releases the mem pool mutex. */
+UNIV_INTERN
+void
+mem_pool_mutex_exit(void)
+/*=====================*/
+{
+	mutex_exit(&(mem_comm_pool->mutex));
+}
+
+/************************************************************************
+Returns memory area size. */
+UNIV_INLINE
+ulint
+mem_area_get_size(
+/*==============*/
+				/* out: size */
+	mem_area_t*	area)	/* in: area */
+{
+	return(area->size_and_free & ~MEM_AREA_FREE);
+}
+
+/************************************************************************
+Sets memory area size. */
+UNIV_INLINE
+void
+mem_area_set_size(
+/*==============*/
+	mem_area_t*	area,	/* in: area */
+	ulint		size)	/* in: size */
+{
+	area->size_and_free = (area->size_and_free & MEM_AREA_FREE)
+		| size;
+}
+
+/************************************************************************
+Returns memory area free bit. */
+UNIV_INLINE
+ibool
+mem_area_get_free(
+/*==============*/
+				/* out: TRUE if free */
+	mem_area_t*	area)	/* in: area */
+{
+#if TRUE != MEM_AREA_FREE
+# error "TRUE != MEM_AREA_FREE"
+#endif
+	return(area->size_and_free & MEM_AREA_FREE);
+}
+
+/************************************************************************
+Sets memory area free bit. */
+UNIV_INLINE
+void
+mem_area_set_free(
+/*==============*/
+	mem_area_t*	area,	/* in: area */
+	ibool		free)	/* in: free bit value */
+{
+#if TRUE != MEM_AREA_FREE
+# error "TRUE != MEM_AREA_FREE"
+#endif
+	area->size_and_free = (area->size_and_free & ~MEM_AREA_FREE)
+		| free;
+}
+
+/************************************************************************
+Creates a memory pool. */
+UNIV_INTERN
+mem_pool_t*
+mem_pool_create(
+/*============*/
+			/* out: memory pool */
+	ulint	size)	/* in: pool size in bytes */
+{
+	mem_pool_t*	pool;
+	mem_area_t*	area;
+	ulint		i;
+	ulint		used;
+
+	pool = ut_malloc(sizeof(mem_pool_t));
+
+	/* We do not set the memory to zero (FALSE) in the pool,
+	but only when allocated at a higher level in mem0mem.c.
+	This is to avoid masking useful Purify warnings. */
+
+	pool->buf = ut_malloc_low(size, FALSE, TRUE);
+	pool->size = size;
+
+	mutex_create(&pool->mutex, SYNC_MEM_POOL);
+
+	/* Initialize the free lists */
+
+	for (i = 0; i < 64; i++) {
+
+		UT_LIST_INIT(pool->free_list[i]);
+	}
+
+	used = 0;
+
+	while (size - used >= MEM_AREA_MIN_SIZE) {
+
+		i = ut_2_log(size - used);
+
+		if (ut_2_exp(i) > size - used) {
+
+			/* ut_2_log rounds upward */
+
+			i--;
+		}
+
+		area = (mem_area_t*)(pool->buf + used);
+
+		mem_area_set_size(area, ut_2_exp(i));
+		mem_area_set_free(area, TRUE);
+		UNIV_MEM_FREE(MEM_AREA_EXTRA_SIZE + (byte*) area,
+			      ut_2_exp(i) - MEM_AREA_EXTRA_SIZE);
+
+		UT_LIST_ADD_FIRST(free_list, pool->free_list[i], area);
+
+		used = used + ut_2_exp(i);
+	}
+
+	ut_ad(size >= used);
+
+	pool->reserved = 0;
+
+	return(pool);
+}
+
+/************************************************************************
+Fills the specified free list. */
+static
+ibool
+mem_pool_fill_free_list(
+/*====================*/
+				/* out: TRUE if we were able to insert a
+				block to the free list */
+	ulint		i,	/* in: free list index */
+	mem_pool_t*	pool)	/* in: memory pool */
+{
+	mem_area_t*	area;
+	mem_area_t*	area2;
+	ibool		ret;
+
+	ut_ad(mutex_own(&(pool->mutex)));
+
+	if (UNIV_UNLIKELY(i >= 63)) {
+		/* We come here when we have run out of space in the
+		memory pool: */
+
+		return(FALSE);
+	}
+
+	area = UT_LIST_GET_FIRST(pool->free_list[i + 1]);
+
+	if (area == NULL) {
+		if (UT_LIST_GET_LEN(pool->free_list[i + 1]) > 0) {
+			ut_print_timestamp(stderr);
+
+			fprintf(stderr,
+				"  InnoDB: Error: mem pool free list %lu"
+				" length is %lu\n"
+				"InnoDB: though the list is empty!\n",
+				(ulong) i + 1,
+				(ulong)
+				UT_LIST_GET_LEN(pool->free_list[i + 1]));
+		}
+
+		ret = mem_pool_fill_free_list(i + 1, pool);
+
+		if (ret == FALSE) {
+
+			return(FALSE);
+		}
+
+		area = UT_LIST_GET_FIRST(pool->free_list[i + 1]);
+	}
+
+	if (UNIV_UNLIKELY(UT_LIST_GET_LEN(pool->free_list[i + 1]) == 0)) {
+		mem_analyze_corruption(area);
+
+		ut_error;
+	}
+
+	UT_LIST_REMOVE(free_list, pool->free_list[i + 1], area);
+
+	area2 = (mem_area_t*)(((byte*)area) + ut_2_exp(i));
+	UNIV_MEM_ALLOC(area2, MEM_AREA_EXTRA_SIZE);
+
+	mem_area_set_size(area2, ut_2_exp(i));
+	mem_area_set_free(area2, TRUE);
+
+	UT_LIST_ADD_FIRST(free_list, pool->free_list[i], area2);
+
+	mem_area_set_size(area, ut_2_exp(i));
+
+	UT_LIST_ADD_FIRST(free_list, pool->free_list[i], area);
+
+	return(TRUE);
+}
+
+/************************************************************************
+Allocates memory from a pool. NOTE: This low-level function should only be
+used in mem0mem.*! */
+UNIV_INTERN
+void*
+mem_area_alloc(
+/*===========*/
+				/* out, own: allocated memory buffer */
+	ulint*		psize,	/* in: requested size in bytes; for optimum
+				space usage, the size should be a power of 2
+				minus MEM_AREA_EXTRA_SIZE;
+				out: allocated size in bytes (greater than
+				or equal to the requested size) */
+	mem_pool_t*	pool)	/* in: memory pool */
+{
+	mem_area_t*	area;
+	ulint		size;
+	ulint		n;
+	ibool		ret;
+
+	/* If we are using os allocator just make a simple call
+	to malloc */
+	if (UNIV_LIKELY(srv_use_sys_malloc)) {
+		return(malloc(*psize));
+	}
+
+	size = *psize;
+	n = ut_2_log(ut_max(size + MEM_AREA_EXTRA_SIZE, MEM_AREA_MIN_SIZE));
+
+	mutex_enter(&(pool->mutex));
+	mem_n_threads_inside++;
+
+	ut_a(mem_n_threads_inside == 1);
+
+	area = UT_LIST_GET_FIRST(pool->free_list[n]);
+
+	if (area == NULL) {
+		ret = mem_pool_fill_free_list(n, pool);
+
+		if (ret == FALSE) {
+			/* Out of memory in memory pool: we try to allocate
+			from the operating system with the regular malloc: */
+
+			mem_n_threads_inside--;
+			mutex_exit(&(pool->mutex));
+
+			return(ut_malloc(size));
+		}
+
+		area = UT_LIST_GET_FIRST(pool->free_list[n]);
+	}
+
+	if (!mem_area_get_free(area)) {
+		fprintf(stderr,
+			"InnoDB: Error: Removing element from mem pool"
+			" free list %lu though the\n"
+			"InnoDB: element is not marked free!\n",
+			(ulong) n);
+
+		mem_analyze_corruption(area);
+
+		/* Try to analyze a strange assertion failure reported at
+		mysql@lists.mysql.com where the free bit IS 1 in the
+		hex dump above */
+
+		if (mem_area_get_free(area)) {
+			fprintf(stderr,
+				"InnoDB: Probably a race condition"
+				" because now the area is marked free!\n");
+		}
+
+		ut_error;
+	}
+
+	if (UT_LIST_GET_LEN(pool->free_list[n]) == 0) {
+		fprintf(stderr,
+			"InnoDB: Error: Removing element from mem pool"
+			" free list %lu\n"
+			"InnoDB: though the list length is 0!\n",
+			(ulong) n);
+		mem_analyze_corruption(area);
+
+		ut_error;
+	}
+
+	ut_ad(mem_area_get_size(area) == ut_2_exp(n));
+
+	mem_area_set_free(area, FALSE);
+
+	UT_LIST_REMOVE(free_list, pool->free_list[n], area);
+
+	pool->reserved += mem_area_get_size(area);
+
+	mem_n_threads_inside--;
+	mutex_exit(&(pool->mutex));
+
+	ut_ad(mem_pool_validate(pool));
+
+	*psize = ut_2_exp(n) - MEM_AREA_EXTRA_SIZE;
+	UNIV_MEM_ALLOC(MEM_AREA_EXTRA_SIZE + (byte*)area, *psize);
+
+	return((void*)(MEM_AREA_EXTRA_SIZE + ((byte*)area)));
+}
+
+/************************************************************************
+Gets the buddy of an area, if it exists in pool. */
+UNIV_INLINE
+mem_area_t*
+mem_area_get_buddy(
+/*===============*/
+				/* out: the buddy, NULL if no buddy in pool */
+	mem_area_t*	area,	/* in: memory area */
+	ulint		size,	/* in: memory area size */
+	mem_pool_t*	pool)	/* in: memory pool */
+{
+	mem_area_t*	buddy;
+
+	ut_ad(size != 0);
+
+	if (((((byte*)area) - pool->buf) % (2 * size)) == 0) {
+
+		/* The buddy is in a higher address */
+
+		buddy = (mem_area_t*)(((byte*)area) + size);
+
+		if ((((byte*)buddy) - pool->buf) + size > pool->size) {
+
+			/* The buddy is not wholly contained in the pool:
+			there is no buddy */
+
+			buddy = NULL;
+		}
+	} else {
+		/* The buddy is in a lower address; NOTE that area cannot
+		be at the pool lower end, because then we would end up to
+		the upper branch in this if-clause: the remainder would be
+		0 */
+
+		buddy = (mem_area_t*)(((byte*)area) - size);
+	}
+
+	return(buddy);
+}
+
+/************************************************************************
+Frees memory to a pool. */
+UNIV_INTERN
+void
+mem_area_free(
+/*==========*/
+	void*		ptr,	/* in, own: pointer to allocated memory
+				buffer */
+	mem_pool_t*	pool)	/* in: memory pool */
+{
+	mem_area_t*	area;
+	mem_area_t*	buddy;
+	void*		new_ptr;
+	ulint		size;
+	ulint		n;
+
+	if (UNIV_LIKELY(srv_use_sys_malloc)) {
+		free(ptr);
+
+		return;
+	}
+
+	/* It may be that the area was really allocated from the OS with
+	regular malloc: check if ptr points within our memory pool */
+
+	if ((byte*)ptr < pool->buf || (byte*)ptr >= pool->buf + pool->size) {
+		ut_free(ptr);
+
+		return;
+	}
+
+	area = (mem_area_t*) (((byte*)ptr) - MEM_AREA_EXTRA_SIZE);
+
+	if (mem_area_get_free(area)) {
+		fprintf(stderr,
+			"InnoDB: Error: Freeing element to mem pool"
+			" free list though the\n"
+			"InnoDB: element is marked free!\n");
+
+		mem_analyze_corruption(area);
+		ut_error;
+	}
+
+	size = mem_area_get_size(area);
+	UNIV_MEM_FREE(ptr, size - MEM_AREA_EXTRA_SIZE);
+
+	if (size == 0) {
+		fprintf(stderr,
+			"InnoDB: Error: Mem area size is 0. Possibly a"
+			" memory overrun of the\n"
+			"InnoDB: previous allocated area!\n");
+
+		mem_analyze_corruption(area);
+		ut_error;
+	}
+
+#ifdef UNIV_LIGHT_MEM_DEBUG
+	if (((byte*)area) + size < pool->buf + pool->size) {
+
+		ulint	next_size;
+
+		next_size = mem_area_get_size(
+			(mem_area_t*)(((byte*)area) + size));
+		if (UNIV_UNLIKELY(!next_size || !ut_is_2pow(next_size))) {
+			fprintf(stderr,
+				"InnoDB: Error: Memory area size %lu,"
+				" next area size %lu not a power of 2!\n"
+				"InnoDB: Possibly a memory overrun of"
+				" the buffer being freed here.\n",
+				(ulong) size, (ulong) next_size);
+			mem_analyze_corruption(area);
+
+			ut_error;
+		}
+	}
+#endif
+	buddy = mem_area_get_buddy(area, size, pool);
+
+	n = ut_2_log(size);
+
+	mutex_enter(&(pool->mutex));
+	mem_n_threads_inside++;
+
+	ut_a(mem_n_threads_inside == 1);
+
+	if (buddy && mem_area_get_free(buddy)
+	    && (size == mem_area_get_size(buddy))) {
+
+		/* The buddy is in a free list */
+
+		if ((byte*)buddy < (byte*)area) {
+			new_ptr = ((byte*)buddy) + MEM_AREA_EXTRA_SIZE;
+
+			mem_area_set_size(buddy, 2 * size);
+			mem_area_set_free(buddy, FALSE);
+		} else {
+			new_ptr = ptr;
+
+			mem_area_set_size(area, 2 * size);
+		}
+
+		/* Remove the buddy from its free list and merge it to area */
+
+		UT_LIST_REMOVE(free_list, pool->free_list[n], buddy);
+
+		pool->reserved += ut_2_exp(n);
+
+		mem_n_threads_inside--;
+		mutex_exit(&(pool->mutex));
+
+		mem_area_free(new_ptr, pool);
+
+		return;
+	} else {
+		UT_LIST_ADD_FIRST(free_list, pool->free_list[n], area);
+
+		mem_area_set_free(area, TRUE);
+
+		ut_ad(pool->reserved >= size);
+
+		pool->reserved -= size;
+	}
+
+	mem_n_threads_inside--;
+	mutex_exit(&(pool->mutex));
+
+	ut_ad(mem_pool_validate(pool));
+}
+
+/************************************************************************
+Validates a memory pool. */
+UNIV_INTERN
+ibool
+mem_pool_validate(
+/*==============*/
+				/* out: TRUE if ok */
+	mem_pool_t*	pool)	/* in: memory pool */
+{
+	mem_area_t*	area;
+	mem_area_t*	buddy;
+	ulint		free;
+	ulint		i;
+
+	mutex_enter(&(pool->mutex));
+
+	free = 0;
+
+	for (i = 0; i < 64; i++) {
+
+		UT_LIST_VALIDATE(free_list, mem_area_t, pool->free_list[i]);
+
+		area = UT_LIST_GET_FIRST(pool->free_list[i]);
+
+		while (area != NULL) {
+			ut_a(mem_area_get_free(area));
+			ut_a(mem_area_get_size(area) == ut_2_exp(i));
+
+			buddy = mem_area_get_buddy(area, ut_2_exp(i), pool);
+
+			ut_a(!buddy || !mem_area_get_free(buddy)
+			     || (ut_2_exp(i) != mem_area_get_size(buddy)));
+
+			area = UT_LIST_GET_NEXT(free_list, area);
+
+			free += ut_2_exp(i);
+		}
+	}
+
+	ut_a(free + pool->reserved == pool->size);
+
+	mutex_exit(&(pool->mutex));
+
+	return(TRUE);
+}
+
+/************************************************************************
+Prints info of a memory pool. */
+UNIV_INTERN
+void
+mem_pool_print_info(
+/*================*/
+	FILE*		outfile,/* in: output file to write to */
+	mem_pool_t*	pool)	/* in: memory pool */
+{
+	ulint		i;
+
+	mem_pool_validate(pool);
+
+	fprintf(outfile, "INFO OF A MEMORY POOL\n");
+
+	mutex_enter(&(pool->mutex));
+
+	for (i = 0; i < 64; i++) {
+		if (UT_LIST_GET_LEN(pool->free_list[i]) > 0) {
+
+			fprintf(outfile,
+				"Free list length %lu for"
+				" blocks of size %lu\n",
+				(ulong) UT_LIST_GET_LEN(pool->free_list[i]),
+				(ulong) ut_2_exp(i));
+		}
+	}
+
+	fprintf(outfile, "Pool size %lu, reserved %lu.\n", (ulong) pool->size,
+		(ulong) pool->reserved);
+	mutex_exit(&(pool->mutex));
+}
+
+/************************************************************************
+Returns the amount of reserved memory. */
+UNIV_INTERN
+ulint
+mem_pool_get_reserved(
+/*==================*/
+				/* out: reserved memory in bytes */
+	mem_pool_t*	pool)	/* in: memory pool */
+{
+	ulint	reserved;
+
+	mutex_enter(&(pool->mutex));
+
+	reserved = pool->reserved;
+
+	mutex_exit(&(pool->mutex));
+
+	return(reserved);
+}
diff --git a/storage/xtradb/mtr/mtr0log.c b/storage/xtradb/mtr/mtr0log.c
new file mode 100644
index 00000000000..0fe66d08c05
--- /dev/null
+++ b/storage/xtradb/mtr/mtr0log.c
@@ -0,0 +1,609 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Mini-transaction log routines
+
+Created 12/7/1995 Heikki Tuuri
+*******************************************************/
+
+#include "mtr0log.h"
+
+#ifdef UNIV_NONINL
+#include "mtr0log.ic"
+#endif
+
+#include "buf0buf.h"
+#include "dict0boot.h"
+#include "log0recv.h"
+#include "page0page.h"
+
+/************************************************************
+Catenates n bytes to the mtr log. */
+UNIV_INTERN
+void
+mlog_catenate_string(
+/*=================*/
+	mtr_t*		mtr,	/* in: mtr */
+	const byte*	str,	/* in: string to write */
+	ulint		len)	/* in: string length */
+{
+	dyn_array_t*	mlog;
+
+	if (mtr_get_log_mode(mtr) == MTR_LOG_NONE) {
+
+		return;
+	}
+
+	mlog = &(mtr->log);
+
+	dyn_push_string(mlog, str, len);
+}
+
+/************************************************************
+Writes the initial part of a log record consisting of one-byte item
+type and four-byte space and page numbers. Also pushes info
+to the mtr memo that a buffer page has been modified. */
+UNIV_INTERN
+void
+mlog_write_initial_log_record(
+/*==========================*/
+	const byte*	ptr,	/* in: pointer to (inside) a buffer
+				frame holding the file page where
+				modification is made */
+	byte		type,	/* in: log item type: MLOG_1BYTE, ... */
+	mtr_t*		mtr)	/* in: mini-transaction handle */
+{
+	byte*	log_ptr;
+
+	ut_ad(type <= MLOG_BIGGEST_TYPE);
+	ut_ad(type > MLOG_8BYTES);
+
+	log_ptr = mlog_open(mtr, 11);
+
+	/* If no logging is requested, we may return now */
+	if (log_ptr == NULL) {
+
+		return;
+	}
+
+	log_ptr = mlog_write_initial_log_record_fast(ptr, type, log_ptr, mtr);
+
+	mlog_close(mtr, log_ptr);
+}
+
+/************************************************************
+Parses an initial log record written by mlog_write_initial_log_record. */
+UNIV_INTERN
+byte*
+mlog_parse_initial_log_record(
+/*==========================*/
+			/* out: parsed record end, NULL if not a complete
+			record */
+	byte*	ptr,	/* in: buffer */
+	byte*	end_ptr,/* in: buffer end */
+	byte*	type,	/* out: log record type: MLOG_1BYTE, ... */
+	ulint*	space,	/* out: space id */
+	ulint*	page_no)/* out: page number */
+{
+	if (end_ptr < ptr + 1) {
+
+		return(NULL);
+	}
+
+	*type = (byte)((ulint)*ptr & ~MLOG_SINGLE_REC_FLAG);
+	ut_ad(*type <= MLOG_BIGGEST_TYPE);
+
+	ptr++;
+
+	if (end_ptr < ptr + 2) {
+
+		return(NULL);
+	}
+
+	ptr = mach_parse_compressed(ptr, end_ptr, space);
+
+	if (ptr == NULL) {
+
+		return(NULL);
+	}
+
+	ptr = mach_parse_compressed(ptr, end_ptr, page_no);
+
+	return(ptr);
+}
+
+/************************************************************
+Parses a log record written by mlog_write_ulint or mlog_write_dulint. */
+UNIV_INTERN
+byte*
+mlog_parse_nbytes(
+/*==============*/
+			/* out: parsed record end, NULL if not a complete
+			record or a corrupt record */
+	ulint	type,	/* in: log record type: MLOG_1BYTE, ... */
+	byte*	ptr,	/* in: buffer */
+	byte*	end_ptr,/* in: buffer end */
+	byte*	page,	/* in: page where to apply the log record, or NULL */
+	void*	page_zip)/* in/out: compressed page, or NULL */
+{
+	ulint	offset;
+	ulint	val;
+	dulint	dval;
+
+	ut_a(type <= MLOG_8BYTES);
+	ut_a(!page || !page_zip || fil_page_get_type(page) != FIL_PAGE_INDEX);
+
+	if (end_ptr < ptr + 2) {
+
+		return(NULL);
+	}
+
+	offset = mach_read_from_2(ptr);
+	ptr += 2;
+
+	if (offset >= UNIV_PAGE_SIZE) {
+		recv_sys->found_corrupt_log = TRUE;
+
+		return(NULL);
+	}
+
+	if (type == MLOG_8BYTES) {
+		ptr = mach_dulint_parse_compressed(ptr, end_ptr, &dval);
+
+		if (ptr == NULL) {
+
+			return(NULL);
+		}
+
+		if (page) {
+			if (UNIV_LIKELY_NULL(page_zip)) {
+				mach_write_to_8
+					(((page_zip_des_t*) page_zip)->data
+					 + offset, dval);
+			}
+			mach_write_to_8(page + offset, dval);
+		}
+
+		return(ptr);
+	}
+
+	ptr = mach_parse_compressed(ptr, end_ptr, &val);
+
+	if (ptr == NULL) {
+
+		return(NULL);
+	}
+
+	switch (type) {
+	case MLOG_1BYTE:
+		if (UNIV_UNLIKELY(val > 0xFFUL)) {
+			goto corrupt;
+		}
+		if (page) {
+			if (UNIV_LIKELY_NULL(page_zip)) {
+				mach_write_to_1
+					(((page_zip_des_t*) page_zip)->data
+					 + offset, val);
+			}
+			mach_write_to_1(page + offset, val);
+		}
+		break;
+	case MLOG_2BYTES:
+		if (UNIV_UNLIKELY(val > 0xFFFFUL)) {
+			goto corrupt;
+		}
+		if (page) {
+			if (UNIV_LIKELY_NULL(page_zip)) {
+				mach_write_to_2
+					(((page_zip_des_t*) page_zip)->data
+					 + offset, val);
+			}
+			mach_write_to_2(page + offset, val);
+		}
+		break;
+	case MLOG_4BYTES:
+		if (page) {
+			if (UNIV_LIKELY_NULL(page_zip)) {
+				mach_write_to_4
+					(((page_zip_des_t*) page_zip)->data
+					 + offset, val);
+			}
+			mach_write_to_4(page + offset, val);
+		}
+		break;
+	default:
+	corrupt:
+		recv_sys->found_corrupt_log = TRUE;
+		ptr = NULL;
+	}
+
+	return(ptr);
+}
+
+/************************************************************
+Writes 1 - 4 bytes to a file page buffered in the buffer pool.
+Writes the corresponding log record to the mini-transaction log. */
+UNIV_INTERN
+void
+mlog_write_ulint(
+/*=============*/
+	byte*	ptr,	/* in: pointer where to write */
+	ulint	val,	/* in: value to write */
+	byte	type,	/* in: MLOG_1BYTE, MLOG_2BYTES, MLOG_4BYTES */
+	mtr_t*	mtr)	/* in: mini-transaction handle */
+{
+	byte*	log_ptr;
+
+	switch (type) {
+	case MLOG_1BYTE:
+		mach_write_to_1(ptr, val);
+		break;
+	case MLOG_2BYTES:
+		mach_write_to_2(ptr, val);
+		break;
+	case MLOG_4BYTES:
+		mach_write_to_4(ptr, val);
+		break;
+	default:
+		ut_error;
+	}
+
+	log_ptr = mlog_open(mtr, 11 + 2 + 5);
+
+	/* If no logging is requested, we may return now */
+	if (log_ptr == NULL) {
+
+		return;
+	}
+
+	log_ptr = mlog_write_initial_log_record_fast(ptr, type, log_ptr, mtr);
+
+	mach_write_to_2(log_ptr, page_offset(ptr));
+	log_ptr += 2;
+
+	log_ptr += mach_write_compressed(log_ptr, val);
+
+	mlog_close(mtr, log_ptr);
+}
+
+/************************************************************
+Writes 8 bytes to a file page buffered in the buffer pool.
+Writes the corresponding log record to the mini-transaction log. */
+UNIV_INTERN
+void
+mlog_write_dulint(
+/*==============*/
+	byte*	ptr,	/* in: pointer where to write */
+	dulint	val,	/* in: value to write */
+	mtr_t*	mtr)	/* in: mini-transaction handle */
+{
+	byte*	log_ptr;
+
+	ut_ad(ptr && mtr);
+
+	mach_write_to_8(ptr, val);
+
+	log_ptr = mlog_open(mtr, 11 + 2 + 9);
+
+	/* If no logging is requested, we may return now */
+	if (log_ptr == NULL) {
+
+		return;
+	}
+
+	log_ptr = mlog_write_initial_log_record_fast(ptr, MLOG_8BYTES,
+						     log_ptr, mtr);
+
+	mach_write_to_2(log_ptr, page_offset(ptr));
+	log_ptr += 2;
+
+	log_ptr += mach_dulint_write_compressed(log_ptr, val);
+
+	mlog_close(mtr, log_ptr);
+}
+
+/************************************************************
+Writes a string to a file page buffered in the buffer pool. Writes the
+corresponding log record to the mini-transaction log. */
+UNIV_INTERN
+void
+mlog_write_string(
+/*==============*/
+	byte*		ptr,	/* in: pointer where to write */
+	const byte*	str,	/* in: string to write */
+	ulint		len,	/* in: string length */
+	mtr_t*		mtr)	/* in: mini-transaction handle */
+{
+	ut_ad(ptr && mtr);
+	ut_a(len < UNIV_PAGE_SIZE);
+
+	memcpy(ptr, str, len);
+
+	mlog_log_string(ptr, len, mtr);
+}
+
+/************************************************************
+Logs a write of a string to a file page buffered in the buffer pool.
+Writes the corresponding log record to the mini-transaction log. */
+UNIV_INTERN
+void
+mlog_log_string(
+/*============*/
+	byte*	ptr,	/* in: pointer written to */
+	ulint	len,	/* in: string length */
+	mtr_t*	mtr)	/* in: mini-transaction handle */
+{
+	byte*	log_ptr;
+
+	ut_ad(ptr && mtr);
+	ut_ad(len <= UNIV_PAGE_SIZE);
+
+	log_ptr = mlog_open(mtr, 30);
+
+	/* If no logging is requested, we may return now */
+	if (log_ptr == NULL) {
+
+		return;
+	}
+
+	log_ptr = mlog_write_initial_log_record_fast(ptr, MLOG_WRITE_STRING,
+						     log_ptr, mtr);
+	mach_write_to_2(log_ptr, page_offset(ptr));
+	log_ptr += 2;
+
+	mach_write_to_2(log_ptr, len);
+	log_ptr += 2;
+
+	mlog_close(mtr, log_ptr);
+
+	mlog_catenate_string(mtr, ptr, len);
+}
+
+/************************************************************
+Parses a log record written by mlog_write_string. */
+UNIV_INTERN
+byte*
+mlog_parse_string(
+/*==============*/
+			/* out: parsed record end, NULL if not a complete
+			record */
+	byte*	ptr,	/* in: buffer */
+	byte*	end_ptr,/* in: buffer end */
+	byte*	page,	/* in: page where to apply the log record, or NULL */
+	void*	page_zip)/* in/out: compressed page, or NULL */
+{
+	ulint	offset;
+	ulint	len;
+
+	ut_a(!page || !page_zip || fil_page_get_type(page) != FIL_PAGE_INDEX);
+
+	if (end_ptr < ptr + 4) {
+
+		return(NULL);
+	}
+
+	offset = mach_read_from_2(ptr);
+	ptr += 2;
+	len = mach_read_from_2(ptr);
+	ptr += 2;
+
+	if (UNIV_UNLIKELY(offset >= UNIV_PAGE_SIZE)
+			|| UNIV_UNLIKELY(len + offset) > UNIV_PAGE_SIZE) {
+		recv_sys->found_corrupt_log = TRUE;
+
+		return(NULL);
+	}
+
+	if (end_ptr < ptr + len) {
+
+		return(NULL);
+	}
+
+	if (page) {
+		if (UNIV_LIKELY_NULL(page_zip)) {
+			memcpy(((page_zip_des_t*) page_zip)->data
+				+ offset, ptr, len);
+		}
+		memcpy(page + offset, ptr, len);
+	}
+
+	return(ptr + len);
+}
+
+/************************************************************
+Opens a buffer for mlog, writes the initial log record and,
+if needed, the field lengths of an index. */
+UNIV_INTERN
+byte*
+mlog_open_and_write_index(
+/*======================*/
+				/* out: buffer, NULL if log mode
+				MTR_LOG_NONE */
+	mtr_t*		mtr,	/* in: mtr */
+	byte*		rec,	/* in: index record or page */
+	dict_index_t*	index,	/* in: record descriptor */
+	byte		type,	/* in: log item type */
+	ulint		size)	/* in: requested buffer size in bytes
+				(if 0, calls mlog_close() and returns NULL) */
+{
+	byte*		log_ptr;
+	const byte*	log_start;
+	const byte*	log_end;
+
+	ut_ad(!!page_rec_is_comp(rec) == dict_table_is_comp(index->table));
+
+	if (!page_rec_is_comp(rec)) {
+		log_start = log_ptr = mlog_open(mtr, 11 + size);
+		if (!log_ptr) {
+			return(NULL); /* logging is disabled */
+		}
+		log_ptr = mlog_write_initial_log_record_fast(rec, type,
+							     log_ptr, mtr);
+		log_end = log_ptr + 11 + size;
+	} else {
+		ulint	i;
+		ulint	n	= dict_index_get_n_fields(index);
+		/* total size needed */
+		ulint	total	= 11 + size + (n + 2) * 2;
+		ulint	alloc	= total;
+		/* allocate at most DYN_ARRAY_DATA_SIZE at a time */
+		if (alloc > DYN_ARRAY_DATA_SIZE) {
+			alloc = DYN_ARRAY_DATA_SIZE;
+		}
+		log_start = log_ptr = mlog_open(mtr, alloc);
+		if (!log_ptr) {
+			return(NULL); /* logging is disabled */
+		}
+		log_end = log_ptr + alloc;
+		log_ptr = mlog_write_initial_log_record_fast(rec, type,
+							     log_ptr, mtr);
+		mach_write_to_2(log_ptr, n);
+		log_ptr += 2;
+		mach_write_to_2(log_ptr,
+				dict_index_get_n_unique_in_tree(index));
+		log_ptr += 2;
+		for (i = 0; i < n; i++) {
+			dict_field_t*		field;
+			const dict_col_t*	col;
+			ulint			len;
+
+			field = dict_index_get_nth_field(index, i);
+			col = dict_field_get_col(field);
+			len = field->fixed_len;
+			ut_ad(len < 0x7fff);
+			if (len == 0
+			    && (col->len > 255 || col->mtype == DATA_BLOB)) {
+				/* variable-length field
+				with maximum length > 255 */
+				len = 0x7fff;
+			}
+			if (col->prtype & DATA_NOT_NULL) {
+				len |= 0x8000;
+			}
+			if (log_ptr + 2 > log_end) {
+				mlog_close(mtr, log_ptr);
+				ut_a(total > (ulint) (log_ptr - log_start));
+				total -= log_ptr - log_start;
+				alloc = total;
+				if (alloc > DYN_ARRAY_DATA_SIZE) {
+					alloc = DYN_ARRAY_DATA_SIZE;
+				}
+				log_start = log_ptr = mlog_open(mtr, alloc);
+				if (!log_ptr) {
+					return(NULL); /* logging is disabled */
+				}
+				log_end = log_ptr + alloc;
+			}
+			mach_write_to_2(log_ptr, len);
+			log_ptr += 2;
+		}
+	}
+	if (size == 0) {
+		mlog_close(mtr, log_ptr);
+		log_ptr = NULL;
+	} else if (log_ptr + size > log_end) {
+		mlog_close(mtr, log_ptr);
+		log_ptr = mlog_open(mtr, size);
+	}
+	return(log_ptr);
+}
+
+/************************************************************
+Parses a log record written by mlog_open_and_write_index. */
+UNIV_INTERN
+byte*
+mlog_parse_index(
+/*=============*/
+				/* out: parsed record end,
+				NULL if not a complete record */
+	byte*		ptr,	/* in: buffer */
+	byte*		end_ptr,/* in: buffer end */
+				/* out: new value of log_ptr */
+	ibool		comp,	/* in: TRUE=compact record format */
+	dict_index_t**	index)	/* out, own: dummy index */
+{
+	ulint		i, n, n_uniq;
+	dict_table_t*	table;
+	dict_index_t*	ind;
+
+	ut_ad(comp == FALSE || comp == TRUE);
+
+	if (comp) {
+		if (end_ptr < ptr + 4) {
+			return(NULL);
+		}
+		n = mach_read_from_2(ptr);
+		ptr += 2;
+		n_uniq = mach_read_from_2(ptr);
+		ptr += 2;
+		ut_ad(n_uniq <= n);
+		if (end_ptr < ptr + n * 2) {
+			return(NULL);
+		}
+	} else {
+		n = n_uniq = 1;
+	}
+	table = dict_mem_table_create("LOG_DUMMY", DICT_HDR_SPACE, n,
+				      comp ? DICT_TF_COMPACT : 0);
+	ind = dict_mem_index_create("LOG_DUMMY", "LOG_DUMMY",
+				    DICT_HDR_SPACE, 0, n);
+	ind->table = table;
+	ind->n_uniq = (unsigned int) n_uniq;
+	if (n_uniq != n) {
+		ut_a(n_uniq + DATA_ROLL_PTR <= n);
+		ind->type = DICT_CLUSTERED;
+	}
+	if (comp) {
+		for (i = 0; i < n; i++) {
+			ulint	len = mach_read_from_2(ptr);
+			ptr += 2;
+			/* The high-order bit of len is the NOT NULL flag;
+			the rest is 0 or 0x7fff for variable-length fields,
+			and 1..0x7ffe for fixed-length fields. */
+			dict_mem_table_add_col(
+				table, NULL, NULL,
+				((len + 1) & 0x7fff) <= 1
+				? DATA_BINARY : DATA_FIXBINARY,
+				len & 0x8000 ? DATA_NOT_NULL : 0,
+				len & 0x7fff);
+
+			dict_index_add_col(ind, table,
+					   dict_table_get_nth_col(table, i),
+					   0);
+		}
+		dict_table_add_system_columns(table, table->heap);
+		if (n_uniq != n) {
+			/* Identify DB_TRX_ID and DB_ROLL_PTR in the index. */
+			ut_a(DATA_TRX_ID_LEN
+			     == dict_index_get_nth_col(ind, DATA_TRX_ID - 1
+						       + n_uniq)->len);
+			ut_a(DATA_ROLL_PTR_LEN
+			     == dict_index_get_nth_col(ind, DATA_ROLL_PTR - 1
+						       + n_uniq)->len);
+			ind->fields[DATA_TRX_ID - 1 + n_uniq].col
+				= &table->cols[n + DATA_TRX_ID];
+			ind->fields[DATA_ROLL_PTR - 1 + n_uniq].col
+				= &table->cols[n + DATA_ROLL_PTR];
+		}
+	}
+	/* avoid ut_ad(index->cached) in dict_index_get_n_unique_in_tree */
+	ind->cached = TRUE;
+	*index = ind;
+	return(ptr);
+}
diff --git a/storage/xtradb/mtr/mtr0mtr.c b/storage/xtradb/mtr/mtr0mtr.c
new file mode 100644
index 00000000000..881751b560e
--- /dev/null
+++ b/storage/xtradb/mtr/mtr0mtr.c
@@ -0,0 +1,345 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Mini-transaction buffer
+
+Created 11/26/1995 Heikki Tuuri
+*******************************************************/
+
+#include "mtr0mtr.h"
+
+#ifdef UNIV_NONINL
+#include "mtr0mtr.ic"
+#endif
+
+#include "buf0buf.h"
+#include "page0types.h"
+#include "mtr0log.h"
+#include "log0log.h"
+
+/*********************************************************************
+Releases the item in the slot given. */
+UNIV_INLINE
+void
+mtr_memo_slot_release(
+/*==================*/
+	mtr_t*			mtr,	/* in: mtr */
+	mtr_memo_slot_t*	slot)	/* in: memo slot */
+{
+	void*	object;
+	ulint	type;
+
+	ut_ad(mtr && slot);
+
+	object = slot->object;
+	type = slot->type;
+
+	if (UNIV_LIKELY(object != NULL)) {
+		if (type <= MTR_MEMO_BUF_FIX) {
+			buf_page_release((buf_block_t*)object, type, mtr);
+		} else if (type == MTR_MEMO_S_LOCK) {
+			rw_lock_s_unlock((rw_lock_t*)object);
+#ifdef UNIV_DEBUG
+		} else if (type != MTR_MEMO_X_LOCK) {
+			ut_ad(type == MTR_MEMO_MODIFY);
+			ut_ad(mtr_memo_contains(mtr, object,
+						MTR_MEMO_PAGE_X_FIX));
+#endif /* UNIV_DEBUG */
+		} else {
+			rw_lock_x_unlock((rw_lock_t*)object);
+		}
+	}
+
+	slot->object = NULL;
+}
+
+/**************************************************************
+Releases the mlocks and other objects stored in an mtr memo. They are released
+in the order opposite to which they were pushed to the memo. NOTE! It is
+essential that the x-rw-lock on a modified buffer page is not released before
+buf_page_note_modification is called for that page! Otherwise, some thread
+might race to modify it, and the flush list sort order on lsn would be
+destroyed. */
+UNIV_INLINE
+void
+mtr_memo_pop_all(
+/*=============*/
+	mtr_t*	mtr)	/* in: mtr */
+{
+	mtr_memo_slot_t* slot;
+	dyn_array_t*	memo;
+	ulint		offset;
+
+	ut_ad(mtr);
+	ut_ad(mtr->magic_n == MTR_MAGIC_N);
+	ut_ad(mtr->state == MTR_COMMITTING); /* Currently only used in
+					     commit */
+	memo = &(mtr->memo);
+
+	offset = dyn_array_get_data_size(memo);
+
+	while (offset > 0) {
+		offset -= sizeof(mtr_memo_slot_t);
+		slot = dyn_array_get_element(memo, offset);
+
+		mtr_memo_slot_release(mtr, slot);
+	}
+}
+
+/****************************************************************
+Writes the contents of a mini-transaction log, if any, to the database log. */
+static
+void
+mtr_log_reserve_and_write(
+/*======================*/
+	mtr_t*	mtr)	/* in: mtr */
+{
+	dyn_array_t*	mlog;
+	dyn_block_t*	block;
+	ulint		data_size;
+	ibool		success;
+	byte*		first_data;
+
+	ut_ad(mtr);
+
+	mlog = &(mtr->log);
+
+	first_data = dyn_block_get_data(mlog);
+
+	if (mtr->n_log_recs > 1) {
+		mlog_catenate_ulint(mtr, MLOG_MULTI_REC_END, MLOG_1BYTE);
+	} else {
+		*first_data = (byte)((ulint)*first_data
+				     | MLOG_SINGLE_REC_FLAG);
+	}
+
+	if (mlog->heap == NULL) {
+		mtr->end_lsn = log_reserve_and_write_fast(
+			first_data, dyn_block_get_used(mlog),
+			&(mtr->start_lsn), &success);
+		if (success) {
+
+			return;
+		}
+	}
+
+	data_size = dyn_array_get_data_size(mlog);
+
+	/* Open the database log for log_write_low */
+	mtr->start_lsn = log_reserve_and_open(data_size);
+
+	if (mtr->log_mode == MTR_LOG_ALL) {
+
+		block = mlog;
+
+		while (block != NULL) {
+			log_write_low(dyn_block_get_data(block),
+				      dyn_block_get_used(block));
+			block = dyn_array_get_next_block(mlog, block);
+		}
+	} else {
+		ut_ad(mtr->log_mode == MTR_LOG_NONE);
+		/* Do nothing */
+	}
+
+	mtr->end_lsn = log_close();
+}
+
+/*******************************************************************
+Commits a mini-transaction. */
+UNIV_INTERN
+void
+mtr_commit(
+/*=======*/
+	mtr_t*	mtr)	/* in: mini-transaction */
+{
+	ibool		write_log;
+
+	ut_ad(mtr);
+	ut_ad(mtr->magic_n == MTR_MAGIC_N);
+	ut_ad(mtr->state == MTR_ACTIVE);
+	ut_d(mtr->state = MTR_COMMITTING);
+
+	write_log = mtr->modifications && mtr->n_log_recs;
+
+	if (write_log) {
+		mtr_log_reserve_and_write(mtr);
+	}
+
+	/* We first update the modification info to buffer pages, and only
+	after that release the log mutex: this guarantees that when the log
+	mutex is free, all buffer pages contain an up-to-date info of their
+	modifications. This fact is used in making a checkpoint when we look
+	at the oldest modification of any page in the buffer pool. It is also
+	required when we insert modified buffer pages in to the flush list
+	which must be sorted on oldest_modification. */
+
+	mtr_memo_pop_all(mtr);
+
+	if (write_log) {
+		log_release();
+	}
+
+	ut_d(mtr->state = MTR_COMMITTED);
+	dyn_array_free(&(mtr->memo));
+	dyn_array_free(&(mtr->log));
+}
+
+/**************************************************************
+Releases the latches stored in an mtr memo down to a savepoint.
+NOTE! The mtr must not have made changes to buffer pages after the
+savepoint, as these can be handled only by mtr_commit. */
+UNIV_INTERN
+void
+mtr_rollback_to_savepoint(
+/*======================*/
+	mtr_t*	mtr,		/* in: mtr */
+	ulint	savepoint)	/* in: savepoint */
+{
+	mtr_memo_slot_t* slot;
+	dyn_array_t*	memo;
+	ulint		offset;
+
+	ut_ad(mtr);
+	ut_ad(mtr->magic_n == MTR_MAGIC_N);
+	ut_ad(mtr->state == MTR_ACTIVE);
+
+	memo = &(mtr->memo);
+
+	offset = dyn_array_get_data_size(memo);
+	ut_ad(offset >= savepoint);
+
+	while (offset > savepoint) {
+		offset -= sizeof(mtr_memo_slot_t);
+
+		slot = dyn_array_get_element(memo, offset);
+
+		ut_ad(slot->type != MTR_MEMO_MODIFY);
+		mtr_memo_slot_release(mtr, slot);
+	}
+}
+
+/*******************************************************
+Releases an object in the memo stack. */
+UNIV_INTERN
+void
+mtr_memo_release(
+/*=============*/
+	mtr_t*	mtr,	/* in: mtr */
+	void*	object,	/* in: object */
+	ulint	type)	/* in: object type: MTR_MEMO_S_LOCK, ... */
+{
+	mtr_memo_slot_t* slot;
+	dyn_array_t*	memo;
+	ulint		offset;
+
+	ut_ad(mtr);
+	ut_ad(mtr->magic_n == MTR_MAGIC_N);
+	ut_ad(mtr->state == MTR_ACTIVE);
+
+	memo = &(mtr->memo);
+
+	offset = dyn_array_get_data_size(memo);
+
+	while (offset > 0) {
+		offset -= sizeof(mtr_memo_slot_t);
+
+		slot = dyn_array_get_element(memo, offset);
+
+		if ((object == slot->object) && (type == slot->type)) {
+
+			mtr_memo_slot_release(mtr, slot);
+
+			break;
+		}
+	}
+}
+
+/************************************************************
+Reads 1 - 4 bytes from a file page buffered in the buffer pool. */
+UNIV_INTERN
+ulint
+mtr_read_ulint(
+/*===========*/
+				/* out: value read */
+	const byte*	ptr,	/* in: pointer from where to read */
+	ulint		type,	/* in: MLOG_1BYTE, MLOG_2BYTES, MLOG_4BYTES */
+	mtr_t*		mtr __attribute__((unused)))
+				/* in: mini-transaction handle */
+{
+	ut_ad(mtr->state == MTR_ACTIVE);
+	ut_ad(mtr_memo_contains_page(mtr, ptr, MTR_MEMO_PAGE_S_FIX)
+	      || mtr_memo_contains_page(mtr, ptr, MTR_MEMO_PAGE_X_FIX));
+	if (type == MLOG_1BYTE) {
+		return(mach_read_from_1(ptr));
+	} else if (type == MLOG_2BYTES) {
+		return(mach_read_from_2(ptr));
+	} else {
+		ut_ad(type == MLOG_4BYTES);
+		return(mach_read_from_4(ptr));
+	}
+}
+
+/************************************************************
+Reads 8 bytes from a file page buffered in the buffer pool. */
+UNIV_INTERN
+dulint
+mtr_read_dulint(
+/*============*/
+				/* out: value read */
+	const byte*	ptr,	/* in: pointer from where to read */
+	mtr_t*		mtr __attribute__((unused)))
+				/* in: mini-transaction handle */
+{
+	ut_ad(mtr->state == MTR_ACTIVE);
+	ut_ad(mtr_memo_contains_page(mtr, ptr, MTR_MEMO_PAGE_S_FIX)
+	      || mtr_memo_contains_page(mtr, ptr, MTR_MEMO_PAGE_X_FIX));
+	return(mach_read_from_8(ptr));
+}
+
+#ifdef UNIV_DEBUG
+/**************************************************************
+Checks if memo contains the given page. */
+UNIV_INTERN
+ibool
+mtr_memo_contains_page(
+/*===================*/
+				/* out: TRUE if contains */
+	mtr_t*		mtr,	/* in: mtr */
+	const byte*	ptr,	/* in: pointer to buffer frame */
+	ulint		type)	/* in: type of object */
+{
+	return(mtr_memo_contains(mtr, buf_block_align(ptr), type));
+}
+
+/*************************************************************
+Prints info of an mtr handle. */
+UNIV_INTERN
+void
+mtr_print(
+/*======*/
+	mtr_t*	mtr)	/* in: mtr */
+{
+	fprintf(stderr,
+		"Mini-transaction handle: memo size %lu bytes"
+		" log size %lu bytes\n",
+		(ulong) dyn_array_get_data_size(&(mtr->memo)),
+		(ulong) dyn_array_get_data_size(&(mtr->log)));
+}
+#endif /* UNIV_DEBUG */
diff --git a/storage/xtradb/os/os0file.c b/storage/xtradb/os/os0file.c
new file mode 100644
index 00000000000..2d130899622
--- /dev/null
+++ b/storage/xtradb/os/os0file.c
@@ -0,0 +1,4248 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The interface to the operating system file i/o primitives
+
+Created 10/21/1995 Heikki Tuuri
+*******************************************************/
+
+#include "os0file.h"
+#include "os0sync.h"
+#include "os0thread.h"
+#include "ut0mem.h"
+#include "srv0srv.h"
+#include "srv0start.h"
+#include "fil0fil.h"
+#include "buf0buf.h"
+
+#if defined(UNIV_HOTBACKUP) && defined(__WIN__)
+/* Add includes for the _stat() call to compile on Windows */
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <errno.h>
+#endif /* UNIV_HOTBACKUP */
+
+/* This specifies the file permissions InnoDB uses when it creates files in
+Unix; the value of os_innodb_umask is initialized in ha_innodb.cc to
+my_umask */
+
+#ifndef __WIN__
+UNIV_INTERN ulint	os_innodb_umask
+			= S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP;
+#else
+UNIV_INTERN ulint	os_innodb_umask		= 0;
+#endif
+
+#ifdef UNIV_DO_FLUSH
+/* If the following is set to TRUE, we do not call os_file_flush in every
+os_file_write. We can set this TRUE when the doublewrite buffer is used. */
+UNIV_INTERN ibool	os_do_not_call_flush_at_each_write	= FALSE;
+#else
+/* We do not call os_file_flush in every os_file_write. */
+#endif /* UNIV_DO_FLUSH */
+
+/* We use these mutexes to protect lseek + file i/o operation, if the
+OS does not provide an atomic pread or pwrite, or similar */
+#define OS_FILE_N_SEEK_MUTEXES	16
+UNIV_INTERN os_mutex_t	os_file_seek_mutexes[OS_FILE_N_SEEK_MUTEXES];
+
+/* In simulated aio, merge at most this many consecutive i/os */
+#define OS_AIO_MERGE_N_CONSECUTIVE	64
+
+/* If this flag is TRUE, then we will use the native aio of the
+OS (provided we compiled Innobase with it in), otherwise we will
+use simulated aio we build below with threads */
+
+UNIV_INTERN ibool	os_aio_use_native_aio	= FALSE;
+
+UNIV_INTERN ibool	os_aio_print_debug	= FALSE;
+
+/* The aio array slot structure */
+typedef struct os_aio_slot_struct	os_aio_slot_t;
+
+struct os_aio_slot_struct{
+	ibool		is_read;	/* TRUE if a read operation */
+	ulint		pos;		/* index of the slot in the aio
+					array */
+	ibool		reserved;	/* TRUE if this slot is reserved */
+	time_t		reservation_time;/* time when reserved */
+	ulint		len;		/* length of the block to read or
+					write */
+	byte*		buf;		/* buffer used in i/o */
+	ulint		type;		/* OS_FILE_READ or OS_FILE_WRITE */
+	ulint		offset;		/* 32 low bits of file offset in
+					bytes */
+	ulint		offset_high;	/* 32 high bits of file offset */
+	os_file_t	file;		/* file where to read or write */
+	const char*	name;		/* file name or path */
+	ibool		io_already_done;/* used only in simulated aio:
+					TRUE if the physical i/o already
+					made and only the slot message
+					needs to be passed to the caller
+					of os_aio_simulated_handle */
+	fil_node_t*	message1;	/* message which is given by the */
+	void*		message2;	/* the requester of an aio operation
+					and which can be used to identify
+					which pending aio operation was
+					completed */
+#ifdef WIN_ASYNC_IO
+	os_event_t	event;		/* event object we need in the
+					OVERLAPPED struct */
+	OVERLAPPED	control;	/* Windows control block for the
+					aio request */
+#endif
+};
+
+/* The aio array structure */
+typedef struct os_aio_array_struct	os_aio_array_t;
+
+struct os_aio_array_struct{
+	os_mutex_t	mutex;	  /* the mutex protecting the aio array */
+	os_event_t	not_full; /* The event which is set to the signaled
+				  state when there is space in the aio
+				  outside the ibuf segment */
+	os_event_t	is_empty; /* The event which is set to the signaled
+				  state when there are no pending i/os
+				  in this array */
+	ulint		n_slots;  /* Total number of slots in the aio array.
+				  This must be divisible by n_threads. */
+	ulint		n_segments;/* Number of segments in the aio array of
+				  pending aio requests. A thread can wait
+				  separately for any one of the segments. */
+	ulint		n_reserved;/* Number of reserved slots in the
+				  aio array outside the ibuf segment */
+	os_aio_slot_t*	slots;	  /* Pointer to the slots in the array */
+#ifdef __WIN__
+	os_native_event_t* native_events;
+				  /* Pointer to an array of OS native event
+				  handles where we copied the handles from
+				  slots, in the same order. This can be used
+				  in WaitForMultipleObjects; used only in
+				  Windows */
+#endif
+};
+
+/* Array of events used in simulated aio */
+static os_event_t*	os_aio_segment_wait_events	= NULL;
+
+/* The aio arrays for non-ibuf i/o and ibuf i/o, as well as sync aio. These
+are NULL when the module has not yet been initialized. */
+static os_aio_array_t*	os_aio_read_array	= NULL;
+static os_aio_array_t*	os_aio_write_array	= NULL;
+static os_aio_array_t*	os_aio_ibuf_array	= NULL;
+static os_aio_array_t*	os_aio_log_array	= NULL;
+static os_aio_array_t*	os_aio_sync_array	= NULL;
+
+static ulint	os_aio_n_segments	= ULINT_UNDEFINED;
+
+/* If the following is TRUE, read i/o handler threads try to
+wait until a batch of new read requests have been posted */
+static ibool	os_aio_recommend_sleep_for_read_threads	= FALSE;
+
+UNIV_INTERN ulint	os_n_file_reads		= 0;
+UNIV_INTERN ulint	os_bytes_read_since_printout = 0;
+UNIV_INTERN ulint	os_n_file_writes	= 0;
+UNIV_INTERN ulint	os_n_fsyncs		= 0;
+UNIV_INTERN ulint	os_n_file_reads_old	= 0;
+UNIV_INTERN ulint	os_n_file_writes_old	= 0;
+UNIV_INTERN ulint	os_n_fsyncs_old		= 0;
+UNIV_INTERN time_t	os_last_printout;
+
+UNIV_INTERN ibool	os_has_said_disk_full	= FALSE;
+
+/* The mutex protecting the following counts of pending I/O operations */
+static os_mutex_t	os_file_count_mutex;
+UNIV_INTERN ulint	os_file_n_pending_preads  = 0;
+UNIV_INTERN ulint	os_file_n_pending_pwrites = 0;
+UNIV_INTERN ulint	os_n_pending_writes = 0;
+UNIV_INTERN ulint	os_n_pending_reads = 0;
+
+/***************************************************************************
+Gets the operating system version. Currently works only on Windows. */
+UNIV_INTERN
+ulint
+os_get_os_version(void)
+/*===================*/
+		  /* out: OS_WIN95, OS_WIN31, OS_WINNT, OS_WIN2000 */
+{
+#ifdef __WIN__
+	OSVERSIONINFO	  os_info;
+
+	os_info.dwOSVersionInfoSize = sizeof(OSVERSIONINFO);
+
+	ut_a(GetVersionEx(&os_info));
+
+	if (os_info.dwPlatformId == VER_PLATFORM_WIN32s) {
+		return(OS_WIN31);
+	} else if (os_info.dwPlatformId == VER_PLATFORM_WIN32_WINDOWS) {
+		return(OS_WIN95);
+	} else if (os_info.dwPlatformId == VER_PLATFORM_WIN32_NT) {
+		if (os_info.dwMajorVersion <= 4) {
+			return(OS_WINNT);
+		} else {
+			return(OS_WIN2000);
+		}
+	} else {
+		ut_error;
+		return(0);
+	}
+#else
+	ut_error;
+
+	return(0);
+#endif
+}
+
+/***************************************************************************
+Retrieves the last error number if an error occurs in a file io function.
+The number should be retrieved before any other OS calls (because they may
+overwrite the error number). If the number is not known to this program,
+the OS error number + 100 is returned. */
+UNIV_INTERN
+ulint
+os_file_get_last_error(
+/*===================*/
+					/* out: error number, or OS error
+					number + 100 */
+	ibool	report_all_errors)	/* in: TRUE if we want an error message
+					printed of all errors */
+{
+	ulint	err;
+
+#ifdef __WIN__
+
+	err = (ulint) GetLastError();
+
+	if (report_all_errors
+	    || (err != ERROR_DISK_FULL && err != ERROR_FILE_EXISTS)) {
+
+		ut_print_timestamp(stderr);
+		fprintf(stderr,
+			"  InnoDB: Operating system error number %lu"
+			" in a file operation.\n", (ulong) err);
+
+		if (err == ERROR_PATH_NOT_FOUND) {
+			fprintf(stderr,
+				"InnoDB: The error means the system"
+				" cannot find the path specified.\n");
+
+			if (srv_is_being_started) {
+				fprintf(stderr,
+					"InnoDB: If you are installing InnoDB,"
+					" remember that you must create\n"
+					"InnoDB: directories yourself, InnoDB"
+					" does not create them.\n");
+			}
+		} else if (err == ERROR_ACCESS_DENIED) {
+			fprintf(stderr,
+				"InnoDB: The error means mysqld does not have"
+				" the access rights to\n"
+				"InnoDB: the directory. It may also be"
+				" you have created a subdirectory\n"
+				"InnoDB: of the same name as a data file.\n");
+		} else if (err == ERROR_SHARING_VIOLATION
+			   || err == ERROR_LOCK_VIOLATION) {
+			fprintf(stderr,
+				"InnoDB: The error means that another program"
+				" is using InnoDB's files.\n"
+				"InnoDB: This might be a backup or antivirus"
+				" software or another instance\n"
+				"InnoDB: of MySQL."
+				" Please close it to get rid of this error.\n");
+		} else {
+			fprintf(stderr,
+				"InnoDB: Some operating system error numbers"
+				" are described at\n"
+				"InnoDB: "
+				"http://dev.mysql.com/doc/refman/5.1/en/"
+				"operating-system-error-codes.html\n");
+		}
+	}
+
+	fflush(stderr);
+
+	if (err == ERROR_FILE_NOT_FOUND) {
+		return(OS_FILE_NOT_FOUND);
+	} else if (err == ERROR_DISK_FULL) {
+		return(OS_FILE_DISK_FULL);
+	} else if (err == ERROR_FILE_EXISTS) {
+		return(OS_FILE_ALREADY_EXISTS);
+	} else if (err == ERROR_SHARING_VIOLATION
+		   || err == ERROR_LOCK_VIOLATION) {
+		return(OS_FILE_SHARING_VIOLATION);
+	} else {
+		return(100 + err);
+	}
+#else
+	err = (ulint) errno;
+
+	if (report_all_errors
+	    || (err != ENOSPC && err != EEXIST)) {
+
+		ut_print_timestamp(stderr);
+		fprintf(stderr,
+			"  InnoDB: Operating system error number %lu"
+			" in a file operation.\n", (ulong) err);
+
+		if (err == ENOENT) {
+			fprintf(stderr,
+				"InnoDB: The error means the system"
+				" cannot find the path specified.\n");
+
+			if (srv_is_being_started) {
+				fprintf(stderr,
+					"InnoDB: If you are installing InnoDB,"
+					" remember that you must create\n"
+					"InnoDB: directories yourself, InnoDB"
+					" does not create them.\n");
+			}
+		} else if (err == EACCES) {
+			fprintf(stderr,
+				"InnoDB: The error means mysqld does not have"
+				" the access rights to\n"
+				"InnoDB: the directory.\n");
+		} else {
+			if (strerror((int)err) != NULL) {
+				fprintf(stderr,
+					"InnoDB: Error number %lu"
+					" means '%s'.\n",
+					err, strerror((int)err));
+			}
+
+			fprintf(stderr,
+				"InnoDB: Some operating system"
+				" error numbers are described at\n"
+				"InnoDB: "
+				"http://dev.mysql.com/doc/refman/5.1/en/"
+				"operating-system-error-codes.html\n");
+		}
+	}
+
+	fflush(stderr);
+
+	if (err == ENOSPC) {
+		return(OS_FILE_DISK_FULL);
+	} else if (err == ENOENT) {
+		return(OS_FILE_NOT_FOUND);
+	} else if (err == EEXIST) {
+		return(OS_FILE_ALREADY_EXISTS);
+	} else if (err == EXDEV || err == ENOTDIR || err == EISDIR) {
+		return(OS_FILE_PATH_ERROR);
+	} else {
+		return(100 + err);
+	}
+#endif
+}
+
+/********************************************************************
+Does error handling when a file operation fails.
+Conditionally exits (calling exit(3)) based on should_exit value and the
+error type */
+static
+ibool
+os_file_handle_error_cond_exit(
+/*===========================*/
+					/* out: TRUE if we should retry the
+					operation */
+	const char*	name,		/* in: name of a file or NULL */
+	const char*	operation,	/* in: operation */
+	ibool		should_exit)	/* in: call exit(3) if unknown error
+					and this parameter is TRUE */
+{
+	ulint	err;
+
+	err = os_file_get_last_error(FALSE);
+
+	if (err == OS_FILE_DISK_FULL) {
+		/* We only print a warning about disk full once */
+
+		if (os_has_said_disk_full) {
+
+			return(FALSE);
+		}
+
+		if (name) {
+			ut_print_timestamp(stderr);
+			fprintf(stderr,
+				"  InnoDB: Encountered a problem with"
+				" file %s\n", name);
+		}
+
+		ut_print_timestamp(stderr);
+		fprintf(stderr,
+			"  InnoDB: Disk is full. Try to clean the disk"
+			" to free space.\n");
+
+		os_has_said_disk_full = TRUE;
+
+		fflush(stderr);
+
+		return(FALSE);
+	} else if (err == OS_FILE_AIO_RESOURCES_RESERVED) {
+
+		return(TRUE);
+	} else if (err == OS_FILE_ALREADY_EXISTS
+		   || err == OS_FILE_PATH_ERROR) {
+
+		return(FALSE);
+	} else if (err == OS_FILE_SHARING_VIOLATION) {
+
+		os_thread_sleep(10000000);  /* 10 sec */
+		return(TRUE);
+	} else {
+		if (name) {
+			fprintf(stderr, "InnoDB: File name %s\n", name);
+		}
+
+		fprintf(stderr, "InnoDB: File operation call: '%s'.\n",
+			operation);
+
+		if (should_exit) {
+			fprintf(stderr, "InnoDB: Cannot continue operation.\n");
+
+			fflush(stderr);
+
+			exit(1);
+		}
+	}
+
+	return(FALSE);
+}
+
+/********************************************************************
+Does error handling when a file operation fails. */
+static
+ibool
+os_file_handle_error(
+/*=================*/
+				/* out: TRUE if we should retry the
+				operation */
+	const char*	name,	/* in: name of a file or NULL */
+	const char*	operation)/* in: operation */
+{
+	/* exit in case of unknown error */
+	return(os_file_handle_error_cond_exit(name, operation, TRUE));
+}
+
+/********************************************************************
+Does error handling when a file operation fails. */
+static
+ibool
+os_file_handle_error_no_exit(
+/*=========================*/
+				/* out: TRUE if we should retry the
+				operation */
+	const char*	name,	/* in: name of a file or NULL */
+	const char*	operation)/* in: operation */
+{
+	/* don't exit in case of unknown error */
+	return(os_file_handle_error_cond_exit(name, operation, FALSE));
+}
+
+#undef USE_FILE_LOCK
+#define USE_FILE_LOCK
+#if defined(UNIV_HOTBACKUP) || defined(__WIN__) || defined(__NETWARE__)
+/* InnoDB Hot Backup does not lock the data files.
+ * On Windows, mandatory locking is used.
+ */
+# undef USE_FILE_LOCK
+#endif
+#ifdef USE_FILE_LOCK
+/********************************************************************
+Obtain an exclusive lock on a file. */
+static
+int
+os_file_lock(
+/*=========*/
+				/* out: 0 on success */
+	int		fd,	/* in: file descriptor */
+	const char*	name)	/* in: file name */
+{
+	struct flock lk;
+	lk.l_type = F_WRLCK;
+	lk.l_whence = SEEK_SET;
+	lk.l_start = lk.l_len = 0;
+	if (fcntl(fd, F_SETLK, &lk) == -1) {
+		fprintf(stderr,
+			"InnoDB: Unable to lock %s, error: %d\n", name, errno);
+
+		if (errno == EAGAIN || errno == EACCES) {
+			fprintf(stderr,
+				"InnoDB: Check that you do not already have"
+				" another mysqld process\n"
+				"InnoDB: using the same InnoDB data"
+				" or log files.\n");
+		}
+
+		return(-1);
+	}
+
+	return(0);
+}
+#endif /* USE_FILE_LOCK */
+
+/********************************************************************
+Creates the seek mutexes used in positioned reads and writes. */
+UNIV_INTERN
+void
+os_io_init_simple(void)
+/*===================*/
+{
+	ulint	i;
+
+	os_file_count_mutex = os_mutex_create(NULL);
+
+	for (i = 0; i < OS_FILE_N_SEEK_MUTEXES; i++) {
+		os_file_seek_mutexes[i] = os_mutex_create(NULL);
+	}
+}
+
+/***************************************************************************
+Creates a temporary file.  This function is like tmpfile(3), but
+the temporary file is created in the MySQL temporary directory.
+On Netware, this function is like tmpfile(3), because the C run-time
+library of Netware does not expose the delete-on-close flag. */
+UNIV_INTERN
+FILE*
+os_file_create_tmpfile(void)
+/*========================*/
+			/* out: temporary file handle, or NULL on error */
+{
+#ifdef UNIV_HOTBACKUP
+	ut_error;
+
+	return(NULL);
+#else
+# ifdef __NETWARE__
+	FILE*	file	= tmpfile();
+# else /* __NETWARE__ */
+	FILE*	file	= NULL;
+	int	fd	= innobase_mysql_tmpfile();
+
+	if (fd >= 0) {
+		file = fdopen(fd, "w+b");
+	}
+# endif /* __NETWARE__ */
+
+	if (!file) {
+		ut_print_timestamp(stderr);
+		fprintf(stderr,
+			"  InnoDB: Error: unable to create temporary file;"
+			" errno: %d\n", errno);
+# ifndef __NETWARE__
+		if (fd >= 0) {
+			close(fd);
+		}
+# endif /* !__NETWARE__ */
+	}
+
+	return(file);
+#endif /* UNIV_HOTBACKUP */
+}
+
+/***************************************************************************
+The os_file_opendir() function opens a directory stream corresponding to the
+directory named by the dirname argument. The directory stream is positioned
+at the first entry. In both Unix and Windows we automatically skip the '.'
+and '..' items at the start of the directory listing. */
+UNIV_INTERN
+os_file_dir_t
+os_file_opendir(
+/*============*/
+					/* out: directory stream, NULL if
+					error */
+	const char*	dirname,	/* in: directory name; it must not
+					contain a trailing '\' or '/' */
+	ibool		error_is_fatal)	/* in: TRUE if we should treat an
+					error as a fatal error; if we try to
+					open symlinks then we do not wish a
+					fatal error if it happens not to be
+					a directory */
+{
+	os_file_dir_t		dir;
+#ifdef __WIN__
+	LPWIN32_FIND_DATA	lpFindFileData;
+	char			path[OS_FILE_MAX_PATH + 3];
+
+	ut_a(strlen(dirname) < OS_FILE_MAX_PATH);
+
+	strcpy(path, dirname);
+	strcpy(path + strlen(path), "\\*");
+
+	/* Note that in Windows opening the 'directory stream' also retrieves
+	the first entry in the directory. Since it is '.', that is no problem,
+	as we will skip over the '.' and '..' entries anyway. */
+
+	lpFindFileData = ut_malloc(sizeof(WIN32_FIND_DATA));
+
+	dir = FindFirstFile((LPCTSTR) path, lpFindFileData);
+
+	ut_free(lpFindFileData);
+
+	if (dir == INVALID_HANDLE_VALUE) {
+
+		if (error_is_fatal) {
+			os_file_handle_error(dirname, "opendir");
+		}
+
+		return(NULL);
+	}
+
+	return(dir);
+#else
+	dir = opendir(dirname);
+
+	if (dir == NULL && error_is_fatal) {
+		os_file_handle_error(dirname, "opendir");
+	}
+
+	return(dir);
+#endif
+}
+
+/***************************************************************************
+Closes a directory stream. */
+UNIV_INTERN
+int
+os_file_closedir(
+/*=============*/
+				/* out: 0 if success, -1 if failure */
+	os_file_dir_t	dir)	/* in: directory stream */
+{
+#ifdef __WIN__
+	BOOL		ret;
+
+	ret = FindClose(dir);
+
+	if (!ret) {
+		os_file_handle_error_no_exit(NULL, "closedir");
+
+		return(-1);
+	}
+
+	return(0);
+#else
+	int	ret;
+
+	ret = closedir(dir);
+
+	if (ret) {
+		os_file_handle_error_no_exit(NULL, "closedir");
+	}
+
+	return(ret);
+#endif
+}
+
+/***************************************************************************
+This function returns information of the next file in the directory. We jump
+over the '.' and '..' entries in the directory. */
+UNIV_INTERN
+int
+os_file_readdir_next_file(
+/*======================*/
+				/* out: 0 if ok, -1 if error, 1 if at the end
+				of the directory */
+	const char*	dirname,/* in: directory name or path */
+	os_file_dir_t	dir,	/* in: directory stream */
+	os_file_stat_t*	info)	/* in/out: buffer where the info is returned */
+{
+#ifdef __WIN__
+	LPWIN32_FIND_DATA	lpFindFileData;
+	BOOL			ret;
+
+	lpFindFileData = ut_malloc(sizeof(WIN32_FIND_DATA));
+next_file:
+	ret = FindNextFile(dir, lpFindFileData);
+
+	if (ret) {
+		ut_a(strlen((char *) lpFindFileData->cFileName)
+		     < OS_FILE_MAX_PATH);
+
+		if (strcmp((char *) lpFindFileData->cFileName, ".") == 0
+		    || strcmp((char *) lpFindFileData->cFileName, "..") == 0) {
+
+			goto next_file;
+		}
+
+		strcpy(info->name, (char *) lpFindFileData->cFileName);
+
+		info->size = (ib_int64_t)(lpFindFileData->nFileSizeLow)
+			+ (((ib_int64_t)(lpFindFileData->nFileSizeHigh))
+			   << 32);
+
+		if (lpFindFileData->dwFileAttributes
+		    & FILE_ATTRIBUTE_REPARSE_POINT) {
+			/* TODO: test Windows symlinks */
+			/* TODO: MySQL has apparently its own symlink
+			implementation in Windows, dbname.sym can
+			redirect a database directory:
+			http://dev.mysql.com/doc/refman/5.1/en/
+			windows-symbolic-links.html */
+			info->type = OS_FILE_TYPE_LINK;
+		} else if (lpFindFileData->dwFileAttributes
+			   & FILE_ATTRIBUTE_DIRECTORY) {
+			info->type = OS_FILE_TYPE_DIR;
+		} else {
+			/* It is probably safest to assume that all other
+			file types are normal. Better to check them rather
+			than blindly skip them. */
+
+			info->type = OS_FILE_TYPE_FILE;
+		}
+	}
+
+	ut_free(lpFindFileData);
+
+	if (ret) {
+		return(0);
+	} else if (GetLastError() == ERROR_NO_MORE_FILES) {
+
+		return(1);
+	} else {
+		os_file_handle_error_no_exit(dirname,
+					     "readdir_next_file");
+		return(-1);
+	}
+#else
+	struct dirent*	ent;
+	char*		full_path;
+	int		ret;
+	struct stat	statinfo;
+#ifdef HAVE_READDIR_R
+	char		dirent_buf[sizeof(struct dirent)
+				   + _POSIX_PATH_MAX + 100];
+	/* In /mysys/my_lib.c, _POSIX_PATH_MAX + 1 is used as
+	the max file name len; but in most standards, the
+	length is NAME_MAX; we add 100 to be even safer */
+#endif
+
+next_file:
+
+#ifdef HAVE_READDIR_R
+	ret = readdir_r(dir, (struct dirent*)dirent_buf, &ent);
+
+	if (ret != 0) {
+		fprintf(stderr,
+			"InnoDB: cannot read directory %s, error %lu\n",
+			dirname, (ulong)ret);
+
+		return(-1);
+	}
+
+	if (ent == NULL) {
+		/* End of directory */
+
+		return(1);
+	}
+
+	ut_a(strlen(ent->d_name) < _POSIX_PATH_MAX + 100 - 1);
+#else
+	ent = readdir(dir);
+
+	if (ent == NULL) {
+
+		return(1);
+	}
+#endif
+	ut_a(strlen(ent->d_name) < OS_FILE_MAX_PATH);
+
+	if (strcmp(ent->d_name, ".") == 0 || strcmp(ent->d_name, "..") == 0) {
+
+		goto next_file;
+	}
+
+	strcpy(info->name, ent->d_name);
+
+	full_path = ut_malloc(strlen(dirname) + strlen(ent->d_name) + 10);
+
+	sprintf(full_path, "%s/%s", dirname, ent->d_name);
+
+	ret = stat(full_path, &statinfo);
+
+	if (ret) {
+		os_file_handle_error_no_exit(full_path, "stat");
+
+		ut_free(full_path);
+
+		return(-1);
+	}
+
+	info->size = (ib_int64_t)statinfo.st_size;
+
+	if (S_ISDIR(statinfo.st_mode)) {
+		info->type = OS_FILE_TYPE_DIR;
+	} else if (S_ISLNK(statinfo.st_mode)) {
+		info->type = OS_FILE_TYPE_LINK;
+	} else if (S_ISREG(statinfo.st_mode)) {
+		info->type = OS_FILE_TYPE_FILE;
+	} else {
+		info->type = OS_FILE_TYPE_UNKNOWN;
+	}
+
+	ut_free(full_path);
+
+	return(0);
+#endif
+}
+
+/*********************************************************************
+This function attempts to create a directory named pathname. The new directory
+gets default permissions. On Unix the permissions are (0770 & ~umask). If the
+directory exists already, nothing is done and the call succeeds, unless the
+fail_if_exists arguments is true. */
+UNIV_INTERN
+ibool
+os_file_create_directory(
+/*=====================*/
+					/* out: TRUE if call succeeds,
+					FALSE on error */
+	const char*	pathname,	/* in: directory name as
+					null-terminated string */
+	ibool		fail_if_exists)	/* in: if TRUE, pre-existing directory
+					is treated as an error. */
+{
+#ifdef __WIN__
+	BOOL	rcode;
+
+	rcode = CreateDirectory((LPCTSTR) pathname, NULL);
+	if (!(rcode != 0
+	      || (GetLastError() == ERROR_ALREADY_EXISTS
+		  && !fail_if_exists))) {
+		/* failure */
+		os_file_handle_error(pathname, "CreateDirectory");
+
+		return(FALSE);
+	}
+
+	return (TRUE);
+#else
+	int	rcode;
+
+	rcode = mkdir(pathname, 0770);
+
+	if (!(rcode == 0 || (errno == EEXIST && !fail_if_exists))) {
+		/* failure */
+		os_file_handle_error(pathname, "mkdir");
+
+		return(FALSE);
+	}
+
+	return (TRUE);
+#endif
+}
+
+/********************************************************************
+A simple function to open or create a file. */
+UNIV_INTERN
+os_file_t
+os_file_create_simple(
+/*==================*/
+				/* out, own: handle to the file, not defined
+				if error, error number can be retrieved with
+				os_file_get_last_error */
+	const char*	name,	/* in: name of the file or path as a
+				null-terminated string */
+	ulint		create_mode,/* in: OS_FILE_OPEN if an existing file is
+				opened (if does not exist, error), or
+				OS_FILE_CREATE if a new file is created
+				(if exists, error), or
+				OS_FILE_CREATE_PATH if new file
+				(if exists, error) and subdirectories along
+				its path are created (if needed)*/
+	ulint		access_type,/* in: OS_FILE_READ_ONLY or
+				OS_FILE_READ_WRITE */
+	ibool*		success)/* out: TRUE if succeed, FALSE if error */
+{
+#ifdef __WIN__
+	os_file_t	file;
+	DWORD		create_flag;
+	DWORD		access;
+	DWORD		attributes	= 0;
+	ibool		retry;
+
+try_again:
+	ut_a(name);
+
+	if (create_mode == OS_FILE_OPEN) {
+		create_flag = OPEN_EXISTING;
+	} else if (create_mode == OS_FILE_CREATE) {
+		create_flag = CREATE_NEW;
+	} else if (create_mode == OS_FILE_CREATE_PATH) {
+		/* create subdirs along the path if needed  */
+		*success = os_file_create_subdirs_if_needed(name);
+		if (!*success) {
+			ut_error;
+		}
+		create_flag = CREATE_NEW;
+		create_mode = OS_FILE_CREATE;
+	} else {
+		create_flag = 0;
+		ut_error;
+	}
+
+	if (access_type == OS_FILE_READ_ONLY) {
+		access = GENERIC_READ;
+	} else if (access_type == OS_FILE_READ_WRITE) {
+		access = GENERIC_READ | GENERIC_WRITE;
+	} else {
+		access = 0;
+		ut_error;
+	}
+
+	file = CreateFile((LPCTSTR) name,
+			  access,
+			  FILE_SHARE_READ | FILE_SHARE_WRITE,
+			  /* file can be read and written also
+			  by other processes */
+			  NULL,	/* default security attributes */
+			  create_flag,
+			  attributes,
+			  NULL);	/* no template file */
+
+	if (file == INVALID_HANDLE_VALUE) {
+		*success = FALSE;
+
+		retry = os_file_handle_error(name,
+					     create_mode == OS_FILE_OPEN ?
+					     "open" : "create");
+		if (retry) {
+			goto try_again;
+		}
+	} else {
+		*success = TRUE;
+	}
+
+	return(file);
+#else /* __WIN__ */
+	os_file_t	file;
+	int		create_flag;
+	ibool		retry;
+
+try_again:
+	ut_a(name);
+
+	if (create_mode == OS_FILE_OPEN) {
+		if (access_type == OS_FILE_READ_ONLY) {
+			create_flag = O_RDONLY;
+		} else {
+			create_flag = O_RDWR;
+		}
+	} else if (create_mode == OS_FILE_CREATE) {
+		create_flag = O_RDWR | O_CREAT | O_EXCL;
+	} else if (create_mode == OS_FILE_CREATE_PATH) {
+		/* create subdirs along the path if needed  */
+		*success = os_file_create_subdirs_if_needed(name);
+		if (!*success) {
+			return (-1);
+		}
+		create_flag = O_RDWR | O_CREAT | O_EXCL;
+		create_mode = OS_FILE_CREATE;
+	} else {
+		create_flag = 0;
+		ut_error;
+	}
+
+	if (create_mode == OS_FILE_CREATE) {
+		file = open(name, create_flag, S_IRUSR | S_IWUSR
+			    | S_IRGRP | S_IWGRP);
+	} else {
+		file = open(name, create_flag);
+	}
+
+	if (file == -1) {
+		*success = FALSE;
+
+		retry = os_file_handle_error(name,
+					     create_mode == OS_FILE_OPEN ?
+					     "open" : "create");
+		if (retry) {
+			goto try_again;
+		}
+#ifdef USE_FILE_LOCK
+	} else if (access_type == OS_FILE_READ_WRITE
+		   && os_file_lock(file, name)) {
+		*success = FALSE;
+		close(file);
+		file = -1;
+#endif
+	} else {
+		*success = TRUE;
+	}
+
+	return(file);
+#endif /* __WIN__ */
+}
+
+/********************************************************************
+A simple function to open or create a file. */
+UNIV_INTERN
+os_file_t
+os_file_create_simple_no_error_handling(
+/*====================================*/
+				/* out, own: handle to the file, not defined
+				if error, error number can be retrieved with
+				os_file_get_last_error */
+	const char*	name,	/* in: name of the file or path as a
+				null-terminated string */
+	ulint		create_mode,/* in: OS_FILE_OPEN if an existing file
+				is opened (if does not exist, error), or
+				OS_FILE_CREATE if a new file is created
+				(if exists, error) */
+	ulint		access_type,/* in: OS_FILE_READ_ONLY,
+				OS_FILE_READ_WRITE, or
+				OS_FILE_READ_ALLOW_DELETE; the last option is
+				used by a backup program reading the file */
+	ibool*		success)/* out: TRUE if succeed, FALSE if error */
+{
+#ifdef __WIN__
+	os_file_t	file;
+	DWORD		create_flag;
+	DWORD		access;
+	DWORD		attributes	= 0;
+	DWORD		share_mode	= FILE_SHARE_READ | FILE_SHARE_WRITE;
+
+	ut_a(name);
+
+	if (create_mode == OS_FILE_OPEN) {
+		create_flag = OPEN_EXISTING;
+	} else if (create_mode == OS_FILE_CREATE) {
+		create_flag = CREATE_NEW;
+	} else {
+		create_flag = 0;
+		ut_error;
+	}
+
+	if (access_type == OS_FILE_READ_ONLY) {
+		access = GENERIC_READ;
+	} else if (access_type == OS_FILE_READ_WRITE) {
+		access = GENERIC_READ | GENERIC_WRITE;
+	} else if (access_type == OS_FILE_READ_ALLOW_DELETE) {
+		access = GENERIC_READ;
+		share_mode = FILE_SHARE_DELETE | FILE_SHARE_READ
+			| FILE_SHARE_WRITE;	/* A backup program has to give
+						mysqld the maximum freedom to
+						do what it likes with the
+						file */
+	} else {
+		access = 0;
+		ut_error;
+	}
+
+	file = CreateFile((LPCTSTR) name,
+			  access,
+			  share_mode,
+			  NULL,	/* default security attributes */
+			  create_flag,
+			  attributes,
+			  NULL);	/* no template file */
+
+	if (file == INVALID_HANDLE_VALUE) {
+		*success = FALSE;
+	} else {
+		*success = TRUE;
+	}
+
+	return(file);
+#else /* __WIN__ */
+	os_file_t	file;
+	int		create_flag;
+
+	ut_a(name);
+
+	if (create_mode == OS_FILE_OPEN) {
+		if (access_type == OS_FILE_READ_ONLY) {
+			create_flag = O_RDONLY;
+		} else {
+			create_flag = O_RDWR;
+		}
+	} else if (create_mode == OS_FILE_CREATE) {
+		create_flag = O_RDWR | O_CREAT | O_EXCL;
+	} else {
+		create_flag = 0;
+		ut_error;
+	}
+
+	if (create_mode == OS_FILE_CREATE) {
+		file = open(name, create_flag, S_IRUSR | S_IWUSR
+			    | S_IRGRP | S_IWGRP);
+	} else {
+		file = open(name, create_flag);
+	}
+
+	if (file == -1) {
+		*success = FALSE;
+#ifdef USE_FILE_LOCK
+	} else if (access_type == OS_FILE_READ_WRITE
+		   && os_file_lock(file, name)) {
+		*success = FALSE;
+		close(file);
+		file = -1;
+#endif
+	} else {
+		*success = TRUE;
+	}
+
+	return(file);
+#endif /* __WIN__ */
+}
+
+/********************************************************************
+Tries to disable OS caching on an opened file descriptor. */
+UNIV_INTERN
+void
+os_file_set_nocache(
+/*================*/
+	int		fd,		/* in: file descriptor to alter */
+	const char*	file_name,	/* in: file name, used in the
+					diagnostic message */
+	const char*	operation_name)	/* in: "open" or "create"; used in the
+					diagnostic message */
+{
+	/* some versions of Solaris may not have DIRECTIO_ON */
+#if defined(UNIV_SOLARIS) && defined(DIRECTIO_ON)
+	if (directio(fd, DIRECTIO_ON) == -1) {
+		int	errno_save;
+		errno_save = (int)errno;
+		ut_print_timestamp(stderr);
+		fprintf(stderr,
+			"  InnoDB: Failed to set DIRECTIO_ON "
+			"on file %s: %s: %s, continuing anyway\n",
+			file_name, operation_name, strerror(errno_save));
+	}
+#elif defined(O_DIRECT)
+	if (fcntl(fd, F_SETFL, O_DIRECT) == -1) {
+		int	errno_save;
+		errno_save = (int)errno;
+		ut_print_timestamp(stderr);
+		fprintf(stderr,
+			"  InnoDB: Failed to set O_DIRECT "
+			"on file %s: %s: %s, continuing anyway\n",
+			file_name, operation_name, strerror(errno_save));
+		if (errno_save == EINVAL) {
+			ut_print_timestamp(stderr);
+			fprintf(stderr,
+				"  InnoDB: O_DIRECT is known to result in "
+				"'Invalid argument' on Linux on tmpfs, "
+				"see MySQL Bug#26662\n");
+		}
+	}
+#endif
+}
+
+/********************************************************************
+Opens an existing file or creates a new. */
+UNIV_INTERN
+os_file_t
+os_file_create(
+/*===========*/
+				/* out, own: handle to the file, not defined
+				if error, error number can be retrieved with
+				os_file_get_last_error */
+	const char*	name,	/* in: name of the file or path as a
+				null-terminated string */
+	ulint		create_mode,/* in: OS_FILE_OPEN if an existing file
+				is opened (if does not exist, error), or
+				OS_FILE_CREATE if a new file is created
+				(if exists, error),
+				OS_FILE_OVERWRITE if a new file is created
+				or an old overwritten;
+				OS_FILE_OPEN_RAW, if a raw device or disk
+				partition should be opened */
+	ulint		purpose,/* in: OS_FILE_AIO, if asynchronous,
+				non-buffered i/o is desired,
+				OS_FILE_NORMAL, if any normal file;
+				NOTE that it also depends on type, os_aio_..
+				and srv_.. variables whether we really use
+				async i/o or unbuffered i/o: look in the
+				function source code for the exact rules */
+	ulint		type,	/* in: OS_DATA_FILE or OS_LOG_FILE */
+	ibool*		success)/* out: TRUE if succeed, FALSE if error */
+{
+#ifdef __WIN__
+	os_file_t	file;
+	DWORD		share_mode	= FILE_SHARE_READ;
+	DWORD		create_flag;
+	DWORD		attributes;
+	ibool		retry;
+try_again:
+	ut_a(name);
+
+	if (create_mode == OS_FILE_OPEN_RAW) {
+		create_flag = OPEN_EXISTING;
+		share_mode = FILE_SHARE_WRITE;
+	} else if (create_mode == OS_FILE_OPEN
+		   || create_mode == OS_FILE_OPEN_RETRY) {
+		create_flag = OPEN_EXISTING;
+	} else if (create_mode == OS_FILE_CREATE) {
+		create_flag = CREATE_NEW;
+	} else if (create_mode == OS_FILE_OVERWRITE) {
+		create_flag = CREATE_ALWAYS;
+	} else {
+		create_flag = 0;
+		ut_error;
+	}
+
+	if (purpose == OS_FILE_AIO) {
+		/* If specified, use asynchronous (overlapped) io and no
+		buffering of writes in the OS */
+		attributes = 0;
+#ifdef WIN_ASYNC_IO
+		if (os_aio_use_native_aio) {
+			attributes = attributes | FILE_FLAG_OVERLAPPED;
+		}
+#endif
+#ifdef UNIV_NON_BUFFERED_IO
+		if (type == OS_LOG_FILE && srv_flush_log_at_trx_commit == 2) {
+			/* Do not use unbuffered i/o to log files because
+			value 2 denotes that we do not flush the log at every
+			commit, but only once per second */
+		} else if (srv_win_file_flush_method
+			   == SRV_WIN_IO_UNBUFFERED) {
+			attributes = attributes | FILE_FLAG_NO_BUFFERING;
+		}
+#endif
+	} else if (purpose == OS_FILE_NORMAL) {
+		attributes = 0;
+#ifdef UNIV_NON_BUFFERED_IO
+		if (type == OS_LOG_FILE && srv_flush_log_at_trx_commit == 2) {
+			/* Do not use unbuffered i/o to log files because
+			value 2 denotes that we do not flush the log at every
+			commit, but only once per second */
+		} else if (srv_win_file_flush_method
+			   == SRV_WIN_IO_UNBUFFERED) {
+			attributes = attributes | FILE_FLAG_NO_BUFFERING;
+		}
+#endif
+	} else {
+		attributes = 0;
+		ut_error;
+	}
+
+	file = CreateFile((LPCTSTR) name,
+			  GENERIC_READ | GENERIC_WRITE, /* read and write
+							access */
+			  share_mode,	/* File can be read also by other
+					processes; we must give the read
+					permission because of ibbackup. We do
+					not give the write permission to
+					others because if one would succeed to
+					start 2 instances of mysqld on the
+					SAME files, that could cause severe
+					database corruption! When opening
+					raw disk partitions, Microsoft manuals
+					say that we must give also the write
+					permission. */
+			  NULL,	/* default security attributes */
+			  create_flag,
+			  attributes,
+			  NULL);	/* no template file */
+
+	if (file == INVALID_HANDLE_VALUE) {
+		*success = FALSE;
+
+		/* When srv_file_per_table is on, file creation failure may not
+		be critical to the whole instance. Do not crash the server in
+		case of unknown errors. */
+		if (srv_file_per_table) {
+			retry = os_file_handle_error_no_exit(name,
+						create_mode == OS_FILE_CREATE ?
+						"create" : "open");
+		} else {
+			retry = os_file_handle_error(name,
+						create_mode == OS_FILE_CREATE ?
+						"create" : "open");
+		}
+
+		if (retry) {
+			goto try_again;
+		}
+	} else {
+		*success = TRUE;
+	}
+
+	return(file);
+#else /* __WIN__ */
+	os_file_t	file;
+	int		create_flag;
+	ibool		retry;
+	const char*	mode_str	= NULL;
+	const char*	type_str	= NULL;
+	const char*	purpose_str	= NULL;
+
+try_again:
+	ut_a(name);
+
+	if (create_mode == OS_FILE_OPEN || create_mode == OS_FILE_OPEN_RAW
+	    || create_mode == OS_FILE_OPEN_RETRY) {
+		mode_str = "OPEN";
+		create_flag = O_RDWR;
+	} else if (create_mode == OS_FILE_CREATE) {
+		mode_str = "CREATE";
+		create_flag = O_RDWR | O_CREAT | O_EXCL;
+	} else if (create_mode == OS_FILE_OVERWRITE) {
+		mode_str = "OVERWRITE";
+		create_flag = O_RDWR | O_CREAT | O_TRUNC;
+	} else {
+		create_flag = 0;
+		ut_error;
+	}
+
+	if (type == OS_LOG_FILE) {
+		type_str = "LOG";
+	} else if (type == OS_DATA_FILE) {
+		type_str = "DATA";
+	} else {
+		ut_error;
+	}
+
+	if (purpose == OS_FILE_AIO) {
+		purpose_str = "AIO";
+	} else if (purpose == OS_FILE_NORMAL) {
+		purpose_str = "NORMAL";
+	} else {
+		ut_error;
+	}
+
+#if 0
+	fprintf(stderr, "Opening file %s, mode %s, type %s, purpose %s\n",
+		name, mode_str, type_str, purpose_str);
+#endif
+#ifdef O_SYNC
+	/* We let O_SYNC only affect log files; note that we map O_DSYNC to
+	O_SYNC because the datasync options seemed to corrupt files in 2001
+	in both Linux and Solaris */
+	if (type == OS_LOG_FILE
+	    && srv_unix_file_flush_method == SRV_UNIX_O_DSYNC) {
+
+# if 0
+		fprintf(stderr, "Using O_SYNC for file %s\n", name);
+# endif
+
+		create_flag = create_flag | O_SYNC;
+	}
+#endif /* O_SYNC */
+
+	file = open(name, create_flag, os_innodb_umask);
+
+	if (file == -1) {
+		*success = FALSE;
+
+		/* When srv_file_per_table is on, file creation failure may not
+		be critical to the whole instance. Do not crash the server in
+		case of unknown errors. */
+		if (srv_file_per_table) {
+			retry = os_file_handle_error_no_exit(name,
+						create_mode == OS_FILE_CREATE ?
+						"create" : "open");
+		} else {
+			retry = os_file_handle_error(name,
+						create_mode == OS_FILE_CREATE ?
+						"create" : "open");
+		}
+
+		if (retry) {
+			goto try_again;
+		} else {
+			return(file /* -1 */);
+		}
+	}
+	/* else */
+
+	*success = TRUE;
+
+	/* We disable OS caching (O_DIRECT) only on data files */
+	if (type != OS_LOG_FILE
+	    && srv_unix_file_flush_method == SRV_UNIX_O_DIRECT) {
+		
+		os_file_set_nocache(file, name, mode_str);
+	}
+
+#ifdef USE_FILE_LOCK
+	if (create_mode != OS_FILE_OPEN_RAW && os_file_lock(file, name)) {
+
+		if (create_mode == OS_FILE_OPEN_RETRY) {
+			int i;
+			ut_print_timestamp(stderr);
+			fputs("  InnoDB: Retrying to lock"
+			      " the first data file\n",
+			      stderr);
+			for (i = 0; i < 100; i++) {
+				os_thread_sleep(1000000);
+				if (!os_file_lock(file, name)) {
+					*success = TRUE;
+					return(file);
+				}
+			}
+			ut_print_timestamp(stderr);
+			fputs("  InnoDB: Unable to open the first data file\n",
+			      stderr);
+		}
+
+		*success = FALSE;
+		close(file);
+		file = -1;
+	}
+#endif /* USE_FILE_LOCK */
+
+	return(file);
+#endif /* __WIN__ */
+}
+
+/***************************************************************************
+Deletes a file if it exists. The file has to be closed before calling this. */
+UNIV_INTERN
+ibool
+os_file_delete_if_exists(
+/*=====================*/
+				/* out: TRUE if success */
+	const char*	name)	/* in: file path as a null-terminated string */
+{
+#ifdef __WIN__
+	BOOL	ret;
+	ulint	count	= 0;
+loop:
+	/* In Windows, deleting an .ibd file may fail if ibbackup is copying
+	it */
+
+	ret = DeleteFile((LPCTSTR)name);
+
+	if (ret) {
+		return(TRUE);
+	}
+
+	if (GetLastError() == ERROR_FILE_NOT_FOUND) {
+		/* the file does not exist, this not an error */
+
+		return(TRUE);
+	}
+
+	count++;
+
+	if (count > 100 && 0 == (count % 10)) {
+		fprintf(stderr,
+			"InnoDB: Warning: cannot delete file %s\n"
+			"InnoDB: Are you running ibbackup"
+			" to back up the file?\n", name);
+
+		os_file_get_last_error(TRUE); /* print error information */
+	}
+
+	os_thread_sleep(1000000);	/* sleep for a second */
+
+	if (count > 2000) {
+
+		return(FALSE);
+	}
+
+	goto loop;
+#else
+	int	ret;
+
+	ret = unlink(name);
+
+	if (ret != 0 && errno != ENOENT) {
+		os_file_handle_error_no_exit(name, "delete");
+
+		return(FALSE);
+	}
+
+	return(TRUE);
+#endif
+}
+
+/***************************************************************************
+Deletes a file. The file has to be closed before calling this. */
+UNIV_INTERN
+ibool
+os_file_delete(
+/*===========*/
+				/* out: TRUE if success */
+	const char*	name)	/* in: file path as a null-terminated string */
+{
+#ifdef __WIN__
+	BOOL	ret;
+	ulint	count	= 0;
+loop:
+	/* In Windows, deleting an .ibd file may fail if ibbackup is copying
+	it */
+
+	ret = DeleteFile((LPCTSTR)name);
+
+	if (ret) {
+		return(TRUE);
+	}
+
+	if (GetLastError() == ERROR_FILE_NOT_FOUND) {
+		/* If the file does not exist, we classify this as a 'mild'
+		error and return */
+
+		return(FALSE);
+	}
+
+	count++;
+
+	if (count > 100 && 0 == (count % 10)) {
+		fprintf(stderr,
+			"InnoDB: Warning: cannot delete file %s\n"
+			"InnoDB: Are you running ibbackup"
+			" to back up the file?\n", name);
+
+		os_file_get_last_error(TRUE); /* print error information */
+	}
+
+	os_thread_sleep(1000000);	/* sleep for a second */
+
+	if (count > 2000) {
+
+		return(FALSE);
+	}
+
+	goto loop;
+#else
+	int	ret;
+
+	ret = unlink(name);
+
+	if (ret != 0) {
+		os_file_handle_error_no_exit(name, "delete");
+
+		return(FALSE);
+	}
+
+	return(TRUE);
+#endif
+}
+
+/***************************************************************************
+Renames a file (can also move it to another directory). It is safest that the
+file is closed before calling this function. */
+UNIV_INTERN
+ibool
+os_file_rename(
+/*===========*/
+				/* out: TRUE if success */
+	const char*	oldpath,/* in: old file path as a null-terminated
+				string */
+	const char*	newpath)/* in: new file path */
+{
+#ifdef __WIN__
+	BOOL	ret;
+
+	ret = MoveFile((LPCTSTR)oldpath, (LPCTSTR)newpath);
+
+	if (ret) {
+		return(TRUE);
+	}
+
+	os_file_handle_error_no_exit(oldpath, "rename");
+
+	return(FALSE);
+#else
+	int	ret;
+
+	ret = rename(oldpath, newpath);
+
+	if (ret != 0) {
+		os_file_handle_error_no_exit(oldpath, "rename");
+
+		return(FALSE);
+	}
+
+	return(TRUE);
+#endif
+}
+
+/***************************************************************************
+Closes a file handle. In case of error, error number can be retrieved with
+os_file_get_last_error. */
+UNIV_INTERN
+ibool
+os_file_close(
+/*==========*/
+				/* out: TRUE if success */
+	os_file_t	file)	/* in, own: handle to a file */
+{
+#ifdef __WIN__
+	BOOL	ret;
+
+	ut_a(file);
+
+	ret = CloseHandle(file);
+
+	if (ret) {
+		return(TRUE);
+	}
+
+	os_file_handle_error(NULL, "close");
+
+	return(FALSE);
+#else
+	int	ret;
+
+	ret = close(file);
+
+	if (ret == -1) {
+		os_file_handle_error(NULL, "close");
+
+		return(FALSE);
+	}
+
+	return(TRUE);
+#endif
+}
+
+/***************************************************************************
+Closes a file handle. */
+UNIV_INTERN
+ibool
+os_file_close_no_error_handling(
+/*============================*/
+				/* out: TRUE if success */
+	os_file_t	file)	/* in, own: handle to a file */
+{
+#ifdef __WIN__
+	BOOL	ret;
+
+	ut_a(file);
+
+	ret = CloseHandle(file);
+
+	if (ret) {
+		return(TRUE);
+	}
+
+	return(FALSE);
+#else
+	int	ret;
+
+	ret = close(file);
+
+	if (ret == -1) {
+
+		return(FALSE);
+	}
+
+	return(TRUE);
+#endif
+}
+
+/***************************************************************************
+Gets a file size. */
+UNIV_INTERN
+ibool
+os_file_get_size(
+/*=============*/
+				/* out: TRUE if success */
+	os_file_t	file,	/* in: handle to a file */
+	ulint*		size,	/* out: least significant 32 bits of file
+				size */
+	ulint*		size_high)/* out: most significant 32 bits of size */
+{
+#ifdef __WIN__
+	DWORD	high;
+	DWORD	low;
+
+	low = GetFileSize(file, &high);
+
+	if ((low == 0xFFFFFFFF) && (GetLastError() != NO_ERROR)) {
+		return(FALSE);
+	}
+
+	*size = low;
+	*size_high = high;
+
+	return(TRUE);
+#else
+	off_t	offs;
+
+	offs = lseek(file, 0, SEEK_END);
+
+	if (offs == ((off_t)-1)) {
+
+		return(FALSE);
+	}
+
+	if (sizeof(off_t) > 4) {
+		*size = (ulint)(offs & 0xFFFFFFFFUL);
+		*size_high = (ulint)(offs >> 32);
+	} else {
+		*size = (ulint) offs;
+		*size_high = 0;
+	}
+
+	return(TRUE);
+#endif
+}
+
+/***************************************************************************
+Gets file size as a 64-bit integer ib_int64_t. */
+UNIV_INTERN
+ib_int64_t
+os_file_get_size_as_iblonglong(
+/*===========================*/
+				/* out: size in bytes, -1 if error */
+	os_file_t	file)	/* in: handle to a file */
+{
+	ulint	size;
+	ulint	size_high;
+	ibool	success;
+
+	success = os_file_get_size(file, &size, &size_high);
+
+	if (!success) {
+
+		return(-1);
+	}
+
+	return((((ib_int64_t)size_high) << 32) + (ib_int64_t)size);
+}
+
+/***************************************************************************
+Write the specified number of zeros to a newly created file. */
+UNIV_INTERN
+ibool
+os_file_set_size(
+/*=============*/
+				/* out: TRUE if success */
+	const char*	name,	/* in: name of the file or path as a
+				null-terminated string */
+	os_file_t	file,	/* in: handle to a file */
+	ulint		size,	/* in: least significant 32 bits of file
+				size */
+	ulint		size_high)/* in: most significant 32 bits of size */
+{
+	ib_int64_t	current_size;
+	ib_int64_t	desired_size;
+	ibool		ret;
+	byte*		buf;
+	byte*		buf2;
+	ulint		buf_size;
+
+	ut_a(size == (size & 0xFFFFFFFF));
+
+	current_size = 0;
+	desired_size = (ib_int64_t)size + (((ib_int64_t)size_high) << 32);
+
+	/* Write up to 1 megabyte at a time. */
+	buf_size = ut_min(64, (ulint) (desired_size / UNIV_PAGE_SIZE))
+		* UNIV_PAGE_SIZE;
+	buf2 = ut_malloc(buf_size + UNIV_PAGE_SIZE);
+
+	/* Align the buffer for possible raw i/o */
+	buf = ut_align(buf2, UNIV_PAGE_SIZE);
+
+	/* Write buffer full of zeros */
+	memset(buf, 0, buf_size);
+
+	if (desired_size >= (ib_int64_t)(100 * 1024 * 1024)) {
+
+		fprintf(stderr, "InnoDB: Progress in MB:");
+	}
+
+	while (current_size < desired_size) {
+		ulint	n_bytes;
+
+		if (desired_size - current_size < (ib_int64_t) buf_size) {
+			n_bytes = (ulint) (desired_size - current_size);
+		} else {
+			n_bytes = buf_size;
+		}
+
+		ret = os_file_write(name, file, buf,
+				    (ulint)(current_size & 0xFFFFFFFF),
+				    (ulint)(current_size >> 32),
+				    n_bytes);
+		if (!ret) {
+			ut_free(buf2);
+			goto error_handling;
+		}
+
+		/* Print about progress for each 100 MB written */
+		if ((ib_int64_t) (current_size + n_bytes) / (ib_int64_t)(100 * 1024 * 1024)
+		    != current_size / (ib_int64_t)(100 * 1024 * 1024)) {
+
+			fprintf(stderr, " %lu00",
+				(ulong) ((current_size + n_bytes)
+					 / (ib_int64_t)(100 * 1024 * 1024)));
+		}
+
+		current_size += n_bytes;
+	}
+
+	if (desired_size >= (ib_int64_t)(100 * 1024 * 1024)) {
+
+		fprintf(stderr, "\n");
+	}
+
+	ut_free(buf2);
+
+	ret = os_file_flush(file);
+
+	if (ret) {
+		return(TRUE);
+	}
+
+error_handling:
+	return(FALSE);
+}
+
+/***************************************************************************
+Truncates a file at its current position. */
+UNIV_INTERN
+ibool
+os_file_set_eof(
+/*============*/
+				/* out: TRUE if success */
+	FILE*		file)	/* in: file to be truncated */
+{
+#ifdef __WIN__
+	HANDLE h = (HANDLE) _get_osfhandle(fileno(file));
+	return(SetEndOfFile(h));
+#else /* __WIN__ */
+	return(!ftruncate(fileno(file), ftell(file)));
+#endif /* __WIN__ */
+}
+
+#ifndef __WIN__
+/***************************************************************************
+Wrapper to fsync(2) that retries the call on some errors.
+Returns the value 0 if successful; otherwise the value -1 is returned and
+the global variable errno is set to indicate the error. */
+
+static
+int
+os_file_fsync(
+/*==========*/
+				/* out: 0 if success, -1 otherwise */
+	os_file_t	file)	/* in: handle to a file */
+{
+	int	ret;
+	int	failures;
+	ibool	retry;
+
+	failures = 0;
+
+	do {
+		ret = fsync(file);
+
+		os_n_fsyncs++;
+
+		if (ret == -1 && errno == ENOLCK) {
+
+			if (failures % 100 == 0) {
+
+				ut_print_timestamp(stderr);
+				fprintf(stderr,
+					"  InnoDB: fsync(): "
+					"No locks available; retrying\n");
+			}
+
+			os_thread_sleep(200000 /* 0.2 sec */);
+
+			failures++;
+
+			retry = TRUE;
+		} else {
+
+			retry = FALSE;
+		}
+	} while (retry);
+
+	return(ret);
+}
+#endif /* !__WIN__ */
+
+/***************************************************************************
+Flushes the write buffers of a given file to the disk. */
+UNIV_INTERN
+ibool
+os_file_flush(
+/*==========*/
+				/* out: TRUE if success */
+	os_file_t	file)	/* in, own: handle to a file */
+{
+#ifdef __WIN__
+	BOOL	ret;
+
+	ut_a(file);
+
+	os_n_fsyncs++;
+
+	ret = FlushFileBuffers(file);
+
+	if (ret) {
+		return(TRUE);
+	}
+
+	/* Since Windows returns ERROR_INVALID_FUNCTION if the 'file' is
+	actually a raw device, we choose to ignore that error if we are using
+	raw disks */
+
+	if (srv_start_raw_disk_in_use && GetLastError()
+	    == ERROR_INVALID_FUNCTION) {
+		return(TRUE);
+	}
+
+	os_file_handle_error(NULL, "flush");
+
+	/* It is a fatal error if a file flush does not succeed, because then
+	the database can get corrupt on disk */
+	ut_error;
+
+	return(FALSE);
+#else
+	int	ret;
+
+#if defined(HAVE_DARWIN_THREADS)
+# ifndef F_FULLFSYNC
+	/* The following definition is from the Mac OS X 10.3 <sys/fcntl.h> */
+#  define F_FULLFSYNC 51 /* fsync + ask the drive to flush to the media */
+# elif F_FULLFSYNC != 51
+#  error "F_FULLFSYNC != 51: ABI incompatibility with Mac OS X 10.3"
+# endif
+	/* Apple has disabled fsync() for internal disk drives in OS X. That
+	caused corruption for a user when he tested a power outage. Let us in
+	OS X use a nonstandard flush method recommended by an Apple
+	engineer. */
+
+	if (!srv_have_fullfsync) {
+		/* If we are not on an operating system that supports this,
+		then fall back to a plain fsync. */
+
+		ret = os_file_fsync(file);
+	} else {
+		ret = fcntl(file, F_FULLFSYNC, NULL);
+
+		if (ret) {
+			/* If we are not on a file system that supports this,
+			then fall back to a plain fsync. */
+			ret = os_file_fsync(file);
+		}
+	}
+#else
+	ret = os_file_fsync(file);
+#endif
+
+	if (ret == 0) {
+		return(TRUE);
+	}
+
+	/* Since Linux returns EINVAL if the 'file' is actually a raw device,
+	we choose to ignore that error if we are using raw disks */
+
+	if (srv_start_raw_disk_in_use && errno == EINVAL) {
+
+		return(TRUE);
+	}
+
+	ut_print_timestamp(stderr);
+
+	fprintf(stderr,
+		"  InnoDB: Error: the OS said file flush did not succeed\n");
+
+	os_file_handle_error(NULL, "flush");
+
+	/* It is a fatal error if a file flush does not succeed, because then
+	the database can get corrupt on disk */
+	ut_error;
+
+	return(FALSE);
+#endif
+}
+
+#ifndef __WIN__
+/***********************************************************************
+Does a synchronous read operation in Posix. */
+static
+ssize_t
+os_file_pread(
+/*==========*/
+				/* out: number of bytes read, -1 if error */
+	os_file_t	file,	/* in: handle to a file */
+	void*		buf,	/* in: buffer where to read */
+	ulint		n,	/* in: number of bytes to read */
+	ulint		offset,	/* in: least significant 32 bits of file
+				offset from where to read */
+	ulint		offset_high) /* in: most significant 32 bits of
+				offset */
+{
+	off_t	offs;
+	ssize_t	n_bytes;
+
+	ut_a((offset & 0xFFFFFFFFUL) == offset);
+
+	/* If off_t is > 4 bytes in size, then we assume we can pass a
+	64-bit address */
+
+	if (sizeof(off_t) > 4) {
+		offs = (off_t)offset + (((off_t)offset_high) << 32);
+
+	} else {
+		offs = (off_t)offset;
+
+		if (offset_high > 0) {
+			fprintf(stderr,
+				"InnoDB: Error: file read at offset > 4 GB\n");
+		}
+	}
+
+	os_n_file_reads++;
+
+#if defined(HAVE_PREAD) && !defined(HAVE_BROKEN_PREAD)
+	os_mutex_enter(os_file_count_mutex);
+	os_file_n_pending_preads++;
+	os_n_pending_reads++;
+	os_mutex_exit(os_file_count_mutex);
+
+	n_bytes = pread(file, buf, (ssize_t)n, offs);
+
+	os_mutex_enter(os_file_count_mutex);
+	os_file_n_pending_preads--;
+	os_n_pending_reads--;
+	os_mutex_exit(os_file_count_mutex);
+
+	return(n_bytes);
+#else
+	{
+		off_t	ret_offset;
+		ssize_t	ret;
+		ulint	i;
+
+		os_mutex_enter(os_file_count_mutex);
+		os_n_pending_reads++;
+		os_mutex_exit(os_file_count_mutex);
+
+		/* Protect the seek / read operation with a mutex */
+		i = ((ulint) file) % OS_FILE_N_SEEK_MUTEXES;
+
+		os_mutex_enter(os_file_seek_mutexes[i]);
+
+		ret_offset = lseek(file, offs, SEEK_SET);
+
+		if (ret_offset < 0) {
+			ret = -1;
+		} else {
+			ret = read(file, buf, (ssize_t)n);
+		}
+
+		os_mutex_exit(os_file_seek_mutexes[i]);
+
+		os_mutex_enter(os_file_count_mutex);
+		os_n_pending_reads--;
+		os_mutex_exit(os_file_count_mutex);
+
+		return(ret);
+	}
+#endif
+}
+
+/***********************************************************************
+Does a synchronous write operation in Posix. */
+static
+ssize_t
+os_file_pwrite(
+/*===========*/
+				/* out: number of bytes written, -1 if error */
+	os_file_t	file,	/* in: handle to a file */
+	const void*	buf,	/* in: buffer from where to write */
+	ulint		n,	/* in: number of bytes to write */
+	ulint		offset,	/* in: least significant 32 bits of file
+				offset where to write */
+	ulint		offset_high) /* in: most significant 32 bits of
+				offset */
+{
+	ssize_t	ret;
+	off_t	offs;
+
+	ut_a((offset & 0xFFFFFFFFUL) == offset);
+
+	/* If off_t is > 4 bytes in size, then we assume we can pass a
+	64-bit address */
+
+	if (sizeof(off_t) > 4) {
+		offs = (off_t)offset + (((off_t)offset_high) << 32);
+	} else {
+		offs = (off_t)offset;
+
+		if (offset_high > 0) {
+			fprintf(stderr,
+				"InnoDB: Error: file write"
+				" at offset > 4 GB\n");
+		}
+	}
+
+	os_n_file_writes++;
+
+#if defined(HAVE_PWRITE) && !defined(HAVE_BROKEN_PREAD)
+	os_mutex_enter(os_file_count_mutex);
+	os_file_n_pending_pwrites++;
+	os_n_pending_writes++;
+	os_mutex_exit(os_file_count_mutex);
+
+	ret = pwrite(file, buf, (ssize_t)n, offs);
+
+	os_mutex_enter(os_file_count_mutex);
+	os_file_n_pending_pwrites--;
+	os_n_pending_writes--;
+	os_mutex_exit(os_file_count_mutex);
+
+# ifdef UNIV_DO_FLUSH
+	if (srv_unix_file_flush_method != SRV_UNIX_LITTLESYNC
+	    && srv_unix_file_flush_method != SRV_UNIX_NOSYNC
+	    && !os_do_not_call_flush_at_each_write) {
+
+		/* Always do fsync to reduce the probability that when
+		the OS crashes, a database page is only partially
+		physically written to disk. */
+
+		ut_a(TRUE == os_file_flush(file));
+	}
+# endif /* UNIV_DO_FLUSH */
+
+	return(ret);
+#else
+	{
+		off_t	ret_offset;
+		ulint	i;
+
+		os_mutex_enter(os_file_count_mutex);
+		os_n_pending_writes++;
+		os_mutex_exit(os_file_count_mutex);
+
+		/* Protect the seek / write operation with a mutex */
+		i = ((ulint) file) % OS_FILE_N_SEEK_MUTEXES;
+
+		os_mutex_enter(os_file_seek_mutexes[i]);
+
+		ret_offset = lseek(file, offs, SEEK_SET);
+
+		if (ret_offset < 0) {
+			ret = -1;
+
+			goto func_exit;
+		}
+
+		ret = write(file, buf, (ssize_t)n);
+
+# ifdef UNIV_DO_FLUSH
+		if (srv_unix_file_flush_method != SRV_UNIX_LITTLESYNC
+		    && srv_unix_file_flush_method != SRV_UNIX_NOSYNC
+		    && !os_do_not_call_flush_at_each_write) {
+
+			/* Always do fsync to reduce the probability that when
+			the OS crashes, a database page is only partially
+			physically written to disk. */
+
+			ut_a(TRUE == os_file_flush(file));
+		}
+# endif /* UNIV_DO_FLUSH */
+
+func_exit:
+		os_mutex_exit(os_file_seek_mutexes[i]);
+
+		os_mutex_enter(os_file_count_mutex);
+		os_n_pending_writes--;
+		os_mutex_exit(os_file_count_mutex);
+
+		return(ret);
+	}
+#endif
+}
+#endif
+
+/***********************************************************************
+Requests a synchronous positioned read operation. */
+UNIV_INTERN
+ibool
+os_file_read(
+/*=========*/
+				/* out: TRUE if request was
+				successful, FALSE if fail */
+	os_file_t	file,	/* in: handle to a file */
+	void*		buf,	/* in: buffer where to read */
+	ulint		offset,	/* in: least significant 32 bits of file
+				offset where to read */
+	ulint		offset_high, /* in: most significant 32 bits of
+				offset */
+	ulint		n)	/* in: number of bytes to read */
+{
+#ifdef __WIN__
+	BOOL		ret;
+	DWORD		len;
+	DWORD		ret2;
+	DWORD		low;
+	DWORD		high;
+	ibool		retry;
+	ulint		i;
+
+	ut_a((offset & 0xFFFFFFFFUL) == offset);
+
+	os_n_file_reads++;
+	os_bytes_read_since_printout += n;
+
+try_again:
+	ut_ad(file);
+	ut_ad(buf);
+	ut_ad(n > 0);
+
+	low = (DWORD) offset;
+	high = (DWORD) offset_high;
+
+	os_mutex_enter(os_file_count_mutex);
+	os_n_pending_reads++;
+	os_mutex_exit(os_file_count_mutex);
+
+	/* Protect the seek / read operation with a mutex */
+	i = ((ulint) file) % OS_FILE_N_SEEK_MUTEXES;
+
+	os_mutex_enter(os_file_seek_mutexes[i]);
+
+	ret2 = SetFilePointer(file, low, &high, FILE_BEGIN);
+
+	if (ret2 == 0xFFFFFFFF && GetLastError() != NO_ERROR) {
+
+		os_mutex_exit(os_file_seek_mutexes[i]);
+
+		os_mutex_enter(os_file_count_mutex);
+		os_n_pending_reads--;
+		os_mutex_exit(os_file_count_mutex);
+
+		goto error_handling;
+	}
+
+	ret = ReadFile(file, buf, (DWORD) n, &len, NULL);
+
+	os_mutex_exit(os_file_seek_mutexes[i]);
+
+	os_mutex_enter(os_file_count_mutex);
+	os_n_pending_reads--;
+	os_mutex_exit(os_file_count_mutex);
+
+	if (ret && len == n) {
+		return(TRUE);
+	}
+#else
+	ibool	retry;
+	ssize_t	ret;
+
+	os_bytes_read_since_printout += n;
+
+try_again:
+	ret = os_file_pread(file, buf, n, offset, offset_high);
+
+	if ((ulint)ret == n) {
+
+		return(TRUE);
+	}
+
+	fprintf(stderr,
+		"InnoDB: Error: tried to read %lu bytes at offset %lu %lu.\n"
+		"InnoDB: Was only able to read %ld.\n",
+		(ulong)n, (ulong)offset_high,
+		(ulong)offset, (long)ret);
+#endif
+#ifdef __WIN__
+error_handling:
+#endif
+	retry = os_file_handle_error(NULL, "read");
+
+	if (retry) {
+		goto try_again;
+	}
+
+	fprintf(stderr,
+		"InnoDB: Fatal error: cannot read from file."
+		" OS error number %lu.\n",
+#ifdef __WIN__
+		(ulong) GetLastError()
+#else
+		(ulong) errno
+#endif
+		);
+	fflush(stderr);
+
+	ut_error;
+
+	return(FALSE);
+}
+
+/***********************************************************************
+Requests a synchronous positioned read operation. This function does not do
+any error handling. In case of error it returns FALSE. */
+UNIV_INTERN
+ibool
+os_file_read_no_error_handling(
+/*===========================*/
+				/* out: TRUE if request was
+				successful, FALSE if fail */
+	os_file_t	file,	/* in: handle to a file */
+	void*		buf,	/* in: buffer where to read */
+	ulint		offset,	/* in: least significant 32 bits of file
+				offset where to read */
+	ulint		offset_high, /* in: most significant 32 bits of
+				offset */
+	ulint		n)	/* in: number of bytes to read */
+{
+#ifdef __WIN__
+	BOOL		ret;
+	DWORD		len;
+	DWORD		ret2;
+	DWORD		low;
+	DWORD		high;
+	ibool		retry;
+	ulint		i;
+
+	ut_a((offset & 0xFFFFFFFFUL) == offset);
+
+	os_n_file_reads++;
+	os_bytes_read_since_printout += n;
+
+try_again:
+	ut_ad(file);
+	ut_ad(buf);
+	ut_ad(n > 0);
+
+	low = (DWORD) offset;
+	high = (DWORD) offset_high;
+
+	os_mutex_enter(os_file_count_mutex);
+	os_n_pending_reads++;
+	os_mutex_exit(os_file_count_mutex);
+
+	/* Protect the seek / read operation with a mutex */
+	i = ((ulint) file) % OS_FILE_N_SEEK_MUTEXES;
+
+	os_mutex_enter(os_file_seek_mutexes[i]);
+
+	ret2 = SetFilePointer(file, low, &high, FILE_BEGIN);
+
+	if (ret2 == 0xFFFFFFFF && GetLastError() != NO_ERROR) {
+
+		os_mutex_exit(os_file_seek_mutexes[i]);
+
+		os_mutex_enter(os_file_count_mutex);
+		os_n_pending_reads--;
+		os_mutex_exit(os_file_count_mutex);
+
+		goto error_handling;
+	}
+
+	ret = ReadFile(file, buf, (DWORD) n, &len, NULL);
+
+	os_mutex_exit(os_file_seek_mutexes[i]);
+
+	os_mutex_enter(os_file_count_mutex);
+	os_n_pending_reads--;
+	os_mutex_exit(os_file_count_mutex);
+
+	if (ret && len == n) {
+		return(TRUE);
+	}
+#else
+	ibool	retry;
+	ssize_t	ret;
+
+	os_bytes_read_since_printout += n;
+
+try_again:
+	ret = os_file_pread(file, buf, n, offset, offset_high);
+
+	if ((ulint)ret == n) {
+
+		return(TRUE);
+	}
+#endif
+#ifdef __WIN__
+error_handling:
+#endif
+	retry = os_file_handle_error_no_exit(NULL, "read");
+
+	if (retry) {
+		goto try_again;
+	}
+
+	return(FALSE);
+}
+
+/***********************************************************************
+Rewind file to its start, read at most size - 1 bytes from it to str, and
+NUL-terminate str. All errors are silently ignored. This function is
+mostly meant to be used with temporary files. */
+UNIV_INTERN
+void
+os_file_read_string(
+/*================*/
+	FILE*	file,	/* in: file to read from */
+	char*	str,	/* in: buffer where to read */
+	ulint	size)	/* in: size of buffer */
+{
+	size_t	flen;
+
+	if (size == 0) {
+		return;
+	}
+
+	rewind(file);
+	flen = fread(str, 1, size - 1, file);
+	str[flen] = '\0';
+}
+
+/***********************************************************************
+Requests a synchronous write operation. */
+UNIV_INTERN
+ibool
+os_file_write(
+/*==========*/
+				/* out: TRUE if request was
+				successful, FALSE if fail */
+	const char*	name,	/* in: name of the file or path as a
+				null-terminated string */
+	os_file_t	file,	/* in: handle to a file */
+	const void*	buf,	/* in: buffer from which to write */
+	ulint		offset,	/* in: least significant 32 bits of file
+				offset where to write */
+	ulint		offset_high, /* in: most significant 32 bits of
+				offset */
+	ulint		n)	/* in: number of bytes to write */
+{
+#ifdef __WIN__
+	BOOL		ret;
+	DWORD		len;
+	DWORD		ret2;
+	DWORD		low;
+	DWORD		high;
+	ulint		i;
+	ulint		n_retries	= 0;
+	ulint		err;
+
+	ut_a((offset & 0xFFFFFFFF) == offset);
+
+	os_n_file_writes++;
+
+	ut_ad(file);
+	ut_ad(buf);
+	ut_ad(n > 0);
+retry:
+	low = (DWORD) offset;
+	high = (DWORD) offset_high;
+
+	os_mutex_enter(os_file_count_mutex);
+	os_n_pending_writes++;
+	os_mutex_exit(os_file_count_mutex);
+
+	/* Protect the seek / write operation with a mutex */
+	i = ((ulint) file) % OS_FILE_N_SEEK_MUTEXES;
+
+	os_mutex_enter(os_file_seek_mutexes[i]);
+
+	ret2 = SetFilePointer(file, low, &high, FILE_BEGIN);
+
+	if (ret2 == 0xFFFFFFFF && GetLastError() != NO_ERROR) {
+
+		os_mutex_exit(os_file_seek_mutexes[i]);
+
+		os_mutex_enter(os_file_count_mutex);
+		os_n_pending_writes--;
+		os_mutex_exit(os_file_count_mutex);
+
+		ut_print_timestamp(stderr);
+
+		fprintf(stderr,
+			"  InnoDB: Error: File pointer positioning to"
+			" file %s failed at\n"
+			"InnoDB: offset %lu %lu. Operating system"
+			" error number %lu.\n"
+			"InnoDB: Some operating system error numbers"
+			" are described at\n"
+			"InnoDB: "
+			"http://dev.mysql.com/doc/refman/5.1/en/"
+			"operating-system-error-codes.html\n",
+			name, (ulong) offset_high, (ulong) offset,
+			(ulong) GetLastError());
+
+		return(FALSE);
+	}
+
+	ret = WriteFile(file, buf, (DWORD) n, &len, NULL);
+
+	/* Always do fsync to reduce the probability that when the OS crashes,
+	a database page is only partially physically written to disk. */
+
+# ifdef UNIV_DO_FLUSH
+	if (!os_do_not_call_flush_at_each_write) {
+		ut_a(TRUE == os_file_flush(file));
+	}
+# endif /* UNIV_DO_FLUSH */
+
+	os_mutex_exit(os_file_seek_mutexes[i]);
+
+	os_mutex_enter(os_file_count_mutex);
+	os_n_pending_writes--;
+	os_mutex_exit(os_file_count_mutex);
+
+	if (ret && len == n) {
+
+		return(TRUE);
+	}
+
+	/* If some background file system backup tool is running, then, at
+	least in Windows 2000, we may get here a specific error. Let us
+	retry the operation 100 times, with 1 second waits. */
+
+	if (GetLastError() == ERROR_LOCK_VIOLATION && n_retries < 100) {
+
+		os_thread_sleep(1000000);
+
+		n_retries++;
+
+		goto retry;
+	}
+
+	if (!os_has_said_disk_full) {
+
+		err = (ulint)GetLastError();
+
+		ut_print_timestamp(stderr);
+
+		fprintf(stderr,
+			"  InnoDB: Error: Write to file %s failed"
+			" at offset %lu %lu.\n"
+			"InnoDB: %lu bytes should have been written,"
+			" only %lu were written.\n"
+			"InnoDB: Operating system error number %lu.\n"
+			"InnoDB: Check that your OS and file system"
+			" support files of this size.\n"
+			"InnoDB: Check also that the disk is not full"
+			" or a disk quota exceeded.\n",
+			name, (ulong) offset_high, (ulong) offset,
+			(ulong) n, (ulong) len, (ulong) err);
+
+		if (strerror((int)err) != NULL) {
+			fprintf(stderr,
+				"InnoDB: Error number %lu means '%s'.\n",
+				(ulong) err, strerror((int)err));
+		}
+
+		fprintf(stderr,
+			"InnoDB: Some operating system error numbers"
+			" are described at\n"
+			"InnoDB: "
+			"http://dev.mysql.com/doc/refman/5.1/en/"
+			"operating-system-error-codes.html\n");
+
+		os_has_said_disk_full = TRUE;
+	}
+
+	return(FALSE);
+#else
+	ssize_t	ret;
+
+	ret = os_file_pwrite(file, buf, n, offset, offset_high);
+
+	if ((ulint)ret == n) {
+
+		return(TRUE);
+	}
+
+	if (!os_has_said_disk_full) {
+
+		ut_print_timestamp(stderr);
+
+		fprintf(stderr,
+			"  InnoDB: Error: Write to file %s failed"
+			" at offset %lu %lu.\n"
+			"InnoDB: %lu bytes should have been written,"
+			" only %ld were written.\n"
+			"InnoDB: Operating system error number %lu.\n"
+			"InnoDB: Check that your OS and file system"
+			" support files of this size.\n"
+			"InnoDB: Check also that the disk is not full"
+			" or a disk quota exceeded.\n",
+			name, offset_high, offset, n, (long int)ret,
+			(ulint)errno);
+		if (strerror(errno) != NULL) {
+			fprintf(stderr,
+				"InnoDB: Error number %lu means '%s'.\n",
+				(ulint)errno, strerror(errno));
+		}
+
+		fprintf(stderr,
+			"InnoDB: Some operating system error numbers"
+			" are described at\n"
+			"InnoDB: "
+			"http://dev.mysql.com/doc/refman/5.1/en/"
+			"operating-system-error-codes.html\n");
+
+		os_has_said_disk_full = TRUE;
+	}
+
+	return(FALSE);
+#endif
+}
+
+/***********************************************************************
+Check the existence and type of the given file. */
+UNIV_INTERN
+ibool
+os_file_status(
+/*===========*/
+				/* out: TRUE if call succeeded */
+	const char*	path,	/* in:	pathname of the file */
+	ibool*		exists,	/* out: TRUE if file exists */
+	os_file_type_t* type)	/* out: type of the file (if it exists) */
+{
+#ifdef __WIN__
+	int		ret;
+	struct _stat	statinfo;
+
+	ret = _stat(path, &statinfo);
+	if (ret && (errno == ENOENT || errno == ENOTDIR)) {
+		/* file does not exist */
+		*exists = FALSE;
+		return(TRUE);
+	} else if (ret) {
+		/* file exists, but stat call failed */
+
+		os_file_handle_error_no_exit(path, "stat");
+
+		return(FALSE);
+	}
+
+	if (_S_IFDIR & statinfo.st_mode) {
+		*type = OS_FILE_TYPE_DIR;
+	} else if (_S_IFREG & statinfo.st_mode) {
+		*type = OS_FILE_TYPE_FILE;
+	} else {
+		*type = OS_FILE_TYPE_UNKNOWN;
+	}
+
+	*exists = TRUE;
+
+	return(TRUE);
+#else
+	int		ret;
+	struct stat	statinfo;
+
+	ret = stat(path, &statinfo);
+	if (ret && (errno == ENOENT || errno == ENOTDIR)) {
+		/* file does not exist */
+		*exists = FALSE;
+		return(TRUE);
+	} else if (ret) {
+		/* file exists, but stat call failed */
+
+		os_file_handle_error_no_exit(path, "stat");
+
+		return(FALSE);
+	}
+
+	if (S_ISDIR(statinfo.st_mode)) {
+		*type = OS_FILE_TYPE_DIR;
+	} else if (S_ISLNK(statinfo.st_mode)) {
+		*type = OS_FILE_TYPE_LINK;
+	} else if (S_ISREG(statinfo.st_mode)) {
+		*type = OS_FILE_TYPE_FILE;
+	} else {
+		*type = OS_FILE_TYPE_UNKNOWN;
+	}
+
+	*exists = TRUE;
+
+	return(TRUE);
+#endif
+}
+
+/***********************************************************************
+This function returns information about the specified file */
+UNIV_INTERN
+ibool
+os_file_get_status(
+/*===============*/
+					/* out: TRUE if stat
+					information found */
+	const char*	path,		/* in:	pathname of the file */
+	os_file_stat_t* stat_info)	/* information of a file in a
+					directory */
+{
+#ifdef __WIN__
+	int		ret;
+	struct _stat	statinfo;
+
+	ret = _stat(path, &statinfo);
+	if (ret && (errno == ENOENT || errno == ENOTDIR)) {
+		/* file does not exist */
+
+		return(FALSE);
+	} else if (ret) {
+		/* file exists, but stat call failed */
+
+		os_file_handle_error_no_exit(path, "stat");
+
+		return(FALSE);
+	}
+	if (_S_IFDIR & statinfo.st_mode) {
+		stat_info->type = OS_FILE_TYPE_DIR;
+	} else if (_S_IFREG & statinfo.st_mode) {
+		stat_info->type = OS_FILE_TYPE_FILE;
+	} else {
+		stat_info->type = OS_FILE_TYPE_UNKNOWN;
+	}
+
+	stat_info->ctime = statinfo.st_ctime;
+	stat_info->atime = statinfo.st_atime;
+	stat_info->mtime = statinfo.st_mtime;
+	stat_info->size	 = statinfo.st_size;
+
+	return(TRUE);
+#else
+	int		ret;
+	struct stat	statinfo;
+
+	ret = stat(path, &statinfo);
+
+	if (ret && (errno == ENOENT || errno == ENOTDIR)) {
+		/* file does not exist */
+
+		return(FALSE);
+	} else if (ret) {
+		/* file exists, but stat call failed */
+
+		os_file_handle_error_no_exit(path, "stat");
+
+		return(FALSE);
+	}
+
+	if (S_ISDIR(statinfo.st_mode)) {
+		stat_info->type = OS_FILE_TYPE_DIR;
+	} else if (S_ISLNK(statinfo.st_mode)) {
+		stat_info->type = OS_FILE_TYPE_LINK;
+	} else if (S_ISREG(statinfo.st_mode)) {
+		stat_info->type = OS_FILE_TYPE_FILE;
+	} else {
+		stat_info->type = OS_FILE_TYPE_UNKNOWN;
+	}
+
+	stat_info->ctime = statinfo.st_ctime;
+	stat_info->atime = statinfo.st_atime;
+	stat_info->mtime = statinfo.st_mtime;
+	stat_info->size	 = statinfo.st_size;
+
+	return(TRUE);
+#endif
+}
+
+/* path name separator character */
+#ifdef __WIN__
+#  define OS_FILE_PATH_SEPARATOR	'\\'
+#else
+#  define OS_FILE_PATH_SEPARATOR	'/'
+#endif
+
+/********************************************************************
+The function os_file_dirname returns a directory component of a
+null-terminated pathname string.  In the usual case, dirname returns
+the string up to, but not including, the final '/', and basename
+is the component following the final '/'.  Trailing '/' charac­
+ters are not counted as part of the pathname.
+
+If path does not contain a slash, dirname returns the string ".".
+
+Concatenating the string returned by dirname, a "/", and the basename
+yields a complete pathname.
+
+The return value is  a copy of the directory component of the pathname.
+The copy is allocated from heap. It is the caller responsibility
+to free it after it is no longer needed.
+
+The following list of examples (taken from SUSv2) shows the strings
+returned by dirname and basename for different paths:
+
+       path	      dirname	     basename
+       "/usr/lib"     "/usr"	     "lib"
+       "/usr/"	      "/"	     "usr"
+       "usr"	      "."	     "usr"
+       "/"	      "/"	     "/"
+       "."	      "."	     "."
+       ".."	      "."	     ".."
+*/
+UNIV_INTERN
+char*
+os_file_dirname(
+/*============*/
+				/* out, own: directory component of the
+				pathname */
+	const char*	path)	/* in: pathname */
+{
+	/* Find the offset of the last slash */
+	const char* last_slash = strrchr(path, OS_FILE_PATH_SEPARATOR);
+	if (!last_slash) {
+		/* No slash in the path, return "." */
+
+		return(mem_strdup("."));
+	}
+
+	/* Ok, there is a slash */
+
+	if (last_slash == path) {
+		/* last slash is the first char of the path */
+
+		return(mem_strdup("/"));
+	}
+
+	/* Non-trivial directory component */
+
+	return(mem_strdupl(path, last_slash - path));
+}
+
+/********************************************************************
+Creates all missing subdirectories along the given path. */
+UNIV_INTERN
+ibool
+os_file_create_subdirs_if_needed(
+/*=============================*/
+				/* out: TRUE if call succeeded
+				   FALSE otherwise */
+	const char*	path)	/* in: path name */
+{
+	char*		subdir;
+	ibool		success, subdir_exists;
+	os_file_type_t	type;
+
+	subdir = os_file_dirname(path);
+	if (strlen(subdir) == 1
+	    && (*subdir == OS_FILE_PATH_SEPARATOR || *subdir == '.')) {
+		/* subdir is root or cwd, nothing to do */
+		mem_free(subdir);
+
+		return(TRUE);
+	}
+
+	/* Test if subdir exists */
+	success = os_file_status(subdir, &subdir_exists, &type);
+	if (success && !subdir_exists) {
+		/* subdir does not exist, create it */
+		success = os_file_create_subdirs_if_needed(subdir);
+		if (!success) {
+			mem_free(subdir);
+
+			return(FALSE);
+		}
+		success = os_file_create_directory(subdir, FALSE);
+	}
+
+	mem_free(subdir);
+
+	return(success);
+}
+
+/********************************************************************
+Returns a pointer to the nth slot in the aio array. */
+static
+os_aio_slot_t*
+os_aio_array_get_nth_slot(
+/*======================*/
+					/* out: pointer to slot */
+	os_aio_array_t*		array,	/* in: aio array */
+	ulint			index)	/* in: index of the slot */
+{
+	ut_a(index < array->n_slots);
+
+	return((array->slots) + index);
+}
+
+/****************************************************************************
+Creates an aio wait array. */
+static
+os_aio_array_t*
+os_aio_array_create(
+/*================*/
+				/* out, own: aio array */
+	ulint	n,		/* in: maximum number of pending aio operations
+				allowed; n must be divisible by n_segments */
+	ulint	n_segments)	/* in: number of segments in the aio array */
+{
+	os_aio_array_t*	array;
+	ulint		i;
+	os_aio_slot_t*	slot;
+#ifdef WIN_ASYNC_IO
+	OVERLAPPED*	over;
+#endif
+	ut_a(n > 0);
+	ut_a(n_segments > 0);
+
+	array = ut_malloc(sizeof(os_aio_array_t));
+
+	array->mutex		= os_mutex_create(NULL);
+	array->not_full		= os_event_create(NULL);
+	array->is_empty		= os_event_create(NULL);
+
+	os_event_set(array->is_empty);
+
+	array->n_slots		= n;
+	array->n_segments	= n_segments;
+	array->n_reserved	= 0;
+	array->slots		= ut_malloc(n * sizeof(os_aio_slot_t));
+#ifdef __WIN__
+	array->native_events	= ut_malloc(n * sizeof(os_native_event_t));
+#endif
+	for (i = 0; i < n; i++) {
+		slot = os_aio_array_get_nth_slot(array, i);
+
+		slot->pos = i;
+		slot->reserved = FALSE;
+#ifdef WIN_ASYNC_IO
+		slot->event = os_event_create(NULL);
+
+		over = &(slot->control);
+
+		over->hEvent = slot->event->handle;
+
+		*((array->native_events) + i) = over->hEvent;
+#endif
+	}
+
+	return(array);
+}
+
+/****************************************************************************
+Initializes the asynchronous io system. Calls also os_io_init_simple.
+Creates a separate aio array for
+non-ibuf read and write, a third aio array for the ibuf i/o, with just one
+segment, two aio arrays for log reads and writes with one segment, and a
+synchronous aio array of the specified size. The combined number of segments
+in the three first aio arrays is the parameter n_segments given to the
+function. The caller must create an i/o handler thread for each segment in
+the four first arrays, but not for the sync aio array. */
+UNIV_INTERN
+void
+os_aio_init(
+/*========*/
+	ulint	n,		/* in: maximum number of pending aio operations
+				allowed; n must be divisible by n_segments */
+//	ulint	n_segments,	/* in: combined number of segments in the four
+//				first aio arrays; must be >= 4 */
+	ulint	n_read_threads,  /* n_segments == 2 + n_read_threads + n_write_threads*/
+	ulint	n_write_threads, /**/
+	ulint	n_slots_sync)	/* in: number of slots in the sync aio array */
+{
+	ulint	n_read_segs;
+	ulint	n_write_segs;
+	ulint	n_per_seg;
+	ulint	i;
+
+ 	ulint	n_segments = 2 + n_read_threads + n_write_threads;
+ 
+	ut_ad(n % n_segments == 0);
+	ut_ad(n_segments >= 4);
+
+	os_io_init_simple();
+
+	for (i = 0; i < n_segments; i++) {
+		srv_set_io_thread_op_info(i, "not started yet");
+	}
+
+	n_per_seg = n / n_segments;
+	n_write_segs = n_write_threads;
+	n_read_segs = n_read_threads;
+
+	/* fprintf(stderr, "Array n per seg %lu\n", n_per_seg); */
+
+	os_aio_ibuf_array = os_aio_array_create(n_per_seg, 1);
+
+	srv_io_thread_function[0] = "insert buffer thread";
+
+	os_aio_log_array = os_aio_array_create(n_per_seg, 1);
+
+	srv_io_thread_function[1] = "log thread";
+
+	os_aio_read_array = os_aio_array_create(n_read_segs * n_per_seg,
+						n_read_segs);
+	for (i = 2; i < 2 + n_read_segs; i++) {
+		ut_a(i < SRV_MAX_N_IO_THREADS);
+		srv_io_thread_function[i] = "read thread";
+	}
+
+	os_aio_write_array = os_aio_array_create(n_write_segs * n_per_seg,
+						 n_write_segs);
+	for (i = 2 + n_read_segs; i < n_segments; i++) {
+		ut_a(i < SRV_MAX_N_IO_THREADS);
+		srv_io_thread_function[i] = "write thread";
+	}
+
+	os_aio_sync_array = os_aio_array_create(n_slots_sync, 1);
+
+	os_aio_n_segments = n_segments;
+
+	os_aio_validate();
+
+	os_aio_segment_wait_events = ut_malloc(n_segments * sizeof(void*));
+
+	for (i = 0; i < n_segments; i++) {
+		os_aio_segment_wait_events[i] = os_event_create(NULL);
+	}
+
+	os_last_printout = time(NULL);
+
+}
+
+#ifdef WIN_ASYNC_IO
+/****************************************************************************
+Wakes up all async i/o threads in the array in Windows async i/o at
+shutdown. */
+static
+void
+os_aio_array_wake_win_aio_at_shutdown(
+/*==================================*/
+	os_aio_array_t*	array)	/* in: aio array */
+{
+	ulint	i;
+
+	for (i = 0; i < array->n_slots; i++) {
+
+		os_event_set((array->slots + i)->event);
+	}
+}
+#endif
+
+/****************************************************************************
+Wakes up all async i/o threads so that they know to exit themselves in
+shutdown. */
+UNIV_INTERN
+void
+os_aio_wake_all_threads_at_shutdown(void)
+/*=====================================*/
+{
+	ulint	i;
+
+#ifdef WIN_ASYNC_IO
+	/* This code wakes up all ai/o threads in Windows native aio */
+	os_aio_array_wake_win_aio_at_shutdown(os_aio_read_array);
+	os_aio_array_wake_win_aio_at_shutdown(os_aio_write_array);
+	os_aio_array_wake_win_aio_at_shutdown(os_aio_ibuf_array);
+	os_aio_array_wake_win_aio_at_shutdown(os_aio_log_array);
+#endif
+	/* This loop wakes up all simulated ai/o threads */
+
+	for (i = 0; i < os_aio_n_segments; i++) {
+
+		os_event_set(os_aio_segment_wait_events[i]);
+	}
+}
+
+/****************************************************************************
+Waits until there are no pending writes in os_aio_write_array. There can
+be other, synchronous, pending writes. */
+UNIV_INTERN
+void
+os_aio_wait_until_no_pending_writes(void)
+/*=====================================*/
+{
+	os_event_wait(os_aio_write_array->is_empty);
+}
+
+/**************************************************************************
+Calculates segment number for a slot. */
+static
+ulint
+os_aio_get_segment_no_from_slot(
+/*============================*/
+				/* out: segment number (which is the number
+				used by, for example, i/o-handler threads) */
+	os_aio_array_t*	array,	/* in: aio wait array */
+	os_aio_slot_t*	slot)	/* in: slot in this array */
+{
+	ulint	segment;
+	ulint	seg_len;
+
+	if (array == os_aio_ibuf_array) {
+		segment = 0;
+
+	} else if (array == os_aio_log_array) {
+		segment = 1;
+
+	} else if (array == os_aio_read_array) {
+		seg_len = os_aio_read_array->n_slots
+			/ os_aio_read_array->n_segments;
+
+		segment = 2 + slot->pos / seg_len;
+	} else {
+		ut_a(array == os_aio_write_array);
+		seg_len = os_aio_write_array->n_slots
+			/ os_aio_write_array->n_segments;
+
+		segment = os_aio_read_array->n_segments + 2
+			+ slot->pos / seg_len;
+	}
+
+	return(segment);
+}
+
+/**************************************************************************
+Calculates local segment number and aio array from global segment number. */
+static
+ulint
+os_aio_get_array_and_local_segment(
+/*===============================*/
+					/* out: local segment number within
+					the aio array */
+	os_aio_array_t** array,		/* out: aio wait array */
+	ulint		 global_segment)/* in: global segment number */
+{
+	ulint	segment;
+
+	ut_a(global_segment < os_aio_n_segments);
+
+	if (global_segment == 0) {
+		*array = os_aio_ibuf_array;
+		segment = 0;
+
+	} else if (global_segment == 1) {
+		*array = os_aio_log_array;
+		segment = 0;
+
+	} else if (global_segment < os_aio_read_array->n_segments + 2) {
+		*array = os_aio_read_array;
+
+		segment = global_segment - 2;
+	} else {
+		*array = os_aio_write_array;
+
+		segment = global_segment - (os_aio_read_array->n_segments + 2);
+	}
+
+	return(segment);
+}
+
+/***********************************************************************
+Requests for a slot in the aio array. If no slot is available, waits until
+not_full-event becomes signaled. */
+static
+os_aio_slot_t*
+os_aio_array_reserve_slot(
+/*======================*/
+				/* out: pointer to slot */
+	ulint		type,	/* in: OS_FILE_READ or OS_FILE_WRITE */
+	os_aio_array_t*	array,	/* in: aio array */
+	fil_node_t*	message1,/* in: message to be passed along with
+				the aio operation */
+	void*		message2,/* in: message to be passed along with
+				the aio operation */
+	os_file_t	file,	/* in: file handle */
+	const char*	name,	/* in: name of the file or path as a
+				null-terminated string */
+	void*		buf,	/* in: buffer where to read or from which
+				to write */
+	ulint		offset,	/* in: least significant 32 bits of file
+				offset */
+	ulint		offset_high, /* in: most significant 32 bits of
+				offset */
+	ulint		len)	/* in: length of the block to read or write */
+{
+	os_aio_slot_t*	slot;
+#ifdef WIN_ASYNC_IO
+	OVERLAPPED*	control;
+#endif
+	ulint		i;
+	ulint		prim_segment;
+	ulint		n;
+
+	n = array->n_slots / array->n_segments;
+	/* 64 blocks' striping ( aligning max(BUF_READ_AHEAD_AREA) ) */
+	prim_segment = ( offset >> (UNIV_PAGE_SIZE_SHIFT + 6) ) % (array->n_segments);
+
+loop:
+	os_mutex_enter(array->mutex);
+
+	if (array->n_reserved == array->n_slots) {
+		os_mutex_exit(array->mutex);
+
+		if (!os_aio_use_native_aio) {
+			/* If the handler threads are suspended, wake them
+			so that we get more slots */
+
+			os_aio_simulated_wake_handler_threads();
+		}
+
+		os_event_wait(array->not_full);
+
+		goto loop;
+	}
+
+	for (i = prim_segment * n; i < array->n_slots; i++) {
+		slot = os_aio_array_get_nth_slot(array, i);
+
+		if (slot->reserved == FALSE) {
+			break;
+		}
+	}
+
+	if (slot->reserved == TRUE){
+		/* Not found after the intended segment. So we should search before. */
+	for (i = 0;; i++) {
+		slot = os_aio_array_get_nth_slot(array, i);
+
+		if (slot->reserved == FALSE) {
+			break;
+		}
+	}
+	}
+
+	array->n_reserved++;
+
+	if (array->n_reserved == 1) {
+		os_event_reset(array->is_empty);
+	}
+
+	if (array->n_reserved == array->n_slots) {
+		os_event_reset(array->not_full);
+	}
+
+	slot->reserved = TRUE;
+	slot->reservation_time = time(NULL);
+	slot->message1 = message1;
+	slot->message2 = message2;
+	slot->file     = file;
+	slot->name     = name;
+	slot->len      = len;
+	slot->type     = type;
+	slot->buf      = buf;
+	slot->offset   = offset;
+	slot->offset_high = offset_high;
+	slot->io_already_done = FALSE;
+
+#ifdef WIN_ASYNC_IO
+	control = &(slot->control);
+	control->Offset = (DWORD)offset;
+	control->OffsetHigh = (DWORD)offset_high;
+	os_event_reset(slot->event);
+#endif
+
+	os_mutex_exit(array->mutex);
+
+	return(slot);
+}
+
+/***********************************************************************
+Frees a slot in the aio array. */
+static
+void
+os_aio_array_free_slot(
+/*===================*/
+	os_aio_array_t*	array,	/* in: aio array */
+	os_aio_slot_t*	slot)	/* in: pointer to slot */
+{
+	ut_ad(array);
+	ut_ad(slot);
+
+	os_mutex_enter(array->mutex);
+
+	ut_ad(slot->reserved);
+
+	slot->reserved = FALSE;
+
+	array->n_reserved--;
+
+	if (array->n_reserved == array->n_slots - 1) {
+		os_event_set(array->not_full);
+	}
+
+	if (array->n_reserved == 0) {
+		os_event_set(array->is_empty);
+	}
+
+#ifdef WIN_ASYNC_IO
+	os_event_reset(slot->event);
+#endif
+	os_mutex_exit(array->mutex);
+}
+
+/**************************************************************************
+Wakes up a simulated aio i/o-handler thread if it has something to do. */
+static
+void
+os_aio_simulated_wake_handler_thread(
+/*=================================*/
+	ulint	global_segment)	/* in: the number of the segment in the aio
+				arrays */
+{
+	os_aio_array_t*	array;
+	os_aio_slot_t*	slot;
+	ulint		segment;
+	ulint		n;
+	ulint		i;
+
+	ut_ad(!os_aio_use_native_aio);
+
+	segment = os_aio_get_array_and_local_segment(&array, global_segment);
+
+	n = array->n_slots / array->n_segments;
+
+	/* Look through n slots after the segment * n'th slot */
+
+	os_mutex_enter(array->mutex);
+
+	for (i = 0; i < n; i++) {
+		slot = os_aio_array_get_nth_slot(array, i + segment * n);
+
+		if (slot->reserved) {
+			/* Found an i/o request */
+
+			break;
+		}
+	}
+
+	os_mutex_exit(array->mutex);
+
+	if (i < n) {
+		os_event_set(os_aio_segment_wait_events[global_segment]);
+	}
+}
+
+/**************************************************************************
+Wakes up simulated aio i/o-handler threads if they have something to do. */
+UNIV_INTERN
+void
+os_aio_simulated_wake_handler_threads(void)
+/*=======================================*/
+{
+	ulint	i;
+
+	if (os_aio_use_native_aio) {
+		/* We do not use simulated aio: do nothing */
+
+		return;
+	}
+
+	os_aio_recommend_sleep_for_read_threads	= FALSE;
+
+	for (i = 0; i < os_aio_n_segments; i++) {
+		os_aio_simulated_wake_handler_thread(i);
+	}
+}
+
+/**************************************************************************
+This function can be called if one wants to post a batch of reads and
+prefers an i/o-handler thread to handle them all at once later. You must
+call os_aio_simulated_wake_handler_threads later to ensure the threads
+are not left sleeping! */
+UNIV_INTERN
+void
+os_aio_simulated_put_read_threads_to_sleep(void)
+/*============================================*/
+{
+	os_aio_array_t*	array;
+	ulint		g;
+
+	os_aio_recommend_sleep_for_read_threads	= TRUE;
+
+	for (g = 0; g < os_aio_n_segments; g++) {
+		os_aio_get_array_and_local_segment(&array, g);
+
+		if (array == os_aio_read_array) {
+
+			os_event_reset(os_aio_segment_wait_events[g]);
+		}
+	}
+}
+
+/***********************************************************************
+Requests an asynchronous i/o operation. */
+UNIV_INTERN
+ibool
+os_aio(
+/*===*/
+				/* out: TRUE if request was queued
+				successfully, FALSE if fail */
+	ulint		type,	/* in: OS_FILE_READ or OS_FILE_WRITE */
+	ulint		mode,	/* in: OS_AIO_NORMAL, ..., possibly ORed
+				to OS_AIO_SIMULATED_WAKE_LATER: the
+				last flag advises this function not to wake
+				i/o-handler threads, but the caller will
+				do the waking explicitly later, in this
+				way the caller can post several requests in
+				a batch; NOTE that the batch must not be
+				so big that it exhausts the slots in aio
+				arrays! NOTE that a simulated batch
+				may introduce hidden chances of deadlocks,
+				because i/os are not actually handled until
+				all have been posted: use with great
+				caution! */
+	const char*	name,	/* in: name of the file or path as a
+				null-terminated string */
+	os_file_t	file,	/* in: handle to a file */
+	void*		buf,	/* in: buffer where to read or from which
+				to write */
+	ulint		offset,	/* in: least significant 32 bits of file
+				offset where to read or write */
+	ulint		offset_high, /* in: most significant 32 bits of
+				offset */
+	ulint		n,	/* in: number of bytes to read or write */
+	fil_node_t*	message1,/* in: messages for the aio handler (these
+				can be used to identify a completed aio
+				operation); if mode is OS_AIO_SYNC, these
+				are ignored */
+	void*		message2)
+{
+	os_aio_array_t*	array;
+	os_aio_slot_t*	slot;
+#ifdef WIN_ASYNC_IO
+	ibool		retval;
+	BOOL		ret		= TRUE;
+	DWORD		len		= (DWORD) n;
+	struct fil_node_struct * dummy_mess1;
+	void*		dummy_mess2;
+	ulint		dummy_type;
+#endif
+	ulint		err		= 0;
+	ibool		retry;
+	ulint		wake_later;
+
+	ut_ad(file);
+	ut_ad(buf);
+	ut_ad(n > 0);
+	ut_ad(n % OS_FILE_LOG_BLOCK_SIZE == 0);
+	ut_ad(offset % OS_FILE_LOG_BLOCK_SIZE == 0);
+	ut_ad(os_aio_validate());
+
+	wake_later = mode & OS_AIO_SIMULATED_WAKE_LATER;
+	mode = mode & (~OS_AIO_SIMULATED_WAKE_LATER);
+
+	if (mode == OS_AIO_SYNC
+#ifdef WIN_ASYNC_IO
+	    && !os_aio_use_native_aio
+#endif
+	    ) {
+		/* This is actually an ordinary synchronous read or write:
+		no need to use an i/o-handler thread. NOTE that if we use
+		Windows async i/o, Windows does not allow us to use
+		ordinary synchronous os_file_read etc. on the same file,
+		therefore we have built a special mechanism for synchronous
+		wait in the Windows case. */
+
+		if (type == OS_FILE_READ) {
+			return(os_file_read(file, buf, offset,
+					    offset_high, n));
+		}
+
+		ut_a(type == OS_FILE_WRITE);
+
+		return(os_file_write(name, file, buf, offset, offset_high, n));
+	}
+
+try_again:
+	if (mode == OS_AIO_NORMAL) {
+		if (type == OS_FILE_READ) {
+			array = os_aio_read_array;
+		} else {
+			array = os_aio_write_array;
+		}
+	} else if (mode == OS_AIO_IBUF) {
+		ut_ad(type == OS_FILE_READ);
+		/* Reduce probability of deadlock bugs in connection with ibuf:
+		do not let the ibuf i/o handler sleep */
+
+		wake_later = FALSE;
+
+		array = os_aio_ibuf_array;
+	} else if (mode == OS_AIO_LOG) {
+
+		array = os_aio_log_array;
+	} else if (mode == OS_AIO_SYNC) {
+		array = os_aio_sync_array;
+	} else {
+		array = NULL; /* Eliminate compiler warning */
+		ut_error;
+	}
+
+	slot = os_aio_array_reserve_slot(type, array, message1, message2, file,
+					 name, buf, offset, offset_high, n);
+	if (type == OS_FILE_READ) {
+		if (os_aio_use_native_aio) {
+#ifdef WIN_ASYNC_IO
+			os_n_file_reads++;
+			os_bytes_read_since_printout += len;
+
+			ret = ReadFile(file, buf, (DWORD)n, &len,
+				       &(slot->control));
+#endif
+		} else {
+			if (!wake_later) {
+				os_aio_simulated_wake_handler_thread(
+					os_aio_get_segment_no_from_slot(
+						array, slot));
+			}
+		}
+	} else if (type == OS_FILE_WRITE) {
+		if (os_aio_use_native_aio) {
+#ifdef WIN_ASYNC_IO
+			os_n_file_writes++;
+			ret = WriteFile(file, buf, (DWORD)n, &len,
+					&(slot->control));
+#endif
+		} else {
+			if (!wake_later) {
+				os_aio_simulated_wake_handler_thread(
+					os_aio_get_segment_no_from_slot(
+						array, slot));
+			}
+		}
+	} else {
+		ut_error;
+	}
+
+#ifdef WIN_ASYNC_IO
+	if (os_aio_use_native_aio) {
+		if ((ret && len == n)
+		    || (!ret && GetLastError() == ERROR_IO_PENDING)) {
+			/* aio was queued successfully! */
+
+			if (mode == OS_AIO_SYNC) {
+				/* We want a synchronous i/o operation on a
+				file where we also use async i/o: in Windows
+				we must use the same wait mechanism as for
+				async i/o */
+
+				retval = os_aio_windows_handle(ULINT_UNDEFINED,
+							       slot->pos,
+							       &dummy_mess1,
+							       &dummy_mess2,
+							       &dummy_type);
+
+				return(retval);
+			}
+
+			return(TRUE);
+		}
+
+		err = 1; /* Fall through the next if */
+	}
+#endif
+	if (err == 0) {
+		/* aio was queued successfully! */
+
+		return(TRUE);
+	}
+
+	os_aio_array_free_slot(array, slot);
+
+	retry = os_file_handle_error(name,
+				     type == OS_FILE_READ
+				     ? "aio read" : "aio write");
+	if (retry) {
+
+		goto try_again;
+	}
+
+	return(FALSE);
+}
+
+#ifdef WIN_ASYNC_IO
+/**************************************************************************
+This function is only used in Windows asynchronous i/o.
+Waits for an aio operation to complete. This function is used to wait the
+for completed requests. The aio array of pending requests is divided
+into segments. The thread specifies which segment or slot it wants to wait
+for. NOTE: this function will also take care of freeing the aio slot,
+therefore no other thread is allowed to do the freeing! */
+UNIV_INTERN
+ibool
+os_aio_windows_handle(
+/*==================*/
+				/* out: TRUE if the aio operation succeeded */
+	ulint	segment,	/* in: the number of the segment in the aio
+				arrays to wait for; segment 0 is the ibuf
+				i/o thread, segment 1 the log i/o thread,
+				then follow the non-ibuf read threads, and as
+				the last are the non-ibuf write threads; if
+				this is ULINT_UNDEFINED, then it means that
+				sync aio is used, and this parameter is
+				ignored */
+	ulint	pos,		/* this parameter is used only in sync aio:
+				wait for the aio slot at this position */
+	fil_node_t**message1,	/* out: the messages passed with the aio
+				request; note that also in the case where
+				the aio operation failed, these output
+				parameters are valid and can be used to
+				restart the operation, for example */
+	void**	message2,
+	ulint*	type)		/* out: OS_FILE_WRITE or ..._READ */
+{
+	ulint		orig_seg	= segment;
+	os_aio_array_t*	array;
+	os_aio_slot_t*	slot;
+	ulint		n;
+	ulint		i;
+	ibool		ret_val;
+	BOOL		ret;
+	DWORD		len;
+
+	if (segment == ULINT_UNDEFINED) {
+		array = os_aio_sync_array;
+		segment = 0;
+	} else {
+		segment = os_aio_get_array_and_local_segment(&array, segment);
+	}
+
+	/* NOTE! We only access constant fields in os_aio_array. Therefore
+	we do not have to acquire the protecting mutex yet */
+
+	ut_ad(os_aio_validate());
+	ut_ad(segment < array->n_segments);
+
+	n = array->n_slots / array->n_segments;
+
+	if (array == os_aio_sync_array) {
+		os_event_wait(os_aio_array_get_nth_slot(array, pos)->event);
+		i = pos;
+	} else {
+		srv_set_io_thread_op_info(orig_seg, "wait Windows aio");
+		i = os_event_wait_multiple(n,
+					   (array->native_events)
+					   + segment * n);
+	}
+
+	os_mutex_enter(array->mutex);
+
+	slot = os_aio_array_get_nth_slot(array, i + segment * n);
+
+	ut_a(slot->reserved);
+
+	if (orig_seg != ULINT_UNDEFINED) {
+		srv_set_io_thread_op_info(orig_seg,
+					  "get windows aio return value");
+	}
+
+	ret = GetOverlappedResult(slot->file, &(slot->control), &len, TRUE);
+
+	*message1 = slot->message1;
+	*message2 = slot->message2;
+
+	*type = slot->type;
+
+	if (ret && len == slot->len) {
+		ret_val = TRUE;
+
+#ifdef UNIV_DO_FLUSH
+		if (slot->type == OS_FILE_WRITE
+		    && !os_do_not_call_flush_at_each_write) {
+			ut_a(TRUE == os_file_flush(slot->file));
+		}
+#endif /* UNIV_DO_FLUSH */
+	} else {
+		os_file_handle_error(slot->name, "Windows aio");
+
+		ret_val = FALSE;
+	}
+
+	os_mutex_exit(array->mutex);
+
+	os_aio_array_free_slot(array, slot);
+
+	return(ret_val);
+}
+#endif
+
+/**************************************************************************
+Does simulated aio. This function should be called by an i/o-handler
+thread. */
+UNIV_INTERN
+ibool
+os_aio_simulated_handle(
+/*====================*/
+				/* out: TRUE if the aio operation succeeded */
+	ulint	global_segment,	/* in: the number of the segment in the aio
+				arrays to wait for; segment 0 is the ibuf
+				i/o thread, segment 1 the log i/o thread,
+				then follow the non-ibuf read threads, and as
+				the last are the non-ibuf write threads */
+	fil_node_t**message1,	/* out: the messages passed with the aio
+				request; note that also in the case where
+				the aio operation failed, these output
+				parameters are valid and can be used to
+				restart the operation, for example */
+	void**	message2,
+	ulint*	type)		/* out: OS_FILE_WRITE or ..._READ */
+{
+	os_aio_array_t*	array;
+	ulint		segment;
+	os_aio_slot_t*	slot;
+	os_aio_slot_t*	slot2;
+	os_aio_slot_t*	consecutive_ios[OS_AIO_MERGE_N_CONSECUTIVE];
+	ulint		n_consecutive;
+	ulint		total_len;
+	ulint		offs;
+	ulint		lowest_offset;
+	ulint		biggest_age;
+	ulint		age;
+	byte*		combined_buf;
+	byte*		combined_buf2;
+	ibool		ret;
+	ulint		n;
+	ulint		i;
+
+	segment = os_aio_get_array_and_local_segment(&array, global_segment);
+
+restart:
+	/* NOTE! We only access constant fields in os_aio_array. Therefore
+	we do not have to acquire the protecting mutex yet */
+
+	srv_set_io_thread_op_info(global_segment,
+				  "looking for i/o requests (a)");
+	ut_ad(os_aio_validate());
+	ut_ad(segment < array->n_segments);
+
+	n = array->n_slots / array->n_segments;
+
+	/* Look through n slots after the segment * n'th slot */
+
+	if (array == os_aio_read_array
+	    && os_aio_recommend_sleep_for_read_threads) {
+
+		/* Give other threads chance to add several i/os to the array
+		at once. */
+
+		goto recommended_sleep;
+	}
+
+	os_mutex_enter(array->mutex);
+
+	srv_set_io_thread_op_info(global_segment,
+				  "looking for i/o requests (b)");
+
+	/* Check if there is a slot for which the i/o has already been
+	done */
+
+	for (i = 0; i < n; i++) {
+		slot = os_aio_array_get_nth_slot(array, i + segment * n);
+
+		if (slot->reserved && slot->io_already_done) {
+
+			if (os_aio_print_debug) {
+				fprintf(stderr,
+					"InnoDB: i/o for slot %lu"
+					" already done, returning\n",
+					(ulong) i);
+			}
+
+			ret = TRUE;
+
+			goto slot_io_done;
+		}
+	}
+
+	n_consecutive = 0;
+
+	/* If there are at least 2 seconds old requests, then pick the oldest
+	one to prevent starvation. If several requests have the same age,
+	then pick the one at the lowest offset. */
+
+	biggest_age = 0;
+	lowest_offset = ULINT_MAX;
+
+	for (i = 0; i < n; i++) {
+		slot = os_aio_array_get_nth_slot(array, i + segment * n);
+
+		if (slot->reserved) {
+			age = (ulint)difftime(time(NULL),
+					      slot->reservation_time);
+
+			if ((age >= 2 && age > biggest_age)
+			    || (age >= 2 && age == biggest_age
+				&& slot->offset < lowest_offset)) {
+
+				/* Found an i/o request */
+				consecutive_ios[0] = slot;
+
+				n_consecutive = 1;
+
+				biggest_age = age;
+				lowest_offset = slot->offset;
+			}
+		}
+	}
+
+	if (n_consecutive == 0) {
+		/* There were no old requests. Look for an i/o request at the
+		lowest offset in the array (we ignore the high 32 bits of the
+		offset in these heuristics) */
+
+		lowest_offset = ULINT_MAX;
+
+		for (i = 0; i < n; i++) {
+			slot = os_aio_array_get_nth_slot(array,
+							 i + segment * n);
+
+			if (slot->reserved && slot->offset < lowest_offset) {
+
+				/* Found an i/o request */
+				consecutive_ios[0] = slot;
+
+				n_consecutive = 1;
+
+				lowest_offset = slot->offset;
+			}
+		}
+	}
+
+	if (n_consecutive == 0) {
+
+		/* No i/o requested at the moment */
+
+		goto wait_for_io;
+	}
+
+	slot = consecutive_ios[0];
+
+	/* Check if there are several consecutive blocks to read or write */
+
+consecutive_loop:
+	for (i = 0; i < n; i++) {
+		slot2 = os_aio_array_get_nth_slot(array, i + segment * n);
+
+		if (slot2->reserved && slot2 != slot
+		    && slot2->offset == slot->offset + slot->len
+		    /* check that sum does not wrap over */
+		    && slot->offset + slot->len > slot->offset
+		    && slot2->offset_high == slot->offset_high
+		    && slot2->type == slot->type
+		    && slot2->file == slot->file) {
+
+			/* Found a consecutive i/o request */
+
+			consecutive_ios[n_consecutive] = slot2;
+			n_consecutive++;
+
+			slot = slot2;
+
+			if (n_consecutive < OS_AIO_MERGE_N_CONSECUTIVE) {
+
+				goto consecutive_loop;
+			} else {
+				break;
+			}
+		}
+	}
+
+	srv_set_io_thread_op_info(global_segment, "consecutive i/o requests");
+
+	/* We have now collected n_consecutive i/o requests in the array;
+	allocate a single buffer which can hold all data, and perform the
+	i/o */
+
+	total_len = 0;
+	slot = consecutive_ios[0];
+
+	for (i = 0; i < n_consecutive; i++) {
+		total_len += consecutive_ios[i]->len;
+	}
+
+	if (n_consecutive == 1) {
+		/* We can use the buffer of the i/o request */
+		combined_buf = slot->buf;
+		combined_buf2 = NULL;
+	} else {
+		combined_buf2 = ut_malloc(total_len + UNIV_PAGE_SIZE);
+
+		ut_a(combined_buf2);
+
+		combined_buf = ut_align(combined_buf2, UNIV_PAGE_SIZE);
+	}
+
+	/* We release the array mutex for the time of the i/o: NOTE that
+	this assumes that there is just one i/o-handler thread serving
+	a single segment of slots! */
+
+	os_mutex_exit(array->mutex);
+
+	if (slot->type == OS_FILE_WRITE && n_consecutive > 1) {
+		/* Copy the buffers to the combined buffer */
+		offs = 0;
+
+		for (i = 0; i < n_consecutive; i++) {
+
+			ut_memcpy(combined_buf + offs, consecutive_ios[i]->buf,
+				  consecutive_ios[i]->len);
+			offs += consecutive_ios[i]->len;
+		}
+	}
+
+	srv_set_io_thread_op_info(global_segment, "doing file i/o");
+
+	if (os_aio_print_debug) {
+		fprintf(stderr,
+			"InnoDB: doing i/o of type %lu at offset %lu %lu,"
+			" length %lu\n",
+			(ulong) slot->type, (ulong) slot->offset_high,
+			(ulong) slot->offset, (ulong) total_len);
+	}
+
+	/* Do the i/o with ordinary, synchronous i/o functions: */
+	if (slot->type == OS_FILE_WRITE) {
+		ret = os_file_write(slot->name, slot->file, combined_buf,
+				    slot->offset, slot->offset_high,
+				    total_len);
+	} else {
+		ret = os_file_read(slot->file, combined_buf,
+				   slot->offset, slot->offset_high, total_len);
+	}
+
+	ut_a(ret);
+	srv_set_io_thread_op_info(global_segment, "file i/o done");
+
+#if 0
+	fprintf(stderr,
+		"aio: %lu consecutive %lu:th segment, first offs %lu blocks\n",
+		n_consecutive, global_segment, slot->offset / UNIV_PAGE_SIZE);
+#endif
+
+	if (slot->type == OS_FILE_READ && n_consecutive > 1) {
+		/* Copy the combined buffer to individual buffers */
+		offs = 0;
+
+		for (i = 0; i < n_consecutive; i++) {
+
+			ut_memcpy(consecutive_ios[i]->buf, combined_buf + offs,
+				  consecutive_ios[i]->len);
+			offs += consecutive_ios[i]->len;
+		}
+	}
+
+	if (combined_buf2) {
+		ut_free(combined_buf2);
+	}
+
+	os_mutex_enter(array->mutex);
+
+	/* Mark the i/os done in slots */
+
+	for (i = 0; i < n_consecutive; i++) {
+		consecutive_ios[i]->io_already_done = TRUE;
+	}
+
+	/* We return the messages for the first slot now, and if there were
+	several slots, the messages will be returned with subsequent calls
+	of this function */
+
+slot_io_done:
+
+	ut_a(slot->reserved);
+
+	*message1 = slot->message1;
+	*message2 = slot->message2;
+
+	*type = slot->type;
+
+	os_mutex_exit(array->mutex);
+
+	os_aio_array_free_slot(array, slot);
+
+	return(ret);
+
+wait_for_io:
+	srv_set_io_thread_op_info(global_segment, "resetting wait event");
+
+	/* We wait here until there again can be i/os in the segment
+	of this thread */
+
+	os_event_reset(os_aio_segment_wait_events[global_segment]);
+
+	os_mutex_exit(array->mutex);
+
+recommended_sleep:
+	srv_set_io_thread_op_info(global_segment, "waiting for i/o request");
+
+	os_event_wait(os_aio_segment_wait_events[global_segment]);
+
+	if (os_aio_print_debug) {
+		fprintf(stderr,
+			"InnoDB: i/o handler thread for i/o"
+			" segment %lu wakes up\n",
+			(ulong) global_segment);
+	}
+
+	goto restart;
+}
+
+/**************************************************************************
+Validates the consistency of an aio array. */
+static
+ibool
+os_aio_array_validate(
+/*==================*/
+				/* out: TRUE if ok */
+	os_aio_array_t*	array)	/* in: aio wait array */
+{
+	os_aio_slot_t*	slot;
+	ulint		n_reserved	= 0;
+	ulint		i;
+
+	ut_a(array);
+
+	os_mutex_enter(array->mutex);
+
+	ut_a(array->n_slots > 0);
+	ut_a(array->n_segments > 0);
+
+	for (i = 0; i < array->n_slots; i++) {
+		slot = os_aio_array_get_nth_slot(array, i);
+
+		if (slot->reserved) {
+			n_reserved++;
+			ut_a(slot->len > 0);
+		}
+	}
+
+	ut_a(array->n_reserved == n_reserved);
+
+	os_mutex_exit(array->mutex);
+
+	return(TRUE);
+}
+
+/**************************************************************************
+Validates the consistency the aio system. */
+UNIV_INTERN
+ibool
+os_aio_validate(void)
+/*=================*/
+				/* out: TRUE if ok */
+{
+	os_aio_array_validate(os_aio_read_array);
+	os_aio_array_validate(os_aio_write_array);
+	os_aio_array_validate(os_aio_ibuf_array);
+	os_aio_array_validate(os_aio_log_array);
+	os_aio_array_validate(os_aio_sync_array);
+
+	return(TRUE);
+}
+
+/**************************************************************************
+Prints info of the aio arrays. */
+UNIV_INTERN
+void
+os_aio_print(
+/*=========*/
+	FILE*	file)	/* in: file where to print */
+{
+	os_aio_array_t*	array;
+	os_aio_slot_t*	slot;
+	ulint		n_reserved;
+	time_t		current_time;
+	double		time_elapsed;
+	double		avg_bytes_read;
+	ulint		i;
+
+	for (i = 0; i < srv_n_file_io_threads; i++) {
+		fprintf(file, "I/O thread %lu state: %s (%s)", (ulong) i,
+			srv_io_thread_op_info[i],
+			srv_io_thread_function[i]);
+
+#ifndef __WIN__
+		if (os_aio_segment_wait_events[i]->is_set) {
+			fprintf(file, " ev set");
+		}
+#endif
+
+		fprintf(file, "\n");
+	}
+
+	fputs("Pending normal aio reads:", file);
+
+	array = os_aio_read_array;
+loop:
+	ut_a(array);
+
+	os_mutex_enter(array->mutex);
+
+	ut_a(array->n_slots > 0);
+	ut_a(array->n_segments > 0);
+
+	n_reserved = 0;
+
+	for (i = 0; i < array->n_slots; i++) {
+		slot = os_aio_array_get_nth_slot(array, i);
+
+		if (slot->reserved) {
+			n_reserved++;
+#if 0
+			fprintf(stderr, "Reserved slot, messages %p %p\n",
+				(void*) slot->message1,
+				(void*) slot->message2);
+#endif
+			ut_a(slot->len > 0);
+		}
+	}
+
+	ut_a(array->n_reserved == n_reserved);
+
+	fprintf(file, " %lu", (ulong) n_reserved);
+
+	os_mutex_exit(array->mutex);
+
+	if (array == os_aio_read_array) {
+		fputs(", aio writes:", file);
+
+		array = os_aio_write_array;
+
+		goto loop;
+	}
+
+	if (array == os_aio_write_array) {
+		fputs(",\n ibuf aio reads:", file);
+		array = os_aio_ibuf_array;
+
+		goto loop;
+	}
+
+	if (array == os_aio_ibuf_array) {
+		fputs(", log i/o's:", file);
+		array = os_aio_log_array;
+
+		goto loop;
+	}
+
+	if (array == os_aio_log_array) {
+		fputs(", sync i/o's:", file);
+		array = os_aio_sync_array;
+
+		goto loop;
+	}
+
+	putc('\n', file);
+	current_time = time(NULL);
+	time_elapsed = 0.001 + difftime(current_time, os_last_printout);
+
+	fprintf(file,
+		"Pending flushes (fsync) log: %lu; buffer pool: %lu\n"
+		"%lu OS file reads, %lu OS file writes, %lu OS fsyncs\n",
+		(ulong) fil_n_pending_log_flushes,
+		(ulong) fil_n_pending_tablespace_flushes,
+		(ulong) os_n_file_reads, (ulong) os_n_file_writes,
+		(ulong) os_n_fsyncs);
+
+	if (os_file_n_pending_preads != 0 || os_file_n_pending_pwrites != 0) {
+		fprintf(file,
+			"%lu pending preads, %lu pending pwrites\n",
+			(ulong) os_file_n_pending_preads,
+			(ulong) os_file_n_pending_pwrites);
+	}
+
+	if (os_n_file_reads == os_n_file_reads_old) {
+		avg_bytes_read = 0.0;
+	} else {
+		avg_bytes_read = (double) os_bytes_read_since_printout
+			/ (os_n_file_reads - os_n_file_reads_old);
+	}
+
+	fprintf(file,
+		"%.2f reads/s, %lu avg bytes/read,"
+		" %.2f writes/s, %.2f fsyncs/s\n",
+		(os_n_file_reads - os_n_file_reads_old)
+		/ time_elapsed,
+		(ulong)avg_bytes_read,
+		(os_n_file_writes - os_n_file_writes_old)
+		/ time_elapsed,
+		(os_n_fsyncs - os_n_fsyncs_old)
+		/ time_elapsed);
+
+	os_n_file_reads_old = os_n_file_reads;
+	os_n_file_writes_old = os_n_file_writes;
+	os_n_fsyncs_old = os_n_fsyncs;
+	os_bytes_read_since_printout = 0;
+
+	os_last_printout = current_time;
+}
+
+/**************************************************************************
+Refreshes the statistics used to print per-second averages. */
+UNIV_INTERN
+void
+os_aio_refresh_stats(void)
+/*======================*/
+{
+	os_n_file_reads_old = os_n_file_reads;
+	os_n_file_writes_old = os_n_file_writes;
+	os_n_fsyncs_old = os_n_fsyncs;
+	os_bytes_read_since_printout = 0;
+
+	os_last_printout = time(NULL);
+}
+
+#ifdef UNIV_DEBUG
+/**************************************************************************
+Checks that all slots in the system have been freed, that is, there are
+no pending io operations. */
+UNIV_INTERN
+ibool
+os_aio_all_slots_free(void)
+/*=======================*/
+				/* out: TRUE if all free */
+{
+	os_aio_array_t*	array;
+	ulint		n_res	= 0;
+
+	array = os_aio_read_array;
+
+	os_mutex_enter(array->mutex);
+
+	n_res += array->n_reserved;
+
+	os_mutex_exit(array->mutex);
+
+	array = os_aio_write_array;
+
+	os_mutex_enter(array->mutex);
+
+	n_res += array->n_reserved;
+
+	os_mutex_exit(array->mutex);
+
+	array = os_aio_ibuf_array;
+
+	os_mutex_enter(array->mutex);
+
+	n_res += array->n_reserved;
+
+	os_mutex_exit(array->mutex);
+
+	array = os_aio_log_array;
+
+	os_mutex_enter(array->mutex);
+
+	n_res += array->n_reserved;
+
+	os_mutex_exit(array->mutex);
+
+	array = os_aio_sync_array;
+
+	os_mutex_enter(array->mutex);
+
+	n_res += array->n_reserved;
+
+	os_mutex_exit(array->mutex);
+
+	if (n_res == 0) {
+
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
+#endif /* UNIV_DEBUG */
diff --git a/storage/xtradb/os/os0proc.c b/storage/xtradb/os/os0proc.c
new file mode 100644
index 00000000000..8d4a71f8c4e
--- /dev/null
+++ b/storage/xtradb/os/os0proc.c
@@ -0,0 +1,262 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The interface to the operating system
+process control primitives
+
+Created 9/30/1995 Heikki Tuuri
+*******************************************************/
+
+#include "os0proc.h"
+#ifdef UNIV_NONINL
+#include "os0proc.ic"
+#endif
+
+#include "ut0mem.h"
+#include "ut0byte.h"
+
+/* FreeBSD for example has only MAP_ANON, Linux has MAP_ANONYMOUS and
+MAP_ANON but MAP_ANON is marked as deprecated */
+#if defined(MAP_ANONYMOUS)
+#define OS_MAP_ANON	MAP_ANONYMOUS
+#elif defined(MAP_ANON)
+#define OS_MAP_ANON	MAP_ANON
+#endif
+
+UNIV_INTERN ibool os_use_large_pages;
+/* Large page size. This may be a boot-time option on some platforms */
+UNIV_INTERN ulint os_large_page_size;
+
+/********************************************************************
+Converts the current process id to a number. It is not guaranteed that the
+number is unique. In Linux returns the 'process number' of the current
+thread. That number is the same as one sees in 'top', for example. In Linux
+the thread id is not the same as one sees in 'top'. */
+UNIV_INTERN
+ulint
+os_proc_get_number(void)
+/*====================*/
+{
+#ifdef __WIN__
+	return((ulint)GetCurrentProcessId());
+#else
+	return((ulint)getpid());
+#endif
+}
+
+/********************************************************************
+Allocates large pages memory. */
+UNIV_INTERN
+void*
+os_mem_alloc_large(
+/*===============*/
+					/* out: allocated memory */
+	ulint*	n)			/* in/out: number of bytes */
+{
+	void*	ptr;
+	ulint	size;
+#if defined HAVE_LARGE_PAGES && defined UNIV_LINUX
+	int shmid;
+	struct shmid_ds buf;
+
+	if (!os_use_large_pages || !os_large_page_size) {
+		goto skip;
+	}
+
+	/* Align block size to os_large_page_size */
+	ut_ad(ut_is_2pow(os_large_page_size));
+	size = ut_2pow_round(*n + (os_large_page_size - 1),
+			     os_large_page_size);
+
+	shmid = shmget(IPC_PRIVATE, (size_t)size, SHM_HUGETLB | SHM_R | SHM_W);
+	if (shmid < 0) {
+		fprintf(stderr, "InnoDB: HugeTLB: Warning: Failed to allocate"
+			" %lu bytes. errno %d\n", size, errno);
+		ptr = NULL;
+	} else {
+		ptr = shmat(shmid, NULL, 0);
+		if (ptr == (void *)-1) {
+			fprintf(stderr, "InnoDB: HugeTLB: Warning: Failed to"
+				" attach shared memory segment, errno %d\n",
+				errno);
+		}
+
+		/* Remove the shared memory segment so that it will be
+		automatically freed after memory is detached or
+		process exits */
+		shmctl(shmid, IPC_RMID, &buf);
+	}
+
+	if (ptr) {
+		*n = size;
+		os_fast_mutex_lock(&ut_list_mutex);
+		ut_total_allocated_memory += size;
+		os_fast_mutex_unlock(&ut_list_mutex);
+# ifdef UNIV_SET_MEM_TO_ZERO
+		memset(ptr, '\0', size);
+# endif
+		UNIV_MEM_ALLOC(ptr, size);
+		return(ptr);
+	}
+
+	fprintf(stderr, "InnoDB HugeTLB: Warning: Using conventional"
+		" memory pool\n");
+skip:
+#endif /* HAVE_LARGE_PAGES && UNIV_LINUX */
+
+#ifdef __WIN__
+	SYSTEM_INFO	system_info;
+	GetSystemInfo(&system_info);
+
+	/* Align block size to system page size */
+	ut_ad(ut_is_2pow(system_info.dwPageSize));
+	/* system_info.dwPageSize is only 32-bit. Casting to ulint is required
+	on 64-bit Windows. */
+	size = *n = ut_2pow_round(*n + (system_info.dwPageSize - 1),
+				  (ulint) system_info.dwPageSize);
+	ptr = VirtualAlloc(NULL, size, MEM_COMMIT | MEM_RESERVE,
+			   PAGE_READWRITE);
+	if (!ptr) {
+		fprintf(stderr, "InnoDB: VirtualAlloc(%lu bytes) failed;"
+			" Windows error %lu\n",
+			(ulong) size, (ulong) GetLastError());
+	} else {
+		os_fast_mutex_lock(&ut_list_mutex);
+		ut_total_allocated_memory += size;
+		os_fast_mutex_unlock(&ut_list_mutex);
+		UNIV_MEM_ALLOC(ptr, size);
+	}
+#elif defined __NETWARE__ || !defined OS_MAP_ANON
+	size = *n;
+	ptr = ut_malloc_low(size, TRUE, FALSE);
+#else
+# ifdef HAVE_GETPAGESIZE
+	size = getpagesize();
+# else
+	size = UNIV_PAGE_SIZE;
+# endif
+	/* Align block size to system page size */
+	ut_ad(ut_is_2pow(size));
+	size = *n = ut_2pow_round(*n + (size - 1), size);
+	ptr = mmap(NULL, size, PROT_READ | PROT_WRITE,
+		   MAP_PRIVATE | OS_MAP_ANON, -1, 0);
+	if (UNIV_UNLIKELY(ptr == (void*) -1)) {
+		fprintf(stderr, "InnoDB: mmap(%lu bytes) failed;"
+			" errno %lu\n",
+			(ulong) size, (ulong) errno);
+		ptr = NULL;
+	} else {
+		os_fast_mutex_lock(&ut_list_mutex);
+		ut_total_allocated_memory += size;
+		os_fast_mutex_unlock(&ut_list_mutex);
+		UNIV_MEM_ALLOC(ptr, size);
+	}
+#endif
+	return(ptr);
+}
+
+/********************************************************************
+Frees large pages memory. */
+UNIV_INTERN
+void
+os_mem_free_large(
+/*==============*/
+	void	*ptr,			/* in: pointer returned by
+					os_mem_alloc_large() */
+	ulint	size)			/* in: size returned by
+					os_mem_alloc_large() */
+{
+	os_fast_mutex_lock(&ut_list_mutex);
+	ut_a(ut_total_allocated_memory >= size);
+	os_fast_mutex_unlock(&ut_list_mutex);
+
+#if defined HAVE_LARGE_PAGES && defined UNIV_LINUX
+	if (os_use_large_pages && os_large_page_size && !shmdt(ptr)) {
+		os_fast_mutex_lock(&ut_list_mutex);
+		ut_a(ut_total_allocated_memory >= size);
+		ut_total_allocated_memory -= size;
+		os_fast_mutex_unlock(&ut_list_mutex);
+		UNIV_MEM_FREE(ptr, size);
+		return;
+	}
+#endif /* HAVE_LARGE_PAGES && UNIV_LINUX */
+#ifdef __WIN__
+	/* When RELEASE memory, the size parameter must be 0.
+	Do not use MEM_RELEASE with MEM_DECOMMIT. */
+	if (!VirtualFree(ptr, 0, MEM_RELEASE)) {
+		fprintf(stderr, "InnoDB: VirtualFree(%p, %lu) failed;"
+			" Windows error %lu\n",
+			ptr, (ulong) size, (ulong) GetLastError());
+	} else {
+		os_fast_mutex_lock(&ut_list_mutex);
+		ut_a(ut_total_allocated_memory >= size);
+		ut_total_allocated_memory -= size;
+		os_fast_mutex_unlock(&ut_list_mutex);
+		UNIV_MEM_FREE(ptr, size);
+	}
+#elif defined __NETWARE__ || !defined OS_MAP_ANON
+	ut_free(ptr);
+#else
+	if (munmap(ptr, size)) {
+		fprintf(stderr, "InnoDB: munmap(%p, %lu) failed;"
+			" errno %lu\n",
+			ptr, (ulong) size, (ulong) errno);
+	} else {
+		os_fast_mutex_lock(&ut_list_mutex);
+		ut_a(ut_total_allocated_memory >= size);
+		ut_total_allocated_memory -= size;
+		os_fast_mutex_unlock(&ut_list_mutex);
+		UNIV_MEM_FREE(ptr, size);
+	}
+#endif
+}
+
+/********************************************************************
+Sets the priority boost for threads released from waiting within the current
+process. */
+UNIV_INTERN
+void
+os_process_set_priority_boost(
+/*==========================*/
+	ibool	do_boost)	/* in: TRUE if priority boost should be done,
+				FALSE if not */
+{
+#ifdef __WIN__
+	ibool	no_boost;
+
+	if (do_boost) {
+		no_boost = FALSE;
+	} else {
+		no_boost = TRUE;
+	}
+
+#if TRUE != 1
+# error "TRUE != 1"
+#endif
+
+	/* Does not do anything currently!
+	SetProcessPriorityBoost(GetCurrentProcess(), no_boost);
+	*/
+	fputs("Warning: process priority boost setting"
+	      " currently not functional!\n",
+	      stderr);
+#else
+	UT_NOT_USED(do_boost);
+#endif
+}
diff --git a/storage/xtradb/os/os0sync.c b/storage/xtradb/os/os0sync.c
new file mode 100644
index 00000000000..78ff74059f8
--- /dev/null
+++ b/storage/xtradb/os/os0sync.c
@@ -0,0 +1,769 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The interface to the operating system
+synchronization primitives.
+
+Created 9/6/1995 Heikki Tuuri
+*******************************************************/
+
+#include "os0sync.h"
+#ifdef UNIV_NONINL
+#include "os0sync.ic"
+#endif
+
+#ifdef __WIN__
+#include <windows.h>
+#endif
+
+#include "ut0mem.h"
+#include "srv0start.h"
+
+/* Type definition for an operating system mutex struct */
+struct os_mutex_struct{
+	os_event_t	event;	/* Used by sync0arr.c for queing threads */
+	void*		handle;	/* OS handle to mutex */
+	ulint		count;	/* we use this counter to check
+				that the same thread does not
+				recursively lock the mutex: we
+				do not assume that the OS mutex
+				supports recursive locking, though
+				NT seems to do that */
+	UT_LIST_NODE_T(os_mutex_str_t) os_mutex_list;
+				/* list of all 'slow' OS mutexes created */
+};
+
+/* Mutex protecting counts and the lists of OS mutexes and events */
+UNIV_INTERN os_mutex_t	os_sync_mutex;
+static ibool		os_sync_mutex_inited	= FALSE;
+static ibool		os_sync_free_called	= FALSE;
+
+/* This is incremented by 1 in os_thread_create and decremented by 1 in
+os_thread_exit */
+UNIV_INTERN ulint	os_thread_count		= 0;
+
+/* The list of all events created */
+static UT_LIST_BASE_NODE_T(os_event_struct_t)	os_event_list;
+
+/* The list of all OS 'slow' mutexes */
+static UT_LIST_BASE_NODE_T(os_mutex_str_t)	os_mutex_list;
+
+UNIV_INTERN ulint	os_event_count		= 0;
+UNIV_INTERN ulint	os_mutex_count		= 0;
+UNIV_INTERN ulint	os_fast_mutex_count	= 0;
+
+/* Because a mutex is embedded inside an event and there is an
+event embedded inside a mutex, on free, this generates a recursive call.
+This version of the free event function doesn't acquire the global lock */
+static void os_event_free_internal(os_event_t	event);
+
+/*************************************************************
+Initializes global event and OS 'slow' mutex lists. */
+UNIV_INTERN
+void
+os_sync_init(void)
+/*==============*/
+{
+	UT_LIST_INIT(os_event_list);
+	UT_LIST_INIT(os_mutex_list);
+
+	os_sync_mutex = os_mutex_create(NULL);
+
+	os_sync_mutex_inited = TRUE;
+}
+
+/*************************************************************
+Frees created events and OS 'slow' mutexes. */
+UNIV_INTERN
+void
+os_sync_free(void)
+/*==============*/
+{
+	os_event_t	event;
+	os_mutex_t	mutex;
+
+	os_sync_free_called = TRUE;
+	event = UT_LIST_GET_FIRST(os_event_list);
+
+	while (event) {
+
+		os_event_free(event);
+
+		event = UT_LIST_GET_FIRST(os_event_list);
+	}
+
+	mutex = UT_LIST_GET_FIRST(os_mutex_list);
+
+	while (mutex) {
+		if (mutex == os_sync_mutex) {
+			/* Set the flag to FALSE so that we do not try to
+			reserve os_sync_mutex any more in remaining freeing
+			operations in shutdown */
+			os_sync_mutex_inited = FALSE;
+		}
+
+		os_mutex_free(mutex);
+
+		mutex = UT_LIST_GET_FIRST(os_mutex_list);
+	}
+	os_sync_free_called = FALSE;
+}
+
+/*************************************************************
+Creates an event semaphore, i.e., a semaphore which may just have two
+states: signaled and nonsignaled. The created event is manual reset: it
+must be reset explicitly by calling sync_os_reset_event. */
+UNIV_INTERN
+os_event_t
+os_event_create(
+/*============*/
+				/* out: the event handle */
+	const char*	name)	/* in: the name of the event, if NULL
+				the event is created without a name */
+{
+#ifdef __WIN__
+	os_event_t event;
+
+	event = ut_malloc(sizeof(struct os_event_struct));
+
+	event->handle = CreateEvent(NULL, /* No security attributes */
+				    TRUE, /* Manual reset */
+				    FALSE, /* Initial state nonsignaled */
+				    (LPCTSTR) name);
+	if (!event->handle) {
+		fprintf(stderr,
+			"InnoDB: Could not create a Windows event semaphore;"
+			" Windows error %lu\n",
+			(ulong) GetLastError());
+	}
+#else /* Unix */
+	os_event_t	event;
+
+	UT_NOT_USED(name);
+
+	event = ut_malloc(sizeof(struct os_event_struct));
+
+	os_fast_mutex_init(&(event->os_mutex));
+
+#if defined(UNIV_HOTBACKUP) && defined(UNIV_HPUX10)
+	ut_a(0 == pthread_cond_init(&(event->cond_var),
+				    pthread_condattr_default));
+#else
+	ut_a(0 == pthread_cond_init(&(event->cond_var), NULL));
+#endif
+	event->is_set = FALSE;
+
+	/* We return this value in os_event_reset(), which can then be
+	be used to pass to the os_event_wait_low(). The value of zero
+	is reserved in os_event_wait_low() for the case when the
+	caller does not want to pass any signal_count value. To
+	distinguish between the two cases we initialize signal_count
+	to 1 here. */
+	event->signal_count = 1;
+#endif /* __WIN__ */
+
+	/* The os_sync_mutex can be NULL because during startup an event
+	can be created [ because it's embedded in the mutex/rwlock ] before
+	this module has been initialized */
+	if (os_sync_mutex != NULL) {
+		os_mutex_enter(os_sync_mutex);
+	}
+
+	/* Put to the list of events */
+	UT_LIST_ADD_FIRST(os_event_list, os_event_list, event);
+
+	os_event_count++;
+
+	if (os_sync_mutex != NULL) {
+		os_mutex_exit(os_sync_mutex);
+	}
+
+	return(event);
+}
+
+#ifdef __WIN__
+/*************************************************************
+Creates an auto-reset event semaphore, i.e., an event which is automatically
+reset when a single thread is released. Works only in Windows. */
+UNIV_INTERN
+os_event_t
+os_event_create_auto(
+/*=================*/
+				/* out: the event handle */
+	const char*	name)	/* in: the name of the event, if NULL
+				the event is created without a name */
+{
+	os_event_t event;
+
+	event = ut_malloc(sizeof(struct os_event_struct));
+
+	event->handle = CreateEvent(NULL, /* No security attributes */
+				    FALSE, /* Auto-reset */
+				    FALSE, /* Initial state nonsignaled */
+				    (LPCTSTR) name);
+
+	if (!event->handle) {
+		fprintf(stderr,
+			"InnoDB: Could not create a Windows auto"
+			" event semaphore; Windows error %lu\n",
+			(ulong) GetLastError());
+	}
+
+	/* Put to the list of events */
+	os_mutex_enter(os_sync_mutex);
+
+	UT_LIST_ADD_FIRST(os_event_list, os_event_list, event);
+
+	os_event_count++;
+
+	os_mutex_exit(os_sync_mutex);
+
+	return(event);
+}
+#endif
+
+/**************************************************************
+Sets an event semaphore to the signaled state: lets waiting threads
+proceed. */
+UNIV_INTERN
+void
+os_event_set(
+/*=========*/
+	os_event_t	event)	/* in: event to set */
+{
+#ifdef __WIN__
+	ut_a(event);
+	ut_a(SetEvent(event->handle));
+#else
+	ut_a(event);
+
+	os_fast_mutex_lock(&(event->os_mutex));
+
+	if (event->is_set) {
+		/* Do nothing */
+	} else {
+		event->is_set = TRUE;
+		event->signal_count += 1;
+		ut_a(0 == pthread_cond_broadcast(&(event->cond_var)));
+	}
+
+	os_fast_mutex_unlock(&(event->os_mutex));
+#endif
+}
+
+/**************************************************************
+Resets an event semaphore to the nonsignaled state. Waiting threads will
+stop to wait for the event.
+The return value should be passed to os_even_wait_low() if it is desired
+that this thread should not wait in case of an intervening call to
+os_event_set() between this os_event_reset() and the
+os_event_wait_low() call. See comments for os_event_wait_low(). */
+UNIV_INTERN
+ib_int64_t
+os_event_reset(
+/*===========*/
+				/* out: current signal_count. */
+	os_event_t	event)	/* in: event to reset */
+{
+	ib_int64_t	ret = 0;
+
+#ifdef __WIN__
+	ut_a(event);
+
+	ut_a(ResetEvent(event->handle));
+#else
+	ut_a(event);
+
+	os_fast_mutex_lock(&(event->os_mutex));
+
+	if (!event->is_set) {
+		/* Do nothing */
+	} else {
+		event->is_set = FALSE;
+	}
+	ret = event->signal_count;
+
+	os_fast_mutex_unlock(&(event->os_mutex));
+#endif
+	return(ret);
+}
+
+/**************************************************************
+Frees an event object, without acquiring the global lock. */
+static
+void
+os_event_free_internal(
+/*===================*/
+	os_event_t	event)	/* in: event to free */
+{
+#ifdef __WIN__
+	ut_a(event);
+
+	ut_a(CloseHandle(event->handle));
+#else
+	ut_a(event);
+
+	/* This is to avoid freeing the mutex twice */
+	os_fast_mutex_free(&(event->os_mutex));
+
+	ut_a(0 == pthread_cond_destroy(&(event->cond_var)));
+#endif
+	/* Remove from the list of events */
+
+	UT_LIST_REMOVE(os_event_list, os_event_list, event);
+
+	os_event_count--;
+
+	ut_free(event);
+}
+
+/**************************************************************
+Frees an event object. */
+UNIV_INTERN
+void
+os_event_free(
+/*==========*/
+	os_event_t	event)	/* in: event to free */
+
+{
+#ifdef __WIN__
+	ut_a(event);
+
+	ut_a(CloseHandle(event->handle));
+#else
+	ut_a(event);
+
+	os_fast_mutex_free(&(event->os_mutex));
+	ut_a(0 == pthread_cond_destroy(&(event->cond_var)));
+#endif
+	/* Remove from the list of events */
+
+	os_mutex_enter(os_sync_mutex);
+
+	UT_LIST_REMOVE(os_event_list, os_event_list, event);
+
+	os_event_count--;
+
+	os_mutex_exit(os_sync_mutex);
+
+	ut_free(event);
+}
+
+/**************************************************************
+Waits for an event object until it is in the signaled state. If
+srv_shutdown_state == SRV_SHUTDOWN_EXIT_THREADS this also exits the
+waiting thread when the event becomes signaled (or immediately if the
+event is already in the signaled state).
+
+Typically, if the event has been signalled after the os_event_reset()
+we'll return immediately because event->is_set == TRUE.
+There are, however, situations (e.g.: sync_array code) where we may
+lose this information. For example:
+
+thread A calls os_event_reset()
+thread B calls os_event_set()   [event->is_set == TRUE]
+thread C calls os_event_reset() [event->is_set == FALSE]
+thread A calls os_event_wait()  [infinite wait!]
+thread C calls os_event_wait()  [infinite wait!]
+
+Where such a scenario is possible, to avoid infinite wait, the
+value returned by os_event_reset() should be passed in as
+reset_sig_count. */
+UNIV_INTERN
+void
+os_event_wait_low(
+/*==============*/
+	os_event_t	event,		/* in: event to wait */
+	ib_int64_t	reset_sig_count)/* in: zero or the value
+					returned by previous call of
+					os_event_reset(). */
+{
+#ifdef __WIN__
+	DWORD	err;
+
+	ut_a(event);
+
+	UT_NOT_USED(reset_sig_count);
+
+	/* Specify an infinite time limit for waiting */
+	err = WaitForSingleObject(event->handle, INFINITE);
+
+	ut_a(err == WAIT_OBJECT_0);
+
+	if (srv_shutdown_state == SRV_SHUTDOWN_EXIT_THREADS) {
+		os_thread_exit(NULL);
+	}
+#else
+	ib_int64_t	old_signal_count;
+
+	os_fast_mutex_lock(&(event->os_mutex));
+
+	if (reset_sig_count) {
+		old_signal_count = reset_sig_count;
+	} else {
+		old_signal_count = event->signal_count;
+	}
+
+	for (;;) {
+		if (event->is_set == TRUE
+		    || event->signal_count != old_signal_count) {
+
+			os_fast_mutex_unlock(&(event->os_mutex));
+
+			if (srv_shutdown_state == SRV_SHUTDOWN_EXIT_THREADS) {
+
+				os_thread_exit(NULL);
+			}
+			/* Ok, we may return */
+
+			return;
+		}
+
+		pthread_cond_wait(&(event->cond_var), &(event->os_mutex));
+
+		/* Solaris manual said that spurious wakeups may occur: we
+		have to check if the event really has been signaled after
+		we came here to wait */
+	}
+#endif
+}
+
+/**************************************************************
+Waits for an event object until it is in the signaled state or
+a timeout is exceeded. In Unix the timeout is always infinite. */
+UNIV_INTERN
+ulint
+os_event_wait_time(
+/*===============*/
+				/* out: 0 if success, OS_SYNC_TIME_EXCEEDED if
+				timeout was exceeded */
+	os_event_t	event,	/* in: event to wait */
+	ulint		time)	/* in: timeout in microseconds, or
+				OS_SYNC_INFINITE_TIME */
+{
+#ifdef __WIN__
+	DWORD	err;
+
+	ut_a(event);
+
+	if (time != OS_SYNC_INFINITE_TIME) {
+		err = WaitForSingleObject(event->handle, (DWORD) time / 1000);
+	} else {
+		err = WaitForSingleObject(event->handle, INFINITE);
+	}
+
+	if (err == WAIT_OBJECT_0) {
+
+		return(0);
+	} else if (err == WAIT_TIMEOUT) {
+
+		return(OS_SYNC_TIME_EXCEEDED);
+	} else {
+		ut_error;
+		return(1000000); /* dummy value to eliminate compiler warn. */
+	}
+#else
+	UT_NOT_USED(time);
+
+	/* In Posix this is just an ordinary, infinite wait */
+
+	os_event_wait(event);
+
+	return(0);
+#endif
+}
+
+#ifdef __WIN__
+/**************************************************************
+Waits for any event in an OS native event array. Returns if even a single
+one is signaled or becomes signaled. */
+UNIV_INTERN
+ulint
+os_event_wait_multiple(
+/*===================*/
+					/* out: index of the event
+					which was signaled */
+	ulint			n,	/* in: number of events in the
+					array */
+	os_native_event_t*	native_event_array)
+					/* in: pointer to an array of event
+					handles */
+{
+	DWORD	index;
+
+	ut_a(native_event_array);
+	ut_a(n > 0);
+
+	index = WaitForMultipleObjects((DWORD) n, native_event_array,
+				       FALSE,	   /* Wait for any 1 event */
+				       INFINITE); /* Infinite wait time
+						  limit */
+	ut_a(index >= WAIT_OBJECT_0);	/* NOTE: Pointless comparision */
+	ut_a(index < WAIT_OBJECT_0 + n);
+
+	if (srv_shutdown_state == SRV_SHUTDOWN_EXIT_THREADS) {
+		os_thread_exit(NULL);
+	}
+
+	return(index - WAIT_OBJECT_0);
+}
+#endif
+
+/*************************************************************
+Creates an operating system mutex semaphore. Because these are slow, the
+mutex semaphore of InnoDB itself (mutex_t) should be used where possible. */
+UNIV_INTERN
+os_mutex_t
+os_mutex_create(
+/*============*/
+				/* out: the mutex handle */
+	const char*	name)	/* in: the name of the mutex, if NULL
+				the mutex is created without a name */
+{
+#ifdef __WIN__
+	HANDLE		mutex;
+	os_mutex_t	mutex_str;
+
+	mutex = CreateMutex(NULL,	/* No security attributes */
+			    FALSE,		/* Initial state: no owner */
+			    (LPCTSTR) name);
+	ut_a(mutex);
+#else
+	os_fast_mutex_t*	mutex;
+	os_mutex_t		mutex_str;
+
+	UT_NOT_USED(name);
+
+	mutex = ut_malloc(sizeof(os_fast_mutex_t));
+
+	os_fast_mutex_init(mutex);
+#endif
+	mutex_str = ut_malloc(sizeof(os_mutex_str_t));
+
+	mutex_str->handle = mutex;
+	mutex_str->count = 0;
+	mutex_str->event = os_event_create(NULL);
+
+	if (UNIV_LIKELY(os_sync_mutex_inited)) {
+		/* When creating os_sync_mutex itself we cannot reserve it */
+		os_mutex_enter(os_sync_mutex);
+	}
+
+	UT_LIST_ADD_FIRST(os_mutex_list, os_mutex_list, mutex_str);
+
+	os_mutex_count++;
+
+	if (UNIV_LIKELY(os_sync_mutex_inited)) {
+		os_mutex_exit(os_sync_mutex);
+	}
+
+	return(mutex_str);
+}
+
+/**************************************************************
+Acquires ownership of a mutex semaphore. */
+UNIV_INTERN
+void
+os_mutex_enter(
+/*===========*/
+	os_mutex_t	mutex)	/* in: mutex to acquire */
+{
+#ifdef __WIN__
+	DWORD	err;
+
+	ut_a(mutex);
+
+	/* Specify infinite time limit for waiting */
+	err = WaitForSingleObject(mutex->handle, INFINITE);
+
+	ut_a(err == WAIT_OBJECT_0);
+
+	(mutex->count)++;
+	ut_a(mutex->count == 1);
+#else
+	os_fast_mutex_lock(mutex->handle);
+
+	(mutex->count)++;
+
+	ut_a(mutex->count == 1);
+#endif
+}
+
+/**************************************************************
+Releases ownership of a mutex. */
+UNIV_INTERN
+void
+os_mutex_exit(
+/*==========*/
+	os_mutex_t	mutex)	/* in: mutex to release */
+{
+	ut_a(mutex);
+
+	ut_a(mutex->count == 1);
+
+	(mutex->count)--;
+#ifdef __WIN__
+	ut_a(ReleaseMutex(mutex->handle));
+#else
+	os_fast_mutex_unlock(mutex->handle);
+#endif
+}
+
+/**************************************************************
+Frees a mutex object. */
+UNIV_INTERN
+void
+os_mutex_free(
+/*==========*/
+	os_mutex_t	mutex)	/* in: mutex to free */
+{
+	ut_a(mutex);
+
+	if (UNIV_LIKELY(!os_sync_free_called)) {
+		os_event_free_internal(mutex->event);
+	}
+
+	if (UNIV_LIKELY(os_sync_mutex_inited)) {
+		os_mutex_enter(os_sync_mutex);
+	}
+
+	UT_LIST_REMOVE(os_mutex_list, os_mutex_list, mutex);
+
+	os_mutex_count--;
+
+	if (UNIV_LIKELY(os_sync_mutex_inited)) {
+		os_mutex_exit(os_sync_mutex);
+	}
+
+#ifdef __WIN__
+	ut_a(CloseHandle(mutex->handle));
+
+	ut_free(mutex);
+#else
+	os_fast_mutex_free(mutex->handle);
+	ut_free(mutex->handle);
+	ut_free(mutex);
+#endif
+}
+
+/*************************************************************
+Initializes an operating system fast mutex semaphore. */
+UNIV_INTERN
+void
+os_fast_mutex_init(
+/*===============*/
+	os_fast_mutex_t*	fast_mutex)	/* in: fast mutex */
+{
+#ifdef __WIN__
+	ut_a(fast_mutex);
+
+	InitializeCriticalSection((LPCRITICAL_SECTION) fast_mutex);
+#else
+#if defined(UNIV_HOTBACKUP) && defined(UNIV_HPUX10)
+	ut_a(0 == pthread_mutex_init(fast_mutex, pthread_mutexattr_default));
+#else
+	ut_a(0 == pthread_mutex_init(fast_mutex, MY_MUTEX_INIT_FAST));
+#endif
+#endif
+	if (UNIV_LIKELY(os_sync_mutex_inited)) {
+		/* When creating os_sync_mutex itself (in Unix) we cannot
+		reserve it */
+
+		os_mutex_enter(os_sync_mutex);
+	}
+
+	os_fast_mutex_count++;
+
+	if (UNIV_LIKELY(os_sync_mutex_inited)) {
+		os_mutex_exit(os_sync_mutex);
+	}
+}
+
+/**************************************************************
+Acquires ownership of a fast mutex. */
+UNIV_INTERN
+void
+os_fast_mutex_lock(
+/*===============*/
+	os_fast_mutex_t*	fast_mutex)	/* in: mutex to acquire */
+{
+#ifdef __WIN__
+	EnterCriticalSection((LPCRITICAL_SECTION) fast_mutex);
+#else
+	pthread_mutex_lock(fast_mutex);
+#endif
+}
+
+/**************************************************************
+Releases ownership of a fast mutex. */
+UNIV_INTERN
+void
+os_fast_mutex_unlock(
+/*=================*/
+	os_fast_mutex_t*	fast_mutex)	/* in: mutex to release */
+{
+#ifdef __WIN__
+	LeaveCriticalSection(fast_mutex);
+#else
+	pthread_mutex_unlock(fast_mutex);
+#endif
+}
+
+/**************************************************************
+Frees a mutex object. */
+UNIV_INTERN
+void
+os_fast_mutex_free(
+/*===============*/
+	os_fast_mutex_t*	fast_mutex)	/* in: mutex to free */
+{
+#ifdef __WIN__
+	ut_a(fast_mutex);
+
+	DeleteCriticalSection((LPCRITICAL_SECTION) fast_mutex);
+#else
+	int	ret;
+
+	ret = pthread_mutex_destroy(fast_mutex);
+
+	if (UNIV_UNLIKELY(ret != 0)) {
+		ut_print_timestamp(stderr);
+		fprintf(stderr,
+			"  InnoDB: error: return value %lu when calling\n"
+			"InnoDB: pthread_mutex_destroy().\n", (ulint)ret);
+		fprintf(stderr,
+			"InnoDB: Byte contents of the pthread mutex at %p:\n",
+			(void*) fast_mutex);
+		ut_print_buf(stderr, fast_mutex, sizeof(os_fast_mutex_t));
+		putc('\n', stderr);
+	}
+#endif
+	if (UNIV_LIKELY(os_sync_mutex_inited)) {
+		/* When freeing the last mutexes, we have
+		already freed os_sync_mutex */
+
+		os_mutex_enter(os_sync_mutex);
+	}
+
+	os_fast_mutex_count--;
+
+	if (UNIV_LIKELY(os_sync_mutex_inited)) {
+		os_mutex_exit(os_sync_mutex);
+	}
+}
diff --git a/storage/xtradb/os/os0thread.c b/storage/xtradb/os/os0thread.c
new file mode 100644
index 00000000000..7d0a57ae17c
--- /dev/null
+++ b/storage/xtradb/os/os0thread.c
@@ -0,0 +1,365 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The interface to the operating system thread control primitives
+
+Created 9/8/1995 Heikki Tuuri
+*******************************************************/
+
+#include "os0thread.h"
+#ifdef UNIV_NONINL
+#include "os0thread.ic"
+#endif
+
+#ifdef __WIN__
+#include <windows.h>
+#endif
+
+#include "srv0srv.h"
+#include "os0sync.h"
+
+/*******************************************************************
+Compares two thread ids for equality. */
+UNIV_INTERN
+ibool
+os_thread_eq(
+/*=========*/
+				/* out: TRUE if equal */
+	os_thread_id_t	a,	/* in: OS thread or thread id */
+	os_thread_id_t	b)	/* in: OS thread or thread id */
+{
+#ifdef __WIN__
+	if (a == b) {
+		return(TRUE);
+	}
+
+	return(FALSE);
+#else
+	if (pthread_equal(a, b)) {
+		return(TRUE);
+	}
+
+	return(FALSE);
+#endif
+}
+
+/********************************************************************
+Converts an OS thread id to a ulint. It is NOT guaranteed that the ulint is
+unique for the thread though! */
+UNIV_INTERN
+ulint
+os_thread_pf(
+/*=========*/
+	os_thread_id_t	a)
+{
+#ifdef UNIV_HPUX10
+	/* In HP-UX-10.20 a pthread_t is a struct of 3 fields: field1, field2,
+	field3. We do not know if field1 determines the thread uniquely. */
+
+	return((ulint)(a.field1));
+#else
+	return((ulint)a);
+#endif
+}
+
+/*********************************************************************
+Returns the thread identifier of current thread. Currently the thread
+identifier in Unix is the thread handle itself. Note that in HP-UX
+pthread_t is a struct of 3 fields. */
+UNIV_INTERN
+os_thread_id_t
+os_thread_get_curr_id(void)
+/*=======================*/
+{
+#ifdef __WIN__
+	return(GetCurrentThreadId());
+#else
+	return(pthread_self());
+#endif
+}
+
+/********************************************************************
+Creates a new thread of execution. The execution starts from
+the function given. The start function takes a void* parameter
+and returns an ulint. */
+UNIV_INTERN
+os_thread_t
+os_thread_create(
+/*=============*/
+						/* out: handle to the thread */
+#ifndef __WIN__
+	os_posix_f_t		start_f,
+#else
+	ulint (*start_f)(void*),		/* in: pointer to function
+						from which to start */
+#endif
+	void*			arg,		/* in: argument to start
+						function */
+	os_thread_id_t*		thread_id)	/* out: id of the created
+						thread, or NULL */
+{
+#ifdef __WIN__
+	os_thread_t	thread;
+	DWORD		win_thread_id;
+
+	os_mutex_enter(os_sync_mutex);
+	os_thread_count++;
+	os_mutex_exit(os_sync_mutex);
+
+	thread = CreateThread(NULL,	/* no security attributes */
+			      0,	/* default size stack */
+			      (LPTHREAD_START_ROUTINE)start_f,
+			      arg,
+			      0,	/* thread runs immediately */
+			      &win_thread_id);
+
+	if (srv_set_thread_priorities) {
+
+		/* Set created thread priority the same as a normal query
+		in MYSQL: we try to prevent starvation of threads by
+		assigning same priority QUERY_PRIOR to all */
+
+		ut_a(SetThreadPriority(thread, srv_query_thread_priority));
+	}
+
+	if (thread_id) {
+		*thread_id = win_thread_id;
+	}
+
+	return(thread);
+#else
+	int		ret;
+	os_thread_t	pthread;
+	pthread_attr_t	attr;
+
+#if !(defined(UNIV_HOTBACKUP) && defined(UNIV_HPUX10))
+	pthread_attr_init(&attr);
+#endif
+
+#ifdef UNIV_AIX
+	/* We must make sure a thread stack is at least 32 kB, otherwise
+	InnoDB might crash; we do not know if the default stack size on
+	AIX is always big enough. An empirical test on AIX-4.3 suggested
+	the size was 96 kB, though. */
+
+	ret = pthread_attr_setstacksize(&attr,
+					(size_t)(PTHREAD_STACK_MIN
+						 + 32 * 1024));
+	if (ret) {
+		fprintf(stderr,
+			"InnoDB: Error: pthread_attr_setstacksize"
+			" returned %d\n", ret);
+		exit(1);
+	}
+#endif
+#ifdef __NETWARE__
+	ret = pthread_attr_setstacksize(&attr,
+					(size_t) NW_THD_STACKSIZE);
+	if (ret) {
+		fprintf(stderr,
+			"InnoDB: Error: pthread_attr_setstacksize"
+			" returned %d\n", ret);
+		exit(1);
+	}
+#endif
+	os_mutex_enter(os_sync_mutex);
+	os_thread_count++;
+	os_mutex_exit(os_sync_mutex);
+
+#if defined(UNIV_HOTBACKUP) && defined(UNIV_HPUX10)
+	ret = pthread_create(&pthread, pthread_attr_default, start_f, arg);
+#else
+	ret = pthread_create(&pthread, &attr, start_f, arg);
+#endif
+	if (ret) {
+		fprintf(stderr,
+			"InnoDB: Error: pthread_create returned %d\n", ret);
+		exit(1);
+	}
+
+#if !(defined(UNIV_HOTBACKUP) && defined(UNIV_HPUX10))
+	pthread_attr_destroy(&attr);
+#endif
+	if (srv_set_thread_priorities) {
+
+		my_pthread_setprio(pthread, srv_query_thread_priority);
+	}
+
+	if (thread_id) {
+		*thread_id = pthread;
+	}
+
+	return(pthread);
+#endif
+}
+
+/*********************************************************************
+Exits the current thread. */
+UNIV_INTERN
+void
+os_thread_exit(
+/*===========*/
+	void*	exit_value)	/* in: exit value; in Windows this void*
+				is cast as a DWORD */
+{
+#ifdef UNIV_DEBUG_THREAD_CREATION
+	fprintf(stderr, "Thread exits, id %lu\n",
+		os_thread_pf(os_thread_get_curr_id()));
+#endif
+	os_mutex_enter(os_sync_mutex);
+	os_thread_count--;
+	os_mutex_exit(os_sync_mutex);
+
+#ifdef __WIN__
+	ExitThread((DWORD)exit_value);
+#else
+	pthread_exit(exit_value);
+#endif
+}
+
+/*********************************************************************
+Returns handle to the current thread. */
+UNIV_INTERN
+os_thread_t
+os_thread_get_curr(void)
+/*====================*/
+{
+#ifdef __WIN__
+	return(GetCurrentThread());
+#else
+	return(pthread_self());
+#endif
+}
+
+/*********************************************************************
+Advises the os to give up remainder of the thread's time slice. */
+UNIV_INTERN
+void
+os_thread_yield(void)
+/*=================*/
+{
+#if defined(__WIN__)
+	Sleep(0);
+#elif (defined(HAVE_SCHED_YIELD) && defined(HAVE_SCHED_H))
+	sched_yield();
+#elif defined(HAVE_PTHREAD_YIELD_ZERO_ARG)
+	pthread_yield();
+#elif defined(HAVE_PTHREAD_YIELD_ONE_ARG)
+	pthread_yield(0);
+#else
+	os_thread_sleep(0);
+#endif
+}
+
+/*********************************************************************
+The thread sleeps at least the time given in microseconds. */
+UNIV_INTERN
+void
+os_thread_sleep(
+/*============*/
+	ulint	tm)	/* in: time in microseconds */
+{
+#ifdef __WIN__
+	Sleep((DWORD) tm / 1000);
+#elif defined(__NETWARE__)
+	delay(tm / 1000);
+#else
+	struct timeval	t;
+
+	t.tv_sec = tm / 1000000;
+	t.tv_usec = tm % 1000000;
+
+	select(0, NULL, NULL, NULL, &t);
+#endif
+}
+
+/**********************************************************************
+Sets a thread priority. */
+UNIV_INTERN
+void
+os_thread_set_priority(
+/*===================*/
+	os_thread_t	handle,	/* in: OS handle to the thread */
+	ulint		pri)	/* in: priority */
+{
+#ifdef __WIN__
+	int	os_pri;
+
+	if (pri == OS_THREAD_PRIORITY_BACKGROUND) {
+		os_pri = THREAD_PRIORITY_BELOW_NORMAL;
+	} else if (pri == OS_THREAD_PRIORITY_NORMAL) {
+		os_pri = THREAD_PRIORITY_NORMAL;
+	} else if (pri == OS_THREAD_PRIORITY_ABOVE_NORMAL) {
+		os_pri = THREAD_PRIORITY_HIGHEST;
+	} else {
+		ut_error;
+	}
+
+	ut_a(SetThreadPriority(handle, os_pri));
+#else
+	UT_NOT_USED(handle);
+	UT_NOT_USED(pri);
+#endif
+}
+
+/**********************************************************************
+Gets a thread priority. */
+UNIV_INTERN
+ulint
+os_thread_get_priority(
+/*===================*/
+				/* out: priority */
+	os_thread_t	handle __attribute__((unused)))
+				/* in: OS handle to the thread */
+{
+#ifdef __WIN__
+	int	os_pri;
+	ulint	pri;
+
+	os_pri = GetThreadPriority(handle);
+
+	if (os_pri == THREAD_PRIORITY_BELOW_NORMAL) {
+		pri = OS_THREAD_PRIORITY_BACKGROUND;
+	} else if (os_pri == THREAD_PRIORITY_NORMAL) {
+		pri = OS_THREAD_PRIORITY_NORMAL;
+	} else if (os_pri == THREAD_PRIORITY_HIGHEST) {
+		pri = OS_THREAD_PRIORITY_ABOVE_NORMAL;
+	} else {
+		ut_error;
+	}
+
+	return(pri);
+#else
+	return(0);
+#endif
+}
+
+/**********************************************************************
+Gets the last operating system error code for the calling thread. */
+UNIV_INTERN
+ulint
+os_thread_get_last_error(void)
+/*==========================*/
+{
+#ifdef __WIN__
+	return(GetLastError());
+#else
+	return(0);
+#endif
+}
diff --git a/storage/xtradb/page/page0cur.c b/storage/xtradb/page/page0cur.c
new file mode 100644
index 00000000000..e810756c1e4
--- /dev/null
+++ b/storage/xtradb/page/page0cur.c
@@ -0,0 +1,1922 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/************************************************************************
+The page cursor
+
+Created 10/4/1994 Heikki Tuuri
+*************************************************************************/
+
+#include "page0cur.h"
+#ifdef UNIV_NONINL
+#include "page0cur.ic"
+#endif
+
+#include "page0zip.h"
+#include "mtr0log.h"
+#include "log0recv.h"
+#include "rem0cmp.h"
+
+static ulint	page_rnd	= 976722341;
+
+#ifdef PAGE_CUR_ADAPT
+# ifdef UNIV_SEARCH_PERF_STAT
+static ulint	page_cur_short_succ	= 0;
+# endif /* UNIV_SEARCH_PERF_STAT */
+
+/********************************************************************
+Tries a search shortcut based on the last insert. */
+UNIV_INLINE
+ibool
+page_cur_try_search_shortcut(
+/*=========================*/
+					/* out: TRUE on success */
+	const buf_block_t*	block,	/* in: index page */
+	const dict_index_t*	index,	/* in: record descriptor */
+	const dtuple_t*		tuple,	/* in: data tuple */
+	ulint*			iup_matched_fields,
+					/* in/out: already matched
+					fields in upper limit record */
+	ulint*			iup_matched_bytes,
+					/* in/out: already matched
+					bytes in a field not yet
+					completely matched */
+	ulint*			ilow_matched_fields,
+					/* in/out: already matched
+					fields in lower limit record */
+	ulint*			ilow_matched_bytes,
+					/* in/out: already matched
+					bytes in a field not yet
+					completely matched */
+	page_cur_t*		cursor) /* out: page cursor */
+{
+	const rec_t*	rec;
+	const rec_t*	next_rec;
+	ulint		low_match;
+	ulint		low_bytes;
+	ulint		up_match;
+	ulint		up_bytes;
+#ifdef UNIV_SEARCH_DEBUG
+	page_cur_t	cursor2;
+#endif
+	ibool		success		= FALSE;
+	const page_t*	page		= buf_block_get_frame(block);
+	mem_heap_t*	heap		= NULL;
+	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint*		offsets		= offsets_;
+	rec_offs_init(offsets_);
+
+	ut_ad(dtuple_check_typed(tuple));
+
+	rec = page_header_get_ptr(page, PAGE_LAST_INSERT);
+	offsets = rec_get_offsets(rec, index, offsets,
+				  dtuple_get_n_fields(tuple), &heap);
+
+	ut_ad(rec);
+	ut_ad(page_rec_is_user_rec(rec));
+
+	ut_pair_min(&low_match, &low_bytes,
+		    *ilow_matched_fields, *ilow_matched_bytes,
+		    *iup_matched_fields, *iup_matched_bytes);
+
+	up_match = low_match;
+	up_bytes = low_bytes;
+
+	if (page_cmp_dtuple_rec_with_match(tuple, rec, offsets,
+					   &low_match, &low_bytes) < 0) {
+		goto exit_func;
+	}
+
+	next_rec = page_rec_get_next_const(rec);
+	offsets = rec_get_offsets(next_rec, index, offsets,
+				  dtuple_get_n_fields(tuple), &heap);
+
+	if (page_cmp_dtuple_rec_with_match(tuple, next_rec, offsets,
+					   &up_match, &up_bytes) >= 0) {
+		goto exit_func;
+	}
+
+	page_cur_position(rec, block, cursor);
+
+#ifdef UNIV_SEARCH_DEBUG
+	page_cur_search_with_match(block, index, tuple, PAGE_CUR_DBG,
+				   iup_matched_fields,
+				   iup_matched_bytes,
+				   ilow_matched_fields,
+				   ilow_matched_bytes,
+				   &cursor2);
+	ut_a(cursor2.rec == cursor->rec);
+
+	if (!page_rec_is_supremum(next_rec)) {
+
+		ut_a(*iup_matched_fields == up_match);
+		ut_a(*iup_matched_bytes == up_bytes);
+	}
+
+	ut_a(*ilow_matched_fields == low_match);
+	ut_a(*ilow_matched_bytes == low_bytes);
+#endif
+	if (!page_rec_is_supremum(next_rec)) {
+
+		*iup_matched_fields = up_match;
+		*iup_matched_bytes = up_bytes;
+	}
+
+	*ilow_matched_fields = low_match;
+	*ilow_matched_bytes = low_bytes;
+
+#ifdef UNIV_SEARCH_PERF_STAT
+	page_cur_short_succ++;
+#endif
+	success = TRUE;
+exit_func:
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+	return(success);
+}
+
+#endif
+
+#ifdef PAGE_CUR_LE_OR_EXTENDS
+/********************************************************************
+Checks if the nth field in a record is a character type field which extends
+the nth field in tuple, i.e., the field is longer or equal in length and has
+common first characters. */
+static
+ibool
+page_cur_rec_field_extends(
+/*=======================*/
+				/* out: TRUE if rec field
+				extends tuple field */
+	const dtuple_t*	tuple,	/* in: data tuple */
+	const rec_t*	rec,	/* in: record */
+	const ulint*	offsets,/* in: array returned by rec_get_offsets() */
+	ulint		n)	/* in: compare nth field */
+{
+	const dtype_t*	type;
+	const dfield_t*	dfield;
+	const byte*	rec_f;
+	ulint		rec_f_len;
+
+	ut_ad(rec_offs_validate(rec, NULL, offsets));
+	dfield = dtuple_get_nth_field(tuple, n);
+
+	type = dfield_get_type(dfield);
+
+	rec_f = rec_get_nth_field(rec, offsets, n, &rec_f_len);
+
+	if (type->mtype == DATA_VARCHAR
+	    || type->mtype == DATA_CHAR
+	    || type->mtype == DATA_FIXBINARY
+	    || type->mtype == DATA_BINARY
+	    || type->mtype == DATA_BLOB
+	    || type->mtype == DATA_VARMYSQL
+	    || type->mtype == DATA_MYSQL) {
+
+		if (dfield_get_len(dfield) != UNIV_SQL_NULL
+		    && rec_f_len != UNIV_SQL_NULL
+		    && rec_f_len >= dfield_get_len(dfield)
+		    && !cmp_data_data_slow(type->mtype, type->prtype,
+					   dfield_get_data(dfield),
+					   dfield_get_len(dfield),
+					   rec_f, dfield_get_len(dfield))) {
+
+			return(TRUE);
+		}
+	}
+
+	return(FALSE);
+}
+#endif /* PAGE_CUR_LE_OR_EXTENDS */
+
+/********************************************************************
+Searches the right position for a page cursor. */
+UNIV_INTERN
+void
+page_cur_search_with_match(
+/*=======================*/
+	const buf_block_t*	block,	/* in: buffer block */
+	const dict_index_t*	index,	/* in: record descriptor */
+	const dtuple_t*		tuple,	/* in: data tuple */
+	ulint			mode,	/* in: PAGE_CUR_L,
+					PAGE_CUR_LE, PAGE_CUR_G, or
+					PAGE_CUR_GE */
+	ulint*			iup_matched_fields,
+					/* in/out: already matched
+					fields in upper limit record */
+	ulint*			iup_matched_bytes,
+					/* in/out: already matched
+					bytes in a field not yet
+					completely matched */
+	ulint*			ilow_matched_fields,
+					/* in/out: already matched
+					fields in lower limit record */
+	ulint*			ilow_matched_bytes,
+					/* in/out: already matched
+					bytes in a field not yet
+					completely matched */
+	page_cur_t*		cursor)	/* out: page cursor */
+{
+	ulint		up;
+	ulint		low;
+	ulint		mid;
+	const page_t*	page;
+	const page_dir_slot_t* slot;
+	const rec_t*	up_rec;
+	const rec_t*	low_rec;
+	const rec_t*	mid_rec;
+	ulint		up_matched_fields;
+	ulint		up_matched_bytes;
+	ulint		low_matched_fields;
+	ulint		low_matched_bytes;
+	ulint		cur_matched_fields;
+	ulint		cur_matched_bytes;
+	int		cmp;
+#ifdef UNIV_SEARCH_DEBUG
+	int		dbg_cmp;
+	ulint		dbg_matched_fields;
+	ulint		dbg_matched_bytes;
+#endif
+#ifdef UNIV_ZIP_DEBUG
+	const page_zip_des_t*	page_zip = buf_block_get_page_zip(block);
+#endif /* UNIV_ZIP_DEBUG */
+	mem_heap_t*	heap		= NULL;
+	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint*		offsets		= offsets_;
+	rec_offs_init(offsets_);
+
+	ut_ad(block && tuple && iup_matched_fields && iup_matched_bytes
+	      && ilow_matched_fields && ilow_matched_bytes && cursor);
+	ut_ad(dtuple_validate(tuple));
+#ifdef UNIV_DEBUG
+# ifdef PAGE_CUR_DBG
+	if (mode != PAGE_CUR_DBG)
+# endif /* PAGE_CUR_DBG */
+# ifdef PAGE_CUR_LE_OR_EXTENDS
+		if (mode != PAGE_CUR_LE_OR_EXTENDS)
+# endif /* PAGE_CUR_LE_OR_EXTENDS */
+			ut_ad(mode == PAGE_CUR_L || mode == PAGE_CUR_LE
+			      || mode == PAGE_CUR_G || mode == PAGE_CUR_GE);
+#endif /* UNIV_DEBUG */
+	page = buf_block_get_frame(block);
+#ifdef UNIV_ZIP_DEBUG
+	ut_a(!page_zip || page_zip_validate(page_zip, page));
+#endif /* UNIV_ZIP_DEBUG */
+
+	page_check_dir(page);
+
+#ifdef PAGE_CUR_ADAPT
+	if (page_is_leaf(page)
+	    && (mode == PAGE_CUR_LE)
+	    && (page_header_get_field(page, PAGE_N_DIRECTION) > 3)
+	    && (page_header_get_ptr(page, PAGE_LAST_INSERT))
+	    && (page_header_get_field(page, PAGE_DIRECTION) == PAGE_RIGHT)) {
+
+		if (page_cur_try_search_shortcut(
+			    block, index, tuple,
+			    iup_matched_fields, iup_matched_bytes,
+			    ilow_matched_fields, ilow_matched_bytes,
+			    cursor)) {
+			return;
+		}
+	}
+# ifdef PAGE_CUR_DBG
+	if (mode == PAGE_CUR_DBG) {
+		mode = PAGE_CUR_LE;
+	}
+# endif
+#endif
+
+	/* The following flag does not work for non-latin1 char sets because
+	cmp_full_field does not tell how many bytes matched */
+#ifdef PAGE_CUR_LE_OR_EXTENDS
+	ut_a(mode != PAGE_CUR_LE_OR_EXTENDS);
+#endif /* PAGE_CUR_LE_OR_EXTENDS */
+
+	/* If mode PAGE_CUR_G is specified, we are trying to position the
+	cursor to answer a query of the form "tuple < X", where tuple is
+	the input parameter, and X denotes an arbitrary physical record on
+	the page. We want to position the cursor on the first X which
+	satisfies the condition. */
+
+	up_matched_fields  = *iup_matched_fields;
+	up_matched_bytes   = *iup_matched_bytes;
+	low_matched_fields = *ilow_matched_fields;
+	low_matched_bytes  = *ilow_matched_bytes;
+
+	/* Perform binary search. First the search is done through the page
+	directory, after that as a linear search in the list of records
+	owned by the upper limit directory slot. */
+
+	low = 0;
+	up = page_dir_get_n_slots(page) - 1;
+
+	/* Perform binary search until the lower and upper limit directory
+	slots come to the distance 1 of each other */
+
+	while (up - low > 1) {
+		mid = (low + up) / 2;
+		slot = page_dir_get_nth_slot(page, mid);
+		mid_rec = page_dir_slot_get_rec(slot);
+
+		ut_pair_min(&cur_matched_fields, &cur_matched_bytes,
+			    low_matched_fields, low_matched_bytes,
+			    up_matched_fields, up_matched_bytes);
+
+		offsets = rec_get_offsets(mid_rec, index, offsets,
+					  dtuple_get_n_fields_cmp(tuple),
+					  &heap);
+
+		cmp = cmp_dtuple_rec_with_match(tuple, mid_rec, offsets,
+						&cur_matched_fields,
+						&cur_matched_bytes);
+		if (UNIV_LIKELY(cmp > 0)) {
+low_slot_match:
+			low = mid;
+			low_matched_fields = cur_matched_fields;
+			low_matched_bytes = cur_matched_bytes;
+
+		} else if (UNIV_EXPECT(cmp, -1)) {
+#ifdef PAGE_CUR_LE_OR_EXTENDS
+			if (mode == PAGE_CUR_LE_OR_EXTENDS
+			    && page_cur_rec_field_extends(
+				    tuple, mid_rec, offsets,
+				    cur_matched_fields)) {
+
+				goto low_slot_match;
+			}
+#endif /* PAGE_CUR_LE_OR_EXTENDS */
+up_slot_match:
+			up = mid;
+			up_matched_fields = cur_matched_fields;
+			up_matched_bytes = cur_matched_bytes;
+
+		} else if (mode == PAGE_CUR_G || mode == PAGE_CUR_LE
+#ifdef PAGE_CUR_LE_OR_EXTENDS
+			   || mode == PAGE_CUR_LE_OR_EXTENDS
+#endif /* PAGE_CUR_LE_OR_EXTENDS */
+			   ) {
+
+			goto low_slot_match;
+		} else {
+
+			goto up_slot_match;
+		}
+	}
+
+	slot = page_dir_get_nth_slot(page, low);
+	low_rec = page_dir_slot_get_rec(slot);
+	slot = page_dir_get_nth_slot(page, up);
+	up_rec = page_dir_slot_get_rec(slot);
+
+	/* Perform linear search until the upper and lower records come to
+	distance 1 of each other. */
+
+	while (page_rec_get_next_const(low_rec) != up_rec) {
+
+		mid_rec = page_rec_get_next_const(low_rec);
+
+		ut_pair_min(&cur_matched_fields, &cur_matched_bytes,
+			    low_matched_fields, low_matched_bytes,
+			    up_matched_fields, up_matched_bytes);
+
+		offsets = rec_get_offsets(mid_rec, index, offsets,
+					  dtuple_get_n_fields_cmp(tuple),
+					  &heap);
+
+		cmp = cmp_dtuple_rec_with_match(tuple, mid_rec, offsets,
+						&cur_matched_fields,
+						&cur_matched_bytes);
+		if (UNIV_LIKELY(cmp > 0)) {
+low_rec_match:
+			low_rec = mid_rec;
+			low_matched_fields = cur_matched_fields;
+			low_matched_bytes = cur_matched_bytes;
+
+		} else if (UNIV_EXPECT(cmp, -1)) {
+#ifdef PAGE_CUR_LE_OR_EXTENDS
+			if (mode == PAGE_CUR_LE_OR_EXTENDS
+			    && page_cur_rec_field_extends(
+				    tuple, mid_rec, offsets,
+				    cur_matched_fields)) {
+
+				goto low_rec_match;
+			}
+#endif /* PAGE_CUR_LE_OR_EXTENDS */
+up_rec_match:
+			up_rec = mid_rec;
+			up_matched_fields = cur_matched_fields;
+			up_matched_bytes = cur_matched_bytes;
+		} else if (mode == PAGE_CUR_G || mode == PAGE_CUR_LE
+#ifdef PAGE_CUR_LE_OR_EXTENDS
+			   || mode == PAGE_CUR_LE_OR_EXTENDS
+#endif /* PAGE_CUR_LE_OR_EXTENDS */
+			   ) {
+
+			goto low_rec_match;
+		} else {
+
+			goto up_rec_match;
+		}
+	}
+
+#ifdef UNIV_SEARCH_DEBUG
+
+	/* Check that the lower and upper limit records have the
+	right alphabetical order compared to tuple. */
+	dbg_matched_fields = 0;
+	dbg_matched_bytes = 0;
+
+	offsets = rec_get_offsets(low_rec, index, offsets,
+				  ULINT_UNDEFINED, &heap);
+	dbg_cmp = page_cmp_dtuple_rec_with_match(tuple, low_rec, offsets,
+						 &dbg_matched_fields,
+						 &dbg_matched_bytes);
+	if (mode == PAGE_CUR_G) {
+		ut_a(dbg_cmp >= 0);
+	} else if (mode == PAGE_CUR_GE) {
+		ut_a(dbg_cmp == 1);
+	} else if (mode == PAGE_CUR_L) {
+		ut_a(dbg_cmp == 1);
+	} else if (mode == PAGE_CUR_LE) {
+		ut_a(dbg_cmp >= 0);
+	}
+
+	if (!page_rec_is_infimum(low_rec)) {
+
+		ut_a(low_matched_fields == dbg_matched_fields);
+		ut_a(low_matched_bytes == dbg_matched_bytes);
+	}
+
+	dbg_matched_fields = 0;
+	dbg_matched_bytes = 0;
+
+	offsets = rec_get_offsets(up_rec, index, offsets,
+				  ULINT_UNDEFINED, &heap);
+	dbg_cmp = page_cmp_dtuple_rec_with_match(tuple, up_rec, offsets,
+						 &dbg_matched_fields,
+						 &dbg_matched_bytes);
+	if (mode == PAGE_CUR_G) {
+		ut_a(dbg_cmp == -1);
+	} else if (mode == PAGE_CUR_GE) {
+		ut_a(dbg_cmp <= 0);
+	} else if (mode == PAGE_CUR_L) {
+		ut_a(dbg_cmp <= 0);
+	} else if (mode == PAGE_CUR_LE) {
+		ut_a(dbg_cmp == -1);
+	}
+
+	if (!page_rec_is_supremum(up_rec)) {
+
+		ut_a(up_matched_fields == dbg_matched_fields);
+		ut_a(up_matched_bytes == dbg_matched_bytes);
+	}
+#endif
+	if (mode <= PAGE_CUR_GE) {
+		page_cur_position(up_rec, block, cursor);
+	} else {
+		page_cur_position(low_rec, block, cursor);
+	}
+
+	*iup_matched_fields  = up_matched_fields;
+	*iup_matched_bytes   = up_matched_bytes;
+	*ilow_matched_fields = low_matched_fields;
+	*ilow_matched_bytes  = low_matched_bytes;
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+}
+
+/***************************************************************
+Positions a page cursor on a randomly chosen user record on a page. If there
+are no user records, sets the cursor on the infimum record. */
+UNIV_INTERN
+void
+page_cur_open_on_rnd_user_rec(
+/*==========================*/
+	buf_block_t*	block,	/* in: page */
+	page_cur_t*	cursor)	/* out: page cursor */
+{
+	ulint	rnd;
+	ulint	n_recs = page_get_n_recs(buf_block_get_frame(block));
+
+	page_cur_set_before_first(block, cursor);
+
+	if (UNIV_UNLIKELY(n_recs == 0)) {
+
+		return;
+	}
+
+	page_rnd += 87584577;
+
+	rnd = page_rnd % n_recs;
+
+	do {
+		page_cur_move_to_next(cursor);
+	} while (rnd--);
+}
+
+/***************************************************************
+Writes the log record of a record insert on a page. */
+static
+void
+page_cur_insert_rec_write_log(
+/*==========================*/
+	rec_t*		insert_rec,	/* in: inserted physical record */
+	ulint		rec_size,	/* in: insert_rec size */
+	rec_t*		cursor_rec,	/* in: record the
+					cursor is pointing to */
+	dict_index_t*	index,		/* in: record descriptor */
+	mtr_t*		mtr)		/* in: mini-transaction handle */
+{
+	ulint	cur_rec_size;
+	ulint	extra_size;
+	ulint	cur_extra_size;
+	const byte* ins_ptr;
+	byte*	log_ptr;
+	const byte* log_end;
+	ulint	i;
+
+	ut_a(rec_size < UNIV_PAGE_SIZE);
+	ut_ad(page_align(insert_rec) == page_align(cursor_rec));
+	ut_ad(!page_rec_is_comp(insert_rec)
+	      == !dict_table_is_comp(index->table));
+
+	{
+		mem_heap_t*	heap		= NULL;
+		ulint		cur_offs_[REC_OFFS_NORMAL_SIZE];
+		ulint		ins_offs_[REC_OFFS_NORMAL_SIZE];
+
+		ulint*		cur_offs;
+		ulint*		ins_offs;
+
+		rec_offs_init(cur_offs_);
+		rec_offs_init(ins_offs_);
+
+		cur_offs = rec_get_offsets(cursor_rec, index, cur_offs_,
+					   ULINT_UNDEFINED, &heap);
+		ins_offs = rec_get_offsets(insert_rec, index, ins_offs_,
+					   ULINT_UNDEFINED, &heap);
+
+		extra_size = rec_offs_extra_size(ins_offs);
+		cur_extra_size = rec_offs_extra_size(cur_offs);
+		ut_ad(rec_size == rec_offs_size(ins_offs));
+		cur_rec_size = rec_offs_size(cur_offs);
+
+		if (UNIV_LIKELY_NULL(heap)) {
+			mem_heap_free(heap);
+		}
+	}
+
+	ins_ptr = insert_rec - extra_size;
+
+	i = 0;
+
+	if (cur_extra_size == extra_size) {
+		ulint		min_rec_size = ut_min(cur_rec_size, rec_size);
+
+		const byte*	cur_ptr = cursor_rec - cur_extra_size;
+
+		/* Find out the first byte in insert_rec which differs from
+		cursor_rec; skip the bytes in the record info */
+
+		do {
+			if (*ins_ptr == *cur_ptr) {
+				i++;
+				ins_ptr++;
+				cur_ptr++;
+			} else if ((i < extra_size)
+				   && (i >= extra_size
+				       - page_rec_get_base_extra_size
+				       (insert_rec))) {
+				i = extra_size;
+				ins_ptr = insert_rec;
+				cur_ptr = cursor_rec;
+			} else {
+				break;
+			}
+		} while (i < min_rec_size);
+	}
+
+	if (mtr_get_log_mode(mtr) != MTR_LOG_SHORT_INSERTS) {
+
+		if (page_rec_is_comp(insert_rec)) {
+			log_ptr = mlog_open_and_write_index(
+				mtr, insert_rec, index, MLOG_COMP_REC_INSERT,
+				2 + 5 + 1 + 5 + 5 + MLOG_BUF_MARGIN);
+			if (UNIV_UNLIKELY(!log_ptr)) {
+				/* Logging in mtr is switched off
+				during crash recovery: in that case
+				mlog_open returns NULL */
+				return;
+			}
+		} else {
+			log_ptr = mlog_open(mtr, 11
+					    + 2 + 5 + 1 + 5 + 5
+					    + MLOG_BUF_MARGIN);
+			if (UNIV_UNLIKELY(!log_ptr)) {
+				/* Logging in mtr is switched off
+				during crash recovery: in that case
+				mlog_open returns NULL */
+				return;
+			}
+
+			log_ptr = mlog_write_initial_log_record_fast(
+				insert_rec, MLOG_REC_INSERT, log_ptr, mtr);
+		}
+
+		log_end = &log_ptr[2 + 5 + 1 + 5 + 5 + MLOG_BUF_MARGIN];
+		/* Write the cursor rec offset as a 2-byte ulint */
+		mach_write_to_2(log_ptr, page_offset(cursor_rec));
+		log_ptr += 2;
+	} else {
+		log_ptr = mlog_open(mtr, 5 + 1 + 5 + 5 + MLOG_BUF_MARGIN);
+		if (!log_ptr) {
+			/* Logging in mtr is switched off during crash
+			recovery: in that case mlog_open returns NULL */
+			return;
+		}
+		log_end = &log_ptr[5 + 1 + 5 + 5 + MLOG_BUF_MARGIN];
+	}
+
+	if (page_rec_is_comp(insert_rec)) {
+		if (UNIV_UNLIKELY
+		    (rec_get_info_and_status_bits(insert_rec, TRUE)
+		     != rec_get_info_and_status_bits(cursor_rec, TRUE))) {
+
+			goto need_extra_info;
+		}
+	} else {
+		if (UNIV_UNLIKELY
+		    (rec_get_info_and_status_bits(insert_rec, FALSE)
+		     != rec_get_info_and_status_bits(cursor_rec, FALSE))) {
+
+			goto need_extra_info;
+		}
+	}
+
+	if (extra_size != cur_extra_size || rec_size != cur_rec_size) {
+need_extra_info:
+		/* Write the record end segment length
+		and the extra info storage flag */
+		log_ptr += mach_write_compressed(log_ptr,
+						 2 * (rec_size - i) + 1);
+
+		/* Write the info bits */
+		mach_write_to_1(log_ptr,
+				rec_get_info_and_status_bits(
+					insert_rec,
+					page_rec_is_comp(insert_rec)));
+		log_ptr++;
+
+		/* Write the record origin offset */
+		log_ptr += mach_write_compressed(log_ptr, extra_size);
+
+		/* Write the mismatch index */
+		log_ptr += mach_write_compressed(log_ptr, i);
+
+		ut_a(i < UNIV_PAGE_SIZE);
+		ut_a(extra_size < UNIV_PAGE_SIZE);
+	} else {
+		/* Write the record end segment length
+		and the extra info storage flag */
+		log_ptr += mach_write_compressed(log_ptr, 2 * (rec_size - i));
+	}
+
+	/* Write to the log the inserted index record end segment which
+	differs from the cursor record */
+
+	rec_size -= i;
+
+	if (log_ptr + rec_size <= log_end) {
+		memcpy(log_ptr, ins_ptr, rec_size);
+		mlog_close(mtr, log_ptr + rec_size);
+	} else {
+		mlog_close(mtr, log_ptr);
+		ut_a(rec_size < UNIV_PAGE_SIZE);
+		mlog_catenate_string(mtr, ins_ptr, rec_size);
+	}
+}
+
+/***************************************************************
+Parses a log record of a record insert on a page. */
+UNIV_INTERN
+byte*
+page_cur_parse_insert_rec(
+/*======================*/
+				/* out: end of log record or NULL */
+	ibool		is_short,/* in: TRUE if short inserts */
+	byte*		ptr,	/* in: buffer */
+	byte*		end_ptr,/* in: buffer end */
+	buf_block_t*	block,	/* in: page or NULL */
+	dict_index_t*	index,	/* in: record descriptor */
+	mtr_t*		mtr)	/* in: mtr or NULL */
+{
+	ulint	origin_offset;
+	ulint	end_seg_len;
+	ulint	mismatch_index;
+	page_t*	page;
+	rec_t*	cursor_rec;
+	byte	buf1[1024];
+	byte*	buf;
+	byte*	ptr2			= ptr;
+	ulint	info_and_status_bits = 0; /* remove warning */
+	page_cur_t cursor;
+	mem_heap_t*	heap		= NULL;
+	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint*		offsets		= offsets_;
+	rec_offs_init(offsets_);
+
+	page = block ? buf_block_get_frame(block) : NULL;
+
+	if (is_short) {
+		cursor_rec = page_rec_get_prev(page_get_supremum_rec(page));
+	} else {
+		ulint	offset;
+
+		/* Read the cursor rec offset as a 2-byte ulint */
+
+		if (UNIV_UNLIKELY(end_ptr < ptr + 2)) {
+
+			return(NULL);
+		}
+
+		offset = mach_read_from_2(ptr);
+		ptr += 2;
+
+		cursor_rec = page + offset;
+
+		if (UNIV_UNLIKELY(offset >= UNIV_PAGE_SIZE)) {
+
+			recv_sys->found_corrupt_log = TRUE;
+
+			return(NULL);
+		}
+	}
+
+	ptr = mach_parse_compressed(ptr, end_ptr, &end_seg_len);
+
+	if (ptr == NULL) {
+
+		return(NULL);
+	}
+
+	if (UNIV_UNLIKELY(end_seg_len >= UNIV_PAGE_SIZE << 1)) {
+		recv_sys->found_corrupt_log = TRUE;
+
+		return(NULL);
+	}
+
+	if (end_seg_len & 0x1UL) {
+		/* Read the info bits */
+
+		if (end_ptr < ptr + 1) {
+
+			return(NULL);
+		}
+
+		info_and_status_bits = mach_read_from_1(ptr);
+		ptr++;
+
+		ptr = mach_parse_compressed(ptr, end_ptr, &origin_offset);
+
+		if (ptr == NULL) {
+
+			return(NULL);
+		}
+
+		ut_a(origin_offset < UNIV_PAGE_SIZE);
+
+		ptr = mach_parse_compressed(ptr, end_ptr, &mismatch_index);
+
+		if (ptr == NULL) {
+
+			return(NULL);
+		}
+
+		ut_a(mismatch_index < UNIV_PAGE_SIZE);
+	}
+
+	if (UNIV_UNLIKELY(end_ptr < ptr + (end_seg_len >> 1))) {
+
+		return(NULL);
+	}
+
+	if (!block) {
+
+		return(ptr + (end_seg_len >> 1));
+	}
+
+	ut_ad(!!page_is_comp(page) == dict_table_is_comp(index->table));
+	ut_ad(!buf_block_get_page_zip(block) || page_is_comp(page));
+
+	/* Read from the log the inserted index record end segment which
+	differs from the cursor record */
+
+	offsets = rec_get_offsets(cursor_rec, index, offsets,
+				  ULINT_UNDEFINED, &heap);
+
+	if (!(end_seg_len & 0x1UL)) {
+		info_and_status_bits = rec_get_info_and_status_bits(
+			cursor_rec, page_is_comp(page));
+		origin_offset = rec_offs_extra_size(offsets);
+		mismatch_index = rec_offs_size(offsets) - (end_seg_len >> 1);
+	}
+
+	end_seg_len >>= 1;
+
+	if (mismatch_index + end_seg_len < sizeof buf1) {
+		buf = buf1;
+	} else {
+		buf = mem_alloc(mismatch_index + end_seg_len);
+	}
+
+	/* Build the inserted record to buf */
+
+        if (UNIV_UNLIKELY(mismatch_index >= UNIV_PAGE_SIZE)) {
+		fprintf(stderr,
+			"Is short %lu, info_and_status_bits %lu, offset %lu, "
+			"o_offset %lu\n"
+			"mismatch index %lu, end_seg_len %lu\n"
+			"parsed len %lu\n",
+			(ulong) is_short, (ulong) info_and_status_bits,
+			(ulong) page_offset(cursor_rec),
+			(ulong) origin_offset,
+			(ulong) mismatch_index, (ulong) end_seg_len,
+			(ulong) (ptr - ptr2));
+
+		fputs("Dump of 300 bytes of log:\n", stderr);
+		ut_print_buf(stderr, ptr2, 300);
+		putc('\n', stderr);
+
+		buf_page_print(page, 0);
+
+		ut_error;
+	}
+
+	ut_memcpy(buf, rec_get_start(cursor_rec, offsets), mismatch_index);
+	ut_memcpy(buf + mismatch_index, ptr, end_seg_len);
+
+	if (page_is_comp(page)) {
+		rec_set_info_and_status_bits(buf + origin_offset,
+				     info_and_status_bits);
+	} else {
+		rec_set_info_bits_old(buf + origin_offset,
+							info_and_status_bits);
+	}
+
+	page_cur_position(cursor_rec, block, &cursor);
+
+	offsets = rec_get_offsets(buf + origin_offset, index, offsets,
+				  ULINT_UNDEFINED, &heap);
+	if (UNIV_UNLIKELY(!page_cur_rec_insert(&cursor,
+					       buf + origin_offset,
+					       index, offsets, mtr))) {
+		/* The redo log record should only have been written
+		after the write was successful. */
+		ut_error;
+	}
+
+	if (buf != buf1) {
+
+		mem_free(buf);
+	}
+
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+
+	return(ptr + end_seg_len);
+}
+
+/***************************************************************
+Inserts a record next to page cursor on an uncompressed page.
+Returns pointer to inserted record if succeed, i.e., enough
+space available, NULL otherwise. The cursor stays at the same position. */
+UNIV_INTERN
+rec_t*
+page_cur_insert_rec_low(
+/*====================*/
+				/* out: pointer to record if succeed, NULL
+				otherwise */
+	rec_t*		current_rec,/* in: pointer to current record after
+				which the new record is inserted */
+	dict_index_t*	index,	/* in: record descriptor */
+	const rec_t*	rec,	/* in: pointer to a physical record */
+	ulint*		offsets,/* in/out: rec_get_offsets(rec, index) */
+	mtr_t*		mtr)	/* in: mini-transaction handle, or NULL */
+{
+	byte*		insert_buf;
+	ulint		rec_size;
+	page_t*		page;		/* the relevant page */
+	rec_t*		last_insert;	/* cursor position at previous
+					insert */
+	rec_t*		free_rec;	/* a free record that was reused,
+					or NULL */
+	rec_t*		insert_rec;	/* inserted record */
+	ulint		heap_no;	/* heap number of the inserted
+					record */
+
+	ut_ad(rec_offs_validate(rec, index, offsets));
+
+	page = page_align(current_rec);
+	ut_ad(dict_table_is_comp(index->table)
+	      == (ibool) !!page_is_comp(page));
+
+	ut_ad(!page_rec_is_supremum(current_rec));
+
+	/* 1. Get the size of the physical record in the page */
+	rec_size = rec_offs_size(offsets);
+
+#ifdef UNIV_DEBUG_VALGRIND
+	{
+		const void*	rec_start
+			= rec - rec_offs_extra_size(offsets);
+		ulint		extra_size
+			= rec_offs_extra_size(offsets)
+			- (rec_offs_comp(offsets)
+			   ? REC_N_NEW_EXTRA_BYTES
+			   : REC_N_OLD_EXTRA_BYTES);
+
+		/* All data bytes of the record must be valid. */
+		UNIV_MEM_ASSERT_RW(rec, rec_offs_data_size(offsets));
+		/* The variable-length header must be valid. */
+		UNIV_MEM_ASSERT_RW(rec_start, extra_size);
+	}
+#endif /* UNIV_DEBUG_VALGRIND */
+
+	/* 2. Try to find suitable space from page memory management */
+
+	free_rec = page_header_get_ptr(page, PAGE_FREE);
+	if (UNIV_LIKELY_NULL(free_rec)) {
+		/* Try to allocate from the head of the free list. */
+		ulint		foffsets_[REC_OFFS_NORMAL_SIZE];
+		ulint*		foffsets	= foffsets_;
+		mem_heap_t*	heap		= NULL;
+
+		rec_offs_init(foffsets_);
+
+		foffsets = rec_get_offsets(free_rec, index, foffsets,
+					ULINT_UNDEFINED, &heap);
+		if (rec_offs_size(foffsets) < rec_size) {
+			if (UNIV_LIKELY_NULL(heap)) {
+				mem_heap_free(heap);
+			}
+
+			goto use_heap;
+		}
+
+		insert_buf = free_rec - rec_offs_extra_size(foffsets);
+
+		if (page_is_comp(page)) {
+			heap_no = rec_get_heap_no_new(free_rec);
+			page_mem_alloc_free(page, NULL,
+					rec_get_next_ptr(free_rec, TRUE),
+					rec_size);
+		} else {
+			heap_no = rec_get_heap_no_old(free_rec);
+			page_mem_alloc_free(page, NULL,
+					rec_get_next_ptr(free_rec, FALSE),
+					rec_size);
+		}
+
+		if (UNIV_LIKELY_NULL(heap)) {
+			mem_heap_free(heap);
+		}
+	} else {
+use_heap:
+		free_rec = NULL;
+		insert_buf = page_mem_alloc_heap(page, NULL,
+						 rec_size, &heap_no);
+
+		if (UNIV_UNLIKELY(insert_buf == NULL)) {
+			return(NULL);
+		}
+	}
+
+	/* 3. Create the record */
+	insert_rec = rec_copy(insert_buf, rec, offsets);
+	rec_offs_make_valid(insert_rec, index, offsets);
+
+	/* 4. Insert the record in the linked list of records */
+	ut_ad(current_rec != insert_rec);
+
+	{
+		/* next record after current before the insertion */
+		rec_t*	next_rec = page_rec_get_next(current_rec);
+#ifdef UNIV_DEBUG
+		if (page_is_comp(page)) {
+			ut_ad(rec_get_status(current_rec)
+				<= REC_STATUS_INFIMUM);
+			ut_ad(rec_get_status(insert_rec) < REC_STATUS_INFIMUM);
+			ut_ad(rec_get_status(next_rec) != REC_STATUS_INFIMUM);
+		}
+#endif
+		page_rec_set_next(insert_rec, next_rec);
+		page_rec_set_next(current_rec, insert_rec);
+	}
+
+	page_header_set_field(page, NULL, PAGE_N_RECS,
+			      1 + page_get_n_recs(page));
+
+	/* 5. Set the n_owned field in the inserted record to zero,
+	and set the heap_no field */
+	if (page_is_comp(page)) {
+		rec_set_n_owned_new(insert_rec, NULL, 0);
+		rec_set_heap_no_new(insert_rec, heap_no);
+	} else {
+		rec_set_n_owned_old(insert_rec, 0);
+		rec_set_heap_no_old(insert_rec, heap_no);
+	}
+
+	UNIV_MEM_ASSERT_RW(rec_get_start(insert_rec, offsets),
+			   rec_offs_size(offsets));
+	/* 6. Update the last insertion info in page header */
+
+	last_insert = page_header_get_ptr(page, PAGE_LAST_INSERT);
+	ut_ad(!last_insert || !page_is_comp(page)
+	      || rec_get_node_ptr_flag(last_insert)
+	      == rec_get_node_ptr_flag(insert_rec));
+
+	if (UNIV_UNLIKELY(last_insert == NULL)) {
+		page_header_set_field(page, NULL, PAGE_DIRECTION,
+				      PAGE_NO_DIRECTION);
+		page_header_set_field(page, NULL, PAGE_N_DIRECTION, 0);
+
+	} else if ((last_insert == current_rec)
+		   && (page_header_get_field(page, PAGE_DIRECTION)
+		       != PAGE_LEFT)) {
+
+		page_header_set_field(page, NULL, PAGE_DIRECTION,
+							PAGE_RIGHT);
+		page_header_set_field(page, NULL, PAGE_N_DIRECTION,
+				      page_header_get_field(
+					      page, PAGE_N_DIRECTION) + 1);
+
+	} else if ((page_rec_get_next(insert_rec) == last_insert)
+		   && (page_header_get_field(page, PAGE_DIRECTION)
+		       != PAGE_RIGHT)) {
+
+		page_header_set_field(page, NULL, PAGE_DIRECTION,
+							PAGE_LEFT);
+		page_header_set_field(page, NULL, PAGE_N_DIRECTION,
+				      page_header_get_field(
+					      page, PAGE_N_DIRECTION) + 1);
+	} else {
+		page_header_set_field(page, NULL, PAGE_DIRECTION,
+							PAGE_NO_DIRECTION);
+		page_header_set_field(page, NULL, PAGE_N_DIRECTION, 0);
+	}
+
+	page_header_set_ptr(page, NULL, PAGE_LAST_INSERT, insert_rec);
+
+	/* 7. It remains to update the owner record. */
+	{
+		rec_t*	owner_rec	= page_rec_find_owner_rec(insert_rec);
+		ulint	n_owned;
+		if (page_is_comp(page)) {
+			n_owned = rec_get_n_owned_new(owner_rec);
+			rec_set_n_owned_new(owner_rec, NULL, n_owned + 1);
+		} else {
+			n_owned = rec_get_n_owned_old(owner_rec);
+			rec_set_n_owned_old(owner_rec, n_owned + 1);
+		}
+
+		/* 8. Now we have incremented the n_owned field of the owner
+		record. If the number exceeds PAGE_DIR_SLOT_MAX_N_OWNED,
+		we have to split the corresponding directory slot in two. */
+
+		if (UNIV_UNLIKELY(n_owned == PAGE_DIR_SLOT_MAX_N_OWNED)) {
+			page_dir_split_slot(
+				page, NULL,
+				page_dir_find_owner_slot(owner_rec));
+		}
+	}
+
+	/* 9. Write log record of the insert */
+	if (UNIV_LIKELY(mtr != NULL)) {
+		page_cur_insert_rec_write_log(insert_rec, rec_size,
+					      current_rec, index, mtr);
+	}
+
+	return(insert_rec);
+}
+
+/***************************************************************
+Compresses or reorganizes a page after an optimistic insert. */
+static
+rec_t*
+page_cur_insert_rec_zip_reorg(
+/*==========================*/
+				/* out: rec if succeed, NULL otherwise */
+	rec_t**		current_rec,/* in/out: pointer to current record after
+				which the new record is inserted */
+	buf_block_t*	block,	/* in: buffer block */
+	dict_index_t*	index,	/* in: record descriptor */
+	rec_t*		rec,	/* in: inserted record */
+	page_t*		page,	/* in: uncompressed page */
+	page_zip_des_t*	page_zip,/* in: compressed page */
+	mtr_t*		mtr)	/* in: mini-transaction, or NULL */
+{
+	ulint		pos;
+
+	/* Recompress or reorganize and recompress the page. */
+	if (UNIV_LIKELY(page_zip_compress(page_zip, page, index, mtr))) {
+		return(rec);
+	}
+
+	/* Before trying to reorganize the page,
+	store the number of preceding records on the page. */
+	pos = page_rec_get_n_recs_before(rec);
+
+	if (page_zip_reorganize(block, index, mtr)) {
+		/* The page was reorganized: Find rec by seeking to pos,
+		and update *current_rec. */
+		rec = page + PAGE_NEW_INFIMUM;
+
+		while (--pos) {
+			rec = page + rec_get_next_offs(rec, TRUE);
+		}
+
+		*current_rec = rec;
+		rec = page + rec_get_next_offs(rec, TRUE);
+
+		return(rec);
+	}
+
+	/* Out of space: restore the page */
+	if (!page_zip_decompress(page_zip, page)) {
+		ut_error; /* Memory corrupted? */
+	}
+	ut_ad(page_validate(page, index));
+	return(NULL);
+}
+
+/***************************************************************
+Inserts a record next to page cursor on a compressed and uncompressed
+page. Returns pointer to inserted record if succeed, i.e.,
+enough space available, NULL otherwise.
+The cursor stays at the same position. */
+UNIV_INTERN
+rec_t*
+page_cur_insert_rec_zip(
+/*====================*/
+				/* out: pointer to record if succeed, NULL
+				otherwise */
+	rec_t**		current_rec,/* in/out: pointer to current record after
+				which the new record is inserted */
+	buf_block_t*	block,	/* in: buffer block of *current_rec */
+	dict_index_t*	index,	/* in: record descriptor */
+	const rec_t*	rec,	/* in: pointer to a physical record */
+	ulint*		offsets,/* in/out: rec_get_offsets(rec, index) */
+	mtr_t*		mtr)	/* in: mini-transaction handle, or NULL */
+{
+	byte*		insert_buf;
+	ulint		rec_size;
+	page_t*		page;		/* the relevant page */
+	rec_t*		last_insert;	/* cursor position at previous
+					insert */
+	rec_t*		free_rec;	/* a free record that was reused,
+					or NULL */
+	rec_t*		insert_rec;	/* inserted record */
+	ulint		heap_no;	/* heap number of the inserted
+					record */
+	page_zip_des_t*	page_zip;
+
+	page_zip = buf_block_get_page_zip(block);
+	ut_ad(page_zip);
+
+	ut_ad(rec_offs_validate(rec, index, offsets));
+
+	page = page_align(*current_rec);
+	ut_ad(dict_table_is_comp(index->table));
+	ut_ad(page_is_comp(page));
+
+	ut_ad(!page_rec_is_supremum(*current_rec));
+#ifdef UNIV_ZIP_DEBUG
+	ut_a(page_zip_validate(page_zip, page));
+#endif /* UNIV_ZIP_DEBUG */
+
+	/* 1. Get the size of the physical record in the page */
+	rec_size = rec_offs_size(offsets);
+
+#ifdef UNIV_DEBUG_VALGRIND
+	{
+		const void*	rec_start
+			= rec - rec_offs_extra_size(offsets);
+		ulint		extra_size
+			= rec_offs_extra_size(offsets)
+			- (rec_offs_comp(offsets)
+			   ? REC_N_NEW_EXTRA_BYTES
+			   : REC_N_OLD_EXTRA_BYTES);
+
+		/* All data bytes of the record must be valid. */
+		UNIV_MEM_ASSERT_RW(rec, rec_offs_data_size(offsets));
+		/* The variable-length header must be valid. */
+		UNIV_MEM_ASSERT_RW(rec_start, extra_size);
+	}
+#endif /* UNIV_DEBUG_VALGRIND */
+
+	/* 2. Try to find suitable space from page memory management */
+	if (!page_zip_available(page_zip, dict_index_is_clust(index),
+				rec_size, 1)) {
+
+		/* Try compressing the whole page afterwards. */
+		insert_rec = page_cur_insert_rec_low(*current_rec,
+						     index, rec, offsets,
+						     NULL);
+
+		if (UNIV_LIKELY(insert_rec != NULL)) {
+			insert_rec = page_cur_insert_rec_zip_reorg(
+				current_rec, block, index, insert_rec,
+				page, page_zip, mtr);
+		}
+
+		return(insert_rec);
+	}
+
+	free_rec = page_header_get_ptr(page, PAGE_FREE);
+	if (UNIV_LIKELY_NULL(free_rec)) {
+		/* Try to allocate from the head of the free list. */
+		lint	extra_size_diff;
+		ulint		foffsets_[REC_OFFS_NORMAL_SIZE];
+		ulint*		foffsets	= foffsets_;
+		mem_heap_t*	heap		= NULL;
+
+		rec_offs_init(foffsets_);
+
+		foffsets = rec_get_offsets(free_rec, index, foffsets,
+					ULINT_UNDEFINED, &heap);
+		if (rec_offs_size(foffsets) < rec_size) {
+too_small:
+			if (UNIV_LIKELY_NULL(heap)) {
+				mem_heap_free(heap);
+			}
+
+			goto use_heap;
+		}
+
+		insert_buf = free_rec - rec_offs_extra_size(foffsets);
+
+		/* On compressed pages, do not relocate records from
+		the free list.  If extra_size would grow, use the heap. */
+		extra_size_diff
+			= rec_offs_extra_size(offsets)
+			- rec_offs_extra_size(foffsets);
+
+		if (UNIV_UNLIKELY(extra_size_diff < 0)) {
+			/* Add an offset to the extra_size. */
+			if (rec_offs_size(foffsets)
+			    < rec_size - extra_size_diff) {
+
+				goto too_small;
+			}
+
+			insert_buf -= extra_size_diff;
+		} else if (UNIV_UNLIKELY(extra_size_diff)) {
+			/* Do not allow extra_size to grow */
+
+			goto too_small;
+		}
+
+		heap_no = rec_get_heap_no_new(free_rec);
+		page_mem_alloc_free(page, page_zip,
+				    rec_get_next_ptr(free_rec, TRUE),
+				    rec_size);
+
+		if (!page_is_leaf(page)) {
+			/* Zero out the node pointer of free_rec,
+			in case it will not be overwritten by
+			insert_rec. */
+
+			ut_ad(rec_size > REC_NODE_PTR_SIZE);
+
+			if (rec_offs_extra_size(foffsets)
+			    + rec_offs_data_size(foffsets) > rec_size) {
+
+				memset(rec_get_end(free_rec, foffsets)
+				       - REC_NODE_PTR_SIZE, 0,
+				       REC_NODE_PTR_SIZE);
+			}
+		} else if (dict_index_is_clust(index)) {
+			/* Zero out the DB_TRX_ID and DB_ROLL_PTR
+			columns of free_rec, in case it will not be
+			overwritten by insert_rec. */
+
+			ulint	trx_id_col;
+			ulint	trx_id_offs;
+			ulint	len;
+
+			trx_id_col = dict_index_get_sys_col_pos(index,
+								DATA_TRX_ID);
+			ut_ad(trx_id_col > 0);
+			ut_ad(trx_id_col != ULINT_UNDEFINED);
+
+			trx_id_offs = rec_get_nth_field_offs(foffsets,
+							     trx_id_col, &len);
+			ut_ad(len == DATA_TRX_ID_LEN);
+
+			if (DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN + trx_id_offs
+			    + rec_offs_extra_size(foffsets) > rec_size) {
+				/* We will have to zero out the
+				DB_TRX_ID and DB_ROLL_PTR, because
+				they will not be fully overwritten by
+				insert_rec. */
+
+				memset(free_rec + trx_id_offs, 0,
+				       DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN);
+			}
+
+			ut_ad(free_rec + trx_id_offs + DATA_TRX_ID_LEN
+			      == rec_get_nth_field(free_rec, foffsets,
+						   trx_id_col + 1, &len));
+			ut_ad(len == DATA_ROLL_PTR_LEN);
+		}
+
+		if (UNIV_LIKELY_NULL(heap)) {
+			mem_heap_free(heap);
+		}
+	} else {
+use_heap:
+		free_rec = NULL;
+		insert_buf = page_mem_alloc_heap(page, page_zip,
+						 rec_size, &heap_no);
+
+		if (UNIV_UNLIKELY(insert_buf == NULL)) {
+			return(NULL);
+		}
+
+		page_zip_dir_add_slot(page_zip, dict_index_is_clust(index));
+	}
+
+	/* 3. Create the record */
+	insert_rec = rec_copy(insert_buf, rec, offsets);
+	rec_offs_make_valid(insert_rec, index, offsets);
+
+	/* 4. Insert the record in the linked list of records */
+	ut_ad(*current_rec != insert_rec);
+
+	{
+		/* next record after current before the insertion */
+		rec_t*	next_rec = page_rec_get_next(*current_rec);
+		ut_ad(rec_get_status(*current_rec)
+		      <= REC_STATUS_INFIMUM);
+		ut_ad(rec_get_status(insert_rec) < REC_STATUS_INFIMUM);
+		ut_ad(rec_get_status(next_rec) != REC_STATUS_INFIMUM);
+
+		page_rec_set_next(insert_rec, next_rec);
+		page_rec_set_next(*current_rec, insert_rec);
+	}
+
+	page_header_set_field(page, page_zip, PAGE_N_RECS,
+			      1 + page_get_n_recs(page));
+
+	/* 5. Set the n_owned field in the inserted record to zero,
+	and set the heap_no field */
+	rec_set_n_owned_new(insert_rec, NULL, 0);
+	rec_set_heap_no_new(insert_rec, heap_no);
+
+	UNIV_MEM_ASSERT_RW(rec_get_start(insert_rec, offsets),
+			   rec_offs_size(offsets));
+
+	page_zip_dir_insert(page_zip, *current_rec, free_rec, insert_rec);
+
+	/* 6. Update the last insertion info in page header */
+
+	last_insert = page_header_get_ptr(page, PAGE_LAST_INSERT);
+	ut_ad(!last_insert
+	      || rec_get_node_ptr_flag(last_insert)
+	      == rec_get_node_ptr_flag(insert_rec));
+
+	if (UNIV_UNLIKELY(last_insert == NULL)) {
+		page_header_set_field(page, page_zip, PAGE_DIRECTION,
+							PAGE_NO_DIRECTION);
+		page_header_set_field(page, page_zip, PAGE_N_DIRECTION, 0);
+
+	} else if ((last_insert == *current_rec)
+		   && (page_header_get_field(page, PAGE_DIRECTION)
+		       != PAGE_LEFT)) {
+
+		page_header_set_field(page, page_zip, PAGE_DIRECTION,
+							PAGE_RIGHT);
+		page_header_set_field(page, page_zip, PAGE_N_DIRECTION,
+				      page_header_get_field(
+					      page, PAGE_N_DIRECTION) + 1);
+
+	} else if ((page_rec_get_next(insert_rec) == last_insert)
+		   && (page_header_get_field(page, PAGE_DIRECTION)
+		       != PAGE_RIGHT)) {
+
+		page_header_set_field(page, page_zip, PAGE_DIRECTION,
+							PAGE_LEFT);
+		page_header_set_field(page, page_zip, PAGE_N_DIRECTION,
+				      page_header_get_field(
+					      page, PAGE_N_DIRECTION) + 1);
+	} else {
+		page_header_set_field(page, page_zip, PAGE_DIRECTION,
+							PAGE_NO_DIRECTION);
+		page_header_set_field(page, page_zip, PAGE_N_DIRECTION, 0);
+	}
+
+	page_header_set_ptr(page, page_zip, PAGE_LAST_INSERT, insert_rec);
+
+	/* 7. It remains to update the owner record. */
+	{
+		rec_t*	owner_rec	= page_rec_find_owner_rec(insert_rec);
+		ulint	n_owned;
+
+		n_owned = rec_get_n_owned_new(owner_rec);
+		rec_set_n_owned_new(owner_rec, page_zip, n_owned + 1);
+
+		/* 8. Now we have incremented the n_owned field of the owner
+		record. If the number exceeds PAGE_DIR_SLOT_MAX_N_OWNED,
+		we have to split the corresponding directory slot in two. */
+
+		if (UNIV_UNLIKELY(n_owned == PAGE_DIR_SLOT_MAX_N_OWNED)) {
+			page_dir_split_slot(
+				page, page_zip,
+				page_dir_find_owner_slot(owner_rec));
+		}
+	}
+
+	page_zip_write_rec(page_zip, insert_rec, index, offsets, 1);
+
+	/* 9. Write log record of the insert */
+	if (UNIV_LIKELY(mtr != NULL)) {
+		page_cur_insert_rec_write_log(insert_rec, rec_size,
+					      *current_rec, index, mtr);
+	}
+
+	return(insert_rec);
+}
+
+/**************************************************************
+Writes a log record of copying a record list end to a new created page. */
+UNIV_INLINE
+byte*
+page_copy_rec_list_to_created_page_write_log(
+/*=========================================*/
+				/* out: 4-byte field where to
+				write the log data length,
+				or NULL if logging is disabled */
+	page_t*		page,	/* in: index page */
+	dict_index_t*	index,	/* in: record descriptor */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	byte*	log_ptr;
+
+	ut_ad(!!page_is_comp(page) == dict_table_is_comp(index->table));
+
+	log_ptr = mlog_open_and_write_index(mtr, page, index,
+					    page_is_comp(page)
+					    ? MLOG_COMP_LIST_END_COPY_CREATED
+					    : MLOG_LIST_END_COPY_CREATED, 4);
+	if (UNIV_LIKELY(log_ptr != NULL)) {
+		mlog_close(mtr, log_ptr + 4);
+	}
+
+	return(log_ptr);
+}
+
+/**************************************************************
+Parses a log record of copying a record list end to a new created page. */
+UNIV_INTERN
+byte*
+page_parse_copy_rec_list_to_created_page(
+/*=====================================*/
+				/* out: end of log record or NULL */
+	byte*		ptr,	/* in: buffer */
+	byte*		end_ptr,/* in: buffer end */
+	buf_block_t*	block,	/* in: page or NULL */
+	dict_index_t*	index,	/* in: record descriptor */
+	mtr_t*		mtr)	/* in: mtr or NULL */
+{
+	byte*		rec_end;
+	ulint		log_data_len;
+	page_t*		page;
+	page_zip_des_t*	page_zip;
+
+	if (ptr + 4 > end_ptr) {
+
+		return(NULL);
+	}
+
+	log_data_len = mach_read_from_4(ptr);
+	ptr += 4;
+
+	rec_end = ptr + log_data_len;
+
+	if (rec_end > end_ptr) {
+
+		return(NULL);
+	}
+
+	if (!block) {
+
+		return(rec_end);
+	}
+
+	while (ptr < rec_end) {
+		ptr = page_cur_parse_insert_rec(TRUE, ptr, end_ptr,
+						block, index, mtr);
+	}
+
+	ut_a(ptr == rec_end);
+
+	page = buf_block_get_frame(block);
+	page_zip = buf_block_get_page_zip(block);
+
+	page_header_set_ptr(page, page_zip, PAGE_LAST_INSERT, NULL);
+	page_header_set_field(page, page_zip, PAGE_DIRECTION,
+							PAGE_NO_DIRECTION);
+	page_header_set_field(page, page_zip, PAGE_N_DIRECTION, 0);
+
+	return(rec_end);
+}
+
+/*****************************************************************
+Copies records from page to a newly created page, from a given record onward,
+including that record. Infimum and supremum records are not copied. */
+UNIV_INTERN
+void
+page_copy_rec_list_end_to_created_page(
+/*===================================*/
+	page_t*		new_page,	/* in/out: index page to copy to */
+	rec_t*		rec,		/* in: first record to copy */
+	dict_index_t*	index,		/* in: record descriptor */
+	mtr_t*		mtr)		/* in: mtr */
+{
+	page_dir_slot_t* slot = 0; /* remove warning */
+	byte*	heap_top;
+	rec_t*	insert_rec = 0; /* remove warning */
+	rec_t*	prev_rec;
+	ulint	count;
+	ulint	n_recs;
+	ulint	slot_index;
+	ulint	rec_size;
+	ulint	log_mode;
+	byte*	log_ptr;
+	ulint	log_data_len;
+	mem_heap_t*	heap		= NULL;
+	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint*		offsets		= offsets_;
+	rec_offs_init(offsets_);
+
+	ut_ad(page_dir_get_n_heap(new_page) == PAGE_HEAP_NO_USER_LOW);
+	ut_ad(page_align(rec) != new_page);
+	ut_ad(page_rec_is_comp(rec) == page_is_comp(new_page));
+
+	if (page_rec_is_infimum(rec)) {
+
+		rec = page_rec_get_next(rec);
+	}
+
+	if (page_rec_is_supremum(rec)) {
+
+		return;
+	}
+
+#ifdef UNIV_DEBUG
+	/* To pass the debug tests we have to set these dummy values
+	in the debug version */
+	page_dir_set_n_slots(new_page, NULL, UNIV_PAGE_SIZE / 2);
+	page_header_set_ptr(new_page, NULL, PAGE_HEAP_TOP,
+			    new_page + UNIV_PAGE_SIZE - 1);
+#endif
+
+	log_ptr = page_copy_rec_list_to_created_page_write_log(new_page,
+							       index, mtr);
+
+	log_data_len = dyn_array_get_data_size(&(mtr->log));
+
+	/* Individual inserts are logged in a shorter form */
+
+	log_mode = mtr_set_log_mode(mtr, MTR_LOG_SHORT_INSERTS);
+
+	prev_rec = page_get_infimum_rec(new_page);
+	if (page_is_comp(new_page)) {
+		heap_top = new_page + PAGE_NEW_SUPREMUM_END;
+	} else {
+		heap_top = new_page + PAGE_OLD_SUPREMUM_END;
+	}
+	count = 0;
+	slot_index = 0;
+	n_recs = 0;
+
+	do {
+		offsets = rec_get_offsets(rec, index, offsets,
+					  ULINT_UNDEFINED, &heap);
+		insert_rec = rec_copy(heap_top, rec, offsets);
+
+		if (page_is_comp(new_page)) {
+			rec_set_next_offs_new(prev_rec,
+					      page_offset(insert_rec));
+
+			rec_set_n_owned_new(insert_rec, NULL, 0);
+			rec_set_heap_no_new(insert_rec,
+					    PAGE_HEAP_NO_USER_LOW + n_recs);
+		} else {
+			rec_set_next_offs_old(prev_rec,
+					      page_offset(insert_rec));
+
+			rec_set_n_owned_old(insert_rec, 0);
+			rec_set_heap_no_old(insert_rec,
+					    PAGE_HEAP_NO_USER_LOW + n_recs);
+		}
+
+		count++;
+		n_recs++;
+
+		if (UNIV_UNLIKELY
+		    (count == (PAGE_DIR_SLOT_MAX_N_OWNED + 1) / 2)) {
+
+			slot_index++;
+
+			slot = page_dir_get_nth_slot(new_page, slot_index);
+
+			page_dir_slot_set_rec(slot, insert_rec);
+			page_dir_slot_set_n_owned(slot, NULL, count);
+
+			count = 0;
+		}
+
+		rec_size = rec_offs_size(offsets);
+
+		ut_ad(heap_top < new_page + UNIV_PAGE_SIZE);
+
+		heap_top += rec_size;
+
+		page_cur_insert_rec_write_log(insert_rec, rec_size, prev_rec,
+					      index, mtr);
+		prev_rec = insert_rec;
+		rec = page_rec_get_next(rec);
+	} while (!page_rec_is_supremum(rec));
+
+	if ((slot_index > 0) && (count + 1
+				 + (PAGE_DIR_SLOT_MAX_N_OWNED + 1) / 2
+				 <= PAGE_DIR_SLOT_MAX_N_OWNED)) {
+		/* We can merge the two last dir slots. This operation is
+		here to make this function imitate exactly the equivalent
+		task made using page_cur_insert_rec, which we use in database
+		recovery to reproduce the task performed by this function.
+		To be able to check the correctness of recovery, it is good
+		that it imitates exactly. */
+
+		count += (PAGE_DIR_SLOT_MAX_N_OWNED + 1) / 2;
+
+		page_dir_slot_set_n_owned(slot, NULL, 0);
+
+		slot_index--;
+	}
+
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+
+	log_data_len = dyn_array_get_data_size(&(mtr->log)) - log_data_len;
+
+	ut_a(log_data_len < 100 * UNIV_PAGE_SIZE);
+
+	if (UNIV_LIKELY(log_ptr != NULL)) {
+		mach_write_to_4(log_ptr, log_data_len);
+	}
+
+	if (page_is_comp(new_page)) {
+		rec_set_next_offs_new(insert_rec, PAGE_NEW_SUPREMUM);
+	} else {
+		rec_set_next_offs_old(insert_rec, PAGE_OLD_SUPREMUM);
+	}
+
+	slot = page_dir_get_nth_slot(new_page, 1 + slot_index);
+
+	page_dir_slot_set_rec(slot, page_get_supremum_rec(new_page));
+	page_dir_slot_set_n_owned(slot, NULL, count + 1);
+
+	page_dir_set_n_slots(new_page, NULL, 2 + slot_index);
+	page_header_set_ptr(new_page, NULL, PAGE_HEAP_TOP, heap_top);
+	page_dir_set_n_heap(new_page, NULL, PAGE_HEAP_NO_USER_LOW + n_recs);
+	page_header_set_field(new_page, NULL, PAGE_N_RECS, n_recs);
+
+	page_header_set_ptr(new_page, NULL, PAGE_LAST_INSERT, NULL);
+	page_header_set_field(new_page, NULL, PAGE_DIRECTION,
+							PAGE_NO_DIRECTION);
+	page_header_set_field(new_page, NULL, PAGE_N_DIRECTION, 0);
+
+	/* Restore the log mode */
+
+	mtr_set_log_mode(mtr, log_mode);
+}
+
+/***************************************************************
+Writes log record of a record delete on a page. */
+UNIV_INLINE
+void
+page_cur_delete_rec_write_log(
+/*==========================*/
+	rec_t*		rec,	/* in: record to be deleted */
+	dict_index_t*	index,	/* in: record descriptor */
+	mtr_t*		mtr)	/* in: mini-transaction handle */
+{
+	byte*	log_ptr;
+
+	ut_ad(!!page_rec_is_comp(rec) == dict_table_is_comp(index->table));
+
+	log_ptr = mlog_open_and_write_index(mtr, rec, index,
+					    page_rec_is_comp(rec)
+					    ? MLOG_COMP_REC_DELETE
+					    : MLOG_REC_DELETE, 2);
+
+	if (!log_ptr) {
+		/* Logging in mtr is switched off during crash recovery:
+		in that case mlog_open returns NULL */
+		return;
+	}
+
+	/* Write the cursor rec offset as a 2-byte ulint */
+	mach_write_to_2(log_ptr, page_offset(rec));
+
+	mlog_close(mtr, log_ptr + 2);
+}
+
+/***************************************************************
+Parses log record of a record delete on a page. */
+UNIV_INTERN
+byte*
+page_cur_parse_delete_rec(
+/*======================*/
+				/* out: pointer to record end or NULL */
+	byte*		ptr,	/* in: buffer */
+	byte*		end_ptr,/* in: buffer end */
+	buf_block_t*	block,	/* in: page or NULL */
+	dict_index_t*	index,	/* in: record descriptor */
+	mtr_t*		mtr)	/* in: mtr or NULL */
+{
+	ulint		offset;
+	page_cur_t	cursor;
+
+	if (end_ptr < ptr + 2) {
+
+		return(NULL);
+	}
+
+	/* Read the cursor rec offset as a 2-byte ulint */
+	offset = mach_read_from_2(ptr);
+	ptr += 2;
+
+	ut_a(offset <= UNIV_PAGE_SIZE);
+
+	if (block) {
+		page_t*		page		= buf_block_get_frame(block);
+		mem_heap_t*	heap		= NULL;
+		ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+		rec_t*		rec		= page + offset;
+		rec_offs_init(offsets_);
+
+		page_cur_position(rec, block, &cursor);
+		ut_ad(!buf_block_get_page_zip(block) || page_is_comp(page));
+
+		page_cur_delete_rec(&cursor, index,
+				    rec_get_offsets(rec, index, offsets_,
+						    ULINT_UNDEFINED, &heap),
+				    mtr);
+		if (UNIV_LIKELY_NULL(heap)) {
+			mem_heap_free(heap);
+		}
+	}
+
+	return(ptr);
+}
+
+/***************************************************************
+Deletes a record at the page cursor. The cursor is moved to the next
+record after the deleted one. */
+UNIV_INTERN
+void
+page_cur_delete_rec(
+/*================*/
+	page_cur_t*	cursor,	/* in/out: a page cursor */
+	dict_index_t*	index,	/* in: record descriptor */
+	const ulint*	offsets,/* in: rec_get_offsets(cursor->rec, index) */
+	mtr_t*		mtr)	/* in: mini-transaction handle */
+{
+	page_dir_slot_t* cur_dir_slot;
+	page_dir_slot_t* prev_slot;
+	page_t*		page;
+	page_zip_des_t*	page_zip;
+	rec_t*		current_rec;
+	rec_t*		prev_rec	= NULL;
+	rec_t*		next_rec;
+	ulint		cur_slot_no;
+	ulint		cur_n_owned;
+	rec_t*		rec;
+
+	ut_ad(cursor && mtr);
+
+	page = page_cur_get_page(cursor);
+	page_zip = page_cur_get_page_zip(cursor);
+
+	/* page_zip_validate() will fail here when
+	btr_cur_pessimistic_delete() invokes btr_set_min_rec_mark().
+	Then, both "page_zip" and "page" would have the min-rec-mark
+	set on the smallest user record, but "page" would additionally
+	have it set on the smallest-but-one record.  Because sloppy
+	page_zip_validate_low() only ignores min-rec-flag differences
+	in the smallest user record, it cannot be used here either. */
+
+	current_rec = cursor->rec;
+	ut_ad(rec_offs_validate(current_rec, index, offsets));
+	ut_ad(!!page_is_comp(page) == dict_table_is_comp(index->table));
+
+	/* The record must not be the supremum or infimum record. */
+	ut_ad(page_rec_is_user_rec(current_rec));
+
+	/* Save to local variables some data associated with current_rec */
+	cur_slot_no = page_dir_find_owner_slot(current_rec);
+	cur_dir_slot = page_dir_get_nth_slot(page, cur_slot_no);
+	cur_n_owned = page_dir_slot_get_n_owned(cur_dir_slot);
+
+	/* 0. Write the log record */
+	page_cur_delete_rec_write_log(current_rec, index, mtr);
+
+	/* 1. Reset the last insert info in the page header and increment
+	the modify clock for the frame */
+
+	page_header_set_ptr(page, page_zip, PAGE_LAST_INSERT, NULL);
+
+	/* The page gets invalid for optimistic searches: increment the
+	frame modify clock */
+
+	buf_block_modify_clock_inc(page_cur_get_block(cursor));
+
+	/* 2. Find the next and the previous record. Note that the cursor is
+	left at the next record. */
+
+	ut_ad(cur_slot_no > 0);
+	prev_slot = page_dir_get_nth_slot(page, cur_slot_no - 1);
+
+	rec = (rec_t*) page_dir_slot_get_rec(prev_slot);
+
+	/* rec now points to the record of the previous directory slot. Look
+	for the immediate predecessor of current_rec in a loop. */
+
+	while(current_rec != rec) {
+		prev_rec = rec;
+		rec = page_rec_get_next(rec);
+	}
+
+	page_cur_move_to_next(cursor);
+	next_rec = cursor->rec;
+
+	/* 3. Remove the record from the linked list of records */
+
+	page_rec_set_next(prev_rec, next_rec);
+
+	/* 4. If the deleted record is pointed to by a dir slot, update the
+	record pointer in slot. In the following if-clause we assume that
+	prev_rec is owned by the same slot, i.e., PAGE_DIR_SLOT_MIN_N_OWNED
+	>= 2. */
+
+#if PAGE_DIR_SLOT_MIN_N_OWNED < 2
+# error "PAGE_DIR_SLOT_MIN_N_OWNED < 2"
+#endif
+	ut_ad(cur_n_owned > 1);
+
+	if (current_rec == page_dir_slot_get_rec(cur_dir_slot)) {
+		page_dir_slot_set_rec(cur_dir_slot, prev_rec);
+	}
+
+	/* 5. Update the number of owned records of the slot */
+
+	page_dir_slot_set_n_owned(cur_dir_slot, page_zip, cur_n_owned - 1);
+
+	/* 6. Free the memory occupied by the record */
+	page_mem_free(page, page_zip, current_rec, index, offsets);
+
+	/* 7. Now we have decremented the number of owned records of the slot.
+	If the number drops below PAGE_DIR_SLOT_MIN_N_OWNED, we balance the
+	slots. */
+
+	if (UNIV_UNLIKELY(cur_n_owned <= PAGE_DIR_SLOT_MIN_N_OWNED)) {
+		page_dir_balance_slot(page, page_zip, cur_slot_no);
+	}
+
+#ifdef UNIV_ZIP_DEBUG
+	ut_a(!page_zip || page_zip_validate(page_zip, page));
+#endif /* UNIV_ZIP_DEBUG */
+}
diff --git a/storage/xtradb/page/page0page.c b/storage/xtradb/page/page0page.c
new file mode 100644
index 00000000000..0ce532068ce
--- /dev/null
+++ b/storage/xtradb/page/page0page.c
@@ -0,0 +1,2568 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Index page routines
+
+Created 2/2/1994 Heikki Tuuri
+*******************************************************/
+
+#define THIS_MODULE
+#include "page0page.h"
+#ifdef UNIV_NONINL
+#include "page0page.ic"
+#endif
+#undef THIS_MODULE
+
+#include "page0cur.h"
+#include "page0zip.h"
+#include "lock0lock.h"
+#include "fut0lst.h"
+#include "btr0sea.h"
+#include "buf0buf.h"
+#include "srv0srv.h"
+#include "btr0btr.h"
+
+/*			THE INDEX PAGE
+			==============
+
+The index page consists of a page header which contains the page's
+id and other information. On top of it are the the index records
+in a heap linked into a one way linear list according to alphabetic order.
+
+Just below page end is an array of pointers which we call page directory,
+to about every sixth record in the list. The pointers are placed in
+the directory in the alphabetical order of the records pointed to,
+enabling us to make binary search using the array. Each slot n:o I
+in the directory points to a record, where a 4-bit field contains a count
+of those records which are in the linear list between pointer I and
+the pointer I - 1 in the directory, including the record
+pointed to by pointer I and not including the record pointed to by I - 1.
+We say that the record pointed to by slot I, or that slot I, owns
+these records. The count is always kept in the range 4 to 8, with
+the exception that it is 1 for the first slot, and 1--8 for the second slot.
+
+An essentially binary search can be performed in the list of index
+records, like we could do if we had pointer to every record in the
+page directory. The data structure is, however, more efficient when
+we are doing inserts, because most inserts are just pushed on a heap.
+Only every 8th insert requires block move in the directory pointer
+table, which itself is quite small. A record is deleted from the page
+by just taking it off the linear list and updating the number of owned
+records-field of the record which owns it, and updating the page directory,
+if necessary. A special case is the one when the record owns itself.
+Because the overhead of inserts is so small, we may also increase the
+page size from the projected default of 8 kB to 64 kB without too
+much loss of efficiency in inserts. Bigger page becomes actual
+when the disk transfer rate compared to seek and latency time rises.
+On the present system, the page size is set so that the page transfer
+time (3 ms) is 20 % of the disk random access time (15 ms).
+
+When the page is split, merged, or becomes full but contains deleted
+records, we have to reorganize the page.
+
+Assuming a page size of 8 kB, a typical index page of a secondary
+index contains 300 index entries, and the size of the page directory
+is 50 x 4 bytes = 200 bytes. */
+
+/*******************************************************************
+Looks for the directory slot which owns the given record. */
+UNIV_INTERN
+ulint
+page_dir_find_owner_slot(
+/*=====================*/
+				/* out: the directory slot number */
+	const rec_t*	rec)	/* in: the physical record */
+{
+	const page_t*			page;
+	register uint16			rec_offs_bytes;
+	register const page_dir_slot_t*	slot;
+	register const page_dir_slot_t*	first_slot;
+	register const rec_t*		r = rec;
+
+	ut_ad(page_rec_check(rec));
+
+	page = page_align(rec);
+	first_slot = page_dir_get_nth_slot(page, 0);
+	slot = page_dir_get_nth_slot(page, page_dir_get_n_slots(page) - 1);
+
+	if (page_is_comp(page)) {
+		while (rec_get_n_owned_new(r) == 0) {
+			r = rec_get_next_ptr_const(r, TRUE);
+			ut_ad(r >= page + PAGE_NEW_SUPREMUM);
+			ut_ad(r < page + (UNIV_PAGE_SIZE - PAGE_DIR));
+		}
+	} else {
+		while (rec_get_n_owned_old(r) == 0) {
+			r = rec_get_next_ptr_const(r, FALSE);
+			ut_ad(r >= page + PAGE_OLD_SUPREMUM);
+			ut_ad(r < page + (UNIV_PAGE_SIZE - PAGE_DIR));
+		}
+	}
+
+	rec_offs_bytes = mach_encode_2(r - page);
+
+	while (UNIV_LIKELY(*(uint16*) slot != rec_offs_bytes)) {
+
+		if (UNIV_UNLIKELY(slot == first_slot)) {
+			fprintf(stderr,
+				"InnoDB: Probable data corruption on"
+				" page %lu\n"
+				"InnoDB: Original record ",
+				(ulong) page_get_page_no(page));
+
+			if (page_is_comp(page)) {
+				fputs("(compact record)", stderr);
+			} else {
+				rec_print_old(stderr, rec);
+			}
+
+			fputs("\n"
+			      "InnoDB: on that page.\n"
+			      "InnoDB: Cannot find the dir slot for record ",
+			      stderr);
+			if (page_is_comp(page)) {
+				fputs("(compact record)", stderr);
+			} else {
+				rec_print_old(stderr, page
+					      + mach_decode_2(rec_offs_bytes));
+			}
+			fputs("\n"
+			      "InnoDB: on that page!\n", stderr);
+
+			buf_page_print(page, 0);
+
+			ut_error;
+		}
+
+		slot += PAGE_DIR_SLOT_SIZE;
+	}
+
+	return(((ulint) (first_slot - slot)) / PAGE_DIR_SLOT_SIZE);
+}
+
+/******************************************************************
+Used to check the consistency of a directory slot. */
+static
+ibool
+page_dir_slot_check(
+/*================*/
+					/* out: TRUE if succeed */
+	page_dir_slot_t*	slot)	/* in: slot */
+{
+	page_t*	page;
+	ulint	n_slots;
+	ulint	n_owned;
+
+	ut_a(slot);
+
+	page = page_align(slot);
+
+	n_slots = page_dir_get_n_slots(page);
+
+	ut_a(slot <= page_dir_get_nth_slot(page, 0));
+	ut_a(slot >= page_dir_get_nth_slot(page, n_slots - 1));
+
+	ut_a(page_rec_check(page_dir_slot_get_rec(slot)));
+
+	if (page_is_comp(page)) {
+		n_owned = rec_get_n_owned_new(page_dir_slot_get_rec(slot));
+	} else {
+		n_owned = rec_get_n_owned_old(page_dir_slot_get_rec(slot));
+	}
+
+	if (slot == page_dir_get_nth_slot(page, 0)) {
+		ut_a(n_owned == 1);
+	} else if (slot == page_dir_get_nth_slot(page, n_slots - 1)) {
+		ut_a(n_owned >= 1);
+		ut_a(n_owned <= PAGE_DIR_SLOT_MAX_N_OWNED);
+	} else {
+		ut_a(n_owned >= PAGE_DIR_SLOT_MIN_N_OWNED);
+		ut_a(n_owned <= PAGE_DIR_SLOT_MAX_N_OWNED);
+	}
+
+	return(TRUE);
+}
+
+/*****************************************************************
+Sets the max trx id field value. */
+UNIV_INTERN
+void
+page_set_max_trx_id(
+/*================*/
+	buf_block_t*	block,	/* in/out: page */
+	page_zip_des_t*	page_zip,/* in/out: compressed page, or NULL */
+	dulint		trx_id)	/* in: transaction id */
+{
+	const ibool	is_hashed	= block->is_hashed;
+	page_t*		page		= buf_block_get_frame(block);
+
+	if (is_hashed) {
+		rw_lock_x_lock(&btr_search_latch);
+	}
+
+	/* It is not necessary to write this change to the redo log, as
+	during a database recovery we assume that the max trx id of every
+	page is the maximum trx id assigned before the crash. */
+
+	mach_write_to_8(page + (PAGE_HEADER + PAGE_MAX_TRX_ID), trx_id);
+	if (UNIV_LIKELY_NULL(page_zip)) {
+		page_zip_write_header(page_zip,
+				      page + (PAGE_HEADER + PAGE_MAX_TRX_ID),
+				      8, NULL);
+	}
+
+	if (is_hashed) {
+		rw_lock_x_unlock(&btr_search_latch);
+	}
+}
+
+/****************************************************************
+Allocates a block of memory from the heap of an index page. */
+UNIV_INTERN
+byte*
+page_mem_alloc_heap(
+/*================*/
+				/* out: pointer to start of allocated
+				buffer, or NULL if allocation fails */
+	page_t*		page,	/* in/out: index page */
+	page_zip_des_t*	page_zip,/* in/out: compressed page with enough
+				space available for inserting the record,
+				or NULL */
+	ulint		need,	/* in: total number of bytes needed */
+	ulint*		heap_no)/* out: this contains the heap number
+				of the allocated record
+				if allocation succeeds */
+{
+	byte*	block;
+	ulint	avl_space;
+
+	ut_ad(page && heap_no);
+
+	avl_space = page_get_max_insert_size(page, 1);
+
+	if (avl_space >= need) {
+		block = page_header_get_ptr(page, PAGE_HEAP_TOP);
+
+		page_header_set_ptr(page, page_zip, PAGE_HEAP_TOP,
+				    block + need);
+		*heap_no = page_dir_get_n_heap(page);
+
+		page_dir_set_n_heap(page, page_zip, 1 + *heap_no);
+
+		return(block);
+	}
+
+	return(NULL);
+}
+
+/**************************************************************
+Writes a log record of page creation. */
+UNIV_INLINE
+void
+page_create_write_log(
+/*==================*/
+	buf_frame_t*	frame,	/* in: a buffer frame where the page is
+				created */
+	mtr_t*		mtr,	/* in: mini-transaction handle */
+	ibool		comp)	/* in: TRUE=compact page format */
+{
+	mlog_write_initial_log_record(frame, comp
+				      ? MLOG_COMP_PAGE_CREATE
+				      : MLOG_PAGE_CREATE, mtr);
+}
+
+/***************************************************************
+Parses a redo log record of creating a page. */
+UNIV_INTERN
+byte*
+page_parse_create(
+/*==============*/
+				/* out: end of log record or NULL */
+	byte*		ptr,	/* in: buffer */
+	byte*		end_ptr __attribute__((unused)), /* in: buffer end */
+	ulint		comp,	/* in: nonzero=compact page format */
+	buf_block_t*	block,	/* in: block or NULL */
+	mtr_t*		mtr)	/* in: mtr or NULL */
+{
+	ut_ad(ptr && end_ptr);
+
+	/* The record is empty, except for the record initial part */
+
+	if (block) {
+		page_create(block, mtr, comp);
+	}
+
+	return(ptr);
+}
+
+/**************************************************************
+The index page creation function. */
+static
+page_t*
+page_create_low(
+/*============*/
+					/* out: pointer to the page */
+	buf_block_t*	block,		/* in: a buffer block where the
+					page is created */
+	ulint		comp)		/* in: nonzero=compact page format */
+{
+	page_dir_slot_t* slot;
+	mem_heap_t*	heap;
+	dtuple_t*	tuple;
+	dfield_t*	field;
+	byte*		heap_top;
+	rec_t*		infimum_rec;
+	rec_t*		supremum_rec;
+	page_t*		page;
+	dict_index_t*	index;
+	ulint*		offsets;
+
+	ut_ad(block);
+#if PAGE_BTR_IBUF_FREE_LIST + FLST_BASE_NODE_SIZE > PAGE_DATA
+# error "PAGE_BTR_IBUF_FREE_LIST + FLST_BASE_NODE_SIZE > PAGE_DATA"
+#endif
+#if PAGE_BTR_IBUF_FREE_LIST_NODE + FLST_NODE_SIZE > PAGE_DATA
+# error "PAGE_BTR_IBUF_FREE_LIST_NODE + FLST_NODE_SIZE > PAGE_DATA"
+#endif
+
+	/* The infimum and supremum records use a dummy index. */
+	if (UNIV_LIKELY(comp)) {
+		index = srv_sys->dummy_ind2;
+	} else {
+		index = srv_sys->dummy_ind1;
+	}
+
+	/* 1. INCREMENT MODIFY CLOCK */
+	buf_block_modify_clock_inc(block);
+
+	page = buf_block_get_frame(block);
+
+	fil_page_set_type(page, FIL_PAGE_INDEX);
+
+	heap = mem_heap_create(200);
+
+	/* 3. CREATE THE INFIMUM AND SUPREMUM RECORDS */
+
+	/* Create first a data tuple for infimum record */
+	tuple = dtuple_create(heap, 1);
+	dtuple_set_info_bits(tuple, REC_STATUS_INFIMUM);
+	field = dtuple_get_nth_field(tuple, 0);
+
+	dfield_set_data(field, "infimum", 8);
+	dtype_set(dfield_get_type(field),
+		  DATA_VARCHAR, DATA_ENGLISH | DATA_NOT_NULL, 8);
+	/* Set the corresponding physical record to its place in the page
+	record heap */
+
+	heap_top = page + PAGE_DATA;
+
+	infimum_rec = rec_convert_dtuple_to_rec(heap_top, index, tuple, 0);
+
+	if (UNIV_LIKELY(comp)) {
+		ut_a(infimum_rec == page + PAGE_NEW_INFIMUM);
+
+		rec_set_n_owned_new(infimum_rec, NULL, 1);
+		rec_set_heap_no_new(infimum_rec, 0);
+	} else {
+		ut_a(infimum_rec == page + PAGE_OLD_INFIMUM);
+
+		rec_set_n_owned_old(infimum_rec, 1);
+		rec_set_heap_no_old(infimum_rec, 0);
+	}
+
+	offsets = rec_get_offsets(infimum_rec, index, NULL,
+				  ULINT_UNDEFINED, &heap);
+
+	heap_top = rec_get_end(infimum_rec, offsets);
+
+	/* Create then a tuple for supremum */
+
+	tuple = dtuple_create(heap, 1);
+	dtuple_set_info_bits(tuple, REC_STATUS_SUPREMUM);
+	field = dtuple_get_nth_field(tuple, 0);
+
+	dfield_set_data(field, "supremum", comp ? 8 : 9);
+	dtype_set(dfield_get_type(field),
+		  DATA_VARCHAR, DATA_ENGLISH | DATA_NOT_NULL, comp ? 8 : 9);
+
+	supremum_rec = rec_convert_dtuple_to_rec(heap_top, index, tuple, 0);
+
+	if (UNIV_LIKELY(comp)) {
+		ut_a(supremum_rec == page + PAGE_NEW_SUPREMUM);
+
+		rec_set_n_owned_new(supremum_rec, NULL, 1);
+		rec_set_heap_no_new(supremum_rec, 1);
+	} else {
+		ut_a(supremum_rec == page + PAGE_OLD_SUPREMUM);
+
+		rec_set_n_owned_old(supremum_rec, 1);
+		rec_set_heap_no_old(supremum_rec, 1);
+	}
+
+	offsets = rec_get_offsets(supremum_rec, index, offsets,
+				  ULINT_UNDEFINED, &heap);
+	heap_top = rec_get_end(supremum_rec, offsets);
+
+	ut_ad(heap_top == page
+	      + (comp ? PAGE_NEW_SUPREMUM_END : PAGE_OLD_SUPREMUM_END));
+
+	mem_heap_free(heap);
+
+	/* 4. INITIALIZE THE PAGE */
+
+	page_header_set_field(page, NULL, PAGE_N_DIR_SLOTS, 2);
+	page_header_set_ptr(page, NULL, PAGE_HEAP_TOP, heap_top);
+	page_header_set_field(page, NULL, PAGE_N_HEAP, comp
+			      ? 0x8000 | PAGE_HEAP_NO_USER_LOW
+			      : PAGE_HEAP_NO_USER_LOW);
+	page_header_set_ptr(page, NULL, PAGE_FREE, NULL);
+	page_header_set_field(page, NULL, PAGE_GARBAGE, 0);
+	page_header_set_ptr(page, NULL, PAGE_LAST_INSERT, NULL);
+	page_header_set_field(page, NULL, PAGE_DIRECTION, PAGE_NO_DIRECTION);
+	page_header_set_field(page, NULL, PAGE_N_DIRECTION, 0);
+	page_header_set_field(page, NULL, PAGE_N_RECS, 0);
+	page_set_max_trx_id(block, NULL, ut_dulint_zero);
+	memset(heap_top, 0, UNIV_PAGE_SIZE - PAGE_EMPTY_DIR_START
+	       - page_offset(heap_top));
+
+	/* 5. SET POINTERS IN RECORDS AND DIR SLOTS */
+
+	/* Set the slots to point to infimum and supremum. */
+
+	slot = page_dir_get_nth_slot(page, 0);
+	page_dir_slot_set_rec(slot, infimum_rec);
+
+	slot = page_dir_get_nth_slot(page, 1);
+	page_dir_slot_set_rec(slot, supremum_rec);
+
+	/* Set the next pointers in infimum and supremum */
+
+	if (UNIV_LIKELY(comp)) {
+		rec_set_next_offs_new(infimum_rec, PAGE_NEW_SUPREMUM);
+		rec_set_next_offs_new(supremum_rec, 0);
+	} else {
+		rec_set_next_offs_old(infimum_rec, PAGE_OLD_SUPREMUM);
+		rec_set_next_offs_old(supremum_rec, 0);
+	}
+
+	return(page);
+}
+
+/**************************************************************
+Create an uncompressed B-tree index page. */
+UNIV_INTERN
+page_t*
+page_create(
+/*========*/
+					/* out: pointer to the page */
+	buf_block_t*	block,		/* in: a buffer block where the
+					page is created */
+	mtr_t*		mtr,		/* in: mini-transaction handle */
+	ulint		comp)		/* in: nonzero=compact page format */
+{
+	page_create_write_log(buf_block_get_frame(block), mtr, comp);
+	return(page_create_low(block, comp));
+}
+
+/**************************************************************
+Create a compressed B-tree index page. */
+UNIV_INTERN
+page_t*
+page_create_zip(
+/*============*/
+					/* out: pointer to the page */
+	buf_block_t*	block,		/* in/out: a buffer frame where the
+					page is created */
+	dict_index_t*	index,		/* in: the index of the page */
+	ulint		level,		/* in: the B-tree level of the page */
+	mtr_t*		mtr)		/* in: mini-transaction handle */
+{
+	page_t*		page;
+	page_zip_des_t*	page_zip	= buf_block_get_page_zip(block);
+
+	ut_ad(block);
+	ut_ad(page_zip);
+	ut_ad(index);
+	ut_ad(dict_table_is_comp(index->table));
+
+	page = page_create_low(block, TRUE);
+	mach_write_to_2(page + PAGE_HEADER + PAGE_LEVEL, level);
+
+	if (UNIV_UNLIKELY(!page_zip_compress(page_zip, page, index, mtr))) {
+		/* The compression of a newly created page
+		should always succeed. */
+		ut_error;
+	}
+
+	return(page);
+}
+
+/*****************************************************************
+Differs from page_copy_rec_list_end, because this function does not
+touch the lock table and max trx id on page or compress the page. */
+UNIV_INTERN
+void
+page_copy_rec_list_end_no_locks(
+/*============================*/
+	buf_block_t*	new_block,	/* in: index page to copy to */
+	buf_block_t*	block,		/* in: index page of rec */
+	rec_t*		rec,		/* in: record on page */
+	dict_index_t*	index,		/* in: record descriptor */
+	mtr_t*		mtr)		/* in: mtr */
+{
+	page_t*		new_page	= buf_block_get_frame(new_block);
+	page_cur_t	cur1;
+	rec_t*		cur2;
+	mem_heap_t*	heap		= NULL;
+	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint*		offsets		= offsets_;
+	rec_offs_init(offsets_);
+
+	page_cur_position(rec, block, &cur1);
+
+	if (page_cur_is_before_first(&cur1)) {
+
+		page_cur_move_to_next(&cur1);
+	}
+
+	ut_a((ibool)!!page_is_comp(new_page)
+	     == dict_table_is_comp(index->table));
+	ut_a(page_is_comp(new_page) == page_rec_is_comp(rec));
+	ut_a(mach_read_from_2(new_page + UNIV_PAGE_SIZE - 10) == (ulint)
+	     (page_is_comp(new_page) ? PAGE_NEW_INFIMUM : PAGE_OLD_INFIMUM));
+
+	cur2 = page_get_infimum_rec(buf_block_get_frame(new_block));
+
+	/* Copy records from the original page to the new page */
+
+	while (!page_cur_is_after_last(&cur1)) {
+		rec_t*	cur1_rec = page_cur_get_rec(&cur1);
+		rec_t*	ins_rec;
+		offsets = rec_get_offsets(cur1_rec, index, offsets,
+					  ULINT_UNDEFINED, &heap);
+		ins_rec = page_cur_insert_rec_low(cur2, index,
+						  cur1_rec, offsets, mtr);
+		if (UNIV_UNLIKELY(!ins_rec)) {
+			/* Track an assertion failure reported on the mailing
+			list on June 18th, 2003 */
+
+			buf_page_print(new_page, 0);
+			buf_page_print(page_align(rec), 0);
+			ut_print_timestamp(stderr);
+
+			fprintf(stderr,
+				"InnoDB: rec offset %lu, cur1 offset %lu,"
+				" cur2 offset %lu\n",
+				(ulong) page_offset(rec),
+				(ulong) page_offset(page_cur_get_rec(&cur1)),
+				(ulong) page_offset(cur2));
+			ut_error;
+		}
+
+		page_cur_move_to_next(&cur1);
+		cur2 = ins_rec;
+	}
+
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+}
+
+/*****************************************************************
+Copies records from page to new_page, from a given record onward,
+including that record. Infimum and supremum records are not copied.
+The records are copied to the start of the record list on new_page. */
+UNIV_INTERN
+rec_t*
+page_copy_rec_list_end(
+/*===================*/
+					/* out: pointer to the original
+					successor of the infimum record
+					on new_page, or NULL on zip overflow
+					(new_block will be decompressed) */
+	buf_block_t*	new_block,	/* in/out: index page to copy to */
+	buf_block_t*	block,		/* in: index page containing rec */
+	rec_t*		rec,		/* in: record on page */
+	dict_index_t*	index,		/* in: record descriptor */
+	mtr_t*		mtr)		/* in: mtr */
+{
+	page_t*		new_page	= buf_block_get_frame(new_block);
+	page_zip_des_t*	new_page_zip	= buf_block_get_page_zip(new_block);
+	page_t*		page		= page_align(rec);
+	rec_t*		ret		= page_rec_get_next(
+		page_get_infimum_rec(new_page));
+	ulint		log_mode	= 0; /* remove warning */
+
+#ifdef UNIV_ZIP_DEBUG
+	if (new_page_zip) {
+		page_zip_des_t*	page_zip = buf_block_get_page_zip(block);
+		ut_a(page_zip);
+
+		/* Strict page_zip_validate() may fail here.
+		Furthermore, btr_compress() may set FIL_PAGE_PREV to
+		FIL_NULL on new_page while leaving it intact on
+		new_page_zip.  So, we cannot validate new_page_zip. */
+		ut_a(page_zip_validate_low(page_zip, page, TRUE));
+	}
+#endif /* UNIV_ZIP_DEBUG */
+	ut_ad(buf_block_get_frame(block) == page);
+	ut_ad(page_is_leaf(page) == page_is_leaf(new_page));
+	ut_ad(page_is_comp(page) == page_is_comp(new_page));
+	/* Here, "ret" may be pointing to a user record or the
+	predefined supremum record. */
+
+	if (UNIV_LIKELY_NULL(new_page_zip)) {
+		log_mode = mtr_set_log_mode(mtr, MTR_LOG_NONE);
+	}
+
+	if (page_dir_get_n_heap(new_page) == PAGE_HEAP_NO_USER_LOW) {
+		page_copy_rec_list_end_to_created_page(new_page, rec,
+						       index, mtr);
+	} else {
+		page_copy_rec_list_end_no_locks(new_block, block, rec,
+						index, mtr);
+	}
+
+	if (UNIV_LIKELY_NULL(new_page_zip)) {
+		mtr_set_log_mode(mtr, log_mode);
+
+		if (UNIV_UNLIKELY
+		    (!page_zip_compress(new_page_zip, new_page, index, mtr))) {
+			/* Before trying to reorganize the page,
+			store the number of preceding records on the page. */
+			ulint	ret_pos
+				= page_rec_get_n_recs_before(ret);
+			/* Before copying, "ret" was the successor of
+			the predefined infimum record.  It must still
+			have at least one predecessor (the predefined
+			infimum record, or a freshly copied record
+			that is smaller than "ret"). */
+			ut_a(ret_pos > 0);
+
+			if (UNIV_UNLIKELY
+			    (!page_zip_reorganize(new_block, index, mtr))) {
+
+				if (UNIV_UNLIKELY
+				    (!page_zip_decompress(new_page_zip,
+							  new_page))) {
+					ut_error;
+				}
+				ut_ad(page_validate(new_page, index));
+				return(NULL);
+			} else {
+				/* The page was reorganized:
+				Seek to ret_pos. */
+				ret = new_page + PAGE_NEW_INFIMUM;
+
+				do {
+					ret = rec_get_next_ptr(ret, TRUE);
+				} while (--ret_pos);
+			}
+		}
+	}
+
+	/* Update the lock table, MAX_TRX_ID, and possible hash index */
+
+	lock_move_rec_list_end(new_block, block, rec);
+
+	page_update_max_trx_id(new_block, new_page_zip,
+			       page_get_max_trx_id(page));
+
+	btr_search_move_or_delete_hash_entries(new_block, block, index);
+
+	return(ret);
+}
+
+/*****************************************************************
+Copies records from page to new_page, up to the given record,
+NOT including that record. Infimum and supremum records are not copied.
+The records are copied to the end of the record list on new_page. */
+UNIV_INTERN
+rec_t*
+page_copy_rec_list_start(
+/*=====================*/
+					/* out: pointer to the original
+					predecessor of the supremum record
+					on new_page, or NULL on zip overflow
+					(new_block will be decompressed) */
+	buf_block_t*	new_block,	/* in/out: index page to copy to */
+	buf_block_t*	block,		/* in: index page containing rec */
+	rec_t*		rec,		/* in: record on page */
+	dict_index_t*	index,		/* in: record descriptor */
+	mtr_t*		mtr)		/* in: mtr */
+{
+	page_t*		new_page	= buf_block_get_frame(new_block);
+	page_zip_des_t*	new_page_zip	= buf_block_get_page_zip(new_block);
+	page_cur_t	cur1;
+	rec_t*		cur2;
+	ulint		log_mode	= 0 /* remove warning */;
+	mem_heap_t*	heap		= NULL;
+	rec_t*		ret
+		= page_rec_get_prev(page_get_supremum_rec(new_page));
+	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint*		offsets		= offsets_;
+	rec_offs_init(offsets_);
+
+	/* Here, "ret" may be pointing to a user record or the
+	predefined infimum record. */
+
+	if (page_rec_is_infimum(rec)) {
+
+		return(ret);
+	}
+
+	if (UNIV_LIKELY_NULL(new_page_zip)) {
+		log_mode = mtr_set_log_mode(mtr, MTR_LOG_NONE);
+	}
+
+	page_cur_set_before_first(block, &cur1);
+	page_cur_move_to_next(&cur1);
+
+	cur2 = ret;
+
+	/* Copy records from the original page to the new page */
+
+	while (page_cur_get_rec(&cur1) != rec) {
+		rec_t*	cur1_rec = page_cur_get_rec(&cur1);
+		offsets = rec_get_offsets(cur1_rec, index, offsets,
+					  ULINT_UNDEFINED, &heap);
+		cur2 = page_cur_insert_rec_low(cur2, index,
+					       cur1_rec, offsets, mtr);
+		ut_a(cur2);
+
+		page_cur_move_to_next(&cur1);
+	}
+
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+
+	if (UNIV_LIKELY_NULL(new_page_zip)) {
+		mtr_set_log_mode(mtr, log_mode);
+
+		if (UNIV_UNLIKELY
+		    (!page_zip_compress(new_page_zip, new_page, index, mtr))) {
+			/* Before trying to reorganize the page,
+			store the number of preceding records on the page. */
+			ulint	ret_pos
+				= page_rec_get_n_recs_before(ret);
+			/* Before copying, "ret" was the predecessor
+			of the predefined supremum record.  If it was
+			the predefined infimum record, then it would
+			still be the infimum.  Thus, the assertion
+			ut_a(ret_pos > 0) would fail here. */
+
+			if (UNIV_UNLIKELY
+			    (!page_zip_reorganize(new_block, index, mtr))) {
+
+				if (UNIV_UNLIKELY
+				    (!page_zip_decompress(new_page_zip,
+							  new_page))) {
+					ut_error;
+				}
+				ut_ad(page_validate(new_page, index));
+				return(NULL);
+			} else {
+				/* The page was reorganized:
+				Seek to ret_pos. */
+				ret = new_page + PAGE_NEW_INFIMUM;
+
+				do {
+					ret = rec_get_next_ptr(ret, TRUE);
+				} while (--ret_pos);
+			}
+		}
+	}
+
+	/* Update MAX_TRX_ID, the lock table, and possible hash index */
+
+	page_update_max_trx_id(new_block, new_page_zip,
+			       page_get_max_trx_id(page_align(rec)));
+
+	lock_move_rec_list_start(new_block, block, rec, ret);
+
+	btr_search_move_or_delete_hash_entries(new_block, block, index);
+
+	return(ret);
+}
+
+/**************************************************************
+Writes a log record of a record list end or start deletion. */
+UNIV_INLINE
+void
+page_delete_rec_list_write_log(
+/*===========================*/
+	rec_t*		rec,	/* in: record on page */
+	dict_index_t*	index,	/* in: record descriptor */
+	byte		type,	/* in: operation type:
+				MLOG_LIST_END_DELETE, ... */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	byte*	log_ptr;
+	ut_ad(type == MLOG_LIST_END_DELETE
+	      || type == MLOG_LIST_START_DELETE
+	      || type == MLOG_COMP_LIST_END_DELETE
+	      || type == MLOG_COMP_LIST_START_DELETE);
+
+	log_ptr = mlog_open_and_write_index(mtr, rec, index, type, 2);
+	if (log_ptr) {
+		/* Write the parameter as a 2-byte ulint */
+		mach_write_to_2(log_ptr, page_offset(rec));
+		mlog_close(mtr, log_ptr + 2);
+	}
+}
+
+/**************************************************************
+Parses a log record of a record list end or start deletion. */
+UNIV_INTERN
+byte*
+page_parse_delete_rec_list(
+/*=======================*/
+				/* out: end of log record or NULL */
+	byte		type,	/* in: MLOG_LIST_END_DELETE,
+				MLOG_LIST_START_DELETE,
+				MLOG_COMP_LIST_END_DELETE or
+				MLOG_COMP_LIST_START_DELETE */
+	byte*		ptr,	/* in: buffer */
+	byte*		end_ptr,/* in: buffer end */
+	buf_block_t*	block,	/* in/out: buffer block or NULL */
+	dict_index_t*	index,	/* in: record descriptor */
+	mtr_t*		mtr)	/* in: mtr or NULL */
+{
+	page_t*	page;
+	ulint	offset;
+
+	ut_ad(type == MLOG_LIST_END_DELETE
+	      || type == MLOG_LIST_START_DELETE
+	      || type == MLOG_COMP_LIST_END_DELETE
+	      || type == MLOG_COMP_LIST_START_DELETE);
+
+	/* Read the record offset as a 2-byte ulint */
+
+	if (end_ptr < ptr + 2) {
+
+		return(NULL);
+	}
+
+	offset = mach_read_from_2(ptr);
+	ptr += 2;
+
+	if (!block) {
+
+		return(ptr);
+	}
+
+	page = buf_block_get_frame(block);
+
+	ut_ad(!!page_is_comp(page) == dict_table_is_comp(index->table));
+
+	if (type == MLOG_LIST_END_DELETE
+	    || type == MLOG_COMP_LIST_END_DELETE) {
+		page_delete_rec_list_end(page + offset, block, index,
+					 ULINT_UNDEFINED, ULINT_UNDEFINED,
+					 mtr);
+	} else {
+		page_delete_rec_list_start(page + offset, block, index, mtr);
+	}
+
+	return(ptr);
+}
+
+/*****************************************************************
+Deletes records from a page from a given record onward, including that record.
+The infimum and supremum records are not deleted. */
+UNIV_INTERN
+void
+page_delete_rec_list_end(
+/*=====================*/
+	rec_t*		rec,	/* in: pointer to record on page */
+	buf_block_t*	block,	/* in: buffer block of the page */
+	dict_index_t*	index,	/* in: record descriptor */
+	ulint		n_recs,	/* in: number of records to delete,
+				or ULINT_UNDEFINED if not known */
+	ulint		size,	/* in: the sum of the sizes of the
+				records in the end of the chain to
+				delete, or ULINT_UNDEFINED if not known */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	page_dir_slot_t*slot;
+	ulint		slot_index;
+	rec_t*		last_rec;
+	rec_t*		prev_rec;
+	ulint		n_owned;
+	page_zip_des_t*	page_zip	= buf_block_get_page_zip(block);
+	page_t*		page		= page_align(rec);
+	mem_heap_t*	heap		= NULL;
+	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint*		offsets		= offsets_;
+	rec_offs_init(offsets_);
+
+	ut_ad(size == ULINT_UNDEFINED || size < UNIV_PAGE_SIZE);
+	ut_ad(!page_zip || page_rec_is_comp(rec));
+#ifdef UNIV_ZIP_DEBUG
+	ut_a(!page_zip || page_zip_validate(page_zip, page));
+#endif /* UNIV_ZIP_DEBUG */
+
+	if (page_rec_is_infimum(rec)) {
+		rec = page_rec_get_next(rec);
+	}
+
+	if (page_rec_is_supremum(rec)) {
+
+		return;
+	}
+
+	/* Reset the last insert info in the page header and increment
+	the modify clock for the frame */
+
+	page_header_set_ptr(page, page_zip, PAGE_LAST_INSERT, NULL);
+
+	/* The page gets invalid for optimistic searches: increment the
+	frame modify clock */
+
+	buf_block_modify_clock_inc(block);
+
+	page_delete_rec_list_write_log(rec, index, page_is_comp(page)
+				       ? MLOG_COMP_LIST_END_DELETE
+				       : MLOG_LIST_END_DELETE, mtr);
+
+	if (UNIV_LIKELY_NULL(page_zip)) {
+		ulint		log_mode;
+
+		ut_a(page_is_comp(page));
+		/* Individual deletes are not logged */
+
+		log_mode = mtr_set_log_mode(mtr, MTR_LOG_NONE);
+
+		do {
+			page_cur_t	cur;
+			page_cur_position(rec, block, &cur);
+
+			offsets = rec_get_offsets(rec, index, offsets,
+						  ULINT_UNDEFINED, &heap);
+			rec = rec_get_next_ptr(rec, TRUE);
+#ifdef UNIV_ZIP_DEBUG
+			ut_a(page_zip_validate(page_zip, page));
+#endif /* UNIV_ZIP_DEBUG */
+			page_cur_delete_rec(&cur, index, offsets, mtr);
+		} while (page_offset(rec) != PAGE_NEW_SUPREMUM);
+
+		if (UNIV_LIKELY_NULL(heap)) {
+			mem_heap_free(heap);
+		}
+
+		/* Restore log mode */
+
+		mtr_set_log_mode(mtr, log_mode);
+		return;
+	}
+
+	prev_rec = page_rec_get_prev(rec);
+
+	last_rec = page_rec_get_prev(page_get_supremum_rec(page));
+
+	if ((size == ULINT_UNDEFINED) || (n_recs == ULINT_UNDEFINED)) {
+		rec_t*		rec2		= rec;
+		/* Calculate the sum of sizes and the number of records */
+		size = 0;
+		n_recs = 0;
+
+		do {
+			ulint	s;
+			offsets = rec_get_offsets(rec2, index, offsets,
+						  ULINT_UNDEFINED, &heap);
+			s = rec_offs_size(offsets);
+			ut_ad(rec2 - page + s - rec_offs_extra_size(offsets)
+			      < UNIV_PAGE_SIZE);
+			ut_ad(size + s < UNIV_PAGE_SIZE);
+			size += s;
+			n_recs++;
+
+			rec2 = page_rec_get_next(rec2);
+		} while (!page_rec_is_supremum(rec2));
+
+		if (UNIV_LIKELY_NULL(heap)) {
+			mem_heap_free(heap);
+		}
+	}
+
+	ut_ad(size < UNIV_PAGE_SIZE);
+
+	/* Update the page directory; there is no need to balance the number
+	of the records owned by the supremum record, as it is allowed to be
+	less than PAGE_DIR_SLOT_MIN_N_OWNED */
+
+	if (page_is_comp(page)) {
+		rec_t*	rec2	= rec;
+		ulint	count	= 0;
+
+		while (rec_get_n_owned_new(rec2) == 0) {
+			count++;
+
+			rec2 = rec_get_next_ptr(rec2, TRUE);
+		}
+
+		ut_ad(rec_get_n_owned_new(rec2) > count);
+
+		n_owned = rec_get_n_owned_new(rec2) - count;
+		slot_index = page_dir_find_owner_slot(rec2);
+		slot = page_dir_get_nth_slot(page, slot_index);
+	} else {
+		rec_t*	rec2	= rec;
+		ulint	count	= 0;
+
+		while (rec_get_n_owned_old(rec2) == 0) {
+			count++;
+
+			rec2 = rec_get_next_ptr(rec2, FALSE);
+		}
+
+		ut_ad(rec_get_n_owned_old(rec2) > count);
+
+		n_owned = rec_get_n_owned_old(rec2) - count;
+		slot_index = page_dir_find_owner_slot(rec2);
+		slot = page_dir_get_nth_slot(page, slot_index);
+	}
+
+	page_dir_slot_set_rec(slot, page_get_supremum_rec(page));
+	page_dir_slot_set_n_owned(slot, NULL, n_owned);
+
+	page_dir_set_n_slots(page, NULL, slot_index + 1);
+
+	/* Remove the record chain segment from the record chain */
+	page_rec_set_next(prev_rec, page_get_supremum_rec(page));
+
+	/* Catenate the deleted chain segment to the page free list */
+
+	page_rec_set_next(last_rec, page_header_get_ptr(page, PAGE_FREE));
+	page_header_set_ptr(page, NULL, PAGE_FREE, rec);
+
+	page_header_set_field(page, NULL, PAGE_GARBAGE, size
+			      + page_header_get_field(page, PAGE_GARBAGE));
+
+	page_header_set_field(page, NULL, PAGE_N_RECS,
+			      (ulint)(page_get_n_recs(page) - n_recs));
+}
+
+/*****************************************************************
+Deletes records from page, up to the given record, NOT including
+that record. Infimum and supremum records are not deleted. */
+UNIV_INTERN
+void
+page_delete_rec_list_start(
+/*=======================*/
+	rec_t*		rec,	/* in: record on page */
+	buf_block_t*	block,	/* in: buffer block of the page */
+	dict_index_t*	index,	/* in: record descriptor */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	page_cur_t	cur1;
+	ulint		log_mode;
+	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint*		offsets		= offsets_;
+	mem_heap_t*	heap		= NULL;
+	byte		type;
+
+	rec_offs_init(offsets_);
+
+	ut_ad((ibool) !!page_rec_is_comp(rec)
+	      == dict_table_is_comp(index->table));
+#ifdef UNIV_ZIP_DEBUG
+	{
+		page_zip_des_t*	page_zip= buf_block_get_page_zip(block);
+		page_t*		page	= buf_block_get_frame(block);
+
+		/* page_zip_validate() would detect a min_rec_mark mismatch
+		in btr_page_split_and_insert()
+		between btr_attach_half_pages() and insert_page = ...
+		when btr_page_get_split_rec_to_left() holds
+		(direction == FSP_DOWN). */
+		ut_a(!page_zip || page_zip_validate_low(page_zip, page, TRUE));
+	}
+#endif /* UNIV_ZIP_DEBUG */
+
+	if (page_rec_is_infimum(rec)) {
+
+		return;
+	}
+
+	if (page_rec_is_comp(rec)) {
+		type = MLOG_COMP_LIST_START_DELETE;
+	} else {
+		type = MLOG_LIST_START_DELETE;
+	}
+
+	page_delete_rec_list_write_log(rec, index, type, mtr);
+
+	page_cur_set_before_first(block, &cur1);
+	page_cur_move_to_next(&cur1);
+
+	/* Individual deletes are not logged */
+
+	log_mode = mtr_set_log_mode(mtr, MTR_LOG_NONE);
+
+	while (page_cur_get_rec(&cur1) != rec) {
+		offsets = rec_get_offsets(page_cur_get_rec(&cur1), index,
+					  offsets, ULINT_UNDEFINED, &heap);
+		page_cur_delete_rec(&cur1, index, offsets, mtr);
+	}
+
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+
+	/* Restore log mode */
+
+	mtr_set_log_mode(mtr, log_mode);
+}
+
+/*****************************************************************
+Moves record list end to another page. Moved records include
+split_rec. */
+UNIV_INTERN
+ibool
+page_move_rec_list_end(
+/*===================*/
+					/* out: TRUE on success; FALSE on
+					compression failure
+					(new_block will be decompressed) */
+	buf_block_t*	new_block,	/* in/out: index page where to move */
+	buf_block_t*	block,		/* in: index page from where to move */
+	rec_t*		split_rec,	/* in: first record to move */
+	dict_index_t*	index,		/* in: record descriptor */
+	mtr_t*		mtr)		/* in: mtr */
+{
+	page_t*		new_page	= buf_block_get_frame(new_block);
+	ulint		old_data_size;
+	ulint		new_data_size;
+	ulint		old_n_recs;
+	ulint		new_n_recs;
+
+	old_data_size = page_get_data_size(new_page);
+	old_n_recs = page_get_n_recs(new_page);
+#ifdef UNIV_ZIP_DEBUG
+	{
+		page_zip_des_t*	new_page_zip
+			= buf_block_get_page_zip(new_block);
+		page_zip_des_t*	page_zip
+			= buf_block_get_page_zip(block);
+		ut_a(!new_page_zip == !page_zip);
+		ut_a(!new_page_zip
+		     || page_zip_validate(new_page_zip, new_page));
+		ut_a(!page_zip
+		     || page_zip_validate(page_zip, page_align(split_rec)));
+	}
+#endif /* UNIV_ZIP_DEBUG */
+
+	if (UNIV_UNLIKELY(!page_copy_rec_list_end(new_block, block,
+						  split_rec, index, mtr))) {
+		return(FALSE);
+	}
+
+	new_data_size = page_get_data_size(new_page);
+	new_n_recs = page_get_n_recs(new_page);
+
+	ut_ad(new_data_size >= old_data_size);
+
+	page_delete_rec_list_end(split_rec, block, index,
+				 new_n_recs - old_n_recs,
+				 new_data_size - old_data_size, mtr);
+
+	return(TRUE);
+}
+
+/*****************************************************************
+Moves record list start to another page. Moved records do not include
+split_rec. */
+UNIV_INTERN
+ibool
+page_move_rec_list_start(
+/*=====================*/
+					/* out: TRUE on success; FALSE on
+					compression failure */
+	buf_block_t*	new_block,	/* in/out: index page where to move */
+	buf_block_t*	block,		/* in/out: page containing split_rec */
+	rec_t*		split_rec,	/* in: first record not to move */
+	dict_index_t*	index,		/* in: record descriptor */
+	mtr_t*		mtr)		/* in: mtr */
+{
+	if (UNIV_UNLIKELY(!page_copy_rec_list_start(new_block, block,
+						    split_rec, index, mtr))) {
+		return(FALSE);
+	}
+
+	page_delete_rec_list_start(split_rec, block, index, mtr);
+
+	return(TRUE);
+}
+
+/***************************************************************************
+This is a low-level operation which is used in a database index creation
+to update the page number of a created B-tree to a data dictionary record. */
+UNIV_INTERN
+void
+page_rec_write_index_page_no(
+/*=========================*/
+	rec_t*	rec,	/* in: record to update */
+	ulint	i,	/* in: index of the field to update */
+	ulint	page_no,/* in: value to write */
+	mtr_t*	mtr)	/* in: mtr */
+{
+	byte*	data;
+	ulint	len;
+
+	data = rec_get_nth_field_old(rec, i, &len);
+
+	ut_ad(len == 4);
+
+	mlog_write_ulint(data, page_no, MLOG_4BYTES, mtr);
+}
+
+/******************************************************************
+Used to delete n slots from the directory. This function updates
+also n_owned fields in the records, so that the first slot after
+the deleted ones inherits the records of the deleted slots. */
+UNIV_INLINE
+void
+page_dir_delete_slot(
+/*=================*/
+	page_t*		page,	/* in/out: the index page */
+	page_zip_des_t*	page_zip,/* in/out: compressed page, or NULL */
+	ulint		slot_no)/* in: slot to be deleted */
+{
+	page_dir_slot_t*	slot;
+	ulint			n_owned;
+	ulint			i;
+	ulint			n_slots;
+
+	ut_ad(!page_zip || page_is_comp(page));
+	ut_ad(slot_no > 0);
+	ut_ad(slot_no + 1 < page_dir_get_n_slots(page));
+
+	n_slots = page_dir_get_n_slots(page);
+
+	/* 1. Reset the n_owned fields of the slots to be
+	deleted */
+	slot = page_dir_get_nth_slot(page, slot_no);
+	n_owned = page_dir_slot_get_n_owned(slot);
+	page_dir_slot_set_n_owned(slot, page_zip, 0);
+
+	/* 2. Update the n_owned value of the first non-deleted slot */
+
+	slot = page_dir_get_nth_slot(page, slot_no + 1);
+	page_dir_slot_set_n_owned(slot, page_zip,
+				  n_owned + page_dir_slot_get_n_owned(slot));
+
+	/* 3. Destroy the slot by copying slots */
+	for (i = slot_no + 1; i < n_slots; i++) {
+		rec_t*	rec = (rec_t*)
+			page_dir_slot_get_rec(page_dir_get_nth_slot(page, i));
+		page_dir_slot_set_rec(page_dir_get_nth_slot(page, i - 1), rec);
+	}
+
+	/* 4. Zero out the last slot, which will be removed */
+	mach_write_to_2(page_dir_get_nth_slot(page, n_slots - 1), 0);
+
+	/* 5. Update the page header */
+	page_header_set_field(page, page_zip, PAGE_N_DIR_SLOTS, n_slots - 1);
+}
+
+/******************************************************************
+Used to add n slots to the directory. Does not set the record pointers
+in the added slots or update n_owned values: this is the responsibility
+of the caller. */
+UNIV_INLINE
+void
+page_dir_add_slot(
+/*==============*/
+	page_t*		page,	/* in/out: the index page */
+	page_zip_des_t*	page_zip,/* in/out: comprssed page, or NULL */
+	ulint		start)	/* in: the slot above which the new slots
+				are added */
+{
+	page_dir_slot_t*	slot;
+	ulint			n_slots;
+
+	n_slots = page_dir_get_n_slots(page);
+
+	ut_ad(start < n_slots - 1);
+
+	/* Update the page header */
+	page_dir_set_n_slots(page, page_zip, n_slots + 1);
+
+	/* Move slots up */
+	slot = page_dir_get_nth_slot(page, n_slots);
+	memmove(slot, slot + PAGE_DIR_SLOT_SIZE,
+		(n_slots - 1 - start) * PAGE_DIR_SLOT_SIZE);
+}
+
+/********************************************************************
+Splits a directory slot which owns too many records. */
+UNIV_INTERN
+void
+page_dir_split_slot(
+/*================*/
+	page_t*		page,	/* in/out: index page */
+	page_zip_des_t*	page_zip,/* in/out: compressed page whose
+				uncompressed part will be written, or NULL */
+	ulint		slot_no)/* in: the directory slot */
+{
+	rec_t*			rec;
+	page_dir_slot_t*	new_slot;
+	page_dir_slot_t*	prev_slot;
+	page_dir_slot_t*	slot;
+	ulint			i;
+	ulint			n_owned;
+
+	ut_ad(page);
+	ut_ad(!page_zip || page_is_comp(page));
+	ut_ad(slot_no > 0);
+
+	slot = page_dir_get_nth_slot(page, slot_no);
+
+	n_owned = page_dir_slot_get_n_owned(slot);
+	ut_ad(n_owned == PAGE_DIR_SLOT_MAX_N_OWNED + 1);
+
+	/* 1. We loop to find a record approximately in the middle of the
+	records owned by the slot. */
+
+	prev_slot = page_dir_get_nth_slot(page, slot_no - 1);
+	rec = (rec_t*) page_dir_slot_get_rec(prev_slot);
+
+	for (i = 0; i < n_owned / 2; i++) {
+		rec = page_rec_get_next(rec);
+	}
+
+	ut_ad(n_owned / 2 >= PAGE_DIR_SLOT_MIN_N_OWNED);
+
+	/* 2. We add one directory slot immediately below the slot to be
+	split. */
+
+	page_dir_add_slot(page, page_zip, slot_no - 1);
+
+	/* The added slot is now number slot_no, and the old slot is
+	now number slot_no + 1 */
+
+	new_slot = page_dir_get_nth_slot(page, slot_no);
+	slot = page_dir_get_nth_slot(page, slot_no + 1);
+
+	/* 3. We store the appropriate values to the new slot. */
+
+	page_dir_slot_set_rec(new_slot, rec);
+	page_dir_slot_set_n_owned(new_slot, page_zip, n_owned / 2);
+
+	/* 4. Finally, we update the number of records field of the
+	original slot */
+
+	page_dir_slot_set_n_owned(slot, page_zip, n_owned - (n_owned / 2));
+}
+
+/*****************************************************************
+Tries to balance the given directory slot with too few records with the upper
+neighbor, so that there are at least the minimum number of records owned by
+the slot; this may result in the merging of two slots. */
+UNIV_INTERN
+void
+page_dir_balance_slot(
+/*==================*/
+	page_t*		page,	/* in/out: index page */
+	page_zip_des_t*	page_zip,/* in/out: compressed page, or NULL */
+	ulint		slot_no)/* in: the directory slot */
+{
+	page_dir_slot_t*	slot;
+	page_dir_slot_t*	up_slot;
+	ulint			n_owned;
+	ulint			up_n_owned;
+	rec_t*			old_rec;
+	rec_t*			new_rec;
+
+	ut_ad(page);
+	ut_ad(!page_zip || page_is_comp(page));
+	ut_ad(slot_no > 0);
+
+	slot = page_dir_get_nth_slot(page, slot_no);
+
+	/* The last directory slot cannot be balanced with the upper
+	neighbor, as there is none. */
+
+	if (UNIV_UNLIKELY(slot_no == page_dir_get_n_slots(page) - 1)) {
+
+		return;
+	}
+
+	up_slot = page_dir_get_nth_slot(page, slot_no + 1);
+
+	n_owned = page_dir_slot_get_n_owned(slot);
+	up_n_owned = page_dir_slot_get_n_owned(up_slot);
+
+	ut_ad(n_owned == PAGE_DIR_SLOT_MIN_N_OWNED - 1);
+
+	/* If the upper slot has the minimum value of n_owned, we will merge
+	the two slots, therefore we assert: */
+	ut_ad(2 * PAGE_DIR_SLOT_MIN_N_OWNED - 1 <= PAGE_DIR_SLOT_MAX_N_OWNED);
+
+	if (up_n_owned > PAGE_DIR_SLOT_MIN_N_OWNED) {
+
+		/* In this case we can just transfer one record owned
+		by the upper slot to the property of the lower slot */
+		old_rec = (rec_t*) page_dir_slot_get_rec(slot);
+
+		if (page_is_comp(page)) {
+			new_rec = rec_get_next_ptr(old_rec, TRUE);
+
+			rec_set_n_owned_new(old_rec, page_zip, 0);
+			rec_set_n_owned_new(new_rec, page_zip, n_owned + 1);
+		} else {
+			new_rec = rec_get_next_ptr(old_rec, FALSE);
+
+			rec_set_n_owned_old(old_rec, 0);
+			rec_set_n_owned_old(new_rec, n_owned + 1);
+		}
+
+		page_dir_slot_set_rec(slot, new_rec);
+
+		page_dir_slot_set_n_owned(up_slot, page_zip, up_n_owned -1);
+	} else {
+		/* In this case we may merge the two slots */
+		page_dir_delete_slot(page, page_zip, slot_no);
+	}
+}
+
+/****************************************************************
+Returns the middle record of the record list. If there are an even number
+of records in the list, returns the first record of the upper half-list. */
+UNIV_INTERN
+rec_t*
+page_get_middle_rec(
+/*================*/
+			/* out: middle record */
+	page_t*	page)	/* in: page */
+{
+	page_dir_slot_t*	slot;
+	ulint			middle;
+	ulint			i;
+	ulint			n_owned;
+	ulint			count;
+	rec_t*			rec;
+
+	/* This many records we must leave behind */
+	middle = (page_get_n_recs(page) + PAGE_HEAP_NO_USER_LOW) / 2;
+
+	count = 0;
+
+	for (i = 0;; i++) {
+
+		slot = page_dir_get_nth_slot(page, i);
+		n_owned = page_dir_slot_get_n_owned(slot);
+
+		if (count + n_owned > middle) {
+			break;
+		} else {
+			count += n_owned;
+		}
+	}
+
+	ut_ad(i > 0);
+	slot = page_dir_get_nth_slot(page, i - 1);
+	rec = (rec_t*) page_dir_slot_get_rec(slot);
+	rec = page_rec_get_next(rec);
+
+	/* There are now count records behind rec */
+
+	for (i = 0; i < middle - count; i++) {
+		rec = page_rec_get_next(rec);
+	}
+
+	return(rec);
+}
+
+/*******************************************************************
+Returns the number of records before the given record in chain.
+The number includes infimum and supremum records. */
+UNIV_INTERN
+ulint
+page_rec_get_n_recs_before(
+/*=======================*/
+				/* out: number of records */
+	const rec_t*	rec)	/* in: the physical record */
+{
+	const page_dir_slot_t*	slot;
+	const rec_t*		slot_rec;
+	const page_t*		page;
+	ulint			i;
+	lint			n	= 0;
+
+	ut_ad(page_rec_check(rec));
+
+	page = page_align(rec);
+	if (page_is_comp(page)) {
+		while (rec_get_n_owned_new(rec) == 0) {
+
+			rec = rec_get_next_ptr_const(rec, TRUE);
+			n--;
+		}
+
+		for (i = 0; ; i++) {
+			slot = page_dir_get_nth_slot(page, i);
+			slot_rec = page_dir_slot_get_rec(slot);
+
+			n += rec_get_n_owned_new(slot_rec);
+
+			if (rec == slot_rec) {
+
+				break;
+			}
+		}
+	} else {
+		while (rec_get_n_owned_old(rec) == 0) {
+
+			rec = rec_get_next_ptr_const(rec, FALSE);
+			n--;
+		}
+
+		for (i = 0; ; i++) {
+			slot = page_dir_get_nth_slot(page, i);
+			slot_rec = page_dir_slot_get_rec(slot);
+
+			n += rec_get_n_owned_old(slot_rec);
+
+			if (rec == slot_rec) {
+
+				break;
+			}
+		}
+	}
+
+	n--;
+
+	ut_ad(n >= 0);
+
+	return((ulint) n);
+}
+
+/****************************************************************
+Prints record contents including the data relevant only in
+the index page context. */
+UNIV_INTERN
+void
+page_rec_print(
+/*===========*/
+	const rec_t*	rec,	/* in: physical record */
+	const ulint*	offsets)/* in: record descriptor */
+{
+	ut_a(!page_rec_is_comp(rec) == !rec_offs_comp(offsets));
+	rec_print_new(stderr, rec, offsets);
+	if (page_rec_is_comp(rec)) {
+		fprintf(stderr,
+			" n_owned: %lu; heap_no: %lu; next rec: %lu\n",
+			(ulong) rec_get_n_owned_new(rec),
+			(ulong) rec_get_heap_no_new(rec),
+			(ulong) rec_get_next_offs(rec, TRUE));
+	} else {
+		fprintf(stderr,
+			" n_owned: %lu; heap_no: %lu; next rec: %lu\n",
+			(ulong) rec_get_n_owned_old(rec),
+			(ulong) rec_get_heap_no_old(rec),
+			(ulong) rec_get_next_offs(rec, TRUE));
+	}
+
+	page_rec_check(rec);
+	rec_validate(rec, offsets);
+}
+
+/*******************************************************************
+This is used to print the contents of the directory for
+debugging purposes. */
+UNIV_INTERN
+void
+page_dir_print(
+/*===========*/
+	page_t*	page,	/* in: index page */
+	ulint	pr_n)	/* in: print n first and n last entries */
+{
+	ulint			n;
+	ulint			i;
+	page_dir_slot_t*	slot;
+
+	n = page_dir_get_n_slots(page);
+
+	fprintf(stderr, "--------------------------------\n"
+		"PAGE DIRECTORY\n"
+		"Page address %p\n"
+		"Directory stack top at offs: %lu; number of slots: %lu\n",
+		page, (ulong) page_offset(page_dir_get_nth_slot(page, n - 1)),
+		(ulong) n);
+	for (i = 0; i < n; i++) {
+		slot = page_dir_get_nth_slot(page, i);
+		if ((i == pr_n) && (i < n - pr_n)) {
+			fputs("    ...   \n", stderr);
+		}
+		if ((i < pr_n) || (i >= n - pr_n)) {
+			fprintf(stderr,
+				"Contents of slot: %lu: n_owned: %lu,"
+				" rec offs: %lu\n",
+				(ulong) i,
+				(ulong) page_dir_slot_get_n_owned(slot),
+				(ulong)
+				page_offset(page_dir_slot_get_rec(slot)));
+		}
+	}
+	fprintf(stderr, "Total of %lu records\n"
+		"--------------------------------\n",
+		(ulong) (PAGE_HEAP_NO_USER_LOW + page_get_n_recs(page)));
+}
+
+/*******************************************************************
+This is used to print the contents of the page record list for
+debugging purposes. */
+UNIV_INTERN
+void
+page_print_list(
+/*============*/
+	buf_block_t*	block,	/* in: index page */
+	dict_index_t*	index,	/* in: dictionary index of the page */
+	ulint		pr_n)	/* in: print n first and n last entries */
+{
+	page_t*		page		= block->frame;
+	page_cur_t	cur;
+	ulint		count;
+	ulint		n_recs;
+	mem_heap_t*	heap		= NULL;
+	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint*		offsets		= offsets_;
+	rec_offs_init(offsets_);
+
+	ut_a((ibool)!!page_is_comp(page) == dict_table_is_comp(index->table));
+
+	fprintf(stderr,
+		"--------------------------------\n"
+		"PAGE RECORD LIST\n"
+		"Page address %p\n", page);
+
+	n_recs = page_get_n_recs(page);
+
+	page_cur_set_before_first(block, &cur);
+	count = 0;
+	for (;;) {
+		offsets = rec_get_offsets(cur.rec, index, offsets,
+					  ULINT_UNDEFINED, &heap);
+		page_rec_print(cur.rec, offsets);
+
+		if (count == pr_n) {
+			break;
+		}
+		if (page_cur_is_after_last(&cur)) {
+			break;
+		}
+		page_cur_move_to_next(&cur);
+		count++;
+	}
+
+	if (n_recs > 2 * pr_n) {
+		fputs(" ... \n", stderr);
+	}
+
+	while (!page_cur_is_after_last(&cur)) {
+		page_cur_move_to_next(&cur);
+
+		if (count + pr_n >= n_recs) {
+			offsets = rec_get_offsets(cur.rec, index, offsets,
+						  ULINT_UNDEFINED, &heap);
+			page_rec_print(cur.rec, offsets);
+		}
+		count++;
+	}
+
+	fprintf(stderr,
+		"Total of %lu records \n"
+		"--------------------------------\n",
+		(ulong) (count + 1));
+
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+}
+
+/*******************************************************************
+Prints the info in a page header. */
+UNIV_INTERN
+void
+page_header_print(
+/*==============*/
+	const page_t*	page)
+{
+	fprintf(stderr,
+		"--------------------------------\n"
+		"PAGE HEADER INFO\n"
+		"Page address %p, n records %lu (%s)\n"
+		"n dir slots %lu, heap top %lu\n"
+		"Page n heap %lu, free %lu, garbage %lu\n"
+		"Page last insert %lu, direction %lu, n direction %lu\n",
+		page, (ulong) page_header_get_field(page, PAGE_N_RECS),
+		page_is_comp(page) ? "compact format" : "original format",
+		(ulong) page_header_get_field(page, PAGE_N_DIR_SLOTS),
+		(ulong) page_header_get_field(page, PAGE_HEAP_TOP),
+		(ulong) page_dir_get_n_heap(page),
+		(ulong) page_header_get_field(page, PAGE_FREE),
+		(ulong) page_header_get_field(page, PAGE_GARBAGE),
+		(ulong) page_header_get_field(page, PAGE_LAST_INSERT),
+		(ulong) page_header_get_field(page, PAGE_DIRECTION),
+		(ulong) page_header_get_field(page, PAGE_N_DIRECTION));
+}
+
+/*******************************************************************
+This is used to print the contents of the page for
+debugging purposes. */
+UNIV_INTERN
+void
+page_print(
+/*=======*/
+	buf_block_t*	block,	/* in: index page */
+	dict_index_t*	index,	/* in: dictionary index of the page */
+	ulint		dn,	/* in: print dn first and last entries
+				in directory */
+	ulint		rn)	/* in: print rn first and last records
+				in directory */
+{
+	page_t*	page = block->frame;
+
+	page_header_print(page);
+	page_dir_print(page, dn);
+	page_print_list(block, index, rn);
+}
+
+/*******************************************************************
+The following is used to validate a record on a page. This function
+differs from rec_validate as it can also check the n_owned field and
+the heap_no field. */
+UNIV_INTERN
+ibool
+page_rec_validate(
+/*==============*/
+				/* out: TRUE if ok */
+	rec_t*		rec,	/* in: physical record */
+	const ulint*	offsets)/* in: array returned by rec_get_offsets() */
+{
+	ulint	n_owned;
+	ulint	heap_no;
+	page_t*	page;
+
+	page = page_align(rec);
+	ut_a(!page_is_comp(page) == !rec_offs_comp(offsets));
+
+	page_rec_check(rec);
+	rec_validate(rec, offsets);
+
+	if (page_rec_is_comp(rec)) {
+		n_owned = rec_get_n_owned_new(rec);
+		heap_no = rec_get_heap_no_new(rec);
+	} else {
+		n_owned = rec_get_n_owned_old(rec);
+		heap_no = rec_get_heap_no_old(rec);
+	}
+
+	if (UNIV_UNLIKELY(!(n_owned <= PAGE_DIR_SLOT_MAX_N_OWNED))) {
+		fprintf(stderr,
+			"InnoDB: Dir slot of rec %lu, n owned too big %lu\n",
+			(ulong) page_offset(rec), (ulong) n_owned);
+		return(FALSE);
+	}
+
+	if (UNIV_UNLIKELY(!(heap_no < page_dir_get_n_heap(page)))) {
+		fprintf(stderr,
+			"InnoDB: Heap no of rec %lu too big %lu %lu\n",
+			(ulong) page_offset(rec), (ulong) heap_no,
+			(ulong) page_dir_get_n_heap(page));
+		return(FALSE);
+	}
+
+	return(TRUE);
+}
+
+/*******************************************************************
+Checks that the first directory slot points to the infimum record and
+the last to the supremum. This function is intended to track if the
+bug fixed in 4.0.14 has caused corruption to users' databases. */
+UNIV_INTERN
+void
+page_check_dir(
+/*===========*/
+	const page_t*	page)	/* in: index page */
+{
+	ulint	n_slots;
+	ulint	infimum_offs;
+	ulint	supremum_offs;
+
+	n_slots = page_dir_get_n_slots(page);
+	infimum_offs = mach_read_from_2(page_dir_get_nth_slot(page, 0));
+	supremum_offs = mach_read_from_2(page_dir_get_nth_slot(page,
+							       n_slots - 1));
+
+	if (UNIV_UNLIKELY(!page_rec_is_infimum_low(infimum_offs))) {
+
+		fprintf(stderr,
+			"InnoDB: Page directory corruption:"
+			" infimum not pointed to\n");
+		buf_page_print(page, 0);
+	}
+
+	if (UNIV_UNLIKELY(!page_rec_is_supremum_low(supremum_offs))) {
+
+		fprintf(stderr,
+			"InnoDB: Page directory corruption:"
+			" supremum not pointed to\n");
+		buf_page_print(page, 0);
+	}
+}
+
+/*******************************************************************
+This function checks the consistency of an index page when we do not
+know the index. This is also resilient so that this should never crash
+even if the page is total garbage. */
+UNIV_INTERN
+ibool
+page_simple_validate_old(
+/*=====================*/
+			/* out: TRUE if ok */
+	page_t*	page)	/* in: old-style index page */
+{
+	page_dir_slot_t* slot;
+	ulint		slot_no;
+	ulint		n_slots;
+	rec_t*		rec;
+	byte*		rec_heap_top;
+	ulint		count;
+	ulint		own_count;
+	ibool		ret	= FALSE;
+
+	ut_a(!page_is_comp(page));
+
+	/* Check first that the record heap and the directory do not
+	overlap. */
+
+	n_slots = page_dir_get_n_slots(page);
+
+	if (UNIV_UNLIKELY(n_slots > UNIV_PAGE_SIZE / 4)) {
+		fprintf(stderr,
+			"InnoDB: Nonsensical number %lu of page dir slots\n",
+			(ulong) n_slots);
+
+		goto func_exit;
+	}
+
+	rec_heap_top = page_header_get_ptr(page, PAGE_HEAP_TOP);
+
+	if (UNIV_UNLIKELY(rec_heap_top
+			  > page_dir_get_nth_slot(page, n_slots - 1))) {
+
+		fprintf(stderr,
+			"InnoDB: Record heap and dir overlap on a page,"
+			" heap top %lu, dir %lu\n",
+			(ulong) page_header_get_field(page, PAGE_HEAP_TOP),
+			(ulong)
+			page_offset(page_dir_get_nth_slot(page, n_slots - 1)));
+
+		goto func_exit;
+	}
+
+	/* Validate the record list in a loop checking also that it is
+	consistent with the page record directory. */
+
+	count = 0;
+	own_count = 1;
+	slot_no = 0;
+	slot = page_dir_get_nth_slot(page, slot_no);
+
+	rec = page_get_infimum_rec(page);
+
+	for (;;) {
+		if (UNIV_UNLIKELY(rec > rec_heap_top)) {
+			fprintf(stderr,
+				"InnoDB: Record %lu is above"
+				" rec heap top %lu\n",
+				(ulong)(rec - page),
+				(ulong)(rec_heap_top - page));
+
+			goto func_exit;
+		}
+
+		if (UNIV_UNLIKELY(rec_get_n_owned_old(rec))) {
+			/* This is a record pointed to by a dir slot */
+			if (UNIV_UNLIKELY(rec_get_n_owned_old(rec)
+					  != own_count)) {
+
+				fprintf(stderr,
+					"InnoDB: Wrong owned count %lu, %lu,"
+					" rec %lu\n",
+					(ulong) rec_get_n_owned_old(rec),
+					(ulong) own_count,
+					(ulong)(rec - page));
+
+				goto func_exit;
+			}
+
+			if (UNIV_UNLIKELY
+			    (page_dir_slot_get_rec(slot) != rec)) {
+				fprintf(stderr,
+					"InnoDB: Dir slot does not point"
+					" to right rec %lu\n",
+					(ulong)(rec - page));
+
+				goto func_exit;
+			}
+
+			own_count = 0;
+
+			if (!page_rec_is_supremum(rec)) {
+				slot_no++;
+				slot = page_dir_get_nth_slot(page, slot_no);
+			}
+		}
+
+		if (page_rec_is_supremum(rec)) {
+
+			break;
+		}
+
+		if (UNIV_UNLIKELY
+		    (rec_get_next_offs(rec, FALSE) < FIL_PAGE_DATA
+		     || rec_get_next_offs(rec, FALSE) >= UNIV_PAGE_SIZE)) {
+			fprintf(stderr,
+				"InnoDB: Next record offset"
+				" nonsensical %lu for rec %lu\n",
+				(ulong) rec_get_next_offs(rec, FALSE),
+				(ulong) (rec - page));
+
+			goto func_exit;
+		}
+
+		count++;
+
+		if (UNIV_UNLIKELY(count > UNIV_PAGE_SIZE)) {
+			fprintf(stderr,
+				"InnoDB: Page record list appears"
+				" to be circular %lu\n",
+				(ulong) count);
+			goto func_exit;
+		}
+
+		rec = page_rec_get_next(rec);
+		own_count++;
+	}
+
+	if (UNIV_UNLIKELY(rec_get_n_owned_old(rec) == 0)) {
+		fprintf(stderr, "InnoDB: n owned is zero in a supremum rec\n");
+
+		goto func_exit;
+	}
+
+	if (UNIV_UNLIKELY(slot_no != n_slots - 1)) {
+		fprintf(stderr, "InnoDB: n slots wrong %lu, %lu\n",
+			(ulong) slot_no, (ulong) (n_slots - 1));
+		goto func_exit;
+	}
+
+	if (UNIV_UNLIKELY(page_header_get_field(page, PAGE_N_RECS)
+			  + PAGE_HEAP_NO_USER_LOW
+			  != count + 1)) {
+		fprintf(stderr, "InnoDB: n recs wrong %lu %lu\n",
+			(ulong) page_header_get_field(page, PAGE_N_RECS)
+			+ PAGE_HEAP_NO_USER_LOW,
+			(ulong) (count + 1));
+
+		goto func_exit;
+	}
+
+	/* Check then the free list */
+	rec = page_header_get_ptr(page, PAGE_FREE);
+
+	while (rec != NULL) {
+		if (UNIV_UNLIKELY(rec < page + FIL_PAGE_DATA
+				  || rec >= page + UNIV_PAGE_SIZE)) {
+			fprintf(stderr,
+				"InnoDB: Free list record has"
+				" a nonsensical offset %lu\n",
+				(ulong) (rec - page));
+
+			goto func_exit;
+		}
+
+		if (UNIV_UNLIKELY(rec > rec_heap_top)) {
+			fprintf(stderr,
+				"InnoDB: Free list record %lu"
+				" is above rec heap top %lu\n",
+				(ulong) (rec - page),
+				(ulong) (rec_heap_top - page));
+
+			goto func_exit;
+		}
+
+		count++;
+
+		if (UNIV_UNLIKELY(count > UNIV_PAGE_SIZE)) {
+			fprintf(stderr,
+				"InnoDB: Page free list appears"
+				" to be circular %lu\n",
+				(ulong) count);
+			goto func_exit;
+		}
+
+		rec = page_rec_get_next(rec);
+	}
+
+	if (UNIV_UNLIKELY(page_dir_get_n_heap(page) != count + 1)) {
+
+		fprintf(stderr, "InnoDB: N heap is wrong %lu, %lu\n",
+			(ulong) page_dir_get_n_heap(page),
+			(ulong) (count + 1));
+
+		goto func_exit;
+	}
+
+	ret = TRUE;
+
+func_exit:
+	return(ret);
+}
+
+/*******************************************************************
+This function checks the consistency of an index page when we do not
+know the index. This is also resilient so that this should never crash
+even if the page is total garbage. */
+UNIV_INTERN
+ibool
+page_simple_validate_new(
+/*=====================*/
+			/* out: TRUE if ok */
+	page_t*	page)	/* in: new-style index page */
+{
+	page_dir_slot_t* slot;
+	ulint		slot_no;
+	ulint		n_slots;
+	rec_t*		rec;
+	byte*		rec_heap_top;
+	ulint		count;
+	ulint		own_count;
+	ibool		ret	= FALSE;
+
+	ut_a(page_is_comp(page));
+
+	/* Check first that the record heap and the directory do not
+	overlap. */
+
+	n_slots = page_dir_get_n_slots(page);
+
+	if (UNIV_UNLIKELY(n_slots > UNIV_PAGE_SIZE / 4)) {
+		fprintf(stderr,
+			"InnoDB: Nonsensical number %lu"
+			" of page dir slots\n", (ulong) n_slots);
+
+		goto func_exit;
+	}
+
+	rec_heap_top = page_header_get_ptr(page, PAGE_HEAP_TOP);
+
+	if (UNIV_UNLIKELY(rec_heap_top
+			  > page_dir_get_nth_slot(page, n_slots - 1))) {
+
+		fprintf(stderr,
+			"InnoDB: Record heap and dir overlap on a page,"
+			" heap top %lu, dir %lu\n",
+			(ulong) page_header_get_field(page, PAGE_HEAP_TOP),
+			(ulong)
+			page_offset(page_dir_get_nth_slot(page, n_slots - 1)));
+
+		goto func_exit;
+	}
+
+	/* Validate the record list in a loop checking also that it is
+	consistent with the page record directory. */
+
+	count = 0;
+	own_count = 1;
+	slot_no = 0;
+	slot = page_dir_get_nth_slot(page, slot_no);
+
+	rec = page_get_infimum_rec(page);
+
+	for (;;) {
+		if (UNIV_UNLIKELY(rec > rec_heap_top)) {
+			fprintf(stderr,
+				"InnoDB: Record %lu is above rec"
+				" heap top %lu\n",
+				(ulong) page_offset(rec),
+				(ulong) page_offset(rec_heap_top));
+
+			goto func_exit;
+		}
+
+		if (UNIV_UNLIKELY(rec_get_n_owned_new(rec))) {
+			/* This is a record pointed to by a dir slot */
+			if (UNIV_UNLIKELY(rec_get_n_owned_new(rec)
+					  != own_count)) {
+
+				fprintf(stderr,
+					"InnoDB: Wrong owned count %lu, %lu,"
+					" rec %lu\n",
+					(ulong) rec_get_n_owned_new(rec),
+					(ulong) own_count,
+					(ulong) page_offset(rec));
+
+				goto func_exit;
+			}
+
+			if (UNIV_UNLIKELY
+			    (page_dir_slot_get_rec(slot) != rec)) {
+				fprintf(stderr,
+					"InnoDB: Dir slot does not point"
+					" to right rec %lu\n",
+					(ulong) page_offset(rec));
+
+				goto func_exit;
+			}
+
+			own_count = 0;
+
+			if (!page_rec_is_supremum(rec)) {
+				slot_no++;
+				slot = page_dir_get_nth_slot(page, slot_no);
+			}
+		}
+
+		if (page_rec_is_supremum(rec)) {
+
+			break;
+		}
+
+		if (UNIV_UNLIKELY
+		    (rec_get_next_offs(rec, TRUE) < FIL_PAGE_DATA
+		     || rec_get_next_offs(rec, TRUE) >= UNIV_PAGE_SIZE)) {
+			fprintf(stderr,
+				"InnoDB: Next record offset nonsensical %lu"
+				" for rec %lu\n",
+				(ulong) rec_get_next_offs(rec, TRUE),
+				(ulong) page_offset(rec));
+
+			goto func_exit;
+		}
+
+		count++;
+
+		if (UNIV_UNLIKELY(count > UNIV_PAGE_SIZE)) {
+			fprintf(stderr,
+				"InnoDB: Page record list appears"
+				" to be circular %lu\n",
+				(ulong) count);
+			goto func_exit;
+		}
+
+		rec = page_rec_get_next(rec);
+		own_count++;
+	}
+
+	if (UNIV_UNLIKELY(rec_get_n_owned_new(rec) == 0)) {
+		fprintf(stderr, "InnoDB: n owned is zero"
+			" in a supremum rec\n");
+
+		goto func_exit;
+	}
+
+	if (UNIV_UNLIKELY(slot_no != n_slots - 1)) {
+		fprintf(stderr, "InnoDB: n slots wrong %lu, %lu\n",
+			(ulong) slot_no, (ulong) (n_slots - 1));
+		goto func_exit;
+	}
+
+	if (UNIV_UNLIKELY(page_header_get_field(page, PAGE_N_RECS)
+			  + PAGE_HEAP_NO_USER_LOW
+			  != count + 1)) {
+		fprintf(stderr, "InnoDB: n recs wrong %lu %lu\n",
+			(ulong) page_header_get_field(page, PAGE_N_RECS)
+			+ PAGE_HEAP_NO_USER_LOW,
+			(ulong) (count + 1));
+
+		goto func_exit;
+	}
+
+	/* Check then the free list */
+	rec = page_header_get_ptr(page, PAGE_FREE);
+
+	while (rec != NULL) {
+		if (UNIV_UNLIKELY(rec < page + FIL_PAGE_DATA
+				  || rec >= page + UNIV_PAGE_SIZE)) {
+			fprintf(stderr,
+				"InnoDB: Free list record has"
+				" a nonsensical offset %lu\n",
+				(ulong) page_offset(rec));
+
+			goto func_exit;
+		}
+
+		if (UNIV_UNLIKELY(rec > rec_heap_top)) {
+			fprintf(stderr,
+				"InnoDB: Free list record %lu"
+				" is above rec heap top %lu\n",
+				(ulong) page_offset(rec),
+				(ulong) page_offset(rec_heap_top));
+
+			goto func_exit;
+		}
+
+		count++;
+
+		if (UNIV_UNLIKELY(count > UNIV_PAGE_SIZE)) {
+			fprintf(stderr,
+				"InnoDB: Page free list appears"
+				" to be circular %lu\n",
+				(ulong) count);
+			goto func_exit;
+		}
+
+		rec = page_rec_get_next(rec);
+	}
+
+	if (UNIV_UNLIKELY(page_dir_get_n_heap(page) != count + 1)) {
+
+		fprintf(stderr, "InnoDB: N heap is wrong %lu, %lu\n",
+			(ulong) page_dir_get_n_heap(page),
+			(ulong) (count + 1));
+
+		goto func_exit;
+	}
+
+	ret = TRUE;
+
+func_exit:
+	return(ret);
+}
+
+/*******************************************************************
+This function checks the consistency of an index page. */
+UNIV_INTERN
+ibool
+page_validate(
+/*==========*/
+				/* out: TRUE if ok */
+	page_t*		page,	/* in: index page */
+	dict_index_t*	index)	/* in: data dictionary index containing
+				the page record type definition */
+{
+	page_dir_slot_t*slot;
+	mem_heap_t*	heap;
+	byte*		buf;
+	ulint		count;
+	ulint		own_count;
+	ulint		rec_own_count;
+	ulint		slot_no;
+	ulint		data_size;
+	rec_t*		rec;
+	rec_t*		old_rec		= NULL;
+	ulint		offs;
+	ulint		n_slots;
+	ibool		ret		= FALSE;
+	ulint		i;
+	ulint*		offsets		= NULL;
+	ulint*		old_offsets	= NULL;
+
+	if (UNIV_UNLIKELY((ibool) !!page_is_comp(page)
+			  != dict_table_is_comp(index->table))) {
+		fputs("InnoDB: 'compact format' flag mismatch\n", stderr);
+		goto func_exit2;
+	}
+	if (page_is_comp(page)) {
+		if (UNIV_UNLIKELY(!page_simple_validate_new(page))) {
+			goto func_exit2;
+		}
+	} else {
+		if (UNIV_UNLIKELY(!page_simple_validate_old(page))) {
+			goto func_exit2;
+		}
+	}
+
+	heap = mem_heap_create(UNIV_PAGE_SIZE + 200);
+
+	/* The following buffer is used to check that the
+	records in the page record heap do not overlap */
+
+	buf = mem_heap_zalloc(heap, UNIV_PAGE_SIZE);
+
+	/* Check first that the record heap and the directory do not
+	overlap. */
+
+	n_slots = page_dir_get_n_slots(page);
+
+	if (UNIV_UNLIKELY(!(page_header_get_ptr(page, PAGE_HEAP_TOP)
+			    <= page_dir_get_nth_slot(page, n_slots - 1)))) {
+
+		fputs("InnoDB: Record heap and dir overlap on a page ",
+		      stderr);
+		dict_index_name_print(stderr, NULL, index);
+		fprintf(stderr, ", %p, %p\n",
+			page_header_get_ptr(page, PAGE_HEAP_TOP),
+			page_dir_get_nth_slot(page, n_slots - 1));
+
+		goto func_exit;
+	}
+
+	/* Validate the record list in a loop checking also that
+	it is consistent with the directory. */
+	count = 0;
+	data_size = 0;
+	own_count = 1;
+	slot_no = 0;
+	slot = page_dir_get_nth_slot(page, slot_no);
+
+	rec = page_get_infimum_rec(page);
+
+	for (;;) {
+		offsets = rec_get_offsets(rec, index, offsets,
+					  ULINT_UNDEFINED, &heap);
+
+		if (page_is_comp(page) && page_rec_is_user_rec(rec)
+		    && UNIV_UNLIKELY(rec_get_node_ptr_flag(rec)
+				     == page_is_leaf(page))) {
+			fputs("InnoDB: node_ptr flag mismatch\n", stderr);
+			goto func_exit;
+		}
+
+		if (UNIV_UNLIKELY(!page_rec_validate(rec, offsets))) {
+			goto func_exit;
+		}
+
+		/* Check that the records are in the ascending order */
+		if (UNIV_LIKELY(count >= PAGE_HEAP_NO_USER_LOW)
+		    && !page_rec_is_supremum(rec)) {
+			if (UNIV_UNLIKELY
+			    (1 != cmp_rec_rec(rec, old_rec,
+					      offsets, old_offsets, index))) {
+				fprintf(stderr,
+					"InnoDB: Records in wrong order"
+					" on page %lu ",
+					(ulong) page_get_page_no(page));
+				dict_index_name_print(stderr, NULL, index);
+				fputs("\nInnoDB: previous record ", stderr);
+				rec_print_new(stderr, old_rec, old_offsets);
+				fputs("\nInnoDB: record ", stderr);
+				rec_print_new(stderr, rec, offsets);
+				putc('\n', stderr);
+
+				goto func_exit;
+			}
+		}
+
+		if (page_rec_is_user_rec(rec)) {
+
+			data_size += rec_offs_size(offsets);
+		}
+
+		offs = page_offset(rec_get_start(rec, offsets));
+
+		for (i = rec_offs_size(offsets); i--; ) {
+			if (UNIV_UNLIKELY(buf[offs + i])) {
+				/* No other record may overlap this */
+
+				fputs("InnoDB: Record overlaps another\n",
+				      stderr);
+				goto func_exit;
+			}
+
+			buf[offs + i] = 1;
+		}
+
+		if (page_is_comp(page)) {
+			rec_own_count = rec_get_n_owned_new(rec);
+		} else {
+			rec_own_count = rec_get_n_owned_old(rec);
+		}
+
+		if (UNIV_UNLIKELY(rec_own_count)) {
+			/* This is a record pointed to by a dir slot */
+			if (UNIV_UNLIKELY(rec_own_count != own_count)) {
+				fprintf(stderr,
+					"InnoDB: Wrong owned count %lu, %lu\n",
+					(ulong) rec_own_count,
+					(ulong) own_count);
+				goto func_exit;
+			}
+
+			if (page_dir_slot_get_rec(slot) != rec) {
+				fputs("InnoDB: Dir slot does not"
+				      " point to right rec\n",
+				      stderr);
+				goto func_exit;
+			}
+
+			page_dir_slot_check(slot);
+
+			own_count = 0;
+			if (!page_rec_is_supremum(rec)) {
+				slot_no++;
+				slot = page_dir_get_nth_slot(page, slot_no);
+			}
+		}
+
+		if (page_rec_is_supremum(rec)) {
+			break;
+		}
+
+		count++;
+		own_count++;
+		old_rec = rec;
+		rec = page_rec_get_next(rec);
+
+		/* set old_offsets to offsets; recycle offsets */
+		{
+			ulint* offs = old_offsets;
+			old_offsets = offsets;
+			offsets = offs;
+		}
+	}
+
+	if (page_is_comp(page)) {
+		if (UNIV_UNLIKELY(rec_get_n_owned_new(rec) == 0)) {
+
+			goto n_owned_zero;
+		}
+	} else if (UNIV_UNLIKELY(rec_get_n_owned_old(rec) == 0)) {
+n_owned_zero:
+		fputs("InnoDB: n owned is zero\n", stderr);
+		goto func_exit;
+	}
+
+	if (UNIV_UNLIKELY(slot_no != n_slots - 1)) {
+		fprintf(stderr, "InnoDB: n slots wrong %lu %lu\n",
+			(ulong) slot_no, (ulong) (n_slots - 1));
+		goto func_exit;
+	}
+
+	if (UNIV_UNLIKELY(page_header_get_field(page, PAGE_N_RECS)
+			  + PAGE_HEAP_NO_USER_LOW
+			  != count + 1)) {
+		fprintf(stderr, "InnoDB: n recs wrong %lu %lu\n",
+			(ulong) page_header_get_field(page, PAGE_N_RECS)
+			+ PAGE_HEAP_NO_USER_LOW,
+			(ulong) (count + 1));
+		goto func_exit;
+	}
+
+	if (UNIV_UNLIKELY(data_size != page_get_data_size(page))) {
+		fprintf(stderr,
+			"InnoDB: Summed data size %lu, returned by func %lu\n",
+			(ulong) data_size, (ulong) page_get_data_size(page));
+		goto func_exit;
+	}
+
+	/* Check then the free list */
+	rec = page_header_get_ptr(page, PAGE_FREE);
+
+	while (rec != NULL) {
+		offsets = rec_get_offsets(rec, index, offsets,
+					  ULINT_UNDEFINED, &heap);
+		if (UNIV_UNLIKELY(!page_rec_validate(rec, offsets))) {
+
+			goto func_exit;
+		}
+
+		count++;
+		offs = page_offset(rec_get_start(rec, offsets));
+
+		for (i = rec_offs_size(offsets); i--; ) {
+
+			if (UNIV_UNLIKELY(buf[offs + i])) {
+				fputs("InnoDB: Record overlaps another"
+				      " in free list\n", stderr);
+				goto func_exit;
+			}
+
+			buf[offs + i] = 1;
+		}
+
+		rec = page_rec_get_next(rec);
+	}
+
+	if (UNIV_UNLIKELY(page_dir_get_n_heap(page) != count + 1)) {
+		fprintf(stderr, "InnoDB: N heap is wrong %lu %lu\n",
+			(ulong) page_dir_get_n_heap(page),
+			(ulong) count + 1);
+		goto func_exit;
+	}
+
+	ret = TRUE;
+
+func_exit:
+	mem_heap_free(heap);
+
+	if (UNIV_UNLIKELY(ret == FALSE)) {
+func_exit2:
+		fprintf(stderr, "InnoDB: Apparent corruption in page %lu in ",
+			(ulong) page_get_page_no(page));
+		dict_index_name_print(stderr, NULL, index);
+		putc('\n', stderr);
+		buf_page_print(page, 0);
+	}
+
+	return(ret);
+}
+
+/*******************************************************************
+Looks in the page record list for a record with the given heap number. */
+UNIV_INTERN
+const rec_t*
+page_find_rec_with_heap_no(
+/*=======================*/
+				/* out: record, NULL if not found */
+	const page_t*	page,	/* in: index page */
+	ulint		heap_no)/* in: heap number */
+{
+	const rec_t*	rec;
+
+	if (page_is_comp(page)) {
+		rec = page + PAGE_NEW_INFIMUM;
+
+		for(;;) {
+			ulint	rec_heap_no = rec_get_heap_no_new(rec);
+
+			if (rec_heap_no == heap_no) {
+
+				return(rec);
+			} else if (rec_heap_no == PAGE_HEAP_NO_SUPREMUM) {
+
+				return(NULL);
+			}
+
+			rec = page + rec_get_next_offs(rec, TRUE);
+		}
+	} else {
+		rec = page + PAGE_OLD_INFIMUM;
+
+		for (;;) {
+			ulint	rec_heap_no = rec_get_heap_no_old(rec);
+
+			if (rec_heap_no == heap_no) {
+
+				return(rec);
+			} else if (rec_heap_no == PAGE_HEAP_NO_SUPREMUM) {
+
+				return(NULL);
+			}
+
+			rec = page + rec_get_next_offs(rec, FALSE);
+		}
+	}
+}
diff --git a/storage/xtradb/page/page0zip.c b/storage/xtradb/page/page0zip.c
new file mode 100644
index 00000000000..56189ce3bad
--- /dev/null
+++ b/storage/xtradb/page/page0zip.c
@@ -0,0 +1,4575 @@
+/*****************************************************************************
+
+Copyright (c) 2005, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Compressed page interface
+
+Created June 2005 by Marko Makela
+*******************************************************/
+
+#define THIS_MODULE
+#include "page0zip.h"
+#ifdef UNIV_NONINL
+# include "page0zip.ic"
+#endif
+#undef THIS_MODULE
+#include "page0page.h"
+#include "mtr0log.h"
+#include "ut0sort.h"
+#include "dict0boot.h"
+#include "dict0dict.h"
+#include "btr0sea.h"
+#include "btr0cur.h"
+#include "page0types.h"
+#include "lock0lock.h"
+#include "log0recv.h"
+#include "zlib.h"
+#include "buf0lru.h"
+
+/** Statistics on compression, indexed by page_zip_des_t::ssize - 1 */
+UNIV_INTERN page_zip_stat_t page_zip_stat[PAGE_ZIP_NUM_SSIZE - 1];
+
+/* Please refer to ../include/page0zip.ic for a description of the
+compressed page format. */
+
+/* The infimum and supremum records are omitted from the compressed page.
+On compress, we compare that the records are there, and on uncompress we
+restore the records. */
+static const byte infimum_extra[] = {
+	0x01,			/* info_bits=0, n_owned=1 */
+	0x00, 0x02		/* heap_no=0, status=2 */
+	/* ?, ?	*/		/* next=(first user rec, or supremum) */
+};
+static const byte infimum_data[] = {
+	0x69, 0x6e, 0x66, 0x69,
+	0x6d, 0x75, 0x6d, 0x00	/* "infimum\0" */
+};
+static const byte supremum_extra_data[] = {
+	/* 0x0?, */		/* info_bits=0, n_owned=1..8 */
+	0x00, 0x0b,		/* heap_no=1, status=3 */
+	0x00, 0x00,		/* next=0 */
+	0x73, 0x75, 0x70, 0x72,
+	0x65, 0x6d, 0x75, 0x6d	/* "supremum" */
+};
+
+/** Assert that a block of memory is filled with zero bytes.
+Compare at most sizeof(field_ref_zero) bytes. */
+#define ASSERT_ZERO(b, s) \
+	ut_ad(!memcmp(b, field_ref_zero, ut_min(s, sizeof field_ref_zero)))
+/** Assert that a BLOB pointer is filled with zero bytes. */
+#define ASSERT_ZERO_BLOB(b) \
+	ut_ad(!memcmp(b, field_ref_zero, sizeof field_ref_zero))
+
+/* Enable some extra debugging output.  This code can be enabled
+independently of any UNIV_ debugging conditions. */
+#if defined UNIV_DEBUG || defined UNIV_ZIP_DEBUG
+# include <stdarg.h>
+__attribute__((format (printf, 1, 2)))
+/**************************************************************************
+Report a failure to decompress or compress. */
+static
+int
+page_zip_fail_func(
+/*===============*/
+				/* out: number of characters printed */
+	const char*	fmt,	/* in: printf(3) format string */
+	...)			/* in: arguments corresponding to fmt */
+{
+	int	res;
+	va_list	ap;
+
+	ut_print_timestamp(stderr);
+	fputs("  InnoDB: ", stderr);
+	va_start(ap, fmt);
+	res = vfprintf(stderr, fmt, ap);
+	va_end(ap);
+
+	return(res);
+}
+# define page_zip_fail(fmt_args) page_zip_fail_func fmt_args
+#else /* UNIV_DEBUG || UNIV_ZIP_DEBUG */
+# define page_zip_fail(fmt_args) /* empty */
+#endif /* UNIV_DEBUG || UNIV_ZIP_DEBUG */
+
+/**************************************************************************
+Determine the guaranteed free space on an empty page. */
+UNIV_INTERN
+ulint
+page_zip_empty_size(
+/*================*/
+				/* out: minimum payload size on the page */
+	ulint	n_fields,	/* in: number of columns in the index */
+	ulint	zip_size)	/* in: compressed page size in bytes */
+{
+	lint	size = zip_size
+		/* subtract the page header and the longest
+		uncompressed data needed for one record */
+		- (PAGE_DATA
+		   + PAGE_ZIP_DIR_SLOT_SIZE
+		   + DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN
+		   + 1/* encoded heap_no==2 in page_zip_write_rec() */
+		   + 1/* end of modification log */
+		   - REC_N_NEW_EXTRA_BYTES/* omitted bytes */)
+		/* subtract the space for page_zip_fields_encode() */
+		- compressBound(2 * (n_fields + 1));
+	return(size > 0 ? (ulint) size : 0);
+}
+
+/*****************************************************************
+Gets the size of the compressed page trailer (the dense page directory),
+including deleted records (the free list). */
+UNIV_INLINE
+ulint
+page_zip_dir_size(
+/*==============*/
+						/* out: length of dense page
+						directory, in bytes */
+	const page_zip_des_t*	page_zip)	/* in: compressed page */
+{
+	/* Exclude the page infimum and supremum from the record count. */
+	ulint	size = PAGE_ZIP_DIR_SLOT_SIZE
+		* (page_dir_get_n_heap(page_zip->data)
+		   - PAGE_HEAP_NO_USER_LOW);
+	return(size);
+}
+
+/*****************************************************************
+Gets the size of the compressed page trailer (the dense page directory),
+only including user records (excluding the free list). */
+UNIV_INLINE
+ulint
+page_zip_dir_user_size(
+/*===================*/
+						/* out: length of dense page
+						directory comprising existing
+						records, in bytes */
+	const page_zip_des_t*	page_zip)	/* in: compressed page */
+{
+	ulint	size = PAGE_ZIP_DIR_SLOT_SIZE
+		* page_get_n_recs(page_zip->data);
+	ut_ad(size <= page_zip_dir_size(page_zip));
+	return(size);
+}
+
+/*****************************************************************
+Find the slot of the given record in the dense page directory. */
+UNIV_INLINE
+byte*
+page_zip_dir_find_low(
+/*==================*/
+					/* out: dense directory slot,
+					or NULL if record not found */
+	byte*	slot,			/* in: start of records */
+	byte*	end,			/* in: end of records */
+	ulint	offset)			/* in: offset of user record */
+{
+	ut_ad(slot <= end);
+
+	for (; slot < end; slot += PAGE_ZIP_DIR_SLOT_SIZE) {
+		if ((mach_read_from_2(slot) & PAGE_ZIP_DIR_SLOT_MASK)
+		    == offset) {
+			return(slot);
+		}
+	}
+
+	return(NULL);
+}
+
+/*****************************************************************
+Find the slot of the given non-free record in the dense page directory. */
+UNIV_INLINE
+byte*
+page_zip_dir_find(
+/*==============*/
+						/* out: dense directory slot,
+						or NULL if record not found */
+	page_zip_des_t*	page_zip,		/* in: compressed page */
+	ulint		offset)			/* in: offset of user record */
+{
+	byte*	end	= page_zip->data + page_zip_get_size(page_zip);
+
+	ut_ad(page_zip_simple_validate(page_zip));
+
+	return(page_zip_dir_find_low(end - page_zip_dir_user_size(page_zip),
+				     end,
+				     offset));
+}
+
+/*****************************************************************
+Find the slot of the given free record in the dense page directory. */
+UNIV_INLINE
+byte*
+page_zip_dir_find_free(
+/*===================*/
+						/* out: dense directory slot,
+						or NULL if record not found */
+	page_zip_des_t*	page_zip,		/* in: compressed page */
+	ulint		offset)			/* in: offset of user record */
+{
+	byte*	end	= page_zip->data + page_zip_get_size(page_zip);
+
+	ut_ad(page_zip_simple_validate(page_zip));
+
+	return(page_zip_dir_find_low(end - page_zip_dir_size(page_zip),
+				     end - page_zip_dir_user_size(page_zip),
+				     offset));
+}
+
+/*****************************************************************
+Read a given slot in the dense page directory. */
+UNIV_INLINE
+ulint
+page_zip_dir_get(
+/*=============*/
+						/* out: record offset
+						on the uncompressed page,
+						possibly ORed with
+						PAGE_ZIP_DIR_SLOT_DEL or
+						PAGE_ZIP_DIR_SLOT_OWNED */
+	const page_zip_des_t*	page_zip,	/* in: compressed page */
+	ulint			slot)		/* in: slot
+						(0=first user record) */
+{
+	ut_ad(page_zip_simple_validate(page_zip));
+	ut_ad(slot < page_zip_dir_size(page_zip) / PAGE_ZIP_DIR_SLOT_SIZE);
+	return(mach_read_from_2(page_zip->data + page_zip_get_size(page_zip)
+				- PAGE_ZIP_DIR_SLOT_SIZE * (slot + 1)));
+}
+
+/**************************************************************************
+Write a log record of compressing an index page. */
+static
+void
+page_zip_compress_write_log(
+/*========================*/
+	const page_zip_des_t*	page_zip,/* in: compressed page */
+	const page_t*		page,	/* in: uncompressed page */
+	dict_index_t*		index,	/* in: index of the B-tree node */
+	mtr_t*			mtr)	/* in: mini-transaction */
+{
+	byte*	log_ptr;
+	ulint	trailer_size;
+
+	log_ptr = mlog_open(mtr, 11 + 2 + 2);
+
+	if (!log_ptr) {
+
+		return;
+	}
+
+	/* Read the number of user records. */
+	trailer_size = page_dir_get_n_heap(page_zip->data)
+		- PAGE_HEAP_NO_USER_LOW;
+	/* Multiply by uncompressed of size stored per record */
+	if (!page_is_leaf(page)) {
+		trailer_size *= PAGE_ZIP_DIR_SLOT_SIZE + REC_NODE_PTR_SIZE;
+	} else if (dict_index_is_clust(index)) {
+		trailer_size *= PAGE_ZIP_DIR_SLOT_SIZE
+			+ DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN;
+	} else {
+		trailer_size *= PAGE_ZIP_DIR_SLOT_SIZE;
+	}
+	/* Add the space occupied by BLOB pointers. */
+	trailer_size += page_zip->n_blobs * BTR_EXTERN_FIELD_REF_SIZE;
+	ut_a(page_zip->m_end > PAGE_DATA);
+#if FIL_PAGE_DATA > PAGE_DATA
+# error "FIL_PAGE_DATA > PAGE_DATA"
+#endif
+	ut_a(page_zip->m_end + trailer_size <= page_zip_get_size(page_zip));
+
+	log_ptr = mlog_write_initial_log_record_fast((page_t*) page,
+						     MLOG_ZIP_PAGE_COMPRESS,
+						     log_ptr, mtr);
+	mach_write_to_2(log_ptr, page_zip->m_end - FIL_PAGE_TYPE);
+	log_ptr += 2;
+	mach_write_to_2(log_ptr, trailer_size);
+	log_ptr += 2;
+	mlog_close(mtr, log_ptr);
+
+	/* Write FIL_PAGE_PREV and FIL_PAGE_NEXT */
+	mlog_catenate_string(mtr, page_zip->data + FIL_PAGE_PREV, 4);
+	mlog_catenate_string(mtr, page_zip->data + FIL_PAGE_NEXT, 4);
+	/* Write most of the page header, the compressed stream and
+	the modification log. */
+	mlog_catenate_string(mtr, page_zip->data + FIL_PAGE_TYPE,
+			     page_zip->m_end - FIL_PAGE_TYPE);
+	/* Write the uncompressed trailer of the compressed page. */
+	mlog_catenate_string(mtr, page_zip->data + page_zip_get_size(page_zip)
+			     - trailer_size, trailer_size);
+}
+
+/**********************************************************
+Determine how many externally stored columns are contained
+in existing records with smaller heap_no than rec. */
+static
+ulint
+page_zip_get_n_prev_extern(
+/*=======================*/
+	const page_zip_des_t*	page_zip,/* in: dense page directory on
+					compressed page */
+	const rec_t*		rec,	/* in: compact physical record
+					on a B-tree leaf page */
+	dict_index_t*		index)	/* in: record descriptor */
+{
+	const page_t*	page	= page_align(rec);
+	ulint		n_ext	= 0;
+	ulint		i;
+	ulint		left;
+	ulint		heap_no;
+	ulint		n_recs	= page_get_n_recs(page_zip->data);
+
+	ut_ad(page_is_leaf(page));
+	ut_ad(page_is_comp(page));
+	ut_ad(dict_table_is_comp(index->table));
+	ut_ad(dict_index_is_clust(index));
+
+	heap_no = rec_get_heap_no_new(rec);
+	ut_ad(heap_no >= PAGE_HEAP_NO_USER_LOW);
+	left = heap_no - PAGE_HEAP_NO_USER_LOW;
+	if (UNIV_UNLIKELY(!left)) {
+		return(0);
+	}
+
+	for (i = 0; i < n_recs; i++) {
+		const rec_t*	r	= page + (page_zip_dir_get(page_zip, i)
+						  & PAGE_ZIP_DIR_SLOT_MASK);
+
+		if (rec_get_heap_no_new(r) < heap_no) {
+			n_ext += rec_get_n_extern_new(r, index,
+						      ULINT_UNDEFINED);
+			if (!--left) {
+				break;
+			}
+		}
+	}
+
+	return(n_ext);
+}
+
+/**************************************************************************
+Encode the length of a fixed-length column. */
+static
+byte*
+page_zip_fixed_field_encode(
+/*========================*/
+			/* out: buf + length of encoded val */
+	byte*	buf,	/* in: pointer to buffer where to write */
+	ulint	val)	/* in: value to write */
+{
+	ut_ad(val >= 2);
+
+	if (UNIV_LIKELY(val < 126)) {
+		/*
+		0 = nullable variable field of at most 255 bytes length;
+		1 = not null variable field of at most 255 bytes length;
+		126 = nullable variable field with maximum length >255;
+		127 = not null variable field with maximum length >255
+		*/
+		*buf++ = (byte) val;
+	} else {
+		*buf++ = (byte) (0x80 | val >> 8);
+		*buf++ = (byte) val;
+	}
+
+	return(buf);
+}
+
+/**************************************************************************
+Write the index information for the compressed page. */
+static
+ulint
+page_zip_fields_encode(
+/*===================*/
+				/* out: used size of buf */
+	ulint		n,	/* in: number of fields to compress */
+	dict_index_t*	index,	/* in: index comprising at least n fields */
+	ulint		trx_id_pos,/* in: position of the trx_id column
+				in the index, or ULINT_UNDEFINED if
+				this is a non-leaf page */
+	byte*		buf)	/* out: buffer of (n + 1) * 2 bytes */
+{
+	const byte*	buf_start	= buf;
+	ulint		i;
+	ulint		col;
+	ulint		trx_id_col	= 0;
+	/* sum of lengths of preceding non-nullable fixed fields, or 0 */
+	ulint		fixed_sum	= 0;
+
+	ut_ad(trx_id_pos == ULINT_UNDEFINED || trx_id_pos < n);
+
+	for (i = col = 0; i < n; i++) {
+		dict_field_t*	field = dict_index_get_nth_field(index, i);
+		ulint		val;
+
+		if (dict_field_get_col(field)->prtype & DATA_NOT_NULL) {
+			val = 1; /* set the "not nullable" flag */
+		} else {
+			val = 0; /* nullable field */
+		}
+
+		if (!field->fixed_len) {
+			/* variable-length field */
+			const dict_col_t*	column
+				= dict_field_get_col(field);
+
+			if (UNIV_UNLIKELY(column->len > 255)
+			    || UNIV_UNLIKELY(column->mtype == DATA_BLOB)) {
+				val |= 0x7e; /* max > 255 bytes */
+			}
+
+			if (fixed_sum) {
+				/* write out the length of any
+				preceding non-nullable fields */
+				buf = page_zip_fixed_field_encode(
+					buf, fixed_sum << 1 | 1);
+				fixed_sum = 0;
+				col++;
+			}
+
+			*buf++ = (byte) val;
+			col++;
+		} else if (val) {
+			/* fixed-length non-nullable field */
+
+			if (fixed_sum && UNIV_UNLIKELY
+			    (fixed_sum + field->fixed_len
+			     > DICT_MAX_INDEX_COL_LEN)) {
+				/* Write out the length of the
+				preceding non-nullable fields,
+				to avoid exceeding the maximum
+				length of a fixed-length column. */
+				buf = page_zip_fixed_field_encode(
+					buf, fixed_sum << 1 | 1);
+				fixed_sum = 0;
+				col++;
+			}
+
+			if (i && UNIV_UNLIKELY(i == trx_id_pos)) {
+				if (fixed_sum) {
+					/* Write out the length of any
+					preceding non-nullable fields,
+					and start a new trx_id column. */
+					buf = page_zip_fixed_field_encode(
+						buf, fixed_sum << 1 | 1);
+					col++;
+				}
+
+				trx_id_col = col;
+				fixed_sum = field->fixed_len;
+			} else {
+				/* add to the sum */
+				fixed_sum += field->fixed_len;
+			}
+		} else {
+			/* fixed-length nullable field */
+
+			if (fixed_sum) {
+				/* write out the length of any
+				preceding non-nullable fields */
+				buf = page_zip_fixed_field_encode(
+					buf, fixed_sum << 1 | 1);
+				fixed_sum = 0;
+				col++;
+			}
+
+			buf = page_zip_fixed_field_encode(
+				buf, field->fixed_len << 1);
+			col++;
+		}
+	}
+
+	if (fixed_sum) {
+		/* Write out the lengths of last fixed-length columns. */
+		buf = page_zip_fixed_field_encode(buf, fixed_sum << 1 | 1);
+	}
+
+	if (trx_id_pos != ULINT_UNDEFINED) {
+		/* Write out the position of the trx_id column */
+		i = trx_id_col;
+	} else {
+		/* Write out the number of nullable fields */
+		i = index->n_nullable;
+	}
+
+	if (i < 128) {
+		*buf++ = (byte) i;
+	} else {
+		*buf++ = (byte) (0x80 | i >> 8);
+		*buf++ = (byte) i;
+	}
+
+	ut_ad((ulint) (buf - buf_start) <= (n + 2) * 2);
+	return((ulint) (buf - buf_start));
+}
+
+/**************************************************************************
+Populate the dense page directory from the sparse directory. */
+static
+void
+page_zip_dir_encode(
+/*================*/
+	const page_t*	page,	/* in: compact page */
+	byte*		buf,	/* in: pointer to dense page directory[-1];
+				out: dense directory on compressed page */
+	const rec_t**	recs)	/* in: pointer to an array of 0, or NULL;
+				out: dense page directory sorted by ascending
+				address (and heap_no) */
+{
+	const byte*	rec;
+	ulint		status;
+	ulint		min_mark;
+	ulint		heap_no;
+	ulint		i;
+	ulint		n_heap;
+	ulint		offs;
+
+	min_mark = 0;
+
+	if (page_is_leaf(page)) {
+		status = REC_STATUS_ORDINARY;
+	} else {
+		status = REC_STATUS_NODE_PTR;
+		if (UNIV_UNLIKELY
+		    (mach_read_from_4(page + FIL_PAGE_PREV) == FIL_NULL)) {
+			min_mark = REC_INFO_MIN_REC_FLAG;
+		}
+	}
+
+	n_heap = page_dir_get_n_heap(page);
+
+	/* Traverse the list of stored records in the collation order,
+	starting from the first user record. */
+
+	rec = page + PAGE_NEW_INFIMUM, TRUE;
+
+	i = 0;
+
+	for (;;) {
+		ulint	info_bits;
+		offs = rec_get_next_offs(rec, TRUE);
+		if (UNIV_UNLIKELY(offs == PAGE_NEW_SUPREMUM)) {
+			break;
+		}
+		rec = page + offs;
+		heap_no = rec_get_heap_no_new(rec);
+		ut_a(heap_no >= PAGE_HEAP_NO_USER_LOW);
+		ut_a(heap_no < n_heap);
+		ut_a(offs < UNIV_PAGE_SIZE - PAGE_DIR);
+		ut_a(offs >= PAGE_ZIP_START);
+#if PAGE_ZIP_DIR_SLOT_MASK & (PAGE_ZIP_DIR_SLOT_MASK + 1)
+# error "PAGE_ZIP_DIR_SLOT_MASK is not 1 less than a power of 2"
+#endif
+#if PAGE_ZIP_DIR_SLOT_MASK < UNIV_PAGE_SIZE - 1
+# error "PAGE_ZIP_DIR_SLOT_MASK < UNIV_PAGE_SIZE - 1"
+#endif
+		if (UNIV_UNLIKELY(rec_get_n_owned_new(rec))) {
+			offs |= PAGE_ZIP_DIR_SLOT_OWNED;
+		}
+
+		info_bits = rec_get_info_bits(rec, TRUE);
+		if (UNIV_UNLIKELY(info_bits & REC_INFO_DELETED_FLAG)) {
+			info_bits &= ~REC_INFO_DELETED_FLAG;
+			offs |= PAGE_ZIP_DIR_SLOT_DEL;
+		}
+		ut_a(info_bits == min_mark);
+		/* Only the smallest user record can have
+		REC_INFO_MIN_REC_FLAG set. */
+		min_mark = 0;
+
+		mach_write_to_2(buf - PAGE_ZIP_DIR_SLOT_SIZE * ++i, offs);
+
+		if (UNIV_LIKELY_NULL(recs)) {
+			/* Ensure that each heap_no occurs at most once. */
+			ut_a(!recs[heap_no - PAGE_HEAP_NO_USER_LOW]);
+			/* exclude infimum and supremum */
+			recs[heap_no - PAGE_HEAP_NO_USER_LOW] = rec;
+		}
+
+		ut_a(rec_get_status(rec) == status);
+	}
+
+	offs = page_header_get_field(page, PAGE_FREE);
+
+	/* Traverse the free list (of deleted records). */
+	while (offs) {
+		ut_ad(!(offs & ~PAGE_ZIP_DIR_SLOT_MASK));
+		rec = page + offs;
+
+		heap_no = rec_get_heap_no_new(rec);
+		ut_a(heap_no >= PAGE_HEAP_NO_USER_LOW);
+		ut_a(heap_no < n_heap);
+
+		ut_a(!rec[-REC_N_NEW_EXTRA_BYTES]); /* info_bits and n_owned */
+		ut_a(rec_get_status(rec) == status);
+
+		mach_write_to_2(buf - PAGE_ZIP_DIR_SLOT_SIZE * ++i, offs);
+
+		if (UNIV_LIKELY_NULL(recs)) {
+			/* Ensure that each heap_no occurs at most once. */
+			ut_a(!recs[heap_no - PAGE_HEAP_NO_USER_LOW]);
+			/* exclude infimum and supremum */
+			recs[heap_no - PAGE_HEAP_NO_USER_LOW] = rec;
+		}
+
+		offs = rec_get_next_offs(rec, TRUE);
+	}
+
+	/* Ensure that each heap no occurs at least once. */
+	ut_a(i + PAGE_HEAP_NO_USER_LOW == n_heap);
+}
+
+/**************************************************************************
+Allocate memory for zlib. */
+static
+void*
+page_zip_malloc(
+/*============*/
+	void*	opaque,
+	uInt	items,
+	uInt	size)
+{
+	return(mem_heap_alloc(opaque, items * size));
+}
+
+/**************************************************************************
+Deallocate memory for zlib. */
+static
+void
+page_zip_free(
+/*==========*/
+	void*	opaque __attribute__((unused)),
+	void*	address __attribute__((unused)))
+{
+}
+
+/**************************************************************************
+Configure the zlib allocator to use the given memory heap. */
+UNIV_INTERN
+void
+page_zip_set_alloc(
+/*===============*/
+	void*		stream,		/* in/out: zlib stream */
+	mem_heap_t*	heap)		/* in: memory heap to use */
+{
+	z_stream*	strm = stream;
+
+	strm->zalloc = page_zip_malloc;
+	strm->zfree = page_zip_free;
+	strm->opaque = heap;
+}
+
+#if 0 || defined UNIV_DEBUG || defined UNIV_ZIP_DEBUG
+# define PAGE_ZIP_COMPRESS_DBG
+#endif
+
+#ifdef PAGE_ZIP_COMPRESS_DBG
+/* Set this variable in a debugger to enable
+excessive logging in page_zip_compress(). */
+UNIV_INTERN ibool	page_zip_compress_dbg;
+/* Set this variable in a debugger to enable
+binary logging of the data passed to deflate().
+When this variable is nonzero, it will act
+as a log file name generator. */
+UNIV_INTERN unsigned	page_zip_compress_log;
+
+/**************************************************************************
+Wrapper for deflate().  Log the operation if page_zip_compress_dbg is set. */
+static
+ibool
+page_zip_compress_deflate(
+/*======================*/
+	FILE*		logfile,/* in: log file, or NULL */
+	z_streamp	strm,	/* in/out: compressed stream for deflate() */
+	int		flush)	/* in: deflate() flushing method */
+{
+	int	status;
+	if (UNIV_UNLIKELY(page_zip_compress_dbg)) {
+		ut_print_buf(stderr, strm->next_in, strm->avail_in);
+	}
+	if (UNIV_LIKELY_NULL(logfile)) {
+		fwrite(strm->next_in, 1, strm->avail_in, logfile);
+	}
+	status = deflate(strm, flush);
+	if (UNIV_UNLIKELY(page_zip_compress_dbg)) {
+		fprintf(stderr, " -> %d\n", status);
+	}
+	return(status);
+}
+
+/* Redefine deflate(). */
+# undef deflate
+# define deflate(strm, flush) page_zip_compress_deflate(logfile, strm, flush)
+# define FILE_LOGFILE FILE* logfile,
+# define LOGFILE logfile,
+#else /* PAGE_ZIP_COMPRESS_DBG */
+# define FILE_LOGFILE
+# define LOGFILE
+#endif /* PAGE_ZIP_COMPRESS_DBG */
+
+/**************************************************************************
+Compress the records of a node pointer page. */
+static
+int
+page_zip_compress_node_ptrs(
+/*========================*/
+					/* out: Z_OK, or a zlib error code */
+	FILE_LOGFILE
+	z_stream*	c_stream,	/* in/out: compressed page stream */
+	const rec_t**	recs,		/* in: dense page directory
+					sorted by address */
+	ulint		n_dense,	/* in: size of recs[] */
+	dict_index_t*	index,		/* in: the index of the page */
+	byte*		storage,	/* in: end of dense page directory */
+	mem_heap_t*	heap)		/* in: temporary memory heap */
+{
+	int	err	= Z_OK;
+	ulint*	offsets = NULL;
+
+	do {
+		const rec_t*	rec = *recs++;
+
+		offsets = rec_get_offsets(rec, index, offsets,
+					  ULINT_UNDEFINED, &heap);
+		/* Only leaf nodes may contain externally stored columns. */
+		ut_ad(!rec_offs_any_extern(offsets));
+
+		UNIV_MEM_ASSERT_RW(rec, rec_offs_data_size(offsets));
+		UNIV_MEM_ASSERT_RW(rec - rec_offs_extra_size(offsets),
+				   rec_offs_extra_size(offsets));
+
+		/* Compress the extra bytes. */
+		c_stream->avail_in = rec - REC_N_NEW_EXTRA_BYTES
+			- c_stream->next_in;
+
+		if (c_stream->avail_in) {
+			err = deflate(c_stream, Z_NO_FLUSH);
+			if (UNIV_UNLIKELY(err != Z_OK)) {
+				break;
+			}
+		}
+		ut_ad(!c_stream->avail_in);
+
+		/* Compress the data bytes, except node_ptr. */
+		c_stream->next_in = (byte*) rec;
+		c_stream->avail_in = rec_offs_data_size(offsets)
+			- REC_NODE_PTR_SIZE;
+		ut_ad(c_stream->avail_in);
+
+		err = deflate(c_stream, Z_NO_FLUSH);
+		if (UNIV_UNLIKELY(err != Z_OK)) {
+			break;
+		}
+
+		ut_ad(!c_stream->avail_in);
+
+		memcpy(storage - REC_NODE_PTR_SIZE
+		       * (rec_get_heap_no_new(rec) - 1),
+		       c_stream->next_in, REC_NODE_PTR_SIZE);
+		c_stream->next_in += REC_NODE_PTR_SIZE;
+	} while (--n_dense);
+
+	return(err);
+}
+
+/**************************************************************************
+Compress the records of a leaf node of a secondary index. */
+static
+int
+page_zip_compress_sec(
+/*==================*/
+					/* out: Z_OK, or a zlib error code */
+	FILE_LOGFILE
+	z_stream*	c_stream,	/* in/out: compressed page stream */
+	const rec_t**	recs,		/* in: dense page directory
+					sorted by address */
+	ulint		n_dense)	/* in: size of recs[] */
+{
+	int		err	= Z_OK;
+
+	ut_ad(n_dense > 0);
+
+	do {
+		const rec_t*	rec = *recs++;
+
+		/* Compress everything up to this record. */
+		c_stream->avail_in = rec - REC_N_NEW_EXTRA_BYTES
+			- c_stream->next_in;
+
+		if (UNIV_LIKELY(c_stream->avail_in)) {
+			UNIV_MEM_ASSERT_RW(c_stream->next_in,
+					   c_stream->avail_in);
+			err = deflate(c_stream, Z_NO_FLUSH);
+			if (UNIV_UNLIKELY(err != Z_OK)) {
+				break;
+			}
+		}
+
+		ut_ad(!c_stream->avail_in);
+		ut_ad(c_stream->next_in == rec - REC_N_NEW_EXTRA_BYTES);
+
+		/* Skip the REC_N_NEW_EXTRA_BYTES. */
+
+		c_stream->next_in = (byte*) rec;
+	} while (--n_dense);
+
+	return(err);
+}
+
+/**************************************************************************
+Compress a record of a leaf node of a clustered index that contains
+externally stored columns. */
+static
+int
+page_zip_compress_clust_ext(
+/*========================*/
+					/* out: Z_OK, or a zlib error code */
+	FILE_LOGFILE
+	z_stream*	c_stream,	/* in/out: compressed page stream */
+	const rec_t*	rec,		/* in: record */
+	const ulint*	offsets,	/* in: rec_get_offsets(rec) */
+	ulint		trx_id_col,	/* in: position of of DB_TRX_ID */
+	byte*		deleted,	/* in: dense directory entry pointing
+					to the head of the free list */
+	byte*		storage,	/* in: end of dense page directory */
+	byte**		externs,	/* in/out: pointer to the next
+					available BLOB pointer */
+	ulint*		n_blobs)	/* in/out: number of
+					externally stored columns */
+{
+	int	err;
+	ulint	i;
+
+	UNIV_MEM_ASSERT_RW(rec, rec_offs_data_size(offsets));
+	UNIV_MEM_ASSERT_RW(rec - rec_offs_extra_size(offsets),
+			   rec_offs_extra_size(offsets));
+
+	for (i = 0; i < rec_offs_n_fields(offsets); i++) {
+		ulint		len;
+		const byte*	src;
+
+		if (UNIV_UNLIKELY(i == trx_id_col)) {
+			ut_ad(!rec_offs_nth_extern(offsets, i));
+			/* Store trx_id and roll_ptr
+			in uncompressed form. */
+			src = rec_get_nth_field(rec, offsets, i, &len);
+			ut_ad(src + DATA_TRX_ID_LEN
+			      == rec_get_nth_field(rec, offsets,
+						   i + 1, &len));
+			ut_ad(len == DATA_ROLL_PTR_LEN);
+
+			/* Compress any preceding bytes. */
+			c_stream->avail_in
+				= src - c_stream->next_in;
+
+			if (c_stream->avail_in) {
+				err = deflate(c_stream, Z_NO_FLUSH);
+				if (UNIV_UNLIKELY(err != Z_OK)) {
+
+					return(err);
+				}
+			}
+
+			ut_ad(!c_stream->avail_in);
+			ut_ad(c_stream->next_in == src);
+
+			memcpy(storage
+			       - (DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN)
+			       * (rec_get_heap_no_new(rec) - 1),
+			       c_stream->next_in,
+			       DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN);
+
+			c_stream->next_in
+				+= DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN;
+
+			/* Skip also roll_ptr */
+			i++;
+		} else if (rec_offs_nth_extern(offsets, i)) {
+			src = rec_get_nth_field(rec, offsets, i, &len);
+			ut_ad(len >= BTR_EXTERN_FIELD_REF_SIZE);
+			src += len - BTR_EXTERN_FIELD_REF_SIZE;
+
+			c_stream->avail_in = src
+				- c_stream->next_in;
+			if (UNIV_LIKELY(c_stream->avail_in)) {
+				err = deflate(c_stream, Z_NO_FLUSH);
+				if (UNIV_UNLIKELY(err != Z_OK)) {
+
+					return(err);
+				}
+			}
+
+			ut_ad(!c_stream->avail_in);
+			ut_ad(c_stream->next_in == src);
+
+			/* Reserve space for the data at
+			the end of the space reserved for
+			the compressed data and the page
+			modification log. */
+
+			if (UNIV_UNLIKELY
+			    (c_stream->avail_out
+			     <= BTR_EXTERN_FIELD_REF_SIZE)) {
+				/* out of space */
+				return(Z_BUF_ERROR);
+			}
+
+			ut_ad(*externs == c_stream->next_out
+			      + c_stream->avail_out
+			      + 1/* end of modif. log */);
+
+			c_stream->next_in
+				+= BTR_EXTERN_FIELD_REF_SIZE;
+
+			/* Skip deleted records. */
+			if (UNIV_LIKELY_NULL
+			    (page_zip_dir_find_low(
+				    storage, deleted,
+				    page_offset(rec)))) {
+				continue;
+			}
+
+			(*n_blobs)++;
+			c_stream->avail_out
+				-= BTR_EXTERN_FIELD_REF_SIZE;
+			*externs -= BTR_EXTERN_FIELD_REF_SIZE;
+
+			/* Copy the BLOB pointer */
+			memcpy(*externs, c_stream->next_in
+			       - BTR_EXTERN_FIELD_REF_SIZE,
+			       BTR_EXTERN_FIELD_REF_SIZE);
+		}
+	}
+
+	return(Z_OK);
+}
+
+/**************************************************************************
+Compress the records of a leaf node of a clustered index. */
+static
+int
+page_zip_compress_clust(
+/*====================*/
+					/* out: Z_OK, or a zlib error code */
+	FILE_LOGFILE
+	z_stream*	c_stream,	/* in/out: compressed page stream */
+	const rec_t**	recs,		/* in: dense page directory
+					sorted by address */
+	ulint		n_dense,	/* in: size of recs[] */
+	dict_index_t*	index,		/* in: the index of the page */
+	ulint*		n_blobs,	/* in: 0; out: number of
+					externally stored columns */
+	ulint		trx_id_col,	/* index of the trx_id column */
+	byte*		deleted,	/* in: dense directory entry pointing
+					to the head of the free list */
+	byte*		storage,	/* in: end of dense page directory */
+	mem_heap_t*	heap)		/* in: temporary memory heap */
+{
+	int	err		= Z_OK;
+	ulint*	offsets		= NULL;
+	/* BTR_EXTERN_FIELD_REF storage */
+	byte*	externs		= storage - n_dense
+		* (DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN);
+
+	ut_ad(*n_blobs == 0);
+
+	do {
+		const rec_t*	rec = *recs++;
+
+		offsets = rec_get_offsets(rec, index, offsets,
+					  ULINT_UNDEFINED, &heap);
+		ut_ad(rec_offs_n_fields(offsets)
+		      == dict_index_get_n_fields(index));
+		UNIV_MEM_ASSERT_RW(rec, rec_offs_data_size(offsets));
+		UNIV_MEM_ASSERT_RW(rec - rec_offs_extra_size(offsets),
+				   rec_offs_extra_size(offsets));
+
+		/* Compress the extra bytes. */
+		c_stream->avail_in = rec - REC_N_NEW_EXTRA_BYTES
+			- c_stream->next_in;
+
+		if (c_stream->avail_in) {
+			err = deflate(c_stream, Z_NO_FLUSH);
+			if (UNIV_UNLIKELY(err != Z_OK)) {
+
+				goto func_exit;
+			}
+		}
+		ut_ad(!c_stream->avail_in);
+		ut_ad(c_stream->next_in == rec - REC_N_NEW_EXTRA_BYTES);
+
+		/* Compress the data bytes. */
+
+		c_stream->next_in = (byte*) rec;
+
+		/* Check if there are any externally stored columns.
+		For each externally stored column, store the
+		BTR_EXTERN_FIELD_REF separately. */
+		if (UNIV_UNLIKELY(rec_offs_any_extern(offsets))) {
+			ut_ad(dict_index_is_clust(index));
+
+			err = page_zip_compress_clust_ext(
+				LOGFILE
+				c_stream, rec, offsets, trx_id_col,
+				deleted, storage, &externs, n_blobs);
+
+			if (UNIV_UNLIKELY(err != Z_OK)) {
+
+				goto func_exit;
+			}
+		} else {
+			ulint		len;
+			const byte*	src;
+
+			/* Store trx_id and roll_ptr in uncompressed form. */
+			src = rec_get_nth_field(rec, offsets,
+						trx_id_col, &len);
+			ut_ad(src + DATA_TRX_ID_LEN
+			      == rec_get_nth_field(rec, offsets,
+						   trx_id_col + 1, &len));
+			ut_ad(len == DATA_ROLL_PTR_LEN);
+			UNIV_MEM_ASSERT_RW(rec, rec_offs_data_size(offsets));
+			UNIV_MEM_ASSERT_RW(rec - rec_offs_extra_size(offsets),
+					   rec_offs_extra_size(offsets));
+
+			/* Compress any preceding bytes. */
+			c_stream->avail_in = src - c_stream->next_in;
+
+			if (c_stream->avail_in) {
+				err = deflate(c_stream, Z_NO_FLUSH);
+				if (UNIV_UNLIKELY(err != Z_OK)) {
+
+					return(err);
+				}
+			}
+
+			ut_ad(!c_stream->avail_in);
+			ut_ad(c_stream->next_in == src);
+
+			memcpy(storage
+			       - (DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN)
+			       * (rec_get_heap_no_new(rec) - 1),
+			       c_stream->next_in,
+			       DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN);
+
+			c_stream->next_in
+				+= DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN;
+
+			/* Skip also roll_ptr */
+			ut_ad(trx_id_col + 1 < rec_offs_n_fields(offsets));
+		}
+
+		/* Compress the last bytes of the record. */
+		c_stream->avail_in = rec + rec_offs_data_size(offsets)
+			- c_stream->next_in;
+
+		if (c_stream->avail_in) {
+			err = deflate(c_stream, Z_NO_FLUSH);
+			if (UNIV_UNLIKELY(err != Z_OK)) {
+
+				goto func_exit;
+			}
+		}
+		ut_ad(!c_stream->avail_in);
+	} while (--n_dense);
+
+func_exit:
+	return(err);
+}
+
+/**************************************************************************
+Compress a page. */
+UNIV_INTERN
+ibool
+page_zip_compress(
+/*==============*/
+				/* out: TRUE on success, FALSE on failure;
+				page_zip will be left intact on failure. */
+	page_zip_des_t*	page_zip,/* in: size; out: data, n_blobs,
+				m_start, m_end, m_nonempty */
+	const page_t*	page,	/* in: uncompressed page */
+	dict_index_t*	index,	/* in: index of the B-tree node */
+	mtr_t*		mtr)	/* in: mini-transaction, or NULL */
+{
+	z_stream	c_stream;
+	int		err;
+	ulint		n_fields;/* number of index fields needed */
+	byte*		fields;	/* index field information */
+	byte*		buf;	/* compressed payload of the page */
+	byte*		buf_end;/* end of buf */
+	ulint		n_dense;
+	ulint		slot_size;/* amount of uncompressed bytes per record */
+	const rec_t**	recs;	/* dense page directory, sorted by address */
+	mem_heap_t*	heap;
+	ulint		trx_id_col;
+	ulint*		offsets	= NULL;
+	ulint		n_blobs	= 0;
+	byte*		storage;/* storage of uncompressed columns */
+	ullint		usec = ut_time_us(NULL);
+#ifdef PAGE_ZIP_COMPRESS_DBG
+	FILE*		logfile = NULL;
+#endif
+
+	ut_a(page_is_comp(page));
+	ut_a(fil_page_get_type(page) == FIL_PAGE_INDEX);
+	ut_ad(page_simple_validate_new((page_t*) page));
+	ut_ad(page_zip_simple_validate(page_zip));
+
+	UNIV_MEM_ASSERT_RW(page, UNIV_PAGE_SIZE);
+
+	/* Check the data that will be omitted. */
+	ut_a(!memcmp(page + (PAGE_NEW_INFIMUM - REC_N_NEW_EXTRA_BYTES),
+		     infimum_extra, sizeof infimum_extra));
+	ut_a(!memcmp(page + PAGE_NEW_INFIMUM,
+		     infimum_data, sizeof infimum_data));
+	ut_a(page[PAGE_NEW_SUPREMUM - REC_N_NEW_EXTRA_BYTES]
+	     /* info_bits == 0, n_owned <= max */
+	     <= PAGE_DIR_SLOT_MAX_N_OWNED);
+	ut_a(!memcmp(page + (PAGE_NEW_SUPREMUM - REC_N_NEW_EXTRA_BYTES + 1),
+		     supremum_extra_data, sizeof supremum_extra_data));
+
+	if (UNIV_UNLIKELY(!page_get_n_recs(page))) {
+		ut_a(rec_get_next_offs(page + PAGE_NEW_INFIMUM, TRUE)
+		     == PAGE_NEW_SUPREMUM);
+	}
+
+	if (page_is_leaf(page)) {
+		n_fields = dict_index_get_n_fields(index);
+	} else {
+		n_fields = dict_index_get_n_unique_in_tree(index);
+	}
+
+	/* The dense directory excludes the infimum and supremum records. */
+	n_dense = page_dir_get_n_heap(page) - PAGE_HEAP_NO_USER_LOW;
+#ifdef PAGE_ZIP_COMPRESS_DBG
+	if (UNIV_UNLIKELY(page_zip_compress_dbg)) {
+		fprintf(stderr, "compress %p %p %lu %lu %lu\n",
+			(void*) page_zip, (void*) page,
+			page_is_leaf(page),
+			n_fields, n_dense);
+	}
+	if (UNIV_UNLIKELY(page_zip_compress_log)) {
+		/* Create a log file for every compression attempt. */
+		char	logfilename[9];
+		ut_snprintf(logfilename, sizeof logfilename,
+			    "%08x", page_zip_compress_log++);
+		logfile = fopen(logfilename, "wb");
+
+		if (logfile) {
+			/* Write the uncompressed page to the log. */
+			fwrite(page, 1, UNIV_PAGE_SIZE, logfile);
+			/* Record the compressed size as zero.
+			This will be overwritten at successful exit. */
+			putc(0, logfile);
+			putc(0, logfile);
+			putc(0, logfile);
+			putc(0, logfile);
+		}
+	}
+#endif /* PAGE_ZIP_COMPRESS_DBG */
+	page_zip_stat[page_zip->ssize - 1].compressed++;
+
+	if (UNIV_UNLIKELY(n_dense * PAGE_ZIP_DIR_SLOT_SIZE
+			  >= page_zip_get_size(page_zip))) {
+
+		goto err_exit;
+	}
+
+	heap = mem_heap_create(page_zip_get_size(page_zip)
+			       + n_fields * (2 + sizeof *offsets)
+			       + n_dense * ((sizeof *recs)
+					    - PAGE_ZIP_DIR_SLOT_SIZE)
+			       + UNIV_PAGE_SIZE * 4
+			       + (512 << MAX_MEM_LEVEL));
+
+	recs = mem_heap_zalloc(heap, n_dense * sizeof *recs);
+
+	fields = mem_heap_alloc(heap, (n_fields + 1) * 2);
+
+	buf = mem_heap_alloc(heap, page_zip_get_size(page_zip) - PAGE_DATA);
+	buf_end = buf + page_zip_get_size(page_zip) - PAGE_DATA;
+
+	/* Compress the data payload. */
+	page_zip_set_alloc(&c_stream, heap);
+
+	err = deflateInit2(&c_stream, Z_DEFAULT_COMPRESSION,
+			   Z_DEFLATED, UNIV_PAGE_SIZE_SHIFT,
+			   MAX_MEM_LEVEL, Z_DEFAULT_STRATEGY);
+	ut_a(err == Z_OK);
+
+	c_stream.next_out = buf;
+	/* Subtract the space reserved for uncompressed data. */
+	/* Page header and the end marker of the modification log */
+	c_stream.avail_out = buf_end - buf - 1;
+	/* Dense page directory and uncompressed columns, if any */
+	if (page_is_leaf(page)) {
+		if (dict_index_is_clust(index)) {
+			trx_id_col = dict_index_get_sys_col_pos(
+				index, DATA_TRX_ID);
+			ut_ad(trx_id_col > 0);
+			ut_ad(trx_id_col != ULINT_UNDEFINED);
+
+			slot_size = PAGE_ZIP_DIR_SLOT_SIZE
+				+ DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN;
+		} else {
+			/* Signal the absence of trx_id
+			in page_zip_fields_encode() */
+			ut_ad(dict_index_get_sys_col_pos(index, DATA_TRX_ID)
+			      == ULINT_UNDEFINED);
+			trx_id_col = 0;
+			slot_size = PAGE_ZIP_DIR_SLOT_SIZE;
+		}
+	} else {
+		slot_size = PAGE_ZIP_DIR_SLOT_SIZE + REC_NODE_PTR_SIZE;
+		trx_id_col = ULINT_UNDEFINED;
+	}
+
+	if (UNIV_UNLIKELY(c_stream.avail_out <= n_dense * slot_size
+			  + 6/* sizeof(zlib header and footer) */)) {
+		goto zlib_error;
+	}
+
+	c_stream.avail_out -= n_dense * slot_size;
+	c_stream.avail_in = page_zip_fields_encode(n_fields, index,
+						   trx_id_col, fields);
+	c_stream.next_in = fields;
+	if (UNIV_LIKELY(!trx_id_col)) {
+		trx_id_col = ULINT_UNDEFINED;
+	}
+
+	UNIV_MEM_ASSERT_RW(c_stream.next_in, c_stream.avail_in);
+	err = deflate(&c_stream, Z_FULL_FLUSH);
+	if (err != Z_OK) {
+		goto zlib_error;
+	}
+
+	ut_ad(!c_stream.avail_in);
+
+	page_zip_dir_encode(page, buf_end, recs);
+
+	c_stream.next_in = (byte*) page + PAGE_ZIP_START;
+
+	storage = buf_end - n_dense * PAGE_ZIP_DIR_SLOT_SIZE;
+
+	/* Compress the records in heap_no order. */
+	if (UNIV_UNLIKELY(!n_dense)) {
+	} else if (!page_is_leaf(page)) {
+		/* This is a node pointer page. */
+		err = page_zip_compress_node_ptrs(LOGFILE
+						  &c_stream, recs, n_dense,
+						  index, storage, heap);
+		if (UNIV_UNLIKELY(err != Z_OK)) {
+			goto zlib_error;
+		}
+	} else if (UNIV_LIKELY(trx_id_col == ULINT_UNDEFINED)) {
+		/* This is a leaf page in a secondary index. */
+		err = page_zip_compress_sec(LOGFILE
+					    &c_stream, recs, n_dense);
+		if (UNIV_UNLIKELY(err != Z_OK)) {
+			goto zlib_error;
+		}
+	} else {
+		/* This is a leaf page in a clustered index. */
+		err = page_zip_compress_clust(LOGFILE
+					      &c_stream, recs, n_dense,
+					      index, &n_blobs, trx_id_col,
+					      buf_end - PAGE_ZIP_DIR_SLOT_SIZE
+					      * page_get_n_recs(page),
+					      storage, heap);
+		if (UNIV_UNLIKELY(err != Z_OK)) {
+			goto zlib_error;
+		}
+	}
+
+	/* Finish the compression. */
+	ut_ad(!c_stream.avail_in);
+	/* Compress any trailing garbage, in case the last record was
+	allocated from an originally longer space on the free list,
+	or the data of the last record from page_zip_compress_sec(). */
+	c_stream.avail_in
+		= page_header_get_field(page, PAGE_HEAP_TOP)
+		- (c_stream.next_in - page);
+	ut_a(c_stream.avail_in <= UNIV_PAGE_SIZE - PAGE_ZIP_START - PAGE_DIR);
+
+	UNIV_MEM_ASSERT_RW(c_stream.next_in, c_stream.avail_in);
+	err = deflate(&c_stream, Z_FINISH);
+
+	if (UNIV_UNLIKELY(err != Z_STREAM_END)) {
+zlib_error:
+		deflateEnd(&c_stream);
+		mem_heap_free(heap);
+err_exit:
+#ifdef PAGE_ZIP_COMPRESS_DBG
+		if (logfile) {
+			fclose(logfile);
+		}
+#endif /* PAGE_ZIP_COMPRESS_DBG */
+		page_zip_stat[page_zip->ssize - 1].compressed_usec
+			+= ut_time_us(NULL) - usec;
+		return(FALSE);
+	}
+
+	err = deflateEnd(&c_stream);
+	ut_a(err == Z_OK);
+
+	ut_ad(buf + c_stream.total_out == c_stream.next_out);
+	ut_ad((ulint) (storage - c_stream.next_out) >= c_stream.avail_out);
+
+	/* Valgrind believes that zlib does not initialize some bits
+	in the last 7 or 8 bytes of the stream.  Make Valgrind happy. */
+	UNIV_MEM_VALID(buf, c_stream.total_out);
+
+	/* Zero out the area reserved for the modification log.
+	Space for the end marker of the modification log is not
+	included in avail_out. */
+	memset(c_stream.next_out, 0, c_stream.avail_out + 1/* end marker */);
+
+#ifdef UNIV_DEBUG
+	page_zip->m_start =
+#endif /* UNIV_DEBUG */
+		page_zip->m_end = PAGE_DATA + c_stream.total_out;
+	page_zip->m_nonempty = FALSE;
+	page_zip->n_blobs = n_blobs;
+	/* Copy those header fields that will not be written
+	in buf_flush_init_for_writing() */
+	memcpy(page_zip->data + FIL_PAGE_PREV, page + FIL_PAGE_PREV,
+	       FIL_PAGE_LSN - FIL_PAGE_PREV);
+	memcpy(page_zip->data + FIL_PAGE_TYPE, page + FIL_PAGE_TYPE, 2);
+	memcpy(page_zip->data + FIL_PAGE_DATA, page + FIL_PAGE_DATA,
+	       PAGE_DATA - FIL_PAGE_DATA);
+	/* Copy the rest of the compressed page */
+	memcpy(page_zip->data + PAGE_DATA, buf,
+	       page_zip_get_size(page_zip) - PAGE_DATA);
+	mem_heap_free(heap);
+#ifdef UNIV_ZIP_DEBUG
+	ut_a(page_zip_validate(page_zip, page));
+#endif /* UNIV_ZIP_DEBUG */
+
+	if (mtr) {
+		page_zip_compress_write_log(page_zip, page, index, mtr);
+	}
+
+	UNIV_MEM_ASSERT_RW(page_zip->data, page_zip_get_size(page_zip));
+
+#ifdef PAGE_ZIP_COMPRESS_DBG
+	if (logfile) {
+		/* Record the compressed size of the block. */
+		byte sz[4];
+		mach_write_to_4(sz, c_stream.total_out);
+		fseek(logfile, UNIV_PAGE_SIZE, SEEK_SET);
+		fwrite(sz, 1, sizeof sz, logfile);
+		fclose(logfile);
+	}
+#endif /* PAGE_ZIP_COMPRESS_DBG */
+	{
+		page_zip_stat_t*	zip_stat
+			= &page_zip_stat[page_zip->ssize - 1];
+		zip_stat->compressed_ok++;
+		zip_stat->compressed_usec += ut_time_us(NULL) - usec;
+	}
+
+	return(TRUE);
+}
+
+/**************************************************************************
+Compare two page directory entries. */
+UNIV_INLINE
+ibool
+page_zip_dir_cmp(
+/*=============*/
+				/* out: positive if rec1 > rec2 */
+	const rec_t*	rec1,	/* in: rec1 */
+	const rec_t*	rec2)	/* in: rec2 */
+{
+	return(rec1 > rec2);
+}
+
+/**************************************************************************
+Sort the dense page directory by address (heap_no). */
+static
+void
+page_zip_dir_sort(
+/*==============*/
+	rec_t**	arr,	/* in/out: dense page directory */
+	rec_t**	aux_arr,/* in/out: work area */
+	ulint	low,	/* in: lower bound of the sorting area, inclusive */
+	ulint	high)	/* in: upper bound of the sorting area, exclusive */
+{
+	UT_SORT_FUNCTION_BODY(page_zip_dir_sort, arr, aux_arr, low, high,
+			      page_zip_dir_cmp);
+}
+
+/**************************************************************************
+Deallocate the index information initialized by page_zip_fields_decode(). */
+static
+void
+page_zip_fields_free(
+/*=================*/
+	dict_index_t*	index)	/* in: dummy index to be freed */
+{
+	if (index) {
+		dict_table_t*	table = index->table;
+		mem_heap_free(index->heap);
+		mutex_free(&(table->autoinc_mutex));
+		mem_heap_free(table->heap);
+	}
+}
+
+/**************************************************************************
+Read the index information for the compressed page. */
+static
+dict_index_t*
+page_zip_fields_decode(
+/*===================*/
+				/* out,own: dummy index describing the page,
+				or NULL on error */
+	const byte*	buf,	/* in: index information */
+	const byte*	end,	/* in: end of buf */
+	ulint*		trx_id_col)/* in: NULL for non-leaf pages;
+				for leaf pages, pointer to where to store
+				the position of the trx_id column */
+{
+	const byte*	b;
+	ulint		n;
+	ulint		i;
+	ulint		val;
+	dict_table_t*	table;
+	dict_index_t*	index;
+
+	/* Determine the number of fields. */
+	for (b = buf, n = 0; b < end; n++) {
+		if (*b++ & 0x80) {
+			b++; /* skip the second byte */
+		}
+	}
+
+	n--; /* n_nullable or trx_id */
+
+	if (UNIV_UNLIKELY(n > REC_MAX_N_FIELDS)) {
+
+		page_zip_fail(("page_zip_fields_decode: n = %lu\n",
+			       (ulong) n));
+		return(NULL);
+	}
+
+	if (UNIV_UNLIKELY(b > end)) {
+
+		page_zip_fail(("page_zip_fields_decode: %p > %p\n",
+			       (const void*) b, (const void*) end));
+		return(NULL);
+	}
+
+	table = dict_mem_table_create("ZIP_DUMMY", DICT_HDR_SPACE, n,
+				      DICT_TF_COMPACT);
+	index = dict_mem_index_create("ZIP_DUMMY", "ZIP_DUMMY",
+				      DICT_HDR_SPACE, 0, n);
+	index->table = table;
+	index->n_uniq = n;
+	/* avoid ut_ad(index->cached) in dict_index_get_n_unique_in_tree */
+	index->cached = TRUE;
+
+	/* Initialize the fields. */
+	for (b = buf, i = 0; i < n; i++) {
+		ulint	mtype;
+		ulint	len;
+
+		val = *b++;
+
+		if (UNIV_UNLIKELY(val & 0x80)) {
+			/* fixed length > 62 bytes */
+			val = (val & 0x7f) << 8 | *b++;
+			len = val >> 1;
+			mtype = DATA_FIXBINARY;
+		} else if (UNIV_UNLIKELY(val >= 126)) {
+			/* variable length with max > 255 bytes */
+			len = 0x7fff;
+			mtype = DATA_BINARY;
+		} else if (val <= 1) {
+			/* variable length with max <= 255 bytes */
+			len = 0;
+			mtype = DATA_BINARY;
+		} else {
+			/* fixed length < 62 bytes */
+			len = val >> 1;
+			mtype = DATA_FIXBINARY;
+		}
+
+		dict_mem_table_add_col(table, NULL, NULL, mtype,
+				       val & 1 ? DATA_NOT_NULL : 0, len);
+		dict_index_add_col(index, table,
+				   dict_table_get_nth_col(table, i), 0);
+	}
+
+	val = *b++;
+	if (UNIV_UNLIKELY(val & 0x80)) {
+		val = (val & 0x7f) << 8 | *b++;
+	}
+
+	/* Decode the position of the trx_id column. */
+	if (trx_id_col) {
+		if (!val) {
+			val = ULINT_UNDEFINED;
+		} else if (UNIV_UNLIKELY(val >= n)) {
+			page_zip_fields_free(index);
+			index = NULL;
+		} else {
+			index->type = DICT_CLUSTERED;
+		}
+
+		*trx_id_col = val;
+	} else {
+		/* Decode the number of nullable fields. */
+		if (UNIV_UNLIKELY(index->n_nullable > val)) {
+			page_zip_fields_free(index);
+			index = NULL;
+		} else {
+			index->n_nullable = val;
+		}
+	}
+
+	ut_ad(b == end);
+
+	return(index);
+}
+
+/**************************************************************************
+Populate the sparse page directory from the dense directory. */
+static
+ibool
+page_zip_dir_decode(
+/*================*/
+					/* out: TRUE on success,
+					FALSE on failure */
+	const page_zip_des_t*	page_zip,/* in: dense page directory on
+					compressed page */
+	page_t*			page,	/* in: compact page with valid header;
+					out: trailer and sparse page directory
+					filled in */
+	rec_t**			recs,	/* out: dense page directory sorted by
+					ascending address (and heap_no) */
+	rec_t**			recs_aux,/* in/out: scratch area */
+	ulint			n_dense)/* in: number of user records, and
+					size of recs[] and recs_aux[] */
+{
+	ulint	i;
+	ulint	n_recs;
+	byte*	slot;
+
+	n_recs = page_get_n_recs(page);
+
+	if (UNIV_UNLIKELY(n_recs > n_dense)) {
+		page_zip_fail(("page_zip_dir_decode 1: %lu > %lu\n",
+			       (ulong) n_recs, (ulong) n_dense));
+		return(FALSE);
+	}
+
+	/* Traverse the list of stored records in the sorting order,
+	starting from the first user record. */
+
+	slot = page + (UNIV_PAGE_SIZE - PAGE_DIR - PAGE_DIR_SLOT_SIZE);
+	UNIV_PREFETCH_RW(slot);
+
+	/* Zero out the page trailer. */
+	memset(slot + PAGE_DIR_SLOT_SIZE, 0, PAGE_DIR);
+
+	mach_write_to_2(slot, PAGE_NEW_INFIMUM);
+	slot -= PAGE_DIR_SLOT_SIZE;
+	UNIV_PREFETCH_RW(slot);
+
+	/* Initialize the sparse directory and copy the dense directory. */
+	for (i = 0; i < n_recs; i++) {
+		ulint	offs = page_zip_dir_get(page_zip, i);
+
+		if (offs & PAGE_ZIP_DIR_SLOT_OWNED) {
+			mach_write_to_2(slot, offs & PAGE_ZIP_DIR_SLOT_MASK);
+			slot -= PAGE_DIR_SLOT_SIZE;
+			UNIV_PREFETCH_RW(slot);
+		}
+
+		if (UNIV_UNLIKELY((offs & PAGE_ZIP_DIR_SLOT_MASK)
+				  < PAGE_ZIP_START + REC_N_NEW_EXTRA_BYTES)) {
+			page_zip_fail(("page_zip_dir_decode 2: %u %u %lx\n",
+				       (unsigned) i, (unsigned) n_recs,
+				       (ulong) offs));
+			return(FALSE);
+		}
+
+		recs[i] = page + (offs & PAGE_ZIP_DIR_SLOT_MASK);
+	}
+
+	mach_write_to_2(slot, PAGE_NEW_SUPREMUM);
+	{
+		const page_dir_slot_t*	last_slot = page_dir_get_nth_slot(
+			page, page_dir_get_n_slots(page) - 1);
+
+		if (UNIV_UNLIKELY(slot != last_slot)) {
+			page_zip_fail(("page_zip_dir_decode 3: %p != %p\n",
+				       (const void*) slot,
+				       (const void*) last_slot));
+			return(FALSE);
+		}
+	}
+
+	/* Copy the rest of the dense directory. */
+	for (; i < n_dense; i++) {
+		ulint	offs = page_zip_dir_get(page_zip, i);
+
+		if (UNIV_UNLIKELY(offs & ~PAGE_ZIP_DIR_SLOT_MASK)) {
+			page_zip_fail(("page_zip_dir_decode 4: %u %u %lx\n",
+				       (unsigned) i, (unsigned) n_dense,
+				       (ulong) offs));
+			return(FALSE);
+		}
+
+		recs[i] = page + offs;
+	}
+
+	if (UNIV_LIKELY(n_dense > 1)) {
+		page_zip_dir_sort(recs, recs_aux, 0, n_dense);
+	}
+	return(TRUE);
+}
+
+/**************************************************************************
+Initialize the REC_N_NEW_EXTRA_BYTES of each record. */
+static
+ibool
+page_zip_set_extra_bytes(
+/*=====================*/
+					/* out: TRUE on success,
+					FALSE on failure */
+	const page_zip_des_t*	page_zip,/* in: compressed page */
+	page_t*			page,	/* in/out: uncompressed page */
+	ulint			info_bits)/* in: REC_INFO_MIN_REC_FLAG or 0 */
+{
+	ulint	n;
+	ulint	i;
+	ulint	n_owned = 1;
+	ulint	offs;
+	rec_t*	rec;
+
+	n = page_get_n_recs(page);
+	rec = page + PAGE_NEW_INFIMUM;
+
+	for (i = 0; i < n; i++) {
+		offs = page_zip_dir_get(page_zip, i);
+
+		if (UNIV_UNLIKELY(offs & PAGE_ZIP_DIR_SLOT_DEL)) {
+			info_bits |= REC_INFO_DELETED_FLAG;
+		}
+		if (UNIV_UNLIKELY(offs & PAGE_ZIP_DIR_SLOT_OWNED)) {
+			info_bits |= n_owned;
+			n_owned = 1;
+		} else {
+			n_owned++;
+		}
+		offs &= PAGE_ZIP_DIR_SLOT_MASK;
+		if (UNIV_UNLIKELY(offs < PAGE_ZIP_START
+				  + REC_N_NEW_EXTRA_BYTES)) {
+			page_zip_fail(("page_zip_set_extra_bytes 1:"
+				       " %u %u %lx\n",
+				       (unsigned) i, (unsigned) n,
+				       (ulong) offs));
+			return(FALSE);
+		}
+
+		rec_set_next_offs_new(rec, offs);
+		rec = page + offs;
+		rec[-REC_N_NEW_EXTRA_BYTES] = (byte) info_bits;
+		info_bits = 0;
+	}
+
+	/* Set the next pointer of the last user record. */
+	rec_set_next_offs_new(rec, PAGE_NEW_SUPREMUM);
+
+	/* Set n_owned of the supremum record. */
+	page[PAGE_NEW_SUPREMUM - REC_N_NEW_EXTRA_BYTES] = (byte) n_owned;
+
+	/* The dense directory excludes the infimum and supremum records. */
+	n = page_dir_get_n_heap(page) - PAGE_HEAP_NO_USER_LOW;
+
+	if (i >= n) {
+		if (UNIV_LIKELY(i == n)) {
+			return(TRUE);
+		}
+
+		page_zip_fail(("page_zip_set_extra_bytes 2: %u != %u\n",
+			       (unsigned) i, (unsigned) n));
+		return(FALSE);
+	}
+
+	offs = page_zip_dir_get(page_zip, i);
+
+	/* Set the extra bytes of deleted records on the free list. */
+	for (;;) {
+		if (UNIV_UNLIKELY(!offs)
+		    || UNIV_UNLIKELY(offs & ~PAGE_ZIP_DIR_SLOT_MASK)) {
+
+			page_zip_fail(("page_zip_set_extra_bytes 3: %lx\n",
+				       (ulong) offs));
+			return(FALSE);
+		}
+
+		rec = page + offs;
+		rec[-REC_N_NEW_EXTRA_BYTES] = 0; /* info_bits and n_owned */
+
+		if (++i == n) {
+			break;
+		}
+
+		offs = page_zip_dir_get(page_zip, i);
+		rec_set_next_offs_new(rec, offs);
+	}
+
+	/* Terminate the free list. */
+	rec[-REC_N_NEW_EXTRA_BYTES] = 0; /* info_bits and n_owned */
+	rec_set_next_offs_new(rec, 0);
+
+	return(TRUE);
+}
+
+/**************************************************************************
+Apply the modification log to a record containing externally stored
+columns.  Do not copy the fields that are stored separately. */
+static
+const byte*
+page_zip_apply_log_ext(
+/*===================*/
+					/* out: pointer to modification log,
+					or NULL on failure */
+	rec_t*		rec,		/* in/out: record */
+	const ulint*	offsets,	/* in: rec_get_offsets(rec) */
+	ulint		trx_id_col,	/* in: position of of DB_TRX_ID */
+	const byte*	data,		/* in: modification log */
+	const byte*	end)		/* in: end of modification log */
+{
+	ulint	i;
+	ulint	len;
+	byte*	next_out = rec;
+
+	/* Check if there are any externally stored columns.
+	For each externally stored column, skip the
+	BTR_EXTERN_FIELD_REF. */
+
+	for (i = 0; i < rec_offs_n_fields(offsets); i++) {
+		byte*	dst;
+
+		if (UNIV_UNLIKELY(i == trx_id_col)) {
+			/* Skip trx_id and roll_ptr */
+			dst = rec_get_nth_field(rec, offsets,
+						i, &len);
+			if (UNIV_UNLIKELY(dst - next_out >= end - data)
+			    || UNIV_UNLIKELY
+			    (len < (DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN))
+			    || rec_offs_nth_extern(offsets, i)) {
+				page_zip_fail(("page_zip_apply_log_ext:"
+					       " trx_id len %lu,"
+					       " %p - %p >= %p - %p\n",
+					       (ulong) len,
+					       (const void*) dst,
+					       (const void*) next_out,
+					       (const void*) end,
+					       (const void*) data));
+				return(NULL);
+			}
+
+			memcpy(next_out, data, dst - next_out);
+			data += dst - next_out;
+			next_out = dst + (DATA_TRX_ID_LEN
+					  + DATA_ROLL_PTR_LEN);
+		} else if (rec_offs_nth_extern(offsets, i)) {
+			dst = rec_get_nth_field(rec, offsets,
+						i, &len);
+			ut_ad(len
+			      >= BTR_EXTERN_FIELD_REF_SIZE);
+
+			len += dst - next_out
+				- BTR_EXTERN_FIELD_REF_SIZE;
+
+			if (UNIV_UNLIKELY(data + len >= end)) {
+				page_zip_fail(("page_zip_apply_log_ext: "
+					       "ext %p+%lu >= %p\n",
+					       (const void*) data,
+					       (ulong) len,
+					       (const void*) end));
+				return(NULL);
+			}
+
+			memcpy(next_out, data, len);
+			data += len;
+			next_out += len
+				+ BTR_EXTERN_FIELD_REF_SIZE;
+		}
+	}
+
+	/* Copy the last bytes of the record. */
+	len = rec_get_end(rec, offsets) - next_out;
+	if (UNIV_UNLIKELY(data + len >= end)) {
+		page_zip_fail(("page_zip_apply_log_ext: "
+			       "last %p+%lu >= %p\n",
+			       (const void*) data,
+			       (ulong) len,
+			       (const void*) end));
+		return(NULL);
+	}
+	memcpy(next_out, data, len);
+	data += len;
+
+	return(data);
+}
+
+/**************************************************************************
+Apply the modification log to an uncompressed page.
+Do not copy the fields that are stored separately. */
+static
+const byte*
+page_zip_apply_log(
+/*===============*/
+				/* out: pointer to end of modification log,
+				or NULL on failure */
+	const byte*	data,	/* in: modification log */
+	ulint		size,	/* in: maximum length of the log, in bytes */
+	rec_t**		recs,	/* in: dense page directory,
+				sorted by address (indexed by
+				heap_no - PAGE_HEAP_NO_USER_LOW) */
+	ulint		n_dense,/* in: size of recs[] */
+	ulint		trx_id_col,/* in: column number of trx_id in the index,
+				or ULINT_UNDEFINED if none */
+	ulint		heap_status,
+				/* in: heap_no and status bits for
+				the next record to uncompress */
+	dict_index_t*	index,	/* in: index of the page */
+	ulint*		offsets)/* in/out: work area for
+				rec_get_offsets_reverse() */
+{
+	const byte* const end = data + size;
+
+	for (;;) {
+		ulint	val;
+		rec_t*	rec;
+		ulint	len;
+		ulint	hs;
+
+		val = *data++;
+		if (UNIV_UNLIKELY(!val)) {
+			return(data - 1);
+		}
+		if (val & 0x80) {
+			val = (val & 0x7f) << 8 | *data++;
+			if (UNIV_UNLIKELY(!val)) {
+				page_zip_fail(("page_zip_apply_log:"
+					       " invalid val %x%x\n",
+					       data[-2], data[-1]));
+				return(NULL);
+			}
+		}
+		if (UNIV_UNLIKELY(data >= end)) {
+			page_zip_fail(("page_zip_apply_log: %p >= %p\n",
+				       (const void*) data,
+				       (const void*) end));
+			return(NULL);
+		}
+		if (UNIV_UNLIKELY((val >> 1) > n_dense)) {
+			page_zip_fail(("page_zip_apply_log: %lu>>1 > %lu\n",
+				       (ulong) val, (ulong) n_dense));
+			return(NULL);
+		}
+
+		/* Determine the heap number and status bits of the record. */
+		rec = recs[(val >> 1) - 1];
+
+		hs = ((val >> 1) + 1) << REC_HEAP_NO_SHIFT;
+		hs |= heap_status & ((1 << REC_HEAP_NO_SHIFT) - 1);
+
+		/* This may either be an old record that is being
+		overwritten (updated in place, or allocated from
+		the free list), or a new record, with the next
+		available_heap_no. */
+		if (UNIV_UNLIKELY(hs > heap_status)) {
+			page_zip_fail(("page_zip_apply_log: %lu > %lu\n",
+				       (ulong) hs, (ulong) heap_status));
+			return(NULL);
+		} else if (hs == heap_status) {
+			/* A new record was allocated from the heap. */
+			if (UNIV_UNLIKELY(val & 1)) {
+				/* Only existing records may be cleared. */
+				page_zip_fail(("page_zip_apply_log:"
+					       " attempting to create"
+					       " deleted rec %lu\n",
+					       (ulong) hs));
+				return(NULL);
+			}
+			heap_status += 1 << REC_HEAP_NO_SHIFT;
+		}
+
+		mach_write_to_2(rec - REC_NEW_HEAP_NO, hs);
+
+		if (val & 1) {
+			/* Clear the data bytes of the record. */
+			mem_heap_t*	heap	= NULL;
+			ulint*		offs;
+			offs = rec_get_offsets(rec, index, offsets,
+					       ULINT_UNDEFINED, &heap);
+			memset(rec, 0, rec_offs_data_size(offs));
+
+			if (UNIV_LIKELY_NULL(heap)) {
+				mem_heap_free(heap);
+			}
+			continue;
+		}
+
+#if REC_STATUS_NODE_PTR != TRUE
+# error "REC_STATUS_NODE_PTR != TRUE"
+#endif
+		rec_get_offsets_reverse(data, index,
+					hs & REC_STATUS_NODE_PTR,
+					offsets);
+		rec_offs_make_valid(rec, index, offsets);
+
+		/* Copy the extra bytes (backwards). */
+		{
+			byte*	start	= rec_get_start(rec, offsets);
+			byte*	b	= rec - REC_N_NEW_EXTRA_BYTES;
+			while (b != start) {
+				*--b = *data++;
+			}
+		}
+
+		/* Copy the data bytes. */
+		if (UNIV_UNLIKELY(rec_offs_any_extern(offsets))) {
+			/* Non-leaf nodes should not contain any
+			externally stored columns. */
+			if (UNIV_UNLIKELY(hs & REC_STATUS_NODE_PTR)) {
+				page_zip_fail(("page_zip_apply_log: "
+					       "%lu&REC_STATUS_NODE_PTR\n",
+					       (ulong) hs));
+				return(NULL);
+			}
+
+			data = page_zip_apply_log_ext(
+				rec, offsets, trx_id_col, data, end);
+
+			if (UNIV_UNLIKELY(!data)) {
+				return(NULL);
+			}
+		} else if (UNIV_UNLIKELY(hs & REC_STATUS_NODE_PTR)) {
+			len = rec_offs_data_size(offsets)
+				- REC_NODE_PTR_SIZE;
+			/* Copy the data bytes, except node_ptr. */
+			if (UNIV_UNLIKELY(data + len >= end)) {
+				page_zip_fail(("page_zip_apply_log: "
+					       "node_ptr %p+%lu >= %p\n",
+					       (const void*) data,
+					       (ulong) len,
+					       (const void*) end));
+				return(NULL);
+			}
+			memcpy(rec, data, len);
+			data += len;
+		} else if (UNIV_LIKELY(trx_id_col == ULINT_UNDEFINED)) {
+			len = rec_offs_data_size(offsets);
+
+			/* Copy all data bytes of
+			a record in a secondary index. */
+			if (UNIV_UNLIKELY(data + len >= end)) {
+				page_zip_fail(("page_zip_apply_log: "
+					       "sec %p+%lu >= %p\n",
+					       (const void*) data,
+					       (ulong) len,
+					       (const void*) end));
+				return(NULL);
+			}
+
+			memcpy(rec, data, len);
+			data += len;
+		} else {
+			/* Skip DB_TRX_ID and DB_ROLL_PTR. */
+			ulint	l = rec_get_nth_field_offs(offsets,
+							   trx_id_col, &len);
+			byte*	b;
+
+			if (UNIV_UNLIKELY(data + l >= end)
+			    || UNIV_UNLIKELY(len < (DATA_TRX_ID_LEN
+						    + DATA_ROLL_PTR_LEN))) {
+				page_zip_fail(("page_zip_apply_log: "
+					       "trx_id %p+%lu >= %p\n",
+					       (const void*) data,
+					       (ulong) l,
+					       (const void*) end));
+				return(NULL);
+			}
+
+			/* Copy any preceding data bytes. */
+			memcpy(rec, data, l);
+			data += l;
+
+			/* Copy any bytes following DB_TRX_ID, DB_ROLL_PTR. */
+			b = rec + l + (DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN);
+			len = rec_get_end(rec, offsets) - b;
+			if (UNIV_UNLIKELY(data + len >= end)) {
+				page_zip_fail(("page_zip_apply_log: "
+					       "clust %p+%lu >= %p\n",
+					       (const void*) data,
+					       (ulong) len,
+					       (const void*) end));
+				return(NULL);
+			}
+			memcpy(b, data, len);
+			data += len;
+		}
+	}
+}
+
+/**************************************************************************
+Decompress the records of a node pointer page. */
+static
+ibool
+page_zip_decompress_node_ptrs(
+/*==========================*/
+					/* out: TRUE on success,
+					FALSE on failure */
+	page_zip_des_t*	page_zip,	/* in/out: compressed page */
+	z_stream*	d_stream,	/* in/out: compressed page stream */
+	rec_t**		recs,		/* in: dense page directory
+					sorted by address */
+	ulint		n_dense,	/* in: size of recs[] */
+	dict_index_t*	index,		/* in: the index of the page */
+	ulint*		offsets,	/* in/out: temporary offsets */
+	mem_heap_t*	heap)		/* in: temporary memory heap */
+{
+	ulint		heap_status = REC_STATUS_NODE_PTR
+		| PAGE_HEAP_NO_USER_LOW << REC_HEAP_NO_SHIFT;
+	ulint		slot;
+	const byte*	storage;
+
+	/* Subtract the space reserved for uncompressed data. */
+	d_stream->avail_in -= n_dense
+		* (PAGE_ZIP_DIR_SLOT_SIZE + REC_NODE_PTR_SIZE);
+
+	/* Decompress the records in heap_no order. */
+	for (slot = 0; slot < n_dense; slot++) {
+		rec_t*	rec = recs[slot];
+
+		d_stream->avail_out = rec - REC_N_NEW_EXTRA_BYTES
+			- d_stream->next_out;
+
+		ut_ad(d_stream->avail_out < UNIV_PAGE_SIZE
+		      - PAGE_ZIP_START - PAGE_DIR);
+		switch (inflate(d_stream, Z_SYNC_FLUSH)) {
+		case Z_STREAM_END:
+			/* Apparently, n_dense has grown
+			since the time the page was last compressed. */
+			goto zlib_done;
+		case Z_OK:
+		case Z_BUF_ERROR:
+			if (!d_stream->avail_out) {
+				break;
+			}
+			/* fall through */
+		default:
+			page_zip_fail(("page_zip_decompress_node_ptrs:"
+				       " 1 inflate(Z_SYNC_FLUSH)=%s\n",
+				       d_stream->msg));
+			goto zlib_error;
+		}
+
+		ut_ad(d_stream->next_out == rec - REC_N_NEW_EXTRA_BYTES);
+		/* Prepare to decompress the data bytes. */
+		d_stream->next_out = rec;
+		/* Set heap_no and the status bits. */
+		mach_write_to_2(rec - REC_NEW_HEAP_NO, heap_status);
+		heap_status += 1 << REC_HEAP_NO_SHIFT;
+
+		/* Read the offsets. The status bits are needed here. */
+		offsets = rec_get_offsets(rec, index, offsets,
+					  ULINT_UNDEFINED, &heap);
+
+		/* Non-leaf nodes should not have any externally
+		stored columns. */
+		ut_ad(!rec_offs_any_extern(offsets));
+
+		/* Decompress the data bytes, except node_ptr. */
+		d_stream->avail_out = rec_offs_data_size(offsets)
+			- REC_NODE_PTR_SIZE;
+
+		switch (inflate(d_stream, Z_SYNC_FLUSH)) {
+		case Z_STREAM_END:
+			goto zlib_done;
+		case Z_OK:
+		case Z_BUF_ERROR:
+			if (!d_stream->avail_out) {
+				break;
+			}
+			/* fall through */
+		default:
+			page_zip_fail(("page_zip_decompress_node_ptrs:"
+				       " 2 inflate(Z_SYNC_FLUSH)=%s\n",
+				       d_stream->msg));
+			goto zlib_error;
+		}
+
+		/* Clear the node pointer in case the record
+		will be deleted and the space will be reallocated
+		to a smaller record. */
+		memset(d_stream->next_out, 0, REC_NODE_PTR_SIZE);
+		d_stream->next_out += REC_NODE_PTR_SIZE;
+
+		ut_ad(d_stream->next_out == rec_get_end(rec, offsets));
+	}
+
+	/* Decompress any trailing garbage, in case the last record was
+	allocated from an originally longer space on the free list. */
+	d_stream->avail_out = page_header_get_field(page_zip->data,
+						    PAGE_HEAP_TOP)
+		- page_offset(d_stream->next_out);
+	if (UNIV_UNLIKELY(d_stream->avail_out > UNIV_PAGE_SIZE
+			  - PAGE_ZIP_START - PAGE_DIR)) {
+
+		page_zip_fail(("page_zip_decompress_node_ptrs:"
+			       " avail_out = %u\n",
+			       d_stream->avail_out));
+		goto zlib_error;
+	}
+
+	if (UNIV_UNLIKELY(inflate(d_stream, Z_FINISH) != Z_STREAM_END)) {
+		page_zip_fail(("page_zip_decompress_node_ptrs:"
+			       " inflate(Z_FINISH)=%s\n",
+			       d_stream->msg));
+zlib_error:
+		inflateEnd(d_stream);
+		return(FALSE);
+	}
+
+	/* Note that d_stream->avail_out > 0 may hold here
+	if the modification log is nonempty. */
+
+zlib_done:
+	if (UNIV_UNLIKELY(inflateEnd(d_stream) != Z_OK)) {
+		ut_error;
+	}
+
+	{
+		page_t*	page = page_align(d_stream->next_out);
+
+		/* Clear the unused heap space on the uncompressed page. */
+		memset(d_stream->next_out, 0,
+		       page_dir_get_nth_slot(page,
+					     page_dir_get_n_slots(page) - 1)
+		       - d_stream->next_out);
+	}
+
+#ifdef UNIV_DEBUG
+	page_zip->m_start = PAGE_DATA + d_stream->total_in;
+#endif /* UNIV_DEBUG */
+
+	/* Apply the modification log. */
+	{
+		const byte*	mod_log_ptr;
+		mod_log_ptr = page_zip_apply_log(d_stream->next_in,
+						 d_stream->avail_in + 1,
+						 recs, n_dense,
+						 ULINT_UNDEFINED, heap_status,
+						 index, offsets);
+
+		if (UNIV_UNLIKELY(!mod_log_ptr)) {
+			return(FALSE);
+		}
+		page_zip->m_end = mod_log_ptr - page_zip->data;
+		page_zip->m_nonempty = mod_log_ptr != d_stream->next_in;
+	}
+
+	if (UNIV_UNLIKELY
+	    (page_zip_get_trailer_len(page_zip,
+				      dict_index_is_clust(index), NULL)
+	     + page_zip->m_end >= page_zip_get_size(page_zip))) {
+		page_zip_fail(("page_zip_decompress_node_ptrs:"
+			       " %lu + %lu >= %lu, %lu\n",
+			       (ulong) page_zip_get_trailer_len(
+				       page_zip, dict_index_is_clust(index),
+				       NULL),
+			       (ulong) page_zip->m_end,
+			       (ulong) page_zip_get_size(page_zip),
+			       (ulong) dict_index_is_clust(index)));
+		return(FALSE);
+	}
+
+	/* Restore the uncompressed columns in heap_no order. */
+	storage	= page_zip->data + page_zip_get_size(page_zip)
+		- n_dense * PAGE_ZIP_DIR_SLOT_SIZE;
+
+	for (slot = 0; slot < n_dense; slot++) {
+		rec_t*		rec	= recs[slot];
+
+		offsets = rec_get_offsets(rec, index, offsets,
+					  ULINT_UNDEFINED, &heap);
+		/* Non-leaf nodes should not have any externally
+		stored columns. */
+		ut_ad(!rec_offs_any_extern(offsets));
+		storage -= REC_NODE_PTR_SIZE;
+
+		memcpy(rec_get_end(rec, offsets) - REC_NODE_PTR_SIZE,
+		       storage, REC_NODE_PTR_SIZE);
+	}
+
+	return(TRUE);
+}
+
+/**************************************************************************
+Decompress the records of a leaf node of a secondary index. */
+static
+ibool
+page_zip_decompress_sec(
+/*====================*/
+					/* out: TRUE on success,
+					FALSE on failure */
+	page_zip_des_t*	page_zip,	/* in/out: compressed page */
+	z_stream*	d_stream,	/* in/out: compressed page stream */
+	rec_t**		recs,		/* in: dense page directory
+					sorted by address */
+	ulint		n_dense,	/* in: size of recs[] */
+	dict_index_t*	index,		/* in: the index of the page */
+	ulint*		offsets)	/* in/out: temporary offsets */
+{
+	ulint	heap_status	= REC_STATUS_ORDINARY
+		| PAGE_HEAP_NO_USER_LOW << REC_HEAP_NO_SHIFT;
+	ulint	slot;
+
+	ut_a(!dict_index_is_clust(index));
+
+	/* Subtract the space reserved for uncompressed data. */
+	d_stream->avail_in -= n_dense * PAGE_ZIP_DIR_SLOT_SIZE;
+
+	for (slot = 0; slot < n_dense; slot++) {
+		rec_t*	rec = recs[slot];
+
+		/* Decompress everything up to this record. */
+		d_stream->avail_out = rec - REC_N_NEW_EXTRA_BYTES
+			- d_stream->next_out;
+
+		if (UNIV_LIKELY(d_stream->avail_out)) {
+			switch (inflate(d_stream, Z_SYNC_FLUSH)) {
+			case Z_STREAM_END:
+				/* Apparently, n_dense has grown
+				since the time the page was last compressed. */
+				goto zlib_done;
+			case Z_OK:
+			case Z_BUF_ERROR:
+				if (!d_stream->avail_out) {
+					break;
+				}
+				/* fall through */
+			default:
+				page_zip_fail(("page_zip_decompress_sec:"
+					       " inflate(Z_SYNC_FLUSH)=%s\n",
+					       d_stream->msg));
+				goto zlib_error;
+			}
+		}
+
+		ut_ad(d_stream->next_out == rec - REC_N_NEW_EXTRA_BYTES);
+
+		/* Skip the REC_N_NEW_EXTRA_BYTES. */
+
+		d_stream->next_out = rec;
+
+		/* Set heap_no and the status bits. */
+		mach_write_to_2(rec - REC_NEW_HEAP_NO, heap_status);
+		heap_status += 1 << REC_HEAP_NO_SHIFT;
+	}
+
+	/* Decompress the data of the last record and any trailing garbage,
+	in case the last record was allocated from an originally longer space
+	on the free list. */
+	d_stream->avail_out = page_header_get_field(page_zip->data,
+						    PAGE_HEAP_TOP)
+		- page_offset(d_stream->next_out);
+	if (UNIV_UNLIKELY(d_stream->avail_out > UNIV_PAGE_SIZE
+			  - PAGE_ZIP_START - PAGE_DIR)) {
+
+		page_zip_fail(("page_zip_decompress_sec:"
+			       " avail_out = %u\n",
+			       d_stream->avail_out));
+		goto zlib_error;
+	}
+
+	if (UNIV_UNLIKELY(inflate(d_stream, Z_FINISH) != Z_STREAM_END)) {
+		page_zip_fail(("page_zip_decompress_sec:"
+			       " inflate(Z_FINISH)=%s\n",
+			       d_stream->msg));
+zlib_error:
+		inflateEnd(d_stream);
+		return(FALSE);
+	}
+
+	/* Note that d_stream->avail_out > 0 may hold here
+	if the modification log is nonempty. */
+
+zlib_done:
+	if (UNIV_UNLIKELY(inflateEnd(d_stream) != Z_OK)) {
+		ut_error;
+	}
+
+	{
+		page_t*	page = page_align(d_stream->next_out);
+
+		/* Clear the unused heap space on the uncompressed page. */
+		memset(d_stream->next_out, 0,
+		       page_dir_get_nth_slot(page,
+					     page_dir_get_n_slots(page) - 1)
+		       - d_stream->next_out);
+	}
+
+#ifdef UNIV_DEBUG
+	page_zip->m_start = PAGE_DATA + d_stream->total_in;
+#endif /* UNIV_DEBUG */
+
+	/* Apply the modification log. */
+	{
+		const byte*	mod_log_ptr;
+		mod_log_ptr = page_zip_apply_log(d_stream->next_in,
+						 d_stream->avail_in + 1,
+						 recs, n_dense,
+						 ULINT_UNDEFINED, heap_status,
+						 index, offsets);
+
+		if (UNIV_UNLIKELY(!mod_log_ptr)) {
+			return(FALSE);
+		}
+		page_zip->m_end = mod_log_ptr - page_zip->data;
+		page_zip->m_nonempty = mod_log_ptr != d_stream->next_in;
+	}
+
+	if (UNIV_UNLIKELY(page_zip_get_trailer_len(page_zip, FALSE, NULL)
+			  + page_zip->m_end >= page_zip_get_size(page_zip))) {
+
+		page_zip_fail(("page_zip_decompress_sec: %lu + %lu >= %lu\n",
+			       (ulong) page_zip_get_trailer_len(
+				       page_zip, FALSE, NULL),
+			       (ulong) page_zip->m_end,
+			       (ulong) page_zip_get_size(page_zip)));
+		return(FALSE);
+	}
+
+	/* There are no uncompressed columns on leaf pages of
+	secondary indexes. */
+
+	return(TRUE);
+}
+
+/**************************************************************************
+Decompress a record of a leaf node of a clustered index that contains
+externally stored columns. */
+static
+ibool
+page_zip_decompress_clust_ext(
+/*==========================*/
+					/* out: TRUE on success */
+	z_stream*	d_stream,	/* in/out: compressed page stream */
+	rec_t*		rec,		/* in/out: record */
+	const ulint*	offsets,	/* in: rec_get_offsets(rec) */
+	ulint		trx_id_col)	/* in: position of of DB_TRX_ID */
+{
+	ulint	i;
+
+	for (i = 0; i < rec_offs_n_fields(offsets); i++) {
+		ulint	len;
+		byte*	dst;
+
+		if (UNIV_UNLIKELY(i == trx_id_col)) {
+			/* Skip trx_id and roll_ptr */
+			dst = rec_get_nth_field(rec, offsets, i, &len);
+			if (UNIV_UNLIKELY(len < DATA_TRX_ID_LEN
+					  + DATA_ROLL_PTR_LEN)) {
+
+				page_zip_fail(("page_zip_decompress_clust_ext:"
+					       " len[%lu] = %lu\n",
+					       (ulong) i, (ulong) len));
+				return(FALSE);
+			}
+
+			if (rec_offs_nth_extern(offsets, i)) {
+
+				page_zip_fail(("page_zip_decompress_clust_ext:"
+					       " DB_TRX_ID at %lu is ext\n",
+					       (ulong) i));
+				return(FALSE);
+			}
+
+			d_stream->avail_out = dst - d_stream->next_out;
+
+			switch (inflate(d_stream, Z_SYNC_FLUSH)) {
+			case Z_STREAM_END:
+			case Z_OK:
+			case Z_BUF_ERROR:
+				if (!d_stream->avail_out) {
+					break;
+				}
+				/* fall through */
+			default:
+				page_zip_fail(("page_zip_decompress_clust_ext:"
+					       " 1 inflate(Z_SYNC_FLUSH)=%s\n",
+					       d_stream->msg));
+				return(FALSE);
+			}
+
+			ut_ad(d_stream->next_out == dst);
+
+			/* Clear DB_TRX_ID and DB_ROLL_PTR in order to
+			avoid uninitialized bytes in case the record
+			is affected by page_zip_apply_log(). */
+			memset(dst, 0, DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN);
+
+			d_stream->next_out += DATA_TRX_ID_LEN
+				+ DATA_ROLL_PTR_LEN;
+		} else if (rec_offs_nth_extern(offsets, i)) {
+			dst = rec_get_nth_field(rec, offsets, i, &len);
+			ut_ad(len >= BTR_EXTERN_FIELD_REF_SIZE);
+			dst += len - BTR_EXTERN_FIELD_REF_SIZE;
+
+			d_stream->avail_out = dst - d_stream->next_out;
+			switch (inflate(d_stream, Z_SYNC_FLUSH)) {
+			case Z_STREAM_END:
+			case Z_OK:
+			case Z_BUF_ERROR:
+				if (!d_stream->avail_out) {
+					break;
+				}
+				/* fall through */
+			default:
+				page_zip_fail(("page_zip_decompress_clust_ext:"
+					       " 2 inflate(Z_SYNC_FLUSH)=%s\n",
+					       d_stream->msg));
+				return(FALSE);
+			}
+
+			ut_ad(d_stream->next_out == dst);
+
+			/* Clear the BLOB pointer in case
+			the record will be deleted and the
+			space will not be reused.  Note that
+			the final initialization of the BLOB
+			pointers (copying from "externs"
+			or clearing) will have to take place
+			only after the page modification log
+			has been applied.  Otherwise, we
+			could end up with an uninitialized
+			BLOB pointer when a record is deleted,
+			reallocated and deleted. */
+			memset(d_stream->next_out, 0,
+			       BTR_EXTERN_FIELD_REF_SIZE);
+			d_stream->next_out
+				+= BTR_EXTERN_FIELD_REF_SIZE;
+		}
+	}
+
+	return(TRUE);
+}
+
+/**************************************************************************
+Compress the records of a leaf node of a clustered index. */
+static
+ibool
+page_zip_decompress_clust(
+/*======================*/
+					/* out: TRUE on success,
+					FALSE on failure */
+	page_zip_des_t*	page_zip,	/* in/out: compressed page */
+	z_stream*	d_stream,	/* in/out: compressed page stream */
+	rec_t**		recs,		/* in: dense page directory
+					sorted by address */
+	ulint		n_dense,	/* in: size of recs[] */
+	dict_index_t*	index,		/* in: the index of the page */
+	ulint		trx_id_col,	/* index of the trx_id column */
+	ulint*		offsets,	/* in/out: temporary offsets */
+	mem_heap_t*	heap)		/* in: temporary memory heap */
+{
+	int		err;
+	ulint		slot;
+	ulint		heap_status	= REC_STATUS_ORDINARY
+		| PAGE_HEAP_NO_USER_LOW << REC_HEAP_NO_SHIFT;
+	const byte*	storage;
+	const byte*	externs;
+
+	ut_a(dict_index_is_clust(index));
+
+	/* Subtract the space reserved for uncompressed data. */
+	d_stream->avail_in -= n_dense * (PAGE_ZIP_DIR_SLOT_SIZE
+					 + DATA_TRX_ID_LEN
+					 + DATA_ROLL_PTR_LEN);
+
+	/* Decompress the records in heap_no order. */
+	for (slot = 0; slot < n_dense; slot++) {
+		rec_t*	rec	= recs[slot];
+
+		d_stream->avail_out = rec - REC_N_NEW_EXTRA_BYTES
+			- d_stream->next_out;
+
+		ut_ad(d_stream->avail_out < UNIV_PAGE_SIZE
+		      - PAGE_ZIP_START - PAGE_DIR);
+		err = inflate(d_stream, Z_SYNC_FLUSH);
+		switch (err) {
+		case Z_STREAM_END:
+			/* Apparently, n_dense has grown
+			since the time the page was last compressed. */
+			goto zlib_done;
+		case Z_OK:
+		case Z_BUF_ERROR:
+			if (UNIV_LIKELY(!d_stream->avail_out)) {
+				break;
+			}
+			/* fall through */
+		default:
+			page_zip_fail(("page_zip_decompress_clust:"
+				       " 1 inflate(Z_SYNC_FLUSH)=%s\n",
+				       d_stream->msg));
+			goto zlib_error;
+		}
+
+		ut_ad(d_stream->next_out == rec - REC_N_NEW_EXTRA_BYTES);
+		/* Prepare to decompress the data bytes. */
+		d_stream->next_out = rec;
+		/* Set heap_no and the status bits. */
+		mach_write_to_2(rec - REC_NEW_HEAP_NO, heap_status);
+		heap_status += 1 << REC_HEAP_NO_SHIFT;
+
+		/* Read the offsets. The status bits are needed here. */
+		offsets = rec_get_offsets(rec, index, offsets,
+					  ULINT_UNDEFINED, &heap);
+
+		/* This is a leaf page in a clustered index. */
+
+		/* Check if there are any externally stored columns.
+		For each externally stored column, restore the
+		BTR_EXTERN_FIELD_REF separately. */
+
+		if (UNIV_UNLIKELY(rec_offs_any_extern(offsets))) {
+			if (UNIV_UNLIKELY
+			    (!page_zip_decompress_clust_ext(
+				    d_stream, rec, offsets, trx_id_col))) {
+
+				goto zlib_error;
+			}
+		} else {
+			/* Skip trx_id and roll_ptr */
+			ulint	len;
+			byte*	dst = rec_get_nth_field(rec, offsets,
+							trx_id_col, &len);
+			if (UNIV_UNLIKELY(len < DATA_TRX_ID_LEN
+					  + DATA_ROLL_PTR_LEN)) {
+
+				page_zip_fail(("page_zip_decompress_clust:"
+					       " len = %lu\n", (ulong) len));
+				goto zlib_error;
+			}
+
+			d_stream->avail_out = dst - d_stream->next_out;
+
+			switch (inflate(d_stream, Z_SYNC_FLUSH)) {
+			case Z_STREAM_END:
+			case Z_OK:
+			case Z_BUF_ERROR:
+				if (!d_stream->avail_out) {
+					break;
+				}
+				/* fall through */
+			default:
+				page_zip_fail(("page_zip_decompress_clust:"
+					       " 2 inflate(Z_SYNC_FLUSH)=%s\n",
+					       d_stream->msg));
+				goto zlib_error;
+			}
+
+			ut_ad(d_stream->next_out == dst);
+
+			/* Clear DB_TRX_ID and DB_ROLL_PTR in order to
+			avoid uninitialized bytes in case the record
+			is affected by page_zip_apply_log(). */
+			memset(dst, 0, DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN);
+
+			d_stream->next_out += DATA_TRX_ID_LEN
+				+ DATA_ROLL_PTR_LEN;
+		}
+
+		/* Decompress the last bytes of the record. */
+		d_stream->avail_out = rec_get_end(rec, offsets)
+			- d_stream->next_out;
+
+		switch (inflate(d_stream, Z_SYNC_FLUSH)) {
+		case Z_STREAM_END:
+		case Z_OK:
+		case Z_BUF_ERROR:
+			if (!d_stream->avail_out) {
+				break;
+			}
+			/* fall through */
+		default:
+			page_zip_fail(("page_zip_decompress_clust:"
+				       " 3 inflate(Z_SYNC_FLUSH)=%s\n",
+				       d_stream->msg));
+			goto zlib_error;
+		}
+	}
+
+	/* Decompress any trailing garbage, in case the last record was
+	allocated from an originally longer space on the free list. */
+	d_stream->avail_out = page_header_get_field(page_zip->data,
+						    PAGE_HEAP_TOP)
+		- page_offset(d_stream->next_out);
+	if (UNIV_UNLIKELY(d_stream->avail_out > UNIV_PAGE_SIZE
+			  - PAGE_ZIP_START - PAGE_DIR)) {
+
+		page_zip_fail(("page_zip_decompress_clust:"
+			       " avail_out = %u\n",
+			       d_stream->avail_out));
+		goto zlib_error;
+	}
+
+	if (UNIV_UNLIKELY(inflate(d_stream, Z_FINISH) != Z_STREAM_END)) {
+		page_zip_fail(("page_zip_decompress_clust:"
+			       " inflate(Z_FINISH)=%s\n",
+			       d_stream->msg));
+zlib_error:
+		inflateEnd(d_stream);
+		return(FALSE);
+	}
+
+	/* Note that d_stream->avail_out > 0 may hold here
+	if the modification log is nonempty. */
+
+zlib_done:
+	if (UNIV_UNLIKELY(inflateEnd(d_stream) != Z_OK)) {
+		ut_error;
+	}
+
+	{
+		page_t*	page = page_align(d_stream->next_out);
+
+		/* Clear the unused heap space on the uncompressed page. */
+		memset(d_stream->next_out, 0,
+		       page_dir_get_nth_slot(page,
+					     page_dir_get_n_slots(page) - 1)
+		       - d_stream->next_out);
+	}
+
+#ifdef UNIV_DEBUG
+	page_zip->m_start = PAGE_DATA + d_stream->total_in;
+#endif /* UNIV_DEBUG */
+
+	/* Apply the modification log. */
+	{
+		const byte*	mod_log_ptr;
+		mod_log_ptr = page_zip_apply_log(d_stream->next_in,
+						 d_stream->avail_in + 1,
+						 recs, n_dense,
+						 trx_id_col, heap_status,
+						 index, offsets);
+
+		if (UNIV_UNLIKELY(!mod_log_ptr)) {
+			return(FALSE);
+		}
+		page_zip->m_end = mod_log_ptr - page_zip->data;
+		page_zip->m_nonempty = mod_log_ptr != d_stream->next_in;
+	}
+
+	if (UNIV_UNLIKELY(page_zip_get_trailer_len(page_zip, TRUE, NULL)
+			  + page_zip->m_end >= page_zip_get_size(page_zip))) {
+
+		page_zip_fail(("page_zip_decompress_clust: %lu + %lu >= %lu\n",
+			       (ulong) page_zip_get_trailer_len(
+				       page_zip, TRUE, NULL),
+			       (ulong) page_zip->m_end,
+			       (ulong) page_zip_get_size(page_zip)));
+		return(FALSE);
+	}
+
+	storage = page_zip->data + page_zip_get_size(page_zip)
+		- n_dense * PAGE_ZIP_DIR_SLOT_SIZE;
+
+	externs = storage - n_dense
+		* (DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN);
+
+	/* Restore the uncompressed columns in heap_no order. */
+
+	for (slot = 0; slot < n_dense; slot++) {
+		ulint	i;
+		ulint	len;
+		byte*	dst;
+		rec_t*	rec	= recs[slot];
+		ibool	exists	= !page_zip_dir_find_free(
+			page_zip, page_offset(rec));
+		offsets = rec_get_offsets(rec, index, offsets,
+					  ULINT_UNDEFINED, &heap);
+
+		dst = rec_get_nth_field(rec, offsets,
+					trx_id_col, &len);
+		ut_ad(len >= DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN);
+		storage -= DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN;
+		memcpy(dst, storage,
+		       DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN);
+
+		/* Check if there are any externally stored
+		columns in this record.  For each externally
+		stored column, restore or clear the
+		BTR_EXTERN_FIELD_REF. */
+		if (!rec_offs_any_extern(offsets)) {
+			continue;
+		}
+
+		for (i = 0; i < rec_offs_n_fields(offsets); i++) {
+			if (!rec_offs_nth_extern(offsets, i)) {
+				continue;
+			}
+			dst = rec_get_nth_field(rec, offsets, i, &len);
+
+			if (UNIV_UNLIKELY(len < BTR_EXTERN_FIELD_REF_SIZE)) {
+				page_zip_fail(("page_zip_decompress_clust:"
+					       " %lu < 20\n",
+					       (ulong) len));
+				return(FALSE);
+			}
+
+			dst += len - BTR_EXTERN_FIELD_REF_SIZE;
+
+			if (UNIV_LIKELY(exists)) {
+				/* Existing record:
+				restore the BLOB pointer */
+				externs -= BTR_EXTERN_FIELD_REF_SIZE;
+
+				if (UNIV_UNLIKELY
+				    (externs < page_zip->data
+				     + page_zip->m_end)) {
+					page_zip_fail(("page_zip_"
+						       "decompress_clust: "
+						       "%p < %p + %lu\n",
+						       (const void*) externs,
+						       (const void*)
+						       page_zip->data,
+						       (ulong)
+						       page_zip->m_end));
+					return(FALSE);
+				}
+
+				memcpy(dst, externs,
+				       BTR_EXTERN_FIELD_REF_SIZE);
+
+				page_zip->n_blobs++;
+			} else {
+				/* Deleted record:
+				clear the BLOB pointer */
+				memset(dst, 0,
+				       BTR_EXTERN_FIELD_REF_SIZE);
+			}
+		}
+	}
+
+	return(TRUE);
+}
+
+/**************************************************************************
+Decompress a page.  This function should tolerate errors on the compressed
+page.  Instead of letting assertions fail, it will return FALSE if an
+inconsistency is detected. */
+UNIV_INTERN
+ibool
+page_zip_decompress(
+/*================*/
+				/* out: TRUE on success, FALSE on failure */
+	page_zip_des_t*	page_zip,/* in: data, ssize;
+				out: m_start, m_end, m_nonempty, n_blobs */
+	page_t*		page)	/* out: uncompressed page, may be trashed */
+{
+	z_stream	d_stream;
+	dict_index_t*	index	= NULL;
+	rec_t**		recs;	/* dense page directory, sorted by address */
+	ulint		n_dense;/* number of user records on the page */
+	ulint		trx_id_col = ULINT_UNDEFINED;
+	mem_heap_t*	heap;
+	ulint*		offsets;
+	ullint		usec = ut_time_us(NULL);
+
+	ut_ad(page_zip_simple_validate(page_zip));
+	UNIV_MEM_ASSERT_W(page, UNIV_PAGE_SIZE);
+	UNIV_MEM_ASSERT_RW(page_zip->data, page_zip_get_size(page_zip));
+
+	/* The dense directory excludes the infimum and supremum records. */
+	n_dense = page_dir_get_n_heap(page_zip->data) - PAGE_HEAP_NO_USER_LOW;
+	if (UNIV_UNLIKELY(n_dense * PAGE_ZIP_DIR_SLOT_SIZE
+			  >= page_zip_get_size(page_zip))) {
+		page_zip_fail(("page_zip_decompress 1: %lu %lu\n",
+			       (ulong) n_dense,
+			       (ulong) page_zip_get_size(page_zip)));
+		return(FALSE);
+	}
+
+	heap = mem_heap_create(n_dense * (3 * sizeof *recs) + UNIV_PAGE_SIZE);
+	recs = mem_heap_alloc(heap, n_dense * (2 * sizeof *recs));
+
+#ifdef UNIV_ZIP_DEBUG
+	/* Clear the page. */
+	memset(page, 0x55, UNIV_PAGE_SIZE);
+#endif /* UNIV_ZIP_DEBUG */
+	UNIV_MEM_INVALID(page, UNIV_PAGE_SIZE);
+	/* Copy the page header. */
+	memcpy(page, page_zip->data, PAGE_DATA);
+
+	/* Copy the page directory. */
+	if (UNIV_UNLIKELY(!page_zip_dir_decode(page_zip, page, recs,
+					       recs + n_dense, n_dense))) {
+zlib_error:
+		mem_heap_free(heap);
+		return(FALSE);
+	}
+
+	/* Copy the infimum and supremum records. */
+	memcpy(page + (PAGE_NEW_INFIMUM - REC_N_NEW_EXTRA_BYTES),
+	       infimum_extra, sizeof infimum_extra);
+	if (UNIV_UNLIKELY(!page_get_n_recs(page))) {
+		rec_set_next_offs_new(page + PAGE_NEW_INFIMUM,
+				      PAGE_NEW_SUPREMUM);
+	} else {
+		rec_set_next_offs_new(page + PAGE_NEW_INFIMUM,
+				      page_zip_dir_get(page_zip, 0)
+				      & PAGE_ZIP_DIR_SLOT_MASK);
+	}
+	memcpy(page + PAGE_NEW_INFIMUM, infimum_data, sizeof infimum_data);
+	memcpy(page + (PAGE_NEW_SUPREMUM - REC_N_NEW_EXTRA_BYTES + 1),
+	       supremum_extra_data, sizeof supremum_extra_data);
+
+	page_zip_set_alloc(&d_stream, heap);
+
+	if (UNIV_UNLIKELY(inflateInit2(&d_stream, UNIV_PAGE_SIZE_SHIFT)
+			  != Z_OK)) {
+		ut_error;
+	}
+
+	d_stream.next_in = page_zip->data + PAGE_DATA;
+	/* Subtract the space reserved for
+	the page header and the end marker of the modification log. */
+	d_stream.avail_in = page_zip_get_size(page_zip) - (PAGE_DATA + 1);
+
+	d_stream.next_out = page + PAGE_ZIP_START;
+	d_stream.avail_out = UNIV_PAGE_SIZE - PAGE_ZIP_START;
+
+	/* Decode the zlib header and the index information. */
+	if (UNIV_UNLIKELY(inflate(&d_stream, Z_BLOCK) != Z_OK)) {
+
+		page_zip_fail(("page_zip_decompress:"
+			       " 1 inflate(Z_BLOCK)=%s\n", d_stream.msg));
+		goto zlib_error;
+	}
+
+	if (UNIV_UNLIKELY(inflate(&d_stream, Z_BLOCK) != Z_OK)) {
+
+		page_zip_fail(("page_zip_decompress:"
+			       " 2 inflate(Z_BLOCK)=%s\n", d_stream.msg));
+		goto zlib_error;
+	}
+
+	index = page_zip_fields_decode(
+		page + PAGE_ZIP_START, d_stream.next_out,
+		page_is_leaf(page) ? &trx_id_col : NULL);
+
+	if (UNIV_UNLIKELY(!index)) {
+
+		goto zlib_error;
+	}
+
+	/* Decompress the user records. */
+	page_zip->n_blobs = 0;
+	d_stream.next_out = page + PAGE_ZIP_START;
+
+	{
+		/* Pre-allocate the offsets for rec_get_offsets_reverse(). */
+		ulint	n = 1 + 1/* node ptr */ + REC_OFFS_HEADER_SIZE
+			+ dict_index_get_n_fields(index);
+		offsets = mem_heap_alloc(heap, n * sizeof(ulint));
+		*offsets = n;
+	}
+
+	/* Decompress the records in heap_no order. */
+	if (!page_is_leaf(page)) {
+		/* This is a node pointer page. */
+		ulint	info_bits;
+
+		if (UNIV_UNLIKELY
+		    (!page_zip_decompress_node_ptrs(page_zip, &d_stream,
+						    recs, n_dense, index,
+						    offsets, heap))) {
+			goto err_exit;
+		}
+
+		info_bits = mach_read_from_4(page + FIL_PAGE_PREV) == FIL_NULL
+			? REC_INFO_MIN_REC_FLAG : 0;
+
+		if (UNIV_UNLIKELY(!page_zip_set_extra_bytes(page_zip, page,
+							    info_bits))) {
+			goto err_exit;
+		}
+	} else if (UNIV_LIKELY(trx_id_col == ULINT_UNDEFINED)) {
+		/* This is a leaf page in a secondary index. */
+		if (UNIV_UNLIKELY(!page_zip_decompress_sec(page_zip, &d_stream,
+							   recs, n_dense,
+							   index, offsets))) {
+			goto err_exit;
+		}
+
+		if (UNIV_UNLIKELY(!page_zip_set_extra_bytes(page_zip,
+							    page, 0))) {
+err_exit:
+			page_zip_fields_free(index);
+			mem_heap_free(heap);
+			return(FALSE);
+		}
+	} else {
+		/* This is a leaf page in a clustered index. */
+		if (UNIV_UNLIKELY(!page_zip_decompress_clust(page_zip,
+							     &d_stream, recs,
+							     n_dense, index,
+							     trx_id_col,
+							     offsets, heap))) {
+			goto err_exit;
+		}
+
+		if (UNIV_UNLIKELY(!page_zip_set_extra_bytes(page_zip,
+							    page, 0))) {
+			goto err_exit;
+		}
+	}
+
+	ut_a(page_is_comp(page));
+	UNIV_MEM_ASSERT_RW(page, UNIV_PAGE_SIZE);
+
+	page_zip_fields_free(index);
+	mem_heap_free(heap);
+	{
+		page_zip_stat_t*	zip_stat
+			= &page_zip_stat[page_zip->ssize - 1];
+		zip_stat->decompressed++;
+		zip_stat->decompressed_usec += ut_time_us(NULL) - usec;
+	}
+
+	/* Update the stat counter for LRU policy. */
+	buf_LRU_stat_inc_unzip();
+
+	return(TRUE);
+}
+
+#ifdef UNIV_ZIP_DEBUG
+/**************************************************************************
+Dump a block of memory on the standard error stream. */
+static
+void
+page_zip_hexdump_func(
+/*==================*/
+	const char*	name,	/* in: name of the data structure */
+	const void*	buf,	/* in: data */
+	ulint		size)	/* in: length of the data, in bytes */
+{
+	const byte*	s	= buf;
+	ulint		addr;
+	const ulint	width	= 32; /* bytes per line */
+
+	fprintf(stderr, "%s:\n", name);
+
+	for (addr = 0; addr < size; addr += width) {
+		ulint	i;
+
+		fprintf(stderr, "%04lx ", (ulong) addr);
+
+		i = ut_min(width, size - addr);
+
+		while (i--) {
+			fprintf(stderr, "%02x", *s++);
+		}
+
+		putc('\n', stderr);
+	}
+}
+
+#define page_zip_hexdump(buf, size) page_zip_hexdump_func(#buf, buf, size)
+
+/* Flag: make page_zip_validate() compare page headers only */
+UNIV_INTERN ibool	page_zip_validate_header_only = FALSE;
+
+/**************************************************************************
+Check that the compressed and decompressed pages match. */
+UNIV_INTERN
+ibool
+page_zip_validate_low(
+/*==================*/
+					/* out: TRUE if valid, FALSE if not */
+	const page_zip_des_t*	page_zip,/* in: compressed page */
+	const page_t*		page,	/* in: uncompressed page */
+	ibool			sloppy)	/* in: FALSE=strict,
+					TRUE=ignore the MIN_REC_FLAG */
+{
+	page_zip_des_t	temp_page_zip;
+	byte*		temp_page_buf;
+	page_t*		temp_page;
+	ibool		valid;
+
+	if (memcmp(page_zip->data + FIL_PAGE_PREV, page + FIL_PAGE_PREV,
+		   FIL_PAGE_LSN - FIL_PAGE_PREV)
+	    || memcmp(page_zip->data + FIL_PAGE_TYPE, page + FIL_PAGE_TYPE, 2)
+	    || memcmp(page_zip->data + FIL_PAGE_DATA, page + FIL_PAGE_DATA,
+		      PAGE_DATA - FIL_PAGE_DATA)) {
+		page_zip_fail(("page_zip_validate: page header\n"));
+		page_zip_hexdump(page_zip, sizeof *page_zip);
+		page_zip_hexdump(page_zip->data, page_zip_get_size(page_zip));
+		page_zip_hexdump(page, UNIV_PAGE_SIZE);
+		return(FALSE);
+	}
+
+	ut_a(page_is_comp(page));
+
+	if (page_zip_validate_header_only) {
+		return(TRUE);
+	}
+
+	/* page_zip_decompress() expects the uncompressed page to be
+	UNIV_PAGE_SIZE aligned. */
+	temp_page_buf = ut_malloc(2 * UNIV_PAGE_SIZE);
+	temp_page = ut_align(temp_page_buf, UNIV_PAGE_SIZE);
+
+#ifdef UNIV_DEBUG_VALGRIND
+	/* Get detailed information on the valid bits in case the
+	UNIV_MEM_ASSERT_RW() checks fail.  The v-bits of page[],
+	page_zip->data[] or page_zip could be viewed at temp_page[] or
+	temp_page_zip in a debugger when running valgrind --db-attach. */
+	VALGRIND_GET_VBITS(page, temp_page, UNIV_PAGE_SIZE);
+	UNIV_MEM_ASSERT_RW(page, UNIV_PAGE_SIZE);
+	VALGRIND_GET_VBITS(page_zip, &temp_page_zip, sizeof temp_page_zip);
+	UNIV_MEM_ASSERT_RW(page_zip, sizeof *page_zip);
+	VALGRIND_GET_VBITS(page_zip->data, temp_page,
+			   page_zip_get_size(page_zip));
+	UNIV_MEM_ASSERT_RW(page_zip->data, page_zip_get_size(page_zip));
+#endif /* UNIV_DEBUG_VALGRIND */
+
+	temp_page_zip = *page_zip;
+	valid = page_zip_decompress(&temp_page_zip, temp_page);
+	if (!valid) {
+		fputs("page_zip_validate(): failed to decompress\n", stderr);
+		goto func_exit;
+	}
+	if (page_zip->n_blobs != temp_page_zip.n_blobs) {
+		page_zip_fail(("page_zip_validate: n_blobs: %u!=%u\n",
+			       page_zip->n_blobs, temp_page_zip.n_blobs));
+		valid = FALSE;
+	}
+#ifdef UNIV_DEBUG
+	if (page_zip->m_start != temp_page_zip.m_start) {
+		page_zip_fail(("page_zip_validate: m_start: %u!=%u\n",
+			       page_zip->m_start, temp_page_zip.m_start));
+		valid = FALSE;
+	}
+#endif /* UNIV_DEBUG */
+	if (page_zip->m_end != temp_page_zip.m_end) {
+		page_zip_fail(("page_zip_validate: m_end: %u!=%u\n",
+			       page_zip->m_end, temp_page_zip.m_end));
+		valid = FALSE;
+	}
+	if (page_zip->m_nonempty != temp_page_zip.m_nonempty) {
+		page_zip_fail(("page_zip_validate(): m_nonempty: %u!=%u\n",
+			       page_zip->m_nonempty,
+			       temp_page_zip.m_nonempty));
+		valid = FALSE;
+	}
+	if (memcmp(page + PAGE_HEADER, temp_page + PAGE_HEADER,
+		   UNIV_PAGE_SIZE - PAGE_HEADER - FIL_PAGE_DATA_END)) {
+
+		/* In crash recovery, the "minimum record" flag may be
+		set incorrectly until the mini-transaction is
+		committed.  Let us tolerate that difference when we
+		are performing a sloppy validation. */
+
+		if (sloppy) {
+			byte	info_bits_diff;
+			ulint	offset
+				= rec_get_next_offs(page + PAGE_NEW_INFIMUM,
+						    TRUE);
+			ut_a(offset >= PAGE_NEW_SUPREMUM);
+			offset -= 5 /* REC_NEW_INFO_BITS */;
+
+			info_bits_diff = page[offset] ^ temp_page[offset];
+
+			if (info_bits_diff == REC_INFO_MIN_REC_FLAG) {
+				temp_page[offset] = page[offset];
+
+				if (!memcmp(page + PAGE_HEADER,
+					    temp_page + PAGE_HEADER,
+					    UNIV_PAGE_SIZE - PAGE_HEADER
+					    - FIL_PAGE_DATA_END)) {
+
+					/* Only the minimum record flag
+					differed.  Let us ignore it. */
+					page_zip_fail(("page_zip_validate: "
+						       "min_rec_flag "
+						       "(ignored, "
+						       "%lu,%lu,0x%02lx)\n",
+						       page_get_space_id(page),
+						       page_get_page_no(page),
+						       (ulong) page[offset]));
+					goto func_exit;
+				}
+			}
+		}
+		page_zip_fail(("page_zip_validate: content\n"));
+		valid = FALSE;
+	}
+
+func_exit:
+	if (!valid) {
+		page_zip_hexdump(page_zip, sizeof *page_zip);
+		page_zip_hexdump(page_zip->data, page_zip_get_size(page_zip));
+		page_zip_hexdump(page, UNIV_PAGE_SIZE);
+		page_zip_hexdump(temp_page, UNIV_PAGE_SIZE);
+	}
+	ut_free(temp_page_buf);
+	return(valid);
+}
+
+/**************************************************************************
+Check that the compressed and decompressed pages match. */
+UNIV_INTERN
+ibool
+page_zip_validate(
+/*==============*/
+					/* out: TRUE if valid, FALSE if not */
+	const page_zip_des_t*	page_zip,/* in: compressed page */
+	const page_t*		page)	/* in: uncompressed page */
+{
+	return(page_zip_validate_low(page_zip, page,
+				     recv_recovery_is_on()));
+}
+#endif /* UNIV_ZIP_DEBUG */
+
+#ifdef UNIV_DEBUG
+static
+ibool
+page_zip_header_cmp(
+/*================*/
+					/* out: TRUE */
+	const page_zip_des_t*	page_zip,/* in: compressed page */
+	const byte*		page)	/* in: uncompressed page */
+{
+	ut_ad(!memcmp(page_zip->data + FIL_PAGE_PREV, page + FIL_PAGE_PREV,
+		      FIL_PAGE_LSN - FIL_PAGE_PREV));
+	ut_ad(!memcmp(page_zip->data + FIL_PAGE_TYPE, page + FIL_PAGE_TYPE,
+		      2));
+	ut_ad(!memcmp(page_zip->data + FIL_PAGE_DATA, page + FIL_PAGE_DATA,
+		      PAGE_DATA - FIL_PAGE_DATA));
+
+	return(TRUE);
+}
+#endif /* UNIV_DEBUG */
+
+/**************************************************************************
+Write a record on the compressed page that contains externally stored
+columns.  The data must already have been written to the uncompressed page. */
+static
+byte*
+page_zip_write_rec_ext(
+/*===================*/
+					/* out: end of modification log */
+	page_zip_des_t*	page_zip,	/* in/out: compressed page */
+	const page_t*	page,		/* in: page containing rec */
+	const byte*	rec,		/* in: record being written */
+	dict_index_t*	index,		/* in: record descriptor */
+	const ulint*	offsets,	/* in: rec_get_offsets(rec, index) */
+	ulint		create,		/* in: nonzero=insert, zero=update */
+	ulint		trx_id_col,	/* in: position of DB_TRX_ID */
+	ulint		heap_no,	/* in: heap number of rec */
+	byte*		storage,	/* in: end of dense page directory */
+	byte*		data)		/* in: end of modification log */
+{
+	const byte*	start	= rec;
+	ulint		i;
+	ulint		len;
+	byte*		externs	= storage;
+	ulint		n_ext	= rec_offs_n_extern(offsets);
+
+	ut_ad(rec_offs_validate(rec, index, offsets));
+	UNIV_MEM_ASSERT_RW(rec, rec_offs_data_size(offsets));
+	UNIV_MEM_ASSERT_RW(rec - rec_offs_extra_size(offsets),
+			   rec_offs_extra_size(offsets));
+
+	externs -= (DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN)
+		* (page_dir_get_n_heap(page) - PAGE_HEAP_NO_USER_LOW);
+
+	/* Note that this will not take into account
+	the BLOB columns of rec if create==TRUE. */
+	ut_ad(data + rec_offs_data_size(offsets)
+	      - (DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN)
+	      - n_ext * BTR_EXTERN_FIELD_REF_SIZE
+	      < externs - BTR_EXTERN_FIELD_REF_SIZE * page_zip->n_blobs);
+
+	{
+		ulint	blob_no = page_zip_get_n_prev_extern(
+			page_zip, rec, index);
+		byte*	ext_end = externs - page_zip->n_blobs
+			* BTR_EXTERN_FIELD_REF_SIZE;
+		ut_ad(blob_no <= page_zip->n_blobs);
+		externs -= blob_no * BTR_EXTERN_FIELD_REF_SIZE;
+
+		if (create) {
+			page_zip->n_blobs += n_ext;
+			ASSERT_ZERO_BLOB(ext_end - n_ext
+					 * BTR_EXTERN_FIELD_REF_SIZE);
+			memmove(ext_end - n_ext
+				* BTR_EXTERN_FIELD_REF_SIZE,
+				ext_end,
+				externs - ext_end);
+		}
+
+		ut_a(blob_no + n_ext <= page_zip->n_blobs);
+	}
+
+	for (i = 0; i < rec_offs_n_fields(offsets); i++) {
+		const byte*	src;
+
+		if (UNIV_UNLIKELY(i == trx_id_col)) {
+			ut_ad(!rec_offs_nth_extern(offsets,
+						   i));
+			ut_ad(!rec_offs_nth_extern(offsets,
+						   i + 1));
+			/* Locate trx_id and roll_ptr. */
+			src = rec_get_nth_field(rec, offsets,
+						i, &len);
+			ut_ad(len == DATA_TRX_ID_LEN);
+			ut_ad(src + DATA_TRX_ID_LEN
+			      == rec_get_nth_field(
+				      rec, offsets,
+				      i + 1, &len));
+			ut_ad(len == DATA_ROLL_PTR_LEN);
+
+			/* Log the preceding fields. */
+			ASSERT_ZERO(data, src - start);
+			memcpy(data, start, src - start);
+			data += src - start;
+			start = src + (DATA_TRX_ID_LEN
+				       + DATA_ROLL_PTR_LEN);
+
+			/* Store trx_id and roll_ptr. */
+			memcpy(storage - (DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN)
+			       * (heap_no - 1),
+			       src, DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN);
+			i++; /* skip also roll_ptr */
+		} else if (rec_offs_nth_extern(offsets, i)) {
+			src = rec_get_nth_field(rec, offsets,
+						i, &len);
+
+			ut_ad(dict_index_is_clust(index));
+			ut_ad(len
+			      >= BTR_EXTERN_FIELD_REF_SIZE);
+			src += len - BTR_EXTERN_FIELD_REF_SIZE;
+
+			ASSERT_ZERO(data, src - start);
+			memcpy(data, start, src - start);
+			data += src - start;
+			start = src + BTR_EXTERN_FIELD_REF_SIZE;
+
+			/* Store the BLOB pointer. */
+			externs -= BTR_EXTERN_FIELD_REF_SIZE;
+			ut_ad(data < externs);
+			memcpy(externs, src, BTR_EXTERN_FIELD_REF_SIZE);
+		}
+	}
+
+	/* Log the last bytes of the record. */
+	len = rec_offs_data_size(offsets) - (start - rec);
+
+	ASSERT_ZERO(data, len);
+	memcpy(data, start, len);
+	data += len;
+
+	return(data);
+}
+
+/**************************************************************************
+Write an entire record on the compressed page.  The data must already
+have been written to the uncompressed page. */
+UNIV_INTERN
+void
+page_zip_write_rec(
+/*===============*/
+	page_zip_des_t*	page_zip,/* in/out: compressed page */
+	const byte*	rec,	/* in: record being written */
+	dict_index_t*	index,	/* in: the index the record belongs to */
+	const ulint*	offsets,/* in: rec_get_offsets(rec, index) */
+	ulint		create)	/* in: nonzero=insert, zero=update */
+{
+	const page_t*	page;
+	byte*		data;
+	byte*		storage;
+	ulint		heap_no;
+	byte*		slot;
+
+	ut_ad(buf_frame_get_page_zip(rec) == page_zip);
+	ut_ad(page_zip_simple_validate(page_zip));
+	ut_ad(page_zip_get_size(page_zip)
+	      > PAGE_DATA + page_zip_dir_size(page_zip));
+	ut_ad(rec_offs_comp(offsets));
+	ut_ad(rec_offs_validate(rec, index, offsets));
+
+	ut_ad(page_zip->m_start >= PAGE_DATA);
+
+	page = page_align(rec);
+
+	ut_ad(page_zip_header_cmp(page_zip, page));
+	ut_ad(page_simple_validate_new((page_t*) page));
+
+	UNIV_MEM_ASSERT_RW(page_zip->data, page_zip_get_size(page_zip));
+	UNIV_MEM_ASSERT_RW(rec, rec_offs_data_size(offsets));
+	UNIV_MEM_ASSERT_RW(rec - rec_offs_extra_size(offsets),
+			   rec_offs_extra_size(offsets));
+
+	slot = page_zip_dir_find(page_zip, page_offset(rec));
+	ut_a(slot);
+	/* Copy the delete mark. */
+	if (rec_get_deleted_flag(rec, TRUE)) {
+		*slot |= PAGE_ZIP_DIR_SLOT_DEL >> 8;
+	} else {
+		*slot &= ~(PAGE_ZIP_DIR_SLOT_DEL >> 8);
+	}
+
+	ut_ad(rec_get_start((rec_t*) rec, offsets) >= page + PAGE_ZIP_START);
+	ut_ad(rec_get_end((rec_t*) rec, offsets) <= page + UNIV_PAGE_SIZE
+	      - PAGE_DIR - PAGE_DIR_SLOT_SIZE
+	      * page_dir_get_n_slots(page));
+
+	heap_no = rec_get_heap_no_new(rec);
+	ut_ad(heap_no >= PAGE_HEAP_NO_USER_LOW); /* not infimum or supremum */
+	ut_ad(heap_no < page_dir_get_n_heap(page));
+
+	/* Append to the modification log. */
+	data = page_zip->data + page_zip->m_end;
+	ut_ad(!*data);
+
+	/* Identify the record by writing its heap number - 1.
+	0 is reserved to indicate the end of the modification log. */
+
+	if (UNIV_UNLIKELY(heap_no - 1 >= 64)) {
+		*data++ = (byte) (0x80 | (heap_no - 1) >> 7);
+		ut_ad(!*data);
+	}
+	*data++ = (byte) ((heap_no - 1) << 1);
+	ut_ad(!*data);
+
+	{
+		const byte*	start	= rec - rec_offs_extra_size(offsets);
+		const byte*	b	= rec - REC_N_NEW_EXTRA_BYTES;
+
+		/* Write the extra bytes backwards, so that
+		rec_offs_extra_size() can be easily computed in
+		page_zip_apply_log() by invoking
+		rec_get_offsets_reverse(). */
+
+		while (b != start) {
+			*data++ = *--b;
+			ut_ad(!*data);
+		}
+	}
+
+	/* Write the data bytes.  Store the uncompressed bytes separately. */
+	storage = page_zip->data + page_zip_get_size(page_zip)
+		- (page_dir_get_n_heap(page) - PAGE_HEAP_NO_USER_LOW)
+		* PAGE_ZIP_DIR_SLOT_SIZE;
+
+	if (page_is_leaf(page)) {
+		ulint		len;
+
+		if (dict_index_is_clust(index)) {
+			ulint		trx_id_col;
+
+			trx_id_col = dict_index_get_sys_col_pos(index,
+								DATA_TRX_ID);
+			ut_ad(trx_id_col != ULINT_UNDEFINED);
+
+			/* Store separately trx_id, roll_ptr and
+			the BTR_EXTERN_FIELD_REF of each BLOB column. */
+			if (rec_offs_any_extern(offsets)) {
+				data = page_zip_write_rec_ext(
+					page_zip, page,
+					rec, index, offsets, create,
+					trx_id_col, heap_no, storage, data);
+			} else {
+				/* Locate trx_id and roll_ptr. */
+				const byte*	src
+					= rec_get_nth_field(rec, offsets,
+							    trx_id_col, &len);
+				ut_ad(len == DATA_TRX_ID_LEN);
+				ut_ad(src + DATA_TRX_ID_LEN
+				      == rec_get_nth_field(
+					      rec, offsets,
+					      trx_id_col + 1, &len));
+				ut_ad(len == DATA_ROLL_PTR_LEN);
+
+				/* Log the preceding fields. */
+				ASSERT_ZERO(data, src - rec);
+				memcpy(data, rec, src - rec);
+				data += src - rec;
+
+				/* Store trx_id and roll_ptr. */
+				memcpy(storage
+				       - (DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN)
+				       * (heap_no - 1),
+				       src,
+				       DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN);
+
+				src += DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN;
+
+				/* Log the last bytes of the record. */
+				len = rec_offs_data_size(offsets)
+					- (src - rec);
+
+				ASSERT_ZERO(data, len);
+				memcpy(data, src, len);
+				data += len;
+			}
+		} else {
+			/* Leaf page of a secondary index:
+			no externally stored columns */
+			ut_ad(dict_index_get_sys_col_pos(index, DATA_TRX_ID)
+			      == ULINT_UNDEFINED);
+			ut_ad(!rec_offs_any_extern(offsets));
+
+			/* Log the entire record. */
+			len = rec_offs_data_size(offsets);
+
+			ASSERT_ZERO(data, len);
+			memcpy(data, rec, len);
+			data += len;
+		}
+	} else {
+		/* This is a node pointer page. */
+		ulint	len;
+
+		/* Non-leaf nodes should not have any externally
+		stored columns. */
+		ut_ad(!rec_offs_any_extern(offsets));
+
+		/* Copy the data bytes, except node_ptr. */
+		len = rec_offs_data_size(offsets) - REC_NODE_PTR_SIZE;
+		ut_ad(data + len < storage - REC_NODE_PTR_SIZE
+		      * (page_dir_get_n_heap(page) - PAGE_HEAP_NO_USER_LOW));
+		ASSERT_ZERO(data, len);
+		memcpy(data, rec, len);
+		data += len;
+
+		/* Copy the node pointer to the uncompressed area. */
+		memcpy(storage - REC_NODE_PTR_SIZE
+		       * (heap_no - 1),
+		       rec + len,
+		       REC_NODE_PTR_SIZE);
+	}
+
+	ut_a(!*data);
+	ut_ad((ulint) (data - page_zip->data) < page_zip_get_size(page_zip));
+	page_zip->m_end = data - page_zip->data;
+	page_zip->m_nonempty = TRUE;
+
+#ifdef UNIV_ZIP_DEBUG
+	ut_a(page_zip_validate(page_zip, page_align(rec)));
+#endif /* UNIV_ZIP_DEBUG */
+}
+
+/***************************************************************
+Parses a log record of writing a BLOB pointer of a record. */
+UNIV_INTERN
+byte*
+page_zip_parse_write_blob_ptr(
+/*==========================*/
+				/* out: end of log record or NULL */
+	byte*		ptr,	/* in: redo log buffer */
+	byte*		end_ptr,/* in: redo log buffer end */
+	page_t*		page,	/* in/out: uncompressed page */
+	page_zip_des_t*	page_zip)/* in/out: compressed page */
+{
+	ulint	offset;
+	ulint	z_offset;
+
+	ut_ad(!page == !page_zip);
+
+	if (UNIV_UNLIKELY
+	    (end_ptr < ptr + (2 + 2 + BTR_EXTERN_FIELD_REF_SIZE))) {
+
+		return(NULL);
+	}
+
+	offset = mach_read_from_2(ptr);
+	z_offset = mach_read_from_2(ptr + 2);
+
+	if (UNIV_UNLIKELY(offset < PAGE_ZIP_START)
+	    || UNIV_UNLIKELY(offset >= UNIV_PAGE_SIZE)
+	    || UNIV_UNLIKELY(z_offset >= UNIV_PAGE_SIZE)) {
+corrupt:
+		recv_sys->found_corrupt_log = TRUE;
+
+		return(NULL);
+	}
+
+	if (page) {
+		if (UNIV_UNLIKELY(!page_zip)
+		    || UNIV_UNLIKELY(!page_is_leaf(page))) {
+
+			goto corrupt;
+		}
+
+#ifdef UNIV_ZIP_DEBUG
+		ut_a(page_zip_validate(page_zip, page));
+#endif /* UNIV_ZIP_DEBUG */
+
+		memcpy(page + offset,
+		       ptr + 4, BTR_EXTERN_FIELD_REF_SIZE);
+		memcpy(page_zip->data + z_offset,
+		       ptr + 4, BTR_EXTERN_FIELD_REF_SIZE);
+
+#ifdef UNIV_ZIP_DEBUG
+		ut_a(page_zip_validate(page_zip, page));
+#endif /* UNIV_ZIP_DEBUG */
+	}
+
+	return(ptr + (2 + 2 + BTR_EXTERN_FIELD_REF_SIZE));
+}
+
+/**************************************************************************
+Write a BLOB pointer of a record on the leaf page of a clustered index.
+The information must already have been updated on the uncompressed page. */
+UNIV_INTERN
+void
+page_zip_write_blob_ptr(
+/*====================*/
+	page_zip_des_t*	page_zip,/* in/out: compressed page */
+	const byte*	rec,	/* in/out: record whose data is being
+				written */
+	dict_index_t*	index,	/* in: index of the page */
+	const ulint*	offsets,/* in: rec_get_offsets(rec, index) */
+	ulint		n,	/* in: column index */
+	mtr_t*		mtr)	/* in: mini-transaction handle,
+				or NULL if no logging is needed */
+{
+	const byte*	field;
+	byte*		externs;
+	const page_t*	page	= page_align(rec);
+	ulint		blob_no;
+	ulint		len;
+
+	ut_ad(buf_frame_get_page_zip(rec) == page_zip);
+	ut_ad(page_simple_validate_new((page_t*) page));
+	ut_ad(page_zip_simple_validate(page_zip));
+	ut_ad(page_zip_get_size(page_zip)
+	      > PAGE_DATA + page_zip_dir_size(page_zip));
+	ut_ad(rec_offs_comp(offsets));
+	ut_ad(rec_offs_validate(rec, NULL, offsets));
+	ut_ad(rec_offs_any_extern(offsets));
+	ut_ad(rec_offs_nth_extern(offsets, n));
+
+	ut_ad(page_zip->m_start >= PAGE_DATA);
+	ut_ad(page_zip_header_cmp(page_zip, page));
+
+	ut_ad(page_is_leaf(page));
+	ut_ad(dict_index_is_clust(index));
+
+	UNIV_MEM_ASSERT_RW(page_zip->data, page_zip_get_size(page_zip));
+	UNIV_MEM_ASSERT_RW(rec, rec_offs_data_size(offsets));
+	UNIV_MEM_ASSERT_RW(rec - rec_offs_extra_size(offsets),
+			   rec_offs_extra_size(offsets));
+
+	blob_no = page_zip_get_n_prev_extern(page_zip, rec, index)
+		+ rec_get_n_extern_new(rec, index, n);
+	ut_a(blob_no < page_zip->n_blobs);
+
+	externs = page_zip->data + page_zip_get_size(page_zip)
+		- (page_dir_get_n_heap(page) - PAGE_HEAP_NO_USER_LOW)
+		* (PAGE_ZIP_DIR_SLOT_SIZE
+		   + DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN);
+
+	field = rec_get_nth_field(rec, offsets, n, &len);
+
+	externs -= (blob_no + 1) * BTR_EXTERN_FIELD_REF_SIZE;
+	field += len - BTR_EXTERN_FIELD_REF_SIZE;
+
+	memcpy(externs, field, BTR_EXTERN_FIELD_REF_SIZE);
+
+#ifdef UNIV_ZIP_DEBUG
+	ut_a(page_zip_validate(page_zip, page));
+#endif /* UNIV_ZIP_DEBUG */
+
+	if (mtr) {
+		byte*	log_ptr	= mlog_open(
+			mtr, 11 + 2 + 2 + BTR_EXTERN_FIELD_REF_SIZE);
+		if (UNIV_UNLIKELY(!log_ptr)) {
+			return;
+		}
+
+		log_ptr = mlog_write_initial_log_record_fast(
+			(byte*) field, MLOG_ZIP_WRITE_BLOB_PTR, log_ptr, mtr);
+		mach_write_to_2(log_ptr, page_offset(field));
+		log_ptr += 2;
+		mach_write_to_2(log_ptr, externs - page_zip->data);
+		log_ptr += 2;
+		memcpy(log_ptr, externs, BTR_EXTERN_FIELD_REF_SIZE);
+		log_ptr += BTR_EXTERN_FIELD_REF_SIZE;
+		mlog_close(mtr, log_ptr);
+	}
+}
+
+/***************************************************************
+Parses a log record of writing the node pointer of a record. */
+UNIV_INTERN
+byte*
+page_zip_parse_write_node_ptr(
+/*==========================*/
+				/* out: end of log record or NULL */
+	byte*		ptr,	/* in: redo log buffer */
+	byte*		end_ptr,/* in: redo log buffer end */
+	page_t*		page,	/* in/out: uncompressed page */
+	page_zip_des_t*	page_zip)/* in/out: compressed page */
+{
+	ulint	offset;
+	ulint	z_offset;
+
+	ut_ad(!page == !page_zip);
+
+	if (UNIV_UNLIKELY(end_ptr < ptr + (2 + 2 + REC_NODE_PTR_SIZE))) {
+
+		return(NULL);
+	}
+
+	offset = mach_read_from_2(ptr);
+	z_offset = mach_read_from_2(ptr + 2);
+
+	if (UNIV_UNLIKELY(offset < PAGE_ZIP_START)
+	    || UNIV_UNLIKELY(offset >= UNIV_PAGE_SIZE)
+	    || UNIV_UNLIKELY(z_offset >= UNIV_PAGE_SIZE)) {
+corrupt:
+		recv_sys->found_corrupt_log = TRUE;
+
+		return(NULL);
+	}
+
+	if (page) {
+		byte*	storage_end;
+		byte*	field;
+		byte*	storage;
+		ulint	heap_no;
+
+		if (UNIV_UNLIKELY(!page_zip)
+		    || UNIV_UNLIKELY(page_is_leaf(page))) {
+
+			goto corrupt;
+		}
+
+#ifdef UNIV_ZIP_DEBUG
+		ut_a(page_zip_validate(page_zip, page));
+#endif /* UNIV_ZIP_DEBUG */
+
+		field = page + offset;
+		storage = page_zip->data + z_offset;
+
+		storage_end = page_zip->data + page_zip_get_size(page_zip)
+			- (page_dir_get_n_heap(page) - PAGE_HEAP_NO_USER_LOW)
+			* PAGE_ZIP_DIR_SLOT_SIZE;
+
+		heap_no = 1 + (storage_end - storage) / REC_NODE_PTR_SIZE;
+
+		if (UNIV_UNLIKELY((storage_end - storage) % REC_NODE_PTR_SIZE)
+		    || UNIV_UNLIKELY(heap_no < PAGE_HEAP_NO_USER_LOW)
+		    || UNIV_UNLIKELY(heap_no >= page_dir_get_n_heap(page))) {
+
+			goto corrupt;
+		}
+
+		memcpy(field, ptr + 4, REC_NODE_PTR_SIZE);
+		memcpy(storage, ptr + 4, REC_NODE_PTR_SIZE);
+
+#ifdef UNIV_ZIP_DEBUG
+		ut_a(page_zip_validate(page_zip, page));
+#endif /* UNIV_ZIP_DEBUG */
+	}
+
+	return(ptr + (2 + 2 + REC_NODE_PTR_SIZE));
+}
+
+/**************************************************************************
+Write the node pointer of a record on a non-leaf compressed page. */
+UNIV_INTERN
+void
+page_zip_write_node_ptr(
+/*====================*/
+	page_zip_des_t*	page_zip,/* in/out: compressed page */
+	byte*		rec,	/* in/out: record */
+	ulint		size,	/* in: data size of rec */
+	ulint		ptr,	/* in: node pointer */
+	mtr_t*		mtr)	/* in: mini-transaction, or NULL */
+{
+	byte*	field;
+	byte*	storage;
+	page_t*	page	= page_align(rec);
+
+	ut_ad(buf_frame_get_page_zip(rec) == page_zip);
+	ut_ad(page_simple_validate_new(page));
+	ut_ad(page_zip_simple_validate(page_zip));
+	ut_ad(page_zip_get_size(page_zip)
+	      > PAGE_DATA + page_zip_dir_size(page_zip));
+	ut_ad(page_rec_is_comp(rec));
+
+	ut_ad(page_zip->m_start >= PAGE_DATA);
+	ut_ad(page_zip_header_cmp(page_zip, page));
+
+	ut_ad(!page_is_leaf(page));
+
+	UNIV_MEM_ASSERT_RW(page_zip->data, page_zip_get_size(page_zip));
+	UNIV_MEM_ASSERT_RW(rec, size);
+
+	storage = page_zip->data + page_zip_get_size(page_zip)
+		- (page_dir_get_n_heap(page) - PAGE_HEAP_NO_USER_LOW)
+		* PAGE_ZIP_DIR_SLOT_SIZE
+		- (rec_get_heap_no_new(rec) - 1) * REC_NODE_PTR_SIZE;
+	field = rec + size - REC_NODE_PTR_SIZE;
+
+#if defined UNIV_DEBUG || defined UNIV_ZIP_DEBUG
+	ut_a(!memcmp(storage, field, REC_NODE_PTR_SIZE));
+#endif /* UNIV_DEBUG || UNIV_ZIP_DEBUG */
+#if REC_NODE_PTR_SIZE != 4
+# error "REC_NODE_PTR_SIZE != 4"
+#endif
+	mach_write_to_4(field, ptr);
+	memcpy(storage, field, REC_NODE_PTR_SIZE);
+
+	if (mtr) {
+		byte*	log_ptr	= mlog_open(mtr,
+					    11 + 2 + 2 + REC_NODE_PTR_SIZE);
+		if (UNIV_UNLIKELY(!log_ptr)) {
+			return;
+		}
+
+		log_ptr = mlog_write_initial_log_record_fast(
+			field, MLOG_ZIP_WRITE_NODE_PTR, log_ptr, mtr);
+		mach_write_to_2(log_ptr, page_offset(field));
+		log_ptr += 2;
+		mach_write_to_2(log_ptr, storage - page_zip->data);
+		log_ptr += 2;
+		memcpy(log_ptr, field, REC_NODE_PTR_SIZE);
+		log_ptr += REC_NODE_PTR_SIZE;
+		mlog_close(mtr, log_ptr);
+	}
+}
+
+/**************************************************************************
+Write the trx_id and roll_ptr of a record on a B-tree leaf node page. */
+UNIV_INTERN
+void
+page_zip_write_trx_id_and_roll_ptr(
+/*===============================*/
+	page_zip_des_t*	page_zip,/* in/out: compressed page */
+	byte*		rec,	/* in/out: record */
+	const ulint*	offsets,/* in: rec_get_offsets(rec, index) */
+	ulint		trx_id_col,/* in: column number of TRX_ID in rec */
+	dulint		trx_id,	/* in: transaction identifier */
+	dulint		roll_ptr)/* in: roll_ptr */
+{
+	byte*	field;
+	byte*	storage;
+	page_t*	page	= page_align(rec);
+	ulint	len;
+
+	ut_ad(buf_frame_get_page_zip(rec) == page_zip);
+	ut_ad(page_simple_validate_new(page));
+	ut_ad(page_zip_simple_validate(page_zip));
+	ut_ad(page_zip_get_size(page_zip)
+	      > PAGE_DATA + page_zip_dir_size(page_zip));
+	ut_ad(rec_offs_validate(rec, NULL, offsets));
+	ut_ad(rec_offs_comp(offsets));
+
+	ut_ad(page_zip->m_start >= PAGE_DATA);
+	ut_ad(page_zip_header_cmp(page_zip, page));
+
+	ut_ad(page_is_leaf(page));
+
+	UNIV_MEM_ASSERT_RW(page_zip->data, page_zip_get_size(page_zip));
+
+	storage = page_zip->data + page_zip_get_size(page_zip)
+		- (page_dir_get_n_heap(page) - PAGE_HEAP_NO_USER_LOW)
+		* PAGE_ZIP_DIR_SLOT_SIZE
+		- (rec_get_heap_no_new(rec) - 1)
+		* (DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN);
+
+#if DATA_TRX_ID + 1 != DATA_ROLL_PTR
+# error "DATA_TRX_ID + 1 != DATA_ROLL_PTR"
+#endif
+	field = rec_get_nth_field(rec, offsets, trx_id_col, &len);
+	ut_ad(len == DATA_TRX_ID_LEN);
+	ut_ad(field + DATA_TRX_ID_LEN
+	      == rec_get_nth_field(rec, offsets, trx_id_col + 1, &len));
+	ut_ad(len == DATA_ROLL_PTR_LEN);
+#if defined UNIV_DEBUG || defined UNIV_ZIP_DEBUG
+	ut_a(!memcmp(storage, field, DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN));
+#endif /* UNIV_DEBUG || UNIV_ZIP_DEBUG */
+#if DATA_TRX_ID_LEN != 6
+# error "DATA_TRX_ID_LEN != 6"
+#endif
+	mach_write_to_6(field, trx_id);
+#if DATA_ROLL_PTR_LEN != 7
+# error "DATA_ROLL_PTR_LEN != 7"
+#endif
+	mach_write_to_7(field + DATA_TRX_ID_LEN, roll_ptr);
+	memcpy(storage, field, DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN);
+
+	UNIV_MEM_ASSERT_RW(rec, rec_offs_data_size(offsets));
+	UNIV_MEM_ASSERT_RW(rec - rec_offs_extra_size(offsets),
+			   rec_offs_extra_size(offsets));
+	UNIV_MEM_ASSERT_RW(page_zip->data, page_zip_get_size(page_zip));
+}
+
+#ifdef UNIV_ZIP_DEBUG
+/* Set this variable in a debugger to disable page_zip_clear_rec().
+The only observable effect should be the compression ratio due to
+deleted records not being zeroed out.  In rare cases, there can be
+page_zip_validate() failures on the node_ptr, trx_id and roll_ptr
+columns if the space is reallocated for a smaller record. */
+UNIV_INTERN ibool	page_zip_clear_rec_disable;
+#endif /* UNIV_ZIP_DEBUG */
+
+/**************************************************************************
+Clear an area on the uncompressed and compressed page, if possible. */
+static
+void
+page_zip_clear_rec(
+/*===============*/
+	page_zip_des_t*	page_zip,/* in/out: compressed page */
+	byte*		rec,	/* in: record to clear */
+	dict_index_t*	index,	/* in: index of rec */
+	const ulint*	offsets)/* in: rec_get_offsets(rec, index) */
+{
+	ulint	heap_no;
+	page_t*	page	= page_align(rec);
+	/* page_zip_validate() would fail here if a record
+	containing externally stored columns is being deleted. */
+	ut_ad(rec_offs_validate(rec, index, offsets));
+	ut_ad(!page_zip_dir_find(page_zip, page_offset(rec)));
+	ut_ad(page_zip_dir_find_free(page_zip, page_offset(rec)));
+	ut_ad(page_zip_header_cmp(page_zip, page));
+
+	heap_no = rec_get_heap_no_new(rec);
+	ut_ad(heap_no >= PAGE_HEAP_NO_USER_LOW);
+
+	UNIV_MEM_ASSERT_RW(page_zip->data, page_zip_get_size(page_zip));
+	UNIV_MEM_ASSERT_RW(rec, rec_offs_data_size(offsets));
+	UNIV_MEM_ASSERT_RW(rec - rec_offs_extra_size(offsets),
+			   rec_offs_extra_size(offsets));
+
+	if (
+#ifdef UNIV_ZIP_DEBUG
+	    !page_zip_clear_rec_disable &&
+#endif /* UNIV_ZIP_DEBUG */
+	    page_zip->m_end
+	    + 1 + ((heap_no - 1) >= 64)/* size of the log entry */
+	    + page_zip_get_trailer_len(page_zip,
+				       dict_index_is_clust(index), NULL)
+	    < page_zip_get_size(page_zip)) {
+		byte*	data;
+
+		/* Clear only the data bytes, because the allocator and
+		the decompressor depend on the extra bytes. */
+		memset(rec, 0, rec_offs_data_size(offsets));
+
+		if (!page_is_leaf(page)) {
+			/* Clear node_ptr on the compressed page. */
+			byte*	storage	= page_zip->data
+				+ page_zip_get_size(page_zip)
+				- (page_dir_get_n_heap(page)
+				   - PAGE_HEAP_NO_USER_LOW)
+				* PAGE_ZIP_DIR_SLOT_SIZE;
+
+			memset(storage - (heap_no - 1) * REC_NODE_PTR_SIZE,
+			       0, REC_NODE_PTR_SIZE);
+		} else if (dict_index_is_clust(index)) {
+			/* Clear trx_id and roll_ptr on the compressed page. */
+			byte*	storage	= page_zip->data
+				+ page_zip_get_size(page_zip)
+				- (page_dir_get_n_heap(page)
+				   - PAGE_HEAP_NO_USER_LOW)
+				* PAGE_ZIP_DIR_SLOT_SIZE;
+
+			memset(storage - (heap_no - 1)
+			       * (DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN),
+			       0, DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN);
+		}
+
+		/* Log that the data was zeroed out. */
+		data = page_zip->data + page_zip->m_end;
+		ut_ad(!*data);
+		if (UNIV_UNLIKELY(heap_no - 1 >= 64)) {
+			*data++ = (byte) (0x80 | (heap_no - 1) >> 7);
+			ut_ad(!*data);
+		}
+		*data++ = (byte) ((heap_no - 1) << 1 | 1);
+		ut_ad(!*data);
+		ut_ad((ulint) (data - page_zip->data)
+		      < page_zip_get_size(page_zip));
+		page_zip->m_end = data - page_zip->data;
+		page_zip->m_nonempty = TRUE;
+	} else if (page_is_leaf(page) && dict_index_is_clust(index)) {
+		/* Do not clear the record, because there is not enough space
+		to log the operation. */
+
+		if (rec_offs_any_extern(offsets)) {
+			ulint	i;
+
+			for (i = rec_offs_n_fields(offsets); i--; ) {
+				/* Clear all BLOB pointers in order to make
+				page_zip_validate() pass. */
+				if (rec_offs_nth_extern(offsets, i)) {
+					ulint	len;
+					byte*	field = rec_get_nth_field(
+						rec, offsets, i, &len);
+					memset(field + len
+					       - BTR_EXTERN_FIELD_REF_SIZE,
+					       0, BTR_EXTERN_FIELD_REF_SIZE);
+				}
+			}
+		}
+	}
+
+#ifdef UNIV_ZIP_DEBUG
+	ut_a(page_zip_validate(page_zip, page));
+#endif /* UNIV_ZIP_DEBUG */
+}
+
+/**************************************************************************
+Write the "deleted" flag of a record on a compressed page.  The flag must
+already have been written on the uncompressed page. */
+UNIV_INTERN
+void
+page_zip_rec_set_deleted(
+/*=====================*/
+	page_zip_des_t*	page_zip,/* in/out: compressed page */
+	const byte*	rec,	/* in: record on the uncompressed page */
+	ulint		flag)	/* in: the deleted flag (nonzero=TRUE) */
+{
+	byte*	slot = page_zip_dir_find(page_zip, page_offset(rec));
+	ut_a(slot);
+	UNIV_MEM_ASSERT_RW(page_zip->data, page_zip_get_size(page_zip));
+	if (flag) {
+		*slot |= (PAGE_ZIP_DIR_SLOT_DEL >> 8);
+	} else {
+		*slot &= ~(PAGE_ZIP_DIR_SLOT_DEL >> 8);
+	}
+#ifdef UNIV_ZIP_DEBUG
+	ut_a(page_zip_validate(page_zip, page_align(rec)));
+#endif /* UNIV_ZIP_DEBUG */
+}
+
+/**************************************************************************
+Write the "owned" flag of a record on a compressed page.  The n_owned field
+must already have been written on the uncompressed page. */
+UNIV_INTERN
+void
+page_zip_rec_set_owned(
+/*===================*/
+	page_zip_des_t*	page_zip,/* in/out: compressed page */
+	const byte*	rec,	/* in: record on the uncompressed page */
+	ulint		flag)	/* in: the owned flag (nonzero=TRUE) */
+{
+	byte*	slot = page_zip_dir_find(page_zip, page_offset(rec));
+	ut_a(slot);
+	UNIV_MEM_ASSERT_RW(page_zip->data, page_zip_get_size(page_zip));
+	if (flag) {
+		*slot |= (PAGE_ZIP_DIR_SLOT_OWNED >> 8);
+	} else {
+		*slot &= ~(PAGE_ZIP_DIR_SLOT_OWNED >> 8);
+	}
+}
+
+/**************************************************************************
+Insert a record to the dense page directory. */
+UNIV_INTERN
+void
+page_zip_dir_insert(
+/*================*/
+	page_zip_des_t*	page_zip,/* in/out: compressed page */
+	const byte*	prev_rec,/* in: record after which to insert */
+	const byte*	free_rec,/* in: record from which rec was
+				allocated, or NULL */
+	byte*		rec)	/* in: record to insert */
+{
+	ulint	n_dense;
+	byte*	slot_rec;
+	byte*	slot_free;
+
+	ut_ad(prev_rec != rec);
+	ut_ad(page_rec_get_next((rec_t*) prev_rec) == rec);
+	ut_ad(page_zip_simple_validate(page_zip));
+
+	UNIV_MEM_ASSERT_RW(page_zip->data, page_zip_get_size(page_zip));
+
+	if (page_rec_is_infimum(prev_rec)) {
+		/* Use the first slot. */
+		slot_rec = page_zip->data + page_zip_get_size(page_zip);
+	} else {
+		byte*	end	= page_zip->data + page_zip_get_size(page_zip);
+		byte*	start	= end - page_zip_dir_user_size(page_zip);
+
+		if (UNIV_LIKELY(!free_rec)) {
+			/* PAGE_N_RECS was already incremented
+			in page_cur_insert_rec_zip(), but the
+			dense directory slot at that position
+			contains garbage.  Skip it. */
+			start += PAGE_ZIP_DIR_SLOT_SIZE;
+		}
+
+		slot_rec = page_zip_dir_find_low(start, end,
+						 page_offset(prev_rec));
+		ut_a(slot_rec);
+	}
+
+	/* Read the old n_dense (n_heap may have been incremented). */
+	n_dense = page_dir_get_n_heap(page_zip->data)
+		- (PAGE_HEAP_NO_USER_LOW + 1);
+
+	if (UNIV_LIKELY_NULL(free_rec)) {
+		/* The record was allocated from the free list.
+		Shift the dense directory only up to that slot.
+		Note that in this case, n_dense is actually
+		off by one, because page_cur_insert_rec_zip()
+		did not increment n_heap. */
+		ut_ad(rec_get_heap_no_new(rec) < n_dense + 1
+		      + PAGE_HEAP_NO_USER_LOW);
+		ut_ad(rec >= free_rec);
+		slot_free = page_zip_dir_find(page_zip, page_offset(free_rec));
+		ut_ad(slot_free);
+		slot_free += PAGE_ZIP_DIR_SLOT_SIZE;
+	} else {
+		/* The record was allocated from the heap.
+		Shift the entire dense directory. */
+		ut_ad(rec_get_heap_no_new(rec) == n_dense
+		      + PAGE_HEAP_NO_USER_LOW);
+
+		/* Shift to the end of the dense page directory. */
+		slot_free = page_zip->data + page_zip_get_size(page_zip)
+			- PAGE_ZIP_DIR_SLOT_SIZE * n_dense;
+	}
+
+	/* Shift the dense directory to allocate place for rec. */
+	memmove(slot_free - PAGE_ZIP_DIR_SLOT_SIZE, slot_free,
+		slot_rec - slot_free);
+
+	/* Write the entry for the inserted record.
+	The "owned" and "deleted" flags must be zero. */
+	mach_write_to_2(slot_rec - PAGE_ZIP_DIR_SLOT_SIZE, page_offset(rec));
+}
+
+/**************************************************************************
+Shift the dense page directory and the array of BLOB pointers
+when a record is deleted. */
+UNIV_INTERN
+void
+page_zip_dir_delete(
+/*================*/
+	page_zip_des_t*	page_zip,/* in/out: compressed page */
+	byte*		rec,	/* in: record to delete */
+	dict_index_t*	index,	/* in: index of rec */
+	const ulint*	offsets,/* in: rec_get_offsets(rec) */
+	const byte*	free)	/* in: previous start of the free list */
+{
+	byte*	slot_rec;
+	byte*	slot_free;
+	ulint	n_ext;
+	page_t*	page	= page_align(rec);
+
+	ut_ad(rec_offs_validate(rec, index, offsets));
+	ut_ad(rec_offs_comp(offsets));
+
+	UNIV_MEM_ASSERT_RW(page_zip->data, page_zip_get_size(page_zip));
+	UNIV_MEM_ASSERT_RW(rec, rec_offs_data_size(offsets));
+	UNIV_MEM_ASSERT_RW(rec - rec_offs_extra_size(offsets),
+			   rec_offs_extra_size(offsets));
+
+	slot_rec = page_zip_dir_find(page_zip, page_offset(rec));
+
+	ut_a(slot_rec);
+
+	/* This could not be done before page_zip_dir_find(). */
+	page_header_set_field(page, page_zip, PAGE_N_RECS,
+			      (ulint)(page_get_n_recs(page) - 1));
+
+	if (UNIV_UNLIKELY(!free)) {
+		/* Make the last slot the start of the free list. */
+		slot_free = page_zip->data + page_zip_get_size(page_zip)
+			- PAGE_ZIP_DIR_SLOT_SIZE
+			* (page_dir_get_n_heap(page_zip->data)
+			   - PAGE_HEAP_NO_USER_LOW);
+	} else {
+		slot_free = page_zip_dir_find_free(page_zip,
+						   page_offset(free));
+		ut_a(slot_free < slot_rec);
+		/* Grow the free list by one slot by moving the start. */
+		slot_free += PAGE_ZIP_DIR_SLOT_SIZE;
+	}
+
+	if (UNIV_LIKELY(slot_rec > slot_free)) {
+		memmove(slot_free + PAGE_ZIP_DIR_SLOT_SIZE,
+			slot_free,
+			slot_rec - slot_free);
+	}
+
+	/* Write the entry for the deleted record.
+	The "owned" and "deleted" flags will be cleared. */
+	mach_write_to_2(slot_free, page_offset(rec));
+
+	if (!page_is_leaf(page) || !dict_index_is_clust(index)) {
+		ut_ad(!rec_offs_any_extern(offsets));
+		goto skip_blobs;
+	}
+
+	n_ext = rec_offs_n_extern(offsets);
+	if (UNIV_UNLIKELY(n_ext)) {
+		/* Shift and zero fill the array of BLOB pointers. */
+		ulint	blob_no;
+		byte*	externs;
+		byte*	ext_end;
+
+		blob_no = page_zip_get_n_prev_extern(page_zip, rec, index);
+		ut_a(blob_no + n_ext <= page_zip->n_blobs);
+
+		externs = page_zip->data + page_zip_get_size(page_zip)
+			- (page_dir_get_n_heap(page) - PAGE_HEAP_NO_USER_LOW)
+			* (PAGE_ZIP_DIR_SLOT_SIZE
+			   + DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN);
+
+		ext_end = externs - page_zip->n_blobs
+			* BTR_EXTERN_FIELD_REF_SIZE;
+		externs -= blob_no * BTR_EXTERN_FIELD_REF_SIZE;
+
+		page_zip->n_blobs -= n_ext;
+		/* Shift and zero fill the array. */
+		memmove(ext_end + n_ext * BTR_EXTERN_FIELD_REF_SIZE, ext_end,
+			(page_zip->n_blobs - blob_no)
+			* BTR_EXTERN_FIELD_REF_SIZE);
+		memset(ext_end, 0, n_ext * BTR_EXTERN_FIELD_REF_SIZE);
+	}
+
+skip_blobs:
+	/* The compression algorithm expects info_bits and n_owned
+	to be 0 for deleted records. */
+	rec[-REC_N_NEW_EXTRA_BYTES] = 0; /* info_bits and n_owned */
+
+	page_zip_clear_rec(page_zip, rec, index, offsets);
+}
+
+/**************************************************************************
+Add a slot to the dense page directory. */
+UNIV_INTERN
+void
+page_zip_dir_add_slot(
+/*==================*/
+	page_zip_des_t*	page_zip,	/* in/out: compressed page */
+	ulint		is_clustered)	/* in: nonzero for clustered index,
+					zero for others */
+{
+	ulint	n_dense;
+	byte*	dir;
+	byte*	stored;
+
+	ut_ad(page_is_comp(page_zip->data));
+	UNIV_MEM_ASSERT_RW(page_zip->data, page_zip_get_size(page_zip));
+
+	/* Read the old n_dense (n_heap has already been incremented). */
+	n_dense = page_dir_get_n_heap(page_zip->data)
+		- (PAGE_HEAP_NO_USER_LOW + 1);
+
+	dir = page_zip->data + page_zip_get_size(page_zip)
+		- PAGE_ZIP_DIR_SLOT_SIZE * n_dense;
+
+	if (!page_is_leaf(page_zip->data)) {
+		ut_ad(!page_zip->n_blobs);
+		stored = dir - n_dense * REC_NODE_PTR_SIZE;
+	} else if (UNIV_UNLIKELY(is_clustered)) {
+		/* Move the BLOB pointer array backwards to make space for the
+		roll_ptr and trx_id columns and the dense directory slot. */
+		byte*	externs;
+
+		stored = dir - n_dense
+			* (DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN);
+		externs = stored
+			- page_zip->n_blobs * BTR_EXTERN_FIELD_REF_SIZE;
+		ASSERT_ZERO(externs
+			    - (PAGE_ZIP_DIR_SLOT_SIZE
+			       + DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN),
+			    PAGE_ZIP_DIR_SLOT_SIZE
+			    + DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN);
+		memmove(externs - (PAGE_ZIP_DIR_SLOT_SIZE
+				   + DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN),
+			externs, stored - externs);
+	} else {
+		stored = dir
+			- page_zip->n_blobs * BTR_EXTERN_FIELD_REF_SIZE;
+		ASSERT_ZERO(stored - PAGE_ZIP_DIR_SLOT_SIZE,
+			    PAGE_ZIP_DIR_SLOT_SIZE);
+	}
+
+	/* Move the uncompressed area backwards to make space
+	for one directory slot. */
+	memmove(stored - PAGE_ZIP_DIR_SLOT_SIZE, stored, dir - stored);
+}
+
+/***************************************************************
+Parses a log record of writing to the header of a page. */
+UNIV_INTERN
+byte*
+page_zip_parse_write_header(
+/*========================*/
+				/* out: end of log record or NULL */
+	byte*		ptr,	/* in: redo log buffer */
+	byte*		end_ptr,/* in: redo log buffer end */
+	page_t*		page,	/* in/out: uncompressed page */
+	page_zip_des_t*	page_zip)/* in/out: compressed page */
+{
+	ulint	offset;
+	ulint	len;
+
+	ut_ad(ptr && end_ptr);
+	ut_ad(!page == !page_zip);
+
+	if (UNIV_UNLIKELY(end_ptr < ptr + (1 + 1))) {
+
+		return(NULL);
+	}
+
+	offset = (ulint) *ptr++;
+	len = (ulint) *ptr++;
+
+	if (UNIV_UNLIKELY(!len) || UNIV_UNLIKELY(offset + len >= PAGE_DATA)) {
+corrupt:
+		recv_sys->found_corrupt_log = TRUE;
+
+		return(NULL);
+	}
+
+	if (UNIV_UNLIKELY(end_ptr < ptr + len)) {
+
+		return(NULL);
+	}
+
+	if (page) {
+		if (UNIV_UNLIKELY(!page_zip)) {
+
+			goto corrupt;
+		}
+#ifdef UNIV_ZIP_DEBUG
+		ut_a(page_zip_validate(page_zip, page));
+#endif /* UNIV_ZIP_DEBUG */
+
+		memcpy(page + offset, ptr, len);
+		memcpy(page_zip->data + offset, ptr, len);
+
+#ifdef UNIV_ZIP_DEBUG
+		ut_a(page_zip_validate(page_zip, page));
+#endif /* UNIV_ZIP_DEBUG */
+	}
+
+	return(ptr + len);
+}
+
+/**************************************************************************
+Write a log record of writing to the uncompressed header portion of a page. */
+UNIV_INTERN
+void
+page_zip_write_header_log(
+/*======================*/
+	const byte*	data,	/* in: data on the uncompressed page */
+	ulint		length,	/* in: length of the data */
+	mtr_t*		mtr)	/* in: mini-transaction */
+{
+	byte*	log_ptr	= mlog_open(mtr, 11 + 1 + 1);
+	ulint	offset	= page_offset(data);
+
+	ut_ad(offset < PAGE_DATA);
+	ut_ad(offset + length < PAGE_DATA);
+#if PAGE_DATA > 255
+# error "PAGE_DATA > 255"
+#endif
+	ut_ad(length < 256);
+
+	/* If no logging is requested, we may return now */
+	if (UNIV_UNLIKELY(!log_ptr)) {
+
+		return;
+	}
+
+	log_ptr = mlog_write_initial_log_record_fast(
+		(byte*) data, MLOG_ZIP_WRITE_HEADER, log_ptr, mtr);
+	*log_ptr++ = (byte) offset;
+	*log_ptr++ = (byte) length;
+	mlog_close(mtr, log_ptr);
+
+	mlog_catenate_string(mtr, data, length);
+}
+
+/**************************************************************************
+Reorganize and compress a page.  This is a low-level operation for
+compressed pages, to be used when page_zip_compress() fails.
+On success, a redo log entry MLOG_ZIP_PAGE_COMPRESS will be written.
+The function btr_page_reorganize() should be preferred whenever possible.
+IMPORTANT: if page_zip_reorganize() is invoked on a leaf page of a
+non-clustered index, the caller must update the insert buffer free
+bits in the same mini-transaction in such a way that the modification
+will be redo-logged. */
+UNIV_INTERN
+ibool
+page_zip_reorganize(
+/*================*/
+				/* out: TRUE on success, FALSE on failure;
+				page and page_zip will be left intact
+				on failure. */
+	buf_block_t*	block,	/* in/out: page with compressed page;
+				on the compressed page, in: size;
+				out: data, n_blobs,
+				m_start, m_end, m_nonempty */
+	dict_index_t*	index,	/* in: index of the B-tree node */
+	mtr_t*		mtr)	/* in: mini-transaction */
+{
+	page_zip_des_t*	page_zip	= buf_block_get_page_zip(block);
+	page_t*		page		= buf_block_get_frame(block);
+	buf_block_t*	temp_block;
+	page_t*		temp_page;
+	ulint		log_mode;
+
+	ut_ad(mtr_memo_contains(mtr, block, MTR_MEMO_PAGE_X_FIX));
+	ut_ad(page_is_comp(page));
+	/* Note that page_zip_validate(page_zip, page) may fail here. */
+	UNIV_MEM_ASSERT_RW(page, UNIV_PAGE_SIZE);
+	UNIV_MEM_ASSERT_RW(page_zip->data, page_zip_get_size(page_zip));
+
+	/* Disable logging */
+	log_mode = mtr_set_log_mode(mtr, MTR_LOG_NONE);
+
+	temp_block = buf_block_alloc(0);
+	temp_page = temp_block->frame;
+
+	btr_search_drop_page_hash_index(block);
+
+	/* Copy the old page to temporary space */
+	buf_frame_copy(temp_page, page);
+
+	/* Recreate the page: note that global data on page (possible
+	segment headers, next page-field, etc.) is preserved intact */
+
+	page_create(block, mtr, TRUE);
+	block->check_index_page_at_flush = TRUE;
+
+	/* Copy the records from the temporary space to the recreated page;
+	do not copy the lock bits yet */
+
+	page_copy_rec_list_end_no_locks(block, temp_block,
+					page_get_infimum_rec(temp_page),
+					index, mtr);
+	/* Copy max trx id to recreated page */
+	page_set_max_trx_id(block, NULL, page_get_max_trx_id(temp_page));
+
+	/* Restore logging. */
+	mtr_set_log_mode(mtr, log_mode);
+
+	if (UNIV_UNLIKELY(!page_zip_compress(page_zip, page, index, mtr))) {
+
+		/* Restore the old page and exit. */
+		buf_frame_copy(page, temp_page);
+
+		buf_block_free(temp_block);
+		return(FALSE);
+	}
+
+	lock_move_reorganize_page(block, temp_block);
+
+	buf_block_free(temp_block);
+	return(TRUE);
+}
+
+/**************************************************************************
+Copy the records of a page byte for byte.  Do not copy the page header
+or trailer, except those B-tree header fields that are directly
+related to the storage of records.  Also copy PAGE_MAX_TRX_ID.
+NOTE: The caller must update the lock table and the adaptive hash index. */
+UNIV_INTERN
+void
+page_zip_copy_recs(
+/*===============*/
+	page_zip_des_t*		page_zip,	/* out: copy of src_zip
+						(n_blobs, m_start, m_end,
+						m_nonempty, data[0..size-1]) */
+	page_t*			page,		/* out: copy of src */
+	const page_zip_des_t*	src_zip,	/* in: compressed page */
+	const page_t*		src,		/* in: page */
+	dict_index_t*		index,		/* in: index of the B-tree */
+	mtr_t*			mtr)		/* in: mini-transaction */
+{
+	ut_ad(mtr_memo_contains_page(mtr, page, MTR_MEMO_PAGE_X_FIX));
+	ut_ad(mtr_memo_contains_page(mtr, (page_t*) src, MTR_MEMO_PAGE_X_FIX));
+#ifdef UNIV_ZIP_DEBUG
+	/* The B-tree operations that call this function may set
+	FIL_PAGE_PREV or PAGE_LEVEL, causing a temporary min_rec_flag
+	mismatch.  A strict page_zip_validate() will be executed later
+	during the B-tree operations. */
+	ut_a(page_zip_validate_low(src_zip, src, TRUE));
+#endif /* UNIV_ZIP_DEBUG */
+	ut_a(page_zip_get_size(page_zip) == page_zip_get_size(src_zip));
+	if (UNIV_UNLIKELY(src_zip->n_blobs)) {
+		ut_a(page_is_leaf(src));
+		ut_a(dict_index_is_clust(index));
+	}
+
+	UNIV_MEM_ASSERT_W(page, UNIV_PAGE_SIZE);
+	UNIV_MEM_ASSERT_W(page_zip->data, page_zip_get_size(page_zip));
+	UNIV_MEM_ASSERT_RW(src, UNIV_PAGE_SIZE);
+	UNIV_MEM_ASSERT_RW(src_zip->data, page_zip_get_size(page_zip));
+
+	/* Copy those B-tree page header fields that are related to
+	the records stored in the page.  Also copy the field
+	PAGE_MAX_TRX_ID.  Skip the rest of the page header and
+	trailer.  On the compressed page, there is no trailer. */
+#if PAGE_MAX_TRX_ID + 8 != PAGE_HEADER_PRIV_END
+# error "PAGE_MAX_TRX_ID + 8 != PAGE_HEADER_PRIV_END"
+#endif
+	memcpy(PAGE_HEADER + page, PAGE_HEADER + src,
+	       PAGE_HEADER_PRIV_END);
+	memcpy(PAGE_DATA + page, PAGE_DATA + src,
+	       UNIV_PAGE_SIZE - PAGE_DATA - FIL_PAGE_DATA_END);
+	memcpy(PAGE_HEADER + page_zip->data, PAGE_HEADER + src_zip->data,
+	       PAGE_HEADER_PRIV_END);
+	memcpy(PAGE_DATA + page_zip->data, PAGE_DATA + src_zip->data,
+	       page_zip_get_size(page_zip) - PAGE_DATA);
+
+	/* Copy all fields of src_zip to page_zip, except the pointer
+	to the compressed data page. */
+	{
+		page_zip_t*	data = page_zip->data;
+		memcpy(page_zip, src_zip, sizeof *page_zip);
+		page_zip->data = data;
+	}
+	ut_ad(page_zip_get_trailer_len(page_zip,
+				       dict_index_is_clust(index), NULL)
+	      + page_zip->m_end < page_zip_get_size(page_zip));
+
+	if (!page_is_leaf(src)
+	    && UNIV_UNLIKELY(mach_read_from_4(src + FIL_PAGE_PREV) == FIL_NULL)
+	    && UNIV_LIKELY(mach_read_from_4(page
+					    + FIL_PAGE_PREV) != FIL_NULL)) {
+		/* Clear the REC_INFO_MIN_REC_FLAG of the first user record. */
+		ulint	offs = rec_get_next_offs(page + PAGE_NEW_INFIMUM,
+						 TRUE);
+		if (UNIV_LIKELY(offs != PAGE_NEW_SUPREMUM)) {
+			rec_t*	rec = page + offs;
+			ut_a(rec[-REC_N_NEW_EXTRA_BYTES]
+			     & REC_INFO_MIN_REC_FLAG);
+			rec[-REC_N_NEW_EXTRA_BYTES] &= ~ REC_INFO_MIN_REC_FLAG;
+		}
+	}
+
+#ifdef UNIV_ZIP_DEBUG
+	ut_a(page_zip_validate(page_zip, page));
+#endif /* UNIV_ZIP_DEBUG */
+
+	page_zip_compress_write_log(page_zip, page, index, mtr);
+}
+
+/**************************************************************************
+Parses a log record of compressing an index page. */
+UNIV_INTERN
+byte*
+page_zip_parse_compress(
+/*====================*/
+				/* out: end of log record or NULL */
+	byte*		ptr,	/* in: buffer */
+	byte*		end_ptr,/* in: buffer end */
+	page_t*		page,	/* out: uncompressed page */
+	page_zip_des_t*	page_zip)/* out: compressed page */
+{
+	ulint	size;
+	ulint	trailer_size;
+
+	ut_ad(ptr && end_ptr);
+	ut_ad(!page == !page_zip);
+
+	if (UNIV_UNLIKELY(ptr + (2 + 2) > end_ptr)) {
+
+		return(NULL);
+	}
+
+	size = mach_read_from_2(ptr);
+	ptr += 2;
+	trailer_size = mach_read_from_2(ptr);
+	ptr += 2;
+
+	if (UNIV_UNLIKELY(ptr + 8 + size + trailer_size > end_ptr)) {
+
+		return(NULL);
+	}
+
+	if (page) {
+		if (UNIV_UNLIKELY(!page_zip)
+		    || UNIV_UNLIKELY(page_zip_get_size(page_zip) < size)) {
+corrupt:
+			recv_sys->found_corrupt_log = TRUE;
+
+			return(NULL);
+		}
+
+		memcpy(page_zip->data + FIL_PAGE_PREV, ptr, 4);
+		memcpy(page_zip->data + FIL_PAGE_NEXT, ptr + 4, 4);
+		memcpy(page_zip->data + FIL_PAGE_TYPE, ptr + 8, size);
+		memset(page_zip->data + FIL_PAGE_TYPE + size, 0,
+		       page_zip_get_size(page_zip) - trailer_size
+		       - (FIL_PAGE_TYPE + size));
+		memcpy(page_zip->data + page_zip_get_size(page_zip)
+		       - trailer_size, ptr + 8 + size, trailer_size);
+
+		if (UNIV_UNLIKELY(!page_zip_decompress(page_zip, page))) {
+
+			goto corrupt;
+		}
+	}
+
+	return(ptr + 8 + size + trailer_size);
+}
+
+/**************************************************************************
+Calculate the compressed page checksum. */
+UNIV_INTERN
+ulint
+page_zip_calc_checksum(
+/*===================*/
+				/* out: page checksum */
+	const void*	data,	/* in: compressed page */
+	ulint		size)	/* in: size of compressed page */
+{
+	/* Exclude FIL_PAGE_SPACE_OR_CHKSUM, FIL_PAGE_LSN,
+	and FIL_PAGE_FILE_FLUSH_LSN from the checksum. */
+
+	const Bytef*	s	= data;
+	uLong		adler;
+
+	ut_ad(size > FIL_PAGE_ARCH_LOG_NO_OR_SPACE_ID);
+
+	adler = adler32(0L, s + FIL_PAGE_OFFSET,
+			FIL_PAGE_LSN - FIL_PAGE_OFFSET);
+	adler = adler32(adler, s + FIL_PAGE_TYPE, 2);
+	adler = adler32(adler, s + FIL_PAGE_ARCH_LOG_NO_OR_SPACE_ID,
+			size - FIL_PAGE_ARCH_LOG_NO_OR_SPACE_ID);
+
+	return((ulint) adler);
+}
diff --git a/storage/xtradb/pars/lexyy.c b/storage/xtradb/pars/lexyy.c
new file mode 100644
index 00000000000..489752a1900
--- /dev/null
+++ b/storage/xtradb/pars/lexyy.c
@@ -0,0 +1,2780 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+#include "univ.i"
+#line 2 "lexyy.c"
+
+#line 4 "lexyy.c"
+
+#define  YY_INT_ALIGNED short int
+
+/* A lexical scanner generated by flex */
+
+#define FLEX_SCANNER
+#define YY_FLEX_MAJOR_VERSION 2
+#define YY_FLEX_MINOR_VERSION 5
+#define YY_FLEX_SUBMINOR_VERSION 31
+#if YY_FLEX_SUBMINOR_VERSION > 0
+#define FLEX_BETA
+#endif
+
+/* First, we deal with  platform-specific or compiler-specific issues. */
+
+/* begin standard C headers. */
+#include <stdio.h>
+#include <string.h>
+#include <errno.h>
+#include <stdlib.h>
+
+/* end standard C headers. */
+
+/* flex integer type definitions */
+
+#ifndef FLEXINT_H
+#define FLEXINT_H
+
+/* C99 systems have <inttypes.h>. Non-C99 systems may or may not. */
+
+#if defined __STDC_VERSION__ && __STDC_VERSION__ >= 199901L
+#include <inttypes.h>
+typedef int8_t flex_int8_t;
+typedef uint8_t flex_uint8_t;
+typedef int16_t flex_int16_t;
+typedef uint16_t flex_uint16_t;
+typedef int32_t flex_int32_t;
+typedef uint32_t flex_uint32_t;
+#else
+typedef signed char flex_int8_t;
+typedef short int flex_int16_t;
+typedef int flex_int32_t;
+typedef unsigned char flex_uint8_t; 
+typedef unsigned short int flex_uint16_t;
+typedef unsigned int flex_uint32_t;
+#endif /* ! C99 */
+
+/* Limits of integral types. */
+#ifndef INT8_MIN
+#define INT8_MIN               (-128)
+#endif
+#ifndef INT16_MIN
+#define INT16_MIN              (-32767-1)
+#endif
+#ifndef INT32_MIN
+#define INT32_MIN              (-2147483647-1)
+#endif
+#ifndef INT8_MAX
+#define INT8_MAX               (127)
+#endif
+#ifndef INT16_MAX
+#define INT16_MAX              (32767)
+#endif
+#ifndef INT32_MAX
+#define INT32_MAX              (2147483647)
+#endif
+#ifndef UINT8_MAX
+#define UINT8_MAX              (255U)
+#endif
+#ifndef UINT16_MAX
+#define UINT16_MAX             (65535U)
+#endif
+#ifndef UINT32_MAX
+#define UINT32_MAX             (4294967295U)
+#endif
+
+#endif /* ! FLEXINT_H */
+
+#ifdef __cplusplus
+
+/* The "const" storage-class-modifier is valid. */
+#define YY_USE_CONST
+
+#else	/* ! __cplusplus */
+
+#if __STDC__
+
+#define YY_USE_CONST
+
+#endif	/* __STDC__ */
+#endif	/* ! __cplusplus */
+
+#ifdef YY_USE_CONST
+#define yyconst const
+#else
+#define yyconst
+#endif
+
+/* Returned upon end-of-file. */
+#define YY_NULL 0
+
+/* Promotes a possibly negative, possibly signed char to an unsigned
+ * integer for use as an array index.  If the signed char is negative,
+ * we want to instead treat it as an 8-bit unsigned char, hence the
+ * double cast.
+ */
+#define YY_SC_TO_UI(c) ((unsigned int) (unsigned char) c)
+
+/* Enter a start condition.  This macro really ought to take a parameter,
+ * but we do it the disgusting crufty way forced on us by the ()-less
+ * definition of BEGIN.
+ */
+#define BEGIN (yy_start) = 1 + 2 *
+
+/* Translate the current start state into a value that can be later handed
+ * to BEGIN to return to the state.  The YYSTATE alias is for lex
+ * compatibility.
+ */
+#define YY_START (((yy_start) - 1) / 2)
+#define YYSTATE YY_START
+
+/* Action number for EOF rule of a given start state. */
+#define YY_STATE_EOF(state) (YY_END_OF_BUFFER + state + 1)
+
+/* Special action meaning "start processing a new file". */
+#define YY_NEW_FILE yyrestart(yyin  )
+
+#define YY_END_OF_BUFFER_CHAR 0
+
+/* Size of default input buffer. */
+#ifndef YY_BUF_SIZE
+#define YY_BUF_SIZE 16384
+#endif
+
+#ifndef YY_TYPEDEF_YY_BUFFER_STATE
+#define YY_TYPEDEF_YY_BUFFER_STATE
+typedef struct yy_buffer_state *YY_BUFFER_STATE;
+#endif
+
+static int yyleng;
+
+static FILE *yyin, *yyout;
+
+#define EOB_ACT_CONTINUE_SCAN 0
+#define EOB_ACT_END_OF_FILE 1
+#define EOB_ACT_LAST_MATCH 2
+
+    #define YY_LESS_LINENO(n)
+    
+/* Return all but the first "n" matched characters back to the input stream. */
+#define yyless(n) \
+	do \
+		{ \
+		/* Undo effects of setting up yytext. */ \
+        int yyless_macro_arg = (n); \
+        YY_LESS_LINENO(yyless_macro_arg);\
+		*yy_cp = (yy_hold_char); \
+		YY_RESTORE_YY_MORE_OFFSET \
+		(yy_c_buf_p) = yy_cp = yy_bp + yyless_macro_arg - YY_MORE_ADJ; \
+		YY_DO_BEFORE_ACTION; /* set up yytext again */ \
+		} \
+	while ( 0 )
+
+#define unput(c) yyunput( c, (yytext_ptr)  )
+
+/* The following is because we cannot portably get our hands on size_t
+ * (without autoconf's help, which isn't available because we want
+ * flex-generated scanners to compile on their own).
+ */
+
+#ifndef YY_TYPEDEF_YY_SIZE_T
+#define YY_TYPEDEF_YY_SIZE_T
+typedef unsigned int yy_size_t;
+#endif
+
+#ifndef YY_STRUCT_YY_BUFFER_STATE
+#define YY_STRUCT_YY_BUFFER_STATE
+struct yy_buffer_state
+	{
+	FILE *yy_input_file;
+
+	char *yy_ch_buf;		/* input buffer */
+	char *yy_buf_pos;		/* current position in input buffer */
+
+	/* Size of input buffer in bytes, not including room for EOB
+	 * characters.
+	 */
+	yy_size_t yy_buf_size;
+
+	/* Number of characters read into yy_ch_buf, not including EOB
+	 * characters.
+	 */
+	int yy_n_chars;
+
+	/* Whether we "own" the buffer - i.e., we know we created it,
+	 * and can realloc() it to grow it, and should free() it to
+	 * delete it.
+	 */
+	int yy_is_our_buffer;
+
+	/* Whether this is an "interactive" input source; if so, and
+	 * if we're using stdio for input, then we want to use getc()
+	 * instead of fread(), to make sure we stop fetching input after
+	 * each newline.
+	 */
+	int yy_is_interactive;
+
+	/* Whether we're considered to be at the beginning of a line.
+	 * If so, '^' rules will be active on the next match, otherwise
+	 * not.
+	 */
+	int yy_at_bol;
+
+    int yy_bs_lineno; /**< The line count. */
+    int yy_bs_column; /**< The column count. */
+    
+	/* Whether to try to fill the input buffer when we reach the
+	 * end of it.
+	 */
+	int yy_fill_buffer;
+
+	int yy_buffer_status;
+
+#define YY_BUFFER_NEW 0
+#define YY_BUFFER_NORMAL 1
+	/* When an EOF's been seen but there's still some text to process
+	 * then we mark the buffer as YY_EOF_PENDING, to indicate that we
+	 * shouldn't try reading from the input source any more.  We might
+	 * still have a bunch of tokens to match, though, because of
+	 * possible backing-up.
+	 *
+	 * When we actually see the EOF, we change the status to "new"
+	 * (via yyrestart()), so that the user can continue scanning by
+	 * just pointing yyin at a new input file.
+	 */
+#define YY_BUFFER_EOF_PENDING 2
+
+	};
+#endif /* !YY_STRUCT_YY_BUFFER_STATE */
+
+/* Stack of input buffers. */
+static size_t yy_buffer_stack_top = 0; /**< index of top of stack. */
+static size_t yy_buffer_stack_max = 0; /**< capacity of stack. */
+static YY_BUFFER_STATE * yy_buffer_stack = 0; /**< Stack as an array. */
+
+/* We provide macros for accessing buffer states in case in the
+ * future we want to put the buffer states in a more general
+ * "scanner state".
+ *
+ * Returns the top of the stack, or NULL.
+ */
+#define YY_CURRENT_BUFFER ( (yy_buffer_stack) \
+                          ? (yy_buffer_stack)[(yy_buffer_stack_top)] \
+                          : NULL)
+
+/* Same as previous macro, but useful when we know that the buffer stack is not
+ * NULL or when we need an lvalue. For internal use only.
+ */
+#define YY_CURRENT_BUFFER_LVALUE (yy_buffer_stack)[(yy_buffer_stack_top)]
+
+/* yy_hold_char holds the character lost when yytext is formed. */
+static char yy_hold_char;
+static int yy_n_chars;		/* number of characters read into yy_ch_buf */
+static int yyleng;
+
+/* Points to current character in buffer. */
+static char *yy_c_buf_p = (char *) 0;
+static int yy_init = 1;		/* whether we need to initialize */
+static int yy_start = 0;	/* start state number */
+
+/* Flag which is used to allow yywrap()'s to do buffer switches
+ * instead of setting up a fresh yyin.  A bit of a hack ...
+ */
+static int yy_did_buffer_switch_on_eof;
+
+static void yyrestart (FILE *input_file  );
+__attribute__((unused)) static void yy_switch_to_buffer (YY_BUFFER_STATE new_buffer  );
+static YY_BUFFER_STATE yy_create_buffer (FILE *file,int size  );
+static void yy_delete_buffer (YY_BUFFER_STATE b  );
+static void yy_flush_buffer (YY_BUFFER_STATE b  );
+__attribute__((unused)) static void yypush_buffer_state (YY_BUFFER_STATE new_buffer  );
+__attribute__((unused)) static void yypop_buffer_state (void );
+
+static void yyensure_buffer_stack (void );
+static void yy_load_buffer_state (void );
+static void yy_init_buffer (YY_BUFFER_STATE b,FILE *file  );
+
+#define YY_FLUSH_BUFFER yy_flush_buffer(YY_CURRENT_BUFFER )
+
+YY_BUFFER_STATE yy_scan_buffer (char *base,yy_size_t size  );
+YY_BUFFER_STATE yy_scan_string (yyconst char *yy_str  );
+YY_BUFFER_STATE yy_scan_bytes (yyconst char *bytes,int len  );
+
+static void *yyalloc (yy_size_t  );
+static void *yyrealloc (void *,yy_size_t  );
+static void yyfree (void *  );
+
+#define yy_new_buffer yy_create_buffer
+
+#define yy_set_interactive(is_interactive) \
+	{ \
+	if ( ! YY_CURRENT_BUFFER ){ \
+        yyensure_buffer_stack (); \
+		YY_CURRENT_BUFFER_LVALUE =    \
+            yy_create_buffer(yyin,YY_BUF_SIZE ); \
+	} \
+	YY_CURRENT_BUFFER_LVALUE->yy_is_interactive = is_interactive; \
+	}
+
+#define yy_set_bol(at_bol) \
+	{ \
+	if ( ! YY_CURRENT_BUFFER ){\
+        yyensure_buffer_stack (); \
+		YY_CURRENT_BUFFER_LVALUE =    \
+            yy_create_buffer(yyin,YY_BUF_SIZE ); \
+	} \
+	YY_CURRENT_BUFFER_LVALUE->yy_at_bol = at_bol; \
+	}
+
+#define YY_AT_BOL() (YY_CURRENT_BUFFER_LVALUE->yy_at_bol)
+
+/* Begin user sect3 */
+
+#define yywrap(n) 1
+#define YY_SKIP_YYWRAP
+
+typedef unsigned char YY_CHAR;
+
+static FILE *yyin = (FILE *) 0, *yyout = (FILE *) 0;
+
+typedef int yy_state_type;
+
+static int yylineno;
+
+static int yylineno = 1;
+
+static char *yytext;
+#define yytext_ptr yytext
+
+static yy_state_type yy_get_previous_state (void );
+static yy_state_type yy_try_NUL_trans (yy_state_type current_state  );
+static int yy_get_next_buffer (void );
+static void yy_fatal_error (yyconst char msg[]  );
+
+/* Done after the current pattern has been matched and before the
+ * corresponding action - sets up yytext.
+ */
+#define YY_DO_BEFORE_ACTION \
+	(yytext_ptr) = yy_bp; \
+	yyleng = (size_t) (yy_cp - yy_bp); \
+	(yy_hold_char) = *yy_cp; \
+	*yy_cp = '\0'; \
+	(yy_c_buf_p) = yy_cp;
+
+#define YY_NUM_RULES 119
+#define YY_END_OF_BUFFER 120
+/* This struct is not used in this scanner,
+   but its presence is necessary. */
+struct yy_trans_info
+	{
+	flex_int32_t yy_verify;
+	flex_int32_t yy_nxt;
+	};
+static yyconst flex_int16_t yy_accept[399] =
+    {   0,
+        0,    0,  114,  114,    0,    0,    0,    0,  120,  118,
+      117,  117,    8,  118,  109,    5,   98,  104,  107,  105,
+      102,  106,  118,  108,    1,  118,  103,  101,   99,  100,
+      112,   92,   92,   92,   92,   92,   92,   92,   92,   92,
+       92,   92,   92,   92,   92,   92,   92,   92,   92,   92,
+      110,  111,  114,  115,    6,    7,    9,   10,  117,    4,
+       93,  113,    2,    1,    3,   94,   95,   97,   96,   92,
+       92,   92,   92,   92,   92,   44,   92,   92,   92,   92,
+       92,   92,   92,   92,   92,   92,   92,   92,   92,   92,
+       92,   92,   28,   17,   25,   92,   92,   92,   92,   92,
+
+       54,   61,   92,   14,   92,   92,   92,   92,   92,   92,
+       92,   92,   92,   92,   92,   92,   92,   92,   92,   92,
+       92,   92,  114,  115,  115,  116,    6,    7,    9,   10,
+        2,   13,   45,   92,   92,   92,   92,   92,   92,   92,
+       92,   92,   92,   92,   92,   92,   92,   92,   92,   92,
+       92,   27,   92,   92,   92,   41,   92,   92,   92,   92,
+       21,   92,   92,   92,   92,   15,   92,   92,   92,   18,
+       92,   92,   92,   92,   92,   80,   92,   92,   92,   51,
+       92,   12,   92,   36,   92,   92,   92,   92,   92,   92,
+       92,   92,   92,   92,   92,   92,   92,   92,   20,   24,
+
+       92,   92,   92,   92,   92,   92,   92,   92,   92,   92,
+       46,   92,   92,   30,   92,   87,   92,   92,   39,   92,
+       92,   92,   92,   92,   48,   92,   89,   32,   91,   92,
+       11,   64,   92,   92,   92,   42,   92,   92,   92,   92,
+       92,   92,   92,   92,   92,   92,   29,   92,   92,   92,
+       92,   92,   92,   92,   92,   92,   85,   92,   26,   92,
+       66,   92,   92,   92,   37,   92,   92,   92,   92,   92,
+       92,   92,   31,   65,   23,   92,   57,   92,   75,   92,
+       92,   92,   43,   92,   92,   92,   92,   92,   92,   92,
+       92,   90,   92,   92,   56,   92,   92,   92,   92,   92,
+
+       92,   92,   40,   33,   79,   19,   92,   83,   74,   55,
+       92,   63,   92,   52,   92,   92,   92,   47,   92,   76,
+       92,   78,   92,   92,   34,   92,   92,   92,   35,   72,
+       92,   92,   92,   92,   58,   92,   50,   49,   92,   92,
+       53,   62,   92,   92,   92,   22,   92,   92,   73,   81,
+       92,   92,   77,   92,   68,   92,   92,   92,   92,   38,
+       92,   88,   67,   92,   84,   92,   92,   92,   86,   92,
+       59,   92,   16,   92,   70,   69,   92,   92,   82,   92,
+       92,   92,   92,   92,   92,   92,   92,   92,   92,   71,
+       92,   92,   92,   92,   92,   92,   60,    0
+
+    } ;
+
+static yyconst flex_int32_t yy_ec[256] =
+    {   0,
+        1,    1,    1,    1,    1,    1,    1,    1,    2,    3,
+        1,    1,    1,    1,    1,    1,    1,    1,    1,    1,
+        1,    1,    1,    1,    1,    1,    1,    1,    1,    1,
+        1,    2,    1,    4,    1,    5,    6,    1,    7,    8,
+        9,   10,   11,   12,   13,   14,   15,   16,   16,   16,
+       16,   16,   16,   16,   16,   16,   16,   17,   18,   19,
+       20,   21,   22,    1,   23,   24,   25,   26,   27,   28,
+       29,   30,   31,   32,   33,   34,   35,   36,   37,   38,
+       39,   40,   41,   42,   43,   44,   45,   46,   47,   32,
+        1,    1,    1,    1,   48,    1,   32,   32,   32,   32,
+
+       32,   32,   32,   32,   32,   32,   32,   32,   32,   32,
+       32,   32,   32,   32,   32,   32,   32,   32,   32,   32,
+       32,   32,   49,    1,   50,    1,    1,    1,    1,    1,
+        1,    1,    1,    1,    1,    1,    1,    1,    1,    1,
+        1,    1,    1,    1,    1,    1,    1,    1,    1,    1,
+        1,    1,    1,    1,    1,    1,    1,    1,    1,    1,
+        1,    1,    1,    1,    1,    1,    1,    1,    1,    1,
+        1,    1,    1,    1,    1,    1,    1,    1,    1,    1,
+        1,    1,    1,    1,    1,    1,    1,    1,    1,    1,
+        1,    1,    1,    1,    1,    1,    1,    1,    1,    1,
+
+        1,    1,    1,    1,    1,    1,    1,    1,    1,    1,
+        1,    1,    1,    1,    1,    1,    1,    1,    1,    1,
+        1,    1,    1,    1,    1,    1,    1,    1,    1,    1,
+        1,    1,    1,    1,    1,    1,    1,    1,    1,    1,
+        1,    1,    1,    1,    1,    1,    1,    1,    1,    1,
+        1,    1,    1,    1,    1
+    } ;
+
+static yyconst flex_int32_t yy_meta[51] =
+    {   0,
+        1,    1,    1,    2,    1,    1,    3,    1,    1,    4,
+        1,    1,    1,    1,    1,    5,    1,    1,    1,    6,
+        1,    1,    5,    5,    5,    5,    5,    5,    5,    5,
+        5,    5,    5,    5,    5,    5,    5,    5,    5,    5,
+        5,    5,    5,    5,    5,    5,    5,    5,    1,    1
+    } ;
+
+static yyconst flex_int16_t yy_base[409] =
+    {   0,
+        0,    0,  437,  436,  438,  437,  439,  438,  441,  448,
+       49,   51,  448,    0,  448,  448,  448,  448,  448,  448,
+      448,  448,  426,  429,   41,  418,  448,   38,  448,  417,
+      448,   20,   33,   32,   46,   40,   44,    0,   54,   52,
+      399,   48,   60,  395,   65,   67,   81,   27,  411,   75,
+      448,  448,    0,   98,    0,  426,    0,  428,  113,    0,
+      448,  448,  415,   54,  410,  448,  448,  448,  448,    0,
+      403,   68,  399,  391,  389,    0,  402,   80,   84,  397,
+      383,   96,  381,  394,  379,  393,  387,  375,  379,  375,
+      377,  377,    0,   98,    0,  376,   97,  385,  368,  375,
+
+        0,    0,  381,  381,  364,   94,  103,  379,   98,   65,
+      381,  369,  109,  361,  377,  373,  351,   97,  372,  363,
+      115,  356,    0,  137,  138,  448,    0,  388,    0,  390,
+      377,    0,    0,  365,  360,  367,  365,  348,  346,  345,
+      350,  359,  347,  359,   95,  347,  353,  354,  336,  336,
+      123,    0,  334,  350,  351,    0,  338,  347,  344,  122,
+      124,  341,  336,  330,  340,  338,  331,  328,  336,    0,
+      326,  336,  334,  325,  315,  309,  322,  307,  327,    0,
+      313,    0,  311,    0,  325,  316,  313,  131,  309,  316,
+      323,  302,  304,  309,  309,  301,  304,  299,    0,    0,
+
+      311,  295,  305,  312,  292,  291,  305,  294,  307,  287,
+        0,  297,  279,    0,  298,    0,  295,  282,    0,  281,
+      276,  281,  280,  290,    0,  276,    0,    0,    0,  280,
+        0,    0,  276,  273,  287,    0,  272,  272,  270,  286,
+      271,  283,  280,  264,  282,  277,    0,  272,  272,  258,
+      257,  270,  256,  270,  269,  268,    0,  252,    0,  246,
+        0,  265,  249,  248,    0,  262,  252,  247,  246,  258,
+      248,  247,    0,    0,    0,  251,    0,  239,    0,  253,
+      249,  235,    0,  249,  250,  233,  238,  231,  249,  231,
+      228,    0,  229,  226,    0,  231,  243,  230,  237,  227,
+
+      235,  220,    0,    0,    0,  212,  219,    0,    0,    0,
+      216,    0,  230,    0,  231,  218,  217,    0,  213,    0,
+      216,    0,  208,  210,    0,  209,  223,  216,    0,    0,
+      219,  222,  204,  219,    0,  215,    0,    0,  199,  213,
+        0,    0,  197,  196,  201,    0,  210,  195,    0,    0,
+      201,  197,    0,  192,    0,  204,  204,  192,  202,    0,
+      179,    0,    0,  199,    0,  183,  177,  183,    0,  174,
+        0,  193,    0,  192,    0,    0,  183,  187,    0,  174,
+      174,  180,  166,  189,  181,  180,  166,  151,  118,    0,
+      130,  136,  127,  123,  119,  111,    0,  448,  167,  173,
+
+      179,  152,  181,  124,  187,  193,  199,  205
+    } ;
+
+static yyconst flex_int16_t yy_def[409] =
+    {   0,
+      398,    1,  399,  399,  400,  400,  401,  401,  398,  398,
+      398,  398,  398,  402,  398,  398,  398,  398,  398,  398,
+      398,  398,  398,  398,  398,  403,  398,  398,  398,  398,
+      398,  404,  404,  404,  404,  404,  404,  404,  404,  404,
+      404,  404,  404,  404,  404,  404,  404,  404,  404,  404,
+      398,  398,  405,  406,  407,  398,  408,  398,  398,  402,
+      398,  398,  398,  398,  403,  398,  398,  398,  398,  404,
+      404,  404,  404,  404,  404,  404,  404,  404,  404,  404,
+      404,  404,  404,  404,  404,  404,  404,  404,  404,  404,
+      404,  404,  404,  404,  404,  404,  404,  404,  404,  404,
+
+      404,  404,  404,  404,  404,  404,  404,  404,  404,  404,
+      404,  404,  404,  404,  404,  404,  404,  404,  404,  404,
+      404,  404,  405,  406,  406,  398,  407,  398,  408,  398,
+      398,  404,  404,  404,  404,  404,  404,  404,  404,  404,
+      404,  404,  404,  404,  404,  404,  404,  404,  404,  404,
+      404,  404,  404,  404,  404,  404,  404,  404,  404,  404,
+      404,  404,  404,  404,  404,  404,  404,  404,  404,  404,
+      404,  404,  404,  404,  404,  404,  404,  404,  404,  404,
+      404,  404,  404,  404,  404,  404,  404,  404,  404,  404,
+      404,  404,  404,  404,  404,  404,  404,  404,  404,  404,
+
+      404,  404,  404,  404,  404,  404,  404,  404,  404,  404,
+      404,  404,  404,  404,  404,  404,  404,  404,  404,  404,
+      404,  404,  404,  404,  404,  404,  404,  404,  404,  404,
+      404,  404,  404,  404,  404,  404,  404,  404,  404,  404,
+      404,  404,  404,  404,  404,  404,  404,  404,  404,  404,
+      404,  404,  404,  404,  404,  404,  404,  404,  404,  404,
+      404,  404,  404,  404,  404,  404,  404,  404,  404,  404,
+      404,  404,  404,  404,  404,  404,  404,  404,  404,  404,
+      404,  404,  404,  404,  404,  404,  404,  404,  404,  404,
+      404,  404,  404,  404,  404,  404,  404,  404,  404,  404,
+
+      404,  404,  404,  404,  404,  404,  404,  404,  404,  404,
+      404,  404,  404,  404,  404,  404,  404,  404,  404,  404,
+      404,  404,  404,  404,  404,  404,  404,  404,  404,  404,
+      404,  404,  404,  404,  404,  404,  404,  404,  404,  404,
+      404,  404,  404,  404,  404,  404,  404,  404,  404,  404,
+      404,  404,  404,  404,  404,  404,  404,  404,  404,  404,
+      404,  404,  404,  404,  404,  404,  404,  404,  404,  404,
+      404,  404,  404,  404,  404,  404,  404,  404,  404,  404,
+      404,  404,  404,  404,  404,  404,  404,  404,  404,  404,
+      404,  404,  404,  404,  404,  404,  404,    0,  398,  398,
+
+      398,  398,  398,  398,  398,  398,  398,  398
+    } ;
+
+static yyconst flex_int16_t yy_nxt[499] =
+    {   0,
+       10,   11,   12,   13,   14,   15,   16,   17,   18,   19,
+       20,   21,   22,   23,   24,   25,   26,   27,   28,   29,
+       30,   31,   32,   33,   34,   35,   36,   37,   38,   38,
+       39,   38,   38,   40,   41,   42,   43,   44,   38,   45,
+       46,   47,   48,   49,   50,   38,   38,   38,   51,   52,
+       59,   59,   59,   59,   63,   71,   64,   67,   68,   73,
+       72,   77,  118,   74,  119,   78,   75,   63,   79,   64,
+       88,   80,   82,   85,   81,   86,   83,   89,   96,   76,
+       90,   93,   84,   91,   99,   87,   92,  101,   97,   94,
+      100,  107,  133,  110,   95,  102,  111,  103,  179,  104,
+
+      108,  109,  105,  115,  121,  112,  180,  125,  134,  113,
+      116,  122,  126,  114,   59,   59,  139,  117,  141,  142,
+      146,  163,  140,  159,  171,  173,  143,  189,   70,  147,
+      172,  177,  183,  164,  207,  208,  148,  190,  160,  161,
+      174,  193,  178,  184,  175,  194,  398,  125,  222,  214,
+      224,  398,  126,  215,  248,  249,   60,  397,  396,  395,
+      225,  394,  393,  223,  392,  391,  250,   53,   53,   53,
+       53,   53,   53,   55,   55,   55,   55,   55,   55,   57,
+       57,   57,   57,   57,   57,   65,   65,  123,  123,  123,
+      390,  123,  123,  124,  124,  124,  124,  124,  124,  127,
+
+      127,  389,  127,  127,  127,  129,  388,  129,  129,  129,
+      129,  387,  386,  385,  384,  383,  382,  381,  380,  379,
+      378,  377,  376,  375,  374,  373,  372,  371,  370,  369,
+      368,  367,  366,  365,  364,  363,  362,  361,  360,  359,
+      358,  357,  356,  355,  354,  353,  352,  351,  350,  349,
+      348,  347,  346,  345,  344,  343,  342,  341,  340,  339,
+      338,  337,  336,  335,  334,  333,  332,  331,  330,  329,
+      328,  327,  326,  325,  324,  323,  322,  321,  320,  319,
+      318,  317,  316,  315,  314,  313,  312,  311,  310,  309,
+      308,  307,  306,  305,  304,  303,  302,  301,  300,  299,
+
+      298,  297,  296,  295,  294,  293,  292,  291,  290,  289,
+      288,  287,  286,  285,  284,  283,  282,  281,  280,  279,
+      278,  277,  276,  275,  274,  273,  272,  271,  270,  269,
+      268,  267,  266,  265,  264,  263,  262,  261,  260,  259,
+      258,  257,  256,  255,  254,  253,  252,  251,  247,  246,
+      245,  244,  243,  242,  241,  240,  239,  238,  237,  236,
+      235,  234,  233,  232,  231,  230,  229,  228,  227,  226,
+      221,  220,  219,  218,  217,  216,  213,  212,  211,  210,
+      209,  206,  205,  204,  203,  202,  201,  200,  199,  198,
+      197,  196,  131,  130,  128,  195,  192,  191,  188,  187,
+
+      186,  185,  182,  181,  176,  170,  169,  168,  167,  166,
+      165,  162,  158,  157,  156,  155,  154,  153,  152,  151,
+      150,  149,  145,  144,  138,  137,  136,  135,  132,  398,
+      131,  130,  128,  120,  106,   98,   69,   66,   62,   61,
+      398,   58,   58,   56,   56,   54,   54,    9,  398,  398,
+      398,  398,  398,  398,  398,  398,  398,  398,  398,  398,
+      398,  398,  398,  398,  398,  398,  398,  398,  398,  398,
+      398,  398,  398,  398,  398,  398,  398,  398,  398,  398,
+      398,  398,  398,  398,  398,  398,  398,  398,  398,  398,
+      398,  398,  398,  398,  398,  398,  398,  398
+
+    } ;
+
+static yyconst flex_int16_t yy_chk[499] =
+    {   0,
+        1,    1,    1,    1,    1,    1,    1,    1,    1,    1,
+        1,    1,    1,    1,    1,    1,    1,    1,    1,    1,
+        1,    1,    1,    1,    1,    1,    1,    1,    1,    1,
+        1,    1,    1,    1,    1,    1,    1,    1,    1,    1,
+        1,    1,    1,    1,    1,    1,    1,    1,    1,    1,
+       11,   11,   12,   12,   25,   32,   25,   28,   28,   33,
+       32,   34,   48,   33,   48,   34,   33,   64,   34,   64,
+       37,   34,   35,   36,   34,   36,   35,   37,   40,   33,
+       37,   39,   35,   37,   42,   36,   37,   43,   40,   39,
+       42,   45,   72,   46,   39,   43,   46,   43,  110,   43,
+
+       45,   45,   43,   47,   50,   46,  110,   54,   72,   46,
+       47,   50,   54,   46,   59,   59,   78,   47,   79,   79,
+       82,   97,   78,   94,  106,  107,   79,  118,  404,   82,
+      106,  109,  113,   97,  145,  145,   82,  118,   94,   94,
+      107,  121,  109,  113,  107,  121,  124,  125,  160,  151,
+      161,  124,  125,  151,  188,  188,  402,  396,  395,  394,
+      161,  393,  392,  160,  391,  389,  188,  399,  399,  399,
+      399,  399,  399,  400,  400,  400,  400,  400,  400,  401,
+      401,  401,  401,  401,  401,  403,  403,  405,  405,  405,
+      388,  405,  405,  406,  406,  406,  406,  406,  406,  407,
+
+      407,  387,  407,  407,  407,  408,  386,  408,  408,  408,
+      408,  385,  384,  383,  382,  381,  380,  378,  377,  374,
+      372,  370,  368,  367,  366,  364,  361,  359,  358,  357,
+      356,  354,  352,  351,  348,  347,  345,  344,  343,  340,
+      339,  336,  334,  333,  332,  331,  328,  327,  326,  324,
+      323,  321,  319,  317,  316,  315,  313,  311,  307,  306,
+      302,  301,  300,  299,  298,  297,  296,  294,  293,  291,
+      290,  289,  288,  287,  286,  285,  284,  282,  281,  280,
+      278,  276,  272,  271,  270,  269,  268,  267,  266,  264,
+      263,  262,  260,  258,  256,  255,  254,  253,  252,  251,
+
+      250,  249,  248,  246,  245,  244,  243,  242,  241,  240,
+      239,  238,  237,  235,  234,  233,  230,  226,  224,  223,
+      222,  221,  220,  218,  217,  215,  213,  212,  210,  209,
+      208,  207,  206,  205,  204,  203,  202,  201,  198,  197,
+      196,  195,  194,  193,  192,  191,  190,  189,  187,  186,
+      185,  183,  181,  179,  178,  177,  176,  175,  174,  173,
+      172,  171,  169,  168,  167,  166,  165,  164,  163,  162,
+      159,  158,  157,  155,  154,  153,  150,  149,  148,  147,
+      146,  144,  143,  142,  141,  140,  139,  138,  137,  136,
+      135,  134,  131,  130,  128,  122,  120,  119,  117,  116,
+
+      115,  114,  112,  111,  108,  105,  104,  103,  100,   99,
+       98,   96,   92,   91,   90,   89,   88,   87,   86,   85,
+       84,   83,   81,   80,   77,   75,   74,   73,   71,   65,
+       63,   58,   56,   49,   44,   41,   30,   26,   24,   23,
+        9,    8,    7,    6,    5,    4,    3,  398,  398,  398,
+      398,  398,  398,  398,  398,  398,  398,  398,  398,  398,
+      398,  398,  398,  398,  398,  398,  398,  398,  398,  398,
+      398,  398,  398,  398,  398,  398,  398,  398,  398,  398,
+      398,  398,  398,  398,  398,  398,  398,  398,  398,  398,
+      398,  398,  398,  398,  398,  398,  398,  398
+
+    } ;
+
+static yy_state_type yy_last_accepting_state;
+static char *yy_last_accepting_cpos;
+
+static int yy_flex_debug;
+static int yy_flex_debug = 0;
+
+/* The intent behind this definition is that it'll catch
+ * any uses of REJECT which flex missed.
+ */
+#define REJECT reject_used_but_not_detected
+#define yymore() yymore_used_but_not_detected
+#define YY_MORE_ADJ 0
+#define YY_RESTORE_YY_MORE_OFFSET
+static char *yytext;
+#line 1 "pars0lex.l"
+/******************************************************
+SQL parser lexical analyzer: input file for the GNU Flex lexer generator
+
+(c) 1997 Innobase Oy
+
+Created 12/14/1997 Heikki Tuuri
+Published under the GPL version 2
+
+The InnoDB parser is frozen because MySQL takes care of SQL parsing.
+Therefore we normally keep the InnoDB parser C files as they are, and do
+not automatically generate them from pars0grm.y and pars0lex.l.
+
+How to make the InnoDB parser and lexer C files:
+
+1. Run ./make_flex.sh to generate lexer files.
+
+2. Run ./make_bison.sh to generate parser files.
+
+These instructions seem to work at least with bison-1.875d and flex-2.5.31 on
+Linux.
+*******************************************************/
+#define YY_NO_INPUT 1
+#define YY_NO_UNISTD_H 1
+#line 38 "pars0lex.l"
+#define YYSTYPE que_node_t*
+
+#include "univ.i"
+#include "pars0pars.h"
+#include "pars0grm.h"
+#include "pars0sym.h"
+#include "mem0mem.h"
+#include "os0proc.h"
+
+#define malloc(A)	ut_malloc(A)
+#define free(A)		ut_free(A)
+#define realloc(P, A)	ut_realloc(P, A)
+#define exit(A) 	ut_error
+
+#define YY_INPUT(buf, result, max_size) pars_get_lex_chars(buf, &result, max_size)
+
+/* String buffer for removing quotes */
+static ulint	stringbuf_len_alloc = 0; /* Allocated length */
+static ulint	stringbuf_len = 0; /* Current length */
+static char*	stringbuf; /* Start of buffer */
+/* Appends a string to the buffer. */
+static
+void
+string_append(
+/*==========*/
+	const char*	str,	/* in: string to be appended */
+	ulint		len)	/* in: length of the string */
+{
+	if (stringbuf == NULL) {
+		stringbuf = malloc(1);
+		stringbuf_len_alloc = 1;
+	}
+
+	if (stringbuf_len + len > stringbuf_len_alloc) {
+		while (stringbuf_len + len > stringbuf_len_alloc) {
+			stringbuf_len_alloc <<= 1;
+		}
+		stringbuf = realloc(stringbuf, stringbuf_len_alloc);
+	}
+
+	memcpy(stringbuf + stringbuf_len, str, len);
+	stringbuf_len += len;
+}
+
+
+
+
+#line 759 "lexyy.c"
+
+#define INITIAL 0
+#define comment 1
+#define quoted 2
+#define id 3
+
+#ifndef YY_NO_UNISTD_H
+/* Special case for "unistd.h", since it is non-ANSI. We include it way
+ * down here because we want the user's section 1 to have been scanned first.
+ * The user has a chance to override it with an option.
+ */
+#include <unistd.h>
+#endif
+
+#ifndef YY_EXTRA_TYPE
+#define YY_EXTRA_TYPE void *
+#endif
+
+/* Macros after this point can all be overridden by user definitions in
+ * section 1.
+ */
+
+#ifndef YY_SKIP_YYWRAP
+#ifdef __cplusplus
+extern "C" int yywrap (void );
+#else
+extern int yywrap (void );
+#endif
+#endif
+
+#ifndef yytext_ptr
+static void yy_flex_strncpy (char *,yyconst char *,int );
+#endif
+
+#ifdef YY_NEED_STRLEN
+static int yy_flex_strlen (yyconst char * );
+#endif
+
+#ifndef YY_NO_INPUT
+
+#ifdef __cplusplus
+static int yyinput (void );
+#else
+static int input (void );
+#endif
+
+#endif
+
+/* Amount of stuff to slurp up with each read. */
+#ifndef YY_READ_BUF_SIZE
+#define YY_READ_BUF_SIZE 8192
+#endif
+
+/* Copy whatever the last rule matched to the standard output. */
+#ifndef ECHO
+/* This used to be an fputs(), but since the string might contain NUL's,
+ * we now use fwrite().
+ */
+#define ECHO (void) fwrite( yytext, yyleng, 1, yyout )
+#endif
+
+/* Gets input and stuffs it into "buf".  number of characters read, or YY_NULL,
+ * is returned in "result".
+ */
+#ifndef YY_INPUT
+#define YY_INPUT(buf,result,max_size) \
+	if ( YY_CURRENT_BUFFER_LVALUE->yy_is_interactive ) \
+		{ \
+		int c = '*'; \
+		size_t n; \
+		for ( n = 0; n < max_size && \
+			     (c = getc( yyin )) != EOF && c != '\n'; ++n ) \
+			buf[n] = (char) c; \
+		if ( c == '\n' ) \
+			buf[n++] = (char) c; \
+		if ( c == EOF && ferror( yyin ) ) \
+			YY_FATAL_ERROR( "input in flex scanner failed" ); \
+		result = n; \
+		} \
+	else \
+		{ \
+		errno=0; \
+		while ( (result = fread(buf, 1, max_size, yyin))==0 && ferror(yyin)) \
+			{ \
+			if( errno != EINTR) \
+				{ \
+				YY_FATAL_ERROR( "input in flex scanner failed" ); \
+				break; \
+				} \
+			errno=0; \
+			clearerr(yyin); \
+			} \
+		}\
+\
+
+#endif
+
+/* No semi-colon after return; correct usage is to write "yyterminate();" -
+ * we don't want an extra ';' after the "return" because that will cause
+ * some compilers to complain about unreachable statements.
+ */
+#ifndef yyterminate
+#define yyterminate() return YY_NULL
+#endif
+
+/* Number of entries by which start-condition stack grows. */
+#ifndef YY_START_STACK_INCR
+#define YY_START_STACK_INCR 25
+#endif
+
+/* Report a fatal error. */
+#ifndef YY_FATAL_ERROR
+#define YY_FATAL_ERROR(msg) yy_fatal_error( msg )
+#endif
+
+/* end tables serialization structures and prototypes */
+
+/* Default declaration of generated scanner - a define so the user can
+ * easily add parameters.
+ */
+#ifndef YY_DECL
+#define YY_DECL_IS_OURS 1
+
+UNIV_INTERN int yylex (void);
+
+#define YY_DECL UNIV_INTERN int yylex (void)
+#endif /* !YY_DECL */
+
+/* Code executed at the beginning of each rule, after yytext and yyleng
+ * have been set up.
+ */
+#ifndef YY_USER_ACTION
+#define YY_USER_ACTION
+#endif
+
+/* Code executed at the end of each rule. */
+#ifndef YY_BREAK
+#define YY_BREAK break;
+#endif
+
+#define YY_RULE_SETUP \
+	YY_USER_ACTION
+
+/** The main scanner function which does all the work.
+ */
+YY_DECL
+{
+	register yy_state_type yy_current_state;
+	register char *yy_cp, *yy_bp;
+	register int yy_act;
+    
+#line 92 "pars0lex.l"
+
+
+#line 914 "lexyy.c"
+
+	if ( (yy_init) )
+		{
+		(yy_init) = 0;
+
+#ifdef YY_USER_INIT
+		YY_USER_INIT;
+#endif
+
+		if ( ! (yy_start) )
+			(yy_start) = 1;	/* first start state */
+
+		if ( ! yyin )
+			yyin = stdin;
+
+		if ( ! yyout )
+			yyout = stdout;
+
+		if ( ! YY_CURRENT_BUFFER ) {
+			yyensure_buffer_stack ();
+			YY_CURRENT_BUFFER_LVALUE =
+				yy_create_buffer(yyin,YY_BUF_SIZE );
+		}
+
+		yy_load_buffer_state( );
+		}
+
+	while ( 1 )		/* loops until end-of-file is reached */
+		{
+		yy_cp = (yy_c_buf_p);
+
+		/* Support of yytext. */
+		*yy_cp = (yy_hold_char);
+
+		/* yy_bp points to the position in yy_ch_buf of the start of
+		 * the current run.
+		 */
+		yy_bp = yy_cp;
+
+		yy_current_state = (yy_start);
+yy_match:
+		do
+			{
+			register YY_CHAR yy_c = yy_ec[YY_SC_TO_UI(*yy_cp)];
+			if ( yy_accept[yy_current_state] )
+				{
+				(yy_last_accepting_state) = yy_current_state;
+				(yy_last_accepting_cpos) = yy_cp;
+				}
+			while ( yy_chk[yy_base[yy_current_state] + yy_c] != yy_current_state )
+				{
+				yy_current_state = (int) yy_def[yy_current_state];
+				if ( yy_current_state >= 399 )
+					yy_c = yy_meta[(unsigned int) yy_c];
+				}
+			yy_current_state = yy_nxt[yy_base[yy_current_state] + (unsigned int) yy_c];
+			++yy_cp;
+			}
+		while ( yy_current_state != 398 );
+		yy_cp = (yy_last_accepting_cpos);
+		yy_current_state = (yy_last_accepting_state);
+
+yy_find_action:
+		yy_act = yy_accept[yy_current_state];
+
+		YY_DO_BEFORE_ACTION;
+
+do_action:	/* This label is used only to access EOF actions. */
+
+		switch ( yy_act )
+	{ /* beginning of action switch */
+			case 0: /* must back up */
+			/* undo the effects of YY_DO_BEFORE_ACTION */
+			*yy_cp = (yy_hold_char);
+			yy_cp = (yy_last_accepting_cpos);
+			yy_current_state = (yy_last_accepting_state);
+			goto yy_find_action;
+
+case 1:
+YY_RULE_SETUP
+#line 94 "pars0lex.l"
+{
+			yylval = sym_tab_add_int_lit(pars_sym_tab_global,
+								atoi(yytext));
+			return(PARS_INT_LIT);
+}
+	YY_BREAK
+case 2:
+YY_RULE_SETUP
+#line 100 "pars0lex.l"
+{
+			ut_error;	/* not implemented */
+
+			return(PARS_FLOAT_LIT);
+}
+	YY_BREAK
+case 3:
+YY_RULE_SETUP
+#line 106 "pars0lex.l"
+{
+			ulint	type;
+
+			yylval = sym_tab_add_bound_lit(pars_sym_tab_global,
+				yytext + 1, &type);
+
+			return((int) type);
+}
+	YY_BREAK
+case 4:
+YY_RULE_SETUP
+#line 115 "pars0lex.l"
+{
+			yylval = sym_tab_add_bound_id(pars_sym_tab_global,
+				yytext + 1);
+
+			return(PARS_ID_TOKEN);
+}
+	YY_BREAK
+case 5:
+YY_RULE_SETUP
+#line 122 "pars0lex.l"
+{
+/* Quoted character string literals are handled in an explicit
+start state 'quoted'.  This state is entered and the buffer for
+the scanned string is emptied upon encountering a starting quote.
+
+In the state 'quoted', only two actions are possible (defined below). */
+			BEGIN(quoted);
+			stringbuf_len = 0;
+}
+	YY_BREAK
+case 6:
+/* rule 6 can match eol */
+YY_RULE_SETUP
+#line 131 "pars0lex.l"
+{
+			/* Got a sequence of characters other than "'":
+			append to string buffer */
+			string_append(yytext, yyleng);
+}
+	YY_BREAK
+case 7:
+YY_RULE_SETUP
+#line 136 "pars0lex.l"
+{
+			/* Got a sequence of "'" characters:
+			append half of them to string buffer,
+			as "''" represents a single "'".
+			We apply truncating division,
+			so that "'''" will result in "'". */
+
+			string_append(yytext, yyleng / 2);
+
+			/* If we got an odd number of quotes, then the
+			last quote we got is the terminating quote.
+			At the end of the string, we return to the
+			initial start state and report the scanned
+			string literal. */
+
+			if (yyleng % 2) {
+				BEGIN(INITIAL);
+				yylval = sym_tab_add_str_lit(
+					pars_sym_tab_global,
+					(byte*) stringbuf, stringbuf_len);
+				return(PARS_STR_LIT);
+			}
+}
+	YY_BREAK
+case 8:
+YY_RULE_SETUP
+#line 160 "pars0lex.l"
+{
+/* Quoted identifiers are handled in an explicit start state 'id'.
+This state is entered and the buffer for the scanned string is emptied
+upon encountering a starting quote.
+
+In the state 'id', only two actions are possible (defined below). */
+			BEGIN(id);
+			stringbuf_len = 0;
+}
+	YY_BREAK
+case 9:
+/* rule 9 can match eol */
+YY_RULE_SETUP
+#line 169 "pars0lex.l"
+{
+			/* Got a sequence of characters other than '"':
+			append to string buffer */
+			string_append(yytext, yyleng);
+}
+	YY_BREAK
+case 10:
+YY_RULE_SETUP
+#line 174 "pars0lex.l"
+{
+			/* Got a sequence of '"' characters:
+			append half of them to string buffer,
+			as '""' represents a single '"'.
+			We apply truncating division,
+			so that '"""' will result in '"'. */
+
+			string_append(yytext, yyleng / 2);
+
+			/* If we got an odd number of quotes, then the
+			last quote we got is the terminating quote.
+			At the end of the string, we return to the
+			initial start state and report the scanned
+			identifier. */
+
+			if (yyleng % 2) {
+				BEGIN(INITIAL);
+				yylval = sym_tab_add_id(
+					pars_sym_tab_global,
+					(byte*) stringbuf, stringbuf_len);
+
+				return(PARS_ID_TOKEN);
+			}
+}
+	YY_BREAK
+case 11:
+YY_RULE_SETUP
+#line 199 "pars0lex.l"
+{
+			yylval = sym_tab_add_null_lit(pars_sym_tab_global);
+
+			return(PARS_NULL_LIT);
+}
+	YY_BREAK
+case 12:
+YY_RULE_SETUP
+#line 205 "pars0lex.l"
+{
+			/* Implicit cursor name */
+			yylval = sym_tab_add_str_lit(pars_sym_tab_global,
+							(byte*) yytext, yyleng);
+			return(PARS_SQL_TOKEN);
+}
+	YY_BREAK
+case 13:
+YY_RULE_SETUP
+#line 212 "pars0lex.l"
+{
+			return(PARS_AND_TOKEN);
+}
+	YY_BREAK
+case 14:
+YY_RULE_SETUP
+#line 216 "pars0lex.l"
+{
+			return(PARS_OR_TOKEN);
+}
+	YY_BREAK
+case 15:
+YY_RULE_SETUP
+#line 220 "pars0lex.l"
+{
+			return(PARS_NOT_TOKEN);
+}
+	YY_BREAK
+case 16:
+YY_RULE_SETUP
+#line 224 "pars0lex.l"
+{
+			return(PARS_PROCEDURE_TOKEN);
+}
+	YY_BREAK
+case 17:
+YY_RULE_SETUP
+#line 228 "pars0lex.l"
+{
+			return(PARS_IN_TOKEN);
+}
+	YY_BREAK
+case 18:
+YY_RULE_SETUP
+#line 232 "pars0lex.l"
+{
+			return(PARS_OUT_TOKEN);
+}
+	YY_BREAK
+case 19:
+YY_RULE_SETUP
+#line 236 "pars0lex.l"
+{
+	 		return(PARS_BINARY_TOKEN);
+}
+	YY_BREAK
+case 20:
+YY_RULE_SETUP
+#line 240 "pars0lex.l"
+{
+	 		return(PARS_BLOB_TOKEN);
+}
+	YY_BREAK
+case 21:
+YY_RULE_SETUP
+#line 244 "pars0lex.l"
+{
+	 		return(PARS_INT_TOKEN);
+}
+	YY_BREAK
+case 22:
+YY_RULE_SETUP
+#line 248 "pars0lex.l"
+{
+	 		return(PARS_INT_TOKEN);
+}
+	YY_BREAK
+case 23:
+YY_RULE_SETUP
+#line 252 "pars0lex.l"
+{
+	 		return(PARS_FLOAT_TOKEN);
+}
+	YY_BREAK
+case 24:
+YY_RULE_SETUP
+#line 256 "pars0lex.l"
+{
+	 		return(PARS_CHAR_TOKEN);
+}
+	YY_BREAK
+case 25:
+YY_RULE_SETUP
+#line 260 "pars0lex.l"
+{
+			return(PARS_IS_TOKEN);
+}
+	YY_BREAK
+case 26:
+YY_RULE_SETUP
+#line 264 "pars0lex.l"
+{
+			return(PARS_BEGIN_TOKEN);
+}
+	YY_BREAK
+case 27:
+YY_RULE_SETUP
+#line 268 "pars0lex.l"
+{
+			return(PARS_END_TOKEN);
+}
+	YY_BREAK
+case 28:
+YY_RULE_SETUP
+#line 272 "pars0lex.l"
+{
+			return(PARS_IF_TOKEN);
+}
+	YY_BREAK
+case 29:
+YY_RULE_SETUP
+#line 276 "pars0lex.l"
+{
+			return(PARS_THEN_TOKEN);
+}
+	YY_BREAK
+case 30:
+YY_RULE_SETUP
+#line 280 "pars0lex.l"
+{
+			return(PARS_ELSE_TOKEN);
+}
+	YY_BREAK
+case 31:
+YY_RULE_SETUP
+#line 284 "pars0lex.l"
+{
+			return(PARS_ELSIF_TOKEN);
+}
+	YY_BREAK
+case 32:
+YY_RULE_SETUP
+#line 288 "pars0lex.l"
+{
+			return(PARS_LOOP_TOKEN);
+}
+	YY_BREAK
+case 33:
+YY_RULE_SETUP
+#line 292 "pars0lex.l"
+{
+			return(PARS_WHILE_TOKEN);
+}
+	YY_BREAK
+case 34:
+YY_RULE_SETUP
+#line 296 "pars0lex.l"
+{
+			return(PARS_RETURN_TOKEN);
+}
+	YY_BREAK
+case 35:
+YY_RULE_SETUP
+#line 300 "pars0lex.l"
+{
+			return(PARS_SELECT_TOKEN);
+}
+	YY_BREAK
+case 36:
+YY_RULE_SETUP
+#line 304 "pars0lex.l"
+{
+			return(PARS_SUM_TOKEN);
+}
+	YY_BREAK
+case 37:
+YY_RULE_SETUP
+#line 308 "pars0lex.l"
+{
+			return(PARS_COUNT_TOKEN);
+}
+	YY_BREAK
+case 38:
+YY_RULE_SETUP
+#line 312 "pars0lex.l"
+{
+			return(PARS_DISTINCT_TOKEN);
+}
+	YY_BREAK
+case 39:
+YY_RULE_SETUP
+#line 316 "pars0lex.l"
+{
+			return(PARS_FROM_TOKEN);
+}
+	YY_BREAK
+case 40:
+YY_RULE_SETUP
+#line 320 "pars0lex.l"
+{
+			return(PARS_WHERE_TOKEN);
+}
+	YY_BREAK
+case 41:
+YY_RULE_SETUP
+#line 324 "pars0lex.l"
+{
+			return(PARS_FOR_TOKEN);
+}
+	YY_BREAK
+case 42:
+YY_RULE_SETUP
+#line 328 "pars0lex.l"
+{
+			return(PARS_READ_TOKEN);
+}
+	YY_BREAK
+case 43:
+YY_RULE_SETUP
+#line 332 "pars0lex.l"
+{
+			return(PARS_ORDER_TOKEN);
+}
+	YY_BREAK
+case 44:
+YY_RULE_SETUP
+#line 336 "pars0lex.l"
+{
+			return(PARS_BY_TOKEN);
+}
+	YY_BREAK
+case 45:
+YY_RULE_SETUP
+#line 340 "pars0lex.l"
+{
+			return(PARS_ASC_TOKEN);
+}
+	YY_BREAK
+case 46:
+YY_RULE_SETUP
+#line 344 "pars0lex.l"
+{
+			return(PARS_DESC_TOKEN);
+}
+	YY_BREAK
+case 47:
+YY_RULE_SETUP
+#line 348 "pars0lex.l"
+{
+			return(PARS_INSERT_TOKEN);
+}
+	YY_BREAK
+case 48:
+YY_RULE_SETUP
+#line 352 "pars0lex.l"
+{
+			return(PARS_INTO_TOKEN);
+}
+	YY_BREAK
+case 49:
+YY_RULE_SETUP
+#line 356 "pars0lex.l"
+{
+			return(PARS_VALUES_TOKEN);
+}
+	YY_BREAK
+case 50:
+YY_RULE_SETUP
+#line 360 "pars0lex.l"
+{
+			return(PARS_UPDATE_TOKEN);
+}
+	YY_BREAK
+case 51:
+YY_RULE_SETUP
+#line 364 "pars0lex.l"
+{
+			return(PARS_SET_TOKEN);
+}
+	YY_BREAK
+case 52:
+YY_RULE_SETUP
+#line 368 "pars0lex.l"
+{
+			return(PARS_DELETE_TOKEN);
+}
+	YY_BREAK
+case 53:
+YY_RULE_SETUP
+#line 372 "pars0lex.l"
+{
+			return(PARS_CURRENT_TOKEN);
+}
+	YY_BREAK
+case 54:
+YY_RULE_SETUP
+#line 376 "pars0lex.l"
+{
+			return(PARS_OF_TOKEN);
+}
+	YY_BREAK
+case 55:
+YY_RULE_SETUP
+#line 380 "pars0lex.l"
+{
+			return(PARS_CREATE_TOKEN);
+}
+	YY_BREAK
+case 56:
+YY_RULE_SETUP
+#line 384 "pars0lex.l"
+{
+			return(PARS_TABLE_TOKEN);
+}
+	YY_BREAK
+case 57:
+YY_RULE_SETUP
+#line 388 "pars0lex.l"
+{
+	 		return(PARS_INDEX_TOKEN);
+}
+	YY_BREAK
+case 58:
+YY_RULE_SETUP
+#line 392 "pars0lex.l"
+{
+	 		return(PARS_UNIQUE_TOKEN);
+}
+	YY_BREAK
+case 59:
+YY_RULE_SETUP
+#line 396 "pars0lex.l"
+{
+	 		return(PARS_CLUSTERED_TOKEN);
+}
+	YY_BREAK
+case 60:
+YY_RULE_SETUP
+#line 400 "pars0lex.l"
+{
+			return(PARS_DOES_NOT_FIT_IN_MEM_TOKEN);
+}
+	YY_BREAK
+case 61:
+YY_RULE_SETUP
+#line 404 "pars0lex.l"
+{
+	 		return(PARS_ON_TOKEN);
+}
+	YY_BREAK
+case 62:
+YY_RULE_SETUP
+#line 408 "pars0lex.l"
+{
+			return(PARS_DECLARE_TOKEN);
+}
+	YY_BREAK
+case 63:
+YY_RULE_SETUP
+#line 412 "pars0lex.l"
+{
+			return(PARS_CURSOR_TOKEN);
+}
+	YY_BREAK
+case 64:
+YY_RULE_SETUP
+#line 416 "pars0lex.l"
+{
+			return(PARS_OPEN_TOKEN);
+}
+	YY_BREAK
+case 65:
+YY_RULE_SETUP
+#line 420 "pars0lex.l"
+{
+			return(PARS_FETCH_TOKEN);
+}
+	YY_BREAK
+case 66:
+YY_RULE_SETUP
+#line 424 "pars0lex.l"
+{
+			return(PARS_CLOSE_TOKEN);
+}
+	YY_BREAK
+case 67:
+YY_RULE_SETUP
+#line 428 "pars0lex.l"
+{
+			return(PARS_NOTFOUND_TOKEN);
+}
+	YY_BREAK
+case 68:
+YY_RULE_SETUP
+#line 432 "pars0lex.l"
+{
+			return(PARS_TO_CHAR_TOKEN);
+}
+	YY_BREAK
+case 69:
+YY_RULE_SETUP
+#line 436 "pars0lex.l"
+{
+			return(PARS_TO_NUMBER_TOKEN);
+}
+	YY_BREAK
+case 70:
+YY_RULE_SETUP
+#line 440 "pars0lex.l"
+{
+			return(PARS_TO_BINARY_TOKEN);
+}
+	YY_BREAK
+case 71:
+YY_RULE_SETUP
+#line 444 "pars0lex.l"
+{
+			return(PARS_BINARY_TO_NUMBER_TOKEN);
+}
+	YY_BREAK
+case 72:
+YY_RULE_SETUP
+#line 448 "pars0lex.l"
+{
+			return(PARS_SUBSTR_TOKEN);
+}
+	YY_BREAK
+case 73:
+YY_RULE_SETUP
+#line 452 "pars0lex.l"
+{
+			return(PARS_REPLSTR_TOKEN);
+}
+	YY_BREAK
+case 74:
+YY_RULE_SETUP
+#line 456 "pars0lex.l"
+{
+			return(PARS_CONCAT_TOKEN);
+}
+	YY_BREAK
+case 75:
+YY_RULE_SETUP
+#line 460 "pars0lex.l"
+{
+			return(PARS_INSTR_TOKEN);
+}
+	YY_BREAK
+case 76:
+YY_RULE_SETUP
+#line 464 "pars0lex.l"
+{
+			return(PARS_LENGTH_TOKEN);
+}
+	YY_BREAK
+case 77:
+YY_RULE_SETUP
+#line 468 "pars0lex.l"
+{
+			return(PARS_SYSDATE_TOKEN);
+}
+	YY_BREAK
+case 78:
+YY_RULE_SETUP
+#line 472 "pars0lex.l"
+{
+			return(PARS_PRINTF_TOKEN);
+}
+	YY_BREAK
+case 79:
+YY_RULE_SETUP
+#line 476 "pars0lex.l"
+{
+			return(PARS_ASSERT_TOKEN);
+}
+	YY_BREAK
+case 80:
+YY_RULE_SETUP
+#line 480 "pars0lex.l"
+{
+			return(PARS_RND_TOKEN);
+}
+	YY_BREAK
+case 81:
+YY_RULE_SETUP
+#line 484 "pars0lex.l"
+{
+			return(PARS_RND_STR_TOKEN);
+}
+	YY_BREAK
+case 82:
+YY_RULE_SETUP
+#line 488 "pars0lex.l"
+{
+			return(PARS_ROW_PRINTF_TOKEN);
+}
+	YY_BREAK
+case 83:
+YY_RULE_SETUP
+#line 492 "pars0lex.l"
+{
+			return(PARS_COMMIT_TOKEN);
+}
+	YY_BREAK
+case 84:
+YY_RULE_SETUP
+#line 496 "pars0lex.l"
+{
+			return(PARS_ROLLBACK_TOKEN);
+}
+	YY_BREAK
+case 85:
+YY_RULE_SETUP
+#line 500 "pars0lex.l"
+{
+			return(PARS_WORK_TOKEN);
+}
+	YY_BREAK
+case 86:
+YY_RULE_SETUP
+#line 504 "pars0lex.l"
+{
+			return(PARS_UNSIGNED_TOKEN);
+}
+	YY_BREAK
+case 87:
+YY_RULE_SETUP
+#line 508 "pars0lex.l"
+{
+			return(PARS_EXIT_TOKEN);
+}
+	YY_BREAK
+case 88:
+YY_RULE_SETUP
+#line 512 "pars0lex.l"
+{
+			return(PARS_FUNCTION_TOKEN);
+}
+	YY_BREAK
+case 89:
+YY_RULE_SETUP
+#line 516 "pars0lex.l"
+{
+			return(PARS_LOCK_TOKEN);
+}
+	YY_BREAK
+case 90:
+YY_RULE_SETUP
+#line 520 "pars0lex.l"
+{
+			return(PARS_SHARE_TOKEN);
+}
+	YY_BREAK
+case 91:
+YY_RULE_SETUP
+#line 524 "pars0lex.l"
+{
+			return(PARS_MODE_TOKEN);
+}
+	YY_BREAK
+case 92:
+YY_RULE_SETUP
+#line 528 "pars0lex.l"
+{
+			yylval = sym_tab_add_id(pars_sym_tab_global,
+							(byte*)yytext,
+							ut_strlen(yytext));
+			return(PARS_ID_TOKEN);
+}
+	YY_BREAK
+case 93:
+YY_RULE_SETUP
+#line 535 "pars0lex.l"
+{
+			return(PARS_DDOT_TOKEN);
+}
+	YY_BREAK
+case 94:
+YY_RULE_SETUP
+#line 539 "pars0lex.l"
+{
+			return(PARS_ASSIGN_TOKEN);
+}
+	YY_BREAK
+case 95:
+YY_RULE_SETUP
+#line 543 "pars0lex.l"
+{
+			return(PARS_LE_TOKEN);
+}
+	YY_BREAK
+case 96:
+YY_RULE_SETUP
+#line 547 "pars0lex.l"
+{
+			return(PARS_GE_TOKEN);
+}
+	YY_BREAK
+case 97:
+YY_RULE_SETUP
+#line 551 "pars0lex.l"
+{
+			return(PARS_NE_TOKEN);
+}
+	YY_BREAK
+case 98:
+YY_RULE_SETUP
+#line 555 "pars0lex.l"
+{
+
+			return((int)(*yytext));
+}
+	YY_BREAK
+case 99:
+YY_RULE_SETUP
+#line 560 "pars0lex.l"
+{
+
+			return((int)(*yytext));
+}
+	YY_BREAK
+case 100:
+YY_RULE_SETUP
+#line 565 "pars0lex.l"
+{
+
+			return((int)(*yytext));
+}
+	YY_BREAK
+case 101:
+YY_RULE_SETUP
+#line 570 "pars0lex.l"
+{
+
+			return((int)(*yytext));
+}
+	YY_BREAK
+case 102:
+YY_RULE_SETUP
+#line 575 "pars0lex.l"
+{
+
+			return((int)(*yytext));
+}
+	YY_BREAK
+case 103:
+YY_RULE_SETUP
+#line 580 "pars0lex.l"
+{
+
+			return((int)(*yytext));
+}
+	YY_BREAK
+case 104:
+YY_RULE_SETUP
+#line 585 "pars0lex.l"
+{
+
+			return((int)(*yytext));
+}
+	YY_BREAK
+case 105:
+YY_RULE_SETUP
+#line 590 "pars0lex.l"
+{
+
+			return((int)(*yytext));
+}
+	YY_BREAK
+case 106:
+YY_RULE_SETUP
+#line 595 "pars0lex.l"
+{
+
+			return((int)(*yytext));
+}
+	YY_BREAK
+case 107:
+YY_RULE_SETUP
+#line 600 "pars0lex.l"
+{
+
+			return((int)(*yytext));
+}
+	YY_BREAK
+case 108:
+YY_RULE_SETUP
+#line 605 "pars0lex.l"
+{
+
+			return((int)(*yytext));
+}
+	YY_BREAK
+case 109:
+YY_RULE_SETUP
+#line 610 "pars0lex.l"
+{
+
+			return((int)(*yytext));
+}
+	YY_BREAK
+case 110:
+YY_RULE_SETUP
+#line 615 "pars0lex.l"
+{
+
+			return((int)(*yytext));
+}
+	YY_BREAK
+case 111:
+YY_RULE_SETUP
+#line 620 "pars0lex.l"
+{
+
+			return((int)(*yytext));
+}
+	YY_BREAK
+case 112:
+YY_RULE_SETUP
+#line 625 "pars0lex.l"
+{
+
+			return((int)(*yytext));
+}
+	YY_BREAK
+case 113:
+YY_RULE_SETUP
+#line 630 "pars0lex.l"
+BEGIN(comment); /* eat up comment */
+	YY_BREAK
+case 114:
+/* rule 114 can match eol */
+YY_RULE_SETUP
+#line 632 "pars0lex.l"
+
+	YY_BREAK
+case 115:
+/* rule 115 can match eol */
+YY_RULE_SETUP
+#line 633 "pars0lex.l"
+
+	YY_BREAK
+case 116:
+YY_RULE_SETUP
+#line 634 "pars0lex.l"
+BEGIN(INITIAL);
+	YY_BREAK
+case 117:
+/* rule 117 can match eol */
+YY_RULE_SETUP
+#line 636 "pars0lex.l"
+/* eat up whitespace */
+	YY_BREAK
+case 118:
+YY_RULE_SETUP
+#line 639 "pars0lex.l"
+{
+			fprintf(stderr,"Unrecognized character: %02x\n",
+				*yytext);
+
+			ut_error;
+
+			return(0);
+}
+	YY_BREAK
+case 119:
+YY_RULE_SETUP
+#line 648 "pars0lex.l"
+YY_FATAL_ERROR( "flex scanner jammed" );
+	YY_BREAK
+#line 1916 "lexyy.c"
+case YY_STATE_EOF(INITIAL):
+case YY_STATE_EOF(comment):
+case YY_STATE_EOF(quoted):
+case YY_STATE_EOF(id):
+	yyterminate();
+
+	case YY_END_OF_BUFFER:
+		{
+		/* Amount of text matched not including the EOB char. */
+		int yy_amount_of_matched_text = (int) (yy_cp - (yytext_ptr)) - 1;
+
+		/* Undo the effects of YY_DO_BEFORE_ACTION. */
+		*yy_cp = (yy_hold_char);
+		YY_RESTORE_YY_MORE_OFFSET
+
+		if ( YY_CURRENT_BUFFER_LVALUE->yy_buffer_status == YY_BUFFER_NEW )
+			{
+			/* We're scanning a new file or input source.  It's
+			 * possible that this happened because the user
+			 * just pointed yyin at a new source and called
+			 * yylex().  If so, then we have to assure
+			 * consistency between YY_CURRENT_BUFFER and our
+			 * globals.  Here is the right place to do so, because
+			 * this is the first action (other than possibly a
+			 * back-up) that will match for the new input source.
+			 */
+			(yy_n_chars) = YY_CURRENT_BUFFER_LVALUE->yy_n_chars;
+			YY_CURRENT_BUFFER_LVALUE->yy_input_file = yyin;
+			YY_CURRENT_BUFFER_LVALUE->yy_buffer_status = YY_BUFFER_NORMAL;
+			}
+
+		/* Note that here we test for yy_c_buf_p "<=" to the position
+		 * of the first EOB in the buffer, since yy_c_buf_p will
+		 * already have been incremented past the NUL character
+		 * (since all states make transitions on EOB to the
+		 * end-of-buffer state).  Contrast this with the test
+		 * in input().
+		 */
+		if ( (yy_c_buf_p) <= &YY_CURRENT_BUFFER_LVALUE->yy_ch_buf[(yy_n_chars)] )
+			{ /* This was really a NUL. */
+			yy_state_type yy_next_state;
+
+			(yy_c_buf_p) = (yytext_ptr) + yy_amount_of_matched_text;
+
+			yy_current_state = yy_get_previous_state(  );
+
+			/* Okay, we're now positioned to make the NUL
+			 * transition.  We couldn't have
+			 * yy_get_previous_state() go ahead and do it
+			 * for us because it doesn't know how to deal
+			 * with the possibility of jamming (and we don't
+			 * want to build jamming into it because then it
+			 * will run more slowly).
+			 */
+
+			yy_next_state = yy_try_NUL_trans( yy_current_state );
+
+			yy_bp = (yytext_ptr) + YY_MORE_ADJ;
+
+			if ( yy_next_state )
+				{
+				/* Consume the NUL. */
+				yy_cp = ++(yy_c_buf_p);
+				yy_current_state = yy_next_state;
+				goto yy_match;
+				}
+
+			else
+				{
+				yy_cp = (yy_last_accepting_cpos);
+				yy_current_state = (yy_last_accepting_state);
+				goto yy_find_action;
+				}
+			}
+
+		else switch ( yy_get_next_buffer(  ) )
+			{
+			case EOB_ACT_END_OF_FILE:
+				{
+				(yy_did_buffer_switch_on_eof) = 0;
+
+				if ( yywrap( ) )
+					{
+					/* Note: because we've taken care in
+					 * yy_get_next_buffer() to have set up
+					 * yytext, we can now set up
+					 * yy_c_buf_p so that if some total
+					 * hoser (like flex itself) wants to
+					 * call the scanner after we return the
+					 * YY_NULL, it'll still work - another
+					 * YY_NULL will get returned.
+					 */
+					(yy_c_buf_p) = (yytext_ptr) + YY_MORE_ADJ;
+
+					yy_act = YY_STATE_EOF(YY_START);
+					goto do_action;
+					}
+
+				else
+					{
+					if ( ! (yy_did_buffer_switch_on_eof) )
+						YY_NEW_FILE;
+					}
+				break;
+				}
+
+			case EOB_ACT_CONTINUE_SCAN:
+				(yy_c_buf_p) =
+					(yytext_ptr) + yy_amount_of_matched_text;
+
+				yy_current_state = yy_get_previous_state(  );
+
+				yy_cp = (yy_c_buf_p);
+				yy_bp = (yytext_ptr) + YY_MORE_ADJ;
+				goto yy_match;
+
+			case EOB_ACT_LAST_MATCH:
+				(yy_c_buf_p) =
+				&YY_CURRENT_BUFFER_LVALUE->yy_ch_buf[(yy_n_chars)];
+
+				yy_current_state = yy_get_previous_state(  );
+
+				yy_cp = (yy_c_buf_p);
+				yy_bp = (yytext_ptr) + YY_MORE_ADJ;
+				goto yy_find_action;
+			}
+		break;
+		}
+
+	default:
+		YY_FATAL_ERROR(
+			"fatal flex scanner internal error--no action found" );
+	} /* end of action switch */
+		} /* end of scanning one token */
+} /* end of yylex */
+
+/* yy_get_next_buffer - try to read in a new buffer
+ *
+ * Returns a code representing an action:
+ *	EOB_ACT_LAST_MATCH -
+ *	EOB_ACT_CONTINUE_SCAN - continue scanning from current position
+ *	EOB_ACT_END_OF_FILE - end of file
+ */
+static int yy_get_next_buffer (void)
+{
+    	register char *dest = YY_CURRENT_BUFFER_LVALUE->yy_ch_buf;
+	register char *source = (yytext_ptr);
+	register int number_to_move, i;
+	int ret_val;
+
+	if ( (yy_c_buf_p) > &YY_CURRENT_BUFFER_LVALUE->yy_ch_buf[(yy_n_chars) + 1] )
+		YY_FATAL_ERROR(
+		"fatal flex scanner internal error--end of buffer missed" );
+
+	if ( YY_CURRENT_BUFFER_LVALUE->yy_fill_buffer == 0 )
+		{ /* Don't try to fill the buffer, so this is an EOF. */
+		if ( (yy_c_buf_p) - (yytext_ptr) - YY_MORE_ADJ == 1 )
+			{
+			/* We matched a single character, the EOB, so
+			 * treat this as a final EOF.
+			 */
+			return EOB_ACT_END_OF_FILE;
+			}
+
+		else
+			{
+			/* We matched some text prior to the EOB, first
+			 * process it.
+			 */
+			return EOB_ACT_LAST_MATCH;
+			}
+		}
+
+	/* Try to read more data. */
+
+	/* First move last chars to start of buffer. */
+	number_to_move = (int) ((yy_c_buf_p) - (yytext_ptr)) - 1;
+
+	for ( i = 0; i < number_to_move; ++i )
+		*(dest++) = *(source++);
+
+	if ( YY_CURRENT_BUFFER_LVALUE->yy_buffer_status == YY_BUFFER_EOF_PENDING )
+		/* don't do the read, it's not guaranteed to return an EOF,
+		 * just force an EOF
+		 */
+		YY_CURRENT_BUFFER_LVALUE->yy_n_chars = (yy_n_chars) = 0;
+
+	else
+		{
+			size_t num_to_read =
+			YY_CURRENT_BUFFER_LVALUE->yy_buf_size - number_to_move - 1;
+
+		while ( num_to_read <= 0 )
+			{ /* Not enough room in the buffer - grow it. */
+
+			/* just a shorter name for the current buffer */
+			YY_BUFFER_STATE b = YY_CURRENT_BUFFER;
+
+			int yy_c_buf_p_offset =
+				(int) ((yy_c_buf_p) - b->yy_ch_buf);
+
+			if ( b->yy_is_our_buffer )
+				{
+				int new_size = b->yy_buf_size * 2;
+
+				if ( new_size <= 0 )
+					b->yy_buf_size += b->yy_buf_size / 8;
+				else
+					b->yy_buf_size *= 2;
+
+				b->yy_ch_buf = (char *)
+					/* Include room in for 2 EOB chars. */
+					yyrealloc((void *) b->yy_ch_buf,b->yy_buf_size + 2  );
+				}
+			else
+				/* Can't grow it, we don't own it. */
+				b->yy_ch_buf = 0;
+
+			if ( ! b->yy_ch_buf )
+				YY_FATAL_ERROR(
+				"fatal error - scanner input buffer overflow" );
+
+			(yy_c_buf_p) = &b->yy_ch_buf[yy_c_buf_p_offset];
+
+			num_to_read = YY_CURRENT_BUFFER_LVALUE->yy_buf_size -
+						number_to_move - 1;
+
+			}
+
+		if ( num_to_read > YY_READ_BUF_SIZE )
+			num_to_read = YY_READ_BUF_SIZE;
+
+		/* Read in more data. */
+		YY_INPUT( (&YY_CURRENT_BUFFER_LVALUE->yy_ch_buf[number_to_move]),
+			(yy_n_chars), num_to_read );
+
+		YY_CURRENT_BUFFER_LVALUE->yy_n_chars = (yy_n_chars);
+		}
+
+	if ( (yy_n_chars) == 0 )
+		{
+		if ( number_to_move == YY_MORE_ADJ )
+			{
+			ret_val = EOB_ACT_END_OF_FILE;
+			yyrestart(yyin  );
+			}
+
+		else
+			{
+			ret_val = EOB_ACT_LAST_MATCH;
+			YY_CURRENT_BUFFER_LVALUE->yy_buffer_status =
+				YY_BUFFER_EOF_PENDING;
+			}
+		}
+
+	else
+		ret_val = EOB_ACT_CONTINUE_SCAN;
+
+	(yy_n_chars) += number_to_move;
+	YY_CURRENT_BUFFER_LVALUE->yy_ch_buf[(yy_n_chars)] = YY_END_OF_BUFFER_CHAR;
+	YY_CURRENT_BUFFER_LVALUE->yy_ch_buf[(yy_n_chars) + 1] = YY_END_OF_BUFFER_CHAR;
+
+	(yytext_ptr) = &YY_CURRENT_BUFFER_LVALUE->yy_ch_buf[0];
+
+	return ret_val;
+}
+
+/* yy_get_previous_state - get the state just before the EOB char was reached */
+
+    static yy_state_type yy_get_previous_state (void)
+{
+	register yy_state_type yy_current_state;
+	register char *yy_cp;
+    
+	yy_current_state = (yy_start);
+
+	for ( yy_cp = (yytext_ptr) + YY_MORE_ADJ; yy_cp < (yy_c_buf_p); ++yy_cp )
+		{
+		register YY_CHAR yy_c = (*yy_cp ? yy_ec[YY_SC_TO_UI(*yy_cp)] : 1);
+		if ( yy_accept[yy_current_state] )
+			{
+			(yy_last_accepting_state) = yy_current_state;
+			(yy_last_accepting_cpos) = yy_cp;
+			}
+		while ( yy_chk[yy_base[yy_current_state] + yy_c] != yy_current_state )
+			{
+			yy_current_state = (int) yy_def[yy_current_state];
+			if ( yy_current_state >= 399 )
+				yy_c = yy_meta[(unsigned int) yy_c];
+			}
+		yy_current_state = yy_nxt[yy_base[yy_current_state] + (unsigned int) yy_c];
+		}
+
+	return yy_current_state;
+}
+
+/* yy_try_NUL_trans - try to make a transition on the NUL character
+ *
+ * synopsis
+ *	next_state = yy_try_NUL_trans( current_state );
+ */
+    static yy_state_type yy_try_NUL_trans  (yy_state_type yy_current_state )
+{
+	register int yy_is_jam;
+    	register char *yy_cp = (yy_c_buf_p);
+
+	register YY_CHAR yy_c = 1;
+	if ( yy_accept[yy_current_state] )
+		{
+		(yy_last_accepting_state) = yy_current_state;
+		(yy_last_accepting_cpos) = yy_cp;
+		}
+	while ( yy_chk[yy_base[yy_current_state] + yy_c] != yy_current_state )
+		{
+		yy_current_state = (int) yy_def[yy_current_state];
+		if ( yy_current_state >= 399 )
+			yy_c = yy_meta[(unsigned int) yy_c];
+		}
+	yy_current_state = yy_nxt[yy_base[yy_current_state] + (unsigned int) yy_c];
+	yy_is_jam = (yy_current_state == 398);
+
+	return yy_is_jam ? 0 : yy_current_state;
+}
+
+#ifndef YY_NO_INPUT
+#ifdef __cplusplus
+    static int yyinput (void)
+#else
+    static int input  (void)
+#endif
+
+{
+	int c;
+    
+	*(yy_c_buf_p) = (yy_hold_char);
+
+	if ( *(yy_c_buf_p) == YY_END_OF_BUFFER_CHAR )
+		{
+		/* yy_c_buf_p now points to the character we want to return.
+		 * If this occurs *before* the EOB characters, then it's a
+		 * valid NUL; if not, then we've hit the end of the buffer.
+		 */
+		if ( (yy_c_buf_p) < &YY_CURRENT_BUFFER_LVALUE->yy_ch_buf[(yy_n_chars)] )
+			/* This was really a NUL. */
+			*(yy_c_buf_p) = '\0';
+
+		else
+			{ /* need more input */
+			int offset = (int)((yy_c_buf_p) - (yytext_ptr));
+			++(yy_c_buf_p);
+
+			switch ( yy_get_next_buffer(  ) )
+				{
+				case EOB_ACT_LAST_MATCH:
+					/* This happens because yy_g_n_b()
+					 * sees that we've accumulated a
+					 * token and flags that we need to
+					 * try matching the token before
+					 * proceeding.  But for input(),
+					 * there's no matching to consider.
+					 * So convert the EOB_ACT_LAST_MATCH
+					 * to EOB_ACT_END_OF_FILE.
+					 */
+
+					/* Reset buffer status. */
+					yyrestart(yyin );
+
+					/*FALLTHROUGH*/
+
+				case EOB_ACT_END_OF_FILE:
+					{
+					if ( yywrap( ) )
+						return EOF;
+
+					if ( ! (yy_did_buffer_switch_on_eof) )
+						YY_NEW_FILE;
+#ifdef __cplusplus
+					return yyinput();
+#else
+					return input();
+#endif
+					}
+
+				case EOB_ACT_CONTINUE_SCAN:
+					(yy_c_buf_p) = (yytext_ptr) + offset;
+					break;
+				}
+			}
+		}
+
+	c = *(unsigned char *) (yy_c_buf_p);	/* cast for 8-bit char's */
+	*(yy_c_buf_p) = '\0';	/* preserve yytext */
+	(yy_hold_char) = *++(yy_c_buf_p);
+
+	return c;
+}
+#endif	/* ifndef YY_NO_INPUT */
+
+/** Immediately switch to a different input stream.
+ * @param input_file A readable stream.
+ * 
+ * @note This function does not reset the start condition to @c INITIAL .
+ */
+    static void yyrestart  (FILE * input_file )
+{
+    
+	if ( ! YY_CURRENT_BUFFER ){
+        yyensure_buffer_stack ();
+		YY_CURRENT_BUFFER_LVALUE =
+            yy_create_buffer(yyin,YY_BUF_SIZE );
+	}
+
+	yy_init_buffer(YY_CURRENT_BUFFER,input_file );
+	yy_load_buffer_state( );
+}
+
+/** Switch to a different input buffer.
+ * @param new_buffer The new input buffer.
+ * 
+ */
+    __attribute__((unused)) static void yy_switch_to_buffer  (YY_BUFFER_STATE  new_buffer )
+{
+    
+	/* TODO. We should be able to replace this entire function body
+	 * with
+	 *		yypop_buffer_state();
+	 *		yypush_buffer_state(new_buffer);
+     */
+	yyensure_buffer_stack ();
+	if ( YY_CURRENT_BUFFER == new_buffer )
+		return;
+
+	if ( YY_CURRENT_BUFFER )
+		{
+		/* Flush out information for old buffer. */
+		*(yy_c_buf_p) = (yy_hold_char);
+		YY_CURRENT_BUFFER_LVALUE->yy_buf_pos = (yy_c_buf_p);
+		YY_CURRENT_BUFFER_LVALUE->yy_n_chars = (yy_n_chars);
+		}
+
+	YY_CURRENT_BUFFER_LVALUE = new_buffer;
+	yy_load_buffer_state( );
+
+	/* We don't actually know whether we did this switch during
+	 * EOF (yywrap()) processing, but the only time this flag
+	 * is looked at is after yywrap() is called, so it's safe
+	 * to go ahead and always set it.
+	 */
+	(yy_did_buffer_switch_on_eof) = 1;
+}
+
+static void yy_load_buffer_state  (void)
+{
+    	(yy_n_chars) = YY_CURRENT_BUFFER_LVALUE->yy_n_chars;
+	(yytext_ptr) = (yy_c_buf_p) = YY_CURRENT_BUFFER_LVALUE->yy_buf_pos;
+	yyin = YY_CURRENT_BUFFER_LVALUE->yy_input_file;
+	(yy_hold_char) = *(yy_c_buf_p);
+}
+
+/** Allocate and initialize an input buffer state.
+ * @param file A readable stream.
+ * @param size The character buffer size in bytes. When in doubt, use @c YY_BUF_SIZE.
+ * 
+ * @return the allocated buffer state.
+ */
+    static YY_BUFFER_STATE yy_create_buffer  (FILE * file, int  size )
+{
+	YY_BUFFER_STATE b;
+    
+	b = (YY_BUFFER_STATE) yyalloc(sizeof( struct yy_buffer_state )  );
+	if ( ! b )
+		YY_FATAL_ERROR( "out of dynamic memory in yy_create_buffer()" );
+
+	b->yy_buf_size = size;
+
+	/* yy_ch_buf has to be 2 characters longer than the size given because
+	 * we need to put in 2 end-of-buffer characters.
+	 */
+	b->yy_ch_buf = (char *) yyalloc(b->yy_buf_size + 2  );
+	if ( ! b->yy_ch_buf )
+		YY_FATAL_ERROR( "out of dynamic memory in yy_create_buffer()" );
+
+	b->yy_is_our_buffer = 1;
+
+	yy_init_buffer(b,file );
+
+	return b;
+}
+
+/** Destroy the buffer.
+ * @param b a buffer created with yy_create_buffer()
+ * 
+ */
+    static void yy_delete_buffer (YY_BUFFER_STATE  b )
+{
+    
+	if ( ! b )
+		return;
+
+	if ( b == YY_CURRENT_BUFFER ) /* Not sure if we should pop here. */
+		YY_CURRENT_BUFFER_LVALUE = (YY_BUFFER_STATE) 0;
+
+	if ( b->yy_is_our_buffer )
+		yyfree((void *) b->yy_ch_buf  );
+
+	yyfree((void *) b  );
+}
+
+/* Initializes or reinitializes a buffer.
+ * This function is sometimes called more than once on the same buffer,
+ * such as during a yyrestart() or at EOF.
+ */
+    static void yy_init_buffer  (YY_BUFFER_STATE  b, FILE * file )
+
+{
+	int oerrno = errno;
+    
+	yy_flush_buffer(b );
+
+	b->yy_input_file = file;
+	b->yy_fill_buffer = 1;
+
+    /* If b is the current buffer, then yy_init_buffer was _probably_
+     * called from yyrestart() or through yy_get_next_buffer.
+     * In that case, we don't want to reset the lineno or column.
+     */
+    if (b != YY_CURRENT_BUFFER){
+        b->yy_bs_lineno = 1;
+        b->yy_bs_column = 0;
+    }
+
+        b->yy_is_interactive = 0;
+    
+	errno = oerrno;
+}
+
+/** Discard all buffered characters. On the next scan, YY_INPUT will be called.
+ * @param b the buffer state to be flushed, usually @c YY_CURRENT_BUFFER.
+ * 
+ */
+    static void yy_flush_buffer (YY_BUFFER_STATE  b )
+{
+    	if ( ! b )
+		return;
+
+	b->yy_n_chars = 0;
+
+	/* We always need two end-of-buffer characters.  The first causes
+	 * a transition to the end-of-buffer state.  The second causes
+	 * a jam in that state.
+	 */
+	b->yy_ch_buf[0] = YY_END_OF_BUFFER_CHAR;
+	b->yy_ch_buf[1] = YY_END_OF_BUFFER_CHAR;
+
+	b->yy_buf_pos = &b->yy_ch_buf[0];
+
+	b->yy_at_bol = 1;
+	b->yy_buffer_status = YY_BUFFER_NEW;
+
+	if ( b == YY_CURRENT_BUFFER )
+		yy_load_buffer_state( );
+}
+
+/** Pushes the new state onto the stack. The new state becomes
+ *  the current state. This function will allocate the stack
+ *  if necessary.
+ *  @param new_buffer The new state.
+ *  
+ */
+__attribute__((unused)) static void yypush_buffer_state (YY_BUFFER_STATE new_buffer )
+{
+    	if (new_buffer == NULL)
+		return;
+
+	yyensure_buffer_stack();
+
+	/* This block is copied from yy_switch_to_buffer. */
+	if ( YY_CURRENT_BUFFER )
+		{
+		/* Flush out information for old buffer. */
+		*(yy_c_buf_p) = (yy_hold_char);
+		YY_CURRENT_BUFFER_LVALUE->yy_buf_pos = (yy_c_buf_p);
+		YY_CURRENT_BUFFER_LVALUE->yy_n_chars = (yy_n_chars);
+		}
+
+	/* Only push if top exists. Otherwise, replace top. */
+	if (YY_CURRENT_BUFFER)
+		(yy_buffer_stack_top)++;
+	YY_CURRENT_BUFFER_LVALUE = new_buffer;
+
+	/* copied from yy_switch_to_buffer. */
+	yy_load_buffer_state( );
+	(yy_did_buffer_switch_on_eof) = 1;
+}
+
+/** Removes and deletes the top of the stack, if present.
+ *  The next element becomes the new top.
+ *  
+ */
+__attribute__((unused)) static void yypop_buffer_state (void)
+{
+    	if (!YY_CURRENT_BUFFER)
+		return;
+
+	yy_delete_buffer(YY_CURRENT_BUFFER );
+	YY_CURRENT_BUFFER_LVALUE = NULL;
+	if ((yy_buffer_stack_top) > 0)
+		--(yy_buffer_stack_top);
+
+	if (YY_CURRENT_BUFFER) {
+		yy_load_buffer_state( );
+		(yy_did_buffer_switch_on_eof) = 1;
+	}
+}
+
+/* Allocates the stack if it does not exist.
+ *  Guarantees space for at least one push.
+ */
+static void yyensure_buffer_stack (void)
+{
+	int num_to_alloc;
+    
+	if (!(yy_buffer_stack)) {
+
+		/* First allocation is just for 2 elements, since we don't know if this
+		 * scanner will even need a stack. We use 2 instead of 1 to avoid an
+		 * immediate realloc on the next call.
+         */
+		num_to_alloc = 1;
+		(yy_buffer_stack) = (struct yy_buffer_state**)yyalloc
+								(num_to_alloc * sizeof(struct yy_buffer_state*)
+								);
+		
+		memset((yy_buffer_stack), 0, num_to_alloc * sizeof(struct yy_buffer_state*));
+				
+		(yy_buffer_stack_max) = num_to_alloc;
+		(yy_buffer_stack_top) = 0;
+		return;
+	}
+
+	if ((yy_buffer_stack_top) >= ((yy_buffer_stack_max)) - 1){
+
+		/* Increase the buffer to prepare for a possible push. */
+		int grow_size = 8 /* arbitrary grow size */;
+
+		num_to_alloc = (yy_buffer_stack_max) + grow_size;
+		(yy_buffer_stack) = (struct yy_buffer_state**)yyrealloc
+								((yy_buffer_stack),
+								num_to_alloc * sizeof(struct yy_buffer_state*)
+								);
+
+		/* zero only the new slots.*/
+		memset((yy_buffer_stack) + (yy_buffer_stack_max), 0, grow_size * sizeof(struct yy_buffer_state*));
+		(yy_buffer_stack_max) = num_to_alloc;
+	}
+}
+
+#ifndef YY_EXIT_FAILURE
+#define YY_EXIT_FAILURE 2
+#endif
+
+static void yy_fatal_error (yyconst char* msg )
+{
+    	(void) fprintf( stderr, "%s\n", msg );
+	exit( YY_EXIT_FAILURE );
+}
+
+/* Redefine yyless() so it works in section 3 code. */
+
+#undef yyless
+#define yyless(n) \
+	do \
+		{ \
+		/* Undo effects of setting up yytext. */ \
+        int yyless_macro_arg = (n); \
+        YY_LESS_LINENO(yyless_macro_arg);\
+		yytext[yyleng] = (yy_hold_char); \
+		(yy_c_buf_p) = yytext + yyless_macro_arg; \
+		(yy_hold_char) = *(yy_c_buf_p); \
+		*(yy_c_buf_p) = '\0'; \
+		yyleng = yyless_macro_arg; \
+		} \
+	while ( 0 )
+
+/* Accessor  methods (get/set functions) to struct members. */
+
+/** Get the current line number.
+ * 
+ */
+__attribute__((unused)) static int yyget_lineno  (void)
+{
+        
+    return yylineno;
+}
+
+/** Get the input stream.
+ * 
+ */
+__attribute__((unused)) static FILE *yyget_in  (void)
+{
+        return yyin;
+}
+
+/** Get the output stream.
+ * 
+ */
+__attribute__((unused)) static FILE *yyget_out  (void)
+{
+        return yyout;
+}
+
+/** Get the length of the current token.
+ * 
+ */
+__attribute__((unused)) static int yyget_leng  (void)
+{
+        return yyleng;
+}
+
+/** Get the current token.
+ * 
+ */
+
+__attribute__((unused)) static char *yyget_text  (void)
+{
+        return yytext;
+}
+
+/** Set the current line number.
+ * @param line_number
+ * 
+ */
+__attribute__((unused)) static void yyset_lineno (int  line_number )
+{
+    
+    yylineno = line_number;
+}
+
+/** Set the input stream. This does not discard the current
+ * input buffer.
+ * @param in_str A readable stream.
+ * 
+ * @see yy_switch_to_buffer
+ */
+__attribute__((unused)) static void yyset_in (FILE *  in_str )
+{
+        yyin = in_str ;
+}
+
+__attribute__((unused)) static void yyset_out (FILE *  out_str )
+{
+        yyout = out_str ;
+}
+
+__attribute__((unused)) static int yyget_debug  (void)
+{
+        return yy_flex_debug;
+}
+
+__attribute__((unused)) static void yyset_debug (int  bdebug )
+{
+        yy_flex_debug = bdebug ;
+}
+
+/* yylex_destroy is for both reentrant and non-reentrant scanners. */
+__attribute__((unused)) static int yylex_destroy  (void)
+{
+    
+    /* Pop the buffer stack, destroying each element. */
+	while(YY_CURRENT_BUFFER){
+		yy_delete_buffer(YY_CURRENT_BUFFER  );
+		YY_CURRENT_BUFFER_LVALUE = NULL;
+		yypop_buffer_state();
+	}
+
+	/* Destroy the stack itself. */
+	yyfree((yy_buffer_stack) );
+	(yy_buffer_stack) = NULL;
+
+    return 0;
+}
+
+/*
+ * Internal utility routines.
+ */
+
+#ifndef yytext_ptr
+static void yy_flex_strncpy (char* s1, yyconst char * s2, int n )
+{
+	register int i;
+    	for ( i = 0; i < n; ++i )
+		s1[i] = s2[i];
+}
+#endif
+
+#ifdef YY_NEED_STRLEN
+static int yy_flex_strlen (yyconst char * s )
+{
+	register int n;
+    	for ( n = 0; s[n]; ++n )
+		;
+
+	return n;
+}
+#endif
+
+static void *yyalloc (yy_size_t  size )
+{
+	return (void *) malloc( size );
+}
+
+static void *yyrealloc  (void * ptr, yy_size_t  size )
+{
+	/* The cast to (char *) in the following accommodates both
+	 * implementations that use char* generic pointers, and those
+	 * that use void* generic pointers.  It works with the latter
+	 * because both ANSI C and C++ allow castless assignment from
+	 * any pointer type to void*, and deal with argument conversions
+	 * as though doing an assignment.
+	 */
+	return (void *) realloc( (char *) ptr, size );
+}
+
+static void yyfree (void * ptr )
+{
+	free( (char *) ptr );	/* see yyrealloc() for (char *) cast */
+}
+
+#define YYTABLES_NAME "yytables"
+
+#undef YY_NEW_FILE
+#undef YY_FLUSH_BUFFER
+#undef yy_set_bol
+#undef yy_new_buffer
+#undef yy_set_interactive
+#undef yytext_ptr
+#undef YY_DO_BEFORE_ACTION
+
+#ifdef YY_DECL_IS_OURS
+#undef YY_DECL_IS_OURS
+#undef YY_DECL
+#endif
+#line 648 "pars0lex.l"
+
+
+
diff --git a/storage/xtradb/pars/make_bison.sh b/storage/xtradb/pars/make_bison.sh
new file mode 100755
index 00000000000..09bb86e3106
--- /dev/null
+++ b/storage/xtradb/pars/make_bison.sh
@@ -0,0 +1,32 @@
+#!/bin/bash
+#
+# Copyright (c) 2006, 2009, Innobase Oy. All Rights Reserved.
+# 
+# This program is free software; you can redistribute it and/or modify it under
+# the terms of the GNU General Public License as published by the Free Software
+# Foundation; version 2 of the License.
+# 
+# This program is distributed in the hope that it will be useful, but WITHOUT
+# ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+# FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+# 
+# You should have received a copy of the GNU General Public License along with
+# this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+# Place, Suite 330, Boston, MA 02111-1307 USA
+#
+# generate parser files from bison input files.
+
+set -eu
+TMPFILE=pars0grm.tab.c
+OUTFILE=pars0grm.c
+
+bison -d pars0grm.y
+mv pars0grm.tab.h ../include/pars0grm.h
+
+sed -e '
+s/'"$TMPFILE"'/'"$OUTFILE"'/;
+s/^\(\(YYSTYPE\|int\) yy\(char\|nerrs\)\)/static \1/;
+s/\(\(YYSTYPE\|int\) yy\(lval\|parse\)\)/UNIV_INTERN \1/;
+' < "$TMPFILE" > "$OUTFILE"
+
+rm "$TMPFILE"
diff --git a/storage/xtradb/pars/make_flex.sh b/storage/xtradb/pars/make_flex.sh
new file mode 100755
index 00000000000..89308a6636f
--- /dev/null
+++ b/storage/xtradb/pars/make_flex.sh
@@ -0,0 +1,48 @@
+#!/bin/bash
+#
+# Copyright (c) 2006, 2009, Innobase Oy. All Rights Reserved.
+# 
+# This program is free software; you can redistribute it and/or modify it under
+# the terms of the GNU General Public License as published by the Free Software
+# Foundation; version 2 of the License.
+# 
+# This program is distributed in the hope that it will be useful, but WITHOUT
+# ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+# FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+# 
+# You should have received a copy of the GNU General Public License along with
+# this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+# Place, Suite 330, Boston, MA 02111-1307 USA
+#
+# generate lexer files from flex input files.
+
+set -eu
+
+TMPFILE=_flex_tmp.c
+OUTFILE=lexyy.c
+
+flex -o $TMPFILE pars0lex.l
+
+# AIX needs its includes done in a certain order, so include "univ.i" first
+# to be sure we get it right.
+echo '#include "univ.i"' > $OUTFILE
+
+# flex assigns a pointer to an int in one place without a cast, resulting in
+# a warning on Win64.  Add the cast.  Also define some symbols as static.
+sed -e '
+s/'"$TMPFILE"'/'"$OUTFILE"'/;
+s/\(int offset = \)\((yy_c_buf_p) - (yytext_ptr)\);/\1(int)(\2);/;
+s/\(void yy\(restart\|_\(delete\|flush\)_buffer\)\)/static \1/;
+s/\(void yy_switch_to_buffer\)/__attribute__((unused)) static \1/;
+s/\(void yy\(push\|pop\)_buffer_state\)/__attribute__((unused)) static \1/;
+s/\(YY_BUFFER_STATE yy_create_buffer\)/static \1/;
+s/\(\(int\|void\) yy[gs]et_\)/__attribute__((unused)) static \1/;
+s/\(void \*\?yy\(\(re\)\?alloc\|free\)\)/static \1/;
+s/\(extern \)\?\(int yy\(leng\|lineno\|_flex_debug\)\)/static \2/;
+s/\(int yylex_destroy\)/__attribute__((unused)) static \1/;
+s/\(extern \)\?\(int yylex \)/UNIV_INTERN \2/;
+s/^\(\(FILE\|char\) *\* *yyget\)/__attribute__((unused)) static \1/;
+s/^\(extern \)\?\(\(FILE\|char\) *\* *yy\)/static \2/;
+' < $TMPFILE >> $OUTFILE
+
+rm $TMPFILE
diff --git a/storage/xtradb/pars/pars0grm.c b/storage/xtradb/pars/pars0grm.c
new file mode 100644
index 00000000000..d667970735e
--- /dev/null
+++ b/storage/xtradb/pars/pars0grm.c
@@ -0,0 +1,2601 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+Copyright (c) 1984, 1989, 1990, 2000, 2001, 2002, 2003, 2004 Free Software
+Foundation, Inc.
+
+As a special exception, when this file is copied by Bison into a
+Bison output file, you may use that output file without restriction.
+This special exception was added by the Free Software Foundation
+in version 1.24 of Bison.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/* A Bison parser, made by GNU Bison 2.0.  */
+
+/* Written by Richard Stallman by simplifying the original so called
+   ``semantic'' parser.  */
+
+/* All symbols defined below should begin with yy or YY, to avoid
+   infringing on user name space.  This should be done even for local
+   variables, as they might otherwise be expanded by user macros.
+   There are some unavoidable exceptions within include files to
+   define necessary library symbols; they are noted "INFRINGES ON
+   USER NAME SPACE" below.  */
+
+/* Identify Bison output.  */
+#define YYBISON 1
+
+/* Skeleton name.  */
+#define YYSKELETON_NAME "yacc.c"
+
+/* Pure parsers.  */
+#define YYPURE 0
+
+/* Using locations.  */
+#define YYLSP_NEEDED 0
+
+
+
+/* Tokens.  */
+#ifndef YYTOKENTYPE
+# define YYTOKENTYPE
+   /* Put the tokens into the symbol table, so that GDB and other debuggers
+      know about them.  */
+   enum yytokentype {
+     PARS_INT_LIT = 258,
+     PARS_FLOAT_LIT = 259,
+     PARS_STR_LIT = 260,
+     PARS_FIXBINARY_LIT = 261,
+     PARS_BLOB_LIT = 262,
+     PARS_NULL_LIT = 263,
+     PARS_ID_TOKEN = 264,
+     PARS_AND_TOKEN = 265,
+     PARS_OR_TOKEN = 266,
+     PARS_NOT_TOKEN = 267,
+     PARS_GE_TOKEN = 268,
+     PARS_LE_TOKEN = 269,
+     PARS_NE_TOKEN = 270,
+     PARS_PROCEDURE_TOKEN = 271,
+     PARS_IN_TOKEN = 272,
+     PARS_OUT_TOKEN = 273,
+     PARS_BINARY_TOKEN = 274,
+     PARS_BLOB_TOKEN = 275,
+     PARS_INT_TOKEN = 276,
+     PARS_INTEGER_TOKEN = 277,
+     PARS_FLOAT_TOKEN = 278,
+     PARS_CHAR_TOKEN = 279,
+     PARS_IS_TOKEN = 280,
+     PARS_BEGIN_TOKEN = 281,
+     PARS_END_TOKEN = 282,
+     PARS_IF_TOKEN = 283,
+     PARS_THEN_TOKEN = 284,
+     PARS_ELSE_TOKEN = 285,
+     PARS_ELSIF_TOKEN = 286,
+     PARS_LOOP_TOKEN = 287,
+     PARS_WHILE_TOKEN = 288,
+     PARS_RETURN_TOKEN = 289,
+     PARS_SELECT_TOKEN = 290,
+     PARS_SUM_TOKEN = 291,
+     PARS_COUNT_TOKEN = 292,
+     PARS_DISTINCT_TOKEN = 293,
+     PARS_FROM_TOKEN = 294,
+     PARS_WHERE_TOKEN = 295,
+     PARS_FOR_TOKEN = 296,
+     PARS_DDOT_TOKEN = 297,
+     PARS_READ_TOKEN = 298,
+     PARS_ORDER_TOKEN = 299,
+     PARS_BY_TOKEN = 300,
+     PARS_ASC_TOKEN = 301,
+     PARS_DESC_TOKEN = 302,
+     PARS_INSERT_TOKEN = 303,
+     PARS_INTO_TOKEN = 304,
+     PARS_VALUES_TOKEN = 305,
+     PARS_UPDATE_TOKEN = 306,
+     PARS_SET_TOKEN = 307,
+     PARS_DELETE_TOKEN = 308,
+     PARS_CURRENT_TOKEN = 309,
+     PARS_OF_TOKEN = 310,
+     PARS_CREATE_TOKEN = 311,
+     PARS_TABLE_TOKEN = 312,
+     PARS_INDEX_TOKEN = 313,
+     PARS_UNIQUE_TOKEN = 314,
+     PARS_CLUSTERED_TOKEN = 315,
+     PARS_DOES_NOT_FIT_IN_MEM_TOKEN = 316,
+     PARS_ON_TOKEN = 317,
+     PARS_ASSIGN_TOKEN = 318,
+     PARS_DECLARE_TOKEN = 319,
+     PARS_CURSOR_TOKEN = 320,
+     PARS_SQL_TOKEN = 321,
+     PARS_OPEN_TOKEN = 322,
+     PARS_FETCH_TOKEN = 323,
+     PARS_CLOSE_TOKEN = 324,
+     PARS_NOTFOUND_TOKEN = 325,
+     PARS_TO_CHAR_TOKEN = 326,
+     PARS_TO_NUMBER_TOKEN = 327,
+     PARS_TO_BINARY_TOKEN = 328,
+     PARS_BINARY_TO_NUMBER_TOKEN = 329,
+     PARS_SUBSTR_TOKEN = 330,
+     PARS_REPLSTR_TOKEN = 331,
+     PARS_CONCAT_TOKEN = 332,
+     PARS_INSTR_TOKEN = 333,
+     PARS_LENGTH_TOKEN = 334,
+     PARS_SYSDATE_TOKEN = 335,
+     PARS_PRINTF_TOKEN = 336,
+     PARS_ASSERT_TOKEN = 337,
+     PARS_RND_TOKEN = 338,
+     PARS_RND_STR_TOKEN = 339,
+     PARS_ROW_PRINTF_TOKEN = 340,
+     PARS_COMMIT_TOKEN = 341,
+     PARS_ROLLBACK_TOKEN = 342,
+     PARS_WORK_TOKEN = 343,
+     PARS_UNSIGNED_TOKEN = 344,
+     PARS_EXIT_TOKEN = 345,
+     PARS_FUNCTION_TOKEN = 346,
+     PARS_LOCK_TOKEN = 347,
+     PARS_SHARE_TOKEN = 348,
+     PARS_MODE_TOKEN = 349,
+     NEG = 350
+   };
+#endif
+#define PARS_INT_LIT 258
+#define PARS_FLOAT_LIT 259
+#define PARS_STR_LIT 260
+#define PARS_FIXBINARY_LIT 261
+#define PARS_BLOB_LIT 262
+#define PARS_NULL_LIT 263
+#define PARS_ID_TOKEN 264
+#define PARS_AND_TOKEN 265
+#define PARS_OR_TOKEN 266
+#define PARS_NOT_TOKEN 267
+#define PARS_GE_TOKEN 268
+#define PARS_LE_TOKEN 269
+#define PARS_NE_TOKEN 270
+#define PARS_PROCEDURE_TOKEN 271
+#define PARS_IN_TOKEN 272
+#define PARS_OUT_TOKEN 273
+#define PARS_BINARY_TOKEN 274
+#define PARS_BLOB_TOKEN 275
+#define PARS_INT_TOKEN 276
+#define PARS_INTEGER_TOKEN 277
+#define PARS_FLOAT_TOKEN 278
+#define PARS_CHAR_TOKEN 279
+#define PARS_IS_TOKEN 280
+#define PARS_BEGIN_TOKEN 281
+#define PARS_END_TOKEN 282
+#define PARS_IF_TOKEN 283
+#define PARS_THEN_TOKEN 284
+#define PARS_ELSE_TOKEN 285
+#define PARS_ELSIF_TOKEN 286
+#define PARS_LOOP_TOKEN 287
+#define PARS_WHILE_TOKEN 288
+#define PARS_RETURN_TOKEN 289
+#define PARS_SELECT_TOKEN 290
+#define PARS_SUM_TOKEN 291
+#define PARS_COUNT_TOKEN 292
+#define PARS_DISTINCT_TOKEN 293
+#define PARS_FROM_TOKEN 294
+#define PARS_WHERE_TOKEN 295
+#define PARS_FOR_TOKEN 296
+#define PARS_DDOT_TOKEN 297
+#define PARS_READ_TOKEN 298
+#define PARS_ORDER_TOKEN 299
+#define PARS_BY_TOKEN 300
+#define PARS_ASC_TOKEN 301
+#define PARS_DESC_TOKEN 302
+#define PARS_INSERT_TOKEN 303
+#define PARS_INTO_TOKEN 304
+#define PARS_VALUES_TOKEN 305
+#define PARS_UPDATE_TOKEN 306
+#define PARS_SET_TOKEN 307
+#define PARS_DELETE_TOKEN 308
+#define PARS_CURRENT_TOKEN 309
+#define PARS_OF_TOKEN 310
+#define PARS_CREATE_TOKEN 311
+#define PARS_TABLE_TOKEN 312
+#define PARS_INDEX_TOKEN 313
+#define PARS_UNIQUE_TOKEN 314
+#define PARS_CLUSTERED_TOKEN 315
+#define PARS_DOES_NOT_FIT_IN_MEM_TOKEN 316
+#define PARS_ON_TOKEN 317
+#define PARS_ASSIGN_TOKEN 318
+#define PARS_DECLARE_TOKEN 319
+#define PARS_CURSOR_TOKEN 320
+#define PARS_SQL_TOKEN 321
+#define PARS_OPEN_TOKEN 322
+#define PARS_FETCH_TOKEN 323
+#define PARS_CLOSE_TOKEN 324
+#define PARS_NOTFOUND_TOKEN 325
+#define PARS_TO_CHAR_TOKEN 326
+#define PARS_TO_NUMBER_TOKEN 327
+#define PARS_TO_BINARY_TOKEN 328
+#define PARS_BINARY_TO_NUMBER_TOKEN 329
+#define PARS_SUBSTR_TOKEN 330
+#define PARS_REPLSTR_TOKEN 331
+#define PARS_CONCAT_TOKEN 332
+#define PARS_INSTR_TOKEN 333
+#define PARS_LENGTH_TOKEN 334
+#define PARS_SYSDATE_TOKEN 335
+#define PARS_PRINTF_TOKEN 336
+#define PARS_ASSERT_TOKEN 337
+#define PARS_RND_TOKEN 338
+#define PARS_RND_STR_TOKEN 339
+#define PARS_ROW_PRINTF_TOKEN 340
+#define PARS_COMMIT_TOKEN 341
+#define PARS_ROLLBACK_TOKEN 342
+#define PARS_WORK_TOKEN 343
+#define PARS_UNSIGNED_TOKEN 344
+#define PARS_EXIT_TOKEN 345
+#define PARS_FUNCTION_TOKEN 346
+#define PARS_LOCK_TOKEN 347
+#define PARS_SHARE_TOKEN 348
+#define PARS_MODE_TOKEN 349
+#define NEG 350
+
+
+
+
+/* Copy the first part of user declarations.  */
+#line 13 "pars0grm.y"
+
+/* The value of the semantic attribute is a pointer to a query tree node
+que_node_t */
+
+#include "univ.i"
+#include <math.h>				/* Can't be before univ.i */
+#include "pars0pars.h"
+#include "mem0mem.h"
+#include "que0types.h"
+#include "que0que.h"
+#include "row0sel.h"
+
+#define YYSTYPE que_node_t*
+
+/* #define __STDC__ */
+
+int
+yylex(void);
+
+
+/* Enabling traces.  */
+#ifndef YYDEBUG
+# define YYDEBUG 0
+#endif
+
+/* Enabling verbose error messages.  */
+#ifdef YYERROR_VERBOSE
+# undef YYERROR_VERBOSE
+# define YYERROR_VERBOSE 1
+#else
+# define YYERROR_VERBOSE 0
+#endif
+
+#if ! defined (YYSTYPE) && ! defined (YYSTYPE_IS_DECLARED)
+typedef int YYSTYPE;
+# define yystype YYSTYPE /* obsolescent; will be withdrawn */
+# define YYSTYPE_IS_DECLARED 1
+# define YYSTYPE_IS_TRIVIAL 1
+#endif
+
+
+
+/* Copy the second part of user declarations.  */
+
+
+/* Line 213 of yacc.c.  */
+#line 297 "pars0grm.c"
+
+#if ! defined (yyoverflow) || YYERROR_VERBOSE
+
+# ifndef YYFREE
+#  define YYFREE free
+# endif
+# ifndef YYMALLOC
+#  define YYMALLOC malloc
+# endif
+
+/* The parser invokes alloca or malloc; define the necessary symbols.  */
+
+# ifdef YYSTACK_USE_ALLOCA
+#  if YYSTACK_USE_ALLOCA
+#   ifdef __GNUC__
+#    define YYSTACK_ALLOC __builtin_alloca
+#   else
+#    define YYSTACK_ALLOC alloca
+#   endif
+#  endif
+# endif
+
+# ifdef YYSTACK_ALLOC
+   /* Pacify GCC's `empty if-body' warning. */
+#  define YYSTACK_FREE(Ptr) do { /* empty */; } while (0)
+# else
+#  if defined (__STDC__) || defined (__cplusplus)
+#   include <stdlib.h> /* INFRINGES ON USER NAME SPACE */
+#   define YYSIZE_T size_t
+#  endif
+#  define YYSTACK_ALLOC YYMALLOC
+#  define YYSTACK_FREE YYFREE
+# endif
+#endif /* ! defined (yyoverflow) || YYERROR_VERBOSE */
+
+
+#if (! defined (yyoverflow) \
+     && (! defined (__cplusplus) \
+	 || (defined (YYSTYPE_IS_TRIVIAL) && YYSTYPE_IS_TRIVIAL)))
+
+/* A type that is properly aligned for any stack member.  */
+union yyalloc
+{
+  short int yyss;
+  YYSTYPE yyvs;
+  };
+
+/* The size of the maximum gap between one aligned stack and the next.  */
+# define YYSTACK_GAP_MAXIMUM (sizeof (union yyalloc) - 1)
+
+/* The size of an array large to enough to hold all stacks, each with
+   N elements.  */
+# define YYSTACK_BYTES(N) \
+     ((N) * (sizeof (short int) + sizeof (YYSTYPE))			\
+      + YYSTACK_GAP_MAXIMUM)
+
+/* Copy COUNT objects from FROM to TO.  The source and destination do
+   not overlap.  */
+# ifndef YYCOPY
+#  if defined (__GNUC__) && 1 < __GNUC__
+#   define YYCOPY(To, From, Count) \
+      __builtin_memcpy (To, From, (Count) * sizeof (*(From)))
+#  else
+#   define YYCOPY(To, From, Count)		\
+      do					\
+	{					\
+	  register YYSIZE_T yyi;		\
+	  for (yyi = 0; yyi < (Count); yyi++)	\
+	    (To)[yyi] = (From)[yyi];		\
+	}					\
+      while (0)
+#  endif
+# endif
+
+/* Relocate STACK from its old location to the new one.  The
+   local variables YYSIZE and YYSTACKSIZE give the old and new number of
+   elements in the stack, and YYPTR gives the new location of the
+   stack.  Advance YYPTR to a properly aligned location for the next
+   stack.  */
+# define YYSTACK_RELOCATE(Stack)					\
+    do									\
+      {									\
+	YYSIZE_T yynewbytes;						\
+	YYCOPY (&yyptr->Stack, Stack, yysize);				\
+	Stack = &yyptr->Stack;						\
+	yynewbytes = yystacksize * sizeof (*Stack) + YYSTACK_GAP_MAXIMUM; \
+	yyptr += yynewbytes / sizeof (*yyptr);				\
+      }									\
+    while (0)
+
+#endif
+
+#if defined (__STDC__) || defined (__cplusplus)
+   typedef signed char yysigned_char;
+#else
+   typedef short int yysigned_char;
+#endif
+
+/* YYFINAL -- State number of the termination state. */
+#define YYFINAL  5
+/* YYLAST -- Last index in YYTABLE.  */
+#define YYLAST   752
+
+/* YYNTOKENS -- Number of terminals. */
+#define YYNTOKENS  111
+/* YYNNTS -- Number of nonterminals. */
+#define YYNNTS  70
+/* YYNRULES -- Number of rules. */
+#define YYNRULES  175
+/* YYNRULES -- Number of states. */
+#define YYNSTATES  339
+
+/* YYTRANSLATE(YYLEX) -- Bison symbol number corresponding to YYLEX.  */
+#define YYUNDEFTOK  2
+#define YYMAXUTOK   350
+
+#define YYTRANSLATE(YYX) 						\
+  ((unsigned int) (YYX) <= YYMAXUTOK ? yytranslate[YYX] : YYUNDEFTOK)
+
+/* YYTRANSLATE[YYLEX] -- Bison symbol number corresponding to YYLEX.  */
+static const unsigned char yytranslate[] =
+{
+       0,     2,     2,     2,     2,     2,     2,     2,     2,     2,
+       2,     2,     2,     2,     2,     2,     2,     2,     2,     2,
+       2,     2,     2,     2,     2,     2,     2,     2,     2,     2,
+       2,     2,     2,     2,     2,     2,     2,   103,     2,     2,
+     105,   106,   100,    99,   108,    98,     2,   101,     2,     2,
+       2,     2,     2,     2,     2,     2,     2,     2,     2,   104,
+      96,    95,    97,   107,     2,     2,     2,     2,     2,     2,
+       2,     2,     2,     2,     2,     2,     2,     2,     2,     2,
+       2,     2,     2,     2,     2,     2,     2,     2,     2,     2,
+       2,     2,     2,     2,     2,     2,     2,     2,     2,     2,
+       2,     2,     2,     2,     2,     2,     2,     2,     2,     2,
+       2,     2,     2,     2,     2,     2,     2,     2,     2,     2,
+       2,     2,     2,   109,     2,   110,     2,     2,     2,     2,
+       2,     2,     2,     2,     2,     2,     2,     2,     2,     2,
+       2,     2,     2,     2,     2,     2,     2,     2,     2,     2,
+       2,     2,     2,     2,     2,     2,     2,     2,     2,     2,
+       2,     2,     2,     2,     2,     2,     2,     2,     2,     2,
+       2,     2,     2,     2,     2,     2,     2,     2,     2,     2,
+       2,     2,     2,     2,     2,     2,     2,     2,     2,     2,
+       2,     2,     2,     2,     2,     2,     2,     2,     2,     2,
+       2,     2,     2,     2,     2,     2,     2,     2,     2,     2,
+       2,     2,     2,     2,     2,     2,     2,     2,     2,     2,
+       2,     2,     2,     2,     2,     2,     2,     2,     2,     2,
+       2,     2,     2,     2,     2,     2,     2,     2,     2,     2,
+       2,     2,     2,     2,     2,     2,     2,     2,     2,     2,
+       2,     2,     2,     2,     2,     2,     1,     2,     3,     4,
+       5,     6,     7,     8,     9,    10,    11,    12,    13,    14,
+      15,    16,    17,    18,    19,    20,    21,    22,    23,    24,
+      25,    26,    27,    28,    29,    30,    31,    32,    33,    34,
+      35,    36,    37,    38,    39,    40,    41,    42,    43,    44,
+      45,    46,    47,    48,    49,    50,    51,    52,    53,    54,
+      55,    56,    57,    58,    59,    60,    61,    62,    63,    64,
+      65,    66,    67,    68,    69,    70,    71,    72,    73,    74,
+      75,    76,    77,    78,    79,    80,    81,    82,    83,    84,
+      85,    86,    87,    88,    89,    90,    91,    92,    93,    94,
+     102
+};
+
+#if YYDEBUG
+/* YYPRHS[YYN] -- Index of the first RHS symbol of rule number YYN in
+   YYRHS.  */
+static const unsigned short int yyprhs[] =
+{
+       0,     0,     3,     6,     8,    11,    14,    17,    20,    23,
+      26,    29,    32,    35,    38,    41,    44,    47,    50,    53,
+      56,    59,    62,    65,    68,    71,    73,    76,    78,    83,
+      85,    87,    89,    91,    93,    95,    97,   101,   105,   109,
+     113,   116,   120,   124,   128,   132,   136,   140,   144,   148,
+     152,   155,   159,   163,   165,   167,   169,   171,   173,   175,
+     177,   179,   181,   183,   185,   186,   188,   192,   199,   204,
+     206,   208,   210,   214,   216,   220,   221,   223,   227,   228,
+     230,   234,   236,   241,   247,   252,   253,   255,   259,   261,
+     265,   267,   268,   271,   272,   275,   276,   281,   282,   284,
+     286,   287,   292,   301,   305,   311,   314,   318,   320,   324,
+     329,   334,   337,   340,   344,   347,   350,   353,   357,   362,
+     364,   367,   368,   371,   373,   381,   388,   399,   401,   403,
+     406,   409,   414,   419,   425,   427,   431,   432,   436,   437,
+     439,   440,   443,   444,   446,   454,   456,   460,   461,   463,
+     464,   466,   477,   480,   483,   485,   487,   489,   491,   493,
+     497,   501,   502,   504,   508,   512,   513,   515,   518,   525,
+     530,   532,   534,   535,   537,   540
+};
+
+/* YYRHS -- A `-1'-separated list of the rules' RHS. */
+static const short int yyrhs[] =
+{
+     112,     0,    -1,   180,   104,    -1,   118,    -1,   119,   104,
+      -1,   151,   104,    -1,   152,   104,    -1,   153,   104,    -1,
+     150,   104,    -1,   154,   104,    -1,   146,   104,    -1,   133,
+     104,    -1,   135,   104,    -1,   145,   104,    -1,   143,   104,
+      -1,   144,   104,    -1,   140,   104,    -1,   141,   104,    -1,
+     155,   104,    -1,   157,   104,    -1,   156,   104,    -1,   169,
+     104,    -1,   170,   104,    -1,   164,   104,    -1,   168,   104,
+      -1,   113,    -1,   114,   113,    -1,     9,    -1,   116,   105,
+     124,   106,    -1,     3,    -1,     4,    -1,     5,    -1,     6,
+      -1,     7,    -1,     8,    -1,    66,    -1,   115,    99,   115,
+      -1,   115,    98,   115,    -1,   115,   100,   115,    -1,   115,
+     101,   115,    -1,    98,   115,    -1,   105,   115,   106,    -1,
+     115,    95,   115,    -1,   115,    96,   115,    -1,   115,    97,
+     115,    -1,   115,    13,   115,    -1,   115,    14,   115,    -1,
+     115,    15,   115,    -1,   115,    10,   115,    -1,   115,    11,
+     115,    -1,    12,   115,    -1,     9,   103,    70,    -1,    66,
+     103,    70,    -1,    71,    -1,    72,    -1,    73,    -1,    74,
+      -1,    75,    -1,    77,    -1,    78,    -1,    79,    -1,    80,
+      -1,    83,    -1,    84,    -1,    -1,   107,    -1,   117,   108,
+     107,    -1,   109,     9,   105,   117,   106,   110,    -1,   120,
+     105,   124,   106,    -1,    76,    -1,    81,    -1,    82,    -1,
+       9,   105,   106,    -1,     9,    -1,   122,   108,     9,    -1,
+      -1,     9,    -1,   123,   108,     9,    -1,    -1,   115,    -1,
+     124,   108,   115,    -1,   115,    -1,    37,   105,   100,   106,
+      -1,    37,   105,    38,     9,   106,    -1,    36,   105,   115,
+     106,    -1,    -1,   125,    -1,   126,   108,   125,    -1,   100,
+      -1,   126,    49,   123,    -1,   126,    -1,    -1,    40,   115,
+      -1,    -1,    41,    51,    -1,    -1,    92,    17,    93,    94,
+      -1,    -1,    46,    -1,    47,    -1,    -1,    44,    45,     9,
+     131,    -1,    35,   127,    39,   122,   128,   129,   130,   132,
+      -1,    48,    49,     9,    -1,   134,    50,   105,   124,   106,
+      -1,   134,   133,    -1,     9,    95,   115,    -1,   136,    -1,
+     137,   108,   136,    -1,    40,    54,    55,     9,    -1,    51,
+       9,    52,   137,    -1,   139,   128,    -1,   139,   138,    -1,
+      53,    39,     9,    -1,   142,   128,    -1,   142,   138,    -1,
+      85,   133,    -1,     9,    63,   115,    -1,    31,   115,    29,
+     114,    -1,   147,    -1,   148,   147,    -1,    -1,    30,   114,
+      -1,   148,    -1,    28,   115,    29,   114,   149,    27,    28,
+      -1,    33,   115,    32,   114,    27,    32,    -1,    41,     9,
+      17,   115,    42,   115,    32,   114,    27,    32,    -1,    90,
+      -1,    34,    -1,    67,     9,    -1,    69,     9,    -1,    68,
+       9,    49,   123,    -1,    68,     9,    49,   121,    -1,     9,
+     171,   160,   161,   162,    -1,   158,    -1,   159,   108,   158,
+      -1,    -1,   105,     3,   106,    -1,    -1,    89,    -1,    -1,
+      12,     8,    -1,    -1,    61,    -1,    56,    57,     9,   105,
+     159,   106,   163,    -1,     9,    -1,   165,   108,     9,    -1,
+      -1,    59,    -1,    -1,    60,    -1,    56,   166,   167,    58,
+       9,    62,     9,   105,   165,   106,    -1,    86,    88,    -1,
+      87,    88,    -1,    21,    -1,    22,    -1,    24,    -1,    19,
+      -1,    20,    -1,     9,    17,   171,    -1,     9,    18,   171,
+      -1,    -1,   172,    -1,   173,   108,   172,    -1,     9,   171,
+     104,    -1,    -1,   174,    -1,   175,   174,    -1,    64,    65,
+       9,    25,   133,   104,    -1,    64,    91,     9,   104,    -1,
+     176,    -1,   177,    -1,    -1,   178,    -1,   179,   178,    -1,
+      16,     9,   105,   173,   106,    25,   175,   179,    26,   114,
+      27,    -1
+};
+
+/* YYRLINE[YYN] -- source line where rule number YYN was defined.  */
+static const unsigned short int yyrline[] =
+{
+       0,   138,   138,   141,   142,   143,   144,   145,   146,   147,
+     148,   149,   150,   151,   152,   153,   154,   155,   156,   157,
+     158,   159,   160,   161,   162,   166,   167,   172,   173,   175,
+     176,   177,   178,   179,   180,   181,   182,   183,   184,   185,
+     186,   187,   188,   189,   190,   191,   192,   193,   194,   195,
+     196,   197,   199,   204,   205,   206,   207,   209,   210,   211,
+     212,   213,   214,   215,   218,   220,   221,   225,   230,   235,
+     236,   237,   241,   245,   246,   251,   252,   253,   258,   259,
+     260,   264,   265,   270,   276,   283,   284,   285,   290,   292,
+     294,   298,   299,   303,   304,   309,   310,   315,   316,   317,
+     321,   322,   327,   337,   342,   344,   349,   353,   354,   359,
+     365,   372,   377,   382,   388,   393,   398,   403,   408,   414,
+     415,   420,   421,   423,   427,   434,   440,   448,   452,   456,
+     462,   468,   470,   475,   480,   481,   486,   487,   492,   493,
+     499,   500,   506,   507,   513,   519,   520,   525,   526,   530,
+     531,   535,   543,   548,   553,   554,   555,   556,   557,   561,
+     564,   570,   571,   572,   577,   581,   583,   584,   588,   594,
+     599,   600,   603,   605,   606,   610
+};
+#endif
+
+#if YYDEBUG || YYERROR_VERBOSE
+/* YYTNME[SYMBOL-NUM] -- String name of the symbol SYMBOL-NUM.
+   First, the terminals, then, starting at YYNTOKENS, nonterminals. */
+static const char *const yytname[] =
+{
+  "$end", "error", "$undefined", "PARS_INT_LIT", "PARS_FLOAT_LIT",
+  "PARS_STR_LIT", "PARS_FIXBINARY_LIT", "PARS_BLOB_LIT", "PARS_NULL_LIT",
+  "PARS_ID_TOKEN", "PARS_AND_TOKEN", "PARS_OR_TOKEN", "PARS_NOT_TOKEN",
+  "PARS_GE_TOKEN", "PARS_LE_TOKEN", "PARS_NE_TOKEN",
+  "PARS_PROCEDURE_TOKEN", "PARS_IN_TOKEN", "PARS_OUT_TOKEN",
+  "PARS_BINARY_TOKEN", "PARS_BLOB_TOKEN", "PARS_INT_TOKEN",
+  "PARS_INTEGER_TOKEN", "PARS_FLOAT_TOKEN", "PARS_CHAR_TOKEN",
+  "PARS_IS_TOKEN", "PARS_BEGIN_TOKEN", "PARS_END_TOKEN", "PARS_IF_TOKEN",
+  "PARS_THEN_TOKEN", "PARS_ELSE_TOKEN", "PARS_ELSIF_TOKEN",
+  "PARS_LOOP_TOKEN", "PARS_WHILE_TOKEN", "PARS_RETURN_TOKEN",
+  "PARS_SELECT_TOKEN", "PARS_SUM_TOKEN", "PARS_COUNT_TOKEN",
+  "PARS_DISTINCT_TOKEN", "PARS_FROM_TOKEN", "PARS_WHERE_TOKEN",
+  "PARS_FOR_TOKEN", "PARS_DDOT_TOKEN", "PARS_READ_TOKEN",
+  "PARS_ORDER_TOKEN", "PARS_BY_TOKEN", "PARS_ASC_TOKEN", "PARS_DESC_TOKEN",
+  "PARS_INSERT_TOKEN", "PARS_INTO_TOKEN", "PARS_VALUES_TOKEN",
+  "PARS_UPDATE_TOKEN", "PARS_SET_TOKEN", "PARS_DELETE_TOKEN",
+  "PARS_CURRENT_TOKEN", "PARS_OF_TOKEN", "PARS_CREATE_TOKEN",
+  "PARS_TABLE_TOKEN", "PARS_INDEX_TOKEN", "PARS_UNIQUE_TOKEN",
+  "PARS_CLUSTERED_TOKEN", "PARS_DOES_NOT_FIT_IN_MEM_TOKEN",
+  "PARS_ON_TOKEN", "PARS_ASSIGN_TOKEN", "PARS_DECLARE_TOKEN",
+  "PARS_CURSOR_TOKEN", "PARS_SQL_TOKEN", "PARS_OPEN_TOKEN",
+  "PARS_FETCH_TOKEN", "PARS_CLOSE_TOKEN", "PARS_NOTFOUND_TOKEN",
+  "PARS_TO_CHAR_TOKEN", "PARS_TO_NUMBER_TOKEN", "PARS_TO_BINARY_TOKEN",
+  "PARS_BINARY_TO_NUMBER_TOKEN", "PARS_SUBSTR_TOKEN", "PARS_REPLSTR_TOKEN",
+  "PARS_CONCAT_TOKEN", "PARS_INSTR_TOKEN", "PARS_LENGTH_TOKEN",
+  "PARS_SYSDATE_TOKEN", "PARS_PRINTF_TOKEN", "PARS_ASSERT_TOKEN",
+  "PARS_RND_TOKEN", "PARS_RND_STR_TOKEN", "PARS_ROW_PRINTF_TOKEN",
+  "PARS_COMMIT_TOKEN", "PARS_ROLLBACK_TOKEN", "PARS_WORK_TOKEN",
+  "PARS_UNSIGNED_TOKEN", "PARS_EXIT_TOKEN", "PARS_FUNCTION_TOKEN",
+  "PARS_LOCK_TOKEN", "PARS_SHARE_TOKEN", "PARS_MODE_TOKEN", "'='", "'<'",
+  "'>'", "'-'", "'+'", "'*'", "'/'", "NEG", "'%'", "';'", "'('", "')'",
+  "'?'", "','", "'{'", "'}'", "$accept", "top_statement", "statement",
+  "statement_list", "exp", "function_name", "question_mark_list",
+  "stored_procedure_call", "predefined_procedure_call",
+  "predefined_procedure_name", "user_function_call", "table_list",
+  "variable_list", "exp_list", "select_item", "select_item_list",
+  "select_list", "search_condition", "for_update_clause",
+  "lock_shared_clause", "order_direction", "order_by_clause",
+  "select_statement", "insert_statement_start", "insert_statement",
+  "column_assignment", "column_assignment_list", "cursor_positioned",
+  "update_statement_start", "update_statement_searched",
+  "update_statement_positioned", "delete_statement_start",
+  "delete_statement_searched", "delete_statement_positioned",
+  "row_printf_statement", "assignment_statement", "elsif_element",
+  "elsif_list", "else_part", "if_statement", "while_statement",
+  "for_statement", "exit_statement", "return_statement",
+  "open_cursor_statement", "close_cursor_statement", "fetch_statement",
+  "column_def", "column_def_list", "opt_column_len", "opt_unsigned",
+  "opt_not_null", "not_fit_in_memory", "create_table", "column_list",
+  "unique_def", "clustered_def", "create_index", "commit_statement",
+  "rollback_statement", "type_name", "parameter_declaration",
+  "parameter_declaration_list", "variable_declaration",
+  "variable_declaration_list", "cursor_declaration",
+  "function_declaration", "declaration", "declaration_list",
+  "procedure_definition", 0
+};
+#endif
+
+# ifdef YYPRINT
+/* YYTOKNUM[YYLEX-NUM] -- Internal token number corresponding to
+   token YYLEX-NUM.  */
+static const unsigned short int yytoknum[] =
+{
+       0,   256,   257,   258,   259,   260,   261,   262,   263,   264,
+     265,   266,   267,   268,   269,   270,   271,   272,   273,   274,
+     275,   276,   277,   278,   279,   280,   281,   282,   283,   284,
+     285,   286,   287,   288,   289,   290,   291,   292,   293,   294,
+     295,   296,   297,   298,   299,   300,   301,   302,   303,   304,
+     305,   306,   307,   308,   309,   310,   311,   312,   313,   314,
+     315,   316,   317,   318,   319,   320,   321,   322,   323,   324,
+     325,   326,   327,   328,   329,   330,   331,   332,   333,   334,
+     335,   336,   337,   338,   339,   340,   341,   342,   343,   344,
+     345,   346,   347,   348,   349,    61,    60,    62,    45,    43,
+      42,    47,   350,    37,    59,    40,    41,    63,    44,   123,
+     125
+};
+# endif
+
+/* YYR1[YYN] -- Symbol number of symbol that rule YYN derives.  */
+static const unsigned char yyr1[] =
+{
+       0,   111,   112,   113,   113,   113,   113,   113,   113,   113,
+     113,   113,   113,   113,   113,   113,   113,   113,   113,   113,
+     113,   113,   113,   113,   113,   114,   114,   115,   115,   115,
+     115,   115,   115,   115,   115,   115,   115,   115,   115,   115,
+     115,   115,   115,   115,   115,   115,   115,   115,   115,   115,
+     115,   115,   115,   116,   116,   116,   116,   116,   116,   116,
+     116,   116,   116,   116,   117,   117,   117,   118,   119,   120,
+     120,   120,   121,   122,   122,   123,   123,   123,   124,   124,
+     124,   125,   125,   125,   125,   126,   126,   126,   127,   127,
+     127,   128,   128,   129,   129,   130,   130,   131,   131,   131,
+     132,   132,   133,   134,   135,   135,   136,   137,   137,   138,
+     139,   140,   141,   142,   143,   144,   145,   146,   147,   148,
+     148,   149,   149,   149,   150,   151,   152,   153,   154,   155,
+     156,   157,   157,   158,   159,   159,   160,   160,   161,   161,
+     162,   162,   163,   163,   164,   165,   165,   166,   166,   167,
+     167,   168,   169,   170,   171,   171,   171,   171,   171,   172,
+     172,   173,   173,   173,   174,   175,   175,   175,   176,   177,
+     178,   178,   179,   179,   179,   180
+};
+
+/* YYR2[YYN] -- Number of symbols composing right hand side of rule YYN.  */
+static const unsigned char yyr2[] =
+{
+       0,     2,     2,     1,     2,     2,     2,     2,     2,     2,
+       2,     2,     2,     2,     2,     2,     2,     2,     2,     2,
+       2,     2,     2,     2,     2,     1,     2,     1,     4,     1,
+       1,     1,     1,     1,     1,     1,     3,     3,     3,     3,
+       2,     3,     3,     3,     3,     3,     3,     3,     3,     3,
+       2,     3,     3,     1,     1,     1,     1,     1,     1,     1,
+       1,     1,     1,     1,     0,     1,     3,     6,     4,     1,
+       1,     1,     3,     1,     3,     0,     1,     3,     0,     1,
+       3,     1,     4,     5,     4,     0,     1,     3,     1,     3,
+       1,     0,     2,     0,     2,     0,     4,     0,     1,     1,
+       0,     4,     8,     3,     5,     2,     3,     1,     3,     4,
+       4,     2,     2,     3,     2,     2,     2,     3,     4,     1,
+       2,     0,     2,     1,     7,     6,    10,     1,     1,     2,
+       2,     4,     4,     5,     1,     3,     0,     3,     0,     1,
+       0,     2,     0,     1,     7,     1,     3,     0,     1,     0,
+       1,    10,     2,     2,     1,     1,     1,     1,     1,     3,
+       3,     0,     1,     3,     3,     0,     1,     2,     6,     4,
+       1,     1,     0,     1,     2,    11
+};
+
+/* YYDEFACT[STATE-NAME] -- Default rule to reduce with in state
+   STATE-NUM when YYTABLE doesn't specify something else to do.  Zero
+   means the default is an error.  */
+static const unsigned char yydefact[] =
+{
+       0,     0,     0,     0,     0,     1,     2,   161,     0,   162,
+       0,     0,     0,     0,     0,   157,   158,   154,   155,   156,
+     159,   160,   165,   163,     0,   166,   172,     0,     0,   167,
+     170,   171,   173,     0,   164,     0,     0,     0,   174,     0,
+       0,     0,     0,     0,   128,    85,     0,     0,     0,     0,
+     147,     0,     0,     0,    69,    70,    71,     0,     0,     0,
+     127,     0,    25,     0,     3,     0,     0,     0,     0,     0,
+      91,     0,     0,    91,     0,     0,     0,     0,     0,     0,
+       0,     0,     0,     0,     0,     0,     0,     0,     0,     0,
+       0,   169,     0,    29,    30,    31,    32,    33,    34,    27,
+       0,    35,    53,    54,    55,    56,    57,    58,    59,    60,
+      61,    62,    63,     0,     0,     0,     0,     0,     0,     0,
+      88,    81,    86,    90,     0,     0,     0,     0,     0,     0,
+     148,   149,   129,     0,   130,   116,   152,   153,     0,   175,
+      26,     4,    78,    11,     0,   105,    12,     0,   111,   112,
+      16,    17,   114,   115,    14,    15,    13,    10,     8,     5,
+       6,     7,     9,    18,    20,    19,    23,    24,    21,    22,
+       0,   117,     0,    50,     0,    40,     0,     0,     0,     0,
+       0,     0,     0,     0,     0,     0,     0,     0,     0,     0,
+      78,     0,     0,     0,    75,     0,     0,     0,   103,     0,
+     113,     0,   150,     0,    75,    64,    79,     0,    78,     0,
+      92,   168,    51,    52,    41,    48,    49,    45,    46,    47,
+     121,    42,    43,    44,    37,    36,    38,    39,     0,     0,
+       0,     0,     0,    76,    89,    87,    73,    91,     0,     0,
+     107,   110,     0,     0,    76,   132,   131,    65,     0,    68,
+       0,     0,     0,     0,     0,   119,   123,     0,    28,     0,
+      84,     0,    82,     0,     0,     0,    93,     0,     0,     0,
+       0,   134,     0,     0,     0,     0,     0,    80,   104,   109,
+     122,     0,   120,     0,   125,    83,    77,    74,     0,    95,
+       0,   106,   108,   136,   142,     0,     0,    72,    67,    66,
+       0,   124,    94,     0,   100,     0,     0,   138,   143,   144,
+     135,     0,   118,     0,     0,   102,     0,     0,   139,   140,
+       0,     0,     0,     0,   137,     0,   133,   145,     0,    96,
+      97,   126,   141,   151,     0,    98,    99,   101,   146
+};
+
+/* YYDEFGOTO[NTERM-NUM]. */
+static const short int yydefgoto[] =
+{
+      -1,     2,    62,    63,   206,   116,   248,    64,    65,    66,
+     245,   237,   234,   207,   122,   123,   124,   148,   289,   304,
+     337,   315,    67,    68,    69,   240,   241,   149,    70,    71,
+      72,    73,    74,    75,    76,    77,   255,   256,   257,    78,
+      79,    80,    81,    82,    83,    84,    85,   271,   272,   307,
+     319,   326,   309,    86,   328,   131,   203,    87,    88,    89,
+      20,     9,    10,    25,    26,    30,    31,    32,    33,     3
+};
+
+/* YYPACT[STATE-NUM] -- Index in YYTABLE of the portion describing
+   STATE-NUM.  */
+#define YYPACT_NINF -177
+static const short int yypact[] =
+{
+      28,    38,    54,   -46,   -29,  -177,  -177,    56,    50,  -177,
+     -75,     8,     8,    46,    56,  -177,  -177,  -177,  -177,  -177,
+    -177,  -177,    63,  -177,     8,  -177,     2,   -26,   -51,  -177,
+    -177,  -177,  -177,   -13,  -177,    71,    72,   587,  -177,    57,
+     -21,    26,   272,   272,  -177,    13,    91,    55,    96,    67,
+     -22,    99,   100,   103,  -177,  -177,  -177,    75,    29,    35,
+    -177,   116,  -177,   396,  -177,    22,    23,    27,    -9,    30,
+      87,    31,    32,    87,    47,    49,    52,    58,    59,    60,
+      61,    62,    65,    66,    74,    77,    78,    86,    89,   102,
+      75,  -177,   272,  -177,  -177,  -177,  -177,  -177,  -177,    39,
+     272,    51,  -177,  -177,  -177,  -177,  -177,  -177,  -177,  -177,
+    -177,  -177,  -177,   272,   272,   361,    25,   489,    45,    90,
+    -177,   651,  -177,   -39,    93,   142,   124,   108,   152,   170,
+    -177,   131,  -177,   143,  -177,  -177,  -177,  -177,    98,  -177,
+    -177,  -177,   272,  -177,   110,  -177,  -177,   256,  -177,  -177,
+    -177,  -177,  -177,  -177,  -177,  -177,  -177,  -177,  -177,  -177,
+    -177,  -177,  -177,  -177,  -177,  -177,  -177,  -177,  -177,  -177,
+     112,   651,   137,   101,   147,   204,    88,   272,   272,   272,
+     272,   272,   587,   272,   272,   272,   272,   272,   272,   272,
+     272,   587,   272,   -30,   211,   168,   212,   272,  -177,   213,
+    -177,   118,  -177,   167,   217,   122,   651,   -63,   272,   175,
+     651,  -177,  -177,  -177,  -177,   101,   101,    21,    21,   651,
+     332,    21,    21,    21,    -6,    -6,   204,   204,   -60,   460,
+     198,   222,   126,  -177,   125,  -177,  -177,   -33,   584,   140,
+    -177,   128,   228,   229,   139,  -177,   125,  -177,   -53,  -177,
+     272,   -49,   240,   587,   272,  -177,   224,   226,  -177,   225,
+    -177,   150,  -177,   258,   272,   260,   230,   272,   272,   213,
+       8,  -177,   -45,   208,   166,   164,   176,   651,  -177,  -177,
+     587,   631,  -177,   254,  -177,  -177,  -177,  -177,   234,   194,
+     638,   651,  -177,   182,   227,   228,   280,  -177,  -177,  -177,
+     587,  -177,  -177,   273,   247,   587,   289,   214,  -177,  -177,
+    -177,   195,   587,   209,   261,  -177,   524,   199,  -177,   295,
+     292,   215,   299,   279,  -177,   304,  -177,  -177,   -44,  -177,
+      -8,  -177,  -177,  -177,   305,  -177,  -177,  -177,  -177
+};
+
+/* YYPGOTO[NTERM-NUM].  */
+static const short int yypgoto[] =
+{
+    -177,  -177,   -62,  -176,   -40,  -177,  -177,  -177,  -177,  -177,
+    -177,  -177,   109,  -166,   120,  -177,  -177,   -69,  -177,  -177,
+    -177,  -177,   -34,  -177,  -177,    48,  -177,   243,  -177,  -177,
+    -177,  -177,  -177,  -177,  -177,  -177,    64,  -177,  -177,  -177,
+    -177,  -177,  -177,  -177,  -177,  -177,  -177,    24,  -177,  -177,
+    -177,  -177,  -177,  -177,  -177,  -177,  -177,  -177,  -177,  -177,
+     -12,   307,  -177,   297,  -177,  -177,  -177,   285,  -177,  -177
+};
+
+/* YYTABLE[YYPACT[STATE-NUM]].  What to do in state STATE-NUM.  If
+   positive, shift that token.  If negative, reduce the rule which
+   number is the opposite.  If zero, do what YYDEFACT says.
+   If YYTABLE_NINF, syntax error.  */
+#define YYTABLE_NINF -1
+static const unsigned short int yytable[] =
+{
+      21,   140,   115,   117,   152,   121,   220,   264,   231,   181,
+     194,    24,    27,    37,    35,   229,    93,    94,    95,    96,
+      97,    98,    99,   135,   228,   100,    45,    15,    16,    17,
+      18,    13,    19,    14,   145,   129,   181,   130,   335,   336,
+      36,   144,   251,   249,     1,   250,   258,     4,   250,   118,
+     119,    28,   171,   275,     5,   276,   170,   278,     6,   250,
+     173,   294,   333,   295,   334,     8,    28,    11,    12,   195,
+     232,    22,    24,   175,   176,   265,     7,   280,    34,   101,
+      39,    40,    90,    91,   102,   103,   104,   105,   106,    92,
+     107,   108,   109,   110,   188,   189,   111,   112,   177,   178,
+     125,   179,   180,   181,   126,   127,   128,   210,   132,   133,
+      45,   113,   134,   120,   179,   180,   181,   136,   114,   186,
+     187,   188,   189,   137,   312,   138,   141,   147,   142,   316,
+     190,   143,   196,   198,   146,   150,   151,   215,   216,   217,
+     218,   219,   172,   221,   222,   223,   224,   225,   226,   227,
+     192,   154,   230,   155,   174,   121,   156,   238,   140,   197,
+     199,   200,   157,   158,   159,   160,   161,   140,   266,   162,
+     163,    93,    94,    95,    96,    97,    98,    99,   164,   201,
+     100,   165,   166,   183,   184,   185,   186,   187,   188,   189,
+     167,   202,   204,   168,   214,   193,   183,   184,   185,   186,
+     187,   188,   189,   205,   118,   119,   169,   212,   177,   178,
+     277,   179,   180,   181,   281,   208,   211,   213,   140,   181,
+     233,   236,   239,   242,   210,   243,   244,   290,   291,   247,
+     252,   261,   262,   263,   101,   268,   269,   270,   273,   102,
+     103,   104,   105,   106,   274,   107,   108,   109,   110,   279,
+     140,   111,   112,   283,   140,   254,   285,   284,   293,    93,
+      94,    95,    96,    97,    98,    99,   113,   286,   100,   287,
+     296,   288,   297,   114,   298,    93,    94,    95,    96,    97,
+      98,    99,   301,   299,   100,   302,   303,   306,   308,   311,
+     313,   314,   317,   183,   184,   185,   186,   187,   188,   189,
+     320,   327,   321,   318,   260,   324,   322,   325,   330,   329,
+     209,   331,   332,   246,   338,   235,   153,   292,    38,   310,
+     282,    23,   101,    29,     0,     0,     0,   102,   103,   104,
+     105,   106,     0,   107,   108,   109,   110,     0,   101,   111,
+     112,    41,     0,   102,   103,   104,   105,   106,     0,   107,
+     108,   109,   110,     0,   113,   111,   112,     0,     0,     0,
+      42,   114,   253,   254,     0,    43,    44,    45,     0,     0,
+     113,   177,   178,    46,   179,   180,   181,   114,     0,     0,
+      47,     0,     0,    48,     0,    49,     0,     0,    50,     0,
+     182,     0,     0,     0,     0,     0,     0,     0,     0,    51,
+      52,    53,     0,     0,     0,    41,     0,     0,    54,     0,
+       0,     0,     0,    55,    56,     0,     0,    57,    58,    59,
+       0,     0,    60,   139,    42,     0,     0,     0,     0,    43,
+      44,    45,     0,     0,     0,     0,     0,    46,     0,     0,
+       0,    61,     0,     0,    47,     0,     0,    48,     0,    49,
+       0,     0,    50,     0,     0,     0,   183,   184,   185,   186,
+     187,   188,   189,    51,    52,    53,     0,     0,     0,    41,
+       0,     0,    54,     0,     0,     0,     0,    55,    56,     0,
+       0,    57,    58,    59,     0,     0,    60,   259,    42,     0,
+       0,     0,     0,    43,    44,    45,     0,     0,     0,   177,
+     178,    46,   179,   180,   181,    61,     0,     0,    47,     0,
+       0,    48,     0,    49,     0,     0,    50,     0,     0,     0,
+       0,   191,     0,     0,     0,     0,     0,    51,    52,    53,
+       0,     0,     0,    41,     0,     0,    54,     0,     0,     0,
+       0,    55,    56,     0,     0,    57,    58,    59,     0,     0,
+      60,   323,    42,     0,     0,     0,     0,    43,    44,    45,
+       0,     0,     0,     0,     0,    46,     0,     0,     0,    61,
+       0,     0,    47,     0,     0,    48,     0,    49,     0,     0,
+      50,     0,     0,     0,   183,   184,   185,   186,   187,   188,
+     189,    51,    52,    53,   177,   178,    41,   179,   180,   181,
+      54,     0,     0,     0,     0,    55,    56,     0,     0,    57,
+      58,    59,     0,     0,    60,    42,     0,     0,     0,     0,
+      43,    44,    45,     0,     0,     0,   267,     0,    46,     0,
+       0,     0,     0,    61,     0,    47,     0,     0,    48,     0,
+      49,   177,   178,    50,   179,   180,   181,     0,   177,   178,
+       0,   179,   180,   181,    51,    52,    53,     0,     0,     0,
+     300,   177,   178,    54,   179,   180,   181,     0,    55,    56,
+     305,     0,    57,    58,    59,     0,     0,    60,     0,   183,
+     184,   185,   186,   187,   188,   189,     0,     0,     0,     0,
+       0,     0,     0,     0,     0,     0,    61,     0,     0,     0,
+       0,     0,     0,     0,     0,     0,     0,     0,     0,     0,
+       0,     0,     0,     0,     0,     0,     0,     0,     0,     0,
+       0,     0,     0,     0,     0,     0,   183,   184,   185,   186,
+     187,   188,   189,   183,   184,   185,   186,   187,   188,   189,
+       0,     0,     0,     0,     0,     0,   183,   184,   185,   186,
+     187,   188,   189
+};
+
+static const short int yycheck[] =
+{
+      12,    63,    42,    43,    73,    45,   182,    40,    38,    15,
+      49,     9,    24,    26,    65,   191,     3,     4,     5,     6,
+       7,     8,     9,    57,   190,    12,    35,    19,    20,    21,
+      22,   106,    24,   108,    68,    57,    15,    59,    46,    47,
+      91,    50,   208,   106,    16,   108,   106,     9,   108,    36,
+      37,    64,    92,   106,     0,   108,    90,   106,   104,   108,
+     100,   106,   106,   108,   108,     9,    64,    17,    18,   108,
+     100,    25,     9,   113,   114,   108,   105,   253,   104,    66,
+       9,     9,    25,   104,    71,    72,    73,    74,    75,    63,
+      77,    78,    79,    80,   100,   101,    83,    84,    10,    11,
+       9,    13,    14,    15,    49,     9,    39,   147,     9,     9,
+      35,    98,     9,   100,    13,    14,    15,    88,   105,    98,
+      99,   100,   101,    88,   300,     9,   104,    40,   105,   305,
+     105,   104,    39,     9,   104,   104,   104,   177,   178,   179,
+     180,   181,   103,   183,   184,   185,   186,   187,   188,   189,
+     105,   104,   192,   104,   103,   195,   104,   197,   220,    17,
+      52,     9,   104,   104,   104,   104,   104,   229,   237,   104,
+     104,     3,     4,     5,     6,     7,     8,     9,   104,     9,
+      12,   104,   104,    95,    96,    97,    98,    99,   100,   101,
+     104,    60,    49,   104,   106,   105,    95,    96,    97,    98,
+      99,   100,   101,   105,    36,    37,   104,    70,    10,    11,
+     250,    13,    14,    15,   254,   105,   104,    70,   280,    15,
+       9,     9,     9,   105,   264,    58,     9,   267,   268,   107,
+      55,     9,   106,   108,    66,    95,   108,     9,     9,    71,
+      72,    73,    74,    75,   105,    77,    78,    79,    80,     9,
+     312,    83,    84,    27,   316,    31,   106,    32,   270,     3,
+       4,     5,     6,     7,     8,     9,    98,     9,    12,     9,
+      62,    41,   106,   105,   110,     3,     4,     5,     6,     7,
+       8,     9,    28,   107,    12,    51,    92,   105,    61,     9,
+      17,    44,     3,    95,    96,    97,    98,    99,   100,   101,
+     105,     9,    93,    89,   106,   106,    45,    12,     9,    94,
+      54,    32,     8,   204,     9,   195,    73,   269,    33,   295,
+     256,    14,    66,    26,    -1,    -1,    -1,    71,    72,    73,
+      74,    75,    -1,    77,    78,    79,    80,    -1,    66,    83,
+      84,     9,    -1,    71,    72,    73,    74,    75,    -1,    77,
+      78,    79,    80,    -1,    98,    83,    84,    -1,    -1,    -1,
+      28,   105,    30,    31,    -1,    33,    34,    35,    -1,    -1,
+      98,    10,    11,    41,    13,    14,    15,   105,    -1,    -1,
+      48,    -1,    -1,    51,    -1,    53,    -1,    -1,    56,    -1,
+      29,    -1,    -1,    -1,    -1,    -1,    -1,    -1,    -1,    67,
+      68,    69,    -1,    -1,    -1,     9,    -1,    -1,    76,    -1,
+      -1,    -1,    -1,    81,    82,    -1,    -1,    85,    86,    87,
+      -1,    -1,    90,    27,    28,    -1,    -1,    -1,    -1,    33,
+      34,    35,    -1,    -1,    -1,    -1,    -1,    41,    -1,    -1,
+      -1,   109,    -1,    -1,    48,    -1,    -1,    51,    -1,    53,
+      -1,    -1,    56,    -1,    -1,    -1,    95,    96,    97,    98,
+      99,   100,   101,    67,    68,    69,    -1,    -1,    -1,     9,
+      -1,    -1,    76,    -1,    -1,    -1,    -1,    81,    82,    -1,
+      -1,    85,    86,    87,    -1,    -1,    90,    27,    28,    -1,
+      -1,    -1,    -1,    33,    34,    35,    -1,    -1,    -1,    10,
+      11,    41,    13,    14,    15,   109,    -1,    -1,    48,    -1,
+      -1,    51,    -1,    53,    -1,    -1,    56,    -1,    -1,    -1,
+      -1,    32,    -1,    -1,    -1,    -1,    -1,    67,    68,    69,
+      -1,    -1,    -1,     9,    -1,    -1,    76,    -1,    -1,    -1,
+      -1,    81,    82,    -1,    -1,    85,    86,    87,    -1,    -1,
+      90,    27,    28,    -1,    -1,    -1,    -1,    33,    34,    35,
+      -1,    -1,    -1,    -1,    -1,    41,    -1,    -1,    -1,   109,
+      -1,    -1,    48,    -1,    -1,    51,    -1,    53,    -1,    -1,
+      56,    -1,    -1,    -1,    95,    96,    97,    98,    99,   100,
+     101,    67,    68,    69,    10,    11,     9,    13,    14,    15,
+      76,    -1,    -1,    -1,    -1,    81,    82,    -1,    -1,    85,
+      86,    87,    -1,    -1,    90,    28,    -1,    -1,    -1,    -1,
+      33,    34,    35,    -1,    -1,    -1,    42,    -1,    41,    -1,
+      -1,    -1,    -1,   109,    -1,    48,    -1,    -1,    51,    -1,
+      53,    10,    11,    56,    13,    14,    15,    -1,    10,    11,
+      -1,    13,    14,    15,    67,    68,    69,    -1,    -1,    -1,
+      29,    10,    11,    76,    13,    14,    15,    -1,    81,    82,
+      32,    -1,    85,    86,    87,    -1,    -1,    90,    -1,    95,
+      96,    97,    98,    99,   100,   101,    -1,    -1,    -1,    -1,
+      -1,    -1,    -1,    -1,    -1,    -1,   109,    -1,    -1,    -1,
+      -1,    -1,    -1,    -1,    -1,    -1,    -1,    -1,    -1,    -1,
+      -1,    -1,    -1,    -1,    -1,    -1,    -1,    -1,    -1,    -1,
+      -1,    -1,    -1,    -1,    -1,    -1,    95,    96,    97,    98,
+      99,   100,   101,    95,    96,    97,    98,    99,   100,   101,
+      -1,    -1,    -1,    -1,    -1,    -1,    95,    96,    97,    98,
+      99,   100,   101
+};
+
+/* YYSTOS[STATE-NUM] -- The (internal number of the) accessing
+   symbol of state STATE-NUM.  */
+static const unsigned char yystos[] =
+{
+       0,    16,   112,   180,     9,     0,   104,   105,     9,   172,
+     173,    17,    18,   106,   108,    19,    20,    21,    22,    24,
+     171,   171,    25,   172,     9,   174,   175,   171,    64,   174,
+     176,   177,   178,   179,   104,    65,    91,    26,   178,     9,
+       9,     9,    28,    33,    34,    35,    41,    48,    51,    53,
+      56,    67,    68,    69,    76,    81,    82,    85,    86,    87,
+      90,   109,   113,   114,   118,   119,   120,   133,   134,   135,
+     139,   140,   141,   142,   143,   144,   145,   146,   150,   151,
+     152,   153,   154,   155,   156,   157,   164,   168,   169,   170,
+      25,   104,    63,     3,     4,     5,     6,     7,     8,     9,
+      12,    66,    71,    72,    73,    74,    75,    77,    78,    79,
+      80,    83,    84,    98,   105,   115,   116,   115,    36,    37,
+     100,   115,   125,   126,   127,     9,    49,     9,    39,    57,
+      59,   166,     9,     9,     9,   133,    88,    88,     9,    27,
+     113,   104,   105,   104,    50,   133,   104,    40,   128,   138,
+     104,   104,   128,   138,   104,   104,   104,   104,   104,   104,
+     104,   104,   104,   104,   104,   104,   104,   104,   104,   104,
+     133,   115,   103,   115,   103,   115,   115,    10,    11,    13,
+      14,    15,    29,    95,    96,    97,    98,    99,   100,   101,
+     105,    32,   105,   105,    49,   108,    39,    17,     9,    52,
+       9,     9,    60,   167,    49,   105,   115,   124,   105,    54,
+     115,   104,    70,    70,   106,   115,   115,   115,   115,   115,
+     114,   115,   115,   115,   115,   115,   115,   115,   124,   114,
+     115,    38,   100,     9,   123,   125,     9,   122,   115,     9,
+     136,   137,   105,    58,     9,   121,   123,   107,   117,   106,
+     108,   124,    55,    30,    31,   147,   148,   149,   106,    27,
+     106,     9,   106,   108,    40,   108,   128,    42,    95,   108,
+       9,   158,   159,     9,   105,   106,   108,   115,   106,     9,
+     114,   115,   147,    27,    32,   106,     9,     9,    41,   129,
+     115,   115,   136,   171,   106,   108,    62,   106,   110,   107,
+      29,    28,    51,    92,   130,    32,   105,   160,    61,   163,
+     158,     9,   114,    17,    44,   132,   114,     3,    89,   161,
+     105,    93,    45,    27,   106,    12,   162,     9,   165,    94,
+       9,    32,     8,   106,   108,    46,    47,   131,     9
+};
+
+#if ! defined (YYSIZE_T) && defined (__SIZE_TYPE__)
+# define YYSIZE_T __SIZE_TYPE__
+#endif
+#if ! defined (YYSIZE_T) && defined (size_t)
+# define YYSIZE_T size_t
+#endif
+#if ! defined (YYSIZE_T)
+# if defined (__STDC__) || defined (__cplusplus)
+#  include <stddef.h> /* INFRINGES ON USER NAME SPACE */
+#  define YYSIZE_T size_t
+# endif
+#endif
+#if ! defined (YYSIZE_T)
+# define YYSIZE_T unsigned int
+#endif
+
+#define yyerrok		(yyerrstatus = 0)
+#define yyclearin	(yychar = YYEMPTY)
+#define YYEMPTY		(-2)
+#define YYEOF		0
+
+#define YYACCEPT	goto yyacceptlab
+#define YYABORT		goto yyabortlab
+#define YYERROR		goto yyerrorlab
+
+
+/* Like YYERROR except do call yyerror.  This remains here temporarily
+   to ease the transition to the new meaning of YYERROR, for GCC.
+   Once GCC version 2 has supplanted version 1, this can go.  */
+
+#define YYFAIL		goto yyerrlab
+
+#define YYRECOVERING()  (!!yyerrstatus)
+
+#define YYBACKUP(Token, Value)					\
+do								\
+  if (yychar == YYEMPTY && yylen == 1)				\
+    {								\
+      yychar = (Token);						\
+      yylval = (Value);						\
+      yytoken = YYTRANSLATE (yychar);				\
+      YYPOPSTACK;						\
+      goto yybackup;						\
+    }								\
+  else								\
+    { 								\
+      yyerror ("syntax error: cannot back up");\
+      YYERROR;							\
+    }								\
+while (0)
+
+
+#define YYTERROR	1
+#define YYERRCODE	256
+
+
+/* YYLLOC_DEFAULT -- Set CURRENT to span from RHS[1] to RHS[N].
+   If N is 0, then set CURRENT to the empty location which ends
+   the previous symbol: RHS[0] (always defined).  */
+
+#define YYRHSLOC(Rhs, K) ((Rhs)[K])
+#ifndef YYLLOC_DEFAULT
+# define YYLLOC_DEFAULT(Current, Rhs, N)				\
+    do									\
+      if (N)								\
+	{								\
+	  (Current).first_line   = YYRHSLOC (Rhs, 1).first_line;	\
+	  (Current).first_column = YYRHSLOC (Rhs, 1).first_column;	\
+	  (Current).last_line    = YYRHSLOC (Rhs, N).last_line;		\
+	  (Current).last_column  = YYRHSLOC (Rhs, N).last_column;	\
+	}								\
+      else								\
+	{								\
+	  (Current).first_line   = (Current).last_line   =		\
+	    YYRHSLOC (Rhs, 0).last_line;				\
+	  (Current).first_column = (Current).last_column =		\
+	    YYRHSLOC (Rhs, 0).last_column;				\
+	}								\
+    while (0)
+#endif
+
+
+/* YY_LOCATION_PRINT -- Print the location on the stream.
+   This macro was not mandated originally: define only if we know
+   we won't break user code: when these are the locations we know.  */
+
+#ifndef YY_LOCATION_PRINT
+# if YYLTYPE_IS_TRIVIAL
+#  define YY_LOCATION_PRINT(File, Loc)			\
+     fprintf (File, "%d.%d-%d.%d",			\
+              (Loc).first_line, (Loc).first_column,	\
+              (Loc).last_line,  (Loc).last_column)
+# else
+#  define YY_LOCATION_PRINT(File, Loc) ((void) 0)
+# endif
+#endif
+
+
+/* YYLEX -- calling `yylex' with the right arguments.  */
+
+#ifdef YYLEX_PARAM
+# define YYLEX yylex (YYLEX_PARAM)
+#else
+# define YYLEX yylex ()
+#endif
+
+/* Enable debugging if requested.  */
+#if YYDEBUG
+
+# ifndef YYFPRINTF
+#  include <stdio.h> /* INFRINGES ON USER NAME SPACE */
+#  define YYFPRINTF fprintf
+# endif
+
+# define YYDPRINTF(Args)			\
+do {						\
+  if (yydebug)					\
+    YYFPRINTF Args;				\
+} while (0)
+
+# define YY_SYMBOL_PRINT(Title, Type, Value, Location)		\
+do {								\
+  if (yydebug)							\
+    {								\
+      YYFPRINTF (stderr, "%s ", Title);				\
+      yysymprint (stderr, 					\
+                  Type, Value);	\
+      YYFPRINTF (stderr, "\n");					\
+    }								\
+} while (0)
+
+/*------------------------------------------------------------------.
+| yy_stack_print -- Print the state stack from its BOTTOM up to its |
+| TOP (included).                                                   |
+`------------------------------------------------------------------*/
+
+#if defined (__STDC__) || defined (__cplusplus)
+static void
+yy_stack_print (short int *bottom, short int *top)
+#else
+static void
+yy_stack_print (bottom, top)
+    short int *bottom;
+    short int *top;
+#endif
+{
+  YYFPRINTF (stderr, "Stack now");
+  for (/* Nothing. */; bottom <= top; ++bottom)
+    YYFPRINTF (stderr, " %d", *bottom);
+  YYFPRINTF (stderr, "\n");
+}
+
+# define YY_STACK_PRINT(Bottom, Top)				\
+do {								\
+  if (yydebug)							\
+    yy_stack_print ((Bottom), (Top));				\
+} while (0)
+
+
+/*------------------------------------------------.
+| Report that the YYRULE is going to be reduced.  |
+`------------------------------------------------*/
+
+#if defined (__STDC__) || defined (__cplusplus)
+static void
+yy_reduce_print (int yyrule)
+#else
+static void
+yy_reduce_print (yyrule)
+    int yyrule;
+#endif
+{
+  int yyi;
+  unsigned int yylno = yyrline[yyrule];
+  YYFPRINTF (stderr, "Reducing stack by rule %d (line %u), ",
+             yyrule - 1, yylno);
+  /* Print the symbols being reduced, and their result.  */
+  for (yyi = yyprhs[yyrule]; 0 <= yyrhs[yyi]; yyi++)
+    YYFPRINTF (stderr, "%s ", yytname [yyrhs[yyi]]);
+  YYFPRINTF (stderr, "-> %s\n", yytname [yyr1[yyrule]]);
+}
+
+# define YY_REDUCE_PRINT(Rule)		\
+do {					\
+  if (yydebug)				\
+    yy_reduce_print (Rule);		\
+} while (0)
+
+/* Nonzero means print parse trace.  It is left uninitialized so that
+   multiple parsers can coexist.  */
+int yydebug;
+#else /* !YYDEBUG */
+# define YYDPRINTF(Args)
+# define YY_SYMBOL_PRINT(Title, Type, Value, Location)
+# define YY_STACK_PRINT(Bottom, Top)
+# define YY_REDUCE_PRINT(Rule)
+#endif /* !YYDEBUG */
+
+
+/* YYINITDEPTH -- initial size of the parser's stacks.  */
+#ifndef	YYINITDEPTH
+# define YYINITDEPTH 200
+#endif
+
+/* YYMAXDEPTH -- maximum size the stacks can grow to (effective only
+   if the built-in stack extension method is used).
+
+   Do not make this value too large; the results are undefined if
+   SIZE_MAX < YYSTACK_BYTES (YYMAXDEPTH)
+   evaluated with infinite-precision integer arithmetic.  */
+
+#ifndef YYMAXDEPTH
+# define YYMAXDEPTH 10000
+#endif
+
+
+
+#if YYERROR_VERBOSE
+
+# ifndef yystrlen
+#  if defined (__GLIBC__) && defined (_STRING_H)
+#   define yystrlen strlen
+#  else
+/* Return the length of YYSTR.  */
+static YYSIZE_T
+#   if defined (__STDC__) || defined (__cplusplus)
+yystrlen (const char *yystr)
+#   else
+yystrlen (yystr)
+     const char *yystr;
+#   endif
+{
+  register const char *yys = yystr;
+
+  while (*yys++ != '\0')
+    continue;
+
+  return yys - yystr - 1;
+}
+#  endif
+# endif
+
+# ifndef yystpcpy
+#  if defined (__GLIBC__) && defined (_STRING_H) && defined (_GNU_SOURCE)
+#   define yystpcpy stpcpy
+#  else
+/* Copy YYSRC to YYDEST, returning the address of the terminating '\0' in
+   YYDEST.  */
+static char *
+#   if defined (__STDC__) || defined (__cplusplus)
+yystpcpy (char *yydest, const char *yysrc)
+#   else
+yystpcpy (yydest, yysrc)
+     char *yydest;
+     const char *yysrc;
+#   endif
+{
+  register char *yyd = yydest;
+  register const char *yys = yysrc;
+
+  while ((*yyd++ = *yys++) != '\0')
+    continue;
+
+  return yyd - 1;
+}
+#  endif
+# endif
+
+#endif /* !YYERROR_VERBOSE */
+
+
+
+#if YYDEBUG
+/*--------------------------------.
+| Print this symbol on YYOUTPUT.  |
+`--------------------------------*/
+
+#if defined (__STDC__) || defined (__cplusplus)
+static void
+yysymprint (FILE *yyoutput, int yytype, YYSTYPE *yyvaluep)
+#else
+static void
+yysymprint (yyoutput, yytype, yyvaluep)
+    FILE *yyoutput;
+    int yytype;
+    YYSTYPE *yyvaluep;
+#endif
+{
+  /* Pacify ``unused variable'' warnings.  */
+  (void) yyvaluep;
+
+  if (yytype < YYNTOKENS)
+    YYFPRINTF (yyoutput, "token %s (", yytname[yytype]);
+  else
+    YYFPRINTF (yyoutput, "nterm %s (", yytname[yytype]);
+
+
+# ifdef YYPRINT
+  if (yytype < YYNTOKENS)
+    YYPRINT (yyoutput, yytoknum[yytype], *yyvaluep);
+# endif
+  switch (yytype)
+    {
+      default:
+        break;
+    }
+  YYFPRINTF (yyoutput, ")");
+}
+
+#endif /* ! YYDEBUG */
+/*-----------------------------------------------.
+| Release the memory associated to this symbol.  |
+`-----------------------------------------------*/
+
+#if defined (__STDC__) || defined (__cplusplus)
+static void
+yydestruct (const char *yymsg, int yytype, YYSTYPE *yyvaluep)
+#else
+static void
+yydestruct (yymsg, yytype, yyvaluep)
+    const char *yymsg;
+    int yytype;
+    YYSTYPE *yyvaluep;
+#endif
+{
+  /* Pacify ``unused variable'' warnings.  */
+  (void) yyvaluep;
+
+  if (!yymsg)
+    yymsg = "Deleting";
+  YY_SYMBOL_PRINT (yymsg, yytype, yyvaluep, yylocationp);
+
+  switch (yytype)
+    {
+
+      default:
+        break;
+    }
+}
+
+
+/* Prevent warnings from -Wmissing-prototypes.  */
+
+#ifdef YYPARSE_PARAM
+# if defined (__STDC__) || defined (__cplusplus)
+UNIV_INTERN int yyparse (void *YYPARSE_PARAM);
+# else
+UNIV_INTERN int yyparse ();
+# endif
+#else /* ! YYPARSE_PARAM */
+#if defined (__STDC__) || defined (__cplusplus)
+UNIV_INTERN int yyparse (void);
+#else
+UNIV_INTERN int yyparse ();
+#endif
+#endif /* ! YYPARSE_PARAM */
+
+
+
+/* The look-ahead symbol.  */
+static int yychar;
+
+/* The semantic value of the look-ahead symbol.  */
+UNIV_INTERN YYSTYPE yylval;
+
+/* Number of syntax errors so far.  */
+static int yynerrs;
+
+
+
+/*----------.
+| yyparse.  |
+`----------*/
+
+#ifdef YYPARSE_PARAM
+# if defined (__STDC__) || defined (__cplusplus)
+UNIV_INTERN int yyparse (void *YYPARSE_PARAM)
+# else
+UNIV_INTERN int yyparse (YYPARSE_PARAM)
+  void *YYPARSE_PARAM;
+# endif
+#else /* ! YYPARSE_PARAM */
+#if defined (__STDC__) || defined (__cplusplus)
+int
+yyparse (void)
+#else
+int
+yyparse ()
+
+#endif
+#endif
+{
+  
+  register int yystate;
+  register int yyn;
+  int yyresult;
+  /* Number of tokens to shift before error messages enabled.  */
+  int yyerrstatus;
+  /* Look-ahead token as an internal (translated) token number.  */
+  int yytoken = 0;
+
+  /* Three stacks and their tools:
+     `yyss': related to states,
+     `yyvs': related to semantic values,
+     `yyls': related to locations.
+
+     Refer to the stacks thru separate pointers, to allow yyoverflow
+     to reallocate them elsewhere.  */
+
+  /* The state stack.  */
+  short int yyssa[YYINITDEPTH];
+  short int *yyss = yyssa;
+  register short int *yyssp;
+
+  /* The semantic value stack.  */
+  YYSTYPE yyvsa[YYINITDEPTH];
+  YYSTYPE *yyvs = yyvsa;
+  register YYSTYPE *yyvsp;
+
+
+
+#define YYPOPSTACK   (yyvsp--, yyssp--)
+
+  YYSIZE_T yystacksize = YYINITDEPTH;
+
+  /* The variables used to return semantic value and location from the
+     action routines.  */
+  YYSTYPE yyval;
+
+
+  /* When reducing, the number of symbols on the RHS of the reduced
+     rule.  */
+  int yylen;
+
+  YYDPRINTF ((stderr, "Starting parse\n"));
+
+  yystate = 0;
+  yyerrstatus = 0;
+  yynerrs = 0;
+  yychar = YYEMPTY;		/* Cause a token to be read.  */
+
+  /* Initialize stack pointers.
+     Waste one element of value and location stack
+     so that they stay on the same level as the state stack.
+     The wasted elements are never initialized.  */
+
+  yyssp = yyss;
+  yyvsp = yyvs;
+
+
+  yyvsp[0] = yylval;
+
+  goto yysetstate;
+
+/*------------------------------------------------------------.
+| yynewstate -- Push a new state, which is found in yystate.  |
+`------------------------------------------------------------*/
+ yynewstate:
+  /* In all cases, when you get here, the value and location stacks
+     have just been pushed. so pushing a state here evens the stacks.
+     */
+  yyssp++;
+
+ yysetstate:
+  *yyssp = yystate;
+
+  if (yyss + yystacksize - 1 <= yyssp)
+    {
+      /* Get the current used size of the three stacks, in elements.  */
+      YYSIZE_T yysize = yyssp - yyss + 1;
+
+#ifdef yyoverflow
+      {
+	/* Give user a chance to reallocate the stack. Use copies of
+	   these so that the &'s don't force the real ones into
+	   memory.  */
+	YYSTYPE *yyvs1 = yyvs;
+	short int *yyss1 = yyss;
+
+
+	/* Each stack pointer address is followed by the size of the
+	   data in use in that stack, in bytes.  This used to be a
+	   conditional around just the two extra args, but that might
+	   be undefined if yyoverflow is a macro.  */
+	yyoverflow ("parser stack overflow",
+		    &yyss1, yysize * sizeof (*yyssp),
+		    &yyvs1, yysize * sizeof (*yyvsp),
+
+		    &yystacksize);
+
+	yyss = yyss1;
+	yyvs = yyvs1;
+      }
+#else /* no yyoverflow */
+# ifndef YYSTACK_RELOCATE
+      goto yyoverflowlab;
+# else
+      /* Extend the stack our own way.  */
+      if (YYMAXDEPTH <= yystacksize)
+	goto yyoverflowlab;
+      yystacksize *= 2;
+      if (YYMAXDEPTH < yystacksize)
+	yystacksize = YYMAXDEPTH;
+
+      {
+	short int *yyss1 = yyss;
+	union yyalloc *yyptr =
+	  (union yyalloc *) YYSTACK_ALLOC (YYSTACK_BYTES (yystacksize));
+	if (! yyptr)
+	  goto yyoverflowlab;
+	YYSTACK_RELOCATE (yyss);
+	YYSTACK_RELOCATE (yyvs);
+
+#  undef YYSTACK_RELOCATE
+	if (yyss1 != yyssa)
+	  YYSTACK_FREE (yyss1);
+      }
+# endif
+#endif /* no yyoverflow */
+
+      yyssp = yyss + yysize - 1;
+      yyvsp = yyvs + yysize - 1;
+
+
+      YYDPRINTF ((stderr, "Stack size increased to %lu\n",
+		  (unsigned long int) yystacksize));
+
+      if (yyss + yystacksize - 1 <= yyssp)
+	YYABORT;
+    }
+
+  YYDPRINTF ((stderr, "Entering state %d\n", yystate));
+
+  goto yybackup;
+
+/*-----------.
+| yybackup.  |
+`-----------*/
+yybackup:
+
+/* Do appropriate processing given the current state.  */
+/* Read a look-ahead token if we need one and don't already have one.  */
+/* yyresume: */
+
+  /* First try to decide what to do without reference to look-ahead token.  */
+
+  yyn = yypact[yystate];
+  if (yyn == YYPACT_NINF)
+    goto yydefault;
+
+  /* Not known => get a look-ahead token if don't already have one.  */
+
+  /* YYCHAR is either YYEMPTY or YYEOF or a valid look-ahead symbol.  */
+  if (yychar == YYEMPTY)
+    {
+      YYDPRINTF ((stderr, "Reading a token: "));
+      yychar = YYLEX;
+    }
+
+  if (yychar <= YYEOF)
+    {
+      yychar = yytoken = YYEOF;
+      YYDPRINTF ((stderr, "Now at end of input.\n"));
+    }
+  else
+    {
+      yytoken = YYTRANSLATE (yychar);
+      YY_SYMBOL_PRINT ("Next token is", yytoken, &yylval, &yylloc);
+    }
+
+  /* If the proper action on seeing token YYTOKEN is to reduce or to
+     detect an error, take that action.  */
+  yyn += yytoken;
+  if (yyn < 0 || YYLAST < yyn || yycheck[yyn] != yytoken)
+    goto yydefault;
+  yyn = yytable[yyn];
+  if (yyn <= 0)
+    {
+      if (yyn == 0 || yyn == YYTABLE_NINF)
+	goto yyerrlab;
+      yyn = -yyn;
+      goto yyreduce;
+    }
+
+  if (yyn == YYFINAL)
+    YYACCEPT;
+
+  /* Shift the look-ahead token.  */
+  YY_SYMBOL_PRINT ("Shifting", yytoken, &yylval, &yylloc);
+
+  /* Discard the token being shifted unless it is eof.  */
+  if (yychar != YYEOF)
+    yychar = YYEMPTY;
+
+  *++yyvsp = yylval;
+
+
+  /* Count tokens shifted since error; after three, turn off error
+     status.  */
+  if (yyerrstatus)
+    yyerrstatus--;
+
+  yystate = yyn;
+  goto yynewstate;
+
+
+/*-----------------------------------------------------------.
+| yydefault -- do the default action for the current state.  |
+`-----------------------------------------------------------*/
+yydefault:
+  yyn = yydefact[yystate];
+  if (yyn == 0)
+    goto yyerrlab;
+  goto yyreduce;
+
+
+/*-----------------------------.
+| yyreduce -- Do a reduction.  |
+`-----------------------------*/
+yyreduce:
+  /* yyn is the number of a rule to reduce with.  */
+  yylen = yyr2[yyn];
+
+  /* If YYLEN is nonzero, implement the default value of the action:
+     `$$ = $1'.
+
+     Otherwise, the following line sets YYVAL to garbage.
+     This behavior is undocumented and Bison
+     users should not rely upon it.  Assigning to YYVAL
+     unconditionally makes the parser a bit smaller, and it avoids a
+     GCC warning that YYVAL may be used uninitialized.  */
+  yyval = yyvsp[1-yylen];
+
+
+  YY_REDUCE_PRINT (yyn);
+  switch (yyn)
+    {
+        case 25:
+#line 166 "pars0grm.y"
+    { (yyval) = que_node_list_add_last(NULL, (yyvsp[0])); ;}
+    break;
+
+  case 26:
+#line 168 "pars0grm.y"
+    { (yyval) = que_node_list_add_last((yyvsp[-1]), (yyvsp[0])); ;}
+    break;
+
+  case 27:
+#line 172 "pars0grm.y"
+    { (yyval) = (yyvsp[0]);;}
+    break;
+
+  case 28:
+#line 174 "pars0grm.y"
+    { (yyval) = pars_func((yyvsp[-3]), (yyvsp[-1])); ;}
+    break;
+
+  case 29:
+#line 175 "pars0grm.y"
+    { (yyval) = (yyvsp[0]);;}
+    break;
+
+  case 30:
+#line 176 "pars0grm.y"
+    { (yyval) = (yyvsp[0]);;}
+    break;
+
+  case 31:
+#line 177 "pars0grm.y"
+    { (yyval) = (yyvsp[0]);;}
+    break;
+
+  case 32:
+#line 178 "pars0grm.y"
+    { (yyval) = (yyvsp[0]);;}
+    break;
+
+  case 33:
+#line 179 "pars0grm.y"
+    { (yyval) = (yyvsp[0]);;}
+    break;
+
+  case 34:
+#line 180 "pars0grm.y"
+    { (yyval) = (yyvsp[0]);;}
+    break;
+
+  case 35:
+#line 181 "pars0grm.y"
+    { (yyval) = (yyvsp[0]);;}
+    break;
+
+  case 36:
+#line 182 "pars0grm.y"
+    { (yyval) = pars_op('+', (yyvsp[-2]), (yyvsp[0])); ;}
+    break;
+
+  case 37:
+#line 183 "pars0grm.y"
+    { (yyval) = pars_op('-', (yyvsp[-2]), (yyvsp[0])); ;}
+    break;
+
+  case 38:
+#line 184 "pars0grm.y"
+    { (yyval) = pars_op('*', (yyvsp[-2]), (yyvsp[0])); ;}
+    break;
+
+  case 39:
+#line 185 "pars0grm.y"
+    { (yyval) = pars_op('/', (yyvsp[-2]), (yyvsp[0])); ;}
+    break;
+
+  case 40:
+#line 186 "pars0grm.y"
+    { (yyval) = pars_op('-', (yyvsp[0]), NULL); ;}
+    break;
+
+  case 41:
+#line 187 "pars0grm.y"
+    { (yyval) = (yyvsp[-1]); ;}
+    break;
+
+  case 42:
+#line 188 "pars0grm.y"
+    { (yyval) = pars_op('=', (yyvsp[-2]), (yyvsp[0])); ;}
+    break;
+
+  case 43:
+#line 189 "pars0grm.y"
+    { (yyval) = pars_op('<', (yyvsp[-2]), (yyvsp[0])); ;}
+    break;
+
+  case 44:
+#line 190 "pars0grm.y"
+    { (yyval) = pars_op('>', (yyvsp[-2]), (yyvsp[0])); ;}
+    break;
+
+  case 45:
+#line 191 "pars0grm.y"
+    { (yyval) = pars_op(PARS_GE_TOKEN, (yyvsp[-2]), (yyvsp[0])); ;}
+    break;
+
+  case 46:
+#line 192 "pars0grm.y"
+    { (yyval) = pars_op(PARS_LE_TOKEN, (yyvsp[-2]), (yyvsp[0])); ;}
+    break;
+
+  case 47:
+#line 193 "pars0grm.y"
+    { (yyval) = pars_op(PARS_NE_TOKEN, (yyvsp[-2]), (yyvsp[0])); ;}
+    break;
+
+  case 48:
+#line 194 "pars0grm.y"
+    { (yyval) = pars_op(PARS_AND_TOKEN, (yyvsp[-2]), (yyvsp[0])); ;}
+    break;
+
+  case 49:
+#line 195 "pars0grm.y"
+    { (yyval) = pars_op(PARS_OR_TOKEN, (yyvsp[-2]), (yyvsp[0])); ;}
+    break;
+
+  case 50:
+#line 196 "pars0grm.y"
+    { (yyval) = pars_op(PARS_NOT_TOKEN, (yyvsp[0]), NULL); ;}
+    break;
+
+  case 51:
+#line 198 "pars0grm.y"
+    { (yyval) = pars_op(PARS_NOTFOUND_TOKEN, (yyvsp[-2]), NULL); ;}
+    break;
+
+  case 52:
+#line 200 "pars0grm.y"
+    { (yyval) = pars_op(PARS_NOTFOUND_TOKEN, (yyvsp[-2]), NULL); ;}
+    break;
+
+  case 53:
+#line 204 "pars0grm.y"
+    { (yyval) = &pars_to_char_token; ;}
+    break;
+
+  case 54:
+#line 205 "pars0grm.y"
+    { (yyval) = &pars_to_number_token; ;}
+    break;
+
+  case 55:
+#line 206 "pars0grm.y"
+    { (yyval) = &pars_to_binary_token; ;}
+    break;
+
+  case 56:
+#line 208 "pars0grm.y"
+    { (yyval) = &pars_binary_to_number_token; ;}
+    break;
+
+  case 57:
+#line 209 "pars0grm.y"
+    { (yyval) = &pars_substr_token; ;}
+    break;
+
+  case 58:
+#line 210 "pars0grm.y"
+    { (yyval) = &pars_concat_token; ;}
+    break;
+
+  case 59:
+#line 211 "pars0grm.y"
+    { (yyval) = &pars_instr_token; ;}
+    break;
+
+  case 60:
+#line 212 "pars0grm.y"
+    { (yyval) = &pars_length_token; ;}
+    break;
+
+  case 61:
+#line 213 "pars0grm.y"
+    { (yyval) = &pars_sysdate_token; ;}
+    break;
+
+  case 62:
+#line 214 "pars0grm.y"
+    { (yyval) = &pars_rnd_token; ;}
+    break;
+
+  case 63:
+#line 215 "pars0grm.y"
+    { (yyval) = &pars_rnd_str_token; ;}
+    break;
+
+  case 67:
+#line 226 "pars0grm.y"
+    { (yyval) = pars_stored_procedure_call((yyvsp[-4])); ;}
+    break;
+
+  case 68:
+#line 231 "pars0grm.y"
+    { (yyval) = pars_procedure_call((yyvsp[-3]), (yyvsp[-1])); ;}
+    break;
+
+  case 69:
+#line 235 "pars0grm.y"
+    { (yyval) = &pars_replstr_token; ;}
+    break;
+
+  case 70:
+#line 236 "pars0grm.y"
+    { (yyval) = &pars_printf_token; ;}
+    break;
+
+  case 71:
+#line 237 "pars0grm.y"
+    { (yyval) = &pars_assert_token; ;}
+    break;
+
+  case 72:
+#line 241 "pars0grm.y"
+    { (yyval) = (yyvsp[-2]); ;}
+    break;
+
+  case 73:
+#line 245 "pars0grm.y"
+    { (yyval) = que_node_list_add_last(NULL, (yyvsp[0])); ;}
+    break;
+
+  case 74:
+#line 247 "pars0grm.y"
+    { (yyval) = que_node_list_add_last((yyvsp[-2]), (yyvsp[0])); ;}
+    break;
+
+  case 75:
+#line 251 "pars0grm.y"
+    { (yyval) = NULL; ;}
+    break;
+
+  case 76:
+#line 252 "pars0grm.y"
+    { (yyval) = que_node_list_add_last(NULL, (yyvsp[0])); ;}
+    break;
+
+  case 77:
+#line 254 "pars0grm.y"
+    { (yyval) = que_node_list_add_last((yyvsp[-2]), (yyvsp[0])); ;}
+    break;
+
+  case 78:
+#line 258 "pars0grm.y"
+    { (yyval) = NULL; ;}
+    break;
+
+  case 79:
+#line 259 "pars0grm.y"
+    { (yyval) = que_node_list_add_last(NULL, (yyvsp[0]));;}
+    break;
+
+  case 80:
+#line 260 "pars0grm.y"
+    { (yyval) = que_node_list_add_last((yyvsp[-2]), (yyvsp[0])); ;}
+    break;
+
+  case 81:
+#line 264 "pars0grm.y"
+    { (yyval) = (yyvsp[0]); ;}
+    break;
+
+  case 82:
+#line 266 "pars0grm.y"
+    { (yyval) = pars_func(&pars_count_token,
+				          que_node_list_add_last(NULL,
+					    sym_tab_add_int_lit(
+						pars_sym_tab_global, 1))); ;}
+    break;
+
+  case 83:
+#line 271 "pars0grm.y"
+    { (yyval) = pars_func(&pars_count_token,
+					    que_node_list_add_last(NULL,
+						pars_func(&pars_distinct_token,
+						     que_node_list_add_last(
+								NULL, (yyvsp[-1]))))); ;}
+    break;
+
+  case 84:
+#line 277 "pars0grm.y"
+    { (yyval) = pars_func(&pars_sum_token,
+						que_node_list_add_last(NULL,
+									(yyvsp[-1]))); ;}
+    break;
+
+  case 85:
+#line 283 "pars0grm.y"
+    { (yyval) = NULL; ;}
+    break;
+
+  case 86:
+#line 284 "pars0grm.y"
+    { (yyval) = que_node_list_add_last(NULL, (yyvsp[0])); ;}
+    break;
+
+  case 87:
+#line 286 "pars0grm.y"
+    { (yyval) = que_node_list_add_last((yyvsp[-2]), (yyvsp[0])); ;}
+    break;
+
+  case 88:
+#line 290 "pars0grm.y"
+    { (yyval) = pars_select_list(&pars_star_denoter,
+								NULL); ;}
+    break;
+
+  case 89:
+#line 293 "pars0grm.y"
+    { (yyval) = pars_select_list((yyvsp[-2]), (yyvsp[0])); ;}
+    break;
+
+  case 90:
+#line 294 "pars0grm.y"
+    { (yyval) = pars_select_list((yyvsp[0]), NULL); ;}
+    break;
+
+  case 91:
+#line 298 "pars0grm.y"
+    { (yyval) = NULL; ;}
+    break;
+
+  case 92:
+#line 299 "pars0grm.y"
+    { (yyval) = (yyvsp[0]); ;}
+    break;
+
+  case 93:
+#line 303 "pars0grm.y"
+    { (yyval) = NULL; ;}
+    break;
+
+  case 94:
+#line 305 "pars0grm.y"
+    { (yyval) = &pars_update_token; ;}
+    break;
+
+  case 95:
+#line 309 "pars0grm.y"
+    { (yyval) = NULL; ;}
+    break;
+
+  case 96:
+#line 311 "pars0grm.y"
+    { yyval = &pars_share_token; ;}
+    break;
+
+  case 97:
+#line 315 "pars0grm.y"
+    { (yyval) = &pars_asc_token; ;}
+    break;
+
+  case 98:
+#line 316 "pars0grm.y"
+    { (yyval) = &pars_asc_token; ;}
+    break;
+
+  case 99:
+#line 317 "pars0grm.y"
+    { (yyval) = &pars_desc_token; ;}
+    break;
+
+  case 100:
+#line 321 "pars0grm.y"
+    { (yyval) = NULL; ;}
+    break;
+
+  case 101:
+#line 323 "pars0grm.y"
+    { (yyval) = pars_order_by((yyvsp[-1]), (yyvsp[0])); ;}
+    break;
+
+  case 102:
+#line 332 "pars0grm.y"
+    { (yyval) = pars_select_statement((yyvsp[-6]), (yyvsp[-4]), (yyvsp[-3]),
+								(yyvsp[-2]), (yyvsp[-1]), (yyvsp[0])); ;}
+    break;
+
+  case 103:
+#line 338 "pars0grm.y"
+    { (yyval) = (yyvsp[0]); ;}
+    break;
+
+  case 104:
+#line 343 "pars0grm.y"
+    { (yyval) = pars_insert_statement((yyvsp[-4]), (yyvsp[-1]), NULL); ;}
+    break;
+
+  case 105:
+#line 345 "pars0grm.y"
+    { (yyval) = pars_insert_statement((yyvsp[-1]), NULL, (yyvsp[0])); ;}
+    break;
+
+  case 106:
+#line 349 "pars0grm.y"
+    { (yyval) = pars_column_assignment((yyvsp[-2]), (yyvsp[0])); ;}
+    break;
+
+  case 107:
+#line 353 "pars0grm.y"
+    { (yyval) = que_node_list_add_last(NULL, (yyvsp[0])); ;}
+    break;
+
+  case 108:
+#line 355 "pars0grm.y"
+    { (yyval) = que_node_list_add_last((yyvsp[-2]), (yyvsp[0])); ;}
+    break;
+
+  case 109:
+#line 361 "pars0grm.y"
+    { (yyval) = (yyvsp[0]); ;}
+    break;
+
+  case 110:
+#line 367 "pars0grm.y"
+    { (yyval) = pars_update_statement_start(FALSE,
+								(yyvsp[-2]), (yyvsp[0])); ;}
+    break;
+
+  case 111:
+#line 373 "pars0grm.y"
+    { (yyval) = pars_update_statement((yyvsp[-1]), NULL, (yyvsp[0])); ;}
+    break;
+
+  case 112:
+#line 378 "pars0grm.y"
+    { (yyval) = pars_update_statement((yyvsp[-1]), (yyvsp[0]), NULL); ;}
+    break;
+
+  case 113:
+#line 383 "pars0grm.y"
+    { (yyval) = pars_update_statement_start(TRUE,
+								(yyvsp[0]), NULL); ;}
+    break;
+
+  case 114:
+#line 389 "pars0grm.y"
+    { (yyval) = pars_update_statement((yyvsp[-1]), NULL, (yyvsp[0])); ;}
+    break;
+
+  case 115:
+#line 394 "pars0grm.y"
+    { (yyval) = pars_update_statement((yyvsp[-1]), (yyvsp[0]), NULL); ;}
+    break;
+
+  case 116:
+#line 399 "pars0grm.y"
+    { (yyval) = pars_row_printf_statement((yyvsp[0])); ;}
+    break;
+
+  case 117:
+#line 404 "pars0grm.y"
+    { (yyval) = pars_assignment_statement((yyvsp[-2]), (yyvsp[0])); ;}
+    break;
+
+  case 118:
+#line 410 "pars0grm.y"
+    { (yyval) = pars_elsif_element((yyvsp[-2]), (yyvsp[0])); ;}
+    break;
+
+  case 119:
+#line 414 "pars0grm.y"
+    { (yyval) = que_node_list_add_last(NULL, (yyvsp[0])); ;}
+    break;
+
+  case 120:
+#line 416 "pars0grm.y"
+    { (yyval) = que_node_list_add_last((yyvsp[-1]), (yyvsp[0])); ;}
+    break;
+
+  case 121:
+#line 420 "pars0grm.y"
+    { (yyval) = NULL; ;}
+    break;
+
+  case 122:
+#line 422 "pars0grm.y"
+    { (yyval) = (yyvsp[0]); ;}
+    break;
+
+  case 123:
+#line 423 "pars0grm.y"
+    { (yyval) = (yyvsp[0]); ;}
+    break;
+
+  case 124:
+#line 430 "pars0grm.y"
+    { (yyval) = pars_if_statement((yyvsp[-5]), (yyvsp[-3]), (yyvsp[-2])); ;}
+    break;
+
+  case 125:
+#line 436 "pars0grm.y"
+    { (yyval) = pars_while_statement((yyvsp[-4]), (yyvsp[-2])); ;}
+    break;
+
+  case 126:
+#line 444 "pars0grm.y"
+    { (yyval) = pars_for_statement((yyvsp[-8]), (yyvsp[-6]), (yyvsp[-4]), (yyvsp[-2])); ;}
+    break;
+
+  case 127:
+#line 448 "pars0grm.y"
+    { (yyval) = pars_exit_statement(); ;}
+    break;
+
+  case 128:
+#line 452 "pars0grm.y"
+    { (yyval) = pars_return_statement(); ;}
+    break;
+
+  case 129:
+#line 457 "pars0grm.y"
+    { (yyval) = pars_open_statement(
+						ROW_SEL_OPEN_CURSOR, (yyvsp[0])); ;}
+    break;
+
+  case 130:
+#line 463 "pars0grm.y"
+    { (yyval) = pars_open_statement(
+						ROW_SEL_CLOSE_CURSOR, (yyvsp[0])); ;}
+    break;
+
+  case 131:
+#line 469 "pars0grm.y"
+    { (yyval) = pars_fetch_statement((yyvsp[-2]), (yyvsp[0]), NULL); ;}
+    break;
+
+  case 132:
+#line 471 "pars0grm.y"
+    { (yyval) = pars_fetch_statement((yyvsp[-2]), NULL, (yyvsp[0])); ;}
+    break;
+
+  case 133:
+#line 476 "pars0grm.y"
+    { (yyval) = pars_column_def((yyvsp[-4]), (yyvsp[-3]), (yyvsp[-2]), (yyvsp[-1]), (yyvsp[0])); ;}
+    break;
+
+  case 134:
+#line 480 "pars0grm.y"
+    { (yyval) = que_node_list_add_last(NULL, (yyvsp[0])); ;}
+    break;
+
+  case 135:
+#line 482 "pars0grm.y"
+    { (yyval) = que_node_list_add_last((yyvsp[-2]), (yyvsp[0])); ;}
+    break;
+
+  case 136:
+#line 486 "pars0grm.y"
+    { (yyval) = NULL; ;}
+    break;
+
+  case 137:
+#line 488 "pars0grm.y"
+    { (yyval) = (yyvsp[-1]); ;}
+    break;
+
+  case 138:
+#line 492 "pars0grm.y"
+    { (yyval) = NULL; ;}
+    break;
+
+  case 139:
+#line 494 "pars0grm.y"
+    { (yyval) = &pars_int_token;
+					/* pass any non-NULL pointer */ ;}
+    break;
+
+  case 140:
+#line 499 "pars0grm.y"
+    { (yyval) = NULL; ;}
+    break;
+
+  case 141:
+#line 501 "pars0grm.y"
+    { (yyval) = &pars_int_token;
+					/* pass any non-NULL pointer */ ;}
+    break;
+
+  case 142:
+#line 506 "pars0grm.y"
+    { (yyval) = NULL; ;}
+    break;
+
+  case 143:
+#line 508 "pars0grm.y"
+    { (yyval) = &pars_int_token;
+					/* pass any non-NULL pointer */ ;}
+    break;
+
+  case 144:
+#line 515 "pars0grm.y"
+    { (yyval) = pars_create_table((yyvsp[-4]), (yyvsp[-2]), (yyvsp[0])); ;}
+    break;
+
+  case 145:
+#line 519 "pars0grm.y"
+    { (yyval) = que_node_list_add_last(NULL, (yyvsp[0])); ;}
+    break;
+
+  case 146:
+#line 521 "pars0grm.y"
+    { (yyval) = que_node_list_add_last((yyvsp[-2]), (yyvsp[0])); ;}
+    break;
+
+  case 147:
+#line 525 "pars0grm.y"
+    { (yyval) = NULL; ;}
+    break;
+
+  case 148:
+#line 526 "pars0grm.y"
+    { (yyval) = &pars_unique_token; ;}
+    break;
+
+  case 149:
+#line 530 "pars0grm.y"
+    { (yyval) = NULL; ;}
+    break;
+
+  case 150:
+#line 531 "pars0grm.y"
+    { (yyval) = &pars_clustered_token; ;}
+    break;
+
+  case 151:
+#line 539 "pars0grm.y"
+    { (yyval) = pars_create_index((yyvsp[-8]), (yyvsp[-7]), (yyvsp[-5]), (yyvsp[-3]), (yyvsp[-1])); ;}
+    break;
+
+  case 152:
+#line 544 "pars0grm.y"
+    { (yyval) = pars_commit_statement(); ;}
+    break;
+
+  case 153:
+#line 549 "pars0grm.y"
+    { (yyval) = pars_rollback_statement(); ;}
+    break;
+
+  case 154:
+#line 553 "pars0grm.y"
+    { (yyval) = &pars_int_token; ;}
+    break;
+
+  case 155:
+#line 554 "pars0grm.y"
+    { (yyval) = &pars_int_token; ;}
+    break;
+
+  case 156:
+#line 555 "pars0grm.y"
+    { (yyval) = &pars_char_token; ;}
+    break;
+
+  case 157:
+#line 556 "pars0grm.y"
+    { (yyval) = &pars_binary_token; ;}
+    break;
+
+  case 158:
+#line 557 "pars0grm.y"
+    { (yyval) = &pars_blob_token; ;}
+    break;
+
+  case 159:
+#line 562 "pars0grm.y"
+    { (yyval) = pars_parameter_declaration((yyvsp[-2]),
+							PARS_INPUT, (yyvsp[0])); ;}
+    break;
+
+  case 160:
+#line 565 "pars0grm.y"
+    { (yyval) = pars_parameter_declaration((yyvsp[-2]),
+							PARS_OUTPUT, (yyvsp[0])); ;}
+    break;
+
+  case 161:
+#line 570 "pars0grm.y"
+    { (yyval) = NULL; ;}
+    break;
+
+  case 162:
+#line 571 "pars0grm.y"
+    { (yyval) = que_node_list_add_last(NULL, (yyvsp[0])); ;}
+    break;
+
+  case 163:
+#line 573 "pars0grm.y"
+    { (yyval) = que_node_list_add_last((yyvsp[-2]), (yyvsp[0])); ;}
+    break;
+
+  case 164:
+#line 578 "pars0grm.y"
+    { (yyval) = pars_variable_declaration((yyvsp[-2]), (yyvsp[-1])); ;}
+    break;
+
+  case 168:
+#line 590 "pars0grm.y"
+    { (yyval) = pars_cursor_declaration((yyvsp[-3]), (yyvsp[-1])); ;}
+    break;
+
+  case 169:
+#line 595 "pars0grm.y"
+    { (yyval) = pars_function_declaration((yyvsp[-1])); ;}
+    break;
+
+  case 175:
+#line 616 "pars0grm.y"
+    { (yyval) = pars_procedure_definition((yyvsp[-9]), (yyvsp[-7]),
+								(yyvsp[-1])); ;}
+    break;
+
+
+    }
+
+/* Line 1010 of yacc.c.  */
+#line 2345 "pars0grm.c"
+
+  yyvsp -= yylen;
+  yyssp -= yylen;
+
+
+  YY_STACK_PRINT (yyss, yyssp);
+
+  *++yyvsp = yyval;
+
+
+  /* Now `shift' the result of the reduction.  Determine what state
+     that goes to, based on the state we popped back to and the rule
+     number reduced by.  */
+
+  yyn = yyr1[yyn];
+
+  yystate = yypgoto[yyn - YYNTOKENS] + *yyssp;
+  if (0 <= yystate && yystate <= YYLAST && yycheck[yystate] == *yyssp)
+    yystate = yytable[yystate];
+  else
+    yystate = yydefgoto[yyn - YYNTOKENS];
+
+  goto yynewstate;
+
+
+/*------------------------------------.
+| yyerrlab -- here on detecting error |
+`------------------------------------*/
+yyerrlab:
+  /* If not already recovering from an error, report this error.  */
+  if (!yyerrstatus)
+    {
+      ++yynerrs;
+#if YYERROR_VERBOSE
+      yyn = yypact[yystate];
+
+      if (YYPACT_NINF < yyn && yyn < YYLAST)
+	{
+	  YYSIZE_T yysize = 0;
+	  int yytype = YYTRANSLATE (yychar);
+	  const char* yyprefix;
+	  char *yymsg;
+	  int yyx;
+
+	  /* Start YYX at -YYN if negative to avoid negative indexes in
+	     YYCHECK.  */
+	  int yyxbegin = yyn < 0 ? -yyn : 0;
+
+	  /* Stay within bounds of both yycheck and yytname.  */
+	  int yychecklim = YYLAST - yyn;
+	  int yyxend = yychecklim < YYNTOKENS ? yychecklim : YYNTOKENS;
+	  int yycount = 0;
+
+	  yyprefix = ", expecting ";
+	  for (yyx = yyxbegin; yyx < yyxend; ++yyx)
+	    if (yycheck[yyx + yyn] == yyx && yyx != YYTERROR)
+	      {
+		yysize += yystrlen (yyprefix) + yystrlen (yytname [yyx]);
+		yycount += 1;
+		if (yycount == 5)
+		  {
+		    yysize = 0;
+		    break;
+		  }
+	      }
+	  yysize += (sizeof ("syntax error, unexpected ")
+		     + yystrlen (yytname[yytype]));
+	  yymsg = (char *) YYSTACK_ALLOC (yysize);
+	  if (yymsg != 0)
+	    {
+	      char *yyp = yystpcpy (yymsg, "syntax error, unexpected ");
+	      yyp = yystpcpy (yyp, yytname[yytype]);
+
+	      if (yycount < 5)
+		{
+		  yyprefix = ", expecting ";
+		  for (yyx = yyxbegin; yyx < yyxend; ++yyx)
+		    if (yycheck[yyx + yyn] == yyx && yyx != YYTERROR)
+		      {
+			yyp = yystpcpy (yyp, yyprefix);
+			yyp = yystpcpy (yyp, yytname[yyx]);
+			yyprefix = " or ";
+		      }
+		}
+	      yyerror (yymsg);
+	      YYSTACK_FREE (yymsg);
+	    }
+	  else
+	    yyerror ("syntax error; also virtual memory exhausted");
+	}
+      else
+#endif /* YYERROR_VERBOSE */
+	yyerror ("syntax error");
+    }
+
+
+
+  if (yyerrstatus == 3)
+    {
+      /* If just tried and failed to reuse look-ahead token after an
+	 error, discard it.  */
+
+      if (yychar <= YYEOF)
+        {
+          /* If at end of input, pop the error token,
+	     then the rest of the stack, then return failure.  */
+	  if (yychar == YYEOF)
+	     for (;;)
+	       {
+
+		 YYPOPSTACK;
+		 if (yyssp == yyss)
+		   YYABORT;
+		 yydestruct ("Error: popping",
+                             yystos[*yyssp], yyvsp);
+	       }
+        }
+      else
+	{
+	  yydestruct ("Error: discarding", yytoken, &yylval);
+	  yychar = YYEMPTY;
+	}
+    }
+
+  /* Else will try to reuse look-ahead token after shifting the error
+     token.  */
+  goto yyerrlab1;
+
+
+/*---------------------------------------------------.
+| yyerrorlab -- error raised explicitly by YYERROR.  |
+`---------------------------------------------------*/
+yyerrorlab:
+
+#ifdef __GNUC__
+  /* Pacify GCC when the user code never invokes YYERROR and the label
+     yyerrorlab therefore never appears in user code.  */
+  if (0)
+     goto yyerrorlab;
+#endif
+
+yyvsp -= yylen;
+  yyssp -= yylen;
+  yystate = *yyssp;
+  goto yyerrlab1;
+
+
+/*-------------------------------------------------------------.
+| yyerrlab1 -- common code for both syntax error and YYERROR.  |
+`-------------------------------------------------------------*/
+yyerrlab1:
+  yyerrstatus = 3;	/* Each real token shifted decrements this.  */
+
+  for (;;)
+    {
+      yyn = yypact[yystate];
+      if (yyn != YYPACT_NINF)
+	{
+	  yyn += YYTERROR;
+	  if (0 <= yyn && yyn <= YYLAST && yycheck[yyn] == YYTERROR)
+	    {
+	      yyn = yytable[yyn];
+	      if (0 < yyn)
+		break;
+	    }
+	}
+
+      /* Pop the current state because it cannot handle the error token.  */
+      if (yyssp == yyss)
+	YYABORT;
+
+
+      yydestruct ("Error: popping", yystos[yystate], yyvsp);
+      YYPOPSTACK;
+      yystate = *yyssp;
+      YY_STACK_PRINT (yyss, yyssp);
+    }
+
+  if (yyn == YYFINAL)
+    YYACCEPT;
+
+  *++yyvsp = yylval;
+
+
+  /* Shift the error token. */
+  YY_SYMBOL_PRINT ("Shifting", yystos[yyn], yyvsp, yylsp);
+
+  yystate = yyn;
+  goto yynewstate;
+
+
+/*-------------------------------------.
+| yyacceptlab -- YYACCEPT comes here.  |
+`-------------------------------------*/
+yyacceptlab:
+  yyresult = 0;
+  goto yyreturn;
+
+/*-----------------------------------.
+| yyabortlab -- YYABORT comes here.  |
+`-----------------------------------*/
+yyabortlab:
+  yydestruct ("Error: discarding lookahead",
+              yytoken, &yylval);
+  yychar = YYEMPTY;
+  yyresult = 1;
+  goto yyreturn;
+
+#ifndef yyoverflow
+/*----------------------------------------------.
+| yyoverflowlab -- parser overflow comes here.  |
+`----------------------------------------------*/
+yyoverflowlab:
+  yyerror ("parser stack overflow");
+  yyresult = 2;
+  /* Fall through.  */
+#endif
+
+yyreturn:
+#ifndef yyoverflow
+  if (yyss != yyssa)
+    YYSTACK_FREE (yyss);
+#endif
+  return yyresult;
+}
+
+
+#line 620 "pars0grm.y"
+
+
diff --git a/storage/xtradb/pars/pars0grm.y b/storage/xtradb/pars/pars0grm.y
new file mode 100644
index 00000000000..14d64f1826f
--- /dev/null
+++ b/storage/xtradb/pars/pars0grm.y
@@ -0,0 +1,635 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+SQL parser: input file for the GNU Bison parser generator
+
+Look from pars0lex.l for instructions how to generate the C files for
+the InnoDB parser.
+
+Created 12/14/1997 Heikki Tuuri
+*******************************************************/
+
+%{
+/* The value of the semantic attribute is a pointer to a query tree node
+que_node_t */
+
+#include "univ.i"
+#include <math.h>				/* Can't be before univ.i */
+#include "pars0pars.h"
+#include "mem0mem.h"
+#include "que0types.h"
+#include "que0que.h"
+#include "row0sel.h"
+
+#define YYSTYPE que_node_t*
+
+/* #define __STDC__ */
+
+int
+yylex(void);
+%}
+
+%token PARS_INT_LIT
+%token PARS_FLOAT_LIT
+%token PARS_STR_LIT
+%token PARS_FIXBINARY_LIT
+%token PARS_BLOB_LIT
+%token PARS_NULL_LIT
+%token PARS_ID_TOKEN
+%token PARS_AND_TOKEN
+%token PARS_OR_TOKEN
+%token PARS_NOT_TOKEN
+%token PARS_GE_TOKEN
+%token PARS_LE_TOKEN
+%token PARS_NE_TOKEN
+%token PARS_PROCEDURE_TOKEN
+%token PARS_IN_TOKEN
+%token PARS_OUT_TOKEN
+%token PARS_BINARY_TOKEN
+%token PARS_BLOB_TOKEN
+%token PARS_INT_TOKEN
+%token PARS_INTEGER_TOKEN
+%token PARS_FLOAT_TOKEN
+%token PARS_CHAR_TOKEN
+%token PARS_IS_TOKEN
+%token PARS_BEGIN_TOKEN
+%token PARS_END_TOKEN
+%token PARS_IF_TOKEN
+%token PARS_THEN_TOKEN
+%token PARS_ELSE_TOKEN
+%token PARS_ELSIF_TOKEN
+%token PARS_LOOP_TOKEN
+%token PARS_WHILE_TOKEN
+%token PARS_RETURN_TOKEN
+%token PARS_SELECT_TOKEN
+%token PARS_SUM_TOKEN
+%token PARS_COUNT_TOKEN
+%token PARS_DISTINCT_TOKEN
+%token PARS_FROM_TOKEN
+%token PARS_WHERE_TOKEN
+%token PARS_FOR_TOKEN
+%token PARS_DDOT_TOKEN
+%token PARS_READ_TOKEN
+%token PARS_ORDER_TOKEN
+%token PARS_BY_TOKEN
+%token PARS_ASC_TOKEN
+%token PARS_DESC_TOKEN
+%token PARS_INSERT_TOKEN
+%token PARS_INTO_TOKEN
+%token PARS_VALUES_TOKEN
+%token PARS_UPDATE_TOKEN
+%token PARS_SET_TOKEN
+%token PARS_DELETE_TOKEN
+%token PARS_CURRENT_TOKEN
+%token PARS_OF_TOKEN
+%token PARS_CREATE_TOKEN
+%token PARS_TABLE_TOKEN
+%token PARS_INDEX_TOKEN
+%token PARS_UNIQUE_TOKEN
+%token PARS_CLUSTERED_TOKEN
+%token PARS_DOES_NOT_FIT_IN_MEM_TOKEN
+%token PARS_ON_TOKEN
+%token PARS_ASSIGN_TOKEN
+%token PARS_DECLARE_TOKEN
+%token PARS_CURSOR_TOKEN
+%token PARS_SQL_TOKEN
+%token PARS_OPEN_TOKEN
+%token PARS_FETCH_TOKEN
+%token PARS_CLOSE_TOKEN
+%token PARS_NOTFOUND_TOKEN
+%token PARS_TO_CHAR_TOKEN
+%token PARS_TO_NUMBER_TOKEN
+%token PARS_TO_BINARY_TOKEN
+%token PARS_BINARY_TO_NUMBER_TOKEN
+%token PARS_SUBSTR_TOKEN
+%token PARS_REPLSTR_TOKEN
+%token PARS_CONCAT_TOKEN
+%token PARS_INSTR_TOKEN
+%token PARS_LENGTH_TOKEN
+%token PARS_SYSDATE_TOKEN
+%token PARS_PRINTF_TOKEN
+%token PARS_ASSERT_TOKEN
+%token PARS_RND_TOKEN
+%token PARS_RND_STR_TOKEN
+%token PARS_ROW_PRINTF_TOKEN
+%token PARS_COMMIT_TOKEN
+%token PARS_ROLLBACK_TOKEN
+%token PARS_WORK_TOKEN
+%token PARS_UNSIGNED_TOKEN
+%token PARS_EXIT_TOKEN
+%token PARS_FUNCTION_TOKEN
+%token PARS_LOCK_TOKEN
+%token PARS_SHARE_TOKEN
+%token PARS_MODE_TOKEN
+
+%left PARS_AND_TOKEN PARS_OR_TOKEN
+%left PARS_NOT_TOKEN
+%left '=' '<' '>' PARS_GE_TOKEN PARS_LE_TOKEN
+%left '-' '+'
+%left '*' '/'
+%left NEG     /* negation--unary minus */
+%left '%'
+
+/* Grammar follows */
+%%
+
+top_statement:
+        procedure_definition ';'
+
+statement:
+	stored_procedure_call
+	| predefined_procedure_call ';'
+	| while_statement ';'
+	| for_statement ';'
+	| exit_statement ';'
+	| if_statement ';'
+	| return_statement ';'
+	| assignment_statement ';'
+	| select_statement ';'
+	| insert_statement ';'
+	| row_printf_statement ';'
+	| delete_statement_searched ';'
+	| delete_statement_positioned ';'
+	| update_statement_searched ';'
+	| update_statement_positioned ';'
+	| open_cursor_statement ';'
+	| fetch_statement ';'
+	| close_cursor_statement ';'
+	| commit_statement ';'
+	| rollback_statement ';'
+	| create_table ';'
+	| create_index ';'
+;
+
+statement_list:
+	statement		{ $$ = que_node_list_add_last(NULL, $1); }
+	| statement_list statement
+				{ $$ = que_node_list_add_last($1, $2); }
+;
+
+exp:
+	PARS_ID_TOKEN		{ $$ = $1;}
+	| function_name '(' exp_list ')'
+				{ $$ = pars_func($1, $3); }
+	| PARS_INT_LIT		{ $$ = $1;}
+	| PARS_FLOAT_LIT	{ $$ = $1;}
+	| PARS_STR_LIT		{ $$ = $1;}
+	| PARS_FIXBINARY_LIT	{ $$ = $1;}
+	| PARS_BLOB_LIT		{ $$ = $1;}
+	| PARS_NULL_LIT		{ $$ = $1;}
+	| PARS_SQL_TOKEN	{ $$ = $1;}
+	| exp '+' exp        	{ $$ = pars_op('+', $1, $3); }
+	| exp '-' exp        	{ $$ = pars_op('-', $1, $3); }
+	| exp '*' exp        	{ $$ = pars_op('*', $1, $3); }
+	| exp '/' exp        	{ $$ = pars_op('/', $1, $3); }
+	| '-' exp %prec NEG 	{ $$ = pars_op('-', $2, NULL); }
+	| '(' exp ')'        	{ $$ = $2; }
+	| exp '=' exp		{ $$ = pars_op('=', $1, $3); }
+	| exp '<' exp		{ $$ = pars_op('<', $1, $3); }
+	| exp '>' exp		{ $$ = pars_op('>', $1, $3); }
+	| exp PARS_GE_TOKEN exp	{ $$ = pars_op(PARS_GE_TOKEN, $1, $3); }
+	| exp PARS_LE_TOKEN exp	{ $$ = pars_op(PARS_LE_TOKEN, $1, $3); }
+	| exp PARS_NE_TOKEN exp	{ $$ = pars_op(PARS_NE_TOKEN, $1, $3); }
+	| exp PARS_AND_TOKEN exp{ $$ = pars_op(PARS_AND_TOKEN, $1, $3); }
+	| exp PARS_OR_TOKEN exp	{ $$ = pars_op(PARS_OR_TOKEN, $1, $3); }
+	| PARS_NOT_TOKEN exp	{ $$ = pars_op(PARS_NOT_TOKEN, $2, NULL); }
+	| PARS_ID_TOKEN '%' PARS_NOTFOUND_TOKEN
+				{ $$ = pars_op(PARS_NOTFOUND_TOKEN, $1, NULL); }
+	| PARS_SQL_TOKEN '%' PARS_NOTFOUND_TOKEN
+				{ $$ = pars_op(PARS_NOTFOUND_TOKEN, $1, NULL); }
+;
+
+function_name:
+	PARS_TO_CHAR_TOKEN	{ $$ = &pars_to_char_token; }
+	| PARS_TO_NUMBER_TOKEN	{ $$ = &pars_to_number_token; }
+	| PARS_TO_BINARY_TOKEN	{ $$ = &pars_to_binary_token; }
+	| PARS_BINARY_TO_NUMBER_TOKEN
+				{ $$ = &pars_binary_to_number_token; }
+	| PARS_SUBSTR_TOKEN	{ $$ = &pars_substr_token; }
+	| PARS_CONCAT_TOKEN	{ $$ = &pars_concat_token; }
+	| PARS_INSTR_TOKEN	{ $$ = &pars_instr_token; }
+	| PARS_LENGTH_TOKEN	{ $$ = &pars_length_token; }
+	| PARS_SYSDATE_TOKEN	{ $$ = &pars_sysdate_token; }
+	| PARS_RND_TOKEN	{ $$ = &pars_rnd_token; }
+	| PARS_RND_STR_TOKEN	{ $$ = &pars_rnd_str_token; }
+;
+
+question_mark_list:
+	/* Nothing */
+	| '?'
+	| question_mark_list ',' '?'
+;
+
+stored_procedure_call:
+	'{' PARS_ID_TOKEN '(' question_mark_list ')' '}'
+				{ $$ = pars_stored_procedure_call($2); }
+;
+
+predefined_procedure_call:
+	predefined_procedure_name '(' exp_list ')'
+				{ $$ = pars_procedure_call($1, $3); }
+;
+
+predefined_procedure_name:
+	PARS_REPLSTR_TOKEN	{ $$ = &pars_replstr_token; }
+	| PARS_PRINTF_TOKEN	{ $$ = &pars_printf_token; }
+	| PARS_ASSERT_TOKEN	{ $$ = &pars_assert_token; }
+;
+
+user_function_call:
+	PARS_ID_TOKEN '(' ')'	{ $$ = $1; }
+;
+
+table_list:
+	PARS_ID_TOKEN		{ $$ = que_node_list_add_last(NULL, $1); }
+	| table_list ',' PARS_ID_TOKEN
+				{ $$ = que_node_list_add_last($1, $3); }
+;
+
+variable_list:
+	/* Nothing */		{ $$ = NULL; }
+	| PARS_ID_TOKEN		{ $$ = que_node_list_add_last(NULL, $1); }
+	| variable_list ',' PARS_ID_TOKEN
+				{ $$ = que_node_list_add_last($1, $3); }
+;
+
+exp_list:
+	/* Nothing */		{ $$ = NULL; }
+	| exp			{ $$ = que_node_list_add_last(NULL, $1);}
+	| exp_list ',' exp	{ $$ = que_node_list_add_last($1, $3); }
+;
+
+select_item:
+	exp			{ $$ = $1; }
+	| PARS_COUNT_TOKEN '(' '*' ')'
+				{ $$ = pars_func(&pars_count_token,
+				          que_node_list_add_last(NULL,
+					    sym_tab_add_int_lit(
+						pars_sym_tab_global, 1))); }
+	| PARS_COUNT_TOKEN '(' PARS_DISTINCT_TOKEN PARS_ID_TOKEN ')'
+				{ $$ = pars_func(&pars_count_token,
+					    que_node_list_add_last(NULL,
+						pars_func(&pars_distinct_token,
+						     que_node_list_add_last(
+								NULL, $4)))); }
+	| PARS_SUM_TOKEN '(' exp ')'
+				{ $$ = pars_func(&pars_sum_token,
+						que_node_list_add_last(NULL,
+									$3)); }
+;
+
+select_item_list:
+	/* Nothing */		{ $$ = NULL; }
+	| select_item		{ $$ = que_node_list_add_last(NULL, $1); }
+	| select_item_list ',' select_item
+				{ $$ = que_node_list_add_last($1, $3); }
+;
+
+select_list:
+	'*'			{ $$ = pars_select_list(&pars_star_denoter,
+								NULL); }
+	| select_item_list PARS_INTO_TOKEN variable_list
+				{ $$ = pars_select_list($1, $3); }
+	| select_item_list	{ $$ = pars_select_list($1, NULL); }
+;
+
+search_condition:
+	/* Nothing */		{ $$ = NULL; }
+	| PARS_WHERE_TOKEN exp	{ $$ = $2; }
+;
+
+for_update_clause:
+	/* Nothing */		{ $$ = NULL; }
+	| PARS_FOR_TOKEN PARS_UPDATE_TOKEN
+				{ $$ = &pars_update_token; }
+;
+
+lock_shared_clause:
+	/* Nothing */		{ $$ = NULL; }
+	| PARS_LOCK_TOKEN PARS_IN_TOKEN PARS_SHARE_TOKEN PARS_MODE_TOKEN
+				{ $$ = &pars_share_token; }
+;
+
+order_direction:
+	/* Nothing */		{ $$ = &pars_asc_token; }
+	| PARS_ASC_TOKEN	{ $$ = &pars_asc_token; }
+	| PARS_DESC_TOKEN	{ $$ = &pars_desc_token; }
+;
+
+order_by_clause:
+	/* Nothing */		{ $$ = NULL; }
+	| PARS_ORDER_TOKEN PARS_BY_TOKEN PARS_ID_TOKEN order_direction
+				{ $$ = pars_order_by($3, $4); }
+;
+
+select_statement:
+	PARS_SELECT_TOKEN select_list
+	PARS_FROM_TOKEN table_list
+	search_condition
+	for_update_clause
+	lock_shared_clause
+	order_by_clause		{ $$ = pars_select_statement($2, $4, $5,
+								$6, $7, $8); }
+;
+
+insert_statement_start:
+	PARS_INSERT_TOKEN PARS_INTO_TOKEN
+	PARS_ID_TOKEN		{ $$ = $3; }
+;
+
+insert_statement:
+	insert_statement_start PARS_VALUES_TOKEN '(' exp_list ')'
+				{ $$ = pars_insert_statement($1, $4, NULL); }
+	| insert_statement_start select_statement
+				{ $$ = pars_insert_statement($1, NULL, $2); }
+;
+
+column_assignment:
+	PARS_ID_TOKEN '=' exp	{ $$ = pars_column_assignment($1, $3); }
+;
+
+column_assignment_list:
+	column_assignment	{ $$ = que_node_list_add_last(NULL, $1); }
+	| column_assignment_list ',' column_assignment
+				{ $$ = que_node_list_add_last($1, $3); }
+;
+
+cursor_positioned:
+	PARS_WHERE_TOKEN
+	PARS_CURRENT_TOKEN PARS_OF_TOKEN
+	PARS_ID_TOKEN 		{ $$ = $4; }
+;
+
+update_statement_start:
+	PARS_UPDATE_TOKEN PARS_ID_TOKEN
+	PARS_SET_TOKEN
+	column_assignment_list	{ $$ = pars_update_statement_start(FALSE,
+								$2, $4); }
+;
+
+update_statement_searched:
+	update_statement_start
+	search_condition	{ $$ = pars_update_statement($1, NULL, $2); }
+;
+
+update_statement_positioned:
+	update_statement_start
+	cursor_positioned	{ $$ = pars_update_statement($1, $2, NULL); }
+;
+
+delete_statement_start:
+	PARS_DELETE_TOKEN PARS_FROM_TOKEN
+	PARS_ID_TOKEN		{ $$ = pars_update_statement_start(TRUE,
+								$3, NULL); }
+;
+
+delete_statement_searched:
+	delete_statement_start
+	search_condition	{ $$ = pars_update_statement($1, NULL, $2); }
+;
+
+delete_statement_positioned:
+	delete_statement_start
+	cursor_positioned	{ $$ = pars_update_statement($1, $2, NULL); }
+;
+
+row_printf_statement:
+	PARS_ROW_PRINTF_TOKEN select_statement
+				{ $$ = pars_row_printf_statement($2); }
+;
+
+assignment_statement:
+	PARS_ID_TOKEN PARS_ASSIGN_TOKEN exp
+				{ $$ = pars_assignment_statement($1, $3); }
+;
+
+elsif_element:
+	PARS_ELSIF_TOKEN
+	exp PARS_THEN_TOKEN statement_list
+				{ $$ = pars_elsif_element($2, $4); }
+;
+
+elsif_list:
+	elsif_element		{ $$ = que_node_list_add_last(NULL, $1); }
+	| elsif_list elsif_element
+				{ $$ = que_node_list_add_last($1, $2); }
+;
+
+else_part:
+	/* Nothing */		{ $$ = NULL; }
+	| PARS_ELSE_TOKEN statement_list
+				{ $$ = $2; }
+	| elsif_list		{ $$ = $1; }
+;
+
+if_statement:
+	PARS_IF_TOKEN exp PARS_THEN_TOKEN statement_list
+	else_part
+	PARS_END_TOKEN PARS_IF_TOKEN
+				{ $$ = pars_if_statement($2, $4, $5); }
+;
+
+while_statement:
+	PARS_WHILE_TOKEN exp PARS_LOOP_TOKEN statement_list
+	PARS_END_TOKEN PARS_LOOP_TOKEN
+				{ $$ = pars_while_statement($2, $4); }
+;
+
+for_statement:
+	PARS_FOR_TOKEN PARS_ID_TOKEN PARS_IN_TOKEN
+	exp PARS_DDOT_TOKEN exp
+	PARS_LOOP_TOKEN statement_list
+	PARS_END_TOKEN PARS_LOOP_TOKEN
+				{ $$ = pars_for_statement($2, $4, $6, $8); }
+;
+
+exit_statement:
+	PARS_EXIT_TOKEN		{ $$ = pars_exit_statement(); }
+;
+
+return_statement:
+	PARS_RETURN_TOKEN	{ $$ = pars_return_statement(); }
+;
+
+open_cursor_statement:
+	PARS_OPEN_TOKEN PARS_ID_TOKEN
+				{ $$ = pars_open_statement(
+						ROW_SEL_OPEN_CURSOR, $2); }
+;
+
+close_cursor_statement:
+	PARS_CLOSE_TOKEN PARS_ID_TOKEN
+				{ $$ = pars_open_statement(
+						ROW_SEL_CLOSE_CURSOR, $2); }
+;
+
+fetch_statement:
+	PARS_FETCH_TOKEN PARS_ID_TOKEN PARS_INTO_TOKEN variable_list
+				{ $$ = pars_fetch_statement($2, $4, NULL); }
+	| PARS_FETCH_TOKEN PARS_ID_TOKEN PARS_INTO_TOKEN user_function_call
+				{ $$ = pars_fetch_statement($2, NULL, $4); }
+;
+
+column_def:
+	PARS_ID_TOKEN type_name	opt_column_len opt_unsigned opt_not_null
+				{ $$ = pars_column_def($1, $2, $3, $4, $5); }
+;
+
+column_def_list:
+	column_def		{ $$ = que_node_list_add_last(NULL, $1); }
+	| column_def_list ',' column_def
+				{ $$ = que_node_list_add_last($1, $3); }
+;
+
+opt_column_len:
+	/* Nothing */		{ $$ = NULL; }
+	| '(' PARS_INT_LIT ')'
+				{ $$ = $2; }
+;
+
+opt_unsigned:
+	/* Nothing */		{ $$ = NULL; }
+	| PARS_UNSIGNED_TOKEN
+				{ $$ = &pars_int_token;
+					/* pass any non-NULL pointer */ }
+;
+
+opt_not_null:
+	/* Nothing */		{ $$ = NULL; }
+	| PARS_NOT_TOKEN PARS_NULL_LIT
+				{ $$ = &pars_int_token;
+					/* pass any non-NULL pointer */ }
+;
+
+not_fit_in_memory:
+	/* Nothing */		{ $$ = NULL; }
+	| PARS_DOES_NOT_FIT_IN_MEM_TOKEN
+				{ $$ = &pars_int_token;
+					/* pass any non-NULL pointer */ }
+;
+
+create_table:
+	PARS_CREATE_TOKEN PARS_TABLE_TOKEN
+	PARS_ID_TOKEN '(' column_def_list ')'
+	not_fit_in_memory	{ $$ = pars_create_table($3, $5, $7); }
+;
+
+column_list:
+	PARS_ID_TOKEN		{ $$ = que_node_list_add_last(NULL, $1); }
+	| column_list ',' PARS_ID_TOKEN
+				{ $$ = que_node_list_add_last($1, $3); }
+;
+
+unique_def:
+	/* Nothing */		{ $$ = NULL; }
+	| PARS_UNIQUE_TOKEN	{ $$ = &pars_unique_token; }
+;
+
+clustered_def:
+	/* Nothing */		{ $$ = NULL; }
+	| PARS_CLUSTERED_TOKEN	{ $$ = &pars_clustered_token; }
+;
+
+create_index:
+	PARS_CREATE_TOKEN unique_def
+	clustered_def
+	PARS_INDEX_TOKEN
+	PARS_ID_TOKEN PARS_ON_TOKEN PARS_ID_TOKEN
+	'(' column_list ')'	{ $$ = pars_create_index($2, $3, $5, $7, $9); }
+;
+
+commit_statement:
+	PARS_COMMIT_TOKEN PARS_WORK_TOKEN
+				{ $$ = pars_commit_statement(); }
+;
+
+rollback_statement:
+	PARS_ROLLBACK_TOKEN PARS_WORK_TOKEN
+				{ $$ = pars_rollback_statement(); }
+;
+
+type_name:
+	PARS_INT_TOKEN		{ $$ = &pars_int_token; }
+	| PARS_INTEGER_TOKEN	{ $$ = &pars_int_token; }
+	| PARS_CHAR_TOKEN	{ $$ = &pars_char_token; }
+	| PARS_BINARY_TOKEN	{ $$ = &pars_binary_token; }
+	| PARS_BLOB_TOKEN	{ $$ = &pars_blob_token; }
+;
+
+parameter_declaration:
+	PARS_ID_TOKEN PARS_IN_TOKEN type_name
+				{ $$ = pars_parameter_declaration($1,
+							PARS_INPUT, $3); }
+	| PARS_ID_TOKEN PARS_OUT_TOKEN type_name
+				{ $$ = pars_parameter_declaration($1,
+							PARS_OUTPUT, $3); }
+;
+
+parameter_declaration_list:
+	/* Nothing */		{ $$ = NULL; }
+	| parameter_declaration	{ $$ = que_node_list_add_last(NULL, $1); }
+	| parameter_declaration_list ',' parameter_declaration
+				{ $$ = que_node_list_add_last($1, $3); }
+;
+
+variable_declaration:
+	PARS_ID_TOKEN type_name ';'
+				{ $$ = pars_variable_declaration($1, $2); }
+;
+
+variable_declaration_list:
+	/* Nothing */
+	| variable_declaration
+	| variable_declaration_list variable_declaration
+;
+
+cursor_declaration:
+	PARS_DECLARE_TOKEN PARS_CURSOR_TOKEN PARS_ID_TOKEN
+	PARS_IS_TOKEN select_statement ';'
+				{ $$ = pars_cursor_declaration($3, $5); }
+;
+
+function_declaration:
+	PARS_DECLARE_TOKEN PARS_FUNCTION_TOKEN PARS_ID_TOKEN ';'
+				{ $$ = pars_function_declaration($3); }
+;
+
+declaration:
+	cursor_declaration
+	| function_declaration
+;
+
+declaration_list:
+	/* Nothing */
+	| declaration
+	| declaration_list declaration
+;
+
+procedure_definition:
+	PARS_PROCEDURE_TOKEN PARS_ID_TOKEN '(' parameter_declaration_list ')'
+	PARS_IS_TOKEN
+	variable_declaration_list
+	declaration_list
+	PARS_BEGIN_TOKEN
+	statement_list
+	PARS_END_TOKEN		{ $$ = pars_procedure_definition($2, $4,
+								$10); }
+;
+
+%%
diff --git a/storage/xtradb/pars/pars0lex.l b/storage/xtradb/pars/pars0lex.l
new file mode 100644
index 00000000000..38cb744bd44
--- /dev/null
+++ b/storage/xtradb/pars/pars0lex.l
@@ -0,0 +1,663 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+SQL parser lexical analyzer: input file for the GNU Flex lexer generator
+
+The InnoDB parser is frozen because MySQL takes care of SQL parsing.
+Therefore we normally keep the InnoDB parser C files as they are, and do
+not automatically generate them from pars0grm.y and pars0lex.l.
+
+How to make the InnoDB parser and lexer C files:
+
+1. Run ./make_flex.sh to generate lexer files.
+
+2. Run ./make_bison.sh to generate parser files.
+
+These instructions seem to work at least with bison-1.875d and flex-2.5.31 on
+Linux.
+
+Created 12/14/1997 Heikki Tuuri
+*******************************************************/
+
+%option nostdinit
+%option 8bit
+%option warn
+%option pointer
+%option never-interactive
+%option nodefault
+%option noinput
+%option nounput
+%option noyywrap
+%option noyy_scan_buffer
+%option noyy_scan_bytes
+%option noyy_scan_string
+%option nounistd
+
+%{
+#define YYSTYPE que_node_t*
+
+#include "univ.i"
+#include "pars0pars.h"
+#include "pars0grm.h"
+#include "pars0sym.h"
+#include "mem0mem.h"
+#include "os0proc.h"
+
+#define malloc(A)	ut_malloc(A)
+#define free(A)		ut_free(A)
+#define realloc(P, A)	ut_realloc(P, A)
+#define exit(A) 	ut_error
+
+#define YY_INPUT(buf, result, max_size) pars_get_lex_chars(buf, &result, max_size)
+
+/* String buffer for removing quotes */
+static ulint	stringbuf_len_alloc = 0; /* Allocated length */
+static ulint	stringbuf_len = 0; /* Current length */
+static char*	stringbuf; /* Start of buffer */
+/* Appends a string to the buffer. */
+static
+void
+string_append(
+/*==========*/
+	const char*	str,	/* in: string to be appended */
+	ulint		len)	/* in: length of the string */
+{
+	if (stringbuf == NULL) {
+		stringbuf = malloc(1);
+		stringbuf_len_alloc = 1;
+	}
+
+	if (stringbuf_len + len > stringbuf_len_alloc) {
+		while (stringbuf_len + len > stringbuf_len_alloc) {
+			stringbuf_len_alloc <<= 1;
+		}
+		stringbuf = realloc(stringbuf, stringbuf_len_alloc);
+	}
+
+	memcpy(stringbuf + stringbuf_len, str, len);
+	stringbuf_len += len;
+}
+
+%}
+
+DIGIT	[0-9]
+ID	[a-z_A-Z][a-z_A-Z0-9]*
+BOUND_LIT	\:[a-z_A-Z0-9]+
+BOUND_ID	\$[a-z_A-Z0-9]+
+
+%x comment
+%x quoted
+%x id
+%%
+
+{DIGIT}+	{
+			yylval = sym_tab_add_int_lit(pars_sym_tab_global,
+								atoi(yytext));
+			return(PARS_INT_LIT);
+}
+
+{DIGIT}+"."{DIGIT}* {
+			ut_error;	/* not implemented */
+
+			return(PARS_FLOAT_LIT);
+}
+
+{BOUND_LIT}	{
+			ulint	type;
+
+			yylval = sym_tab_add_bound_lit(pars_sym_tab_global,
+				yytext + 1, &type);
+
+			return((int) type);
+}
+
+{BOUND_ID}	{
+			yylval = sym_tab_add_bound_id(pars_sym_tab_global,
+				yytext + 1);
+
+			return(PARS_ID_TOKEN);
+}
+
+"'"		{
+/* Quoted character string literals are handled in an explicit
+start state 'quoted'.  This state is entered and the buffer for
+the scanned string is emptied upon encountering a starting quote.
+
+In the state 'quoted', only two actions are possible (defined below). */
+			BEGIN(quoted);
+			stringbuf_len = 0;
+}
+<quoted>[^\']+	{
+			/* Got a sequence of characters other than "'":
+			append to string buffer */
+			string_append(yytext, yyleng);
+}
+<quoted>"'"+	{
+			/* Got a sequence of "'" characters:
+			append half of them to string buffer,
+			as "''" represents a single "'".
+			We apply truncating division,
+			so that "'''" will result in "'". */
+
+			string_append(yytext, yyleng / 2);
+
+			/* If we got an odd number of quotes, then the
+			last quote we got is the terminating quote.
+			At the end of the string, we return to the
+			initial start state and report the scanned
+			string literal. */
+
+			if (yyleng % 2) {
+				BEGIN(INITIAL);
+				yylval = sym_tab_add_str_lit(
+					pars_sym_tab_global,
+					(byte*) stringbuf, stringbuf_len);
+				return(PARS_STR_LIT);
+			}
+}
+
+\"		{
+/* Quoted identifiers are handled in an explicit start state 'id'.
+This state is entered and the buffer for the scanned string is emptied
+upon encountering a starting quote.
+
+In the state 'id', only two actions are possible (defined below). */
+			BEGIN(id);
+			stringbuf_len = 0;
+}
+<id>[^\"]+	{
+			/* Got a sequence of characters other than '"':
+			append to string buffer */
+			string_append(yytext, yyleng);
+}
+<id>\"+	{
+			/* Got a sequence of '"' characters:
+			append half of them to string buffer,
+			as '""' represents a single '"'.
+			We apply truncating division,
+			so that '"""' will result in '"'. */
+
+			string_append(yytext, yyleng / 2);
+
+			/* If we got an odd number of quotes, then the
+			last quote we got is the terminating quote.
+			At the end of the string, we return to the
+			initial start state and report the scanned
+			identifier. */
+
+			if (yyleng % 2) {
+				BEGIN(INITIAL);
+				yylval = sym_tab_add_id(
+					pars_sym_tab_global,
+					(byte*) stringbuf, stringbuf_len);
+
+				return(PARS_ID_TOKEN);
+			}
+}
+
+"NULL"		{
+			yylval = sym_tab_add_null_lit(pars_sym_tab_global);
+
+			return(PARS_NULL_LIT);
+}
+
+"SQL"		{
+			/* Implicit cursor name */
+			yylval = sym_tab_add_str_lit(pars_sym_tab_global,
+							(byte*) yytext, yyleng);
+			return(PARS_SQL_TOKEN);
+}
+
+"AND"		{
+			return(PARS_AND_TOKEN);
+}
+
+"OR"		{
+			return(PARS_OR_TOKEN);
+}
+
+"NOT"		{
+			return(PARS_NOT_TOKEN);
+}
+
+"PROCEDURE"	{
+			return(PARS_PROCEDURE_TOKEN);
+}
+
+"IN"		{
+			return(PARS_IN_TOKEN);
+}
+
+"OUT"		{
+			return(PARS_OUT_TOKEN);
+}
+
+"BINARY"	{
+	 		return(PARS_BINARY_TOKEN);
+}
+
+"BLOB"		{
+	 		return(PARS_BLOB_TOKEN);
+}
+
+"INT"		{
+	 		return(PARS_INT_TOKEN);
+}
+
+"INTEGER"	{
+	 		return(PARS_INT_TOKEN);
+}
+
+"FLOAT"		{
+	 		return(PARS_FLOAT_TOKEN);
+}
+
+"CHAR"		{
+	 		return(PARS_CHAR_TOKEN);
+}
+
+"IS"		{
+			return(PARS_IS_TOKEN);
+}
+
+"BEGIN"		{
+			return(PARS_BEGIN_TOKEN);
+}
+
+"END"		{
+			return(PARS_END_TOKEN);
+}
+
+"IF"		{
+			return(PARS_IF_TOKEN);
+}
+
+"THEN"		{
+			return(PARS_THEN_TOKEN);
+}
+
+"ELSE"		{
+			return(PARS_ELSE_TOKEN);
+}
+
+"ELSIF"		{
+			return(PARS_ELSIF_TOKEN);
+}
+
+"LOOP"		{
+			return(PARS_LOOP_TOKEN);
+}
+
+"WHILE"		{
+			return(PARS_WHILE_TOKEN);
+}
+
+"RETURN"	{
+			return(PARS_RETURN_TOKEN);
+}
+
+"SELECT"	{
+			return(PARS_SELECT_TOKEN);
+}
+
+"SUM"		{
+			return(PARS_SUM_TOKEN);
+}
+
+"COUNT"		{
+			return(PARS_COUNT_TOKEN);
+}
+
+"DISTINCT"	{
+			return(PARS_DISTINCT_TOKEN);
+}
+
+"FROM"		{
+			return(PARS_FROM_TOKEN);
+}
+
+"WHERE"		{
+			return(PARS_WHERE_TOKEN);
+}
+
+"FOR"		{
+			return(PARS_FOR_TOKEN);
+}
+
+"READ"		{
+			return(PARS_READ_TOKEN);
+}
+
+"ORDER"		{
+			return(PARS_ORDER_TOKEN);
+}
+
+"BY"		{
+			return(PARS_BY_TOKEN);
+}
+
+"ASC"		{
+			return(PARS_ASC_TOKEN);
+}
+
+"DESC"		{
+			return(PARS_DESC_TOKEN);
+}
+
+"INSERT"	{
+			return(PARS_INSERT_TOKEN);
+}
+
+"INTO"		{
+			return(PARS_INTO_TOKEN);
+}
+
+"VALUES"	{
+			return(PARS_VALUES_TOKEN);
+}
+
+"UPDATE"	{
+			return(PARS_UPDATE_TOKEN);
+}
+
+"SET"		{
+			return(PARS_SET_TOKEN);
+}
+
+"DELETE"	{
+			return(PARS_DELETE_TOKEN);
+}
+
+"CURRENT"	{
+			return(PARS_CURRENT_TOKEN);
+}
+
+"OF"		{
+			return(PARS_OF_TOKEN);
+}
+
+"CREATE"	{
+			return(PARS_CREATE_TOKEN);
+}
+
+"TABLE"		{
+			return(PARS_TABLE_TOKEN);
+}
+
+"INDEX"		{
+	 		return(PARS_INDEX_TOKEN);
+}
+
+"UNIQUE"	{
+	 		return(PARS_UNIQUE_TOKEN);
+}
+
+"CLUSTERED"	{
+	 		return(PARS_CLUSTERED_TOKEN);
+}
+
+"DOES_NOT_FIT_IN_MEMORY"	{
+			return(PARS_DOES_NOT_FIT_IN_MEM_TOKEN);
+}
+
+"ON"		{
+	 		return(PARS_ON_TOKEN);
+}
+
+"DECLARE"	{
+			return(PARS_DECLARE_TOKEN);
+}
+
+"CURSOR"	{
+			return(PARS_CURSOR_TOKEN);
+}
+
+"OPEN"	{
+			return(PARS_OPEN_TOKEN);
+}
+
+"FETCH"	{
+			return(PARS_FETCH_TOKEN);
+}
+
+"CLOSE"	{
+			return(PARS_CLOSE_TOKEN);
+}
+
+"NOTFOUND"	{
+			return(PARS_NOTFOUND_TOKEN);
+}
+
+"TO_CHAR"	{
+			return(PARS_TO_CHAR_TOKEN);
+}
+
+"TO_NUMBER"	{
+			return(PARS_TO_NUMBER_TOKEN);
+}
+
+"TO_BINARY"	{
+			return(PARS_TO_BINARY_TOKEN);
+}
+
+"BINARY_TO_NUMBER" {
+			return(PARS_BINARY_TO_NUMBER_TOKEN);
+}
+
+"SUBSTR"	{
+			return(PARS_SUBSTR_TOKEN);
+}
+
+"REPLSTR"	{
+			return(PARS_REPLSTR_TOKEN);
+}
+
+"CONCAT"	{
+			return(PARS_CONCAT_TOKEN);
+}
+
+"INSTR"		{
+			return(PARS_INSTR_TOKEN);
+}
+
+"LENGTH"	{
+			return(PARS_LENGTH_TOKEN);
+}
+
+"SYSDATE"	{
+			return(PARS_SYSDATE_TOKEN);
+}
+
+"PRINTF"	{
+			return(PARS_PRINTF_TOKEN);
+}
+
+"ASSERT"	{
+			return(PARS_ASSERT_TOKEN);
+}
+
+"RND"		{
+			return(PARS_RND_TOKEN);
+}
+
+"RND_STR"	{
+			return(PARS_RND_STR_TOKEN);
+}
+
+"ROW_PRINTF"	{
+			return(PARS_ROW_PRINTF_TOKEN);
+}
+
+"COMMIT"	{
+			return(PARS_COMMIT_TOKEN);
+}
+
+"ROLLBACK"	{
+			return(PARS_ROLLBACK_TOKEN);
+}
+
+"WORK"		{
+			return(PARS_WORK_TOKEN);
+}
+
+"UNSIGNED"	{
+			return(PARS_UNSIGNED_TOKEN);
+}
+
+"EXIT"		{
+			return(PARS_EXIT_TOKEN);
+}
+
+"FUNCTION"	{
+			return(PARS_FUNCTION_TOKEN);
+}
+
+"LOCK"	{
+			return(PARS_LOCK_TOKEN);
+}
+
+"SHARE"	{
+			return(PARS_SHARE_TOKEN);
+}
+
+"MODE"	{
+			return(PARS_MODE_TOKEN);
+}
+
+{ID}		{
+			yylval = sym_tab_add_id(pars_sym_tab_global,
+							(byte*)yytext,
+							ut_strlen(yytext));
+			return(PARS_ID_TOKEN);
+}
+
+".."		{
+			return(PARS_DDOT_TOKEN);
+}
+
+":="		{
+			return(PARS_ASSIGN_TOKEN);
+}
+
+"<="		{
+			return(PARS_LE_TOKEN);
+}
+
+">="		{
+			return(PARS_GE_TOKEN);
+}
+
+"<>"		{
+			return(PARS_NE_TOKEN);
+}
+
+"("		{
+
+			return((int)(*yytext));
+}
+
+"="		{
+
+			return((int)(*yytext));
+}
+
+">"		{
+
+			return((int)(*yytext));
+}
+
+"<"		{
+
+			return((int)(*yytext));
+}
+
+","		{
+
+			return((int)(*yytext));
+}
+
+";"		{
+
+			return((int)(*yytext));
+}
+
+")"		{
+
+			return((int)(*yytext));
+}
+
+"+" 		{
+
+			return((int)(*yytext));
+}
+
+"-"		{
+
+			return((int)(*yytext));
+}
+
+"*"		{
+
+			return((int)(*yytext));
+}
+
+"/"		{
+
+			return((int)(*yytext));
+}
+
+"%"		{
+
+			return((int)(*yytext));
+}
+
+"{"		{
+
+			return((int)(*yytext));
+}
+
+"}"		{
+
+			return((int)(*yytext));
+}
+
+"?"		{
+
+			return((int)(*yytext));
+}
+
+"/*"			BEGIN(comment); /* eat up comment */
+
+<comment>[^*]*
+<comment>"*"+[^*/]*
+<comment>"*"+"/"        BEGIN(INITIAL);
+
+[ \t\n]+		/* eat up whitespace */
+
+
+.		{
+			fprintf(stderr,"Unrecognized character: %02x\n",
+				*yytext);
+
+			ut_error;
+
+			return(0);
+}
+
+%%
diff --git a/storage/xtradb/pars/pars0opt.c b/storage/xtradb/pars/pars0opt.c
new file mode 100644
index 00000000000..34246929c53
--- /dev/null
+++ b/storage/xtradb/pars/pars0opt.c
@@ -0,0 +1,1224 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Simple SQL optimizer
+
+Created 12/21/1997 Heikki Tuuri
+*******************************************************/
+
+#include "pars0opt.h"
+
+#ifdef UNIV_NONINL
+#include "pars0opt.ic"
+#endif
+
+#include "row0sel.h"
+#include "row0ins.h"
+#include "row0upd.h"
+#include "dict0dict.h"
+#include "dict0mem.h"
+#include "que0que.h"
+#include "pars0grm.h"
+#include "pars0pars.h"
+#include "lock0lock.h"
+
+#define OPT_EQUAL	1	/* comparison by = */
+#define OPT_COMPARISON	2	/* comparison by <, >, <=, or >= */
+
+#define OPT_NOT_COND	1
+#define OPT_END_COND	2
+#define OPT_TEST_COND	3
+#define OPT_SCROLL_COND	4
+
+
+/***********************************************************************
+Inverts a comparison operator. */
+static
+int
+opt_invert_cmp_op(
+/*==============*/
+			/* out: the equivalent operator when the order of
+			the arguments is switched */
+	int	op)	/* in: operator */
+{
+	if (op == '<') {
+		return('>');
+	} else if (op == '>') {
+		return('<');
+	} else if (op == '=') {
+		return('=');
+	} else if (op == PARS_LE_TOKEN) {
+		return(PARS_GE_TOKEN);
+	} else if (op == PARS_GE_TOKEN) {
+		return(PARS_LE_TOKEN);
+	} else {
+		ut_error;
+	}
+
+	return(0);
+}
+
+/***********************************************************************
+Checks if the value of an expression can be calculated BEFORE the nth table
+in a join is accessed. If this is the case, it can possibly be used in an
+index search for the nth table. */
+static
+ibool
+opt_check_exp_determined_before(
+/*============================*/
+					/* out: TRUE if already determined */
+	que_node_t*	exp,		/* in: expression */
+	sel_node_t*	sel_node,	/* in: select node */
+	ulint		nth_table)	/* in: nth table will be accessed */
+{
+	func_node_t*	func_node;
+	sym_node_t*	sym_node;
+	dict_table_t*	table;
+	que_node_t*	arg;
+	ulint		i;
+
+	ut_ad(exp && sel_node);
+
+	if (que_node_get_type(exp) == QUE_NODE_FUNC) {
+		func_node = exp;
+
+		arg = func_node->args;
+
+		while (arg) {
+			if (!opt_check_exp_determined_before(arg, sel_node,
+							     nth_table)) {
+				return(FALSE);
+			}
+
+			arg = que_node_get_next(arg);
+		}
+
+		return(TRUE);
+	}
+
+	ut_a(que_node_get_type(exp) == QUE_NODE_SYMBOL);
+
+	sym_node = exp;
+
+	if (sym_node->token_type != SYM_COLUMN) {
+
+		return(TRUE);
+	}
+
+	for (i = 0; i < nth_table; i++) {
+
+		table = sel_node_get_nth_plan(sel_node, i)->table;
+
+		if (sym_node->table == table) {
+
+			return(TRUE);
+		}
+	}
+
+	return(FALSE);
+}
+
+/***********************************************************************
+Looks in a comparison condition if a column value is already restricted by
+it BEFORE the nth table is accessed. */
+static
+que_node_t*
+opt_look_for_col_in_comparison_before(
+/*==================================*/
+					/* out: expression restricting the
+					value of the column, or NULL if not
+					known */
+	ulint		cmp_type,	/* in: OPT_EQUAL, OPT_COMPARISON */
+	ulint		col_no,		/* in: column number */
+	func_node_t*	search_cond,	/* in: comparison condition */
+	sel_node_t*	sel_node,	/* in: select node */
+	ulint		nth_table,	/* in: nth table in a join (a query
+					from a single table is considered a
+					join of 1 table) */
+	ulint*		op)		/* out: comparison operator ('=',
+					PARS_GE_TOKEN, ... ); this is inverted
+					if the column appears on the right
+					side */
+{
+	sym_node_t*	sym_node;
+	dict_table_t*	table;
+	que_node_t*	exp;
+	que_node_t*	arg;
+
+	ut_ad(search_cond);
+
+	ut_a((search_cond->func == '<')
+	     || (search_cond->func == '>')
+	     || (search_cond->func == '=')
+	     || (search_cond->func == PARS_GE_TOKEN)
+	     || (search_cond->func == PARS_LE_TOKEN));
+
+	table = sel_node_get_nth_plan(sel_node, nth_table)->table;
+
+	if ((cmp_type == OPT_EQUAL) && (search_cond->func != '=')) {
+
+		return(NULL);
+
+	} else if ((cmp_type == OPT_COMPARISON)
+		   && (search_cond->func != '<')
+		   && (search_cond->func != '>')
+		   && (search_cond->func != PARS_GE_TOKEN)
+		   && (search_cond->func != PARS_LE_TOKEN)) {
+
+		return(NULL);
+	}
+
+	arg = search_cond->args;
+
+	if (que_node_get_type(arg) == QUE_NODE_SYMBOL) {
+		sym_node = arg;
+
+		if ((sym_node->token_type == SYM_COLUMN)
+		    && (sym_node->table == table)
+		    && (sym_node->col_no == col_no)) {
+
+			/* sym_node contains the desired column id */
+
+			/* Check if the expression on the right side of the
+			operator is already determined */
+
+			exp = que_node_get_next(arg);
+
+			if (opt_check_exp_determined_before(exp, sel_node,
+							    nth_table)) {
+				*op = search_cond->func;
+
+				return(exp);
+			}
+		}
+	}
+
+	exp = search_cond->args;
+	arg = que_node_get_next(arg);
+
+	if (que_node_get_type(arg) == QUE_NODE_SYMBOL) {
+		sym_node = arg;
+
+		if ((sym_node->token_type == SYM_COLUMN)
+		    && (sym_node->table == table)
+		    && (sym_node->col_no == col_no)) {
+
+			if (opt_check_exp_determined_before(exp, sel_node,
+							    nth_table)) {
+				*op = opt_invert_cmp_op(search_cond->func);
+
+				return(exp);
+			}
+		}
+	}
+
+	return(NULL);
+}
+
+/***********************************************************************
+Looks in a search condition if a column value is already restricted by the
+search condition BEFORE the nth table is accessed. Takes into account that
+if we will fetch in an ascending order, we cannot utilize an upper limit for
+a column value; in a descending order, respectively, a lower limit. */
+static
+que_node_t*
+opt_look_for_col_in_cond_before(
+/*============================*/
+					/* out: expression restricting the
+					value of the column, or NULL if not
+					known */
+	ulint		cmp_type,	/* in: OPT_EQUAL, OPT_COMPARISON */
+	ulint		col_no,		/* in: column number */
+	func_node_t*	search_cond,	/* in: search condition or NULL */
+	sel_node_t*	sel_node,	/* in: select node */
+	ulint		nth_table,	/* in: nth table in a join (a query
+					from a single table is considered a
+					join of 1 table) */
+	ulint*		op)		/* out: comparison operator ('=',
+					PARS_GE_TOKEN, ... ) */
+{
+	func_node_t*	new_cond;
+	que_node_t*	exp;
+
+	if (search_cond == NULL) {
+
+		return(NULL);
+	}
+
+	ut_a(que_node_get_type(search_cond) == QUE_NODE_FUNC);
+	ut_a(search_cond->func != PARS_OR_TOKEN);
+	ut_a(search_cond->func != PARS_NOT_TOKEN);
+
+	if (search_cond->func == PARS_AND_TOKEN) {
+		new_cond = search_cond->args;
+
+		exp = opt_look_for_col_in_cond_before(cmp_type, col_no,
+						      new_cond, sel_node,
+						      nth_table, op);
+		if (exp) {
+
+			return(exp);
+		}
+
+		new_cond = que_node_get_next(new_cond);
+
+		exp = opt_look_for_col_in_cond_before(cmp_type, col_no,
+						      new_cond, sel_node,
+						      nth_table, op);
+		return(exp);
+	}
+
+	exp = opt_look_for_col_in_comparison_before(cmp_type, col_no,
+						    search_cond, sel_node,
+						    nth_table, op);
+	if (exp == NULL) {
+
+		return(NULL);
+	}
+
+	/* If we will fetch in an ascending order, we cannot utilize an upper
+	limit for a column value; in a descending order, respectively, a lower
+	limit */
+
+	if (sel_node->asc && ((*op == '<') || (*op == PARS_LE_TOKEN))) {
+
+		return(NULL);
+
+	} else if (!sel_node->asc
+		   && ((*op == '>') || (*op == PARS_GE_TOKEN))) {
+
+		return(NULL);
+	}
+
+	return(exp);
+}
+
+/***********************************************************************
+Calculates the goodness for an index according to a select node. The
+goodness is 4 times the number of first fields in index whose values we
+already know exactly in the query. If we have a comparison condition for
+an additional field, 2 point are added. If the index is unique, and we know
+all the unique fields for the index we add 1024 points. For a clustered index
+we add 1 point. */
+static
+ulint
+opt_calc_index_goodness(
+/*====================*/
+					/* out: goodness */
+	dict_index_t*	index,		/* in: index */
+	sel_node_t*	sel_node,	/* in: parsed select node */
+	ulint		nth_table,	/* in: nth table in a join */
+	que_node_t**	index_plan,	/* in/out: comparison expressions for
+					this index */
+	ulint*		last_op)	/* out: last comparison operator, if
+					goodness > 1 */
+{
+	que_node_t*	exp;
+	ulint		goodness;
+	ulint		n_fields;
+	ulint		col_no;
+	ulint		op;
+	ulint		j;
+
+	goodness = 0;
+
+	/* Note that as higher level node pointers in the B-tree contain
+	page addresses as the last field, we must not put more fields in
+	the search tuple than dict_index_get_n_unique_in_tree(index); see
+	the note in btr_cur_search_to_nth_level. */
+
+	n_fields = dict_index_get_n_unique_in_tree(index);
+
+	for (j = 0; j < n_fields; j++) {
+
+		col_no = dict_index_get_nth_col_no(index, j);
+
+		exp = opt_look_for_col_in_cond_before(
+			OPT_EQUAL, col_no, sel_node->search_cond,
+			sel_node, nth_table, &op);
+		if (exp) {
+			/* The value for this column is exactly known already
+			at this stage of the join */
+
+			index_plan[j] = exp;
+			*last_op = op;
+			goodness += 4;
+		} else {
+			/* Look for non-equality comparisons */
+
+			exp = opt_look_for_col_in_cond_before(
+				OPT_COMPARISON, col_no, sel_node->search_cond,
+				sel_node, nth_table, &op);
+			if (exp) {
+				index_plan[j] = exp;
+				*last_op = op;
+				goodness += 2;
+			}
+
+			break;
+		}
+	}
+
+	if (goodness >= 4 * dict_index_get_n_unique(index)) {
+		goodness += 1024;
+
+		if (dict_index_is_clust(index)) {
+
+			goodness += 1024;
+		}
+	}
+
+	/* We have to test for goodness here, as last_op may note be set */
+	if (goodness && dict_index_is_clust(index)) {
+
+		goodness++;
+	}
+
+	return(goodness);
+}
+
+/***********************************************************************
+Calculates the number of matched fields based on an index goodness. */
+UNIV_INLINE
+ulint
+opt_calc_n_fields_from_goodness(
+/*============================*/
+				/* out: number of excatly or partially matched
+				fields */
+	ulint	goodness)	/* in: goodness */
+{
+	return(((goodness % 1024) + 2) / 4);
+}
+
+/***********************************************************************
+Converts a comparison operator to the corresponding search mode PAGE_CUR_GE,
+... */
+UNIV_INLINE
+ulint
+opt_op_to_search_mode(
+/*==================*/
+			/* out: search mode */
+	ibool	asc,	/* in: TRUE if the rows should be fetched in an
+			ascending order */
+	ulint	op)	/* in: operator '=', PARS_GE_TOKEN, ... */
+{
+	if (op == '=') {
+		if (asc) {
+			return(PAGE_CUR_GE);
+		} else {
+			return(PAGE_CUR_LE);
+		}
+	} else if (op == '<') {
+		ut_a(!asc);
+		return(PAGE_CUR_L);
+	} else if (op == '>') {
+		ut_a(asc);
+		return(PAGE_CUR_G);
+	} else if (op == PARS_GE_TOKEN) {
+		ut_a(asc);
+		return(PAGE_CUR_GE);
+	} else if (op == PARS_LE_TOKEN) {
+		ut_a(!asc);
+		return(PAGE_CUR_LE);
+	} else {
+		ut_error;
+	}
+
+	return(0);
+}
+
+/***********************************************************************
+Determines if a node is an argument node of a function node. */
+static
+ibool
+opt_is_arg(
+/*=======*/
+					/* out: TRUE if is an argument */
+	que_node_t*	arg_node,	/* in: possible argument node */
+	func_node_t*	func_node)	/* in: function node */
+{
+	que_node_t*	arg;
+
+	arg = func_node->args;
+
+	while (arg) {
+		if (arg == arg_node) {
+
+			return(TRUE);
+		}
+
+		arg = que_node_get_next(arg);
+	}
+
+	return(FALSE);
+}
+
+/***********************************************************************
+Decides if the fetching of rows should be made in a descending order, and
+also checks that the chosen query plan produces a result which satisfies
+the order-by. */
+static
+void
+opt_check_order_by(
+/*===============*/
+	sel_node_t*	sel_node)	/* in: select node; asserts an error
+					if the plan does not agree with the
+					order-by */
+{
+	order_node_t*	order_node;
+	dict_table_t*	order_table;
+	ulint		order_col_no;
+	plan_t*		plan;
+	ulint		i;
+
+	if (!sel_node->order_by) {
+
+		return;
+	}
+
+	order_node = sel_node->order_by;
+	order_col_no = order_node->column->col_no;
+	order_table = order_node->column->table;
+
+	/* If there is an order-by clause, the first non-exactly matched field
+	in the index used for the last table in the table list should be the
+	column defined in the order-by clause, and for all the other tables
+	we should get only at most a single row, otherwise we cannot presently
+	calculate the order-by, as we have no sort utility */
+
+	for (i = 0; i < sel_node->n_tables; i++) {
+
+		plan = sel_node_get_nth_plan(sel_node, i);
+
+		if (i < sel_node->n_tables - 1) {
+			ut_a(dict_index_get_n_unique(plan->index)
+			     <= plan->n_exact_match);
+		} else {
+			ut_a(plan->table == order_table);
+
+			ut_a((dict_index_get_n_unique(plan->index)
+			      <= plan->n_exact_match)
+			     || (dict_index_get_nth_col_no(plan->index,
+							   plan->n_exact_match)
+				 == order_col_no));
+		}
+	}
+}
+
+/***********************************************************************
+Optimizes a select. Decides which indexes to tables to use. The tables
+are accessed in the order that they were written to the FROM part in the
+select statement. */
+static
+void
+opt_search_plan_for_table(
+/*======================*/
+	sel_node_t*	sel_node,	/* in: parsed select node */
+	ulint		i,		/* in: this is the ith table */
+	dict_table_t*	table)		/* in: table */
+{
+	plan_t*		plan;
+	dict_index_t*	index;
+	dict_index_t*	best_index;
+	ulint		n_fields;
+	ulint		goodness;
+	ulint		last_op		= 75946965;	/* Eliminate a Purify
+							warning */
+	ulint		best_goodness;
+	ulint		best_last_op = 0; /* remove warning */
+	que_node_t*	index_plan[256];
+	que_node_t*	best_index_plan[256];
+
+	plan = sel_node_get_nth_plan(sel_node, i);
+
+	plan->table = table;
+	plan->asc = sel_node->asc;
+	plan->pcur_is_open = FALSE;
+	plan->cursor_at_end = FALSE;
+
+	/* Calculate goodness for each index of the table */
+
+	index = dict_table_get_first_index(table);
+	best_index = index; /* Eliminate compiler warning */
+	best_goodness = 0;
+
+	/* should be do ... until ? comment by Jani */
+	while (index) {
+		goodness = opt_calc_index_goodness(index, sel_node, i,
+						   index_plan, &last_op);
+		if (goodness > best_goodness) {
+
+			best_index = index;
+			best_goodness = goodness;
+			n_fields = opt_calc_n_fields_from_goodness(goodness);
+
+			ut_memcpy(best_index_plan, index_plan,
+				  n_fields * sizeof(void*));
+			best_last_op = last_op;
+		}
+
+		index = dict_table_get_next_index(index);
+	}
+
+	plan->index = best_index;
+
+	n_fields = opt_calc_n_fields_from_goodness(best_goodness);
+
+	if (n_fields == 0) {
+		plan->tuple = NULL;
+		plan->n_exact_match = 0;
+	} else {
+		plan->tuple = dtuple_create(pars_sym_tab_global->heap,
+					    n_fields);
+		dict_index_copy_types(plan->tuple, plan->index, n_fields);
+
+		plan->tuple_exps = mem_heap_alloc(pars_sym_tab_global->heap,
+						  n_fields * sizeof(void*));
+
+		ut_memcpy(plan->tuple_exps, best_index_plan,
+			  n_fields * sizeof(void*));
+		if (best_last_op == '=') {
+			plan->n_exact_match = n_fields;
+		} else {
+			plan->n_exact_match = n_fields - 1;
+		}
+
+		plan->mode = opt_op_to_search_mode(sel_node->asc,
+						   best_last_op);
+	}
+
+	if (dict_index_is_clust(best_index)
+	    && (plan->n_exact_match >= dict_index_get_n_unique(best_index))) {
+
+		plan->unique_search = TRUE;
+	} else {
+		plan->unique_search = FALSE;
+	}
+
+	plan->old_vers_heap = NULL;
+
+	btr_pcur_init(&(plan->pcur));
+	btr_pcur_init(&(plan->clust_pcur));
+}
+
+/***********************************************************************
+Looks at a comparison condition and decides if it can, and need, be tested for
+a table AFTER the table has been accessed. */
+static
+ulint
+opt_classify_comparison(
+/*====================*/
+					/* out: OPT_NOT_COND if not for this
+					table, else OPT_END_COND,
+					OPT_TEST_COND, or OPT_SCROLL_COND,
+					where the last means that the
+					condition need not be tested, except
+					when scroll cursors are used */
+	sel_node_t*	sel_node,	/* in: select node */
+	ulint		i,		/* in: ith table in the join */
+	func_node_t*	cond)		/* in: comparison condition */
+{
+	plan_t*	plan;
+	ulint	n_fields;
+	ulint	op;
+	ulint	j;
+
+	ut_ad(cond && sel_node);
+
+	plan = sel_node_get_nth_plan(sel_node, i);
+
+	/* Check if the condition is determined after the ith table has been
+	accessed, but not after the i - 1:th */
+
+	if (!opt_check_exp_determined_before(cond, sel_node, i + 1)) {
+
+		return(OPT_NOT_COND);
+	}
+
+	if ((i > 0) && opt_check_exp_determined_before(cond, sel_node, i)) {
+
+		return(OPT_NOT_COND);
+	}
+
+	/* If the condition is an exact match condition used in constructing
+	the search tuple, it is classified as OPT_END_COND */
+
+	if (plan->tuple) {
+		n_fields = dtuple_get_n_fields(plan->tuple);
+	} else {
+		n_fields = 0;
+	}
+
+	for (j = 0; j < plan->n_exact_match; j++) {
+
+		if (opt_is_arg(plan->tuple_exps[j], cond)) {
+
+			return(OPT_END_COND);
+		}
+	}
+
+	/* If the condition is an non-exact match condition used in
+	constructing the search tuple, it is classified as OPT_SCROLL_COND.
+	When the cursor is positioned, and if a non-scroll cursor is used,
+	there is no need to test this condition; if a scroll cursor is used
+	the testing is necessary when the cursor is reversed. */
+
+	if ((n_fields > plan->n_exact_match)
+	    && opt_is_arg(plan->tuple_exps[n_fields - 1], cond)) {
+
+		return(OPT_SCROLL_COND);
+	}
+
+	/* If the condition is a non-exact match condition on the first field
+	in index for which there is no exact match, and it limits the search
+	range from the opposite side of the search tuple already BEFORE we
+	access the table, it is classified as OPT_END_COND */
+
+	if ((dict_index_get_n_fields(plan->index) > plan->n_exact_match)
+	    && opt_look_for_col_in_comparison_before(
+		    OPT_COMPARISON,
+		    dict_index_get_nth_col_no(plan->index,
+					      plan->n_exact_match),
+		    cond, sel_node, i, &op)) {
+
+		if (sel_node->asc && ((op == '<') || (op == PARS_LE_TOKEN))) {
+
+			return(OPT_END_COND);
+		}
+
+		if (!sel_node->asc && ((op == '>') || (op == PARS_GE_TOKEN))) {
+
+			return(OPT_END_COND);
+		}
+	}
+
+	/* Otherwise, cond is classified as OPT_TEST_COND */
+
+	return(OPT_TEST_COND);
+}
+
+/***********************************************************************
+Recursively looks for test conditions for a table in a join. */
+static
+void
+opt_find_test_conds(
+/*================*/
+	sel_node_t*	sel_node,	/* in: select node */
+	ulint		i,		/* in: ith table in the join */
+	func_node_t*	cond)		/* in: conjunction of search
+					conditions or NULL */
+{
+	func_node_t*	new_cond;
+	ulint		class;
+	plan_t*		plan;
+
+	if (cond == NULL) {
+
+		return;
+	}
+
+	if (cond->func == PARS_AND_TOKEN) {
+		new_cond = cond->args;
+
+		opt_find_test_conds(sel_node, i, new_cond);
+
+		new_cond = que_node_get_next(new_cond);
+
+		opt_find_test_conds(sel_node, i, new_cond);
+
+		return;
+	}
+
+	plan = sel_node_get_nth_plan(sel_node, i);
+
+	class = opt_classify_comparison(sel_node, i, cond);
+
+	if (class == OPT_END_COND) {
+		UT_LIST_ADD_LAST(cond_list, plan->end_conds, cond);
+
+	} else if (class == OPT_TEST_COND) {
+		UT_LIST_ADD_LAST(cond_list, plan->other_conds, cond);
+
+	}
+}
+
+/***********************************************************************
+Normalizes a list of comparison conditions so that a column of the table
+appears on the left side of the comparison if possible. This is accomplished
+by switching the arguments of the operator. */
+static
+void
+opt_normalize_cmp_conds(
+/*====================*/
+	func_node_t*	cond,	/* in: first in a list of comparison
+				conditions, or NULL */
+	dict_table_t*	table)	/* in: table */
+{
+	que_node_t*	arg1;
+	que_node_t*	arg2;
+	sym_node_t*	sym_node;
+
+	while (cond) {
+		arg1 = cond->args;
+		arg2 = que_node_get_next(arg1);
+
+		if (que_node_get_type(arg2) == QUE_NODE_SYMBOL) {
+
+			sym_node = arg2;
+
+			if ((sym_node->token_type == SYM_COLUMN)
+			    && (sym_node->table == table)) {
+
+				/* Switch the order of the arguments */
+
+				cond->args = arg2;
+				que_node_list_add_last(NULL, arg2);
+				que_node_list_add_last(arg2, arg1);
+
+				/* Invert the operator */
+				cond->func = opt_invert_cmp_op(cond->func);
+			}
+		}
+
+		cond = UT_LIST_GET_NEXT(cond_list, cond);
+	}
+}
+
+/***********************************************************************
+Finds out the search condition conjuncts we can, and need, to test as the ith
+table in a join is accessed. The search tuple can eliminate the need to test
+some conjuncts. */
+static
+void
+opt_determine_and_normalize_test_conds(
+/*===================================*/
+	sel_node_t*	sel_node,	/* in: select node */
+	ulint		i)		/* in: ith table in the join */
+{
+	plan_t*	plan;
+
+	plan = sel_node_get_nth_plan(sel_node, i);
+
+	UT_LIST_INIT(plan->end_conds);
+	UT_LIST_INIT(plan->other_conds);
+
+	/* Recursively go through the conjuncts and classify them */
+
+	opt_find_test_conds(sel_node, i, sel_node->search_cond);
+
+	opt_normalize_cmp_conds(UT_LIST_GET_FIRST(plan->end_conds),
+				plan->table);
+
+	ut_a(UT_LIST_GET_LEN(plan->end_conds) >= plan->n_exact_match);
+}
+
+/***********************************************************************
+Looks for occurrences of the columns of the table in the query subgraph and
+adds them to the list of columns if an occurrence of the same column does not
+already exist in the list. If the column is already in the list, puts a value
+indirection to point to the occurrence in the column list, except if the
+column occurrence we are looking at is in the column list, in which case
+nothing is done. */
+UNIV_INTERN
+void
+opt_find_all_cols(
+/*==============*/
+	ibool		copy_val,	/* in: if TRUE, new found columns are
+					added as columns to copy */
+	dict_index_t*	index,		/* in: index of the table to use */
+	sym_node_list_t* col_list,	/* in: base node of a list where
+					to add new found columns */
+	plan_t*		plan,		/* in: plan or NULL */
+	que_node_t*	exp)		/* in: expression or condition or
+					NULL */
+{
+	func_node_t*	func_node;
+	que_node_t*	arg;
+	sym_node_t*	sym_node;
+	sym_node_t*	col_node;
+	ulint		col_pos;
+
+	if (exp == NULL) {
+
+		return;
+	}
+
+	if (que_node_get_type(exp) == QUE_NODE_FUNC) {
+		func_node = exp;
+
+		arg = func_node->args;
+
+		while (arg) {
+			opt_find_all_cols(copy_val, index, col_list, plan,
+					  arg);
+			arg = que_node_get_next(arg);
+		}
+
+		return;
+	}
+
+	ut_a(que_node_get_type(exp) == QUE_NODE_SYMBOL);
+
+	sym_node = exp;
+
+	if (sym_node->token_type != SYM_COLUMN) {
+
+		return;
+	}
+
+	if (sym_node->table != index->table) {
+
+		return;
+	}
+
+	/* Look for an occurrence of the same column in the plan column
+	list */
+
+	col_node = UT_LIST_GET_FIRST(*col_list);
+
+	while (col_node) {
+		if (col_node->col_no == sym_node->col_no) {
+
+			if (col_node == sym_node) {
+				/* sym_node was already in a list: do
+				nothing */
+
+				return;
+			}
+
+			/* Put an indirection */
+			sym_node->indirection = col_node;
+			sym_node->alias = col_node;
+
+			return;
+		}
+
+		col_node = UT_LIST_GET_NEXT(col_var_list, col_node);
+	}
+
+	/* The same column did not occur in the list: add it */
+
+	UT_LIST_ADD_LAST(col_var_list, *col_list, sym_node);
+
+	sym_node->copy_val = copy_val;
+
+	/* Fill in the field_no fields in sym_node */
+
+	sym_node->field_nos[SYM_CLUST_FIELD_NO] = dict_index_get_nth_col_pos(
+		dict_table_get_first_index(index->table), sym_node->col_no);
+	if (!dict_index_is_clust(index)) {
+
+		ut_a(plan);
+
+		col_pos = dict_index_get_nth_col_pos(index, sym_node->col_no);
+
+		if (col_pos == ULINT_UNDEFINED) {
+
+			plan->must_get_clust = TRUE;
+		}
+
+		sym_node->field_nos[SYM_SEC_FIELD_NO] = col_pos;
+	}
+}
+
+/***********************************************************************
+Looks for occurrences of the columns of the table in conditions which are
+not yet determined AFTER the join operation has fetched a row in the ith
+table. The values for these column must be copied to dynamic memory for
+later use. */
+static
+void
+opt_find_copy_cols(
+/*===============*/
+	sel_node_t*	sel_node,	/* in: select node */
+	ulint		i,		/* in: ith table in the join */
+	func_node_t*	search_cond)	/* in: search condition or NULL */
+{
+	func_node_t*	new_cond;
+	plan_t*		plan;
+
+	if (search_cond == NULL) {
+
+		return;
+	}
+
+	ut_ad(que_node_get_type(search_cond) == QUE_NODE_FUNC);
+
+	if (search_cond->func == PARS_AND_TOKEN) {
+		new_cond = search_cond->args;
+
+		opt_find_copy_cols(sel_node, i, new_cond);
+
+		new_cond = que_node_get_next(new_cond);
+
+		opt_find_copy_cols(sel_node, i, new_cond);
+
+		return;
+	}
+
+	if (!opt_check_exp_determined_before(search_cond, sel_node, i + 1)) {
+
+		/* Any ith table columns occurring in search_cond should be
+		copied, as this condition cannot be tested already on the
+		fetch from the ith table */
+
+		plan = sel_node_get_nth_plan(sel_node, i);
+
+		opt_find_all_cols(TRUE, plan->index, &(plan->columns), plan,
+				  search_cond);
+	}
+}
+
+/***********************************************************************
+Classifies the table columns according to whether we use the column only while
+holding the latch on the page, or whether we have to copy the column value to
+dynamic memory. Puts the first occurrence of a column to either list in the
+plan node, and puts indirections to later occurrences of the column. */
+static
+void
+opt_classify_cols(
+/*==============*/
+	sel_node_t*	sel_node,	/* in: select node */
+	ulint		i)		/* in: ith table in the join */
+{
+	plan_t*		plan;
+	que_node_t*	exp;
+
+	plan = sel_node_get_nth_plan(sel_node, i);
+
+	/* The final value of the following field will depend on the
+	environment of the select statement: */
+
+	plan->must_get_clust = FALSE;
+
+	UT_LIST_INIT(plan->columns);
+
+	/* All select list columns should be copied: therefore TRUE as the
+	first argument */
+
+	exp = sel_node->select_list;
+
+	while (exp) {
+		opt_find_all_cols(TRUE, plan->index, &(plan->columns), plan,
+				  exp);
+		exp = que_node_get_next(exp);
+	}
+
+	opt_find_copy_cols(sel_node, i, sel_node->search_cond);
+
+	/* All remaining columns in the search condition are temporary
+	columns: therefore FALSE */
+
+	opt_find_all_cols(FALSE, plan->index, &(plan->columns), plan,
+			  sel_node->search_cond);
+}
+
+/***********************************************************************
+Fills in the info in plan which is used in accessing a clustered index
+record. The columns must already be classified for the plan node. */
+static
+void
+opt_clust_access(
+/*=============*/
+	sel_node_t*	sel_node,	/* in: select node */
+	ulint		n)		/* in: nth table in select */
+{
+	plan_t*		plan;
+	dict_table_t*	table;
+	dict_index_t*	clust_index;
+	dict_index_t*	index;
+	mem_heap_t*	heap;
+	ulint		n_fields;
+	ulint		pos;
+	ulint		i;
+
+	plan = sel_node_get_nth_plan(sel_node, n);
+
+	index = plan->index;
+
+	/* The final value of the following field depends on the environment
+	of the select statement: */
+
+	plan->no_prefetch = FALSE;
+
+	if (dict_index_is_clust(index)) {
+		plan->clust_map = NULL;
+		plan->clust_ref = NULL;
+
+		return;
+	}
+
+	table = index->table;
+
+	clust_index = dict_table_get_first_index(table);
+
+	n_fields = dict_index_get_n_unique(clust_index);
+
+	heap = pars_sym_tab_global->heap;
+
+	plan->clust_ref = dtuple_create(heap, n_fields);
+
+	dict_index_copy_types(plan->clust_ref, clust_index, n_fields);
+
+	plan->clust_map = mem_heap_alloc(heap, n_fields * sizeof(ulint));
+
+	for (i = 0; i < n_fields; i++) {
+		pos = dict_index_get_nth_field_pos(index, clust_index, i);
+
+		ut_a(pos != ULINT_UNDEFINED);
+
+		/* We optimize here only queries to InnoDB's internal system
+		tables, and they should not contain column prefix indexes. */
+
+		if (dict_index_get_nth_field(index, pos)->prefix_len != 0
+		    || dict_index_get_nth_field(clust_index, i)
+		    ->prefix_len != 0) {
+			fprintf(stderr,
+				"InnoDB: Error in pars0opt.c:"
+				" table %s has prefix_len != 0\n",
+				index->table_name);
+		}
+
+		*(plan->clust_map + i) = pos;
+
+		ut_ad(pos != ULINT_UNDEFINED);
+	}
+}
+
+/***********************************************************************
+Optimizes a select. Decides which indexes to tables to use. The tables
+are accessed in the order that they were written to the FROM part in the
+select statement. */
+UNIV_INTERN
+void
+opt_search_plan(
+/*============*/
+	sel_node_t*	sel_node)	/* in: parsed select node */
+{
+	sym_node_t*	table_node;
+	dict_table_t*	table;
+	order_node_t*	order_by;
+	ulint		i;
+
+	sel_node->plans = mem_heap_alloc(pars_sym_tab_global->heap,
+					 sel_node->n_tables * sizeof(plan_t));
+
+	/* Analyze the search condition to find out what we know at each
+	join stage about the conditions that the columns of a table should
+	satisfy */
+
+	table_node = sel_node->table_list;
+
+	if (sel_node->order_by == NULL) {
+		sel_node->asc = TRUE;
+	} else {
+		order_by = sel_node->order_by;
+
+		sel_node->asc = order_by->asc;
+	}
+
+	for (i = 0; i < sel_node->n_tables; i++) {
+
+		table = table_node->table;
+
+		/* Choose index through which to access the table */
+
+		opt_search_plan_for_table(sel_node, i, table);
+
+		/* Determine the search condition conjuncts we can test at
+		this table; normalize the end conditions */
+
+		opt_determine_and_normalize_test_conds(sel_node, i);
+
+		table_node = que_node_get_next(table_node);
+	}
+
+	table_node = sel_node->table_list;
+
+	for (i = 0; i < sel_node->n_tables; i++) {
+
+		/* Classify the table columns into those we only need to access
+		but not copy, and to those we must copy to dynamic memory */
+
+		opt_classify_cols(sel_node, i);
+
+		/* Calculate possible info for accessing the clustered index
+		record */
+
+		opt_clust_access(sel_node, i);
+
+		table_node = que_node_get_next(table_node);
+	}
+
+	/* Check that the plan obeys a possible order-by clause: if not,
+	an assertion error occurs */
+
+	opt_check_order_by(sel_node);
+
+#ifdef UNIV_SQL_DEBUG
+	opt_print_query_plan(sel_node);
+#endif
+}
+
+/************************************************************************
+Prints info of a query plan. */
+UNIV_INTERN
+void
+opt_print_query_plan(
+/*=================*/
+	sel_node_t*	sel_node)	/* in: select node */
+{
+	plan_t*	plan;
+	ulint	n_fields;
+	ulint	i;
+
+	fputs("QUERY PLAN FOR A SELECT NODE\n", stderr);
+
+	fputs(sel_node->asc ? "Asc. search; " : "Desc. search; ", stderr);
+
+	if (sel_node->set_x_locks) {
+		fputs("sets row x-locks; ", stderr);
+		ut_a(sel_node->row_lock_mode == LOCK_X);
+		ut_a(!sel_node->consistent_read);
+	} else if (sel_node->consistent_read) {
+		fputs("consistent read; ", stderr);
+	} else {
+		ut_a(sel_node->row_lock_mode == LOCK_S);
+		fputs("sets row s-locks; ", stderr);
+	}
+
+	putc('\n', stderr);
+
+	for (i = 0; i < sel_node->n_tables; i++) {
+		plan = sel_node_get_nth_plan(sel_node, i);
+
+		if (plan->tuple) {
+			n_fields = dtuple_get_n_fields(plan->tuple);
+		} else {
+			n_fields = 0;
+		}
+
+		fputs("Table ", stderr);
+		dict_index_name_print(stderr, NULL, plan->index);
+		fprintf(stderr,"; exact m. %lu, match %lu, end conds %lu\n",
+			(unsigned long) plan->n_exact_match,
+			(unsigned long) n_fields,
+			(unsigned long) UT_LIST_GET_LEN(plan->end_conds));
+	}
+}
diff --git a/storage/xtradb/pars/pars0pars.c b/storage/xtradb/pars/pars0pars.c
new file mode 100644
index 00000000000..62ae3b3d09b
--- /dev/null
+++ b/storage/xtradb/pars/pars0pars.c
@@ -0,0 +1,2205 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+SQL parser
+
+Created 11/19/1996 Heikki Tuuri
+*******************************************************/
+
+/* Historical note: Innobase executed its first SQL string (CREATE TABLE)
+on 1/27/1998 */
+
+#include "pars0pars.h"
+
+#ifdef UNIV_NONINL
+#include "pars0pars.ic"
+#endif
+
+#include "row0sel.h"
+#include "row0ins.h"
+#include "row0upd.h"
+#include "dict0dict.h"
+#include "dict0mem.h"
+#include "dict0crea.h"
+#include "que0que.h"
+#include "pars0grm.h"
+#include "pars0opt.h"
+#include "data0data.h"
+#include "data0type.h"
+#include "trx0trx.h"
+#include "trx0roll.h"
+#include "lock0lock.h"
+#include "eval0eval.h"
+
+#ifdef UNIV_SQL_DEBUG
+/* If the following is set TRUE, the lexer will print the SQL string
+as it tokenizes it */
+
+ibool	pars_print_lexed	= FALSE;
+#endif /* UNIV_SQL_DEBUG */
+
+/* Global variable used while parsing a single procedure or query : the code is
+NOT re-entrant */
+UNIV_INTERN sym_tab_t*	pars_sym_tab_global;
+
+/* Global variables used to denote certain reserved words, used in
+constructing the parsing tree */
+
+UNIV_INTERN pars_res_word_t	pars_to_char_token = {PARS_TO_CHAR_TOKEN};
+UNIV_INTERN pars_res_word_t	pars_to_number_token = {PARS_TO_NUMBER_TOKEN};
+UNIV_INTERN pars_res_word_t	pars_to_binary_token = {PARS_TO_BINARY_TOKEN};
+UNIV_INTERN pars_res_word_t	pars_binary_to_number_token = {PARS_BINARY_TO_NUMBER_TOKEN};
+UNIV_INTERN pars_res_word_t	pars_substr_token = {PARS_SUBSTR_TOKEN};
+UNIV_INTERN pars_res_word_t	pars_replstr_token = {PARS_REPLSTR_TOKEN};
+UNIV_INTERN pars_res_word_t	pars_concat_token = {PARS_CONCAT_TOKEN};
+UNIV_INTERN pars_res_word_t	pars_instr_token = {PARS_INSTR_TOKEN};
+UNIV_INTERN pars_res_word_t	pars_length_token = {PARS_LENGTH_TOKEN};
+UNIV_INTERN pars_res_word_t	pars_sysdate_token = {PARS_SYSDATE_TOKEN};
+UNIV_INTERN pars_res_word_t	pars_printf_token = {PARS_PRINTF_TOKEN};
+UNIV_INTERN pars_res_word_t	pars_assert_token = {PARS_ASSERT_TOKEN};
+UNIV_INTERN pars_res_word_t	pars_rnd_token = {PARS_RND_TOKEN};
+UNIV_INTERN pars_res_word_t	pars_rnd_str_token = {PARS_RND_STR_TOKEN};
+UNIV_INTERN pars_res_word_t	pars_count_token = {PARS_COUNT_TOKEN};
+UNIV_INTERN pars_res_word_t	pars_sum_token = {PARS_SUM_TOKEN};
+UNIV_INTERN pars_res_word_t	pars_distinct_token = {PARS_DISTINCT_TOKEN};
+UNIV_INTERN pars_res_word_t	pars_binary_token = {PARS_BINARY_TOKEN};
+UNIV_INTERN pars_res_word_t	pars_blob_token = {PARS_BLOB_TOKEN};
+UNIV_INTERN pars_res_word_t	pars_int_token = {PARS_INT_TOKEN};
+UNIV_INTERN pars_res_word_t	pars_char_token = {PARS_CHAR_TOKEN};
+UNIV_INTERN pars_res_word_t	pars_float_token = {PARS_FLOAT_TOKEN};
+UNIV_INTERN pars_res_word_t	pars_update_token = {PARS_UPDATE_TOKEN};
+UNIV_INTERN pars_res_word_t	pars_asc_token = {PARS_ASC_TOKEN};
+UNIV_INTERN pars_res_word_t	pars_desc_token = {PARS_DESC_TOKEN};
+UNIV_INTERN pars_res_word_t	pars_open_token = {PARS_OPEN_TOKEN};
+UNIV_INTERN pars_res_word_t	pars_close_token = {PARS_CLOSE_TOKEN};
+UNIV_INTERN pars_res_word_t	pars_share_token = {PARS_SHARE_TOKEN};
+UNIV_INTERN pars_res_word_t	pars_unique_token = {PARS_UNIQUE_TOKEN};
+UNIV_INTERN pars_res_word_t	pars_clustered_token = {PARS_CLUSTERED_TOKEN};
+
+/* Global variable used to denote the '*' in SELECT * FROM.. */
+#define PARS_STAR_DENOTER	12345678
+UNIV_INTERN ulint	pars_star_denoter	= PARS_STAR_DENOTER;
+
+
+/*************************************************************************
+Determines the class of a function code. */
+static
+ulint
+pars_func_get_class(
+/*================*/
+			/* out: function class: PARS_FUNC_ARITH, ... */
+	int	func)	/* in: function code: '=', PARS_GE_TOKEN, ... */
+{
+	switch (func) {
+	case '+': case '-': case '*': case '/':
+		return(PARS_FUNC_ARITH);
+
+	case '=': case '<': case '>':
+	case PARS_GE_TOKEN: case PARS_LE_TOKEN: case PARS_NE_TOKEN:
+		return(PARS_FUNC_CMP);
+
+	case PARS_AND_TOKEN: case PARS_OR_TOKEN: case PARS_NOT_TOKEN:
+		return(PARS_FUNC_LOGICAL);
+
+	case PARS_COUNT_TOKEN: case PARS_SUM_TOKEN:
+		return(PARS_FUNC_AGGREGATE);
+
+	case PARS_TO_CHAR_TOKEN:
+	case PARS_TO_NUMBER_TOKEN:
+	case PARS_TO_BINARY_TOKEN:
+	case PARS_BINARY_TO_NUMBER_TOKEN:
+	case PARS_SUBSTR_TOKEN:
+	case PARS_CONCAT_TOKEN:
+	case PARS_LENGTH_TOKEN:
+	case PARS_INSTR_TOKEN:
+	case PARS_SYSDATE_TOKEN:
+	case PARS_NOTFOUND_TOKEN:
+	case PARS_PRINTF_TOKEN:
+	case PARS_ASSERT_TOKEN:
+	case PARS_RND_TOKEN:
+	case PARS_RND_STR_TOKEN:
+	case PARS_REPLSTR_TOKEN:
+		return(PARS_FUNC_PREDEFINED);
+
+	default:
+		return(PARS_FUNC_OTHER);
+	}
+}
+
+/*************************************************************************
+Parses an operator or predefined function expression. */
+static
+func_node_t*
+pars_func_low(
+/*==========*/
+				/* out, own: function node in a query tree */
+	int		func,	/* in: function token code */
+	que_node_t*	arg)	/* in: first argument in the argument list */
+{
+	func_node_t*	node;
+
+	node = mem_heap_alloc(pars_sym_tab_global->heap, sizeof(func_node_t));
+
+	node->common.type = QUE_NODE_FUNC;
+	dfield_set_data(&(node->common.val), NULL, 0);
+	node->common.val_buf_size = 0;
+
+	node->func = func;
+
+	node->class = pars_func_get_class(func);
+
+	node->args = arg;
+
+	UT_LIST_ADD_LAST(func_node_list, pars_sym_tab_global->func_node_list,
+			 node);
+	return(node);
+}
+
+/*************************************************************************
+Parses a function expression. */
+UNIV_INTERN
+func_node_t*
+pars_func(
+/*======*/
+				/* out, own: function node in a query tree */
+	que_node_t*	res_word,/* in: function name reserved word */
+	que_node_t*	arg)	/* in: first argument in the argument list */
+{
+	return(pars_func_low(((pars_res_word_t*)res_word)->code, arg));
+}
+
+/*************************************************************************
+Parses an operator expression. */
+UNIV_INTERN
+func_node_t*
+pars_op(
+/*====*/
+				/* out, own: function node in a query tree */
+	int		func,	/* in: operator token code */
+	que_node_t*	arg1,	/* in: first argument */
+	que_node_t*	arg2)	/* in: second argument or NULL for an unary
+				operator */
+{
+	que_node_list_add_last(NULL, arg1);
+
+	if (arg2) {
+		que_node_list_add_last(arg1, arg2);
+	}
+
+	return(pars_func_low(func, arg1));
+}
+
+/*************************************************************************
+Parses an ORDER BY clause. Order by a single column only is supported. */
+UNIV_INTERN
+order_node_t*
+pars_order_by(
+/*==========*/
+				/* out, own: order-by node in a query tree */
+	sym_node_t*	column,	/* in: column name */
+	pars_res_word_t* asc)	/* in: &pars_asc_token or pars_desc_token */
+{
+	order_node_t*	node;
+
+	node = mem_heap_alloc(pars_sym_tab_global->heap, sizeof(order_node_t));
+
+	node->common.type = QUE_NODE_ORDER;
+
+	node->column = column;
+
+	if (asc == &pars_asc_token) {
+		node->asc = TRUE;
+	} else {
+		ut_a(asc == &pars_desc_token);
+		node->asc = FALSE;
+	}
+
+	return(node);
+}
+
+/*************************************************************************
+Determine if a data type is a built-in string data type of the InnoDB
+SQL parser. */
+static
+ibool
+pars_is_string_type(
+/*================*/
+			/* out: TRUE if string data type */
+	ulint	mtype)	/* in: main data type */
+{
+	switch (mtype) {
+	case DATA_VARCHAR: case DATA_CHAR:
+	case DATA_FIXBINARY: case DATA_BINARY:
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
+
+/*************************************************************************
+Resolves the data type of a function in an expression. The argument data
+types must already be resolved. */
+static
+void
+pars_resolve_func_data_type(
+/*========================*/
+	func_node_t*	node)	/* in: function node */
+{
+	que_node_t*	arg;
+
+	ut_a(que_node_get_type(node) == QUE_NODE_FUNC);
+
+	arg = node->args;
+
+	switch (node->func) {
+	case PARS_SUM_TOKEN:
+	case '+': case '-': case '*': case '/':
+		/* Inherit the data type from the first argument (which must
+		not be the SQL null literal whose type is DATA_ERROR) */
+
+		dtype_copy(que_node_get_data_type(node),
+			   que_node_get_data_type(arg));
+
+		ut_a(dtype_get_mtype(que_node_get_data_type(node))
+		     == DATA_INT);
+		break;
+
+	case PARS_COUNT_TOKEN:
+		ut_a(arg);
+		dtype_set(que_node_get_data_type(node), DATA_INT, 0, 4);
+		break;
+
+	case PARS_TO_CHAR_TOKEN:
+	case PARS_RND_STR_TOKEN:
+		ut_a(dtype_get_mtype(que_node_get_data_type(arg)) == DATA_INT);
+		dtype_set(que_node_get_data_type(node), DATA_VARCHAR,
+			  DATA_ENGLISH, 0);
+		break;
+
+	case PARS_TO_BINARY_TOKEN:
+		if (dtype_get_mtype(que_node_get_data_type(arg)) == DATA_INT) {
+			dtype_set(que_node_get_data_type(node), DATA_VARCHAR,
+				  DATA_ENGLISH, 0);
+		} else {
+			dtype_set(que_node_get_data_type(node), DATA_BINARY,
+				  0, 0);
+		}
+		break;
+
+	case PARS_TO_NUMBER_TOKEN:
+	case PARS_BINARY_TO_NUMBER_TOKEN:
+	case PARS_LENGTH_TOKEN:
+	case PARS_INSTR_TOKEN:
+		ut_a(pars_is_string_type(que_node_get_data_type(arg)->mtype));
+		dtype_set(que_node_get_data_type(node), DATA_INT, 0, 4);
+		break;
+
+	case PARS_SYSDATE_TOKEN:
+		ut_a(arg == NULL);
+		dtype_set(que_node_get_data_type(node), DATA_INT, 0, 4);
+		break;
+
+	case PARS_SUBSTR_TOKEN:
+	case PARS_CONCAT_TOKEN:
+		ut_a(pars_is_string_type(que_node_get_data_type(arg)->mtype));
+		dtype_set(que_node_get_data_type(node), DATA_VARCHAR,
+			  DATA_ENGLISH, 0);
+		break;
+
+	case '>': case '<': case '=':
+	case PARS_GE_TOKEN:
+	case PARS_LE_TOKEN:
+	case PARS_NE_TOKEN:
+	case PARS_AND_TOKEN:
+	case PARS_OR_TOKEN:
+	case PARS_NOT_TOKEN:
+	case PARS_NOTFOUND_TOKEN:
+
+		/* We currently have no iboolean type: use integer type */
+		dtype_set(que_node_get_data_type(node), DATA_INT, 0, 4);
+		break;
+
+	case PARS_RND_TOKEN:
+		ut_a(dtype_get_mtype(que_node_get_data_type(arg)) == DATA_INT);
+		dtype_set(que_node_get_data_type(node), DATA_INT, 0, 4);
+		break;
+
+	default:
+		ut_error;
+	}
+}
+
+/*************************************************************************
+Resolves the meaning of variables in an expression and the data types of
+functions. It is an error if some identifier cannot be resolved here. */
+static
+void
+pars_resolve_exp_variables_and_types(
+/*=================================*/
+	sel_node_t*	select_node,	/* in: select node or NULL; if
+					this is not NULL then the variable
+					sym nodes are added to the
+					copy_variables list of select_node */
+	que_node_t*	exp_node)	/* in: expression */
+{
+	func_node_t*	func_node;
+	que_node_t*	arg;
+	sym_node_t*	sym_node;
+	sym_node_t*	node;
+
+	ut_a(exp_node);
+
+	if (que_node_get_type(exp_node) == QUE_NODE_FUNC) {
+		func_node = exp_node;
+
+		arg = func_node->args;
+
+		while (arg) {
+			pars_resolve_exp_variables_and_types(select_node, arg);
+
+			arg = que_node_get_next(arg);
+		}
+
+		pars_resolve_func_data_type(func_node);
+
+		return;
+	}
+
+	ut_a(que_node_get_type(exp_node) == QUE_NODE_SYMBOL);
+
+	sym_node = exp_node;
+
+	if (sym_node->resolved) {
+
+		return;
+	}
+
+	/* Not resolved yet: look in the symbol table for a variable
+	or a cursor or a function with the same name */
+
+	node = UT_LIST_GET_FIRST(pars_sym_tab_global->sym_list);
+
+	while (node) {
+		if (node->resolved
+		    && ((node->token_type == SYM_VAR)
+			|| (node->token_type == SYM_CURSOR)
+			|| (node->token_type == SYM_FUNCTION))
+		    && node->name
+		    && (sym_node->name_len == node->name_len)
+		    && (ut_memcmp(sym_node->name, node->name,
+				  node->name_len) == 0)) {
+
+			/* Found a variable or a cursor declared with
+			the same name */
+
+			break;
+		}
+
+		node = UT_LIST_GET_NEXT(sym_list, node);
+	}
+
+	if (!node) {
+		fprintf(stderr, "PARSER ERROR: Unresolved identifier %s\n",
+			sym_node->name);
+	}
+
+	ut_a(node);
+
+	sym_node->resolved = TRUE;
+	sym_node->token_type = SYM_IMPLICIT_VAR;
+	sym_node->alias = node;
+	sym_node->indirection = node;
+
+	if (select_node) {
+		UT_LIST_ADD_LAST(col_var_list, select_node->copy_variables,
+				 sym_node);
+	}
+
+	dfield_set_type(que_node_get_val(sym_node),
+			que_node_get_data_type(node));
+}
+
+/*************************************************************************
+Resolves the meaning of variables in an expression list. It is an error if
+some identifier cannot be resolved here. Resolves also the data types of
+functions. */
+static
+void
+pars_resolve_exp_list_variables_and_types(
+/*======================================*/
+	sel_node_t*	select_node,	/* in: select node or NULL */
+	que_node_t*	exp_node)	/* in: expression list first node, or
+					NULL */
+{
+	while (exp_node) {
+		pars_resolve_exp_variables_and_types(select_node, exp_node);
+
+		exp_node = que_node_get_next(exp_node);
+	}
+}
+
+/*************************************************************************
+Resolves the columns in an expression. */
+static
+void
+pars_resolve_exp_columns(
+/*=====================*/
+	sym_node_t*	table_node,	/* in: first node in a table list */
+	que_node_t*	exp_node)	/* in: expression */
+{
+	func_node_t*	func_node;
+	que_node_t*	arg;
+	sym_node_t*	sym_node;
+	dict_table_t*	table;
+	sym_node_t*	t_node;
+	ulint		n_cols;
+	ulint		i;
+
+	ut_a(exp_node);
+
+	if (que_node_get_type(exp_node) == QUE_NODE_FUNC) {
+		func_node = exp_node;
+
+		arg = func_node->args;
+
+		while (arg) {
+			pars_resolve_exp_columns(table_node, arg);
+
+			arg = que_node_get_next(arg);
+		}
+
+		return;
+	}
+
+	ut_a(que_node_get_type(exp_node) == QUE_NODE_SYMBOL);
+
+	sym_node = exp_node;
+
+	if (sym_node->resolved) {
+
+		return;
+	}
+
+	/* Not resolved yet: look in the table list for a column with the
+	same name */
+
+	t_node = table_node;
+
+	while (t_node) {
+		table = t_node->table;
+
+		n_cols = dict_table_get_n_cols(table);
+
+		for (i = 0; i < n_cols; i++) {
+			const dict_col_t*	col
+				= dict_table_get_nth_col(table, i);
+			const char*		col_name
+				= dict_table_get_col_name(table, i);
+
+			if ((sym_node->name_len == ut_strlen(col_name))
+			    && (0 == ut_memcmp(sym_node->name, col_name,
+					       sym_node->name_len))) {
+				/* Found */
+				sym_node->resolved = TRUE;
+				sym_node->token_type = SYM_COLUMN;
+				sym_node->table = table;
+				sym_node->col_no = i;
+				sym_node->prefetch_buf = NULL;
+
+				dict_col_copy_type(
+					col,
+					dfield_get_type(&sym_node
+							->common.val));
+
+				return;
+			}
+		}
+
+		t_node = que_node_get_next(t_node);
+	}
+}
+
+/*************************************************************************
+Resolves the meaning of columns in an expression list. */
+static
+void
+pars_resolve_exp_list_columns(
+/*==========================*/
+	sym_node_t*	table_node,	/* in: first node in a table list */
+	que_node_t*	exp_node)	/* in: expression list first node, or
+					NULL */
+{
+	while (exp_node) {
+		pars_resolve_exp_columns(table_node, exp_node);
+
+		exp_node = que_node_get_next(exp_node);
+	}
+}
+
+/*************************************************************************
+Retrieves the table definition for a table name id. */
+static
+void
+pars_retrieve_table_def(
+/*====================*/
+	sym_node_t*	sym_node)	/* in: table node */
+{
+	const char*	table_name;
+
+	ut_a(sym_node);
+	ut_a(que_node_get_type(sym_node) == QUE_NODE_SYMBOL);
+
+	sym_node->resolved = TRUE;
+	sym_node->token_type = SYM_TABLE;
+
+	table_name = (const char*) sym_node->name;
+
+	sym_node->table = dict_table_get_low(table_name);
+
+	ut_a(sym_node->table);
+}
+
+/*************************************************************************
+Retrieves the table definitions for a list of table name ids. */
+static
+ulint
+pars_retrieve_table_list_defs(
+/*==========================*/
+					/* out: number of tables */
+	sym_node_t*	sym_node)	/* in: first table node in list */
+{
+	ulint		count		= 0;
+
+	if (sym_node == NULL) {
+
+		return(count);
+	}
+
+	while (sym_node) {
+		pars_retrieve_table_def(sym_node);
+
+		count++;
+
+		sym_node = que_node_get_next(sym_node);
+	}
+
+	return(count);
+}
+
+/*************************************************************************
+Adds all columns to the select list if the query is SELECT * FROM ... */
+static
+void
+pars_select_all_columns(
+/*====================*/
+	sel_node_t*	select_node)	/* in: select node already containing
+					the table list */
+{
+	sym_node_t*	col_node;
+	sym_node_t*	table_node;
+	dict_table_t*	table;
+	ulint		i;
+
+	select_node->select_list = NULL;
+
+	table_node = select_node->table_list;
+
+	while (table_node) {
+		table = table_node->table;
+
+		for (i = 0; i < dict_table_get_n_user_cols(table); i++) {
+			const char*	col_name = dict_table_get_col_name(
+				table, i);
+
+			col_node = sym_tab_add_id(pars_sym_tab_global,
+						  (byte*)col_name,
+						  ut_strlen(col_name));
+
+			select_node->select_list = que_node_list_add_last(
+				select_node->select_list, col_node);
+		}
+
+		table_node = que_node_get_next(table_node);
+	}
+}
+
+/*************************************************************************
+Parses a select list; creates a query graph node for the whole SELECT
+statement. */
+UNIV_INTERN
+sel_node_t*
+pars_select_list(
+/*=============*/
+					/* out, own: select node in a query
+					tree */
+	que_node_t*	select_list,	/* in: select list */
+	sym_node_t*	into_list)	/* in: variables list or NULL */
+{
+	sel_node_t*	node;
+
+	node = sel_node_create(pars_sym_tab_global->heap);
+
+	node->select_list = select_list;
+	node->into_list = into_list;
+
+	pars_resolve_exp_list_variables_and_types(NULL, into_list);
+
+	return(node);
+}
+
+/*************************************************************************
+Checks if the query is an aggregate query, in which case the selct list must
+contain only aggregate function items. */
+static
+void
+pars_check_aggregate(
+/*=================*/
+	sel_node_t*	select_node)	/* in: select node already containing
+					the select list */
+{
+	que_node_t*	exp_node;
+	func_node_t*	func_node;
+	ulint		n_nodes			= 0;
+	ulint		n_aggregate_nodes	= 0;
+
+	exp_node = select_node->select_list;
+
+	while (exp_node) {
+
+		n_nodes++;
+
+		if (que_node_get_type(exp_node) == QUE_NODE_FUNC) {
+
+			func_node = exp_node;
+
+			if (func_node->class == PARS_FUNC_AGGREGATE) {
+
+				n_aggregate_nodes++;
+			}
+		}
+
+		exp_node = que_node_get_next(exp_node);
+	}
+
+	if (n_aggregate_nodes > 0) {
+		ut_a(n_nodes == n_aggregate_nodes);
+
+		select_node->is_aggregate = TRUE;
+	} else {
+		select_node->is_aggregate = FALSE;
+	}
+}
+
+/*************************************************************************
+Parses a select statement. */
+UNIV_INTERN
+sel_node_t*
+pars_select_statement(
+/*==================*/
+					/* out, own: select node in a query
+					tree */
+	sel_node_t*	select_node,	/* in: select node already containing
+					the select list */
+	sym_node_t*	table_list,	/* in: table list */
+	que_node_t*	search_cond,	/* in: search condition or NULL */
+	pars_res_word_t* for_update,	/* in: NULL or &pars_update_token */
+	pars_res_word_t* lock_shared,	/* in: NULL or &pars_share_token */
+	order_node_t*	order_by)	/* in: NULL or an order-by node */
+{
+	select_node->state = SEL_NODE_OPEN;
+
+	select_node->table_list = table_list;
+	select_node->n_tables = pars_retrieve_table_list_defs(table_list);
+
+	if (select_node->select_list == &pars_star_denoter) {
+
+		/* SELECT * FROM ... */
+		pars_select_all_columns(select_node);
+	}
+
+	if (select_node->into_list) {
+		ut_a(que_node_list_get_len(select_node->into_list)
+		     == que_node_list_get_len(select_node->select_list));
+	}
+
+	UT_LIST_INIT(select_node->copy_variables);
+
+	pars_resolve_exp_list_columns(table_list, select_node->select_list);
+	pars_resolve_exp_list_variables_and_types(select_node,
+						  select_node->select_list);
+	pars_check_aggregate(select_node);
+
+	select_node->search_cond = search_cond;
+
+	if (search_cond) {
+		pars_resolve_exp_columns(table_list, search_cond);
+		pars_resolve_exp_variables_and_types(select_node, search_cond);
+	}
+
+	if (for_update) {
+		ut_a(!lock_shared);
+
+		select_node->set_x_locks = TRUE;
+		select_node->row_lock_mode = LOCK_X;
+
+		select_node->consistent_read = FALSE;
+		select_node->read_view = NULL;
+	} else if (lock_shared){
+		select_node->set_x_locks = FALSE;
+		select_node->row_lock_mode = LOCK_S;
+
+		select_node->consistent_read = FALSE;
+		select_node->read_view = NULL;
+	} else {
+		select_node->set_x_locks = FALSE;
+		select_node->row_lock_mode = LOCK_S;
+
+		select_node->consistent_read = TRUE;
+	}
+
+	select_node->order_by = order_by;
+
+	if (order_by) {
+		pars_resolve_exp_columns(table_list, order_by->column);
+	}
+
+	/* The final value of the following fields depend on the environment
+	where the select statement appears: */
+
+	select_node->can_get_updated = FALSE;
+	select_node->explicit_cursor = NULL;
+
+	opt_search_plan(select_node);
+
+	return(select_node);
+}
+
+/*************************************************************************
+Parses a cursor declaration. */
+UNIV_INTERN
+que_node_t*
+pars_cursor_declaration(
+/*====================*/
+					/* out: sym_node */
+	sym_node_t*	sym_node,	/* in: cursor id node in the symbol
+					table */
+	sel_node_t*	select_node)	/* in: select node */
+{
+	sym_node->resolved = TRUE;
+	sym_node->token_type = SYM_CURSOR;
+	sym_node->cursor_def = select_node;
+
+	select_node->state = SEL_NODE_CLOSED;
+	select_node->explicit_cursor = sym_node;
+
+	return(sym_node);
+}
+
+/*************************************************************************
+Parses a function declaration. */
+UNIV_INTERN
+que_node_t*
+pars_function_declaration(
+/*======================*/
+					/* out: sym_node */
+	sym_node_t*	sym_node)	/* in: function id node in the symbol
+					table */
+{
+	sym_node->resolved = TRUE;
+	sym_node->token_type = SYM_FUNCTION;
+
+	/* Check that the function exists. */
+	ut_a(pars_info_get_user_func(pars_sym_tab_global->info,
+				     sym_node->name));
+
+	return(sym_node);
+}
+
+/*************************************************************************
+Parses a delete or update statement start. */
+UNIV_INTERN
+upd_node_t*
+pars_update_statement_start(
+/*========================*/
+					/* out, own: update node in a query
+					tree */
+	ibool		is_delete,	/* in: TRUE if delete */
+	sym_node_t*	table_sym,	/* in: table name node */
+	col_assign_node_t* col_assign_list)/* in: column assignment list, NULL
+					if delete */
+{
+	upd_node_t*	node;
+
+	node = upd_node_create(pars_sym_tab_global->heap);
+
+	node->is_delete = is_delete;
+
+	node->table_sym = table_sym;
+	node->col_assign_list = col_assign_list;
+
+	return(node);
+}
+
+/*************************************************************************
+Parses a column assignment in an update. */
+UNIV_INTERN
+col_assign_node_t*
+pars_column_assignment(
+/*===================*/
+				/* out: column assignment node */
+	sym_node_t*	column,	/* in: column to assign */
+	que_node_t*	exp)	/* in: value to assign */
+{
+	col_assign_node_t*	node;
+
+	node = mem_heap_alloc(pars_sym_tab_global->heap,
+			      sizeof(col_assign_node_t));
+	node->common.type = QUE_NODE_COL_ASSIGNMENT;
+
+	node->col = column;
+	node->val = exp;
+
+	return(node);
+}
+
+/*************************************************************************
+Processes an update node assignment list. */
+static
+void
+pars_process_assign_list(
+/*=====================*/
+	upd_node_t*	node)	/* in: update node */
+{
+	col_assign_node_t*	col_assign_list;
+	sym_node_t*		table_sym;
+	col_assign_node_t*	assign_node;
+	upd_field_t*		upd_field;
+	dict_index_t*		clust_index;
+	sym_node_t*		col_sym;
+	ulint			changes_ord_field;
+	ulint			changes_field_size;
+	ulint			n_assigns;
+	ulint			i;
+
+	table_sym = node->table_sym;
+	col_assign_list = node->col_assign_list;
+	clust_index = dict_table_get_first_index(node->table);
+
+	assign_node = col_assign_list;
+	n_assigns = 0;
+
+	while (assign_node) {
+		pars_resolve_exp_columns(table_sym, assign_node->col);
+		pars_resolve_exp_columns(table_sym, assign_node->val);
+		pars_resolve_exp_variables_and_types(NULL, assign_node->val);
+#if 0
+		ut_a(dtype_get_mtype(
+			     dfield_get_type(que_node_get_val(
+						     assign_node->col)))
+		     == dtype_get_mtype(
+			     dfield_get_type(que_node_get_val(
+						     assign_node->val))));
+#endif
+
+		/* Add to the update node all the columns found in assignment
+		values as columns to copy: therefore, TRUE */
+
+		opt_find_all_cols(TRUE, clust_index, &(node->columns), NULL,
+				  assign_node->val);
+		n_assigns++;
+
+		assign_node = que_node_get_next(assign_node);
+	}
+
+	node->update = upd_create(n_assigns, pars_sym_tab_global->heap);
+
+	assign_node = col_assign_list;
+
+	changes_field_size = UPD_NODE_NO_SIZE_CHANGE;
+
+	for (i = 0; i < n_assigns; i++) {
+		upd_field = upd_get_nth_field(node->update, i);
+
+		col_sym = assign_node->col;
+
+		upd_field_set_field_no(upd_field, dict_index_get_nth_col_pos(
+					       clust_index, col_sym->col_no),
+				       clust_index, NULL);
+		upd_field->exp = assign_node->val;
+
+		if (!dict_col_get_fixed_size(
+			    dict_index_get_nth_col(clust_index,
+						   upd_field->field_no))) {
+			changes_field_size = 0;
+		}
+
+		assign_node = que_node_get_next(assign_node);
+	}
+
+	/* Find out if the update can modify an ordering field in any index */
+
+	changes_ord_field = UPD_NODE_NO_ORD_CHANGE;
+
+	if (row_upd_changes_some_index_ord_field_binary(node->table,
+							node->update)) {
+		changes_ord_field = 0;
+	}
+
+	node->cmpl_info = changes_ord_field | changes_field_size;
+}
+
+/*************************************************************************
+Parses an update or delete statement. */
+UNIV_INTERN
+upd_node_t*
+pars_update_statement(
+/*==================*/
+					/* out, own: update node in a query
+					tree */
+	upd_node_t*	node,		/* in: update node */
+	sym_node_t*	cursor_sym,	/* in: pointer to a cursor entry in
+					the symbol table or NULL */
+	que_node_t*	search_cond)	/* in: search condition or NULL */
+{
+	sym_node_t*	table_sym;
+	sel_node_t*	sel_node;
+	plan_t*		plan;
+
+	table_sym = node->table_sym;
+
+	pars_retrieve_table_def(table_sym);
+	node->table = table_sym->table;
+
+	UT_LIST_INIT(node->columns);
+
+	/* Make the single table node into a list of table nodes of length 1 */
+
+	que_node_list_add_last(NULL, table_sym);
+
+	if (cursor_sym) {
+		pars_resolve_exp_variables_and_types(NULL, cursor_sym);
+
+		sel_node = cursor_sym->alias->cursor_def;
+
+		node->searched_update = FALSE;
+	} else {
+		sel_node = pars_select_list(NULL, NULL);
+
+		pars_select_statement(sel_node, table_sym, search_cond, NULL,
+				      &pars_share_token, NULL);
+		node->searched_update = TRUE;
+		sel_node->common.parent = node;
+	}
+
+	node->select = sel_node;
+
+	ut_a(!node->is_delete || (node->col_assign_list == NULL));
+	ut_a(node->is_delete || (node->col_assign_list != NULL));
+
+	if (node->is_delete) {
+		node->cmpl_info = 0;
+	} else {
+		pars_process_assign_list(node);
+	}
+
+	if (node->searched_update) {
+		node->has_clust_rec_x_lock = TRUE;
+		sel_node->set_x_locks = TRUE;
+		sel_node->row_lock_mode = LOCK_X;
+	} else {
+		node->has_clust_rec_x_lock = sel_node->set_x_locks;
+	}
+
+	ut_a(sel_node->n_tables == 1);
+	ut_a(sel_node->consistent_read == FALSE);
+	ut_a(sel_node->order_by == NULL);
+	ut_a(sel_node->is_aggregate == FALSE);
+
+	sel_node->can_get_updated = TRUE;
+
+	node->state = UPD_NODE_UPDATE_CLUSTERED;
+
+	plan = sel_node_get_nth_plan(sel_node, 0);
+
+	plan->no_prefetch = TRUE;
+
+	if (!dict_index_is_clust(plan->index)) {
+
+		plan->must_get_clust = TRUE;
+
+		node->pcur = &(plan->clust_pcur);
+	} else {
+		node->pcur = &(plan->pcur);
+	}
+
+	return(node);
+}
+
+/*************************************************************************
+Parses an insert statement. */
+UNIV_INTERN
+ins_node_t*
+pars_insert_statement(
+/*==================*/
+					/* out, own: update node in a query
+					tree */
+	sym_node_t*	table_sym,	/* in: table name node */
+	que_node_t*	values_list,	/* in: value expression list or NULL */
+	sel_node_t*	select)		/* in: select condition or NULL */
+{
+	ins_node_t*	node;
+	dtuple_t*	row;
+	ulint		ins_type;
+
+	ut_a(values_list || select);
+	ut_a(!values_list || !select);
+
+	if (values_list) {
+		ins_type = INS_VALUES;
+	} else {
+		ins_type = INS_SEARCHED;
+	}
+
+	pars_retrieve_table_def(table_sym);
+
+	node = ins_node_create(ins_type, table_sym->table,
+			       pars_sym_tab_global->heap);
+
+	row = dtuple_create(pars_sym_tab_global->heap,
+			    dict_table_get_n_cols(node->table));
+
+	dict_table_copy_types(row, table_sym->table);
+
+	ins_node_set_new_row(node, row);
+
+	node->select = select;
+
+	if (select) {
+		select->common.parent = node;
+
+		ut_a(que_node_list_get_len(select->select_list)
+		     == dict_table_get_n_user_cols(table_sym->table));
+	}
+
+	node->values_list = values_list;
+
+	if (node->values_list) {
+		pars_resolve_exp_list_variables_and_types(NULL, values_list);
+
+		ut_a(que_node_list_get_len(values_list)
+		     == dict_table_get_n_user_cols(table_sym->table));
+	}
+
+	return(node);
+}
+
+/*************************************************************************
+Set the type of a dfield. */
+static
+void
+pars_set_dfield_type(
+/*=================*/
+	dfield_t*		dfield,		/* in: dfield */
+	pars_res_word_t*	type,		/* in: pointer to a type
+						token */
+	ulint			len,		/* in: length, or 0 */
+	ibool			is_unsigned,	/* in: if TRUE, column is
+						UNSIGNED. */
+	ibool			is_not_null)	/* in: if TRUE, column is
+						NOT NULL. */
+{
+	ulint flags = 0;
+
+	if (is_not_null) {
+		flags |= DATA_NOT_NULL;
+	}
+
+	if (is_unsigned) {
+		flags |= DATA_UNSIGNED;
+	}
+
+	if (type == &pars_int_token) {
+		ut_a(len == 0);
+
+		dtype_set(dfield_get_type(dfield), DATA_INT, flags, 4);
+
+	} else if (type == &pars_char_token) {
+		ut_a(len == 0);
+
+		dtype_set(dfield_get_type(dfield), DATA_VARCHAR,
+			  DATA_ENGLISH | flags, 0);
+	} else if (type == &pars_binary_token) {
+		ut_a(len != 0);
+
+		dtype_set(dfield_get_type(dfield), DATA_FIXBINARY,
+			  DATA_BINARY_TYPE | flags, len);
+	} else if (type == &pars_blob_token) {
+		ut_a(len == 0);
+
+		dtype_set(dfield_get_type(dfield), DATA_BLOB,
+			  DATA_BINARY_TYPE | flags, 0);
+	} else {
+		ut_error;
+	}
+}
+
+/*************************************************************************
+Parses a variable declaration. */
+UNIV_INTERN
+sym_node_t*
+pars_variable_declaration(
+/*======================*/
+				/* out, own: symbol table node of type
+				SYM_VAR */
+	sym_node_t*	node,	/* in: symbol table node allocated for the
+				id of the variable */
+	pars_res_word_t* type)	/* in: pointer to a type token */
+{
+	node->resolved = TRUE;
+	node->token_type = SYM_VAR;
+
+	node->param_type = PARS_NOT_PARAM;
+
+	pars_set_dfield_type(que_node_get_val(node), type, 0, FALSE, FALSE);
+
+	return(node);
+}
+
+/*************************************************************************
+Parses a procedure parameter declaration. */
+UNIV_INTERN
+sym_node_t*
+pars_parameter_declaration(
+/*=======================*/
+				/* out, own: symbol table node of type
+				SYM_VAR */
+	sym_node_t*	node,	/* in: symbol table node allocated for the
+				id of the parameter */
+	ulint		param_type,
+				/* in: PARS_INPUT or PARS_OUTPUT */
+	pars_res_word_t* type)	/* in: pointer to a type token */
+{
+	ut_a((param_type == PARS_INPUT) || (param_type == PARS_OUTPUT));
+
+	pars_variable_declaration(node, type);
+
+	node->param_type = param_type;
+
+	return(node);
+}
+
+/*************************************************************************
+Sets the parent field in a query node list. */
+static
+void
+pars_set_parent_in_list(
+/*====================*/
+	que_node_t*	node_list,	/* in: first node in a list */
+	que_node_t*	parent)		/* in: parent value to set in all
+					nodes of the list */
+{
+	que_common_t*	common;
+
+	common = node_list;
+
+	while (common) {
+		common->parent = parent;
+
+		common = que_node_get_next(common);
+	}
+}
+
+/*************************************************************************
+Parses an elsif element. */
+UNIV_INTERN
+elsif_node_t*
+pars_elsif_element(
+/*===============*/
+					/* out: elsif node */
+	que_node_t*	cond,		/* in: if-condition */
+	que_node_t*	stat_list)	/* in: statement list */
+{
+	elsif_node_t*	node;
+
+	node = mem_heap_alloc(pars_sym_tab_global->heap, sizeof(elsif_node_t));
+
+	node->common.type = QUE_NODE_ELSIF;
+
+	node->cond = cond;
+
+	pars_resolve_exp_variables_and_types(NULL, cond);
+
+	node->stat_list = stat_list;
+
+	return(node);
+}
+
+/*************************************************************************
+Parses an if-statement. */
+UNIV_INTERN
+if_node_t*
+pars_if_statement(
+/*==============*/
+					/* out: if-statement node */
+	que_node_t*	cond,		/* in: if-condition */
+	que_node_t*	stat_list,	/* in: statement list */
+	que_node_t*	else_part)	/* in: else-part statement list
+					or elsif element list */
+{
+	if_node_t*	node;
+	elsif_node_t*	elsif_node;
+
+	node = mem_heap_alloc(pars_sym_tab_global->heap, sizeof(if_node_t));
+
+	node->common.type = QUE_NODE_IF;
+
+	node->cond = cond;
+
+	pars_resolve_exp_variables_and_types(NULL, cond);
+
+	node->stat_list = stat_list;
+
+	if (else_part && (que_node_get_type(else_part) == QUE_NODE_ELSIF)) {
+
+		/* There is a list of elsif conditions */
+
+		node->else_part = NULL;
+		node->elsif_list = else_part;
+
+		elsif_node = else_part;
+
+		while (elsif_node) {
+			pars_set_parent_in_list(elsif_node->stat_list, node);
+
+			elsif_node = que_node_get_next(elsif_node);
+		}
+	} else {
+		node->else_part = else_part;
+		node->elsif_list = NULL;
+
+		pars_set_parent_in_list(else_part, node);
+	}
+
+	pars_set_parent_in_list(stat_list, node);
+
+	return(node);
+}
+
+/*************************************************************************
+Parses a while-statement. */
+UNIV_INTERN
+while_node_t*
+pars_while_statement(
+/*=================*/
+					/* out: while-statement node */
+	que_node_t*	cond,		/* in: while-condition */
+	que_node_t*	stat_list)	/* in: statement list */
+{
+	while_node_t*	node;
+
+	node = mem_heap_alloc(pars_sym_tab_global->heap, sizeof(while_node_t));
+
+	node->common.type = QUE_NODE_WHILE;
+
+	node->cond = cond;
+
+	pars_resolve_exp_variables_and_types(NULL, cond);
+
+	node->stat_list = stat_list;
+
+	pars_set_parent_in_list(stat_list, node);
+
+	return(node);
+}
+
+/*************************************************************************
+Parses a for-loop-statement. */
+UNIV_INTERN
+for_node_t*
+pars_for_statement(
+/*===============*/
+					/* out: for-statement node */
+	sym_node_t*	loop_var,	/* in: loop variable */
+	que_node_t*	loop_start_limit,/* in: loop start expression */
+	que_node_t*	loop_end_limit,	/* in: loop end expression */
+	que_node_t*	stat_list)	/* in: statement list */
+{
+	for_node_t*	node;
+
+	node = mem_heap_alloc(pars_sym_tab_global->heap, sizeof(for_node_t));
+
+	node->common.type = QUE_NODE_FOR;
+
+	pars_resolve_exp_variables_and_types(NULL, loop_var);
+	pars_resolve_exp_variables_and_types(NULL, loop_start_limit);
+	pars_resolve_exp_variables_and_types(NULL, loop_end_limit);
+
+	node->loop_var = loop_var->indirection;
+
+	ut_a(loop_var->indirection);
+
+	node->loop_start_limit = loop_start_limit;
+	node->loop_end_limit = loop_end_limit;
+
+	node->stat_list = stat_list;
+
+	pars_set_parent_in_list(stat_list, node);
+
+	return(node);
+}
+
+/*************************************************************************
+Parses an exit statement. */
+UNIV_INTERN
+exit_node_t*
+pars_exit_statement(void)
+/*=====================*/
+					/* out: exit statement node */
+{
+	exit_node_t*	node;
+
+	node = mem_heap_alloc(pars_sym_tab_global->heap, sizeof(exit_node_t));
+	node->common.type = QUE_NODE_EXIT;
+
+	return(node);
+}
+
+/*************************************************************************
+Parses a return-statement. */
+UNIV_INTERN
+return_node_t*
+pars_return_statement(void)
+/*=======================*/
+					/* out: return-statement node */
+{
+	return_node_t*	node;
+
+	node = mem_heap_alloc(pars_sym_tab_global->heap,
+			      sizeof(return_node_t));
+	node->common.type = QUE_NODE_RETURN;
+
+	return(node);
+}
+
+/*************************************************************************
+Parses an assignment statement. */
+UNIV_INTERN
+assign_node_t*
+pars_assignment_statement(
+/*======================*/
+				/* out: assignment statement node */
+	sym_node_t*	var,	/* in: variable to assign */
+	que_node_t*	val)	/* in: value to assign */
+{
+	assign_node_t*	node;
+
+	node = mem_heap_alloc(pars_sym_tab_global->heap,
+			      sizeof(assign_node_t));
+	node->common.type = QUE_NODE_ASSIGNMENT;
+
+	node->var = var;
+	node->val = val;
+
+	pars_resolve_exp_variables_and_types(NULL, var);
+	pars_resolve_exp_variables_and_types(NULL, val);
+
+	ut_a(dtype_get_mtype(dfield_get_type(que_node_get_val(var)))
+	     == dtype_get_mtype(dfield_get_type(que_node_get_val(val))));
+
+	return(node);
+}
+
+/*************************************************************************
+Parses a procedure call. */
+UNIV_INTERN
+func_node_t*
+pars_procedure_call(
+/*================*/
+				/* out: function node */
+	que_node_t*	res_word,/* in: procedure name reserved word */
+	que_node_t*	args)	/* in: argument list */
+{
+	func_node_t*	node;
+
+	node = pars_func(res_word, args);
+
+	pars_resolve_exp_list_variables_and_types(NULL, args);
+
+	return(node);
+}
+
+/*************************************************************************
+Parses a fetch statement. into_list or user_func (but not both) must be
+non-NULL. */
+UNIV_INTERN
+fetch_node_t*
+pars_fetch_statement(
+/*=================*/
+					/* out: fetch statement node */
+	sym_node_t*	cursor,		/* in: cursor node */
+	sym_node_t*	into_list,	/* in: variables to set, or NULL */
+	sym_node_t*	user_func)	/* in: user function name, or NULL */
+{
+	sym_node_t*	cursor_decl;
+	fetch_node_t*	node;
+
+	/* Logical XOR. */
+	ut_a(!into_list != !user_func);
+
+	node = mem_heap_alloc(pars_sym_tab_global->heap, sizeof(fetch_node_t));
+
+	node->common.type = QUE_NODE_FETCH;
+
+	pars_resolve_exp_variables_and_types(NULL, cursor);
+
+	if (into_list) {
+		pars_resolve_exp_list_variables_and_types(NULL, into_list);
+		node->into_list = into_list;
+		node->func = NULL;
+	} else {
+		pars_resolve_exp_variables_and_types(NULL, user_func);
+
+		node->func = pars_info_get_user_func(pars_sym_tab_global->info,
+						     user_func->name);
+		ut_a(node->func);
+
+		node->into_list = NULL;
+	}
+
+	cursor_decl = cursor->alias;
+
+	ut_a(cursor_decl->token_type == SYM_CURSOR);
+
+	node->cursor_def = cursor_decl->cursor_def;
+
+	if (into_list) {
+		ut_a(que_node_list_get_len(into_list)
+		     == que_node_list_get_len(node->cursor_def->select_list));
+	}
+
+	return(node);
+}
+
+/*************************************************************************
+Parses an open or close cursor statement. */
+UNIV_INTERN
+open_node_t*
+pars_open_statement(
+/*================*/
+				/* out: fetch statement node */
+	ulint		type,	/* in: ROW_SEL_OPEN_CURSOR
+				or ROW_SEL_CLOSE_CURSOR */
+	sym_node_t*	cursor)	/* in: cursor node */
+{
+	sym_node_t*	cursor_decl;
+	open_node_t*	node;
+
+	node = mem_heap_alloc(pars_sym_tab_global->heap, sizeof(open_node_t));
+
+	node->common.type = QUE_NODE_OPEN;
+
+	pars_resolve_exp_variables_and_types(NULL, cursor);
+
+	cursor_decl = cursor->alias;
+
+	ut_a(cursor_decl->token_type == SYM_CURSOR);
+
+	node->op_type = type;
+	node->cursor_def = cursor_decl->cursor_def;
+
+	return(node);
+}
+
+/*************************************************************************
+Parses a row_printf-statement. */
+UNIV_INTERN
+row_printf_node_t*
+pars_row_printf_statement(
+/*======================*/
+					/* out: row_printf-statement node */
+	sel_node_t*	sel_node)	/* in: select node */
+{
+	row_printf_node_t*	node;
+
+	node = mem_heap_alloc(pars_sym_tab_global->heap,
+			      sizeof(row_printf_node_t));
+	node->common.type = QUE_NODE_ROW_PRINTF;
+
+	node->sel_node = sel_node;
+
+	sel_node->common.parent = node;
+
+	return(node);
+}
+
+/*************************************************************************
+Parses a commit statement. */
+UNIV_INTERN
+commit_node_t*
+pars_commit_statement(void)
+/*=======================*/
+{
+	return(commit_node_create(pars_sym_tab_global->heap));
+}
+
+/*************************************************************************
+Parses a rollback statement. */
+UNIV_INTERN
+roll_node_t*
+pars_rollback_statement(void)
+/*=========================*/
+{
+	return(roll_node_create(pars_sym_tab_global->heap));
+}
+
+/*************************************************************************
+Parses a column definition at a table creation. */
+UNIV_INTERN
+sym_node_t*
+pars_column_def(
+/*============*/
+						/* out: column sym table
+						node */
+	sym_node_t*		sym_node,	/* in: column node in the
+						symbol table */
+	pars_res_word_t*	type,		/* in: data type */
+	sym_node_t*		len,		/* in: length of column, or
+						NULL */
+	void*			is_unsigned,	/* in: if not NULL, column
+						is of type UNSIGNED. */
+	void*			is_not_null)	/* in: if not NULL, column
+						is of type NOT NULL. */
+{
+	ulint len2;
+
+	if (len) {
+		len2 = eval_node_get_int_val(len);
+	} else {
+		len2 = 0;
+	}
+
+	pars_set_dfield_type(que_node_get_val(sym_node), type, len2,
+			     is_unsigned != NULL, is_not_null != NULL);
+
+	return(sym_node);
+}
+
+/*************************************************************************
+Parses a table creation operation. */
+UNIV_INTERN
+tab_node_t*
+pars_create_table(
+/*==============*/
+					/* out: table create subgraph */
+	sym_node_t*	table_sym,	/* in: table name node in the symbol
+					table */
+	sym_node_t*	column_defs,	/* in: list of column names */
+	void*		not_fit_in_memory __attribute__((unused)))
+					/* in: a non-NULL pointer means that
+					this is a table which in simulations
+					should be simulated as not fitting
+					in memory; thread is put to sleep
+					to simulate disk accesses; NOTE that
+					this flag is not stored to the data
+					dictionary on disk, and the database
+					will forget about non-NULL value if
+					it has to reload the table definition
+					from disk */
+{
+	dict_table_t*	table;
+	sym_node_t*	column;
+	tab_node_t*	node;
+	const dtype_t*	dtype;
+	ulint		n_cols;
+
+	n_cols = que_node_list_get_len(column_defs);
+
+	/* As the InnoDB SQL parser is for internal use only,
+	for creating some system tables, this function will only
+	create tables in the old (not compact) record format. */
+	table = dict_mem_table_create(table_sym->name, 0, n_cols, 0);
+
+#ifdef UNIV_DEBUG
+	if (not_fit_in_memory != NULL) {
+		table->does_not_fit_in_memory = TRUE;
+	}
+#endif /* UNIV_DEBUG */
+	column = column_defs;
+
+	while (column) {
+		dtype = dfield_get_type(que_node_get_val(column));
+
+		dict_mem_table_add_col(table, table->heap,
+				       column->name, dtype->mtype,
+				       dtype->prtype, dtype->len);
+		column->resolved = TRUE;
+		column->token_type = SYM_COLUMN;
+
+		column = que_node_get_next(column);
+	}
+
+	node = tab_create_graph_create(table, pars_sym_tab_global->heap);
+
+	table_sym->resolved = TRUE;
+	table_sym->token_type = SYM_TABLE;
+
+	return(node);
+}
+
+/*************************************************************************
+Parses an index creation operation. */
+UNIV_INTERN
+ind_node_t*
+pars_create_index(
+/*==============*/
+					/* out: index create subgraph */
+	pars_res_word_t* unique_def,	/* in: not NULL if a unique index */
+	pars_res_word_t* clustered_def,	/* in: not NULL if a clustered index */
+	sym_node_t*	index_sym,	/* in: index name node in the symbol
+					table */
+	sym_node_t*	table_sym,	/* in: table name node in the symbol
+					table */
+	sym_node_t*	column_list)	/* in: list of column names */
+{
+	dict_index_t*	index;
+	sym_node_t*	column;
+	ind_node_t*	node;
+	ulint		n_fields;
+	ulint		ind_type;
+
+	n_fields = que_node_list_get_len(column_list);
+
+	ind_type = 0;
+
+	if (unique_def) {
+		ind_type = ind_type | DICT_UNIQUE;
+	}
+
+	if (clustered_def) {
+		ind_type = ind_type | DICT_CLUSTERED;
+	}
+
+	index = dict_mem_index_create(table_sym->name, index_sym->name, 0,
+				      ind_type, n_fields);
+	column = column_list;
+
+	while (column) {
+		dict_mem_index_add_field(index, column->name, 0);
+
+		column->resolved = TRUE;
+		column->token_type = SYM_COLUMN;
+
+		column = que_node_get_next(column);
+	}
+
+	node = ind_create_graph_create(index, pars_sym_tab_global->heap);
+
+	table_sym->resolved = TRUE;
+	table_sym->token_type = SYM_TABLE;
+
+	index_sym->resolved = TRUE;
+	index_sym->token_type = SYM_TABLE;
+
+	return(node);
+}
+
+/*************************************************************************
+Parses a procedure definition. */
+UNIV_INTERN
+que_fork_t*
+pars_procedure_definition(
+/*======================*/
+					/* out: query fork node */
+	sym_node_t*	sym_node,	/* in: procedure id node in the symbol
+					table */
+	sym_node_t*	param_list,	/* in: parameter declaration list */
+	que_node_t*	stat_list)	/* in: statement list */
+{
+	proc_node_t*	node;
+	que_fork_t*	fork;
+	que_thr_t*	thr;
+	mem_heap_t*	heap;
+
+	heap = pars_sym_tab_global->heap;
+
+	fork = que_fork_create(NULL, NULL, QUE_FORK_PROCEDURE, heap);
+	fork->trx = NULL;
+
+	thr = que_thr_create(fork, heap);
+
+	node = mem_heap_alloc(heap, sizeof(proc_node_t));
+
+	node->common.type = QUE_NODE_PROC;
+	node->common.parent = thr;
+
+	sym_node->token_type = SYM_PROCEDURE_NAME;
+	sym_node->resolved = TRUE;
+
+	node->proc_id = sym_node;
+	node->param_list = param_list;
+	node->stat_list = stat_list;
+
+	pars_set_parent_in_list(stat_list, node);
+
+	node->sym_tab = pars_sym_tab_global;
+
+	thr->child = node;
+
+	pars_sym_tab_global->query_graph = fork;
+
+	return(fork);
+}
+
+/*****************************************************************
+Parses a stored procedure call, when this is not within another stored
+procedure, that is, the client issues a procedure call directly.
+In MySQL/InnoDB, stored InnoDB procedures are invoked via the
+parsed procedure tree, not via InnoDB SQL, so this function is not used. */
+UNIV_INTERN
+que_fork_t*
+pars_stored_procedure_call(
+/*=======================*/
+					/* out: query graph */
+	sym_node_t*	sym_node __attribute__((unused)))
+					/* in: stored procedure name */
+{
+	ut_error;
+	return(NULL);
+}
+
+/*****************************************************************
+Retrieves characters to the lexical analyzer. */
+UNIV_INTERN
+void
+pars_get_lex_chars(
+/*===============*/
+	char*	buf,		/* in/out: buffer where to copy */
+	int*	result,		/* out: number of characters copied or EOF */
+	int	max_size)	/* in: maximum number of characters which fit
+				in the buffer */
+{
+	int	len;
+
+	len = pars_sym_tab_global->string_len
+		- pars_sym_tab_global->next_char_pos;
+	if (len == 0) {
+#ifdef YYDEBUG
+		/* fputs("SQL string ends\n", stderr); */
+#endif
+		*result = 0;
+
+		return;
+	}
+
+	if (len > max_size) {
+		len = max_size;
+	}
+
+#ifdef UNIV_SQL_DEBUG
+	if (pars_print_lexed) {
+
+		if (len >= 5) {
+			len = 5;
+		}
+
+		fwrite(pars_sym_tab_global->sql_string
+		       + pars_sym_tab_global->next_char_pos,
+		       1, len, stderr);
+	}
+#endif /* UNIV_SQL_DEBUG */
+
+	ut_memcpy(buf, pars_sym_tab_global->sql_string
+		  + pars_sym_tab_global->next_char_pos, len);
+	*result = len;
+
+	pars_sym_tab_global->next_char_pos += len;
+}
+
+/*****************************************************************
+Called by yyparse on error. */
+UNIV_INTERN
+void
+yyerror(
+/*====*/
+	const char*	s __attribute__((unused)))
+				/* in: error message string */
+{
+	ut_ad(s);
+
+	fputs("PARSER ERROR: Syntax error in SQL string\n", stderr);
+
+	ut_error;
+}
+
+/*****************************************************************
+Parses an SQL string returning the query graph. */
+UNIV_INTERN
+que_t*
+pars_sql(
+/*=====*/
+				/* out, own: the query graph */
+	pars_info_t*	info,	/* in: extra information, or NULL */
+	const char*	str)	/* in: SQL string */
+{
+	sym_node_t*	sym_node;
+	mem_heap_t*	heap;
+	que_t*		graph;
+
+	ut_ad(str);
+
+	heap = mem_heap_create(256);
+
+	/* Currently, the parser is not reentrant: */
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+
+	pars_sym_tab_global = sym_tab_create(heap);
+
+	pars_sym_tab_global->string_len = strlen(str);
+	pars_sym_tab_global->sql_string = mem_heap_dup(
+		heap, str, pars_sym_tab_global->string_len + 1);
+	pars_sym_tab_global->next_char_pos = 0;
+	pars_sym_tab_global->info = info;
+
+	yyparse();
+
+	sym_node = UT_LIST_GET_FIRST(pars_sym_tab_global->sym_list);
+
+	while (sym_node) {
+		ut_a(sym_node->resolved);
+
+		sym_node = UT_LIST_GET_NEXT(sym_list, sym_node);
+	}
+
+	graph = pars_sym_tab_global->query_graph;
+
+	graph->sym_tab = pars_sym_tab_global;
+	graph->info = info;
+
+	/* fprintf(stderr, "SQL graph size %lu\n", mem_heap_get_size(heap)); */
+
+	return(graph);
+}
+
+/**********************************************************************
+Completes a query graph by adding query thread and fork nodes
+above it and prepares the graph for running. The fork created is of
+type QUE_FORK_MYSQL_INTERFACE. */
+UNIV_INTERN
+que_thr_t*
+pars_complete_graph_for_exec(
+/*=========================*/
+				/* out: query thread node to run */
+	que_node_t*	node,	/* in: root node for an incomplete
+				query graph */
+	trx_t*		trx,	/* in: transaction handle */
+	mem_heap_t*	heap)	/* in: memory heap from which allocated */
+{
+	que_fork_t*	fork;
+	que_thr_t*	thr;
+
+	fork = que_fork_create(NULL, NULL, QUE_FORK_MYSQL_INTERFACE, heap);
+	fork->trx = trx;
+
+	thr = que_thr_create(fork, heap);
+
+	thr->child = node;
+
+	que_node_set_parent(node, thr);
+
+	trx->graph = NULL;
+
+	return(thr);
+}
+
+/********************************************************************
+Create parser info struct.*/
+UNIV_INTERN
+pars_info_t*
+pars_info_create(void)
+/*==================*/
+		/* out, own: info struct */
+{
+	pars_info_t*	info;
+	mem_heap_t*	heap;
+
+	heap = mem_heap_create(512);
+
+	info = mem_heap_alloc(heap, sizeof(*info));
+
+	info->heap = heap;
+	info->funcs = NULL;
+	info->bound_lits = NULL;
+	info->bound_ids = NULL;
+	info->graph_owns_us = TRUE;
+
+	return(info);
+}
+
+/********************************************************************
+Free info struct and everything it contains.*/
+UNIV_INTERN
+void
+pars_info_free(
+/*===========*/
+	pars_info_t*	info)	/* in: info struct */
+{
+	mem_heap_free(info->heap);
+}
+
+/********************************************************************
+Add bound literal. */
+UNIV_INTERN
+void
+pars_info_add_literal(
+/*==================*/
+	pars_info_t*	info,		/* in: info struct */
+	const char*	name,		/* in: name */
+	const void*	address,	/* in: address */
+	ulint		length,		/* in: length of data */
+	ulint		type,		/* in: type, e.g. DATA_FIXBINARY */
+	ulint		prtype)		/* in: precise type, e.g.
+					DATA_UNSIGNED */
+{
+	pars_bound_lit_t*	pbl;
+
+	ut_ad(!pars_info_get_bound_lit(info, name));
+
+	pbl = mem_heap_alloc(info->heap, sizeof(*pbl));
+
+	pbl->name = name;
+	pbl->address = address;
+	pbl->length = length;
+	pbl->type = type;
+	pbl->prtype = prtype;
+
+	if (!info->bound_lits) {
+		info->bound_lits = ib_vector_create(info->heap, 8);
+	}
+
+	ib_vector_push(info->bound_lits, pbl);
+}
+
+/********************************************************************
+Equivalent to pars_info_add_literal(info, name, str, strlen(str),
+DATA_VARCHAR, DATA_ENGLISH). */
+UNIV_INTERN
+void
+pars_info_add_str_literal(
+/*======================*/
+	pars_info_t*	info,		/* in: info struct */
+	const char*	name,		/* in: name */
+	const char*	str)		/* in: string */
+{
+	pars_info_add_literal(info, name, str, strlen(str),
+			      DATA_VARCHAR, DATA_ENGLISH);
+}
+
+/********************************************************************
+Equivalent to:
+
+char buf[4];
+mach_write_to_4(buf, val);
+pars_info_add_literal(info, name, buf, 4, DATA_INT, 0);
+
+except that the buffer is dynamically allocated from the info struct's
+heap. */
+UNIV_INTERN
+void
+pars_info_add_int4_literal(
+/*=======================*/
+	pars_info_t*	info,		/* in: info struct */
+	const char*	name,		/* in: name */
+	lint		val)		/* in: value */
+{
+	byte*	buf = mem_heap_alloc(info->heap, 4);
+
+	mach_write_to_4(buf, val);
+	pars_info_add_literal(info, name, buf, 4, DATA_INT, 0);
+}
+
+/********************************************************************
+Equivalent to:
+
+char buf[8];
+mach_write_to_8(buf, val);
+pars_info_add_literal(info, name, buf, 8, DATA_FIXBINARY, 0);
+
+except that the buffer is dynamically allocated from the info struct's
+heap. */
+UNIV_INTERN
+void
+pars_info_add_dulint_literal(
+/*=========================*/
+	pars_info_t*	info,		/* in: info struct */
+	const char*	name,		/* in: name */
+	dulint		val)		/* in: value */
+{
+	byte*	buf = mem_heap_alloc(info->heap, 8);
+
+	mach_write_to_8(buf, val);
+
+	pars_info_add_literal(info, name, buf, 8, DATA_FIXBINARY, 0);
+}
+
+/********************************************************************
+Add user function. */
+UNIV_INTERN
+void
+pars_info_add_function(
+/*===================*/
+	pars_info_t*		info,	/* in: info struct */
+	const char*		name,	/* in: function name */
+	pars_user_func_cb_t	func,	/* in: function address */
+	void*			arg)	/* in: user-supplied argument */
+{
+	pars_user_func_t*	puf;
+
+	ut_ad(!pars_info_get_user_func(info, name));
+
+	puf = mem_heap_alloc(info->heap, sizeof(*puf));
+
+	puf->name = name;
+	puf->func = func;
+	puf->arg = arg;
+
+	if (!info->funcs) {
+		info->funcs = ib_vector_create(info->heap, 8);
+	}
+
+	ib_vector_push(info->funcs, puf);
+}
+
+/********************************************************************
+Add bound id. */
+UNIV_INTERN
+void
+pars_info_add_id(
+/*=============*/
+	pars_info_t*	info,		/* in: info struct */
+	const char*	name,		/* in: name */
+	const char*	id)		/* in: id */
+{
+	pars_bound_id_t*	bid;
+
+	ut_ad(!pars_info_get_bound_id(info, name));
+
+	bid = mem_heap_alloc(info->heap, sizeof(*bid));
+
+	bid->name = name;
+	bid->id = id;
+
+	if (!info->bound_ids) {
+		info->bound_ids = ib_vector_create(info->heap, 8);
+	}
+
+	ib_vector_push(info->bound_ids, bid);
+}
+
+/********************************************************************
+Get user function with the given name.*/
+UNIV_INTERN
+pars_user_func_t*
+pars_info_get_user_func(
+/*====================*/
+					/* out: user func, or NULL if not
+					found */
+	pars_info_t*		info,	/* in: info struct */
+	const char*		name)	/* in: function name to find*/
+{
+	ulint		i;
+	ib_vector_t*	vec;
+
+	if (!info || !info->funcs) {
+		return(NULL);
+	}
+
+	vec = info->funcs;
+
+	for (i = 0; i < ib_vector_size(vec); i++) {
+		pars_user_func_t*	puf = ib_vector_get(vec, i);
+
+		if (strcmp(puf->name, name) == 0) {
+			return(puf);
+		}
+	}
+
+	return(NULL);
+}
+
+/********************************************************************
+Get bound literal with the given name.*/
+UNIV_INTERN
+pars_bound_lit_t*
+pars_info_get_bound_lit(
+/*====================*/
+					/* out: bound literal, or NULL if
+					not found */
+	pars_info_t*		info,	/* in: info struct */
+	const char*		name)	/* in: bound literal name to find */
+{
+	ulint		i;
+	ib_vector_t*	vec;
+
+	if (!info || !info->bound_lits) {
+		return(NULL);
+	}
+
+	vec = info->bound_lits;
+
+	for (i = 0; i < ib_vector_size(vec); i++) {
+		pars_bound_lit_t*	pbl = ib_vector_get(vec, i);
+
+		if (strcmp(pbl->name, name) == 0) {
+			return(pbl);
+		}
+	}
+
+	return(NULL);
+}
+
+/********************************************************************
+Get bound id with the given name.*/
+UNIV_INTERN
+pars_bound_id_t*
+pars_info_get_bound_id(
+/*===================*/
+					/* out: bound id, or NULL if not
+					found */
+	pars_info_t*		info,	/* in: info struct */
+	const char*		name)	/* in: bound id name to find */
+{
+	ulint		i;
+	ib_vector_t*	vec;
+
+	if (!info || !info->bound_ids) {
+		return(NULL);
+	}
+
+	vec = info->bound_ids;
+
+	for (i = 0; i < ib_vector_size(vec); i++) {
+		pars_bound_id_t*	bid = ib_vector_get(vec, i);
+
+		if (strcmp(bid->name, name) == 0) {
+			return(bid);
+		}
+	}
+
+	return(NULL);
+}
diff --git a/storage/xtradb/pars/pars0sym.c b/storage/xtradb/pars/pars0sym.c
new file mode 100644
index 00000000000..fb23547e767
--- /dev/null
+++ b/storage/xtradb/pars/pars0sym.c
@@ -0,0 +1,370 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+SQL parser symbol table
+
+Created 12/15/1997 Heikki Tuuri
+*******************************************************/
+
+#include "pars0sym.h"
+
+#ifdef UNIV_NONINL
+#include "pars0sym.ic"
+#endif
+
+#include "mem0mem.h"
+#include "data0type.h"
+#include "data0data.h"
+#include "pars0grm.h"
+#include "pars0pars.h"
+#include "que0que.h"
+#include "eval0eval.h"
+#include "row0sel.h"
+
+/**********************************************************************
+Creates a symbol table for a single stored procedure or query. */
+UNIV_INTERN
+sym_tab_t*
+sym_tab_create(
+/*===========*/
+				/* out, own: symbol table */
+	mem_heap_t*	heap)	/* in: memory heap where to create */
+{
+	sym_tab_t*	sym_tab;
+
+	sym_tab = mem_heap_alloc(heap, sizeof(sym_tab_t));
+
+	UT_LIST_INIT(sym_tab->sym_list);
+	UT_LIST_INIT(sym_tab->func_node_list);
+
+	sym_tab->heap = heap;
+
+	return(sym_tab);
+}
+
+/**********************************************************************
+Frees the memory allocated dynamically AFTER parsing phase for variables
+etc. in the symbol table. Does not free the mem heap where the table was
+originally created. Frees also SQL explicit cursor definitions. */
+UNIV_INTERN
+void
+sym_tab_free_private(
+/*=================*/
+	sym_tab_t*	sym_tab)	/* in, own: symbol table */
+{
+	sym_node_t*	sym;
+	func_node_t*	func;
+
+	sym = UT_LIST_GET_FIRST(sym_tab->sym_list);
+
+	while (sym) {
+		eval_node_free_val_buf(sym);
+
+		if (sym->prefetch_buf) {
+			sel_col_prefetch_buf_free(sym->prefetch_buf);
+		}
+
+		if (sym->cursor_def) {
+			que_graph_free_recursive(sym->cursor_def);
+		}
+
+		sym = UT_LIST_GET_NEXT(sym_list, sym);
+	}
+
+	func = UT_LIST_GET_FIRST(sym_tab->func_node_list);
+
+	while (func) {
+		eval_node_free_val_buf(func);
+
+		func = UT_LIST_GET_NEXT(func_node_list, func);
+	}
+}
+
+/**********************************************************************
+Adds an integer literal to a symbol table. */
+UNIV_INTERN
+sym_node_t*
+sym_tab_add_int_lit(
+/*================*/
+					/* out: symbol table node */
+	sym_tab_t*	sym_tab,	/* in: symbol table */
+	ulint		val)		/* in: integer value */
+{
+	sym_node_t*	node;
+	byte*		data;
+
+	node = mem_heap_alloc(sym_tab->heap, sizeof(sym_node_t));
+
+	node->common.type = QUE_NODE_SYMBOL;
+
+	node->resolved = TRUE;
+	node->token_type = SYM_LIT;
+
+	node->indirection = NULL;
+
+	dtype_set(dfield_get_type(&node->common.val), DATA_INT, 0, 4);
+
+	data = mem_heap_alloc(sym_tab->heap, 4);
+	mach_write_to_4(data, val);
+
+	dfield_set_data(&(node->common.val), data, 4);
+
+	node->common.val_buf_size = 0;
+	node->prefetch_buf = NULL;
+	node->cursor_def = NULL;
+
+	UT_LIST_ADD_LAST(sym_list, sym_tab->sym_list, node);
+
+	node->sym_table = sym_tab;
+
+	return(node);
+}
+
+/**********************************************************************
+Adds a string literal to a symbol table. */
+UNIV_INTERN
+sym_node_t*
+sym_tab_add_str_lit(
+/*================*/
+					/* out: symbol table node */
+	sym_tab_t*	sym_tab,	/* in: symbol table */
+	byte*		str,		/* in: string with no quotes around
+					it */
+	ulint		len)		/* in: string length */
+{
+	sym_node_t*	node;
+	byte*		data;
+
+	node = mem_heap_alloc(sym_tab->heap, sizeof(sym_node_t));
+
+	node->common.type = QUE_NODE_SYMBOL;
+
+	node->resolved = TRUE;
+	node->token_type = SYM_LIT;
+
+	node->indirection = NULL;
+
+	dtype_set(dfield_get_type(&node->common.val),
+		  DATA_VARCHAR, DATA_ENGLISH, 0);
+
+	if (len) {
+		data = mem_heap_alloc(sym_tab->heap, len);
+		ut_memcpy(data, str, len);
+	} else {
+		data = NULL;
+	}
+
+	dfield_set_data(&(node->common.val), data, len);
+
+	node->common.val_buf_size = 0;
+	node->prefetch_buf = NULL;
+	node->cursor_def = NULL;
+
+	UT_LIST_ADD_LAST(sym_list, sym_tab->sym_list, node);
+
+	node->sym_table = sym_tab;
+
+	return(node);
+}
+
+/**********************************************************************
+Add a bound literal to a symbol table. */
+UNIV_INTERN
+sym_node_t*
+sym_tab_add_bound_lit(
+/*==================*/
+					/* out: symbol table node */
+	sym_tab_t*	sym_tab,	/* in: symbol table */
+	const char*	name,		/* in: name of bound literal */
+	ulint*		lit_type)	/* out: type of literal (PARS_*_LIT) */
+{
+	sym_node_t*		node;
+	pars_bound_lit_t*	blit;
+	ulint			len = 0;
+
+	blit = pars_info_get_bound_lit(sym_tab->info, name);
+	ut_a(blit);
+
+	node = mem_heap_alloc(sym_tab->heap, sizeof(sym_node_t));
+
+	node->common.type = QUE_NODE_SYMBOL;
+
+	node->resolved = TRUE;
+	node->token_type = SYM_LIT;
+
+	node->indirection = NULL;
+
+	switch (blit->type) {
+	case DATA_FIXBINARY:
+		len = blit->length;
+		*lit_type = PARS_FIXBINARY_LIT;
+		break;
+
+	case DATA_BLOB:
+		*lit_type = PARS_BLOB_LIT;
+		break;
+
+	case DATA_VARCHAR:
+		*lit_type = PARS_STR_LIT;
+		break;
+
+	case DATA_CHAR:
+		ut_a(blit->length > 0);
+
+		len = blit->length;
+		*lit_type = PARS_STR_LIT;
+		break;
+
+	case DATA_INT:
+		ut_a(blit->length > 0);
+		ut_a(blit->length <= 8);
+
+		len = blit->length;
+		*lit_type = PARS_INT_LIT;
+		break;
+
+	default:
+		ut_error;
+	}
+
+	dtype_set(dfield_get_type(&node->common.val),
+		  blit->type, blit->prtype, len);
+
+	dfield_set_data(&(node->common.val), blit->address, blit->length);
+
+	node->common.val_buf_size = 0;
+	node->prefetch_buf = NULL;
+	node->cursor_def = NULL;
+
+	UT_LIST_ADD_LAST(sym_list, sym_tab->sym_list, node);
+
+	node->sym_table = sym_tab;
+
+	return(node);
+}
+
+/**********************************************************************
+Adds an SQL null literal to a symbol table. */
+UNIV_INTERN
+sym_node_t*
+sym_tab_add_null_lit(
+/*=================*/
+					/* out: symbol table node */
+	sym_tab_t*	sym_tab)	/* in: symbol table */
+{
+	sym_node_t*	node;
+
+	node = mem_heap_alloc(sym_tab->heap, sizeof(sym_node_t));
+
+	node->common.type = QUE_NODE_SYMBOL;
+
+	node->resolved = TRUE;
+	node->token_type = SYM_LIT;
+
+	node->indirection = NULL;
+
+	dfield_get_type(&node->common.val)->mtype = DATA_ERROR;
+
+	dfield_set_null(&node->common.val);
+
+	node->common.val_buf_size = 0;
+	node->prefetch_buf = NULL;
+	node->cursor_def = NULL;
+
+	UT_LIST_ADD_LAST(sym_list, sym_tab->sym_list, node);
+
+	node->sym_table = sym_tab;
+
+	return(node);
+}
+
+/**********************************************************************
+Adds an identifier to a symbol table. */
+UNIV_INTERN
+sym_node_t*
+sym_tab_add_id(
+/*===========*/
+					/* out: symbol table node */
+	sym_tab_t*	sym_tab,	/* in: symbol table */
+	byte*		name,		/* in: identifier name */
+	ulint		len)		/* in: identifier length */
+{
+	sym_node_t*	node;
+
+	node = mem_heap_alloc(sym_tab->heap, sizeof(sym_node_t));
+
+	node->common.type = QUE_NODE_SYMBOL;
+
+	node->resolved = FALSE;
+	node->indirection = NULL;
+
+	node->name = mem_heap_strdupl(sym_tab->heap, (char*) name, len);
+	node->name_len = len;
+
+	UT_LIST_ADD_LAST(sym_list, sym_tab->sym_list, node);
+
+	dfield_set_null(&node->common.val);
+
+	node->common.val_buf_size = 0;
+	node->prefetch_buf = NULL;
+	node->cursor_def = NULL;
+
+	node->sym_table = sym_tab;
+
+	return(node);
+}
+
+/**********************************************************************
+Add a bound identifier to a symbol table. */
+UNIV_INTERN
+sym_node_t*
+sym_tab_add_bound_id(
+/*===========*/
+					/* out: symbol table node */
+	sym_tab_t*	sym_tab,	/* in: symbol table */
+	const char*	name)		/* in: name of bound id */
+{
+	sym_node_t*		node;
+	pars_bound_id_t*	bid;
+
+	bid = pars_info_get_bound_id(sym_tab->info, name);
+	ut_a(bid);
+
+	node = mem_heap_alloc(sym_tab->heap, sizeof(sym_node_t));
+
+	node->common.type = QUE_NODE_SYMBOL;
+
+	node->resolved = FALSE;
+	node->indirection = NULL;
+
+	node->name = mem_heap_strdup(sym_tab->heap, bid->id);
+	node->name_len = strlen(node->name);
+
+	UT_LIST_ADD_LAST(sym_list, sym_tab->sym_list, node);
+
+	dfield_set_null(&node->common.val);
+
+	node->common.val_buf_size = 0;
+	node->prefetch_buf = NULL;
+	node->cursor_def = NULL;
+
+	node->sym_table = sym_tab;
+
+	return(node);
+}
diff --git a/storage/xtradb/plug.in b/storage/xtradb/plug.in
new file mode 100644
index 00000000000..4d5e792bfb1
--- /dev/null
+++ b/storage/xtradb/plug.in
@@ -0,0 +1,80 @@
+#
+# Copyright (c) 2006, 2009, Innobase Oy. All Rights Reserved.
+# 
+# This program is free software; you can redistribute it and/or modify it under
+# the terms of the GNU General Public License as published by the Free Software
+# Foundation; version 2 of the License.
+# 
+# This program is distributed in the hope that it will be useful, but WITHOUT
+# ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+# FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+# 
+# You should have received a copy of the GNU General Public License along with
+# this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+# Place, Suite 330, Boston, MA 02111-1307 USA
+#
+
+MYSQL_STORAGE_ENGINE(innobase,  innodb, [InnoDB Storage Engine],
+        [Transactional Tables using InnoDB], [max,max-no-ndb])
+MYSQL_PLUGIN_DIRECTORY(innobase, [storage/xtradb])
+MYSQL_PLUGIN_STATIC(innobase,   [libinnobase.a])
+MYSQL_PLUGIN_DYNAMIC(innobase,  [ha_innodb.la])
+MYSQL_PLUGIN_ACTIONS(innobase,  [
+  AC_CHECK_HEADERS(sched.h)
+  AC_CHECK_SIZEOF(int, 4)
+  AC_CHECK_SIZEOF(long, 4)
+  AC_CHECK_SIZEOF(void*, 4)
+  AC_CHECK_FUNCS(sched_yield fdatasync localtime_r)
+  AC_C_BIGENDIAN
+  case "$target_os" in
+	lin*)
+		CFLAGS="$CFLAGS -DUNIV_LINUX";;
+	hpux10*)
+		CFLAGS="$CFLAGS -DUNIV_MUST_NOT_INLINE -DUNIV_HPUX -DUNIV_HPUX10";;
+	hp*)
+		CFLAGS="$CFLAGS -DUNIV_MUST_NOT_INLINE -DUNIV_HPUX";;
+	aix*)
+		CFLAGS="$CFLAGS -DUNIV_AIX";;
+	irix*|osf*|sysv5uw7*|openbsd*)
+		CFLAGS="$CFLAGS -DUNIV_MUST_NOT_INLINE";;
+	*solaris*|*SunOS*)
+		CFLAGS="$CFLAGS -DUNIV_SOLARIS";;
+  esac
+  INNODB_DYNAMIC_CFLAGS="-DMYSQL_DYNAMIC_PLUGIN"
+  case "$target_cpu" in
+	x86_64)
+		# The AMD64 ABI forbids absolute addresses in shared libraries
+		;;
+	*86)
+		# Use absolute addresses on IA-32
+		INNODB_DYNAMIC_CFLAGS="$INNODB_DYNAMIC_CFLAGS -prefer-non-pic"
+		;;
+  esac
+  AC_SUBST(INNODB_DYNAMIC_CFLAGS)
+  AC_MSG_CHECKING(whether pthread_t can be used by GCC atomic builtins)
+  AC_TRY_RUN(
+    [
+      #include <pthread.h>
+
+      int main(int argc, char** argv) {
+        pthread_t       x1;
+        pthread_t       x2;
+        pthread_t       x3;
+
+        __sync_bool_compare_and_swap(&x1, x2, x3);
+
+        return(0);
+      }
+    ],
+    [
+      AC_DEFINE([HAVE_ATOMIC_PTHREAD_T], [1],
+                [pthread_t can be used by GCC atomic builtins])
+      AC_MSG_RESULT(yes)
+    ],
+    [
+      AC_MSG_RESULT(no)
+    ]
+    )
+  ])
+
+# vim: set ft=config:
diff --git a/storage/xtradb/que/que0que.c b/storage/xtradb/que/que0que.c
new file mode 100644
index 00000000000..91a9d30ec4c
--- /dev/null
+++ b/storage/xtradb/que/que0que.c
@@ -0,0 +1,1461 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Query graph
+
+Created 5/27/1996 Heikki Tuuri
+*******************************************************/
+
+#include "que0que.h"
+
+#ifdef UNIV_NONINL
+#include "que0que.ic"
+#endif
+
+#include "srv0que.h"
+#include "usr0sess.h"
+#include "trx0trx.h"
+#include "trx0roll.h"
+#include "row0undo.h"
+#include "row0ins.h"
+#include "row0upd.h"
+#include "row0sel.h"
+#include "row0purge.h"
+#include "dict0crea.h"
+#include "log0log.h"
+#include "eval0proc.h"
+#include "eval0eval.h"
+#include "pars0types.h"
+
+#define QUE_PARALLELIZE_LIMIT	(64 * 256 * 256 * 256)
+#define QUE_ROUND_ROBIN_LIMIT	(64 * 256 * 256 * 256)
+#define QUE_MAX_LOOPS_WITHOUT_CHECK	16
+
+#ifdef UNIV_DEBUG
+/* If the following flag is set TRUE, the module will print trace info
+of SQL execution in the UNIV_SQL_DEBUG version */
+UNIV_INTERN ibool	que_trace_on		= FALSE;
+#endif /* UNIV_DEBUG */
+
+/* Short introduction to query graphs
+   ==================================
+
+A query graph consists of nodes linked to each other in various ways. The
+execution starts at que_run_threads() which takes a que_thr_t parameter.
+que_thr_t contains two fields that control query graph execution: run_node
+and prev_node. run_node is the next node to execute and prev_node is the
+last node executed.
+
+Each node has a pointer to a 'next' statement, i.e., its brother, and a
+pointer to its parent node. The next pointer is NULL in the last statement
+of a block.
+
+Loop nodes contain a link to the first statement of the enclosed statement
+list. While the loop runs, que_thr_step() checks if execution to the loop
+node came from its parent or from one of the statement nodes in the loop. If
+it came from the parent of the loop node it starts executing the first
+statement node in the loop. If it came from one of the statement nodes in
+the loop, then it checks if the statement node has another statement node
+following it, and runs it if so.
+
+To signify loop ending, the loop statements (see e.g. while_step()) set
+que_thr_t->run_node to the loop node's parent node. This is noticed on the
+next call of que_thr_step() and execution proceeds to the node pointed to by
+the loop node's 'next' pointer.
+
+For example, the code:
+
+X := 1;
+WHILE X < 5 LOOP
+ X := X + 1;
+ X := X + 1;
+X := 5
+
+will result in the following node hierarchy, with the X-axis indicating
+'next' links and the Y-axis indicating parent/child links:
+
+A - W - A
+    |
+    |
+    A - A
+
+A = assign_node_t, W = while_node_t. */
+
+/* How a stored procedure containing COMMIT or ROLLBACK commands
+is executed?
+
+The commit or rollback can be seen as a subprocedure call.
+The problem is that if there are several query threads
+currently running within the transaction, their action could
+mess the commit or rollback operation. Or, at the least, the
+operation would be difficult to visualize and keep in control.
+
+Therefore the query thread requesting a commit or a rollback
+sends to the transaction a signal, which moves the transaction
+to TRX_QUE_SIGNALED state. All running query threads of the
+transaction will eventually notice that the transaction is now in
+this state and voluntarily suspend themselves. Only the last
+query thread which suspends itself will trigger handling of
+the signal.
+
+When the transaction starts to handle a rollback or commit
+signal, it builds a query graph which, when executed, will
+roll back or commit the incomplete transaction. The transaction
+is moved to the TRX_QUE_ROLLING_BACK or TRX_QUE_COMMITTING state.
+If specified, the SQL cursors opened by the transaction are closed.
+When the execution of the graph completes, it is like returning
+from a subprocedure: the query thread which requested the operation
+starts running again. */
+
+/**************************************************************************
+Moves a thread from another state to the QUE_THR_RUNNING state. Increments
+the n_active_thrs counters of the query graph and transaction.
+***NOTE***: This is the only function in which such a transition is allowed
+to happen! */
+static
+void
+que_thr_move_to_run_state(
+/*======================*/
+	que_thr_t*	thr);	/* in: an query thread */
+
+/***************************************************************************
+Adds a query graph to the session's list of graphs. */
+UNIV_INTERN
+void
+que_graph_publish(
+/*==============*/
+	que_t*	graph,	/* in: graph */
+	sess_t*	sess)	/* in: session */
+{
+	ut_ad(mutex_own(&kernel_mutex));
+
+	UT_LIST_ADD_LAST(graphs, sess->graphs, graph);
+}
+
+/***************************************************************************
+Creates a query graph fork node. */
+UNIV_INTERN
+que_fork_t*
+que_fork_create(
+/*============*/
+					/* out, own: fork node */
+	que_t*		graph,		/* in: graph, if NULL then this
+					fork node is assumed to be the
+					graph root */
+	que_node_t*	parent,		/* in: parent node */
+	ulint		fork_type,	/* in: fork type */
+	mem_heap_t*	heap)		/* in: memory heap where created */
+{
+	que_fork_t*	fork;
+
+	ut_ad(heap);
+
+	fork = mem_heap_alloc(heap, sizeof(que_fork_t));
+
+	fork->common.type = QUE_NODE_FORK;
+	fork->n_active_thrs = 0;
+
+	fork->state = QUE_FORK_COMMAND_WAIT;
+
+	if (graph != NULL) {
+		fork->graph = graph;
+	} else {
+		fork->graph = fork;
+	}
+
+	fork->common.parent = parent;
+	fork->fork_type = fork_type;
+
+	fork->caller = NULL;
+
+	UT_LIST_INIT(fork->thrs);
+
+	fork->sym_tab = NULL;
+	fork->info = NULL;
+
+	fork->heap = heap;
+
+	return(fork);
+}
+
+/***************************************************************************
+Creates a query graph thread node. */
+UNIV_INTERN
+que_thr_t*
+que_thr_create(
+/*===========*/
+				/* out, own: query thread node */
+	que_fork_t*	parent,	/* in: parent node, i.e., a fork node */
+	mem_heap_t*	heap)	/* in: memory heap where created */
+{
+	que_thr_t*	thr;
+
+	ut_ad(parent && heap);
+
+	thr = mem_heap_alloc(heap, sizeof(que_thr_t));
+
+	thr->common.type = QUE_NODE_THR;
+	thr->common.parent = parent;
+
+	thr->magic_n = QUE_THR_MAGIC_N;
+
+	thr->graph = parent->graph;
+
+	thr->state = QUE_THR_COMMAND_WAIT;
+
+	thr->is_active = FALSE;
+
+	thr->run_node = NULL;
+	thr->resource = 0;
+	thr->lock_state = QUE_THR_LOCK_NOLOCK;
+
+	UT_LIST_ADD_LAST(thrs, parent->thrs, thr);
+
+	return(thr);
+}
+
+/**************************************************************************
+Moves a suspended query thread to the QUE_THR_RUNNING state and may release
+a single worker thread to execute it. This function should be used to end
+the wait state of a query thread waiting for a lock or a stored procedure
+completion. */
+UNIV_INTERN
+void
+que_thr_end_wait(
+/*=============*/
+	que_thr_t*	thr,		/* in: query thread in the
+					QUE_THR_LOCK_WAIT,
+					or QUE_THR_PROCEDURE_WAIT, or
+					QUE_THR_SIG_REPLY_WAIT state */
+	que_thr_t**	next_thr)	/* in/out: next query thread to run;
+					if the value which is passed in is
+					a pointer to a NULL pointer, then the
+					calling function can start running
+					a new query thread; if NULL is passed
+					as the parameter, it is ignored */
+{
+	ibool	was_active;
+
+	ut_ad(mutex_own(&kernel_mutex));
+	ut_ad(thr);
+	ut_ad((thr->state == QUE_THR_LOCK_WAIT)
+	      || (thr->state == QUE_THR_PROCEDURE_WAIT)
+	      || (thr->state == QUE_THR_SIG_REPLY_WAIT));
+	ut_ad(thr->run_node);
+
+	thr->prev_node = thr->run_node;
+
+	was_active = thr->is_active;
+
+	que_thr_move_to_run_state(thr);
+
+	if (was_active) {
+
+		return;
+	}
+
+	if (next_thr && *next_thr == NULL) {
+		*next_thr = thr;
+	} else {
+		ut_a(0);
+		srv_que_task_enqueue_low(thr);
+	}
+}
+
+/**************************************************************************
+Same as que_thr_end_wait, but no parameter next_thr available. */
+UNIV_INTERN
+void
+que_thr_end_wait_no_next_thr(
+/*=========================*/
+	que_thr_t*	thr)	/* in: query thread in the QUE_THR_LOCK_WAIT,
+				or QUE_THR_PROCEDURE_WAIT, or
+				QUE_THR_SIG_REPLY_WAIT state */
+{
+	ibool	was_active;
+
+	ut_a(thr->state == QUE_THR_LOCK_WAIT);	/* In MySQL this is the
+						only possible state here */
+	ut_ad(mutex_own(&kernel_mutex));
+	ut_ad(thr);
+	ut_ad((thr->state == QUE_THR_LOCK_WAIT)
+	      || (thr->state == QUE_THR_PROCEDURE_WAIT)
+	      || (thr->state == QUE_THR_SIG_REPLY_WAIT));
+
+	was_active = thr->is_active;
+
+	que_thr_move_to_run_state(thr);
+
+	if (was_active) {
+
+		return;
+	}
+
+	/* In MySQL we let the OS thread (not just the query thread) to wait
+	for the lock to be released: */
+
+	srv_release_mysql_thread_if_suspended(thr);
+
+	/* srv_que_task_enqueue_low(thr); */
+}
+
+/**************************************************************************
+Inits a query thread for a command. */
+UNIV_INLINE
+void
+que_thr_init_command(
+/*=================*/
+	que_thr_t*	thr)	/* in: query thread */
+{
+	thr->run_node = thr;
+	thr->prev_node = thr->common.parent;
+
+	que_thr_move_to_run_state(thr);
+}
+
+/**************************************************************************
+Starts execution of a command in a query fork. Picks a query thread which
+is not in the QUE_THR_RUNNING state and moves it to that state. If none
+can be chosen, a situation which may arise in parallelized fetches, NULL
+is returned. */
+UNIV_INTERN
+que_thr_t*
+que_fork_start_command(
+/*===================*/
+				/* out: a query thread of the graph moved to
+				QUE_THR_RUNNING state, or NULL; the query
+				thread should be executed by que_run_threads
+				by the caller */
+	que_fork_t*	fork)	/* in: a query fork */
+{
+	que_thr_t*	thr;
+	que_thr_t*	suspended_thr = NULL;
+	que_thr_t*	completed_thr = NULL;
+
+	fork->state = QUE_FORK_ACTIVE;
+
+	fork->last_sel_node = NULL;
+
+	suspended_thr = NULL;
+	completed_thr = NULL;
+
+	/* Choose the query thread to run: usually there is just one thread,
+	but in a parallelized select, which necessarily is non-scrollable,
+	there may be several to choose from */
+
+	/* First we try to find a query thread in the QUE_THR_COMMAND_WAIT
+	state. Then we try to find a query thread in the QUE_THR_SUSPENDED
+	state, finally we try to find a query thread in the QUE_THR_COMPLETED
+	state */
+
+	thr = UT_LIST_GET_FIRST(fork->thrs);
+
+	/* We make a single pass over the thr list within which we note which
+	threads are ready to run. */
+	while (thr) {
+		switch (thr->state) {
+		case QUE_THR_COMMAND_WAIT:
+
+			/* We have to send the initial message to query thread
+			to start it */
+
+			que_thr_init_command(thr);
+
+			return(thr);
+
+		case QUE_THR_SUSPENDED:
+			/* In this case the execution of the thread was
+			suspended: no initial message is needed because
+			execution can continue from where it was left */
+			if (!suspended_thr) {
+				suspended_thr = thr;
+			}
+
+			break;
+
+		case QUE_THR_COMPLETED:
+			if (!completed_thr) {
+				completed_thr = thr;
+			}
+
+			break;
+
+		case QUE_THR_LOCK_WAIT:
+			ut_error;
+
+		}
+
+		thr = UT_LIST_GET_NEXT(thrs, thr);
+	}
+
+	if (suspended_thr) {
+
+		thr = suspended_thr;
+		que_thr_move_to_run_state(thr);
+
+	} else if (completed_thr) {
+
+		thr = completed_thr;
+		que_thr_init_command(thr);
+	}
+
+	return(thr);
+}
+
+/**************************************************************************
+After signal handling is finished, returns control to a query graph error
+handling routine. (Currently, just returns the control to the root of the
+graph so that the graph can communicate an error message to the client.) */
+UNIV_INTERN
+void
+que_fork_error_handle(
+/*==================*/
+	trx_t*	trx __attribute__((unused)),	/* in: trx */
+	que_t*	fork)	/* in: query graph which was run before signal
+			handling started, NULL not allowed */
+{
+	que_thr_t*	thr;
+
+	ut_ad(mutex_own(&kernel_mutex));
+	ut_ad(trx->sess->state == SESS_ERROR);
+	ut_ad(UT_LIST_GET_LEN(trx->reply_signals) == 0);
+	ut_ad(UT_LIST_GET_LEN(trx->wait_thrs) == 0);
+
+	thr = UT_LIST_GET_FIRST(fork->thrs);
+
+	while (thr != NULL) {
+		ut_ad(!thr->is_active);
+		ut_ad(thr->state != QUE_THR_SIG_REPLY_WAIT);
+		ut_ad(thr->state != QUE_THR_LOCK_WAIT);
+
+		thr->run_node = thr;
+		thr->prev_node = thr->child;
+		thr->state = QUE_THR_COMPLETED;
+
+		thr = UT_LIST_GET_NEXT(thrs, thr);
+	}
+
+	thr = UT_LIST_GET_FIRST(fork->thrs);
+
+	que_thr_move_to_run_state(thr);
+
+	ut_a(0);
+	srv_que_task_enqueue_low(thr);
+}
+
+/********************************************************************
+Tests if all the query threads in the same fork have a given state. */
+UNIV_INLINE
+ibool
+que_fork_all_thrs_in_state(
+/*=======================*/
+				/* out: TRUE if all the query threads in the
+				same fork were in the given state */
+	que_fork_t*	fork,	/* in: query fork */
+	ulint		state)	/* in: state */
+{
+	que_thr_t*	thr_node;
+
+	thr_node = UT_LIST_GET_FIRST(fork->thrs);
+
+	while (thr_node != NULL) {
+		if (thr_node->state != state) {
+
+			return(FALSE);
+		}
+
+		thr_node = UT_LIST_GET_NEXT(thrs, thr_node);
+	}
+
+	return(TRUE);
+}
+
+/**************************************************************************
+Calls que_graph_free_recursive for statements in a statement list. */
+static
+void
+que_graph_free_stat_list(
+/*=====================*/
+	que_node_t*	node)	/* in: first query graph node in the list */
+{
+	while (node) {
+		que_graph_free_recursive(node);
+
+		node = que_node_get_next(node);
+	}
+}
+
+/**************************************************************************
+Frees a query graph, but not the heap where it was created. Does not free
+explicit cursor declarations, they are freed in que_graph_free. */
+UNIV_INTERN
+void
+que_graph_free_recursive(
+/*=====================*/
+	que_node_t*	node)	/* in: query graph node */
+{
+	que_fork_t*	fork;
+	que_thr_t*	thr;
+	undo_node_t*	undo;
+	sel_node_t*	sel;
+	ins_node_t*	ins;
+	upd_node_t*	upd;
+	tab_node_t*	cre_tab;
+	ind_node_t*	cre_ind;
+
+	if (node == NULL) {
+
+		return;
+	}
+
+	switch (que_node_get_type(node)) {
+
+	case QUE_NODE_FORK:
+		fork = node;
+
+		thr = UT_LIST_GET_FIRST(fork->thrs);
+
+		while (thr) {
+			que_graph_free_recursive(thr);
+
+			thr = UT_LIST_GET_NEXT(thrs, thr);
+		}
+
+		break;
+	case QUE_NODE_THR:
+
+		thr = node;
+
+		if (thr->magic_n != QUE_THR_MAGIC_N) {
+			fprintf(stderr,
+				"que_thr struct appears corrupt;"
+				" magic n %lu\n",
+				(unsigned long) thr->magic_n);
+			mem_analyze_corruption(thr);
+			ut_error;
+		}
+
+		thr->magic_n = QUE_THR_MAGIC_FREED;
+
+		que_graph_free_recursive(thr->child);
+
+		break;
+	case QUE_NODE_UNDO:
+
+		undo = node;
+
+		mem_heap_free(undo->heap);
+
+		break;
+	case QUE_NODE_SELECT:
+
+		sel = node;
+
+		sel_node_free_private(sel);
+
+		break;
+	case QUE_NODE_INSERT:
+
+		ins = node;
+
+		que_graph_free_recursive(ins->select);
+
+		mem_heap_free(ins->entry_sys_heap);
+
+		break;
+	case QUE_NODE_UPDATE:
+
+		upd = node;
+
+		if (upd->in_mysql_interface) {
+
+			btr_pcur_free_for_mysql(upd->pcur);
+		}
+
+		que_graph_free_recursive(upd->cascade_node);
+
+		if (upd->cascade_heap) {
+			mem_heap_free(upd->cascade_heap);
+		}
+
+		que_graph_free_recursive(upd->select);
+
+		mem_heap_free(upd->heap);
+
+		break;
+	case QUE_NODE_CREATE_TABLE:
+		cre_tab = node;
+
+		que_graph_free_recursive(cre_tab->tab_def);
+		que_graph_free_recursive(cre_tab->col_def);
+		que_graph_free_recursive(cre_tab->commit_node);
+
+		mem_heap_free(cre_tab->heap);
+
+		break;
+	case QUE_NODE_CREATE_INDEX:
+		cre_ind = node;
+
+		que_graph_free_recursive(cre_ind->ind_def);
+		que_graph_free_recursive(cre_ind->field_def);
+		que_graph_free_recursive(cre_ind->commit_node);
+
+		mem_heap_free(cre_ind->heap);
+
+		break;
+	case QUE_NODE_PROC:
+		que_graph_free_stat_list(((proc_node_t*)node)->stat_list);
+
+		break;
+	case QUE_NODE_IF:
+		que_graph_free_stat_list(((if_node_t*)node)->stat_list);
+		que_graph_free_stat_list(((if_node_t*)node)->else_part);
+		que_graph_free_stat_list(((if_node_t*)node)->elsif_list);
+
+		break;
+	case QUE_NODE_ELSIF:
+		que_graph_free_stat_list(((elsif_node_t*)node)->stat_list);
+
+		break;
+	case QUE_NODE_WHILE:
+		que_graph_free_stat_list(((while_node_t*)node)->stat_list);
+
+		break;
+	case QUE_NODE_FOR:
+		que_graph_free_stat_list(((for_node_t*)node)->stat_list);
+
+		break;
+
+	case QUE_NODE_ASSIGNMENT:
+	case QUE_NODE_EXIT:
+	case QUE_NODE_RETURN:
+	case QUE_NODE_COMMIT:
+	case QUE_NODE_ROLLBACK:
+	case QUE_NODE_LOCK:
+	case QUE_NODE_FUNC:
+	case QUE_NODE_ORDER:
+	case QUE_NODE_ROW_PRINTF:
+	case QUE_NODE_OPEN:
+	case QUE_NODE_FETCH:
+		/* No need to do anything */
+
+		break;
+	default:
+		fprintf(stderr,
+			"que_node struct appears corrupt; type %lu\n",
+			(unsigned long) que_node_get_type(node));
+		mem_analyze_corruption(node);
+		ut_error;
+	}
+}
+
+/**************************************************************************
+Frees a query graph. */
+UNIV_INTERN
+void
+que_graph_free(
+/*===========*/
+	que_t*	graph)	/* in: query graph; we assume that the memory
+			heap where this graph was created is private
+			to this graph: if not, then use
+			que_graph_free_recursive and free the heap
+			afterwards! */
+{
+	ut_ad(graph);
+
+	if (graph->sym_tab) {
+		/* The following call frees dynamic memory allocated
+		for variables etc. during execution. Frees also explicit
+		cursor definitions. */
+
+		sym_tab_free_private(graph->sym_tab);
+	}
+
+	if (graph->info && graph->info->graph_owns_us) {
+		pars_info_free(graph->info);
+	}
+
+	que_graph_free_recursive(graph);
+
+	mem_heap_free(graph->heap);
+}
+
+/**************************************************************************
+Checks if the query graph is in a state where it should be freed, and
+frees it in that case. If the session is in a state where it should be
+closed, also this is done. */
+UNIV_INTERN
+ibool
+que_graph_try_free(
+/*===============*/
+			/* out: TRUE if freed */
+	que_t*	graph)	/* in: query graph */
+{
+	sess_t*	sess;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	sess = (graph->trx)->sess;
+
+	if ((graph->state == QUE_FORK_BEING_FREED)
+	    && (graph->n_active_thrs == 0)) {
+
+		UT_LIST_REMOVE(graphs, sess->graphs, graph);
+		que_graph_free(graph);
+
+		sess_try_close(sess);
+
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
+
+/********************************************************************
+Performs an execution step on a thr node. */
+static
+que_thr_t*
+que_thr_node_step(
+/*==============*/
+				/* out: query thread to run next, or NULL
+				if none */
+	que_thr_t*	thr)	/* in: query thread where run_node must
+				be the thread node itself */
+{
+	ut_ad(thr->run_node == thr);
+
+	if (thr->prev_node == thr->common.parent) {
+		/* If control to the node came from above, it is just passed
+		on */
+
+		thr->run_node = thr->child;
+
+		return(thr);
+	}
+
+	mutex_enter(&kernel_mutex);
+
+	if (que_thr_peek_stop(thr)) {
+
+		mutex_exit(&kernel_mutex);
+
+		return(thr);
+	}
+
+	/* Thread execution completed */
+
+	thr->state = QUE_THR_COMPLETED;
+
+	mutex_exit(&kernel_mutex);
+
+	return(NULL);
+}
+
+/**************************************************************************
+Moves a thread from another state to the QUE_THR_RUNNING state. Increments
+the n_active_thrs counters of the query graph and transaction if thr was
+not active.
+***NOTE***: This and ..._mysql are  the only functions in which such a
+transition is allowed to happen! */
+static
+void
+que_thr_move_to_run_state(
+/*======================*/
+	que_thr_t*	thr)	/* in: an query thread */
+{
+	trx_t*	trx;
+
+	ut_ad(thr->state != QUE_THR_RUNNING);
+
+	trx = thr_get_trx(thr);
+
+	if (!thr->is_active) {
+
+		(thr->graph)->n_active_thrs++;
+
+		trx->n_active_thrs++;
+
+		thr->is_active = TRUE;
+
+		ut_ad((thr->graph)->n_active_thrs == 1);
+		ut_ad(trx->n_active_thrs == 1);
+	}
+
+	thr->state = QUE_THR_RUNNING;
+}
+
+/**************************************************************************
+Decrements the query thread reference counts in the query graph and the
+transaction. May start signal handling, e.g., a rollback.
+*** NOTE ***:
+This and que_thr_stop_for_mysql are the only functions where the reference
+count can be decremented and this function may only be called from inside
+que_run_threads or que_thr_check_if_switch! These restrictions exist to make
+the rollback code easier to maintain. */
+static
+void
+que_thr_dec_refer_count(
+/*====================*/
+	que_thr_t*	thr,		/* in: query thread */
+	que_thr_t**	next_thr)	/* in/out: next query thread to run;
+					if the value which is passed in is
+					a pointer to a NULL pointer, then the
+					calling function can start running
+					a new query thread */
+{
+	que_fork_t*	fork;
+	trx_t*		trx;
+	ulint		fork_type;
+	ibool		stopped;
+
+	fork = thr->common.parent;
+	trx = thr_get_trx(thr);
+
+	mutex_enter(&kernel_mutex);
+
+	ut_a(thr->is_active);
+
+	if (thr->state == QUE_THR_RUNNING) {
+
+		stopped = que_thr_stop(thr);
+
+		if (!stopped) {
+			/* The reason for the thr suspension or wait was
+			already canceled before we came here: continue
+			running the thread */
+
+			/* fputs("!!!!!!!! Wait already ended: continue thr\n",
+			stderr); */
+
+			if (next_thr && *next_thr == NULL) {
+				/* Normally srv_suspend_mysql_thread resets
+				the state to DB_SUCCESS before waiting, but
+				in this case we have to do it here,
+				otherwise nobody does it. */
+				trx->error_state = DB_SUCCESS;
+
+				*next_thr = thr;
+			} else {
+				ut_error;
+				srv_que_task_enqueue_low(thr);
+			}
+
+			mutex_exit(&kernel_mutex);
+
+			return;
+		}
+	}
+
+	ut_ad(fork->n_active_thrs == 1);
+	ut_ad(trx->n_active_thrs == 1);
+
+	fork->n_active_thrs--;
+	trx->n_active_thrs--;
+
+	thr->is_active = FALSE;
+
+	if (trx->n_active_thrs > 0) {
+
+		mutex_exit(&kernel_mutex);
+
+		return;
+	}
+
+	fork_type = fork->fork_type;
+
+	/* Check if all query threads in the same fork are completed */
+
+	if (que_fork_all_thrs_in_state(fork, QUE_THR_COMPLETED)) {
+
+		switch (fork_type) {
+		case QUE_FORK_ROLLBACK:
+			/* This is really the undo graph used in rollback,
+			no roll_node in this graph */
+
+			ut_ad(UT_LIST_GET_LEN(trx->signals) > 0);
+			ut_ad(trx->handling_signals == TRUE);
+
+			trx_finish_rollback_off_kernel(fork, trx, next_thr);
+			break;
+
+		case QUE_FORK_PURGE:
+		case QUE_FORK_RECOVERY:
+		case QUE_FORK_MYSQL_INTERFACE:
+
+			/* Do nothing */
+			break;
+
+		default:
+			ut_error;	/* not used in MySQL */
+		}
+	}
+
+	if (UT_LIST_GET_LEN(trx->signals) > 0 && trx->n_active_thrs == 0) {
+
+		/* If the trx is signaled and its query thread count drops to
+		zero, then we start processing a signal; from it we may get
+		a new query thread to run */
+
+		trx_sig_start_handle(trx, next_thr);
+	}
+
+	if (trx->handling_signals && UT_LIST_GET_LEN(trx->signals) == 0) {
+
+		trx_end_signal_handling(trx);
+	}
+
+	mutex_exit(&kernel_mutex);
+}
+
+/**************************************************************************
+Stops a query thread if graph or trx is in a state requiring it. The
+conditions are tested in the order (1) graph, (2) trx. The kernel mutex has
+to be reserved. */
+UNIV_INTERN
+ibool
+que_thr_stop(
+/*=========*/
+				/* out: TRUE if stopped */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	trx_t*	trx;
+	que_t*	graph;
+	ibool	ret	= TRUE;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	graph = thr->graph;
+	trx = graph->trx;
+
+	if (graph->state == QUE_FORK_COMMAND_WAIT) {
+		thr->state = QUE_THR_SUSPENDED;
+
+	} else if (trx->que_state == TRX_QUE_LOCK_WAIT) {
+
+		UT_LIST_ADD_FIRST(trx_thrs, trx->wait_thrs, thr);
+		thr->state = QUE_THR_LOCK_WAIT;
+
+	} else if (trx->error_state != DB_SUCCESS
+		   && trx->error_state != DB_LOCK_WAIT) {
+
+		/* Error handling built for the MySQL interface */
+		thr->state = QUE_THR_COMPLETED;
+
+	} else if (UT_LIST_GET_LEN(trx->signals) > 0
+		   && graph->fork_type != QUE_FORK_ROLLBACK) {
+
+		thr->state = QUE_THR_SUSPENDED;
+	} else {
+		ut_ad(graph->state == QUE_FORK_ACTIVE);
+
+		ret = FALSE;
+	}
+
+	return(ret);
+}
+
+/**************************************************************************
+A patch for MySQL used to 'stop' a dummy query thread used in MySQL. The
+query thread is stopped and made inactive, except in the case where
+it was put to the lock wait state in lock0lock.c, but the lock has already
+been granted or the transaction chosen as a victim in deadlock resolution. */
+UNIV_INTERN
+void
+que_thr_stop_for_mysql(
+/*===================*/
+	que_thr_t*	thr)	/* in: query thread */
+{
+	trx_t*	trx;
+
+	trx = thr_get_trx(thr);
+
+	mutex_enter(&kernel_mutex);
+
+	if (thr->state == QUE_THR_RUNNING) {
+
+		if (trx->error_state != DB_SUCCESS
+		    && trx->error_state != DB_LOCK_WAIT) {
+
+			/* Error handling built for the MySQL interface */
+			thr->state = QUE_THR_COMPLETED;
+		} else {
+			/* It must have been a lock wait but the lock was
+			already released, or this transaction was chosen
+			as a victim in selective deadlock resolution */
+
+			mutex_exit(&kernel_mutex);
+
+			return;
+		}
+	}
+
+	ut_ad(thr->is_active == TRUE);
+	ut_ad(trx->n_active_thrs == 1);
+	ut_ad(thr->graph->n_active_thrs == 1);
+
+	thr->is_active = FALSE;
+	(thr->graph)->n_active_thrs--;
+
+	trx->n_active_thrs--;
+
+	mutex_exit(&kernel_mutex);
+}
+
+/**************************************************************************
+Moves a thread from another state to the QUE_THR_RUNNING state. Increments
+the n_active_thrs counters of the query graph and transaction if thr was
+not active. */
+UNIV_INTERN
+void
+que_thr_move_to_run_state_for_mysql(
+/*================================*/
+	que_thr_t*	thr,	/* in: an query thread */
+	trx_t*		trx)	/* in: transaction */
+{
+	if (thr->magic_n != QUE_THR_MAGIC_N) {
+		fprintf(stderr,
+			"que_thr struct appears corrupt; magic n %lu\n",
+			(unsigned long) thr->magic_n);
+
+		mem_analyze_corruption(thr);
+
+		ut_error;
+	}
+
+	if (!thr->is_active) {
+
+		thr->graph->n_active_thrs++;
+
+		trx->n_active_thrs++;
+
+		thr->is_active = TRUE;
+	}
+
+	thr->state = QUE_THR_RUNNING;
+}
+
+/**************************************************************************
+A patch for MySQL used to 'stop' a dummy query thread used in MySQL
+select, when there is no error or lock wait. */
+UNIV_INTERN
+void
+que_thr_stop_for_mysql_no_error(
+/*============================*/
+	que_thr_t*	thr,	/* in: query thread */
+	trx_t*		trx)	/* in: transaction */
+{
+	ut_ad(thr->state == QUE_THR_RUNNING);
+	ut_ad(thr->is_active == TRUE);
+	ut_ad(trx->n_active_thrs == 1);
+	ut_ad(thr->graph->n_active_thrs == 1);
+
+	if (thr->magic_n != QUE_THR_MAGIC_N) {
+		fprintf(stderr,
+			"que_thr struct appears corrupt; magic n %lu\n",
+			(unsigned long) thr->magic_n);
+
+		mem_analyze_corruption(thr);
+
+		ut_error;
+	}
+
+	thr->state = QUE_THR_COMPLETED;
+
+	thr->is_active = FALSE;
+	(thr->graph)->n_active_thrs--;
+
+	trx->n_active_thrs--;
+}
+
+/********************************************************************
+Get the first containing loop node (e.g. while_node_t or for_node_t) for the
+given node, or NULL if the node is not within a loop. */
+UNIV_INTERN
+que_node_t*
+que_node_get_containing_loop_node(
+/*==============================*/
+				/* out: containing loop node, or NULL. */
+	que_node_t*	node)	/* in: node */
+{
+	ut_ad(node);
+
+	for (;;) {
+		ulint	type;
+
+		node = que_node_get_parent(node);
+
+		if (!node) {
+			break;
+		}
+
+		type = que_node_get_type(node);
+
+		if ((type == QUE_NODE_FOR) || (type == QUE_NODE_WHILE)) {
+			break;
+		}
+	}
+
+	return(node);
+}
+
+/**************************************************************************
+Prints info of an SQL query graph node. */
+UNIV_INTERN
+void
+que_node_print_info(
+/*================*/
+	que_node_t*	node)	/* in: query graph node */
+{
+	ulint		type;
+	const char*	str;
+
+	type = que_node_get_type(node);
+
+	if (type == QUE_NODE_SELECT) {
+		str = "SELECT";
+	} else if (type == QUE_NODE_INSERT) {
+		str = "INSERT";
+	} else if (type == QUE_NODE_UPDATE) {
+		str = "UPDATE";
+	} else if (type == QUE_NODE_WHILE) {
+		str = "WHILE";
+	} else if (type == QUE_NODE_ASSIGNMENT) {
+		str = "ASSIGNMENT";
+	} else if (type == QUE_NODE_IF) {
+		str = "IF";
+	} else if (type == QUE_NODE_FETCH) {
+		str = "FETCH";
+	} else if (type == QUE_NODE_OPEN) {
+		str = "OPEN";
+	} else if (type == QUE_NODE_PROC) {
+		str = "STORED PROCEDURE";
+	} else if (type == QUE_NODE_FUNC) {
+		str = "FUNCTION";
+	} else if (type == QUE_NODE_LOCK) {
+		str = "LOCK";
+	} else if (type == QUE_NODE_THR) {
+		str = "QUERY THREAD";
+	} else if (type == QUE_NODE_COMMIT) {
+		str = "COMMIT";
+	} else if (type == QUE_NODE_UNDO) {
+		str = "UNDO ROW";
+	} else if (type == QUE_NODE_PURGE) {
+		str = "PURGE ROW";
+	} else if (type == QUE_NODE_ROLLBACK) {
+		str = "ROLLBACK";
+	} else if (type == QUE_NODE_CREATE_TABLE) {
+		str = "CREATE TABLE";
+	} else if (type == QUE_NODE_CREATE_INDEX) {
+		str = "CREATE INDEX";
+	} else if (type == QUE_NODE_FOR) {
+		str = "FOR LOOP";
+	} else if (type == QUE_NODE_RETURN) {
+		str = "RETURN";
+	} else if (type == QUE_NODE_EXIT) {
+		str = "EXIT";
+	} else {
+		str = "UNKNOWN NODE TYPE";
+	}
+
+	fprintf(stderr, "Node type %lu: %s, address %p\n",
+		(ulong) type, str, (void*) node);
+}
+
+/**************************************************************************
+Performs an execution step on a query thread. */
+UNIV_INLINE
+que_thr_t*
+que_thr_step(
+/*=========*/
+				/* out: query thread to run next: it may
+				differ from the input parameter if, e.g., a
+				subprocedure call is made */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	que_node_t*	node;
+	que_thr_t*	old_thr;
+	trx_t*		trx;
+	ulint		type;
+
+	trx = thr_get_trx(thr);
+
+	ut_ad(thr->state == QUE_THR_RUNNING);
+	ut_a(trx->error_state == DB_SUCCESS);
+
+	thr->resource++;
+
+	node = thr->run_node;
+	type = que_node_get_type(node);
+
+	old_thr = thr;
+
+#ifdef UNIV_DEBUG
+	if (que_trace_on) {
+		fputs("To execute: ", stderr);
+		que_node_print_info(node);
+	}
+#endif
+	if (type & QUE_NODE_CONTROL_STAT) {
+		if ((thr->prev_node != que_node_get_parent(node))
+		    && que_node_get_next(thr->prev_node)) {
+
+			/* The control statements, like WHILE, always pass the
+			control to the next child statement if there is any
+			child left */
+
+			thr->run_node = que_node_get_next(thr->prev_node);
+
+		} else if (type == QUE_NODE_IF) {
+			if_step(thr);
+		} else if (type == QUE_NODE_FOR) {
+			for_step(thr);
+		} else if (type == QUE_NODE_PROC) {
+
+			/* We can access trx->undo_no without reserving
+			trx->undo_mutex, because there cannot be active query
+			threads doing updating or inserting at the moment! */
+
+			if (thr->prev_node == que_node_get_parent(node)) {
+				trx->last_sql_stat_start.least_undo_no
+					= trx->undo_no;
+			}
+
+			proc_step(thr);
+		} else if (type == QUE_NODE_WHILE) {
+			while_step(thr);
+		} else {
+			ut_error;
+		}
+	} else if (type == QUE_NODE_ASSIGNMENT) {
+		assign_step(thr);
+	} else if (type == QUE_NODE_SELECT) {
+		thr = row_sel_step(thr);
+	} else if (type == QUE_NODE_INSERT) {
+		thr = row_ins_step(thr);
+	} else if (type == QUE_NODE_UPDATE) {
+		thr = row_upd_step(thr);
+	} else if (type == QUE_NODE_FETCH) {
+		thr = fetch_step(thr);
+	} else if (type == QUE_NODE_OPEN) {
+		thr = open_step(thr);
+	} else if (type == QUE_NODE_FUNC) {
+		proc_eval_step(thr);
+
+	} else if (type == QUE_NODE_LOCK) {
+
+		ut_error;
+		/*
+		thr = que_lock_step(thr);
+		*/
+	} else if (type == QUE_NODE_THR) {
+		thr = que_thr_node_step(thr);
+	} else if (type == QUE_NODE_COMMIT) {
+		thr = trx_commit_step(thr);
+	} else if (type == QUE_NODE_UNDO) {
+		thr = row_undo_step(thr);
+	} else if (type == QUE_NODE_PURGE) {
+		thr = row_purge_step(thr);
+	} else if (type == QUE_NODE_RETURN) {
+		thr = return_step(thr);
+	} else if (type == QUE_NODE_EXIT) {
+		thr = exit_step(thr);
+	} else if (type == QUE_NODE_ROLLBACK) {
+		thr = trx_rollback_step(thr);
+	} else if (type == QUE_NODE_CREATE_TABLE) {
+		thr = dict_create_table_step(thr);
+	} else if (type == QUE_NODE_CREATE_INDEX) {
+		thr = dict_create_index_step(thr);
+	} else if (type == QUE_NODE_ROW_PRINTF) {
+		thr = row_printf_step(thr);
+	} else {
+		ut_error;
+	}
+
+	if (type == QUE_NODE_EXIT) {
+		old_thr->prev_node = que_node_get_containing_loop_node(node);
+	} else {
+		old_thr->prev_node = node;
+	}
+
+	if (thr) {
+		ut_a(thr_get_trx(thr)->error_state == DB_SUCCESS);
+	}
+
+	return(thr);
+}
+
+/**************************************************************************
+Run a query thread until it finishes or encounters e.g. a lock wait. */
+static
+void
+que_run_threads_low(
+/*================*/
+	que_thr_t*	thr)	/* in: query thread */
+{
+	que_thr_t*	next_thr;
+	ulint		cumul_resource;
+	ulint		loop_count;
+
+	ut_ad(thr->state == QUE_THR_RUNNING);
+	ut_a(thr_get_trx(thr)->error_state == DB_SUCCESS);
+	ut_ad(!mutex_own(&kernel_mutex));
+
+	/* cumul_resource counts how much resources the OS thread (NOT the
+	query thread) has spent in this function */
+
+	loop_count = QUE_MAX_LOOPS_WITHOUT_CHECK;
+	cumul_resource = 0;
+loop:
+	/* Check that there is enough space in the log to accommodate
+	possible log entries by this query step; if the operation can touch
+	more than about 4 pages, checks must be made also within the query
+	step! */
+
+	log_free_check();
+
+	/* Perform the actual query step: note that the query thread
+	may change if, e.g., a subprocedure call is made */
+
+	/*-------------------------*/
+	next_thr = que_thr_step(thr);
+	/*-------------------------*/
+
+	ut_a(!next_thr || (thr_get_trx(next_thr)->error_state == DB_SUCCESS));
+
+	loop_count++;
+
+	if (next_thr != thr) {
+		ut_a(next_thr == NULL);
+
+		/* This can change next_thr to a non-NULL value if there was
+		a lock wait that already completed. */
+		que_thr_dec_refer_count(thr, &next_thr);
+
+		if (next_thr == NULL) {
+
+			return;
+		}
+
+		loop_count = QUE_MAX_LOOPS_WITHOUT_CHECK;
+
+		thr = next_thr;
+	}
+
+	goto loop;
+}
+
+/**************************************************************************
+Run a query thread. Handles lock waits. */
+UNIV_INTERN
+void
+que_run_threads(
+/*============*/
+	que_thr_t*	thr)	/* in: query thread */
+{
+loop:
+	ut_a(thr_get_trx(thr)->error_state == DB_SUCCESS);
+	que_run_threads_low(thr);
+
+	mutex_enter(&kernel_mutex);
+
+	switch (thr->state) {
+
+	case QUE_THR_RUNNING:
+		/* There probably was a lock wait, but it already ended
+		before we came here: continue running thr */
+
+		mutex_exit(&kernel_mutex);
+
+		goto loop;
+
+	case QUE_THR_LOCK_WAIT:
+		mutex_exit(&kernel_mutex);
+
+		/* The ..._mysql_... function works also for InnoDB's
+		internal threads. Let us wait that the lock wait ends. */
+
+		srv_suspend_mysql_thread(thr);
+
+		if (thr_get_trx(thr)->error_state != DB_SUCCESS) {
+			/* thr was chosen as a deadlock victim or there was
+			a lock wait timeout */
+
+			que_thr_dec_refer_count(thr, NULL);
+
+			return;
+		}
+
+		goto loop;
+
+	case QUE_THR_COMPLETED:
+	case QUE_THR_COMMAND_WAIT:
+		/* Do nothing */
+		break;
+
+	default:
+		ut_error;
+	}
+
+	mutex_exit(&kernel_mutex);
+}
+
+/*************************************************************************
+Evaluate the given SQL. */
+UNIV_INTERN
+ulint
+que_eval_sql(
+/*=========*/
+				/* out: error code or DB_SUCCESS */
+	pars_info_t*	info,	/* in: info struct, or NULL */
+	const char*	sql,	/* in: SQL string */
+	ibool		reserve_dict_mutex,
+				/* in: if TRUE, acquire/release
+				dict_sys->mutex around call to pars_sql. */
+	trx_t*		trx)	/* in: trx */
+{
+	que_thr_t*	thr;
+	que_t*		graph;
+
+	ut_a(trx->error_state == DB_SUCCESS);
+
+	if (reserve_dict_mutex) {
+		mutex_enter(&dict_sys->mutex);
+	}
+
+	graph = pars_sql(info, sql);
+
+	if (reserve_dict_mutex) {
+		mutex_exit(&dict_sys->mutex);
+	}
+
+	ut_a(graph);
+
+	graph->trx = trx;
+	trx->graph = NULL;
+
+	graph->fork_type = QUE_FORK_MYSQL_INTERFACE;
+
+	ut_a(thr = que_fork_start_command(graph));
+
+	que_run_threads(thr);
+
+	que_graph_free(graph);
+
+	return(trx->error_state);
+}
diff --git a/storage/xtradb/read/read0read.c b/storage/xtradb/read/read0read.c
new file mode 100644
index 00000000000..e3e5ee5d623
--- /dev/null
+++ b/storage/xtradb/read/read0read.c
@@ -0,0 +1,538 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Cursor read
+
+Created 2/16/1997 Heikki Tuuri
+*******************************************************/
+
+#include "read0read.h"
+
+#ifdef UNIV_NONINL
+#include "read0read.ic"
+#endif
+
+#include "srv0srv.h"
+#include "trx0sys.h"
+
+/*
+-------------------------------------------------------------------------------
+FACT A: Cursor read view on a secondary index sees only committed versions
+-------
+of the records in the secondary index or those versions of rows created
+by transaction which created a cursor before cursor was created even
+if transaction which created the cursor has changed that clustered index page.
+
+PROOF: We must show that read goes always to the clustered index record
+to see that record is visible in the cursor read view. Consider e.g.
+following table and SQL-clauses:
+
+create table t1(a int not null, b int, primary key(a), index(b));
+insert into t1 values (1,1),(2,2);
+commit;
+
+Now consider that we have a cursor for a query
+
+select b from t1 where b >= 1;
+
+This query will use secondary key on the table t1. Now after the first fetch
+on this cursor if we do a update:
+
+update t1 set b = 5 where b = 2;
+
+Now second fetch of the cursor should not see record (2,5) instead it should
+see record (2,2).
+
+We also should show that if we have delete t1 where b = 5; we still
+can see record (2,2).
+
+When we access a secondary key record maximum transaction id is fetched
+from this record and this trx_id is compared to up_limit_id in the view.
+If trx_id in the record is greater or equal than up_limit_id in the view
+cluster record is accessed.  Because trx_id of the creating
+transaction is stored when this view was created to the list of
+trx_ids not seen by this read view previous version of the
+record is requested to be built. This is build using clustered record.
+If the secondary key record is delete  marked it's corresponding
+clustered record can be already be purged only if records
+trx_id < low_limit_no. Purge can't remove any record deleted by a
+transaction which was active when cursor was created. But, we still
+may have a deleted secondary key record but no clustered record. But,
+this is not a problem because this case is handled in
+row_sel_get_clust_rec() function which is called
+whenever we note that this read view does not see trx_id in the
+record. Thus, we see correct version. Q. E. D.
+
+-------------------------------------------------------------------------------
+FACT B: Cursor read view on a clustered index sees only committed versions
+-------
+of the records in the clustered index or those versions of rows created
+by transaction which created a cursor before cursor was created even
+if transaction which created the cursor has changed that clustered index page.
+
+PROOF:  Consider e.g.following table and SQL-clauses:
+
+create table t1(a int not null, b int, primary key(a));
+insert into t1 values (1),(2);
+commit;
+
+Now consider that we have a cursor for a query
+
+select a from t1 where a >= 1;
+
+This query will use clustered key on the table t1. Now after the first fetch
+on this cursor if we do a update:
+
+update t1 set a = 5 where a = 2;
+
+Now second fetch of the cursor should not see record (5) instead it should
+see record (2).
+
+We also should show that if we have execute delete t1 where a = 5; after
+the cursor is opened we still can see record (2).
+
+When accessing clustered record we always check if this read view sees
+trx_id stored to clustered record. By default we don't see any changes
+if record trx_id >= low_limit_id i.e. change was made transaction
+which started after transaction which created the cursor. If row
+was changed by the future transaction a previous version of the
+clustered record is created. Thus we see only committed version in
+this case. We see all changes made by committed transactions i.e.
+record trx_id < up_limit_id. In this case we don't need to do anything,
+we already see correct version of the record. We don't see any changes
+made by active transaction except creating transaction. We have stored
+trx_id of creating transaction to list of trx_ids when this view was
+created. Thus we can easily see if this record was changed by the
+creating transaction. Because we already have clustered record we can
+access roll_ptr. Using this roll_ptr we can fetch undo record.
+We can now check that undo_no of the undo record is less than undo_no of the
+trancaction which created a view when cursor was created. We see this
+clustered record only in case when record undo_no is less than undo_no
+in the view. If this is not true we build based on undo_rec previous
+version of the record. This record is found because purge can't remove
+records accessed by active transaction. Thus we see correct version. Q. E. D.
+-------------------------------------------------------------------------------
+FACT C: Purge does not remove any delete marked row that is visible
+-------
+to cursor view.
+
+TODO: proof this
+
+*/
+
+/*************************************************************************
+Creates a read view object. */
+UNIV_INLINE
+read_view_t*
+read_view_create_low(
+/*=================*/
+				/* out, own: read view struct */
+	ulint		n,	/* in: number of cells in the trx_ids array */
+	mem_heap_t*	heap)	/* in: memory heap from which allocated */
+{
+	read_view_t*	view;
+
+	view = mem_heap_alloc(heap, sizeof(read_view_t));
+
+	view->n_trx_ids = n;
+	view->trx_ids = mem_heap_alloc(heap, n * sizeof(dulint));
+
+	return(view);
+}
+
+/*************************************************************************
+Makes a copy of the oldest existing read view, with the exception that also
+the creating trx of the oldest view is set as not visible in the 'copied'
+view. Opens a new view if no views currently exist. The view must be closed
+with ..._close. This is used in purge. */
+UNIV_INTERN
+read_view_t*
+read_view_oldest_copy_or_open_new(
+/*==============================*/
+					/* out, own: read view struct */
+	dulint		cr_trx_id,	/* in: trx_id of creating
+					transaction, or (0, 0) used in purge*/
+	mem_heap_t*	heap)		/* in: memory heap from which
+					allocated */
+{
+	read_view_t*	old_view;
+	read_view_t*	view_copy;
+	ibool		needs_insert	= TRUE;
+	ulint		insert_done	= 0;
+	ulint		n;
+	ulint		i;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	old_view = UT_LIST_GET_LAST(trx_sys->view_list);
+
+	if (old_view == NULL) {
+
+		return(read_view_open_now(cr_trx_id, heap));
+	}
+
+	n = old_view->n_trx_ids;
+
+	if (!ut_dulint_is_zero(old_view->creator_trx_id)) {
+		n++;
+	} else {
+		needs_insert = FALSE;
+	}
+
+	view_copy = read_view_create_low(n, heap);
+
+	/* Insert the id of the creator in the right place of the descending
+	array of ids, if needs_insert is TRUE: */
+
+	i = 0;
+	while (i < n) {
+		if (needs_insert
+		    && (i >= old_view->n_trx_ids
+			|| ut_dulint_cmp(old_view->creator_trx_id,
+					 read_view_get_nth_trx_id(old_view, i))
+			> 0)) {
+
+			read_view_set_nth_trx_id(view_copy, i,
+						 old_view->creator_trx_id);
+			needs_insert = FALSE;
+			insert_done = 1;
+		} else {
+			read_view_set_nth_trx_id(view_copy, i,
+						 read_view_get_nth_trx_id(
+							 old_view,
+							 i - insert_done));
+		}
+
+		i++;
+	}
+
+	view_copy->creator_trx_id = cr_trx_id;
+
+	view_copy->low_limit_no = old_view->low_limit_no;
+	view_copy->low_limit_id = old_view->low_limit_id;
+
+
+	if (n > 0) {
+		/* The last active transaction has the smallest id: */
+		view_copy->up_limit_id = read_view_get_nth_trx_id(
+			view_copy, n - 1);
+	} else {
+		view_copy->up_limit_id = old_view->up_limit_id;
+	}
+
+	UT_LIST_ADD_LAST(view_list, trx_sys->view_list, view_copy);
+
+	return(view_copy);
+}
+
+/*************************************************************************
+Opens a read view where exactly the transactions serialized before this
+point in time are seen in the view. */
+UNIV_INTERN
+read_view_t*
+read_view_open_now(
+/*===============*/
+					/* out, own: read view struct */
+	dulint		cr_trx_id,	/* in: trx_id of creating
+					transaction, or (0, 0) used in
+					purge */
+	mem_heap_t*	heap)		/* in: memory heap from which
+					allocated */
+{
+	read_view_t*	view;
+	trx_t*		trx;
+	ulint		n;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	view = read_view_create_low(UT_LIST_GET_LEN(trx_sys->trx_list), heap);
+
+	view->creator_trx_id = cr_trx_id;
+	view->type = VIEW_NORMAL;
+	view->undo_no = ut_dulint_zero;
+
+	/* No future transactions should be visible in the view */
+
+	view->low_limit_no = trx_sys->max_trx_id;
+	view->low_limit_id = view->low_limit_no;
+
+	n = 0;
+	trx = UT_LIST_GET_FIRST(trx_sys->trx_list);
+
+	/* No active transaction should be visible, except cr_trx */
+
+	while (trx) {
+		if (ut_dulint_cmp(trx->id, cr_trx_id) != 0
+		    && (trx->conc_state == TRX_ACTIVE
+			|| trx->conc_state == TRX_PREPARED)) {
+
+			read_view_set_nth_trx_id(view, n, trx->id);
+
+			n++;
+
+			/* NOTE that a transaction whose trx number is <
+			trx_sys->max_trx_id can still be active, if it is
+			in the middle of its commit! Note that when a
+			transaction starts, we initialize trx->no to
+			ut_dulint_max. */
+
+			if (ut_dulint_cmp(view->low_limit_no, trx->no) > 0) {
+
+				view->low_limit_no = trx->no;
+			}
+		}
+
+		trx = UT_LIST_GET_NEXT(trx_list, trx);
+	}
+
+	view->n_trx_ids = n;
+
+	if (n > 0) {
+		/* The last active transaction has the smallest id: */
+		view->up_limit_id = read_view_get_nth_trx_id(view, n - 1);
+	} else {
+		view->up_limit_id = view->low_limit_id;
+	}
+
+
+	UT_LIST_ADD_FIRST(view_list, trx_sys->view_list, view);
+
+	return(view);
+}
+
+/*************************************************************************
+Closes a read view. */
+UNIV_INTERN
+void
+read_view_close(
+/*============*/
+	read_view_t*	view)	/* in: read view */
+{
+	ut_ad(mutex_own(&kernel_mutex));
+
+	UT_LIST_REMOVE(view_list, trx_sys->view_list, view);
+}
+
+/*************************************************************************
+Closes a consistent read view for MySQL. This function is called at an SQL
+statement end if the trx isolation level is <= TRX_ISO_READ_COMMITTED. */
+UNIV_INTERN
+void
+read_view_close_for_mysql(
+/*======================*/
+	trx_t*	trx)	/* in: trx which has a read view */
+{
+	ut_a(trx->global_read_view);
+
+	mutex_enter(&kernel_mutex);
+
+	read_view_close(trx->global_read_view);
+
+	mem_heap_empty(trx->global_read_view_heap);
+
+	trx->read_view = NULL;
+	trx->global_read_view = NULL;
+
+	mutex_exit(&kernel_mutex);
+}
+
+/*************************************************************************
+Prints a read view to stderr. */
+UNIV_INTERN
+void
+read_view_print(
+/*============*/
+	read_view_t*	view)	/* in: read view */
+{
+	ulint	n_ids;
+	ulint	i;
+
+	if (view->type == VIEW_HIGH_GRANULARITY) {
+		fprintf(stderr,
+			"High-granularity read view undo_n:o %lu %lu\n",
+			(ulong) ut_dulint_get_high(view->undo_no),
+			(ulong) ut_dulint_get_low(view->undo_no));
+	} else {
+		fprintf(stderr, "Normal read view\n");
+	}
+
+	fprintf(stderr, "Read view low limit trx n:o %lu %lu\n",
+		(ulong) ut_dulint_get_high(view->low_limit_no),
+		(ulong) ut_dulint_get_low(view->low_limit_no));
+
+	fprintf(stderr, "Read view up limit trx id " TRX_ID_FMT "\n",
+		TRX_ID_PREP_PRINTF(view->up_limit_id));
+
+	fprintf(stderr, "Read view low limit trx id " TRX_ID_FMT "\n",
+		TRX_ID_PREP_PRINTF(view->low_limit_id));
+
+	fprintf(stderr, "Read view individually stored trx ids:\n");
+
+	n_ids = view->n_trx_ids;
+
+	for (i = 0; i < n_ids; i++) {
+		fprintf(stderr, "Read view trx id " TRX_ID_FMT "\n",
+			TRX_ID_PREP_PRINTF(
+				read_view_get_nth_trx_id(view, i)));
+	}
+}
+
+/*************************************************************************
+Create a high-granularity consistent cursor view for mysql to be used
+in cursors. In this consistent read view modifications done by the
+creating transaction after the cursor is created or future transactions
+are not visible. */
+UNIV_INTERN
+cursor_view_t*
+read_cursor_view_create_for_mysql(
+/*==============================*/
+	trx_t*	cr_trx)	/* in: trx where cursor view is created */
+{
+	cursor_view_t*	curview;
+	read_view_t*	view;
+	mem_heap_t*	heap;
+	trx_t*		trx;
+	ulint		n;
+
+	ut_a(cr_trx);
+
+	/* Use larger heap than in trx_create when creating a read_view
+	because cursors are quite long. */
+
+	heap = mem_heap_create(512);
+
+	curview = (cursor_view_t*) mem_heap_alloc(heap, sizeof(cursor_view_t));
+	curview->heap = heap;
+
+	/* Drop cursor tables from consideration when evaluating the need of
+	auto-commit */
+	curview->n_mysql_tables_in_use = cr_trx->n_mysql_tables_in_use;
+	cr_trx->n_mysql_tables_in_use = 0;
+
+	mutex_enter(&kernel_mutex);
+
+	curview->read_view = read_view_create_low(
+		UT_LIST_GET_LEN(trx_sys->trx_list), curview->heap);
+
+	view = curview->read_view;
+	view->creator_trx_id = cr_trx->id;
+	view->type = VIEW_HIGH_GRANULARITY;
+	view->undo_no = cr_trx->undo_no;
+
+	/* No future transactions should be visible in the view */
+
+	view->low_limit_no = trx_sys->max_trx_id;
+	view->low_limit_id = view->low_limit_no;
+
+	n = 0;
+	trx = UT_LIST_GET_FIRST(trx_sys->trx_list);
+
+	/* No active transaction should be visible */
+
+	while (trx) {
+
+		if (trx->conc_state == TRX_ACTIVE
+		    || trx->conc_state == TRX_PREPARED) {
+
+			read_view_set_nth_trx_id(view, n, trx->id);
+
+			n++;
+
+			/* NOTE that a transaction whose trx number is <
+			trx_sys->max_trx_id can still be active, if it is
+			in the middle of its commit! Note that when a
+			transaction starts, we initialize trx->no to
+			ut_dulint_max. */
+
+			if (ut_dulint_cmp(view->low_limit_no, trx->no) > 0) {
+
+				view->low_limit_no = trx->no;
+			}
+		}
+
+		trx = UT_LIST_GET_NEXT(trx_list, trx);
+	}
+
+	view->n_trx_ids = n;
+
+	if (n > 0) {
+		/* The last active transaction has the smallest id: */
+		view->up_limit_id = read_view_get_nth_trx_id(view, n - 1);
+	} else {
+		view->up_limit_id = view->low_limit_id;
+	}
+
+	UT_LIST_ADD_FIRST(view_list, trx_sys->view_list, view);
+
+	mutex_exit(&kernel_mutex);
+
+	return(curview);
+}
+
+/*************************************************************************
+Close a given consistent cursor view for mysql and restore global read view
+back to a transaction read view. */
+UNIV_INTERN
+void
+read_cursor_view_close_for_mysql(
+/*=============================*/
+	trx_t*		trx,	/* in: trx */
+	cursor_view_t*	curview)/* in: cursor view to be closed */
+{
+	ut_a(curview);
+	ut_a(curview->read_view);
+	ut_a(curview->heap);
+
+	/* Add cursor's tables to the global count of active tables that
+	belong to this transaction */
+	trx->n_mysql_tables_in_use += curview->n_mysql_tables_in_use;
+
+	mutex_enter(&kernel_mutex);
+
+	read_view_close(curview->read_view);
+	trx->read_view = trx->global_read_view;
+
+	mutex_exit(&kernel_mutex);
+
+	mem_heap_free(curview->heap);
+}
+
+/*************************************************************************
+This function sets a given consistent cursor view to a transaction
+read view if given consistent cursor view is not NULL. Otherwise, function
+restores a global read view to a transaction read view. */
+UNIV_INTERN
+void
+read_cursor_set_for_mysql(
+/*======================*/
+	trx_t*		trx,	/* in: transaction where cursor is set */
+	cursor_view_t*	curview)/* in: consistent cursor view to be set */
+{
+	ut_a(trx);
+
+	mutex_enter(&kernel_mutex);
+
+	if (UNIV_LIKELY(curview != NULL)) {
+		trx->read_view = curview->read_view;
+	} else {
+		trx->read_view = trx->global_read_view;
+	}
+
+	mutex_exit(&kernel_mutex);
+}
diff --git a/storage/xtradb/rem/rem0cmp.c b/storage/xtradb/rem/rem0cmp.c
new file mode 100644
index 00000000000..39fcb6f19dd
--- /dev/null
+++ b/storage/xtradb/rem/rem0cmp.c
@@ -0,0 +1,1234 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/***********************************************************************
+Comparison services for records
+
+Created 7/1/1994 Heikki Tuuri
+************************************************************************/
+
+#include "rem0cmp.h"
+
+#ifdef UNIV_NONINL
+#include "rem0cmp.ic"
+#endif
+
+#include "srv0srv.h"
+
+/*		ALPHABETICAL ORDER
+		==================
+
+The records are put into alphabetical order in the following
+way: let F be the first field where two records disagree.
+If there is a character in some position n where the the
+records disagree, the order is determined by comparison of
+the characters at position n, possibly after
+collating transformation. If there is no such character,
+but the corresponding fields have different lengths, then
+if the data type of the fields is paddable,
+shorter field is padded with a padding character. If the
+data type is not paddable, longer field is considered greater.
+Finally, the SQL null is bigger than any other value.
+
+At the present, the comparison functions return 0 in the case,
+where two records disagree only in the way that one
+has more fields than the other. */
+
+#ifdef UNIV_DEBUG
+/*****************************************************************
+Used in debug checking of cmp_dtuple_... .
+This function is used to compare a data tuple to a physical record. If
+dtuple has n fields then rec must have either m >= n fields, or it must
+differ from dtuple in some of the m fields rec has. */
+static
+int
+cmp_debug_dtuple_rec_with_match(
+/*============================*/
+				/* out: 1, 0, -1, if dtuple is greater, equal,
+				less than rec, respectively, when only the
+				common first fields are compared */
+	const dtuple_t*	dtuple,	/* in: data tuple */
+	const rec_t*	rec,	/* in: physical record which differs from
+				dtuple in some of the common fields, or which
+				has an equal number or more fields than
+				dtuple */
+	const ulint*	offsets,/* in: array returned by rec_get_offsets() */
+	ulint*		matched_fields);/* in/out: number of already
+				completely  matched fields; when function
+				returns, contains the value for current
+				comparison */
+#endif /* UNIV_DEBUG */
+#ifndef UNIV_HOTBACKUP
+/*****************************************************************
+This function is used to compare two data fields for which the data type
+is such that we must use MySQL code to compare them. The prototype here
+must be a copy of the the one in ha_innobase.cc! */
+extern
+int
+innobase_mysql_cmp(
+/*===============*/
+					/* out: 1, 0, -1, if a is greater,
+					equal, less than b, respectively */
+	int		mysql_type,	/* in: MySQL type */
+	uint		charset_number,	/* in: number of the charset */
+	const unsigned char* a,		/* in: data field */
+	unsigned int	a_length,	/* in: data field length,
+					not UNIV_SQL_NULL */
+	const unsigned char* b,		/* in: data field */
+	unsigned int	b_length);	/* in: data field length,
+					not UNIV_SQL_NULL */
+#endif /* !UNIV_HOTBACKUP */
+/*************************************************************************
+Transforms the character code so that it is ordered appropriately for the
+language. This is only used for the latin1 char set. MySQL does the
+comparisons for other char sets. */
+UNIV_INLINE
+ulint
+cmp_collate(
+/*========*/
+			/* out: collation order position */
+	ulint	code)	/* in: code of a character stored in database record */
+{
+	return((ulint) srv_latin1_ordering[code]);
+}
+
+/*****************************************************************
+Returns TRUE if two columns are equal for comparison purposes. */
+UNIV_INTERN
+ibool
+cmp_cols_are_equal(
+/*===============*/
+					/* out: TRUE if the columns are
+					considered equal in comparisons */
+	const dict_col_t*	col1,	/* in: column 1 */
+	const dict_col_t*	col2,	/* in: column 2 */
+	ibool			check_charsets)
+					/* in: whether to check charsets */
+{
+	if (dtype_is_non_binary_string_type(col1->mtype, col1->prtype)
+	    && dtype_is_non_binary_string_type(col2->mtype, col2->prtype)) {
+
+		/* Both are non-binary string types: they can be compared if
+		and only if the charset-collation is the same */
+
+		if (check_charsets) {
+			return(dtype_get_charset_coll(col1->prtype)
+			       == dtype_get_charset_coll(col2->prtype));
+		} else {
+			return(TRUE);
+		}
+	}
+
+	if (dtype_is_binary_string_type(col1->mtype, col1->prtype)
+	    && dtype_is_binary_string_type(col2->mtype, col2->prtype)) {
+
+		/* Both are binary string types: they can be compared */
+
+		return(TRUE);
+	}
+
+	if (col1->mtype != col2->mtype) {
+
+		return(FALSE);
+	}
+
+	if (col1->mtype == DATA_INT
+	    && (col1->prtype & DATA_UNSIGNED)
+	    != (col2->prtype & DATA_UNSIGNED)) {
+
+		/* The storage format of an unsigned integer is different
+		from a signed integer: in a signed integer we OR
+		0x8000... to the value of positive integers. */
+
+		return(FALSE);
+	}
+
+	return(col1->mtype != DATA_INT || col1->len == col2->len);
+}
+
+#ifndef UNIV_HOTBACKUP
+/*****************************************************************
+Innobase uses this function to compare two data fields for which the data type
+is such that we must compare whole fields or call MySQL to do the comparison */
+static
+int
+cmp_whole_field(
+/*============*/
+					/* out: 1, 0, -1, if a is greater,
+					equal, less than b, respectively */
+	ulint		mtype,		/* in: main type */
+	ulint		prtype,		/* in: precise type */
+	const byte*	a,		/* in: data field */
+	unsigned int	a_length,	/* in: data field length,
+					not UNIV_SQL_NULL */
+	const byte*	b,		/* in: data field */
+	unsigned int	b_length)	/* in: data field length,
+					not UNIV_SQL_NULL */
+{
+	float		f_1;
+	float		f_2;
+	double		d_1;
+	double		d_2;
+	int		swap_flag	= 1;
+
+	switch (mtype) {
+
+	case DATA_DECIMAL:
+		/* Remove preceding spaces */
+		for (; a_length && *a == ' '; a++, a_length--);
+		for (; b_length && *b == ' '; b++, b_length--);
+
+		if (*a == '-') {
+			if (*b != '-') {
+				return(-1);
+			}
+
+			a++; b++;
+			a_length--;
+			b_length--;
+
+			swap_flag = -1;
+
+		} else if (*b == '-') {
+
+			return(1);
+		}
+
+		while (a_length > 0 && (*a == '+' || *a == '0')) {
+			a++; a_length--;
+		}
+
+		while (b_length > 0 && (*b == '+' || *b == '0')) {
+			b++; b_length--;
+		}
+
+		if (a_length != b_length) {
+			if (a_length < b_length) {
+				return(-swap_flag);
+			}
+
+			return(swap_flag);
+		}
+
+		while (a_length > 0 && *a == *b) {
+
+			a++; b++; a_length--;
+		}
+
+		if (a_length == 0) {
+
+			return(0);
+		}
+
+		if (*a > *b) {
+			return(swap_flag);
+		}
+
+		return(-swap_flag);
+	case DATA_DOUBLE:
+		d_1 = mach_double_read(a);
+		d_2 = mach_double_read(b);
+
+		if (d_1 > d_2) {
+			return(1);
+		} else if (d_2 > d_1) {
+			return(-1);
+		}
+
+		return(0);
+
+	case DATA_FLOAT:
+		f_1 = mach_float_read(a);
+		f_2 = mach_float_read(b);
+
+		if (f_1 > f_2) {
+			return(1);
+		} else if (f_2 > f_1) {
+			return(-1);
+		}
+
+		return(0);
+	case DATA_BLOB:
+		if (prtype & DATA_BINARY_TYPE) {
+
+			ut_print_timestamp(stderr);
+			fprintf(stderr,
+				"  InnoDB: Error: comparing a binary BLOB"
+				" with a character set sensitive\n"
+				"InnoDB: comparison!\n");
+		}
+		/* fall through */
+	case DATA_VARMYSQL:
+	case DATA_MYSQL:
+		return(innobase_mysql_cmp(
+			       (int)(prtype & DATA_MYSQL_TYPE_MASK),
+			       (uint)dtype_get_charset_coll(prtype),
+			       a, a_length, b, b_length));
+	default:
+		fprintf(stderr,
+			"InnoDB: unknown type number %lu\n",
+			(ulong) mtype);
+		ut_error;
+	}
+
+	return(0);
+}
+#endif /* !UNIV_HOTBACKUP */
+
+/*****************************************************************
+This function is used to compare two data fields for which we know the
+data type. */
+UNIV_INTERN
+int
+cmp_data_data_slow(
+/*===============*/
+				/* out: 1, 0, -1, if data1 is greater, equal,
+				less than data2, respectively */
+	ulint		mtype,	/* in: main type */
+	ulint		prtype,	/* in: precise type */
+	const byte*	data1,	/* in: data field (== a pointer to a memory
+				buffer) */
+	ulint		len1,	/* in: data field length or UNIV_SQL_NULL */
+	const byte*	data2,	/* in: data field (== a pointer to a memory
+				buffer) */
+	ulint		len2)	/* in: data field length or UNIV_SQL_NULL */
+{
+#ifndef UNIV_HOTBACKUP
+	ulint	data1_byte;
+	ulint	data2_byte;
+	ulint	cur_bytes;
+
+	if (len1 == UNIV_SQL_NULL || len2 == UNIV_SQL_NULL) {
+
+		if (len1 == len2) {
+
+			return(0);
+		}
+
+		if (len1 == UNIV_SQL_NULL) {
+			/* We define the SQL null to be the smallest possible
+			value of a field in the alphabetical order */
+
+			return(-1);
+		}
+
+		return(1);
+	}
+
+	if (mtype >= DATA_FLOAT
+	    || (mtype == DATA_BLOB
+		&& 0 == (prtype & DATA_BINARY_TYPE)
+		&& dtype_get_charset_coll(prtype)
+		!= DATA_MYSQL_LATIN1_SWEDISH_CHARSET_COLL)) {
+
+		return(cmp_whole_field(mtype, prtype,
+				       data1, (unsigned) len1,
+				       data2, (unsigned) len2));
+	}
+
+	/* Compare then the fields */
+
+	cur_bytes = 0;
+
+	for (;;) {
+		if (len1 <= cur_bytes) {
+			if (len2 <= cur_bytes) {
+
+				return(0);
+			}
+
+			data1_byte = dtype_get_pad_char(mtype, prtype);
+
+			if (data1_byte == ULINT_UNDEFINED) {
+
+				return(-1);
+			}
+		} else {
+			data1_byte = *data1;
+		}
+
+		if (len2 <= cur_bytes) {
+			data2_byte = dtype_get_pad_char(mtype, prtype);
+
+			if (data2_byte == ULINT_UNDEFINED) {
+
+				return(1);
+			}
+		} else {
+			data2_byte = *data2;
+		}
+
+		if (data1_byte == data2_byte) {
+			/* If the bytes are equal, they will remain such even
+			after the collation transformation below */
+
+			goto next_byte;
+		}
+
+		if (mtype <= DATA_CHAR
+		    || (mtype == DATA_BLOB
+			&& 0 == (prtype & DATA_BINARY_TYPE))) {
+
+			data1_byte = cmp_collate(data1_byte);
+			data2_byte = cmp_collate(data2_byte);
+		}
+
+		if (data1_byte > data2_byte) {
+
+			return(1);
+		} else if (data1_byte < data2_byte) {
+
+			return(-1);
+		}
+next_byte:
+		/* Next byte */
+		cur_bytes++;
+		data1++;
+		data2++;
+	}
+#else /* !UNIV_HOTBACKUP */
+	/* This function depends on MySQL code that is not included in
+	InnoDB Hot Backup builds.  Besides, this function should never
+	be called in InnoDB Hot Backup. */
+	ut_error;
+#endif /* !UNIV_HOTBACKUP */
+
+	return(0);		/* Not reached */
+}
+
+/*****************************************************************
+This function is used to compare a data tuple to a physical record.
+Only dtuple->n_fields_cmp first fields are taken into account for
+the the data tuple! If we denote by n = n_fields_cmp, then rec must
+have either m >= n fields, or it must differ from dtuple in some of
+the m fields rec has. If rec has an externally stored field we do not
+compare it but return with value 0 if such a comparison should be
+made. */
+UNIV_INTERN
+int
+cmp_dtuple_rec_with_match(
+/*======================*/
+				/* out: 1, 0, -1, if dtuple is greater, equal,
+				less than rec, respectively, when only the
+				common first fields are compared, or
+				until the first externally stored field in
+				rec */
+	const dtuple_t*	dtuple,	/* in: data tuple */
+	const rec_t*	rec,	/* in: physical record which differs from
+				dtuple in some of the common fields, or which
+				has an equal number or more fields than
+				dtuple */
+	const ulint*	offsets,/* in: array returned by rec_get_offsets() */
+	ulint*		matched_fields, /* in/out: number of already completely
+				matched fields; when function returns,
+				contains the value for current comparison */
+	ulint*		matched_bytes) /* in/out: number of already matched
+				bytes within the first field not completely
+				matched; when function returns, contains the
+				value for current comparison */
+{
+#ifndef UNIV_HOTBACKUP
+	const dfield_t*	dtuple_field;	/* current field in logical record */
+	ulint		dtuple_f_len;	/* the length of the current field
+					in the logical record */
+	const byte*	dtuple_b_ptr;	/* pointer to the current byte in
+					logical field data */
+	ulint		dtuple_byte;	/* value of current byte to be compared
+					in dtuple*/
+	ulint		rec_f_len;	/* length of current field in rec */
+	const byte*	rec_b_ptr;	/* pointer to the current byte in
+					rec field */
+	ulint		rec_byte;	/* value of current byte to be
+					compared in rec */
+	ulint		cur_field;	/* current field number */
+	ulint		cur_bytes;	/* number of already matched bytes
+					in current field */
+	int		ret = 3333;	/* return value */
+
+	ut_ad(dtuple && rec && matched_fields && matched_bytes);
+	ut_ad(dtuple_check_typed(dtuple));
+	ut_ad(rec_offs_validate(rec, NULL, offsets));
+
+	cur_field = *matched_fields;
+	cur_bytes = *matched_bytes;
+
+	ut_ad(cur_field <= dtuple_get_n_fields_cmp(dtuple));
+	ut_ad(cur_field <= rec_offs_n_fields(offsets));
+
+	if (cur_bytes == 0 && cur_field == 0) {
+		ulint	rec_info = rec_get_info_bits(rec,
+						     rec_offs_comp(offsets));
+		ulint	tup_info = dtuple_get_info_bits(dtuple);
+
+		if (UNIV_UNLIKELY(rec_info & REC_INFO_MIN_REC_FLAG)) {
+			ret = !(tup_info & REC_INFO_MIN_REC_FLAG);
+			goto order_resolved;
+		} else if (UNIV_UNLIKELY(tup_info & REC_INFO_MIN_REC_FLAG)) {
+			ret = -1;
+			goto order_resolved;
+		}
+	}
+
+	/* Match fields in a loop; stop if we run out of fields in dtuple
+	or find an externally stored field */
+
+	while (cur_field < dtuple_get_n_fields_cmp(dtuple)) {
+
+		ulint	mtype;
+		ulint	prtype;
+
+		dtuple_field = dtuple_get_nth_field(dtuple, cur_field);
+		{
+			const dtype_t*	type
+				= dfield_get_type(dtuple_field);
+
+			mtype = type->mtype;
+			prtype = type->prtype;
+		}
+
+		dtuple_f_len = dfield_get_len(dtuple_field);
+
+		rec_b_ptr = rec_get_nth_field(rec, offsets,
+					      cur_field, &rec_f_len);
+
+		/* If we have matched yet 0 bytes, it may be that one or
+		both the fields are SQL null, or the record or dtuple may be
+		the predefined minimum record, or the field is externally
+		stored */
+
+		if (UNIV_LIKELY(cur_bytes == 0)) {
+			if (rec_offs_nth_extern(offsets, cur_field)) {
+				/* We do not compare to an externally
+				stored field */
+
+				ret = 0;
+
+				goto order_resolved;
+			}
+
+			if (dtuple_f_len == UNIV_SQL_NULL) {
+				if (rec_f_len == UNIV_SQL_NULL) {
+
+					goto next_field;
+				}
+
+				ret = -1;
+				goto order_resolved;
+			} else if (rec_f_len == UNIV_SQL_NULL) {
+				/* We define the SQL null to be the
+				smallest possible value of a field
+				in the alphabetical order */
+
+				ret = 1;
+				goto order_resolved;
+			}
+		}
+
+		if (mtype >= DATA_FLOAT
+		    || (mtype == DATA_BLOB
+			&& 0 == (prtype & DATA_BINARY_TYPE)
+			&& dtype_get_charset_coll(prtype)
+			!= DATA_MYSQL_LATIN1_SWEDISH_CHARSET_COLL)) {
+
+			ret = cmp_whole_field(mtype, prtype,
+					      dfield_get_data(dtuple_field),
+					      (unsigned) dtuple_f_len,
+					      rec_b_ptr, (unsigned) rec_f_len);
+
+			if (ret != 0) {
+				cur_bytes = 0;
+
+				goto order_resolved;
+			} else {
+				goto next_field;
+			}
+		}
+
+		/* Set the pointers at the current byte */
+
+		rec_b_ptr = rec_b_ptr + cur_bytes;
+		dtuple_b_ptr = (byte*)dfield_get_data(dtuple_field)
+			+ cur_bytes;
+		/* Compare then the fields */
+
+		for (;;) {
+			if (UNIV_UNLIKELY(rec_f_len <= cur_bytes)) {
+				if (dtuple_f_len <= cur_bytes) {
+
+					goto next_field;
+				}
+
+				rec_byte = dtype_get_pad_char(mtype, prtype);
+
+				if (rec_byte == ULINT_UNDEFINED) {
+					ret = 1;
+
+					goto order_resolved;
+				}
+			} else {
+				rec_byte = *rec_b_ptr;
+			}
+
+			if (UNIV_UNLIKELY(dtuple_f_len <= cur_bytes)) {
+				dtuple_byte = dtype_get_pad_char(mtype,
+								 prtype);
+
+				if (dtuple_byte == ULINT_UNDEFINED) {
+					ret = -1;
+
+					goto order_resolved;
+				}
+			} else {
+				dtuple_byte = *dtuple_b_ptr;
+			}
+
+			if (dtuple_byte == rec_byte) {
+				/* If the bytes are equal, they will
+				remain such even after the collation
+				transformation below */
+
+				goto next_byte;
+			}
+
+			if (mtype <= DATA_CHAR
+			    || (mtype == DATA_BLOB
+				&& !(prtype & DATA_BINARY_TYPE))) {
+
+				rec_byte = cmp_collate(rec_byte);
+				dtuple_byte = cmp_collate(dtuple_byte);
+			}
+
+			ret = (int) (dtuple_byte - rec_byte);
+			if (UNIV_LIKELY(ret)) {
+				if (ret < 0) {
+					ret = -1;
+					goto order_resolved;
+				} else {
+					ret = 1;
+					goto order_resolved;
+				}
+			}
+next_byte:
+			/* Next byte */
+			cur_bytes++;
+			rec_b_ptr++;
+			dtuple_b_ptr++;
+		}
+
+next_field:
+		cur_field++;
+		cur_bytes = 0;
+	}
+
+	ut_ad(cur_bytes == 0);
+
+	ret = 0;	/* If we ran out of fields, dtuple was equal to rec
+			up to the common fields */
+order_resolved:
+	ut_ad((ret >= - 1) && (ret <= 1));
+	ut_ad(ret == cmp_debug_dtuple_rec_with_match(dtuple, rec, offsets,
+						     matched_fields));
+	ut_ad(*matched_fields == cur_field); /* In the debug version, the
+					     above cmp_debug_... sets
+					     *matched_fields to a value */
+	*matched_fields = cur_field;
+	*matched_bytes = cur_bytes;
+
+	return(ret);
+#else /* !UNIV_HOTBACKUP */
+	/* This function depends on MySQL code that is not included in
+	InnoDB Hot Backup builds.  Besides, this function should never
+	be called in InnoDB Hot Backup. */
+	ut_error;
+	return(0);
+#endif /* !UNIV_HOTBACKUP */
+}
+
+/******************************************************************
+Compares a data tuple to a physical record. */
+UNIV_INTERN
+int
+cmp_dtuple_rec(
+/*===========*/
+				/* out: 1, 0, -1, if dtuple is greater, equal,
+				less than rec, respectively; see the comments
+				for cmp_dtuple_rec_with_match */
+	const dtuple_t*	dtuple,	/* in: data tuple */
+	const rec_t*	rec,	/* in: physical record */
+	const ulint*	offsets)/* in: array returned by rec_get_offsets() */
+{
+	ulint	matched_fields	= 0;
+	ulint	matched_bytes	= 0;
+
+	ut_ad(rec_offs_validate(rec, NULL, offsets));
+	return(cmp_dtuple_rec_with_match(dtuple, rec, offsets,
+					 &matched_fields, &matched_bytes));
+}
+
+/******************************************************************
+Checks if a dtuple is a prefix of a record. The last field in dtuple
+is allowed to be a prefix of the corresponding field in the record. */
+UNIV_INTERN
+ibool
+cmp_dtuple_is_prefix_of_rec(
+/*========================*/
+				/* out: TRUE if prefix */
+	const dtuple_t*	dtuple,	/* in: data tuple */
+	const rec_t*	rec,	/* in: physical record */
+	const ulint*	offsets)/* in: array returned by rec_get_offsets() */
+{
+	ulint	n_fields;
+	ulint	matched_fields	= 0;
+	ulint	matched_bytes	= 0;
+
+	ut_ad(rec_offs_validate(rec, NULL, offsets));
+	n_fields = dtuple_get_n_fields(dtuple);
+
+	if (n_fields > rec_offs_n_fields(offsets)) {
+
+		return(FALSE);
+	}
+
+	cmp_dtuple_rec_with_match(dtuple, rec, offsets,
+				  &matched_fields, &matched_bytes);
+	if (matched_fields == n_fields) {
+
+		return(TRUE);
+	}
+
+	if (matched_fields == n_fields - 1
+	    && matched_bytes == dfield_get_len(
+		    dtuple_get_nth_field(dtuple, n_fields - 1))) {
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
+
+#ifndef UNIV_HOTBACKUP
+/*****************************************************************
+Compare two physical records that contain the same number of columns,
+none of which are stored externally. */
+UNIV_INTERN
+int
+cmp_rec_rec_simple(
+/*===============*/
+					/* out: 1, 0 , -1 if rec1 is greater,
+					equal, less, respectively, than rec2 */
+	const rec_t*		rec1,	/* in: physical record */
+	const rec_t*		rec2,	/* in: physical record */
+	const ulint*		offsets1,/* in: rec_get_offsets(rec1, index) */
+	const ulint*		offsets2,/* in: rec_get_offsets(rec2, index) */
+	const dict_index_t*	index)	/* in: data dictionary index */
+{
+	ulint		rec1_f_len;	/* length of current field in rec1 */
+	const byte*	rec1_b_ptr;	/* pointer to the current byte
+					in rec1 field */
+	ulint		rec1_byte;	/* value of current byte to be
+					compared in rec1 */
+	ulint		rec2_f_len;	/* length of current field in rec2 */
+	const byte*	rec2_b_ptr;	/* pointer to the current byte
+					in rec2 field */
+	ulint		rec2_byte;	/* value of current byte to be
+					compared in rec2 */
+	ulint		cur_field;	/* current field number */
+	ulint		n_uniq;
+
+	n_uniq = dict_index_get_n_unique(index);
+	ut_ad(rec_offs_n_fields(offsets1) >= n_uniq);
+	ut_ad(rec_offs_n_fields(offsets2) >= n_uniq);
+
+	ut_ad(rec_offs_comp(offsets1) == rec_offs_comp(offsets2));
+
+	for (cur_field = 0; cur_field < n_uniq; cur_field++) {
+
+		ulint	cur_bytes;
+		ulint	mtype;
+		ulint	prtype;
+
+		{
+			const dict_col_t*	col
+				= dict_index_get_nth_col(index, cur_field);
+
+			mtype = col->mtype;
+			prtype = col->prtype;
+		}
+
+		ut_ad(!rec_offs_nth_extern(offsets1, cur_field));
+		ut_ad(!rec_offs_nth_extern(offsets2, cur_field));
+
+		rec1_b_ptr = rec_get_nth_field(rec1, offsets1,
+					       cur_field, &rec1_f_len);
+		rec2_b_ptr = rec_get_nth_field(rec2, offsets2,
+					       cur_field, &rec2_f_len);
+
+		if (rec1_f_len == UNIV_SQL_NULL
+		    || rec2_f_len == UNIV_SQL_NULL) {
+
+			if (rec1_f_len == rec2_f_len) {
+
+				goto next_field;
+
+			} else if (rec2_f_len == UNIV_SQL_NULL) {
+
+				/* We define the SQL null to be the
+				smallest possible value of a field
+				in the alphabetical order */
+
+				return(1);
+			} else {
+				return(-1);
+			}
+		}
+
+		if (mtype >= DATA_FLOAT
+		    || (mtype == DATA_BLOB
+			&& 0 == (prtype & DATA_BINARY_TYPE)
+			&& dtype_get_charset_coll(prtype)
+			!= DATA_MYSQL_LATIN1_SWEDISH_CHARSET_COLL)) {
+			int ret = cmp_whole_field(mtype, prtype,
+						  rec1_b_ptr,
+						  (unsigned) rec1_f_len,
+						  rec2_b_ptr,
+						  (unsigned) rec2_f_len);
+			if (ret) {
+				return(ret);
+			}
+
+			goto next_field;
+		}
+
+		/* Compare the fields */
+		for (cur_bytes = 0;; cur_bytes++, rec1_b_ptr++, rec2_b_ptr++) {
+			if (rec2_f_len <= cur_bytes) {
+
+				if (rec1_f_len <= cur_bytes) {
+
+					goto next_field;
+				}
+
+				rec2_byte = dtype_get_pad_char(mtype, prtype);
+
+				if (rec2_byte == ULINT_UNDEFINED) {
+					return(1);
+				}
+			} else {
+				rec2_byte = *rec2_b_ptr;
+			}
+
+			if (rec1_f_len <= cur_bytes) {
+				rec1_byte = dtype_get_pad_char(mtype, prtype);
+
+				if (rec1_byte == ULINT_UNDEFINED) {
+					return(-1);
+				}
+			} else {
+				rec1_byte = *rec1_b_ptr;
+			}
+
+			if (rec1_byte == rec2_byte) {
+				/* If the bytes are equal, they will remain
+				such even after the collation transformation
+				below */
+
+				continue;
+			}
+
+			if (mtype <= DATA_CHAR
+			    || (mtype == DATA_BLOB
+				&& !(prtype & DATA_BINARY_TYPE))) {
+
+				rec1_byte = cmp_collate(rec1_byte);
+				rec2_byte = cmp_collate(rec2_byte);
+			}
+
+			if (rec1_byte < rec2_byte) {
+				return(-1);
+			} else if (rec1_byte > rec2_byte) {
+				return(1);
+			}
+		}
+next_field:
+		continue;
+	}
+
+	/* If we ran out of fields, rec1 was equal to rec2. */
+	return(0);
+}
+#endif /* !UNIV_HOTBACKUP */
+
+/*****************************************************************
+This function is used to compare two physical records. Only the common
+first fields are compared, and if an externally stored field is
+encountered, then 0 is returned. */
+UNIV_INTERN
+int
+cmp_rec_rec_with_match(
+/*===================*/
+				/* out: 1, 0 , -1 if rec1 is greater, equal,
+				less, respectively, than rec2; only the common
+				first fields are compared */
+	const rec_t*	rec1,	/* in: physical record */
+	const rec_t*	rec2,	/* in: physical record */
+	const ulint*	offsets1,/* in: rec_get_offsets(rec1, index) */
+	const ulint*	offsets2,/* in: rec_get_offsets(rec2, index) */
+	dict_index_t*	index,	/* in: data dictionary index */
+	ulint*		matched_fields, /* in/out: number of already completely
+				matched fields; when the function returns,
+				contains the value the for current
+				comparison */
+	ulint*		matched_bytes) /* in/out: number of already matched
+				bytes within the first field not completely
+				matched; when the function returns, contains
+				the value for the current comparison */
+{
+#ifndef UNIV_HOTBACKUP
+	ulint		rec1_n_fields;	/* the number of fields in rec */
+	ulint		rec1_f_len;	/* length of current field in rec */
+	const byte*	rec1_b_ptr;	/* pointer to the current byte
+					in rec field */
+	ulint		rec1_byte;	/* value of current byte to be
+					compared in rec */
+	ulint		rec2_n_fields;	/* the number of fields in rec */
+	ulint		rec2_f_len;	/* length of current field in rec */
+	const byte*	rec2_b_ptr;	/* pointer to the current byte
+					in rec field */
+	ulint		rec2_byte;	/* value of current byte to be
+					compared in rec */
+	ulint		cur_field;	/* current field number */
+	ulint		cur_bytes;	/* number of already matched
+					bytes in current field */
+	int		ret = 0;	/* return value */
+	ulint		comp;
+
+	ut_ad(rec1 && rec2 && index);
+	ut_ad(rec_offs_validate(rec1, index, offsets1));
+	ut_ad(rec_offs_validate(rec2, index, offsets2));
+	ut_ad(rec_offs_comp(offsets1) == rec_offs_comp(offsets2));
+
+	comp = rec_offs_comp(offsets1);
+	rec1_n_fields = rec_offs_n_fields(offsets1);
+	rec2_n_fields = rec_offs_n_fields(offsets2);
+
+	cur_field = *matched_fields;
+	cur_bytes = *matched_bytes;
+
+	/* Match fields in a loop */
+
+	while ((cur_field < rec1_n_fields) && (cur_field < rec2_n_fields)) {
+
+		ulint	mtype;
+		ulint	prtype;
+
+		if (UNIV_UNLIKELY(index->type & DICT_UNIVERSAL)) {
+			/* This is for the insert buffer B-tree. */
+			mtype = DATA_BINARY;
+			prtype = 0;
+		} else {
+			const dict_col_t*	col
+				= dict_index_get_nth_col(index, cur_field);
+
+			mtype = col->mtype;
+			prtype = col->prtype;
+		}
+
+		rec1_b_ptr = rec_get_nth_field(rec1, offsets1,
+					       cur_field, &rec1_f_len);
+		rec2_b_ptr = rec_get_nth_field(rec2, offsets2,
+					       cur_field, &rec2_f_len);
+
+		if (cur_bytes == 0) {
+			if (cur_field == 0) {
+				/* Test if rec is the predefined minimum
+				record */
+				if (UNIV_UNLIKELY(rec_get_info_bits(rec1, comp)
+						  & REC_INFO_MIN_REC_FLAG)) {
+
+					if (!(rec_get_info_bits(rec2, comp)
+					      & REC_INFO_MIN_REC_FLAG)) {
+						ret = -1;
+					}
+
+					goto order_resolved;
+
+				} else if (UNIV_UNLIKELY
+					   (rec_get_info_bits(rec2, comp)
+					    & REC_INFO_MIN_REC_FLAG)) {
+
+					ret = 1;
+
+					goto order_resolved;
+				}
+			}
+
+			if (rec_offs_nth_extern(offsets1, cur_field)
+			    || rec_offs_nth_extern(offsets2, cur_field)) {
+				/* We do not compare to an externally
+				stored field */
+
+				goto order_resolved;
+			}
+
+			if (rec1_f_len == UNIV_SQL_NULL
+			    || rec2_f_len == UNIV_SQL_NULL) {
+
+				if (rec1_f_len == rec2_f_len) {
+
+					goto next_field;
+
+				} else if (rec2_f_len == UNIV_SQL_NULL) {
+
+					/* We define the SQL null to be the
+					smallest possible value of a field
+					in the alphabetical order */
+
+					ret = 1;
+				} else {
+					ret = -1;
+				}
+
+				goto order_resolved;
+			}
+		}
+
+		if (mtype >= DATA_FLOAT
+		    || (mtype == DATA_BLOB
+			&& 0 == (prtype & DATA_BINARY_TYPE)
+			&& dtype_get_charset_coll(prtype)
+			!= DATA_MYSQL_LATIN1_SWEDISH_CHARSET_COLL)) {
+
+			ret = cmp_whole_field(mtype, prtype,
+					      rec1_b_ptr,
+					      (unsigned) rec1_f_len,
+					      rec2_b_ptr,
+					      (unsigned) rec2_f_len);
+			if (ret != 0) {
+				cur_bytes = 0;
+
+				goto order_resolved;
+			} else {
+				goto next_field;
+			}
+		}
+
+		/* Set the pointers at the current byte */
+		rec1_b_ptr = rec1_b_ptr + cur_bytes;
+		rec2_b_ptr = rec2_b_ptr + cur_bytes;
+
+		/* Compare then the fields */
+		for (;;) {
+			if (rec2_f_len <= cur_bytes) {
+
+				if (rec1_f_len <= cur_bytes) {
+
+					goto next_field;
+				}
+
+				rec2_byte = dtype_get_pad_char(mtype, prtype);
+
+				if (rec2_byte == ULINT_UNDEFINED) {
+					ret = 1;
+
+					goto order_resolved;
+				}
+			} else {
+				rec2_byte = *rec2_b_ptr;
+			}
+
+			if (rec1_f_len <= cur_bytes) {
+				rec1_byte = dtype_get_pad_char(mtype, prtype);
+
+				if (rec1_byte == ULINT_UNDEFINED) {
+					ret = -1;
+
+					goto order_resolved;
+				}
+			} else {
+				rec1_byte = *rec1_b_ptr;
+			}
+
+			if (rec1_byte == rec2_byte) {
+				/* If the bytes are equal, they will remain
+				such even after the collation transformation
+				below */
+
+				goto next_byte;
+			}
+
+			if (mtype <= DATA_CHAR
+			    || (mtype == DATA_BLOB
+				&& !(prtype & DATA_BINARY_TYPE))) {
+
+				rec1_byte = cmp_collate(rec1_byte);
+				rec2_byte = cmp_collate(rec2_byte);
+			}
+
+			if (rec1_byte < rec2_byte) {
+				ret = -1;
+				goto order_resolved;
+			} else if (rec1_byte > rec2_byte) {
+				ret = 1;
+				goto order_resolved;
+			}
+next_byte:
+			/* Next byte */
+
+			cur_bytes++;
+			rec1_b_ptr++;
+			rec2_b_ptr++;
+		}
+
+next_field:
+		cur_field++;
+		cur_bytes = 0;
+	}
+
+	ut_ad(cur_bytes == 0);
+
+	/* If we ran out of fields, rec1 was equal to rec2 up
+	to the common fields */
+	ut_ad(ret == 0);
+order_resolved:
+
+	ut_ad((ret >= - 1) && (ret <= 1));
+
+	*matched_fields = cur_field;
+	*matched_bytes = cur_bytes;
+
+	return(ret);
+#else /* !UNIV_HOTBACKUP */
+	/* This function depends on MySQL code that is not included in
+	InnoDB Hot Backup builds.  Besides, this function should never
+	be called in InnoDB Hot Backup. */
+	ut_error;
+	return(0);
+#endif /* !UNIV_HOTBACKUP */
+}
+
+#ifdef UNIV_DEBUG
+/*****************************************************************
+Used in debug checking of cmp_dtuple_... .
+This function is used to compare a data tuple to a physical record. If
+dtuple has n fields then rec must have either m >= n fields, or it must
+differ from dtuple in some of the m fields rec has. If encounters an
+externally stored field, returns 0. */
+static
+int
+cmp_debug_dtuple_rec_with_match(
+/*============================*/
+				/* out: 1, 0, -1, if dtuple is greater, equal,
+				less than rec, respectively, when only the
+				common first fields are compared */
+	const dtuple_t*	dtuple,	/* in: data tuple */
+	const rec_t*	rec,	/* in: physical record which differs from
+				dtuple in some of the common fields, or which
+				has an equal number or more fields than
+				dtuple */
+	const ulint*	offsets,/* in: array returned by rec_get_offsets() */
+	ulint*		matched_fields) /* in/out: number of already
+				completely matched fields; when function
+				returns, contains the value for current
+				comparison */
+{
+	const dfield_t*	dtuple_field;	/* current field in logical record */
+	ulint		dtuple_f_len;	/* the length of the current field
+					in the logical record */
+	const byte*	dtuple_f_data;	/* pointer to the current logical
+					field data */
+	ulint		rec_f_len;	/* length of current field in rec */
+	const byte*	rec_f_data;	/* pointer to the current rec field */
+	int		ret = 3333;	/* return value */
+	ulint		cur_field;	/* current field number */
+
+	ut_ad(dtuple && rec && matched_fields);
+	ut_ad(dtuple_check_typed(dtuple));
+	ut_ad(rec_offs_validate(rec, NULL, offsets));
+
+	ut_ad(*matched_fields <= dtuple_get_n_fields_cmp(dtuple));
+	ut_ad(*matched_fields <= rec_offs_n_fields(offsets));
+
+	cur_field = *matched_fields;
+
+	if (cur_field == 0) {
+		if (UNIV_UNLIKELY
+		    (rec_get_info_bits(rec, rec_offs_comp(offsets))
+		     & REC_INFO_MIN_REC_FLAG)) {
+
+			ret = !(dtuple_get_info_bits(dtuple)
+				& REC_INFO_MIN_REC_FLAG);
+
+			goto order_resolved;
+		}
+
+		if (UNIV_UNLIKELY
+		    (dtuple_get_info_bits(dtuple) & REC_INFO_MIN_REC_FLAG)) {
+			ret = -1;
+
+			goto order_resolved;
+		}
+	}
+
+	/* Match fields in a loop; stop if we run out of fields in dtuple */
+
+	while (cur_field < dtuple_get_n_fields_cmp(dtuple)) {
+
+		ulint	mtype;
+		ulint	prtype;
+
+		dtuple_field = dtuple_get_nth_field(dtuple, cur_field);
+		{
+			const dtype_t*	type
+				= dfield_get_type(dtuple_field);
+
+			mtype = type->mtype;
+			prtype = type->prtype;
+		}
+
+		dtuple_f_data = dfield_get_data(dtuple_field);
+		dtuple_f_len = dfield_get_len(dtuple_field);
+
+		rec_f_data = rec_get_nth_field(rec, offsets,
+					       cur_field, &rec_f_len);
+
+		if (rec_offs_nth_extern(offsets, cur_field)) {
+			/* We do not compare to an externally stored field */
+
+			ret = 0;
+
+			goto order_resolved;
+		}
+
+		ret = cmp_data_data(mtype, prtype, dtuple_f_data, dtuple_f_len,
+				    rec_f_data, rec_f_len);
+		if (ret != 0) {
+			goto order_resolved;
+		}
+
+		cur_field++;
+	}
+
+	ret = 0;	/* If we ran out of fields, dtuple was equal to rec
+			up to the common fields */
+order_resolved:
+	ut_ad((ret >= - 1) && (ret <= 1));
+
+	*matched_fields = cur_field;
+
+	return(ret);
+}
+#endif /* UNIV_DEBUG */
diff --git a/storage/xtradb/rem/rem0rec.c b/storage/xtradb/rem/rem0rec.c
new file mode 100644
index 00000000000..d6899c810e3
--- /dev/null
+++ b/storage/xtradb/rem/rem0rec.c
@@ -0,0 +1,1720 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/************************************************************************
+Record manager
+
+Created 5/30/1994 Heikki Tuuri
+*************************************************************************/
+
+#include "rem0rec.h"
+
+#ifdef UNIV_NONINL
+#include "rem0rec.ic"
+#endif
+
+#include "mtr0mtr.h"
+#include "mtr0log.h"
+
+/*			PHYSICAL RECORD (OLD STYLE)
+			===========================
+
+The physical record, which is the data type of all the records
+found in index pages of the database, has the following format
+(lower addresses and more significant bits inside a byte are below
+represented on a higher text line):
+
+| offset of the end of the last field of data, the most significant
+  bit is set to 1 if and only if the field is SQL-null,
+  if the offset is 2-byte, then the second most significant
+  bit is set to 1 if the field is stored on another page:
+  mostly this will occur in the case of big BLOB fields |
+...
+| offset of the end of the first field of data + the SQL-null bit |
+| 4 bits used to delete mark a record, and mark a predefined
+  minimum record in alphabetical order |
+| 4 bits giving the number of records owned by this record
+  (this term is explained in page0page.h) |
+| 13 bits giving the order number of this record in the
+  heap of the index page |
+| 10 bits giving the number of fields in this record |
+| 1 bit which is set to 1 if the offsets above are given in
+  one byte format, 0 if in two byte format |
+| two bytes giving an absolute pointer to the next record in the page |
+ORIGIN of the record
+| first field of data |
+...
+| last field of data |
+
+The origin of the record is the start address of the first field
+of data. The offsets are given relative to the origin.
+The offsets of the data fields are stored in an inverted
+order because then the offset of the first fields are near the
+origin, giving maybe a better processor cache hit rate in searches.
+
+The offsets of the data fields are given as one-byte
+(if there are less than 127 bytes of data in the record)
+or two-byte unsigned integers. The most significant bit
+is not part of the offset, instead it indicates the SQL-null
+if the bit is set to 1. */
+
+/*			PHYSICAL RECORD (NEW STYLE)
+			===========================
+
+The physical record, which is the data type of all the records
+found in index pages of the database, has the following format
+(lower addresses and more significant bits inside a byte are below
+represented on a higher text line):
+
+| length of the last non-null variable-length field of data:
+  if the maximum length is 255, one byte; otherwise,
+  0xxxxxxx (one byte, length=0..127), or 1exxxxxxxxxxxxxx (two bytes,
+  length=128..16383, extern storage flag) |
+...
+| length of first variable-length field of data |
+| SQL-null flags (1 bit per nullable field), padded to full bytes |
+| 4 bits used to delete mark a record, and mark a predefined
+  minimum record in alphabetical order |
+| 4 bits giving the number of records owned by this record
+  (this term is explained in page0page.h) |
+| 13 bits giving the order number of this record in the
+  heap of the index page |
+| 3 bits record type: 000=conventional, 001=node pointer (inside B-tree),
+  010=infimum, 011=supremum, 1xx=reserved |
+| two bytes giving a relative pointer to the next record in the page |
+ORIGIN of the record
+| first field of data |
+...
+| last field of data |
+
+The origin of the record is the start address of the first field
+of data. The offsets are given relative to the origin.
+The offsets of the data fields are stored in an inverted
+order because then the offset of the first fields are near the
+origin, giving maybe a better processor cache hit rate in searches.
+
+The offsets of the data fields are given as one-byte
+(if there are less than 127 bytes of data in the record)
+or two-byte unsigned integers. The most significant bit
+is not part of the offset, instead it indicates the SQL-null
+if the bit is set to 1. */
+
+/* CANONICAL COORDINATES. A record can be seen as a single
+string of 'characters' in the following way: catenate the bytes
+in each field, in the order of fields. An SQL-null field
+is taken to be an empty sequence of bytes. Then after
+the position of each field insert in the string
+the 'character' <FIELD-END>, except that after an SQL-null field
+insert <NULL-FIELD-END>. Now the ordinal position of each
+byte in this canonical string is its canonical coordinate.
+So, for the record ("AA", SQL-NULL, "BB", ""), the canonical
+string is "AA<FIELD_END><NULL-FIELD-END>BB<FIELD-END><FIELD-END>".
+We identify prefixes (= initial segments) of a record
+with prefixes of the canonical string. The canonical
+length of the prefix is the length of the corresponding
+prefix of the canonical string. The canonical length of
+a record is the length of its canonical string.
+
+For example, the maximal common prefix of records
+("AA", SQL-NULL, "BB", "C") and ("AA", SQL-NULL, "B", "C")
+is "AA<FIELD-END><NULL-FIELD-END>B", and its canonical
+length is 5.
+
+A complete-field prefix of a record is a prefix which ends at the
+end of some field (containing also <FIELD-END>).
+A record is a complete-field prefix of another record, if
+the corresponding canonical strings have the same property. */
+
+/* this is used to fool compiler in rec_validate */
+UNIV_INTERN ulint	rec_dummy;
+
+/*******************************************************************
+Validates the consistency of an old-style physical record. */
+static
+ibool
+rec_validate_old(
+/*=============*/
+				/* out: TRUE if ok */
+	const rec_t*	rec);	/* in: physical record */
+
+/**********************************************************
+Determine how many of the first n columns in a compact
+physical record are stored externally. */
+UNIV_INTERN
+ulint
+rec_get_n_extern_new(
+/*=================*/
+				/* out: number of externally stored columns */
+	const rec_t*	rec,	/* in: compact physical record */
+	dict_index_t*	index,	/* in: record descriptor */
+	ulint		n)	/* in: number of columns to scan */
+{
+	const byte*	nulls;
+	const byte*	lens;
+	dict_field_t*	field;
+	ulint		null_mask;
+	ulint		n_extern;
+	ulint		i;
+
+	ut_ad(dict_table_is_comp(index->table));
+	ut_ad(rec_get_status(rec) == REC_STATUS_ORDINARY);
+	ut_ad(n == ULINT_UNDEFINED || n <= dict_index_get_n_fields(index));
+
+	if (n == ULINT_UNDEFINED) {
+		n = dict_index_get_n_fields(index);
+	}
+
+	nulls = rec - (REC_N_NEW_EXTRA_BYTES + 1);
+	lens = nulls - UT_BITS_IN_BYTES(index->n_nullable);
+	null_mask = 1;
+	n_extern = 0;
+	i = 0;
+
+	/* read the lengths of fields 0..n */
+	do {
+		ulint	len;
+
+		field = dict_index_get_nth_field(index, i);
+		if (!(dict_field_get_col(field)->prtype & DATA_NOT_NULL)) {
+			/* nullable field => read the null flag */
+
+			if (UNIV_UNLIKELY(!(byte) null_mask)) {
+				nulls--;
+				null_mask = 1;
+			}
+
+			if (*nulls & null_mask) {
+				null_mask <<= 1;
+				/* No length is stored for NULL fields. */
+				continue;
+			}
+			null_mask <<= 1;
+		}
+
+		if (UNIV_UNLIKELY(!field->fixed_len)) {
+			/* Variable-length field: read the length */
+			const dict_col_t*	col
+				= dict_field_get_col(field);
+			len = *lens--;
+			if (UNIV_UNLIKELY(col->len > 255)
+			    || UNIV_UNLIKELY(col->mtype == DATA_BLOB)) {
+				if (len & 0x80) {
+					/* 1exxxxxxx xxxxxxxx */
+					if (len & 0x40) {
+						n_extern++;
+					}
+					lens--;
+				}
+			}
+		}
+	} while (++i < n);
+
+	return(n_extern);
+}
+
+/**********************************************************
+Determine the offset to each field in a leaf-page record
+in ROW_FORMAT=COMPACT.  This is a special case of
+rec_init_offsets() and rec_get_offsets_func(). */
+UNIV_INTERN
+void
+rec_init_offsets_comp_ordinary(
+/*===========================*/
+	const rec_t*		rec,	/* in: physical record in
+					ROW_FORMAT=COMPACT */
+	ulint			extra,	/* in: number of bytes to reserve
+					between the record header and
+					the data payload
+					(usually REC_N_NEW_EXTRA_BYTES) */
+	const dict_index_t*	index,	/* in: record descriptor */
+	ulint*			offsets)/* in/out: array of offsets;
+					in: n=rec_offs_n_fields(offsets) */
+{
+	ulint		i		= 0;
+	ulint		offs		= 0;
+	ulint		any_ext		= 0;
+	const byte*	nulls		= rec - (extra + 1);
+	const byte*	lens		= nulls
+		- UT_BITS_IN_BYTES(index->n_nullable);
+	dict_field_t*	field;
+	ulint		null_mask	= 1;
+
+#ifdef UNIV_DEBUG
+	/* We cannot invoke rec_offs_make_valid() here, because it can hold
+	that extra != REC_N_NEW_EXTRA_BYTES.  Similarly, rec_offs_validate()
+	will fail in that case, because it invokes rec_get_status(). */
+	offsets[2] = (ulint) rec;
+	offsets[3] = (ulint) index;
+#endif /* UNIV_DEBUG */
+
+	/* read the lengths of fields 0..n */
+	do {
+		ulint	len;
+
+		field = dict_index_get_nth_field(index, i);
+		if (!(dict_field_get_col(field)->prtype
+		      & DATA_NOT_NULL)) {
+			/* nullable field => read the null flag */
+
+			if (UNIV_UNLIKELY(!(byte) null_mask)) {
+				nulls--;
+				null_mask = 1;
+			}
+
+			if (*nulls & null_mask) {
+				null_mask <<= 1;
+				/* No length is stored for NULL fields.
+				We do not advance offs, and we set
+				the length to zero and enable the
+				SQL NULL flag in offsets[]. */
+				len = offs | REC_OFFS_SQL_NULL;
+				goto resolved;
+			}
+			null_mask <<= 1;
+		}
+
+		if (UNIV_UNLIKELY(!field->fixed_len)) {
+			/* Variable-length field: read the length */
+			const dict_col_t*	col
+				= dict_field_get_col(field);
+			len = *lens--;
+			if (UNIV_UNLIKELY(col->len > 255)
+			    || UNIV_UNLIKELY(col->mtype
+					     == DATA_BLOB)) {
+				if (len & 0x80) {
+					/* 1exxxxxxx xxxxxxxx */
+					len <<= 8;
+					len |= *lens--;
+
+					offs += len & 0x3fff;
+					if (UNIV_UNLIKELY(len
+							  & 0x4000)) {
+						ut_ad(dict_index_is_clust
+						      (index));
+						any_ext = REC_OFFS_EXTERNAL;
+						len = offs
+							| REC_OFFS_EXTERNAL;
+					} else {
+						len = offs;
+					}
+
+					goto resolved;
+				}
+			}
+
+			len = offs += len;
+		} else {
+			len = offs += field->fixed_len;
+		}
+resolved:
+		rec_offs_base(offsets)[i + 1] = len;
+	} while (++i < rec_offs_n_fields(offsets));
+
+	*rec_offs_base(offsets)
+		= (rec - (lens + 1)) | REC_OFFS_COMPACT | any_ext;
+}
+
+/**********************************************************
+The following function determines the offsets to each field in the
+record.	 The offsets are written to a previously allocated array of
+ulint, where rec_offs_n_fields(offsets) has been initialized to the
+number of fields in the record.	 The rest of the array will be
+initialized by this function.  rec_offs_base(offsets)[0] will be set
+to the extra size (if REC_OFFS_COMPACT is set, the record is in the
+new format; if REC_OFFS_EXTERNAL is set, the record contains externally
+stored columns), and rec_offs_base(offsets)[1..n_fields] will be set to
+offsets past the end of fields 0..n_fields, or to the beginning of
+fields 1..n_fields+1.  When the high-order bit of the offset at [i+1]
+is set (REC_OFFS_SQL_NULL), the field i is NULL.  When the second
+high-order bit of the offset at [i+1] is set (REC_OFFS_EXTERNAL), the
+field i is being stored externally. */
+static
+void
+rec_init_offsets(
+/*=============*/
+	const rec_t*		rec,	/* in: physical record */
+	const dict_index_t*	index,	/* in: record descriptor */
+	ulint*			offsets)/* in/out: array of offsets;
+					in: n=rec_offs_n_fields(offsets) */
+{
+	ulint	i	= 0;
+	ulint	offs;
+
+	rec_offs_make_valid(rec, index, offsets);
+
+	if (dict_table_is_comp(index->table)) {
+		const byte*	nulls;
+		const byte*	lens;
+		dict_field_t*	field;
+		ulint		null_mask;
+		ulint		status = rec_get_status(rec);
+		ulint		n_node_ptr_field = ULINT_UNDEFINED;
+
+		switch (UNIV_EXPECT(status, REC_STATUS_ORDINARY)) {
+		case REC_STATUS_INFIMUM:
+		case REC_STATUS_SUPREMUM:
+			/* the field is 8 bytes long */
+			rec_offs_base(offsets)[0]
+				= REC_N_NEW_EXTRA_BYTES | REC_OFFS_COMPACT;
+			rec_offs_base(offsets)[1] = 8;
+			return;
+		case REC_STATUS_NODE_PTR:
+			n_node_ptr_field
+				= dict_index_get_n_unique_in_tree(index);
+			break;
+		case REC_STATUS_ORDINARY:
+			rec_init_offsets_comp_ordinary(rec,
+						       REC_N_NEW_EXTRA_BYTES,
+						       index, offsets);
+			return;
+		}
+
+		nulls = rec - (REC_N_NEW_EXTRA_BYTES + 1);
+		lens = nulls - UT_BITS_IN_BYTES(index->n_nullable);
+		offs = 0;
+		null_mask = 1;
+
+		/* read the lengths of fields 0..n */
+		do {
+			ulint	len;
+			if (UNIV_UNLIKELY(i == n_node_ptr_field)) {
+				len = offs += 4;
+				goto resolved;
+			}
+
+			field = dict_index_get_nth_field(index, i);
+			if (!(dict_field_get_col(field)->prtype
+			      & DATA_NOT_NULL)) {
+				/* nullable field => read the null flag */
+
+				if (UNIV_UNLIKELY(!(byte) null_mask)) {
+					nulls--;
+					null_mask = 1;
+				}
+
+				if (*nulls & null_mask) {
+					null_mask <<= 1;
+					/* No length is stored for NULL fields.
+					We do not advance offs, and we set
+					the length to zero and enable the
+					SQL NULL flag in offsets[]. */
+					len = offs | REC_OFFS_SQL_NULL;
+					goto resolved;
+				}
+				null_mask <<= 1;
+			}
+
+			if (UNIV_UNLIKELY(!field->fixed_len)) {
+				/* Variable-length field: read the length */
+				const dict_col_t*	col
+					= dict_field_get_col(field);
+				len = *lens--;
+				if (UNIV_UNLIKELY(col->len > 255)
+				    || UNIV_UNLIKELY(col->mtype
+						     == DATA_BLOB)) {
+					if (len & 0x80) {
+						/* 1exxxxxxx xxxxxxxx */
+
+						len <<= 8;
+						len |= *lens--;
+
+						/* B-tree node pointers
+						must not contain externally
+						stored columns.  Thus
+						the "e" flag must be 0. */
+						ut_a(!(len & 0x4000));
+						offs += len & 0x3fff;
+						len = offs;
+
+						goto resolved;
+					}
+				}
+
+				len = offs += len;
+			} else {
+				len = offs += field->fixed_len;
+			}
+resolved:
+			rec_offs_base(offsets)[i + 1] = len;
+		} while (++i < rec_offs_n_fields(offsets));
+
+		*rec_offs_base(offsets)
+			= (rec - (lens + 1)) | REC_OFFS_COMPACT;
+	} else {
+		/* Old-style record: determine extra size and end offsets */
+		offs = REC_N_OLD_EXTRA_BYTES;
+		if (rec_get_1byte_offs_flag(rec)) {
+			offs += rec_offs_n_fields(offsets);
+			*rec_offs_base(offsets) = offs;
+			/* Determine offsets to fields */
+			do {
+				offs = rec_1_get_field_end_info(rec, i);
+				if (offs & REC_1BYTE_SQL_NULL_MASK) {
+					offs &= ~REC_1BYTE_SQL_NULL_MASK;
+					offs |= REC_OFFS_SQL_NULL;
+				}
+				rec_offs_base(offsets)[1 + i] = offs;
+			} while (++i < rec_offs_n_fields(offsets));
+		} else {
+			offs += 2 * rec_offs_n_fields(offsets);
+			*rec_offs_base(offsets) = offs;
+			/* Determine offsets to fields */
+			do {
+				offs = rec_2_get_field_end_info(rec, i);
+				if (offs & REC_2BYTE_SQL_NULL_MASK) {
+					offs &= ~REC_2BYTE_SQL_NULL_MASK;
+					offs |= REC_OFFS_SQL_NULL;
+				}
+				if (offs & REC_2BYTE_EXTERN_MASK) {
+					offs &= ~REC_2BYTE_EXTERN_MASK;
+					offs |= REC_OFFS_EXTERNAL;
+					*rec_offs_base(offsets) |= REC_OFFS_EXTERNAL;
+				}
+				rec_offs_base(offsets)[1 + i] = offs;
+			} while (++i < rec_offs_n_fields(offsets));
+		}
+	}
+}
+
+/**********************************************************
+The following function determines the offsets to each field
+in the record.	It can reuse a previously returned array. */
+UNIV_INTERN
+ulint*
+rec_get_offsets_func(
+/*=================*/
+					/* out: the new offsets */
+	const rec_t*		rec,	/* in: physical record */
+	const dict_index_t*	index,	/* in: record descriptor */
+	ulint*			offsets,/* in/out: array consisting of
+					offsets[0] allocated elements,
+					or an array from rec_get_offsets(),
+					or NULL */
+	ulint			n_fields,/* in: maximum number of
+					initialized fields
+					 (ULINT_UNDEFINED if all fields) */
+	mem_heap_t**		heap,	/* in/out: memory heap */
+	const char*		file,	/* in: file name where called */
+	ulint			line)	/* in: line number where called */
+{
+	ulint	n;
+	ulint	size;
+
+	ut_ad(rec);
+	ut_ad(index);
+	ut_ad(heap);
+
+	if (dict_table_is_comp(index->table)) {
+		switch (UNIV_EXPECT(rec_get_status(rec),
+				    REC_STATUS_ORDINARY)) {
+		case REC_STATUS_ORDINARY:
+			n = dict_index_get_n_fields(index);
+			break;
+		case REC_STATUS_NODE_PTR:
+			n = dict_index_get_n_unique_in_tree(index) + 1;
+			break;
+		case REC_STATUS_INFIMUM:
+		case REC_STATUS_SUPREMUM:
+			/* infimum or supremum record */
+			n = 1;
+			break;
+		default:
+			ut_error;
+			return(NULL);
+		}
+	} else {
+		n = rec_get_n_fields_old(rec);
+	}
+
+	if (UNIV_UNLIKELY(n_fields < n)) {
+		n = n_fields;
+	}
+
+	size = n + (1 + REC_OFFS_HEADER_SIZE);
+
+	if (UNIV_UNLIKELY(!offsets)
+	    || UNIV_UNLIKELY(rec_offs_get_n_alloc(offsets) < size)) {
+		if (UNIV_UNLIKELY(!*heap)) {
+			*heap = mem_heap_create_func(size * sizeof(ulint),
+						     MEM_HEAP_DYNAMIC,
+						     file, line);
+		}
+		offsets = mem_heap_alloc(*heap, size * sizeof(ulint));
+		rec_offs_set_n_alloc(offsets, size);
+	}
+
+	rec_offs_set_n_fields(offsets, n);
+	rec_init_offsets(rec, index, offsets);
+	return(offsets);
+}
+
+/**********************************************************
+The following function determines the offsets to each field
+in the record.  It can reuse a previously allocated array. */
+UNIV_INTERN
+void
+rec_get_offsets_reverse(
+/*====================*/
+	const byte*		extra,	/* in: the extra bytes of a
+					compact record in reverse order,
+					excluding the fixed-size
+					REC_N_NEW_EXTRA_BYTES */
+	const dict_index_t*	index,	/* in: record descriptor */
+	ulint			node_ptr,/* in: nonzero=node pointer,
+					0=leaf node */
+	ulint*			offsets)/* in/out: array consisting of
+					offsets[0] allocated elements */
+{
+	ulint		n;
+	ulint		i;
+	ulint		offs;
+	ulint		any_ext;
+	const byte*	nulls;
+	const byte*	lens;
+	dict_field_t*	field;
+	ulint		null_mask;
+	ulint		n_node_ptr_field;
+
+	ut_ad(extra);
+	ut_ad(index);
+	ut_ad(offsets);
+	ut_ad(dict_table_is_comp(index->table));
+
+	if (UNIV_UNLIKELY(node_ptr)) {
+		n_node_ptr_field = dict_index_get_n_unique_in_tree(index);
+		n = n_node_ptr_field + 1;
+	} else {
+		n_node_ptr_field = ULINT_UNDEFINED;
+		n = dict_index_get_n_fields(index);
+	}
+
+	ut_a(rec_offs_get_n_alloc(offsets) >= n + (1 + REC_OFFS_HEADER_SIZE));
+	rec_offs_set_n_fields(offsets, n);
+
+	nulls = extra;
+	lens = nulls + UT_BITS_IN_BYTES(index->n_nullable);
+	i = offs = 0;
+	null_mask = 1;
+	any_ext = 0;
+
+	/* read the lengths of fields 0..n */
+	do {
+		ulint	len;
+		if (UNIV_UNLIKELY(i == n_node_ptr_field)) {
+			len = offs += 4;
+			goto resolved;
+		}
+
+		field = dict_index_get_nth_field(index, i);
+		if (!(dict_field_get_col(field)->prtype & DATA_NOT_NULL)) {
+			/* nullable field => read the null flag */
+
+			if (UNIV_UNLIKELY(!(byte) null_mask)) {
+				nulls++;
+				null_mask = 1;
+			}
+
+			if (*nulls & null_mask) {
+				null_mask <<= 1;
+				/* No length is stored for NULL fields.
+				We do not advance offs, and we set
+				the length to zero and enable the
+				SQL NULL flag in offsets[]. */
+				len = offs | REC_OFFS_SQL_NULL;
+				goto resolved;
+			}
+			null_mask <<= 1;
+		}
+
+		if (UNIV_UNLIKELY(!field->fixed_len)) {
+			/* Variable-length field: read the length */
+			const dict_col_t*	col
+				= dict_field_get_col(field);
+			len = *lens++;
+			if (UNIV_UNLIKELY(col->len > 255)
+			    || UNIV_UNLIKELY(col->mtype == DATA_BLOB)) {
+				if (len & 0x80) {
+					/* 1exxxxxxx xxxxxxxx */
+					len <<= 8;
+					len |= *lens++;
+
+					offs += len & 0x3fff;
+					if (UNIV_UNLIKELY(len & 0x4000)) {
+						any_ext = REC_OFFS_EXTERNAL;
+						len = offs | REC_OFFS_EXTERNAL;
+					} else {
+						len = offs;
+					}
+
+					goto resolved;
+				}
+			}
+
+			len = offs += len;
+		} else {
+			len = offs += field->fixed_len;
+		}
+resolved:
+		rec_offs_base(offsets)[i + 1] = len;
+	} while (++i < rec_offs_n_fields(offsets));
+
+	ut_ad(lens >= extra);
+	*rec_offs_base(offsets) = (lens - extra + REC_N_NEW_EXTRA_BYTES)
+		| REC_OFFS_COMPACT | any_ext;
+}
+
+/****************************************************************
+The following function is used to get the offset to the nth
+data field in an old-style record. */
+UNIV_INTERN
+ulint
+rec_get_nth_field_offs_old(
+/*=======================*/
+				/* out: offset to the field */
+	const rec_t*	rec,	/* in: record */
+	ulint		n,	/* in: index of the field */
+	ulint*		len)	/* out: length of the field;
+				UNIV_SQL_NULL if SQL null */
+{
+	ulint	os;
+	ulint	next_os;
+
+	ut_ad(rec && len);
+	ut_ad(n < rec_get_n_fields_old(rec));
+
+	if (UNIV_UNLIKELY(n > REC_MAX_N_FIELDS)) {
+		fprintf(stderr, "Error: trying to access field %lu in rec\n",
+			(ulong) n);
+		ut_error;
+	}
+
+	if (UNIV_UNLIKELY(rec == NULL)) {
+		fputs("Error: rec is NULL pointer\n", stderr);
+		ut_error;
+	}
+
+	if (rec_get_1byte_offs_flag(rec)) {
+		os = rec_1_get_field_start_offs(rec, n);
+
+		next_os = rec_1_get_field_end_info(rec, n);
+
+		if (next_os & REC_1BYTE_SQL_NULL_MASK) {
+			*len = UNIV_SQL_NULL;
+
+			return(os);
+		}
+
+		next_os = next_os & ~REC_1BYTE_SQL_NULL_MASK;
+	} else {
+		os = rec_2_get_field_start_offs(rec, n);
+
+		next_os = rec_2_get_field_end_info(rec, n);
+
+		if (next_os & REC_2BYTE_SQL_NULL_MASK) {
+			*len = UNIV_SQL_NULL;
+
+			return(os);
+		}
+
+		next_os = next_os & ~(REC_2BYTE_SQL_NULL_MASK
+				      | REC_2BYTE_EXTERN_MASK);
+	}
+
+	*len = next_os - os;
+
+	ut_ad(*len < UNIV_PAGE_SIZE);
+
+	return(os);
+}
+
+/**************************************************************
+Determines the size of a data tuple prefix in ROW_FORMAT=COMPACT. */
+UNIV_INTERN
+ulint
+rec_get_converted_size_comp_prefix(
+/*===============================*/
+					/* out: total size */
+	const dict_index_t*	index,	/* in: record descriptor;
+					dict_table_is_comp() is
+					assumed to hold, even if
+					it does not */
+	const dfield_t*		fields,	/* in: array of data fields */
+	ulint			n_fields,/* in: number of data fields */
+	ulint*			extra)	/* out: extra size */
+{
+	ulint	extra_size;
+	ulint	data_size;
+	ulint	i;
+	ut_ad(index);
+	ut_ad(fields);
+	ut_ad(n_fields > 0);
+	ut_ad(n_fields <= dict_index_get_n_fields(index));
+
+	extra_size = REC_N_NEW_EXTRA_BYTES
+		+ UT_BITS_IN_BYTES(index->n_nullable);
+	data_size = 0;
+
+	/* read the lengths of fields 0..n */
+	for (i = 0; i < n_fields; i++) {
+		const dict_field_t*	field;
+		ulint			len;
+		const dict_col_t*	col;
+
+		field = dict_index_get_nth_field(index, i);
+		len = dfield_get_len(&fields[i]);
+		col = dict_field_get_col(field);
+
+		ut_ad(dict_col_type_assert_equal(col,
+						 dfield_get_type(&fields[i])));
+
+		if (dfield_is_null(&fields[i])) {
+			/* No length is stored for NULL fields. */
+			ut_ad(!(col->prtype & DATA_NOT_NULL));
+			continue;
+		}
+
+		ut_ad(len <= col->len || col->mtype == DATA_BLOB);
+
+		if (field->fixed_len) {
+			ut_ad(len == field->fixed_len);
+			/* dict_index_add_col() should guarantee this */
+			ut_ad(!field->prefix_len
+			      || field->fixed_len == field->prefix_len);
+		} else if (dfield_is_ext(&fields[i])) {
+			extra_size += 2;
+		} else if (len < 128
+			   || (col->len < 256 && col->mtype != DATA_BLOB)) {
+			extra_size++;
+		} else {
+			/* For variable-length columns, we look up the
+			maximum length from the column itself.  If this
+			is a prefix index column shorter than 256 bytes,
+			this will waste one byte. */
+			extra_size += 2;
+		}
+		data_size += len;
+	}
+
+	if (UNIV_LIKELY_NULL(extra)) {
+		*extra = extra_size;
+	}
+
+	return(extra_size + data_size);
+}
+
+/**************************************************************
+Determines the size of a data tuple in ROW_FORMAT=COMPACT. */
+UNIV_INTERN
+ulint
+rec_get_converted_size_comp(
+/*========================*/
+					/* out: total size */
+	const dict_index_t*	index,	/* in: record descriptor;
+					dict_table_is_comp() is
+					assumed to hold, even if
+					it does not */
+	ulint			status,	/* in: status bits of the record */
+	const dfield_t*		fields,	/* in: array of data fields */
+	ulint			n_fields,/* in: number of data fields */
+	ulint*			extra)	/* out: extra size */
+{
+	ulint	size;
+	ut_ad(index);
+	ut_ad(fields);
+	ut_ad(n_fields > 0);
+
+	switch (UNIV_EXPECT(status, REC_STATUS_ORDINARY)) {
+	case REC_STATUS_ORDINARY:
+		ut_ad(n_fields == dict_index_get_n_fields(index));
+		size = 0;
+		break;
+	case REC_STATUS_NODE_PTR:
+		n_fields--;
+		ut_ad(n_fields == dict_index_get_n_unique_in_tree(index));
+		ut_ad(dfield_get_len(&fields[n_fields]) == REC_NODE_PTR_SIZE);
+		size = REC_NODE_PTR_SIZE; /* child page number */
+		break;
+	case REC_STATUS_INFIMUM:
+	case REC_STATUS_SUPREMUM:
+		/* infimum or supremum record, 8 data bytes */
+		if (UNIV_LIKELY_NULL(extra)) {
+			*extra = REC_N_NEW_EXTRA_BYTES;
+		}
+		return(REC_N_NEW_EXTRA_BYTES + 8);
+	default:
+		ut_error;
+		return(ULINT_UNDEFINED);
+	}
+
+	return(size + rec_get_converted_size_comp_prefix(index, fields,
+							 n_fields, extra));
+}
+
+/***************************************************************
+Sets the value of the ith field SQL null bit of an old-style record. */
+UNIV_INTERN
+void
+rec_set_nth_field_null_bit(
+/*=======================*/
+	rec_t*	rec,	/* in: record */
+	ulint	i,	/* in: ith field */
+	ibool	val)	/* in: value to set */
+{
+	ulint	info;
+
+	if (rec_get_1byte_offs_flag(rec)) {
+
+		info = rec_1_get_field_end_info(rec, i);
+
+		if (val) {
+			info = info | REC_1BYTE_SQL_NULL_MASK;
+		} else {
+			info = info & ~REC_1BYTE_SQL_NULL_MASK;
+		}
+
+		rec_1_set_field_end_info(rec, i, info);
+
+		return;
+	}
+
+	info = rec_2_get_field_end_info(rec, i);
+
+	if (val) {
+		info = info | REC_2BYTE_SQL_NULL_MASK;
+	} else {
+		info = info & ~REC_2BYTE_SQL_NULL_MASK;
+	}
+
+	rec_2_set_field_end_info(rec, i, info);
+}
+
+/***************************************************************
+Sets an old-style record field to SQL null.
+The physical size of the field is not changed. */
+UNIV_INTERN
+void
+rec_set_nth_field_sql_null(
+/*=======================*/
+	rec_t*	rec,	/* in: record */
+	ulint	n)	/* in: index of the field */
+{
+	ulint	offset;
+
+	offset = rec_get_field_start_offs(rec, n);
+
+	data_write_sql_null(rec + offset, rec_get_nth_field_size(rec, n));
+
+	rec_set_nth_field_null_bit(rec, n, TRUE);
+}
+
+/*************************************************************
+Builds an old-style physical record out of a data tuple and
+stores it beginning from the start of the given buffer. */
+static
+rec_t*
+rec_convert_dtuple_to_rec_old(
+/*==========================*/
+				/* out: pointer to the origin of
+				physical record */
+	byte*		buf,	/* in: start address of the physical record */
+	const dtuple_t*	dtuple,	/* in: data tuple */
+	ulint		n_ext)	/* in: number of externally stored columns */
+{
+	const dfield_t*	field;
+	ulint		n_fields;
+	ulint		data_size;
+	rec_t*		rec;
+	ulint		end_offset;
+	ulint		ored_offset;
+	ulint		len;
+	ulint		i;
+
+	ut_ad(buf && dtuple);
+	ut_ad(dtuple_validate(dtuple));
+	ut_ad(dtuple_check_typed(dtuple));
+
+	n_fields = dtuple_get_n_fields(dtuple);
+	data_size = dtuple_get_data_size(dtuple);
+
+	ut_ad(n_fields > 0);
+
+	/* Calculate the offset of the origin in the physical record */
+
+	rec = buf + rec_get_converted_extra_size(data_size, n_fields, n_ext);
+#ifdef UNIV_DEBUG
+	/* Suppress Valgrind warnings of ut_ad()
+	in mach_write_to_1(), mach_write_to_2() et al. */
+	memset(buf, 0xff, rec - buf + data_size);
+#endif /* UNIV_DEBUG */
+	/* Store the number of fields */
+	rec_set_n_fields_old(rec, n_fields);
+
+	/* Set the info bits of the record */
+	rec_set_info_bits_old(rec, dtuple_get_info_bits(dtuple)
+			      & REC_INFO_BITS_MASK);
+
+	/* Store the data and the offsets */
+
+	end_offset = 0;
+
+	if (!n_ext && data_size <= REC_1BYTE_OFFS_LIMIT) {
+
+		rec_set_1byte_offs_flag(rec, TRUE);
+
+		for (i = 0; i < n_fields; i++) {
+
+			field = dtuple_get_nth_field(dtuple, i);
+
+			if (dfield_is_null(field)) {
+				len = dtype_get_sql_null_size(
+					dfield_get_type(field));
+				data_write_sql_null(rec + end_offset, len);
+
+				end_offset += len;
+				ored_offset = end_offset
+					| REC_1BYTE_SQL_NULL_MASK;
+			} else {
+				/* If the data is not SQL null, store it */
+				len = dfield_get_len(field);
+
+				memcpy(rec + end_offset,
+				       dfield_get_data(field), len);
+
+				end_offset += len;
+				ored_offset = end_offset;
+			}
+
+			rec_1_set_field_end_info(rec, i, ored_offset);
+		}
+	} else {
+		rec_set_1byte_offs_flag(rec, FALSE);
+
+		for (i = 0; i < n_fields; i++) {
+
+			field = dtuple_get_nth_field(dtuple, i);
+
+			if (dfield_is_null(field)) {
+				len = dtype_get_sql_null_size(
+					dfield_get_type(field));
+				data_write_sql_null(rec + end_offset, len);
+
+				end_offset += len;
+				ored_offset = end_offset
+					| REC_2BYTE_SQL_NULL_MASK;
+			} else {
+				/* If the data is not SQL null, store it */
+				len = dfield_get_len(field);
+
+				memcpy(rec + end_offset,
+				       dfield_get_data(field), len);
+
+				end_offset += len;
+				ored_offset = end_offset;
+
+				if (dfield_is_ext(field)) {
+					ored_offset |= REC_2BYTE_EXTERN_MASK;
+				}
+			}
+
+			rec_2_set_field_end_info(rec, i, ored_offset);
+		}
+	}
+
+	return(rec);
+}
+
+/*************************************************************
+Builds a ROW_FORMAT=COMPACT record out of a data tuple. */
+UNIV_INTERN
+void
+rec_convert_dtuple_to_rec_comp(
+/*===========================*/
+	rec_t*			rec,	/* in: origin of record */
+	ulint			extra,	/* in: number of bytes to
+					reserve between the record
+					header and the data payload
+					(normally REC_N_NEW_EXTRA_BYTES) */
+	const dict_index_t*	index,	/* in: record descriptor */
+	ulint			status,	/* in: status bits of the record */
+	const dfield_t*		fields,	/* in: array of data fields */
+	ulint			n_fields)/* in: number of data fields */
+{
+	const dfield_t*	field;
+	const dtype_t*	type;
+	byte*		end;
+	byte*		nulls;
+	byte*		lens;
+	ulint		len;
+	ulint		i;
+	ulint		n_node_ptr_field;
+	ulint		fixed_len;
+	ulint		null_mask	= 1;
+	ut_ad(extra == 0 || dict_table_is_comp(index->table));
+	ut_ad(extra == 0 || extra == REC_N_NEW_EXTRA_BYTES);
+	ut_ad(n_fields > 0);
+
+	switch (UNIV_EXPECT(status, REC_STATUS_ORDINARY)) {
+	case REC_STATUS_ORDINARY:
+		ut_ad(n_fields <= dict_index_get_n_fields(index));
+		n_node_ptr_field = ULINT_UNDEFINED;
+		break;
+	case REC_STATUS_NODE_PTR:
+		ut_ad(n_fields == dict_index_get_n_unique_in_tree(index) + 1);
+		n_node_ptr_field = n_fields - 1;
+		break;
+	case REC_STATUS_INFIMUM:
+	case REC_STATUS_SUPREMUM:
+		ut_ad(n_fields == 1);
+		n_node_ptr_field = ULINT_UNDEFINED;
+		break;
+	default:
+		ut_error;
+		return;
+	}
+
+	end = rec;
+	nulls = rec - (extra + 1);
+	lens = nulls - UT_BITS_IN_BYTES(index->n_nullable);
+	/* clear the SQL-null flags */
+	memset(lens + 1, 0, nulls - lens);
+
+	/* Store the data and the offsets */
+
+	for (i = 0, field = fields; i < n_fields; i++, field++) {
+		type = dfield_get_type(field);
+		len = dfield_get_len(field);
+
+		if (UNIV_UNLIKELY(i == n_node_ptr_field)) {
+			ut_ad(dtype_get_prtype(type) & DATA_NOT_NULL);
+			ut_ad(len == 4);
+			memcpy(end, dfield_get_data(field), len);
+			end += 4;
+			break;
+		}
+
+		if (!(dtype_get_prtype(type) & DATA_NOT_NULL)) {
+			/* nullable field */
+			ut_ad(index->n_nullable > 0);
+
+			if (UNIV_UNLIKELY(!(byte) null_mask)) {
+				nulls--;
+				null_mask = 1;
+			}
+
+			ut_ad(*nulls < null_mask);
+
+			/* set the null flag if necessary */
+			if (dfield_is_null(field)) {
+				*nulls |= null_mask;
+				null_mask <<= 1;
+				continue;
+			}
+
+			null_mask <<= 1;
+		}
+		/* only nullable fields can be null */
+		ut_ad(!dfield_is_null(field));
+
+		fixed_len = dict_index_get_nth_field(index, i)->fixed_len;
+
+		if (fixed_len) {
+			ut_ad(len == fixed_len);
+			ut_ad(!dfield_is_ext(field));
+		} else if (dfield_is_ext(field)) {
+			ut_ad(len <= REC_MAX_INDEX_COL_LEN
+			      + BTR_EXTERN_FIELD_REF_SIZE);
+			*lens-- = (byte) (len >> 8) | 0xc0;
+			*lens-- = (byte) len;
+		} else {
+			ut_ad(len <= dtype_get_len(type)
+			      || dtype_get_mtype(type) == DATA_BLOB);
+			if (len < 128
+			    || (dtype_get_len(type) < 256
+				&& dtype_get_mtype(type) != DATA_BLOB)) {
+
+				*lens-- = (byte) len;
+			} else {
+				ut_ad(len < 16384);
+				*lens-- = (byte) (len >> 8) | 0x80;
+				*lens-- = (byte) len;
+			}
+		}
+
+		memcpy(end, dfield_get_data(field), len);
+		end += len;
+	}
+}
+
+/*************************************************************
+Builds a new-style physical record out of a data tuple and
+stores it beginning from the start of the given buffer. */
+static
+rec_t*
+rec_convert_dtuple_to_rec_new(
+/*==========================*/
+					/* out: pointer to the origin
+					of physical record */
+	byte*			buf,	/* in: start address of
+					the physical record */
+	const dict_index_t*	index,	/* in: record descriptor */
+	const dtuple_t*		dtuple)	/* in: data tuple */
+{
+	ulint	extra_size;
+	ulint	status;
+	rec_t*	rec;
+
+	status = dtuple_get_info_bits(dtuple) & REC_NEW_STATUS_MASK;
+	rec_get_converted_size_comp(index, status,
+				    dtuple->fields, dtuple->n_fields,
+				    &extra_size);
+	rec = buf + extra_size;
+
+	rec_convert_dtuple_to_rec_comp(
+		rec, REC_N_NEW_EXTRA_BYTES, index, status,
+		dtuple->fields, dtuple->n_fields);
+
+	/* Set the info bits of the record */
+	rec_set_info_and_status_bits(rec, dtuple_get_info_bits(dtuple));
+
+	return(rec);
+}
+
+/*************************************************************
+Builds a physical record out of a data tuple and
+stores it beginning from the start of the given buffer. */
+UNIV_INTERN
+rec_t*
+rec_convert_dtuple_to_rec(
+/*======================*/
+					/* out: pointer to the origin
+					of physical record */
+	byte*			buf,	/* in: start address of the
+					physical record */
+	const dict_index_t*	index,	/* in: record descriptor */
+	const dtuple_t*		dtuple,	/* in: data tuple */
+	ulint			n_ext)	/* in: number of
+					externally stored columns */
+{
+	rec_t*	rec;
+
+	ut_ad(buf && index && dtuple);
+	ut_ad(dtuple_validate(dtuple));
+	ut_ad(dtuple_check_typed(dtuple));
+
+	if (dict_table_is_comp(index->table)) {
+		rec = rec_convert_dtuple_to_rec_new(buf, index, dtuple);
+	} else {
+		rec = rec_convert_dtuple_to_rec_old(buf, dtuple, n_ext);
+	}
+
+#ifdef UNIV_DEBUG
+	{
+		mem_heap_t*	heap	= NULL;
+		ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+		const ulint*	offsets;
+		rec_offs_init(offsets_);
+
+		offsets = rec_get_offsets(rec, index,
+					  offsets_, ULINT_UNDEFINED, &heap);
+		ut_ad(rec_validate(rec, offsets));
+		if (UNIV_LIKELY_NULL(heap)) {
+			mem_heap_free(heap);
+		}
+	}
+#endif /* UNIV_DEBUG */
+	return(rec);
+}
+
+/******************************************************************
+Copies the first n fields of a physical record to a data tuple. The fields
+are copied to the memory heap. */
+UNIV_INTERN
+void
+rec_copy_prefix_to_dtuple(
+/*======================*/
+	dtuple_t*		tuple,		/* out: data tuple */
+	const rec_t*		rec,		/* in: physical record */
+	const dict_index_t*	index,		/* in: record descriptor */
+	ulint			n_fields,	/* in: number of fields
+						to copy */
+	mem_heap_t*		heap)		/* in: memory heap */
+{
+	ulint	i;
+	ulint	offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint*	offsets	= offsets_;
+	rec_offs_init(offsets_);
+
+	offsets = rec_get_offsets(rec, index, offsets, n_fields, &heap);
+
+	ut_ad(rec_validate(rec, offsets));
+	ut_ad(dtuple_check_typed(tuple));
+
+	dtuple_set_info_bits(tuple, rec_get_info_bits(
+				     rec, dict_table_is_comp(index->table)));
+
+	for (i = 0; i < n_fields; i++) {
+		dfield_t*	field;
+		const byte*	data;
+		ulint		len;
+
+		field = dtuple_get_nth_field(tuple, i);
+		data = rec_get_nth_field(rec, offsets, i, &len);
+
+		if (len != UNIV_SQL_NULL) {
+			dfield_set_data(field,
+					mem_heap_dup(heap, data, len), len);
+			ut_ad(!rec_offs_nth_extern(offsets, i));
+		} else {
+			dfield_set_null(field);
+		}
+	}
+}
+
+/******************************************************************
+Copies the first n fields of an old-style physical record
+to a new physical record in a buffer. */
+static
+rec_t*
+rec_copy_prefix_to_buf_old(
+/*=======================*/
+					/* out, own: copied record */
+	const rec_t*	rec,		/* in: physical record */
+	ulint		n_fields,	/* in: number of fields to copy */
+	ulint		area_end,	/* in: end of the prefix data */
+	byte**		buf,		/* in/out: memory buffer for
+					the copied prefix, or NULL */
+	ulint*		buf_size)	/* in/out: buffer size */
+{
+	rec_t*	copy_rec;
+	ulint	area_start;
+	ulint	prefix_len;
+
+	if (rec_get_1byte_offs_flag(rec)) {
+		area_start = REC_N_OLD_EXTRA_BYTES + n_fields;
+	} else {
+		area_start = REC_N_OLD_EXTRA_BYTES + 2 * n_fields;
+	}
+
+	prefix_len = area_start + area_end;
+
+	if ((*buf == NULL) || (*buf_size < prefix_len)) {
+		if (*buf != NULL) {
+			mem_free(*buf);
+		}
+
+		*buf = mem_alloc2(prefix_len, buf_size);
+	}
+
+	ut_memcpy(*buf, rec - area_start, prefix_len);
+
+	copy_rec = *buf + area_start;
+
+	rec_set_n_fields_old(copy_rec, n_fields);
+
+	return(copy_rec);
+}
+
+/******************************************************************
+Copies the first n fields of a physical record to a new physical record in
+a buffer. */
+UNIV_INTERN
+rec_t*
+rec_copy_prefix_to_buf(
+/*===================*/
+						/* out, own: copied record */
+	const rec_t*		rec,		/* in: physical record */
+	const dict_index_t*	index,		/* in: record descriptor */
+	ulint			n_fields,	/* in: number of fields
+						to copy */
+	byte**			buf,		/* in/out: memory buffer
+						for the copied prefix,
+						or NULL */
+	ulint*			buf_size)	/* in/out: buffer size */
+{
+	const byte*	nulls;
+	const byte*	lens;
+	ulint		i;
+	ulint		prefix_len;
+	ulint		null_mask;
+	ulint		status;
+
+	UNIV_PREFETCH_RW(*buf);
+
+	if (!dict_table_is_comp(index->table)) {
+		ut_ad(rec_validate_old(rec));
+		return(rec_copy_prefix_to_buf_old(
+			       rec, n_fields,
+			       rec_get_field_start_offs(rec, n_fields),
+			       buf, buf_size));
+	}
+
+	status = rec_get_status(rec);
+
+	switch (status) {
+	case REC_STATUS_ORDINARY:
+		ut_ad(n_fields <= dict_index_get_n_fields(index));
+		break;
+	case REC_STATUS_NODE_PTR:
+		/* it doesn't make sense to copy the child page number field */
+		ut_ad(n_fields <= dict_index_get_n_unique_in_tree(index));
+		break;
+	case REC_STATUS_INFIMUM:
+	case REC_STATUS_SUPREMUM:
+		/* infimum or supremum record: no sense to copy anything */
+	default:
+		ut_error;
+		return(NULL);
+	}
+
+	nulls = rec - (REC_N_NEW_EXTRA_BYTES + 1);
+	lens = nulls - UT_BITS_IN_BYTES(index->n_nullable);
+	UNIV_PREFETCH_R(lens);
+	prefix_len = 0;
+	null_mask = 1;
+
+	/* read the lengths of fields 0..n */
+	for (i = 0; i < n_fields; i++) {
+		const dict_field_t*	field;
+		const dict_col_t*	col;
+
+		field = dict_index_get_nth_field(index, i);
+		col = dict_field_get_col(field);
+
+		if (!(col->prtype & DATA_NOT_NULL)) {
+			/* nullable field => read the null flag */
+			if (UNIV_UNLIKELY(!(byte) null_mask)) {
+				nulls--;
+				null_mask = 1;
+			}
+
+			if (*nulls & null_mask) {
+				null_mask <<= 1;
+				continue;
+			}
+
+			null_mask <<= 1;
+		}
+
+		if (field->fixed_len) {
+			prefix_len += field->fixed_len;
+		} else {
+			ulint	len = *lens--;
+			if (col->len > 255 || col->mtype == DATA_BLOB) {
+				if (len & 0x80) {
+					/* 1exxxxxx */
+					len &= 0x3f;
+					len <<= 8;
+					len |= *lens--;
+					UNIV_PREFETCH_R(lens);
+				}
+			}
+			prefix_len += len;
+		}
+	}
+
+	UNIV_PREFETCH_R(rec + prefix_len);
+
+	prefix_len += rec - (lens + 1);
+
+	if ((*buf == NULL) || (*buf_size < prefix_len)) {
+		if (*buf != NULL) {
+			mem_free(*buf);
+		}
+
+		*buf = mem_alloc2(prefix_len, buf_size);
+	}
+
+	memcpy(*buf, lens + 1, prefix_len);
+
+	return(*buf + (rec - (lens + 1)));
+}
+
+/*******************************************************************
+Validates the consistency of an old-style physical record. */
+static
+ibool
+rec_validate_old(
+/*=============*/
+				/* out: TRUE if ok */
+	const rec_t*	rec)	/* in: physical record */
+{
+	const byte*	data;
+	ulint		len;
+	ulint		n_fields;
+	ulint		len_sum		= 0;
+	ulint		sum		= 0;
+	ulint		i;
+
+	ut_a(rec);
+	n_fields = rec_get_n_fields_old(rec);
+
+	if ((n_fields == 0) || (n_fields > REC_MAX_N_FIELDS)) {
+		fprintf(stderr, "InnoDB: Error: record has %lu fields\n",
+			(ulong) n_fields);
+		return(FALSE);
+	}
+
+	for (i = 0; i < n_fields; i++) {
+		data = rec_get_nth_field_old(rec, i, &len);
+
+		if (!((len < UNIV_PAGE_SIZE) || (len == UNIV_SQL_NULL))) {
+			fprintf(stderr,
+				"InnoDB: Error: record field %lu len %lu\n",
+				(ulong) i,
+				(ulong) len);
+			return(FALSE);
+		}
+
+		if (len != UNIV_SQL_NULL) {
+			len_sum += len;
+			sum += *(data + len -1); /* dereference the
+						 end of the field to
+						 cause a memory trap
+						 if possible */
+		} else {
+			len_sum += rec_get_nth_field_size(rec, i);
+		}
+	}
+
+	if (len_sum != rec_get_data_size_old(rec)) {
+		fprintf(stderr,
+			"InnoDB: Error: record len should be %lu, len %lu\n",
+			(ulong) len_sum,
+			rec_get_data_size_old(rec));
+		return(FALSE);
+	}
+
+	rec_dummy = sum; /* This is here only to fool the compiler */
+
+	return(TRUE);
+}
+
+/*******************************************************************
+Validates the consistency of a physical record. */
+UNIV_INTERN
+ibool
+rec_validate(
+/*=========*/
+				/* out: TRUE if ok */
+	const rec_t*	rec,	/* in: physical record */
+	const ulint*	offsets)/* in: array returned by rec_get_offsets() */
+{
+	const byte*	data;
+	ulint		len;
+	ulint		n_fields;
+	ulint		len_sum		= 0;
+	ulint		sum		= 0;
+	ulint		i;
+
+	ut_a(rec);
+	n_fields = rec_offs_n_fields(offsets);
+
+	if ((n_fields == 0) || (n_fields > REC_MAX_N_FIELDS)) {
+		fprintf(stderr, "InnoDB: Error: record has %lu fields\n",
+			(ulong) n_fields);
+		return(FALSE);
+	}
+
+	ut_a(rec_offs_comp(offsets) || n_fields <= rec_get_n_fields_old(rec));
+
+	for (i = 0; i < n_fields; i++) {
+		data = rec_get_nth_field(rec, offsets, i, &len);
+
+		if (!((len < UNIV_PAGE_SIZE) || (len == UNIV_SQL_NULL))) {
+			fprintf(stderr,
+				"InnoDB: Error: record field %lu len %lu\n",
+				(ulong) i,
+				(ulong) len);
+			return(FALSE);
+		}
+
+		if (len != UNIV_SQL_NULL) {
+			len_sum += len;
+			sum += *(data + len -1); /* dereference the
+						 end of the field to
+						 cause a memory trap
+						 if possible */
+		} else if (!rec_offs_comp(offsets)) {
+			len_sum += rec_get_nth_field_size(rec, i);
+		}
+	}
+
+	if (len_sum != rec_offs_data_size(offsets)) {
+		fprintf(stderr,
+			"InnoDB: Error: record len should be %lu, len %lu\n",
+			(ulong) len_sum,
+			(ulong) rec_offs_data_size(offsets));
+		return(FALSE);
+	}
+
+	rec_dummy = sum; /* This is here only to fool the compiler */
+
+	if (!rec_offs_comp(offsets)) {
+		ut_a(rec_validate_old(rec));
+	}
+
+	return(TRUE);
+}
+
+/*******************************************************************
+Prints an old-style physical record. */
+UNIV_INTERN
+void
+rec_print_old(
+/*==========*/
+	FILE*		file,	/* in: file where to print */
+	const rec_t*	rec)	/* in: physical record */
+{
+	const byte*	data;
+	ulint		len;
+	ulint		n;
+	ulint		i;
+
+	ut_ad(rec);
+
+	n = rec_get_n_fields_old(rec);
+
+	fprintf(file, "PHYSICAL RECORD: n_fields %lu;"
+		" %u-byte offsets; info bits %lu\n",
+		(ulong) n,
+		rec_get_1byte_offs_flag(rec) ? 1 : 2,
+		(ulong) rec_get_info_bits(rec, FALSE));
+
+	for (i = 0; i < n; i++) {
+
+		data = rec_get_nth_field_old(rec, i, &len);
+
+		fprintf(file, " %lu:", (ulong) i);
+
+		if (len != UNIV_SQL_NULL) {
+			if (len <= 30) {
+
+				ut_print_buf(file, data, len);
+			} else {
+				ut_print_buf(file, data, 30);
+
+				fprintf(file, " (total %lu bytes)",
+					(ulong) len);
+			}
+		} else {
+			fprintf(file, " SQL NULL, size %lu ",
+				rec_get_nth_field_size(rec, i));
+		}
+
+		putc(';', file);
+		putc('\n', file);
+	}
+
+	rec_validate_old(rec);
+}
+
+/*******************************************************************
+Prints a physical record in ROW_FORMAT=COMPACT.  Ignores the
+record header. */
+UNIV_INTERN
+void
+rec_print_comp(
+/*===========*/
+	FILE*		file,	/* in: file where to print */
+	const rec_t*	rec,	/* in: physical record */
+	const ulint*	offsets)/* in: array returned by rec_get_offsets() */
+{
+	ulint	i;
+
+	for (i = 0; i < rec_offs_n_fields(offsets); i++) {
+		const byte*	data;
+		ulint		len;
+
+		data = rec_get_nth_field(rec, offsets, i, &len);
+
+		fprintf(file, " %lu:", (ulong) i);
+
+		if (len != UNIV_SQL_NULL) {
+			if (len <= 30) {
+
+				ut_print_buf(file, data, len);
+			} else {
+				ut_print_buf(file, data, 30);
+
+				fprintf(file, " (total %lu bytes)",
+					(ulong) len);
+			}
+		} else {
+			fputs(" SQL NULL", file);
+		}
+		putc(';', file);
+		putc('\n', file);
+	}
+}
+
+/*******************************************************************
+Prints a physical record. */
+UNIV_INTERN
+void
+rec_print_new(
+/*==========*/
+	FILE*		file,	/* in: file where to print */
+	const rec_t*	rec,	/* in: physical record */
+	const ulint*	offsets)/* in: array returned by rec_get_offsets() */
+{
+	ut_ad(rec);
+	ut_ad(offsets);
+	ut_ad(rec_offs_validate(rec, NULL, offsets));
+
+	if (!rec_offs_comp(offsets)) {
+		rec_print_old(file, rec);
+		return;
+	}
+
+	fprintf(file, "PHYSICAL RECORD: n_fields %lu;"
+		" compact format; info bits %lu\n",
+		(ulong) rec_offs_n_fields(offsets),
+		(ulong) rec_get_info_bits(rec, TRUE));
+
+	rec_print_comp(file, rec, offsets);
+	rec_validate(rec, offsets);
+}
+
+/*******************************************************************
+Prints a physical record. */
+UNIV_INTERN
+void
+rec_print(
+/*======*/
+	FILE*		file,	/* in: file where to print */
+	const rec_t*	rec,	/* in: physical record */
+	dict_index_t*	index)	/* in: record descriptor */
+{
+	ut_ad(index);
+
+	if (!dict_table_is_comp(index->table)) {
+		rec_print_old(file, rec);
+		return;
+	} else {
+		mem_heap_t*	heap	= NULL;
+		ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+		rec_offs_init(offsets_);
+
+		rec_print_new(file, rec,
+			      rec_get_offsets(rec, index, offsets_,
+					      ULINT_UNDEFINED, &heap));
+		if (UNIV_LIKELY_NULL(heap)) {
+			mem_heap_free(heap);
+		}
+	}
+}
diff --git a/storage/xtradb/row/row0ext.c b/storage/xtradb/row/row0ext.c
new file mode 100644
index 00000000000..83dfa024ffc
--- /dev/null
+++ b/storage/xtradb/row/row0ext.c
@@ -0,0 +1,114 @@
+/*****************************************************************************
+
+Copyright (c) 2006, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Caching of externally stored column prefixes
+
+Created September 2006 Marko Makela
+*******************************************************/
+
+#include "row0ext.h"
+
+#ifdef UNIV_NONINL
+#include "row0ext.ic"
+#endif
+
+#include "btr0cur.h"
+
+/************************************************************************
+Fills the column prefix cache of an externally stored column. */
+static
+void
+row_ext_cache_fill(
+/*===============*/
+	row_ext_t*	ext,	/* in/out: column prefix cache */
+	ulint		i,	/* in: index of ext->ext[] */
+	ulint		zip_size,/* compressed page size in bytes, or 0 */
+	const dfield_t*	dfield)	/* in: data field */
+{
+	const byte*	field	= dfield_get_data(dfield);
+	ulint		f_len	= dfield_get_len(dfield);
+	byte*		buf	= ext->buf + i * REC_MAX_INDEX_COL_LEN;
+
+	ut_ad(i < ext->n_ext);
+	ut_ad(dfield_is_ext(dfield));
+	ut_a(f_len >= BTR_EXTERN_FIELD_REF_SIZE);
+
+	if (UNIV_UNLIKELY(!memcmp(field_ref_zero,
+				  field + f_len - BTR_EXTERN_FIELD_REF_SIZE,
+				  BTR_EXTERN_FIELD_REF_SIZE))) {
+		/* The BLOB pointer is not set: we cannot fetch it */
+		ext->len[i] = 0;
+	} else {
+		/* Fetch at most REC_MAX_INDEX_COL_LEN of the column.
+		The column should be non-empty.  However,
+		trx_rollback_or_clean_all_recovered() may try to
+		access a half-deleted BLOB if the server previously
+		crashed during the execution of
+		btr_free_externally_stored_field(). */
+		ext->len[i] = btr_copy_externally_stored_field_prefix(
+			buf, REC_MAX_INDEX_COL_LEN, zip_size, field, f_len);
+	}
+}
+
+/************************************************************************
+Creates a cache of column prefixes of externally stored columns. */
+UNIV_INTERN
+row_ext_t*
+row_ext_create(
+/*===========*/
+				/* out,own: column prefix cache */
+	ulint		n_ext,	/* in: number of externally stored columns */
+	const ulint*	ext,	/* in: col_no's of externally stored columns
+				in the InnoDB table object, as reported by
+				dict_col_get_no(); NOT relative to the records
+				in the clustered index */
+	const dtuple_t*	tuple,	/* in: data tuple containing the field
+				references of the externally stored
+				columns; must be indexed by col_no;
+				the clustered index record must be
+				covered by a lock or a page latch
+				to prevent deletion (rollback or purge). */
+	ulint		zip_size,/* compressed page size in bytes, or 0 */
+	mem_heap_t*	heap)	/* in: heap where created */
+{
+	ulint		i;
+	row_ext_t*	ret = mem_heap_alloc(heap, (sizeof *ret)
+					     + (n_ext - 1) * sizeof ret->len);
+
+	ut_ad(ut_is_2pow(zip_size));
+	ut_ad(zip_size <= UNIV_PAGE_SIZE);
+
+	ret->n_ext = n_ext;
+	ret->ext = ext;
+	ret->buf = mem_heap_alloc(heap, n_ext * REC_MAX_INDEX_COL_LEN);
+#ifdef UNIV_DEBUG
+	memset(ret->buf, 0xaa, n_ext * REC_MAX_INDEX_COL_LEN);
+	UNIV_MEM_ALLOC(ret->buf, n_ext * REC_MAX_INDEX_COL_LEN);
+#endif
+
+	/* Fetch the BLOB prefixes */
+	for (i = 0; i < n_ext; i++) {
+		const dfield_t*	dfield;
+
+		dfield = dtuple_get_nth_field(tuple, ext[i]);
+		row_ext_cache_fill(ret, i, zip_size, dfield);
+	}
+
+	return(ret);
+}
diff --git a/storage/xtradb/row/row0ins.c b/storage/xtradb/row/row0ins.c
new file mode 100644
index 00000000000..ceb0f7d75a2
--- /dev/null
+++ b/storage/xtradb/row/row0ins.c
@@ -0,0 +1,2558 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Insert into a table
+
+Created 4/20/1996 Heikki Tuuri
+*******************************************************/
+
+#include "row0ins.h"
+
+#ifdef UNIV_NONINL
+#include "row0ins.ic"
+#endif
+
+#include "dict0dict.h"
+#include "dict0boot.h"
+#include "trx0undo.h"
+#include "btr0btr.h"
+#include "btr0cur.h"
+#include "mach0data.h"
+#include "que0que.h"
+#include "row0upd.h"
+#include "row0sel.h"
+#include "row0row.h"
+#include "rem0cmp.h"
+#include "lock0lock.h"
+#include "log0log.h"
+#include "eval0eval.h"
+#include "data0data.h"
+#include "usr0sess.h"
+#include "buf0lru.h"
+
+#define	ROW_INS_PREV	1
+#define	ROW_INS_NEXT	2
+
+
+/*********************************************************************
+This prototype is copied from /mysql/sql/ha_innodb.cc.
+Invalidates the MySQL query cache for the table.
+NOTE that the exact prototype of this function has to be in
+/xtradb/row/row0ins.c! */
+extern
+void
+innobase_invalidate_query_cache(
+/*============================*/
+	trx_t*	trx,		/* in: transaction which modifies the table */
+	char*	full_name,	/* in: concatenation of database name, null
+				char '\0', table name, null char'\0';
+				NOTE that in Windows this is always
+				in LOWER CASE! */
+	ulint	full_name_len);	/* in: full name length where also the null
+				chars count */
+
+/*************************************************************************
+Creates an insert node struct. */
+UNIV_INTERN
+ins_node_t*
+ins_node_create(
+/*============*/
+					/* out, own: insert node struct */
+	ulint		ins_type,	/* in: INS_VALUES, ... */
+	dict_table_t*	table,		/* in: table where to insert */
+	mem_heap_t*	heap)		/* in: mem heap where created */
+{
+	ins_node_t*	node;
+
+	node = mem_heap_alloc(heap, sizeof(ins_node_t));
+
+	node->common.type = QUE_NODE_INSERT;
+
+	node->ins_type = ins_type;
+
+	node->state = INS_NODE_SET_IX_LOCK;
+	node->table = table;
+	node->index = NULL;
+	node->entry = NULL;
+
+	node->select = NULL;
+
+	node->trx_id = ut_dulint_zero;
+
+	node->entry_sys_heap = mem_heap_create(128);
+
+	node->magic_n = INS_NODE_MAGIC_N;
+
+	return(node);
+}
+
+/***************************************************************
+Creates an entry template for each index of a table. */
+UNIV_INTERN
+void
+ins_node_create_entry_list(
+/*=======================*/
+	ins_node_t*	node)	/* in: row insert node */
+{
+	dict_index_t*	index;
+	dtuple_t*	entry;
+
+	ut_ad(node->entry_sys_heap);
+
+	UT_LIST_INIT(node->entry_list);
+
+	index = dict_table_get_first_index(node->table);
+
+	while (index != NULL) {
+		entry = row_build_index_entry(node->row, NULL, index,
+					      node->entry_sys_heap);
+		UT_LIST_ADD_LAST(tuple_list, node->entry_list, entry);
+
+		index = dict_table_get_next_index(index);
+	}
+}
+
+/*********************************************************************
+Adds system field buffers to a row. */
+static
+void
+row_ins_alloc_sys_fields(
+/*=====================*/
+	ins_node_t*	node)	/* in: insert node */
+{
+	dtuple_t*		row;
+	dict_table_t*		table;
+	mem_heap_t*		heap;
+	const dict_col_t*	col;
+	dfield_t*		dfield;
+	byte*			ptr;
+
+	row = node->row;
+	table = node->table;
+	heap = node->entry_sys_heap;
+
+	ut_ad(row && table && heap);
+	ut_ad(dtuple_get_n_fields(row) == dict_table_get_n_cols(table));
+
+	/* 1. Allocate buffer for row id */
+
+	col = dict_table_get_sys_col(table, DATA_ROW_ID);
+
+	dfield = dtuple_get_nth_field(row, dict_col_get_no(col));
+
+	ptr = mem_heap_alloc(heap, DATA_ROW_ID_LEN);
+
+	dfield_set_data(dfield, ptr, DATA_ROW_ID_LEN);
+
+	node->row_id_buf = ptr;
+
+	/* 3. Allocate buffer for trx id */
+
+	col = dict_table_get_sys_col(table, DATA_TRX_ID);
+
+	dfield = dtuple_get_nth_field(row, dict_col_get_no(col));
+	ptr = mem_heap_alloc(heap, DATA_TRX_ID_LEN);
+
+	dfield_set_data(dfield, ptr, DATA_TRX_ID_LEN);
+
+	node->trx_id_buf = ptr;
+
+	/* 4. Allocate buffer for roll ptr */
+
+	col = dict_table_get_sys_col(table, DATA_ROLL_PTR);
+
+	dfield = dtuple_get_nth_field(row, dict_col_get_no(col));
+	ptr = mem_heap_alloc(heap, DATA_ROLL_PTR_LEN);
+
+	dfield_set_data(dfield, ptr, DATA_ROLL_PTR_LEN);
+}
+
+/*************************************************************************
+Sets a new row to insert for an INS_DIRECT node. This function is only used
+if we have constructed the row separately, which is a rare case; this
+function is quite slow. */
+UNIV_INTERN
+void
+ins_node_set_new_row(
+/*=================*/
+	ins_node_t*	node,	/* in: insert node */
+	dtuple_t*	row)	/* in: new row (or first row) for the node */
+{
+	node->state = INS_NODE_SET_IX_LOCK;
+	node->index = NULL;
+	node->entry = NULL;
+
+	node->row = row;
+
+	mem_heap_empty(node->entry_sys_heap);
+
+	/* Create templates for index entries */
+
+	ins_node_create_entry_list(node);
+
+	/* Allocate from entry_sys_heap buffers for sys fields */
+
+	row_ins_alloc_sys_fields(node);
+
+	/* As we allocated a new trx id buf, the trx id should be written
+	there again: */
+
+	node->trx_id = ut_dulint_zero;
+}
+
+/***********************************************************************
+Does an insert operation by updating a delete-marked existing record
+in the index. This situation can occur if the delete-marked record is
+kept in the index for consistent reads. */
+static
+ulint
+row_ins_sec_index_entry_by_modify(
+/*==============================*/
+				/* out: DB_SUCCESS or error code */
+	ulint		mode,	/* in: BTR_MODIFY_LEAF or BTR_MODIFY_TREE,
+				depending on whether mtr holds just a leaf
+				latch or also a tree latch */
+	btr_cur_t*	cursor,	/* in: B-tree cursor */
+	const dtuple_t*	entry,	/* in: index entry to insert */
+	que_thr_t*	thr,	/* in: query thread */
+	mtr_t*		mtr)	/* in: mtr; must be committed before
+				latching any further pages */
+{
+	big_rec_t*	dummy_big_rec;
+	mem_heap_t*	heap;
+	upd_t*		update;
+	rec_t*		rec;
+	ulint		err;
+
+	rec = btr_cur_get_rec(cursor);
+
+	ut_ad(!dict_index_is_clust(cursor->index));
+	ut_ad(rec_get_deleted_flag(rec,
+				   dict_table_is_comp(cursor->index->table)));
+
+	/* We know that in the alphabetical ordering, entry and rec are
+	identified. But in their binary form there may be differences if
+	there are char fields in them. Therefore we have to calculate the
+	difference. */
+
+	heap = mem_heap_create(1024);
+
+	update = row_upd_build_sec_rec_difference_binary(
+		cursor->index, entry, rec, thr_get_trx(thr), heap);
+	if (mode == BTR_MODIFY_LEAF) {
+		/* Try an optimistic updating of the record, keeping changes
+		within the page */
+
+		err = btr_cur_optimistic_update(BTR_KEEP_SYS_FLAG, cursor,
+						update, 0, thr, mtr);
+		switch (err) {
+		case DB_OVERFLOW:
+		case DB_UNDERFLOW:
+		case DB_ZIP_OVERFLOW:
+			err = DB_FAIL;
+		}
+	} else {
+		ut_a(mode == BTR_MODIFY_TREE);
+		if (buf_LRU_buf_pool_running_out()) {
+
+			err = DB_LOCK_TABLE_FULL;
+
+			goto func_exit;
+		}
+
+		err = btr_cur_pessimistic_update(BTR_KEEP_SYS_FLAG, cursor,
+						 &heap, &dummy_big_rec, update,
+						 0, thr, mtr);
+		ut_ad(!dummy_big_rec);
+	}
+func_exit:
+	mem_heap_free(heap);
+
+	return(err);
+}
+
+/***********************************************************************
+Does an insert operation by delete unmarking and updating a delete marked
+existing record in the index. This situation can occur if the delete marked
+record is kept in the index for consistent reads. */
+static
+ulint
+row_ins_clust_index_entry_by_modify(
+/*================================*/
+				/* out: DB_SUCCESS, DB_FAIL, or error code */
+	ulint		mode,	/* in: BTR_MODIFY_LEAF or BTR_MODIFY_TREE,
+				depending on whether mtr holds just a leaf
+				latch or also a tree latch */
+	btr_cur_t*	cursor,	/* in: B-tree cursor */
+	mem_heap_t**	heap,	/* in/out: pointer to memory heap, or NULL */
+	big_rec_t**	big_rec,/* out: possible big rec vector of fields
+				which have to be stored externally by the
+				caller */
+	const dtuple_t*	entry,	/* in: index entry to insert */
+	que_thr_t*	thr,	/* in: query thread */
+	mtr_t*		mtr)	/* in: mtr; must be committed before
+				latching any further pages */
+{
+	rec_t*		rec;
+	upd_t*		update;
+	ulint		err;
+
+	ut_ad(dict_index_is_clust(cursor->index));
+
+	*big_rec = NULL;
+
+	rec = btr_cur_get_rec(cursor);
+
+	ut_ad(rec_get_deleted_flag(rec,
+				   dict_table_is_comp(cursor->index->table)));
+
+	if (!*heap) {
+		*heap = mem_heap_create(1024);
+	}
+
+	/* Build an update vector containing all the fields to be modified;
+	NOTE that this vector may NOT contain system columns trx_id or
+	roll_ptr */
+
+	update = row_upd_build_difference_binary(cursor->index, entry, rec,
+						 thr_get_trx(thr), *heap);
+	if (mode == BTR_MODIFY_LEAF) {
+		/* Try optimistic updating of the record, keeping changes
+		within the page */
+
+		err = btr_cur_optimistic_update(0, cursor, update, 0, thr,
+						mtr);
+		switch (err) {
+		case DB_OVERFLOW:
+		case DB_UNDERFLOW:
+		case DB_ZIP_OVERFLOW:
+			err = DB_FAIL;
+		}
+	} else {
+		ut_a(mode == BTR_MODIFY_TREE);
+		if (buf_LRU_buf_pool_running_out()) {
+
+			return(DB_LOCK_TABLE_FULL);
+
+		}
+		err = btr_cur_pessimistic_update(0, cursor,
+						 heap, big_rec, update,
+						 0, thr, mtr);
+	}
+
+	return(err);
+}
+
+/*************************************************************************
+Returns TRUE if in a cascaded update/delete an ancestor node of node
+updates (not DELETE, but UPDATE) table. */
+static
+ibool
+row_ins_cascade_ancestor_updates_table(
+/*===================================*/
+				/* out: TRUE if an ancestor updates table */
+	que_node_t*	node,	/* in: node in a query graph */
+	dict_table_t*	table)	/* in: table */
+{
+	que_node_t*	parent;
+	upd_node_t*	upd_node;
+
+	parent = que_node_get_parent(node);
+
+	while (que_node_get_type(parent) == QUE_NODE_UPDATE) {
+
+		upd_node = parent;
+
+		if (upd_node->table == table && upd_node->is_delete == FALSE) {
+
+			return(TRUE);
+		}
+
+		parent = que_node_get_parent(parent);
+
+		ut_a(parent);
+	}
+
+	return(FALSE);
+}
+
+/*************************************************************************
+Returns the number of ancestor UPDATE or DELETE nodes of a
+cascaded update/delete node. */
+static
+ulint
+row_ins_cascade_n_ancestors(
+/*========================*/
+				/* out: number of ancestors */
+	que_node_t*	node)	/* in: node in a query graph */
+{
+	que_node_t*	parent;
+	ulint		n_ancestors = 0;
+
+	parent = que_node_get_parent(node);
+
+	while (que_node_get_type(parent) == QUE_NODE_UPDATE) {
+		n_ancestors++;
+
+		parent = que_node_get_parent(parent);
+
+		ut_a(parent);
+	}
+
+	return(n_ancestors);
+}
+
+/**********************************************************************
+Calculates the update vector node->cascade->update for a child table in
+a cascaded update. */
+static
+ulint
+row_ins_cascade_calc_update_vec(
+/*============================*/
+					/* out: number of fields in the
+					calculated update vector; the value
+					can also be 0 if no foreign key
+					fields changed; the returned value
+					is ULINT_UNDEFINED if the column
+					type in the child table is too short
+					to fit the new value in the parent
+					table: that means the update fails */
+	upd_node_t*	node,		/* in: update node of the parent
+					table */
+	dict_foreign_t*	foreign,	/* in: foreign key constraint whose
+					type is != 0 */
+	mem_heap_t*	heap)		/* in: memory heap to use as
+					temporary storage */
+{
+	upd_node_t*	cascade		= node->cascade_node;
+	dict_table_t*	table		= foreign->foreign_table;
+	dict_index_t*	index		= foreign->foreign_index;
+	upd_t*		update;
+	upd_field_t*	ufield;
+	dict_table_t*	parent_table;
+	dict_index_t*	parent_index;
+	upd_t*		parent_update;
+	upd_field_t*	parent_ufield;
+	ulint		n_fields_updated;
+	ulint		parent_field_no;
+	ulint		i;
+	ulint		j;
+
+	ut_a(node);
+	ut_a(foreign);
+	ut_a(cascade);
+	ut_a(table);
+	ut_a(index);
+
+	/* Calculate the appropriate update vector which will set the fields
+	in the child index record to the same value (possibly padded with
+	spaces if the column is a fixed length CHAR or FIXBINARY column) as
+	the referenced index record will get in the update. */
+
+	parent_table = node->table;
+	ut_a(parent_table == foreign->referenced_table);
+	parent_index = foreign->referenced_index;
+	parent_update = node->update;
+
+	update = cascade->update;
+
+	update->info_bits = 0;
+	update->n_fields = foreign->n_fields;
+
+	n_fields_updated = 0;
+
+	for (i = 0; i < foreign->n_fields; i++) {
+
+		parent_field_no = dict_table_get_nth_col_pos(
+			parent_table,
+			dict_index_get_nth_col_no(parent_index, i));
+
+		for (j = 0; j < parent_update->n_fields; j++) {
+			parent_ufield = parent_update->fields + j;
+
+			if (parent_ufield->field_no == parent_field_no) {
+
+				ulint			min_size;
+				const dict_col_t*	col;
+				ulint			ufield_len;
+
+				col = dict_index_get_nth_col(index, i);
+
+				/* A field in the parent index record is
+				updated. Let us make the update vector
+				field for the child table. */
+
+				ufield = update->fields + n_fields_updated;
+
+				ufield->field_no
+					= dict_table_get_nth_col_pos(
+					table, dict_col_get_no(col));
+				ufield->exp = NULL;
+
+				ufield->new_val = parent_ufield->new_val;
+				ufield_len = dfield_get_len(&ufield->new_val);
+
+				/* Clear the "external storage" flag */
+				dfield_set_len(&ufield->new_val, ufield_len);
+
+				/* Do not allow a NOT NULL column to be
+				updated as NULL */
+
+				if (dfield_is_null(&ufield->new_val)
+				    && (col->prtype & DATA_NOT_NULL)) {
+
+					return(ULINT_UNDEFINED);
+				}
+
+				/* If the new value would not fit in the
+				column, do not allow the update */
+
+				if (!dfield_is_null(&ufield->new_val)
+				    && dtype_get_at_most_n_mbchars(
+					col->prtype,
+					col->mbminlen, col->mbmaxlen,
+					col->len,
+					ufield_len,
+					dfield_get_data(&ufield->new_val))
+				    < ufield_len) {
+
+					return(ULINT_UNDEFINED);
+				}
+
+				/* If the parent column type has a different
+				length than the child column type, we may
+				need to pad with spaces the new value of the
+				child column */
+
+				min_size = dict_col_get_min_size(col);
+
+				/* Because UNIV_SQL_NULL (the marker
+				of SQL NULL values) exceeds all possible
+				values of min_size, the test below will
+				not hold for SQL NULL columns. */
+
+				if (min_size > ufield_len) {
+
+					char*		pad_start;
+					const char*	pad_end;
+					char*		padded_data
+						= mem_heap_alloc(
+							heap, min_size);
+					pad_start = padded_data + ufield_len;
+					pad_end = padded_data + min_size;
+
+					memcpy(padded_data,
+					       dfield_get_data(&ufield
+							       ->new_val),
+					       dfield_get_len(&ufield
+							      ->new_val));
+
+					switch (UNIV_EXPECT(col->mbminlen,1)) {
+					default:
+						ut_error;
+						return(ULINT_UNDEFINED);
+					case 1:
+						if (UNIV_UNLIKELY
+						    (dtype_get_charset_coll(
+							    col->prtype)
+						     == DATA_MYSQL_BINARY_CHARSET_COLL)) {
+							/* Do not pad BINARY
+							columns. */
+							return(ULINT_UNDEFINED);
+						}
+
+						/* space=0x20 */
+						memset(pad_start, 0x20,
+						       pad_end - pad_start);
+						break;
+					case 2:
+						/* space=0x0020 */
+						ut_a(!(ufield_len % 2));
+						ut_a(!(min_size % 2));
+						do {
+							*pad_start++ = 0x00;
+							*pad_start++ = 0x20;
+						} while (pad_start < pad_end);
+						break;
+					}
+
+					dfield_set_data(&ufield->new_val,
+							padded_data, min_size);
+				}
+
+				n_fields_updated++;
+			}
+		}
+	}
+
+	update->n_fields = n_fields_updated;
+
+	return(n_fields_updated);
+}
+
+/*************************************************************************
+Set detailed error message associated with foreign key errors for
+the given transaction. */
+static
+void
+row_ins_set_detailed(
+/*=================*/
+	trx_t*		trx,		/* in: transaction */
+	dict_foreign_t*	foreign)	/* in: foreign key constraint */
+{
+	mutex_enter(&srv_misc_tmpfile_mutex);
+	rewind(srv_misc_tmpfile);
+
+	if (os_file_set_eof(srv_misc_tmpfile)) {
+		ut_print_name(srv_misc_tmpfile, trx, TRUE,
+			      foreign->foreign_table_name);
+		dict_print_info_on_foreign_key_in_create_format(
+			srv_misc_tmpfile, trx, foreign, FALSE);
+		trx_set_detailed_error_from_file(trx, srv_misc_tmpfile);
+	} else {
+		trx_set_detailed_error(trx, "temp file operation failed");
+	}
+
+	mutex_exit(&srv_misc_tmpfile_mutex);
+}
+
+/*************************************************************************
+Reports a foreign key error associated with an update or a delete of a
+parent table index entry. */
+static
+void
+row_ins_foreign_report_err(
+/*=======================*/
+	const char*	errstr,		/* in: error string from the viewpoint
+					of the parent table */
+	que_thr_t*	thr,		/* in: query thread whose run_node
+					is an update node */
+	dict_foreign_t*	foreign,	/* in: foreign key constraint */
+	const rec_t*	rec,		/* in: a matching index record in the
+					child table */
+	const dtuple_t*	entry)		/* in: index entry in the parent
+					table */
+{
+	FILE*	ef	= dict_foreign_err_file;
+	trx_t*	trx	= thr_get_trx(thr);
+
+	row_ins_set_detailed(trx, foreign);
+
+	mutex_enter(&dict_foreign_err_mutex);
+	rewind(ef);
+	ut_print_timestamp(ef);
+	fputs(" Transaction:\n", ef);
+	trx_print(ef, trx, 600);
+
+	fputs("Foreign key constraint fails for table ", ef);
+	ut_print_name(ef, trx, TRUE, foreign->foreign_table_name);
+	fputs(":\n", ef);
+	dict_print_info_on_foreign_key_in_create_format(ef, trx, foreign,
+							TRUE);
+	putc('\n', ef);
+	fputs(errstr, ef);
+	fputs(" in parent table, in index ", ef);
+	ut_print_name(ef, trx, FALSE, foreign->referenced_index->name);
+	if (entry) {
+		fputs(" tuple:\n", ef);
+		dtuple_print(ef, entry);
+	}
+	fputs("\nBut in child table ", ef);
+	ut_print_name(ef, trx, TRUE, foreign->foreign_table_name);
+	fputs(", in index ", ef);
+	ut_print_name(ef, trx, FALSE, foreign->foreign_index->name);
+	if (rec) {
+		fputs(", there is a record:\n", ef);
+		rec_print(ef, rec, foreign->foreign_index);
+	} else {
+		fputs(", the record is not available\n", ef);
+	}
+	putc('\n', ef);
+
+	mutex_exit(&dict_foreign_err_mutex);
+}
+
+/*************************************************************************
+Reports a foreign key error to dict_foreign_err_file when we are trying
+to add an index entry to a child table. Note that the adding may be the result
+of an update, too. */
+static
+void
+row_ins_foreign_report_add_err(
+/*===========================*/
+	trx_t*		trx,		/* in: transaction */
+	dict_foreign_t*	foreign,	/* in: foreign key constraint */
+	const rec_t*	rec,		/* in: a record in the parent table:
+					it does not match entry because we
+					have an error! */
+	const dtuple_t*	entry)		/* in: index entry to insert in the
+					child table */
+{
+	FILE*	ef	= dict_foreign_err_file;
+
+	row_ins_set_detailed(trx, foreign);
+
+	mutex_enter(&dict_foreign_err_mutex);
+	rewind(ef);
+	ut_print_timestamp(ef);
+	fputs(" Transaction:\n", ef);
+	trx_print(ef, trx, 600);
+	fputs("Foreign key constraint fails for table ", ef);
+	ut_print_name(ef, trx, TRUE, foreign->foreign_table_name);
+	fputs(":\n", ef);
+	dict_print_info_on_foreign_key_in_create_format(ef, trx, foreign,
+							TRUE);
+	fputs("\nTrying to add in child table, in index ", ef);
+	ut_print_name(ef, trx, FALSE, foreign->foreign_index->name);
+	if (entry) {
+		fputs(" tuple:\n", ef);
+		/* TODO: DB_TRX_ID and DB_ROLL_PTR may be uninitialized.
+		It would be better to only display the user columns. */
+		dtuple_print(ef, entry);
+	}
+	fputs("\nBut in parent table ", ef);
+	ut_print_name(ef, trx, TRUE, foreign->referenced_table_name);
+	fputs(", in index ", ef);
+	ut_print_name(ef, trx, FALSE, foreign->referenced_index->name);
+	fputs(",\nthe closest match we can find is record:\n", ef);
+	if (rec && page_rec_is_supremum(rec)) {
+		/* If the cursor ended on a supremum record, it is better
+		to report the previous record in the error message, so that
+		the user gets a more descriptive error message. */
+		rec = page_rec_get_prev_const(rec);
+	}
+
+	if (rec) {
+		rec_print(ef, rec, foreign->referenced_index);
+	}
+	putc('\n', ef);
+
+	mutex_exit(&dict_foreign_err_mutex);
+}
+
+/*************************************************************************
+Invalidate the query cache for the given table. */
+static
+void
+row_ins_invalidate_query_cache(
+/*===========================*/
+	que_thr_t*	thr,		/* in: query thread whose run_node
+					is an update node */
+	const char*	name)		/* in: table name prefixed with
+					database name and a '/' character */
+{
+	char*	buf;
+	char*	ptr;
+	ulint	len = strlen(name) + 1;
+
+	buf = mem_strdupl(name, len);
+
+	ptr = strchr(buf, '/');
+	ut_a(ptr);
+	*ptr = '\0';
+
+	/* We call a function in ha_innodb.cc */
+#ifndef UNIV_HOTBACKUP
+	innobase_invalidate_query_cache(thr_get_trx(thr), buf, len);
+#endif
+	mem_free(buf);
+}
+
+/*************************************************************************
+Perform referential actions or checks when a parent row is deleted or updated
+and the constraint had an ON DELETE or ON UPDATE condition which was not
+RESTRICT. */
+static
+ulint
+row_ins_foreign_check_on_constraint(
+/*================================*/
+					/* out: DB_SUCCESS, DB_LOCK_WAIT,
+					or error code */
+	que_thr_t*	thr,		/* in: query thread whose run_node
+					is an update node */
+	dict_foreign_t*	foreign,	/* in: foreign key constraint whose
+					type is != 0 */
+	btr_pcur_t*	pcur,		/* in: cursor placed on a matching
+					index record in the child table */
+	dtuple_t*	entry,		/* in: index entry in the parent
+					table */
+	mtr_t*		mtr)		/* in: mtr holding the latch of pcur
+					page */
+{
+	upd_node_t*	node;
+	upd_node_t*	cascade;
+	dict_table_t*	table		= foreign->foreign_table;
+	dict_index_t*	index;
+	dict_index_t*	clust_index;
+	dtuple_t*	ref;
+	mem_heap_t*	upd_vec_heap	= NULL;
+	const rec_t*	rec;
+	const rec_t*	clust_rec;
+	const buf_block_t* clust_block;
+	upd_t*		update;
+	ulint		n_to_update;
+	ulint		err;
+	ulint		i;
+	trx_t*		trx;
+	mem_heap_t*	tmp_heap	= NULL;
+
+	ut_a(thr);
+	ut_a(foreign);
+	ut_a(pcur);
+	ut_a(mtr);
+
+	trx = thr_get_trx(thr);
+
+	/* Since we are going to delete or update a row, we have to invalidate
+	the MySQL query cache for table. A deadlock of threads is not possible
+	here because the caller of this function does not hold any latches with
+	the sync0sync.h rank above the kernel mutex. The query cache mutex has
+	a rank just above the kernel mutex. */
+
+	row_ins_invalidate_query_cache(thr, table->name);
+
+	node = thr->run_node;
+
+	if (node->is_delete && 0 == (foreign->type
+				     & (DICT_FOREIGN_ON_DELETE_CASCADE
+					| DICT_FOREIGN_ON_DELETE_SET_NULL))) {
+
+		row_ins_foreign_report_err("Trying to delete",
+					   thr, foreign,
+					   btr_pcur_get_rec(pcur), entry);
+
+		return(DB_ROW_IS_REFERENCED);
+	}
+
+	if (!node->is_delete && 0 == (foreign->type
+				      & (DICT_FOREIGN_ON_UPDATE_CASCADE
+					 | DICT_FOREIGN_ON_UPDATE_SET_NULL))) {
+
+		/* This is an UPDATE */
+
+		row_ins_foreign_report_err("Trying to update",
+					   thr, foreign,
+					   btr_pcur_get_rec(pcur), entry);
+
+		return(DB_ROW_IS_REFERENCED);
+	}
+
+	if (node->cascade_node == NULL) {
+		/* Extend our query graph by creating a child to current
+		update node. The child is used in the cascade or set null
+		operation. */
+
+		node->cascade_heap = mem_heap_create(128);
+		node->cascade_node = row_create_update_node_for_mysql(
+			table, node->cascade_heap);
+		que_node_set_parent(node->cascade_node, node);
+	}
+
+	/* Initialize cascade_node to do the operation we want. Note that we
+	use the SAME cascade node to do all foreign key operations of the
+	SQL DELETE: the table of the cascade node may change if there are
+	several child tables to the table where the delete is done! */
+
+	cascade = node->cascade_node;
+
+	cascade->table = table;
+
+	cascade->foreign = foreign;
+
+	if (node->is_delete
+	    && (foreign->type & DICT_FOREIGN_ON_DELETE_CASCADE)) {
+		cascade->is_delete = TRUE;
+	} else {
+		cascade->is_delete = FALSE;
+
+		if (foreign->n_fields > cascade->update_n_fields) {
+			/* We have to make the update vector longer */
+
+			cascade->update = upd_create(foreign->n_fields,
+						     node->cascade_heap);
+			cascade->update_n_fields = foreign->n_fields;
+		}
+	}
+
+	/* We do not allow cyclic cascaded updating (DELETE is allowed,
+	but not UPDATE) of the same table, as this can lead to an infinite
+	cycle. Check that we are not updating the same table which is
+	already being modified in this cascade chain. We have to check
+	this also because the modification of the indexes of a 'parent'
+	table may still be incomplete, and we must avoid seeing the indexes
+	of the parent table in an inconsistent state! */
+
+	if (!cascade->is_delete
+	    && row_ins_cascade_ancestor_updates_table(cascade, table)) {
+
+		/* We do not know if this would break foreign key
+		constraints, but play safe and return an error */
+
+		err = DB_ROW_IS_REFERENCED;
+
+		row_ins_foreign_report_err(
+			"Trying an update, possibly causing a cyclic"
+			" cascaded update\n"
+			"in the child table,", thr, foreign,
+			btr_pcur_get_rec(pcur), entry);
+
+		goto nonstandard_exit_func;
+	}
+
+	if (row_ins_cascade_n_ancestors(cascade) >= 15) {
+		err = DB_ROW_IS_REFERENCED;
+
+		row_ins_foreign_report_err(
+			"Trying a too deep cascaded delete or update\n",
+			thr, foreign, btr_pcur_get_rec(pcur), entry);
+
+		goto nonstandard_exit_func;
+	}
+
+	index = btr_pcur_get_btr_cur(pcur)->index;
+
+	ut_a(index == foreign->foreign_index);
+
+	rec = btr_pcur_get_rec(pcur);
+
+	if (dict_index_is_clust(index)) {
+		/* pcur is already positioned in the clustered index of
+		the child table */
+
+		clust_index = index;
+		clust_rec = rec;
+		clust_block = btr_pcur_get_block(pcur);
+	} else {
+		/* We have to look for the record in the clustered index
+		in the child table */
+
+		clust_index = dict_table_get_first_index(table);
+
+		tmp_heap = mem_heap_create(256);
+
+		ref = row_build_row_ref(ROW_COPY_POINTERS, index, rec,
+					tmp_heap);
+		btr_pcur_open_with_no_init(clust_index, ref,
+					   PAGE_CUR_LE, BTR_SEARCH_LEAF,
+					   cascade->pcur, 0, mtr);
+
+		clust_rec = btr_pcur_get_rec(cascade->pcur);
+		clust_block = btr_pcur_get_block(cascade->pcur);
+
+		if (!page_rec_is_user_rec(clust_rec)
+		    || btr_pcur_get_low_match(cascade->pcur)
+		    < dict_index_get_n_unique(clust_index)) {
+
+			fputs("InnoDB: error in cascade of a foreign key op\n"
+			      "InnoDB: ", stderr);
+			dict_index_name_print(stderr, trx, index);
+
+			fputs("\n"
+			      "InnoDB: record ", stderr);
+			rec_print(stderr, rec, index);
+			fputs("\n"
+			      "InnoDB: clustered record ", stderr);
+			rec_print(stderr, clust_rec, clust_index);
+			fputs("\n"
+			      "InnoDB: Submit a detailed bug report to"
+			      " http://bugs.mysql.com\n", stderr);
+
+			err = DB_SUCCESS;
+
+			goto nonstandard_exit_func;
+		}
+	}
+
+	/* Set an X-lock on the row to delete or update in the child table */
+
+	err = lock_table(0, table, LOCK_IX, thr);
+
+	if (err == DB_SUCCESS) {
+		/* Here it suffices to use a LOCK_REC_NOT_GAP type lock;
+		we already have a normal shared lock on the appropriate
+		gap if the search criterion was not unique */
+
+		err = lock_clust_rec_read_check_and_lock_alt(
+			0, clust_block, clust_rec, clust_index,
+			LOCK_X, LOCK_REC_NOT_GAP, thr);
+	}
+
+	if (err != DB_SUCCESS) {
+
+		goto nonstandard_exit_func;
+	}
+
+	if (rec_get_deleted_flag(clust_rec, dict_table_is_comp(table))) {
+		/* This can happen if there is a circular reference of
+		rows such that cascading delete comes to delete a row
+		already in the process of being delete marked */
+		err = DB_SUCCESS;
+
+		goto nonstandard_exit_func;
+	}
+
+	if ((node->is_delete
+	     && (foreign->type & DICT_FOREIGN_ON_DELETE_SET_NULL))
+	    || (!node->is_delete
+		&& (foreign->type & DICT_FOREIGN_ON_UPDATE_SET_NULL))) {
+
+		/* Build the appropriate update vector which sets
+		foreign->n_fields first fields in rec to SQL NULL */
+
+		update = cascade->update;
+
+		update->info_bits = 0;
+		update->n_fields = foreign->n_fields;
+
+		for (i = 0; i < foreign->n_fields; i++) {
+			upd_field_t*	ufield = &update->fields[i];
+
+			ufield->field_no = dict_table_get_nth_col_pos(
+				table,
+				dict_index_get_nth_col_no(index, i));
+			ufield->orig_len = 0;
+			ufield->exp = NULL;
+			dfield_set_null(&ufield->new_val);
+		}
+	}
+
+	if (!node->is_delete
+	    && (foreign->type & DICT_FOREIGN_ON_UPDATE_CASCADE)) {
+
+		/* Build the appropriate update vector which sets changing
+		foreign->n_fields first fields in rec to new values */
+
+		upd_vec_heap = mem_heap_create(256);
+
+		n_to_update = row_ins_cascade_calc_update_vec(node, foreign,
+							      upd_vec_heap);
+		if (n_to_update == ULINT_UNDEFINED) {
+			err = DB_ROW_IS_REFERENCED;
+
+			row_ins_foreign_report_err(
+				"Trying a cascaded update where the"
+				" updated value in the child\n"
+				"table would not fit in the length"
+				" of the column, or the value would\n"
+				"be NULL and the column is"
+				" declared as not NULL in the child table,",
+				thr, foreign, btr_pcur_get_rec(pcur), entry);
+
+			goto nonstandard_exit_func;
+		}
+
+		if (cascade->update->n_fields == 0) {
+
+			/* The update does not change any columns referred
+			to in this foreign key constraint: no need to do
+			anything */
+
+			err = DB_SUCCESS;
+
+			goto nonstandard_exit_func;
+		}
+	}
+
+	/* Store pcur position and initialize or store the cascade node
+	pcur stored position */
+
+	btr_pcur_store_position(pcur, mtr);
+
+	if (index == clust_index) {
+		btr_pcur_copy_stored_position(cascade->pcur, pcur);
+	} else {
+		btr_pcur_store_position(cascade->pcur, mtr);
+	}
+
+	mtr_commit(mtr);
+
+	ut_a(cascade->pcur->rel_pos == BTR_PCUR_ON);
+
+	cascade->state = UPD_NODE_UPDATE_CLUSTERED;
+
+	err = row_update_cascade_for_mysql(thr, cascade,
+					   foreign->foreign_table);
+
+	if (foreign->foreign_table->n_foreign_key_checks_running == 0) {
+		fprintf(stderr,
+			"InnoDB: error: table %s has the counter 0"
+			" though there is\n"
+			"InnoDB: a FOREIGN KEY check running on it.\n",
+			foreign->foreign_table->name);
+	}
+
+	/* Release the data dictionary latch for a while, so that we do not
+	starve other threads from doing CREATE TABLE etc. if we have a huge
+	cascaded operation running. The counter n_foreign_key_checks_running
+	will prevent other users from dropping or ALTERing the table when we
+	release the latch. */
+
+	row_mysql_unfreeze_data_dictionary(thr_get_trx(thr));
+	row_mysql_freeze_data_dictionary(thr_get_trx(thr));
+
+	mtr_start(mtr);
+
+	/* Restore pcur position */
+
+	btr_pcur_restore_position(BTR_SEARCH_LEAF, pcur, mtr);
+
+	if (tmp_heap) {
+		mem_heap_free(tmp_heap);
+	}
+
+	if (upd_vec_heap) {
+		mem_heap_free(upd_vec_heap);
+	}
+
+	return(err);
+
+nonstandard_exit_func:
+	if (tmp_heap) {
+		mem_heap_free(tmp_heap);
+	}
+
+	if (upd_vec_heap) {
+		mem_heap_free(upd_vec_heap);
+	}
+
+	btr_pcur_store_position(pcur, mtr);
+
+	mtr_commit(mtr);
+	mtr_start(mtr);
+
+	btr_pcur_restore_position(BTR_SEARCH_LEAF, pcur, mtr);
+
+	return(err);
+}
+
+/*************************************************************************
+Sets a shared lock on a record. Used in locking possible duplicate key
+records and also in checking foreign key constraints. */
+static
+ulint
+row_ins_set_shared_rec_lock(
+/*========================*/
+					/* out: DB_SUCCESS or error code */
+	ulint			type,	/* in: LOCK_ORDINARY, LOCK_GAP, or
+					LOCK_REC_NOT_GAP type lock */
+	const buf_block_t*	block,	/* in: buffer block of rec */
+	const rec_t*		rec,	/* in: record */
+	dict_index_t*		index,	/* in: index */
+	const ulint*		offsets,/* in: rec_get_offsets(rec, index) */
+	que_thr_t*		thr)	/* in: query thread */
+{
+	ulint	err;
+
+	ut_ad(rec_offs_validate(rec, index, offsets));
+
+	if (dict_index_is_clust(index)) {
+		err = lock_clust_rec_read_check_and_lock(
+			0, block, rec, index, offsets, LOCK_S, type, thr);
+	} else {
+		err = lock_sec_rec_read_check_and_lock(
+			0, block, rec, index, offsets, LOCK_S, type, thr);
+	}
+
+	return(err);
+}
+
+#ifndef UNIV_HOTBACKUP
+/*************************************************************************
+Sets a exclusive lock on a record. Used in locking possible duplicate key
+records */
+static
+ulint
+row_ins_set_exclusive_rec_lock(
+/*===========================*/
+					/* out: DB_SUCCESS or error code */
+	ulint			type,	/* in: LOCK_ORDINARY, LOCK_GAP, or
+					LOCK_REC_NOT_GAP type lock */
+	const buf_block_t*	block,	/* in: buffer block of rec */
+	const rec_t*		rec,	/* in: record */
+	dict_index_t*		index,	/* in: index */
+	const ulint*		offsets,/* in: rec_get_offsets(rec, index) */
+	que_thr_t*		thr)	/* in: query thread */
+{
+	ulint	err;
+
+	ut_ad(rec_offs_validate(rec, index, offsets));
+
+	if (dict_index_is_clust(index)) {
+		err = lock_clust_rec_read_check_and_lock(
+			0, block, rec, index, offsets, LOCK_X, type, thr);
+	} else {
+		err = lock_sec_rec_read_check_and_lock(
+			0, block, rec, index, offsets, LOCK_X, type, thr);
+	}
+
+	return(err);
+}
+#endif /* !UNIV_HOTBACKUP */
+
+/*******************************************************************
+Checks if foreign key constraint fails for an index entry. Sets shared locks
+which lock either the success or the failure of the constraint. NOTE that
+the caller must have a shared latch on dict_operation_lock. */
+UNIV_INTERN
+ulint
+row_ins_check_foreign_constraint(
+/*=============================*/
+				/* out: DB_SUCCESS,
+				DB_NO_REFERENCED_ROW,
+				or DB_ROW_IS_REFERENCED */
+	ibool		check_ref,/* in: TRUE if we want to check that
+				the referenced table is ok, FALSE if we
+				want to to check the foreign key table */
+	dict_foreign_t*	foreign,/* in: foreign constraint; NOTE that the
+				tables mentioned in it must be in the
+				dictionary cache if they exist at all */
+	dict_table_t*	table,	/* in: if check_ref is TRUE, then the foreign
+				table, else the referenced table */
+	dtuple_t*	entry,	/* in: index entry for index */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	upd_node_t*	upd_node;
+	dict_table_t*	check_table;
+	dict_index_t*	check_index;
+	ulint		n_fields_cmp;
+	btr_pcur_t	pcur;
+	ibool		moved;
+	int		cmp;
+	ulint		err;
+	ulint		i;
+	mtr_t		mtr;
+	trx_t*		trx		= thr_get_trx(thr);
+	mem_heap_t*	heap		= NULL;
+	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint*		offsets		= offsets_;
+	rec_offs_init(offsets_);
+
+run_again:
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(rw_lock_own(&dict_operation_lock, RW_LOCK_SHARED));
+#endif /* UNIV_SYNC_DEBUG */
+
+	err = DB_SUCCESS;
+
+	if (trx->check_foreigns == FALSE) {
+		/* The user has suppressed foreign key checks currently for
+		this session */
+		goto exit_func;
+	}
+
+	/* If any of the foreign key fields in entry is SQL NULL, we
+	suppress the foreign key check: this is compatible with Oracle,
+	for example */
+
+	for (i = 0; i < foreign->n_fields; i++) {
+		if (UNIV_SQL_NULL == dfield_get_len(
+			    dtuple_get_nth_field(entry, i))) {
+
+			goto exit_func;
+		}
+	}
+
+	if (que_node_get_type(thr->run_node) == QUE_NODE_UPDATE) {
+		upd_node = thr->run_node;
+
+		if (!(upd_node->is_delete) && upd_node->foreign == foreign) {
+			/* If a cascaded update is done as defined by a
+			foreign key constraint, do not check that
+			constraint for the child row. In ON UPDATE CASCADE
+			the update of the parent row is only half done when
+			we come here: if we would check the constraint here
+			for the child row it would fail.
+
+			A QUESTION remains: if in the child table there are
+			several constraints which refer to the same parent
+			table, we should merge all updates to the child as
+			one update? And the updates can be contradictory!
+			Currently we just perform the update associated
+			with each foreign key constraint, one after
+			another, and the user has problems predicting in
+			which order they are performed. */
+
+			goto exit_func;
+		}
+	}
+
+	if (check_ref) {
+		check_table = foreign->referenced_table;
+		check_index = foreign->referenced_index;
+	} else {
+		check_table = foreign->foreign_table;
+		check_index = foreign->foreign_index;
+	}
+
+	if (check_table == NULL || check_table->ibd_file_missing) {
+		if (check_ref) {
+			FILE*	ef = dict_foreign_err_file;
+
+			row_ins_set_detailed(trx, foreign);
+
+			mutex_enter(&dict_foreign_err_mutex);
+			rewind(ef);
+			ut_print_timestamp(ef);
+			fputs(" Transaction:\n", ef);
+			trx_print(ef, trx, 600);
+			fputs("Foreign key constraint fails for table ", ef);
+			ut_print_name(ef, trx, TRUE,
+				      foreign->foreign_table_name);
+			fputs(":\n", ef);
+			dict_print_info_on_foreign_key_in_create_format(
+				ef, trx, foreign, TRUE);
+			fputs("\nTrying to add to index ", ef);
+			ut_print_name(ef, trx, FALSE,
+				      foreign->foreign_index->name);
+			fputs(" tuple:\n", ef);
+			dtuple_print(ef, entry);
+			fputs("\nBut the parent table ", ef);
+			ut_print_name(ef, trx, TRUE,
+				      foreign->referenced_table_name);
+			fputs("\nor its .ibd file does"
+			      " not currently exist!\n", ef);
+			mutex_exit(&dict_foreign_err_mutex);
+
+			err = DB_NO_REFERENCED_ROW;
+		}
+
+		goto exit_func;
+	}
+
+	ut_a(check_table);
+	ut_a(check_index);
+
+	if (check_table != table) {
+		/* We already have a LOCK_IX on table, but not necessarily
+		on check_table */
+
+		err = lock_table(0, check_table, LOCK_IS, thr);
+
+		if (err != DB_SUCCESS) {
+
+			goto do_possible_lock_wait;
+		}
+	}
+
+	mtr_start(&mtr);
+
+	/* Store old value on n_fields_cmp */
+
+	n_fields_cmp = dtuple_get_n_fields_cmp(entry);
+
+	dtuple_set_n_fields_cmp(entry, foreign->n_fields);
+
+	btr_pcur_open(check_index, entry, PAGE_CUR_GE,
+		      BTR_SEARCH_LEAF, &pcur, &mtr);
+
+	/* Scan index records and check if there is a matching record */
+
+	for (;;) {
+		const rec_t*		rec = btr_pcur_get_rec(&pcur);
+		const buf_block_t*	block = btr_pcur_get_block(&pcur);
+
+		if (page_rec_is_infimum(rec)) {
+
+			goto next_rec;
+		}
+
+		offsets = rec_get_offsets(rec, check_index,
+					  offsets, ULINT_UNDEFINED, &heap);
+
+		if (page_rec_is_supremum(rec)) {
+
+			err = row_ins_set_shared_rec_lock(LOCK_ORDINARY, block,
+							  rec, check_index,
+							  offsets, thr);
+			if (err != DB_SUCCESS) {
+
+				break;
+			}
+
+			goto next_rec;
+		}
+
+		cmp = cmp_dtuple_rec(entry, rec, offsets);
+
+		if (cmp == 0) {
+			if (rec_get_deleted_flag(rec,
+						 rec_offs_comp(offsets))) {
+				err = row_ins_set_shared_rec_lock(
+					LOCK_ORDINARY, block,
+					rec, check_index, offsets, thr);
+				if (err != DB_SUCCESS) {
+
+					break;
+				}
+			} else {
+				/* Found a matching record. Lock only
+				a record because we can allow inserts
+				into gaps */
+
+				err = row_ins_set_shared_rec_lock(
+					LOCK_REC_NOT_GAP, block,
+					rec, check_index, offsets, thr);
+
+				if (err != DB_SUCCESS) {
+
+					break;
+				}
+
+				if (check_ref) {
+					err = DB_SUCCESS;
+
+					break;
+				} else if (foreign->type != 0) {
+					/* There is an ON UPDATE or ON DELETE
+					condition: check them in a separate
+					function */
+
+					err = row_ins_foreign_check_on_constraint(
+						thr, foreign, &pcur, entry,
+						&mtr);
+					if (err != DB_SUCCESS) {
+						/* Since reporting a plain
+						"duplicate key" error
+						message to the user in
+						cases where a long CASCADE
+						operation would lead to a
+						duplicate key in some
+						other table is very
+						confusing, map duplicate
+						key errors resulting from
+						FK constraints to a
+						separate error code. */
+
+						if (err == DB_DUPLICATE_KEY) {
+							err = DB_FOREIGN_DUPLICATE_KEY;
+						}
+
+						break;
+					}
+
+					/* row_ins_foreign_check_on_constraint
+					may have repositioned pcur on a
+					different block */
+					block = btr_pcur_get_block(&pcur);
+				} else {
+					row_ins_foreign_report_err(
+						"Trying to delete or update",
+						thr, foreign, rec, entry);
+
+					err = DB_ROW_IS_REFERENCED;
+					break;
+				}
+			}
+		}
+
+		if (cmp < 0) {
+			err = row_ins_set_shared_rec_lock(
+				LOCK_GAP, block,
+				rec, check_index, offsets, thr);
+			if (err != DB_SUCCESS) {
+
+				break;
+			}
+
+			if (check_ref) {
+				err = DB_NO_REFERENCED_ROW;
+				row_ins_foreign_report_add_err(
+					trx, foreign, rec, entry);
+			} else {
+				err = DB_SUCCESS;
+			}
+
+			break;
+		}
+
+		ut_a(cmp == 0);
+next_rec:
+		moved = btr_pcur_move_to_next(&pcur, &mtr);
+
+		if (!moved) {
+			if (check_ref) {
+				rec = btr_pcur_get_rec(&pcur);
+				row_ins_foreign_report_add_err(
+					trx, foreign, rec, entry);
+				err = DB_NO_REFERENCED_ROW;
+			} else {
+				err = DB_SUCCESS;
+			}
+
+			break;
+		}
+	}
+
+	btr_pcur_close(&pcur);
+
+	mtr_commit(&mtr);
+
+	/* Restore old value */
+	dtuple_set_n_fields_cmp(entry, n_fields_cmp);
+
+do_possible_lock_wait:
+	if (err == DB_LOCK_WAIT) {
+		trx->error_state = err;
+
+		que_thr_stop_for_mysql(thr);
+
+		srv_suspend_mysql_thread(thr);
+
+		if (trx->error_state == DB_SUCCESS) {
+
+			goto run_again;
+		}
+
+		err = trx->error_state;
+	}
+
+exit_func:
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+	return(err);
+}
+
+/*******************************************************************
+Checks if foreign key constraints fail for an index entry. If index
+is not mentioned in any constraint, this function does nothing,
+Otherwise does searches to the indexes of referenced tables and
+sets shared locks which lock either the success or the failure of
+a constraint. */
+static
+ulint
+row_ins_check_foreign_constraints(
+/*==============================*/
+				/* out: DB_SUCCESS or error code */
+	dict_table_t*	table,	/* in: table */
+	dict_index_t*	index,	/* in: index */
+	dtuple_t*	entry,	/* in: index entry for index */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	dict_foreign_t*	foreign;
+	ulint		err;
+	trx_t*		trx;
+	ibool		got_s_lock	= FALSE;
+
+	trx = thr_get_trx(thr);
+
+	foreign = UT_LIST_GET_FIRST(table->foreign_list);
+
+	while (foreign) {
+		if (foreign->foreign_index == index) {
+
+			if (foreign->referenced_table == NULL) {
+				dict_table_get(foreign->referenced_table_name,
+					       FALSE);
+			}
+
+			if (0 == trx->dict_operation_lock_mode) {
+				got_s_lock = TRUE;
+
+				row_mysql_freeze_data_dictionary(trx);
+			}
+
+			if (foreign->referenced_table) {
+				mutex_enter(&(dict_sys->mutex));
+
+				(foreign->referenced_table
+				 ->n_foreign_key_checks_running)++;
+
+				mutex_exit(&(dict_sys->mutex));
+			}
+
+			/* NOTE that if the thread ends up waiting for a lock
+			we will release dict_operation_lock temporarily!
+			But the counter on the table protects the referenced
+			table from being dropped while the check is running. */
+
+			err = row_ins_check_foreign_constraint(
+				TRUE, foreign, table, entry, thr);
+
+			if (foreign->referenced_table) {
+				mutex_enter(&(dict_sys->mutex));
+
+				ut_a(foreign->referenced_table
+				     ->n_foreign_key_checks_running > 0);
+				(foreign->referenced_table
+				 ->n_foreign_key_checks_running)--;
+
+				mutex_exit(&(dict_sys->mutex));
+			}
+
+			if (got_s_lock) {
+				row_mysql_unfreeze_data_dictionary(trx);
+			}
+
+			if (err != DB_SUCCESS) {
+				return(err);
+			}
+		}
+
+		foreign = UT_LIST_GET_NEXT(foreign_list, foreign);
+	}
+
+	return(DB_SUCCESS);
+}
+
+#ifndef UNIV_HOTBACKUP
+/*******************************************************************
+Checks if a unique key violation to rec would occur at the index entry
+insert. */
+static
+ibool
+row_ins_dupl_error_with_rec(
+/*========================*/
+				/* out: TRUE if error */
+	const rec_t*	rec,	/* in: user record; NOTE that we assume
+				that the caller already has a record lock on
+				the record! */
+	const dtuple_t*	entry,	/* in: entry to insert */
+	dict_index_t*	index,	/* in: index */
+	const ulint*	offsets)/* in: rec_get_offsets(rec, index) */
+{
+	ulint	matched_fields;
+	ulint	matched_bytes;
+	ulint	n_unique;
+	ulint	i;
+
+	ut_ad(rec_offs_validate(rec, index, offsets));
+
+	n_unique = dict_index_get_n_unique(index);
+
+	matched_fields = 0;
+	matched_bytes = 0;
+
+	cmp_dtuple_rec_with_match(entry, rec, offsets,
+				  &matched_fields, &matched_bytes);
+
+	if (matched_fields < n_unique) {
+
+		return(FALSE);
+	}
+
+	/* In a unique secondary index we allow equal key values if they
+	contain SQL NULLs */
+
+	if (!dict_index_is_clust(index)) {
+
+		for (i = 0; i < n_unique; i++) {
+			if (UNIV_SQL_NULL == dfield_get_len(
+				    dtuple_get_nth_field(entry, i))) {
+
+				return(FALSE);
+			}
+		}
+	}
+
+	return(!rec_get_deleted_flag(rec, rec_offs_comp(offsets)));
+}
+#endif /* !UNIV_HOTBACKUP */
+
+/*******************************************************************
+Scans a unique non-clustered index at a given index entry to determine
+whether a uniqueness violation has occurred for the key value of the entry.
+Set shared locks on possible duplicate records. */
+static
+ulint
+row_ins_scan_sec_index_for_duplicate(
+/*=================================*/
+				/* out: DB_SUCCESS, DB_DUPLICATE_KEY, or
+				DB_LOCK_WAIT */
+	dict_index_t*	index,	/* in: non-clustered unique index */
+	dtuple_t*	entry,	/* in: index entry */
+	que_thr_t*	thr)	/* in: query thread */
+{
+#ifndef UNIV_HOTBACKUP
+	ulint		n_unique;
+	ulint		i;
+	int		cmp;
+	ulint		n_fields_cmp;
+	btr_pcur_t	pcur;
+	ulint		err		= DB_SUCCESS;
+	unsigned	allow_duplicates;
+	mtr_t		mtr;
+	mem_heap_t*	heap		= NULL;
+	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint*		offsets		= offsets_;
+	rec_offs_init(offsets_);
+
+	n_unique = dict_index_get_n_unique(index);
+
+	/* If the secondary index is unique, but one of the fields in the
+	n_unique first fields is NULL, a unique key violation cannot occur,
+	since we define NULL != NULL in this case */
+
+	for (i = 0; i < n_unique; i++) {
+		if (UNIV_SQL_NULL == dfield_get_len(
+			    dtuple_get_nth_field(entry, i))) {
+
+			return(DB_SUCCESS);
+		}
+	}
+
+	mtr_start(&mtr);
+
+	/* Store old value on n_fields_cmp */
+
+	n_fields_cmp = dtuple_get_n_fields_cmp(entry);
+
+	dtuple_set_n_fields_cmp(entry, dict_index_get_n_unique(index));
+
+	btr_pcur_open(index, entry, PAGE_CUR_GE, BTR_SEARCH_LEAF, &pcur, &mtr);
+
+	allow_duplicates = thr_get_trx(thr)->duplicates & TRX_DUP_IGNORE;
+
+	/* Scan index records and check if there is a duplicate */
+
+	do {
+		const rec_t*		rec	= btr_pcur_get_rec(&pcur);
+		const buf_block_t*	block	= btr_pcur_get_block(&pcur);
+
+		if (page_rec_is_infimum(rec)) {
+
+			continue;
+		}
+
+		offsets = rec_get_offsets(rec, index, offsets,
+					  ULINT_UNDEFINED, &heap);
+
+		if (allow_duplicates) {
+
+			/* If the SQL-query will update or replace
+			duplicate key we will take X-lock for
+			duplicates ( REPLACE, LOAD DATAFILE REPLACE,
+			INSERT ON DUPLICATE KEY UPDATE). */
+
+			err = row_ins_set_exclusive_rec_lock(
+				LOCK_ORDINARY, block,
+				rec, index, offsets, thr);
+		} else {
+
+			err = row_ins_set_shared_rec_lock(
+				LOCK_ORDINARY, block,
+				rec, index, offsets, thr);
+		}
+
+		if (err != DB_SUCCESS) {
+
+			break;
+		}
+
+		if (page_rec_is_supremum(rec)) {
+
+			continue;
+		}
+
+		cmp = cmp_dtuple_rec(entry, rec, offsets);
+
+		if (cmp == 0) {
+			if (row_ins_dupl_error_with_rec(rec, entry,
+							index, offsets)) {
+				err = DB_DUPLICATE_KEY;
+
+				thr_get_trx(thr)->error_info = index;
+
+				break;
+			}
+		}
+
+		if (cmp < 0) {
+			break;
+		}
+
+		ut_a(cmp == 0);
+	} while (btr_pcur_move_to_next(&pcur, &mtr));
+
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+	mtr_commit(&mtr);
+
+	/* Restore old value */
+	dtuple_set_n_fields_cmp(entry, n_fields_cmp);
+
+	return(err);
+#else /* UNIV_HOTBACKUP */
+	/* This function depends on MySQL code that is not included in
+	InnoDB Hot Backup builds.  Besides, this function should never
+	be called in InnoDB Hot Backup. */
+	ut_error;
+	return(DB_FAIL);
+#endif /* UNIV_HOTBACKUP */
+}
+
+/*******************************************************************
+Checks if a unique key violation error would occur at an index entry
+insert. Sets shared locks on possible duplicate records. Works only
+for a clustered index! */
+static
+ulint
+row_ins_duplicate_error_in_clust(
+/*=============================*/
+				/* out: DB_SUCCESS if no error,
+				DB_DUPLICATE_KEY if error, DB_LOCK_WAIT if we
+				have to wait for a lock on a possible
+				duplicate record */
+	btr_cur_t*	cursor,	/* in: B-tree cursor */
+	dtuple_t*	entry,	/* in: entry to insert */
+	que_thr_t*	thr,	/* in: query thread */
+	mtr_t*		mtr)	/* in: mtr */
+{
+#ifndef UNIV_HOTBACKUP
+	ulint	err;
+	rec_t*	rec;
+	ulint	n_unique;
+	trx_t*	trx		= thr_get_trx(thr);
+	mem_heap_t*heap		= NULL;
+	ulint	offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint*	offsets		= offsets_;
+	rec_offs_init(offsets_);
+
+	UT_NOT_USED(mtr);
+
+	ut_a(dict_index_is_clust(cursor->index));
+	ut_ad(dict_index_is_unique(cursor->index));
+
+	/* NOTE: For unique non-clustered indexes there may be any number
+	of delete marked records with the same value for the non-clustered
+	index key (remember multiversioning), and which differ only in
+	the row refererence part of the index record, containing the
+	clustered index key fields. For such a secondary index record,
+	to avoid race condition, we must FIRST do the insertion and after
+	that check that the uniqueness condition is not breached! */
+
+	/* NOTE: A problem is that in the B-tree node pointers on an
+	upper level may match more to the entry than the actual existing
+	user records on the leaf level. So, even if low_match would suggest
+	that a duplicate key violation may occur, this may not be the case. */
+
+	n_unique = dict_index_get_n_unique(cursor->index);
+
+	if (cursor->low_match >= n_unique) {
+
+		rec = btr_cur_get_rec(cursor);
+
+		if (!page_rec_is_infimum(rec)) {
+			offsets = rec_get_offsets(rec, cursor->index, offsets,
+						  ULINT_UNDEFINED, &heap);
+
+			/* We set a lock on the possible duplicate: this
+			is needed in logical logging of MySQL to make
+			sure that in roll-forward we get the same duplicate
+			errors as in original execution */
+
+			if (trx->duplicates & TRX_DUP_IGNORE) {
+
+				/* If the SQL-query will update or replace
+				duplicate key we will take X-lock for
+				duplicates ( REPLACE, LOAD DATAFILE REPLACE,
+				INSERT ON DUPLICATE KEY UPDATE). */
+
+				err = row_ins_set_exclusive_rec_lock(
+					LOCK_REC_NOT_GAP,
+					btr_cur_get_block(cursor),
+					rec, cursor->index, offsets, thr);
+			} else {
+
+				err = row_ins_set_shared_rec_lock(
+					LOCK_REC_NOT_GAP,
+					btr_cur_get_block(cursor), rec,
+					cursor->index, offsets, thr);
+			}
+
+			if (err != DB_SUCCESS) {
+				goto func_exit;
+			}
+
+			if (row_ins_dupl_error_with_rec(
+				    rec, entry, cursor->index, offsets)) {
+				trx->error_info = cursor->index;
+				err = DB_DUPLICATE_KEY;
+				goto func_exit;
+			}
+		}
+	}
+
+	if (cursor->up_match >= n_unique) {
+
+		rec = page_rec_get_next(btr_cur_get_rec(cursor));
+
+		if (!page_rec_is_supremum(rec)) {
+			offsets = rec_get_offsets(rec, cursor->index, offsets,
+						  ULINT_UNDEFINED, &heap);
+
+			if (trx->duplicates & TRX_DUP_IGNORE) {
+
+				/* If the SQL-query will update or replace
+				duplicate key we will take X-lock for
+				duplicates ( REPLACE, LOAD DATAFILE REPLACE,
+				INSERT ON DUPLICATE KEY UPDATE). */
+
+				err = row_ins_set_exclusive_rec_lock(
+					LOCK_REC_NOT_GAP,
+					btr_cur_get_block(cursor),
+					rec, cursor->index, offsets, thr);
+			} else {
+
+				err = row_ins_set_shared_rec_lock(
+					LOCK_REC_NOT_GAP,
+					btr_cur_get_block(cursor),
+					rec, cursor->index, offsets, thr);
+			}
+
+			if (err != DB_SUCCESS) {
+				goto func_exit;
+			}
+
+			if (row_ins_dupl_error_with_rec(
+				    rec, entry, cursor->index, offsets)) {
+				trx->error_info = cursor->index;
+				err = DB_DUPLICATE_KEY;
+				goto func_exit;
+			}
+		}
+
+		ut_a(!dict_index_is_clust(cursor->index));
+		/* This should never happen */
+	}
+
+	err = DB_SUCCESS;
+func_exit:
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+	return(err);
+#else /* UNIV_HOTBACKUP */
+	/* This function depends on MySQL code that is not included in
+	InnoDB Hot Backup builds.  Besides, this function should never
+	be called in InnoDB Hot Backup. */
+	ut_error;
+	return(DB_FAIL);
+#endif /* UNIV_HOTBACKUP */
+}
+
+/*******************************************************************
+Checks if an index entry has long enough common prefix with an existing
+record so that the intended insert of the entry must be changed to a modify of
+the existing record. In the case of a clustered index, the prefix must be
+n_unique fields long, and in the case of a secondary index, all fields must be
+equal. */
+UNIV_INLINE
+ulint
+row_ins_must_modify(
+/*================*/
+				/* out: 0 if no update, ROW_INS_PREV if
+				previous should be updated; currently we
+				do the search so that only the low_match
+				record can match enough to the search tuple,
+				not the next record */
+	btr_cur_t*	cursor)	/* in: B-tree cursor */
+{
+	ulint	enough_match;
+	rec_t*	rec;
+
+	/* NOTE: (compare to the note in row_ins_duplicate_error) Because node
+	pointers on upper levels of the B-tree may match more to entry than
+	to actual user records on the leaf level, we have to check if the
+	candidate record is actually a user record. In a clustered index
+	node pointers contain index->n_unique first fields, and in the case
+	of a secondary index, all fields of the index. */
+
+	enough_match = dict_index_get_n_unique_in_tree(cursor->index);
+
+	if (cursor->low_match >= enough_match) {
+
+		rec = btr_cur_get_rec(cursor);
+
+		if (!page_rec_is_infimum(rec)) {
+
+			return(ROW_INS_PREV);
+		}
+	}
+
+	return(0);
+}
+
+/*******************************************************************
+Tries to insert an index entry to an index. If the index is clustered
+and a record with the same unique key is found, the other record is
+necessarily marked deleted by a committed transaction, or a unique key
+violation error occurs. The delete marked record is then updated to an
+existing record, and we must write an undo log record on the delete
+marked record. If the index is secondary, and a record with exactly the
+same fields is found, the other record is necessarily marked deleted.
+It is then unmarked. Otherwise, the entry is just inserted to the index. */
+static
+ulint
+row_ins_index_entry_low(
+/*====================*/
+				/* out: DB_SUCCESS, DB_LOCK_WAIT, DB_FAIL
+				if pessimistic retry needed, or error code */
+	ulint		mode,	/* in: BTR_MODIFY_LEAF or BTR_MODIFY_TREE,
+				depending on whether we wish optimistic or
+				pessimistic descent down the index tree */
+	dict_index_t*	index,	/* in: index */
+	dtuple_t*	entry,	/* in: index entry to insert */
+	ulint		n_ext,	/* in: number of externally stored columns */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	btr_cur_t	cursor;
+	ulint		ignore_sec_unique	= 0;
+	ulint		modify = 0; /* remove warning */
+	rec_t*		insert_rec;
+	rec_t*		rec;
+	ulint		err;
+	ulint		n_unique;
+	big_rec_t*	big_rec			= NULL;
+	mtr_t		mtr;
+	mem_heap_t*	heap			= NULL;
+
+	log_free_check();
+
+	mtr_start(&mtr);
+
+	cursor.thr = thr;
+
+	/* Note that we use PAGE_CUR_LE as the search mode, because then
+	the function will return in both low_match and up_match of the
+	cursor sensible values */
+
+	if (!(thr_get_trx(thr)->check_unique_secondary)) {
+		ignore_sec_unique = BTR_IGNORE_SEC_UNIQUE;
+	}
+
+	btr_cur_search_to_nth_level(index, 0, entry, PAGE_CUR_LE,
+				    mode | BTR_INSERT | ignore_sec_unique,
+				    &cursor, 0, &mtr);
+
+	if (cursor.flag == BTR_CUR_INSERT_TO_IBUF) {
+		/* The insertion was made to the insert buffer already during
+		the search: we are done */
+
+		err = DB_SUCCESS;
+
+		goto function_exit;
+	}
+
+#ifdef UNIV_DEBUG
+	{
+		page_t*	page = btr_cur_get_page(&cursor);
+		rec_t*	first_rec = page_rec_get_next(
+			page_get_infimum_rec(page));
+
+		ut_ad(page_rec_is_supremum(first_rec)
+		      || rec_get_n_fields(first_rec, index)
+		      == dtuple_get_n_fields(entry));
+	}
+#endif
+
+	n_unique = dict_index_get_n_unique(index);
+
+	if (dict_index_is_unique(index) && (cursor.up_match >= n_unique
+					    || cursor.low_match >= n_unique)) {
+
+		if (dict_index_is_clust(index)) {
+			/* Note that the following may return also
+			DB_LOCK_WAIT */
+
+			err = row_ins_duplicate_error_in_clust(
+				&cursor, entry, thr, &mtr);
+			if (err != DB_SUCCESS) {
+
+				goto function_exit;
+			}
+		} else {
+			mtr_commit(&mtr);
+			err = row_ins_scan_sec_index_for_duplicate(
+				index, entry, thr);
+			mtr_start(&mtr);
+
+			if (err != DB_SUCCESS) {
+
+				goto function_exit;
+			}
+
+			/* We did not find a duplicate and we have now
+			locked with s-locks the necessary records to
+			prevent any insertion of a duplicate by another
+			transaction. Let us now reposition the cursor and
+			continue the insertion. */
+
+			btr_cur_search_to_nth_level(index, 0, entry,
+						    PAGE_CUR_LE,
+						    mode | BTR_INSERT,
+						    &cursor, 0, &mtr);
+		}
+	}
+
+	modify = row_ins_must_modify(&cursor);
+
+	if (modify != 0) {
+		/* There is already an index entry with a long enough common
+		prefix, we must convert the insert into a modify of an
+		existing record */
+
+		if (modify == ROW_INS_NEXT) {
+			rec = page_rec_get_next(btr_cur_get_rec(&cursor));
+
+			btr_cur_position(index, rec,
+					 btr_cur_get_block(&cursor),&cursor);
+		}
+
+		if (dict_index_is_clust(index)) {
+			err = row_ins_clust_index_entry_by_modify(
+				mode, &cursor, &heap, &big_rec, entry,
+				thr, &mtr);
+		} else {
+			ut_ad(!n_ext);
+			err = row_ins_sec_index_entry_by_modify(
+				mode, &cursor, entry, thr, &mtr);
+		}
+	} else {
+		if (mode == BTR_MODIFY_LEAF) {
+			err = btr_cur_optimistic_insert(
+				0, &cursor, entry, &insert_rec, &big_rec,
+				n_ext, thr, &mtr);
+		} else {
+			ut_a(mode == BTR_MODIFY_TREE);
+			if (buf_LRU_buf_pool_running_out()) {
+
+				err = DB_LOCK_TABLE_FULL;
+
+				goto function_exit;
+			}
+			err = btr_cur_pessimistic_insert(
+				0, &cursor, entry, &insert_rec, &big_rec,
+				n_ext, thr, &mtr);
+		}
+	}
+
+function_exit:
+	mtr_commit(&mtr);
+
+	if (UNIV_LIKELY_NULL(big_rec)) {
+		rec_t*	rec;
+		ulint*	offsets;
+		mtr_start(&mtr);
+
+		btr_cur_search_to_nth_level(index, 0, entry, PAGE_CUR_LE,
+					    BTR_MODIFY_TREE, &cursor, 0, &mtr);
+		rec = btr_cur_get_rec(&cursor);
+		offsets = rec_get_offsets(rec, index, NULL,
+					  ULINT_UNDEFINED, &heap);
+
+		err = btr_store_big_rec_extern_fields(
+			index, btr_cur_get_block(&cursor),
+			rec, offsets, big_rec, &mtr);
+
+		if (modify) {
+			dtuple_big_rec_free(big_rec);
+		} else {
+			dtuple_convert_back_big_rec(index, entry, big_rec);
+		}
+
+		mtr_commit(&mtr);
+	}
+
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+	return(err);
+}
+
+/*******************************************************************
+Inserts an index entry to index. Tries first optimistic, then pessimistic
+descent down the tree. If the entry matches enough to a delete marked record,
+performs the insert by updating or delete unmarking the delete marked
+record. */
+UNIV_INTERN
+ulint
+row_ins_index_entry(
+/*================*/
+				/* out: DB_SUCCESS, DB_LOCK_WAIT,
+				DB_DUPLICATE_KEY, or some other error code */
+	dict_index_t*	index,	/* in: index */
+	dtuple_t*	entry,	/* in: index entry to insert */
+	ulint		n_ext,	/* in: number of externally stored columns */
+	ibool		foreign,/* in: TRUE=check foreign key constraints */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	ulint	err;
+
+	if (foreign && UT_LIST_GET_FIRST(index->table->foreign_list)) {
+		err = row_ins_check_foreign_constraints(index->table, index,
+							entry, thr);
+		if (err != DB_SUCCESS) {
+
+			return(err);
+		}
+	}
+
+	/* Try first optimistic descent to the B-tree */
+
+	err = row_ins_index_entry_low(BTR_MODIFY_LEAF, index, entry,
+				      n_ext, thr);
+	if (err != DB_FAIL) {
+
+		return(err);
+	}
+
+	/* Try then pessimistic descent to the B-tree */
+
+	err = row_ins_index_entry_low(BTR_MODIFY_TREE, index, entry,
+				      n_ext, thr);
+	return(err);
+}
+
+/***************************************************************
+Sets the values of the dtuple fields in entry from the values of appropriate
+columns in row. */
+static
+void
+row_ins_index_entry_set_vals(
+/*=========================*/
+	dict_index_t*	index,	/* in: index */
+	dtuple_t*	entry,	/* in: index entry to make */
+	const dtuple_t*	row)	/* in: row */
+{
+	ulint	n_fields;
+	ulint	i;
+
+	ut_ad(entry && row);
+
+	n_fields = dtuple_get_n_fields(entry);
+
+	for (i = 0; i < n_fields; i++) {
+		dict_field_t*	ind_field;
+		dfield_t*	field;
+		const dfield_t*	row_field;
+		ulint		len;
+
+		field = dtuple_get_nth_field(entry, i);
+		ind_field = dict_index_get_nth_field(index, i);
+		row_field = dtuple_get_nth_field(row, ind_field->col->ind);
+		len = dfield_get_len(row_field);
+
+		/* Check column prefix indexes */
+		if (ind_field->prefix_len > 0
+		    && dfield_get_len(row_field) != UNIV_SQL_NULL) {
+
+			const	dict_col_t*	col
+				= dict_field_get_col(ind_field);
+
+			len = dtype_get_at_most_n_mbchars(
+				col->prtype, col->mbminlen, col->mbmaxlen,
+				ind_field->prefix_len,
+				len, dfield_get_data(row_field));
+
+			ut_ad(!dfield_is_ext(row_field));
+		}
+
+		dfield_set_data(field, dfield_get_data(row_field), len);
+		if (dfield_is_ext(row_field)) {
+			ut_ad(dict_index_is_clust(index));
+			dfield_set_ext(field);
+		}
+	}
+}
+
+/***************************************************************
+Inserts a single index entry to the table. */
+static
+ulint
+row_ins_index_entry_step(
+/*=====================*/
+				/* out: DB_SUCCESS if operation successfully
+				completed, else error code or DB_LOCK_WAIT */
+	ins_node_t*	node,	/* in: row insert node */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	ulint	err;
+
+	ut_ad(dtuple_check_typed(node->row));
+
+	row_ins_index_entry_set_vals(node->index, node->entry, node->row);
+
+	ut_ad(dtuple_check_typed(node->entry));
+
+	err = row_ins_index_entry(node->index, node->entry, 0, TRUE, thr);
+
+	return(err);
+}
+
+/***************************************************************
+Allocates a row id for row and inits the node->index field. */
+UNIV_INLINE
+void
+row_ins_alloc_row_id_step(
+/*======================*/
+	ins_node_t*	node)	/* in: row insert node */
+{
+	dulint	row_id;
+
+	ut_ad(node->state == INS_NODE_ALLOC_ROW_ID);
+
+	if (dict_index_is_unique(dict_table_get_first_index(node->table))) {
+
+		/* No row id is stored if the clustered index is unique */
+
+		return;
+	}
+
+	/* Fill in row id value to row */
+
+	row_id = dict_sys_get_new_row_id();
+
+	dict_sys_write_row_id(node->row_id_buf, row_id);
+}
+
+/***************************************************************
+Gets a row to insert from the values list. */
+UNIV_INLINE
+void
+row_ins_get_row_from_values(
+/*========================*/
+	ins_node_t*	node)	/* in: row insert node */
+{
+	que_node_t*	list_node;
+	dfield_t*	dfield;
+	dtuple_t*	row;
+	ulint		i;
+
+	/* The field values are copied in the buffers of the select node and
+	it is safe to use them until we fetch from select again: therefore
+	we can just copy the pointers */
+
+	row = node->row;
+
+	i = 0;
+	list_node = node->values_list;
+
+	while (list_node) {
+		eval_exp(list_node);
+
+		dfield = dtuple_get_nth_field(row, i);
+		dfield_copy_data(dfield, que_node_get_val(list_node));
+
+		i++;
+		list_node = que_node_get_next(list_node);
+	}
+}
+
+/***************************************************************
+Gets a row to insert from the select list. */
+UNIV_INLINE
+void
+row_ins_get_row_from_select(
+/*========================*/
+	ins_node_t*	node)	/* in: row insert node */
+{
+	que_node_t*	list_node;
+	dfield_t*	dfield;
+	dtuple_t*	row;
+	ulint		i;
+
+	/* The field values are copied in the buffers of the select node and
+	it is safe to use them until we fetch from select again: therefore
+	we can just copy the pointers */
+
+	row = node->row;
+
+	i = 0;
+	list_node = node->select->select_list;
+
+	while (list_node) {
+		dfield = dtuple_get_nth_field(row, i);
+		dfield_copy_data(dfield, que_node_get_val(list_node));
+
+		i++;
+		list_node = que_node_get_next(list_node);
+	}
+}
+
+/***************************************************************
+Inserts a row to a table. */
+static
+ulint
+row_ins(
+/*====*/
+				/* out: DB_SUCCESS if operation successfully
+				completed, else error code or DB_LOCK_WAIT */
+	ins_node_t*	node,	/* in: row insert node */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	ulint	err;
+
+	ut_ad(node && thr);
+
+	if (node->state == INS_NODE_ALLOC_ROW_ID) {
+
+		row_ins_alloc_row_id_step(node);
+
+		node->index = dict_table_get_first_index(node->table);
+		node->entry = UT_LIST_GET_FIRST(node->entry_list);
+
+		if (node->ins_type == INS_SEARCHED) {
+
+			row_ins_get_row_from_select(node);
+
+		} else if (node->ins_type == INS_VALUES) {
+
+			row_ins_get_row_from_values(node);
+		}
+
+		node->state = INS_NODE_INSERT_ENTRIES;
+	}
+
+	ut_ad(node->state == INS_NODE_INSERT_ENTRIES);
+
+	while (node->index != NULL) {
+		err = row_ins_index_entry_step(node, thr);
+
+		if (err != DB_SUCCESS) {
+
+			return(err);
+		}
+
+		node->index = dict_table_get_next_index(node->index);
+		node->entry = UT_LIST_GET_NEXT(tuple_list, node->entry);
+	}
+
+	ut_ad(node->entry == NULL);
+
+	node->state = INS_NODE_ALLOC_ROW_ID;
+
+	return(DB_SUCCESS);
+}
+
+/***************************************************************
+Inserts a row to a table. This is a high-level function used in SQL execution
+graphs. */
+UNIV_INTERN
+que_thr_t*
+row_ins_step(
+/*=========*/
+				/* out: query thread to run next or NULL */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	ins_node_t*	node;
+	que_node_t*	parent;
+	sel_node_t*	sel_node;
+	trx_t*		trx;
+	ulint		err;
+
+	ut_ad(thr);
+
+	trx = thr_get_trx(thr);
+
+	trx_start_if_not_started(trx);
+
+	node = thr->run_node;
+
+	ut_ad(que_node_get_type(node) == QUE_NODE_INSERT);
+
+	parent = que_node_get_parent(node);
+	sel_node = node->select;
+
+	if (thr->prev_node == parent) {
+		node->state = INS_NODE_SET_IX_LOCK;
+	}
+
+	/* If this is the first time this node is executed (or when
+	execution resumes after wait for the table IX lock), set an
+	IX lock on the table and reset the possible select node. MySQL's
+	partitioned table code may also call an insert within the same
+	SQL statement AFTER it has used this table handle to do a search.
+	This happens, for example, when a row update moves it to another
+	partition. In that case, we have already set the IX lock on the
+	table during the search operation, and there is no need to set
+	it again here. But we must write trx->id to node->trx_id_buf. */
+
+	trx_write_trx_id(node->trx_id_buf, trx->id);
+
+	if (node->state == INS_NODE_SET_IX_LOCK) {
+
+		/* It may be that the current session has not yet started
+		its transaction, or it has been committed: */
+
+		if (UT_DULINT_EQ(trx->id, node->trx_id)) {
+			/* No need to do IX-locking */
+
+			goto same_trx;
+		}
+
+		err = lock_table(0, node->table, LOCK_IX, thr);
+
+		if (err != DB_SUCCESS) {
+
+			goto error_handling;
+		}
+
+		node->trx_id = trx->id;
+same_trx:
+		node->state = INS_NODE_ALLOC_ROW_ID;
+
+		if (node->ins_type == INS_SEARCHED) {
+			/* Reset the cursor */
+			sel_node->state = SEL_NODE_OPEN;
+
+			/* Fetch a row to insert */
+
+			thr->run_node = sel_node;
+
+			return(thr);
+		}
+	}
+
+	if ((node->ins_type == INS_SEARCHED)
+	    && (sel_node->state != SEL_NODE_FETCH)) {
+
+		ut_ad(sel_node->state == SEL_NODE_NO_MORE_ROWS);
+
+		/* No more rows to insert */
+		thr->run_node = parent;
+
+		return(thr);
+	}
+
+	/* DO THE CHECKS OF THE CONSISTENCY CONSTRAINTS HERE */
+
+	err = row_ins(node, thr);
+
+error_handling:
+	trx->error_state = err;
+
+	if (err != DB_SUCCESS) {
+		/* err == DB_LOCK_WAIT or SQL error detected */
+		return(NULL);
+	}
+
+	/* DO THE TRIGGER ACTIONS HERE */
+
+	if (node->ins_type == INS_SEARCHED) {
+		/* Fetch a row to insert */
+
+		thr->run_node = sel_node;
+	} else {
+		thr->run_node = que_node_get_parent(node);
+	}
+
+	return(thr);
+}
diff --git a/storage/xtradb/row/row0merge.c b/storage/xtradb/row/row0merge.c
new file mode 100644
index 00000000000..efed3d26e5b
--- /dev/null
+++ b/storage/xtradb/row/row0merge.c
@@ -0,0 +1,2358 @@
+/*****************************************************************************
+
+Copyright (c) 2005, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+New index creation routines using a merge sort
+
+Created 12/4/2005 Jan Lindstrom
+Completed by Sunny Bains and Marko Makela
+*******************************************************/
+
+#include "row0merge.h"
+#include "row0ext.h"
+#include "row0row.h"
+#include "row0upd.h"
+#include "row0ins.h"
+#include "row0sel.h"
+#include "dict0dict.h"
+#include "dict0mem.h"
+#include "dict0boot.h"
+#include "dict0crea.h"
+#include "dict0load.h"
+#include "btr0btr.h"
+#include "mach0data.h"
+#include "trx0rseg.h"
+#include "trx0trx.h"
+#include "trx0roll.h"
+#include "trx0undo.h"
+#include "trx0purge.h"
+#include "trx0rec.h"
+#include "que0que.h"
+#include "rem0cmp.h"
+#include "read0read.h"
+#include "os0file.h"
+#include "lock0lock.h"
+#include "data0data.h"
+#include "data0type.h"
+#include "que0que.h"
+#include "pars0pars.h"
+#include "mem0mem.h"
+#include "log0log.h"
+#include "ut0sort.h"
+#include "handler0alter.h"
+
+#ifdef UNIV_DEBUG
+/* Set these in order ot enable debug printout. */
+static ibool	row_merge_print_cmp;
+static ibool	row_merge_print_read;
+static ibool	row_merge_print_write;
+#endif /* UNIV_DEBUG */
+
+/* Block size for I/O operations in merge sort.  The minimum is
+UNIV_PAGE_SIZE, or page_get_free_space_of_empty() rounded to a power of 2.
+
+When not creating a PRIMARY KEY that contains column prefixes, this
+can be set as small as UNIV_PAGE_SIZE / 2.  See the comment above
+ut_ad(data_size < sizeof(row_merge_block_t)). */
+
+typedef byte	row_merge_block_t[1048576];
+
+/* Secondary buffer for I/O operations of merge records.  This buffer
+is used for writing or reading a record that spans two row_merge_block_t.
+Thus, it must be able to hold one merge record, whose maximum size is
+the same as the minimum size of row_merge_block_t. */
+
+typedef byte	mrec_buf_t[UNIV_PAGE_SIZE];
+
+/* Merge record in row_merge_block_t.  The format is the same as a
+record in ROW_FORMAT=COMPACT with the exception that the
+REC_N_NEW_EXTRA_BYTES are omitted. */
+typedef byte	mrec_t;
+
+/* Buffer for sorting in main memory. */
+struct row_merge_buf_struct {
+	mem_heap_t*	heap;		/* memory heap where allocated */
+	dict_index_t*	index;		/* the index the tuples belong to */
+	ulint		total_size;	/* total amount of data bytes */
+	ulint		n_tuples;	/* number of data tuples */
+	ulint		max_tuples;	/* maximum number of data tuples */
+	const dfield_t**tuples;		/* array of pointers to
+					arrays of fields that form
+					the data tuples */
+	const dfield_t**tmp_tuples;	/* temporary copy of tuples,
+					for sorting */
+};
+
+typedef struct row_merge_buf_struct row_merge_buf_t;
+
+/* Information about temporary files used in merge sort are stored
+to this structure */
+
+struct merge_file_struct {
+	int	fd;		/* File descriptor */
+	ulint	offset;		/* File offset */
+};
+
+typedef struct merge_file_struct merge_file_t;
+
+#ifdef UNIV_DEBUG
+/**********************************************************
+Display a merge tuple. */
+static
+void
+row_merge_tuple_print(
+/*==================*/
+	FILE*		f,	/* in: output stream */
+	const dfield_t*	entry,	/* in: tuple to print */
+	ulint		n_fields)/* in: number of fields in the tuple */
+{
+	ulint	j;
+
+	for (j = 0; j < n_fields; j++) {
+		const dfield_t*	field = &entry[j];
+
+		if (dfield_is_null(field)) {
+			fputs("\n NULL;", f);
+		} else {
+			ulint	field_len	= dfield_get_len(field);
+			ulint	len		= ut_min(field_len, 20);
+			if (dfield_is_ext(field)) {
+				fputs("\nE", f);
+			} else {
+				fputs("\n ", f);
+			}
+			ut_print_buf(f, dfield_get_data(field), len);
+			if (len != field_len) {
+				fprintf(f, " (total %lu bytes)", field_len);
+			}
+		}
+	}
+	putc('\n', f);
+}
+#endif /* UNIV_DEBUG */
+
+/**********************************************************
+Allocate a sort buffer. */
+static
+row_merge_buf_t*
+row_merge_buf_create_low(
+/*=====================*/
+					/* out,own: sort buffer */
+	mem_heap_t*	heap,		/* in: heap where allocated */
+	dict_index_t*	index,		/* in: secondary index */
+	ulint		max_tuples,	/* in: maximum number of data tuples */
+	ulint		buf_size)	/* in: size of the buffer, in bytes */
+{
+	row_merge_buf_t*	buf;
+
+	ut_ad(max_tuples > 0);
+	ut_ad(max_tuples <= sizeof(row_merge_block_t));
+	ut_ad(max_tuples < buf_size);
+
+	buf = mem_heap_zalloc(heap, buf_size);
+	buf->heap = heap;
+	buf->index = index;
+	buf->max_tuples = max_tuples;
+	buf->tuples = mem_heap_alloc(heap,
+				     2 * max_tuples * sizeof *buf->tuples);
+	buf->tmp_tuples = buf->tuples + max_tuples;
+
+	return(buf);
+}
+
+/**********************************************************
+Allocate a sort buffer. */
+static
+row_merge_buf_t*
+row_merge_buf_create(
+/*=================*/
+				/* out,own: sort buffer */
+	dict_index_t*	index)	/* in: secondary index */
+{
+	row_merge_buf_t*	buf;
+	ulint			max_tuples;
+	ulint			buf_size;
+	mem_heap_t*		heap;
+
+	max_tuples = sizeof(row_merge_block_t)
+		/ ut_max(1, dict_index_get_min_size(index));
+
+	buf_size = (sizeof *buf) + (max_tuples - 1) * sizeof *buf->tuples;
+
+	heap = mem_heap_create(buf_size + sizeof(row_merge_block_t));
+
+	buf = row_merge_buf_create_low(heap, index, max_tuples, buf_size);
+
+	return(buf);
+}
+
+/**********************************************************
+Empty a sort buffer. */
+static
+row_merge_buf_t*
+row_merge_buf_empty(
+/*================*/
+					/* out: sort buffer */
+	row_merge_buf_t*	buf)	/* in,own: sort buffer */
+{
+	ulint		buf_size;
+	ulint		max_tuples	= buf->max_tuples;
+	mem_heap_t*	heap		= buf->heap;
+	dict_index_t*	index		= buf->index;
+
+	buf_size = (sizeof *buf) + (max_tuples - 1) * sizeof *buf->tuples;
+
+	mem_heap_empty(heap);
+
+	return(row_merge_buf_create_low(heap, index, max_tuples, buf_size));
+}
+
+/**********************************************************
+Deallocate a sort buffer. */
+static
+void
+row_merge_buf_free(
+/*===============*/
+	row_merge_buf_t*	buf)	/* in,own: sort buffer, to be freed */
+{
+	mem_heap_free(buf->heap);
+}
+
+/**********************************************************
+Insert a data tuple into a sort buffer. */
+static
+ibool
+row_merge_buf_add(
+/*==============*/
+					/* out: TRUE if added,
+					FALSE if out of space */
+	row_merge_buf_t*	buf,	/* in/out: sort buffer */
+	const dtuple_t*		row,	/* in: row in clustered index */
+	const row_ext_t*	ext)	/* in: cache of externally stored
+					column prefixes, or NULL */
+{
+	ulint			i;
+	ulint			n_fields;
+	ulint			data_size;
+	ulint			extra_size;
+	const dict_index_t*	index;
+	dfield_t*		entry;
+	dfield_t*		field;
+
+	if (buf->n_tuples >= buf->max_tuples) {
+		return(FALSE);
+	}
+
+	UNIV_PREFETCH_R(row->fields);
+
+	index = buf->index;
+
+	n_fields = dict_index_get_n_fields(index);
+
+	entry = mem_heap_alloc(buf->heap, n_fields * sizeof *entry);
+	buf->tuples[buf->n_tuples] = entry;
+	field = entry;
+
+	data_size = 0;
+	extra_size = UT_BITS_IN_BYTES(index->n_nullable);
+
+	for (i = 0; i < n_fields; i++, field++) {
+		const dict_field_t*	ifield;
+		const dict_col_t*	col;
+		ulint			col_no;
+		const dfield_t*		row_field;
+		ulint			len;
+
+		ifield = dict_index_get_nth_field(index, i);
+		col = ifield->col;
+		col_no = dict_col_get_no(col);
+		row_field = dtuple_get_nth_field(row, col_no);
+		dfield_copy(field, row_field);
+		len = dfield_get_len(field);
+
+		if (dfield_is_null(field)) {
+			ut_ad(!(col->prtype & DATA_NOT_NULL));
+			continue;
+		} else if (UNIV_LIKELY(!ext)) {
+		} else if (dict_index_is_clust(index)) {
+			/* Flag externally stored fields. */
+			const byte*	buf = row_ext_lookup(ext, col_no,
+							     &len);
+			if (UNIV_LIKELY_NULL(buf)) {
+				ut_a(buf != field_ref_zero);
+				if (i < dict_index_get_n_unique(index)) {
+					dfield_set_data(field, buf, len);
+				} else {
+					dfield_set_ext(field);
+					len = dfield_get_len(field);
+				}
+			}
+		} else {
+			const byte*	buf = row_ext_lookup(ext, col_no,
+							     &len);
+			if (UNIV_LIKELY_NULL(buf)) {
+				ut_a(buf != field_ref_zero);
+				dfield_set_data(field, buf, len);
+			}
+		}
+
+		/* If a column prefix index, take only the prefix */
+
+		if (ifield->prefix_len) {
+			len = dtype_get_at_most_n_mbchars(
+				col->prtype,
+				col->mbminlen, col->mbmaxlen,
+				ifield->prefix_len,
+				len, dfield_get_data(field));
+			dfield_set_len(field, len);
+		}
+
+		ut_ad(len <= col->len || col->mtype == DATA_BLOB);
+
+		if (ifield->fixed_len) {
+			ut_ad(len == ifield->fixed_len);
+			ut_ad(!dfield_is_ext(field));
+		} else if (dfield_is_ext(field)) {
+			extra_size += 2;
+		} else if (len < 128
+			   || (col->len < 256 && col->mtype != DATA_BLOB)) {
+			extra_size++;
+		} else {
+			/* For variable-length columns, we look up the
+			maximum length from the column itself.  If this
+			is a prefix index column shorter than 256 bytes,
+			this will waste one byte. */
+			extra_size += 2;
+		}
+		data_size += len;
+	}
+
+#ifdef UNIV_DEBUG
+	{
+		ulint	size;
+		ulint	extra;
+
+		size = rec_get_converted_size_comp(index,
+						   REC_STATUS_ORDINARY,
+						   entry, n_fields, &extra);
+
+		ut_ad(data_size + extra_size + REC_N_NEW_EXTRA_BYTES == size);
+		ut_ad(extra_size + REC_N_NEW_EXTRA_BYTES == extra);
+	}
+#endif /* UNIV_DEBUG */
+
+	/* Add to the total size of the record in row_merge_block_t
+	the encoded length of extra_size and the extra bytes (extra_size).
+	See row_merge_buf_write() for the variable-length encoding
+	of extra_size. */
+	data_size += (extra_size + 1) + ((extra_size + 1) >= 0x80);
+
+	/* The following assertion may fail if row_merge_block_t is
+	declared very small and a PRIMARY KEY is being created with
+	many prefix columns.  In that case, the record may exceed the
+	page_zip_rec_needs_ext() limit.  However, no further columns
+	will be moved to external storage until the record is inserted
+	to the clustered index B-tree. */
+	ut_ad(data_size < sizeof(row_merge_block_t));
+
+	/* Reserve one byte for the end marker of row_merge_block_t. */
+	if (buf->total_size + data_size >= sizeof(row_merge_block_t) - 1) {
+		return(FALSE);
+	}
+
+	buf->total_size += data_size;
+	buf->n_tuples++;
+
+	field = entry;
+
+	/* Copy the data fields. */
+
+	do {
+		dfield_dup(field++, buf->heap);
+	} while (--n_fields);
+
+	return(TRUE);
+}
+
+/* Structure for reporting duplicate records. */
+struct row_merge_dup_struct {
+	const dict_index_t*	index;		/* index being sorted */
+	TABLE*			table;		/* MySQL table object */
+	ulint			n_dup;		/* number of duplicates */
+};
+
+typedef struct row_merge_dup_struct row_merge_dup_t;
+
+/*****************************************************************
+Report a duplicate key. */
+static
+void
+row_merge_dup_report(
+/*=================*/
+	row_merge_dup_t*	dup,	/* in/out: for reporting duplicates */
+	const dfield_t*		entry)	/* in: duplicate index entry */
+{
+	mrec_buf_t 		buf;
+	const dtuple_t*		tuple;
+	dtuple_t		tuple_store;
+	const rec_t*		rec;
+	const dict_index_t*	index	= dup->index;
+	ulint			n_fields= dict_index_get_n_fields(index);
+	mem_heap_t*		heap	= NULL;
+	ulint			offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint*			offsets;
+	ulint			n_ext;
+
+	if (dup->n_dup++) {
+		/* Only report the first duplicate record,
+		but count all duplicate records. */
+		return;
+	}
+
+	rec_offs_init(offsets_);
+
+	/* Convert the tuple to a record and then to MySQL format. */
+
+	tuple = dtuple_from_fields(&tuple_store, entry, n_fields);
+	n_ext = dict_index_is_clust(index) ? dtuple_get_n_ext(tuple) : 0;
+
+	rec = rec_convert_dtuple_to_rec(buf, index, tuple, n_ext);
+	offsets = rec_get_offsets(rec, index, offsets_, ULINT_UNDEFINED,
+				  &heap);
+
+	innobase_rec_to_mysql(dup->table, rec, index, offsets);
+
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+}
+
+/*****************************************************************
+Compare two tuples. */
+static
+int
+row_merge_tuple_cmp(
+/*================*/
+					/* out: 1, 0, -1 if a is greater,
+					equal, less, respectively, than b */
+	ulint			n_field,/* in: number of fields */
+	const dfield_t*		a,	/* in: first tuple to be compared */
+	const dfield_t*		b,	/* in: second tuple to be compared */
+	row_merge_dup_t*	dup)	/* in/out: for reporting duplicates */
+{
+	int		cmp;
+	const dfield_t*	field	= a;
+
+	/* Compare the fields of the tuples until a difference is
+	found or we run out of fields to compare.  If !cmp at the
+	end, the tuples are equal. */
+	do {
+		cmp = cmp_dfield_dfield(a++, b++);
+	} while (!cmp && --n_field);
+
+	if (UNIV_UNLIKELY(!cmp) && UNIV_LIKELY_NULL(dup)) {
+		/* Report a duplicate value error if the tuples are
+		logically equal.  NULL columns are logically inequal,
+		although they are equal in the sorting order.  Find
+		out if any of the fields are NULL. */
+		for (b = field; b != a; b++) {
+			if (dfield_is_null(b)) {
+
+				goto func_exit;
+			}
+		}
+
+		row_merge_dup_report(dup, field);
+	}
+
+func_exit:
+	return(cmp);
+}
+
+/**************************************************************************
+Merge sort the tuple buffer in main memory. */
+static
+void
+row_merge_tuple_sort(
+/*=================*/
+	ulint			n_field,/* in: number of fields */
+	row_merge_dup_t*	dup,	/* in/out: for reporting duplicates */
+	const dfield_t**	tuples,	/* in/out: tuples */
+	const dfield_t**	aux,	/* in/out: work area */
+	ulint			low,	/* in: lower bound of the
+					sorting area, inclusive */
+	ulint			high)	/* in: upper bound of the
+					sorting area, exclusive */
+{
+#define row_merge_tuple_sort_ctx(a,b,c,d) \
+	row_merge_tuple_sort(n_field, dup, a, b, c, d)
+#define row_merge_tuple_cmp_ctx(a,b) row_merge_tuple_cmp(n_field, a, b, dup)
+
+	UT_SORT_FUNCTION_BODY(row_merge_tuple_sort_ctx,
+			      tuples, aux, low, high, row_merge_tuple_cmp_ctx);
+}
+
+/**********************************************************
+Sort a buffer. */
+static
+void
+row_merge_buf_sort(
+/*===============*/
+	row_merge_buf_t*	buf,	/* in/out: sort buffer */
+	row_merge_dup_t*	dup)	/* in/out: for reporting duplicates */
+{
+	row_merge_tuple_sort(dict_index_get_n_unique(buf->index), dup,
+			     buf->tuples, buf->tmp_tuples, 0, buf->n_tuples);
+}
+
+/**********************************************************
+Write a buffer to a block. */
+static
+void
+row_merge_buf_write(
+/*================*/
+	const row_merge_buf_t*	buf,	/* in: sorted buffer */
+#ifdef UNIV_DEBUG
+	const merge_file_t*	of,	/* in: output file */
+#endif /* UNIV_DEBUG */
+	row_merge_block_t*	block)	/* out: buffer for writing to file */
+#ifndef UNIV_DEBUG
+# define row_merge_buf_write(buf, of, block) row_merge_buf_write(buf, block)
+#endif /* !UNIV_DEBUG */
+{
+	const dict_index_t*	index	= buf->index;
+	ulint			n_fields= dict_index_get_n_fields(index);
+	byte*			b	= &(*block)[0];
+
+	ulint		i;
+
+	for (i = 0; i < buf->n_tuples; i++) {
+		ulint		size;
+		ulint		extra_size;
+		const dfield_t*	entry		= buf->tuples[i];
+
+		size = rec_get_converted_size_comp(index,
+						   REC_STATUS_ORDINARY,
+						   entry, n_fields,
+						   &extra_size);
+		ut_ad(size > extra_size);
+		ut_ad(extra_size >= REC_N_NEW_EXTRA_BYTES);
+		extra_size -= REC_N_NEW_EXTRA_BYTES;
+		size -= REC_N_NEW_EXTRA_BYTES;
+
+		/* Encode extra_size + 1 */
+		if (extra_size + 1 < 0x80) {
+			*b++ = (byte) (extra_size + 1);
+		} else {
+			ut_ad((extra_size + 1) < 0x8000);
+			*b++ = (byte) (0x80 | ((extra_size + 1) >> 8));
+			*b++ = (byte) (extra_size + 1);
+		}
+
+		ut_ad(b + size < block[1]);
+
+		rec_convert_dtuple_to_rec_comp(b + extra_size, 0, index,
+					       REC_STATUS_ORDINARY,
+					       entry, n_fields);
+
+		b += size;
+
+#ifdef UNIV_DEBUG
+		if (row_merge_print_write) {
+			fprintf(stderr, "row_merge_buf_write %p,%d,%lu %lu",
+				(void*) b, of->fd, (ulong) of->offset,
+				(ulong) i);
+			row_merge_tuple_print(stderr, entry, n_fields);
+		}
+#endif /* UNIV_DEBUG */
+	}
+
+	/* Write an "end-of-chunk" marker. */
+	ut_a(b < block[1]);
+	ut_a(b == block[0] + buf->total_size);
+	*b++ = 0;
+#ifdef UNIV_DEBUG_VALGRIND
+	/* The rest of the block is uninitialized.  Initialize it
+	to avoid bogus warnings. */
+	memset(b, 0xff, block[1] - b);
+#endif /* UNIV_DEBUG_VALGRIND */
+#ifdef UNIV_DEBUG
+	if (row_merge_print_write) {
+		fprintf(stderr, "row_merge_buf_write %p,%d,%lu EOF\n",
+			(void*) b, of->fd, (ulong) of->offset);
+	}
+#endif /* UNIV_DEBUG */
+}
+
+/**********************************************************
+Create a memory heap and allocate space for row_merge_rec_offsets(). */
+static
+mem_heap_t*
+row_merge_heap_create(
+/*==================*/
+						/* out: memory heap */
+	const dict_index_t*	index,		/* in: record descriptor */
+	ulint**			offsets1,	/* out: offsets */
+	ulint**			offsets2)	/* out: offsets */
+{
+	ulint		i	= 1 + REC_OFFS_HEADER_SIZE
+		+ dict_index_get_n_fields(index);
+	mem_heap_t*	heap	= mem_heap_create(2 * i * sizeof *offsets1);
+
+	*offsets1 = mem_heap_alloc(heap, i * sizeof *offsets1);
+	*offsets2 = mem_heap_alloc(heap, i * sizeof *offsets2);
+
+	(*offsets1)[0] = (*offsets2)[0] = i;
+	(*offsets1)[1] = (*offsets2)[1] = dict_index_get_n_fields(index);
+
+	return(heap);
+}
+
+/**************************************************************************
+Search an index object by name and column names.  If several indexes match,
+return the index with the max id. */
+static
+dict_index_t*
+row_merge_dict_table_get_index(
+/*===========================*/
+						/* out: matching index,
+						NULL if not found */
+	dict_table_t*		table,		/* in: table */
+	const merge_index_def_t*index_def)	/* in: index definition */
+{
+	ulint		i;
+	dict_index_t*	index;
+	const char**	column_names;
+
+	column_names = mem_alloc(index_def->n_fields * sizeof *column_names);
+
+	for (i = 0; i < index_def->n_fields; ++i) {
+		column_names[i] = index_def->fields[i].field_name;
+	}
+
+	index = dict_table_get_index_by_max_id(
+		table, index_def->name, column_names, index_def->n_fields);
+
+	mem_free((void*) column_names);
+
+	return(index);
+}
+
+/************************************************************************
+Read a merge block from the file system. */
+static
+ibool
+row_merge_read(
+/*===========*/
+					/* out: TRUE if request was
+					successful, FALSE if fail */
+	int			fd,	/* in: file descriptor */
+	ulint			offset,	/* in: offset where to read */
+	row_merge_block_t*	buf)	/* out: data */
+{
+	ib_uint64_t	ofs = ((ib_uint64_t) offset) * sizeof *buf;
+	ibool		success;
+
+	success = os_file_read_no_error_handling(OS_FILE_FROM_FD(fd), buf,
+						 (ulint) (ofs & 0xFFFFFFFF),
+						 (ulint) (ofs >> 32),
+						 sizeof *buf);
+	if (UNIV_UNLIKELY(!success)) {
+		ut_print_timestamp(stderr);
+		fprintf(stderr,
+			"  InnoDB: failed to read merge block at %llu\n", ofs);
+	}
+
+	return(UNIV_LIKELY(success));
+}
+
+/************************************************************************
+Read a merge block from the file system. */
+static
+ibool
+row_merge_write(
+/*============*/
+				/* out: TRUE if request was
+				successful, FALSE if fail */
+	int		fd,	/* in: file descriptor */
+	ulint		offset,	/* in: offset where to write */
+	const void*	buf)	/* in: data */
+{
+	ib_uint64_t	ofs = ((ib_uint64_t) offset)
+		* sizeof(row_merge_block_t);
+
+	return(UNIV_LIKELY(os_file_write("(merge)", OS_FILE_FROM_FD(fd), buf,
+					 (ulint) (ofs & 0xFFFFFFFF),
+					 (ulint) (ofs >> 32),
+					 sizeof(row_merge_block_t))));
+}
+
+/************************************************************************
+Read a merge record. */
+static
+const byte*
+row_merge_read_rec(
+/*===============*/
+					/* out: pointer to next record,
+					or NULL on I/O error
+					or end of list */
+	row_merge_block_t*	block,	/* in/out: file buffer */
+	mrec_buf_t*		buf,	/* in/out: secondary buffer */
+	const byte*		b,	/* in: pointer to record */
+	const dict_index_t*	index,	/* in: index of the record */
+	int			fd,	/* in: file descriptor */
+	ulint*			foffs,	/* in/out: file offset */
+	const mrec_t**		mrec,	/* out: pointer to merge record,
+					or NULL on end of list
+					(non-NULL on I/O error) */
+	ulint*			offsets)/* out: offsets of mrec */
+{
+	ulint	extra_size;
+	ulint	data_size;
+	ulint	avail_size;
+
+	ut_ad(block);
+	ut_ad(buf);
+	ut_ad(b >= block[0]);
+	ut_ad(b < block[1]);
+	ut_ad(index);
+	ut_ad(foffs);
+	ut_ad(mrec);
+	ut_ad(offsets);
+
+	ut_ad(*offsets == 1 + REC_OFFS_HEADER_SIZE
+	      + dict_index_get_n_fields(index));
+
+	extra_size = *b++;
+
+	if (UNIV_UNLIKELY(!extra_size)) {
+		/* End of list */
+		*mrec = NULL;
+#ifdef UNIV_DEBUG
+		if (row_merge_print_read) {
+			fprintf(stderr, "row_merge_read %p,%p,%d,%lu EOF\n",
+				(const void*) b, (const void*) block,
+				fd, (ulong) *foffs);
+		}
+#endif /* UNIV_DEBUG */
+		return(NULL);
+	}
+
+	if (extra_size >= 0x80) {
+		/* Read another byte of extra_size. */
+
+		if (UNIV_UNLIKELY(b >= block[1])) {
+			if (!row_merge_read(fd, ++(*foffs), block)) {
+err_exit:
+				/* Signal I/O error. */
+				*mrec = b;
+				return(NULL);
+			}
+
+			/* Wrap around to the beginning of the buffer. */
+			b = block[0];
+		}
+
+		extra_size = (extra_size & 0x7f) << 8;
+		extra_size |= *b++;
+	}
+
+	/* Normalize extra_size.  Above, value 0 signals "end of list". */
+	extra_size--;
+
+	/* Read the extra bytes. */
+
+	if (UNIV_UNLIKELY(b + extra_size >= block[1])) {
+		/* The record spans two blocks.  Copy the entire record
+		to the auxiliary buffer and handle this as a special
+		case. */
+
+		avail_size = block[1] - b;
+
+		memcpy(*buf, b, avail_size);
+
+		if (!row_merge_read(fd, ++(*foffs), block)) {
+
+			goto err_exit;
+		}
+
+		/* Wrap around to the beginning of the buffer. */
+		b = block[0];
+
+		/* Copy the record. */
+		memcpy(*buf + avail_size, b, extra_size - avail_size);
+		b += extra_size - avail_size;
+
+		*mrec = *buf + extra_size;
+
+		rec_init_offsets_comp_ordinary(*mrec, 0, index, offsets);
+
+		data_size = rec_offs_data_size(offsets);
+
+		/* These overflows should be impossible given that
+		records are much smaller than either buffer, and
+		the record starts near the beginning of each buffer. */
+		ut_a(extra_size + data_size < sizeof *buf);
+		ut_a(b + data_size < block[1]);
+
+		/* Copy the data bytes. */
+		memcpy(*buf + extra_size, b, data_size);
+		b += data_size;
+
+		goto func_exit;
+	}
+
+	*mrec = b + extra_size;
+
+	rec_init_offsets_comp_ordinary(*mrec, 0, index, offsets);
+
+	data_size = rec_offs_data_size(offsets);
+	ut_ad(extra_size + data_size < sizeof *buf);
+
+	b += extra_size + data_size;
+
+	if (UNIV_LIKELY(b < block[1])) {
+		/* The record fits entirely in the block.
+		This is the normal case. */
+		goto func_exit;
+	}
+
+	/* The record spans two blocks.  Copy it to buf. */
+
+	b -= extra_size + data_size;
+	avail_size = block[1] - b;
+	memcpy(*buf, b, avail_size);
+	*mrec = *buf + extra_size;
+	rec_offs_make_valid(*mrec, index, offsets);
+
+	if (!row_merge_read(fd, ++(*foffs), block)) {
+
+		goto err_exit;
+	}
+
+	/* Wrap around to the beginning of the buffer. */
+	b = block[0];
+
+	/* Copy the rest of the record. */
+	memcpy(*buf + avail_size, b, extra_size + data_size - avail_size);
+	b += extra_size + data_size - avail_size;
+
+func_exit:
+#ifdef UNIV_DEBUG
+	if (row_merge_print_read) {
+		fprintf(stderr, "row_merge_read %p,%p,%d,%lu ",
+			(const void*) b, (const void*) block,
+			fd, (ulong) *foffs);
+		rec_print_comp(stderr, *mrec, offsets);
+		putc('\n', stderr);
+	}
+#endif /* UNIV_DEBUG */
+
+	return(b);
+}
+
+/************************************************************************
+Write a merge record. */
+static
+void
+row_merge_write_rec_low(
+/*====================*/
+	byte*		b,	/* out: buffer */
+	ulint		e,	/* in: encoded extra_size */
+#ifdef UNIV_DEBUG
+	ulint		size,	/* in: total size to write */
+	int		fd,	/* in: file descriptor */
+	ulint		foffs,	/* in: file offset */
+#endif /* UNIV_DEBUG */
+	const mrec_t*	mrec,	/* in: record to write */
+	const ulint*	offsets)/* in: offsets of mrec */
+#ifndef UNIV_DEBUG
+# define row_merge_write_rec_low(b, e, size, fd, foffs, mrec, offsets)	\
+	row_merge_write_rec_low(b, e, mrec, offsets)
+#endif /* !UNIV_DEBUG */
+{
+#ifdef UNIV_DEBUG
+	const byte* const end = b + size;
+	ut_ad(e == rec_offs_extra_size(offsets) + 1);
+
+	if (row_merge_print_write) {
+		fprintf(stderr, "row_merge_write %p,%d,%lu ",
+			(void*) b, fd, (ulong) foffs);
+		rec_print_comp(stderr, mrec, offsets);
+		putc('\n', stderr);
+	}
+#endif /* UNIV_DEBUG */
+
+	if (e < 0x80) {
+		*b++ = (byte) e;
+	} else {
+		*b++ = (byte) (0x80 | (e >> 8));
+		*b++ = (byte) e;
+	}
+
+	memcpy(b, mrec - rec_offs_extra_size(offsets), rec_offs_size(offsets));
+	ut_ad(b + rec_offs_size(offsets) == end);
+}
+
+/************************************************************************
+Write a merge record. */
+static
+byte*
+row_merge_write_rec(
+/*================*/
+					/* out: pointer to end of block,
+					or NULL on error */
+	row_merge_block_t*	block,	/* in/out: file buffer */
+	mrec_buf_t*		buf,	/* in/out: secondary buffer */
+	byte*			b,	/* in: pointer to end of block */
+	int			fd,	/* in: file descriptor */
+	ulint*			foffs,	/* in/out: file offset */
+	const mrec_t*		mrec,	/* in: record to write */
+	const ulint*		offsets)/* in: offsets of mrec */
+{
+	ulint	extra_size;
+	ulint	size;
+	ulint	avail_size;
+
+	ut_ad(block);
+	ut_ad(buf);
+	ut_ad(b >= block[0]);
+	ut_ad(b < block[1]);
+	ut_ad(mrec);
+	ut_ad(foffs);
+	ut_ad(mrec < block[0] || mrec > block[1]);
+	ut_ad(mrec < buf[0] || mrec > buf[1]);
+
+	/* Normalize extra_size.  Value 0 signals "end of list". */
+	extra_size = rec_offs_extra_size(offsets) + 1;
+
+	size = extra_size + (extra_size >= 0x80)
+		+ rec_offs_data_size(offsets);
+
+	if (UNIV_UNLIKELY(b + size >= block[1])) {
+		/* The record spans two blocks.
+		Copy it to the temporary buffer first. */
+		avail_size = block[1] - b;
+
+		row_merge_write_rec_low(buf[0],
+					extra_size, size, fd, *foffs,
+					mrec, offsets);
+
+		/* Copy the head of the temporary buffer, write
+		the completed block, and copy the tail of the
+		record to the head of the new block. */
+		memcpy(b, buf[0], avail_size);
+
+		if (!row_merge_write(fd, (*foffs)++, block)) {
+			return(NULL);
+		}
+
+		UNIV_MEM_INVALID(block[0], sizeof block[0]);
+
+		/* Copy the rest. */
+		b = block[0];
+		memcpy(b, buf[0] + avail_size, size - avail_size);
+		b += size - avail_size;
+	} else {
+		row_merge_write_rec_low(b, extra_size, size, fd, *foffs,
+					mrec, offsets);
+		b += size;
+	}
+
+	return(b);
+}
+
+/************************************************************************
+Write an end-of-list marker. */
+static
+byte*
+row_merge_write_eof(
+/*================*/
+					/* out: pointer to end of block,
+					or NULL on error */
+	row_merge_block_t*	block,	/* in/out: file buffer */
+	byte*			b,	/* in: pointer to end of block */
+	int			fd,	/* in: file descriptor */
+	ulint*			foffs)	/* in/out: file offset */
+{
+	ut_ad(block);
+	ut_ad(b >= block[0]);
+	ut_ad(b < block[1]);
+	ut_ad(foffs);
+#ifdef UNIV_DEBUG
+	if (row_merge_print_write) {
+		fprintf(stderr, "row_merge_write %p,%p,%d,%lu EOF\n",
+			(void*) b, (void*) block, fd, (ulong) *foffs);
+	}
+#endif /* UNIV_DEBUG */
+
+	*b++ = 0;
+	UNIV_MEM_ASSERT_RW(block[0], b - block[0]);
+	UNIV_MEM_ASSERT_W(block[0], sizeof block[0]);
+#ifdef UNIV_DEBUG_VALGRIND
+	/* The rest of the block is uninitialized.  Initialize it
+	to avoid bogus warnings. */
+	memset(b, 0xff, block[1] - b);
+#endif /* UNIV_DEBUG_VALGRIND */
+
+	if (!row_merge_write(fd, (*foffs)++, block)) {
+		return(NULL);
+	}
+
+	UNIV_MEM_INVALID(block[0], sizeof block[0]);
+	return(block[0]);
+}
+
+/*****************************************************************
+Compare two merge records. */
+static
+int
+row_merge_cmp(
+/*==========*/
+						/* out: 1, 0, -1 if
+						mrec1 is greater, equal, less,
+						respectively, than mrec2 */
+	const mrec_t*		mrec1,		/* in: first merge
+						record to be compared */
+	const mrec_t*		mrec2,		/* in: second merge
+						record to be compared */
+	const ulint*		offsets1,	/* in: first record offsets */
+	const ulint*		offsets2,	/* in: second record offsets */
+	const dict_index_t*	index)		/* in: index */
+{
+	int	cmp;
+
+	cmp = cmp_rec_rec_simple(mrec1, mrec2, offsets1, offsets2, index);
+
+#ifdef UNIV_DEBUG
+	if (row_merge_print_cmp) {
+		fputs("row_merge_cmp1 ", stderr);
+		rec_print_comp(stderr, mrec1, offsets1);
+		fputs("\nrow_merge_cmp2 ", stderr);
+		rec_print_comp(stderr, mrec2, offsets2);
+		fprintf(stderr, "\nrow_merge_cmp=%d\n", cmp);
+	}
+#endif /* UNIV_DEBUG */
+
+	return(cmp);
+}
+
+/************************************************************************
+Reads clustered index of the table and create temporary files
+containing the index entries for the indexes to be built. */
+static
+ulint
+row_merge_read_clustered_index(
+/*===========================*/
+					/* out: DB_SUCCESS or error */
+	trx_t*			trx,	/* in: transaction */
+	TABLE*			table,	/* in/out: MySQL table object,
+					for reporting erroneous records */
+	const dict_table_t*	old_table,/* in: table where rows are
+					read from */
+	const dict_table_t*	new_table,/* in: table where indexes are
+					created; identical to old_table
+					unless creating a PRIMARY KEY */
+	dict_index_t**		index,	/* in: indexes to be created */
+	merge_file_t*		files,	/* in: temporary files */
+	ulint			n_index,/* in: number of indexes to create */
+	row_merge_block_t*	block)	/* in/out: file buffer */
+{
+	dict_index_t*		clust_index;	/* Clustered index */
+	mem_heap_t*		row_heap;	/* Heap memory to create
+						clustered index records */
+	row_merge_buf_t**	merge_buf;	/* Temporary list for records*/
+	btr_pcur_t		pcur;		/* Persistent cursor on the
+						clustered index */
+	mtr_t			mtr;		/* Mini transaction */
+	ulint			err = DB_SUCCESS;/* Return code */
+	ulint			i;
+	ulint			n_nonnull = 0;	/* number of columns
+						changed to NOT NULL */
+	ulint*			nonnull = NULL;	/* NOT NULL columns */
+
+	trx->op_info = "reading clustered index";
+
+	ut_ad(trx);
+	ut_ad(old_table);
+	ut_ad(new_table);
+	ut_ad(index);
+	ut_ad(files);
+
+	/* Create and initialize memory for record buffers */
+
+	merge_buf = mem_alloc(n_index * sizeof *merge_buf);
+
+	for (i = 0; i < n_index; i++) {
+		merge_buf[i] = row_merge_buf_create(index[i]);
+	}
+
+	mtr_start(&mtr);
+
+	/* Find the clustered index and create a persistent cursor
+	based on that. */
+
+	clust_index = dict_table_get_first_index(old_table);
+
+	btr_pcur_open_at_index_side(
+		TRUE, clust_index, BTR_SEARCH_LEAF, &pcur, TRUE, &mtr);
+
+	if (UNIV_UNLIKELY(old_table != new_table)) {
+		ulint	n_cols = dict_table_get_n_cols(old_table);
+
+		/* A primary key will be created.  Identify the
+		columns that were flagged NOT NULL in the new table,
+		so that we can quickly check that the records in the
+		(old) clustered index do not violate the added NOT
+		NULL constraints. */
+
+		ut_a(n_cols == dict_table_get_n_cols(new_table));
+
+		nonnull = mem_alloc(n_cols * sizeof *nonnull);
+
+		for (i = 0; i < n_cols; i++) {
+			if (dict_table_get_nth_col(old_table, i)->prtype
+			    & DATA_NOT_NULL) {
+
+				continue;
+			}
+
+			if (dict_table_get_nth_col(new_table, i)->prtype
+			    & DATA_NOT_NULL) {
+
+				nonnull[n_nonnull++] = i;
+			}
+		}
+
+		if (!n_nonnull) {
+			mem_free(nonnull);
+			nonnull = NULL;
+		}
+	}
+
+	row_heap = mem_heap_create(sizeof(mrec_buf_t));
+
+	/* Scan the clustered index. */
+	for (;;) {
+		const rec_t*	rec;
+		ulint*		offsets;
+		dtuple_t*	row		= NULL;
+		row_ext_t*	ext;
+		ibool		has_next	= TRUE;
+
+		btr_pcur_move_to_next_on_page(&pcur);
+
+		/* When switching pages, commit the mini-transaction
+		in order to release the latch on the old page. */
+
+		if (btr_pcur_is_after_last_on_page(&pcur)) {
+			btr_pcur_store_position(&pcur, &mtr);
+			mtr_commit(&mtr);
+			mtr_start(&mtr);
+			btr_pcur_restore_position(BTR_SEARCH_LEAF,
+						  &pcur, &mtr);
+			has_next = btr_pcur_move_to_next_user_rec(&pcur, &mtr);
+		}
+
+		if (UNIV_LIKELY(has_next)) {
+			rec = btr_pcur_get_rec(&pcur);
+			offsets = rec_get_offsets(rec, clust_index, NULL,
+						  ULINT_UNDEFINED, &row_heap);
+
+			/* Skip delete marked records. */
+			if (rec_get_deleted_flag(
+				    rec, dict_table_is_comp(old_table))) {
+				continue;
+			}
+
+			srv_n_rows_inserted++;
+
+			/* Build a row based on the clustered index. */
+
+			row = row_build(ROW_COPY_POINTERS, clust_index,
+					rec, offsets,
+					new_table, &ext, row_heap);
+
+			if (UNIV_LIKELY_NULL(nonnull)) {
+				for (i = 0; i < n_nonnull; i++) {
+					dfield_t*	field
+						= &row->fields[nonnull[i]];
+					dtype_t*	field_type
+						= dfield_get_type(field);
+
+					ut_a(!(field_type->prtype
+					       & DATA_NOT_NULL));
+
+					if (dfield_is_null(field)) {
+						err = DB_PRIMARY_KEY_IS_NULL;
+						i = 0;
+						goto err_exit;
+					}
+
+					field_type->prtype |= DATA_NOT_NULL;
+				}
+			}
+		}
+
+		/* Build all entries for all the indexes to be created
+		in a single scan of the clustered index. */
+
+		for (i = 0; i < n_index; i++) {
+			row_merge_buf_t*	buf	= merge_buf[i];
+			merge_file_t*		file	= &files[i];
+			const dict_index_t*	index	= buf->index;
+
+			if (UNIV_LIKELY
+			    (row && row_merge_buf_add(buf, row, ext))) {
+				continue;
+			}
+
+			/* The buffer must be sufficiently large
+			to hold at least one record. */
+			ut_ad(buf->n_tuples || !has_next);
+
+			/* We have enough data tuples to form a block.
+			Sort them and write to disk. */
+
+			if (buf->n_tuples) {
+				if (dict_index_is_unique(index)) {
+					row_merge_dup_t	dup;
+					dup.index = buf->index;
+					dup.table = table;
+					dup.n_dup = 0;
+
+					row_merge_buf_sort(buf, &dup);
+
+					if (dup.n_dup) {
+						err = DB_DUPLICATE_KEY;
+err_exit:
+						trx->error_key_num = i;
+						goto func_exit;
+					}
+				} else {
+					row_merge_buf_sort(buf, NULL);
+				}
+			}
+
+			row_merge_buf_write(buf, file, block);
+
+			if (!row_merge_write(file->fd, file->offset++,
+					     block)) {
+				err = DB_OUT_OF_FILE_SPACE;
+				goto err_exit;
+			}
+
+			UNIV_MEM_INVALID(block[0], sizeof block[0]);
+			merge_buf[i] = row_merge_buf_empty(buf);
+
+			/* Try writing the record again, now that
+			the buffer has been written out and emptied. */
+
+			if (UNIV_UNLIKELY
+			    (row && !row_merge_buf_add(buf, row, ext))) {
+				/* An empty buffer should have enough
+				room for at least one record. */
+				ut_error;
+			}
+		}
+
+		mem_heap_empty(row_heap);
+
+		if (UNIV_UNLIKELY(!has_next)) {
+			goto func_exit;
+		}
+	}
+
+func_exit:
+	btr_pcur_close(&pcur);
+	mtr_commit(&mtr);
+	mem_heap_free(row_heap);
+
+	if (UNIV_LIKELY_NULL(nonnull)) {
+		mem_free(nonnull);
+	}
+
+	for (i = 0; i < n_index; i++) {
+		row_merge_buf_free(merge_buf[i]);
+	}
+
+	mem_free(merge_buf);
+
+	trx->op_info = "";
+
+	return(err);
+}
+
+/*****************************************************************
+Merge two blocks of linked lists on disk and write a bigger block. */
+static
+ulint
+row_merge_blocks(
+/*=============*/
+					/* out: DB_SUCCESS or error code */
+	const dict_index_t*	index,	/* in: index being created */
+	merge_file_t*		file,	/* in/out: file containing
+					index entries */
+	row_merge_block_t*	block,	/* in/out: 3 buffers */
+	ulint*			foffs0,	/* in/out: offset of first
+					source list in the file */
+	ulint*			foffs1,	/* in/out: offset of second
+					source list in the file */
+	merge_file_t*		of,	/* in/out: output file */
+	TABLE*			table)	/* in/out: MySQL table, for
+					reporting erroneous key value
+					if applicable */
+{
+	mem_heap_t*	heap;	/* memory heap for offsets0, offsets1 */
+
+	mrec_buf_t	buf[3];	/* buffer for handling split mrec in block[] */
+	const byte*	b0;	/* pointer to block[0] */
+	const byte*	b1;	/* pointer to block[1] */
+	byte*		b2;	/* pointer to block[2] */
+	const mrec_t*	mrec0;	/* merge rec, points to block[0] or buf[0] */
+	const mrec_t*	mrec1;	/* merge rec, points to block[1] or buf[1] */
+	ulint*		offsets0;/* offsets of mrec0 */
+	ulint*		offsets1;/* offsets of mrec1 */
+
+	heap = row_merge_heap_create(index, &offsets0, &offsets1);
+
+	/* Write a record and read the next record.  Split the output
+	file in two halves, which can be merged on the following pass. */
+#define ROW_MERGE_WRITE_GET_NEXT(N, AT_END)				\
+	do {								\
+		b2 = row_merge_write_rec(&block[2], &buf[2], b2,	\
+					 of->fd, &of->offset,		\
+					 mrec##N, offsets##N);		\
+		if (UNIV_UNLIKELY(!b2)) {				\
+			goto corrupt;					\
+		}							\
+		b##N = row_merge_read_rec(&block[N], &buf[N],		\
+					  b##N, index,			\
+					  file->fd, foffs##N,		\
+					  &mrec##N, offsets##N);	\
+		if (UNIV_UNLIKELY(!b##N)) {				\
+			if (mrec##N) {					\
+				goto corrupt;				\
+			}						\
+			AT_END;						\
+		}							\
+	} while (0)
+
+	if (!row_merge_read(file->fd, *foffs0, &block[0])
+	    || !row_merge_read(file->fd, *foffs1, &block[1])) {
+corrupt:
+		mem_heap_free(heap);
+		return(DB_CORRUPTION);
+	}
+
+	b0 = block[0];
+	b1 = block[1];
+	b2 = block[2];
+
+	b0 = row_merge_read_rec(&block[0], &buf[0], b0, index, file->fd,
+				foffs0, &mrec0, offsets0);
+	b1 = row_merge_read_rec(&block[1], &buf[1], b1, index, file->fd,
+				foffs1, &mrec1, offsets1);
+	if (UNIV_UNLIKELY(!b0 && mrec0)
+	    || UNIV_UNLIKELY(!b1 && mrec1)) {
+
+		goto corrupt;
+	}
+
+	while (mrec0 && mrec1) {
+		switch (row_merge_cmp(mrec0, mrec1,
+				      offsets0, offsets1, index)) {
+		case 0:
+			if (UNIV_UNLIKELY
+			    (dict_index_is_unique(index))) {
+				innobase_rec_to_mysql(table, mrec0,
+						      index, offsets0);
+				mem_heap_free(heap);
+				return(DB_DUPLICATE_KEY);
+			}
+			/* fall through */
+		case -1:
+			ROW_MERGE_WRITE_GET_NEXT(0, goto merged);
+			break;
+		case 1:
+			ROW_MERGE_WRITE_GET_NEXT(1, goto merged);
+			break;
+		default:
+			ut_error;
+		}
+
+	}
+
+merged:
+	if (mrec0) {
+		/* append all mrec0 to output */
+		for (;;) {
+			ROW_MERGE_WRITE_GET_NEXT(0, goto done0);
+		}
+	}
+done0:
+	if (mrec1) {
+		/* append all mrec1 to output */
+		for (;;) {
+			ROW_MERGE_WRITE_GET_NEXT(1, goto done1);
+		}
+	}
+done1:
+
+	mem_heap_free(heap);
+	b2 = row_merge_write_eof(&block[2], b2, of->fd, &of->offset);
+	return(b2 ? DB_SUCCESS : DB_CORRUPTION);
+}
+
+/*****************************************************************
+Merge disk files. */
+static
+ulint
+row_merge(
+/*======*/
+					/* out: DB_SUCCESS or error code */
+	const dict_index_t*	index,	/* in: index being created */
+	merge_file_t*		file,	/* in/out: file containing
+					index entries */
+	ulint			half,	/* in: half the file */
+	row_merge_block_t*	block,	/* in/out: 3 buffers */
+	int*			tmpfd,	/* in/out: temporary file handle */
+	TABLE*			table)	/* in/out: MySQL table, for
+					reporting erroneous key value
+					if applicable */
+{
+	ulint		foffs0;	/* first input offset */
+	ulint		foffs1;	/* second input offset */
+	ulint		error;	/* error code */
+	merge_file_t	of;	/* output file */
+
+	UNIV_MEM_ASSERT_W(block[0], 3 * sizeof block[0]);
+	ut_ad(half > 0);
+
+	of.fd = *tmpfd;
+	of.offset = 0;
+
+	/* Merge blocks to the output file. */
+	foffs0 = 0;
+	foffs1 = half;
+
+	for (; foffs0 < half && foffs1 < file->offset; foffs0++, foffs1++) {
+		error = row_merge_blocks(index, file, block,
+					 &foffs0, &foffs1, &of, table);
+
+		if (error != DB_SUCCESS) {
+			return(error);
+		}
+	}
+
+	/* Copy the last block, if there is one. */
+	while (foffs0 < half) {
+		if (!row_merge_read(file->fd, foffs0++, block)
+		    || !row_merge_write(of.fd, of.offset++, block)) {
+			return(DB_CORRUPTION);
+		}
+	}
+	while (foffs1 < file->offset) {
+		if (!row_merge_read(file->fd, foffs1++, block)
+		    || !row_merge_write(of.fd, of.offset++, block)) {
+			return(DB_CORRUPTION);
+		}
+	}
+
+	/* Swap file descriptors for the next pass. */
+	*tmpfd = file->fd;
+	*file = of;
+
+	UNIV_MEM_INVALID(block[0], 3 * sizeof block[0]);
+
+	return(DB_SUCCESS);
+}
+
+/*****************************************************************
+Merge disk files. */
+static
+ulint
+row_merge_sort(
+/*===========*/
+					/* out: DB_SUCCESS or error code */
+	const dict_index_t*	index,	/* in: index being created */
+	merge_file_t*		file,	/* in/out: file containing
+					index entries */
+	row_merge_block_t*	block,	/* in/out: 3 buffers */
+	int*			tmpfd,	/* in/out: temporary file handle */
+	TABLE*			table)	/* in/out: MySQL table, for
+					reporting erroneous key value
+					if applicable */
+{
+	ulint	blksz;	/* block size */
+
+	for (blksz = 1; blksz < file->offset; blksz *= 2) {
+		ulint	half;
+		ulint	error;
+
+		ut_ad(ut_is_2pow(blksz));
+		half = ut_2pow_round((file->offset + (blksz - 1)) / 2, blksz);
+		error = row_merge(index, file, half, block, tmpfd, table);
+
+		if (error != DB_SUCCESS) {
+			return(error);
+		}
+	}
+
+	return(DB_SUCCESS);
+}
+
+/*****************************************************************
+Copy externally stored columns to the data tuple. */
+static
+void
+row_merge_copy_blobs(
+/*=================*/
+	const mrec_t*	mrec,	/* in: merge record */
+	const ulint*	offsets,/* in: offsets of mrec */
+	ulint		zip_size,/* in: compressed page size in bytes, or 0 */
+	dtuple_t*	tuple,	/* in/out: data tuple */
+	mem_heap_t*	heap)	/* in/out: memory heap */
+{
+	ulint	i;
+	ulint	n_fields = dtuple_get_n_fields(tuple);
+
+	for (i = 0; i < n_fields; i++) {
+		ulint		len;
+		const void*	data;
+		dfield_t*	field = dtuple_get_nth_field(tuple, i);
+
+		if (!dfield_is_ext(field)) {
+			continue;
+		}
+
+		ut_ad(!dfield_is_null(field));
+
+		/* The table is locked during index creation.
+		Therefore, externally stored columns cannot possibly
+		be freed between the time the BLOB pointers are read
+		(row_merge_read_clustered_index()) and dereferenced
+		(below). */
+		data = btr_rec_copy_externally_stored_field(
+			mrec, offsets, zip_size, i, &len, heap);
+
+		dfield_set_data(field, data, len);
+	}
+}
+
+/************************************************************************
+Read sorted file containing index data tuples and insert these data
+tuples to the index */
+static
+ulint
+row_merge_insert_index_tuples(
+/*==========================*/
+					/* out: DB_SUCCESS or error number */
+	trx_t*			trx,	/* in: transaction */
+	dict_index_t*		index,	/* in: index */
+	dict_table_t*		table,	/* in: new table */
+	ulint			zip_size,/* in: compressed page size of
+					 the old table, or 0 if uncompressed */
+	int			fd,	/* in: file descriptor */
+	row_merge_block_t*	block)	/* in/out: file buffer */
+{
+	mrec_buf_t		buf;
+	const byte*		b;
+	que_thr_t*		thr;
+	ins_node_t*		node;
+	mem_heap_t*		tuple_heap;
+	mem_heap_t*		graph_heap;
+	ulint			error = DB_SUCCESS;
+	ulint			foffs = 0;
+	ulint*			offsets;
+
+	ut_ad(trx);
+	ut_ad(index);
+	ut_ad(table);
+
+	/* We use the insert query graph as the dummy graph
+	needed in the row module call */
+
+	trx->op_info = "inserting index entries";
+
+	graph_heap = mem_heap_create(500);
+	node = ins_node_create(INS_DIRECT, table, graph_heap);
+
+	thr = pars_complete_graph_for_exec(node, trx, graph_heap);
+
+	que_thr_move_to_run_state_for_mysql(thr, trx);
+
+	tuple_heap = mem_heap_create(1000);
+
+	{
+		ulint i	= 1 + REC_OFFS_HEADER_SIZE
+			+ dict_index_get_n_fields(index);
+		offsets = mem_heap_alloc(graph_heap, i * sizeof *offsets);
+		offsets[0] = i;
+		offsets[1] = dict_index_get_n_fields(index);
+	}
+
+	b = *block;
+
+	if (!row_merge_read(fd, foffs, block)) {
+		error = DB_CORRUPTION;
+	} else {
+		for (;;) {
+			const mrec_t*	mrec;
+			dtuple_t*	dtuple;
+			ulint		n_ext;
+
+			b = row_merge_read_rec(block, &buf, b, index,
+					       fd, &foffs, &mrec, offsets);
+			if (UNIV_UNLIKELY(!b)) {
+				/* End of list, or I/O error */
+				if (mrec) {
+					error = DB_CORRUPTION;
+				}
+				break;
+			}
+
+			dtuple = row_rec_to_index_entry_low(
+				mrec, index, offsets, &n_ext, tuple_heap);
+
+			if (UNIV_UNLIKELY(n_ext)) {
+				row_merge_copy_blobs(mrec, offsets, zip_size,
+						     dtuple, tuple_heap);
+			}
+
+			node->row = dtuple;
+			node->table = table;
+			node->trx_id = trx->id;
+
+			ut_ad(dtuple_validate(dtuple));
+
+			do {
+				thr->run_node = thr;
+				thr->prev_node = thr->common.parent;
+
+				error = row_ins_index_entry(index, dtuple,
+							    0, FALSE, thr);
+
+				if (UNIV_LIKELY(error == DB_SUCCESS)) {
+
+					goto next_rec;
+				}
+
+				thr->lock_state = QUE_THR_LOCK_ROW;
+				trx->error_state = error;
+				que_thr_stop_for_mysql(thr);
+				thr->lock_state = QUE_THR_LOCK_NOLOCK;
+			} while (row_mysql_handle_errors(&error, trx,
+							 thr, NULL));
+
+			goto err_exit;
+next_rec:
+			mem_heap_empty(tuple_heap);
+		}
+	}
+
+	que_thr_stop_for_mysql_no_error(thr, trx);
+err_exit:
+	que_graph_free(thr->graph);
+
+	trx->op_info = "";
+
+	mem_heap_free(tuple_heap);
+
+	return(error);
+}
+
+/*************************************************************************
+Sets an exclusive lock on a table, for the duration of creating indexes. */
+UNIV_INTERN
+ulint
+row_merge_lock_table(
+/*=================*/
+					/* out: error code or DB_SUCCESS */
+	trx_t*		trx,		/* in/out: transaction */
+	dict_table_t*	table,		/* in: table to lock */
+	enum lock_mode	mode)		/* in: LOCK_X or LOCK_S */
+{
+	mem_heap_t*	heap;
+	que_thr_t*	thr;
+	ulint		err;
+	sel_node_t*	node;
+
+	ut_ad(trx);
+	ut_ad(trx->mysql_thread_id == os_thread_get_curr_id());
+	ut_ad(mode == LOCK_X || mode == LOCK_S);
+
+	heap = mem_heap_create(512);
+
+	trx->op_info = "setting table lock for creating or dropping index";
+
+	node = sel_node_create(heap);
+	thr = pars_complete_graph_for_exec(node, trx, heap);
+	thr->graph->state = QUE_FORK_ACTIVE;
+
+	/* We use the select query graph as the dummy graph needed
+	in the lock module call */
+
+	thr = que_fork_get_first_thr(que_node_get_parent(thr));
+	que_thr_move_to_run_state_for_mysql(thr, trx);
+
+run_again:
+	thr->run_node = thr;
+	thr->prev_node = thr->common.parent;
+
+	err = lock_table(0, table, mode, thr);
+
+	trx->error_state = err;
+
+	if (UNIV_LIKELY(err == DB_SUCCESS)) {
+		que_thr_stop_for_mysql_no_error(thr, trx);
+	} else {
+		que_thr_stop_for_mysql(thr);
+
+		if (err != DB_QUE_THR_SUSPENDED) {
+			ibool	was_lock_wait;
+
+			was_lock_wait = row_mysql_handle_errors(
+				&err, trx, thr, NULL);
+
+			if (was_lock_wait) {
+				goto run_again;
+			}
+		} else {
+			que_thr_t*	run_thr;
+			que_node_t*	parent;
+
+			parent = que_node_get_parent(thr);
+			run_thr = que_fork_start_command(parent);
+
+			ut_a(run_thr == thr);
+
+			/* There was a lock wait but the thread was not
+			in a ready to run or running state. */
+			trx->error_state = DB_LOCK_WAIT;
+
+			goto run_again;
+		}
+	}
+
+	que_graph_free(thr->graph);
+	trx->op_info = "";
+
+	return(err);
+}
+
+/*************************************************************************
+Drop an index from the InnoDB system tables.  The data dictionary must
+have been locked exclusively by the caller, because the transaction
+will not be committed. */
+UNIV_INTERN
+void
+row_merge_drop_index(
+/*=================*/
+	dict_index_t*	index,	/* in: index to be removed */
+	dict_table_t*	table,	/* in: table */
+	trx_t*		trx)	/* in: transaction handle */
+{
+	ulint		err;
+	pars_info_t*	info = pars_info_create();
+
+	/* We use the private SQL parser of Innobase to generate the
+	query graphs needed in deleting the dictionary data from system
+	tables in Innobase. Deleting a row from SYS_INDEXES table also
+	frees the file segments of the B-tree associated with the index. */
+
+	static const char str1[] =
+		"PROCEDURE DROP_INDEX_PROC () IS\n"
+		"BEGIN\n"
+		"DELETE FROM SYS_FIELDS WHERE INDEX_ID = :indexid;\n"
+		"DELETE FROM SYS_INDEXES WHERE ID = :indexid\n"
+		"		AND TABLE_ID = :tableid;\n"
+		"END;\n";
+
+	ut_ad(index && table && trx);
+
+	pars_info_add_dulint_literal(info, "indexid", index->id);
+	pars_info_add_dulint_literal(info, "tableid", table->id);
+
+	trx_start_if_not_started(trx);
+	trx->op_info = "dropping index";
+
+	ut_a(trx->dict_operation_lock_mode == RW_X_LATCH);
+
+	err = que_eval_sql(info, str1, FALSE, trx);
+
+	ut_a(err == DB_SUCCESS);
+
+	/* Replace this index with another equivalent index for all
+	foreign key constraints on this table where this index is used */
+
+	dict_table_replace_index_in_foreign_list(table, index);
+	dict_index_remove_from_cache(table, index);
+
+	trx->op_info = "";
+}
+
+/*************************************************************************
+Drop those indexes which were created before an error occurred when
+building an index.  The data dictionary must have been locked
+exclusively by the caller, because the transaction will not be
+committed. */
+UNIV_INTERN
+void
+row_merge_drop_indexes(
+/*===================*/
+	trx_t*		trx,		/* in: transaction */
+	dict_table_t*	table,		/* in: table containing the indexes */
+	dict_index_t**	index,		/* in: indexes to drop */
+	ulint		num_created)	/* in: number of elements in index[] */
+{
+	ulint	key_num;
+
+	for (key_num = 0; key_num < num_created; key_num++) {
+		row_merge_drop_index(index[key_num], table, trx);
+	}
+}
+
+/*************************************************************************
+Drop all partially created indexes during crash recovery. */
+UNIV_INTERN
+void
+row_merge_drop_temp_indexes(void)
+/*=============================*/
+{
+	trx_t*		trx;
+	ulint		err;
+
+	/* We use the private SQL parser of Innobase to generate the
+	query graphs needed in deleting the dictionary data from system
+	tables in Innobase. Deleting a row from SYS_INDEXES table also
+	frees the file segments of the B-tree associated with the index. */
+#if TEMP_INDEX_PREFIX != '\377'
+# error "TEMP_INDEX_PREFIX != '\377'"
+#endif
+	static const char drop_temp_indexes[] =
+		"PROCEDURE DROP_TEMP_INDEXES_PROC () IS\n"
+		"indexid CHAR;\n"
+		"DECLARE CURSOR c IS SELECT ID FROM SYS_INDEXES\n"
+		"WHERE SUBSTR(NAME,0,1)='\377';\n"
+		"BEGIN\n"
+		"\tOPEN c;\n"
+		"\tWHILE 1=1 LOOP\n"
+		"\t\tFETCH c INTO indexid;\n"
+		"\t\tIF (SQL % NOTFOUND) THEN\n"
+		"\t\t\tEXIT;\n"
+		"\t\tEND IF;\n"
+		"\t\tDELETE FROM SYS_FIELDS WHERE INDEX_ID = indexid;\n"
+		"\t\tDELETE FROM SYS_INDEXES WHERE ID = indexid;\n"
+		"\tEND LOOP;\n"
+		"\tCLOSE c;\n"
+		"\tCOMMIT WORK;\n"
+		"END;\n";
+
+	trx = trx_allocate_for_background();
+	trx->op_info = "dropping partially created indexes";
+	row_mysql_lock_data_dictionary(trx);
+
+	/* Incomplete transactions may be holding some locks on the
+	data dictionary tables.  However, they should never have been
+	able to lock the records corresponding to the partially
+	created indexes that we are attempting to delete, because the
+	table was locked when the indexes were being created.  We will
+	drop the partially created indexes before the rollback of
+	incomplete transactions is initiated.  Thus, this should not
+	interfere with the incomplete transactions. */
+	trx->isolation_level = TRX_ISO_READ_UNCOMMITTED;
+	err = que_eval_sql(NULL, drop_temp_indexes, FALSE, trx);
+	ut_a(err == DB_SUCCESS);
+
+	row_mysql_unlock_data_dictionary(trx);
+	trx_free_for_background(trx);
+}
+
+/*************************************************************************
+Create a merge file. */
+static
+void
+row_merge_file_create(
+/*==================*/
+	merge_file_t*	merge_file)	/* out: merge file structure */
+{
+	merge_file->fd = innobase_mysql_tmpfile();
+	merge_file->offset = 0;
+}
+
+/*************************************************************************
+Destroy a merge file. */
+static
+void
+row_merge_file_destroy(
+/*===================*/
+	merge_file_t*	merge_file)	/* out: merge file structure */
+{
+	if (merge_file->fd != -1) {
+		close(merge_file->fd);
+		merge_file->fd = -1;
+	}
+}
+
+/*************************************************************************
+Determine the precise type of a column that is added to a tem
+if a column must be constrained NOT NULL. */
+UNIV_INLINE
+ulint
+row_merge_col_prtype(
+/*=================*/
+						/* out: col->prtype, possibly
+						ORed with DATA_NOT_NULL */
+	const dict_col_t*	col,		/* in: column */
+	const char*		col_name,	/* in: name of the column */
+	const merge_index_def_t*index_def)	/* in: the index definition
+						of the primary key */
+{
+	ulint	prtype = col->prtype;
+	ulint	i;
+
+	ut_ad(index_def->ind_type & DICT_CLUSTERED);
+
+	if (prtype & DATA_NOT_NULL) {
+
+		return(prtype);
+	}
+
+	/* All columns that are included
+	in the PRIMARY KEY must be NOT NULL. */
+
+	for (i = 0; i < index_def->n_fields; i++) {
+		if (!strcmp(col_name, index_def->fields[i].field_name)) {
+			return(prtype | DATA_NOT_NULL);
+		}
+	}
+
+	return(prtype);
+}
+
+/*************************************************************************
+Create a temporary table for creating a primary key, using the definition
+of an existing table. */
+UNIV_INTERN
+dict_table_t*
+row_merge_create_temporary_table(
+/*=============================*/
+						/* out: table,
+						or NULL on error */
+	const char*		table_name,	/* in: new table name */
+	const merge_index_def_t*index_def,	/* in: the index definition
+						of the primary key */
+	const dict_table_t*	table,		/* in: old table definition */
+	trx_t*			trx)		/* in/out: transaction
+						(sets error_state) */
+{
+	ulint		i;
+	dict_table_t*	new_table = NULL;
+	ulint		n_cols = dict_table_get_n_user_cols(table);
+	ulint		error;
+	mem_heap_t*	heap = mem_heap_create(1000);
+
+	ut_ad(table_name);
+	ut_ad(index_def);
+	ut_ad(table);
+	ut_ad(mutex_own(&dict_sys->mutex));
+
+	new_table = dict_mem_table_create(table_name, 0, n_cols, table->flags);
+
+	for (i = 0; i < n_cols; i++) {
+		const dict_col_t*	col;
+		const char*		col_name;
+
+		col = dict_table_get_nth_col(table, i);
+		col_name = dict_table_get_col_name(table, i);
+
+		dict_mem_table_add_col(new_table, heap, col_name, col->mtype,
+				       row_merge_col_prtype(col, col_name,
+							    index_def),
+				       col->len);
+	}
+
+	error = row_create_table_for_mysql(new_table, trx);
+	mem_heap_free(heap);
+
+	if (error != DB_SUCCESS) {
+		trx->error_state = error;
+		new_table = NULL;
+	}
+
+	return(new_table);
+}
+
+/*************************************************************************
+Rename the temporary indexes in the dictionary to permanent ones.  The
+data dictionary must have been locked exclusively by the caller,
+because the transaction will not be committed. */
+UNIV_INTERN
+ulint
+row_merge_rename_indexes(
+/*=====================*/
+					/* out: DB_SUCCESS if all OK */
+	trx_t*		trx,		/* in/out: transaction */
+	dict_table_t*	table)		/* in/out: table with new indexes */
+{
+	ulint		err = DB_SUCCESS;
+	pars_info_t*	info = pars_info_create();
+
+	/* We use the private SQL parser of Innobase to generate the
+	query graphs needed in renaming indexes. */
+
+#if TEMP_INDEX_PREFIX != '\377'
+# error "TEMP_INDEX_PREFIX != '\377'"
+#endif
+
+	static const char rename_indexes[] =
+		"PROCEDURE RENAME_INDEXES_PROC () IS\n"
+		"BEGIN\n"
+		"UPDATE SYS_INDEXES SET NAME=SUBSTR(NAME,1,LENGTH(NAME)-1)\n"
+		"WHERE TABLE_ID = :tableid AND SUBSTR(NAME,0,1)='\377';\n"
+		"END;\n";
+
+	ut_ad(table);
+	ut_ad(trx);
+	ut_a(trx->dict_operation_lock_mode == RW_X_LATCH);
+
+	trx->op_info = "renaming indexes";
+
+	pars_info_add_dulint_literal(info, "tableid", table->id);
+
+	err = que_eval_sql(info, rename_indexes, FALSE, trx);
+
+	if (err == DB_SUCCESS) {
+		dict_index_t*	index = dict_table_get_first_index(table);
+		do {
+			if (*index->name == TEMP_INDEX_PREFIX) {
+				index->name++;
+			}
+			index = dict_table_get_next_index(index);
+		} while (index);
+	}
+
+	trx->op_info = "";
+
+	return(err);
+}
+
+/*************************************************************************
+Rename the tables in the data dictionary.  The data dictionary must
+have been locked exclusively by the caller, because the transaction
+will not be committed. */
+UNIV_INTERN
+ulint
+row_merge_rename_tables(
+/*====================*/
+					/* out: error code or DB_SUCCESS */
+	dict_table_t*	old_table,	/* in/out: old table, renamed to
+					tmp_name */
+	dict_table_t*	new_table,	/* in/out: new table, renamed to
+					old_table->name */
+	const char*	tmp_name,	/* in: new name for old_table */
+	trx_t*		trx)		/* in: transaction handle */
+{
+	ulint		err	= DB_ERROR;
+	pars_info_t*	info;
+	const char*	old_name= old_table->name;
+
+	ut_ad(trx->mysql_thread_id == os_thread_get_curr_id());
+	ut_ad(old_table != new_table);
+	ut_ad(mutex_own(&dict_sys->mutex));
+
+	ut_a(trx->dict_operation_lock_mode == RW_X_LATCH);
+
+	trx->op_info = "renaming tables";
+
+	/* We use the private SQL parser of Innobase to generate the query
+	graphs needed in updating the dictionary data in system tables. */
+
+	info = pars_info_create();
+
+	pars_info_add_str_literal(info, "new_name", new_table->name);
+	pars_info_add_str_literal(info, "old_name", old_name);
+	pars_info_add_str_literal(info, "tmp_name", tmp_name);
+
+	err = que_eval_sql(info,
+			   "PROCEDURE RENAME_TABLES () IS\n"
+			   "BEGIN\n"
+			   "UPDATE SYS_TABLES SET NAME = :tmp_name\n"
+			   " WHERE NAME = :old_name;\n"
+			   "UPDATE SYS_TABLES SET NAME = :old_name\n"
+			   " WHERE NAME = :new_name;\n"
+			   "END;\n", FALSE, trx);
+
+	if (err != DB_SUCCESS) {
+
+		goto err_exit;
+	}
+
+	/* The following calls will also rename the .ibd data files if
+	the tables are stored in a single-table tablespace */
+
+	if (!dict_table_rename_in_cache(old_table, tmp_name, FALSE)
+	    || !dict_table_rename_in_cache(new_table, old_name, FALSE)) {
+
+		err = DB_ERROR;
+		goto err_exit;
+	}
+
+	err = dict_load_foreigns(old_name, TRUE);
+
+	if (err != DB_SUCCESS) {
+err_exit:
+		trx->error_state = DB_SUCCESS;
+		trx_general_rollback_for_mysql(trx, FALSE, NULL);
+		trx->error_state = DB_SUCCESS;
+	}
+
+	trx->op_info = "";
+
+	return(err);
+}
+
+/*************************************************************************
+Create and execute a query graph for creating an index. */
+static
+ulint
+row_merge_create_index_graph(
+/*=========================*/
+					/* out: DB_SUCCESS or error code */
+	trx_t*		trx,		/* in: trx */
+	dict_table_t*	table,		/* in: table */
+	dict_index_t*	index)		/* in: index */
+{
+	ind_node_t*	node;		/* Index creation node */
+	mem_heap_t*	heap;		/* Memory heap */
+	que_thr_t*	thr;		/* Query thread */
+	ulint		err;
+
+	ut_ad(trx);
+	ut_ad(table);
+	ut_ad(index);
+
+	heap = mem_heap_create(512);
+
+	index->table = table;
+	node = ind_create_graph_create(index, heap);
+	thr = pars_complete_graph_for_exec(node, trx, heap);
+
+	ut_a(thr == que_fork_start_command(que_node_get_parent(thr)));
+
+	que_run_threads(thr);
+
+	err = trx->error_state;
+
+	que_graph_free((que_t*) que_node_get_parent(thr));
+
+	return(err);
+}
+
+/*************************************************************************
+Create the index and load in to the dictionary. */
+UNIV_INTERN
+dict_index_t*
+row_merge_create_index(
+/*===================*/
+					/* out: index, or NULL on error */
+	trx_t*		trx,		/* in/out: trx (sets error_state) */
+	dict_table_t*	table,		/* in: the index is on this table */
+	const merge_index_def_t*	/* in: the index definition */
+			index_def)
+{
+	dict_index_t*	index;
+	ulint		err;
+	ulint		n_fields = index_def->n_fields;
+	ulint		i;
+
+	/* Create the index prototype, using the passed in def, this is not
+	a persistent operation. We pass 0 as the space id, and determine at
+	a lower level the space id where to store the table. */
+
+	index = dict_mem_index_create(table->name, index_def->name,
+				      0, index_def->ind_type, n_fields);
+
+	ut_a(index);
+
+	for (i = 0; i < n_fields; i++) {
+		merge_index_field_t*	ifield = &index_def->fields[i];
+
+		dict_mem_index_add_field(index, ifield->field_name,
+					 ifield->prefix_len);
+	}
+
+	/* Add the index to SYS_INDEXES, using the index prototype. */
+	err = row_merge_create_index_graph(trx, table, index);
+
+	if (err == DB_SUCCESS) {
+
+		index = row_merge_dict_table_get_index(
+			table, index_def);
+
+		ut_a(index);
+
+#ifdef ROW_MERGE_IS_INDEX_USABLE
+		/* Note the id of the transaction that created this
+		index, we use it to restrict readers from accessing
+		this index, to ensure read consistency. */
+		index->trx_id = trx->id;
+#endif /* ROW_MERGE_IS_INDEX_USABLE */
+	} else {
+		index = NULL;
+	}
+
+	return(index);
+}
+
+#ifdef ROW_MERGE_IS_INDEX_USABLE
+/*************************************************************************
+Check if a transaction can use an index. */
+UNIV_INTERN
+ibool
+row_merge_is_index_usable(
+/*======================*/
+	const trx_t*		trx,	/* in: transaction */
+	const dict_index_t*	index)	/* in: index to check */
+{
+	if (!trx->read_view) {
+		return(TRUE);
+	}
+
+	return(ut_dulint_cmp(index->trx_id, trx->read_view->low_limit_id) < 0);
+}
+#endif /* ROW_MERGE_IS_INDEX_USABLE */
+
+/*************************************************************************
+Drop the old table. */
+UNIV_INTERN
+ulint
+row_merge_drop_table(
+/*=================*/
+					/* out: DB_SUCCESS or error code */
+	trx_t*		trx,		/* in: transaction */
+	dict_table_t*	table)		/* in: table to drop */
+{
+	/* There must be no open transactions on the table. */
+	ut_a(table->n_mysql_handles_opened == 0);
+
+	return(row_drop_table_for_mysql(table->name, trx, FALSE));
+}
+
+/*************************************************************************
+Build indexes on a table by reading a clustered index,
+creating a temporary file containing index entries, merge sorting
+these index entries and inserting sorted index entries to indexes. */
+UNIV_INTERN
+ulint
+row_merge_build_indexes(
+/*====================*/
+					/* out: DB_SUCCESS or error code */
+	trx_t*		trx,		/* in: transaction */
+	dict_table_t*	old_table,	/* in: table where rows are
+					read from */
+	dict_table_t*	new_table,	/* in: table where indexes are
+					created; identical to old_table
+					unless creating a PRIMARY KEY */
+	dict_index_t**	indexes,	/* in: indexes to be created */
+	ulint		n_indexes,	/* in: size of indexes[] */
+	TABLE*		table)		/* in/out: MySQL table, for
+					reporting erroneous key value
+					if applicable */
+{
+	merge_file_t*		merge_files;
+	row_merge_block_t*	block;
+	ulint			block_size;
+	ulint			i;
+	ulint			error;
+	int			tmpfd;
+
+	ut_ad(trx);
+	ut_ad(old_table);
+	ut_ad(new_table);
+	ut_ad(indexes);
+	ut_ad(n_indexes);
+
+	trx_start_if_not_started(trx);
+
+	/* Allocate memory for merge file data structure and initialize
+	fields */
+
+	merge_files = mem_alloc(n_indexes * sizeof *merge_files);
+	block_size = 3 * sizeof *block;
+	block = os_mem_alloc_large(&block_size);
+
+	for (i = 0; i < n_indexes; i++) {
+
+		row_merge_file_create(&merge_files[i]);
+	}
+
+	tmpfd = innobase_mysql_tmpfile();
+
+	/* Reset the MySQL row buffer that is used when reporting
+	duplicate keys. */
+	innobase_rec_reset(table);
+
+	/* Read clustered index of the table and create files for
+	secondary index entries for merge sort */
+
+	error = row_merge_read_clustered_index(
+		trx, table, old_table, new_table, indexes,
+		merge_files, n_indexes, block);
+
+	if (error != DB_SUCCESS) {
+
+		goto func_exit;
+	}
+
+	/* Now we have files containing index entries ready for
+	sorting and inserting. */
+
+	for (i = 0; i < n_indexes; i++) {
+		error = row_merge_sort(indexes[i], &merge_files[i],
+				       block, &tmpfd, table);
+
+		if (error == DB_SUCCESS) {
+			error = row_merge_insert_index_tuples(
+				trx, indexes[i], new_table,
+				dict_table_zip_size(old_table),
+				merge_files[i].fd, block);
+		}
+
+		/* Close the temporary file to free up space. */
+		row_merge_file_destroy(&merge_files[i]);
+
+		if (error != DB_SUCCESS) {
+			trx->error_key_num = i;
+			goto func_exit;
+		}
+	}
+
+func_exit:
+	close(tmpfd);
+
+	for (i = 0; i < n_indexes; i++) {
+		row_merge_file_destroy(&merge_files[i]);
+	}
+
+	mem_free(merge_files);
+	os_mem_free_large(block, block_size);
+
+	return(error);
+}
diff --git a/storage/xtradb/row/row0mysql.c b/storage/xtradb/row/row0mysql.c
new file mode 100644
index 00000000000..8204285cc38
--- /dev/null
+++ b/storage/xtradb/row/row0mysql.c
@@ -0,0 +1,4209 @@
+/*****************************************************************************
+
+Copyright (c) 2000, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Interface between Innobase row operations and MySQL.
+Contains also create table and other data dictionary operations.
+
+Created 9/17/2000 Heikki Tuuri
+*******************************************************/
+
+#include "row0mysql.h"
+
+#ifdef UNIV_NONINL
+#include "row0mysql.ic"
+#endif
+
+#include "row0ins.h"
+#include "row0sel.h"
+#include "row0upd.h"
+#include "row0row.h"
+#include "que0que.h"
+#include "pars0pars.h"
+#include "dict0dict.h"
+#include "dict0crea.h"
+#include "dict0load.h"
+#include "dict0boot.h"
+#include "trx0roll.h"
+#include "trx0purge.h"
+#include "trx0rec.h"
+#include "trx0undo.h"
+#include "lock0lock.h"
+#include "rem0cmp.h"
+#include "log0log.h"
+#include "btr0sea.h"
+#include "fil0fil.h"
+#include "ibuf0ibuf.h"
+
+/* Provide optional 4.x backwards compatibility for 5.0 and above */
+UNIV_INTERN ibool	row_rollback_on_timeout	= FALSE;
+
+/* List of tables we should drop in background. ALTER TABLE in MySQL requires
+that the table handler can drop the table in background when there are no
+queries to it any more. Protected by the kernel mutex. */
+typedef struct row_mysql_drop_struct	row_mysql_drop_t;
+struct row_mysql_drop_struct{
+	char*				table_name;
+	UT_LIST_NODE_T(row_mysql_drop_t) row_mysql_drop_list;
+};
+
+static UT_LIST_BASE_NODE_T(row_mysql_drop_t)	row_mysql_drop_list;
+static ibool	row_mysql_drop_list_inited	= FALSE;
+
+/* Magic table names for invoking various monitor threads */
+static const char S_innodb_monitor[] = "innodb_monitor";
+static const char S_innodb_lock_monitor[] = "innodb_lock_monitor";
+static const char S_innodb_tablespace_monitor[] = "innodb_tablespace_monitor";
+static const char S_innodb_table_monitor[] = "innodb_table_monitor";
+static const char S_innodb_mem_validate[] = "innodb_mem_validate";
+
+/* Evaluates to true if str1 equals str2_onstack, used for comparing
+the above strings. */
+#define STR_EQ(str1, str1_len, str2_onstack) \
+	((str1_len) == sizeof(str2_onstack) \
+	 && memcmp(str1, str2_onstack, sizeof(str2_onstack)) == 0)
+
+#ifndef UNIV_HOTBACKUP
+/***********************************************************************
+Determine if the given name is a name reserved for MySQL system tables. */
+static
+ibool
+row_mysql_is_system_table(
+/*======================*/
+				 /* out: TRUE if name is a MySQL
+				 system table name */
+	const char*	name)
+{
+	if (strncmp(name, "mysql/", 6) != 0) {
+
+		return(FALSE);
+	}
+
+	return(0 == strcmp(name + 6, "host")
+	       || 0 == strcmp(name + 6, "user")
+	       || 0 == strcmp(name + 6, "db"));
+}
+#endif /* !UNIV_HOTBACKUP */
+
+/*************************************************************************
+If a table is not yet in the drop list, adds the table to the list of tables
+which the master thread drops in background. We need this on Unix because in
+ALTER TABLE MySQL may call drop table even if the table has running queries on
+it. Also, if there are running foreign key checks on the table, we drop the
+table lazily. */
+static
+ibool
+row_add_table_to_background_drop_list(
+/*==================================*/
+				/* out: TRUE if the table was not yet in the
+				drop list, and was added there */
+	const char*	name);	/* in: table name */
+
+/***********************************************************************
+Delays an INSERT, DELETE or UPDATE operation if the purge is lagging. */
+static
+void
+row_mysql_delay_if_needed(void)
+/*===========================*/
+{
+	if (srv_dml_needed_delay) {
+		os_thread_sleep(srv_dml_needed_delay);
+	}
+}
+
+/***********************************************************************
+Frees the blob heap in prebuilt when no longer needed. */
+UNIV_INTERN
+void
+row_mysql_prebuilt_free_blob_heap(
+/*==============================*/
+	row_prebuilt_t*	prebuilt)	/* in: prebuilt struct of a
+					ha_innobase:: table handle */
+{
+	mem_heap_free(prebuilt->blob_heap);
+	prebuilt->blob_heap = NULL;
+}
+
+/***********************************************************************
+Stores a >= 5.0.3 format true VARCHAR length to dest, in the MySQL row
+format. */
+UNIV_INTERN
+byte*
+row_mysql_store_true_var_len(
+/*=========================*/
+			/* out: pointer to the data, we skip the 1 or 2 bytes
+			at the start that are used to store the len */
+	byte*	dest,	/* in: where to store */
+	ulint	len,	/* in: length, must fit in two bytes */
+	ulint	lenlen)	/* in: storage length of len: either 1 or 2 bytes */
+{
+	if (lenlen == 2) {
+		ut_a(len < 256 * 256);
+
+		mach_write_to_2_little_endian(dest, len);
+
+		return(dest + 2);
+	}
+
+	ut_a(lenlen == 1);
+	ut_a(len < 256);
+
+	mach_write_to_1(dest, len);
+
+	return(dest + 1);
+}
+
+/***********************************************************************
+Reads a >= 5.0.3 format true VARCHAR length, in the MySQL row format, and
+returns a pointer to the data. */
+UNIV_INTERN
+const byte*
+row_mysql_read_true_varchar(
+/*========================*/
+				/* out: pointer to the data, we skip
+				the 1 or 2 bytes at the start that are
+				used to store the len */
+	ulint*		len,	/* out: variable-length field length */
+	const byte*	field,	/* in: field in the MySQL format */
+	ulint		lenlen)	/* in: storage length of len: either 1
+				or 2 bytes */
+{
+	if (lenlen == 2) {
+		*len = mach_read_from_2_little_endian(field);
+
+		return(field + 2);
+	}
+
+	ut_a(lenlen == 1);
+
+	*len = mach_read_from_1(field);
+
+	return(field + 1);
+}
+
+/***********************************************************************
+Stores a reference to a BLOB in the MySQL format. */
+UNIV_INTERN
+void
+row_mysql_store_blob_ref(
+/*=====================*/
+	byte*		dest,	/* in: where to store */
+	ulint		col_len,/* in: dest buffer size: determines into
+				how many bytes the BLOB length is stored,
+				the space for the length may vary from 1
+				to 4 bytes */
+	const void*	data,	/* in: BLOB data; if the value to store
+				is SQL NULL this should be NULL pointer */
+	ulint		len)	/* in: BLOB length; if the value to store
+				is SQL NULL this should be 0; remember
+				also to set the NULL bit in the MySQL record
+				header! */
+{
+	/* MySQL might assume the field is set to zero except the length and
+	the pointer fields */
+
+	memset(dest, '\0', col_len);
+
+	/* In dest there are 1 - 4 bytes reserved for the BLOB length,
+	and after that 8 bytes reserved for the pointer to the data.
+	In 32-bit architectures we only use the first 4 bytes of the pointer
+	slot. */
+
+	ut_a(col_len - 8 > 1 || len < 256);
+	ut_a(col_len - 8 > 2 || len < 256 * 256);
+	ut_a(col_len - 8 > 3 || len < 256 * 256 * 256);
+
+	mach_write_to_n_little_endian(dest, col_len - 8, len);
+
+	memcpy(dest + col_len - 8, &data, sizeof data);
+}
+
+/***********************************************************************
+Reads a reference to a BLOB in the MySQL format. */
+UNIV_INTERN
+const byte*
+row_mysql_read_blob_ref(
+/*====================*/
+					/* out: pointer to BLOB data */
+	ulint*		len,		/* out: BLOB length */
+	const byte*	ref,		/* in: BLOB reference in the
+					MySQL format */
+	ulint		col_len)	/* in: BLOB reference length
+					(not BLOB length) */
+{
+	byte*	data;
+
+	*len = mach_read_from_n_little_endian(ref, col_len - 8);
+
+	memcpy(&data, ref + col_len - 8, sizeof data);
+
+	return(data);
+}
+
+/******************************************************************
+Stores a non-SQL-NULL field given in the MySQL format in the InnoDB format.
+The counterpart of this function is row_sel_field_store_in_mysql_format() in
+row0sel.c. */
+UNIV_INTERN
+byte*
+row_mysql_store_col_in_innobase_format(
+/*===================================*/
+					/* out: up to which byte we used
+					buf in the conversion */
+	dfield_t*	dfield,		/* in/out: dfield where dtype
+					information must be already set when
+					this function is called! */
+	byte*		buf,		/* in/out: buffer for a converted
+					integer value; this must be at least
+					col_len long then! */
+	ibool		row_format_col,	/* TRUE if the mysql_data is from
+					a MySQL row, FALSE if from a MySQL
+					key value;
+					in MySQL, a true VARCHAR storage
+					format differs in a row and in a
+					key value: in a key value the length
+					is always stored in 2 bytes! */
+	const byte*	mysql_data,	/* in: MySQL column value, not
+					SQL NULL; NOTE that dfield may also
+					get a pointer to mysql_data,
+					therefore do not discard this as long
+					as dfield is used! */
+	ulint		col_len,	/* in: MySQL column length; NOTE that
+					this is the storage length of the
+					column in the MySQL format row, not
+					necessarily the length of the actual
+					payload data; if the column is a true
+					VARCHAR then this is irrelevant */
+	ulint		comp)		/* in: nonzero=compact format */
+{
+	const byte*	ptr	= mysql_data;
+	const dtype_t*	dtype;
+	ulint		type;
+	ulint		lenlen;
+
+	dtype = dfield_get_type(dfield);
+
+	type = dtype->mtype;
+
+	if (type == DATA_INT) {
+		/* Store integer data in Innobase in a big-endian format,
+		sign bit negated if the data is a signed integer. In MySQL,
+		integers are stored in a little-endian format. */
+
+		byte*	p = buf + col_len;
+
+		for (;;) {
+			p--;
+			*p = *mysql_data;
+			if (p == buf) {
+				break;
+			}
+			mysql_data++;
+		}
+
+		if (!(dtype->prtype & DATA_UNSIGNED)) {
+
+			*buf ^= 128;
+		}
+
+		ptr = buf;
+		buf += col_len;
+	} else if ((type == DATA_VARCHAR
+		    || type == DATA_VARMYSQL
+		    || type == DATA_BINARY)) {
+
+		if (dtype_get_mysql_type(dtype) == DATA_MYSQL_TRUE_VARCHAR) {
+			/* The length of the actual data is stored to 1 or 2
+			bytes at the start of the field */
+
+			if (row_format_col) {
+				if (dtype->prtype & DATA_LONG_TRUE_VARCHAR) {
+					lenlen = 2;
+				} else {
+					lenlen = 1;
+				}
+			} else {
+				/* In a MySQL key value, lenlen is always 2 */
+				lenlen = 2;
+			}
+
+			ptr = row_mysql_read_true_varchar(&col_len, mysql_data,
+							  lenlen);
+		} else {
+			/* Remove trailing spaces from old style VARCHAR
+			columns. */
+
+			/* Handle UCS2 strings differently. */
+			ulint	mbminlen	= dtype_get_mbminlen(dtype);
+
+			ptr = mysql_data;
+
+			if (mbminlen == 2) {
+				/* space=0x0020 */
+				/* Trim "half-chars", just in case. */
+				col_len &= ~1;
+
+				while (col_len >= 2 && ptr[col_len - 2] == 0x00
+				       && ptr[col_len - 1] == 0x20) {
+					col_len -= 2;
+				}
+			} else {
+				ut_a(mbminlen == 1);
+				/* space=0x20 */
+				while (col_len > 0
+				       && ptr[col_len - 1] == 0x20) {
+					col_len--;
+				}
+			}
+		}
+	} else if (comp && type == DATA_MYSQL
+		   && dtype_get_mbminlen(dtype) == 1
+		   && dtype_get_mbmaxlen(dtype) > 1) {
+		/* In some cases we strip trailing spaces from UTF-8 and other
+		multibyte charsets, from FIXED-length CHAR columns, to save
+		space. UTF-8 would otherwise normally use 3 * the string length
+		bytes to store an ASCII string! */
+
+		/* We assume that this CHAR field is encoded in a
+		variable-length character set where spaces have
+		1:1 correspondence to 0x20 bytes, such as UTF-8.
+
+		Consider a CHAR(n) field, a field of n characters.
+		It will contain between n * mbminlen and n * mbmaxlen bytes.
+		We will try to truncate it to n bytes by stripping
+		space padding.	If the field contains single-byte
+		characters only, it will be truncated to n characters.
+		Consider a CHAR(5) field containing the string ".a   "
+		where "." denotes a 3-byte character represented by
+		the bytes "$%&".  After our stripping, the string will
+		be stored as "$%&a " (5 bytes).	 The string ".abc "
+		will be stored as "$%&abc" (6 bytes).
+
+		The space padding will be restored in row0sel.c, function
+		row_sel_field_store_in_mysql_format(). */
+
+		ulint		n_chars;
+
+		ut_a(!(dtype_get_len(dtype) % dtype_get_mbmaxlen(dtype)));
+
+		n_chars = dtype_get_len(dtype) / dtype_get_mbmaxlen(dtype);
+
+		/* Strip space padding. */
+		while (col_len > n_chars && ptr[col_len - 1] == 0x20) {
+			col_len--;
+		}
+	} else if (type == DATA_BLOB && row_format_col) {
+
+		ptr = row_mysql_read_blob_ref(&col_len, mysql_data, col_len);
+	}
+
+	dfield_set_data(dfield, ptr, col_len);
+
+	return(buf);
+}
+
+/******************************************************************
+Convert a row in the MySQL format to a row in the Innobase format. Note that
+the function to convert a MySQL format key value to an InnoDB dtuple is
+row_sel_convert_mysql_key_to_innobase() in row0sel.c. */
+static
+void
+row_mysql_convert_row_to_innobase(
+/*==============================*/
+	dtuple_t*	row,		/* in/out: Innobase row where the
+					field type information is already
+					copied there! */
+	row_prebuilt_t*	prebuilt,	/* in: prebuilt struct where template
+					must be of type ROW_MYSQL_WHOLE_ROW */
+	byte*		mysql_rec)	/* in: row in the MySQL format;
+					NOTE: do not discard as long as
+					row is used, as row may contain
+					pointers to this record! */
+{
+	mysql_row_templ_t*	templ;
+	dfield_t*		dfield;
+	ulint			i;
+
+	ut_ad(prebuilt->template_type == ROW_MYSQL_WHOLE_ROW);
+	ut_ad(prebuilt->mysql_template);
+
+	for (i = 0; i < prebuilt->n_template; i++) {
+
+		templ = prebuilt->mysql_template + i;
+		dfield = dtuple_get_nth_field(row, i);
+
+		if (templ->mysql_null_bit_mask != 0) {
+			/* Column may be SQL NULL */
+
+			if (mysql_rec[templ->mysql_null_byte_offset]
+			    & (byte) (templ->mysql_null_bit_mask)) {
+
+				/* It is SQL NULL */
+
+				dfield_set_null(dfield);
+
+				goto next_column;
+			}
+		}
+
+		row_mysql_store_col_in_innobase_format(
+			dfield,
+			prebuilt->ins_upd_rec_buff + templ->mysql_col_offset,
+			TRUE, /* MySQL row format data */
+			mysql_rec + templ->mysql_col_offset,
+			templ->mysql_col_len,
+			dict_table_is_comp(prebuilt->table));
+next_column:
+		;
+	}
+}
+
+/********************************************************************
+Handles user errors and lock waits detected by the database engine. */
+UNIV_INTERN
+ibool
+row_mysql_handle_errors(
+/*====================*/
+				/* out: TRUE if it was a lock wait and
+				we should continue running the query thread */
+	ulint*		new_err,/* out: possible new error encountered in
+				lock wait, or if no new error, the value
+				of trx->error_state at the entry of this
+				function */
+	trx_t*		trx,	/* in: transaction */
+	que_thr_t*	thr,	/* in: query thread */
+	trx_savept_t*	savept)	/* in: savepoint or NULL */
+{
+#ifndef UNIV_HOTBACKUP
+	ulint	err;
+
+handle_new_error:
+	err = trx->error_state;
+
+	ut_a(err != DB_SUCCESS);
+
+	trx->error_state = DB_SUCCESS;
+
+	switch (err) {
+	case DB_LOCK_WAIT_TIMEOUT:
+		if (row_rollback_on_timeout) {
+			trx_general_rollback_for_mysql(trx, FALSE, NULL);
+			break;
+		}
+		/* fall through */
+	case DB_DUPLICATE_KEY:
+	case DB_FOREIGN_DUPLICATE_KEY:
+	case DB_TOO_BIG_RECORD:
+	case DB_ROW_IS_REFERENCED:
+	case DB_NO_REFERENCED_ROW:
+	case DB_CANNOT_ADD_CONSTRAINT:
+	case DB_TOO_MANY_CONCURRENT_TRXS:
+	case DB_OUT_OF_FILE_SPACE:
+		if (savept) {
+			/* Roll back the latest, possibly incomplete
+			insertion or update */
+
+			trx_general_rollback_for_mysql(trx, TRUE, savept);
+		}
+		/* MySQL will roll back the latest SQL statement */
+		break;
+	case DB_LOCK_WAIT:
+		srv_suspend_mysql_thread(thr);
+
+		if (trx->error_state != DB_SUCCESS) {
+			que_thr_stop_for_mysql(thr);
+
+			goto handle_new_error;
+		}
+
+		*new_err = err;
+
+		return(TRUE);
+
+	case DB_DEADLOCK:
+	case DB_LOCK_TABLE_FULL:
+		/* Roll back the whole transaction; this resolution was added
+		to version 3.23.43 */
+
+		trx_general_rollback_for_mysql(trx, FALSE, NULL);
+		break;
+
+	case DB_MUST_GET_MORE_FILE_SPACE:
+		fputs("InnoDB: The database cannot continue"
+		      " operation because of\n"
+		      "InnoDB: lack of space. You must add"
+		      " a new data file to\n"
+		      "InnoDB: my.cnf and restart the database.\n", stderr);
+
+		exit(1);
+
+	case DB_CORRUPTION:
+		fputs("InnoDB: We detected index corruption"
+		      " in an InnoDB type table.\n"
+		      "InnoDB: You have to dump + drop + reimport"
+		      " the table or, in\n"
+		      "InnoDB: a case of widespread corruption,"
+		      " dump all InnoDB\n"
+		      "InnoDB: tables and recreate the"
+		      " whole InnoDB tablespace.\n"
+		      "InnoDB: If the mysqld server crashes"
+		      " after the startup or when\n"
+		      "InnoDB: you dump the tables, look at\n"
+		      "InnoDB: http://dev.mysql.com/doc/refman/5.1/en/"
+		      "forcing-recovery.html"
+		      " for help.\n", stderr);
+		break;
+	default:
+		fprintf(stderr, "InnoDB: unknown error code %lu\n",
+			(ulong) err);
+		ut_error;
+	}
+
+	if (trx->error_state != DB_SUCCESS) {
+		*new_err = trx->error_state;
+	} else {
+		*new_err = err;
+	}
+
+	trx->error_state = DB_SUCCESS;
+
+	return(FALSE);
+#else /* UNIV_HOTBACKUP */
+	/* This function depends on MySQL code that is not included in
+	InnoDB Hot Backup builds.  Besides, this function should never
+	be called in InnoDB Hot Backup. */
+	ut_error;
+	return(FALSE);
+#endif /* UNIV_HOTBACKUP */
+}
+
+/************************************************************************
+Create a prebuilt struct for a MySQL table handle. */
+UNIV_INTERN
+row_prebuilt_t*
+row_create_prebuilt(
+/*================*/
+				/* out, own: a prebuilt struct */
+	dict_table_t*	table)	/* in: Innobase table handle */
+{
+	row_prebuilt_t*	prebuilt;
+	mem_heap_t*	heap;
+	dict_index_t*	clust_index;
+	dtuple_t*	ref;
+	ulint		ref_len;
+
+	heap = mem_heap_create(sizeof *prebuilt + 128);
+
+	prebuilt = mem_heap_zalloc(heap, sizeof *prebuilt);
+
+	prebuilt->magic_n = ROW_PREBUILT_ALLOCATED;
+	prebuilt->magic_n2 = ROW_PREBUILT_ALLOCATED;
+
+	prebuilt->table = table;
+
+	prebuilt->sql_stat_start = TRUE;
+	prebuilt->heap = heap;
+
+	prebuilt->pcur = btr_pcur_create_for_mysql();
+	prebuilt->clust_pcur = btr_pcur_create_for_mysql();
+
+	prebuilt->select_lock_type = LOCK_NONE;
+	prebuilt->stored_select_lock_type = 99999999;
+
+	prebuilt->search_tuple = dtuple_create(
+		heap, 2 * dict_table_get_n_cols(table));
+
+	clust_index = dict_table_get_first_index(table);
+
+	/* Make sure that search_tuple is long enough for clustered index */
+	ut_a(2 * dict_table_get_n_cols(table) >= clust_index->n_fields);
+
+	ref_len = dict_index_get_n_unique(clust_index);
+
+	ref = dtuple_create(heap, ref_len);
+
+	dict_index_copy_types(ref, clust_index, ref_len);
+
+	prebuilt->clust_ref = ref;
+
+	prebuilt->autoinc_error = 0;
+	prebuilt->autoinc_offset = 0;
+
+	/* Default to 1, we will set the actual value later in 
+	ha_innobase::get_auto_increment(). */
+	prebuilt->autoinc_increment = 1;
+
+	prebuilt->autoinc_last_value = 0;
+
+	return(prebuilt);
+}
+
+/************************************************************************
+Free a prebuilt struct for a MySQL table handle. */
+UNIV_INTERN
+void
+row_prebuilt_free(
+/*==============*/
+	row_prebuilt_t*	prebuilt,	/* in, own: prebuilt struct */
+	ibool		dict_locked)	/* in: TRUE=data dictionary locked */
+{
+	ulint	i;
+
+	if (UNIV_UNLIKELY
+	    (prebuilt->magic_n != ROW_PREBUILT_ALLOCATED
+	     || prebuilt->magic_n2 != ROW_PREBUILT_ALLOCATED)) {
+
+		fprintf(stderr,
+			"InnoDB: Error: trying to free a corrupt\n"
+			"InnoDB: table handle. Magic n %lu,"
+			" magic n2 %lu, table name ",
+			(ulong) prebuilt->magic_n,
+			(ulong) prebuilt->magic_n2);
+		ut_print_name(stderr, NULL, TRUE, prebuilt->table->name);
+		putc('\n', stderr);
+
+		mem_analyze_corruption(prebuilt);
+
+		ut_error;
+	}
+
+	prebuilt->magic_n = ROW_PREBUILT_FREED;
+	prebuilt->magic_n2 = ROW_PREBUILT_FREED;
+
+	btr_pcur_free_for_mysql(prebuilt->pcur);
+	btr_pcur_free_for_mysql(prebuilt->clust_pcur);
+
+	if (prebuilt->mysql_template) {
+		mem_free(prebuilt->mysql_template);
+	}
+
+	if (prebuilt->ins_graph) {
+		que_graph_free_recursive(prebuilt->ins_graph);
+	}
+
+	if (prebuilt->sel_graph) {
+		que_graph_free_recursive(prebuilt->sel_graph);
+	}
+
+	if (prebuilt->upd_graph) {
+		que_graph_free_recursive(prebuilt->upd_graph);
+	}
+
+	if (prebuilt->blob_heap) {
+		mem_heap_free(prebuilt->blob_heap);
+	}
+
+	if (prebuilt->old_vers_heap) {
+		mem_heap_free(prebuilt->old_vers_heap);
+	}
+
+	for (i = 0; i < MYSQL_FETCH_CACHE_SIZE; i++) {
+		if (prebuilt->fetch_cache[i] != NULL) {
+
+			if ((ROW_PREBUILT_FETCH_MAGIC_N != mach_read_from_4(
+				     (prebuilt->fetch_cache[i]) - 4))
+			    || (ROW_PREBUILT_FETCH_MAGIC_N != mach_read_from_4(
+					(prebuilt->fetch_cache[i])
+					+ prebuilt->mysql_row_len))) {
+				fputs("InnoDB: Error: trying to free"
+				      " a corrupt fetch buffer.\n", stderr);
+
+				mem_analyze_corruption(
+					prebuilt->fetch_cache[i]);
+
+				ut_error;
+			}
+
+			mem_free((prebuilt->fetch_cache[i]) - 4);
+		}
+	}
+
+	dict_table_decrement_handle_count(prebuilt->table, dict_locked);
+
+	mem_heap_free(prebuilt->heap);
+}
+
+/*************************************************************************
+Updates the transaction pointers in query graphs stored in the prebuilt
+struct. */
+UNIV_INTERN
+void
+row_update_prebuilt_trx(
+/*====================*/
+					/* out: prebuilt dtuple */
+	row_prebuilt_t*	prebuilt,	/* in: prebuilt struct in MySQL
+					handle */
+	trx_t*		trx)		/* in: transaction handle */
+{
+	if (trx->magic_n != TRX_MAGIC_N) {
+		fprintf(stderr,
+			"InnoDB: Error: trying to use a corrupt\n"
+			"InnoDB: trx handle. Magic n %lu\n",
+			(ulong) trx->magic_n);
+
+		mem_analyze_corruption(trx);
+
+		ut_error;
+	}
+
+	if (prebuilt->magic_n != ROW_PREBUILT_ALLOCATED) {
+		fprintf(stderr,
+			"InnoDB: Error: trying to use a corrupt\n"
+			"InnoDB: table handle. Magic n %lu, table name ",
+			(ulong) prebuilt->magic_n);
+		ut_print_name(stderr, trx, TRUE, prebuilt->table->name);
+		putc('\n', stderr);
+
+		mem_analyze_corruption(prebuilt);
+
+		ut_error;
+	}
+
+	prebuilt->trx = trx;
+
+	if (prebuilt->ins_graph) {
+		prebuilt->ins_graph->trx = trx;
+	}
+
+	if (prebuilt->upd_graph) {
+		prebuilt->upd_graph->trx = trx;
+	}
+
+	if (prebuilt->sel_graph) {
+		prebuilt->sel_graph->trx = trx;
+	}
+}
+
+/*************************************************************************
+Gets pointer to a prebuilt dtuple used in insertions. If the insert graph
+has not yet been built in the prebuilt struct, then this function first
+builds it. */
+static
+dtuple_t*
+row_get_prebuilt_insert_row(
+/*========================*/
+					/* out: prebuilt dtuple; the column
+					type information is also set in it */
+	row_prebuilt_t*	prebuilt)	/* in: prebuilt struct in MySQL
+					handle */
+{
+	ins_node_t*	node;
+	dtuple_t*	row;
+	dict_table_t*	table	= prebuilt->table;
+
+	ut_ad(prebuilt && table && prebuilt->trx);
+
+	if (prebuilt->ins_node == NULL) {
+
+		/* Not called before for this handle: create an insert node
+		and query graph to the prebuilt struct */
+
+		node = ins_node_create(INS_DIRECT, table, prebuilt->heap);
+
+		prebuilt->ins_node = node;
+
+		if (prebuilt->ins_upd_rec_buff == NULL) {
+			prebuilt->ins_upd_rec_buff = mem_heap_alloc(
+				prebuilt->heap, prebuilt->mysql_row_len);
+		}
+
+		row = dtuple_create(prebuilt->heap,
+				    dict_table_get_n_cols(table));
+
+		dict_table_copy_types(row, table);
+
+		ins_node_set_new_row(node, row);
+
+		prebuilt->ins_graph = que_node_get_parent(
+			pars_complete_graph_for_exec(node,
+						     prebuilt->trx,
+						     prebuilt->heap));
+		prebuilt->ins_graph->state = QUE_FORK_ACTIVE;
+	}
+
+	return(prebuilt->ins_node->row);
+}
+
+/*************************************************************************
+Updates the table modification counter and calculates new estimates
+for table and index statistics if necessary. */
+UNIV_INLINE
+void
+row_update_statistics_if_needed(
+/*============================*/
+	dict_table_t*	table)	/* in: table */
+{
+	ulint	counter;
+
+	counter = table->stat_modified_counter;
+
+	table->stat_modified_counter = counter + 1;
+
+	/* Calculate new statistics if 1 / 16 of table has been modified
+	since the last time a statistics batch was run, or if
+	stat_modified_counter > 2 000 000 000 (to avoid wrap-around).
+	We calculate statistics at most every 16th round, since we may have
+	a counter table which is very small and updated very often. */
+
+	if (counter > 2000000000
+	    || ((ib_int64_t)counter > 16 + table->stat_n_rows / 16)) {
+
+		dict_update_statistics(table);
+	}
+}
+
+/*************************************************************************
+Unlocks AUTO_INC type locks that were possibly reserved by a trx. */
+UNIV_INTERN
+void
+row_unlock_table_autoinc_for_mysql(
+/*===============================*/
+	trx_t*	trx)	/* in/out: transaction */
+{
+	mutex_enter(&kernel_mutex);
+
+	lock_release_autoinc_locks(trx);
+
+	mutex_exit(&kernel_mutex);
+}
+
+/*************************************************************************
+Sets an AUTO_INC type lock on the table mentioned in prebuilt. The
+AUTO_INC lock gives exclusive access to the auto-inc counter of the
+table. The lock is reserved only for the duration of an SQL statement.
+It is not compatible with another AUTO_INC or exclusive lock on the
+table. */
+UNIV_INTERN
+int
+row_lock_table_autoinc_for_mysql(
+/*=============================*/
+					/* out: error code or DB_SUCCESS */
+	row_prebuilt_t*	prebuilt)	/* in: prebuilt struct in the MySQL
+					table handle */
+{
+	trx_t*			trx	= prebuilt->trx;
+	ins_node_t*		node	= prebuilt->ins_node;
+	const dict_table_t*	table	= prebuilt->table;
+	que_thr_t*		thr;
+	ulint			err;
+	ibool			was_lock_wait;
+
+	ut_ad(trx);
+	ut_ad(trx->mysql_thread_id == os_thread_get_curr_id());
+
+	/* If we already hold an AUTOINC lock on the table then do nothing.
+        Note: We peek at the value of the current owner without acquiring
+	the kernel mutex. **/
+	if (trx == table->autoinc_trx) {
+
+		return(DB_SUCCESS);
+	}
+
+	trx->op_info = "setting auto-inc lock";
+
+	if (node == NULL) {
+		row_get_prebuilt_insert_row(prebuilt);
+		node = prebuilt->ins_node;
+	}
+
+	/* We use the insert query graph as the dummy graph needed
+	in the lock module call */
+
+	thr = que_fork_get_first_thr(prebuilt->ins_graph);
+
+	que_thr_move_to_run_state_for_mysql(thr, trx);
+
+run_again:
+	thr->run_node = node;
+	thr->prev_node = node;
+
+	/* It may be that the current session has not yet started
+	its transaction, or it has been committed: */
+
+	trx_start_if_not_started(trx);
+
+	err = lock_table(0, prebuilt->table, LOCK_AUTO_INC, thr);
+
+	trx->error_state = err;
+
+	if (err != DB_SUCCESS) {
+		que_thr_stop_for_mysql(thr);
+
+		was_lock_wait = row_mysql_handle_errors(&err, trx, thr, NULL);
+
+		if (was_lock_wait) {
+			goto run_again;
+		}
+
+		trx->op_info = "";
+
+		return((int) err);
+	}
+
+	que_thr_stop_for_mysql_no_error(thr, trx);
+
+	trx->op_info = "";
+
+	return((int) err);
+}
+
+/*************************************************************************
+Sets a table lock on the table mentioned in prebuilt. */
+UNIV_INTERN
+int
+row_lock_table_for_mysql(
+/*=====================*/
+					/* out: error code or DB_SUCCESS */
+	row_prebuilt_t*	prebuilt,	/* in: prebuilt struct in the MySQL
+					table handle */
+	dict_table_t*	table,		/* in: table to lock, or NULL
+					if prebuilt->table should be
+					locked as
+					prebuilt->select_lock_type */
+	ulint		mode)		/* in: lock mode of table
+					(ignored if table==NULL) */
+{
+	trx_t*		trx		= prebuilt->trx;
+	que_thr_t*	thr;
+	ulint		err;
+	ibool		was_lock_wait;
+
+	ut_ad(trx);
+	ut_ad(trx->mysql_thread_id == os_thread_get_curr_id());
+
+	trx->op_info = "setting table lock";
+
+	if (prebuilt->sel_graph == NULL) {
+		/* Build a dummy select query graph */
+		row_prebuild_sel_graph(prebuilt);
+	}
+
+	/* We use the select query graph as the dummy graph needed
+	in the lock module call */
+
+	thr = que_fork_get_first_thr(prebuilt->sel_graph);
+
+	que_thr_move_to_run_state_for_mysql(thr, trx);
+
+run_again:
+	thr->run_node = thr;
+	thr->prev_node = thr->common.parent;
+
+	/* It may be that the current session has not yet started
+	its transaction, or it has been committed: */
+
+	trx_start_if_not_started(trx);
+
+	if (table) {
+		err = lock_table(0, table, mode, thr);
+	} else {
+		err = lock_table(0, prebuilt->table,
+				 prebuilt->select_lock_type, thr);
+	}
+
+	trx->error_state = err;
+
+	if (err != DB_SUCCESS) {
+		que_thr_stop_for_mysql(thr);
+
+		was_lock_wait = row_mysql_handle_errors(&err, trx, thr, NULL);
+
+		if (was_lock_wait) {
+			goto run_again;
+		}
+
+		trx->op_info = "";
+
+		return((int) err);
+	}
+
+	que_thr_stop_for_mysql_no_error(thr, trx);
+
+	trx->op_info = "";
+
+	return((int) err);
+}
+
+/*************************************************************************
+Does an insert for MySQL. */
+UNIV_INTERN
+int
+row_insert_for_mysql(
+/*=================*/
+					/* out: error code or DB_SUCCESS */
+	byte*		mysql_rec,	/* in: row in the MySQL format */
+	row_prebuilt_t*	prebuilt)	/* in: prebuilt struct in MySQL
+					handle */
+{
+	trx_savept_t	savept;
+	que_thr_t*	thr;
+	ulint		err;
+	ibool		was_lock_wait;
+	trx_t*		trx		= prebuilt->trx;
+	ins_node_t*	node		= prebuilt->ins_node;
+
+	ut_ad(trx);
+	ut_ad(trx->mysql_thread_id == os_thread_get_curr_id());
+
+	if (prebuilt->table->ibd_file_missing) {
+		ut_print_timestamp(stderr);
+		fprintf(stderr, "  InnoDB: Error:\n"
+			"InnoDB: MySQL is trying to use a table handle"
+			" but the .ibd file for\n"
+			"InnoDB: table %s does not exist.\n"
+			"InnoDB: Have you deleted the .ibd file"
+			" from the database directory under\n"
+			"InnoDB: the MySQL datadir, or have you"
+			" used DISCARD TABLESPACE?\n"
+			"InnoDB: Look from\n"
+			"InnoDB: http://dev.mysql.com/doc/refman/5.1/en/"
+			"innodb-troubleshooting.html\n"
+			"InnoDB: how you can resolve the problem.\n",
+			prebuilt->table->name);
+		return(DB_ERROR);
+	}
+
+	if (UNIV_UNLIKELY(prebuilt->magic_n != ROW_PREBUILT_ALLOCATED)) {
+		fprintf(stderr,
+			"InnoDB: Error: trying to free a corrupt\n"
+			"InnoDB: table handle. Magic n %lu, table name ",
+			(ulong) prebuilt->magic_n);
+		ut_print_name(stderr, trx, TRUE, prebuilt->table->name);
+		putc('\n', stderr);
+
+		mem_analyze_corruption(prebuilt);
+
+		ut_error;
+	}
+
+	if (UNIV_UNLIKELY(srv_created_new_raw || srv_force_recovery)) {
+		fputs("InnoDB: A new raw disk partition was initialized or\n"
+		      "InnoDB: innodb_force_recovery is on: we do not allow\n"
+		      "InnoDB: database modifications by the user. Shut down\n"
+		      "InnoDB: mysqld and edit my.cnf so that"
+		      " newraw is replaced\n"
+		      "InnoDB: with raw, and innodb_force_... is removed.\n",
+		      stderr);
+
+		return(DB_ERROR);
+	}
+
+	trx->op_info = "inserting";
+
+	row_mysql_delay_if_needed();
+
+	trx_start_if_not_started(trx);
+
+	if (node == NULL) {
+		row_get_prebuilt_insert_row(prebuilt);
+		node = prebuilt->ins_node;
+	}
+
+	row_mysql_convert_row_to_innobase(node->row, prebuilt, mysql_rec);
+
+	savept = trx_savept_take(trx);
+
+	thr = que_fork_get_first_thr(prebuilt->ins_graph);
+
+	if (prebuilt->sql_stat_start) {
+		node->state = INS_NODE_SET_IX_LOCK;
+		prebuilt->sql_stat_start = FALSE;
+	} else {
+		node->state = INS_NODE_ALLOC_ROW_ID;
+	}
+
+	que_thr_move_to_run_state_for_mysql(thr, trx);
+
+run_again:
+	thr->run_node = node;
+	thr->prev_node = node;
+
+	row_ins_step(thr);
+
+	err = trx->error_state;
+
+	if (err != DB_SUCCESS) {
+		que_thr_stop_for_mysql(thr);
+
+		/* TODO: what is this? */ thr->lock_state= QUE_THR_LOCK_ROW;
+
+		was_lock_wait = row_mysql_handle_errors(&err, trx, thr,
+							&savept);
+		thr->lock_state= QUE_THR_LOCK_NOLOCK;
+
+		if (was_lock_wait) {
+			goto run_again;
+		}
+
+		trx->op_info = "";
+
+		return((int) err);
+	}
+
+	que_thr_stop_for_mysql_no_error(thr, trx);
+
+	prebuilt->table->stat_n_rows++;
+
+	srv_n_rows_inserted++;
+
+	if (prebuilt->table->stat_n_rows == 0) {
+		/* Avoid wrap-over */
+		prebuilt->table->stat_n_rows--;
+	}
+
+	row_update_statistics_if_needed(prebuilt->table);
+	trx->op_info = "";
+
+	return((int) err);
+}
+
+/*************************************************************************
+Builds a dummy query graph used in selects. */
+UNIV_INTERN
+void
+row_prebuild_sel_graph(
+/*===================*/
+	row_prebuilt_t*	prebuilt)	/* in: prebuilt struct in MySQL
+					handle */
+{
+	sel_node_t*	node;
+
+	ut_ad(prebuilt && prebuilt->trx);
+
+	if (prebuilt->sel_graph == NULL) {
+
+		node = sel_node_create(prebuilt->heap);
+
+		prebuilt->sel_graph = que_node_get_parent(
+			pars_complete_graph_for_exec(node,
+						     prebuilt->trx,
+						     prebuilt->heap));
+
+		prebuilt->sel_graph->state = QUE_FORK_ACTIVE;
+	}
+}
+
+/*************************************************************************
+Creates an query graph node of 'update' type to be used in the MySQL
+interface. */
+UNIV_INTERN
+upd_node_t*
+row_create_update_node_for_mysql(
+/*=============================*/
+				/* out, own: update node */
+	dict_table_t*	table,	/* in: table to update */
+	mem_heap_t*	heap)	/* in: mem heap from which allocated */
+{
+	upd_node_t*	node;
+
+	node = upd_node_create(heap);
+
+	node->in_mysql_interface = TRUE;
+	node->is_delete = FALSE;
+	node->searched_update = FALSE;
+	node->select = NULL;
+	node->pcur = btr_pcur_create_for_mysql();
+	node->table = table;
+
+	node->update = upd_create(dict_table_get_n_cols(table), heap);
+
+	node->update_n_fields = dict_table_get_n_cols(table);
+
+	UT_LIST_INIT(node->columns);
+	node->has_clust_rec_x_lock = TRUE;
+	node->cmpl_info = 0;
+
+	node->table_sym = NULL;
+	node->col_assign_list = NULL;
+
+	return(node);
+}
+
+/*************************************************************************
+Gets pointer to a prebuilt update vector used in updates. If the update
+graph has not yet been built in the prebuilt struct, then this function
+first builds it. */
+UNIV_INTERN
+upd_t*
+row_get_prebuilt_update_vector(
+/*===========================*/
+					/* out: prebuilt update vector */
+	row_prebuilt_t*	prebuilt)	/* in: prebuilt struct in MySQL
+					handle */
+{
+	dict_table_t*	table	= prebuilt->table;
+	upd_node_t*	node;
+
+	ut_ad(prebuilt && table && prebuilt->trx);
+
+	if (prebuilt->upd_node == NULL) {
+
+		/* Not called before for this handle: create an update node
+		and query graph to the prebuilt struct */
+
+		node = row_create_update_node_for_mysql(table, prebuilt->heap);
+
+		prebuilt->upd_node = node;
+
+		prebuilt->upd_graph = que_node_get_parent(
+			pars_complete_graph_for_exec(node,
+						     prebuilt->trx,
+						     prebuilt->heap));
+		prebuilt->upd_graph->state = QUE_FORK_ACTIVE;
+	}
+
+	return(prebuilt->upd_node->update);
+}
+
+/*************************************************************************
+Does an update or delete of a row for MySQL. */
+UNIV_INTERN
+int
+row_update_for_mysql(
+/*=================*/
+					/* out: error code or DB_SUCCESS */
+	byte*		mysql_rec,	/* in: the row to be updated, in
+					the MySQL format */
+	row_prebuilt_t*	prebuilt)	/* in: prebuilt struct in MySQL
+					handle */
+{
+	trx_savept_t	savept;
+	ulint		err;
+	que_thr_t*	thr;
+	ibool		was_lock_wait;
+	dict_index_t*	clust_index;
+	/*	ulint		ref_len; */
+	upd_node_t*	node;
+	dict_table_t*	table		= prebuilt->table;
+	trx_t*		trx		= prebuilt->trx;
+
+	ut_ad(prebuilt && trx);
+	ut_ad(trx->mysql_thread_id == os_thread_get_curr_id());
+	UT_NOT_USED(mysql_rec);
+
+	if (prebuilt->table->ibd_file_missing) {
+		ut_print_timestamp(stderr);
+		fprintf(stderr, "  InnoDB: Error:\n"
+			"InnoDB: MySQL is trying to use a table handle"
+			" but the .ibd file for\n"
+			"InnoDB: table %s does not exist.\n"
+			"InnoDB: Have you deleted the .ibd file"
+			" from the database directory under\n"
+			"InnoDB: the MySQL datadir, or have you"
+			" used DISCARD TABLESPACE?\n"
+			"InnoDB: Look from\n"
+			"InnoDB: http://dev.mysql.com/doc/refman/5.1/en/"
+			"innodb-troubleshooting.html\n"
+			"InnoDB: how you can resolve the problem.\n",
+			prebuilt->table->name);
+		return(DB_ERROR);
+	}
+
+	if (UNIV_UNLIKELY(prebuilt->magic_n != ROW_PREBUILT_ALLOCATED)) {
+		fprintf(stderr,
+			"InnoDB: Error: trying to free a corrupt\n"
+			"InnoDB: table handle. Magic n %lu, table name ",
+			(ulong) prebuilt->magic_n);
+		ut_print_name(stderr, trx, TRUE, prebuilt->table->name);
+		putc('\n', stderr);
+
+		mem_analyze_corruption(prebuilt);
+
+		ut_error;
+	}
+
+	if (UNIV_UNLIKELY(srv_created_new_raw || srv_force_recovery)) {
+		fputs("InnoDB: A new raw disk partition was initialized or\n"
+		      "InnoDB: innodb_force_recovery is on: we do not allow\n"
+		      "InnoDB: database modifications by the user. Shut down\n"
+		      "InnoDB: mysqld and edit my.cnf so that newraw"
+		      " is replaced\n"
+		      "InnoDB: with raw, and innodb_force_... is removed.\n",
+		      stderr);
+
+		return(DB_ERROR);
+	}
+
+	trx->op_info = "updating or deleting";
+
+	row_mysql_delay_if_needed();
+
+	trx_start_if_not_started(trx);
+
+	node = prebuilt->upd_node;
+
+	clust_index = dict_table_get_first_index(table);
+
+	if (prebuilt->pcur->btr_cur.index == clust_index) {
+		btr_pcur_copy_stored_position(node->pcur, prebuilt->pcur);
+	} else {
+		btr_pcur_copy_stored_position(node->pcur,
+					      prebuilt->clust_pcur);
+	}
+
+	ut_a(node->pcur->rel_pos == BTR_PCUR_ON);
+
+	/* MySQL seems to call rnd_pos before updating each row it
+	has cached: we can get the correct cursor position from
+	prebuilt->pcur; NOTE that we cannot build the row reference
+	from mysql_rec if the clustered index was automatically
+	generated for the table: MySQL does not know anything about
+	the row id used as the clustered index key */
+
+	savept = trx_savept_take(trx);
+
+	thr = que_fork_get_first_thr(prebuilt->upd_graph);
+
+	node->state = UPD_NODE_UPDATE_CLUSTERED;
+
+	ut_ad(!prebuilt->sql_stat_start);
+
+	que_thr_move_to_run_state_for_mysql(thr, trx);
+
+run_again:
+	thr->run_node = node;
+	thr->prev_node = node;
+
+	row_upd_step(thr);
+
+	err = trx->error_state;
+
+	if (err != DB_SUCCESS) {
+		que_thr_stop_for_mysql(thr);
+
+		if (err == DB_RECORD_NOT_FOUND) {
+			trx->error_state = DB_SUCCESS;
+			trx->op_info = "";
+
+			return((int) err);
+		}
+
+		thr->lock_state= QUE_THR_LOCK_ROW;
+		was_lock_wait = row_mysql_handle_errors(&err, trx, thr,
+							&savept);
+		thr->lock_state= QUE_THR_LOCK_NOLOCK;
+
+		if (was_lock_wait) {
+			goto run_again;
+		}
+
+		trx->op_info = "";
+
+		return((int) err);
+	}
+
+	que_thr_stop_for_mysql_no_error(thr, trx);
+
+	if (node->is_delete) {
+		if (prebuilt->table->stat_n_rows > 0) {
+			prebuilt->table->stat_n_rows--;
+		}
+
+		srv_n_rows_deleted++;
+	} else {
+		srv_n_rows_updated++;
+	}
+
+	row_update_statistics_if_needed(prebuilt->table);
+
+	trx->op_info = "";
+
+	return((int) err);
+}
+
+/*************************************************************************
+This can only be used when srv_locks_unsafe_for_binlog is TRUE or
+this session is using a READ COMMITTED isolation level. Before
+calling this function we must use trx_reset_new_rec_lock_info() and
+trx_register_new_rec_lock() to store the information which new record locks
+really were set. This function removes a newly set lock under prebuilt->pcur,
+and also under prebuilt->clust_pcur. Currently, this is only used and tested
+in the case of an UPDATE or a DELETE statement, where the row lock is of the
+LOCK_X type.
+Thus, this implements a 'mini-rollback' that releases the latest record
+locks we set. */
+UNIV_INTERN
+int
+row_unlock_for_mysql(
+/*=================*/
+					/* out: error code or DB_SUCCESS */
+	row_prebuilt_t*	prebuilt,	/* in: prebuilt struct in MySQL
+					handle */
+	ibool		has_latches_on_recs)/* TRUE if called so that we have
+					the latches on the records under pcur
+					and clust_pcur, and we do not need to
+					reposition the cursors. */
+{
+	dict_index_t*	index;
+	btr_pcur_t*	pcur		= prebuilt->pcur;
+	btr_pcur_t*	clust_pcur	= prebuilt->clust_pcur;
+	trx_t*		trx		= prebuilt->trx;
+	rec_t*		rec;
+	mtr_t		mtr;
+
+	ut_ad(prebuilt && trx);
+	ut_ad(trx->mysql_thread_id == os_thread_get_curr_id());
+
+	if (UNIV_UNLIKELY
+	    (!srv_locks_unsafe_for_binlog
+	     && trx->isolation_level != TRX_ISO_READ_COMMITTED)) {
+
+		fprintf(stderr,
+			"InnoDB: Error: calling row_unlock_for_mysql though\n"
+			"InnoDB: innodb_locks_unsafe_for_binlog is FALSE and\n"
+			"InnoDB: this session is not using"
+			" READ COMMITTED isolation level.\n");
+
+		return(DB_SUCCESS);
+	}
+
+	trx->op_info = "unlock_row";
+
+	index = btr_pcur_get_btr_cur(pcur)->index;
+
+	if (index != NULL && trx_new_rec_locks_contain(trx, index)) {
+
+		mtr_start(&mtr);
+
+		/* Restore the cursor position and find the record */
+
+		if (!has_latches_on_recs) {
+			btr_pcur_restore_position(BTR_SEARCH_LEAF, pcur, &mtr);
+		}
+
+		rec = btr_pcur_get_rec(pcur);
+
+		lock_rec_unlock(trx, btr_pcur_get_block(pcur),
+				rec, prebuilt->select_lock_type);
+
+		mtr_commit(&mtr);
+
+		/* If the search was done through the clustered index, then
+		we have not used clust_pcur at all, and we must NOT try to
+		reset locks on clust_pcur. The values in clust_pcur may be
+		garbage! */
+
+		if (dict_index_is_clust(index)) {
+
+			goto func_exit;
+		}
+	}
+
+	index = btr_pcur_get_btr_cur(clust_pcur)->index;
+
+	if (index != NULL && trx_new_rec_locks_contain(trx, index)) {
+
+		mtr_start(&mtr);
+
+		/* Restore the cursor position and find the record */
+
+		if (!has_latches_on_recs) {
+			btr_pcur_restore_position(BTR_SEARCH_LEAF, clust_pcur,
+						  &mtr);
+		}
+
+		rec = btr_pcur_get_rec(clust_pcur);
+
+		lock_rec_unlock(trx, btr_pcur_get_block(clust_pcur),
+				rec, prebuilt->select_lock_type);
+
+		mtr_commit(&mtr);
+	}
+
+func_exit:
+	trx->op_info = "";
+
+	return(DB_SUCCESS);
+}
+
+/**************************************************************************
+Does a cascaded delete or set null in a foreign key operation. */
+UNIV_INTERN
+ulint
+row_update_cascade_for_mysql(
+/*=========================*/
+				/* out: error code or DB_SUCCESS */
+	que_thr_t*	thr,	/* in: query thread */
+	upd_node_t*	node,	/* in: update node used in the cascade
+				or set null operation */
+	dict_table_t*	table)	/* in: table where we do the operation */
+{
+	ulint	err;
+	trx_t*	trx;
+
+	trx = thr_get_trx(thr);
+run_again:
+	thr->run_node = node;
+	thr->prev_node = node;
+
+	row_upd_step(thr);
+
+	err = trx->error_state;
+
+	/* Note that the cascade node is a subnode of another InnoDB
+	query graph node. We do a normal lock wait in this node, but
+	all errors are handled by the parent node. */
+
+	if (err == DB_LOCK_WAIT) {
+		/* Handle lock wait here */
+
+		que_thr_stop_for_mysql(thr);
+
+		srv_suspend_mysql_thread(thr);
+
+		/* Note that a lock wait may also end in a lock wait timeout,
+		or this transaction is picked as a victim in selective
+		deadlock resolution */
+
+		if (trx->error_state != DB_SUCCESS) {
+
+			return(trx->error_state);
+		}
+
+		/* Retry operation after a normal lock wait */
+
+		goto run_again;
+	}
+
+	if (err != DB_SUCCESS) {
+
+		return(err);
+	}
+
+	if (node->is_delete) {
+		if (table->stat_n_rows > 0) {
+			table->stat_n_rows--;
+		}
+
+		srv_n_rows_deleted++;
+	} else {
+		srv_n_rows_updated++;
+	}
+
+	row_update_statistics_if_needed(table);
+
+	return(err);
+}
+
+/*************************************************************************
+Checks if a table is such that we automatically created a clustered
+index on it (on row id). */
+UNIV_INTERN
+ibool
+row_table_got_default_clust_index(
+/*==============================*/
+	const dict_table_t*	table)
+{
+	const dict_index_t*	clust_index;
+
+	clust_index = dict_table_get_first_index(table);
+
+	return(dict_index_get_nth_col(clust_index, 0)->mtype == DATA_SYS);
+}
+
+/*************************************************************************
+Calculates the key number used inside MySQL for an Innobase index. We have
+to take into account if we generated a default clustered index for the table */
+UNIV_INTERN
+ulint
+row_get_mysql_key_number_for_index(
+/*===============================*/
+	const dict_index_t*	index)
+{
+	const dict_index_t*	ind;
+	ulint			i;
+
+	ut_a(index);
+
+	i = 0;
+	ind = dict_table_get_first_index(index->table);
+
+	while (index != ind) {
+		ind = dict_table_get_next_index(ind);
+		i++;
+	}
+
+	if (row_table_got_default_clust_index(index->table)) {
+		ut_a(i > 0);
+		i--;
+	}
+
+	return(i);
+}
+
+/*************************************************************************
+Locks the data dictionary in shared mode from modifications, for performing
+foreign key check, rollback, or other operation invisible to MySQL. */
+UNIV_INTERN
+void
+row_mysql_freeze_data_dictionary_func(
+/*==================================*/
+	trx_t*		trx,	/* in/out: transaction */
+	const char*	file,	/* in: file name */
+	ulint		line)	/* in: line number */
+{
+	ut_a(trx->dict_operation_lock_mode == 0);
+
+	rw_lock_s_lock_func(&dict_operation_lock, 0, file, line);
+
+	trx->dict_operation_lock_mode = RW_S_LATCH;
+}
+
+/*************************************************************************
+Unlocks the data dictionary shared lock. */
+UNIV_INTERN
+void
+row_mysql_unfreeze_data_dictionary(
+/*===============================*/
+	trx_t*	trx)	/* in/out: transaction */
+{
+	ut_a(trx->dict_operation_lock_mode == RW_S_LATCH);
+
+	rw_lock_s_unlock(&dict_operation_lock);
+
+	trx->dict_operation_lock_mode = 0;
+}
+
+/*************************************************************************
+Locks the data dictionary exclusively for performing a table create or other
+data dictionary modification operation. */
+UNIV_INTERN
+void
+row_mysql_lock_data_dictionary_func(
+/*================================*/
+	trx_t*		trx,	/* in/out: transaction */
+	const char*	file,	/* in: file name */
+	ulint		line)	/* in: line number */
+{
+	ut_a(trx->dict_operation_lock_mode == 0
+	     || trx->dict_operation_lock_mode == RW_X_LATCH);
+
+	/* Serialize data dictionary operations with dictionary mutex:
+	no deadlocks or lock waits can occur then in these operations */
+
+	rw_lock_x_lock_func(&dict_operation_lock, 0, file, line);
+	trx->dict_operation_lock_mode = RW_X_LATCH;
+
+	mutex_enter(&(dict_sys->mutex));
+}
+
+/*************************************************************************
+Unlocks the data dictionary exclusive lock. */
+UNIV_INTERN
+void
+row_mysql_unlock_data_dictionary(
+/*=============================*/
+	trx_t*	trx)	/* in/out: transaction */
+{
+	ut_a(trx->dict_operation_lock_mode == RW_X_LATCH);
+
+	/* Serialize data dictionary operations with dictionary mutex:
+	no deadlocks can occur then in these operations */
+
+	mutex_exit(&(dict_sys->mutex));
+	rw_lock_x_unlock(&dict_operation_lock);
+
+	trx->dict_operation_lock_mode = 0;
+}
+
+#ifndef UNIV_HOTBACKUP
+/*************************************************************************
+Creates a table for MySQL. If the name of the table ends in
+one of "innodb_monitor", "innodb_lock_monitor", "innodb_tablespace_monitor",
+"innodb_table_monitor", then this will also start the printing of monitor
+output by the master thread. If the table name ends in "innodb_mem_validate",
+InnoDB will try to invoke mem_validate(). */
+UNIV_INTERN
+int
+row_create_table_for_mysql(
+/*=======================*/
+				/* out: error code or DB_SUCCESS */
+	dict_table_t*	table,	/* in, own: table definition
+				(will be freed) */
+	trx_t*		trx)	/* in: transaction handle */
+{
+	tab_node_t*	node;
+	mem_heap_t*	heap;
+	que_thr_t*	thr;
+	const char*	table_name;
+	ulint		table_name_len;
+	ulint		err;
+	ulint		i;
+
+	ut_ad(trx->mysql_thread_id == os_thread_get_curr_id());
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(rw_lock_own(&dict_operation_lock, RW_LOCK_EX));
+#endif /* UNIV_SYNC_DEBUG */
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+	ut_ad(trx->dict_operation_lock_mode == RW_X_LATCH);
+
+	if (srv_created_new_raw) {
+		fputs("InnoDB: A new raw disk partition was initialized:\n"
+		      "InnoDB: we do not allow database modifications"
+		      " by the user.\n"
+		      "InnoDB: Shut down mysqld and edit my.cnf so that newraw"
+		      " is replaced with raw.\n", stderr);
+err_exit:
+		dict_mem_table_free(table);
+		trx_commit_for_mysql(trx);
+
+		return(DB_ERROR);
+	}
+
+	trx->op_info = "creating table";
+
+	if (row_mysql_is_system_table(table->name)) {
+
+		fprintf(stderr,
+			"InnoDB: Error: trying to create a MySQL system"
+			" table %s of type InnoDB.\n"
+			"InnoDB: MySQL system tables must be"
+			" of the MyISAM type!\n",
+			table->name);
+		goto err_exit;
+	}
+
+	/* Check that no reserved column names are used. */
+	for (i = 0; i < dict_table_get_n_user_cols(table); i++) {
+		if (dict_col_name_is_reserved(
+			    dict_table_get_col_name(table, i))) {
+
+			goto err_exit;
+		}
+	}
+
+	trx_start_if_not_started(trx);
+
+	/* The table name is prefixed with the database name and a '/'.
+	Certain table names starting with 'innodb_' have their special
+	meaning regardless of the database name.  Thus, we need to
+	ignore the database name prefix in the comparisons. */
+	table_name = strchr(table->name, '/');
+	ut_a(table_name);
+	table_name++;
+	table_name_len = strlen(table_name) + 1;
+
+	if (STR_EQ(table_name, table_name_len, S_innodb_monitor)) {
+
+		/* Table equals "innodb_monitor":
+		start monitor prints */
+
+		srv_print_innodb_monitor = TRUE;
+
+		/* The lock timeout monitor thread also takes care
+		of InnoDB monitor prints */
+
+		os_event_set(srv_lock_timeout_thread_event);
+	} else if (STR_EQ(table_name, table_name_len,
+			  S_innodb_lock_monitor)) {
+
+		srv_print_innodb_monitor = TRUE;
+		srv_print_innodb_lock_monitor = TRUE;
+		os_event_set(srv_lock_timeout_thread_event);
+	} else if (STR_EQ(table_name, table_name_len,
+			  S_innodb_tablespace_monitor)) {
+
+		srv_print_innodb_tablespace_monitor = TRUE;
+		os_event_set(srv_lock_timeout_thread_event);
+	} else if (STR_EQ(table_name, table_name_len,
+			  S_innodb_table_monitor)) {
+
+		srv_print_innodb_table_monitor = TRUE;
+		os_event_set(srv_lock_timeout_thread_event);
+	} else if (STR_EQ(table_name, table_name_len,
+			  S_innodb_mem_validate)) {
+		/* We define here a debugging feature intended for
+		developers */
+
+		fputs("Validating InnoDB memory:\n"
+		      "to use this feature you must compile InnoDB with\n"
+		      "UNIV_MEM_DEBUG defined in univ.i and"
+		      " the server must be\n"
+		      "quiet because allocation from a mem heap"
+		      " is not protected\n"
+		      "by any semaphore.\n", stderr);
+#ifdef UNIV_MEM_DEBUG
+		ut_a(mem_validate());
+		fputs("Memory validated\n", stderr);
+#else /* UNIV_MEM_DEBUG */
+		fputs("Memory NOT validated (recompile with UNIV_MEM_DEBUG)\n",
+		      stderr);
+#endif /* UNIV_MEM_DEBUG */
+	}
+
+	heap = mem_heap_create(512);
+
+	trx_set_dict_operation(trx, TRX_DICT_OP_TABLE);
+
+	node = tab_create_graph_create(table, heap);
+
+	thr = pars_complete_graph_for_exec(node, trx, heap);
+
+	ut_a(thr == que_fork_start_command(que_node_get_parent(thr)));
+	que_run_threads(thr);
+
+	err = trx->error_state;
+
+	if (UNIV_UNLIKELY(err != DB_SUCCESS)) {
+		trx->error_state = DB_SUCCESS;
+		trx_general_rollback_for_mysql(trx, FALSE, NULL);
+	}
+
+	switch (err) {
+	case DB_OUT_OF_FILE_SPACE:
+		ut_print_timestamp(stderr);
+		fputs("  InnoDB: Warning: cannot create table ",
+		      stderr);
+		ut_print_name(stderr, trx, TRUE, table->name);
+		fputs(" because tablespace full\n", stderr);
+
+		if (dict_table_get_low(table->name)) {
+
+			row_drop_table_for_mysql(table->name, trx, FALSE);
+			trx_commit_for_mysql(trx);
+		}
+		break;
+
+	case DB_DUPLICATE_KEY:
+		ut_print_timestamp(stderr);
+		fputs("  InnoDB: Error: table ", stderr);
+		ut_print_name(stderr, trx, TRUE, table->name);
+		fputs(" already exists in InnoDB internal\n"
+		      "InnoDB: data dictionary. Have you deleted"
+		      " the .frm file\n"
+		      "InnoDB: and not used DROP TABLE?"
+		      " Have you used DROP DATABASE\n"
+		      "InnoDB: for InnoDB tables in"
+		      " MySQL version <= 3.23.43?\n"
+		      "InnoDB: See the Restrictions section"
+		      " of the InnoDB manual.\n"
+		      "InnoDB: You can drop the orphaned table"
+		      " inside InnoDB by\n"
+		      "InnoDB: creating an InnoDB table with"
+		      " the same name in another\n"
+		      "InnoDB: database and copying the .frm file"
+		      " to the current database.\n"
+		      "InnoDB: Then MySQL thinks the table exists,"
+		      " and DROP TABLE will\n"
+		      "InnoDB: succeed.\n"
+		      "InnoDB: You can look for further help from\n"
+		      "InnoDB: "
+		      "http://dev.mysql.com/doc/refman/5.1/en/"
+		      "innodb-troubleshooting.html\n", stderr);
+
+		/* We may also get err == DB_ERROR if the .ibd file for the
+		table already exists */
+
+		break;
+	}
+
+	que_graph_free((que_t*) que_node_get_parent(thr));
+
+	trx->op_info = "";
+
+	return((int) err);
+}
+
+/*************************************************************************
+Does an index creation operation for MySQL. TODO: currently failure
+to create an index results in dropping the whole table! This is no problem
+currently as all indexes must be created at the same time as the table. */
+UNIV_INTERN
+int
+row_create_index_for_mysql(
+/*=======================*/
+					/* out: error number or DB_SUCCESS */
+	dict_index_t*	index,		/* in, own: index definition
+					(will be freed) */
+	trx_t*		trx,		/* in: transaction handle */
+	const ulint*	field_lengths)	/* in: if not NULL, must contain
+					dict_index_get_n_fields(index)
+					actual field lengths for the
+					index columns, which are
+					then checked for not being too
+					large. */
+{
+	ind_node_t*	node;
+	mem_heap_t*	heap;
+	que_thr_t*	thr;
+	ulint		err;
+	ulint		i;
+	ulint		len;
+	char*		table_name;
+
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(rw_lock_own(&dict_operation_lock, RW_LOCK_EX));
+#endif /* UNIV_SYNC_DEBUG */
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+	ut_ad(trx->mysql_thread_id == os_thread_get_curr_id());
+
+	trx->op_info = "creating index";
+
+	/* Copy the table name because we may want to drop the
+	table later, after the index object is freed (inside
+	que_run_threads()) and thus index->table_name is not available. */
+	table_name = mem_strdup(index->table_name);
+
+	trx_start_if_not_started(trx);
+
+	/* Check that the same column does not appear twice in the index.
+	Starting from 4.0.14, InnoDB should be able to cope with that, but
+	safer not to allow them. */
+
+	for (i = 0; i < dict_index_get_n_fields(index); i++) {
+		ulint		j;
+
+		for (j = 0; j < i; j++) {
+			if (0 == ut_strcmp(
+				    dict_index_get_nth_field(index, j)->name,
+				    dict_index_get_nth_field(index, i)->name)) {
+				ut_print_timestamp(stderr);
+
+				fputs("  InnoDB: Error: column ", stderr);
+				ut_print_name(stderr, trx, FALSE,
+					      dict_index_get_nth_field(
+						      index, i)->name);
+				fputs(" appears twice in ", stderr);
+				dict_index_name_print(stderr, trx, index);
+				fputs("\n"
+				      "InnoDB: This is not allowed"
+				      " in InnoDB.\n", stderr);
+
+				err = DB_COL_APPEARS_TWICE_IN_INDEX;
+
+				goto error_handling;
+			}
+		}
+
+		/* Check also that prefix_len and actual length
+		< DICT_MAX_INDEX_COL_LEN */
+
+		len = dict_index_get_nth_field(index, i)->prefix_len;
+
+		if (field_lengths) {
+			len = ut_max(len, field_lengths[i]);
+		}
+
+		if (len >= DICT_MAX_INDEX_COL_LEN) {
+			err = DB_TOO_BIG_RECORD;
+
+			goto error_handling;
+		}
+	}
+
+	heap = mem_heap_create(512);
+
+	trx_set_dict_operation(trx, TRX_DICT_OP_TABLE);
+
+	/* Note that the space id where we store the index is inherited from
+	the table in dict_build_index_def_step() in dict0crea.c. */
+
+	node = ind_create_graph_create(index, heap);
+
+	thr = pars_complete_graph_for_exec(node, trx, heap);
+
+	ut_a(thr == que_fork_start_command(que_node_get_parent(thr)));
+	que_run_threads(thr);
+
+	err = trx->error_state;
+
+	que_graph_free((que_t*) que_node_get_parent(thr));
+
+error_handling:
+	if (err != DB_SUCCESS) {
+		/* We have special error handling here */
+
+		trx->error_state = DB_SUCCESS;
+
+		trx_general_rollback_for_mysql(trx, FALSE, NULL);
+
+		row_drop_table_for_mysql(table_name, trx, FALSE);
+
+		trx_commit_for_mysql(trx);
+
+		trx->error_state = DB_SUCCESS;
+	}
+
+	trx->op_info = "";
+
+	mem_free(table_name);
+
+	return((int) err);
+}
+
+/*************************************************************************
+Scans a table create SQL string and adds to the data dictionary
+the foreign key constraints declared in the string. This function
+should be called after the indexes for a table have been created.
+Each foreign key constraint must be accompanied with indexes in
+bot participating tables. The indexes are allowed to contain more
+fields than mentioned in the constraint. Check also that foreign key
+constraints which reference this table are ok. */
+UNIV_INTERN
+int
+row_table_add_foreign_constraints(
+/*==============================*/
+					/* out: error code or DB_SUCCESS */
+	trx_t*		trx,		/* in: transaction */
+	const char*	sql_string,	/* in: table create statement where
+					foreign keys are declared like:
+				FOREIGN KEY (a, b) REFERENCES table2(c, d),
+					table2 can be written also with the
+					database name before it: test.table2 */
+	const char*	name,		/* in: table full name in the
+					normalized form
+					database_name/table_name */
+	ibool		reject_fks)	/* in: if TRUE, fail with error
+					code DB_CANNOT_ADD_CONSTRAINT if
+					any foreign keys are found. */
+{
+	ulint	err;
+
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(rw_lock_own(&dict_operation_lock, RW_LOCK_EX));
+#endif /* UNIV_SYNC_DEBUG */
+	ut_a(sql_string);
+
+	trx->op_info = "adding foreign keys";
+
+	trx_start_if_not_started(trx);
+
+	trx_set_dict_operation(trx, TRX_DICT_OP_TABLE);
+
+	err = dict_create_foreign_constraints(trx, sql_string, name,
+					      reject_fks);
+#ifndef UNIV_HOTBACKUP
+	if (err == DB_SUCCESS) {
+		/* Check that also referencing constraints are ok */
+		err = dict_load_foreigns(name, TRUE);
+	}
+#endif /* !UNIV_HOTBACKUP */
+	if (err != DB_SUCCESS) {
+		/* We have special error handling here */
+
+		trx->error_state = DB_SUCCESS;
+
+		trx_general_rollback_for_mysql(trx, FALSE, NULL);
+
+		row_drop_table_for_mysql(name, trx, FALSE);
+
+		trx_commit_for_mysql(trx);
+
+		trx->error_state = DB_SUCCESS;
+	}
+
+	return((int) err);
+}
+
+/*************************************************************************
+Drops a table for MySQL as a background operation. MySQL relies on Unix
+in ALTER TABLE to the fact that the table handler does not remove the
+table before all handles to it has been removed. Furhermore, the MySQL's
+call to drop table must be non-blocking. Therefore we do the drop table
+as a background operation, which is taken care of by the master thread
+in srv0srv.c. */
+static
+int
+row_drop_table_for_mysql_in_background(
+/*===================================*/
+				/* out: error code or DB_SUCCESS */
+	const char*	name)	/* in: table name */
+{
+	ulint	error;
+	trx_t*	trx;
+
+	trx = trx_allocate_for_background();
+
+	/* If the original transaction was dropping a table referenced by
+	foreign keys, we must set the following to be able to drop the
+	table: */
+
+	trx->check_foreigns = FALSE;
+
+	/*	fputs("InnoDB: Error: Dropping table ", stderr);
+	ut_print_name(stderr, trx, TRUE, name);
+	fputs(" in background drop list\n", stderr); */
+
+	/* Try to drop the table in InnoDB */
+
+	error = row_drop_table_for_mysql(name, trx, FALSE);
+
+	/* Flush the log to reduce probability that the .frm files and
+	the InnoDB data dictionary get out-of-sync if the user runs
+	with innodb_flush_log_at_trx_commit = 0 */
+
+	log_buffer_flush_to_disk();
+
+	trx_commit_for_mysql(trx);
+
+	trx_free_for_background(trx);
+
+	return((int) error);
+}
+
+/*************************************************************************
+The master thread in srv0srv.c calls this regularly to drop tables which
+we must drop in background after queries to them have ended. Such lazy
+dropping of tables is needed in ALTER TABLE on Unix. */
+UNIV_INTERN
+ulint
+row_drop_tables_for_mysql_in_background(void)
+/*=========================================*/
+					/* out: how many tables dropped
+					+ remaining tables in list */
+{
+	row_mysql_drop_t*	drop;
+	dict_table_t*		table;
+	ulint			n_tables;
+	ulint			n_tables_dropped = 0;
+loop:
+	mutex_enter(&kernel_mutex);
+
+	if (!row_mysql_drop_list_inited) {
+
+		UT_LIST_INIT(row_mysql_drop_list);
+		row_mysql_drop_list_inited = TRUE;
+	}
+
+	drop = UT_LIST_GET_FIRST(row_mysql_drop_list);
+
+	n_tables = UT_LIST_GET_LEN(row_mysql_drop_list);
+
+	mutex_exit(&kernel_mutex);
+
+	if (drop == NULL) {
+		/* All tables dropped */
+
+		return(n_tables + n_tables_dropped);
+	}
+
+	mutex_enter(&(dict_sys->mutex));
+	table = dict_table_get_low(drop->table_name);
+	mutex_exit(&(dict_sys->mutex));
+
+	if (table == NULL) {
+		/* If for some reason the table has already been dropped
+		through some other mechanism, do not try to drop it */
+
+		goto already_dropped;
+	}
+
+	if (DB_SUCCESS != row_drop_table_for_mysql_in_background(
+		    drop->table_name)) {
+		/* If the DROP fails for some table, we return, and let the
+		main thread retry later */
+
+		return(n_tables + n_tables_dropped);
+	}
+
+	n_tables_dropped++;
+
+already_dropped:
+	mutex_enter(&kernel_mutex);
+
+	UT_LIST_REMOVE(row_mysql_drop_list, row_mysql_drop_list, drop);
+
+	ut_print_timestamp(stderr);
+	fputs("  InnoDB: Dropped table ", stderr);
+	ut_print_name(stderr, NULL, TRUE, drop->table_name);
+	fputs(" in background drop queue.\n", stderr);
+
+	mem_free(drop->table_name);
+
+	mem_free(drop);
+
+	mutex_exit(&kernel_mutex);
+
+	goto loop;
+}
+
+/*************************************************************************
+Get the background drop list length. NOTE: the caller must own the kernel
+mutex! */
+UNIV_INTERN
+ulint
+row_get_background_drop_list_len_low(void)
+/*======================================*/
+					/* out: how many tables in list */
+{
+	ut_ad(mutex_own(&kernel_mutex));
+
+	if (!row_mysql_drop_list_inited) {
+
+		UT_LIST_INIT(row_mysql_drop_list);
+		row_mysql_drop_list_inited = TRUE;
+	}
+
+	return(UT_LIST_GET_LEN(row_mysql_drop_list));
+}
+
+/*************************************************************************
+If a table is not yet in the drop list, adds the table to the list of tables
+which the master thread drops in background. We need this on Unix because in
+ALTER TABLE MySQL may call drop table even if the table has running queries on
+it. Also, if there are running foreign key checks on the table, we drop the
+table lazily. */
+static
+ibool
+row_add_table_to_background_drop_list(
+/*==================================*/
+				/* out: TRUE if the table was not yet in the
+				drop list, and was added there */
+	const char*	name)	/* in: table name */
+{
+	row_mysql_drop_t*	drop;
+
+	mutex_enter(&kernel_mutex);
+
+	if (!row_mysql_drop_list_inited) {
+
+		UT_LIST_INIT(row_mysql_drop_list);
+		row_mysql_drop_list_inited = TRUE;
+	}
+
+	/* Look if the table already is in the drop list */
+	drop = UT_LIST_GET_FIRST(row_mysql_drop_list);
+
+	while (drop != NULL) {
+		if (strcmp(drop->table_name, name) == 0) {
+			/* Already in the list */
+
+			mutex_exit(&kernel_mutex);
+
+			return(FALSE);
+		}
+
+		drop = UT_LIST_GET_NEXT(row_mysql_drop_list, drop);
+	}
+
+	drop = mem_alloc(sizeof(row_mysql_drop_t));
+
+	drop->table_name = mem_strdup(name);
+
+	UT_LIST_ADD_LAST(row_mysql_drop_list, row_mysql_drop_list, drop);
+
+	/*	fputs("InnoDB: Adding table ", stderr);
+	ut_print_name(stderr, trx, TRUE, drop->table_name);
+	fputs(" to background drop list\n", stderr); */
+
+	mutex_exit(&kernel_mutex);
+
+	return(TRUE);
+}
+
+/*************************************************************************
+Discards the tablespace of a table which stored in an .ibd file. Discarding
+means that this function deletes the .ibd file and assigns a new table id for
+the table. Also the flag table->ibd_file_missing is set TRUE. */
+UNIV_INTERN
+int
+row_discard_tablespace_for_mysql(
+/*=============================*/
+				/* out: error code or DB_SUCCESS */
+	const char*	name,	/* in: table name */
+	trx_t*		trx)	/* in: transaction handle */
+{
+	dict_foreign_t*	foreign;
+	dulint		new_id;
+	dict_table_t*	table;
+	ibool		success;
+	ulint		err;
+	pars_info_t*	info = NULL;
+
+	/* How do we prevent crashes caused by ongoing operations on
+	the table? Old operations could try to access non-existent
+	pages.
+
+	1) SQL queries, INSERT, SELECT, ...: we must get an exclusive
+	MySQL table lock on the table before we can do DISCARD
+	TABLESPACE. Then there are no running queries on the table.
+
+	2) Purge and rollback: we assign a new table id for the
+	table. Since purge and rollback look for the table based on
+	the table id, they see the table as 'dropped' and discard
+	their operations.
+
+	3) Insert buffer: we remove all entries for the tablespace in
+	the insert buffer tree; as long as the tablespace mem object
+	does not exist, ongoing insert buffer page merges are
+	discarded in buf0rea.c. If we recreate the tablespace mem
+	object with IMPORT TABLESPACE later, then the tablespace will
+	have the same id, but the tablespace_version field in the mem
+	object is different, and ongoing old insert buffer page merges
+	get discarded.
+
+	4) Linear readahead and random readahead: we use the same
+	method as in 3) to discard ongoing operations.
+
+	5) FOREIGN KEY operations: if
+	table->n_foreign_key_checks_running > 0, we do not allow the
+	discard. We also reserve the data dictionary latch. */
+
+	ut_ad(trx->mysql_thread_id == os_thread_get_curr_id());
+
+	trx->op_info = "discarding tablespace";
+	trx_start_if_not_started(trx);
+
+	/* Serialize data dictionary operations with dictionary mutex:
+	no deadlocks can occur then in these operations */
+
+	row_mysql_lock_data_dictionary(trx);
+
+	table = dict_table_get_low(name);
+
+	if (!table) {
+		err = DB_TABLE_NOT_FOUND;
+
+		goto funct_exit;
+	}
+
+	if (table->space == 0) {
+		ut_print_timestamp(stderr);
+		fputs("  InnoDB: Error: table ", stderr);
+		ut_print_name(stderr, trx, TRUE, name);
+		fputs("\n"
+		      "InnoDB: is in the system tablespace 0"
+		      " which cannot be discarded\n", stderr);
+		err = DB_ERROR;
+
+		goto funct_exit;
+	}
+
+	if (table->n_foreign_key_checks_running > 0) {
+
+		ut_print_timestamp(stderr);
+		fputs("  InnoDB: You are trying to DISCARD table ", stderr);
+		ut_print_name(stderr, trx, TRUE, table->name);
+		fputs("\n"
+		      "InnoDB: though there is a foreign key check"
+		      " running on it.\n"
+		      "InnoDB: Cannot discard the table.\n",
+		      stderr);
+
+		err = DB_ERROR;
+
+		goto funct_exit;
+	}
+
+	/* Check if the table is referenced by foreign key constraints from
+	some other table (not the table itself) */
+
+	foreign = UT_LIST_GET_FIRST(table->referenced_list);
+
+	while (foreign && foreign->foreign_table == table) {
+		foreign = UT_LIST_GET_NEXT(referenced_list, foreign);
+	}
+
+	if (foreign && trx->check_foreigns) {
+
+		FILE*	ef	= dict_foreign_err_file;
+
+		/* We only allow discarding a referenced table if
+		FOREIGN_KEY_CHECKS is set to 0 */
+
+		err = DB_CANNOT_DROP_CONSTRAINT;
+
+		mutex_enter(&dict_foreign_err_mutex);
+		rewind(ef);
+		ut_print_timestamp(ef);
+
+		fputs("  Cannot DISCARD table ", ef);
+		ut_print_name(stderr, trx, TRUE, name);
+		fputs("\n"
+		      "because it is referenced by ", ef);
+		ut_print_name(stderr, trx, TRUE, foreign->foreign_table_name);
+		putc('\n', ef);
+		mutex_exit(&dict_foreign_err_mutex);
+
+		goto funct_exit;
+	}
+
+	new_id = dict_hdr_get_new_id(DICT_HDR_TABLE_ID);
+
+	/* Remove all locks except the table-level S and X locks. */
+	lock_remove_all_on_table(table, FALSE);
+
+	info = pars_info_create();
+
+	pars_info_add_str_literal(info, "table_name", name);
+	pars_info_add_dulint_literal(info, "new_id", new_id);
+
+	err = que_eval_sql(info,
+			   "PROCEDURE DISCARD_TABLESPACE_PROC () IS\n"
+			   "old_id CHAR;\n"
+			   "BEGIN\n"
+			   "SELECT ID INTO old_id\n"
+			   "FROM SYS_TABLES\n"
+			   "WHERE NAME = :table_name\n"
+			   "LOCK IN SHARE MODE;\n"
+			   "IF (SQL % NOTFOUND) THEN\n"
+			   "       COMMIT WORK;\n"
+			   "       RETURN;\n"
+			   "END IF;\n"
+			   "UPDATE SYS_TABLES SET ID = :new_id\n"
+			   " WHERE ID = old_id;\n"
+			   "UPDATE SYS_COLUMNS SET TABLE_ID = :new_id\n"
+			   " WHERE TABLE_ID = old_id;\n"
+			   "UPDATE SYS_INDEXES SET TABLE_ID = :new_id\n"
+			   " WHERE TABLE_ID = old_id;\n"
+			   "COMMIT WORK;\n"
+			   "END;\n"
+			   , FALSE, trx);
+
+	if (err != DB_SUCCESS) {
+		trx->error_state = DB_SUCCESS;
+		trx_general_rollback_for_mysql(trx, FALSE, NULL);
+		trx->error_state = DB_SUCCESS;
+	} else {
+		dict_table_change_id_in_cache(table, new_id);
+
+		success = fil_discard_tablespace(table->space);
+
+		if (!success) {
+			trx->error_state = DB_SUCCESS;
+			trx_general_rollback_for_mysql(trx, FALSE, NULL);
+			trx->error_state = DB_SUCCESS;
+
+			err = DB_ERROR;
+		} else {
+			/* Set the flag which tells that now it is legal to
+			IMPORT a tablespace for this table */
+			table->tablespace_discarded = TRUE;
+			table->ibd_file_missing = TRUE;
+		}
+	}
+
+funct_exit:
+	trx_commit_for_mysql(trx);
+
+	row_mysql_unlock_data_dictionary(trx);
+
+	trx->op_info = "";
+
+	return((int) err);
+}
+
+/*********************************************************************
+Imports a tablespace. The space id in the .ibd file must match the space id
+of the table in the data dictionary. */
+UNIV_INTERN
+int
+row_import_tablespace_for_mysql(
+/*============================*/
+				/* out: error code or DB_SUCCESS */
+	const char*	name,	/* in: table name */
+	trx_t*		trx)	/* in: transaction handle */
+{
+	dict_table_t*	table;
+	ibool		success;
+	ib_uint64_t	current_lsn;
+	ulint		err		= DB_SUCCESS;
+
+	ut_ad(trx->mysql_thread_id == os_thread_get_curr_id());
+
+	trx_start_if_not_started(trx);
+
+	trx->op_info = "importing tablespace";
+
+	current_lsn = log_get_lsn();
+
+	/* It is possible, though very improbable, that the lsn's in the
+	tablespace to be imported have risen above the current system lsn, if
+	a lengthy purge, ibuf merge, or rollback was performed on a backup
+	taken with ibbackup. If that is the case, reset page lsn's in the
+	file. We assume that mysqld was shut down after it performed these
+	cleanup operations on the .ibd file, so that it stamped the latest lsn
+	to the FIL_PAGE_FILE_FLUSH_LSN in the first page of the .ibd file.
+
+	TODO: reset also the trx id's in clustered index records and write
+	a new space id to each data page. That would allow us to import clean
+	.ibd files from another MySQL installation. */
+
+	success = fil_reset_too_high_lsns(name, current_lsn);
+
+	if (!success) {
+		ut_print_timestamp(stderr);
+		fputs("  InnoDB: Error: cannot reset lsn's in table ", stderr);
+		ut_print_name(stderr, trx, TRUE, name);
+		fputs("\n"
+		      "InnoDB: in ALTER TABLE ... IMPORT TABLESPACE\n",
+		      stderr);
+
+		err = DB_ERROR;
+
+		row_mysql_lock_data_dictionary(trx);
+
+		goto funct_exit;
+	}
+
+	/* Serialize data dictionary operations with dictionary mutex:
+	no deadlocks can occur then in these operations */
+
+	row_mysql_lock_data_dictionary(trx);
+
+	table = dict_table_get_low(name);
+
+	if (!table) {
+		ut_print_timestamp(stderr);
+		fputs("  InnoDB: table ", stderr);
+		ut_print_name(stderr, trx, TRUE, name);
+		fputs("\n"
+		      "InnoDB: does not exist in the InnoDB data dictionary\n"
+		      "InnoDB: in ALTER TABLE ... IMPORT TABLESPACE\n",
+		      stderr);
+
+		err = DB_TABLE_NOT_FOUND;
+
+		goto funct_exit;
+	}
+
+	if (table->space == 0) {
+		ut_print_timestamp(stderr);
+		fputs("  InnoDB: Error: table ", stderr);
+		ut_print_name(stderr, trx, TRUE, name);
+		fputs("\n"
+		      "InnoDB: is in the system tablespace 0"
+		      " which cannot be imported\n", stderr);
+		err = DB_ERROR;
+
+		goto funct_exit;
+	}
+
+	if (!table->tablespace_discarded) {
+		ut_print_timestamp(stderr);
+		fputs("  InnoDB: Error: you are trying to"
+		      " IMPORT a tablespace\n"
+		      "InnoDB: ", stderr);
+		ut_print_name(stderr, trx, TRUE, name);
+		fputs(", though you have not called DISCARD on it yet\n"
+		      "InnoDB: during the lifetime of the mysqld process!\n",
+		      stderr);
+
+		err = DB_ERROR;
+
+		goto funct_exit;
+	}
+
+	/* Play safe and remove all insert buffer entries, though we should
+	have removed them already when DISCARD TABLESPACE was called */
+
+	ibuf_delete_for_discarded_space(table->space);
+
+	success = fil_open_single_table_tablespace(
+		TRUE, table->space,
+		table->flags == DICT_TF_COMPACT ? 0 : table->flags,
+		table->name);
+	if (success) {
+		table->ibd_file_missing = FALSE;
+		table->tablespace_discarded = FALSE;
+	} else {
+		if (table->ibd_file_missing) {
+			ut_print_timestamp(stderr);
+			fputs("  InnoDB: cannot find or open in the"
+			      " database directory the .ibd file of\n"
+			      "InnoDB: table ", stderr);
+			ut_print_name(stderr, trx, TRUE, name);
+			fputs("\n"
+			      "InnoDB: in ALTER TABLE ... IMPORT TABLESPACE\n",
+			      stderr);
+		}
+
+		err = DB_ERROR;
+	}
+
+funct_exit:
+	trx_commit_for_mysql(trx);
+
+	row_mysql_unlock_data_dictionary(trx);
+
+	trx->op_info = "";
+
+	return((int) err);
+}
+
+/*************************************************************************
+Truncates a table for MySQL. */
+UNIV_INTERN
+int
+row_truncate_table_for_mysql(
+/*=========================*/
+				/* out: error code or DB_SUCCESS */
+	dict_table_t*	table,	/* in: table handle */
+	trx_t*		trx)	/* in: transaction handle */
+{
+	dict_foreign_t*	foreign;
+	ulint		err;
+	mem_heap_t*	heap;
+	byte*		buf;
+	dtuple_t*	tuple;
+	dfield_t*	dfield;
+	dict_index_t*	sys_index;
+	btr_pcur_t	pcur;
+	mtr_t		mtr;
+	dulint		new_id;
+	ulint		recreate_space = 0;
+	pars_info_t*	info = NULL;
+
+	/* How do we prevent crashes caused by ongoing operations on
+	the table? Old operations could try to access non-existent
+	pages.
+
+	1) SQL queries, INSERT, SELECT, ...: we must get an exclusive
+	MySQL table lock on the table before we can do TRUNCATE
+	TABLE. Then there are no running queries on the table. This is
+	guaranteed, because in ha_innobase::store_lock(), we do not
+	weaken the TL_WRITE lock requested by MySQL when executing
+	SQLCOM_TRUNCATE.
+
+	2) Purge and rollback: we assign a new table id for the
+	table. Since purge and rollback look for the table based on
+	the table id, they see the table as 'dropped' and discard
+	their operations.
+
+	3) Insert buffer: TRUNCATE TABLE is analogous to DROP TABLE,
+	so we do not have to remove insert buffer records, as the
+	insert buffer works at a low level. If a freed page is later
+	reallocated, the allocator will remove the ibuf entries for
+	it.
+
+	When we truncate *.ibd files by recreating them (analogous to
+	DISCARD TABLESPACE), we remove all entries for the table in the
+	insert buffer tree.  This is not strictly necessary, because
+	in 6) we will assign a new tablespace identifier, but we can
+	free up some space in the system tablespace.
+
+	4) Linear readahead and random readahead: we use the same
+	method as in 3) to discard ongoing operations. (This is only
+	relevant for TRUNCATE TABLE by DISCARD TABLESPACE.)
+
+	5) FOREIGN KEY operations: if
+	table->n_foreign_key_checks_running > 0, we do not allow the
+	TRUNCATE. We also reserve the data dictionary latch.
+
+	6) Crash recovery: To prevent the application of pre-truncation
+	redo log records on the truncated tablespace, we will assign
+	a new tablespace identifier to the truncated tablespace. */
+
+	ut_ad(trx->mysql_thread_id == os_thread_get_curr_id());
+	ut_ad(table);
+
+	if (srv_created_new_raw) {
+		fputs("InnoDB: A new raw disk partition was initialized:\n"
+		      "InnoDB: we do not allow database modifications"
+		      " by the user.\n"
+		      "InnoDB: Shut down mysqld and edit my.cnf so that newraw"
+		      " is replaced with raw.\n", stderr);
+
+		return(DB_ERROR);
+	}
+
+	trx->op_info = "truncating table";
+
+	trx_start_if_not_started(trx);
+
+	/* Serialize data dictionary operations with dictionary mutex:
+	no deadlocks can occur then in these operations */
+
+	ut_a(trx->dict_operation_lock_mode == 0);
+	/* Prevent foreign key checks etc. while we are truncating the
+	table */
+
+	row_mysql_lock_data_dictionary(trx);
+
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(rw_lock_own(&dict_operation_lock, RW_LOCK_EX));
+#endif /* UNIV_SYNC_DEBUG */
+
+	/* Check if the table is referenced by foreign key constraints from
+	some other table (not the table itself) */
+
+	foreign = UT_LIST_GET_FIRST(table->referenced_list);
+
+	while (foreign && foreign->foreign_table == table) {
+		foreign = UT_LIST_GET_NEXT(referenced_list, foreign);
+	}
+
+	if (foreign && trx->check_foreigns) {
+		FILE*	ef	= dict_foreign_err_file;
+
+		/* We only allow truncating a referenced table if
+		FOREIGN_KEY_CHECKS is set to 0 */
+
+		mutex_enter(&dict_foreign_err_mutex);
+		rewind(ef);
+		ut_print_timestamp(ef);
+
+		fputs("  Cannot truncate table ", ef);
+		ut_print_name(ef, trx, TRUE, table->name);
+		fputs(" by DROP+CREATE\n"
+		      "InnoDB: because it is referenced by ", ef);
+		ut_print_name(ef, trx, TRUE, foreign->foreign_table_name);
+		putc('\n', ef);
+		mutex_exit(&dict_foreign_err_mutex);
+
+		err = DB_ERROR;
+		goto funct_exit;
+	}
+
+	/* TODO: could we replace the counter n_foreign_key_checks_running
+	with lock checks on the table? Acquire here an exclusive lock on the
+	table, and rewrite lock0lock.c and the lock wait in srv0srv.c so that
+	they can cope with the table having been truncated here? Foreign key
+	checks take an IS or IX lock on the table. */
+
+	if (table->n_foreign_key_checks_running > 0) {
+		ut_print_timestamp(stderr);
+		fputs("  InnoDB: Cannot truncate table ", stderr);
+		ut_print_name(stderr, trx, TRUE, table->name);
+		fputs(" by DROP+CREATE\n"
+		      "InnoDB: because there is a foreign key check"
+		      " running on it.\n",
+		      stderr);
+		err = DB_ERROR;
+
+		goto funct_exit;
+	}
+
+	/* Remove all locks except the table-level S and X locks. */
+	lock_remove_all_on_table(table, FALSE);
+
+	trx->table_id = table->id;
+
+	if (table->space && !table->dir_path_of_temp_table) {
+		/* Discard and create the single-table tablespace. */
+		ulint	space	= table->space;
+		ulint	flags	= fil_space_get_flags(space);
+
+		if (flags != ULINT_UNDEFINED
+		    && fil_discard_tablespace(space)) {
+
+			dict_index_t*	index;
+
+			space = 0;
+
+			if (fil_create_new_single_table_tablespace(
+				    &space, table->name, FALSE, flags,
+				    FIL_IBD_FILE_INITIAL_SIZE) != DB_SUCCESS) {
+				ut_print_timestamp(stderr);
+				fprintf(stderr,
+					"  InnoDB: TRUNCATE TABLE %s failed to"
+					" create a new tablespace\n",
+					table->name);
+				table->ibd_file_missing = 1;
+				err = DB_ERROR;
+				goto funct_exit;
+			}
+
+			recreate_space = space;
+
+			/* Replace the space_id in the data dictionary cache.
+			The persisent data dictionary (SYS_TABLES.SPACE
+			and SYS_INDEXES.SPACE) are updated later in this
+			function. */
+			table->space = space;
+			index = dict_table_get_first_index(table);
+			do {
+				index->space = space;
+				index = dict_table_get_next_index(index);
+			} while (index);
+
+			mtr_start(&mtr);
+			fsp_header_init(space,
+					FIL_IBD_FILE_INITIAL_SIZE, &mtr);
+			mtr_commit(&mtr);
+		}
+	}
+
+	/* scan SYS_INDEXES for all indexes of the table */
+	heap = mem_heap_create(800);
+
+	tuple = dtuple_create(heap, 1);
+	dfield = dtuple_get_nth_field(tuple, 0);
+
+	buf = mem_heap_alloc(heap, 8);
+	mach_write_to_8(buf, table->id);
+
+	dfield_set_data(dfield, buf, 8);
+	sys_index = dict_table_get_first_index(dict_sys->sys_indexes);
+	dict_index_copy_types(tuple, sys_index, 1);
+
+	mtr_start(&mtr);
+	btr_pcur_open_on_user_rec(sys_index, tuple, PAGE_CUR_GE,
+				  BTR_MODIFY_LEAF, &pcur, &mtr);
+	for (;;) {
+		rec_t*		rec;
+		const byte*	field;
+		ulint		len;
+		ulint		root_page_no;
+
+		if (!btr_pcur_is_on_user_rec(&pcur)) {
+			/* The end of SYS_INDEXES has been reached. */
+			break;
+		}
+
+		rec = btr_pcur_get_rec(&pcur);
+
+		field = rec_get_nth_field_old(rec, 0, &len);
+		ut_ad(len == 8);
+
+		if (memcmp(buf, field, len) != 0) {
+			/* End of indexes for the table (TABLE_ID mismatch). */
+			break;
+		}
+
+		if (rec_get_deleted_flag(rec, FALSE)) {
+			/* The index has been dropped. */
+			goto next_rec;
+		}
+
+		/* This call may commit and restart mtr
+		and reposition pcur. */
+		root_page_no = dict_truncate_index_tree(table, recreate_space,
+							&pcur, &mtr);
+
+		rec = btr_pcur_get_rec(&pcur);
+
+		if (root_page_no != FIL_NULL) {
+			page_rec_write_index_page_no(
+				rec, DICT_SYS_INDEXES_PAGE_NO_FIELD,
+				root_page_no, &mtr);
+			/* We will need to commit and restart the
+			mini-transaction in order to avoid deadlocks.
+			The dict_truncate_index_tree() call has allocated
+			a page in this mini-transaction, and the rest of
+			this loop could latch another index page. */
+			mtr_commit(&mtr);
+			mtr_start(&mtr);
+			btr_pcur_restore_position(BTR_MODIFY_LEAF,
+						  &pcur, &mtr);
+		}
+
+next_rec:
+		btr_pcur_move_to_next_user_rec(&pcur, &mtr);
+	}
+
+	btr_pcur_close(&pcur);
+	mtr_commit(&mtr);
+
+	mem_heap_free(heap);
+
+	new_id = dict_hdr_get_new_id(DICT_HDR_TABLE_ID);
+
+	info = pars_info_create();
+
+	pars_info_add_int4_literal(info, "space", (lint) table->space);
+	pars_info_add_dulint_literal(info, "old_id", table->id);
+	pars_info_add_dulint_literal(info, "new_id", new_id);
+
+	err = que_eval_sql(info,
+			   "PROCEDURE RENUMBER_TABLESPACE_PROC () IS\n"
+			   "BEGIN\n"
+			   "UPDATE SYS_TABLES"
+			   " SET ID = :new_id, SPACE = :space\n"
+			   " WHERE ID = :old_id;\n"
+			   "UPDATE SYS_COLUMNS SET TABLE_ID = :new_id\n"
+			   " WHERE TABLE_ID = :old_id;\n"
+			   "UPDATE SYS_INDEXES"
+			   " SET TABLE_ID = :new_id, SPACE = :space\n"
+			   " WHERE TABLE_ID = :old_id;\n"
+			   "COMMIT WORK;\n"
+			   "END;\n"
+			   , FALSE, trx);
+
+	if (err != DB_SUCCESS) {
+		trx->error_state = DB_SUCCESS;
+		trx_general_rollback_for_mysql(trx, FALSE, NULL);
+		trx->error_state = DB_SUCCESS;
+		ut_print_timestamp(stderr);
+		fputs("  InnoDB: Unable to assign a new identifier to table ",
+		      stderr);
+		ut_print_name(stderr, trx, TRUE, table->name);
+		fputs("\n"
+		      "InnoDB: after truncating it.  Background processes"
+		      " may corrupt the table!\n", stderr);
+		err = DB_ERROR;
+	} else {
+		dict_table_change_id_in_cache(table, new_id);
+	}
+
+	/* MySQL calls ha_innobase::reset_auto_increment() which does
+	the same thing. */
+	dict_table_autoinc_lock(table);
+	dict_table_autoinc_initialize(table, 1);
+	dict_table_autoinc_unlock(table);
+	dict_update_statistics(table);
+
+	trx_commit_for_mysql(trx);
+
+funct_exit:
+
+	row_mysql_unlock_data_dictionary(trx);
+
+	trx->op_info = "";
+
+	srv_wake_master_thread();
+
+	return((int) err);
+}
+
+/*************************************************************************
+Drops a table for MySQL.  If the name of the dropped table ends in
+one of "innodb_monitor", "innodb_lock_monitor", "innodb_tablespace_monitor",
+"innodb_table_monitor", then this will also stop the printing of monitor
+output by the master thread.  If the data dictionary was not already locked
+by the transaction, the transaction will be committed.  Otherwise, the
+data dictionary will remain locked. */
+UNIV_INTERN
+int
+row_drop_table_for_mysql(
+/*=====================*/
+				/* out: error code or DB_SUCCESS */
+	const char*	name,	/* in: table name */
+	trx_t*		trx,	/* in: transaction handle */
+	ibool		drop_db)/* in: TRUE=dropping whole database */
+{
+	dict_foreign_t*	foreign;
+	dict_table_t*	table;
+	ulint		space_id;
+	ulint		err;
+	const char*	table_name;
+	ulint		namelen;
+	ibool		locked_dictionary	= FALSE;
+	pars_info_t*    info			= NULL;
+
+	ut_a(name != NULL);
+
+	if (srv_created_new_raw) {
+		fputs("InnoDB: A new raw disk partition was initialized:\n"
+		      "InnoDB: we do not allow database modifications"
+		      " by the user.\n"
+		      "InnoDB: Shut down mysqld and edit my.cnf so that newraw"
+		      " is replaced with raw.\n", stderr);
+
+		return(DB_ERROR);
+	}
+
+	trx->op_info = "dropping table";
+
+	trx_start_if_not_started(trx);
+
+	/* The table name is prefixed with the database name and a '/'.
+	Certain table names starting with 'innodb_' have their special
+	meaning regardless of the database name.  Thus, we need to
+	ignore the database name prefix in the comparisons. */
+	table_name = strchr(name, '/');
+	ut_a(table_name);
+	table_name++;
+	namelen = strlen(table_name) + 1;
+
+	if (namelen == sizeof S_innodb_monitor
+	    && !memcmp(table_name, S_innodb_monitor,
+		       sizeof S_innodb_monitor)) {
+
+		/* Table name equals "innodb_monitor":
+		stop monitor prints */
+
+		srv_print_innodb_monitor = FALSE;
+		srv_print_innodb_lock_monitor = FALSE;
+	} else if (namelen == sizeof S_innodb_lock_monitor
+		   && !memcmp(table_name, S_innodb_lock_monitor,
+			      sizeof S_innodb_lock_monitor)) {
+		srv_print_innodb_monitor = FALSE;
+		srv_print_innodb_lock_monitor = FALSE;
+	} else if (namelen == sizeof S_innodb_tablespace_monitor
+		   && !memcmp(table_name, S_innodb_tablespace_monitor,
+			      sizeof S_innodb_tablespace_monitor)) {
+
+		srv_print_innodb_tablespace_monitor = FALSE;
+	} else if (namelen == sizeof S_innodb_table_monitor
+		   && !memcmp(table_name, S_innodb_table_monitor,
+			      sizeof S_innodb_table_monitor)) {
+
+		srv_print_innodb_table_monitor = FALSE;
+	}
+
+	/* Serialize data dictionary operations with dictionary mutex:
+	no deadlocks can occur then in these operations */
+
+	if (trx->dict_operation_lock_mode != RW_X_LATCH) {
+		/* Prevent foreign key checks etc. while we are dropping the
+		table */
+
+		row_mysql_lock_data_dictionary(trx);
+
+		locked_dictionary = TRUE;
+	}
+
+	ut_ad(mutex_own(&(dict_sys->mutex)));
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(rw_lock_own(&dict_operation_lock, RW_LOCK_EX));
+#endif /* UNIV_SYNC_DEBUG */
+
+	table = dict_table_get_low(name);
+
+	if (!table) {
+		err = DB_TABLE_NOT_FOUND;
+		ut_print_timestamp(stderr);
+
+		fputs("  InnoDB: Error: table ", stderr);
+		ut_print_name(stderr, trx, TRUE, name);
+		fputs(" does not exist in the InnoDB internal\n"
+		      "InnoDB: data dictionary though MySQL is"
+		      " trying to drop it.\n"
+		      "InnoDB: Have you copied the .frm file"
+		      " of the table to the\n"
+		      "InnoDB: MySQL database directory"
+		      " from another database?\n"
+		      "InnoDB: You can look for further help from\n"
+		      "InnoDB: http://dev.mysql.com/doc/refman/5.1/en/"
+		      "innodb-troubleshooting.html\n",
+		      stderr);
+		goto funct_exit;
+	}
+
+	/* Check if the table is referenced by foreign key constraints from
+	some other table (not the table itself) */
+
+	foreign = UT_LIST_GET_FIRST(table->referenced_list);
+
+	while (foreign && foreign->foreign_table == table) {
+check_next_foreign:
+		foreign = UT_LIST_GET_NEXT(referenced_list, foreign);
+	}
+
+	if (foreign && trx->check_foreigns
+	    && !(drop_db && dict_tables_have_same_db(
+			 name, foreign->foreign_table_name))) {
+		FILE*	ef	= dict_foreign_err_file;
+
+		/* We only allow dropping a referenced table if
+		FOREIGN_KEY_CHECKS is set to 0 */
+
+		err = DB_CANNOT_DROP_CONSTRAINT;
+
+		mutex_enter(&dict_foreign_err_mutex);
+		rewind(ef);
+		ut_print_timestamp(ef);
+
+		fputs("  Cannot drop table ", ef);
+		ut_print_name(ef, trx, TRUE, name);
+		fputs("\n"
+		      "because it is referenced by ", ef);
+		ut_print_name(ef, trx, TRUE, foreign->foreign_table_name);
+		putc('\n', ef);
+		mutex_exit(&dict_foreign_err_mutex);
+
+		goto funct_exit;
+	}
+
+	if (foreign && trx->check_foreigns) {
+		goto check_next_foreign;
+	}
+
+	if (table->n_mysql_handles_opened > 0) {
+		ibool	added;
+
+		added = row_add_table_to_background_drop_list(table->name);
+
+		if (added) {
+			ut_print_timestamp(stderr);
+			fputs("  InnoDB: Warning: MySQL is"
+			      " trying to drop table ", stderr);
+			ut_print_name(stderr, trx, TRUE, table->name);
+			fputs("\n"
+			      "InnoDB: though there are still"
+			      " open handles to it.\n"
+			      "InnoDB: Adding the table to the"
+			      " background drop queue.\n",
+			      stderr);
+
+			/* We return DB_SUCCESS to MySQL though the drop will
+			happen lazily later */
+			err = DB_SUCCESS;
+		} else {
+			/* The table is already in the background drop list */
+			err = DB_ERROR;
+		}
+
+		goto funct_exit;
+	}
+
+	/* TODO: could we replace the counter n_foreign_key_checks_running
+	with lock checks on the table? Acquire here an exclusive lock on the
+	table, and rewrite lock0lock.c and the lock wait in srv0srv.c so that
+	they can cope with the table having been dropped here? Foreign key
+	checks take an IS or IX lock on the table. */
+
+	if (table->n_foreign_key_checks_running > 0) {
+
+		const char*	table_name = table->name;
+		ibool		added;
+
+		added = row_add_table_to_background_drop_list(table_name);
+
+		if (added) {
+			ut_print_timestamp(stderr);
+			fputs("  InnoDB: You are trying to drop table ",
+			      stderr);
+			ut_print_name(stderr, trx, TRUE, table_name);
+			fputs("\n"
+			      "InnoDB: though there is a"
+			      " foreign key check running on it.\n"
+			      "InnoDB: Adding the table to"
+			      " the background drop queue.\n",
+			      stderr);
+
+			/* We return DB_SUCCESS to MySQL though the drop will
+			happen lazily later */
+
+			err = DB_SUCCESS;
+		} else {
+			/* The table is already in the background drop list */
+			err = DB_ERROR;
+		}
+
+		goto funct_exit;
+	}
+
+	/* Remove all locks there are on the table or its records */
+	lock_remove_all_on_table(table, TRUE);
+
+	trx_set_dict_operation(trx, TRX_DICT_OP_TABLE);
+	trx->table_id = table->id;
+
+	/* We use the private SQL parser of Innobase to generate the
+	query graphs needed in deleting the dictionary data from system
+	tables in Innobase. Deleting a row from SYS_INDEXES table also
+	frees the file segments of the B-tree associated with the index. */
+
+	info = pars_info_create();
+
+	pars_info_add_str_literal(info, "table_name", name);
+
+	err = que_eval_sql(info,
+			   "PROCEDURE DROP_TABLE_PROC () IS\n"
+			   "sys_foreign_id CHAR;\n"
+			   "table_id CHAR;\n"
+			   "index_id CHAR;\n"
+			   "foreign_id CHAR;\n"
+			   "found INT;\n"
+			   "BEGIN\n"
+			   "SELECT ID INTO table_id\n"
+			   "FROM SYS_TABLES\n"
+			   "WHERE NAME = :table_name\n"
+			   "LOCK IN SHARE MODE;\n"
+			   "IF (SQL % NOTFOUND) THEN\n"
+			   "       RETURN;\n"
+			   "END IF;\n"
+			   "found := 1;\n"
+			   "SELECT ID INTO sys_foreign_id\n"
+			   "FROM SYS_TABLES\n"
+			   "WHERE NAME = 'SYS_FOREIGN'\n"
+			   "LOCK IN SHARE MODE;\n"
+			   "IF (SQL % NOTFOUND) THEN\n"
+			   "       found := 0;\n"
+			   "END IF;\n"
+			   "IF (:table_name = 'SYS_FOREIGN') THEN\n"
+			   "       found := 0;\n"
+			   "END IF;\n"
+			   "IF (:table_name = 'SYS_FOREIGN_COLS') THEN\n"
+			   "       found := 0;\n"
+			   "END IF;\n"
+			   "WHILE found = 1 LOOP\n"
+			   "       SELECT ID INTO foreign_id\n"
+			   "       FROM SYS_FOREIGN\n"
+			   "       WHERE FOR_NAME = :table_name\n"
+			   "               AND TO_BINARY(FOR_NAME)\n"
+			   "                 = TO_BINARY(:table_name)\n"
+			   "               LOCK IN SHARE MODE;\n"
+			   "       IF (SQL % NOTFOUND) THEN\n"
+			   "               found := 0;\n"
+			   "       ELSE\n"
+			   "               DELETE FROM SYS_FOREIGN_COLS\n"
+			   "               WHERE ID = foreign_id;\n"
+			   "               DELETE FROM SYS_FOREIGN\n"
+			   "               WHERE ID = foreign_id;\n"
+			   "       END IF;\n"
+			   "END LOOP;\n"
+			   "found := 1;\n"
+			   "WHILE found = 1 LOOP\n"
+			   "       SELECT ID INTO index_id\n"
+			   "       FROM SYS_INDEXES\n"
+			   "       WHERE TABLE_ID = table_id\n"
+			   "       LOCK IN SHARE MODE;\n"
+			   "       IF (SQL % NOTFOUND) THEN\n"
+			   "               found := 0;\n"
+			   "       ELSE\n"
+			   "               DELETE FROM SYS_FIELDS\n"
+			   "               WHERE INDEX_ID = index_id;\n"
+			   "               DELETE FROM SYS_INDEXES\n"
+			   "               WHERE ID = index_id\n"
+			   "               AND TABLE_ID = table_id;\n"
+			   "       END IF;\n"
+			   "END LOOP;\n"
+			   "DELETE FROM SYS_COLUMNS\n"
+			   "WHERE TABLE_ID = table_id;\n"
+			   "DELETE FROM SYS_TABLES\n"
+			   "WHERE ID = table_id;\n"
+			   "END;\n"
+			   , FALSE, trx);
+
+	if (err != DB_SUCCESS) {
+		ut_a(err == DB_OUT_OF_FILE_SPACE);
+
+		err = DB_MUST_GET_MORE_FILE_SPACE;
+
+		row_mysql_handle_errors(&err, trx, NULL, NULL);
+
+		ut_error;
+	} else {
+		ibool		is_path;
+		const char*	name_or_path;
+		mem_heap_t*	heap;
+
+		heap = mem_heap_create(200);
+
+		/* Clone the name, in case it has been allocated
+		from table->heap, which will be freed by
+		dict_table_remove_from_cache(table) below. */
+		name = mem_heap_strdup(heap, name);
+		space_id = table->space;
+
+		if (table->dir_path_of_temp_table != NULL) {
+			is_path = TRUE;
+			name_or_path = mem_heap_strdup(
+				heap, table->dir_path_of_temp_table);
+		} else {
+			is_path = FALSE;
+			name_or_path = name;
+		}
+
+		dict_table_remove_from_cache(table);
+
+		if (dict_load_table(name) != NULL) {
+			ut_print_timestamp(stderr);
+			fputs("  InnoDB: Error: not able to remove table ",
+			      stderr);
+			ut_print_name(stderr, trx, TRUE, name);
+			fputs(" from the dictionary cache!\n", stderr);
+			err = DB_ERROR;
+		}
+
+		/* Do not drop possible .ibd tablespace if something went
+		wrong: we do not want to delete valuable data of the user */
+
+		if (err == DB_SUCCESS && space_id > 0) {
+			if (!fil_space_for_table_exists_in_mem(space_id,
+							       name_or_path,
+							       is_path,
+							       FALSE, TRUE)) {
+				err = DB_SUCCESS;
+
+				fprintf(stderr,
+					"InnoDB: We removed now the InnoDB"
+					" internal data dictionary entry\n"
+					"InnoDB: of table ");
+				ut_print_name(stderr, trx, TRUE, name);
+				fprintf(stderr, ".\n");
+			} else if (!fil_delete_tablespace(space_id)) {
+				fprintf(stderr,
+					"InnoDB: We removed now the InnoDB"
+					" internal data dictionary entry\n"
+					"InnoDB: of table ");
+				ut_print_name(stderr, trx, TRUE, name);
+				fprintf(stderr, ".\n");
+
+				ut_print_timestamp(stderr);
+				fprintf(stderr,
+					"  InnoDB: Error: not able to"
+					" delete tablespace %lu of table ",
+					(ulong) space_id);
+				ut_print_name(stderr, trx, TRUE, name);
+				fputs("!\n", stderr);
+				err = DB_ERROR;
+			}
+		}
+
+		mem_heap_free(heap);
+	}
+funct_exit:
+
+	if (locked_dictionary) {
+		trx_commit_for_mysql(trx);
+
+		row_mysql_unlock_data_dictionary(trx);
+	}
+
+	trx->op_info = "";
+
+#ifndef UNIV_HOTBACKUP
+	srv_wake_master_thread();
+#endif /* !UNIV_HOTBACKUP */
+
+	return((int) err);
+}
+
+/***********************************************************************
+Drop all foreign keys in a database, see Bug#18942.
+Called at the end of row_drop_database_for_mysql(). */
+static
+ulint
+drop_all_foreign_keys_in_db(
+/*========================*/
+				/* out: error code or DB_SUCCESS */
+	const char*	name,	/* in: database name which ends to '/' */
+	trx_t*		trx)	/* in: transaction handle */
+{
+	pars_info_t*	pinfo;
+	ulint		err;
+
+	ut_a(name[strlen(name) - 1] == '/');
+
+	pinfo = pars_info_create();
+
+	pars_info_add_str_literal(pinfo, "dbname", name);
+
+/* true if for_name is not prefixed with dbname */
+#define TABLE_NOT_IN_THIS_DB \
+"SUBSTR(for_name, 0, LENGTH(:dbname)) <> :dbname"
+
+	err = que_eval_sql(pinfo,
+			   "PROCEDURE DROP_ALL_FOREIGN_KEYS_PROC () IS\n"
+			   "foreign_id CHAR;\n"
+			   "for_name CHAR;\n"
+			   "found INT;\n"
+			   "DECLARE CURSOR cur IS\n"
+			   "SELECT ID, FOR_NAME FROM SYS_FOREIGN\n"
+			   "WHERE FOR_NAME >= :dbname\n"
+			   "LOCK IN SHARE MODE\n"
+			   "ORDER BY FOR_NAME;\n"
+			   "BEGIN\n"
+			   "found := 1;\n"
+			   "OPEN cur;\n"
+			   "WHILE found = 1 LOOP\n"
+			   "        FETCH cur INTO foreign_id, for_name;\n"
+			   "        IF (SQL % NOTFOUND) THEN\n"
+			   "                found := 0;\n"
+			   "        ELSIF (" TABLE_NOT_IN_THIS_DB ") THEN\n"
+			   "                found := 0;\n"
+			   "        ELSIF (1=1) THEN\n"
+			   "                DELETE FROM SYS_FOREIGN_COLS\n"
+			   "                WHERE ID = foreign_id;\n"
+			   "                DELETE FROM SYS_FOREIGN\n"
+			   "                WHERE ID = foreign_id;\n"
+			   "        END IF;\n"
+			   "END LOOP;\n"
+			   "CLOSE cur;\n"
+			   "COMMIT WORK;\n"
+			   "END;\n",
+			   FALSE, /* do not reserve dict mutex,
+				  we are already holding it */
+			   trx);
+
+	return(err);
+}
+
+/*************************************************************************
+Drops a database for MySQL. */
+UNIV_INTERN
+int
+row_drop_database_for_mysql(
+/*========================*/
+				/* out: error code or DB_SUCCESS */
+	const char*	name,	/* in: database name which ends to '/' */
+	trx_t*		trx)	/* in: transaction handle */
+{
+	dict_table_t* table;
+	char*	table_name;
+	int	err	= DB_SUCCESS;
+	ulint	namelen	= strlen(name);
+
+	ut_ad(trx->mysql_thread_id == os_thread_get_curr_id());
+	ut_a(name != NULL);
+	ut_a(name[namelen - 1] == '/');
+
+	trx->op_info = "dropping database";
+
+	trx_start_if_not_started(trx);
+loop:
+	row_mysql_lock_data_dictionary(trx);
+
+	while ((table_name = dict_get_first_table_name_in_db(name))) {
+		ut_a(memcmp(table_name, name, namelen) == 0);
+
+		table = dict_table_get_low(table_name);
+
+		ut_a(table);
+
+		/* Wait until MySQL does not have any queries running on
+		the table */
+
+		if (table->n_mysql_handles_opened > 0) {
+			row_mysql_unlock_data_dictionary(trx);
+
+			ut_print_timestamp(stderr);
+			fputs("  InnoDB: Warning: MySQL is trying to"
+			      " drop database ", stderr);
+			ut_print_name(stderr, trx, TRUE, name);
+			fputs("\n"
+			      "InnoDB: though there are still"
+			      " open handles to table ", stderr);
+			ut_print_name(stderr, trx, TRUE, table_name);
+			fputs(".\n", stderr);
+
+			os_thread_sleep(1000000);
+
+			mem_free(table_name);
+
+			goto loop;
+		}
+
+		err = row_drop_table_for_mysql(table_name, trx, TRUE);
+		trx_commit_for_mysql(trx);
+
+		if (err != DB_SUCCESS) {
+			fputs("InnoDB: DROP DATABASE ", stderr);
+			ut_print_name(stderr, trx, TRUE, name);
+			fprintf(stderr, " failed with error %lu for table ",
+				(ulint) err);
+			ut_print_name(stderr, trx, TRUE, table_name);
+			putc('\n', stderr);
+			mem_free(table_name);
+			break;
+		}
+
+		mem_free(table_name);
+	}
+
+	if (err == DB_SUCCESS) {
+		/* after dropping all tables try to drop all leftover
+		foreign keys in case orphaned ones exist */
+		err = (int) drop_all_foreign_keys_in_db(name, trx);
+
+		if (err != DB_SUCCESS) {
+			fputs("InnoDB: DROP DATABASE ", stderr);
+			ut_print_name(stderr, trx, TRUE, name);
+			fprintf(stderr, " failed with error %d while "
+				"dropping all foreign keys", err);
+		}
+	}
+
+	trx_commit_for_mysql(trx);
+
+	row_mysql_unlock_data_dictionary(trx);
+
+	trx->op_info = "";
+
+	return(err);
+}
+
+/*************************************************************************
+Checks if a table name contains the string "/#sql" which denotes temporary
+tables in MySQL. */
+static
+ibool
+row_is_mysql_tmp_table_name(
+/*========================*/
+				/* out: TRUE if temporary table */
+	const char*	name)	/* in: table name in the form
+				'database/tablename' */
+{
+	return(strstr(name, "/#sql") != NULL);
+	/* return(strstr(name, "/@0023sql") != NULL); */
+}
+
+/********************************************************************
+Delete a single constraint. */
+static
+int
+row_delete_constraint_low(
+/*======================*/
+					/* out: error code or DB_SUCCESS */
+	const char*	id,		/* in: constraint id */
+	trx_t*		trx)		/* in: transaction handle */
+{
+	pars_info_t*	info = pars_info_create();
+
+	pars_info_add_str_literal(info, "id", id);
+
+	return((int) que_eval_sql(info,
+			    "PROCEDURE DELETE_CONSTRAINT () IS\n"
+			    "BEGIN\n"
+			    "DELETE FROM SYS_FOREIGN_COLS WHERE ID = :id;\n"
+			    "DELETE FROM SYS_FOREIGN WHERE ID = :id;\n"
+			    "END;\n"
+			    , FALSE, trx));
+}
+
+/********************************************************************
+Delete a single constraint. */
+static
+int
+row_delete_constraint(
+/*==================*/
+					/* out: error code or DB_SUCCESS */
+	const char*	id,		/* in: constraint id */
+	const char*	database_name,	/* in: database name, with the
+					trailing '/' */
+	mem_heap_t*	heap,		/* in: memory heap */
+	trx_t*		trx)		/* in: transaction handle */
+{
+	ulint		err;
+
+	/* New format constraints have ids <databasename>/<constraintname>. */
+	err = row_delete_constraint_low(
+		mem_heap_strcat(heap, database_name, id), trx);
+
+	if ((err == DB_SUCCESS) && !strchr(id, '/')) {
+		/* Old format < 4.0.18 constraints have constraint ids
+		<number>_<number>. We only try deleting them if the
+		constraint name does not contain a '/' character, otherwise
+		deleting a new format constraint named 'foo/bar' from
+		database 'baz' would remove constraint 'bar' from database
+		'foo', if it existed. */
+
+		err = row_delete_constraint_low(id, trx);
+	}
+
+	return((int) err);
+}
+
+/*************************************************************************
+Renames a table for MySQL. */
+UNIV_INTERN
+ulint
+row_rename_table_for_mysql(
+/*=======================*/
+					/* out: error code or DB_SUCCESS */
+	const char*	old_name,	/* in: old table name */
+	const char*	new_name,	/* in: new table name */
+	trx_t*		trx,		/* in: transaction handle */
+	ibool		commit)		/* in: if TRUE then commit trx */
+{
+	dict_table_t*	table;
+	ulint		err			= DB_ERROR;
+	mem_heap_t*	heap			= NULL;
+	const char**	constraints_to_drop	= NULL;
+	ulint		n_constraints_to_drop	= 0;
+	ibool		old_is_tmp, new_is_tmp;
+	pars_info_t*	info			= NULL;
+
+	ut_ad(trx->mysql_thread_id == os_thread_get_curr_id());
+	ut_a(old_name != NULL);
+	ut_a(new_name != NULL);
+
+	if (srv_created_new_raw || srv_force_recovery) {
+		fputs("InnoDB: A new raw disk partition was initialized or\n"
+		      "InnoDB: innodb_force_recovery is on: we do not allow\n"
+		      "InnoDB: database modifications by the user. Shut down\n"
+		      "InnoDB: mysqld and edit my.cnf so that newraw"
+		      " is replaced\n"
+		      "InnoDB: with raw, and innodb_force_... is removed.\n",
+		      stderr);
+
+		goto funct_exit;
+	} else if (row_mysql_is_system_table(new_name)) {
+
+		fprintf(stderr,
+			"InnoDB: Error: trying to create a MySQL"
+			" system table %s of type InnoDB.\n"
+			"InnoDB: MySQL system tables must be"
+			" of the MyISAM type!\n",
+			new_name);
+
+		goto funct_exit;
+	}
+
+	trx->op_info = "renaming table";
+	trx_start_if_not_started(trx);
+
+	old_is_tmp = row_is_mysql_tmp_table_name(old_name);
+	new_is_tmp = row_is_mysql_tmp_table_name(new_name);
+
+	table = dict_table_get_low(old_name);
+
+	if (!table) {
+		err = DB_TABLE_NOT_FOUND;
+		ut_print_timestamp(stderr);
+
+		fputs("  InnoDB: Error: table ", stderr);
+		ut_print_name(stderr, trx, TRUE, old_name);
+		fputs(" does not exist in the InnoDB internal\n"
+		      "InnoDB: data dictionary though MySQL is"
+		      " trying to rename the table.\n"
+		      "InnoDB: Have you copied the .frm file"
+		      " of the table to the\n"
+		      "InnoDB: MySQL database directory"
+		      " from another database?\n"
+		      "InnoDB: You can look for further help from\n"
+		      "InnoDB: http://dev.mysql.com/doc/refman/5.1/en/"
+		      "innodb-troubleshooting.html\n",
+		      stderr);
+		goto funct_exit;
+	} else if (table->ibd_file_missing) {
+		err = DB_TABLE_NOT_FOUND;
+		ut_print_timestamp(stderr);
+
+		fputs("  InnoDB: Error: table ", stderr);
+		ut_print_name(stderr, trx, TRUE, old_name);
+		fputs(" does not have an .ibd file"
+		      " in the database directory.\n"
+		      "InnoDB: You can look for further help from\n"
+		      "InnoDB: http://dev.mysql.com/doc/refman/5.1/en/"
+		      "innodb-troubleshooting.html\n",
+		      stderr);
+		goto funct_exit;
+	} else if (new_is_tmp) {
+		/* MySQL is doing an ALTER TABLE command and it renames the
+		original table to a temporary table name. We want to preserve
+		the original foreign key constraint definitions despite the
+		name change. An exception is those constraints for which
+		the ALTER TABLE contained DROP FOREIGN KEY <foreign key id>.*/
+
+		heap = mem_heap_create(100);
+
+		err = dict_foreign_parse_drop_constraints(
+			heap, trx, table, &n_constraints_to_drop,
+			&constraints_to_drop);
+
+		if (err != DB_SUCCESS) {
+
+			goto funct_exit;
+		}
+	}
+
+	/* We use the private SQL parser of Innobase to generate the query
+	graphs needed in updating the dictionary data from system tables. */
+
+	info = pars_info_create();
+
+	pars_info_add_str_literal(info, "new_table_name", new_name);
+	pars_info_add_str_literal(info, "old_table_name", old_name);
+
+	err = que_eval_sql(info,
+			   "PROCEDURE RENAME_TABLE () IS\n"
+			   "BEGIN\n"
+			   "UPDATE SYS_TABLES SET NAME = :new_table_name\n"
+			   " WHERE NAME = :old_table_name;\n"
+			   "END;\n"
+			   , FALSE, trx);
+
+	if (err != DB_SUCCESS) {
+
+		goto end;
+	} else if (!new_is_tmp) {
+		/* Rename all constraints. */
+
+		info = pars_info_create();
+
+		pars_info_add_str_literal(info, "new_table_name", new_name);
+		pars_info_add_str_literal(info, "old_table_name", old_name);
+
+		err = que_eval_sql(
+			info,
+			"PROCEDURE RENAME_CONSTRAINT_IDS () IS\n"
+			"gen_constr_prefix CHAR;\n"
+			"new_db_name CHAR;\n"
+			"foreign_id CHAR;\n"
+			"new_foreign_id CHAR;\n"
+			"old_db_name_len INT;\n"
+			"old_t_name_len INT;\n"
+			"new_db_name_len INT;\n"
+			"id_len INT;\n"
+			"found INT;\n"
+			"BEGIN\n"
+			"found := 1;\n"
+			"old_db_name_len := INSTR(:old_table_name, '/')-1;\n"
+			"new_db_name_len := INSTR(:new_table_name, '/')-1;\n"
+			"new_db_name := SUBSTR(:new_table_name, 0,\n"
+			"                      new_db_name_len);\n"
+			"old_t_name_len := LENGTH(:old_table_name);\n"
+			"gen_constr_prefix := CONCAT(:old_table_name,\n"
+			"                            '_ibfk_');\n"
+			"WHILE found = 1 LOOP\n"
+			"       SELECT ID INTO foreign_id\n"
+			"        FROM SYS_FOREIGN\n"
+			"        WHERE FOR_NAME = :old_table_name\n"
+			"         AND TO_BINARY(FOR_NAME)\n"
+			"           = TO_BINARY(:old_table_name)\n"
+			"         LOCK IN SHARE MODE;\n"
+			"       IF (SQL % NOTFOUND) THEN\n"
+			"        found := 0;\n"
+			"       ELSE\n"
+			"        UPDATE SYS_FOREIGN\n"
+			"        SET FOR_NAME = :new_table_name\n"
+			"         WHERE ID = foreign_id;\n"
+			"        id_len := LENGTH(foreign_id);\n"
+			"        IF (INSTR(foreign_id, '/') > 0) THEN\n"
+			"               IF (INSTR(foreign_id,\n"
+			"                         gen_constr_prefix) > 0)\n"
+			"               THEN\n"
+			"                new_foreign_id :=\n"
+			"                CONCAT(:new_table_name,\n"
+			"                SUBSTR(foreign_id, old_t_name_len,\n"
+			"                       id_len - old_t_name_len));\n"
+			"               ELSE\n"
+			"                new_foreign_id :=\n"
+			"                CONCAT(new_db_name,\n"
+			"                SUBSTR(foreign_id,\n"
+			"                       old_db_name_len,\n"
+			"                       id_len - old_db_name_len));\n"
+			"               END IF;\n"
+			"               UPDATE SYS_FOREIGN\n"
+			"                SET ID = new_foreign_id\n"
+			"                WHERE ID = foreign_id;\n"
+			"               UPDATE SYS_FOREIGN_COLS\n"
+			"                SET ID = new_foreign_id\n"
+			"                WHERE ID = foreign_id;\n"
+			"        END IF;\n"
+			"       END IF;\n"
+			"END LOOP;\n"
+			"UPDATE SYS_FOREIGN SET REF_NAME = :new_table_name\n"
+			"WHERE REF_NAME = :old_table_name\n"
+			"  AND TO_BINARY(REF_NAME)\n"
+			"    = TO_BINARY(:old_table_name);\n"
+			"END;\n"
+			, FALSE, trx);
+
+	} else if (n_constraints_to_drop > 0) {
+		/* Drop some constraints of tmp tables. */
+
+		ulint	db_name_len = dict_get_db_name_len(old_name) + 1;
+		char*	db_name = mem_heap_strdupl(heap, old_name,
+						   db_name_len);
+		ulint	i;
+
+		for (i = 0; i < n_constraints_to_drop; i++) {
+			err = row_delete_constraint(constraints_to_drop[i],
+						    db_name, heap, trx);
+
+			if (err != DB_SUCCESS) {
+				break;
+			}
+		}
+	}
+
+end:
+	if (err != DB_SUCCESS) {
+		if (err == DB_DUPLICATE_KEY) {
+			ut_print_timestamp(stderr);
+			fputs("  InnoDB: Error; possible reasons:\n"
+			      "InnoDB: 1) Table rename would cause"
+			      " two FOREIGN KEY constraints\n"
+			      "InnoDB: to have the same internal name"
+			      " in case-insensitive comparison.\n"
+			      "InnoDB: 2) table ", stderr);
+			ut_print_name(stderr, trx, TRUE, new_name);
+			fputs(" exists in the InnoDB internal data\n"
+			      "InnoDB: dictionary though MySQL is"
+			      " trying to rename table ", stderr);
+			ut_print_name(stderr, trx, TRUE, old_name);
+			fputs(" to it.\n"
+			      "InnoDB: Have you deleted the .frm file"
+			      " and not used DROP TABLE?\n"
+			      "InnoDB: You can look for further help from\n"
+			      "InnoDB: http://dev.mysql.com/doc/refman/5.1/en/"
+			      "innodb-troubleshooting.html\n"
+			      "InnoDB: If table ", stderr);
+			ut_print_name(stderr, trx, TRUE, new_name);
+			fputs(" is a temporary table #sql..., then"
+			      " it can be that\n"
+			      "InnoDB: there are still queries running"
+			      " on the table, and it will be\n"
+			      "InnoDB: dropped automatically when"
+			      " the queries end.\n"
+			      "InnoDB: You can drop the orphaned table"
+			      " inside InnoDB by\n"
+			      "InnoDB: creating an InnoDB table with"
+			      " the same name in another\n"
+			      "InnoDB: database and copying the .frm file"
+			      " to the current database.\n"
+			      "InnoDB: Then MySQL thinks the table exists,"
+			      " and DROP TABLE will\n"
+			      "InnoDB: succeed.\n", stderr);
+		}
+		trx->error_state = DB_SUCCESS;
+		trx_general_rollback_for_mysql(trx, FALSE, NULL);
+		trx->error_state = DB_SUCCESS;
+	} else {
+		/* The following call will also rename the .ibd data file if
+		the table is stored in a single-table tablespace */
+
+		if (!dict_table_rename_in_cache(table, new_name,
+						!new_is_tmp)) {
+			trx->error_state = DB_SUCCESS;
+			trx_general_rollback_for_mysql(trx, FALSE, NULL);
+			trx->error_state = DB_SUCCESS;
+			goto funct_exit;
+		}
+
+		/* We only want to switch off some of the type checking in
+		an ALTER, not in a RENAME. */
+
+		err = dict_load_foreigns(
+			new_name, !old_is_tmp || trx->check_foreigns);
+
+		if (err != DB_SUCCESS) {
+			ut_print_timestamp(stderr);
+
+			if (old_is_tmp) {
+				fputs("  InnoDB: Error: in ALTER TABLE ",
+				      stderr);
+				ut_print_name(stderr, trx, TRUE, new_name);
+				fputs("\n"
+				      "InnoDB: has or is referenced"
+				      " in foreign key constraints\n"
+				      "InnoDB: which are not compatible"
+				      " with the new table definition.\n",
+				      stderr);
+			} else {
+				fputs("  InnoDB: Error: in RENAME TABLE"
+				      " table ",
+				      stderr);
+				ut_print_name(stderr, trx, TRUE, new_name);
+				fputs("\n"
+				      "InnoDB: is referenced in"
+				      " foreign key constraints\n"
+				      "InnoDB: which are not compatible"
+				      " with the new table definition.\n",
+				      stderr);
+			}
+
+			ut_a(dict_table_rename_in_cache(table,
+							old_name, FALSE));
+			trx->error_state = DB_SUCCESS;
+			trx_general_rollback_for_mysql(trx, FALSE, NULL);
+			trx->error_state = DB_SUCCESS;
+		}
+	}
+
+funct_exit:
+
+	if (commit) {
+		trx_commit_for_mysql(trx);
+	}
+
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+
+	trx->op_info = "";
+
+	return(err);
+}
+
+/*************************************************************************
+Checks that the index contains entries in an ascending order, unique
+constraint is not broken, and calculates the number of index entries
+in the read view of the current transaction. */
+static
+ibool
+row_scan_and_check_index(
+/*=====================*/
+					/* out: TRUE if ok */
+	row_prebuilt_t*	prebuilt,	/* in: prebuilt struct in MySQL */
+	dict_index_t*	index,		/* in: index */
+	ulint*		n_rows)		/* out: number of entries seen in the
+					current consistent read */
+{
+	dtuple_t*	prev_entry	= NULL;
+	ulint		matched_fields;
+	ulint		matched_bytes;
+	byte*		buf;
+	ulint		ret;
+	rec_t*		rec;
+	ibool		is_ok		= TRUE;
+	int		cmp;
+	ibool		contains_null;
+	ulint		i;
+	ulint		cnt;
+	mem_heap_t*	heap		= NULL;
+	ulint		n_ext;
+	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint*		offsets;
+	rec_offs_init(offsets_);
+
+	*n_rows = 0;
+
+	buf = mem_alloc(UNIV_PAGE_SIZE);
+	heap = mem_heap_create(100);
+
+	/* Make a dummy template in prebuilt, which we will use
+	in scanning the index entries */
+
+	prebuilt->index = index;
+	prebuilt->sql_stat_start = TRUE;
+	prebuilt->template_type = ROW_MYSQL_DUMMY_TEMPLATE;
+	prebuilt->n_template = 0;
+	prebuilt->need_to_access_clustered = FALSE;
+
+	dtuple_set_n_fields(prebuilt->search_tuple, 0);
+
+	prebuilt->select_lock_type = LOCK_NONE;
+	cnt = 1000;
+
+	ret = row_search_for_mysql(buf, PAGE_CUR_G, prebuilt, 0, 0);
+loop:
+	/* Check thd->killed every 1,000 scanned rows */
+	if (--cnt == 0) {
+		if (trx_is_interrupted(prebuilt->trx)) {
+			goto func_exit;
+		}
+		cnt = 1000;
+	}
+	if (ret != DB_SUCCESS) {
+func_exit:
+		mem_free(buf);
+		mem_heap_free(heap);
+
+		return(is_ok);
+	}
+
+	*n_rows = *n_rows + 1;
+
+	/* row_search... returns the index record in buf, record origin offset
+	within buf stored in the first 4 bytes, because we have built a dummy
+	template */
+
+	rec = buf + mach_read_from_4(buf);
+
+	offsets = rec_get_offsets(rec, index, offsets_,
+				  ULINT_UNDEFINED, &heap);
+
+	if (prev_entry != NULL) {
+		matched_fields = 0;
+		matched_bytes = 0;
+
+		cmp = cmp_dtuple_rec_with_match(prev_entry, rec, offsets,
+						&matched_fields,
+						&matched_bytes);
+		contains_null = FALSE;
+
+		/* In a unique secondary index we allow equal key values if
+		they contain SQL NULLs */
+
+		for (i = 0;
+		     i < dict_index_get_n_ordering_defined_by_user(index);
+		     i++) {
+			if (UNIV_SQL_NULL == dfield_get_len(
+				    dtuple_get_nth_field(prev_entry, i))) {
+
+				contains_null = TRUE;
+			}
+		}
+
+		if (cmp > 0) {
+			fputs("InnoDB: index records in a wrong order in ",
+			      stderr);
+not_ok:
+			dict_index_name_print(stderr,
+					      prebuilt->trx, index);
+			fputs("\n"
+			      "InnoDB: prev record ", stderr);
+			dtuple_print(stderr, prev_entry);
+			fputs("\n"
+			      "InnoDB: record ", stderr);
+			rec_print_new(stderr, rec, offsets);
+			putc('\n', stderr);
+			is_ok = FALSE;
+		} else if (dict_index_is_unique(index)
+			   && !contains_null
+			   && matched_fields
+			   >= dict_index_get_n_ordering_defined_by_user(
+				   index)) {
+
+			fputs("InnoDB: duplicate key in ", stderr);
+			goto not_ok;
+		}
+	}
+
+	{
+		mem_heap_t*	tmp_heap = NULL;
+
+		/* Empty the heap on each round.  But preserve offsets[]
+		for the row_rec_to_index_entry() call, by copying them
+		into a separate memory heap when needed. */
+		if (UNIV_UNLIKELY(offsets != offsets_)) {
+			ulint	size = rec_offs_get_n_alloc(offsets)
+				* sizeof *offsets;
+
+			tmp_heap = mem_heap_create(size);
+			offsets = mem_heap_dup(tmp_heap, offsets, size);
+		}
+
+		mem_heap_empty(heap);
+
+		prev_entry = row_rec_to_index_entry(ROW_COPY_DATA, rec,
+						    index, offsets,
+						    &n_ext, heap);
+
+		if (UNIV_LIKELY_NULL(tmp_heap)) {
+			mem_heap_free(tmp_heap);
+		}
+	}
+
+	ret = row_search_for_mysql(buf, PAGE_CUR_G, prebuilt, 0, ROW_SEL_NEXT);
+
+	goto loop;
+}
+
+/*************************************************************************
+Checks a table for corruption. */
+UNIV_INTERN
+ulint
+row_check_table_for_mysql(
+/*======================*/
+					/* out: DB_ERROR or DB_SUCCESS */
+	row_prebuilt_t*	prebuilt)	/* in: prebuilt struct in MySQL
+					handle */
+{
+	dict_table_t*	table		= prebuilt->table;
+	dict_index_t*	index;
+	ulint		n_rows;
+	ulint		n_rows_in_table	= ULINT_UNDEFINED;
+	ulint		ret		= DB_SUCCESS;
+	ulint		old_isolation_level;
+
+	if (table->ibd_file_missing) {
+		ut_print_timestamp(stderr);
+		fprintf(stderr, "  InnoDB: Error:\n"
+			"InnoDB: MySQL is trying to use a table handle"
+			" but the .ibd file for\n"
+			"InnoDB: table %s does not exist.\n"
+			"InnoDB: Have you deleted the .ibd file"
+			" from the database directory under\n"
+			"InnoDB: the MySQL datadir, or have you"
+			" used DISCARD TABLESPACE?\n"
+			"InnoDB: Look from\n"
+			"InnoDB: http://dev.mysql.com/doc/refman/5.1/en/"
+			"innodb-troubleshooting.html\n"
+			"InnoDB: how you can resolve the problem.\n",
+			table->name);
+		return(DB_ERROR);
+	}
+
+	prebuilt->trx->op_info = "checking table";
+
+	old_isolation_level = prebuilt->trx->isolation_level;
+
+	/* We must run the index record counts at an isolation level
+	>= READ COMMITTED, because a dirty read can see a wrong number
+	of records in some index; to play safe, we use always
+	REPEATABLE READ here */
+
+	prebuilt->trx->isolation_level = TRX_ISO_REPEATABLE_READ;
+
+	/* Enlarge the fatal lock wait timeout during CHECK TABLE. */
+	mutex_enter(&kernel_mutex);
+	srv_fatal_semaphore_wait_threshold += 7200; /* 2 hours */
+	mutex_exit(&kernel_mutex);
+
+	index = dict_table_get_first_index(table);
+
+	while (index != NULL) {
+		/* fputs("Validating index ", stderr);
+		ut_print_name(stderr, trx, FALSE, index->name);
+		putc('\n', stderr); */
+
+		if (!btr_validate_index(index, prebuilt->trx)) {
+			ret = DB_ERROR;
+		} else {
+			if (!row_scan_and_check_index(prebuilt,index, &n_rows)){
+				ret = DB_ERROR;
+			}
+
+			if (trx_is_interrupted(prebuilt->trx)) {
+				break;
+			}
+
+			/* fprintf(stderr, "%lu entries in index %s\n", n_rows,
+			index->name); */
+
+			if (index == dict_table_get_first_index(table)) {
+				n_rows_in_table = n_rows;
+			} else if (n_rows != n_rows_in_table) {
+
+				ret = DB_ERROR;
+
+				fputs("Error: ", stderr);
+				dict_index_name_print(stderr,
+						      prebuilt->trx, index);
+				fprintf(stderr,
+					" contains %lu entries,"
+					" should be %lu\n",
+					(ulong) n_rows,
+					(ulong) n_rows_in_table);
+			}
+		}
+
+		index = dict_table_get_next_index(index);
+	}
+
+	/* Restore the original isolation level */
+	prebuilt->trx->isolation_level = old_isolation_level;
+
+	/* We validate also the whole adaptive hash index for all tables
+	at every CHECK TABLE */
+
+	if (!btr_search_validate()) {
+
+		ret = DB_ERROR;
+	}
+
+	/* Restore the fatal lock wait timeout after CHECK TABLE. */
+	mutex_enter(&kernel_mutex);
+	srv_fatal_semaphore_wait_threshold -= 7200; /* 2 hours */
+	mutex_exit(&kernel_mutex);
+
+	prebuilt->trx->op_info = "";
+
+	return(ret);
+}
+#endif /* !UNIV_HOTBACKUP */
+
+/*************************************************************************
+Determines if a table is a magic monitor table. */
+UNIV_INTERN
+ibool
+row_is_magic_monitor_table(
+/*=======================*/
+					/* out: TRUE if monitor table */
+	const char*	table_name)	/* in: name of the table, in the
+					form database/table_name */
+{
+	const char*	name; /* table_name without database/ */
+	ulint		len;
+
+	name = strchr(table_name, '/');
+	ut_a(name != NULL);
+	name++;
+	len = strlen(name) + 1;
+
+	if (STR_EQ(name, len, S_innodb_monitor)
+	    || STR_EQ(name, len, S_innodb_lock_monitor)
+	    || STR_EQ(name, len, S_innodb_tablespace_monitor)
+	    || STR_EQ(name, len, S_innodb_table_monitor)
+	    || STR_EQ(name, len, S_innodb_mem_validate)) {
+
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
diff --git a/storage/xtradb/row/row0purge.c b/storage/xtradb/row/row0purge.c
new file mode 100644
index 00000000000..8c3f9b993ba
--- /dev/null
+++ b/storage/xtradb/row/row0purge.c
@@ -0,0 +1,690 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Purge obsolete records
+
+Created 3/14/1997 Heikki Tuuri
+*******************************************************/
+
+#include "row0purge.h"
+
+#ifdef UNIV_NONINL
+#include "row0purge.ic"
+#endif
+
+#include "fsp0fsp.h"
+#include "mach0data.h"
+#include "trx0rseg.h"
+#include "trx0trx.h"
+#include "trx0roll.h"
+#include "trx0undo.h"
+#include "trx0purge.h"
+#include "trx0rec.h"
+#include "que0que.h"
+#include "row0row.h"
+#include "row0upd.h"
+#include "row0vers.h"
+#include "row0mysql.h"
+#include "log0log.h"
+
+/************************************************************************
+Creates a purge node to a query graph. */
+UNIV_INTERN
+purge_node_t*
+row_purge_node_create(
+/*==================*/
+				/* out, own: purge node */
+	que_thr_t*	parent,	/* in: parent node, i.e., a thr node */
+	mem_heap_t*	heap)	/* in: memory heap where created */
+{
+	purge_node_t*	node;
+
+	ut_ad(parent && heap);
+
+	node = mem_heap_alloc(heap, sizeof(purge_node_t));
+
+	node->common.type = QUE_NODE_PURGE;
+	node->common.parent = parent;
+
+	node->heap = mem_heap_create(256);
+
+	return(node);
+}
+
+/***************************************************************
+Repositions the pcur in the purge node on the clustered index record,
+if found. */
+static
+ibool
+row_purge_reposition_pcur(
+/*======================*/
+				/* out: TRUE if the record was found */
+	ulint		mode,	/* in: latching mode */
+	purge_node_t*	node,	/* in: row purge node */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	ibool	found;
+
+	if (node->found_clust) {
+		found = btr_pcur_restore_position(mode, &(node->pcur), mtr);
+
+		return(found);
+	}
+
+	found = row_search_on_row_ref(&(node->pcur), mode, node->table,
+				      node->ref, mtr);
+	node->found_clust = found;
+
+	if (found) {
+		btr_pcur_store_position(&(node->pcur), mtr);
+	}
+
+	return(found);
+}
+
+/***************************************************************
+Removes a delete marked clustered index record if possible. */
+static
+ibool
+row_purge_remove_clust_if_poss_low(
+/*===============================*/
+				/* out: TRUE if success, or if not found, or
+				if modified after the delete marking */
+	purge_node_t*	node,	/* in: row purge node */
+	ulint		mode)	/* in: BTR_MODIFY_LEAF or BTR_MODIFY_TREE */
+{
+	dict_index_t*	index;
+	btr_pcur_t*	pcur;
+	btr_cur_t*	btr_cur;
+	ibool		success;
+	ulint		err;
+	mtr_t		mtr;
+	rec_t*		rec;
+	mem_heap_t*	heap		= NULL;
+	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+	rec_offs_init(offsets_);
+
+	index = dict_table_get_first_index(node->table);
+
+	pcur = &(node->pcur);
+	btr_cur = btr_pcur_get_btr_cur(pcur);
+
+	mtr_start(&mtr);
+
+	success = row_purge_reposition_pcur(mode, node, &mtr);
+
+	if (!success) {
+		/* The record is already removed */
+
+		btr_pcur_commit_specify_mtr(pcur, &mtr);
+
+		return(TRUE);
+	}
+
+	rec = btr_pcur_get_rec(pcur);
+
+	if (0 != ut_dulint_cmp(node->roll_ptr, row_get_rec_roll_ptr(
+				       rec, index, rec_get_offsets(
+					       rec, index, offsets_,
+					       ULINT_UNDEFINED, &heap)))) {
+		if (UNIV_LIKELY_NULL(heap)) {
+			mem_heap_free(heap);
+		}
+		/* Someone else has modified the record later: do not remove */
+		btr_pcur_commit_specify_mtr(pcur, &mtr);
+
+		return(TRUE);
+	}
+
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+
+	if (mode == BTR_MODIFY_LEAF) {
+		success = btr_cur_optimistic_delete(btr_cur, &mtr);
+	} else {
+		ut_ad(mode == BTR_MODIFY_TREE);
+		btr_cur_pessimistic_delete(&err, FALSE, btr_cur,
+					   RB_NONE, &mtr);
+
+		if (err == DB_SUCCESS) {
+			success = TRUE;
+		} else if (err == DB_OUT_OF_FILE_SPACE) {
+			success = FALSE;
+		} else {
+			ut_error;
+		}
+	}
+
+	btr_pcur_commit_specify_mtr(pcur, &mtr);
+
+	return(success);
+}
+
+/***************************************************************
+Removes a clustered index record if it has not been modified after the delete
+marking. */
+static
+void
+row_purge_remove_clust_if_poss(
+/*===========================*/
+	purge_node_t*	node)	/* in: row purge node */
+{
+	ibool	success;
+	ulint	n_tries	= 0;
+
+	/*	fputs("Purge: Removing clustered record\n", stderr); */
+
+	success = row_purge_remove_clust_if_poss_low(node, BTR_MODIFY_LEAF);
+	if (success) {
+
+		return;
+	}
+retry:
+	success = row_purge_remove_clust_if_poss_low(node, BTR_MODIFY_TREE);
+	/* The delete operation may fail if we have little
+	file space left: TODO: easiest to crash the database
+	and restart with more file space */
+
+	if (!success && n_tries < BTR_CUR_RETRY_DELETE_N_TIMES) {
+		n_tries++;
+
+		os_thread_sleep(BTR_CUR_RETRY_SLEEP_TIME);
+
+		goto retry;
+	}
+
+	ut_a(success);
+}
+
+/***************************************************************
+Removes a secondary index entry if possible. */
+static
+ibool
+row_purge_remove_sec_if_poss_low(
+/*=============================*/
+				/* out: TRUE if success or if not found */
+	purge_node_t*	node,	/* in: row purge node */
+	dict_index_t*	index,	/* in: index */
+	const dtuple_t*	entry,	/* in: index entry */
+	ulint		mode)	/* in: latch mode BTR_MODIFY_LEAF or
+				BTR_MODIFY_TREE */
+{
+	btr_pcur_t	pcur;
+	btr_cur_t*	btr_cur;
+	ibool		success;
+	ibool		old_has = 0; /* remove warning */
+	ibool		found;
+	ulint		err;
+	mtr_t		mtr;
+	mtr_t		mtr_vers;
+
+	log_free_check();
+	mtr_start(&mtr);
+
+	found = row_search_index_entry(index, entry, mode, &pcur, &mtr);
+
+	if (!found) {
+		/* Not found.  This is a legitimate condition.  In a
+		rollback, InnoDB will remove secondary recs that would
+		be purged anyway.  Then the actual purge will not find
+		the secondary index record.  Also, the purge itself is
+		eager: if it comes to consider a secondary index
+		record, and notices it does not need to exist in the
+		index, it will remove it.  Then if/when the purge
+		comes to consider the secondary index record a second
+		time, it will not exist any more in the index. */
+
+		/* fputs("PURGE:........sec entry not found\n", stderr); */
+		/* dtuple_print(stderr, entry); */
+
+		btr_pcur_close(&pcur);
+		mtr_commit(&mtr);
+
+		return(TRUE);
+	}
+
+	btr_cur = btr_pcur_get_btr_cur(&pcur);
+
+	/* We should remove the index record if no later version of the row,
+	which cannot be purged yet, requires its existence. If some requires,
+	we should do nothing. */
+
+	mtr_start(&mtr_vers);
+
+	success = row_purge_reposition_pcur(BTR_SEARCH_LEAF, node, &mtr_vers);
+
+	if (success) {
+		old_has = row_vers_old_has_index_entry(
+			TRUE, btr_pcur_get_rec(&(node->pcur)),
+			&mtr_vers, index, entry);
+	}
+
+	btr_pcur_commit_specify_mtr(&(node->pcur), &mtr_vers);
+
+	if (!success || !old_has) {
+		/* Remove the index record */
+
+		if (mode == BTR_MODIFY_LEAF) {
+			success = btr_cur_optimistic_delete(btr_cur, &mtr);
+		} else {
+			ut_ad(mode == BTR_MODIFY_TREE);
+			btr_cur_pessimistic_delete(&err, FALSE, btr_cur,
+						   RB_NONE, &mtr);
+			success = err == DB_SUCCESS;
+			ut_a(success || err == DB_OUT_OF_FILE_SPACE);
+		}
+	}
+
+	btr_pcur_close(&pcur);
+	mtr_commit(&mtr);
+
+	return(success);
+}
+
+/***************************************************************
+Removes a secondary index entry if possible. */
+UNIV_INLINE
+void
+row_purge_remove_sec_if_poss(
+/*=========================*/
+	purge_node_t*	node,	/* in: row purge node */
+	dict_index_t*	index,	/* in: index */
+	dtuple_t*	entry)	/* in: index entry */
+{
+	ibool	success;
+	ulint	n_tries		= 0;
+
+	/*	fputs("Purge: Removing secondary record\n", stderr); */
+
+	success = row_purge_remove_sec_if_poss_low(node, index, entry,
+						   BTR_MODIFY_LEAF);
+	if (success) {
+
+		return;
+	}
+retry:
+	success = row_purge_remove_sec_if_poss_low(node, index, entry,
+						   BTR_MODIFY_TREE);
+	/* The delete operation may fail if we have little
+	file space left: TODO: easiest to crash the database
+	and restart with more file space */
+
+	if (!success && n_tries < BTR_CUR_RETRY_DELETE_N_TIMES) {
+
+		n_tries++;
+
+		os_thread_sleep(BTR_CUR_RETRY_SLEEP_TIME);
+
+		goto retry;
+	}
+
+	ut_a(success);
+}
+
+/***************************************************************
+Purges a delete marking of a record. */
+static
+void
+row_purge_del_mark(
+/*===============*/
+	purge_node_t*	node)	/* in: row purge node */
+{
+	mem_heap_t*	heap;
+	dtuple_t*	entry;
+	dict_index_t*	index;
+
+	ut_ad(node);
+
+	heap = mem_heap_create(1024);
+
+	while (node->index != NULL) {
+		index = node->index;
+
+		/* Build the index entry */
+		entry = row_build_index_entry(node->row, NULL, index, heap);
+		ut_a(entry);
+		row_purge_remove_sec_if_poss(node, index, entry);
+
+		node->index = dict_table_get_next_index(node->index);
+	}
+
+	mem_heap_free(heap);
+
+	row_purge_remove_clust_if_poss(node);
+}
+
+/***************************************************************
+Purges an update of an existing record. Also purges an update of a delete
+marked record if that record contained an externally stored field. */
+static
+void
+row_purge_upd_exist_or_extern(
+/*==========================*/
+	purge_node_t*	node)	/* in: row purge node */
+{
+	mem_heap_t*	heap;
+	dtuple_t*	entry;
+	dict_index_t*	index;
+	ibool		is_insert;
+	ulint		rseg_id;
+	ulint		page_no;
+	ulint		offset;
+	ulint		i;
+	mtr_t		mtr;
+
+	ut_ad(node);
+
+	if (node->rec_type == TRX_UNDO_UPD_DEL_REC) {
+
+		goto skip_secondaries;
+	}
+
+	heap = mem_heap_create(1024);
+
+	while (node->index != NULL) {
+		index = node->index;
+
+		if (row_upd_changes_ord_field_binary(NULL, node->index,
+						     node->update)) {
+			/* Build the older version of the index entry */
+			entry = row_build_index_entry(node->row, NULL,
+						      index, heap);
+			ut_a(entry);
+			row_purge_remove_sec_if_poss(node, index, entry);
+		}
+
+		node->index = dict_table_get_next_index(node->index);
+	}
+
+	mem_heap_free(heap);
+
+skip_secondaries:
+	/* Free possible externally stored fields */
+	for (i = 0; i < upd_get_n_fields(node->update); i++) {
+
+		const upd_field_t*	ufield
+			= upd_get_nth_field(node->update, i);
+
+		if (dfield_is_ext(&ufield->new_val)) {
+			buf_block_t*	block;
+			ulint		internal_offset;
+			byte*		data_field;
+
+			/* We use the fact that new_val points to
+			node->undo_rec and get thus the offset of
+			dfield data inside the undo record. Then we
+			can calculate from node->roll_ptr the file
+			address of the new_val data */
+
+			internal_offset
+				= ((const byte*)
+				   dfield_get_data(&ufield->new_val))
+				- node->undo_rec;
+
+			ut_a(internal_offset < UNIV_PAGE_SIZE);
+
+			trx_undo_decode_roll_ptr(node->roll_ptr,
+						 &is_insert, &rseg_id,
+						 &page_no, &offset);
+			mtr_start(&mtr);
+
+			/* We have to acquire an X-latch to the clustered
+			index tree */
+
+			index = dict_table_get_first_index(node->table);
+
+			mtr_x_lock(dict_index_get_lock(index), &mtr);
+
+			/* NOTE: we must also acquire an X-latch to the
+			root page of the tree. We will need it when we
+			free pages from the tree. If the tree is of height 1,
+			the tree X-latch does NOT protect the root page,
+			because it is also a leaf page. Since we will have a
+			latch on an undo log page, we would break the
+			latching order if we would only later latch the
+			root page of such a tree! */
+
+			btr_root_get(index, &mtr);
+
+			/* We assume in purge of externally stored fields
+			that the space id of the undo log record is 0! */
+
+			block = buf_page_get(0, 0, page_no, RW_X_LATCH, &mtr);
+			buf_block_dbg_add_level(block, SYNC_TRX_UNDO_PAGE);
+
+			data_field = buf_block_get_frame(block)
+				+ offset + internal_offset;
+
+			ut_a(dfield_get_len(&ufield->new_val)
+			     >= BTR_EXTERN_FIELD_REF_SIZE);
+			btr_free_externally_stored_field(
+				index,
+				data_field + dfield_get_len(&ufield->new_val)
+				- BTR_EXTERN_FIELD_REF_SIZE,
+				NULL, NULL, NULL, 0, RB_NONE, &mtr);
+			mtr_commit(&mtr);
+		}
+	}
+}
+
+/***************************************************************
+Parses the row reference and other info in a modify undo log record. */
+static
+ibool
+row_purge_parse_undo_rec(
+/*=====================*/
+				/* out: TRUE if purge operation required:
+				NOTE that then the CALLER must unfreeze
+				data dictionary! */
+	purge_node_t*	node,	/* in: row undo node */
+	ibool*		updated_extern,
+				/* out: TRUE if an externally stored field
+				was updated */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	dict_index_t*	clust_index;
+	byte*		ptr;
+	trx_t*		trx;
+	dulint		undo_no;
+	dulint		table_id;
+	dulint		trx_id;
+	dulint		roll_ptr;
+	ulint		info_bits;
+	ulint		type;
+	ulint		cmpl_info;
+
+	ut_ad(node && thr);
+
+	trx = thr_get_trx(thr);
+
+	ptr = trx_undo_rec_get_pars(node->undo_rec, &type, &cmpl_info,
+				    updated_extern, &undo_no, &table_id);
+	node->rec_type = type;
+
+	if (type == TRX_UNDO_UPD_DEL_REC && !(*updated_extern)) {
+
+		return(FALSE);
+	}
+
+	ptr = trx_undo_update_rec_get_sys_cols(ptr, &trx_id, &roll_ptr,
+					       &info_bits);
+	node->table = NULL;
+
+	if (type == TRX_UNDO_UPD_EXIST_REC
+	    && cmpl_info & UPD_NODE_NO_ORD_CHANGE && !(*updated_extern)) {
+
+		/* Purge requires no changes to indexes: we may return */
+
+		return(FALSE);
+	}
+
+	/* Prevent DROP TABLE etc. from running when we are doing the purge
+	for this row */
+
+	row_mysql_freeze_data_dictionary(trx);
+
+	mutex_enter(&(dict_sys->mutex));
+
+	node->table = dict_table_get_on_id_low(table_id);
+
+	mutex_exit(&(dict_sys->mutex));
+
+	if (node->table == NULL) {
+		/* The table has been dropped: no need to do purge */
+err_exit:
+		row_mysql_unfreeze_data_dictionary(trx);
+		return(FALSE);
+	}
+
+	if (node->table->ibd_file_missing) {
+		/* We skip purge of missing .ibd files */
+
+		node->table = NULL;
+
+		goto err_exit;
+	}
+
+	clust_index = dict_table_get_first_index(node->table);
+
+	if (clust_index == NULL) {
+		/* The table was corrupt in the data dictionary */
+
+		goto err_exit;
+	}
+
+	ptr = trx_undo_rec_get_row_ref(ptr, clust_index, &(node->ref),
+				       node->heap);
+
+	ptr = trx_undo_update_rec_get_update(ptr, clust_index, type, trx_id,
+					     roll_ptr, info_bits, trx,
+					     node->heap, &(node->update));
+
+	/* Read to the partial row the fields that occur in indexes */
+
+	if (!(cmpl_info & UPD_NODE_NO_ORD_CHANGE)) {
+		ptr = trx_undo_rec_get_partial_row(
+			ptr, clust_index, &node->row,
+			type == TRX_UNDO_UPD_DEL_REC,
+			node->heap);
+	}
+
+	return(TRUE);
+}
+
+/***************************************************************
+Fetches an undo log record and does the purge for the recorded operation.
+If none left, or the current purge completed, returns the control to the
+parent node, which is always a query thread node. */
+static
+ulint
+row_purge(
+/*======*/
+				/* out: DB_SUCCESS if operation successfully
+				completed, else error code */
+	purge_node_t*	node,	/* in: row purge node */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	dulint	roll_ptr;
+	ibool	purge_needed;
+	ibool	updated_extern;
+	trx_t*	trx;
+
+	ut_ad(node && thr);
+
+	trx = thr_get_trx(thr);
+
+	node->undo_rec = trx_purge_fetch_next_rec(&roll_ptr,
+						  &(node->reservation),
+						  node->heap);
+	if (!node->undo_rec) {
+		/* Purge completed for this query thread */
+
+		thr->run_node = que_node_get_parent(node);
+
+		return(DB_SUCCESS);
+	}
+
+	node->roll_ptr = roll_ptr;
+
+	if (node->undo_rec == &trx_purge_dummy_rec) {
+		purge_needed = FALSE;
+	} else {
+		purge_needed = row_purge_parse_undo_rec(node, &updated_extern,
+							thr);
+		/* If purge_needed == TRUE, we must also remember to unfreeze
+		data dictionary! */
+	}
+
+	if (purge_needed) {
+		node->found_clust = FALSE;
+
+		node->index = dict_table_get_next_index(
+			dict_table_get_first_index(node->table));
+
+		if (node->rec_type == TRX_UNDO_DEL_MARK_REC) {
+			row_purge_del_mark(node);
+
+		} else if (updated_extern
+			   || node->rec_type == TRX_UNDO_UPD_EXIST_REC) {
+
+			row_purge_upd_exist_or_extern(node);
+		}
+
+		if (node->found_clust) {
+			btr_pcur_close(&(node->pcur));
+		}
+
+		row_mysql_unfreeze_data_dictionary(trx);
+	}
+
+	/* Do some cleanup */
+	trx_purge_rec_release(node->reservation);
+	mem_heap_empty(node->heap);
+
+	thr->run_node = node;
+
+	return(DB_SUCCESS);
+}
+
+/***************************************************************
+Does the purge operation for a single undo log record. This is a high-level
+function used in an SQL execution graph. */
+UNIV_INTERN
+que_thr_t*
+row_purge_step(
+/*===========*/
+				/* out: query thread to run next or NULL */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	purge_node_t*	node;
+	ulint		err;
+
+	ut_ad(thr);
+
+	node = thr->run_node;
+
+	ut_ad(que_node_get_type(node) == QUE_NODE_PURGE);
+
+	err = row_purge(node, thr);
+
+	ut_ad(err == DB_SUCCESS);
+
+	return(thr);
+}
diff --git a/storage/xtradb/row/row0row.c b/storage/xtradb/row/row0row.c
new file mode 100644
index 00000000000..4343ee2b009
--- /dev/null
+++ b/storage/xtradb/row/row0row.c
@@ -0,0 +1,1241 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+General row routines
+
+Created 4/20/1996 Heikki Tuuri
+*******************************************************/
+
+#include "row0row.h"
+
+#ifdef UNIV_NONINL
+#include "row0row.ic"
+#endif
+
+#include "data0type.h"
+#include "dict0dict.h"
+#include "btr0btr.h"
+#include "ha_prototypes.h"
+#include "mach0data.h"
+#include "trx0rseg.h"
+#include "trx0trx.h"
+#include "trx0roll.h"
+#include "trx0undo.h"
+#include "trx0purge.h"
+#include "trx0rec.h"
+#include "que0que.h"
+#include "row0ext.h"
+#include "row0upd.h"
+#include "rem0cmp.h"
+#include "read0read.h"
+#include "ut0mem.h"
+
+/*************************************************************************
+Gets the offset of trx id field, in bytes relative to the origin of
+a clustered index record. */
+UNIV_INTERN
+ulint
+row_get_trx_id_offset(
+/*==================*/
+				/* out: offset of DATA_TRX_ID */
+	const rec_t*	rec __attribute__((unused)),
+				/* in: record */
+	dict_index_t*	index,	/* in: clustered index */
+	const ulint*	offsets)/* in: rec_get_offsets(rec, index) */
+{
+	ulint	pos;
+	ulint	offset;
+	ulint	len;
+
+	ut_ad(dict_index_is_clust(index));
+	ut_ad(rec_offs_validate(rec, index, offsets));
+
+	pos = dict_index_get_sys_col_pos(index, DATA_TRX_ID);
+
+	offset = rec_get_nth_field_offs(offsets, pos, &len);
+
+	ut_ad(len == DATA_TRX_ID_LEN);
+
+	return(offset);
+}
+
+/*********************************************************************
+When an insert or purge to a table is performed, this function builds
+the entry to be inserted into or purged from an index on the table. */
+UNIV_INTERN
+dtuple_t*
+row_build_index_entry(
+/*==================*/
+				/* out: index entry which should be
+				inserted or purged, or NULL if the
+				externally stored columns in the
+				clustered index record are unavailable
+				and ext != NULL */
+	const dtuple_t*	row,	/* in: row which should be
+				inserted or purged */
+	row_ext_t*	ext,	/* in: externally stored column prefixes,
+				or NULL */
+	dict_index_t*	index,	/* in: index on the table */
+	mem_heap_t*	heap)	/* in: memory heap from which the memory for
+				the index entry is allocated */
+{
+	dtuple_t*	entry;
+	ulint		entry_len;
+	ulint		i;
+
+	ut_ad(row && index && heap);
+	ut_ad(dtuple_check_typed(row));
+
+	entry_len = dict_index_get_n_fields(index);
+	entry = dtuple_create(heap, entry_len);
+
+	if (UNIV_UNLIKELY(index->type & DICT_UNIVERSAL)) {
+		dtuple_set_n_fields_cmp(entry, entry_len);
+		/* There may only be externally stored columns
+		in a clustered index B-tree of a user table. */
+		ut_a(!ext);
+	} else {
+		dtuple_set_n_fields_cmp(
+			entry, dict_index_get_n_unique_in_tree(index));
+	}
+
+	for (i = 0; i < entry_len; i++) {
+		const dict_field_t*	ind_field
+			= dict_index_get_nth_field(index, i);
+		const dict_col_t*	col
+			= ind_field->col;
+		ulint			col_no
+			= dict_col_get_no(col);
+		dfield_t*		dfield
+			= dtuple_get_nth_field(entry, i);
+		const dfield_t*		dfield2
+			= dtuple_get_nth_field(row, col_no);
+		ulint			len
+			= dfield_get_len(dfield2);
+
+		dfield_copy(dfield, dfield2);
+
+		if (dfield_is_null(dfield) || ind_field->prefix_len == 0) {
+			continue;
+		}
+
+		/* If a column prefix index, take only the prefix.
+		Prefix-indexed columns may be externally stored. */
+		ut_ad(col->ord_part);
+
+		if (UNIV_LIKELY_NULL(ext)) {
+			/* See if the column is stored externally. */
+			const byte*	buf = row_ext_lookup(ext, col_no,
+							     &len);
+			if (UNIV_LIKELY_NULL(buf)) {
+				if (UNIV_UNLIKELY(buf == field_ref_zero)) {
+					return(NULL);
+				}
+				dfield_set_data(dfield, buf, len);
+			}
+		} else if (dfield_is_ext(dfield)) {
+			ut_a(len >= BTR_EXTERN_FIELD_REF_SIZE);
+			len -= BTR_EXTERN_FIELD_REF_SIZE;
+			ut_a(ind_field->prefix_len <= len
+			     || dict_index_is_clust(index));
+		}
+
+		len = dtype_get_at_most_n_mbchars(
+			col->prtype, col->mbminlen, col->mbmaxlen,
+			ind_field->prefix_len, len, dfield_get_data(dfield));
+		dfield_set_len(dfield, len);
+	}
+
+	ut_ad(dtuple_check_typed(entry));
+
+	return(entry);
+}
+
+/***********************************************************************
+An inverse function to row_build_index_entry. Builds a row from a
+record in a clustered index. */
+UNIV_INTERN
+dtuple_t*
+row_build(
+/*======*/
+					/* out, own: row built;
+					see the NOTE below! */
+	ulint			type,	/* in: ROW_COPY_POINTERS or
+					ROW_COPY_DATA; the latter
+					copies also the data fields to
+					heap while the first only
+					places pointers to data fields
+					on the index page, and thus is
+					more efficient */
+	const dict_index_t*	index,	/* in: clustered index */
+	const rec_t*		rec,	/* in: record in the clustered
+					index; NOTE: in the case
+					ROW_COPY_POINTERS the data
+					fields in the row will point
+					directly into this record,
+					therefore, the buffer page of
+					this record must be at least
+					s-latched and the latch held
+					as long as the row dtuple is used! */
+	const ulint*		offsets,/* in: rec_get_offsets(rec,index)
+					or NULL, in which case this function
+					will invoke rec_get_offsets() */
+	const dict_table_t*	col_table,
+					/* in: table, to check which
+					externally stored columns
+					occur in the ordering columns
+					of an index, or NULL if
+					index->table should be
+					consulted instead */
+	row_ext_t**		ext,	/* out, own: cache of
+					externally stored column
+					prefixes, or NULL */
+	mem_heap_t*		heap)	/* in: memory heap from which
+					the memory needed is allocated */
+{
+	dtuple_t*		row;
+	const dict_table_t*	table;
+	ulint			n_fields;
+	ulint			n_ext_cols;
+	ulint*			ext_cols	= NULL; /* remove warning */
+	ulint			len;
+	ulint			row_len;
+	byte*			buf;
+	ulint			i;
+	ulint			j;
+	mem_heap_t*		tmp_heap	= NULL;
+	ulint			offsets_[REC_OFFS_NORMAL_SIZE];
+	rec_offs_init(offsets_);
+
+	ut_ad(index && rec && heap);
+	ut_ad(dict_index_is_clust(index));
+
+	if (!offsets) {
+		offsets = rec_get_offsets(rec, index, offsets_,
+					  ULINT_UNDEFINED, &tmp_heap);
+	} else {
+		ut_ad(rec_offs_validate(rec, index, offsets));
+	}
+
+	if (type != ROW_COPY_POINTERS) {
+		/* Take a copy of rec to heap */
+		buf = mem_heap_alloc(heap, rec_offs_size(offsets));
+		rec = rec_copy(buf, rec, offsets);
+		/* Avoid a debug assertion in rec_offs_validate(). */
+		rec_offs_make_valid(rec, index, (ulint*) offsets);
+	}
+
+	table = index->table;
+	row_len = dict_table_get_n_cols(table);
+
+	row = dtuple_create(heap, row_len);
+
+	dict_table_copy_types(row, table);
+
+	dtuple_set_info_bits(row, rec_get_info_bits(
+				     rec, dict_table_is_comp(table)));
+
+	n_fields = rec_offs_n_fields(offsets);
+	n_ext_cols = rec_offs_n_extern(offsets);
+	if (n_ext_cols) {
+		ext_cols = mem_heap_alloc(heap, n_ext_cols * sizeof *ext_cols);
+	}
+
+	for (i = j = 0; i < n_fields; i++) {
+		dict_field_t*		ind_field
+			= dict_index_get_nth_field(index, i);
+		const dict_col_t*	col
+			= dict_field_get_col(ind_field);
+		ulint			col_no
+			= dict_col_get_no(col);
+		dfield_t*		dfield
+			= dtuple_get_nth_field(row, col_no);
+
+		if (ind_field->prefix_len == 0) {
+
+			const byte*	field = rec_get_nth_field(
+				rec, offsets, i, &len);
+
+			dfield_set_data(dfield, field, len);
+		}
+
+		if (rec_offs_nth_extern(offsets, i)) {
+			dfield_set_ext(dfield);
+
+			if (UNIV_LIKELY_NULL(col_table)) {
+				ut_a(col_no
+				     < dict_table_get_n_cols(col_table));
+				col = dict_table_get_nth_col(
+					col_table, col_no);
+			}
+
+			if (col->ord_part) {
+				/* We will have to fetch prefixes of
+				externally stored columns that are
+				referenced by column prefixes. */
+				ext_cols[j++] = col_no;
+			}
+		}
+	}
+
+	ut_ad(dtuple_check_typed(row));
+
+	if (j) {
+		*ext = row_ext_create(j, ext_cols, row,
+				      dict_table_zip_size(index->table),
+				      heap);
+	} else {
+		*ext = NULL;
+	}
+
+	if (tmp_heap) {
+		mem_heap_free(tmp_heap);
+	}
+
+	return(row);
+}
+
+/***********************************************************************
+Converts an index record to a typed data tuple. */
+UNIV_INTERN
+dtuple_t*
+row_rec_to_index_entry_low(
+/*=======================*/
+					/* out: index entry built; does not
+					set info_bits, and the data fields in
+					the entry will point directly to rec */
+	const rec_t*		rec,	/* in: record in the index */
+	const dict_index_t*	index,	/* in: index */
+	const ulint*		offsets,/* in: rec_get_offsets(rec, index) */
+	ulint*			n_ext,	/* out: number of externally
+					stored columns */
+	mem_heap_t*		heap)	/* in: memory heap from which
+					the memory needed is allocated */
+{
+	dtuple_t*	entry;
+	dfield_t*	dfield;
+	ulint		i;
+	const byte*	field;
+	ulint		len;
+	ulint		rec_len;
+
+	ut_ad(rec && heap && index);
+	/* Because this function may be invoked by row0merge.c
+	on a record whose header is in different format, the check
+	rec_offs_validate(rec, index, offsets) must be avoided here. */
+	ut_ad(n_ext);
+	*n_ext = 0;
+
+	rec_len = rec_offs_n_fields(offsets);
+
+	entry = dtuple_create(heap, rec_len);
+
+	dtuple_set_n_fields_cmp(entry,
+				dict_index_get_n_unique_in_tree(index));
+	ut_ad(rec_len == dict_index_get_n_fields(index));
+
+	dict_index_copy_types(entry, index, rec_len);
+
+	for (i = 0; i < rec_len; i++) {
+
+		dfield = dtuple_get_nth_field(entry, i);
+		field = rec_get_nth_field(rec, offsets, i, &len);
+
+		dfield_set_data(dfield, field, len);
+
+		if (rec_offs_nth_extern(offsets, i)) {
+			dfield_set_ext(dfield);
+			(*n_ext)++;
+		}
+	}
+
+	ut_ad(dtuple_check_typed(entry));
+
+	return(entry);
+}
+
+/***********************************************************************
+Converts an index record to a typed data tuple. NOTE that externally
+stored (often big) fields are NOT copied to heap. */
+UNIV_INTERN
+dtuple_t*
+row_rec_to_index_entry(
+/*===================*/
+					/* out, own: index entry
+					built; see the NOTE below! */
+	ulint			type,	/* in: ROW_COPY_DATA, or
+					ROW_COPY_POINTERS: the former
+					copies also the data fields to
+					heap as the latter only places
+					pointers to data fields on the
+					index page */
+	const rec_t*		rec,	/* in: record in the index;
+					NOTE: in the case
+					ROW_COPY_POINTERS the data
+					fields in the row will point
+					directly into this record,
+					therefore, the buffer page of
+					this record must be at least
+					s-latched and the latch held
+					as long as the dtuple is used! */
+	const dict_index_t*	index,	/* in: index */
+	ulint*			offsets,/* in/out: rec_get_offsets(rec) */
+	ulint*			n_ext,	/* out: number of externally
+					stored columns */
+	mem_heap_t*		heap)	/* in: memory heap from which
+					the memory needed is allocated */
+{
+	dtuple_t*	entry;
+	byte*		buf;
+
+	ut_ad(rec && heap && index);
+	ut_ad(rec_offs_validate(rec, index, offsets));
+
+	if (type == ROW_COPY_DATA) {
+		/* Take a copy of rec to heap */
+		buf = mem_heap_alloc(heap, rec_offs_size(offsets));
+		rec = rec_copy(buf, rec, offsets);
+		/* Avoid a debug assertion in rec_offs_validate(). */
+		rec_offs_make_valid(rec, index, offsets);
+	}
+
+	entry = row_rec_to_index_entry_low(rec, index, offsets, n_ext, heap);
+
+	dtuple_set_info_bits(entry,
+			     rec_get_info_bits(rec, rec_offs_comp(offsets)));
+
+	return(entry);
+}
+
+/***********************************************************************
+Builds from a secondary index record a row reference with which we can
+search the clustered index record. */
+UNIV_INTERN
+dtuple_t*
+row_build_row_ref(
+/*==============*/
+				/* out, own: row reference built; see the
+				NOTE below! */
+	ulint		type,	/* in: ROW_COPY_DATA, or ROW_COPY_POINTERS:
+				the former copies also the data fields to
+				heap, whereas the latter only places pointers
+				to data fields on the index page */
+	dict_index_t*	index,	/* in: secondary index */
+	const rec_t*	rec,	/* in: record in the index;
+				NOTE: in the case ROW_COPY_POINTERS
+				the data fields in the row will point
+				directly into this record, therefore,
+				the buffer page of this record must be
+				at least s-latched and the latch held
+				as long as the row reference is used! */
+	mem_heap_t*	heap)	/* in: memory heap from which the memory
+				needed is allocated */
+{
+	dict_table_t*	table;
+	dict_index_t*	clust_index;
+	dfield_t*	dfield;
+	dtuple_t*	ref;
+	const byte*	field;
+	ulint		len;
+	ulint		ref_len;
+	ulint		pos;
+	byte*		buf;
+	ulint		clust_col_prefix_len;
+	ulint		i;
+	mem_heap_t*	tmp_heap	= NULL;
+	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint*		offsets		= offsets_;
+	rec_offs_init(offsets_);
+
+	ut_ad(index && rec && heap);
+	ut_ad(!dict_index_is_clust(index));
+
+	offsets = rec_get_offsets(rec, index, offsets,
+				  ULINT_UNDEFINED, &tmp_heap);
+	/* Secondary indexes must not contain externally stored columns. */
+	ut_ad(!rec_offs_any_extern(offsets));
+
+	if (type == ROW_COPY_DATA) {
+		/* Take a copy of rec to heap */
+
+		buf = mem_heap_alloc(heap, rec_offs_size(offsets));
+
+		rec = rec_copy(buf, rec, offsets);
+		/* Avoid a debug assertion in rec_offs_validate(). */
+		rec_offs_make_valid(rec, index, offsets);
+	}
+
+	table = index->table;
+
+	clust_index = dict_table_get_first_index(table);
+
+	ref_len = dict_index_get_n_unique(clust_index);
+
+	ref = dtuple_create(heap, ref_len);
+
+	dict_index_copy_types(ref, clust_index, ref_len);
+
+	for (i = 0; i < ref_len; i++) {
+		dfield = dtuple_get_nth_field(ref, i);
+
+		pos = dict_index_get_nth_field_pos(index, clust_index, i);
+
+		ut_a(pos != ULINT_UNDEFINED);
+
+		field = rec_get_nth_field(rec, offsets, pos, &len);
+
+		dfield_set_data(dfield, field, len);
+
+		/* If the primary key contains a column prefix, then the
+		secondary index may contain a longer prefix of the same
+		column, or the full column, and we must adjust the length
+		accordingly. */
+
+		clust_col_prefix_len = dict_index_get_nth_field(
+			clust_index, i)->prefix_len;
+
+		if (clust_col_prefix_len > 0) {
+			if (len != UNIV_SQL_NULL) {
+
+				const dtype_t*	dtype
+					= dfield_get_type(dfield);
+
+				dfield_set_len(dfield,
+					       dtype_get_at_most_n_mbchars(
+						       dtype->prtype,
+						       dtype->mbminlen,
+						       dtype->mbmaxlen,
+						       clust_col_prefix_len,
+						       len, (char*) field));
+			}
+		}
+	}
+
+	ut_ad(dtuple_check_typed(ref));
+	if (tmp_heap) {
+		mem_heap_free(tmp_heap);
+	}
+
+	return(ref);
+}
+
+/***********************************************************************
+Builds from a secondary index record a row reference with which we can
+search the clustered index record. */
+UNIV_INTERN
+void
+row_build_row_ref_in_tuple(
+/*=======================*/
+	dtuple_t*		ref,	/* in/out: row reference built;
+					see the NOTE below! */
+	const rec_t*		rec,	/* in: record in the index;
+					NOTE: the data fields in ref
+					will point directly into this
+					record, therefore, the buffer
+					page of this record must be at
+					least s-latched and the latch
+					held as long as the row
+					reference is used! */
+	const dict_index_t*	index,	/* in: secondary index */
+	ulint*			offsets,/* in: rec_get_offsets(rec, index)
+					or NULL */
+	trx_t*			trx)	/* in: transaction */
+{
+	const dict_index_t*	clust_index;
+	dfield_t*		dfield;
+	const byte*		field;
+	ulint			len;
+	ulint			ref_len;
+	ulint			pos;
+	ulint			clust_col_prefix_len;
+	ulint			i;
+	mem_heap_t*		heap		= NULL;
+	ulint			offsets_[REC_OFFS_NORMAL_SIZE];
+	rec_offs_init(offsets_);
+
+	ut_a(ref);
+	ut_a(index);
+	ut_a(rec);
+	ut_ad(!dict_index_is_clust(index));
+
+	if (UNIV_UNLIKELY(!index->table)) {
+		fputs("InnoDB: table ", stderr);
+notfound:
+		ut_print_name(stderr, trx, TRUE, index->table_name);
+		fputs(" for index ", stderr);
+		ut_print_name(stderr, trx, FALSE, index->name);
+		fputs(" not found\n", stderr);
+		ut_error;
+	}
+
+	clust_index = dict_table_get_first_index(index->table);
+
+	if (UNIV_UNLIKELY(!clust_index)) {
+		fputs("InnoDB: clust index for table ", stderr);
+		goto notfound;
+	}
+
+	if (!offsets) {
+		offsets = rec_get_offsets(rec, index, offsets_,
+					  ULINT_UNDEFINED, &heap);
+	} else {
+		ut_ad(rec_offs_validate(rec, index, offsets));
+	}
+
+	/* Secondary indexes must not contain externally stored columns. */
+	ut_ad(!rec_offs_any_extern(offsets));
+	ref_len = dict_index_get_n_unique(clust_index);
+
+	ut_ad(ref_len == dtuple_get_n_fields(ref));
+
+	dict_index_copy_types(ref, clust_index, ref_len);
+
+	for (i = 0; i < ref_len; i++) {
+		dfield = dtuple_get_nth_field(ref, i);
+
+		pos = dict_index_get_nth_field_pos(index, clust_index, i);
+
+		ut_a(pos != ULINT_UNDEFINED);
+
+		field = rec_get_nth_field(rec, offsets, pos, &len);
+
+		dfield_set_data(dfield, field, len);
+
+		/* If the primary key contains a column prefix, then the
+		secondary index may contain a longer prefix of the same
+		column, or the full column, and we must adjust the length
+		accordingly. */
+
+		clust_col_prefix_len = dict_index_get_nth_field(
+			clust_index, i)->prefix_len;
+
+		if (clust_col_prefix_len > 0) {
+			if (len != UNIV_SQL_NULL) {
+
+				const dtype_t*	dtype
+					= dfield_get_type(dfield);
+
+				dfield_set_len(dfield,
+					       dtype_get_at_most_n_mbchars(
+						       dtype->prtype,
+						       dtype->mbminlen,
+						       dtype->mbmaxlen,
+						       clust_col_prefix_len,
+						       len, (char*) field));
+			}
+		}
+	}
+
+	ut_ad(dtuple_check_typed(ref));
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+}
+
+/***********************************************************************
+From a row build a row reference with which we can search the clustered
+index record. */
+UNIV_INTERN
+void
+row_build_row_ref_from_row(
+/*=======================*/
+	dtuple_t*		ref,	/* in/out: row reference built;
+					see the NOTE below!
+					ref must have the right number
+					of fields! */
+	const dict_table_t*	table,	/* in: table */
+	const dtuple_t*		row)	/* in: row
+					NOTE: the data fields in ref will point
+					directly into data of this row */
+{
+	const dict_index_t*	clust_index;
+	ulint			ref_len;
+	ulint			i;
+
+	ut_ad(ref && table && row);
+
+	clust_index = dict_table_get_first_index(table);
+
+	ref_len = dict_index_get_n_unique(clust_index);
+
+	ut_ad(ref_len == dtuple_get_n_fields(ref));
+
+	for (i = 0; i < ref_len; i++) {
+		const dict_col_t*	col;
+		const dict_field_t*	field;
+		dfield_t*		dfield;
+		const dfield_t*		dfield2;
+
+		dfield = dtuple_get_nth_field(ref, i);
+
+		field = dict_index_get_nth_field(clust_index, i);
+
+		col = dict_field_get_col(field);
+
+		dfield2 = dtuple_get_nth_field(row, dict_col_get_no(col));
+
+		dfield_copy(dfield, dfield2);
+		ut_ad(!dfield_is_ext(dfield));
+
+		if (field->prefix_len > 0 && !dfield_is_null(dfield)) {
+
+			ulint	len = dfield_get_len(dfield);
+
+			len = dtype_get_at_most_n_mbchars(
+				col->prtype, col->mbminlen, col->mbmaxlen,
+				field->prefix_len,
+				len, dfield_get_data(dfield));
+
+			dfield_set_len(dfield, len);
+		}
+	}
+
+	ut_ad(dtuple_check_typed(ref));
+}
+
+/*******************************************************************
+Searches the clustered index record for a row, if we have the row reference. */
+UNIV_INTERN
+ibool
+row_search_on_row_ref(
+/*==================*/
+					/* out: TRUE if found */
+	btr_pcur_t*		pcur,	/* out: persistent cursor, which must
+					be closed by the caller */
+	ulint			mode,	/* in: BTR_MODIFY_LEAF, ... */
+	const dict_table_t*	table,	/* in: table */
+	const dtuple_t*		ref,	/* in: row reference */
+	mtr_t*			mtr)	/* in/out: mtr */
+{
+	ulint		low_match;
+	rec_t*		rec;
+	dict_index_t*	index;
+
+	ut_ad(dtuple_check_typed(ref));
+
+	index = dict_table_get_first_index(table);
+
+	ut_a(dtuple_get_n_fields(ref) == dict_index_get_n_unique(index));
+
+	btr_pcur_open(index, ref, PAGE_CUR_LE, mode, pcur, mtr);
+
+	low_match = btr_pcur_get_low_match(pcur);
+
+	rec = btr_pcur_get_rec(pcur);
+
+	if (page_rec_is_infimum(rec)) {
+
+		return(FALSE);
+	}
+
+	if (low_match != dtuple_get_n_fields(ref)) {
+
+		return(FALSE);
+	}
+
+	return(TRUE);
+}
+
+/*************************************************************************
+Fetches the clustered index record for a secondary index record. The latches
+on the secondary index record are preserved. */
+UNIV_INTERN
+rec_t*
+row_get_clust_rec(
+/*==============*/
+				/* out: record or NULL, if no record found */
+	ulint		mode,	/* in: BTR_MODIFY_LEAF, ... */
+	const rec_t*	rec,	/* in: record in a secondary index */
+	dict_index_t*	index,	/* in: secondary index */
+	dict_index_t**	clust_index,/* out: clustered index */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	mem_heap_t*	heap;
+	dtuple_t*	ref;
+	dict_table_t*	table;
+	btr_pcur_t	pcur;
+	ibool		found;
+	rec_t*		clust_rec;
+
+	ut_ad(!dict_index_is_clust(index));
+
+	table = index->table;
+
+	heap = mem_heap_create(256);
+
+	ref = row_build_row_ref(ROW_COPY_POINTERS, index, rec, heap);
+
+	found = row_search_on_row_ref(&pcur, mode, table, ref, mtr);
+
+	clust_rec = found ? btr_pcur_get_rec(&pcur) : NULL;
+
+	mem_heap_free(heap);
+
+	btr_pcur_close(&pcur);
+
+	*clust_index = dict_table_get_first_index(table);
+
+	return(clust_rec);
+}
+
+/*******************************************************************
+Searches an index record. */
+UNIV_INTERN
+ibool
+row_search_index_entry(
+/*===================*/
+				/* out: TRUE if found */
+	dict_index_t*	index,	/* in: index */
+	const dtuple_t*	entry,	/* in: index entry */
+	ulint		mode,	/* in: BTR_MODIFY_LEAF, ... */
+	btr_pcur_t*	pcur,	/* in/out: persistent cursor, which must
+				be closed by the caller */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	ulint	n_fields;
+	ulint	low_match;
+	rec_t*	rec;
+
+	ut_ad(dtuple_check_typed(entry));
+
+	btr_pcur_open(index, entry, PAGE_CUR_LE, mode, pcur, mtr);
+	low_match = btr_pcur_get_low_match(pcur);
+
+	rec = btr_pcur_get_rec(pcur);
+
+	n_fields = dtuple_get_n_fields(entry);
+
+	return(!page_rec_is_infimum(rec) && low_match == n_fields);
+}
+
+#ifndef UNIV_HOTBACKUP
+
+#include <my_sys.h>
+
+/***********************************************************************
+Formats the raw data in "data" (in InnoDB on-disk format) that is of
+type DATA_INT using "prtype" and writes the result to "buf".
+If the data is in unknown format, then nothing is written to "buf",
+0 is returned and "format_in_hex" is set to TRUE, otherwise
+"format_in_hex" is left untouched.
+Not more than "buf_size" bytes are written to "buf".
+The result is always '\0'-terminated (provided buf_size > 0) and the
+number of bytes that were written to "buf" is returned (including the
+terminating '\0'). */
+static
+ulint
+row_raw_format_int(
+/*===============*/
+					/* out: number of bytes
+					that were written */
+	const char*	data,		/* in: raw data */
+	ulint		data_len,	/* in: raw data length
+					in bytes */
+	ulint		prtype,		/* in: precise type */
+	char*		buf,		/* out: output buffer */
+	ulint		buf_size,	/* in: output buffer size
+					in bytes */
+	ibool*		format_in_hex)	/* out: should the data be
+					formated in hex */
+{
+	ulint	ret;
+
+	if (data_len <= sizeof(ullint)) {
+
+		ullint		value;
+		ibool		unsigned_type = prtype & DATA_UNSIGNED;
+
+		value = mach_read_int_type((const byte*) data,
+					   data_len, unsigned_type);
+
+		if (unsigned_type) {
+
+			ret = ut_snprintf(buf, buf_size, "%llu",
+					  value) + 1;
+		} else {
+
+			ret = ut_snprintf(buf, buf_size, "%lld",
+					  (long long) value) + 1;
+		}
+
+	} else {
+
+		*format_in_hex = TRUE;
+		ret = 0;
+	}
+
+	return(ut_min(ret, buf_size));
+}
+
+/***********************************************************************
+Formats the raw data in "data" (in InnoDB on-disk format) that is of
+type DATA_(CHAR|VARCHAR|MYSQL|VARMYSQL) using "prtype" and writes the
+result to "buf".
+If the data is in binary format, then nothing is written to "buf",
+0 is returned and "format_in_hex" is set to TRUE, otherwise
+"format_in_hex" is left untouched.
+Not more than "buf_size" bytes are written to "buf".
+The result is always '\0'-terminated (provided buf_size > 0) and the
+number of bytes that were written to "buf" is returned (including the
+terminating '\0'). */
+static
+ulint
+row_raw_format_str(
+/*===============*/
+					/* out: number of bytes
+					that were written */
+	const char*	data,		/* in: raw data */
+	ulint		data_len,	/* in: raw data length
+					in bytes */
+	ulint		prtype,		/* in: precise type */
+	char*		buf,		/* out: output buffer */
+	ulint		buf_size,	/* in: output buffer size
+					in bytes */
+	ibool*		format_in_hex)	/* out: should the data be
+					formated in hex */
+{
+	ulint	charset_coll;
+
+	if (buf_size == 0) {
+
+		return(0);
+	}
+
+	/* we assume system_charset_info is UTF-8 */
+
+	charset_coll = dtype_get_charset_coll(prtype);
+
+	if (UNIV_LIKELY(dtype_is_utf8(prtype))) {
+
+		return(ut_str_sql_format(data, data_len, buf, buf_size));
+	}
+	/* else */
+
+	if (charset_coll == DATA_MYSQL_BINARY_CHARSET_COLL) {
+
+		*format_in_hex = TRUE;
+		return(0);
+	}
+	/* else */
+
+	return(innobase_raw_format(data, data_len, charset_coll,
+					  buf, buf_size));
+}
+
+/***********************************************************************
+Formats the raw data in "data" (in InnoDB on-disk format) using
+"dict_field" and writes the result to "buf".
+Not more than "buf_size" bytes are written to "buf".
+The result is always '\0'-terminated (provided buf_size > 0) and the
+number of bytes that were written to "buf" is returned (including the
+terminating '\0'). */
+UNIV_INTERN
+ulint
+row_raw_format(
+/*===========*/
+						/* out: number of bytes
+						that were written */
+	const char*		data,		/* in: raw data */
+	ulint			data_len,	/* in: raw data length
+						in bytes */
+	const dict_field_t*	dict_field,	/* in: index field */
+	char*			buf,		/* out: output buffer */
+	ulint			buf_size)	/* in: output buffer size
+						in bytes */
+{
+	ulint	mtype;
+	ulint	prtype;
+	ulint	ret;
+	ibool	format_in_hex;
+
+	if (buf_size == 0) {
+
+		return(0);
+	}
+
+	if (data_len == UNIV_SQL_NULL) {
+
+		ret = ut_snprintf((char*) buf, buf_size, "NULL") + 1;
+
+		return(ut_min(ret, buf_size));
+	}
+
+	mtype = dict_field->col->mtype;
+	prtype = dict_field->col->prtype;
+
+	format_in_hex = FALSE;
+
+	switch (mtype) {
+	case DATA_INT:
+
+		ret = row_raw_format_int(data, data_len, prtype,
+					 buf, buf_size, &format_in_hex);
+		break;
+	case DATA_CHAR:
+	case DATA_VARCHAR:
+	case DATA_MYSQL:
+	case DATA_VARMYSQL:
+
+		ret = row_raw_format_str(data, data_len, prtype,
+					 buf, buf_size, &format_in_hex);
+		break;
+	/* XXX support more data types */
+	default:
+
+		format_in_hex = TRUE;
+	}
+
+	if (format_in_hex) {
+
+		if (UNIV_LIKELY(buf_size > 2)) {
+
+			memcpy(buf, "0x", 2);
+			buf += 2;
+			buf_size -= 2;
+			ret = 2 + ut_raw_to_hex(data, data_len,
+						buf, buf_size);
+		} else {
+
+			buf[0] = '\0';
+			ret = 1;
+		}
+	}
+
+	return(ret);
+}
+
+#endif /* !UNIV_HOTBACKUP */
+
+#ifdef UNIV_COMPILE_TEST_FUNCS
+
+#include "ut0dbg.h"
+
+void
+test_row_raw_format_int()
+{
+	ulint	ret;
+	char	buf[128];
+	ibool	format_in_hex;
+
+#define CALL_AND_TEST(data, data_len, prtype, buf, buf_size,\
+		      ret_expected, buf_expected, format_in_hex_expected)\
+	do {\
+		ibool	ok = TRUE;\
+		ulint	i;\
+		memset(buf, 'x', 10);\
+		buf[10] = '\0';\
+		format_in_hex = FALSE;\
+		fprintf(stderr, "TESTING \"\\x");\
+		for (i = 0; i < data_len; i++) {\
+			fprintf(stderr, "%02hhX", data[i]);\
+		}\
+		fprintf(stderr, "\", %lu, %lu, %lu\n",\
+                        (ulint) data_len, (ulint) prtype,\
+			(ulint) buf_size);\
+		ret = row_raw_format_int(data, data_len, prtype,\
+					 buf, buf_size, &format_in_hex);\
+		if (ret != ret_expected) {\
+			fprintf(stderr, "expected ret %lu, got %lu\n",\
+				(ulint) ret_expected, ret);\
+			ok = FALSE;\
+                }\
+                if (strcmp((char*) buf, buf_expected) != 0) {\
+                        fprintf(stderr, "expected buf \"%s\", got \"%s\"\n",\
+                                buf_expected, buf);\
+                        ok = FALSE;\
+                }\
+                if (format_in_hex != format_in_hex_expected) {\
+                        fprintf(stderr, "expected format_in_hex %d, got %d\n",\
+                                (int) format_in_hex_expected,\
+				(int) format_in_hex);\
+                        ok = FALSE;\
+                }\
+                if (ok) {\
+                        fprintf(stderr, "OK: %lu, \"%s\" %d\n\n",\
+                                (ulint) ret, buf, (int) format_in_hex);\
+                } else {\
+                        return;\
+                }\
+        } while (0)
+
+#if 1
+	/* min values for signed 1-8 byte integers */
+
+	CALL_AND_TEST("\x00", 1, 0,
+		      buf, sizeof(buf), 5, "-128", 0);
+
+	CALL_AND_TEST("\x00\x00", 2, 0,
+		      buf, sizeof(buf), 7, "-32768", 0);
+
+	CALL_AND_TEST("\x00\x00\x00", 3, 0,
+		      buf, sizeof(buf), 9, "-8388608", 0);
+
+	CALL_AND_TEST("\x00\x00\x00\x00", 4, 0,
+		      buf, sizeof(buf), 12, "-2147483648", 0);
+
+	CALL_AND_TEST("\x00\x00\x00\x00\x00", 5, 0,
+		      buf, sizeof(buf), 14, "-549755813888", 0);
+
+	CALL_AND_TEST("\x00\x00\x00\x00\x00\x00", 6, 0,
+		      buf, sizeof(buf), 17, "-140737488355328", 0);
+
+	CALL_AND_TEST("\x00\x00\x00\x00\x00\x00\x00", 7, 0,
+		      buf, sizeof(buf), 19, "-36028797018963968", 0);
+
+	CALL_AND_TEST("\x00\x00\x00\x00\x00\x00\x00\x00", 8, 0,
+		      buf, sizeof(buf), 21, "-9223372036854775808", 0);
+
+	/* min values for unsigned 1-8 byte integers */
+
+	CALL_AND_TEST("\x00", 1, DATA_UNSIGNED,
+		      buf, sizeof(buf), 2, "0", 0);
+
+	CALL_AND_TEST("\x00\x00", 2, DATA_UNSIGNED,
+		      buf, sizeof(buf), 2, "0", 0);
+
+	CALL_AND_TEST("\x00\x00\x00", 3, DATA_UNSIGNED,
+		      buf, sizeof(buf), 2, "0", 0);
+
+	CALL_AND_TEST("\x00\x00\x00\x00", 4, DATA_UNSIGNED,
+		      buf, sizeof(buf), 2, "0", 0);
+
+	CALL_AND_TEST("\x00\x00\x00\x00\x00", 5, DATA_UNSIGNED,
+		      buf, sizeof(buf), 2, "0", 0);
+
+	CALL_AND_TEST("\x00\x00\x00\x00\x00\x00", 6, DATA_UNSIGNED,
+		      buf, sizeof(buf), 2, "0", 0);
+
+	CALL_AND_TEST("\x00\x00\x00\x00\x00\x00\x00", 7, DATA_UNSIGNED,
+		      buf, sizeof(buf), 2, "0", 0);
+
+	CALL_AND_TEST("\x00\x00\x00\x00\x00\x00\x00\x00", 8, DATA_UNSIGNED,
+		      buf, sizeof(buf), 2, "0", 0);
+
+	/* max values for signed 1-8 byte integers */
+
+	CALL_AND_TEST("\xFF", 1, 0,
+		      buf, sizeof(buf), 4, "127", 0);
+
+	CALL_AND_TEST("\xFF\xFF", 2, 0,
+		      buf, sizeof(buf), 6, "32767", 0);
+
+	CALL_AND_TEST("\xFF\xFF\xFF", 3, 0,
+		      buf, sizeof(buf), 8, "8388607", 0);
+
+	CALL_AND_TEST("\xFF\xFF\xFF\xFF", 4, 0,
+		      buf, sizeof(buf), 11, "2147483647", 0);
+
+	CALL_AND_TEST("\xFF\xFF\xFF\xFF\xFF", 5, 0,
+		      buf, sizeof(buf), 13, "549755813887", 0);
+
+	CALL_AND_TEST("\xFF\xFF\xFF\xFF\xFF\xFF", 6, 0,
+		      buf, sizeof(buf), 16, "140737488355327", 0);
+
+	CALL_AND_TEST("\xFF\xFF\xFF\xFF\xFF\xFF\xFF", 7, 0,
+		      buf, sizeof(buf), 18, "36028797018963967", 0);
+
+	CALL_AND_TEST("\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF", 8, 0,
+		      buf, sizeof(buf), 20, "9223372036854775807", 0);
+
+	/* max values for unsigned 1-8 byte integers */
+
+	CALL_AND_TEST("\xFF", 1, DATA_UNSIGNED,
+		      buf, sizeof(buf), 4, "255", 0);
+
+	CALL_AND_TEST("\xFF\xFF", 2, DATA_UNSIGNED,
+		      buf, sizeof(buf), 6, "65535", 0);
+
+	CALL_AND_TEST("\xFF\xFF\xFF", 3, DATA_UNSIGNED,
+		      buf, sizeof(buf), 9, "16777215", 0);
+
+	CALL_AND_TEST("\xFF\xFF\xFF\xFF", 4, DATA_UNSIGNED,
+		      buf, sizeof(buf), 11, "4294967295", 0);
+
+	CALL_AND_TEST("\xFF\xFF\xFF\xFF\xFF", 5, DATA_UNSIGNED,
+		      buf, sizeof(buf), 14, "1099511627775", 0);
+
+	CALL_AND_TEST("\xFF\xFF\xFF\xFF\xFF\xFF", 6, DATA_UNSIGNED,
+		      buf, sizeof(buf), 16, "281474976710655", 0);
+
+	CALL_AND_TEST("\xFF\xFF\xFF\xFF\xFF\xFF\xFF", 7, DATA_UNSIGNED,
+		      buf, sizeof(buf), 18, "72057594037927935", 0);
+
+	CALL_AND_TEST("\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF", 8, DATA_UNSIGNED,
+		      buf, sizeof(buf), 21, "18446744073709551615", 0);
+
+	/* some random values */
+
+	CALL_AND_TEST("\x52", 1, 0,
+		      buf, sizeof(buf), 4, "-46", 0);
+
+	CALL_AND_TEST("\x0E", 1, DATA_UNSIGNED,
+		      buf, sizeof(buf), 3, "14", 0);
+
+	CALL_AND_TEST("\x62\xCE", 2, 0,
+		      buf, sizeof(buf), 6, "-7474", 0);
+
+	CALL_AND_TEST("\x29\xD6", 2, DATA_UNSIGNED,
+		      buf, sizeof(buf), 6, "10710", 0);
+
+	CALL_AND_TEST("\x7F\xFF\x90", 3, 0,
+		      buf, sizeof(buf), 5, "-112", 0);
+
+	CALL_AND_TEST("\x00\xA1\x16", 3, DATA_UNSIGNED,
+		      buf, sizeof(buf), 6, "41238", 0);
+
+	CALL_AND_TEST("\x7F\xFF\xFF\xF7", 4, 0,
+		      buf, sizeof(buf), 3, "-9", 0);
+
+	CALL_AND_TEST("\x00\x00\x00\x5C", 4, DATA_UNSIGNED,
+		      buf, sizeof(buf), 3, "92", 0);
+
+	CALL_AND_TEST("\x7F\xFF\xFF\xFF\xFF\xFF\xDC\x63", 8, 0,
+		      buf, sizeof(buf), 6, "-9117", 0);
+
+	CALL_AND_TEST("\x00\x00\x00\x00\x00\x01\x64\x62", 8, DATA_UNSIGNED,
+		      buf, sizeof(buf), 6, "91234", 0);
+#endif
+
+	/* speed test */
+
+	speedo_t	speedo;
+	ulint		i;
+
+	speedo_reset(&speedo);
+
+	for (i = 0; i < 1000000; i++) {
+		row_raw_format_int("\x23", 1,
+				   0, buf, sizeof(buf),
+				   &format_in_hex);
+		row_raw_format_int("\x23", 1,
+				   DATA_UNSIGNED, buf, sizeof(buf),
+				   &format_in_hex);
+
+		row_raw_format_int("\x00\x00\x00\x00\x00\x01\x64\x62", 8,
+				   0, buf, sizeof(buf),
+				   &format_in_hex);
+		row_raw_format_int("\x00\x00\x00\x00\x00\x01\x64\x62", 8,
+				   DATA_UNSIGNED, buf, sizeof(buf),
+				   &format_in_hex);
+	}
+
+	speedo_show(&speedo);
+}
+
+#endif /* UNIV_COMPILE_TEST_FUNCS */
diff --git a/storage/xtradb/row/row0sel.c b/storage/xtradb/row/row0sel.c
new file mode 100644
index 00000000000..ebd7bf4a2ce
--- /dev/null
+++ b/storage/xtradb/row/row0sel.c
@@ -0,0 +1,4736 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+Copyright (c) 2008, Google Inc.
+
+Portions of this file contain modifications contributed and copyrighted by
+Google, Inc. Those modifications are gratefully acknowledged and are described
+briefly in the InnoDB documentation. The contributions by Google are
+incorporated with their permission, and subject to the conditions contained in
+the file COPYING.Google.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/*******************************************************
+Select
+
+Created 12/19/1997 Heikki Tuuri
+*******************************************************/
+
+#include "row0sel.h"
+
+#ifdef UNIV_NONINL
+#include "row0sel.ic"
+#endif
+
+#include "dict0dict.h"
+#include "dict0boot.h"
+#include "trx0undo.h"
+#include "trx0trx.h"
+#include "btr0btr.h"
+#include "btr0cur.h"
+#include "btr0sea.h"
+#include "mach0data.h"
+#include "que0que.h"
+#include "row0upd.h"
+#include "row0row.h"
+#include "row0vers.h"
+#include "rem0cmp.h"
+#include "lock0lock.h"
+#include "eval0eval.h"
+#include "pars0sym.h"
+#include "pars0pars.h"
+#include "row0mysql.h"
+#include "read0read.h"
+#include "buf0lru.h"
+#include "ha_prototypes.h"
+
+/* Maximum number of rows to prefetch; MySQL interface has another parameter */
+#define SEL_MAX_N_PREFETCH	16
+
+/* Number of rows fetched, after which to start prefetching; MySQL interface
+has another parameter */
+#define SEL_PREFETCH_LIMIT	1
+
+/* When a select has accessed about this many pages, it returns control back
+to que_run_threads: this is to allow canceling runaway queries */
+
+#define SEL_COST_LIMIT	100
+
+/* Flags for search shortcut */
+#define SEL_FOUND	0
+#define	SEL_EXHAUSTED	1
+#define SEL_RETRY	2
+
+/************************************************************************
+Returns TRUE if the user-defined column in a secondary index record
+is alphabetically the same as the corresponding BLOB column in the clustered
+index record.
+NOTE: the comparison is NOT done as a binary comparison, but character
+fields are compared with collation! */
+static
+ibool
+row_sel_sec_rec_is_for_blob(
+/*========================*/
+					/* out: TRUE if the columns
+					are equal */
+	ulint		mtype,		/* in: main type */
+	ulint		prtype,		/* in: precise type */
+	ulint		mbminlen,	/* in: minimum length of a
+					multi-byte character */
+	ulint		mbmaxlen,	/* in: maximum length of a
+					multi-byte character */
+	const byte*	clust_field,	/* in: the locally stored part of
+					the clustered index column, including
+					the BLOB pointer; the clustered
+					index record must be covered by
+					a lock or a page latch to protect it
+					against deletion (rollback or purge) */
+	ulint		clust_len,	/* in: length of clust_field */
+	const byte*	sec_field,	/* in: column in secondary index */
+	ulint		sec_len,	/* in: length of sec_field */
+	ulint		zip_size)	/* in: compressed page size, or 0 */
+{
+	ulint	len;
+	byte	buf[DICT_MAX_INDEX_COL_LEN];
+
+	len = btr_copy_externally_stored_field_prefix(buf, sizeof buf,
+						      zip_size,
+						      clust_field, clust_len);
+
+	if (UNIV_UNLIKELY(len == 0)) {
+		/* The BLOB was being deleted as the server crashed.
+		There should not be any secondary index records
+		referring to this clustered index record, because
+		btr_free_externally_stored_field() is called after all
+		secondary index entries of the row have been purged. */
+		return(FALSE);
+	}
+
+	len = dtype_get_at_most_n_mbchars(prtype, mbminlen, mbmaxlen,
+					  sec_len, len, (const char*) buf);
+
+	return(!cmp_data_data(mtype, prtype, buf, len, sec_field, sec_len));
+}
+
+/************************************************************************
+Returns TRUE if the user-defined column values in a secondary index record
+are alphabetically the same as the corresponding columns in the clustered
+index record.
+NOTE: the comparison is NOT done as a binary comparison, but character
+fields are compared with collation! */
+static
+ibool
+row_sel_sec_rec_is_for_clust_rec(
+/*=============================*/
+					/* out: TRUE if the secondary
+					record is equal to the corresponding
+					fields in the clustered record,
+					when compared with collation */
+	const rec_t*	sec_rec,	/* in: secondary index record */
+	dict_index_t*	sec_index,	/* in: secondary index */
+	const rec_t*	clust_rec,	/* in: clustered index record;
+					must be protected by a lock or
+					a page latch against deletion
+					in rollback or purge */
+	dict_index_t*	clust_index)	/* in: clustered index */
+{
+	const byte*	sec_field;
+	ulint		sec_len;
+	const byte*	clust_field;
+	ulint		n;
+	ulint		i;
+	mem_heap_t*	heap		= NULL;
+	ulint		clust_offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint		sec_offsets_[REC_OFFS_SMALL_SIZE];
+	ulint*		clust_offs	= clust_offsets_;
+	ulint*		sec_offs	= sec_offsets_;
+	ibool		is_equal	= TRUE;
+
+	rec_offs_init(clust_offsets_);
+	rec_offs_init(sec_offsets_);
+
+	if (rec_get_deleted_flag(clust_rec,
+				 dict_table_is_comp(clust_index->table))) {
+
+		/* The clustered index record is delete-marked;
+		it is not visible in the read view.  Besides,
+		if there are any externally stored columns,
+		some of them may have already been purged. */
+		return(FALSE);
+	}
+
+	clust_offs = rec_get_offsets(clust_rec, clust_index, clust_offs,
+				     ULINT_UNDEFINED, &heap);
+	sec_offs = rec_get_offsets(sec_rec, sec_index, sec_offs,
+				   ULINT_UNDEFINED, &heap);
+
+	n = dict_index_get_n_ordering_defined_by_user(sec_index);
+
+	for (i = 0; i < n; i++) {
+		const dict_field_t*	ifield;
+		const dict_col_t*	col;
+		ulint			clust_pos;
+		ulint			clust_len;
+		ulint			len;
+
+		ifield = dict_index_get_nth_field(sec_index, i);
+		col = dict_field_get_col(ifield);
+		clust_pos = dict_col_get_clust_pos(col, clust_index);
+
+		clust_field = rec_get_nth_field(
+			clust_rec, clust_offs, clust_pos, &clust_len);
+		sec_field = rec_get_nth_field(sec_rec, sec_offs, i, &sec_len);
+
+		len = clust_len;
+
+		if (ifield->prefix_len > 0 && len != UNIV_SQL_NULL) {
+
+			if (rec_offs_nth_extern(clust_offs, clust_pos)) {
+				len -= BTR_EXTERN_FIELD_REF_SIZE;
+			}
+
+			len = dtype_get_at_most_n_mbchars(
+				col->prtype, col->mbminlen, col->mbmaxlen,
+				ifield->prefix_len, len, (char*) clust_field);
+
+			if (rec_offs_nth_extern(clust_offs, clust_pos)
+			    && len < sec_len) {
+				if (!row_sel_sec_rec_is_for_blob(
+					    col->mtype, col->prtype,
+					    col->mbminlen, col->mbmaxlen,
+					    clust_field, clust_len,
+					    sec_field, sec_len,
+					    dict_table_zip_size(
+						    clust_index->table))) {
+					goto inequal;
+				}
+
+				continue;
+			}
+		}
+
+		if (0 != cmp_data_data(col->mtype, col->prtype,
+				       clust_field, len,
+				       sec_field, sec_len)) {
+inequal:
+			is_equal = FALSE;
+			goto func_exit;
+		}
+	}
+
+func_exit:
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+	return(is_equal);
+}
+
+/*************************************************************************
+Creates a select node struct. */
+UNIV_INTERN
+sel_node_t*
+sel_node_create(
+/*============*/
+				/* out, own: select node struct */
+	mem_heap_t*	heap)	/* in: memory heap where created */
+{
+	sel_node_t*	node;
+
+	node = mem_heap_alloc(heap, sizeof(sel_node_t));
+	node->common.type = QUE_NODE_SELECT;
+	node->state = SEL_NODE_OPEN;
+
+	node->plans = NULL;
+
+	return(node);
+}
+
+/*************************************************************************
+Frees the memory private to a select node when a query graph is freed,
+does not free the heap where the node was originally created. */
+UNIV_INTERN
+void
+sel_node_free_private(
+/*==================*/
+	sel_node_t*	node)	/* in: select node struct */
+{
+	ulint	i;
+	plan_t*	plan;
+
+	if (node->plans != NULL) {
+		for (i = 0; i < node->n_tables; i++) {
+			plan = sel_node_get_nth_plan(node, i);
+
+			btr_pcur_close(&(plan->pcur));
+			btr_pcur_close(&(plan->clust_pcur));
+
+			if (plan->old_vers_heap) {
+				mem_heap_free(plan->old_vers_heap);
+			}
+		}
+	}
+}
+
+/*************************************************************************
+Evaluates the values in a select list. If there are aggregate functions,
+their argument value is added to the aggregate total. */
+UNIV_INLINE
+void
+sel_eval_select_list(
+/*=================*/
+	sel_node_t*	node)	/* in: select node */
+{
+	que_node_t*	exp;
+
+	exp = node->select_list;
+
+	while (exp) {
+		eval_exp(exp);
+
+		exp = que_node_get_next(exp);
+	}
+}
+
+/*************************************************************************
+Assigns the values in the select list to the possible into-variables in
+SELECT ... INTO ... */
+UNIV_INLINE
+void
+sel_assign_into_var_values(
+/*=======================*/
+	sym_node_t*	var,	/* in: first variable in a list of variables */
+	sel_node_t*	node)	/* in: select node */
+{
+	que_node_t*	exp;
+
+	if (var == NULL) {
+
+		return;
+	}
+
+	exp = node->select_list;
+
+	while (var) {
+		ut_ad(exp);
+
+		eval_node_copy_val(var->alias, exp);
+
+		exp = que_node_get_next(exp);
+		var = que_node_get_next(var);
+	}
+}
+
+/*************************************************************************
+Resets the aggregate value totals in the select list of an aggregate type
+query. */
+UNIV_INLINE
+void
+sel_reset_aggregate_vals(
+/*=====================*/
+	sel_node_t*	node)	/* in: select node */
+{
+	func_node_t*	func_node;
+
+	ut_ad(node->is_aggregate);
+
+	func_node = node->select_list;
+
+	while (func_node) {
+		eval_node_set_int_val(func_node, 0);
+
+		func_node = que_node_get_next(func_node);
+	}
+
+	node->aggregate_already_fetched = FALSE;
+}
+
+/*************************************************************************
+Copies the input variable values when an explicit cursor is opened. */
+UNIV_INLINE
+void
+row_sel_copy_input_variable_vals(
+/*=============================*/
+	sel_node_t*	node)	/* in: select node */
+{
+	sym_node_t*	var;
+
+	var = UT_LIST_GET_FIRST(node->copy_variables);
+
+	while (var) {
+		eval_node_copy_val(var, var->alias);
+
+		var->indirection = NULL;
+
+		var = UT_LIST_GET_NEXT(col_var_list, var);
+	}
+}
+
+/*************************************************************************
+Fetches the column values from a record. */
+static
+void
+row_sel_fetch_columns(
+/*==================*/
+	dict_index_t*	index,	/* in: record index */
+	const rec_t*	rec,	/* in: record in a clustered or non-clustered
+				index; must be protected by a page latch */
+	const ulint*	offsets,/* in: rec_get_offsets(rec, index) */
+	sym_node_t*	column)	/* in: first column in a column list, or
+				NULL */
+{
+	dfield_t*	val;
+	ulint		index_type;
+	ulint		field_no;
+	const byte*	data;
+	ulint		len;
+
+	ut_ad(rec_offs_validate(rec, index, offsets));
+
+	if (dict_index_is_clust(index)) {
+		index_type = SYM_CLUST_FIELD_NO;
+	} else {
+		index_type = SYM_SEC_FIELD_NO;
+	}
+
+	while (column) {
+		mem_heap_t*	heap = NULL;
+		ibool		needs_copy;
+
+		field_no = column->field_nos[index_type];
+
+		if (field_no != ULINT_UNDEFINED) {
+
+			if (UNIV_UNLIKELY(rec_offs_nth_extern(offsets,
+							      field_no))) {
+
+				/* Copy an externally stored field to the
+				temporary heap */
+
+				heap = mem_heap_create(1);
+
+				data = btr_rec_copy_externally_stored_field(
+					rec, offsets,
+					dict_table_zip_size(index->table),
+					field_no, &len, heap);
+
+				ut_a(len != UNIV_SQL_NULL);
+
+				needs_copy = TRUE;
+			} else {
+				data = rec_get_nth_field(rec, offsets,
+							 field_no, &len);
+
+				if (len == UNIV_SQL_NULL) {
+					len = UNIV_SQL_NULL;
+				}
+
+				needs_copy = column->copy_val;
+			}
+
+			if (needs_copy) {
+				eval_node_copy_and_alloc_val(column, data,
+							     len);
+			} else {
+				val = que_node_get_val(column);
+				dfield_set_data(val, data, len);
+			}
+
+			if (UNIV_LIKELY_NULL(heap)) {
+				mem_heap_free(heap);
+			}
+		}
+
+		column = UT_LIST_GET_NEXT(col_var_list, column);
+	}
+}
+
+/*************************************************************************
+Allocates a prefetch buffer for a column when prefetch is first time done. */
+static
+void
+sel_col_prefetch_buf_alloc(
+/*=======================*/
+	sym_node_t*	column)	/* in: symbol table node for a column */
+{
+	sel_buf_t*	sel_buf;
+	ulint		i;
+
+	ut_ad(que_node_get_type(column) == QUE_NODE_SYMBOL);
+
+	column->prefetch_buf = mem_alloc(SEL_MAX_N_PREFETCH
+					 * sizeof(sel_buf_t));
+	for (i = 0; i < SEL_MAX_N_PREFETCH; i++) {
+		sel_buf = column->prefetch_buf + i;
+
+		sel_buf->data = NULL;
+
+		sel_buf->val_buf_size = 0;
+	}
+}
+
+/*************************************************************************
+Frees a prefetch buffer for a column, including the dynamically allocated
+memory for data stored there. */
+UNIV_INTERN
+void
+sel_col_prefetch_buf_free(
+/*======================*/
+	sel_buf_t*	prefetch_buf)	/* in, own: prefetch buffer */
+{
+	sel_buf_t*	sel_buf;
+	ulint		i;
+
+	for (i = 0; i < SEL_MAX_N_PREFETCH; i++) {
+		sel_buf = prefetch_buf + i;
+
+		if (sel_buf->val_buf_size > 0) {
+
+			mem_free(sel_buf->data);
+		}
+	}
+}
+
+/*************************************************************************
+Pops the column values for a prefetched, cached row from the column prefetch
+buffers and places them to the val fields in the column nodes. */
+static
+void
+sel_pop_prefetched_row(
+/*===================*/
+	plan_t*	plan)	/* in: plan node for a table */
+{
+	sym_node_t*	column;
+	sel_buf_t*	sel_buf;
+	dfield_t*	val;
+	byte*		data;
+	ulint		len;
+	ulint		val_buf_size;
+
+	ut_ad(plan->n_rows_prefetched > 0);
+
+	column = UT_LIST_GET_FIRST(plan->columns);
+
+	while (column) {
+		val = que_node_get_val(column);
+
+		if (!column->copy_val) {
+			/* We did not really push any value for the
+			column */
+
+			ut_ad(!column->prefetch_buf);
+			ut_ad(que_node_get_val_buf_size(column) == 0);
+			ut_d(dfield_set_null(val));
+
+			goto next_col;
+		}
+
+		ut_ad(column->prefetch_buf);
+		ut_ad(!dfield_is_ext(val));
+
+		sel_buf = column->prefetch_buf + plan->first_prefetched;
+
+		data = sel_buf->data;
+		len = sel_buf->len;
+		val_buf_size = sel_buf->val_buf_size;
+
+		/* We must keep track of the allocated memory for
+		column values to be able to free it later: therefore
+		we swap the values for sel_buf and val */
+
+		sel_buf->data = dfield_get_data(val);
+		sel_buf->len = dfield_get_len(val);
+		sel_buf->val_buf_size = que_node_get_val_buf_size(column);
+
+		dfield_set_data(val, data, len);
+		que_node_set_val_buf_size(column, val_buf_size);
+next_col:
+		column = UT_LIST_GET_NEXT(col_var_list, column);
+	}
+
+	plan->n_rows_prefetched--;
+
+	plan->first_prefetched++;
+}
+
+/*************************************************************************
+Pushes the column values for a prefetched, cached row to the column prefetch
+buffers from the val fields in the column nodes. */
+UNIV_INLINE
+void
+sel_push_prefetched_row(
+/*====================*/
+	plan_t*	plan)	/* in: plan node for a table */
+{
+	sym_node_t*	column;
+	sel_buf_t*	sel_buf;
+	dfield_t*	val;
+	byte*		data;
+	ulint		len;
+	ulint		pos;
+	ulint		val_buf_size;
+
+	if (plan->n_rows_prefetched == 0) {
+		pos = 0;
+		plan->first_prefetched = 0;
+	} else {
+		pos = plan->n_rows_prefetched;
+
+		/* We have the convention that pushing new rows starts only
+		after the prefetch stack has been emptied: */
+
+		ut_ad(plan->first_prefetched == 0);
+	}
+
+	plan->n_rows_prefetched++;
+
+	ut_ad(pos < SEL_MAX_N_PREFETCH);
+
+	column = UT_LIST_GET_FIRST(plan->columns);
+
+	while (column) {
+		if (!column->copy_val) {
+			/* There is no sense to push pointers to database
+			page fields when we do not keep latch on the page! */
+
+			goto next_col;
+		}
+
+		if (!column->prefetch_buf) {
+			/* Allocate a new prefetch buffer */
+
+			sel_col_prefetch_buf_alloc(column);
+		}
+
+		sel_buf = column->prefetch_buf + pos;
+
+		val = que_node_get_val(column);
+
+		data = dfield_get_data(val);
+		len = dfield_get_len(val);
+		val_buf_size = que_node_get_val_buf_size(column);
+
+		/* We must keep track of the allocated memory for
+		column values to be able to free it later: therefore
+		we swap the values for sel_buf and val */
+
+		dfield_set_data(val, sel_buf->data, sel_buf->len);
+		que_node_set_val_buf_size(column, sel_buf->val_buf_size);
+
+		sel_buf->data = data;
+		sel_buf->len = len;
+		sel_buf->val_buf_size = val_buf_size;
+next_col:
+		column = UT_LIST_GET_NEXT(col_var_list, column);
+	}
+}
+
+/*************************************************************************
+Builds a previous version of a clustered index record for a consistent read */
+static
+ulint
+row_sel_build_prev_vers(
+/*====================*/
+					/* out: DB_SUCCESS or error code */
+	read_view_t*	read_view,	/* in: read view */
+	dict_index_t*	index,		/* in: plan node for table */
+	rec_t*		rec,		/* in: record in a clustered index */
+	ulint**		offsets,	/* in/out: offsets returned by
+					rec_get_offsets(rec, plan->index) */
+	mem_heap_t**	offset_heap,	/* in/out: memory heap from which
+					the offsets are allocated */
+	mem_heap_t**    old_vers_heap,  /* out: old version heap to use */
+	rec_t**		old_vers,	/* out: old version, or NULL if the
+					record does not exist in the view:
+					i.e., it was freshly inserted
+					afterwards */
+	mtr_t*		mtr)		/* in: mtr */
+{
+	ulint	err;
+
+	if (*old_vers_heap) {
+		mem_heap_empty(*old_vers_heap);
+	} else {
+		*old_vers_heap = mem_heap_create(512);
+	}
+
+	err = row_vers_build_for_consistent_read(
+		rec, mtr, index, offsets, read_view, offset_heap,
+		*old_vers_heap, old_vers);
+	return(err);
+}
+
+/*************************************************************************
+Builds the last committed version of a clustered index record for a
+semi-consistent read. */
+static
+ulint
+row_sel_build_committed_vers_for_mysql(
+/*===================================*/
+					/* out: DB_SUCCESS or error code */
+	dict_index_t*	clust_index,	/* in: clustered index */
+	row_prebuilt_t*	prebuilt,	/* in: prebuilt struct */
+	const rec_t*	rec,		/* in: record in a clustered index */
+	ulint**		offsets,	/* in/out: offsets returned by
+					rec_get_offsets(rec, clust_index) */
+	mem_heap_t**	offset_heap,	/* in/out: memory heap from which
+					the offsets are allocated */
+	const rec_t**	old_vers,	/* out: old version, or NULL if the
+					record does not exist in the view:
+					i.e., it was freshly inserted
+					afterwards */
+	mtr_t*		mtr)		/* in: mtr */
+{
+	ulint	err;
+
+	if (prebuilt->old_vers_heap) {
+		mem_heap_empty(prebuilt->old_vers_heap);
+	} else {
+		prebuilt->old_vers_heap = mem_heap_create(200);
+	}
+
+	err = row_vers_build_for_semi_consistent_read(
+		rec, mtr, clust_index, offsets, offset_heap,
+		prebuilt->old_vers_heap, old_vers);
+	return(err);
+}
+
+/*************************************************************************
+Tests the conditions which determine when the index segment we are searching
+through has been exhausted. */
+UNIV_INLINE
+ibool
+row_sel_test_end_conds(
+/*===================*/
+			/* out: TRUE if row passed the tests */
+	plan_t*	plan)	/* in: plan for the table; the column values must
+			already have been retrieved and the right sides of
+			comparisons evaluated */
+{
+	func_node_t*	cond;
+
+	/* All conditions in end_conds are comparisons of a column to an
+	expression */
+
+	cond = UT_LIST_GET_FIRST(plan->end_conds);
+
+	while (cond) {
+		/* Evaluate the left side of the comparison, i.e., get the
+		column value if there is an indirection */
+
+		eval_sym(cond->args);
+
+		/* Do the comparison */
+
+		if (!eval_cmp(cond)) {
+
+			return(FALSE);
+		}
+
+		cond = UT_LIST_GET_NEXT(cond_list, cond);
+	}
+
+	return(TRUE);
+}
+
+/*************************************************************************
+Tests the other conditions. */
+UNIV_INLINE
+ibool
+row_sel_test_other_conds(
+/*=====================*/
+			/* out: TRUE if row passed the tests */
+	plan_t*	plan)	/* in: plan for the table; the column values must
+			already have been retrieved */
+{
+	func_node_t*	cond;
+
+	cond = UT_LIST_GET_FIRST(plan->other_conds);
+
+	while (cond) {
+		eval_exp(cond);
+
+		if (!eval_node_get_ibool_val(cond)) {
+
+			return(FALSE);
+		}
+
+		cond = UT_LIST_GET_NEXT(cond_list, cond);
+	}
+
+	return(TRUE);
+}
+
+/*************************************************************************
+Retrieves the clustered index record corresponding to a record in a
+non-clustered index. Does the necessary locking. */
+static
+ulint
+row_sel_get_clust_rec(
+/*==================*/
+				/* out: DB_SUCCESS or error code */
+	sel_node_t*	node,	/* in: select_node */
+	plan_t*		plan,	/* in: plan node for table */
+	rec_t*		rec,	/* in: record in a non-clustered index */
+	que_thr_t*	thr,	/* in: query thread */
+	rec_t**		out_rec,/* out: clustered record or an old version of
+				it, NULL if the old version did not exist
+				in the read view, i.e., it was a fresh
+				inserted version */
+	mtr_t*		mtr)	/* in: mtr used to get access to the
+				non-clustered record; the same mtr is used to
+				access the clustered index */
+{
+	dict_index_t*	index;
+	rec_t*		clust_rec;
+	rec_t*		old_vers;
+	ulint		err;
+	mem_heap_t*	heap		= NULL;
+	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint*		offsets		= offsets_;
+	rec_offs_init(offsets_);
+
+	*out_rec = NULL;
+
+	offsets = rec_get_offsets(rec,
+				  btr_pcur_get_btr_cur(&plan->pcur)->index,
+				  offsets, ULINT_UNDEFINED, &heap);
+
+	row_build_row_ref_fast(plan->clust_ref, plan->clust_map, rec, offsets);
+
+	index = dict_table_get_first_index(plan->table);
+
+	btr_pcur_open_with_no_init(index, plan->clust_ref, PAGE_CUR_LE,
+				   BTR_SEARCH_LEAF, &plan->clust_pcur,
+				   0, mtr);
+
+	clust_rec = btr_pcur_get_rec(&(plan->clust_pcur));
+
+	/* Note: only if the search ends up on a non-infimum record is the
+	low_match value the real match to the search tuple */
+
+	if (!page_rec_is_user_rec(clust_rec)
+	    || btr_pcur_get_low_match(&(plan->clust_pcur))
+	    < dict_index_get_n_unique(index)) {
+
+		ut_a(rec_get_deleted_flag(rec,
+					  dict_table_is_comp(plan->table)));
+		ut_a(node->read_view);
+
+		/* In a rare case it is possible that no clust rec is found
+		for a delete-marked secondary index record: if in row0umod.c
+		in row_undo_mod_remove_clust_low() we have already removed
+		the clust rec, while purge is still cleaning and removing
+		secondary index records associated with earlier versions of
+		the clustered index record. In that case we know that the
+		clustered index record did not exist in the read view of
+		trx. */
+
+		goto func_exit;
+	}
+
+	offsets = rec_get_offsets(clust_rec, index, offsets,
+				  ULINT_UNDEFINED, &heap);
+
+	if (!node->read_view) {
+		/* Try to place a lock on the index record */
+
+		/* If innodb_locks_unsafe_for_binlog option is used
+		or this session is using READ COMMITTED isolation level
+		we lock only the record, i.e., next-key locking is
+		not used. */
+		ulint	lock_type;
+		trx_t*	trx;
+
+		trx = thr_get_trx(thr);
+
+		if (srv_locks_unsafe_for_binlog
+		    || trx->isolation_level == TRX_ISO_READ_COMMITTED) {
+			lock_type = LOCK_REC_NOT_GAP;
+		} else {
+			lock_type = LOCK_ORDINARY;
+		}
+
+		err = lock_clust_rec_read_check_and_lock(
+			0, btr_pcur_get_block(&plan->clust_pcur),
+			clust_rec, index, offsets,
+			node->row_lock_mode, lock_type, thr);
+
+		if (err != DB_SUCCESS) {
+
+			goto err_exit;
+		}
+	} else {
+		/* This is a non-locking consistent read: if necessary, fetch
+		a previous version of the record */
+
+		old_vers = NULL;
+
+		if (!lock_clust_rec_cons_read_sees(clust_rec, index, offsets,
+						   node->read_view)) {
+
+			err = row_sel_build_prev_vers(
+				node->read_view, index, clust_rec,
+				&offsets, &heap, &plan->old_vers_heap,
+				&old_vers, mtr);
+
+			if (err != DB_SUCCESS) {
+
+				goto err_exit;
+			}
+
+			clust_rec = old_vers;
+
+			if (clust_rec == NULL) {
+				goto func_exit;
+			}
+		}
+
+		/* If we had to go to an earlier version of row or the
+		secondary index record is delete marked, then it may be that
+		the secondary index record corresponding to clust_rec
+		(or old_vers) is not rec; in that case we must ignore
+		such row because in our snapshot rec would not have existed.
+		Remember that from rec we cannot see directly which transaction
+		id corresponds to it: we have to go to the clustered index
+		record. A query where we want to fetch all rows where
+		the secondary index value is in some interval would return
+		a wrong result if we would not drop rows which we come to
+		visit through secondary index records that would not really
+		exist in our snapshot. */
+
+		if ((old_vers
+		     || rec_get_deleted_flag(rec, dict_table_is_comp(
+						     plan->table)))
+		    && !row_sel_sec_rec_is_for_clust_rec(rec, plan->index,
+							 clust_rec, index)) {
+			goto func_exit;
+		}
+	}
+
+	/* Fetch the columns needed in test conditions.  The clustered
+	index record is protected by a page latch that was acquired
+	when plan->clust_pcur was positioned.  The latch will not be
+	released until mtr_commit(mtr). */
+
+	row_sel_fetch_columns(index, clust_rec, offsets,
+			      UT_LIST_GET_FIRST(plan->columns));
+	*out_rec = clust_rec;
+func_exit:
+	err = DB_SUCCESS;
+err_exit:
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+	return(err);
+}
+
+/*************************************************************************
+Sets a lock on a record. */
+UNIV_INLINE
+ulint
+sel_set_rec_lock(
+/*=============*/
+					/* out: DB_SUCCESS or error code */
+	const buf_block_t*	block,	/* in: buffer block of rec */
+	const rec_t*		rec,	/* in: record */
+	dict_index_t*		index,	/* in: index */
+	const ulint*		offsets,/* in: rec_get_offsets(rec, index) */
+	ulint			mode,	/* in: lock mode */
+	ulint			type,	/* in: LOCK_ORDINARY, LOCK_GAP, or
+					LOC_REC_NOT_GAP */
+	que_thr_t*		thr)	/* in: query thread */
+{
+	trx_t*	trx;
+	ulint	err;
+
+	trx = thr_get_trx(thr);
+
+	if (UT_LIST_GET_LEN(trx->trx_locks) > 10000) {
+		if (buf_LRU_buf_pool_running_out()) {
+
+			return(DB_LOCK_TABLE_FULL);
+		}
+	}
+
+	if (dict_index_is_clust(index)) {
+		err = lock_clust_rec_read_check_and_lock(
+			0, block, rec, index, offsets, mode, type, thr);
+	} else {
+		err = lock_sec_rec_read_check_and_lock(
+			0, block, rec, index, offsets, mode, type, thr);
+	}
+
+	return(err);
+}
+
+/*************************************************************************
+Opens a pcur to a table index. */
+static
+void
+row_sel_open_pcur(
+/*==============*/
+	plan_t*		plan,		/* in: table plan */
+	ibool		search_latch_locked,
+					/* in: TRUE if the thread currently
+					has the search latch locked in
+					s-mode */
+	mtr_t*		mtr)		/* in: mtr */
+{
+	dict_index_t*	index;
+	func_node_t*	cond;
+	que_node_t*	exp;
+	ulint		n_fields;
+	ulint		has_search_latch = 0;	/* RW_S_LATCH or 0 */
+	ulint		i;
+
+	if (search_latch_locked) {
+		has_search_latch = RW_S_LATCH;
+	}
+
+	index = plan->index;
+
+	/* Calculate the value of the search tuple: the exact match columns
+	get their expressions evaluated when we evaluate the right sides of
+	end_conds */
+
+	cond = UT_LIST_GET_FIRST(plan->end_conds);
+
+	while (cond) {
+		eval_exp(que_node_get_next(cond->args));
+
+		cond = UT_LIST_GET_NEXT(cond_list, cond);
+	}
+
+	if (plan->tuple) {
+		n_fields = dtuple_get_n_fields(plan->tuple);
+
+		if (plan->n_exact_match < n_fields) {
+			/* There is a non-exact match field which must be
+			evaluated separately */
+
+			eval_exp(plan->tuple_exps[n_fields - 1]);
+		}
+
+		for (i = 0; i < n_fields; i++) {
+			exp = plan->tuple_exps[i];
+
+			dfield_copy_data(dtuple_get_nth_field(plan->tuple, i),
+					 que_node_get_val(exp));
+		}
+
+		/* Open pcur to the index */
+
+		btr_pcur_open_with_no_init(index, plan->tuple, plan->mode,
+					   BTR_SEARCH_LEAF, &plan->pcur,
+					   has_search_latch, mtr);
+	} else {
+		/* Open the cursor to the start or the end of the index
+		(FALSE: no init) */
+
+		btr_pcur_open_at_index_side(plan->asc, index, BTR_SEARCH_LEAF,
+					    &(plan->pcur), FALSE, mtr);
+	}
+
+	ut_ad(plan->n_rows_prefetched == 0);
+	ut_ad(plan->n_rows_fetched == 0);
+	ut_ad(plan->cursor_at_end == FALSE);
+
+	plan->pcur_is_open = TRUE;
+}
+
+/*************************************************************************
+Restores a stored pcur position to a table index. */
+static
+ibool
+row_sel_restore_pcur_pos(
+/*=====================*/
+				/* out: TRUE if the cursor should be moved to
+				the next record after we return from this
+				function (moved to the previous, in the case
+				of a descending cursor) without processing
+				again the current cursor record */
+	plan_t*		plan,	/* in: table plan */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	ibool	equal_position;
+	ulint	relative_position;
+
+	ut_ad(!plan->cursor_at_end);
+
+	relative_position = btr_pcur_get_rel_pos(&(plan->pcur));
+
+	equal_position = btr_pcur_restore_position(BTR_SEARCH_LEAF,
+						   &(plan->pcur), mtr);
+
+	/* If the cursor is traveling upwards, and relative_position is
+
+	(1) BTR_PCUR_BEFORE: this is not allowed, as we did not have a lock
+	yet on the successor of the page infimum;
+	(2) BTR_PCUR_AFTER: btr_pcur_restore_position placed the cursor on the
+	first record GREATER than the predecessor of a page supremum; we have
+	not yet processed the cursor record: no need to move the cursor to the
+	next record;
+	(3) BTR_PCUR_ON: btr_pcur_restore_position placed the cursor on the
+	last record LESS or EQUAL to the old stored user record; (a) if
+	equal_position is FALSE, this means that the cursor is now on a record
+	less than the old user record, and we must move to the next record;
+	(b) if equal_position is TRUE, then if
+	plan->stored_cursor_rec_processed is TRUE, we must move to the next
+	record, else there is no need to move the cursor. */
+
+	if (plan->asc) {
+		if (relative_position == BTR_PCUR_ON) {
+
+			if (equal_position) {
+
+				return(plan->stored_cursor_rec_processed);
+			}
+
+			return(TRUE);
+		}
+
+		ut_ad(relative_position == BTR_PCUR_AFTER
+		      || relative_position == BTR_PCUR_AFTER_LAST_IN_TREE);
+
+		return(FALSE);
+	}
+
+	/* If the cursor is traveling downwards, and relative_position is
+
+	(1) BTR_PCUR_BEFORE: btr_pcur_restore_position placed the cursor on
+	the last record LESS than the successor of a page infimum; we have not
+	processed the cursor record: no need to move the cursor;
+	(2) BTR_PCUR_AFTER: btr_pcur_restore_position placed the cursor on the
+	first record GREATER than the predecessor of a page supremum; we have
+	processed the cursor record: we should move the cursor to the previous
+	record;
+	(3) BTR_PCUR_ON: btr_pcur_restore_position placed the cursor on the
+	last record LESS or EQUAL to the old stored user record; (a) if
+	equal_position is FALSE, this means that the cursor is now on a record
+	less than the old user record, and we need not move to the previous
+	record; (b) if equal_position is TRUE, then if
+	plan->stored_cursor_rec_processed is TRUE, we must move to the previous
+	record, else there is no need to move the cursor. */
+
+	if (relative_position == BTR_PCUR_BEFORE
+	    || relative_position == BTR_PCUR_BEFORE_FIRST_IN_TREE) {
+
+		return(FALSE);
+	}
+
+	if (relative_position == BTR_PCUR_ON) {
+
+		if (equal_position) {
+
+			return(plan->stored_cursor_rec_processed);
+		}
+
+		return(FALSE);
+	}
+
+	ut_ad(relative_position == BTR_PCUR_AFTER
+	      || relative_position == BTR_PCUR_AFTER_LAST_IN_TREE);
+
+	return(TRUE);
+}
+
+/*************************************************************************
+Resets a plan cursor to a closed state. */
+UNIV_INLINE
+void
+plan_reset_cursor(
+/*==============*/
+	plan_t*	plan)	/* in: plan */
+{
+	plan->pcur_is_open = FALSE;
+	plan->cursor_at_end = FALSE;
+	plan->n_rows_fetched = 0;
+	plan->n_rows_prefetched = 0;
+}
+
+/*************************************************************************
+Tries to do a shortcut to fetch a clustered index record with a unique key,
+using the hash index if possible (not always). */
+static
+ulint
+row_sel_try_search_shortcut(
+/*========================*/
+				/* out: SEL_FOUND, SEL_EXHAUSTED, SEL_RETRY */
+	sel_node_t*	node,	/* in: select node for a consistent read */
+	plan_t*		plan,	/* in: plan for a unique search in clustered
+				index */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	dict_index_t*	index;
+	rec_t*		rec;
+	mem_heap_t*	heap		= NULL;
+	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint*		offsets		= offsets_;
+	ulint		ret;
+	rec_offs_init(offsets_);
+
+	index = plan->index;
+
+	ut_ad(node->read_view);
+	ut_ad(plan->unique_search);
+	ut_ad(!plan->must_get_clust);
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(rw_lock_own(&btr_search_latch, RW_LOCK_SHARED));
+#endif /* UNIV_SYNC_DEBUG */
+
+	row_sel_open_pcur(plan, TRUE, mtr);
+
+	rec = btr_pcur_get_rec(&(plan->pcur));
+
+	if (!page_rec_is_user_rec(rec)) {
+
+		return(SEL_RETRY);
+	}
+
+	ut_ad(plan->mode == PAGE_CUR_GE);
+
+	/* As the cursor is now placed on a user record after a search with
+	the mode PAGE_CUR_GE, the up_match field in the cursor tells how many
+	fields in the user record matched to the search tuple */
+
+	if (btr_pcur_get_up_match(&(plan->pcur)) < plan->n_exact_match) {
+
+		return(SEL_EXHAUSTED);
+	}
+
+	/* This is a non-locking consistent read: if necessary, fetch
+	a previous version of the record */
+
+	offsets = rec_get_offsets(rec, index, offsets, ULINT_UNDEFINED, &heap);
+
+	if (dict_index_is_clust(index)) {
+		if (!lock_clust_rec_cons_read_sees(rec, index, offsets,
+						   node->read_view)) {
+			ret = SEL_RETRY;
+			goto func_exit;
+		}
+	} else if (!lock_sec_rec_cons_read_sees(rec, node->read_view)) {
+
+		ret = SEL_RETRY;
+		goto func_exit;
+	}
+
+	/* Test the deleted flag. */
+
+	if (rec_get_deleted_flag(rec, dict_table_is_comp(plan->table))) {
+
+		ret = SEL_EXHAUSTED;
+		goto func_exit;
+	}
+
+	/* Fetch the columns needed in test conditions.  The index
+	record is protected by a page latch that was acquired when
+	plan->pcur was positioned.  The latch will not be released
+	until mtr_commit(mtr). */
+
+	row_sel_fetch_columns(index, rec, offsets,
+			      UT_LIST_GET_FIRST(plan->columns));
+
+	/* Test the rest of search conditions */
+
+	if (!row_sel_test_other_conds(plan)) {
+
+		ret = SEL_EXHAUSTED;
+		goto func_exit;
+	}
+
+	ut_ad(plan->pcur.latch_mode == BTR_SEARCH_LEAF);
+
+	plan->n_rows_fetched++;
+	ret = SEL_FOUND;
+func_exit:
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+	return(ret);
+}
+
+/*************************************************************************
+Performs a select step. */
+static
+ulint
+row_sel(
+/*====*/
+				/* out: DB_SUCCESS or error code */
+	sel_node_t*	node,	/* in: select node */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	dict_index_t*	index;
+	plan_t*		plan;
+	mtr_t		mtr;
+	ibool		moved;
+	rec_t*		rec;
+	rec_t*		old_vers;
+	rec_t*		clust_rec;
+	ibool		search_latch_locked;
+	ibool		consistent_read;
+
+	/* The following flag becomes TRUE when we are doing a
+	consistent read from a non-clustered index and we must look
+	at the clustered index to find out the previous delete mark
+	state of the non-clustered record: */
+
+	ibool		cons_read_requires_clust_rec	= FALSE;
+	ulint		cost_counter			= 0;
+	ibool		cursor_just_opened;
+	ibool		must_go_to_next;
+	ibool		mtr_has_extra_clust_latch	= FALSE;
+	/* TRUE if the search was made using
+	a non-clustered index, and we had to
+	access the clustered record: now &mtr
+	contains a clustered index latch, and
+	&mtr must be committed before we move
+	to the next non-clustered record */
+	ulint		found_flag;
+	ulint		err;
+	mem_heap_t*	heap				= NULL;
+	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint*		offsets				= offsets_;
+	rec_offs_init(offsets_);
+
+	ut_ad(thr->run_node == node);
+
+	search_latch_locked = FALSE;
+
+	if (node->read_view) {
+		/* In consistent reads, we try to do with the hash index and
+		not to use the buffer page get. This is to reduce memory bus
+		load resulting from semaphore operations. The search latch
+		will be s-locked when we access an index with a unique search
+		condition, but not locked when we access an index with a
+		less selective search condition. */
+
+		consistent_read = TRUE;
+	} else {
+		consistent_read = FALSE;
+	}
+
+table_loop:
+	/* TABLE LOOP
+	----------
+	This is the outer major loop in calculating a join. We come here when
+	node->fetch_table changes, and after adding a row to aggregate totals
+	and, of course, when this function is called. */
+
+	ut_ad(mtr_has_extra_clust_latch == FALSE);
+
+	plan = sel_node_get_nth_plan(node, node->fetch_table);
+	index = plan->index;
+
+	if (plan->n_rows_prefetched > 0) {
+		sel_pop_prefetched_row(plan);
+
+		goto next_table_no_mtr;
+	}
+
+	if (plan->cursor_at_end) {
+		/* The cursor has already reached the result set end: no more
+		rows to process for this table cursor, as also the prefetch
+		stack was empty */
+
+		ut_ad(plan->pcur_is_open);
+
+		goto table_exhausted_no_mtr;
+	}
+
+	/* Open a cursor to index, or restore an open cursor position */
+
+	mtr_start(&mtr);
+
+	if (consistent_read && plan->unique_search && !plan->pcur_is_open
+	    && !plan->must_get_clust
+	    && !plan->table->big_rows) {
+		if (!search_latch_locked) {
+			rw_lock_s_lock(&btr_search_latch);
+
+			search_latch_locked = TRUE;
+		} else if (rw_lock_get_writer(&btr_search_latch) == RW_LOCK_WAIT_EX) {
+
+			/* There is an x-latch request waiting: release the
+			s-latch for a moment; as an s-latch here is often
+			kept for some 10 searches before being released,
+			a waiting x-latch request would block other threads
+			from acquiring an s-latch for a long time, lowering
+			performance significantly in multiprocessors. */
+
+			rw_lock_s_unlock(&btr_search_latch);
+			rw_lock_s_lock(&btr_search_latch);
+		}
+
+		found_flag = row_sel_try_search_shortcut(node, plan, &mtr);
+
+		if (found_flag == SEL_FOUND) {
+
+			goto next_table;
+
+		} else if (found_flag == SEL_EXHAUSTED) {
+
+			goto table_exhausted;
+		}
+
+		ut_ad(found_flag == SEL_RETRY);
+
+		plan_reset_cursor(plan);
+
+		mtr_commit(&mtr);
+		mtr_start(&mtr);
+	}
+
+	if (search_latch_locked) {
+		rw_lock_s_unlock(&btr_search_latch);
+
+		search_latch_locked = FALSE;
+	}
+
+	if (!plan->pcur_is_open) {
+		/* Evaluate the expressions to build the search tuple and
+		open the cursor */
+
+		row_sel_open_pcur(plan, search_latch_locked, &mtr);
+
+		cursor_just_opened = TRUE;
+
+		/* A new search was made: increment the cost counter */
+		cost_counter++;
+	} else {
+		/* Restore pcur position to the index */
+
+		must_go_to_next = row_sel_restore_pcur_pos(plan, &mtr);
+
+		cursor_just_opened = FALSE;
+
+		if (must_go_to_next) {
+			/* We have already processed the cursor record: move
+			to the next */
+
+			goto next_rec;
+		}
+	}
+
+rec_loop:
+	/* RECORD LOOP
+	-----------
+	In this loop we use pcur and try to fetch a qualifying row, and
+	also fill the prefetch buffer for this table if n_rows_fetched has
+	exceeded a threshold. While we are inside this loop, the following
+	holds:
+	(1) &mtr is started,
+	(2) pcur is positioned and open.
+
+	NOTE that if cursor_just_opened is TRUE here, it means that we came
+	to this point right after row_sel_open_pcur. */
+
+	ut_ad(mtr_has_extra_clust_latch == FALSE);
+
+	rec = btr_pcur_get_rec(&(plan->pcur));
+
+	/* PHASE 1: Set a lock if specified */
+
+	if (!node->asc && cursor_just_opened
+	    && !page_rec_is_supremum(rec)) {
+
+		/* When we open a cursor for a descending search, we must set
+		a next-key lock on the successor record: otherwise it would
+		be possible to insert new records next to the cursor position,
+		and it might be that these new records should appear in the
+		search result set, resulting in the phantom problem. */
+
+		if (!consistent_read) {
+
+			/* If innodb_locks_unsafe_for_binlog option is used
+			or this session is using READ COMMITTED isolation
+			level, we lock only the record, i.e., next-key
+			locking is not used. */
+
+			rec_t*	next_rec = page_rec_get_next(rec);
+			ulint	lock_type;
+			trx_t*	trx;
+
+			trx = thr_get_trx(thr);
+
+			offsets = rec_get_offsets(next_rec, index, offsets,
+						  ULINT_UNDEFINED, &heap);
+
+			if (srv_locks_unsafe_for_binlog
+			    || trx->isolation_level
+			    == TRX_ISO_READ_COMMITTED) {
+
+				if (page_rec_is_supremum(next_rec)) {
+
+					goto skip_lock;
+				}
+
+				lock_type = LOCK_REC_NOT_GAP;
+			} else {
+				lock_type = LOCK_ORDINARY;
+			}
+
+			err = sel_set_rec_lock(btr_pcur_get_block(&plan->pcur),
+					       next_rec, index, offsets,
+					       node->row_lock_mode,
+					       lock_type, thr);
+
+			if (err != DB_SUCCESS) {
+				/* Note that in this case we will store in pcur
+				the PREDECESSOR of the record we are waiting
+				the lock for */
+
+				goto lock_wait_or_error;
+			}
+		}
+	}
+
+skip_lock:
+	if (page_rec_is_infimum(rec)) {
+
+		/* The infimum record on a page cannot be in the result set,
+		and neither can a record lock be placed on it: we skip such
+		a record. We also increment the cost counter as we may have
+		processed yet another page of index. */
+
+		cost_counter++;
+
+		goto next_rec;
+	}
+
+	if (!consistent_read) {
+		/* Try to place a lock on the index record */
+
+		/* If innodb_locks_unsafe_for_binlog option is used
+		or this session is using READ COMMITTED isolation level,
+		we lock only the record, i.e., next-key locking is
+		not used. */
+
+		ulint	lock_type;
+		trx_t*	trx;
+
+		offsets = rec_get_offsets(rec, index, offsets,
+					  ULINT_UNDEFINED, &heap);
+
+		trx = thr_get_trx(thr);
+
+		if (srv_locks_unsafe_for_binlog
+		    || trx->isolation_level == TRX_ISO_READ_COMMITTED) {
+
+			if (page_rec_is_supremum(rec)) {
+
+				goto next_rec;
+			}
+
+			lock_type = LOCK_REC_NOT_GAP;
+		} else {
+			lock_type = LOCK_ORDINARY;
+		}
+
+		err = sel_set_rec_lock(btr_pcur_get_block(&plan->pcur),
+				       rec, index, offsets,
+				       node->row_lock_mode, lock_type, thr);
+
+		if (err != DB_SUCCESS) {
+
+			goto lock_wait_or_error;
+		}
+	}
+
+	if (page_rec_is_supremum(rec)) {
+
+		/* A page supremum record cannot be in the result set: skip
+		it now when we have placed a possible lock on it */
+
+		goto next_rec;
+	}
+
+	ut_ad(page_rec_is_user_rec(rec));
+
+	if (cost_counter > SEL_COST_LIMIT) {
+
+		/* Now that we have placed the necessary locks, we can stop
+		for a while and store the cursor position; NOTE that if we
+		would store the cursor position BEFORE placing a record lock,
+		it might happen that the cursor would jump over some records
+		that another transaction could meanwhile insert adjacent to
+		the cursor: this would result in the phantom problem. */
+
+		goto stop_for_a_while;
+	}
+
+	/* PHASE 2: Check a mixed index mix id if needed */
+
+	if (plan->unique_search && cursor_just_opened) {
+
+		ut_ad(plan->mode == PAGE_CUR_GE);
+
+		/* As the cursor is now placed on a user record after a search
+		with the mode PAGE_CUR_GE, the up_match field in the cursor
+		tells how many fields in the user record matched to the search
+		tuple */
+
+		if (btr_pcur_get_up_match(&(plan->pcur))
+		    < plan->n_exact_match) {
+			goto table_exhausted;
+		}
+
+		/* Ok, no need to test end_conds or mix id */
+
+	}
+
+	/* We are ready to look at a possible new index entry in the result
+	set: the cursor is now placed on a user record */
+
+	/* PHASE 3: Get previous version in a consistent read */
+
+	cons_read_requires_clust_rec = FALSE;
+	offsets = rec_get_offsets(rec, index, offsets, ULINT_UNDEFINED, &heap);
+
+	if (consistent_read) {
+		/* This is a non-locking consistent read: if necessary, fetch
+		a previous version of the record */
+
+		if (dict_index_is_clust(index)) {
+
+			if (!lock_clust_rec_cons_read_sees(rec, index, offsets,
+							   node->read_view)) {
+
+				err = row_sel_build_prev_vers(
+					node->read_view, index, rec,
+					&offsets, &heap, &plan->old_vers_heap,
+					&old_vers, &mtr);
+
+				if (err != DB_SUCCESS) {
+
+					goto lock_wait_or_error;
+				}
+
+				if (old_vers == NULL) {
+					offsets = rec_get_offsets(
+						rec, index, offsets,
+						ULINT_UNDEFINED, &heap);
+
+					/* Fetch the columns needed in
+					test conditions. The clustered
+					index record is protected by a
+					page latch that was acquired
+					by row_sel_open_pcur() or
+					row_sel_restore_pcur_pos().
+					The latch will not be released
+					until mtr_commit(mtr). */
+
+					row_sel_fetch_columns(
+						index, rec, offsets,
+						UT_LIST_GET_FIRST(
+							plan->columns));
+
+					if (!row_sel_test_end_conds(plan)) {
+
+						goto table_exhausted;
+					}
+
+					goto next_rec;
+				}
+
+				rec = old_vers;
+			}
+		} else if (!lock_sec_rec_cons_read_sees(rec,
+							node->read_view)) {
+			cons_read_requires_clust_rec = TRUE;
+		}
+	}
+
+	/* PHASE 4: Test search end conditions and deleted flag */
+
+	/* Fetch the columns needed in test conditions.  The record is
+	protected by a page latch that was acquired by
+	row_sel_open_pcur() or row_sel_restore_pcur_pos().  The latch
+	will not be released until mtr_commit(mtr). */
+
+	row_sel_fetch_columns(index, rec, offsets,
+			      UT_LIST_GET_FIRST(plan->columns));
+
+	/* Test the selection end conditions: these can only contain columns
+	which already are found in the index, even though the index might be
+	non-clustered */
+
+	if (plan->unique_search && cursor_just_opened) {
+
+		/* No test necessary: the test was already made above */
+
+	} else if (!row_sel_test_end_conds(plan)) {
+
+		goto table_exhausted;
+	}
+
+	if (rec_get_deleted_flag(rec, dict_table_is_comp(plan->table))
+	    && !cons_read_requires_clust_rec) {
+
+		/* The record is delete marked: we can skip it if this is
+		not a consistent read which might see an earlier version
+		of a non-clustered index record */
+
+		if (plan->unique_search) {
+
+			goto table_exhausted;
+		}
+
+		goto next_rec;
+	}
+
+	/* PHASE 5: Get the clustered index record, if needed and if we did
+	not do the search using the clustered index */
+
+	if (plan->must_get_clust || cons_read_requires_clust_rec) {
+
+		/* It was a non-clustered index and we must fetch also the
+		clustered index record */
+
+		err = row_sel_get_clust_rec(node, plan, rec, thr, &clust_rec,
+					    &mtr);
+		mtr_has_extra_clust_latch = TRUE;
+
+		if (err != DB_SUCCESS) {
+
+			goto lock_wait_or_error;
+		}
+
+		/* Retrieving the clustered record required a search:
+		increment the cost counter */
+
+		cost_counter++;
+
+		if (clust_rec == NULL) {
+			/* The record did not exist in the read view */
+			ut_ad(consistent_read);
+
+			goto next_rec;
+		}
+
+		if (rec_get_deleted_flag(clust_rec,
+					 dict_table_is_comp(plan->table))) {
+
+			/* The record is delete marked: we can skip it */
+
+			goto next_rec;
+		}
+
+		if (node->can_get_updated) {
+
+			btr_pcur_store_position(&(plan->clust_pcur), &mtr);
+		}
+	}
+
+	/* PHASE 6: Test the rest of search conditions */
+
+	if (!row_sel_test_other_conds(plan)) {
+
+		if (plan->unique_search) {
+
+			goto table_exhausted;
+		}
+
+		goto next_rec;
+	}
+
+	/* PHASE 7: We found a new qualifying row for the current table; push
+	the row if prefetch is on, or move to the next table in the join */
+
+	plan->n_rows_fetched++;
+
+	ut_ad(plan->pcur.latch_mode == BTR_SEARCH_LEAF);
+
+	if ((plan->n_rows_fetched <= SEL_PREFETCH_LIMIT)
+	    || plan->unique_search || plan->no_prefetch
+	    || plan->table->big_rows) {
+
+		/* No prefetch in operation: go to the next table */
+
+		goto next_table;
+	}
+
+	sel_push_prefetched_row(plan);
+
+	if (plan->n_rows_prefetched == SEL_MAX_N_PREFETCH) {
+
+		/* The prefetch buffer is now full */
+
+		sel_pop_prefetched_row(plan);
+
+		goto next_table;
+	}
+
+next_rec:
+	ut_ad(!search_latch_locked);
+
+	if (mtr_has_extra_clust_latch) {
+
+		/* We must commit &mtr if we are moving to the next
+		non-clustered index record, because we could break the
+		latching order if we would access a different clustered
+		index page right away without releasing the previous. */
+
+		goto commit_mtr_for_a_while;
+	}
+
+	if (node->asc) {
+		moved = btr_pcur_move_to_next(&(plan->pcur), &mtr);
+	} else {
+		moved = btr_pcur_move_to_prev(&(plan->pcur), &mtr);
+	}
+
+	if (!moved) {
+
+		goto table_exhausted;
+	}
+
+	cursor_just_opened = FALSE;
+
+	/* END OF RECORD LOOP
+	------------------ */
+	goto rec_loop;
+
+next_table:
+	/* We found a record which satisfies the conditions: we can move to
+	the next table or return a row in the result set */
+
+	ut_ad(btr_pcur_is_on_user_rec(&plan->pcur));
+
+	if (plan->unique_search && !node->can_get_updated) {
+
+		plan->cursor_at_end = TRUE;
+	} else {
+		ut_ad(!search_latch_locked);
+
+		plan->stored_cursor_rec_processed = TRUE;
+
+		btr_pcur_store_position(&(plan->pcur), &mtr);
+	}
+
+	mtr_commit(&mtr);
+
+	mtr_has_extra_clust_latch = FALSE;
+
+next_table_no_mtr:
+	/* If we use 'goto' to this label, it means that the row was popped
+	from the prefetched rows stack, and &mtr is already committed */
+
+	if (node->fetch_table + 1 == node->n_tables) {
+
+		sel_eval_select_list(node);
+
+		if (node->is_aggregate) {
+
+			goto table_loop;
+		}
+
+		sel_assign_into_var_values(node->into_list, node);
+
+		thr->run_node = que_node_get_parent(node);
+
+		err = DB_SUCCESS;
+		goto func_exit;
+	}
+
+	node->fetch_table++;
+
+	/* When we move to the next table, we first reset the plan cursor:
+	we do not care about resetting it when we backtrack from a table */
+
+	plan_reset_cursor(sel_node_get_nth_plan(node, node->fetch_table));
+
+	goto table_loop;
+
+table_exhausted:
+	/* The table cursor pcur reached the result set end: backtrack to the
+	previous table in the join if we do not have cached prefetched rows */
+
+	plan->cursor_at_end = TRUE;
+
+	mtr_commit(&mtr);
+
+	mtr_has_extra_clust_latch = FALSE;
+
+	if (plan->n_rows_prefetched > 0) {
+		/* The table became exhausted during a prefetch */
+
+		sel_pop_prefetched_row(plan);
+
+		goto next_table_no_mtr;
+	}
+
+table_exhausted_no_mtr:
+	if (node->fetch_table == 0) {
+		err = DB_SUCCESS;
+
+		if (node->is_aggregate && !node->aggregate_already_fetched) {
+
+			node->aggregate_already_fetched = TRUE;
+
+			sel_assign_into_var_values(node->into_list, node);
+
+			thr->run_node = que_node_get_parent(node);
+		} else {
+			node->state = SEL_NODE_NO_MORE_ROWS;
+
+			thr->run_node = que_node_get_parent(node);
+		}
+
+		goto func_exit;
+	}
+
+	node->fetch_table--;
+
+	goto table_loop;
+
+stop_for_a_while:
+	/* Return control for a while to que_run_threads, so that runaway
+	queries can be canceled. NOTE that when we come here, we must, in a
+	locking read, have placed the necessary (possibly waiting request)
+	record lock on the cursor record or its successor: when we reposition
+	the cursor, this record lock guarantees that nobody can meanwhile have
+	inserted new records which should have appeared in the result set,
+	which would result in the phantom problem. */
+
+	ut_ad(!search_latch_locked);
+
+	plan->stored_cursor_rec_processed = FALSE;
+	btr_pcur_store_position(&(plan->pcur), &mtr);
+
+	mtr_commit(&mtr);
+
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(sync_thread_levels_empty_gen(TRUE));
+#endif /* UNIV_SYNC_DEBUG */
+	err = DB_SUCCESS;
+	goto func_exit;
+
+commit_mtr_for_a_while:
+	/* Stores the cursor position and commits &mtr; this is used if
+	&mtr may contain latches which would break the latching order if
+	&mtr would not be committed and the latches released. */
+
+	plan->stored_cursor_rec_processed = TRUE;
+
+	ut_ad(!search_latch_locked);
+	btr_pcur_store_position(&(plan->pcur), &mtr);
+
+	mtr_commit(&mtr);
+
+	mtr_has_extra_clust_latch = FALSE;
+
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(sync_thread_levels_empty_gen(TRUE));
+#endif /* UNIV_SYNC_DEBUG */
+
+	goto table_loop;
+
+lock_wait_or_error:
+	/* See the note at stop_for_a_while: the same holds for this case */
+
+	ut_ad(!btr_pcur_is_before_first_on_page(&plan->pcur) || !node->asc);
+	ut_ad(!search_latch_locked);
+
+	plan->stored_cursor_rec_processed = FALSE;
+	btr_pcur_store_position(&(plan->pcur), &mtr);
+
+	mtr_commit(&mtr);
+
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(sync_thread_levels_empty_gen(TRUE));
+#endif /* UNIV_SYNC_DEBUG */
+
+func_exit:
+	if (search_latch_locked) {
+		rw_lock_s_unlock(&btr_search_latch);
+	}
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+	return(err);
+}
+
+/**************************************************************************
+Performs a select step. This is a high-level function used in SQL execution
+graphs. */
+UNIV_INTERN
+que_thr_t*
+row_sel_step(
+/*=========*/
+				/* out: query thread to run next or NULL */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	ulint		i_lock_mode;
+	sym_node_t*	table_node;
+	sel_node_t*	node;
+	ulint		err;
+
+	ut_ad(thr);
+
+	node = thr->run_node;
+
+	ut_ad(que_node_get_type(node) == QUE_NODE_SELECT);
+
+	/* If this is a new time this node is executed (or when execution
+	resumes after wait for a table intention lock), set intention locks
+	on the tables, or assign a read view */
+
+	if (node->into_list && (thr->prev_node == que_node_get_parent(node))) {
+
+		node->state = SEL_NODE_OPEN;
+	}
+
+	if (node->state == SEL_NODE_OPEN) {
+
+		/* It may be that the current session has not yet started
+		its transaction, or it has been committed: */
+
+		trx_start_if_not_started(thr_get_trx(thr));
+
+		plan_reset_cursor(sel_node_get_nth_plan(node, 0));
+
+		if (node->consistent_read) {
+			/* Assign a read view for the query */
+			node->read_view = trx_assign_read_view(
+				thr_get_trx(thr));
+		} else {
+			if (node->set_x_locks) {
+				i_lock_mode = LOCK_IX;
+			} else {
+				i_lock_mode = LOCK_IS;
+			}
+
+			table_node = node->table_list;
+
+			while (table_node) {
+				err = lock_table(0, table_node->table,
+						 i_lock_mode, thr);
+				if (err != DB_SUCCESS) {
+					thr_get_trx(thr)->error_state = err;
+
+					return(NULL);
+				}
+
+				table_node = que_node_get_next(table_node);
+			}
+		}
+
+		/* If this is an explicit cursor, copy stored procedure
+		variable values, so that the values cannot change between
+		fetches (currently, we copy them also for non-explicit
+		cursors) */
+
+		if (node->explicit_cursor
+		    && UT_LIST_GET_FIRST(node->copy_variables)) {
+
+			row_sel_copy_input_variable_vals(node);
+		}
+
+		node->state = SEL_NODE_FETCH;
+		node->fetch_table = 0;
+
+		if (node->is_aggregate) {
+			/* Reset the aggregate total values */
+			sel_reset_aggregate_vals(node);
+		}
+	}
+
+	err = row_sel(node, thr);
+
+	/* NOTE! if queries are parallelized, the following assignment may
+	have problems; the assignment should be made only if thr is the
+	only top-level thr in the graph: */
+
+	thr->graph->last_sel_node = node;
+
+	if (err != DB_SUCCESS) {
+		thr_get_trx(thr)->error_state = err;
+
+		return(NULL);
+	}
+
+	return(thr);
+}
+
+/**************************************************************************
+Performs a fetch for a cursor. */
+UNIV_INTERN
+que_thr_t*
+fetch_step(
+/*=======*/
+				/* out: query thread to run next or NULL */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	sel_node_t*	sel_node;
+	fetch_node_t*	node;
+
+	ut_ad(thr);
+
+	node = thr->run_node;
+	sel_node = node->cursor_def;
+
+	ut_ad(que_node_get_type(node) == QUE_NODE_FETCH);
+
+	if (thr->prev_node != que_node_get_parent(node)) {
+
+		if (sel_node->state != SEL_NODE_NO_MORE_ROWS) {
+
+			if (node->into_list) {
+				sel_assign_into_var_values(node->into_list,
+							   sel_node);
+			} else {
+				void* ret = (*node->func->func)(
+					sel_node, node->func->arg);
+
+				if (!ret) {
+					sel_node->state
+						= SEL_NODE_NO_MORE_ROWS;
+				}
+			}
+		}
+
+		thr->run_node = que_node_get_parent(node);
+
+		return(thr);
+	}
+
+	/* Make the fetch node the parent of the cursor definition for
+	the time of the fetch, so that execution knows to return to this
+	fetch node after a row has been selected or we know that there is
+	no row left */
+
+	sel_node->common.parent = node;
+
+	if (sel_node->state == SEL_NODE_CLOSED) {
+		fprintf(stderr,
+			"InnoDB: Error: fetch called on a closed cursor\n");
+
+		thr_get_trx(thr)->error_state = DB_ERROR;
+
+		return(NULL);
+	}
+
+	thr->run_node = sel_node;
+
+	return(thr);
+}
+
+/********************************************************************
+Sample callback function for fetch that prints each row.*/
+UNIV_INTERN
+void*
+row_fetch_print(
+/*============*/
+				/* out: always returns non-NULL */
+	void*	row,		/* in:  sel_node_t* */
+	void*	user_arg)	/* in:  not used */
+{
+	sel_node_t*	node = row;
+	que_node_t*	exp;
+	ulint		i = 0;
+
+	UT_NOT_USED(user_arg);
+
+	fprintf(stderr, "row_fetch_print: row %p\n", row);
+
+	exp = node->select_list;
+
+	while (exp) {
+		dfield_t*	dfield = que_node_get_val(exp);
+		const dtype_t*	type = dfield_get_type(dfield);
+
+		fprintf(stderr, " column %lu:\n", (ulong)i);
+
+		dtype_print(type);
+		putc('\n', stderr);
+
+		if (dfield_get_len(dfield) != UNIV_SQL_NULL) {
+			ut_print_buf(stderr, dfield_get_data(dfield),
+				     dfield_get_len(dfield));
+			putc('\n', stderr);
+		} else {
+			fputs(" <NULL>;\n", stderr);
+		}
+
+		exp = que_node_get_next(exp);
+		i++;
+	}
+
+	return((void*)42);
+}
+
+/********************************************************************
+Callback function for fetch that stores an unsigned 4 byte integer to the
+location pointed. The column's type must be DATA_INT, DATA_UNSIGNED, length
+= 4. */
+UNIV_INTERN
+void*
+row_fetch_store_uint4(
+/*==================*/
+				/* out: always returns NULL */
+	void*	row,		/* in:  sel_node_t* */
+	void*	user_arg)	/* in:  data pointer */
+{
+	sel_node_t*	node = row;
+	ib_uint32_t*	val = user_arg;
+	ulint		tmp;
+
+	dfield_t*	dfield = que_node_get_val(node->select_list);
+	const dtype_t*	type = dfield_get_type(dfield);
+	ulint		len = dfield_get_len(dfield);
+
+	ut_a(dtype_get_mtype(type) == DATA_INT);
+	ut_a(dtype_get_prtype(type) & DATA_UNSIGNED);
+	ut_a(len == 4);
+
+	tmp = mach_read_from_4(dfield_get_data(dfield));
+	*val = (ib_uint32_t) tmp;
+
+	return(NULL);
+}
+
+/***************************************************************
+Prints a row in a select result. */
+UNIV_INTERN
+que_thr_t*
+row_printf_step(
+/*============*/
+				/* out: query thread to run next or NULL */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	row_printf_node_t*	node;
+	sel_node_t*		sel_node;
+	que_node_t*		arg;
+
+	ut_ad(thr);
+
+	node = thr->run_node;
+
+	sel_node = node->sel_node;
+
+	ut_ad(que_node_get_type(node) == QUE_NODE_ROW_PRINTF);
+
+	if (thr->prev_node == que_node_get_parent(node)) {
+
+		/* Reset the cursor */
+		sel_node->state = SEL_NODE_OPEN;
+
+		/* Fetch next row to print */
+
+		thr->run_node = sel_node;
+
+		return(thr);
+	}
+
+	if (sel_node->state != SEL_NODE_FETCH) {
+
+		ut_ad(sel_node->state == SEL_NODE_NO_MORE_ROWS);
+
+		/* No more rows to print */
+
+		thr->run_node = que_node_get_parent(node);
+
+		return(thr);
+	}
+
+	arg = sel_node->select_list;
+
+	while (arg) {
+		dfield_print_also_hex(que_node_get_val(arg));
+
+		fputs(" ::: ", stderr);
+
+		arg = que_node_get_next(arg);
+	}
+
+	putc('\n', stderr);
+
+	/* Fetch next row to print */
+
+	thr->run_node = sel_node;
+
+	return(thr);
+}
+
+/********************************************************************
+Converts a key value stored in MySQL format to an Innobase dtuple. The last
+field of the key value may be just a prefix of a fixed length field: hence
+the parameter key_len. But currently we do not allow search keys where the
+last field is only a prefix of the full key field len and print a warning if
+such appears. A counterpart of this function is
+ha_innobase::store_key_val_for_row() in ha_innodb.cc. */
+UNIV_INTERN
+void
+row_sel_convert_mysql_key_to_innobase(
+/*==================================*/
+	dtuple_t*	tuple,		/* in/out: tuple where to build;
+					NOTE: we assume that the type info
+					in the tuple is already according
+					to index! */
+	byte*		buf,		/* in: buffer to use in field
+					conversions */
+	ulint		buf_len,	/* in: buffer length */
+	dict_index_t*	index,		/* in: index of the key value */
+	const byte*	key_ptr,	/* in: MySQL key value */
+	ulint		key_len,	/* in: MySQL key value length */
+	trx_t*		trx)		/* in: transaction */
+{
+	byte*		original_buf	= buf;
+	const byte*	original_key_ptr = key_ptr;
+	dict_field_t*	field;
+	dfield_t*	dfield;
+	ulint		data_offset;
+	ulint		data_len;
+	ulint		data_field_len;
+	ibool		is_null;
+	const byte*	key_end;
+	ulint		n_fields = 0;
+
+	/* For documentation of the key value storage format in MySQL, see
+	ha_innobase::store_key_val_for_row() in ha_innodb.cc. */
+
+	key_end = key_ptr + key_len;
+
+	/* Permit us to access any field in the tuple (ULINT_MAX): */
+
+	dtuple_set_n_fields(tuple, ULINT_MAX);
+
+	dfield = dtuple_get_nth_field(tuple, 0);
+	field = dict_index_get_nth_field(index, 0);
+
+	if (UNIV_UNLIKELY(dfield_get_type(dfield)->mtype == DATA_SYS)) {
+		/* A special case: we are looking for a position in the
+		generated clustered index which InnoDB automatically added
+		to a table with no primary key: the first and the only
+		ordering column is ROW_ID which InnoDB stored to the key_ptr
+		buffer. */
+
+		ut_a(key_len == DATA_ROW_ID_LEN);
+
+		dfield_set_data(dfield, key_ptr, DATA_ROW_ID_LEN);
+
+		dtuple_set_n_fields(tuple, 1);
+
+		return;
+	}
+
+	while (key_ptr < key_end) {
+
+		ulint	type = dfield_get_type(dfield)->mtype;
+		ut_a(field->col->mtype == type);
+
+		data_offset = 0;
+		is_null = FALSE;
+
+		if (!(dfield_get_type(dfield)->prtype & DATA_NOT_NULL)) {
+			/* The first byte in the field tells if this is
+			an SQL NULL value */
+
+			data_offset = 1;
+
+			if (*key_ptr != 0) {
+				dfield_set_null(dfield);
+
+				is_null = TRUE;
+			}
+		}
+
+		/* Calculate data length and data field total length */
+
+		if (type == DATA_BLOB) {
+			/* The key field is a column prefix of a BLOB or
+			TEXT */
+
+			ut_a(field->prefix_len > 0);
+
+			/* MySQL stores the actual data length to the first 2
+			bytes after the optional SQL NULL marker byte. The
+			storage format is little-endian, that is, the most
+			significant byte at a higher address. In UTF-8, MySQL
+			seems to reserve field->prefix_len bytes for
+			storing this field in the key value buffer, even
+			though the actual value only takes data_len bytes
+			from the start. */
+
+			data_len = key_ptr[data_offset]
+				+ 256 * key_ptr[data_offset + 1];
+			data_field_len = data_offset + 2 + field->prefix_len;
+
+			data_offset += 2;
+
+			/* Now that we know the length, we store the column
+			value like it would be a fixed char field */
+
+		} else if (field->prefix_len > 0) {
+			/* Looks like MySQL pads unused end bytes in the
+			prefix with space. Therefore, also in UTF-8, it is ok
+			to compare with a prefix containing full prefix_len
+			bytes, and no need to take at most prefix_len / 3
+			UTF-8 characters from the start.
+			If the prefix is used as the upper end of a LIKE
+			'abc%' query, then MySQL pads the end with chars
+			0xff. TODO: in that case does it any harm to compare
+			with the full prefix_len bytes. How do characters
+			0xff in UTF-8 behave? */
+
+			data_len = field->prefix_len;
+			data_field_len = data_offset + data_len;
+		} else {
+			data_len = dfield_get_type(dfield)->len;
+			data_field_len = data_offset + data_len;
+		}
+
+		if (UNIV_UNLIKELY
+		    (dtype_get_mysql_type(dfield_get_type(dfield))
+		     == DATA_MYSQL_TRUE_VARCHAR)
+		    && UNIV_LIKELY(type != DATA_INT)) {
+			/* In a MySQL key value format, a true VARCHAR is
+			always preceded by 2 bytes of a length field.
+			dfield_get_type(dfield)->len returns the maximum
+			'payload' len in bytes. That does not include the
+			2 bytes that tell the actual data length.
+
+			We added the check != DATA_INT to make sure we do
+			not treat MySQL ENUM or SET as a true VARCHAR! */
+
+			data_len += 2;
+			data_field_len += 2;
+		}
+
+		/* Storing may use at most data_len bytes of buf */
+
+		if (UNIV_LIKELY(!is_null)) {
+			row_mysql_store_col_in_innobase_format(
+				dfield, buf,
+				FALSE, /* MySQL key value format col */
+				key_ptr + data_offset, data_len,
+				dict_table_is_comp(index->table));
+			buf += data_len;
+		}
+
+		key_ptr += data_field_len;
+
+		if (UNIV_UNLIKELY(key_ptr > key_end)) {
+			/* The last field in key was not a complete key field
+			but a prefix of it.
+
+			Print a warning about this! HA_READ_PREFIX_LAST does
+			not currently work in InnoDB with partial-field key
+			value prefixes. Since MySQL currently uses a padding
+			trick to calculate LIKE 'abc%' type queries there
+			should never be partial-field prefixes in searches. */
+
+			ut_print_timestamp(stderr);
+
+			fputs("  InnoDB: Warning: using a partial-field"
+			      " key prefix in search.\n"
+			      "InnoDB: ", stderr);
+			dict_index_name_print(stderr, trx, index);
+			fprintf(stderr, ". Last data field length %lu bytes,\n"
+				"InnoDB: key ptr now exceeds"
+				" key end by %lu bytes.\n"
+				"InnoDB: Key value in the MySQL format:\n",
+				(ulong) data_field_len,
+				(ulong) (key_ptr - key_end));
+			fflush(stderr);
+			ut_print_buf(stderr, original_key_ptr, key_len);
+			putc('\n', stderr);
+
+			if (!is_null) {
+				ulint	len = dfield_get_len(dfield);
+				dfield_set_len(dfield, len
+					       - (ulint) (key_ptr - key_end));
+			}
+		}
+
+		n_fields++;
+		field++;
+		dfield++;
+	}
+
+	ut_a(buf <= original_buf + buf_len);
+
+	/* We set the length of tuple to n_fields: we assume that the memory
+	area allocated for it is big enough (usually bigger than n_fields). */
+
+	dtuple_set_n_fields(tuple, n_fields);
+}
+
+/******************************************************************
+Stores the row id to the prebuilt struct. */
+static
+void
+row_sel_store_row_id_to_prebuilt(
+/*=============================*/
+	row_prebuilt_t*		prebuilt,	/* in/out: prebuilt */
+	const rec_t*		index_rec,	/* in: record */
+	const dict_index_t*	index,		/* in: index of the record */
+	const ulint*		offsets)	/* in: rec_get_offsets
+						(index_rec, index) */
+{
+	const byte*	data;
+	ulint		len;
+
+	ut_ad(rec_offs_validate(index_rec, index, offsets));
+
+	data = rec_get_nth_field(
+		index_rec, offsets,
+		dict_index_get_sys_col_pos(index, DATA_ROW_ID), &len);
+
+	if (UNIV_UNLIKELY(len != DATA_ROW_ID_LEN)) {
+		fprintf(stderr,
+			"InnoDB: Error: Row id field is"
+			" wrong length %lu in ", (ulong) len);
+		dict_index_name_print(stderr, prebuilt->trx, index);
+		fprintf(stderr, "\n"
+			"InnoDB: Field number %lu, record:\n",
+			(ulong) dict_index_get_sys_col_pos(index,
+							   DATA_ROW_ID));
+		rec_print_new(stderr, index_rec, offsets);
+		putc('\n', stderr);
+		ut_error;
+	}
+
+	ut_memcpy(prebuilt->row_id, data, len);
+}
+
+/******************************************************************
+Stores a non-SQL-NULL field in the MySQL format. The counterpart of this
+function is row_mysql_store_col_in_innobase_format() in row0mysql.c. */
+static
+void
+row_sel_field_store_in_mysql_format(
+/*================================*/
+	byte*		dest,	/* in/out: buffer where to store; NOTE
+				that BLOBs are not in themselves
+				stored here: the caller must allocate
+				and copy the BLOB into buffer before,
+				and pass the pointer to the BLOB in
+				'data' */
+	const mysql_row_templ_t* templ,
+				/* in: MySQL column template.
+				Its following fields are referenced:
+				type, is_unsigned, mysql_col_len,
+				mbminlen, mbmaxlen */
+	const byte*	data,	/* in: data to store */
+	ulint		len)	/* in: length of the data */
+{
+	byte*	ptr;
+	byte*	field_end;
+	byte*	pad_ptr;
+
+	ut_ad(len != UNIV_SQL_NULL);
+
+	switch (templ->type) {
+	case DATA_INT:
+		/* Convert integer data from Innobase to a little-endian
+		format, sign bit restored to normal */
+
+		ptr = dest + len;
+
+		for (;;) {
+			ptr--;
+			*ptr = *data;
+			if (ptr == dest) {
+				break;
+			}
+			data++;
+		}
+
+		if (!templ->is_unsigned) {
+			dest[len - 1] = (byte) (dest[len - 1] ^ 128);
+		}
+
+		ut_ad(templ->mysql_col_len == len);
+		break;
+
+	case DATA_VARCHAR:
+	case DATA_VARMYSQL:
+	case DATA_BINARY:
+		field_end = dest + templ->mysql_col_len;
+
+		if (templ->mysql_type == DATA_MYSQL_TRUE_VARCHAR) {
+			/* This is a >= 5.0.3 type true VARCHAR. Store the
+			length of the data to the first byte or the first
+			two bytes of dest. */
+
+			dest = row_mysql_store_true_var_len(
+				dest, len, templ->mysql_length_bytes);
+		}
+
+		/* Copy the actual data */
+		ut_memcpy(dest, data, len);
+
+		/* Pad with trailing spaces. We pad with spaces also the
+		unused end of a >= 5.0.3 true VARCHAR column, just in case
+		MySQL expects its contents to be deterministic. */
+
+		pad_ptr = dest + len;
+
+		ut_ad(templ->mbminlen <= templ->mbmaxlen);
+
+		/* We handle UCS2 charset strings differently. */
+		if (templ->mbminlen == 2) {
+			/* A space char is two bytes, 0x0020 in UCS2 */
+
+			if (len & 1) {
+				/* A 0x20 has been stripped from the column.
+				Pad it back. */
+
+				if (pad_ptr < field_end) {
+					*pad_ptr = 0x20;
+					pad_ptr++;
+				}
+			}
+
+			/* Pad the rest of the string with 0x0020 */
+
+			while (pad_ptr < field_end) {
+				*pad_ptr = 0x00;
+				pad_ptr++;
+				*pad_ptr = 0x20;
+				pad_ptr++;
+			}
+		} else {
+			ut_ad(templ->mbminlen == 1);
+			/* space=0x20 */
+
+			memset(pad_ptr, 0x20, field_end - pad_ptr);
+		}
+		break;
+
+	case DATA_BLOB:
+		/* Store a pointer to the BLOB buffer to dest: the BLOB was
+		already copied to the buffer in row_sel_store_mysql_rec */
+
+		row_mysql_store_blob_ref(dest, templ->mysql_col_len, data,
+					 len);
+		break;
+
+	case DATA_MYSQL:
+		memcpy(dest, data, len);
+
+		ut_ad(templ->mysql_col_len >= len);
+		ut_ad(templ->mbmaxlen >= templ->mbminlen);
+
+		ut_ad(templ->mbmaxlen > templ->mbminlen
+		      || templ->mysql_col_len == len);
+		/* The following assertion would fail for old tables
+		containing UTF-8 ENUM columns due to Bug #9526. */
+		ut_ad(!templ->mbmaxlen
+		      || !(templ->mysql_col_len % templ->mbmaxlen));
+		ut_ad(len * templ->mbmaxlen >= templ->mysql_col_len);
+
+		if (templ->mbminlen != templ->mbmaxlen) {
+			/* Pad with spaces. This undoes the stripping
+			done in row0mysql.ic, function
+			row_mysql_store_col_in_innobase_format(). */
+
+			memset(dest + len, 0x20, templ->mysql_col_len - len);
+		}
+		break;
+
+	default:
+#ifdef UNIV_DEBUG
+	case DATA_SYS_CHILD:
+	case DATA_SYS:
+		/* These column types should never be shipped to MySQL. */
+		ut_ad(0);
+
+	case DATA_CHAR:
+	case DATA_FIXBINARY:
+	case DATA_FLOAT:
+	case DATA_DOUBLE:
+	case DATA_DECIMAL:
+		/* Above are the valid column types for MySQL data. */
+#endif /* UNIV_DEBUG */
+		ut_ad(templ->mysql_col_len == len);
+		memcpy(dest, data, len);
+	}
+}
+
+/******************************************************************
+Convert a row in the Innobase format to a row in the MySQL format.
+Note that the template in prebuilt may advise us to copy only a few
+columns to mysql_rec, other columns are left blank. All columns may not
+be needed in the query. */
+static
+ibool
+row_sel_store_mysql_rec(
+/*====================*/
+					/* out: TRUE if success, FALSE if
+					could not allocate memory for a BLOB
+					(though we may also assert in that
+					case) */
+	byte*		mysql_rec,	/* out: row in the MySQL format */
+	row_prebuilt_t*	prebuilt,	/* in: prebuilt struct */
+	const rec_t*	rec,		/* in: Innobase record in the index
+					which was described in prebuilt's
+					template; must be protected by
+					a page latch */
+	const ulint*	offsets)	/* in: array returned by
+					rec_get_offsets() */
+{
+	mysql_row_templ_t*	templ;
+	mem_heap_t*		extern_field_heap	= NULL;
+	mem_heap_t*		heap;
+	const byte*		data;
+	ulint			len;
+	ulint			i;
+
+	ut_ad(prebuilt->mysql_template);
+	ut_ad(prebuilt->default_rec);
+	ut_ad(rec_offs_validate(rec, NULL, offsets));
+
+	if (UNIV_LIKELY_NULL(prebuilt->blob_heap)) {
+		mem_heap_free(prebuilt->blob_heap);
+		prebuilt->blob_heap = NULL;
+	}
+
+	for (i = 0; i < prebuilt->n_template; i++) {
+
+		templ = prebuilt->mysql_template + i;
+
+		if (UNIV_UNLIKELY(rec_offs_nth_extern(offsets,
+						      templ->rec_field_no))) {
+
+			/* Copy an externally stored field to the temporary
+			heap */
+
+			ut_a(!prebuilt->trx->has_search_latch);
+
+			if (UNIV_UNLIKELY(templ->type == DATA_BLOB)) {
+				if (prebuilt->blob_heap == NULL) {
+					prebuilt->blob_heap = mem_heap_create(
+						UNIV_PAGE_SIZE);
+				}
+
+				heap = prebuilt->blob_heap;
+			} else {
+				extern_field_heap
+					= mem_heap_create(UNIV_PAGE_SIZE);
+
+				heap = extern_field_heap;
+			}
+
+			/* NOTE: if we are retrieving a big BLOB, we may
+			already run out of memory in the next call, which
+			causes an assert */
+
+			data = btr_rec_copy_externally_stored_field(
+				rec, offsets,
+				dict_table_zip_size(prebuilt->table),
+				templ->rec_field_no, &len, heap);
+
+			ut_a(len != UNIV_SQL_NULL);
+		} else {
+			/* Field is stored in the row. */
+
+			data = rec_get_nth_field(rec, offsets,
+						 templ->rec_field_no, &len);
+
+			if (UNIV_UNLIKELY(templ->type == DATA_BLOB)
+			    && len != UNIV_SQL_NULL) {
+
+				/* It is a BLOB field locally stored in the
+				InnoDB record: we MUST copy its contents to
+				prebuilt->blob_heap here because later code
+				assumes all BLOB values have been copied to a
+				safe place. */
+
+				if (prebuilt->blob_heap == NULL) {
+					prebuilt->blob_heap = mem_heap_create(
+						UNIV_PAGE_SIZE);
+				}
+
+				data = memcpy(mem_heap_alloc(
+						prebuilt->blob_heap, len),
+						data, len);
+			}
+		}
+
+		if (len != UNIV_SQL_NULL) {
+			row_sel_field_store_in_mysql_format(
+				mysql_rec + templ->mysql_col_offset,
+				templ, data, len);
+
+			/* Cleanup */
+			if (extern_field_heap) {
+				mem_heap_free(extern_field_heap);
+				extern_field_heap = NULL;
+			}
+
+			if (templ->mysql_null_bit_mask) {
+				/* It is a nullable column with a non-NULL
+				value */
+				mysql_rec[templ->mysql_null_byte_offset]
+					&= ~(byte) templ->mysql_null_bit_mask;
+			}
+		} else {
+			/* MySQL assumes that the field for an SQL
+			NULL value is set to the default value. */
+
+			mysql_rec[templ->mysql_null_byte_offset]
+				|= (byte) templ->mysql_null_bit_mask;
+			memcpy(mysql_rec + templ->mysql_col_offset,
+			       prebuilt->default_rec + templ->mysql_col_offset,
+			       templ->mysql_col_len);
+		}
+	}
+
+	return(TRUE);
+}
+
+/*************************************************************************
+Builds a previous version of a clustered index record for a consistent read */
+static
+ulint
+row_sel_build_prev_vers_for_mysql(
+/*==============================*/
+					/* out: DB_SUCCESS or error code */
+	read_view_t*	read_view,	/* in: read view */
+	dict_index_t*	clust_index,	/* in: clustered index */
+	row_prebuilt_t*	prebuilt,	/* in: prebuilt struct */
+	const rec_t*	rec,		/* in: record in a clustered index */
+	ulint**		offsets,	/* in/out: offsets returned by
+					rec_get_offsets(rec, clust_index) */
+	mem_heap_t**	offset_heap,	/* in/out: memory heap from which
+					the offsets are allocated */
+	rec_t**		old_vers,	/* out: old version, or NULL if the
+					record does not exist in the view:
+					i.e., it was freshly inserted
+					afterwards */
+	mtr_t*		mtr)		/* in: mtr */
+{
+	ulint	err;
+
+	if (prebuilt->old_vers_heap) {
+		mem_heap_empty(prebuilt->old_vers_heap);
+	} else {
+		prebuilt->old_vers_heap = mem_heap_create(200);
+	}
+
+	err = row_vers_build_for_consistent_read(
+		rec, mtr, clust_index, offsets, read_view, offset_heap,
+		prebuilt->old_vers_heap, old_vers);
+	return(err);
+}
+
+/*************************************************************************
+Retrieves the clustered index record corresponding to a record in a
+non-clustered index. Does the necessary locking. Used in the MySQL
+interface. */
+static
+ulint
+row_sel_get_clust_rec_for_mysql(
+/*============================*/
+				/* out: DB_SUCCESS or error code */
+	row_prebuilt_t*	prebuilt,/* in: prebuilt struct in the handle */
+	dict_index_t*	sec_index,/* in: secondary index where rec resides */
+	const rec_t*	rec,	/* in: record in a non-clustered index; if
+				this is a locking read, then rec is not
+				allowed to be delete-marked, and that would
+				not make sense either */
+	que_thr_t*	thr,	/* in: query thread */
+	const rec_t**	out_rec,/* out: clustered record or an old version of
+				it, NULL if the old version did not exist
+				in the read view, i.e., it was a fresh
+				inserted version */
+	ulint**		offsets,/* in: offsets returned by
+				rec_get_offsets(rec, sec_index);
+				out: offsets returned by
+				rec_get_offsets(out_rec, clust_index) */
+	mem_heap_t**	offset_heap,/* in/out: memory heap from which
+				the offsets are allocated */
+	mtr_t*		mtr)	/* in: mtr used to get access to the
+				non-clustered record; the same mtr is used to
+				access the clustered index */
+{
+	dict_index_t*	clust_index;
+	const rec_t*	clust_rec;
+	rec_t*		old_vers;
+	ulint		err;
+	trx_t*		trx;
+
+	*out_rec = NULL;
+	trx = thr_get_trx(thr);
+
+	row_build_row_ref_in_tuple(prebuilt->clust_ref, rec,
+				   sec_index, *offsets, trx);
+
+	clust_index = dict_table_get_first_index(sec_index->table);
+
+	btr_pcur_open_with_no_init(clust_index, prebuilt->clust_ref,
+				   PAGE_CUR_LE, BTR_SEARCH_LEAF,
+				   prebuilt->clust_pcur, 0, mtr);
+
+	clust_rec = btr_pcur_get_rec(prebuilt->clust_pcur);
+
+	prebuilt->clust_pcur->trx_if_known = trx;
+
+	/* Note: only if the search ends up on a non-infimum record is the
+	low_match value the real match to the search tuple */
+
+	if (!page_rec_is_user_rec(clust_rec)
+	    || btr_pcur_get_low_match(prebuilt->clust_pcur)
+	    < dict_index_get_n_unique(clust_index)) {
+
+		/* In a rare case it is possible that no clust rec is found
+		for a delete-marked secondary index record: if in row0umod.c
+		in row_undo_mod_remove_clust_low() we have already removed
+		the clust rec, while purge is still cleaning and removing
+		secondary index records associated with earlier versions of
+		the clustered index record. In that case we know that the
+		clustered index record did not exist in the read view of
+		trx. */
+
+		if (!rec_get_deleted_flag(rec,
+					  dict_table_is_comp(sec_index->table))
+		    || prebuilt->select_lock_type != LOCK_NONE) {
+			ut_print_timestamp(stderr);
+			fputs("  InnoDB: error clustered record"
+			      " for sec rec not found\n"
+			      "InnoDB: ", stderr);
+			dict_index_name_print(stderr, trx, sec_index);
+			fputs("\n"
+			      "InnoDB: sec index record ", stderr);
+			rec_print(stderr, rec, sec_index);
+			fputs("\n"
+			      "InnoDB: clust index record ", stderr);
+			rec_print(stderr, clust_rec, clust_index);
+			putc('\n', stderr);
+			trx_print(stderr, trx, 600);
+
+			fputs("\n"
+			      "InnoDB: Submit a detailed bug report"
+			      " to http://bugs.mysql.com\n", stderr);
+		}
+
+		clust_rec = NULL;
+
+		goto func_exit;
+	}
+
+	*offsets = rec_get_offsets(clust_rec, clust_index, *offsets,
+				   ULINT_UNDEFINED, offset_heap);
+
+	if (prebuilt->select_lock_type != LOCK_NONE) {
+		/* Try to place a lock on the index record; we are searching
+		the clust rec with a unique condition, hence
+		we set a LOCK_REC_NOT_GAP type lock */
+
+		err = lock_clust_rec_read_check_and_lock(
+			0, btr_pcur_get_block(prebuilt->clust_pcur),
+			clust_rec, clust_index, *offsets,
+			prebuilt->select_lock_type, LOCK_REC_NOT_GAP, thr);
+		if (err != DB_SUCCESS) {
+
+			goto err_exit;
+		}
+	} else {
+		/* This is a non-locking consistent read: if necessary, fetch
+		a previous version of the record */
+
+		old_vers = NULL;
+
+		/* If the isolation level allows reading of uncommitted data,
+		then we never look for an earlier version */
+
+		if (trx->isolation_level > TRX_ISO_READ_UNCOMMITTED
+		    && !lock_clust_rec_cons_read_sees(
+			    clust_rec, clust_index, *offsets,
+			    trx->read_view)) {
+
+			/* The following call returns 'offsets' associated with
+			'old_vers' */
+			err = row_sel_build_prev_vers_for_mysql(
+				trx->read_view, clust_index, prebuilt,
+				clust_rec, offsets, offset_heap, &old_vers,
+				mtr);
+
+			if (err != DB_SUCCESS || old_vers == NULL) {
+
+				goto err_exit;
+			}
+
+			clust_rec = old_vers;
+		}
+
+		/* If we had to go to an earlier version of row or the
+		secondary index record is delete marked, then it may be that
+		the secondary index record corresponding to clust_rec
+		(or old_vers) is not rec; in that case we must ignore
+		such row because in our snapshot rec would not have existed.
+		Remember that from rec we cannot see directly which transaction
+		id corresponds to it: we have to go to the clustered index
+		record. A query where we want to fetch all rows where
+		the secondary index value is in some interval would return
+		a wrong result if we would not drop rows which we come to
+		visit through secondary index records that would not really
+		exist in our snapshot. */
+
+		if (clust_rec
+		    && (old_vers
+			|| rec_get_deleted_flag(rec, dict_table_is_comp(
+							sec_index->table)))
+		    && !row_sel_sec_rec_is_for_clust_rec(
+			    rec, sec_index, clust_rec, clust_index)) {
+			clust_rec = NULL;
+#ifdef UNIV_SEARCH_DEBUG
+		} else {
+			ut_a(clust_rec == NULL
+			     || row_sel_sec_rec_is_for_clust_rec(
+				     rec, sec_index, clust_rec, clust_index));
+#endif
+		}
+	}
+
+func_exit:
+	*out_rec = clust_rec;
+
+	if (prebuilt->select_lock_type == LOCK_X) {
+		/* We may use the cursor in update: store its position */
+
+		btr_pcur_store_position(prebuilt->clust_pcur, mtr);
+	}
+
+	err = DB_SUCCESS;
+err_exit:
+	return(err);
+}
+
+/************************************************************************
+Restores cursor position after it has been stored. We have to take into
+account that the record cursor was positioned on may have been deleted.
+Then we may have to move the cursor one step up or down. */
+static
+ibool
+sel_restore_position_for_mysql(
+/*===========================*/
+					/* out: TRUE if we may need to
+					process the record the cursor is
+					now positioned on (i.e. we should
+					not go to the next record yet) */
+	ibool*		same_user_rec,	/* out: TRUE if we were able to restore
+					the cursor on a user record with the
+					same ordering prefix in in the
+					B-tree index */
+	ulint		latch_mode,	/* in: latch mode wished in
+					restoration */
+	btr_pcur_t*	pcur,		/* in: cursor whose position
+					has been stored */
+	ibool		moves_up,	/* in: TRUE if the cursor moves up
+					in the index */
+	mtr_t*		mtr)		/* in: mtr; CAUTION: may commit
+					mtr temporarily! */
+{
+	ibool	success;
+	ulint	relative_position;
+
+	relative_position = pcur->rel_pos;
+
+	success = btr_pcur_restore_position(latch_mode, pcur, mtr);
+
+	*same_user_rec = success;
+
+	if (relative_position == BTR_PCUR_ON) {
+		if (success) {
+			return(FALSE);
+		}
+
+		if (moves_up) {
+			btr_pcur_move_to_next(pcur, mtr);
+		}
+
+		return(TRUE);
+	}
+
+	if (relative_position == BTR_PCUR_AFTER
+	    || relative_position == BTR_PCUR_AFTER_LAST_IN_TREE) {
+
+		if (moves_up) {
+			return(TRUE);
+		}
+
+		if (btr_pcur_is_on_user_rec(pcur)) {
+			btr_pcur_move_to_prev(pcur, mtr);
+		}
+
+		return(TRUE);
+	}
+
+	ut_ad(relative_position == BTR_PCUR_BEFORE
+	      || relative_position == BTR_PCUR_BEFORE_FIRST_IN_TREE);
+
+	if (moves_up && btr_pcur_is_on_user_rec(pcur)) {
+		btr_pcur_move_to_next(pcur, mtr);
+	}
+
+	return(TRUE);
+}
+
+/************************************************************************
+Pops a cached row for MySQL from the fetch cache. */
+UNIV_INLINE
+void
+row_sel_pop_cached_row_for_mysql(
+/*=============================*/
+	byte*		buf,		/* in/out: buffer where to copy the
+					row */
+	row_prebuilt_t*	prebuilt)	/* in: prebuilt struct */
+{
+	ulint			i;
+	mysql_row_templ_t*	templ;
+	byte*			cached_rec;
+	ut_ad(prebuilt->n_fetch_cached > 0);
+	ut_ad(prebuilt->mysql_prefix_len <= prebuilt->mysql_row_len);
+
+	if (UNIV_UNLIKELY(prebuilt->keep_other_fields_on_keyread)) {
+		/* Copy cache record field by field, don't touch fields that
+		are not covered by current key */
+		cached_rec = prebuilt->fetch_cache[
+			prebuilt->fetch_cache_first];
+
+		for (i = 0; i < prebuilt->n_template; i++) {
+			templ = prebuilt->mysql_template + i;
+			ut_memcpy(buf + templ->mysql_col_offset,
+				  cached_rec + templ->mysql_col_offset,
+				  templ->mysql_col_len);
+			/* Copy NULL bit of the current field from cached_rec
+			to buf */
+			if (templ->mysql_null_bit_mask) {
+				buf[templ->mysql_null_byte_offset]
+					^= (buf[templ->mysql_null_byte_offset]
+					    ^ cached_rec[templ->mysql_null_byte_offset])
+					& (byte)templ->mysql_null_bit_mask;
+			}
+		}
+	}
+	else {
+		ut_memcpy(buf,
+			  prebuilt->fetch_cache[prebuilt->fetch_cache_first],
+			  prebuilt->mysql_prefix_len);
+	}
+	prebuilt->n_fetch_cached--;
+	prebuilt->fetch_cache_first++;
+
+	if (prebuilt->n_fetch_cached == 0) {
+		prebuilt->fetch_cache_first = 0;
+	}
+}
+
+/************************************************************************
+Pushes a row for MySQL to the fetch cache. */
+UNIV_INLINE
+void
+row_sel_push_cache_row_for_mysql(
+/*=============================*/
+	row_prebuilt_t*	prebuilt,	/* in: prebuilt struct */
+	const rec_t*	rec,		/* in: record to push; must
+					be protected by a page latch */
+	const ulint*	offsets)	/* in: rec_get_offsets() */
+{
+	byte*	buf;
+	ulint	i;
+
+	ut_ad(prebuilt->n_fetch_cached < MYSQL_FETCH_CACHE_SIZE);
+	ut_ad(rec_offs_validate(rec, NULL, offsets));
+	ut_a(!prebuilt->templ_contains_blob);
+
+	if (prebuilt->fetch_cache[0] == NULL) {
+		/* Allocate memory for the fetch cache */
+
+		for (i = 0; i < MYSQL_FETCH_CACHE_SIZE; i++) {
+
+			/* A user has reported memory corruption in these
+			buffers in Linux. Put magic numbers there to help
+			to track a possible bug. */
+
+			buf = mem_alloc(prebuilt->mysql_row_len + 8);
+
+			prebuilt->fetch_cache[i] = buf + 4;
+
+			mach_write_to_4(buf, ROW_PREBUILT_FETCH_MAGIC_N);
+			mach_write_to_4(buf + 4 + prebuilt->mysql_row_len,
+					ROW_PREBUILT_FETCH_MAGIC_N);
+		}
+	}
+
+	ut_ad(prebuilt->fetch_cache_first == 0);
+
+	if (UNIV_UNLIKELY(!row_sel_store_mysql_rec(
+				  prebuilt->fetch_cache[
+					  prebuilt->n_fetch_cached],
+				  prebuilt, rec, offsets))) {
+		ut_error;
+	}
+
+	prebuilt->n_fetch_cached++;
+}
+
+/*************************************************************************
+Tries to do a shortcut to fetch a clustered index record with a unique key,
+using the hash index if possible (not always). We assume that the search
+mode is PAGE_CUR_GE, it is a consistent read, there is a read view in trx,
+btr search latch has been locked in S-mode. */
+static
+ulint
+row_sel_try_search_shortcut_for_mysql(
+/*==================================*/
+				/* out: SEL_FOUND, SEL_EXHAUSTED, SEL_RETRY */
+	const rec_t**	out_rec,/* out: record if found */
+	row_prebuilt_t*	prebuilt,/* in: prebuilt struct */
+	ulint**		offsets,/* in/out: for rec_get_offsets(*out_rec) */
+	mem_heap_t**	heap,	/* in/out: heap for rec_get_offsets() */
+	mtr_t*		mtr)	/* in: started mtr */
+{
+	dict_index_t*	index		= prebuilt->index;
+	const dtuple_t*	search_tuple	= prebuilt->search_tuple;
+	btr_pcur_t*	pcur		= prebuilt->pcur;
+	trx_t*		trx		= prebuilt->trx;
+	const rec_t*	rec;
+
+	ut_ad(dict_index_is_clust(index));
+	ut_ad(!prebuilt->templ_contains_blob);
+
+	btr_pcur_open_with_no_init(index, search_tuple, PAGE_CUR_GE,
+				   BTR_SEARCH_LEAF, pcur,
+#ifndef UNIV_SEARCH_DEBUG
+				   RW_S_LATCH,
+#else
+				   0,
+#endif
+				   mtr);
+	rec = btr_pcur_get_rec(pcur);
+
+	if (!page_rec_is_user_rec(rec)) {
+
+		return(SEL_RETRY);
+	}
+
+	/* As the cursor is now placed on a user record after a search with
+	the mode PAGE_CUR_GE, the up_match field in the cursor tells how many
+	fields in the user record matched to the search tuple */
+
+	if (btr_pcur_get_up_match(pcur) < dtuple_get_n_fields(search_tuple)) {
+
+		return(SEL_EXHAUSTED);
+	}
+
+	/* This is a non-locking consistent read: if necessary, fetch
+	a previous version of the record */
+
+	*offsets = rec_get_offsets(rec, index, *offsets,
+				   ULINT_UNDEFINED, heap);
+
+	if (!lock_clust_rec_cons_read_sees(rec, index,
+					   *offsets, trx->read_view)) {
+
+		return(SEL_RETRY);
+	}
+
+	if (rec_get_deleted_flag(rec, dict_table_is_comp(index->table))) {
+
+		return(SEL_EXHAUSTED);
+	}
+
+	*out_rec = rec;
+
+	return(SEL_FOUND);
+}
+
+/************************************************************************
+Searches for rows in the database. This is used in the interface to
+MySQL. This function opens a cursor, and also implements fetch next
+and fetch prev. NOTE that if we do a search with a full key value
+from a unique index (ROW_SEL_EXACT), then we will not store the cursor
+position and fetch next or fetch prev must not be tried to the cursor! */
+UNIV_INTERN
+ulint
+row_search_for_mysql(
+/*=================*/
+					/* out: DB_SUCCESS,
+					DB_RECORD_NOT_FOUND,
+					DB_END_OF_INDEX, DB_DEADLOCK,
+					DB_LOCK_TABLE_FULL, DB_CORRUPTION,
+					or DB_TOO_BIG_RECORD */
+	byte*		buf,		/* in/out: buffer for the fetched
+					row in the MySQL format */
+	ulint		mode,		/* in: search mode PAGE_CUR_L, ... */
+	row_prebuilt_t*	prebuilt,	/* in: prebuilt struct for the
+					table handle; this contains the info
+					of search_tuple, index; if search
+					tuple contains 0 fields then we
+					position the cursor at the start or
+					the end of the index, depending on
+					'mode' */
+	ulint		match_mode,	/* in: 0 or ROW_SEL_EXACT or
+					ROW_SEL_EXACT_PREFIX */
+	ulint		direction)	/* in: 0 or ROW_SEL_NEXT or
+					ROW_SEL_PREV; NOTE: if this is != 0,
+					then prebuilt must have a pcur
+					with stored position! In opening of a
+					cursor 'direction' should be 0. */
+{
+	dict_index_t*	index		= prebuilt->index;
+	ibool		comp		= dict_table_is_comp(index->table);
+	const dtuple_t*	search_tuple	= prebuilt->search_tuple;
+	btr_pcur_t*	pcur		= prebuilt->pcur;
+	trx_t*		trx		= prebuilt->trx;
+	dict_index_t*	clust_index;
+	que_thr_t*	thr;
+	const rec_t*	rec;
+	const rec_t*	result_rec;
+	const rec_t*	clust_rec;
+	ulint		err				= DB_SUCCESS;
+	ibool		unique_search			= FALSE;
+	ibool		unique_search_from_clust_index	= FALSE;
+	ibool		mtr_has_extra_clust_latch	= FALSE;
+	ibool		moves_up			= FALSE;
+	ibool		set_also_gap_locks		= TRUE;
+	/* if the query is a plain locking SELECT, and the isolation level
+	is <= TRX_ISO_READ_COMMITTED, then this is set to FALSE */
+	ibool		did_semi_consistent_read	= FALSE;
+	/* if the returned record was locked and we did a semi-consistent
+	read (fetch the newest committed version), then this is set to
+	TRUE */
+#ifdef UNIV_SEARCH_DEBUG
+	ulint		cnt				= 0;
+#endif /* UNIV_SEARCH_DEBUG */
+	ulint		next_offs;
+	ibool		same_user_rec;
+	mtr_t		mtr;
+	mem_heap_t*	heap				= NULL;
+	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint*		offsets				= offsets_;
+
+	rec_offs_init(offsets_);
+
+	ut_ad(index && pcur && search_tuple);
+	ut_ad(trx->mysql_thread_id == os_thread_get_curr_id());
+
+	if (UNIV_UNLIKELY(prebuilt->table->ibd_file_missing)) {
+		ut_print_timestamp(stderr);
+		fprintf(stderr, "  InnoDB: Error:\n"
+			"InnoDB: MySQL is trying to use a table handle"
+			" but the .ibd file for\n"
+			"InnoDB: table %s does not exist.\n"
+			"InnoDB: Have you deleted the .ibd file"
+			" from the database directory under\n"
+			"InnoDB: the MySQL datadir, or have you used"
+			" DISCARD TABLESPACE?\n"
+			"InnoDB: Look from\n"
+			"InnoDB: http://dev.mysql.com/doc/refman/5.1/en/"
+			"innodb-troubleshooting.html\n"
+			"InnoDB: how you can resolve the problem.\n",
+			prebuilt->table->name);
+
+		return(DB_ERROR);
+	}
+
+	if (UNIV_UNLIKELY(prebuilt->magic_n != ROW_PREBUILT_ALLOCATED)) {
+		fprintf(stderr,
+			"InnoDB: Error: trying to free a corrupt\n"
+			"InnoDB: table handle. Magic n %lu, table name ",
+			(ulong) prebuilt->magic_n);
+		ut_print_name(stderr, trx, TRUE, prebuilt->table->name);
+		putc('\n', stderr);
+
+		mem_analyze_corruption(prebuilt);
+
+		ut_error;
+	}
+
+#if 0
+	/* August 19, 2005 by Heikki: temporarily disable this error
+	print until the cursor lock count is done correctly.
+	See bugs #12263 and #12456!*/
+
+	if (trx->n_mysql_tables_in_use == 0
+	    && UNIV_UNLIKELY(prebuilt->select_lock_type == LOCK_NONE)) {
+		/* Note that if MySQL uses an InnoDB temp table that it
+		created inside LOCK TABLES, then n_mysql_tables_in_use can
+		be zero; in that case select_lock_type is set to LOCK_X in
+		::start_stmt. */
+
+		fputs("InnoDB: Error: MySQL is trying to perform a SELECT\n"
+		      "InnoDB: but it has not locked"
+		      " any tables in ::external_lock()!\n",
+		      stderr);
+		trx_print(stderr, trx, 600);
+		fputc('\n', stderr);
+	}
+#endif
+
+#if 0
+	fprintf(stderr, "Match mode %lu\n search tuple ",
+		(ulong) match_mode);
+	dtuple_print(search_tuple);
+	fprintf(stderr, "N tables locked %lu\n",
+		(ulong) trx->mysql_n_tables_locked);
+#endif
+	/*-------------------------------------------------------------*/
+	/* PHASE 0: Release a possible s-latch we are holding on the
+	adaptive hash index latch if there is someone waiting behind */
+
+	if (UNIV_UNLIKELY(rw_lock_get_writer(&btr_search_latch) != RW_LOCK_NOT_LOCKED)
+	    && trx->has_search_latch) {
+
+		/* There is an x-latch request on the adaptive hash index:
+		release the s-latch to reduce starvation and wait for
+		BTR_SEA_TIMEOUT rounds before trying to keep it again over
+		calls from MySQL */
+
+		rw_lock_s_unlock(&btr_search_latch);
+		trx->has_search_latch = FALSE;
+
+		trx->search_latch_timeout = BTR_SEA_TIMEOUT;
+	}
+
+	/* Reset the new record lock info if srv_locks_unsafe_for_binlog
+	is set or session is using a READ COMMITED isolation level. Then
+	we are able to remove the record locks set here on an individual
+	row. */
+
+	if ((srv_locks_unsafe_for_binlog
+	     || trx->isolation_level == TRX_ISO_READ_COMMITTED)
+	    && prebuilt->select_lock_type != LOCK_NONE) {
+
+		trx_reset_new_rec_lock_info(trx);
+	}
+
+	/*-------------------------------------------------------------*/
+	/* PHASE 1: Try to pop the row from the prefetch cache */
+
+	if (UNIV_UNLIKELY(direction == 0)) {
+		trx->op_info = "starting index read";
+
+		prebuilt->n_rows_fetched = 0;
+		prebuilt->n_fetch_cached = 0;
+		prebuilt->fetch_cache_first = 0;
+
+		if (prebuilt->sel_graph == NULL) {
+			/* Build a dummy select query graph */
+			row_prebuild_sel_graph(prebuilt);
+		}
+	} else {
+		trx->op_info = "fetching rows";
+
+		if (prebuilt->n_rows_fetched == 0) {
+			prebuilt->fetch_direction = direction;
+		}
+
+		if (UNIV_UNLIKELY(direction != prebuilt->fetch_direction)) {
+			if (UNIV_UNLIKELY(prebuilt->n_fetch_cached > 0)) {
+				ut_error;
+				/* TODO: scrollable cursor: restore cursor to
+				the place of the latest returned row,
+				or better: prevent caching for a scroll
+				cursor! */
+			}
+
+			prebuilt->n_rows_fetched = 0;
+			prebuilt->n_fetch_cached = 0;
+			prebuilt->fetch_cache_first = 0;
+
+		} else if (UNIV_LIKELY(prebuilt->n_fetch_cached > 0)) {
+			row_sel_pop_cached_row_for_mysql(buf, prebuilt);
+
+			prebuilt->n_rows_fetched++;
+
+			srv_n_rows_read++;
+			err = DB_SUCCESS;
+			goto func_exit;
+		}
+
+		if (prebuilt->fetch_cache_first > 0
+		    && prebuilt->fetch_cache_first < MYSQL_FETCH_CACHE_SIZE) {
+
+			/* The previous returned row was popped from the fetch
+			cache, but the cache was not full at the time of the
+			popping: no more rows can exist in the result set */
+
+			err = DB_RECORD_NOT_FOUND;
+			goto func_exit;
+		}
+
+		prebuilt->n_rows_fetched++;
+
+		if (prebuilt->n_rows_fetched > 1000000000) {
+			/* Prevent wrap-over */
+			prebuilt->n_rows_fetched = 500000000;
+		}
+
+		mode = pcur->search_mode;
+	}
+
+	/* In a search where at most one record in the index may match, we
+	can use a LOCK_REC_NOT_GAP type record lock when locking a
+	non-delete-marked matching record.
+
+	Note that in a unique secondary index there may be different
+	delete-marked versions of a record where only the primary key
+	values differ: thus in a secondary index we must use next-key
+	locks when locking delete-marked records. */
+
+	if (match_mode == ROW_SEL_EXACT
+	    && dict_index_is_unique(index)
+	    && dtuple_get_n_fields(search_tuple)
+	    == dict_index_get_n_unique(index)
+	    && (dict_index_is_clust(index)
+		|| !dtuple_contains_null(search_tuple))) {
+
+		/* Note above that a UNIQUE secondary index can contain many
+		rows with the same key value if one of the columns is the SQL
+		null. A clustered index under MySQL can never contain null
+		columns because we demand that all the columns in primary key
+		are non-null. */
+
+		unique_search = TRUE;
+
+		/* Even if the condition is unique, MySQL seems to try to
+		retrieve also a second row if a primary key contains more than
+		1 column. Return immediately if this is not a HANDLER
+		command. */
+
+		if (UNIV_UNLIKELY(direction != 0
+				  && !prebuilt->used_in_HANDLER)) {
+
+			err = DB_RECORD_NOT_FOUND;
+			goto func_exit;
+		}
+	}
+
+	mtr_start(&mtr);
+
+	/*-------------------------------------------------------------*/
+	/* PHASE 2: Try fast adaptive hash index search if possible */
+
+	/* Next test if this is the special case where we can use the fast
+	adaptive hash index to try the search. Since we must release the
+	search system latch when we retrieve an externally stored field, we
+	cannot use the adaptive hash index in a search in the case the row
+	may be long and there may be externally stored fields */
+
+	if (UNIV_UNLIKELY(direction == 0)
+	    && unique_search
+	    && dict_index_is_clust(index)
+	    && !prebuilt->templ_contains_blob
+	    && !prebuilt->used_in_HANDLER
+	    && (prebuilt->mysql_row_len < UNIV_PAGE_SIZE / 8)) {
+
+		mode = PAGE_CUR_GE;
+
+		unique_search_from_clust_index = TRUE;
+
+		if (trx->mysql_n_tables_locked == 0
+		    && prebuilt->select_lock_type == LOCK_NONE
+		    && trx->isolation_level > TRX_ISO_READ_UNCOMMITTED
+		    && trx->read_view) {
+
+			/* This is a SELECT query done as a consistent read,
+			and the read view has already been allocated:
+			let us try a search shortcut through the hash
+			index.
+			NOTE that we must also test that
+			mysql_n_tables_locked == 0, because this might
+			also be INSERT INTO ... SELECT ... or
+			CREATE TABLE ... SELECT ... . Our algorithm is
+			NOT prepared to inserts interleaved with the SELECT,
+			and if we try that, we can deadlock on the adaptive
+			hash index semaphore! */
+
+#ifndef UNIV_SEARCH_DEBUG
+			if (!trx->has_search_latch) {
+				rw_lock_s_lock(&btr_search_latch);
+				trx->has_search_latch = TRUE;
+			}
+#endif
+			switch (row_sel_try_search_shortcut_for_mysql(
+					&rec, prebuilt, &offsets, &heap,
+					&mtr)) {
+			case SEL_FOUND:
+#ifdef UNIV_SEARCH_DEBUG
+				ut_a(0 == cmp_dtuple_rec(search_tuple,
+							 rec, offsets));
+#endif
+				/* At this point, rec is protected by
+				a page latch that was acquired by
+				row_sel_try_search_shortcut_for_mysql().
+				The latch will not be released until
+				mtr_commit(&mtr). */
+
+				if (!row_sel_store_mysql_rec(buf, prebuilt,
+							     rec, offsets)) {
+					err = DB_TOO_BIG_RECORD;
+
+					/* We let the main loop to do the
+					error handling */
+					goto shortcut_fails_too_big_rec;
+				}
+
+				mtr_commit(&mtr);
+
+				/* ut_print_name(stderr, index->name);
+				fputs(" shortcut\n", stderr); */
+
+				srv_n_rows_read++;
+
+				err = DB_SUCCESS;
+				goto release_search_latch_if_needed;
+
+			case SEL_EXHAUSTED:
+				mtr_commit(&mtr);
+
+				/* ut_print_name(stderr, index->name);
+				fputs(" record not found 2\n", stderr); */
+
+				err = DB_RECORD_NOT_FOUND;
+release_search_latch_if_needed:
+				if (trx->search_latch_timeout > 0
+				    && trx->has_search_latch) {
+
+					trx->search_latch_timeout--;
+
+					rw_lock_s_unlock(&btr_search_latch);
+					trx->has_search_latch = FALSE;
+				}
+
+				/* NOTE that we do NOT store the cursor
+				position */
+				goto func_exit;
+
+			case SEL_RETRY:
+				break;
+
+			default:
+				ut_ad(0);
+			}
+shortcut_fails_too_big_rec:
+			mtr_commit(&mtr);
+			mtr_start(&mtr);
+		}
+	}
+
+	/*-------------------------------------------------------------*/
+	/* PHASE 3: Open or restore index cursor position */
+
+	if (trx->has_search_latch) {
+		rw_lock_s_unlock(&btr_search_latch);
+		trx->has_search_latch = FALSE;
+	}
+
+	trx_start_if_not_started(trx);
+
+	if (trx->isolation_level <= TRX_ISO_READ_COMMITTED
+	    && prebuilt->select_lock_type != LOCK_NONE
+	    && trx->mysql_thd != NULL
+	    && thd_is_select(trx->mysql_thd)) {
+		/* It is a plain locking SELECT and the isolation
+		level is low: do not lock gaps */
+
+		set_also_gap_locks = FALSE;
+	}
+
+	/* Note that if the search mode was GE or G, then the cursor
+	naturally moves upward (in fetch next) in alphabetical order,
+	otherwise downward */
+
+	if (UNIV_UNLIKELY(direction == 0)) {
+		if (mode == PAGE_CUR_GE || mode == PAGE_CUR_G) {
+			moves_up = TRUE;
+		}
+	} else if (direction == ROW_SEL_NEXT) {
+		moves_up = TRUE;
+	}
+
+	thr = que_fork_get_first_thr(prebuilt->sel_graph);
+
+	que_thr_move_to_run_state_for_mysql(thr, trx);
+
+	clust_index = dict_table_get_first_index(index->table);
+
+	if (UNIV_LIKELY(direction != 0)) {
+		ibool	need_to_process = sel_restore_position_for_mysql(
+			&same_user_rec, BTR_SEARCH_LEAF,
+			pcur, moves_up, &mtr);
+
+		if (UNIV_UNLIKELY(need_to_process)) {
+			if (UNIV_UNLIKELY(prebuilt->row_read_type
+					  == ROW_READ_DID_SEMI_CONSISTENT)) {
+				/* We did a semi-consistent read,
+				but the record was removed in
+				the meantime. */
+				prebuilt->row_read_type
+					= ROW_READ_TRY_SEMI_CONSISTENT;
+			}
+		} else if (UNIV_LIKELY(prebuilt->row_read_type
+				       != ROW_READ_DID_SEMI_CONSISTENT)) {
+
+			/* The cursor was positioned on the record
+			that we returned previously.  If we need
+			to repeat a semi-consistent read as a
+			pessimistic locking read, the record
+			cannot be skipped. */
+
+			goto next_rec;
+		}
+
+	} else if (dtuple_get_n_fields(search_tuple) > 0) {
+
+		btr_pcur_open_with_no_init(index, search_tuple, mode,
+					   BTR_SEARCH_LEAF,
+					   pcur, 0, &mtr);
+
+		pcur->trx_if_known = trx;
+
+		rec = btr_pcur_get_rec(pcur);
+
+		if (!moves_up
+		    && !page_rec_is_supremum(rec)
+		    && set_also_gap_locks
+		    && !(srv_locks_unsafe_for_binlog
+			 || trx->isolation_level == TRX_ISO_READ_COMMITTED)
+		    && prebuilt->select_lock_type != LOCK_NONE) {
+
+			/* Try to place a gap lock on the next index record
+			to prevent phantoms in ORDER BY ... DESC queries */
+			const rec_t*	next = page_rec_get_next_const(rec);
+
+			offsets = rec_get_offsets(next, index, offsets,
+						  ULINT_UNDEFINED, &heap);
+			err = sel_set_rec_lock(btr_pcur_get_block(pcur),
+					       next, index, offsets,
+					       prebuilt->select_lock_type,
+					       LOCK_GAP, thr);
+
+			if (err != DB_SUCCESS) {
+
+				goto lock_wait_or_error;
+			}
+		}
+	} else {
+		if (mode == PAGE_CUR_G) {
+			btr_pcur_open_at_index_side(
+				TRUE, index, BTR_SEARCH_LEAF, pcur, FALSE,
+				&mtr);
+		} else if (mode == PAGE_CUR_L) {
+			btr_pcur_open_at_index_side(
+				FALSE, index, BTR_SEARCH_LEAF, pcur, FALSE,
+				&mtr);
+		}
+	}
+
+	if (!prebuilt->sql_stat_start) {
+		/* No need to set an intention lock or assign a read view */
+
+		if (trx->read_view == NULL
+		    && prebuilt->select_lock_type == LOCK_NONE) {
+
+			fputs("InnoDB: Error: MySQL is trying to"
+			      " perform a consistent read\n"
+			      "InnoDB: but the read view is not assigned!\n",
+			      stderr);
+			trx_print(stderr, trx, 600);
+			fputc('\n', stderr);
+			ut_a(0);
+		}
+	} else if (prebuilt->select_lock_type == LOCK_NONE) {
+		/* This is a consistent read */
+		/* Assign a read view for the query */
+
+		trx_assign_read_view(trx);
+		prebuilt->sql_stat_start = FALSE;
+	} else {
+		ulint	lock_mode;
+		if (prebuilt->select_lock_type == LOCK_S) {
+			lock_mode = LOCK_IS;
+		} else {
+			lock_mode = LOCK_IX;
+		}
+		err = lock_table(0, index->table, lock_mode, thr);
+
+		if (err != DB_SUCCESS) {
+
+			goto lock_wait_or_error;
+		}
+		prebuilt->sql_stat_start = FALSE;
+	}
+
+rec_loop:
+	/*-------------------------------------------------------------*/
+	/* PHASE 4: Look for matching records in a loop */
+
+	rec = btr_pcur_get_rec(pcur);
+	ut_ad(!!page_rec_is_comp(rec) == comp);
+#ifdef UNIV_SEARCH_DEBUG
+	/*
+	fputs("Using ", stderr);
+	dict_index_name_print(stderr, index);
+	fprintf(stderr, " cnt %lu ; Page no %lu\n", cnt,
+	page_get_page_no(page_align(rec)));
+	rec_print(rec);
+	*/
+#endif /* UNIV_SEARCH_DEBUG */
+
+	if (page_rec_is_infimum(rec)) {
+
+		/* The infimum record on a page cannot be in the result set,
+		and neither can a record lock be placed on it: we skip such
+		a record. */
+
+		goto next_rec;
+	}
+
+	if (page_rec_is_supremum(rec)) {
+
+		if (set_also_gap_locks
+		    && !(srv_locks_unsafe_for_binlog
+			 || trx->isolation_level == TRX_ISO_READ_COMMITTED)
+		    && prebuilt->select_lock_type != LOCK_NONE) {
+
+			/* Try to place a lock on the index record */
+
+			/* If innodb_locks_unsafe_for_binlog option is used
+			or this session is using a READ COMMITTED isolation
+			level we do not lock gaps. Supremum record is really
+			a gap and therefore we do not set locks there. */
+
+			offsets = rec_get_offsets(rec, index, offsets,
+						  ULINT_UNDEFINED, &heap);
+			err = sel_set_rec_lock(btr_pcur_get_block(pcur),
+					       rec, index, offsets,
+					       prebuilt->select_lock_type,
+					       LOCK_ORDINARY, thr);
+
+			if (err != DB_SUCCESS) {
+
+				goto lock_wait_or_error;
+			}
+		}
+		/* A page supremum record cannot be in the result set: skip
+		it now that we have placed a possible lock on it */
+
+		goto next_rec;
+	}
+
+	/*-------------------------------------------------------------*/
+	/* Do sanity checks in case our cursor has bumped into page
+	corruption */
+
+	if (comp) {
+		next_offs = rec_get_next_offs(rec, TRUE);
+		if (UNIV_UNLIKELY(next_offs < PAGE_NEW_SUPREMUM)) {
+
+			goto wrong_offs;
+		}
+	} else {
+		next_offs = rec_get_next_offs(rec, FALSE);
+		if (UNIV_UNLIKELY(next_offs < PAGE_OLD_SUPREMUM)) {
+
+			goto wrong_offs;
+		}
+	}
+
+	if (UNIV_UNLIKELY(next_offs >= UNIV_PAGE_SIZE - PAGE_DIR)) {
+
+wrong_offs:
+		if (srv_force_recovery == 0 || moves_up == FALSE) {
+			ut_print_timestamp(stderr);
+			buf_page_print(page_align(rec), 0);
+			fprintf(stderr,
+				"\nInnoDB: rec address %p,"
+				" buf block fix count %lu\n",
+				(void*) rec, (ulong)
+				btr_cur_get_block(btr_pcur_get_btr_cur(pcur))
+				->page.buf_fix_count);
+			fprintf(stderr,
+				"InnoDB: Index corruption: rec offs %lu"
+				" next offs %lu, page no %lu,\n"
+				"InnoDB: ",
+				(ulong) page_offset(rec),
+				(ulong) next_offs,
+				(ulong) page_get_page_no(page_align(rec)));
+			dict_index_name_print(stderr, trx, index);
+			fputs(". Run CHECK TABLE. You may need to\n"
+			      "InnoDB: restore from a backup, or"
+			      " dump + drop + reimport the table.\n",
+			      stderr);
+
+			err = DB_CORRUPTION;
+
+			goto lock_wait_or_error;
+		} else {
+			/* The user may be dumping a corrupt table. Jump
+			over the corruption to recover as much as possible. */
+
+			fprintf(stderr,
+				"InnoDB: Index corruption: rec offs %lu"
+				" next offs %lu, page no %lu,\n"
+				"InnoDB: ",
+				(ulong) page_offset(rec),
+				(ulong) next_offs,
+				(ulong) page_get_page_no(page_align(rec)));
+			dict_index_name_print(stderr, trx, index);
+			fputs(". We try to skip the rest of the page.\n",
+			      stderr);
+
+			btr_pcur_move_to_last_on_page(pcur, &mtr);
+
+			goto next_rec;
+		}
+	}
+	/*-------------------------------------------------------------*/
+
+	/* Calculate the 'offsets' associated with 'rec' */
+
+	offsets = rec_get_offsets(rec, index, offsets, ULINT_UNDEFINED, &heap);
+
+	if (UNIV_UNLIKELY(srv_force_recovery > 0)) {
+		if (!rec_validate(rec, offsets)
+		    || !btr_index_rec_validate(rec, index, FALSE)) {
+			fprintf(stderr,
+				"InnoDB: Index corruption: rec offs %lu"
+				" next offs %lu, page no %lu,\n"
+				"InnoDB: ",
+				(ulong) page_offset(rec),
+				(ulong) next_offs,
+				(ulong) page_get_page_no(page_align(rec)));
+			dict_index_name_print(stderr, trx, index);
+			fputs(". We try to skip the record.\n",
+			      stderr);
+
+			goto next_rec;
+		}
+	}
+
+	/* Note that we cannot trust the up_match value in the cursor at this
+	place because we can arrive here after moving the cursor! Thus
+	we have to recompare rec and search_tuple to determine if they
+	match enough. */
+
+	if (match_mode == ROW_SEL_EXACT) {
+		/* Test if the index record matches completely to search_tuple
+		in prebuilt: if not, then we return with DB_RECORD_NOT_FOUND */
+
+		/* fputs("Comparing rec and search tuple\n", stderr); */
+
+		if (0 != cmp_dtuple_rec(search_tuple, rec, offsets)) {
+
+			if (set_also_gap_locks
+			    && !(srv_locks_unsafe_for_binlog
+				 || trx->isolation_level
+				 == TRX_ISO_READ_COMMITTED)
+			    && prebuilt->select_lock_type != LOCK_NONE) {
+
+				/* Try to place a gap lock on the index
+				record only if innodb_locks_unsafe_for_binlog
+				option is not set or this session is not
+				using a READ COMMITTED isolation level. */
+
+				err = sel_set_rec_lock(
+					btr_pcur_get_block(pcur),
+					rec, index, offsets,
+					prebuilt->select_lock_type, LOCK_GAP,
+					thr);
+
+				if (err != DB_SUCCESS) {
+
+					goto lock_wait_or_error;
+				}
+			}
+
+			btr_pcur_store_position(pcur, &mtr);
+
+			err = DB_RECORD_NOT_FOUND;
+			/* ut_print_name(stderr, index->name);
+			fputs(" record not found 3\n", stderr); */
+
+			goto normal_return;
+		}
+
+	} else if (match_mode == ROW_SEL_EXACT_PREFIX) {
+
+		if (!cmp_dtuple_is_prefix_of_rec(search_tuple, rec, offsets)) {
+
+			if (set_also_gap_locks
+			    && !(srv_locks_unsafe_for_binlog
+				 || trx->isolation_level
+				 == TRX_ISO_READ_COMMITTED)
+			    && prebuilt->select_lock_type != LOCK_NONE) {
+
+				/* Try to place a gap lock on the index
+				record only if innodb_locks_unsafe_for_binlog
+				option is not set or this session is not
+				using a READ COMMITTED isolation level. */
+
+				err = sel_set_rec_lock(
+					btr_pcur_get_block(pcur),
+					rec, index, offsets,
+					prebuilt->select_lock_type, LOCK_GAP,
+					thr);
+
+				if (err != DB_SUCCESS) {
+
+					goto lock_wait_or_error;
+				}
+			}
+
+			btr_pcur_store_position(pcur, &mtr);
+
+			err = DB_RECORD_NOT_FOUND;
+			/* ut_print_name(stderr, index->name);
+			fputs(" record not found 4\n", stderr); */
+
+			goto normal_return;
+		}
+	}
+
+	/* We are ready to look at a possible new index entry in the result
+	set: the cursor is now placed on a user record */
+
+	if (prebuilt->select_lock_type != LOCK_NONE) {
+		/* Try to place a lock on the index record; note that delete
+		marked records are a special case in a unique search. If there
+		is a non-delete marked record, then it is enough to lock its
+		existence with LOCK_REC_NOT_GAP. */
+
+		/* If innodb_locks_unsafe_for_binlog option is used
+		or this session is using a READ COMMITED isolation
+		level we lock only the record, i.e., next-key locking is
+		not used. */
+
+		ulint	lock_type;
+
+		if (!set_also_gap_locks
+		    || srv_locks_unsafe_for_binlog
+		    || trx->isolation_level == TRX_ISO_READ_COMMITTED
+		    || (unique_search
+			&& !UNIV_UNLIKELY(rec_get_deleted_flag(rec, comp)))) {
+
+			goto no_gap_lock;
+		} else {
+			lock_type = LOCK_ORDINARY;
+		}
+
+		/* If we are doing a 'greater or equal than a primary key
+		value' search from a clustered index, and we find a record
+		that has that exact primary key value, then there is no need
+		to lock the gap before the record, because no insert in the
+		gap can be in our search range. That is, no phantom row can
+		appear that way.
+
+		An example: if col1 is the primary key, the search is WHERE
+		col1 >= 100, and we find a record where col1 = 100, then no
+		need to lock the gap before that record. */
+
+		if (index == clust_index
+		    && mode == PAGE_CUR_GE
+		    && direction == 0
+		    && dtuple_get_n_fields_cmp(search_tuple)
+		    == dict_index_get_n_unique(index)
+		    && 0 == cmp_dtuple_rec(search_tuple, rec, offsets)) {
+no_gap_lock:
+			lock_type = LOCK_REC_NOT_GAP;
+		}
+
+		err = sel_set_rec_lock(btr_pcur_get_block(pcur),
+				       rec, index, offsets,
+				       prebuilt->select_lock_type,
+				       lock_type, thr);
+
+		switch (err) {
+			const rec_t*	old_vers;
+		case DB_SUCCESS:
+			break;
+		case DB_LOCK_WAIT:
+			if (UNIV_LIKELY(prebuilt->row_read_type
+					!= ROW_READ_TRY_SEMI_CONSISTENT)
+			    || index != clust_index) {
+
+				goto lock_wait_or_error;
+			}
+
+			/* The following call returns 'offsets'
+			associated with 'old_vers' */
+			err = row_sel_build_committed_vers_for_mysql(
+				clust_index, prebuilt, rec,
+				&offsets, &heap, &old_vers, &mtr);
+
+			if (err != DB_SUCCESS) {
+
+				goto lock_wait_or_error;
+			}
+
+			mutex_enter(&kernel_mutex);
+			if (trx->was_chosen_as_deadlock_victim) {
+				mutex_exit(&kernel_mutex);
+				err = DB_DEADLOCK;
+
+				goto lock_wait_or_error;
+			}
+			if (UNIV_LIKELY(trx->wait_lock != NULL)) {
+				lock_cancel_waiting_and_release(
+					trx->wait_lock);
+				trx_reset_new_rec_lock_info(trx);
+			} else {
+				mutex_exit(&kernel_mutex);
+
+				/* The lock was granted while we were
+				searching for the last committed version.
+				Do a normal locking read. */
+
+				offsets = rec_get_offsets(rec, index, offsets,
+							  ULINT_UNDEFINED,
+							  &heap);
+				err = DB_SUCCESS;
+				break;
+			}
+			mutex_exit(&kernel_mutex);
+
+			if (old_vers == NULL) {
+				/* The row was not yet committed */
+
+				goto next_rec;
+			}
+
+			did_semi_consistent_read = TRUE;
+			rec = old_vers;
+			break;
+		default:
+
+			goto lock_wait_or_error;
+		}
+	} else {
+		/* This is a non-locking consistent read: if necessary, fetch
+		a previous version of the record */
+
+		if (trx->isolation_level == TRX_ISO_READ_UNCOMMITTED) {
+
+			/* Do nothing: we let a non-locking SELECT read the
+			latest version of the record */
+
+		} else if (index == clust_index) {
+
+			/* Fetch a previous version of the row if the current
+			one is not visible in the snapshot; if we have a very
+			high force recovery level set, we try to avoid crashes
+			by skipping this lookup */
+
+			if (UNIV_LIKELY(srv_force_recovery < 5)
+			    && !lock_clust_rec_cons_read_sees(
+				    rec, index, offsets, trx->read_view)) {
+
+				rec_t*	old_vers;
+				/* The following call returns 'offsets'
+				associated with 'old_vers' */
+				err = row_sel_build_prev_vers_for_mysql(
+					trx->read_view, clust_index,
+					prebuilt, rec, &offsets, &heap,
+					&old_vers, &mtr);
+
+				if (err != DB_SUCCESS) {
+
+					goto lock_wait_or_error;
+				}
+
+				if (old_vers == NULL) {
+					/* The row did not exist yet in
+					the read view */
+
+					goto next_rec;
+				}
+
+				rec = old_vers;
+			}
+		} else if (!lock_sec_rec_cons_read_sees(rec, trx->read_view)) {
+			/* We are looking into a non-clustered index,
+			and to get the right version of the record we
+			have to look also into the clustered index: this
+			is necessary, because we can only get the undo
+			information via the clustered index record. */
+
+			ut_ad(index != clust_index);
+
+			goto requires_clust_rec;
+		}
+	}
+
+	/* NOTE that at this point rec can be an old version of a clustered
+	index record built for a consistent read. We cannot assume after this
+	point that rec is on a buffer pool page. Functions like
+	page_rec_is_comp() cannot be used! */
+
+	if (UNIV_UNLIKELY(rec_get_deleted_flag(rec, comp))) {
+
+		/* The record is delete-marked: we can skip it */
+
+		if ((srv_locks_unsafe_for_binlog
+		     || trx->isolation_level == TRX_ISO_READ_COMMITTED)
+		    && prebuilt->select_lock_type != LOCK_NONE
+		    && !did_semi_consistent_read) {
+
+			/* No need to keep a lock on a delete-marked record
+			if we do not want to use next-key locking. */
+
+			row_unlock_for_mysql(prebuilt, TRUE);
+		}
+
+		/* This is an optimization to skip setting the next key lock
+		on the record that follows this delete-marked record. This
+		optimization works because of the unique search criteria
+		which precludes the presence of a range lock between this
+		delete marked record and the record following it.
+
+		For now this is applicable only to clustered indexes while
+		doing a unique search. There is scope for further optimization
+		applicable to unique secondary indexes. Current behaviour is
+		to widen the scope of a lock on an already delete marked record
+		if the same record is deleted twice by the same transaction */
+		if (index == clust_index && unique_search) {
+			err = DB_RECORD_NOT_FOUND;
+
+			goto normal_return;
+		}
+
+		goto next_rec;
+	}
+
+	/* Get the clustered index record if needed, if we did not do the
+	search using the clustered index. */
+
+	if (index != clust_index && prebuilt->need_to_access_clustered) {
+
+requires_clust_rec:
+		/* We use a 'goto' to the preceding label if a consistent
+		read of a secondary index record requires us to look up old
+		versions of the associated clustered index record. */
+
+		ut_ad(rec_offs_validate(rec, index, offsets));
+
+		/* It was a non-clustered index and we must fetch also the
+		clustered index record */
+
+		mtr_has_extra_clust_latch = TRUE;
+
+		/* The following call returns 'offsets' associated with
+		'clust_rec'. Note that 'clust_rec' can be an old version
+		built for a consistent read. */
+
+		err = row_sel_get_clust_rec_for_mysql(prebuilt, index, rec,
+						      thr, &clust_rec,
+						      &offsets, &heap, &mtr);
+		if (err != DB_SUCCESS) {
+
+			goto lock_wait_or_error;
+		}
+
+		if (clust_rec == NULL) {
+			/* The record did not exist in the read view */
+			ut_ad(prebuilt->select_lock_type == LOCK_NONE);
+
+			goto next_rec;
+		}
+
+		if (UNIV_UNLIKELY(rec_get_deleted_flag(clust_rec, comp))) {
+
+			/* The record is delete marked: we can skip it */
+
+			if ((srv_locks_unsafe_for_binlog
+			     || trx->isolation_level == TRX_ISO_READ_COMMITTED)
+			    && prebuilt->select_lock_type != LOCK_NONE) {
+
+				/* No need to keep a lock on a delete-marked
+				record if we do not want to use next-key
+				locking. */
+
+				row_unlock_for_mysql(prebuilt, TRUE);
+			}
+
+			goto next_rec;
+		}
+
+		if (prebuilt->need_to_access_clustered) {
+
+			result_rec = clust_rec;
+
+			ut_ad(rec_offs_validate(result_rec, clust_index,
+						offsets));
+		} else {
+			/* We used 'offsets' for the clust rec, recalculate
+			them for 'rec' */
+			offsets = rec_get_offsets(rec, index, offsets,
+						  ULINT_UNDEFINED, &heap);
+			result_rec = rec;
+		}
+	} else {
+		result_rec = rec;
+	}
+
+	/* We found a qualifying record 'result_rec'. At this point,
+	'offsets' are associated with 'result_rec'. */
+
+	ut_ad(rec_offs_validate(result_rec,
+				result_rec != rec ? clust_index : index,
+				offsets));
+
+	/* At this point, the clustered index record is protected
+	by a page latch that was acquired when pcur was positioned.
+	The latch will not be released until mtr_commit(&mtr). */
+
+	if ((match_mode == ROW_SEL_EXACT
+	     || prebuilt->n_rows_fetched >= MYSQL_FETCH_CACHE_THRESHOLD)
+	    && prebuilt->select_lock_type == LOCK_NONE
+	    && !prebuilt->templ_contains_blob
+	    && !prebuilt->clust_index_was_generated
+	    && !prebuilt->used_in_HANDLER
+	    && prebuilt->template_type
+	    != ROW_MYSQL_DUMMY_TEMPLATE) {
+
+		/* Inside an update, for example, we do not cache rows,
+		since we may use the cursor position to do the actual
+		update, that is why we require ...lock_type == LOCK_NONE.
+		Since we keep space in prebuilt only for the BLOBs of
+		a single row, we cannot cache rows in the case there
+		are BLOBs in the fields to be fetched. In HANDLER we do
+		not cache rows because there the cursor is a scrollable
+		cursor. */
+
+		row_sel_push_cache_row_for_mysql(prebuilt, result_rec,
+						 offsets);
+		if (prebuilt->n_fetch_cached == MYSQL_FETCH_CACHE_SIZE) {
+
+			goto got_row;
+		}
+
+		goto next_rec;
+	} else {
+		if (prebuilt->template_type == ROW_MYSQL_DUMMY_TEMPLATE) {
+			memcpy(buf + 4, result_rec
+			       - rec_offs_extra_size(offsets),
+			       rec_offs_size(offsets));
+			mach_write_to_4(buf,
+					rec_offs_extra_size(offsets) + 4);
+		} else {
+			if (!row_sel_store_mysql_rec(buf, prebuilt,
+						     result_rec, offsets)) {
+				err = DB_TOO_BIG_RECORD;
+
+				goto lock_wait_or_error;
+			}
+		}
+
+		if (prebuilt->clust_index_was_generated) {
+			if (result_rec != rec) {
+				offsets = rec_get_offsets(
+					rec, index, offsets, ULINT_UNDEFINED,
+					&heap);
+			}
+			row_sel_store_row_id_to_prebuilt(prebuilt, rec,
+							 index, offsets);
+		}
+	}
+
+	/* From this point on, 'offsets' are invalid. */
+
+got_row:
+	/* We have an optimization to save CPU time: if this is a consistent
+	read on a unique condition on the clustered index, then we do not
+	store the pcur position, because any fetch next or prev will anyway
+	return 'end of file'. Exceptions are locking reads and the MySQL
+	HANDLER command where the user can move the cursor with PREV or NEXT
+	even after a unique search. */
+
+	if (!unique_search_from_clust_index
+	    || prebuilt->select_lock_type != LOCK_NONE
+	    || prebuilt->used_in_HANDLER) {
+
+		/* Inside an update always store the cursor position */
+
+		btr_pcur_store_position(pcur, &mtr);
+	}
+
+	err = DB_SUCCESS;
+
+	goto normal_return;
+
+next_rec:
+	/* Reset the old and new "did semi-consistent read" flags. */
+	if (UNIV_UNLIKELY(prebuilt->row_read_type
+			  == ROW_READ_DID_SEMI_CONSISTENT)) {
+		prebuilt->row_read_type = ROW_READ_TRY_SEMI_CONSISTENT;
+	}
+	did_semi_consistent_read = FALSE;
+
+	if (UNIV_UNLIKELY(srv_locks_unsafe_for_binlog
+			  || trx->isolation_level == TRX_ISO_READ_COMMITTED)
+	    && prebuilt->select_lock_type != LOCK_NONE) {
+
+		trx_reset_new_rec_lock_info(trx);
+	}
+
+	/*-------------------------------------------------------------*/
+	/* PHASE 5: Move the cursor to the next index record */
+
+	if (UNIV_UNLIKELY(mtr_has_extra_clust_latch)) {
+		/* We must commit mtr if we are moving to the next
+		non-clustered index record, because we could break the
+		latching order if we would access a different clustered
+		index page right away without releasing the previous. */
+
+		btr_pcur_store_position(pcur, &mtr);
+
+		mtr_commit(&mtr);
+		mtr_has_extra_clust_latch = FALSE;
+
+		mtr_start(&mtr);
+		if (sel_restore_position_for_mysql(&same_user_rec,
+						   BTR_SEARCH_LEAF,
+						   pcur, moves_up, &mtr)) {
+#ifdef UNIV_SEARCH_DEBUG
+			cnt++;
+#endif /* UNIV_SEARCH_DEBUG */
+
+			goto rec_loop;
+		}
+	}
+
+	if (moves_up) {
+		if (UNIV_UNLIKELY(!btr_pcur_move_to_next(pcur, &mtr))) {
+not_moved:
+			btr_pcur_store_position(pcur, &mtr);
+
+			if (match_mode != 0) {
+				err = DB_RECORD_NOT_FOUND;
+			} else {
+				err = DB_END_OF_INDEX;
+			}
+
+			goto normal_return;
+		}
+	} else {
+		if (UNIV_UNLIKELY(!btr_pcur_move_to_prev(pcur, &mtr))) {
+			goto not_moved;
+		}
+	}
+
+#ifdef UNIV_SEARCH_DEBUG
+	cnt++;
+#endif /* UNIV_SEARCH_DEBUG */
+
+	goto rec_loop;
+
+lock_wait_or_error:
+	/* Reset the old and new "did semi-consistent read" flags. */
+	if (UNIV_UNLIKELY(prebuilt->row_read_type
+			  == ROW_READ_DID_SEMI_CONSISTENT)) {
+		prebuilt->row_read_type = ROW_READ_TRY_SEMI_CONSISTENT;
+	}
+	did_semi_consistent_read = FALSE;
+
+	/*-------------------------------------------------------------*/
+
+	btr_pcur_store_position(pcur, &mtr);
+
+	mtr_commit(&mtr);
+	mtr_has_extra_clust_latch = FALSE;
+
+	trx->error_state = err;
+
+	/* The following is a patch for MySQL */
+
+	que_thr_stop_for_mysql(thr);
+
+	thr->lock_state = QUE_THR_LOCK_ROW;
+
+	if (row_mysql_handle_errors(&err, trx, thr, NULL)) {
+		/* It was a lock wait, and it ended */
+
+		thr->lock_state = QUE_THR_LOCK_NOLOCK;
+		mtr_start(&mtr);
+
+		sel_restore_position_for_mysql(&same_user_rec,
+					       BTR_SEARCH_LEAF, pcur,
+					       moves_up, &mtr);
+
+		if ((srv_locks_unsafe_for_binlog
+		     || trx->isolation_level == TRX_ISO_READ_COMMITTED)
+		    && !same_user_rec) {
+
+			/* Since we were not able to restore the cursor
+			on the same user record, we cannot use
+			row_unlock_for_mysql() to unlock any records, and
+			we must thus reset the new rec lock info. Since
+			in lock0lock.c we have blocked the inheriting of gap
+			X-locks, we actually do not have any new record locks
+			set in this case.
+
+			Note that if we were able to restore on the 'same'
+			user record, it is still possible that we were actually
+			waiting on a delete-marked record, and meanwhile
+			it was removed by purge and inserted again by some
+			other user. But that is no problem, because in
+			rec_loop we will again try to set a lock, and
+			new_rec_lock_info in trx will be right at the end. */
+
+			trx_reset_new_rec_lock_info(trx);
+		}
+
+		mode = pcur->search_mode;
+
+		goto rec_loop;
+	}
+
+	thr->lock_state = QUE_THR_LOCK_NOLOCK;
+
+#ifdef UNIV_SEARCH_DEBUG
+	/*	fputs("Using ", stderr);
+	dict_index_name_print(stderr, index);
+	fprintf(stderr, " cnt %lu ret value %lu err\n", cnt, err); */
+#endif /* UNIV_SEARCH_DEBUG */
+	goto func_exit;
+
+normal_return:
+	/*-------------------------------------------------------------*/
+	que_thr_stop_for_mysql_no_error(thr, trx);
+
+	mtr_commit(&mtr);
+
+	if (prebuilt->n_fetch_cached > 0) {
+		row_sel_pop_cached_row_for_mysql(buf, prebuilt);
+
+		err = DB_SUCCESS;
+	}
+
+#ifdef UNIV_SEARCH_DEBUG
+	/*	fputs("Using ", stderr);
+	dict_index_name_print(stderr, index);
+	fprintf(stderr, " cnt %lu ret value %lu err\n", cnt, err); */
+#endif /* UNIV_SEARCH_DEBUG */
+	if (err == DB_SUCCESS) {
+		srv_n_rows_read++;
+	}
+
+func_exit:
+	trx->op_info = "";
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+
+	/* Set or reset the "did semi-consistent read" flag on return.
+	The flag did_semi_consistent_read is set if and only if
+	the record being returned was fetched with a semi-consistent read. */
+	ut_ad(prebuilt->row_read_type != ROW_READ_WITH_LOCKS
+	      || !did_semi_consistent_read);
+
+	if (UNIV_UNLIKELY(prebuilt->row_read_type != ROW_READ_WITH_LOCKS)) {
+		if (UNIV_UNLIKELY(did_semi_consistent_read)) {
+			prebuilt->row_read_type = ROW_READ_DID_SEMI_CONSISTENT;
+		} else {
+			prebuilt->row_read_type = ROW_READ_TRY_SEMI_CONSISTENT;
+		}
+	}
+	return(err);
+}
+
+/***********************************************************************
+Checks if MySQL at the moment is allowed for this table to retrieve a
+consistent read result, or store it to the query cache. */
+UNIV_INTERN
+ibool
+row_search_check_if_query_cache_permitted(
+/*======================================*/
+					/* out: TRUE if storing or retrieving
+					from the query cache is permitted */
+	trx_t*		trx,		/* in: transaction object */
+	const char*	norm_name)	/* in: concatenation of database name,
+					'/' char, table name */
+{
+	dict_table_t*	table;
+	ibool		ret	= FALSE;
+
+	table = dict_table_get(norm_name, FALSE);
+
+	if (table == NULL) {
+
+		return(FALSE);
+	}
+
+	mutex_enter(&kernel_mutex);
+
+	/* Start the transaction if it is not started yet */
+
+	trx_start_if_not_started_low(trx);
+
+	/* If there are locks on the table or some trx has invalidated the
+	cache up to our trx id, then ret = FALSE.
+	We do not check what type locks there are on the table, though only
+	IX type locks actually would require ret = FALSE. */
+
+	if (UT_LIST_GET_LEN(table->locks) == 0
+	    && ut_dulint_cmp(trx->id,
+			     table->query_cache_inv_trx_id) >= 0) {
+
+		ret = TRUE;
+
+		/* If the isolation level is high, assign a read view for the
+		transaction if it does not yet have one */
+
+		if (trx->isolation_level >= TRX_ISO_REPEATABLE_READ
+		    && !trx->read_view) {
+
+			trx->read_view = read_view_open_now(
+				trx->id, trx->global_read_view_heap);
+			trx->global_read_view = trx->read_view;
+		}
+	}
+
+	mutex_exit(&kernel_mutex);
+
+	return(ret);
+}
+
+/***********************************************************************
+Read the AUTOINC column from the current row. If the value is less than
+0 and the type is not unsigned then we reset the value to 0. */
+static
+ib_uint64_t
+row_search_autoinc_read_column(
+/*===========================*/
+					/* out: value read from the column */
+	dict_index_t*	index,		/* in: index to read from */
+	const rec_t*	rec,		/* in: current rec */
+	ulint		col_no,		/* in: column number */
+	ibool		unsigned_type)	/* in: signed or unsigned flag */
+{
+	ulint		len;
+	const byte*	data;
+	ib_uint64_t	value;
+	mem_heap_t*	heap = NULL;
+	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint*		offsets	= offsets_;
+
+	rec_offs_init(offsets_);
+
+	offsets = rec_get_offsets(rec, index, offsets, ULINT_UNDEFINED, &heap);
+
+	data = rec_get_nth_field(rec, offsets, col_no, &len);
+
+	ut_a(len != UNIV_SQL_NULL);
+	ut_a(len <= sizeof value);
+
+	/* we assume AUTOINC value cannot be negative */
+	value = mach_read_int_type(data, len, unsigned_type);
+
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+
+	if (!unsigned_type && (ib_int64_t) value < 0) {
+		value = 0;
+	}
+
+	return(value);
+}
+
+/***********************************************************************
+Get the last row. */
+static
+const rec_t*
+row_search_autoinc_get_rec(
+/*=======================*/
+					/* out: current rec or NULL */
+	btr_pcur_t*	pcur,		/* in: the current cursor */
+	mtr_t*		mtr)		/* in: mini transaction */
+{
+	do {
+		const rec_t* rec = btr_pcur_get_rec(pcur);
+
+		if (page_rec_is_user_rec(rec)) {
+			return(rec);
+		}
+	} while (btr_pcur_move_to_prev(pcur, mtr));
+
+	return(NULL);
+}
+
+/***********************************************************************
+Read the max AUTOINC value from an index. */
+UNIV_INTERN
+ulint
+row_search_max_autoinc(
+/*===================*/
+					/* out: DB_SUCCESS if all OK else
+					error code, DB_RECORD_NOT_FOUND if
+					column name can't be found in index */
+	dict_index_t*	index,		/* in: index to search */
+	const char*	col_name,	/* in: name of autoinc column */
+	ib_uint64_t*	value)		/* out: AUTOINC value read */
+{
+	ulint		i;
+	ulint		n_cols;
+	dict_field_t*	dfield = NULL;
+	ulint		error = DB_SUCCESS;
+
+	n_cols = dict_index_get_n_ordering_defined_by_user(index);
+
+	/* Search the index for the AUTOINC column name */
+	for (i = 0; i < n_cols; ++i) {
+		dfield = dict_index_get_nth_field(index, i);
+
+		if (strcmp(col_name, dfield->name) == 0) {
+			break;
+		}
+	}
+
+	*value = 0;
+
+	/* Must find the AUTOINC column name */
+	if (i < n_cols && dfield) {
+		mtr_t		mtr;
+		btr_pcur_t	pcur;
+
+		mtr_start(&mtr);
+
+		/* Open at the high/right end (FALSE), and INIT
+		cursor (TRUE) */
+		btr_pcur_open_at_index_side(
+			FALSE, index, BTR_SEARCH_LEAF, &pcur, TRUE, &mtr);
+
+		if (page_get_n_recs(btr_pcur_get_page(&pcur)) > 0) {
+			const rec_t*	rec;
+
+			rec = row_search_autoinc_get_rec(&pcur, &mtr);
+
+			if (rec != NULL) {
+				ibool unsigned_type = (
+					dfield->col->prtype & DATA_UNSIGNED);
+
+				*value = row_search_autoinc_read_column(
+					index, rec, i, unsigned_type);
+			}
+		}
+
+		btr_pcur_close(&pcur);
+
+		mtr_commit(&mtr);
+	} else {
+		error = DB_RECORD_NOT_FOUND;
+	}
+
+	return(error);
+}
diff --git a/storage/xtradb/row/row0uins.c b/storage/xtradb/row/row0uins.c
new file mode 100644
index 00000000000..69d6b2e6c2a
--- /dev/null
+++ b/storage/xtradb/row/row0uins.c
@@ -0,0 +1,350 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Fresh insert undo
+
+Created 2/25/1997 Heikki Tuuri
+*******************************************************/
+
+#include "row0uins.h"
+
+#ifdef UNIV_NONINL
+#include "row0uins.ic"
+#endif
+
+#include "dict0dict.h"
+#include "dict0boot.h"
+#include "dict0crea.h"
+#include "trx0undo.h"
+#include "trx0roll.h"
+#include "btr0btr.h"
+#include "mach0data.h"
+#include "row0undo.h"
+#include "row0vers.h"
+#include "trx0trx.h"
+#include "trx0rec.h"
+#include "row0row.h"
+#include "row0upd.h"
+#include "que0que.h"
+#include "ibuf0ibuf.h"
+#include "log0log.h"
+
+/*******************************************************************
+Removes a clustered index record. The pcur in node was positioned on the
+record, now it is detached. */
+static
+ulint
+row_undo_ins_remove_clust_rec(
+/*==========================*/
+				/* out: DB_SUCCESS or DB_OUT_OF_FILE_SPACE */
+	undo_node_t*	node)	/* in: undo node */
+{
+	btr_cur_t*	btr_cur;
+	ibool		success;
+	ulint		err;
+	ulint		n_tries		= 0;
+	mtr_t		mtr;
+
+	mtr_start(&mtr);
+
+	success = btr_pcur_restore_position(BTR_MODIFY_LEAF, &(node->pcur),
+					    &mtr);
+	ut_a(success);
+
+	if (ut_dulint_cmp(node->table->id, DICT_INDEXES_ID) == 0) {
+		ut_ad(node->trx->dict_operation_lock_mode == RW_X_LATCH);
+
+		/* Drop the index tree associated with the row in
+		SYS_INDEXES table: */
+
+		dict_drop_index_tree(btr_pcur_get_rec(&(node->pcur)), &mtr);
+
+		mtr_commit(&mtr);
+
+		mtr_start(&mtr);
+
+		success = btr_pcur_restore_position(BTR_MODIFY_LEAF,
+						    &(node->pcur), &mtr);
+		ut_a(success);
+	}
+
+	btr_cur = btr_pcur_get_btr_cur(&(node->pcur));
+
+	success = btr_cur_optimistic_delete(btr_cur, &mtr);
+
+	btr_pcur_commit_specify_mtr(&(node->pcur), &mtr);
+
+	if (success) {
+		trx_undo_rec_release(node->trx, node->undo_no);
+
+		return(DB_SUCCESS);
+	}
+retry:
+	/* If did not succeed, try pessimistic descent to tree */
+	mtr_start(&mtr);
+
+	success = btr_pcur_restore_position(BTR_MODIFY_TREE,
+					    &(node->pcur), &mtr);
+	ut_a(success);
+
+	btr_cur_pessimistic_delete(&err, FALSE, btr_cur,
+				   trx_is_recv(node->trx)
+				   ? RB_RECOVERY
+				   : RB_NORMAL, &mtr);
+
+	/* The delete operation may fail if we have little
+	file space left: TODO: easiest to crash the database
+	and restart with more file space */
+
+	if (err == DB_OUT_OF_FILE_SPACE
+	    && n_tries < BTR_CUR_RETRY_DELETE_N_TIMES) {
+
+		btr_pcur_commit_specify_mtr(&(node->pcur), &mtr);
+
+		n_tries++;
+
+		os_thread_sleep(BTR_CUR_RETRY_SLEEP_TIME);
+
+		goto retry;
+	}
+
+	btr_pcur_commit_specify_mtr(&(node->pcur), &mtr);
+
+	trx_undo_rec_release(node->trx, node->undo_no);
+
+	return(err);
+}
+
+/*******************************************************************
+Removes a secondary index entry if found. */
+static
+ulint
+row_undo_ins_remove_sec_low(
+/*========================*/
+				/* out: DB_SUCCESS, DB_FAIL, or
+				DB_OUT_OF_FILE_SPACE */
+	ulint		mode,	/* in: BTR_MODIFY_LEAF or BTR_MODIFY_TREE,
+				depending on whether we wish optimistic or
+				pessimistic descent down the index tree */
+	dict_index_t*	index,	/* in: index */
+	dtuple_t*	entry)	/* in: index entry to remove */
+{
+	btr_pcur_t	pcur;
+	btr_cur_t*	btr_cur;
+	ibool		found;
+	ibool		success;
+	ulint		err;
+	mtr_t		mtr;
+
+	log_free_check();
+	mtr_start(&mtr);
+
+	found = row_search_index_entry(index, entry, mode, &pcur, &mtr);
+
+	btr_cur = btr_pcur_get_btr_cur(&pcur);
+
+	if (!found) {
+		/* Not found */
+
+		btr_pcur_close(&pcur);
+		mtr_commit(&mtr);
+
+		return(DB_SUCCESS);
+	}
+
+	if (mode == BTR_MODIFY_LEAF) {
+		success = btr_cur_optimistic_delete(btr_cur, &mtr);
+
+		if (success) {
+			err = DB_SUCCESS;
+		} else {
+			err = DB_FAIL;
+		}
+	} else {
+		ut_ad(mode == BTR_MODIFY_TREE);
+
+		/* No need to distinguish RB_RECOVERY here, because we
+		are deleting a secondary index record: the distinction
+		between RB_NORMAL and RB_RECOVERY only matters when
+		deleting a record that contains externally stored
+		columns. */
+		ut_ad(!dict_index_is_clust(index));
+		btr_cur_pessimistic_delete(&err, FALSE, btr_cur,
+					   RB_NORMAL, &mtr);
+	}
+
+	btr_pcur_close(&pcur);
+	mtr_commit(&mtr);
+
+	return(err);
+}
+
+/*******************************************************************
+Removes a secondary index entry from the index if found. Tries first
+optimistic, then pessimistic descent down the tree. */
+static
+ulint
+row_undo_ins_remove_sec(
+/*====================*/
+				/* out: DB_SUCCESS or DB_OUT_OF_FILE_SPACE */
+	dict_index_t*	index,	/* in: index */
+	dtuple_t*	entry)	/* in: index entry to insert */
+{
+	ulint	err;
+	ulint	n_tries	= 0;
+
+	/* Try first optimistic descent to the B-tree */
+
+	err = row_undo_ins_remove_sec_low(BTR_MODIFY_LEAF, index, entry);
+
+	if (err == DB_SUCCESS) {
+
+		return(err);
+	}
+
+	/* Try then pessimistic descent to the B-tree */
+retry:
+	err = row_undo_ins_remove_sec_low(BTR_MODIFY_TREE, index, entry);
+
+	/* The delete operation may fail if we have little
+	file space left: TODO: easiest to crash the database
+	and restart with more file space */
+
+	if (err != DB_SUCCESS && n_tries < BTR_CUR_RETRY_DELETE_N_TIMES) {
+
+		n_tries++;
+
+		os_thread_sleep(BTR_CUR_RETRY_SLEEP_TIME);
+
+		goto retry;
+	}
+
+	return(err);
+}
+
+/***************************************************************
+Parses the row reference and other info in a fresh insert undo record. */
+static
+void
+row_undo_ins_parse_undo_rec(
+/*========================*/
+	undo_node_t*	node)	/* in/out: row undo node */
+{
+	dict_index_t*	clust_index;
+	byte*		ptr;
+	dulint		undo_no;
+	dulint		table_id;
+	ulint		type;
+	ulint		dummy;
+	ibool		dummy_extern;
+
+	ut_ad(node);
+
+	ptr = trx_undo_rec_get_pars(node->undo_rec, &type, &dummy,
+				    &dummy_extern, &undo_no, &table_id);
+	ut_ad(type == TRX_UNDO_INSERT_REC);
+	node->rec_type = type;
+
+	node->update = NULL;
+	node->table = dict_table_get_on_id(table_id, node->trx);
+
+	/* Skip the UNDO if we can't find the table or the .ibd file. */
+	if (UNIV_UNLIKELY(node->table == NULL)) {
+	} else if (UNIV_UNLIKELY(node->table->ibd_file_missing)) {
+		node->table = NULL;
+	} else {
+		clust_index = dict_table_get_first_index(node->table);
+
+		if (clust_index != NULL) {
+			ptr = trx_undo_rec_get_row_ref(
+				ptr, clust_index, &node->ref, node->heap);
+		} else {
+			ut_print_timestamp(stderr);
+			fprintf(stderr, "  InnoDB: table ");
+			ut_print_name(stderr, node->trx, TRUE,
+				      node->table->name);
+			fprintf(stderr, " has no indexes, "
+				"ignoring the table\n");
+
+			node->table = NULL;
+		}
+	}
+}
+
+/***************************************************************
+Undoes a fresh insert of a row to a table. A fresh insert means that
+the same clustered index unique key did not have any record, even delete
+marked, at the time of the insert.  InnoDB is eager in a rollback:
+if it figures out that an index record will be removed in the purge
+anyway, it will remove it in the rollback. */
+UNIV_INTERN
+ulint
+row_undo_ins(
+/*=========*/
+				/* out: DB_SUCCESS or DB_OUT_OF_FILE_SPACE */
+	undo_node_t*	node)	/* in: row undo node */
+{
+	ut_ad(node);
+	ut_ad(node->state == UNDO_NODE_INSERT);
+
+	row_undo_ins_parse_undo_rec(node);
+
+	if (!node->table || !row_undo_search_clust_to_pcur(node)) {
+		trx_undo_rec_release(node->trx, node->undo_no);
+
+		return(DB_SUCCESS);
+	}
+
+	/* Iterate over all the indexes and undo the insert.*/
+
+	/* Skip the clustered index (the first index) */
+	node->index = dict_table_get_next_index(
+		dict_table_get_first_index(node->table));
+
+	while (node->index != NULL) {
+		dtuple_t*	entry;
+		ulint		err;
+
+		entry = row_build_index_entry(node->row, node->ext,
+					      node->index, node->heap);
+		if (UNIV_UNLIKELY(!entry)) {
+			/* The database must have crashed after
+			inserting a clustered index record but before
+			writing all the externally stored columns of
+			that record.  Because secondary index entries
+			are inserted after the clustered index record,
+			we may assume that the secondary index record
+			does not exist.  However, this situation may
+			only occur during the rollback of incomplete
+			transactions. */
+			ut_a(trx_is_recv(node->trx));
+		} else {
+			err = row_undo_ins_remove_sec(node->index, entry);
+
+			if (err != DB_SUCCESS) {
+
+				return(err);
+			}
+		}
+
+		node->index = dict_table_get_next_index(node->index);
+	}
+
+	return(row_undo_ins_remove_clust_rec(node));
+}
diff --git a/storage/xtradb/row/row0umod.c b/storage/xtradb/row/row0umod.c
new file mode 100644
index 00000000000..835f357fc8d
--- /dev/null
+++ b/storage/xtradb/row/row0umod.c
@@ -0,0 +1,820 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Undo modify of a row
+
+Created 2/27/1997 Heikki Tuuri
+*******************************************************/
+
+#include "row0umod.h"
+
+#ifdef UNIV_NONINL
+#include "row0umod.ic"
+#endif
+
+#include "dict0dict.h"
+#include "dict0boot.h"
+#include "trx0undo.h"
+#include "trx0roll.h"
+#include "btr0btr.h"
+#include "mach0data.h"
+#include "row0undo.h"
+#include "row0vers.h"
+#include "trx0trx.h"
+#include "trx0rec.h"
+#include "row0row.h"
+#include "row0upd.h"
+#include "que0que.h"
+#include "log0log.h"
+
+/* Considerations on undoing a modify operation.
+(1) Undoing a delete marking: all index records should be found. Some of
+them may have delete mark already FALSE, if the delete mark operation was
+stopped underway, or if the undo operation ended prematurely because of a
+system crash.
+(2) Undoing an update of a delete unmarked record: the newer version of
+an updated secondary index entry should be removed if no prior version
+of the clustered index record requires its existence. Otherwise, it should
+be delete marked.
+(3) Undoing an update of a delete marked record. In this kind of update a
+delete marked clustered index record was delete unmarked and possibly also
+some of its fields were changed. Now, it is possible that the delete marked
+version has become obsolete at the time the undo is started. */
+
+/***************************************************************
+Checks if also the previous version of the clustered index record was
+modified or inserted by the same transaction, and its undo number is such
+that it should be undone in the same rollback. */
+UNIV_INLINE
+ibool
+row_undo_mod_undo_also_prev_vers(
+/*=============================*/
+				/* out: TRUE if also previous modify or
+				insert of this row should be undone */
+	undo_node_t*	node,	/* in: row undo node */
+	dulint*		undo_no)/* out: the undo number */
+{
+	trx_undo_rec_t*	undo_rec;
+	trx_t*		trx;
+
+	trx = node->trx;
+
+	if (0 != ut_dulint_cmp(node->new_trx_id, trx->id)) {
+
+		*undo_no = ut_dulint_zero;
+		return(FALSE);
+	}
+
+	undo_rec = trx_undo_get_undo_rec_low(node->new_roll_ptr, node->heap);
+
+	*undo_no = trx_undo_rec_get_undo_no(undo_rec);
+
+	return(ut_dulint_cmp(trx->roll_limit, *undo_no) <= 0);
+}
+
+/***************************************************************
+Undoes a modify in a clustered index record. */
+static
+ulint
+row_undo_mod_clust_low(
+/*===================*/
+				/* out: DB_SUCCESS, DB_FAIL, or error code:
+				we may run out of file space */
+	undo_node_t*	node,	/* in: row undo node */
+	que_thr_t*	thr,	/* in: query thread */
+	mtr_t*		mtr,	/* in: mtr; must be committed before
+				latching any further pages */
+	ulint		mode)	/* in: BTR_MODIFY_LEAF or BTR_MODIFY_TREE */
+{
+	btr_pcur_t*	pcur;
+	btr_cur_t*	btr_cur;
+	ulint		err;
+	ibool		success;
+
+	pcur = &(node->pcur);
+	btr_cur = btr_pcur_get_btr_cur(pcur);
+
+	success = btr_pcur_restore_position(mode, pcur, mtr);
+
+	ut_ad(success);
+
+	if (mode == BTR_MODIFY_LEAF) {
+
+		err = btr_cur_optimistic_update(BTR_NO_LOCKING_FLAG
+						| BTR_NO_UNDO_LOG_FLAG
+						| BTR_KEEP_SYS_FLAG,
+						btr_cur, node->update,
+						node->cmpl_info, thr, mtr);
+	} else {
+		mem_heap_t*	heap		= NULL;
+		big_rec_t*	dummy_big_rec;
+
+		ut_ad(mode == BTR_MODIFY_TREE);
+
+		err = btr_cur_pessimistic_update(
+			BTR_NO_LOCKING_FLAG
+			| BTR_NO_UNDO_LOG_FLAG
+			| BTR_KEEP_SYS_FLAG,
+			btr_cur, &heap, &dummy_big_rec, node->update,
+			node->cmpl_info, thr, mtr);
+
+		ut_a(!dummy_big_rec);
+		if (UNIV_LIKELY_NULL(heap)) {
+			mem_heap_free(heap);
+		}
+	}
+
+	return(err);
+}
+
+/***************************************************************
+Removes a clustered index record after undo if possible. */
+static
+ulint
+row_undo_mod_remove_clust_low(
+/*==========================*/
+				/* out: DB_SUCCESS, DB_FAIL, or error code:
+				we may run out of file space */
+	undo_node_t*	node,	/* in: row undo node */
+	que_thr_t*	thr __attribute__((unused)), /* in: query thread */
+	mtr_t*		mtr,	/* in: mtr */
+	ulint		mode)	/* in: BTR_MODIFY_LEAF or BTR_MODIFY_TREE */
+{
+	btr_pcur_t*	pcur;
+	btr_cur_t*	btr_cur;
+	ulint		err;
+	ibool		success;
+
+	pcur = &(node->pcur);
+	btr_cur = btr_pcur_get_btr_cur(pcur);
+
+	success = btr_pcur_restore_position(mode, pcur, mtr);
+
+	if (!success) {
+
+		return(DB_SUCCESS);
+	}
+
+	/* Find out if we can remove the whole clustered index record */
+
+	if (node->rec_type == TRX_UNDO_UPD_DEL_REC
+	    && !row_vers_must_preserve_del_marked(node->new_trx_id, mtr)) {
+
+		/* Ok, we can remove */
+	} else {
+		return(DB_SUCCESS);
+	}
+
+	if (mode == BTR_MODIFY_LEAF) {
+		success = btr_cur_optimistic_delete(btr_cur, mtr);
+
+		if (success) {
+			err = DB_SUCCESS;
+		} else {
+			err = DB_FAIL;
+		}
+	} else {
+		ut_ad(mode == BTR_MODIFY_TREE);
+
+		/* Note that since this operation is analogous to purge,
+		we can free also inherited externally stored fields:
+		hence the RB_NONE in the call below */
+
+		btr_cur_pessimistic_delete(&err, FALSE, btr_cur, RB_NONE, mtr);
+
+		/* The delete operation may fail if we have little
+		file space left: TODO: easiest to crash the database
+		and restart with more file space */
+	}
+
+	return(err);
+}
+
+/***************************************************************
+Undoes a modify in a clustered index record. Sets also the node state for the
+next round of undo. */
+static
+ulint
+row_undo_mod_clust(
+/*===============*/
+				/* out: DB_SUCCESS or error code: we may run
+				out of file space */
+	undo_node_t*	node,	/* in: row undo node */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	btr_pcur_t*	pcur;
+	mtr_t		mtr;
+	ulint		err;
+	ibool		success;
+	ibool		more_vers;
+	dulint		new_undo_no;
+
+	ut_ad(node && thr);
+
+	/* Check if also the previous version of the clustered index record
+	should be undone in this same rollback operation */
+
+	more_vers = row_undo_mod_undo_also_prev_vers(node, &new_undo_no);
+
+	pcur = &(node->pcur);
+
+	mtr_start(&mtr);
+
+	/* Try optimistic processing of the record, keeping changes within
+	the index page */
+
+	err = row_undo_mod_clust_low(node, thr, &mtr, BTR_MODIFY_LEAF);
+
+	if (err != DB_SUCCESS) {
+		btr_pcur_commit_specify_mtr(pcur, &mtr);
+
+		/* We may have to modify tree structure: do a pessimistic
+		descent down the index tree */
+
+		mtr_start(&mtr);
+
+		err = row_undo_mod_clust_low(node, thr, &mtr, BTR_MODIFY_TREE);
+	}
+
+	btr_pcur_commit_specify_mtr(pcur, &mtr);
+
+	if (err == DB_SUCCESS && node->rec_type == TRX_UNDO_UPD_DEL_REC) {
+
+		mtr_start(&mtr);
+
+		err = row_undo_mod_remove_clust_low(node, thr, &mtr,
+						    BTR_MODIFY_LEAF);
+		if (err != DB_SUCCESS) {
+			btr_pcur_commit_specify_mtr(pcur, &mtr);
+
+			/* We may have to modify tree structure: do a
+			pessimistic descent down the index tree */
+
+			mtr_start(&mtr);
+
+			err = row_undo_mod_remove_clust_low(node, thr, &mtr,
+							    BTR_MODIFY_TREE);
+		}
+
+		btr_pcur_commit_specify_mtr(pcur, &mtr);
+	}
+
+	node->state = UNDO_NODE_FETCH_NEXT;
+
+	trx_undo_rec_release(node->trx, node->undo_no);
+
+	if (more_vers && err == DB_SUCCESS) {
+
+		/* Reserve the undo log record to the prior version after
+		committing &mtr: this is necessary to comply with the latching
+		order, as &mtr may contain the fsp latch which is lower in
+		the latch hierarchy than trx->undo_mutex. */
+
+		success = trx_undo_rec_reserve(node->trx, new_undo_no);
+
+		if (success) {
+			node->state = UNDO_NODE_PREV_VERS;
+		}
+	}
+
+	return(err);
+}
+
+/***************************************************************
+Delete marks or removes a secondary index entry if found. */
+static
+ulint
+row_undo_mod_del_mark_or_remove_sec_low(
+/*====================================*/
+				/* out: DB_SUCCESS, DB_FAIL, or
+				DB_OUT_OF_FILE_SPACE */
+	undo_node_t*	node,	/* in: row undo node */
+	que_thr_t*	thr,	/* in: query thread */
+	dict_index_t*	index,	/* in: index */
+	dtuple_t*	entry,	/* in: index entry */
+	ulint		mode)	/* in: latch mode BTR_MODIFY_LEAF or
+				BTR_MODIFY_TREE */
+{
+	ibool		found;
+	btr_pcur_t	pcur;
+	btr_cur_t*	btr_cur;
+	ibool		success;
+	ibool		old_has;
+	ulint		err;
+	mtr_t		mtr;
+	mtr_t		mtr_vers;
+
+	log_free_check();
+	mtr_start(&mtr);
+
+	found = row_search_index_entry(index, entry, mode, &pcur, &mtr);
+
+	btr_cur = btr_pcur_get_btr_cur(&pcur);
+
+	if (!found) {
+		/* In crash recovery, the secondary index record may
+		be missing if the UPDATE did not have time to insert
+		the secondary index records before the crash.  When we
+		are undoing that UPDATE in crash recovery, the record
+		may be missing.
+
+		In normal processing, if an update ends in a deadlock
+		before it has inserted all updated secondary index
+		records, then the undo will not find those records. */
+
+		btr_pcur_close(&pcur);
+		mtr_commit(&mtr);
+
+		return(DB_SUCCESS);
+	}
+
+	/* We should remove the index record if no prior version of the row,
+	which cannot be purged yet, requires its existence. If some requires,
+	we should delete mark the record. */
+
+	mtr_start(&mtr_vers);
+
+	success = btr_pcur_restore_position(BTR_SEARCH_LEAF, &(node->pcur),
+					    &mtr_vers);
+	ut_a(success);
+
+	old_has = row_vers_old_has_index_entry(FALSE,
+					       btr_pcur_get_rec(&(node->pcur)),
+					       &mtr_vers, index, entry);
+	if (old_has) {
+		err = btr_cur_del_mark_set_sec_rec(BTR_NO_LOCKING_FLAG,
+						   btr_cur, TRUE, thr, &mtr);
+		ut_ad(err == DB_SUCCESS);
+	} else {
+		/* Remove the index record */
+
+		if (mode == BTR_MODIFY_LEAF) {
+			success = btr_cur_optimistic_delete(btr_cur, &mtr);
+			if (success) {
+				err = DB_SUCCESS;
+			} else {
+				err = DB_FAIL;
+			}
+		} else {
+			ut_ad(mode == BTR_MODIFY_TREE);
+
+			/* No need to distinguish RB_RECOVERY here, because we
+			are deleting a secondary index record: the distinction
+			between RB_NORMAL and RB_RECOVERY only matters when
+			deleting a record that contains externally stored
+			columns. */
+			ut_ad(!dict_index_is_clust(index));
+			btr_cur_pessimistic_delete(&err, FALSE, btr_cur,
+						   RB_NORMAL, &mtr);
+
+			/* The delete operation may fail if we have little
+			file space left: TODO: easiest to crash the database
+			and restart with more file space */
+		}
+	}
+
+	btr_pcur_commit_specify_mtr(&(node->pcur), &mtr_vers);
+	btr_pcur_close(&pcur);
+	mtr_commit(&mtr);
+
+	return(err);
+}
+
+/***************************************************************
+Delete marks or removes a secondary index entry if found.
+NOTE that if we updated the fields of a delete-marked secondary index record
+so that alphabetically they stayed the same, e.g., 'abc' -> 'aBc', we cannot
+return to the original values because we do not know them. But this should
+not cause problems because in row0sel.c, in queries we always retrieve the
+clustered index record or an earlier version of it, if the secondary index
+record through which we do the search is delete-marked. */
+static
+ulint
+row_undo_mod_del_mark_or_remove_sec(
+/*================================*/
+				/* out: DB_SUCCESS or DB_OUT_OF_FILE_SPACE */
+	undo_node_t*	node,	/* in: row undo node */
+	que_thr_t*	thr,	/* in: query thread */
+	dict_index_t*	index,	/* in: index */
+	dtuple_t*	entry)	/* in: index entry */
+{
+	ulint	err;
+
+	err = row_undo_mod_del_mark_or_remove_sec_low(node, thr, index,
+						      entry, BTR_MODIFY_LEAF);
+	if (err == DB_SUCCESS) {
+
+		return(err);
+	}
+
+	err = row_undo_mod_del_mark_or_remove_sec_low(node, thr, index,
+						      entry, BTR_MODIFY_TREE);
+	return(err);
+}
+
+/***************************************************************
+Delete unmarks a secondary index entry which must be found. It might not be
+delete-marked at the moment, but it does not harm to unmark it anyway. We also
+need to update the fields of the secondary index record if we updated its
+fields but alphabetically they stayed the same, e.g., 'abc' -> 'aBc'. */
+static
+ulint
+row_undo_mod_del_unmark_sec_and_undo_update(
+/*========================================*/
+				/* out: DB_FAIL or DB_SUCCESS or
+				DB_OUT_OF_FILE_SPACE */
+	ulint		mode,	/* in: search mode: BTR_MODIFY_LEAF or
+				BTR_MODIFY_TREE */
+	que_thr_t*	thr,	/* in: query thread */
+	dict_index_t*	index,	/* in: index */
+	dtuple_t*	entry)	/* in: index entry */
+{
+	mem_heap_t*	heap;
+	btr_pcur_t	pcur;
+	upd_t*		update;
+	ulint		err		= DB_SUCCESS;
+	big_rec_t*	dummy_big_rec;
+	mtr_t		mtr;
+	trx_t*		trx		= thr_get_trx(thr);
+
+	/* Ignore indexes that are being created. */
+	if (UNIV_UNLIKELY(*index->name == TEMP_INDEX_PREFIX)) {
+
+		return(DB_SUCCESS);
+	}
+
+	log_free_check();
+	mtr_start(&mtr);
+
+	if (UNIV_UNLIKELY(!row_search_index_entry(index, entry,
+						  mode, &pcur, &mtr))) {
+		fputs("InnoDB: error in sec index entry del undo in\n"
+		      "InnoDB: ", stderr);
+		dict_index_name_print(stderr, trx, index);
+		fputs("\n"
+		      "InnoDB: tuple ", stderr);
+		dtuple_print(stderr, entry);
+		fputs("\n"
+		      "InnoDB: record ", stderr);
+		rec_print(stderr, btr_pcur_get_rec(&pcur), index);
+		putc('\n', stderr);
+		trx_print(stderr, trx, 0);
+		fputs("\n"
+		      "InnoDB: Submit a detailed bug report"
+		      " to http://bugs.mysql.com\n", stderr);
+	} else {
+		btr_cur_t*	btr_cur = btr_pcur_get_btr_cur(&pcur);
+
+		err = btr_cur_del_mark_set_sec_rec(BTR_NO_LOCKING_FLAG,
+						   btr_cur, FALSE, thr, &mtr);
+		ut_a(err == DB_SUCCESS);
+		heap = mem_heap_create(100);
+
+		update = row_upd_build_sec_rec_difference_binary(
+			index, entry, btr_cur_get_rec(btr_cur), trx, heap);
+		if (upd_get_n_fields(update) == 0) {
+
+			/* Do nothing */
+
+		} else if (mode == BTR_MODIFY_LEAF) {
+			/* Try an optimistic updating of the record, keeping
+			changes within the page */
+
+			err = btr_cur_optimistic_update(
+				BTR_KEEP_SYS_FLAG | BTR_NO_LOCKING_FLAG,
+				btr_cur, update, 0, thr, &mtr);
+			switch (err) {
+			case DB_OVERFLOW:
+			case DB_UNDERFLOW:
+			case DB_ZIP_OVERFLOW:
+				err = DB_FAIL;
+			}
+		} else {
+			ut_a(mode == BTR_MODIFY_TREE);
+			err = btr_cur_pessimistic_update(
+				BTR_KEEP_SYS_FLAG | BTR_NO_LOCKING_FLAG,
+				btr_cur, &heap, &dummy_big_rec,
+				update, 0, thr, &mtr);
+			ut_a(!dummy_big_rec);
+		}
+
+		mem_heap_free(heap);
+	}
+
+	btr_pcur_close(&pcur);
+	mtr_commit(&mtr);
+
+	return(err);
+}
+
+/***************************************************************
+Undoes a modify in secondary indexes when undo record type is UPD_DEL. */
+static
+ulint
+row_undo_mod_upd_del_sec(
+/*=====================*/
+				/* out: DB_SUCCESS or DB_OUT_OF_FILE_SPACE */
+	undo_node_t*	node,	/* in: row undo node */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	mem_heap_t*	heap;
+	dtuple_t*	entry;
+	dict_index_t*	index;
+	ulint		err	= DB_SUCCESS;
+
+	heap = mem_heap_create(1024);
+
+	while (node->index != NULL) {
+		index = node->index;
+
+		entry = row_build_index_entry(node->row, node->ext,
+					      index, heap);
+		if (UNIV_UNLIKELY(!entry)) {
+			/* The database must have crashed after
+			inserting a clustered index record but before
+			writing all the externally stored columns of
+			that record.  Because secondary index entries
+			are inserted after the clustered index record,
+			we may assume that the secondary index record
+			does not exist.  However, this situation may
+			only occur during the rollback of incomplete
+			transactions. */
+			ut_a(trx_is_recv(thr_get_trx(thr)));
+		} else {
+			err = row_undo_mod_del_mark_or_remove_sec(
+				node, thr, index, entry);
+
+			if (err != DB_SUCCESS) {
+
+				break;
+			}
+		}
+
+		mem_heap_empty(heap);
+
+		node->index = dict_table_get_next_index(node->index);
+	}
+
+	mem_heap_free(heap);
+
+	return(err);
+}
+
+/***************************************************************
+Undoes a modify in secondary indexes when undo record type is DEL_MARK. */
+static
+ulint
+row_undo_mod_del_mark_sec(
+/*======================*/
+				/* out: DB_SUCCESS or DB_OUT_OF_FILE_SPACE */
+	undo_node_t*	node,	/* in: row undo node */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	mem_heap_t*	heap;
+	dtuple_t*	entry;
+	dict_index_t*	index;
+	ulint		err;
+
+	heap = mem_heap_create(1024);
+
+	while (node->index != NULL) {
+		index = node->index;
+
+		entry = row_build_index_entry(node->row, node->ext,
+					      index, heap);
+		ut_a(entry);
+		err = row_undo_mod_del_unmark_sec_and_undo_update(
+			BTR_MODIFY_LEAF, thr, index, entry);
+		if (err == DB_FAIL) {
+			err = row_undo_mod_del_unmark_sec_and_undo_update(
+				BTR_MODIFY_TREE, thr, index, entry);
+		}
+
+		if (err != DB_SUCCESS) {
+
+			mem_heap_free(heap);
+
+			return(err);
+		}
+
+		node->index = dict_table_get_next_index(node->index);
+	}
+
+	mem_heap_free(heap);
+
+	return(DB_SUCCESS);
+}
+
+/***************************************************************
+Undoes a modify in secondary indexes when undo record type is UPD_EXIST. */
+static
+ulint
+row_undo_mod_upd_exist_sec(
+/*=======================*/
+				/* out: DB_SUCCESS or DB_OUT_OF_FILE_SPACE */
+	undo_node_t*	node,	/* in: row undo node */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	mem_heap_t*	heap;
+	dtuple_t*	entry;
+	dict_index_t*	index;
+	ulint		err;
+
+	if (node->cmpl_info & UPD_NODE_NO_ORD_CHANGE) {
+		/* No change in secondary indexes */
+
+		return(DB_SUCCESS);
+	}
+
+	heap = mem_heap_create(1024);
+
+	while (node->index != NULL) {
+		index = node->index;
+
+		if (row_upd_changes_ord_field_binary(node->row, node->index,
+						     node->update)) {
+
+			/* Build the newest version of the index entry */
+			entry = row_build_index_entry(node->row, node->ext,
+						      index, heap);
+			ut_a(entry);
+			/* NOTE that if we updated the fields of a
+			delete-marked secondary index record so that
+			alphabetically they stayed the same, e.g.,
+			'abc' -> 'aBc', we cannot return to the original
+			values because we do not know them. But this should
+			not cause problems because in row0sel.c, in queries
+			we always retrieve the clustered index record or an
+			earlier version of it, if the secondary index record
+			through which we do the search is delete-marked. */
+
+			err = row_undo_mod_del_mark_or_remove_sec(node, thr,
+								  index,
+								  entry);
+			if (err != DB_SUCCESS) {
+				mem_heap_free(heap);
+
+				return(err);
+			}
+
+			/* We may have to update the delete mark in the
+			secondary index record of the previous version of
+			the row. We also need to update the fields of
+			the secondary index record if we updated its fields
+			but alphabetically they stayed the same, e.g.,
+			'abc' -> 'aBc'. */
+			mem_heap_empty(heap);
+			entry = row_build_index_entry(node->undo_row,
+						      node->undo_ext,
+						      index, heap);
+			ut_a(entry);
+
+			err = row_undo_mod_del_unmark_sec_and_undo_update(
+				BTR_MODIFY_LEAF, thr, index, entry);
+			if (err == DB_FAIL) {
+				err = row_undo_mod_del_unmark_sec_and_undo_update(
+					BTR_MODIFY_TREE, thr, index, entry);
+			}
+
+			if (err != DB_SUCCESS) {
+				mem_heap_free(heap);
+
+				return(err);
+			}
+		}
+
+		node->index = dict_table_get_next_index(node->index);
+	}
+
+	mem_heap_free(heap);
+
+	return(DB_SUCCESS);
+}
+
+/***************************************************************
+Parses the row reference and other info in a modify undo log record. */
+static
+void
+row_undo_mod_parse_undo_rec(
+/*========================*/
+	undo_node_t*	node,	/* in: row undo node */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	dict_index_t*	clust_index;
+	byte*		ptr;
+	dulint		undo_no;
+	dulint		table_id;
+	dulint		trx_id;
+	dulint		roll_ptr;
+	ulint		info_bits;
+	ulint		type;
+	ulint		cmpl_info;
+	ibool		dummy_extern;
+	trx_t*		trx;
+
+	ut_ad(node && thr);
+	trx = thr_get_trx(thr);
+	ptr = trx_undo_rec_get_pars(node->undo_rec, &type, &cmpl_info,
+				    &dummy_extern, &undo_no, &table_id);
+	node->rec_type = type;
+
+	node->table = dict_table_get_on_id(table_id, trx);
+
+	/* TODO: other fixes associated with DROP TABLE + rollback in the
+	same table by another user */
+
+	if (node->table == NULL) {
+		/* Table was dropped */
+		return;
+	}
+
+	if (node->table->ibd_file_missing) {
+		/* We skip undo operations to missing .ibd files */
+		node->table = NULL;
+
+		return;
+	}
+
+	clust_index = dict_table_get_first_index(node->table);
+
+	ptr = trx_undo_update_rec_get_sys_cols(ptr, &trx_id, &roll_ptr,
+					       &info_bits);
+
+	ptr = trx_undo_rec_get_row_ref(ptr, clust_index, &(node->ref),
+				       node->heap);
+
+	trx_undo_update_rec_get_update(ptr, clust_index, type, trx_id,
+				       roll_ptr, info_bits, trx,
+				       node->heap, &(node->update));
+	node->new_roll_ptr = roll_ptr;
+	node->new_trx_id = trx_id;
+	node->cmpl_info = cmpl_info;
+}
+
+/***************************************************************
+Undoes a modify operation on a row of a table. */
+UNIV_INTERN
+ulint
+row_undo_mod(
+/*=========*/
+				/* out: DB_SUCCESS or error code */
+	undo_node_t*	node,	/* in: row undo node */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	ulint	err;
+
+	ut_ad(node && thr);
+	ut_ad(node->state == UNDO_NODE_MODIFY);
+
+	row_undo_mod_parse_undo_rec(node, thr);
+
+	if (!node->table || !row_undo_search_clust_to_pcur(node)) {
+		/* It is already undone, or will be undone by another query
+		thread, or table was dropped */
+
+		trx_undo_rec_release(node->trx, node->undo_no);
+		node->state = UNDO_NODE_FETCH_NEXT;
+
+		return(DB_SUCCESS);
+	}
+
+	node->index = dict_table_get_next_index(
+		dict_table_get_first_index(node->table));
+
+	if (node->rec_type == TRX_UNDO_UPD_EXIST_REC) {
+
+		err = row_undo_mod_upd_exist_sec(node, thr);
+
+	} else if (node->rec_type == TRX_UNDO_DEL_MARK_REC) {
+
+		err = row_undo_mod_del_mark_sec(node, thr);
+	} else {
+		ut_ad(node->rec_type == TRX_UNDO_UPD_DEL_REC);
+		err = row_undo_mod_upd_del_sec(node, thr);
+	}
+
+	if (err != DB_SUCCESS) {
+
+		return(err);
+	}
+
+	err = row_undo_mod_clust(node, thr);
+
+	return(err);
+}
diff --git a/storage/xtradb/row/row0undo.c b/storage/xtradb/row/row0undo.c
new file mode 100644
index 00000000000..d372f88e207
--- /dev/null
+++ b/storage/xtradb/row/row0undo.c
@@ -0,0 +1,378 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Row undo
+
+Created 1/8/1997 Heikki Tuuri
+*******************************************************/
+
+#include "row0undo.h"
+
+#ifdef UNIV_NONINL
+#include "row0undo.ic"
+#endif
+
+#include "fsp0fsp.h"
+#include "mach0data.h"
+#include "trx0rseg.h"
+#include "trx0trx.h"
+#include "trx0roll.h"
+#include "trx0undo.h"
+#include "trx0purge.h"
+#include "trx0rec.h"
+#include "que0que.h"
+#include "row0row.h"
+#include "row0uins.h"
+#include "row0umod.h"
+#include "row0upd.h"
+#include "row0mysql.h"
+#include "srv0srv.h"
+
+/* How to undo row operations?
+(1) For an insert, we have stored a prefix of the clustered index record
+in the undo log. Using it, we look for the clustered record, and using
+that we look for the records in the secondary indexes. The insert operation
+may have been left incomplete, if the database crashed, for example.
+We may have look at the trx id and roll ptr to make sure the record in the
+clustered index is really the one for which the undo log record was
+written. We can use the framework we get from the original insert op.
+(2) Delete marking: We can use the framework we get from the original
+delete mark op. We only have to check the trx id.
+(3) Update: This may be the most complicated. We have to use the framework
+we get from the original update op.
+
+What if the same trx repeatedly deletes and inserts an identical row.
+Then the row id changes and also roll ptr. What if the row id was not
+part of the ordering fields in the clustered index? Maybe we have to write
+it to undo log. Well, maybe not, because if we order the row id and trx id
+in descending order, then the only undeleted copy is the first in the
+index. Our searches in row operations always position the cursor before
+the first record in the result set. But, if there is no key defined for
+a table, then it would be desirable that row id is in ascending order.
+So, lets store row id in descending order only if it is not an ordering
+field in the clustered index.
+
+NOTE: Deletes and inserts may lead to situation where there are identical
+records in a secondary index. Is that a problem in the B-tree? Yes.
+Also updates can lead to this, unless trx id and roll ptr are included in
+ord fields.
+(1) Fix in clustered indexes: include row id, trx id, and roll ptr
+in node pointers of B-tree.
+(2) Fix in secondary indexes: include all fields in node pointers, and
+if an entry is inserted, check if it is equal to the right neighbor,
+in which case update the right neighbor: the neighbor must be delete
+marked, set it unmarked and write the trx id of the current transaction.
+
+What if the same trx repeatedly updates the same row, updating a secondary
+index field or not? Updating a clustered index ordering field?
+
+(1) If it does not update the secondary index and not the clustered index
+ord field. Then the secondary index record stays unchanged, but the
+trx id in the secondary index record may be smaller than in the clustered
+index record. This is no problem?
+(2) If it updates secondary index ord field but not clustered: then in
+secondary index there are delete marked records, which differ in an
+ord field. No problem.
+(3) Updates clustered ord field but not secondary, and secondary index
+is unique. Then the record in secondary index is just updated at the
+clustered ord field.
+(4)
+
+Problem with duplicate records:
+Fix 1: Add a trx op no field to all indexes. A problem: if a trx with a
+bigger trx id has inserted and delete marked a similar row, our trx inserts
+again a similar row, and a trx with an even bigger id delete marks it. Then
+the position of the row should change in the index if the trx id affects
+the alphabetical ordering.
+
+Fix 2: If an insert encounters a similar row marked deleted, we turn the
+insert into an 'update' of the row marked deleted. Then we must write undo
+info on the update. A problem: what if a purge operation tries to remove
+the delete marked row?
+
+We can think of the database row versions as a linked list which starts
+from the record in the clustered index, and is linked by roll ptrs
+through undo logs. The secondary index records are references which tell
+what kinds of records can be found in this linked list for a record
+in the clustered index.
+
+How to do the purge? A record can be removed from the clustered index
+if its linked list becomes empty, i.e., the row has been marked deleted
+and its roll ptr points to the record in the undo log we are going through,
+doing the purge. Similarly, during a rollback, a record can be removed
+if the stored roll ptr in the undo log points to a trx already (being) purged,
+or if the roll ptr is NULL, i.e., it was a fresh insert. */
+
+/************************************************************************
+Creates a row undo node to a query graph. */
+UNIV_INTERN
+undo_node_t*
+row_undo_node_create(
+/*=================*/
+				/* out, own: undo node */
+	trx_t*		trx,	/* in: transaction */
+	que_thr_t*	parent,	/* in: parent node, i.e., a thr node */
+	mem_heap_t*	heap)	/* in: memory heap where created */
+{
+	undo_node_t*	undo;
+
+	ut_ad(trx && parent && heap);
+
+	undo = mem_heap_alloc(heap, sizeof(undo_node_t));
+
+	undo->common.type = QUE_NODE_UNDO;
+	undo->common.parent = parent;
+
+	undo->state = UNDO_NODE_FETCH_NEXT;
+	undo->trx = trx;
+
+	btr_pcur_init(&(undo->pcur));
+
+	undo->heap = mem_heap_create(256);
+
+	return(undo);
+}
+
+/***************************************************************
+Looks for the clustered index record when node has the row reference.
+The pcur in node is used in the search. If found, stores the row to node,
+and stores the position of pcur, and detaches it. The pcur must be closed
+by the caller in any case. */
+UNIV_INTERN
+ibool
+row_undo_search_clust_to_pcur(
+/*==========================*/
+				/* out: TRUE if found; NOTE the node->pcur
+				must be closed by the caller, regardless of
+				the return value */
+	undo_node_t*	node)	/* in: row undo node */
+{
+	dict_index_t*	clust_index;
+	ibool		found;
+	mtr_t		mtr;
+	ibool		ret;
+	rec_t*		rec;
+	mem_heap_t*	heap		= NULL;
+	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint*		offsets		= offsets_;
+	rec_offs_init(offsets_);
+
+	mtr_start(&mtr);
+
+	clust_index = dict_table_get_first_index(node->table);
+
+	found = row_search_on_row_ref(&(node->pcur), BTR_MODIFY_LEAF,
+				      node->table, node->ref, &mtr);
+
+	rec = btr_pcur_get_rec(&(node->pcur));
+
+	offsets = rec_get_offsets(rec, clust_index, offsets,
+				  ULINT_UNDEFINED, &heap);
+
+	if (!found || 0 != ut_dulint_cmp(node->roll_ptr,
+					 row_get_rec_roll_ptr(rec, clust_index,
+							      offsets))) {
+
+		/* We must remove the reservation on the undo log record
+		BEFORE releasing the latch on the clustered index page: this
+		is to make sure that some thread will eventually undo the
+		modification corresponding to node->roll_ptr. */
+
+		/* fputs("--------------------undoing a previous version\n",
+		stderr); */
+
+		ret = FALSE;
+	} else {
+		node->row = row_build(ROW_COPY_DATA, clust_index, rec,
+				      offsets, NULL, &node->ext, node->heap);
+		if (node->update) {
+			node->undo_row = dtuple_copy(node->row, node->heap);
+			row_upd_replace(node->undo_row, &node->undo_ext,
+					clust_index, node->update, node->heap);
+		} else {
+			node->undo_row = NULL;
+			node->undo_ext = NULL;
+		}
+
+		btr_pcur_store_position(&(node->pcur), &mtr);
+
+		ret = TRUE;
+	}
+
+	btr_pcur_commit_specify_mtr(&(node->pcur), &mtr);
+
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+	return(ret);
+}
+
+/***************************************************************
+Fetches an undo log record and does the undo for the recorded operation.
+If none left, or a partial rollback completed, returns control to the
+parent node, which is always a query thread node. */
+static
+ulint
+row_undo(
+/*=====*/
+				/* out: DB_SUCCESS if operation successfully
+				completed, else error code */
+	undo_node_t*	node,	/* in: row undo node */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	ulint	err;
+	trx_t*	trx;
+	dulint	roll_ptr;
+	ibool	locked_data_dict;
+
+	ut_ad(node && thr);
+
+	trx = node->trx;
+
+	if (node->state == UNDO_NODE_FETCH_NEXT) {
+
+		node->undo_rec = trx_roll_pop_top_rec_of_trx(trx,
+							     trx->roll_limit,
+							     &roll_ptr,
+							     node->heap);
+		if (!node->undo_rec) {
+			/* Rollback completed for this query thread */
+
+			thr->run_node = que_node_get_parent(node);
+
+			return(DB_SUCCESS);
+		}
+
+		node->roll_ptr = roll_ptr;
+		node->undo_no = trx_undo_rec_get_undo_no(node->undo_rec);
+
+		if (trx_undo_roll_ptr_is_insert(roll_ptr)) {
+
+			node->state = UNDO_NODE_INSERT;
+		} else {
+			node->state = UNDO_NODE_MODIFY;
+		}
+
+	} else if (node->state == UNDO_NODE_PREV_VERS) {
+
+		/* Undo should be done to the same clustered index record
+		again in this same rollback, restoring the previous version */
+
+		roll_ptr = node->new_roll_ptr;
+
+		node->undo_rec = trx_undo_get_undo_rec_low(roll_ptr,
+							   node->heap);
+		node->roll_ptr = roll_ptr;
+		node->undo_no = trx_undo_rec_get_undo_no(node->undo_rec);
+
+		if (trx_undo_roll_ptr_is_insert(roll_ptr)) {
+
+			node->state = UNDO_NODE_INSERT;
+		} else {
+			node->state = UNDO_NODE_MODIFY;
+		}
+	}
+
+	/* Prevent DROP TABLE etc. while we are rolling back this row.
+	If we are doing a TABLE CREATE or some other dictionary operation,
+	then we already have dict_operation_lock locked in x-mode. Do not
+	try to lock again, because that would cause a hang. */
+
+	locked_data_dict = (trx->dict_operation_lock_mode == 0);
+
+	if (locked_data_dict) {
+
+		row_mysql_lock_data_dictionary(trx);
+	}
+
+	if (node->state == UNDO_NODE_INSERT) {
+
+		err = row_undo_ins(node);
+
+		node->state = UNDO_NODE_FETCH_NEXT;
+	} else {
+		ut_ad(node->state == UNDO_NODE_MODIFY);
+		err = row_undo_mod(node, thr);
+	}
+
+	if (locked_data_dict) {
+
+		row_mysql_unlock_data_dictionary(trx);
+	}
+
+	/* Do some cleanup */
+	btr_pcur_close(&(node->pcur));
+
+	mem_heap_empty(node->heap);
+
+	thr->run_node = node;
+
+	return(err);
+}
+
+/***************************************************************
+Undoes a row operation in a table. This is a high-level function used
+in SQL execution graphs. */
+UNIV_INTERN
+que_thr_t*
+row_undo_step(
+/*==========*/
+				/* out: query thread to run next or NULL */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	ulint		err;
+	undo_node_t*	node;
+	trx_t*		trx;
+
+	ut_ad(thr);
+
+	srv_activity_count++;
+
+	trx = thr_get_trx(thr);
+
+	node = thr->run_node;
+
+	ut_ad(que_node_get_type(node) == QUE_NODE_UNDO);
+
+	err = row_undo(node, thr);
+
+	trx->error_state = err;
+
+	if (err != DB_SUCCESS) {
+		/* SQL error detected */
+
+		fprintf(stderr, "InnoDB: Fatal error %lu in rollback.\n",
+			(ulong) err);
+
+		if (err == DB_OUT_OF_FILE_SPACE) {
+			fprintf(stderr,
+				"InnoDB: Error 13 means out of tablespace.\n"
+				"InnoDB: Consider increasing"
+				" your tablespace.\n");
+
+			exit(1);
+		}
+
+		ut_error;
+
+		return(NULL);
+	}
+
+	return(thr);
+}
diff --git a/storage/xtradb/row/row0upd.c b/storage/xtradb/row/row0upd.c
new file mode 100644
index 00000000000..740f1ee593d
--- /dev/null
+++ b/storage/xtradb/row/row0upd.c
@@ -0,0 +1,2168 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Update of a row
+
+Created 12/27/1996 Heikki Tuuri
+*******************************************************/
+
+#include "row0upd.h"
+
+#ifdef UNIV_NONINL
+#include "row0upd.ic"
+#endif
+
+#include "dict0dict.h"
+#include "dict0boot.h"
+#include "dict0crea.h"
+#include "mach0data.h"
+#include "trx0undo.h"
+#include "btr0btr.h"
+#include "btr0cur.h"
+#include "que0que.h"
+#include "row0ext.h"
+#include "row0ins.h"
+#include "row0sel.h"
+#include "row0row.h"
+#include "rem0cmp.h"
+#include "lock0lock.h"
+#include "log0log.h"
+#include "pars0sym.h"
+#include "eval0eval.h"
+#include "buf0lru.h"
+
+
+/* What kind of latch and lock can we assume when the control comes to
+   -------------------------------------------------------------------
+an update node?
+--------------
+Efficiency of massive updates would require keeping an x-latch on a
+clustered index page through many updates, and not setting an explicit
+x-lock on clustered index records, as they anyway will get an implicit
+x-lock when they are updated. A problem is that the read nodes in the
+graph should know that they must keep the latch when passing the control
+up to the update node, and not set any record lock on the record which
+will be updated. Another problem occurs if the execution is stopped,
+as the kernel switches to another query thread, or the transaction must
+wait for a lock. Then we should be able to release the latch and, maybe,
+acquire an explicit x-lock on the record.
+	Because this seems too complicated, we conclude that the less
+efficient solution of releasing all the latches when the control is
+transferred to another node, and acquiring explicit x-locks, is better. */
+
+/* How is a delete performed? If there is a delete without an
+explicit cursor, i.e., a searched delete, there are at least
+two different situations:
+the implicit select cursor may run on (1) the clustered index or
+on (2) a secondary index. The delete is performed by setting
+the delete bit in the record and substituting the id of the
+deleting transaction for the original trx id, and substituting a
+new roll ptr for previous roll ptr. The old trx id and roll ptr
+are saved in the undo log record. Thus, no physical changes occur
+in the index tree structure at the time of the delete. Only
+when the undo log is purged, the index records will be physically
+deleted from the index trees.
+
+The query graph executing a searched delete would consist of
+a delete node which has as a subtree a select subgraph.
+The select subgraph should return a (persistent) cursor
+in the clustered index, placed on page which is x-latched.
+The delete node should look for all secondary index records for
+this clustered index entry and mark them as deleted. When is
+the x-latch freed? The most efficient way for performing a
+searched delete is obviously to keep the x-latch for several
+steps of query graph execution. */
+
+/***************************************************************
+Checks if an update vector changes some of the first ordering fields of an
+index record. This is only used in foreign key checks and we can assume
+that index does not contain column prefixes. */
+static
+ibool
+row_upd_changes_first_fields_binary(
+/*================================*/
+				/* out: TRUE if changes */
+	dtuple_t*	entry,	/* in: old value of index entry */
+	dict_index_t*	index,	/* in: index of entry */
+	const upd_t*	update,	/* in: update vector for the row */
+	ulint		n);	/* in: how many first fields to check */
+
+
+/*************************************************************************
+Checks if index currently is mentioned as a referenced index in a foreign
+key constraint. */
+static
+ibool
+row_upd_index_is_referenced(
+/*========================*/
+				/* out: TRUE if referenced; NOTE that since
+				we do not hold dict_operation_lock
+				when leaving the function, it may be that
+				the referencing table has been dropped when
+				we leave this function: this function is only
+				for heuristic use! */
+	dict_index_t*	index,	/* in: index */
+	trx_t*		trx)	/* in: transaction */
+{
+	dict_table_t*	table		= index->table;
+	dict_foreign_t*	foreign;
+	ibool		froze_data_dict	= FALSE;
+	ibool		is_referenced	= FALSE;
+
+	if (!UT_LIST_GET_FIRST(table->referenced_list)) {
+
+		return(FALSE);
+	}
+
+	if (trx->dict_operation_lock_mode == 0) {
+		row_mysql_freeze_data_dictionary(trx);
+		froze_data_dict = TRUE;
+	}
+
+	foreign = UT_LIST_GET_FIRST(table->referenced_list);
+
+	while (foreign) {
+		if (foreign->referenced_index == index) {
+
+			is_referenced = TRUE;
+			goto func_exit;
+		}
+
+		foreign = UT_LIST_GET_NEXT(referenced_list, foreign);
+	}
+
+func_exit:
+	if (froze_data_dict) {
+		row_mysql_unfreeze_data_dictionary(trx);
+	}
+
+	return(is_referenced);
+}
+
+/*************************************************************************
+Checks if possible foreign key constraints hold after a delete of the record
+under pcur. NOTE that this function will temporarily commit mtr and lose the
+pcur position! */
+static
+ulint
+row_upd_check_references_constraints(
+/*=================================*/
+				/* out: DB_SUCCESS or an error code */
+	upd_node_t*	node,	/* in: row update node */
+	btr_pcur_t*	pcur,	/* in: cursor positioned on a record; NOTE: the
+				cursor position is lost in this function! */
+	dict_table_t*	table,	/* in: table in question */
+	dict_index_t*	index,	/* in: index of the cursor */
+	ulint*		offsets,/* in/out: rec_get_offsets(pcur.rec, index) */
+	que_thr_t*	thr,	/* in: query thread */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	dict_foreign_t*	foreign;
+	mem_heap_t*	heap;
+	dtuple_t*	entry;
+	trx_t*		trx;
+	const rec_t*	rec;
+	ulint		n_ext;
+	ulint		err;
+	ibool		got_s_lock	= FALSE;
+
+	if (UT_LIST_GET_FIRST(table->referenced_list) == NULL) {
+
+		return(DB_SUCCESS);
+	}
+
+	trx = thr_get_trx(thr);
+
+	rec = btr_pcur_get_rec(pcur);
+	ut_ad(rec_offs_validate(rec, index, offsets));
+
+	heap = mem_heap_create(500);
+
+	entry = row_rec_to_index_entry(ROW_COPY_DATA, rec, index, offsets,
+				       &n_ext, heap);
+
+	mtr_commit(mtr);
+
+	mtr_start(mtr);
+
+	if (trx->dict_operation_lock_mode == 0) {
+		got_s_lock = TRUE;
+
+		row_mysql_freeze_data_dictionary(trx);
+	}
+
+	foreign = UT_LIST_GET_FIRST(table->referenced_list);
+
+	while (foreign) {
+		/* Note that we may have an update which updates the index
+		record, but does NOT update the first fields which are
+		referenced in a foreign key constraint. Then the update does
+		NOT break the constraint. */
+
+		if (foreign->referenced_index == index
+		    && (node->is_delete
+			|| row_upd_changes_first_fields_binary(
+				entry, index, node->update,
+				foreign->n_fields))) {
+
+			if (foreign->foreign_table == NULL) {
+				dict_table_get(foreign->foreign_table_name,
+					       FALSE);
+			}
+
+			if (foreign->foreign_table) {
+				mutex_enter(&(dict_sys->mutex));
+
+				(foreign->foreign_table
+				 ->n_foreign_key_checks_running)++;
+
+				mutex_exit(&(dict_sys->mutex));
+			}
+
+			/* NOTE that if the thread ends up waiting for a lock
+			we will release dict_operation_lock temporarily!
+			But the counter on the table protects 'foreign' from
+			being dropped while the check is running. */
+
+			err = row_ins_check_foreign_constraint(
+				FALSE, foreign, table, entry, thr);
+
+			if (foreign->foreign_table) {
+				mutex_enter(&(dict_sys->mutex));
+
+				ut_a(foreign->foreign_table
+				     ->n_foreign_key_checks_running > 0);
+
+				(foreign->foreign_table
+				 ->n_foreign_key_checks_running)--;
+
+				mutex_exit(&(dict_sys->mutex));
+			}
+
+			if (err != DB_SUCCESS) {
+
+				goto func_exit;
+			}
+		}
+
+		foreign = UT_LIST_GET_NEXT(referenced_list, foreign);
+	}
+
+	err = DB_SUCCESS;
+
+func_exit:
+	if (got_s_lock) {
+		row_mysql_unfreeze_data_dictionary(trx);
+	}
+
+	mem_heap_free(heap);
+
+	return(err);
+}
+
+/*************************************************************************
+Creates an update node for a query graph. */
+UNIV_INTERN
+upd_node_t*
+upd_node_create(
+/*============*/
+				/* out, own: update node */
+	mem_heap_t*	heap)	/* in: mem heap where created */
+{
+	upd_node_t*	node;
+
+	node = mem_heap_alloc(heap, sizeof(upd_node_t));
+	node->common.type = QUE_NODE_UPDATE;
+
+	node->state = UPD_NODE_UPDATE_CLUSTERED;
+	node->in_mysql_interface = FALSE;
+
+	node->row = NULL;
+	node->ext = NULL;
+	node->upd_row = NULL;
+	node->upd_ext = NULL;
+	node->index = NULL;
+	node->update = NULL;
+
+	node->foreign = NULL;
+	node->cascade_heap = NULL;
+	node->cascade_node = NULL;
+
+	node->select = NULL;
+
+	node->heap = mem_heap_create(128);
+	node->magic_n = UPD_NODE_MAGIC_N;
+
+	node->cmpl_info = 0;
+
+	return(node);
+}
+
+/*************************************************************************
+Updates the trx id and roll ptr field in a clustered index record in database
+recovery. */
+UNIV_INTERN
+void
+row_upd_rec_sys_fields_in_recovery(
+/*===============================*/
+	rec_t*		rec,	/* in/out: record */
+	page_zip_des_t*	page_zip,/* in/out: compressed page, or NULL */
+	const ulint*	offsets,/* in: array returned by rec_get_offsets() */
+	ulint		pos,	/* in: TRX_ID position in rec */
+	dulint		trx_id,	/* in: transaction id */
+	dulint		roll_ptr)/* in: roll ptr of the undo log record */
+{
+	ut_ad(rec_offs_validate(rec, NULL, offsets));
+
+	if (UNIV_LIKELY_NULL(page_zip)) {
+		page_zip_write_trx_id_and_roll_ptr(
+			page_zip, rec, offsets, pos, trx_id, roll_ptr);
+	} else {
+		byte*	field;
+		ulint	len;
+
+		field = rec_get_nth_field(rec, offsets, pos, &len);
+		ut_ad(len == DATA_TRX_ID_LEN);
+#if DATA_TRX_ID + 1 != DATA_ROLL_PTR
+# error "DATA_TRX_ID + 1 != DATA_ROLL_PTR"
+#endif
+		trx_write_trx_id(field, trx_id);
+		trx_write_roll_ptr(field + DATA_TRX_ID_LEN, roll_ptr);
+	}
+}
+
+/*************************************************************************
+Sets the trx id or roll ptr field of a clustered index entry. */
+UNIV_INTERN
+void
+row_upd_index_entry_sys_field(
+/*==========================*/
+	const dtuple_t*	entry,	/* in: index entry, where the memory buffers
+				for sys fields are already allocated:
+				the function just copies the new values to
+				them */
+	dict_index_t*	index,	/* in: clustered index */
+	ulint		type,	/* in: DATA_TRX_ID or DATA_ROLL_PTR */
+	dulint		val)	/* in: value to write */
+{
+	dfield_t*	dfield;
+	byte*		field;
+	ulint		pos;
+
+	ut_ad(dict_index_is_clust(index));
+
+	pos = dict_index_get_sys_col_pos(index, type);
+
+	dfield = dtuple_get_nth_field(entry, pos);
+	field = dfield_get_data(dfield);
+
+	if (type == DATA_TRX_ID) {
+		trx_write_trx_id(field, val);
+	} else {
+		ut_ad(type == DATA_ROLL_PTR);
+		trx_write_roll_ptr(field, val);
+	}
+}
+
+/***************************************************************
+Returns TRUE if row update changes size of some field in index or if some
+field to be updated is stored externally in rec or update. */
+UNIV_INTERN
+ibool
+row_upd_changes_field_size_or_external(
+/*===================================*/
+				/* out: TRUE if the update changes the size of
+				some field in index or the field is external
+				in rec or update */
+	dict_index_t*	index,	/* in: index */
+	const ulint*	offsets,/* in: rec_get_offsets(rec, index) */
+	const upd_t*	update)	/* in: update vector */
+{
+	const upd_field_t*	upd_field;
+	const dfield_t*		new_val;
+	ulint			old_len;
+	ulint			new_len;
+	ulint			n_fields;
+	ulint			i;
+
+	ut_ad(rec_offs_validate(NULL, index, offsets));
+	n_fields = upd_get_n_fields(update);
+
+	for (i = 0; i < n_fields; i++) {
+		upd_field = upd_get_nth_field(update, i);
+
+		new_val = &(upd_field->new_val);
+		new_len = dfield_get_len(new_val);
+
+		if (dfield_is_null(new_val) && !rec_offs_comp(offsets)) {
+			/* A bug fixed on Dec 31st, 2004: we looked at the
+			SQL NULL size from the wrong field! We may backport
+			this fix also to 4.0. The merge to 5.0 will be made
+			manually immediately after we commit this to 4.1. */
+
+			new_len = dict_col_get_sql_null_size(
+				dict_index_get_nth_col(index,
+						       upd_field->field_no));
+		}
+
+		old_len = rec_offs_nth_size(offsets, upd_field->field_no);
+
+		if (rec_offs_comp(offsets)
+		    && rec_offs_nth_sql_null(offsets,
+					     upd_field->field_no)) {
+			/* Note that in the compact table format, for a
+			variable length field, an SQL NULL will use zero
+			bytes in the offset array at the start of the physical
+			record, but a zero-length value (empty string) will
+			use one byte! Thus, we cannot use update-in-place
+			if we update an SQL NULL varchar to an empty string! */
+
+			old_len = UNIV_SQL_NULL;
+		}
+
+		if (dfield_is_ext(new_val) || old_len != new_len
+		    || rec_offs_nth_extern(offsets, upd_field->field_no)) {
+
+			return(TRUE);
+		}
+	}
+
+	return(FALSE);
+}
+
+/***************************************************************
+Replaces the new column values stored in the update vector to the record
+given. No field size changes are allowed. */
+UNIV_INTERN
+void
+row_upd_rec_in_place(
+/*=================*/
+	rec_t*		rec,	/* in/out: record where replaced */
+	dict_index_t*	index,	/* in: the index the record belongs to */
+	const ulint*	offsets,/* in: array returned by rec_get_offsets() */
+	const upd_t*	update,	/* in: update vector */
+	page_zip_des_t*	page_zip)/* in: compressed page with enough space
+				available, or NULL */
+{
+	const upd_field_t*	upd_field;
+	const dfield_t*		new_val;
+	ulint			n_fields;
+	ulint			i;
+
+	ut_ad(rec_offs_validate(rec, index, offsets));
+
+	if (rec_offs_comp(offsets)) {
+		rec_set_info_bits_new(rec, update->info_bits);
+	} else {
+		rec_set_info_bits_old(rec, update->info_bits);
+	}
+
+	n_fields = upd_get_n_fields(update);
+
+	for (i = 0; i < n_fields; i++) {
+		upd_field = upd_get_nth_field(update, i);
+		new_val = &(upd_field->new_val);
+		ut_ad(!dfield_is_ext(new_val) ==
+		      !rec_offs_nth_extern(offsets, upd_field->field_no));
+
+		rec_set_nth_field(rec, offsets, upd_field->field_no,
+				  dfield_get_data(new_val),
+				  dfield_get_len(new_val));
+	}
+
+	if (UNIV_LIKELY_NULL(page_zip)) {
+		page_zip_write_rec(page_zip, rec, index, offsets, 0);
+	}
+}
+
+/*************************************************************************
+Writes into the redo log the values of trx id and roll ptr and enough info
+to determine their positions within a clustered index record. */
+UNIV_INTERN
+byte*
+row_upd_write_sys_vals_to_log(
+/*==========================*/
+				/* out: new pointer to mlog */
+	dict_index_t*	index,	/* in: clustered index */
+	trx_t*		trx,	/* in: transaction */
+	dulint		roll_ptr,/* in: roll ptr of the undo log record */
+	byte*		log_ptr,/* pointer to a buffer of size > 20 opened
+				in mlog */
+	mtr_t*		mtr __attribute__((unused))) /* in: mtr */
+{
+	ut_ad(dict_index_is_clust(index));
+	ut_ad(mtr);
+
+	log_ptr += mach_write_compressed(log_ptr,
+					 dict_index_get_sys_col_pos(
+						 index, DATA_TRX_ID));
+
+	trx_write_roll_ptr(log_ptr, roll_ptr);
+	log_ptr += DATA_ROLL_PTR_LEN;
+
+	log_ptr += mach_dulint_write_compressed(log_ptr, trx->id);
+
+	return(log_ptr);
+}
+
+/*************************************************************************
+Parses the log data of system field values. */
+UNIV_INTERN
+byte*
+row_upd_parse_sys_vals(
+/*===================*/
+			/* out: log data end or NULL */
+	byte*	ptr,	/* in: buffer */
+	byte*	end_ptr,/* in: buffer end */
+	ulint*	pos,	/* out: TRX_ID position in record */
+	dulint*	trx_id,	/* out: trx id */
+	dulint*	roll_ptr)/* out: roll ptr */
+{
+	ptr = mach_parse_compressed(ptr, end_ptr, pos);
+
+	if (ptr == NULL) {
+
+		return(NULL);
+	}
+
+	if (end_ptr < ptr + DATA_ROLL_PTR_LEN) {
+
+		return(NULL);
+	}
+
+	*roll_ptr = trx_read_roll_ptr(ptr);
+	ptr += DATA_ROLL_PTR_LEN;
+
+	ptr = mach_dulint_parse_compressed(ptr, end_ptr, trx_id);
+
+	return(ptr);
+}
+
+/***************************************************************
+Writes to the redo log the new values of the fields occurring in the index. */
+UNIV_INTERN
+void
+row_upd_index_write_log(
+/*====================*/
+	const upd_t*	update,	/* in: update vector */
+	byte*		log_ptr,/* in: pointer to mlog buffer: must
+				contain at least MLOG_BUF_MARGIN bytes
+				of free space; the buffer is closed
+				within this function */
+	mtr_t*		mtr)	/* in: mtr into whose log to write */
+{
+	const upd_field_t*	upd_field;
+	const dfield_t*		new_val;
+	ulint			len;
+	ulint			n_fields;
+	byte*			buf_end;
+	ulint			i;
+
+	n_fields = upd_get_n_fields(update);
+
+	buf_end = log_ptr + MLOG_BUF_MARGIN;
+
+	mach_write_to_1(log_ptr, update->info_bits);
+	log_ptr++;
+	log_ptr += mach_write_compressed(log_ptr, n_fields);
+
+	for (i = 0; i < n_fields; i++) {
+
+#if MLOG_BUF_MARGIN <= 30
+# error "MLOG_BUF_MARGIN <= 30"
+#endif
+
+		if (log_ptr + 30 > buf_end) {
+			mlog_close(mtr, log_ptr);
+
+			log_ptr = mlog_open(mtr, MLOG_BUF_MARGIN);
+			buf_end = log_ptr + MLOG_BUF_MARGIN;
+		}
+
+		upd_field = upd_get_nth_field(update, i);
+
+		new_val = &(upd_field->new_val);
+
+		len = dfield_get_len(new_val);
+
+		log_ptr += mach_write_compressed(log_ptr, upd_field->field_no);
+		log_ptr += mach_write_compressed(log_ptr, len);
+
+		if (len != UNIV_SQL_NULL) {
+			if (log_ptr + len < buf_end) {
+				memcpy(log_ptr, dfield_get_data(new_val), len);
+
+				log_ptr += len;
+			} else {
+				mlog_close(mtr, log_ptr);
+
+				mlog_catenate_string(mtr,
+						     dfield_get_data(new_val),
+						     len);
+
+				log_ptr = mlog_open(mtr, MLOG_BUF_MARGIN);
+				buf_end = log_ptr + MLOG_BUF_MARGIN;
+			}
+		}
+	}
+
+	mlog_close(mtr, log_ptr);
+}
+
+/*************************************************************************
+Parses the log data written by row_upd_index_write_log. */
+UNIV_INTERN
+byte*
+row_upd_index_parse(
+/*================*/
+				/* out: log data end or NULL */
+	byte*		ptr,	/* in: buffer */
+	byte*		end_ptr,/* in: buffer end */
+	mem_heap_t*	heap,	/* in: memory heap where update vector is
+				built */
+	upd_t**		update_out)/* out: update vector */
+{
+	upd_t*		update;
+	upd_field_t*	upd_field;
+	dfield_t*	new_val;
+	ulint		len;
+	ulint		n_fields;
+	ulint		info_bits;
+	ulint		i;
+
+	if (end_ptr < ptr + 1) {
+
+		return(NULL);
+	}
+
+	info_bits = mach_read_from_1(ptr);
+	ptr++;
+	ptr = mach_parse_compressed(ptr, end_ptr, &n_fields);
+
+	if (ptr == NULL) {
+
+		return(NULL);
+	}
+
+	update = upd_create(n_fields, heap);
+	update->info_bits = info_bits;
+
+	for (i = 0; i < n_fields; i++) {
+		ulint	field_no;
+		upd_field = upd_get_nth_field(update, i);
+		new_val = &(upd_field->new_val);
+
+		ptr = mach_parse_compressed(ptr, end_ptr, &field_no);
+
+		if (ptr == NULL) {
+
+			return(NULL);
+		}
+
+		upd_field->field_no = field_no;
+
+		ptr = mach_parse_compressed(ptr, end_ptr, &len);
+
+		if (ptr == NULL) {
+
+			return(NULL);
+		}
+
+		if (len != UNIV_SQL_NULL) {
+
+			if (end_ptr < ptr + len) {
+
+				return(NULL);
+			}
+
+			dfield_set_data(new_val,
+					mem_heap_dup(heap, ptr, len), len);
+			ptr += len;
+		} else {
+			dfield_set_null(new_val);
+		}
+	}
+
+	*update_out = update;
+
+	return(ptr);
+}
+
+/*******************************************************************
+Builds an update vector from those fields which in a secondary index entry
+differ from a record that has the equal ordering fields. NOTE: we compare
+the fields as binary strings! */
+UNIV_INTERN
+upd_t*
+row_upd_build_sec_rec_difference_binary(
+/*====================================*/
+				/* out, own: update vector of differing
+				fields */
+	dict_index_t*	index,	/* in: index */
+	const dtuple_t*	entry,	/* in: entry to insert */
+	const rec_t*	rec,	/* in: secondary index record */
+	trx_t*		trx,	/* in: transaction */
+	mem_heap_t*	heap)	/* in: memory heap from which allocated */
+{
+	upd_field_t*	upd_field;
+	const dfield_t*	dfield;
+	const byte*	data;
+	ulint		len;
+	upd_t*		update;
+	ulint		n_diff;
+	ulint		i;
+	ulint		offsets_[REC_OFFS_SMALL_SIZE];
+	const ulint*	offsets;
+	rec_offs_init(offsets_);
+
+	/* This function is used only for a secondary index */
+	ut_a(!dict_index_is_clust(index));
+
+	update = upd_create(dtuple_get_n_fields(entry), heap);
+
+	n_diff = 0;
+	offsets = rec_get_offsets(rec, index, offsets_,
+				  ULINT_UNDEFINED, &heap);
+
+	for (i = 0; i < dtuple_get_n_fields(entry); i++) {
+
+		data = rec_get_nth_field(rec, offsets, i, &len);
+
+		dfield = dtuple_get_nth_field(entry, i);
+
+		/* NOTE that it may be that len != dfield_get_len(dfield) if we
+		are updating in a character set and collation where strings of
+		different length can be equal in an alphabetical comparison,
+		and also in the case where we have a column prefix index
+		and the last characters in the index field are spaces; the
+		latter case probably caused the assertion failures reported at
+		row0upd.c line 713 in versions 4.0.14 - 4.0.16. */
+
+		/* NOTE: we compare the fields as binary strings!
+		(No collation) */
+
+		if (!dfield_data_is_binary_equal(dfield, len, data)) {
+
+			upd_field = upd_get_nth_field(update, n_diff);
+
+			dfield_copy(&(upd_field->new_val), dfield);
+
+			upd_field_set_field_no(upd_field, i, index, trx);
+
+			n_diff++;
+		}
+	}
+
+	update->n_fields = n_diff;
+
+	return(update);
+}
+
+/*******************************************************************
+Builds an update vector from those fields, excluding the roll ptr and
+trx id fields, which in an index entry differ from a record that has
+the equal ordering fields. NOTE: we compare the fields as binary strings! */
+UNIV_INTERN
+upd_t*
+row_upd_build_difference_binary(
+/*============================*/
+				/* out, own: update vector of differing
+				fields, excluding roll ptr and trx id */
+	dict_index_t*	index,	/* in: clustered index */
+	const dtuple_t*	entry,	/* in: entry to insert */
+	const rec_t*	rec,	/* in: clustered index record */
+	trx_t*		trx,	/* in: transaction */
+	mem_heap_t*	heap)	/* in: memory heap from which allocated */
+{
+	upd_field_t*	upd_field;
+	const dfield_t*	dfield;
+	const byte*	data;
+	ulint		len;
+	upd_t*		update;
+	ulint		n_diff;
+	ulint		roll_ptr_pos;
+	ulint		trx_id_pos;
+	ulint		i;
+	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+	const ulint*	offsets;
+	rec_offs_init(offsets_);
+
+	/* This function is used only for a clustered index */
+	ut_a(dict_index_is_clust(index));
+
+	update = upd_create(dtuple_get_n_fields(entry), heap);
+
+	n_diff = 0;
+
+	roll_ptr_pos = dict_index_get_sys_col_pos(index, DATA_ROLL_PTR);
+	trx_id_pos = dict_index_get_sys_col_pos(index, DATA_TRX_ID);
+
+	offsets = rec_get_offsets(rec, index, offsets_,
+				  ULINT_UNDEFINED, &heap);
+
+	for (i = 0; i < dtuple_get_n_fields(entry); i++) {
+
+		data = rec_get_nth_field(rec, offsets, i, &len);
+
+		dfield = dtuple_get_nth_field(entry, i);
+
+		/* NOTE: we compare the fields as binary strings!
+		(No collation) */
+
+		if (i == trx_id_pos || i == roll_ptr_pos) {
+
+			goto skip_compare;
+		}
+
+		if (UNIV_UNLIKELY(!dfield_is_ext(dfield)
+				  != !rec_offs_nth_extern(offsets, i))
+		    || !dfield_data_is_binary_equal(dfield, len, data)) {
+
+			upd_field = upd_get_nth_field(update, n_diff);
+
+			dfield_copy(&(upd_field->new_val), dfield);
+
+			upd_field_set_field_no(upd_field, i, index, trx);
+
+			n_diff++;
+		}
+skip_compare:
+		;
+	}
+
+	update->n_fields = n_diff;
+
+	return(update);
+}
+
+/***************************************************************
+Fetch a prefix of an externally stored column.  This is similar
+to row_ext_lookup(), but the row_ext_t holds the old values
+of the column and must not be poisoned with the new values. */
+static
+byte*
+row_upd_ext_fetch(
+/*==============*/
+					/* out: BLOB prefix */
+	const byte*	data,		/* in: 'internally' stored part of the
+					field containing also the reference to
+					the external part */
+	ulint		local_len,	/* in: length of data, in bytes */
+	ulint		zip_size,	/* in: nonzero=compressed BLOB
+					page size, zero for uncompressed
+					BLOBs */
+	ulint*		len,		/* in: length of prefix to fetch;
+					out: fetched length of the prefix */
+	mem_heap_t*	heap)		/* in: heap where to allocate */
+{
+	byte*	buf = mem_heap_alloc(heap, *len);
+
+	*len = btr_copy_externally_stored_field_prefix(buf, *len,
+						       zip_size,
+						       data, local_len);
+	/* We should never update records containing a half-deleted BLOB. */
+	ut_a(*len);
+
+	return(buf);
+}
+
+/***************************************************************
+Replaces the new column value stored in the update vector in
+the given index entry field. */
+static
+void
+row_upd_index_replace_new_col_val(
+/*==============================*/
+	dfield_t*		dfield,	/* in/out: data field
+					of the index entry */
+	const dict_field_t*	field,	/* in: index field */
+	const dict_col_t*	col,	/* in: field->col */
+	const upd_field_t*	uf,	/* in: update field */
+	mem_heap_t*		heap,	/* in: memory heap for allocating
+					and copying the new value */
+	ulint			zip_size)/* in: compressed page
+					 size of the table, or 0 */
+{
+	ulint		len;
+	const byte*	data;
+
+	dfield_copy_data(dfield, &uf->new_val);
+
+	if (dfield_is_null(dfield)) {
+		return;
+	}
+
+	len = dfield_get_len(dfield);
+	data = dfield_get_data(dfield);
+
+	if (field->prefix_len > 0) {
+		ibool		fetch_ext = dfield_is_ext(dfield)
+			&& len < (ulint) field->prefix_len
+			+ BTR_EXTERN_FIELD_REF_SIZE;
+
+		if (fetch_ext) {
+			ulint	l = len;
+
+			len = field->prefix_len;
+
+			data = row_upd_ext_fetch(data, l, zip_size,
+						 &len, heap);
+		}
+
+		len = dtype_get_at_most_n_mbchars(col->prtype,
+						  col->mbminlen, col->mbmaxlen,
+						  field->prefix_len, len,
+						  (const char*) data);
+
+		dfield_set_data(dfield, data, len);
+
+		if (!fetch_ext) {
+			dfield_dup(dfield, heap);
+		}
+
+		return;
+	}
+
+	switch (uf->orig_len) {
+		byte*	buf;
+	case BTR_EXTERN_FIELD_REF_SIZE:
+		/* Restore the original locally stored
+		part of the column.  In the undo log,
+		InnoDB writes a longer prefix of externally
+		stored columns, so that column prefixes
+		in secondary indexes can be reconstructed. */
+		dfield_set_data(dfield,
+				data + len - BTR_EXTERN_FIELD_REF_SIZE,
+				BTR_EXTERN_FIELD_REF_SIZE);
+		dfield_set_ext(dfield);
+		/* fall through */
+	case 0:
+		dfield_dup(dfield, heap);
+		break;
+	default:
+		/* Reconstruct the original locally
+		stored part of the column.  The data
+		will have to be copied. */
+		ut_a(uf->orig_len > BTR_EXTERN_FIELD_REF_SIZE);
+		buf = mem_heap_alloc(heap, uf->orig_len);
+		/* Copy the locally stored prefix. */
+		memcpy(buf, data,
+		       uf->orig_len - BTR_EXTERN_FIELD_REF_SIZE);
+		/* Copy the BLOB pointer. */
+		memcpy(buf + uf->orig_len - BTR_EXTERN_FIELD_REF_SIZE,
+		       data + len - BTR_EXTERN_FIELD_REF_SIZE,
+		       BTR_EXTERN_FIELD_REF_SIZE);
+
+		dfield_set_data(dfield, buf, uf->orig_len);
+		dfield_set_ext(dfield);
+		break;
+	}
+}
+
+/***************************************************************
+Replaces the new column values stored in the update vector to the index entry
+given. */
+UNIV_INTERN
+void
+row_upd_index_replace_new_col_vals_index_pos(
+/*=========================================*/
+	dtuple_t*	entry,	/* in/out: index entry where replaced;
+				the clustered index record must be
+				covered by a lock or a page latch to
+				prevent deletion (rollback or purge) */
+	dict_index_t*	index,	/* in: index; NOTE that this may also be a
+				non-clustered index */
+	const upd_t*	update,	/* in: an update vector built for the index so
+				that the field number in an upd_field is the
+				index position */
+	ibool		order_only,
+				/* in: if TRUE, limit the replacement to
+				ordering fields of index; note that this
+				does not work for non-clustered indexes. */
+	mem_heap_t*	heap)	/* in: memory heap for allocating and
+				copying the new values */
+{
+	ulint		i;
+	ulint		n_fields;
+	const ulint	zip_size	= dict_table_zip_size(index->table);
+
+	ut_ad(index);
+
+	dtuple_set_info_bits(entry, update->info_bits);
+
+	if (order_only) {
+		n_fields = dict_index_get_n_unique(index);
+	} else {
+		n_fields = dict_index_get_n_fields(index);
+	}
+
+	for (i = 0; i < n_fields; i++) {
+		const dict_field_t*	field;
+		const dict_col_t*	col;
+		const upd_field_t*	uf;
+
+		field = dict_index_get_nth_field(index, i);
+		col = dict_field_get_col(field);
+		uf = upd_get_field_by_field_no(update, i);
+
+		if (uf) {
+			row_upd_index_replace_new_col_val(
+				dtuple_get_nth_field(entry, i),
+				field, col, uf, heap, zip_size);
+		}
+	}
+}
+
+/***************************************************************
+Replaces the new column values stored in the update vector to the index entry
+given. */
+UNIV_INTERN
+void
+row_upd_index_replace_new_col_vals(
+/*===============================*/
+	dtuple_t*	entry,	/* in/out: index entry where replaced;
+				the clustered index record must be
+				covered by a lock or a page latch to
+				prevent deletion (rollback or purge) */
+	dict_index_t*	index,	/* in: index; NOTE that this may also be a
+				non-clustered index */
+	const upd_t*	update,	/* in: an update vector built for the
+				CLUSTERED index so that the field number in
+				an upd_field is the clustered index position */
+	mem_heap_t*	heap)	/* in: memory heap for allocating and
+				copying the new values */
+{
+	ulint			i;
+	const dict_index_t*	clust_index
+		= dict_table_get_first_index(index->table);
+	const ulint		zip_size
+		= dict_table_zip_size(index->table);
+
+	dtuple_set_info_bits(entry, update->info_bits);
+
+	for (i = 0; i < dict_index_get_n_fields(index); i++) {
+		const dict_field_t*	field;
+		const dict_col_t*	col;
+		const upd_field_t*	uf;
+
+		field = dict_index_get_nth_field(index, i);
+		col = dict_field_get_col(field);
+		uf = upd_get_field_by_field_no(
+			update, dict_col_get_clust_pos(col, clust_index));
+
+		if (uf) {
+			row_upd_index_replace_new_col_val(
+				dtuple_get_nth_field(entry, i),
+				field, col, uf, heap, zip_size);
+		}
+	}
+}
+
+/***************************************************************
+Replaces the new column values stored in the update vector. */
+UNIV_INTERN
+void
+row_upd_replace(
+/*============*/
+	dtuple_t*		row,	/* in/out: row where replaced,
+					indexed by col_no;
+					the clustered index record must be
+					covered by a lock or a page latch to
+					prevent deletion (rollback or purge) */
+	row_ext_t**		ext,	/* out, own: NULL, or externally
+					stored column prefixes */
+	const dict_index_t*	index,	/* in: clustered index */
+	const upd_t*		update,	/* in: an update vector built for the
+					clustered index */
+	mem_heap_t*		heap)	/* in: memory heap */
+{
+	ulint			col_no;
+	ulint			i;
+	ulint			n_cols;
+	ulint			n_ext_cols;
+	ulint*			ext_cols;
+	const dict_table_t*	table;
+
+	ut_ad(row);
+	ut_ad(ext);
+	ut_ad(index);
+	ut_ad(dict_index_is_clust(index));
+	ut_ad(update);
+	ut_ad(heap);
+
+	n_cols = dtuple_get_n_fields(row);
+	table = index->table;
+	ut_ad(n_cols == dict_table_get_n_cols(table));
+
+	ext_cols = mem_heap_alloc(heap, n_cols * sizeof *ext_cols);
+	n_ext_cols = 0;
+
+	dtuple_set_info_bits(row, update->info_bits);
+
+	for (col_no = 0; col_no < n_cols; col_no++) {
+
+		const dict_col_t*	col
+			= dict_table_get_nth_col(table, col_no);
+		const ulint		clust_pos
+			= dict_col_get_clust_pos(col, index);
+		dfield_t*		dfield;
+
+		if (UNIV_UNLIKELY(clust_pos == ULINT_UNDEFINED)) {
+
+			continue;
+		}
+
+		dfield = dtuple_get_nth_field(row, col_no);
+
+		for (i = 0; i < upd_get_n_fields(update); i++) {
+
+			const upd_field_t*	upd_field
+				= upd_get_nth_field(update, i);
+
+			if (upd_field->field_no != clust_pos) {
+
+				continue;
+			}
+
+			dfield_copy_data(dfield, &upd_field->new_val);
+			break;
+		}
+
+		if (dfield_is_ext(dfield) && col->ord_part) {
+			ext_cols[n_ext_cols++] = col_no;
+		}
+	}
+
+	if (n_ext_cols) {
+		*ext = row_ext_create(n_ext_cols, ext_cols, row,
+				      dict_table_zip_size(table), heap);
+	} else {
+		*ext = NULL;
+	}
+}
+
+/***************************************************************
+Checks if an update vector changes an ordering field of an index record.
+This function is fast if the update vector is short or the number of ordering
+fields in the index is small. Otherwise, this can be quadratic.
+NOTE: we compare the fields as binary strings! */
+UNIV_INTERN
+ibool
+row_upd_changes_ord_field_binary(
+/*=============================*/
+				/* out: TRUE if update vector changes
+				an ordering field in the index record;
+				NOTE: the fields are compared as binary
+				strings */
+	const dtuple_t*	row,	/* in: old value of row, or NULL if the
+				row and the data values in update are not
+				known when this function is called, e.g., at
+				compile time */
+	dict_index_t*	index,	/* in: index of the record */
+	const upd_t*	update)	/* in: update vector for the row; NOTE: the
+				field numbers in this MUST be clustered index
+				positions! */
+{
+	ulint		n_unique;
+	ulint		n_upd_fields;
+	ulint		i, j;
+	dict_index_t*	clust_index;
+
+	ut_ad(update && index);
+
+	n_unique = dict_index_get_n_unique(index);
+	n_upd_fields = upd_get_n_fields(update);
+
+	clust_index = dict_table_get_first_index(index->table);
+
+	for (i = 0; i < n_unique; i++) {
+
+		const dict_field_t*	ind_field;
+		const dict_col_t*	col;
+		ulint			col_pos;
+		ulint			col_no;
+
+		ind_field = dict_index_get_nth_field(index, i);
+		col = dict_field_get_col(ind_field);
+		col_pos = dict_col_get_clust_pos(col, clust_index);
+		col_no = dict_col_get_no(col);
+
+		for (j = 0; j < n_upd_fields; j++) {
+
+			const upd_field_t*	upd_field
+				= upd_get_nth_field(update, j);
+
+			/* Note that if the index field is a column prefix
+			then it may be that row does not contain an externally
+			stored part of the column value, and we cannot compare
+			the datas */
+
+			if (col_pos == upd_field->field_no
+			    && (row == NULL
+				|| ind_field->prefix_len > 0
+				|| !dfield_datas_are_binary_equal(
+					dtuple_get_nth_field(row, col_no),
+					&(upd_field->new_val)))) {
+
+				return(TRUE);
+			}
+		}
+	}
+
+	return(FALSE);
+}
+
+/***************************************************************
+Checks if an update vector changes an ordering field of an index record.
+NOTE: we compare the fields as binary strings! */
+UNIV_INTERN
+ibool
+row_upd_changes_some_index_ord_field_binary(
+/*========================================*/
+					/* out: TRUE if update vector
+					may change an ordering field
+					in an index record */
+	const dict_table_t*	table,	/* in: table */
+	const upd_t*		update)	/* in: update vector for the row */
+{
+	upd_field_t*	upd_field;
+	dict_index_t*	index;
+	ulint		i;
+
+	index = dict_table_get_first_index(table);
+
+	for (i = 0; i < upd_get_n_fields(update); i++) {
+
+		upd_field = upd_get_nth_field(update, i);
+
+		if (dict_field_get_col(dict_index_get_nth_field(
+					       index, upd_field->field_no))
+		    ->ord_part) {
+
+			return(TRUE);
+		}
+	}
+
+	return(FALSE);
+}
+
+/***************************************************************
+Checks if an update vector changes some of the first ordering fields of an
+index record. This is only used in foreign key checks and we can assume
+that index does not contain column prefixes. */
+static
+ibool
+row_upd_changes_first_fields_binary(
+/*================================*/
+				/* out: TRUE if changes */
+	dtuple_t*	entry,	/* in: index entry */
+	dict_index_t*	index,	/* in: index of entry */
+	const upd_t*	update,	/* in: update vector for the row */
+	ulint		n)	/* in: how many first fields to check */
+{
+	ulint		n_upd_fields;
+	ulint		i, j;
+	dict_index_t*	clust_index;
+
+	ut_ad(update && index);
+	ut_ad(n <= dict_index_get_n_fields(index));
+
+	n_upd_fields = upd_get_n_fields(update);
+	clust_index = dict_table_get_first_index(index->table);
+
+	for (i = 0; i < n; i++) {
+
+		const dict_field_t*	ind_field;
+		const dict_col_t*	col;
+		ulint			col_pos;
+
+		ind_field = dict_index_get_nth_field(index, i);
+		col = dict_field_get_col(ind_field);
+		col_pos = dict_col_get_clust_pos(col, clust_index);
+
+		ut_a(ind_field->prefix_len == 0);
+
+		for (j = 0; j < n_upd_fields; j++) {
+
+			upd_field_t*	upd_field
+				= upd_get_nth_field(update, j);
+
+			if (col_pos == upd_field->field_no
+			    && !dfield_datas_are_binary_equal(
+				    dtuple_get_nth_field(entry, i),
+				    &(upd_field->new_val))) {
+
+				return(TRUE);
+			}
+		}
+	}
+
+	return(FALSE);
+}
+
+/*************************************************************************
+Copies the column values from a record. */
+UNIV_INLINE
+void
+row_upd_copy_columns(
+/*=================*/
+	rec_t*		rec,	/* in: record in a clustered index */
+	const ulint*	offsets,/* in: array returned by rec_get_offsets() */
+	sym_node_t*	column)	/* in: first column in a column list, or
+				NULL */
+{
+	byte*	data;
+	ulint	len;
+
+	while (column) {
+		data = rec_get_nth_field(rec, offsets,
+					 column->field_nos[SYM_CLUST_FIELD_NO],
+					 &len);
+		if (len == UNIV_SQL_NULL) {
+			len = UNIV_SQL_NULL;
+		}
+		eval_node_copy_and_alloc_val(column, data, len);
+
+		column = UT_LIST_GET_NEXT(col_var_list, column);
+	}
+}
+
+/*************************************************************************
+Calculates the new values for fields to update. Note that row_upd_copy_columns
+must have been called first. */
+UNIV_INLINE
+void
+row_upd_eval_new_vals(
+/*==================*/
+	upd_t*	update)	/* in/out: update vector */
+{
+	que_node_t*	exp;
+	upd_field_t*	upd_field;
+	ulint		n_fields;
+	ulint		i;
+
+	n_fields = upd_get_n_fields(update);
+
+	for (i = 0; i < n_fields; i++) {
+		upd_field = upd_get_nth_field(update, i);
+
+		exp = upd_field->exp;
+
+		eval_exp(exp);
+
+		dfield_copy_data(&(upd_field->new_val), que_node_get_val(exp));
+	}
+}
+
+/***************************************************************
+Stores to the heap the row on which the node->pcur is positioned. */
+static
+void
+row_upd_store_row(
+/*==============*/
+	upd_node_t*	node)	/* in: row update node */
+{
+	dict_index_t*	clust_index;
+	rec_t*		rec;
+	mem_heap_t*	heap		= NULL;
+	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+	const ulint*	offsets;
+	rec_offs_init(offsets_);
+
+	ut_ad(node->pcur->latch_mode != BTR_NO_LATCHES);
+
+	if (node->row != NULL) {
+		mem_heap_empty(node->heap);
+	}
+
+	clust_index = dict_table_get_first_index(node->table);
+
+	rec = btr_pcur_get_rec(node->pcur);
+
+	offsets = rec_get_offsets(rec, clust_index, offsets_,
+				  ULINT_UNDEFINED, &heap);
+	node->row = row_build(ROW_COPY_DATA, clust_index, rec, offsets,
+			      NULL, &node->ext, node->heap);
+	if (node->is_delete) {
+		node->upd_row = NULL;
+		node->upd_ext = NULL;
+	} else {
+		node->upd_row = dtuple_copy(node->row, node->heap);
+		row_upd_replace(node->upd_row, &node->upd_ext,
+				clust_index, node->update, node->heap);
+	}
+
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+}
+
+/***************************************************************
+Updates a secondary index entry of a row. */
+static
+ulint
+row_upd_sec_index_entry(
+/*====================*/
+				/* out: DB_SUCCESS if operation successfully
+				completed, else error code or DB_LOCK_WAIT */
+	upd_node_t*	node,	/* in: row update node */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	ibool		check_ref;
+	ibool		found;
+	dict_index_t*	index;
+	dtuple_t*	entry;
+	btr_pcur_t	pcur;
+	btr_cur_t*	btr_cur;
+	mem_heap_t*	heap;
+	rec_t*		rec;
+	ulint		err	= DB_SUCCESS;
+	mtr_t		mtr;
+	trx_t*		trx	= thr_get_trx(thr);
+
+	index = node->index;
+
+	check_ref = row_upd_index_is_referenced(index, trx);
+
+	heap = mem_heap_create(1024);
+
+	/* Build old index entry */
+	entry = row_build_index_entry(node->row, node->ext, index, heap);
+	ut_a(entry);
+
+	log_free_check();
+	mtr_start(&mtr);
+
+	found = row_search_index_entry(index, entry, BTR_MODIFY_LEAF, &pcur,
+				       &mtr);
+	btr_cur = btr_pcur_get_btr_cur(&pcur);
+
+	rec = btr_cur_get_rec(btr_cur);
+
+	if (UNIV_UNLIKELY(!found)) {
+		fputs("InnoDB: error in sec index entry update in\n"
+		      "InnoDB: ", stderr);
+		dict_index_name_print(stderr, trx, index);
+		fputs("\n"
+		      "InnoDB: tuple ", stderr);
+		dtuple_print(stderr, entry);
+		fputs("\n"
+		      "InnoDB: record ", stderr);
+		rec_print(stderr, rec, index);
+		putc('\n', stderr);
+
+		trx_print(stderr, trx, 0);
+
+		fputs("\n"
+		      "InnoDB: Submit a detailed bug report"
+		      " to http://bugs.mysql.com\n", stderr);
+	} else {
+		/* Delete mark the old index record; it can already be
+		delete marked if we return after a lock wait in
+		row_ins_index_entry below */
+
+		if (!rec_get_deleted_flag(rec,
+					  dict_table_is_comp(index->table))) {
+			err = btr_cur_del_mark_set_sec_rec(0, btr_cur, TRUE,
+							   thr, &mtr);
+			if (err == DB_SUCCESS && check_ref) {
+
+				ulint*	offsets = rec_get_offsets(
+					rec, index, NULL,
+					ULINT_UNDEFINED, &heap);
+				/* NOTE that the following call loses
+				the position of pcur ! */
+				err = row_upd_check_references_constraints(
+					node, &pcur, index->table,
+					index, offsets, thr, &mtr);
+			}
+		}
+	}
+
+	btr_pcur_close(&pcur);
+	mtr_commit(&mtr);
+
+	if (node->is_delete || err != DB_SUCCESS) {
+
+		goto func_exit;
+	}
+
+	/* Build a new index entry */
+	entry = row_build_index_entry(node->upd_row, node->upd_ext,
+				      index, heap);
+	ut_a(entry);
+
+	/* Insert new index entry */
+	err = row_ins_index_entry(index, entry, 0, TRUE, thr);
+
+func_exit:
+	mem_heap_free(heap);
+
+	return(err);
+}
+
+/***************************************************************
+Updates the secondary index record if it is changed in the row update or
+deletes it if this is a delete. */
+UNIV_INLINE
+ulint
+row_upd_sec_step(
+/*=============*/
+				/* out: DB_SUCCESS if operation successfully
+				completed, else error code or DB_LOCK_WAIT */
+	upd_node_t*	node,	/* in: row update node */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	ut_ad((node->state == UPD_NODE_UPDATE_ALL_SEC)
+	      || (node->state == UPD_NODE_UPDATE_SOME_SEC));
+	ut_ad(!dict_index_is_clust(node->index));
+
+	if (node->state == UPD_NODE_UPDATE_ALL_SEC
+	    || row_upd_changes_ord_field_binary(node->row, node->index,
+						node->update)) {
+		return(row_upd_sec_index_entry(node, thr));
+	}
+
+	return(DB_SUCCESS);
+}
+
+/***************************************************************
+Marks the clustered index record deleted and inserts the updated version
+of the record to the index. This function should be used when the ordering
+fields of the clustered index record change. This should be quite rare in
+database applications. */
+static
+ulint
+row_upd_clust_rec_by_insert(
+/*========================*/
+				/* out: DB_SUCCESS if operation successfully
+				completed, else error code or DB_LOCK_WAIT */
+	upd_node_t*	node,	/* in: row update node */
+	dict_index_t*	index,	/* in: clustered index of the record */
+	que_thr_t*	thr,	/* in: query thread */
+	ibool		check_ref,/* in: TRUE if index may be referenced in
+				a foreign key constraint */
+	mtr_t*		mtr)	/* in: mtr; gets committed here */
+{
+	mem_heap_t*	heap	= NULL;
+	btr_pcur_t*	pcur;
+	btr_cur_t*	btr_cur;
+	trx_t*		trx;
+	dict_table_t*	table;
+	dtuple_t*	entry;
+	ulint		err;
+
+	ut_ad(node);
+	ut_ad(dict_index_is_clust(index));
+
+	trx = thr_get_trx(thr);
+	table = node->table;
+	pcur = node->pcur;
+	btr_cur	= btr_pcur_get_btr_cur(pcur);
+
+	if (node->state != UPD_NODE_INSERT_CLUSTERED) {
+		rec_t*		rec;
+		dict_index_t*	index;
+		ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+		ulint*		offsets;
+		rec_offs_init(offsets_);
+
+		err = btr_cur_del_mark_set_clust_rec(BTR_NO_LOCKING_FLAG,
+						     btr_cur, TRUE, thr, mtr);
+		if (err != DB_SUCCESS) {
+			mtr_commit(mtr);
+			return(err);
+		}
+
+		/* Mark as not-owned the externally stored fields which the new
+		row inherits from the delete marked record: purge should not
+		free those externally stored fields even if the delete marked
+		record is removed from the index tree, or updated. */
+
+		rec = btr_cur_get_rec(btr_cur);
+		index = dict_table_get_first_index(table);
+		offsets = rec_get_offsets(rec, index, offsets_,
+					  ULINT_UNDEFINED, &heap);
+		btr_cur_mark_extern_inherited_fields(
+			btr_cur_get_page_zip(btr_cur),
+			rec, index, offsets, node->update, mtr);
+		if (check_ref) {
+			/* NOTE that the following call loses
+			the position of pcur ! */
+			err = row_upd_check_references_constraints(
+				node, pcur, table, index, offsets, thr, mtr);
+			if (err != DB_SUCCESS) {
+				mtr_commit(mtr);
+				if (UNIV_LIKELY_NULL(heap)) {
+					mem_heap_free(heap);
+				}
+				return(err);
+			}
+		}
+	}
+
+	mtr_commit(mtr);
+
+	if (!heap) {
+		heap = mem_heap_create(500);
+	}
+	node->state = UPD_NODE_INSERT_CLUSTERED;
+
+	entry = row_build_index_entry(node->upd_row, node->upd_ext,
+				      index, heap);
+	ut_a(entry);
+
+	row_upd_index_entry_sys_field(entry, index, DATA_TRX_ID, trx->id);
+
+	if (node->upd_ext) {
+		/* If we return from a lock wait, for example, we may have
+		extern fields marked as not-owned in entry (marked in the
+		if-branch above). We must unmark them. */
+
+		btr_cur_unmark_dtuple_extern_fields(entry);
+
+		/* We must mark non-updated extern fields in entry as
+		inherited, so that a possible rollback will not free them. */
+
+		btr_cur_mark_dtuple_inherited_extern(entry, node->update);
+	}
+
+	err = row_ins_index_entry(index, entry,
+				  node->upd_ext ? node->upd_ext->n_ext : 0,
+				  TRUE, thr);
+	mem_heap_free(heap);
+
+	return(err);
+}
+
+/***************************************************************
+Updates a clustered index record of a row when the ordering fields do
+not change. */
+static
+ulint
+row_upd_clust_rec(
+/*==============*/
+				/* out: DB_SUCCESS if operation successfully
+				completed, else error code or DB_LOCK_WAIT */
+	upd_node_t*	node,	/* in: row update node */
+	dict_index_t*	index,	/* in: clustered index */
+	que_thr_t*	thr,	/* in: query thread */
+	mtr_t*		mtr)	/* in: mtr; gets committed here */
+{
+	mem_heap_t*	heap	= NULL;
+	big_rec_t*	big_rec	= NULL;
+	btr_pcur_t*	pcur;
+	btr_cur_t*	btr_cur;
+	ulint		err;
+
+	ut_ad(node);
+	ut_ad(dict_index_is_clust(index));
+
+	pcur = node->pcur;
+	btr_cur = btr_pcur_get_btr_cur(pcur);
+
+	ut_ad(!rec_get_deleted_flag(btr_pcur_get_rec(pcur),
+				    dict_table_is_comp(index->table)));
+
+	/* Try optimistic updating of the record, keeping changes within
+	the page; we do not check locks because we assume the x-lock on the
+	record to update */
+
+	if (node->cmpl_info & UPD_NODE_NO_SIZE_CHANGE) {
+		err = btr_cur_update_in_place(BTR_NO_LOCKING_FLAG,
+					      btr_cur, node->update,
+					      node->cmpl_info, thr, mtr);
+	} else {
+		err = btr_cur_optimistic_update(BTR_NO_LOCKING_FLAG,
+						btr_cur, node->update,
+						node->cmpl_info, thr, mtr);
+	}
+
+	mtr_commit(mtr);
+
+	if (UNIV_LIKELY(err == DB_SUCCESS)) {
+
+		return(DB_SUCCESS);
+	}
+
+	if (buf_LRU_buf_pool_running_out()) {
+
+		return(DB_LOCK_TABLE_FULL);
+	}
+	/* We may have to modify the tree structure: do a pessimistic descent
+	down the index tree */
+
+	mtr_start(mtr);
+
+	/* NOTE: this transaction has an s-lock or x-lock on the record and
+	therefore other transactions cannot modify the record when we have no
+	latch on the page. In addition, we assume that other query threads of
+	the same transaction do not modify the record in the meantime.
+	Therefore we can assert that the restoration of the cursor succeeds. */
+
+	ut_a(btr_pcur_restore_position(BTR_MODIFY_TREE, pcur, mtr));
+
+	ut_ad(!rec_get_deleted_flag(btr_pcur_get_rec(pcur),
+				    dict_table_is_comp(index->table)));
+
+	err = btr_cur_pessimistic_update(BTR_NO_LOCKING_FLAG, btr_cur,
+					 &heap, &big_rec, node->update,
+					 node->cmpl_info, thr, mtr);
+	mtr_commit(mtr);
+
+	if (err == DB_SUCCESS && big_rec) {
+		ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+		rec_t*		rec;
+		rec_offs_init(offsets_);
+
+		mtr_start(mtr);
+
+		ut_a(btr_pcur_restore_position(BTR_MODIFY_TREE, pcur, mtr));
+		rec = btr_cur_get_rec(btr_cur);
+		err = btr_store_big_rec_extern_fields(
+			index, btr_cur_get_block(btr_cur), rec,
+			rec_get_offsets(rec, index, offsets_,
+					ULINT_UNDEFINED, &heap),
+			big_rec, mtr);
+		mtr_commit(mtr);
+	}
+
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+
+	if (big_rec) {
+		dtuple_big_rec_free(big_rec);
+	}
+
+	return(err);
+}
+
+/***************************************************************
+Delete marks a clustered index record. */
+static
+ulint
+row_upd_del_mark_clust_rec(
+/*=======================*/
+				/* out: DB_SUCCESS if operation successfully
+				completed, else error code */
+	upd_node_t*	node,	/* in: row update node */
+	dict_index_t*	index,	/* in: clustered index */
+	ulint*		offsets,/* in/out: rec_get_offsets() for the
+				record under the cursor */
+	que_thr_t*	thr,	/* in: query thread */
+	ibool		check_ref,/* in: TRUE if index may be referenced in
+				a foreign key constraint */
+	mtr_t*		mtr)	/* in: mtr; gets committed here */
+{
+	btr_pcur_t*	pcur;
+	btr_cur_t*	btr_cur;
+	ulint		err;
+
+	ut_ad(node);
+	ut_ad(dict_index_is_clust(index));
+	ut_ad(node->is_delete);
+
+	pcur = node->pcur;
+	btr_cur = btr_pcur_get_btr_cur(pcur);
+
+	/* Store row because we have to build also the secondary index
+	entries */
+
+	row_upd_store_row(node);
+
+	/* Mark the clustered index record deleted; we do not have to check
+	locks, because we assume that we have an x-lock on the record */
+
+	err = btr_cur_del_mark_set_clust_rec(BTR_NO_LOCKING_FLAG,
+					     btr_cur, TRUE, thr, mtr);
+	if (err == DB_SUCCESS && check_ref) {
+		/* NOTE that the following call loses the position of pcur ! */
+
+		err = row_upd_check_references_constraints(node,
+							   pcur, index->table,
+							   index, offsets,
+							   thr, mtr);
+	}
+
+	mtr_commit(mtr);
+
+	return(err);
+}
+
+/***************************************************************
+Updates the clustered index record. */
+static
+ulint
+row_upd_clust_step(
+/*===============*/
+				/* out: DB_SUCCESS if operation successfully
+				completed, DB_LOCK_WAIT in case of a lock wait,
+				else error code */
+	upd_node_t*	node,	/* in: row update node */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	dict_index_t*	index;
+	btr_pcur_t*	pcur;
+	ibool		success;
+	ibool		check_ref;
+	ulint		err;
+	mtr_t*		mtr;
+	mtr_t		mtr_buf;
+	rec_t*		rec;
+	mem_heap_t*	heap		= NULL;
+	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint*		offsets;
+	rec_offs_init(offsets_);
+
+	index = dict_table_get_first_index(node->table);
+
+	check_ref = row_upd_index_is_referenced(index, thr_get_trx(thr));
+
+	pcur = node->pcur;
+
+	/* We have to restore the cursor to its position */
+	mtr = &mtr_buf;
+
+	mtr_start(mtr);
+
+	/* If the restoration does not succeed, then the same
+	transaction has deleted the record on which the cursor was,
+	and that is an SQL error. If the restoration succeeds, it may
+	still be that the same transaction has successively deleted
+	and inserted a record with the same ordering fields, but in
+	that case we know that the transaction has at least an
+	implicit x-lock on the record. */
+
+	ut_a(pcur->rel_pos == BTR_PCUR_ON);
+
+	success = btr_pcur_restore_position(BTR_MODIFY_LEAF, pcur, mtr);
+
+	if (!success) {
+		err = DB_RECORD_NOT_FOUND;
+
+		mtr_commit(mtr);
+
+		return(err);
+	}
+
+	/* If this is a row in SYS_INDEXES table of the data dictionary,
+	then we have to free the file segments of the index tree associated
+	with the index */
+
+	if (node->is_delete
+	    && ut_dulint_cmp(node->table->id, DICT_INDEXES_ID) == 0) {
+
+		dict_drop_index_tree(btr_pcur_get_rec(pcur), mtr);
+
+		mtr_commit(mtr);
+
+		mtr_start(mtr);
+
+		success = btr_pcur_restore_position(BTR_MODIFY_LEAF, pcur,
+						    mtr);
+		if (!success) {
+			err = DB_ERROR;
+
+			mtr_commit(mtr);
+
+			return(err);
+		}
+	}
+
+	rec = btr_pcur_get_rec(pcur);
+	offsets = rec_get_offsets(rec, index, offsets_,
+				  ULINT_UNDEFINED, &heap);
+
+	if (!node->has_clust_rec_x_lock) {
+		err = lock_clust_rec_modify_check_and_lock(
+			0, btr_pcur_get_block(pcur),
+			rec, index, offsets, thr);
+		if (err != DB_SUCCESS) {
+			mtr_commit(mtr);
+			goto exit_func;
+		}
+	}
+
+	/* NOTE: the following function calls will also commit mtr */
+
+	if (node->is_delete) {
+		err = row_upd_del_mark_clust_rec(node, index, offsets,
+						 thr, check_ref, mtr);
+		if (err == DB_SUCCESS) {
+			node->state = UPD_NODE_UPDATE_ALL_SEC;
+			node->index = dict_table_get_next_index(index);
+		}
+exit_func:
+		if (UNIV_LIKELY_NULL(heap)) {
+			mem_heap_free(heap);
+		}
+		return(err);
+	}
+
+	/* If the update is made for MySQL, we already have the update vector
+	ready, else we have to do some evaluation: */
+
+	if (UNIV_UNLIKELY(!node->in_mysql_interface)) {
+		/* Copy the necessary columns from clust_rec and calculate the
+		new values to set */
+		row_upd_copy_columns(rec, offsets,
+				     UT_LIST_GET_FIRST(node->columns));
+		row_upd_eval_new_vals(node->update);
+	}
+
+	if (UNIV_LIKELY_NULL(heap)) {
+		mem_heap_free(heap);
+	}
+
+	if (node->cmpl_info & UPD_NODE_NO_ORD_CHANGE) {
+
+		err = row_upd_clust_rec(node, index, thr, mtr);
+		return(err);
+	}
+
+	row_upd_store_row(node);
+
+	if (row_upd_changes_ord_field_binary(node->row, index, node->update)) {
+
+		/* Update causes an ordering field (ordering fields within
+		the B-tree) of the clustered index record to change: perform
+		the update by delete marking and inserting.
+
+		TODO! What to do to the 'Halloween problem', where an update
+		moves the record forward in index so that it is again
+		updated when the cursor arrives there? Solution: the
+		read operation must check the undo record undo number when
+		choosing records to update. MySQL solves now the problem
+		externally! */
+
+		err = row_upd_clust_rec_by_insert(node, index, thr, check_ref,
+						  mtr);
+		if (err != DB_SUCCESS) {
+
+			return(err);
+		}
+
+		node->state = UPD_NODE_UPDATE_ALL_SEC;
+	} else {
+		err = row_upd_clust_rec(node, index, thr, mtr);
+
+		if (err != DB_SUCCESS) {
+
+			return(err);
+		}
+
+		node->state = UPD_NODE_UPDATE_SOME_SEC;
+	}
+
+	node->index = dict_table_get_next_index(index);
+
+	return(err);
+}
+
+/***************************************************************
+Updates the affected index records of a row. When the control is transferred
+to this node, we assume that we have a persistent cursor which was on a
+record, and the position of the cursor is stored in the cursor. */
+static
+ulint
+row_upd(
+/*====*/
+				/* out: DB_SUCCESS if operation successfully
+				completed, else error code or DB_LOCK_WAIT */
+	upd_node_t*	node,	/* in: row update node */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	ulint	err	= DB_SUCCESS;
+
+	ut_ad(node && thr);
+
+	if (UNIV_LIKELY(node->in_mysql_interface)) {
+
+		/* We do not get the cmpl_info value from the MySQL
+		interpreter: we must calculate it on the fly: */
+
+		if (node->is_delete
+		    || row_upd_changes_some_index_ord_field_binary(
+			    node->table, node->update)) {
+			node->cmpl_info = 0;
+		} else {
+			node->cmpl_info = UPD_NODE_NO_ORD_CHANGE;
+		}
+	}
+
+	if (node->state == UPD_NODE_UPDATE_CLUSTERED
+	    || node->state == UPD_NODE_INSERT_CLUSTERED) {
+
+		err = row_upd_clust_step(node, thr);
+
+		if (err != DB_SUCCESS) {
+
+			goto function_exit;
+		}
+	}
+
+	if (!node->is_delete && (node->cmpl_info & UPD_NODE_NO_ORD_CHANGE)) {
+
+		goto function_exit;
+	}
+
+	while (node->index != NULL) {
+		err = row_upd_sec_step(node, thr);
+
+		if (err != DB_SUCCESS) {
+
+			goto function_exit;
+		}
+
+		node->index = dict_table_get_next_index(node->index);
+	}
+
+function_exit:
+	if (err == DB_SUCCESS) {
+		/* Do some cleanup */
+
+		if (node->row != NULL) {
+			node->row = NULL;
+			node->ext = NULL;
+			node->upd_row = NULL;
+			node->upd_ext = NULL;
+			mem_heap_empty(node->heap);
+		}
+
+		node->state = UPD_NODE_UPDATE_CLUSTERED;
+	}
+
+	return(err);
+}
+
+/***************************************************************
+Updates a row in a table. This is a high-level function used in SQL execution
+graphs. */
+UNIV_INTERN
+que_thr_t*
+row_upd_step(
+/*=========*/
+				/* out: query thread to run next or NULL */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	upd_node_t*	node;
+	sel_node_t*	sel_node;
+	que_node_t*	parent;
+	ulint		err		= DB_SUCCESS;
+	trx_t*		trx;
+
+	ut_ad(thr);
+
+	trx = thr_get_trx(thr);
+
+	trx_start_if_not_started(trx);
+
+	node = thr->run_node;
+
+	sel_node = node->select;
+
+	parent = que_node_get_parent(node);
+
+	ut_ad(que_node_get_type(node) == QUE_NODE_UPDATE);
+
+	if (thr->prev_node == parent) {
+		node->state = UPD_NODE_SET_IX_LOCK;
+	}
+
+	if (node->state == UPD_NODE_SET_IX_LOCK) {
+
+		if (!node->has_clust_rec_x_lock) {
+			/* It may be that the current session has not yet
+			started its transaction, or it has been committed: */
+
+			err = lock_table(0, node->table, LOCK_IX, thr);
+
+			if (err != DB_SUCCESS) {
+
+				goto error_handling;
+			}
+		}
+
+		node->state = UPD_NODE_UPDATE_CLUSTERED;
+
+		if (node->searched_update) {
+			/* Reset the cursor */
+			sel_node->state = SEL_NODE_OPEN;
+
+			/* Fetch a row to update */
+
+			thr->run_node = sel_node;
+
+			return(thr);
+		}
+	}
+
+	/* sel_node is NULL if we are in the MySQL interface */
+
+	if (sel_node && (sel_node->state != SEL_NODE_FETCH)) {
+
+		if (!node->searched_update) {
+			/* An explicit cursor should be positioned on a row
+			to update */
+
+			ut_error;
+
+			err = DB_ERROR;
+
+			goto error_handling;
+		}
+
+		ut_ad(sel_node->state == SEL_NODE_NO_MORE_ROWS);
+
+		/* No more rows to update, or the select node performed the
+		updates directly in-place */
+
+		thr->run_node = parent;
+
+		return(thr);
+	}
+
+	/* DO THE CHECKS OF THE CONSISTENCY CONSTRAINTS HERE */
+
+	err = row_upd(node, thr);
+
+error_handling:
+	trx->error_state = err;
+
+	if (err != DB_SUCCESS) {
+		return(NULL);
+	}
+
+	/* DO THE TRIGGER ACTIONS HERE */
+
+	if (node->searched_update) {
+		/* Fetch next row to update */
+
+		thr->run_node = sel_node;
+	} else {
+		/* It was an explicit cursor update */
+
+		thr->run_node = parent;
+	}
+
+	node->state = UPD_NODE_UPDATE_CLUSTERED;
+
+	return(thr);
+}
diff --git a/storage/xtradb/row/row0vers.c b/storage/xtradb/row/row0vers.c
new file mode 100644
index 00000000000..3abba6d6fb8
--- /dev/null
+++ b/storage/xtradb/row/row0vers.c
@@ -0,0 +1,741 @@
+/*****************************************************************************
+
+Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Row versions
+
+Created 2/6/1997 Heikki Tuuri
+*******************************************************/
+
+#include "row0vers.h"
+
+#ifdef UNIV_NONINL
+#include "row0vers.ic"
+#endif
+
+#include "dict0dict.h"
+#include "dict0boot.h"
+#include "btr0btr.h"
+#include "mach0data.h"
+#include "trx0rseg.h"
+#include "trx0trx.h"
+#include "trx0roll.h"
+#include "trx0undo.h"
+#include "trx0purge.h"
+#include "trx0rec.h"
+#include "que0que.h"
+#include "row0row.h"
+#include "row0upd.h"
+#include "rem0cmp.h"
+#include "read0read.h"
+#include "lock0lock.h"
+
+/*********************************************************************
+Finds out if an active transaction has inserted or modified a secondary
+index record. NOTE: the kernel mutex is temporarily released in this
+function! */
+UNIV_INTERN
+trx_t*
+row_vers_impl_x_locked_off_kernel(
+/*==============================*/
+				/* out: NULL if committed, else the active
+				transaction; NOTE that the kernel mutex is
+				temporarily released! */
+	const rec_t*	rec,	/* in: record in a secondary index */
+	dict_index_t*	index,	/* in: the secondary index */
+	const ulint*	offsets)/* in: rec_get_offsets(rec, index) */
+{
+	dict_index_t*	clust_index;
+	rec_t*		clust_rec;
+	ulint*		clust_offsets;
+	rec_t*		version;
+	dulint		trx_id;
+	mem_heap_t*	heap;
+	mem_heap_t*	heap2;
+	dtuple_t*	row;
+	dtuple_t*	entry	= NULL; /* assignment to eliminate compiler
+					warning */
+	trx_t*		trx;
+	ulint		rec_del;
+	ulint		err;
+	mtr_t		mtr;
+	ulint		comp;
+
+	ut_ad(mutex_own(&kernel_mutex));
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(!rw_lock_own(&(purge_sys->latch), RW_LOCK_SHARED));
+#endif /* UNIV_SYNC_DEBUG */
+
+	mutex_exit(&kernel_mutex);
+
+	mtr_start(&mtr);
+
+	/* Search for the clustered index record: this is a time-consuming
+	operation: therefore we release the kernel mutex; also, the release
+	is required by the latching order convention. The latch on the
+	clustered index locks the top of the stack of versions. We also
+	reserve purge_latch to lock the bottom of the version stack. */
+
+	clust_rec = row_get_clust_rec(BTR_SEARCH_LEAF, rec, index,
+				      &clust_index, &mtr);
+	if (!clust_rec) {
+		/* In a rare case it is possible that no clust rec is found
+		for a secondary index record: if in row0umod.c
+		row_undo_mod_remove_clust_low() we have already removed the
+		clust rec, while purge is still cleaning and removing
+		secondary index records associated with earlier versions of
+		the clustered index record. In that case there cannot be
+		any implicit lock on the secondary index record, because
+		an active transaction which has modified the secondary index
+		record has also modified the clustered index record. And in
+		a rollback we always undo the modifications to secondary index
+		records before the clustered index record. */
+
+		mutex_enter(&kernel_mutex);
+		mtr_commit(&mtr);
+
+		return(NULL);
+	}
+
+	heap = mem_heap_create(1024);
+	clust_offsets = rec_get_offsets(clust_rec, clust_index, NULL,
+					ULINT_UNDEFINED, &heap);
+	trx_id = row_get_rec_trx_id(clust_rec, clust_index, clust_offsets);
+
+	mtr_s_lock(&(purge_sys->latch), &mtr);
+
+	mutex_enter(&kernel_mutex);
+
+	trx = NULL;
+	if (!trx_is_active(trx_id)) {
+		/* The transaction that modified or inserted clust_rec is no
+		longer active: no implicit lock on rec */
+		goto exit_func;
+	}
+
+	if (!lock_check_trx_id_sanity(trx_id, clust_rec, clust_index,
+				      clust_offsets, TRUE)) {
+		/* Corruption noticed: try to avoid a crash by returning */
+		goto exit_func;
+	}
+
+	comp = page_rec_is_comp(rec);
+	ut_ad(index->table == clust_index->table);
+	ut_ad(!!comp == dict_table_is_comp(index->table));
+	ut_ad(!comp == !page_rec_is_comp(clust_rec));
+
+	/* We look up if some earlier version, which was modified by the trx_id
+	transaction, of the clustered index record would require rec to be in
+	a different state (delete marked or unmarked, or have different field
+	values, or not existing). If there is such a version, then rec was
+	modified by the trx_id transaction, and it has an implicit x-lock on
+	rec. Note that if clust_rec itself would require rec to be in a
+	different state, then the trx_id transaction has not yet had time to
+	modify rec, and does not necessarily have an implicit x-lock on rec. */
+
+	rec_del = rec_get_deleted_flag(rec, comp);
+	trx = NULL;
+
+	version = clust_rec;
+
+	for (;;) {
+		rec_t*		prev_version;
+		ulint		vers_del;
+		row_ext_t*	ext;
+		dulint		prev_trx_id;
+
+		mutex_exit(&kernel_mutex);
+
+		/* While we retrieve an earlier version of clust_rec, we
+		release the kernel mutex, because it may take time to access
+		the disk. After the release, we have to check if the trx_id
+		transaction is still active. We keep the semaphore in mtr on
+		the clust_rec page, so that no other transaction can update
+		it and get an implicit x-lock on rec. */
+
+		heap2 = heap;
+		heap = mem_heap_create(1024);
+		err = trx_undo_prev_version_build(clust_rec, &mtr, version,
+						  clust_index, clust_offsets,
+						  heap, &prev_version);
+		mem_heap_free(heap2); /* free version and clust_offsets */
+
+		if (prev_version == NULL) {
+			mutex_enter(&kernel_mutex);
+
+			if (!trx_is_active(trx_id)) {
+				/* Transaction no longer active: no
+				implicit x-lock */
+
+				break;
+			}
+
+			/* If the transaction is still active,
+			clust_rec must be a fresh insert, because no
+			previous version was found. */
+			ut_ad(err == DB_SUCCESS);
+
+			/* It was a freshly inserted version: there is an
+			implicit x-lock on rec */
+
+			trx = trx_get_on_id(trx_id);
+
+			break;
+		}
+
+		clust_offsets = rec_get_offsets(prev_version, clust_index,
+						NULL, ULINT_UNDEFINED, &heap);
+
+		vers_del = rec_get_deleted_flag(prev_version, comp);
+		prev_trx_id = row_get_rec_trx_id(prev_version, clust_index,
+						 clust_offsets);
+
+		/* If the trx_id and prev_trx_id are different and if
+		the prev_version is marked deleted then the
+		prev_trx_id must have already committed for the trx_id
+		to be able to modify the row. Therefore, prev_trx_id
+		cannot hold any implicit lock. */
+		if (vers_del && 0 != ut_dulint_cmp(trx_id, prev_trx_id)) {
+
+			mutex_enter(&kernel_mutex);
+			break;
+		}
+
+		/* The stack of versions is locked by mtr.  Thus, it
+		is safe to fetch the prefixes for externally stored
+		columns. */
+		row = row_build(ROW_COPY_POINTERS, clust_index, prev_version,
+				clust_offsets, NULL, &ext, heap);
+		entry = row_build_index_entry(row, ext, index, heap);
+		/* entry may be NULL if a record was inserted in place
+		of a deleted record, and the BLOB pointers of the new
+		record were not initialized yet.  But in that case,
+		prev_version should be NULL. */
+		ut_a(entry);
+
+		mutex_enter(&kernel_mutex);
+
+		if (!trx_is_active(trx_id)) {
+			/* Transaction no longer active: no implicit x-lock */
+
+			break;
+		}
+
+		/* If we get here, we know that the trx_id transaction is
+		still active and it has modified prev_version. Let us check
+		if prev_version would require rec to be in a different
+		state. */
+
+		/* The previous version of clust_rec must be
+		accessible, because the transaction is still active
+		and clust_rec was not a fresh insert. */
+		ut_ad(err == DB_SUCCESS);
+
+		/* We check if entry and rec are identified in the alphabetical
+		ordering */
+		if (0 == cmp_dtuple_rec(entry, rec, offsets)) {
+			/* The delete marks of rec and prev_version should be
+			equal for rec to be in the state required by
+			prev_version */
+
+			if (rec_del != vers_del) {
+				trx = trx_get_on_id(trx_id);
+
+				break;
+			}
+
+			/* It is possible that the row was updated so that the
+			secondary index record remained the same in
+			alphabetical ordering, but the field values changed
+			still. For example, 'abc' -> 'ABC'. Check also that. */
+
+			dtuple_set_types_binary(entry,
+						dtuple_get_n_fields(entry));
+			if (0 != cmp_dtuple_rec(entry, rec, offsets)) {
+
+				trx = trx_get_on_id(trx_id);
+
+				break;
+			}
+		} else if (!rec_del) {
+			/* The delete mark should be set in rec for it to be
+			in the state required by prev_version */
+
+			trx = trx_get_on_id(trx_id);
+
+			break;
+		}
+
+		if (0 != ut_dulint_cmp(trx_id, prev_trx_id)) {
+			/* The versions modified by the trx_id transaction end
+			to prev_version: no implicit x-lock */
+
+			break;
+		}
+
+		version = prev_version;
+	}/* for (;;) */
+
+exit_func:
+	mtr_commit(&mtr);
+	mem_heap_free(heap);
+
+	return(trx);
+}
+
+/*********************************************************************
+Finds out if we must preserve a delete marked earlier version of a clustered
+index record, because it is >= the purge view. */
+UNIV_INTERN
+ibool
+row_vers_must_preserve_del_marked(
+/*==============================*/
+			/* out: TRUE if earlier version should be preserved */
+	dulint	trx_id,	/* in: transaction id in the version */
+	mtr_t*	mtr)	/* in: mtr holding the latch on the clustered index
+			record; it will also hold the latch on purge_view */
+{
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(!rw_lock_own(&(purge_sys->latch), RW_LOCK_SHARED));
+#endif /* UNIV_SYNC_DEBUG */
+
+	mtr_s_lock(&(purge_sys->latch), mtr);
+
+	if (trx_purge_update_undo_must_exist(trx_id)) {
+
+		/* A purge operation is not yet allowed to remove this
+		delete marked record */
+
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
+
+/*********************************************************************
+Finds out if a version of the record, where the version >= the current
+purge view, should have ientry as its secondary index entry. We check
+if there is any not delete marked version of the record where the trx
+id >= purge view, and the secondary index entry and ientry are identified in
+the alphabetical ordering; exactly in this case we return TRUE. */
+UNIV_INTERN
+ibool
+row_vers_old_has_index_entry(
+/*=========================*/
+				/* out: TRUE if earlier version should have */
+	ibool		also_curr,/* in: TRUE if also rec is included in the
+				versions to search; otherwise only versions
+				prior to it are searched */
+	const rec_t*	rec,	/* in: record in the clustered index; the
+				caller must have a latch on the page */
+	mtr_t*		mtr,	/* in: mtr holding the latch on rec; it will
+				also hold the latch on purge_view */
+	dict_index_t*	index,	/* in: the secondary index */
+	const dtuple_t*	ientry)	/* in: the secondary index entry */
+{
+	const rec_t*	version;
+	rec_t*		prev_version;
+	dict_index_t*	clust_index;
+	ulint*		clust_offsets;
+	mem_heap_t*	heap;
+	mem_heap_t*	heap2;
+	const dtuple_t*	row;
+	const dtuple_t*	entry;
+	ulint		err;
+	ulint		comp;
+
+	ut_ad(mtr_memo_contains_page(mtr, rec, MTR_MEMO_PAGE_X_FIX)
+	      || mtr_memo_contains_page(mtr, rec, MTR_MEMO_PAGE_S_FIX));
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(!rw_lock_own(&(purge_sys->latch), RW_LOCK_SHARED));
+#endif /* UNIV_SYNC_DEBUG */
+	mtr_s_lock(&(purge_sys->latch), mtr);
+
+	clust_index = dict_table_get_first_index(index->table);
+
+	comp = page_rec_is_comp(rec);
+	ut_ad(!dict_table_is_comp(index->table) == !comp);
+	heap = mem_heap_create(1024);
+	clust_offsets = rec_get_offsets(rec, clust_index, NULL,
+					ULINT_UNDEFINED, &heap);
+
+	if (also_curr && !rec_get_deleted_flag(rec, comp)) {
+		row_ext_t*	ext;
+
+		/* The stack of versions is locked by mtr.
+		Thus, it is safe to fetch the prefixes for
+		externally stored columns. */
+		row = row_build(ROW_COPY_POINTERS, clust_index,
+				rec, clust_offsets, NULL, &ext, heap);
+		entry = row_build_index_entry(row, ext, index, heap);
+
+		/* If entry == NULL, the record contains unset BLOB
+		pointers.  This must be a freshly inserted record.  If
+		this is called from
+		row_purge_remove_sec_if_poss_low(), the thread will
+		hold latches on the clustered index and the secondary
+		index.  Because the insert works in three steps:
+
+			(1) insert the record to clustered index
+			(2) store the BLOBs and update BLOB pointers
+			(3) insert records to secondary indexes
+
+		the purge thread can safely ignore freshly inserted
+		records and delete the secondary index record.  The
+		thread that inserted the new record will be inserting
+		the secondary index records. */
+
+		/* NOTE that we cannot do the comparison as binary
+		fields because the row is maybe being modified so that
+		the clustered index record has already been updated to
+		a different binary value in a char field, but the
+		collation identifies the old and new value anyway! */
+		if (entry && !dtuple_coll_cmp(ientry, entry)) {
+
+			mem_heap_free(heap);
+
+			return(TRUE);
+		}
+	}
+
+	version = rec;
+
+	for (;;) {
+		heap2 = heap;
+		heap = mem_heap_create(1024);
+		err = trx_undo_prev_version_build(rec, mtr, version,
+						  clust_index, clust_offsets,
+						  heap, &prev_version);
+		mem_heap_free(heap2); /* free version and clust_offsets */
+
+		if (err != DB_SUCCESS || !prev_version) {
+			/* Versions end here */
+
+			mem_heap_free(heap);
+
+			return(FALSE);
+		}
+
+		clust_offsets = rec_get_offsets(prev_version, clust_index,
+						NULL, ULINT_UNDEFINED, &heap);
+
+		if (!rec_get_deleted_flag(prev_version, comp)) {
+			row_ext_t*	ext;
+
+			/* The stack of versions is locked by mtr.
+			Thus, it is safe to fetch the prefixes for
+			externally stored columns. */
+			row = row_build(ROW_COPY_POINTERS, clust_index,
+					prev_version, clust_offsets,
+					NULL, &ext, heap);
+			entry = row_build_index_entry(row, ext, index, heap);
+
+			/* If entry == NULL, the record contains unset
+			BLOB pointers.  This must be a freshly
+			inserted record that we can safely ignore.
+			For the justification, see the comments after
+			the previous row_build_index_entry() call. */
+
+			/* NOTE that we cannot do the comparison as binary
+			fields because maybe the secondary index record has
+			already been updated to a different binary value in
+			a char field, but the collation identifies the old
+			and new value anyway! */
+
+			if (entry && !dtuple_coll_cmp(ientry, entry)) {
+
+				mem_heap_free(heap);
+
+				return(TRUE);
+			}
+		}
+
+		version = prev_version;
+	}
+}
+
+/*********************************************************************
+Constructs the version of a clustered index record which a consistent
+read should see. We assume that the trx id stored in rec is such that
+the consistent read should not see rec in its present version. */
+UNIV_INTERN
+ulint
+row_vers_build_for_consistent_read(
+/*===============================*/
+				/* out: DB_SUCCESS or DB_MISSING_HISTORY */
+	const rec_t*	rec,	/* in: record in a clustered index; the
+				caller must have a latch on the page; this
+				latch locks the top of the stack of versions
+				of this records */
+	mtr_t*		mtr,	/* in: mtr holding the latch on rec */
+	dict_index_t*	index,	/* in: the clustered index */
+	ulint**		offsets,/* in/out: offsets returned by
+				rec_get_offsets(rec, index) */
+	read_view_t*	view,	/* in: the consistent read view */
+	mem_heap_t**	offset_heap,/* in/out: memory heap from which
+				the offsets are allocated */
+	mem_heap_t*	in_heap,/* in: memory heap from which the memory for
+				*old_vers is allocated; memory for possible
+				intermediate versions is allocated and freed
+				locally within the function */
+	rec_t**		old_vers)/* out, own: old version, or NULL if the
+				record does not exist in the view, that is,
+				it was freshly inserted afterwards */
+{
+	const rec_t*	version;
+	rec_t*		prev_version;
+	dulint		trx_id;
+	mem_heap_t*	heap		= NULL;
+	byte*		buf;
+	ulint		err;
+
+	ut_ad(dict_index_is_clust(index));
+	ut_ad(mtr_memo_contains_page(mtr, rec, MTR_MEMO_PAGE_X_FIX)
+	      || mtr_memo_contains_page(mtr, rec, MTR_MEMO_PAGE_S_FIX));
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(!rw_lock_own(&(purge_sys->latch), RW_LOCK_SHARED));
+#endif /* UNIV_SYNC_DEBUG */
+
+	ut_ad(rec_offs_validate(rec, index, *offsets));
+
+	trx_id = row_get_rec_trx_id(rec, index, *offsets);
+
+	ut_ad(!read_view_sees_trx_id(view, trx_id));
+
+	rw_lock_s_lock(&(purge_sys->latch));
+	version = rec;
+
+	for (;;) {
+		mem_heap_t*	heap2	= heap;
+		trx_undo_rec_t* undo_rec;
+		dulint		roll_ptr;
+		dulint		undo_no;
+		heap = mem_heap_create(1024);
+
+		/* If we have high-granularity consistent read view and
+		creating transaction of the view is the same as trx_id in
+		the record we see this record only in the case when
+		undo_no of the record is < undo_no in the view. */
+
+		if (view->type == VIEW_HIGH_GRANULARITY
+		    && ut_dulint_cmp(view->creator_trx_id, trx_id) == 0) {
+
+			roll_ptr = row_get_rec_roll_ptr(version, index,
+							*offsets);
+			undo_rec = trx_undo_get_undo_rec_low(roll_ptr, heap);
+			undo_no = trx_undo_rec_get_undo_no(undo_rec);
+			mem_heap_empty(heap);
+
+			if (ut_dulint_cmp(view->undo_no, undo_no) > 0) {
+				/* The view already sees this version: we can
+				copy it to in_heap and return */
+
+				buf = mem_heap_alloc(in_heap,
+						     rec_offs_size(*offsets));
+				*old_vers = rec_copy(buf, version, *offsets);
+				rec_offs_make_valid(*old_vers, index,
+						    *offsets);
+				err = DB_SUCCESS;
+
+				break;
+			}
+		}
+
+		err = trx_undo_prev_version_build(rec, mtr, version, index,
+						  *offsets, heap,
+						  &prev_version);
+		if (heap2) {
+			mem_heap_free(heap2); /* free version */
+		}
+
+		if (err != DB_SUCCESS) {
+			break;
+		}
+
+		if (prev_version == NULL) {
+			/* It was a freshly inserted version */
+			*old_vers = NULL;
+			err = DB_SUCCESS;
+
+			break;
+		}
+
+		*offsets = rec_get_offsets(prev_version, index, *offsets,
+					   ULINT_UNDEFINED, offset_heap);
+
+		trx_id = row_get_rec_trx_id(prev_version, index, *offsets);
+
+		if (read_view_sees_trx_id(view, trx_id)) {
+
+			/* The view already sees this version: we can copy
+			it to in_heap and return */
+
+			buf = mem_heap_alloc(in_heap, rec_offs_size(*offsets));
+			*old_vers = rec_copy(buf, prev_version, *offsets);
+			rec_offs_make_valid(*old_vers, index, *offsets);
+			err = DB_SUCCESS;
+
+			break;
+		}
+
+		version = prev_version;
+	}/* for (;;) */
+
+	mem_heap_free(heap);
+	rw_lock_s_unlock(&(purge_sys->latch));
+
+	return(err);
+}
+
+/*********************************************************************
+Constructs the last committed version of a clustered index record,
+which should be seen by a semi-consistent read. */
+UNIV_INTERN
+ulint
+row_vers_build_for_semi_consistent_read(
+/*====================================*/
+				/* out: DB_SUCCESS or DB_MISSING_HISTORY */
+	const rec_t*	rec,	/* in: record in a clustered index; the
+				caller must have a latch on the page; this
+				latch locks the top of the stack of versions
+				of this records */
+	mtr_t*		mtr,	/* in: mtr holding the latch on rec */
+	dict_index_t*	index,	/* in: the clustered index */
+	ulint**		offsets,/* in/out: offsets returned by
+				rec_get_offsets(rec, index) */
+	mem_heap_t**	offset_heap,/* in/out: memory heap from which
+				the offsets are allocated */
+	mem_heap_t*	in_heap,/* in: memory heap from which the memory for
+				*old_vers is allocated; memory for possible
+				intermediate versions is allocated and freed
+				locally within the function */
+	const rec_t**	old_vers)/* out: rec, old version, or NULL if the
+				record does not exist in the view, that is,
+				it was freshly inserted afterwards */
+{
+	const rec_t*	version;
+	mem_heap_t*	heap		= NULL;
+	byte*		buf;
+	ulint		err;
+	dulint		rec_trx_id	= ut_dulint_zero;
+
+	ut_ad(dict_index_is_clust(index));
+	ut_ad(mtr_memo_contains_page(mtr, rec, MTR_MEMO_PAGE_X_FIX)
+	      || mtr_memo_contains_page(mtr, rec, MTR_MEMO_PAGE_S_FIX));
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(!rw_lock_own(&(purge_sys->latch), RW_LOCK_SHARED));
+#endif /* UNIV_SYNC_DEBUG */
+
+	ut_ad(rec_offs_validate(rec, index, *offsets));
+
+	rw_lock_s_lock(&(purge_sys->latch));
+	/* The S-latch on purge_sys prevents the purge view from
+	changing.  Thus, if we have an uncommitted transaction at
+	this point, then purge cannot remove its undo log even if
+	the transaction could commit now. */
+
+	version = rec;
+
+	for (;;) {
+		trx_t*		version_trx;
+		mem_heap_t*	heap2;
+		rec_t*		prev_version;
+		dulint		version_trx_id;
+
+		version_trx_id = row_get_rec_trx_id(version, index, *offsets);
+		if (rec == version) {
+			rec_trx_id = version_trx_id;
+		}
+
+		mutex_enter(&kernel_mutex);
+		version_trx = trx_get_on_id(version_trx_id);
+		mutex_exit(&kernel_mutex);
+
+		if (!version_trx
+		    || version_trx->conc_state == TRX_NOT_STARTED
+		    || version_trx->conc_state == TRX_COMMITTED_IN_MEMORY) {
+
+			/* We found a version that belongs to a
+			committed transaction: return it. */
+
+			if (rec == version) {
+				*old_vers = rec;
+				err = DB_SUCCESS;
+				break;
+			}
+
+			/* We assume that a rolled-back transaction stays in
+			TRX_ACTIVE state until all the changes have been
+			rolled back and the transaction is removed from
+			the global list of transactions. */
+
+			if (!ut_dulint_cmp(rec_trx_id, version_trx_id)) {
+				/* The transaction was committed while
+				we searched for earlier versions.
+				Return the current version as a
+				semi-consistent read. */
+
+				version = rec;
+				*offsets = rec_get_offsets(version,
+							   index, *offsets,
+							   ULINT_UNDEFINED,
+							   offset_heap);
+			}
+
+			buf = mem_heap_alloc(in_heap, rec_offs_size(*offsets));
+			*old_vers = rec_copy(buf, version, *offsets);
+			rec_offs_make_valid(*old_vers, index, *offsets);
+			err = DB_SUCCESS;
+
+			break;
+		}
+
+		heap2 = heap;
+		heap = mem_heap_create(1024);
+
+		err = trx_undo_prev_version_build(rec, mtr, version, index,
+						  *offsets, heap,
+						  &prev_version);
+		if (heap2) {
+			mem_heap_free(heap2); /* free version */
+		}
+
+		if (UNIV_UNLIKELY(err != DB_SUCCESS)) {
+			break;
+		}
+
+		if (prev_version == NULL) {
+			/* It was a freshly inserted version */
+			*old_vers = NULL;
+			err = DB_SUCCESS;
+
+			break;
+		}
+
+		version = prev_version;
+		*offsets = rec_get_offsets(version, index, *offsets,
+					   ULINT_UNDEFINED, offset_heap);
+	}/* for (;;) */
+
+	if (heap) {
+		mem_heap_free(heap);
+	}
+	rw_lock_s_unlock(&(purge_sys->latch));
+
+	return(err);
+}
diff --git a/storage/xtradb/scripts/install_innodb_plugins.sql b/storage/xtradb/scripts/install_innodb_plugins.sql
new file mode 100644
index 00000000000..b1a6f8e2ba0
--- /dev/null
+++ b/storage/xtradb/scripts/install_innodb_plugins.sql
@@ -0,0 +1,14 @@
+-- execute these to install InnoDB if it is built as a dynamic plugin
+INSTALL PLUGIN innodb SONAME 'ha_innodb.so';
+INSTALL PLUGIN innodb_trx SONAME 'ha_innodb.so';
+INSTALL PLUGIN innodb_locks SONAME 'ha_innodb.so';
+INSTALL PLUGIN innodb_lock_waits SONAME 'ha_innodb.so';
+INSTALL PLUGIN innodb_cmp SONAME 'ha_innodb.so';
+INSTALL PLUGIN innodb_cmp_reset SONAME 'ha_innodb.so';
+INSTALL PLUGIN innodb_cmpmem SONAME 'ha_innodb.so';
+INSTALL PLUGIN innodb_cmpmem_reset SONAME 'ha_innodb.so';
+INSTALL PLUGIN XTRADB_ENHANCEMENTS SONAME 'ha_innodb.so';
+INSTALL PLUGIN INNODB_BUFFER_POOL_PAGES SONAME 'ha_innodb.so';
+INSTALL PLUGIN INNODB_BUFFER_POOL_PAGES_BLOB SONAME 'ha_innodb.so';
+INSTALL PLUGIN INNODB_BUFFER_POOL_PAGES_INDEX SONAME 'ha_innodb.so';
+INSTALL PLUGIN innodb_rseg SONAME 'ha_innodb.so';
diff --git a/storage/xtradb/scripts/install_innodb_plugins_win.sql b/storage/xtradb/scripts/install_innodb_plugins_win.sql
new file mode 100644
index 00000000000..8c94b4e240d
--- /dev/null
+++ b/storage/xtradb/scripts/install_innodb_plugins_win.sql
@@ -0,0 +1,9 @@
+-- execute these to install InnoDB if it is built as a dynamic plugin
+INSTALL PLUGIN innodb SONAME 'ha_innodb.dll';
+INSTALL PLUGIN innodb_trx SONAME 'ha_innodb.dll';
+INSTALL PLUGIN innodb_locks SONAME 'ha_innodb.dll';
+INSTALL PLUGIN innodb_lock_waits SONAME 'ha_innodb.dll';
+INSTALL PLUGIN innodb_cmp SONAME 'ha_innodb.dll';
+INSTALL PLUGIN innodb_cmp_reset SONAME 'ha_innodb.dll';
+INSTALL PLUGIN innodb_cmpmem SONAME 'ha_innodb.dll';
+INSTALL PLUGIN innodb_cmpmem_reset SONAME 'ha_innodb.dll';
diff --git a/storage/xtradb/srv/srv0que.c b/storage/xtradb/srv/srv0que.c
new file mode 100644
index 00000000000..344aaed8775
--- /dev/null
+++ b/storage/xtradb/srv/srv0que.c
@@ -0,0 +1,126 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Server query execution
+
+Created 6/5/1996 Heikki Tuuri
+*******************************************************/
+
+#include "srv0que.h"
+
+#include "srv0srv.h"
+#include "sync0sync.h"
+#include "os0thread.h"
+#include "usr0sess.h"
+#include "que0que.h"
+
+/**************************************************************************
+Checks if there is work to do in the server task queue. If there is, the
+thread starts processing a task. Before leaving, it again checks the task
+queue and picks a new task if any exists. This is called by a SRV_WORKER
+thread. */
+UNIV_INTERN
+void
+srv_que_task_queue_check(void)
+/*==========================*/
+{
+	que_thr_t*	thr;
+
+	for (;;) {
+		mutex_enter(&kernel_mutex);
+
+		thr = UT_LIST_GET_FIRST(srv_sys->tasks);
+
+		if (thr == NULL) {
+			mutex_exit(&kernel_mutex);
+
+			return;
+		}
+
+		UT_LIST_REMOVE(queue, srv_sys->tasks, thr);
+
+		mutex_exit(&kernel_mutex);
+
+		que_run_threads(thr);
+	}
+}
+
+/**************************************************************************
+Performs round-robin on the server tasks. This is called by a SRV_WORKER
+thread every second or so. */
+UNIV_INTERN
+que_thr_t*
+srv_que_round_robin(
+/*================*/
+				/* out: the new (may be == thr) query thread
+				to run */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	que_thr_t*	new_thr;
+
+	ut_ad(thr);
+	ut_ad(thr->state == QUE_THR_RUNNING);
+
+	mutex_enter(&kernel_mutex);
+
+	UT_LIST_ADD_LAST(queue, srv_sys->tasks, thr);
+
+	new_thr = UT_LIST_GET_FIRST(srv_sys->tasks);
+
+	mutex_exit(&kernel_mutex);
+
+	return(new_thr);
+}
+
+/**************************************************************************
+Enqueues a task to server task queue and releases a worker thread, if there
+is a suspended one. */
+UNIV_INTERN
+void
+srv_que_task_enqueue_low(
+/*=====================*/
+	que_thr_t*	thr)	/* in: query thread */
+{
+	ut_ad(thr);
+	ut_ad(mutex_own(&kernel_mutex));
+
+	UT_LIST_ADD_LAST(queue, srv_sys->tasks, thr);
+
+	srv_release_threads(SRV_WORKER, 1);
+}
+
+/**************************************************************************
+Enqueues a task to server task queue and releases a worker thread, if there
+is a suspended one. */
+UNIV_INTERN
+void
+srv_que_task_enqueue(
+/*=================*/
+	que_thr_t*	thr)	/* in: query thread */
+{
+	ut_ad(thr);
+
+	ut_a(0);	/* Under MySQL this is never called */
+
+	mutex_enter(&kernel_mutex);
+
+	srv_que_task_enqueue_low(thr);
+
+	mutex_exit(&kernel_mutex);
+}
diff --git a/storage/xtradb/srv/srv0srv.c b/storage/xtradb/srv/srv0srv.c
new file mode 100644
index 00000000000..f448afb0dad
--- /dev/null
+++ b/storage/xtradb/srv/srv0srv.c
@@ -0,0 +1,2814 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+Copyright (c) 2008, Google Inc.
+
+Portions of this file contain modifications contributed and copyrighted by
+Google, Inc. Those modifications are gratefully acknowledged and are described
+briefly in the InnoDB documentation. The contributions by Google are
+incorporated with their permission, and subject to the conditions contained in
+the file COPYING.Google.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The database server main program
+
+NOTE: SQL Server 7 uses something which the documentation
+calls user mode scheduled threads (UMS threads). One such
+thread is usually allocated per processor. Win32
+documentation does not know any UMS threads, which suggests
+that the concept is internal to SQL Server 7. It may mean that
+SQL Server 7 does all the scheduling of threads itself, even
+in i/o waits. We should maybe modify InnoDB to use the same
+technique, because thread switches within NT may be too slow.
+
+SQL Server 7 also mentions fibers, which are cooperatively
+scheduled threads. They can boost performance by 5 %,
+according to the Delaney and Soukup's book.
+
+Windows 2000 will have something called thread pooling
+(see msdn website), which we could possibly use.
+
+Another possibility could be to use some very fast user space
+thread library. This might confuse NT though.
+
+Created 10/8/1995 Heikki Tuuri
+*******************************************************/
+
+/* Dummy comment */
+#include "srv0srv.h"
+
+#include "ut0mem.h"
+#include "ut0ut.h"
+#include "os0proc.h"
+#include "mem0mem.h"
+#include "mem0pool.h"
+#include "sync0sync.h"
+#include "thr0loc.h"
+#include "que0que.h"
+#include "srv0que.h"
+#include "log0recv.h"
+#include "pars0pars.h"
+#include "usr0sess.h"
+#include "lock0lock.h"
+#include "trx0purge.h"
+#include "ibuf0ibuf.h"
+#include "buf0flu.h"
+#include "buf0lru.h"
+#include "btr0sea.h"
+#include "dict0load.h"
+#include "dict0boot.h"
+#include "srv0start.h"
+#include "row0mysql.h"
+#include "ha_prototypes.h"
+#include "trx0i_s.h"
+
+/* This is set to TRUE if the MySQL user has set it in MySQL; currently
+affects only FOREIGN KEY definition parsing */
+UNIV_INTERN ibool	srv_lower_case_table_names	= FALSE;
+
+/* The following counter is incremented whenever there is some user activity
+in the server */
+UNIV_INTERN ulint	srv_activity_count	= 0;
+
+/* The following is the maximum allowed duration of a lock wait. */
+UNIV_INTERN ulint	srv_fatal_semaphore_wait_threshold = 600;
+
+/* How much data manipulation language (DML) statements need to be delayed,
+in microseconds, in order to reduce the lagging of the purge thread. */
+UNIV_INTERN ulint	srv_dml_needed_delay = 0;
+
+UNIV_INTERN ibool	srv_lock_timeout_and_monitor_active = FALSE;
+UNIV_INTERN ibool	srv_error_monitor_active = FALSE;
+
+UNIV_INTERN const char*	srv_main_thread_op_info = "";
+
+/* Prefix used by MySQL to indicate pre-5.1 table name encoding */
+UNIV_INTERN const char	srv_mysql50_table_name_prefix[9] = "#mysql50#";
+
+/* Server parameters which are read from the initfile */
+
+/* The following three are dir paths which are catenated before file
+names, where the file name itself may also contain a path */
+
+UNIV_INTERN char*	srv_data_home	= NULL;
+#ifdef UNIV_LOG_ARCHIVE
+UNIV_INTERN char*	srv_arch_dir	= NULL;
+#endif /* UNIV_LOG_ARCHIVE */
+
+/* store to its own file each table created by an user; data
+dictionary tables are in the system tablespace 0 */
+UNIV_INTERN my_bool	srv_file_per_table;
+/* The file format to use on new *.ibd files. */
+UNIV_INTERN ulint	srv_file_format = 0;
+/* Whether to check file format during startup a value of 
+DICT_TF_FORMAT_MAX + 1 means no checking ie. FALSE.  The default is to
+set it to the highest format we support. */
+UNIV_INTERN ulint	srv_check_file_format_at_startup = DICT_TF_FORMAT_MAX;
+
+#if DICT_TF_FORMAT_51
+# error "DICT_TF_FORMAT_51 must be 0!"
+#endif
+/* Place locks to records only i.e. do not use next-key locking except
+on duplicate key checking and foreign key checking */
+UNIV_INTERN ibool	srv_locks_unsafe_for_binlog = FALSE;
+
+UNIV_INTERN ulint	srv_n_data_files = 0;
+UNIV_INTERN char**	srv_data_file_names = NULL;
+/* size in database pages */
+UNIV_INTERN ulint*	srv_data_file_sizes = NULL;
+
+UNIV_INTERN ibool	srv_extra_undoslots = FALSE;
+
+/* if TRUE, then we auto-extend the last data file */
+UNIV_INTERN ibool	srv_auto_extend_last_data_file	= FALSE;
+/* if != 0, this tells the max size auto-extending may increase the
+last data file size */
+UNIV_INTERN ulint	srv_last_file_size_max	= 0;
+/* If the last data file is auto-extended, we add this
+many pages to it at a time */
+UNIV_INTERN ulong	srv_auto_extend_increment = 8;
+UNIV_INTERN ulint*	srv_data_file_is_raw_partition = NULL;
+
+/* If the following is TRUE we do not allow inserts etc. This protects
+the user from forgetting the 'newraw' keyword to my.cnf */
+
+UNIV_INTERN ibool	srv_created_new_raw	= FALSE;
+
+UNIV_INTERN char**	srv_log_group_home_dirs = NULL;
+
+UNIV_INTERN ulint	srv_n_log_groups	= ULINT_MAX;
+UNIV_INTERN ulint	srv_n_log_files		= ULINT_MAX;
+/* size in database pages */
+UNIV_INTERN ulint	srv_log_file_size	= ULINT_MAX;
+/* size in database pages */
+UNIV_INTERN ulint	srv_log_buffer_size	= ULINT_MAX;
+UNIV_INTERN ulong	srv_flush_log_at_trx_commit = 1;
+
+UNIV_INTERN ulint  srv_show_locks_held     = 10;
+UNIV_INTERN ulint  srv_show_verbose_locks  = 0;
+
+
+/* The sort order table of the MySQL latin1_swedish_ci character set
+collation */
+UNIV_INTERN const byte*	srv_latin1_ordering;
+
+/* use os/external memory allocator */
+UNIV_INTERN my_bool	srv_use_sys_malloc	= TRUE;
+/* requested size in kilobytes */
+UNIV_INTERN ulint	srv_buf_pool_size	= ULINT_MAX;
+/* previously requested size */
+UNIV_INTERN ulint	srv_buf_pool_old_size;
+/* current size in kilobytes */
+UNIV_INTERN ulint	srv_buf_pool_curr_size	= 0;
+/* size in bytes */
+UNIV_INTERN ulint	srv_mem_pool_size	= ULINT_MAX;
+UNIV_INTERN ulint	srv_lock_table_size	= ULINT_MAX;
+
+UNIV_INTERN ulint	srv_n_file_io_threads	= ULINT_MAX;
+UNIV_INTERN ulint	srv_n_read_io_threads	= 1;
+UNIV_INTERN ulint	srv_n_write_io_threads	= 1;
+
+#ifdef UNIV_LOG_ARCHIVE
+UNIV_INTERN ibool		srv_log_archive_on	= FALSE;
+UNIV_INTERN ibool		srv_archive_recovery	= 0;
+UNIV_INTERN ib_uint64_t	srv_archive_recovery_limit_lsn;
+#endif /* UNIV_LOG_ARCHIVE */
+
+/* This parameter is used to throttle the number of insert buffers that are
+merged in a batch. By increasing this parameter on a faster disk you can
+possibly reduce the number of I/O operations performed to complete the
+merge operation. The value of this parameter is used as is by the
+background loop when the system is idle (low load), on a busy system
+the parameter is scaled down by a factor of 4, this is to avoid putting
+a heavier load on the I/O sub system. */
+
+UNIV_INTERN ulong	srv_insert_buffer_batch_size = 20;
+
+UNIV_INTERN char*	srv_file_flush_method_str = NULL;
+UNIV_INTERN ulint	srv_unix_file_flush_method = SRV_UNIX_FSYNC;
+UNIV_INTERN ulint	srv_win_file_flush_method = SRV_WIN_IO_UNBUFFERED;
+
+UNIV_INTERN ulint	srv_max_n_open_files	  = 300;
+
+/* The InnoDB main thread tries to keep the ratio of modified pages
+in the buffer pool to all database pages in the buffer pool smaller than
+the following number. But it is not guaranteed that the value stays below
+that during a time of heavy update/insert activity. */
+
+UNIV_INTERN ulong	srv_max_buf_pool_modified_pct	= 90;
+
+/* variable counts amount of data read in total (in bytes) */
+UNIV_INTERN ulint srv_data_read = 0;
+
+/* here we count the amount of data written in total (in bytes) */
+UNIV_INTERN ulint srv_data_written = 0;
+
+/* the number of the log write requests done */
+UNIV_INTERN ulint srv_log_write_requests = 0;
+
+/* the number of physical writes to the log performed */
+UNIV_INTERN ulint srv_log_writes = 0;
+
+/* amount of data written to the log files in bytes */
+UNIV_INTERN ulint srv_os_log_written = 0;
+
+/* amount of writes being done to the log files */
+UNIV_INTERN ulint srv_os_log_pending_writes = 0;
+
+/* we increase this counter, when there we don't have enough space in the
+log buffer and have to flush it */
+UNIV_INTERN ulint srv_log_waits = 0;
+
+/* this variable counts the amount of times, when the doublewrite buffer
+was flushed */
+UNIV_INTERN ulint srv_dblwr_writes = 0;
+
+/* here we store the number of pages that have been flushed to the
+doublewrite buffer */
+UNIV_INTERN ulint srv_dblwr_pages_written = 0;
+
+/* in this variable we store the number of write requests issued */
+UNIV_INTERN ulint srv_buf_pool_write_requests = 0;
+
+/* here we store the number of times when we had to wait for a free page
+in the buffer pool. It happens when the buffer pool is full and we need
+to make a flush, in order to be able to read or create a page. */
+UNIV_INTERN ulint srv_buf_pool_wait_free = 0;
+
+/* variable to count the number of pages that were written from buffer
+pool to the disk */
+UNIV_INTERN ulint srv_buf_pool_flushed = 0;
+
+/* variable to count the number of buffer pool reads that led to the
+reading of a disk page */
+UNIV_INTERN ulint srv_buf_pool_reads = 0;
+
+/* variable to count the number of sequential read-aheads */
+UNIV_INTERN ulint srv_read_ahead_seq = 0;
+
+/* variable to count the number of random read-aheads */
+UNIV_INTERN ulint srv_read_ahead_rnd = 0;
+
+/* structure to pass status variables to MySQL */
+UNIV_INTERN export_struc export_vars;
+
+/* If the following is != 0 we do not allow inserts etc. This protects
+the user from forgetting the innodb_force_recovery keyword to my.cnf */
+
+UNIV_INTERN ulint	srv_force_recovery	= 0;
+/*-----------------------*/
+/* We are prepared for a situation that we have this many threads waiting for
+a semaphore inside InnoDB. innobase_start_or_create_for_mysql() sets the
+value. */
+
+UNIV_INTERN ulint	srv_max_n_threads	= 0;
+
+/* The following controls how many threads we let inside InnoDB concurrently:
+threads waiting for locks are not counted into the number because otherwise
+we could get a deadlock. MySQL creates a thread for each user session, and
+semaphore contention and convoy problems can occur withput this restriction.
+Value 10 should be good if there are less than 4 processors + 4 disks in the
+computer. Bigger computers need bigger values. Value 0 will disable the
+concurrency check. */
+
+UNIV_INTERN ulong	srv_thread_concurrency	= 0;
+UNIV_INTERN ulong	srv_commit_concurrency	= 0;
+
+/* this mutex protects srv_conc data structures */
+UNIV_INTERN os_fast_mutex_t	srv_conc_mutex;
+/* number of transactions that have declared_to_be_inside_innodb set.
+It used to be a non-error for this value to drop below zero temporarily.
+This is no longer true. We'll, however, keep the lint datatype to add
+assertions to catch any corner cases that we may have missed. */
+UNIV_INTERN lint	srv_conc_n_threads	= 0;
+/* number of OS threads waiting in the FIFO for a permission to enter
+InnoDB */
+UNIV_INTERN ulint	srv_conc_n_waiting_threads = 0;
+
+typedef struct srv_conc_slot_struct	srv_conc_slot_t;
+struct srv_conc_slot_struct{
+	os_event_t			event;		/* event to wait */
+	ibool				reserved;	/* TRUE if slot
+							reserved */
+	ibool				wait_ended;	/* TRUE when another
+							thread has already set
+							the event and the
+							thread in this slot is
+							free to proceed; but
+							reserved may still be
+							TRUE at that point */
+	UT_LIST_NODE_T(srv_conc_slot_t)	srv_conc_queue;	/* queue node */
+};
+
+/* queue of threads waiting to get in */
+UNIV_INTERN UT_LIST_BASE_NODE_T(srv_conc_slot_t)	srv_conc_queue;
+/* array of wait slots */
+UNIV_INTERN srv_conc_slot_t* srv_conc_slots;
+
+/* Number of times a thread is allowed to enter InnoDB within the same
+SQL query after it has once got the ticket at srv_conc_enter_innodb */
+#define SRV_FREE_TICKETS_TO_ENTER srv_n_free_tickets_to_enter
+#define SRV_THREAD_SLEEP_DELAY srv_thread_sleep_delay
+/*-----------------------*/
+/* If the following is set to 1 then we do not run purge and insert buffer
+merge to completion before shutdown. If it is set to 2, do not even flush the
+buffer pool to data files at the shutdown: we effectively 'crash'
+InnoDB (but lose no committed transactions). */
+UNIV_INTERN ulint	srv_fast_shutdown	= 0;
+
+/* Generate a innodb_status.<pid> file */
+UNIV_INTERN ibool	srv_innodb_status	= FALSE;
+
+/* When estimating number of different key values in an index, sample
+this many index pages */
+UNIV_INTERN unsigned long long	srv_stats_sample_pages = 8;
+
+UNIV_INTERN ibool	srv_use_doublewrite_buf	= TRUE;
+UNIV_INTERN ibool	srv_use_checksums = TRUE;
+
+UNIV_INTERN ibool	srv_set_thread_priorities = TRUE;
+UNIV_INTERN int	srv_query_thread_priority = 0;
+
+UNIV_INTERN ulong	srv_replication_delay		= 0;
+
+UNIV_INTERN ulint	srv_io_capacity = 100;
+
+/* Returns the number of IO operations that is X percent of the capacity.
+PCT_IO(5) -> returns the number of IO operations that is 5% of the max
+where max is srv_io_capacity. */
+#define PCT_IO(pct) ((ulint) (srv_io_capacity * ((double) pct / 100.0)))
+
+UNIV_INTERN long long	srv_ibuf_max_size = 0;
+UNIV_INTERN ulint	srv_ibuf_active_contract = 0; /* 0:disable 1:enable */
+UNIV_INTERN ulint	srv_ibuf_accel_rate = 100;
+#define PCT_IBUF_IO(pct) ((ulint) (srv_io_capacity * srv_ibuf_accel_rate * ((double) pct / 10000.0)))
+
+UNIV_INTERN ulint	srv_flush_neighbor_pages = 1; /* 0:disable 1:enable */
+
+UNIV_INTERN ulint	srv_enable_unsafe_group_commit = 0; /* 0:disable 1:enable */
+UNIV_INTERN ulint	srv_read_ahead = 3; /* 1: random  2: linear  3: Both */
+UNIV_INTERN ulint	srv_adaptive_checkpoint = 0; /* 0:disable 1:enable */
+
+UNIV_INTERN ulint	srv_extra_rsegments = 0; /* extra rseg for users */
+/*-------------------------------------------*/
+UNIV_INTERN ulong	srv_n_spin_wait_rounds	= 20;
+UNIV_INTERN ulong	srv_n_free_tickets_to_enter = 500;
+UNIV_INTERN ulong	srv_thread_sleep_delay = 10000;
+UNIV_INTERN ulint	srv_spin_wait_delay	= 5;
+UNIV_INTERN ibool	srv_priority_boost	= TRUE;
+
+#ifdef UNIV_DEBUG
+UNIV_INTERN ibool	srv_print_thread_releases	= FALSE;
+UNIV_INTERN ibool	srv_print_lock_waits		= FALSE;
+UNIV_INTERN ibool	srv_print_buf_io		= FALSE;
+UNIV_INTERN ibool	srv_print_log_io		= FALSE;
+UNIV_INTERN ibool	srv_print_latch_waits		= FALSE;
+#endif /* UNIV_DEBUG */
+
+UNIV_INTERN ulint		srv_n_rows_inserted		= 0;
+UNIV_INTERN ulint		srv_n_rows_updated		= 0;
+UNIV_INTERN ulint		srv_n_rows_deleted		= 0;
+UNIV_INTERN ulint		srv_n_rows_read			= 0;
+#ifndef UNIV_HOTBACKUP
+static ulint	srv_n_rows_inserted_old		= 0;
+static ulint	srv_n_rows_updated_old		= 0;
+static ulint	srv_n_rows_deleted_old		= 0;
+static ulint	srv_n_rows_read_old		= 0;
+#endif /* !UNIV_HOTBACKUP */
+
+UNIV_INTERN ulint		srv_n_lock_wait_count		= 0;
+UNIV_INTERN ulint		srv_n_lock_wait_current_count	= 0;
+UNIV_INTERN ib_int64_t	srv_n_lock_wait_time		= 0;
+UNIV_INTERN ulint		srv_n_lock_max_wait_time	= 0;
+
+
+/*
+  Set the following to 0 if you want InnoDB to write messages on
+  stderr on startup/shutdown
+*/
+UNIV_INTERN ibool	srv_print_verbose_log		= TRUE;
+UNIV_INTERN ibool	srv_print_innodb_monitor	= FALSE;
+UNIV_INTERN ibool	srv_print_innodb_lock_monitor	= FALSE;
+UNIV_INTERN ibool	srv_print_innodb_tablespace_monitor = FALSE;
+UNIV_INTERN ibool	srv_print_innodb_table_monitor = FALSE;
+
+/* Array of English strings describing the current state of an
+i/o handler thread */
+
+UNIV_INTERN const char* srv_io_thread_op_info[SRV_MAX_N_IO_THREADS];
+UNIV_INTERN const char* srv_io_thread_function[SRV_MAX_N_IO_THREADS];
+
+UNIV_INTERN time_t	srv_last_monitor_time;
+
+UNIV_INTERN mutex_t	srv_innodb_monitor_mutex;
+
+/* Mutex for locking srv_monitor_file */
+UNIV_INTERN mutex_t	srv_monitor_file_mutex;
+/* Temporary file for innodb monitor output */
+UNIV_INTERN FILE*	srv_monitor_file;
+/* Mutex for locking srv_dict_tmpfile.
+This mutex has a very high rank; threads reserving it should not
+be holding any InnoDB latches. */
+UNIV_INTERN mutex_t	srv_dict_tmpfile_mutex;
+/* Temporary file for output from the data dictionary */
+UNIV_INTERN FILE*	srv_dict_tmpfile;
+/* Mutex for locking srv_misc_tmpfile.
+This mutex has a very low rank; threads reserving it should not
+acquire any further latches or sleep before releasing this one. */
+UNIV_INTERN mutex_t	srv_misc_tmpfile_mutex;
+/* Temporary file for miscellanous diagnostic output */
+UNIV_INTERN FILE*	srv_misc_tmpfile;
+
+UNIV_INTERN ulint	srv_main_thread_process_no	= 0;
+UNIV_INTERN ulint	srv_main_thread_id		= 0;
+
+/*
+	IMPLEMENTATION OF THE SERVER MAIN PROGRAM
+	=========================================
+
+There is the following analogue between this database
+server and an operating system kernel:
+
+DB concept			equivalent OS concept
+----------			---------------------
+transaction		--	process;
+
+query thread		--	thread;
+
+lock			--	semaphore;
+
+transaction set to
+the rollback state	--	kill signal delivered to a process;
+
+kernel			--	kernel;
+
+query thread execution:
+(a) without kernel mutex
+reserved		--	process executing in user mode;
+(b) with kernel mutex reserved
+			--	process executing in kernel mode;
+
+The server is controlled by a master thread which runs at
+a priority higher than normal, that is, higher than user threads.
+It sleeps most of the time, and wakes up, say, every 300 milliseconds,
+to check whether there is anything happening in the server which
+requires intervention of the master thread. Such situations may be,
+for example, when flushing of dirty blocks is needed in the buffer
+pool or old version of database rows have to be cleaned away.
+
+The threads which we call user threads serve the queries of
+the clients and input from the console of the server.
+They run at normal priority. The server may have several
+communications endpoints. A dedicated set of user threads waits
+at each of these endpoints ready to receive a client request.
+Each request is taken by a single user thread, which then starts
+processing and, when the result is ready, sends it to the client
+and returns to wait at the same endpoint the thread started from.
+
+So, we do not have dedicated communication threads listening at
+the endpoints and dealing the jobs to dedicated worker threads.
+Our architecture saves one thread swithch per request, compared
+to the solution with dedicated communication threads
+which amounts to 15 microseconds on 100 MHz Pentium
+running NT. If the client
+is communicating over a network, this saving is negligible, but
+if the client resides in the same machine, maybe in an SMP machine
+on a different processor from the server thread, the saving
+can be important as the threads can communicate over shared
+memory with an overhead of a few microseconds.
+
+We may later implement a dedicated communication thread solution
+for those endpoints which communicate over a network.
+
+Our solution with user threads has two problems: for each endpoint
+there has to be a number of listening threads. If there are many
+communication endpoints, it may be difficult to set the right number
+of concurrent threads in the system, as many of the threads
+may always be waiting at less busy endpoints. Another problem
+is queuing of the messages, as the server internally does not
+offer any queue for jobs.
+
+Another group of user threads is intended for splitting the
+queries and processing them in parallel. Let us call these
+parallel communication threads. These threads are waiting for
+parallelized tasks, suspended on event semaphores.
+
+A single user thread waits for input from the console,
+like a command to shut the database.
+
+Utility threads are a different group of threads which takes
+care of the buffer pool flushing and other, mainly background
+operations, in the server.
+Some of these utility threads always run at a lower than normal
+priority, so that they are always in background. Some of them
+may dynamically boost their priority by the pri_adjust function,
+even to higher than normal priority, if their task becomes urgent.
+The running of utilities is controlled by high- and low-water marks
+of urgency. The urgency may be measured by the number of dirty blocks
+in the buffer pool, in the case of the flush thread, for example.
+When the high-water mark is exceeded, an utility starts running, until
+the urgency drops under the low-water mark. Then the utility thread
+suspend itself to wait for an event. The master thread is
+responsible of signaling this event when the utility thread is
+again needed.
+
+For each individual type of utility, some threads always remain
+at lower than normal priority. This is because pri_adjust is implemented
+so that the threads at normal or higher priority control their
+share of running time by calling sleep. Thus, if the load of the
+system sudenly drops, these threads cannot necessarily utilize
+the system fully. The background priority threads make up for this,
+starting to run when the load drops.
+
+When there is no activity in the system, also the master thread
+suspends itself to wait for an event making
+the server totally silent. The responsibility to signal this
+event is on the user thread which again receives a message
+from a client.
+
+There is still one complication in our server design. If a
+background utility thread obtains a resource (e.g., mutex) needed by a user
+thread, and there is also some other user activity in the system,
+the user thread may have to wait indefinitely long for the
+resource, as the OS does not schedule a background thread if
+there is some other runnable user thread. This problem is called
+priority inversion in real-time programming.
+
+One solution to the priority inversion problem would be to
+keep record of which thread owns which resource and
+in the above case boost the priority of the background thread
+so that it will be scheduled and it can release the resource.
+This solution is called priority inheritance in real-time programming.
+A drawback of this solution is that the overhead of acquiring a mutex
+increases slightly, maybe 0.2 microseconds on a 100 MHz Pentium, because
+the thread has to call os_thread_get_curr_id.
+This may be compared to 0.5 microsecond overhead for a mutex lock-unlock
+pair. Note that the thread
+cannot store the information in the resource, say mutex, itself,
+because competing threads could wipe out the information if it is
+stored before acquiring the mutex, and if it stored afterwards,
+the information is outdated for the time of one machine instruction,
+at least. (To be precise, the information could be stored to
+lock_word in mutex if the machine supports atomic swap.)
+
+The above solution with priority inheritance may become actual in the
+future, but at the moment we plan to implement a more coarse solution,
+which could be called a global priority inheritance. If a thread
+has to wait for a long time, say 300 milliseconds, for a resource,
+we just guess that it may be waiting for a resource owned by a background
+thread, and boost the the priority of all runnable background threads
+to the normal level. The background threads then themselves adjust
+their fixed priority back to background after releasing all resources
+they had (or, at some fixed points in their program code).
+
+What is the performance of the global priority inheritance solution?
+We may weigh the length of the wait time 300 milliseconds, during
+which the system processes some other thread
+to the cost of boosting the priority of each runnable background
+thread, rescheduling it, and lowering the priority again.
+On 100 MHz Pentium + NT this overhead may be of the order 100
+microseconds per thread. So, if the number of runnable background
+threads is not very big, say < 100, the cost is tolerable.
+Utility threads probably will access resources used by
+user threads not very often, so collisions of user threads
+to preempted utility threads should not happen very often.
+
+The thread table contains
+information of the current status of each thread existing in the system,
+and also the event semaphores used in suspending the master thread
+and utility and parallel communication threads when they have nothing to do.
+The thread table can be seen as an analogue to the process table
+in a traditional Unix implementation.
+
+The thread table is also used in the global priority inheritance
+scheme. This brings in one additional complication: threads accessing
+the thread table must have at least normal fixed priority,
+because the priority inheritance solution does not work if a background
+thread is preempted while possessing the mutex protecting the thread table.
+So, if a thread accesses the thread table, its priority has to be
+boosted at least to normal. This priority requirement can be seen similar to
+the privileged mode used when processing the kernel calls in traditional
+Unix.*/
+
+/* Thread slot in the thread table */
+struct srv_slot_struct{
+	os_thread_id_t	id;		/* thread id */
+	os_thread_t	handle;		/* thread handle */
+	unsigned	type:3;		/* thread type: user, utility etc. */
+	unsigned	in_use:1;	/* TRUE if this slot is in use */
+	unsigned	suspended:1;	/* TRUE if the thread is waiting
+					for the event of this slot */
+	ib_time_t	suspend_time;	/* time when the thread was
+					suspended */
+	os_event_t	event;		/* event used in suspending the
+					thread when it has nothing to do */
+	que_thr_t*	thr;		/* suspended query thread (only
+					used for MySQL threads) */
+};
+
+/* Table for MySQL threads where they will be suspended to wait for locks */
+UNIV_INTERN srv_slot_t*	srv_mysql_table = NULL;
+
+UNIV_INTERN os_event_t	srv_lock_timeout_thread_event;
+
+UNIV_INTERN srv_sys_t*	srv_sys	= NULL;
+
+/* padding to prevent other memory update hotspots from residing on
+the same memory cache line */
+UNIV_INTERN byte	srv_pad1[64];
+/* mutex protecting the server, trx structs, query threads, and lock table */
+UNIV_INTERN mutex_t*	kernel_mutex_temp;
+/* padding to prevent other memory update hotspots from residing on
+the same memory cache line */
+UNIV_INTERN byte	srv_pad2[64];
+
+#if 0
+/* The following three values measure the urgency of the jobs of
+buffer, version, and insert threads. They may vary from 0 - 1000.
+The server mutex protects all these variables. The low-water values
+tell that the server can acquiesce the utility when the value
+drops below this low-water mark. */
+
+static ulint	srv_meter[SRV_MASTER + 1];
+static ulint	srv_meter_low_water[SRV_MASTER + 1];
+static ulint	srv_meter_high_water[SRV_MASTER + 1];
+static ulint	srv_meter_high_water2[SRV_MASTER + 1];
+static ulint	srv_meter_foreground[SRV_MASTER + 1];
+#endif
+
+/* The following values give info about the activity going on in
+the database. They are protected by the server mutex. The arrays
+are indexed by the type of the thread. */
+
+UNIV_INTERN ulint	srv_n_threads_active[SRV_MASTER + 1];
+UNIV_INTERN ulint	srv_n_threads[SRV_MASTER + 1];
+
+/*************************************************************************
+Sets the info describing an i/o thread current state. */
+UNIV_INTERN
+void
+srv_set_io_thread_op_info(
+/*======================*/
+	ulint		i,	/* in: the 'segment' of the i/o thread */
+	const char*	str)	/* in: constant char string describing the
+				state */
+{
+	ut_a(i < SRV_MAX_N_IO_THREADS);
+
+	srv_io_thread_op_info[i] = str;
+}
+
+/*************************************************************************
+Accessor function to get pointer to n'th slot in the server thread
+table. */
+static
+srv_slot_t*
+srv_table_get_nth_slot(
+/*===================*/
+				/* out: pointer to the slot */
+	ulint	index)		/* in: index of the slot */
+{
+	ut_a(index < OS_THREAD_MAX_N);
+
+	return(srv_sys->threads + index);
+}
+
+#ifndef UNIV_HOTBACKUP
+/*************************************************************************
+Gets the number of threads in the system. */
+UNIV_INTERN
+ulint
+srv_get_n_threads(void)
+/*===================*/
+{
+	ulint	i;
+	ulint	n_threads	= 0;
+
+	mutex_enter(&kernel_mutex);
+
+	for (i = SRV_COM; i < SRV_MASTER + 1; i++) {
+
+		n_threads += srv_n_threads[i];
+	}
+
+	mutex_exit(&kernel_mutex);
+
+	return(n_threads);
+}
+
+/*************************************************************************
+Reserves a slot in the thread table for the current thread. Also creates the
+thread local storage struct for the current thread. NOTE! The server mutex
+has to be reserved by the caller! */
+static
+ulint
+srv_table_reserve_slot(
+/*===================*/
+					/* out: reserved slot index */
+	enum srv_thread_type	type)	/* in: type of the thread */
+{
+	srv_slot_t*	slot;
+	ulint		i;
+
+	ut_a(type > 0);
+	ut_a(type <= SRV_MASTER);
+
+	i = 0;
+	slot = srv_table_get_nth_slot(i);
+
+	while (slot->in_use) {
+		i++;
+		slot = srv_table_get_nth_slot(i);
+	}
+
+	ut_a(slot->in_use == FALSE);
+
+	slot->in_use = TRUE;
+	slot->suspended = FALSE;
+	slot->type = type;
+	slot->id = os_thread_get_curr_id();
+	slot->handle = os_thread_get_curr();
+
+	thr_local_create();
+
+	thr_local_set_slot_no(os_thread_get_curr_id(), i);
+
+	return(i);
+}
+
+/*************************************************************************
+Suspends the calling thread to wait for the event in its thread slot.
+NOTE! The server mutex has to be reserved by the caller! */
+static
+os_event_t
+srv_suspend_thread(void)
+/*====================*/
+			/* out: event for the calling thread to wait */
+{
+	srv_slot_t*		slot;
+	os_event_t		event;
+	ulint			slot_no;
+	enum srv_thread_type	type;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	slot_no = thr_local_get_slot_no(os_thread_get_curr_id());
+
+	if (srv_print_thread_releases) {
+		fprintf(stderr,
+			"Suspending thread %lu to slot %lu\n",
+			(ulong) os_thread_get_curr_id(), (ulong) slot_no);
+	}
+
+	slot = srv_table_get_nth_slot(slot_no);
+
+	type = slot->type;
+
+	ut_ad(type >= SRV_WORKER);
+	ut_ad(type <= SRV_MASTER);
+
+	event = slot->event;
+
+	slot->suspended = TRUE;
+
+	ut_ad(srv_n_threads_active[type] > 0);
+
+	srv_n_threads_active[type]--;
+
+	os_event_reset(event);
+
+	return(event);
+}
+#endif /* !UNIV_HOTBACKUP */
+
+/*************************************************************************
+Releases threads of the type given from suspension in the thread table.
+NOTE! The server mutex has to be reserved by the caller! */
+UNIV_INTERN
+ulint
+srv_release_threads(
+/*================*/
+					/* out: number of threads
+					released: this may be < n if
+					not enough threads were
+					suspended at the moment */
+	enum srv_thread_type	type,	/* in: thread type */
+	ulint			n)	/* in: number of threads to release */
+{
+	srv_slot_t*	slot;
+	ulint		i;
+	ulint		count	= 0;
+
+	ut_ad(type >= SRV_WORKER);
+	ut_ad(type <= SRV_MASTER);
+	ut_ad(n > 0);
+	ut_ad(mutex_own(&kernel_mutex));
+
+	for (i = 0; i < OS_THREAD_MAX_N; i++) {
+
+		slot = srv_table_get_nth_slot(i);
+
+		if (slot->in_use && slot->type == type && slot->suspended) {
+
+			slot->suspended = FALSE;
+
+			srv_n_threads_active[type]++;
+
+			os_event_set(slot->event);
+
+			if (srv_print_thread_releases) {
+				fprintf(stderr,
+					"Releasing thread %lu type %lu"
+					" from slot %lu\n",
+					(ulong) slot->id, (ulong) type,
+					(ulong) i);
+			}
+
+			count++;
+
+			if (count == n) {
+				break;
+			}
+		}
+	}
+
+	return(count);
+}
+
+/*************************************************************************
+Returns the calling thread type. */
+UNIV_INTERN
+enum srv_thread_type
+srv_get_thread_type(void)
+/*=====================*/
+			/* out: SRV_COM, ... */
+{
+	ulint			slot_no;
+	srv_slot_t*		slot;
+	enum srv_thread_type	type;
+
+	mutex_enter(&kernel_mutex);
+
+	slot_no = thr_local_get_slot_no(os_thread_get_curr_id());
+
+	slot = srv_table_get_nth_slot(slot_no);
+
+	type = slot->type;
+
+	ut_ad(type >= SRV_WORKER);
+	ut_ad(type <= SRV_MASTER);
+
+	mutex_exit(&kernel_mutex);
+
+	return(type);
+}
+
+/*************************************************************************
+Initializes the server. */
+UNIV_INTERN
+void
+srv_init(void)
+/*==========*/
+{
+	srv_conc_slot_t*	conc_slot;
+	srv_slot_t*		slot;
+	dict_table_t*		table;
+	ulint			i;
+
+	srv_sys = mem_alloc(sizeof(srv_sys_t));
+
+	kernel_mutex_temp = mem_alloc(sizeof(mutex_t));
+	mutex_create(&kernel_mutex, SYNC_KERNEL);
+
+	mutex_create(&srv_innodb_monitor_mutex, SYNC_NO_ORDER_CHECK);
+
+	srv_sys->threads = mem_alloc(OS_THREAD_MAX_N * sizeof(srv_slot_t));
+
+	for (i = 0; i < OS_THREAD_MAX_N; i++) {
+		slot = srv_table_get_nth_slot(i);
+		slot->in_use = FALSE;
+		slot->type=0;	/* Avoid purify errors */
+		slot->event = os_event_create(NULL);
+		ut_a(slot->event);
+	}
+
+	srv_mysql_table = mem_alloc(OS_THREAD_MAX_N * sizeof(srv_slot_t));
+
+	for (i = 0; i < OS_THREAD_MAX_N; i++) {
+		slot = srv_mysql_table + i;
+		slot->in_use = FALSE;
+		slot->type = 0;
+		slot->event = os_event_create(NULL);
+		ut_a(slot->event);
+	}
+
+	srv_lock_timeout_thread_event = os_event_create(NULL);
+
+	for (i = 0; i < SRV_MASTER + 1; i++) {
+		srv_n_threads_active[i] = 0;
+		srv_n_threads[i] = 0;
+#if 0
+		srv_meter[i] = 30;
+		srv_meter_low_water[i] = 50;
+		srv_meter_high_water[i] = 100;
+		srv_meter_high_water2[i] = 200;
+		srv_meter_foreground[i] = 250;
+#endif
+	}
+
+	UT_LIST_INIT(srv_sys->tasks);
+
+	/* create dummy table and index for old-style infimum and supremum */
+	table = dict_mem_table_create("SYS_DUMMY1",
+				      DICT_HDR_SPACE, 1, 0);
+	dict_mem_table_add_col(table, NULL, NULL, DATA_CHAR,
+			       DATA_ENGLISH | DATA_NOT_NULL, 8);
+
+	srv_sys->dummy_ind1 = dict_mem_index_create(
+		"SYS_DUMMY1", "SYS_DUMMY1", DICT_HDR_SPACE, 0, 1);
+	dict_index_add_col(srv_sys->dummy_ind1, table,
+			   dict_table_get_nth_col(table, 0), 0);
+	srv_sys->dummy_ind1->table = table;
+	/* create dummy table and index for new-style infimum and supremum */
+	table = dict_mem_table_create("SYS_DUMMY2",
+				      DICT_HDR_SPACE, 1, DICT_TF_COMPACT);
+	dict_mem_table_add_col(table, NULL, NULL, DATA_CHAR,
+			       DATA_ENGLISH | DATA_NOT_NULL, 8);
+	srv_sys->dummy_ind2 = dict_mem_index_create(
+		"SYS_DUMMY2", "SYS_DUMMY2", DICT_HDR_SPACE, 0, 1);
+	dict_index_add_col(srv_sys->dummy_ind2, table,
+			   dict_table_get_nth_col(table, 0), 0);
+	srv_sys->dummy_ind2->table = table;
+
+	/* avoid ut_ad(index->cached) in dict_index_get_n_unique_in_tree */
+	srv_sys->dummy_ind1->cached = srv_sys->dummy_ind2->cached = TRUE;
+
+	/* Init the server concurrency restriction data structures */
+
+	os_fast_mutex_init(&srv_conc_mutex);
+
+	UT_LIST_INIT(srv_conc_queue);
+
+	srv_conc_slots = mem_alloc(OS_THREAD_MAX_N * sizeof(srv_conc_slot_t));
+
+	for (i = 0; i < OS_THREAD_MAX_N; i++) {
+		conc_slot = srv_conc_slots + i;
+		conc_slot->reserved = FALSE;
+		conc_slot->event = os_event_create(NULL);
+		ut_a(conc_slot->event);
+	}
+
+	/* Initialize some INFORMATION SCHEMA internal structures */
+	trx_i_s_cache_init(trx_i_s_cache);
+}
+
+/*************************************************************************
+Frees the OS fast mutex created in srv_init(). */
+UNIV_INTERN
+void
+srv_free(void)
+/*==========*/
+{
+	os_fast_mutex_free(&srv_conc_mutex);
+}
+
+/*************************************************************************
+Initializes the synchronization primitives, memory system, and the thread
+local storage. */
+UNIV_INTERN
+void
+srv_general_init(void)
+/*==================*/
+{
+	ut_mem_init();
+	os_sync_init();
+	sync_init();
+	mem_init(srv_mem_pool_size);
+	thr_local_init();
+}
+
+/*======================= InnoDB Server FIFO queue =======================*/
+
+/* Maximum allowable purge history length.  <=0 means 'infinite'. */
+UNIV_INTERN ulong	srv_max_purge_lag		= 0;
+
+/*************************************************************************
+Puts an OS thread to wait if there are too many concurrent threads
+(>= srv_thread_concurrency) inside InnoDB. The threads wait in a FIFO queue. */
+UNIV_INTERN
+void
+srv_conc_enter_innodb(
+/*==================*/
+	trx_t*	trx)	/* in: transaction object associated with the
+			thread */
+{
+	ibool			has_slept = FALSE;
+	srv_conc_slot_t*	slot	  = NULL;
+	ulint			i;
+
+	if (trx->mysql_thd != NULL
+	    && thd_is_replication_slave_thread(trx->mysql_thd)) {
+
+		UT_WAIT_FOR(srv_conc_n_threads
+			    < (lint)srv_thread_concurrency,
+			    srv_replication_delay * 1000);
+
+		return;
+	}
+
+	/* If trx has 'free tickets' to enter the engine left, then use one
+	such ticket */
+
+	if (trx->n_tickets_to_enter_innodb > 0) {
+		trx->n_tickets_to_enter_innodb--;
+
+		return;
+	}
+
+	os_fast_mutex_lock(&srv_conc_mutex);
+retry:
+	if (trx->declared_to_be_inside_innodb) {
+		ut_print_timestamp(stderr);
+		fputs("  InnoDB: Error: trying to declare trx"
+		      " to enter InnoDB, but\n"
+		      "InnoDB: it already is declared.\n", stderr);
+		trx_print(stderr, trx, 0);
+		putc('\n', stderr);
+		os_fast_mutex_unlock(&srv_conc_mutex);
+
+		return;
+	}
+
+	ut_ad(srv_conc_n_threads >= 0);
+
+	if (srv_conc_n_threads < (lint)srv_thread_concurrency) {
+
+		srv_conc_n_threads++;
+		trx->declared_to_be_inside_innodb = TRUE;
+		trx->n_tickets_to_enter_innodb = SRV_FREE_TICKETS_TO_ENTER;
+
+		os_fast_mutex_unlock(&srv_conc_mutex);
+
+		return;
+	}
+
+	/* If the transaction is not holding resources, let it sleep
+	for SRV_THREAD_SLEEP_DELAY microseconds, and try again then */
+
+	if (!has_slept && !trx->has_search_latch
+	    && NULL == UT_LIST_GET_FIRST(trx->trx_locks)) {
+
+		has_slept = TRUE; /* We let it sleep only once to avoid
+				  starvation */
+
+		srv_conc_n_waiting_threads++;
+
+		os_fast_mutex_unlock(&srv_conc_mutex);
+
+		trx->op_info = "sleeping before joining InnoDB queue";
+
+		/* Peter Zaitsev suggested that we take the sleep away
+		altogether. But the sleep may be good in pathological
+		situations of lots of thread switches. Simply put some
+		threads aside for a while to reduce the number of thread
+		switches. */
+		if (SRV_THREAD_SLEEP_DELAY > 0) {
+			os_thread_sleep(SRV_THREAD_SLEEP_DELAY);
+		}
+
+		trx->op_info = "";
+
+		os_fast_mutex_lock(&srv_conc_mutex);
+
+		srv_conc_n_waiting_threads--;
+
+		goto retry;
+	}
+
+	/* Too many threads inside: put the current thread to a queue */
+
+	for (i = 0; i < OS_THREAD_MAX_N; i++) {
+		slot = srv_conc_slots + i;
+
+		if (!slot->reserved) {
+
+			break;
+		}
+	}
+
+	if (i == OS_THREAD_MAX_N) {
+		/* Could not find a free wait slot, we must let the
+		thread enter */
+
+		srv_conc_n_threads++;
+		trx->declared_to_be_inside_innodb = TRUE;
+		trx->n_tickets_to_enter_innodb = 0;
+
+		os_fast_mutex_unlock(&srv_conc_mutex);
+
+		return;
+	}
+
+	/* Release possible search system latch this thread has */
+	if (trx->has_search_latch) {
+		trx_search_latch_release_if_reserved(trx);
+	}
+
+	/* Add to the queue */
+	slot->reserved = TRUE;
+	slot->wait_ended = FALSE;
+
+	UT_LIST_ADD_LAST(srv_conc_queue, srv_conc_queue, slot);
+
+	os_event_reset(slot->event);
+
+	srv_conc_n_waiting_threads++;
+
+	os_fast_mutex_unlock(&srv_conc_mutex);
+
+	/* Go to wait for the event; when a thread leaves InnoDB it will
+	release this thread */
+
+	trx->op_info = "waiting in InnoDB queue";
+
+	os_event_wait(slot->event);
+
+	trx->op_info = "";
+
+	os_fast_mutex_lock(&srv_conc_mutex);
+
+	srv_conc_n_waiting_threads--;
+
+	/* NOTE that the thread which released this thread already
+	incremented the thread counter on behalf of this thread */
+
+	slot->reserved = FALSE;
+
+	UT_LIST_REMOVE(srv_conc_queue, srv_conc_queue, slot);
+
+	trx->declared_to_be_inside_innodb = TRUE;
+	trx->n_tickets_to_enter_innodb = SRV_FREE_TICKETS_TO_ENTER;
+
+	os_fast_mutex_unlock(&srv_conc_mutex);
+}
+
+/*************************************************************************
+This lets a thread enter InnoDB regardless of the number of threads inside
+InnoDB. This must be called when a thread ends a lock wait. */
+UNIV_INTERN
+void
+srv_conc_force_enter_innodb(
+/*========================*/
+	trx_t*	trx)	/* in: transaction object associated with the
+			thread */
+{
+	if (UNIV_LIKELY(!srv_thread_concurrency)) {
+
+		return;
+	}
+
+	ut_ad(srv_conc_n_threads >= 0);
+
+	os_fast_mutex_lock(&srv_conc_mutex);
+
+	srv_conc_n_threads++;
+	trx->declared_to_be_inside_innodb = TRUE;
+	trx->n_tickets_to_enter_innodb = 1;
+
+	os_fast_mutex_unlock(&srv_conc_mutex);
+}
+
+/*************************************************************************
+This must be called when a thread exits InnoDB in a lock wait or at the
+end of an SQL statement. */
+UNIV_INTERN
+void
+srv_conc_force_exit_innodb(
+/*=======================*/
+	trx_t*	trx)	/* in: transaction object associated with the
+			thread */
+{
+	srv_conc_slot_t*	slot	= NULL;
+
+	if (trx->mysql_thd != NULL
+	    && thd_is_replication_slave_thread(trx->mysql_thd)) {
+
+		return;
+	}
+
+	if (trx->declared_to_be_inside_innodb == FALSE) {
+
+		return;
+	}
+
+	os_fast_mutex_lock(&srv_conc_mutex);
+
+	ut_ad(srv_conc_n_threads > 0);
+	srv_conc_n_threads--;
+	trx->declared_to_be_inside_innodb = FALSE;
+	trx->n_tickets_to_enter_innodb = 0;
+
+	if (srv_conc_n_threads < (lint)srv_thread_concurrency) {
+		/* Look for a slot where a thread is waiting and no other
+		thread has yet released the thread */
+
+		slot = UT_LIST_GET_FIRST(srv_conc_queue);
+
+		while (slot && slot->wait_ended == TRUE) {
+			slot = UT_LIST_GET_NEXT(srv_conc_queue, slot);
+		}
+
+		if (slot != NULL) {
+			slot->wait_ended = TRUE;
+
+			/* We increment the count on behalf of the released
+			thread */
+
+			srv_conc_n_threads++;
+		}
+	}
+
+	os_fast_mutex_unlock(&srv_conc_mutex);
+
+	if (slot != NULL) {
+		os_event_set(slot->event);
+	}
+}
+
+/*************************************************************************
+This must be called when a thread exits InnoDB. */
+UNIV_INTERN
+void
+srv_conc_exit_innodb(
+/*=================*/
+	trx_t*	trx)	/* in: transaction object associated with the
+			thread */
+{
+	if (trx->n_tickets_to_enter_innodb > 0) {
+		/* We will pretend the thread is still inside InnoDB though it
+		now leaves the InnoDB engine. In this way we save
+		a lot of semaphore operations. srv_conc_force_exit_innodb is
+		used to declare the thread definitely outside InnoDB. It
+		should be called when there is a lock wait or an SQL statement
+		ends. */
+
+		return;
+	}
+
+	srv_conc_force_exit_innodb(trx);
+}
+
+/*========================================================================*/
+
+/*************************************************************************
+Normalizes init parameter values to use units we use inside InnoDB. */
+static
+ulint
+srv_normalize_init_values(void)
+/*===========================*/
+				/* out: DB_SUCCESS or error code */
+{
+	ulint	n;
+	ulint	i;
+
+	n = srv_n_data_files;
+
+	for (i = 0; i < n; i++) {
+		srv_data_file_sizes[i] = srv_data_file_sizes[i]
+			* ((1024 * 1024) / UNIV_PAGE_SIZE);
+	}
+
+	srv_last_file_size_max = srv_last_file_size_max
+		* ((1024 * 1024) / UNIV_PAGE_SIZE);
+
+	srv_log_file_size = srv_log_file_size / UNIV_PAGE_SIZE;
+
+	srv_log_buffer_size = srv_log_buffer_size / UNIV_PAGE_SIZE;
+
+	srv_lock_table_size = 5 * (srv_buf_pool_size / UNIV_PAGE_SIZE);
+
+	return(DB_SUCCESS);
+}
+
+/*************************************************************************
+Boots the InnoDB server. */
+UNIV_INTERN
+ulint
+srv_boot(void)
+/*==========*/
+			/* out: DB_SUCCESS or error code */
+{
+	ulint	err;
+
+	/* Transform the init parameter values given by MySQL to
+	use units we use inside InnoDB: */
+
+	err = srv_normalize_init_values();
+
+	if (err != DB_SUCCESS) {
+		return(err);
+	}
+
+	/* Initialize synchronization primitives, memory management, and thread
+	local storage */
+
+	srv_general_init();
+
+	/* Initialize this module */
+
+	srv_init();
+
+	return(DB_SUCCESS);
+}
+
+#ifndef UNIV_HOTBACKUP
+/*************************************************************************
+Reserves a slot in the thread table for the current MySQL OS thread.
+NOTE! The kernel mutex has to be reserved by the caller! */
+static
+srv_slot_t*
+srv_table_reserve_slot_for_mysql(void)
+/*==================================*/
+			/* out: reserved slot */
+{
+	srv_slot_t*	slot;
+	ulint		i;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	i = 0;
+	slot = srv_mysql_table + i;
+
+	while (slot->in_use) {
+		i++;
+
+		if (i >= OS_THREAD_MAX_N) {
+
+			ut_print_timestamp(stderr);
+
+			fprintf(stderr,
+				"  InnoDB: There appear to be %lu MySQL"
+				" threads currently waiting\n"
+				"InnoDB: inside InnoDB, which is the"
+				" upper limit. Cannot continue operation.\n"
+				"InnoDB: We intentionally generate"
+				" a seg fault to print a stack trace\n"
+				"InnoDB: on Linux. But first we print"
+				" a list of waiting threads.\n", (ulong) i);
+
+			for (i = 0; i < OS_THREAD_MAX_N; i++) {
+
+				slot = srv_mysql_table + i;
+
+				fprintf(stderr,
+					"Slot %lu: thread id %lu, type %lu,"
+					" in use %lu, susp %lu, time %lu\n",
+					(ulong) i,
+					(ulong) os_thread_pf(slot->id),
+					(ulong) slot->type,
+					(ulong) slot->in_use,
+					(ulong) slot->suspended,
+					(ulong) difftime(ut_time(),
+							 slot->suspend_time));
+			}
+
+			ut_error;
+		}
+
+		slot = srv_mysql_table + i;
+	}
+
+	ut_a(slot->in_use == FALSE);
+
+	slot->in_use = TRUE;
+	slot->id = os_thread_get_curr_id();
+	slot->handle = os_thread_get_curr();
+
+	return(slot);
+}
+#endif /* !UNIV_HOTBACKUP */
+
+/*******************************************************************
+Puts a MySQL OS thread to wait for a lock to be released. If an error
+occurs during the wait trx->error_state associated with thr is
+!= DB_SUCCESS when we return. DB_LOCK_WAIT_TIMEOUT and DB_DEADLOCK
+are possible errors. DB_DEADLOCK is returned if selective deadlock
+resolution chose this transaction as a victim. */
+UNIV_INTERN
+void
+srv_suspend_mysql_thread(
+/*=====================*/
+	que_thr_t*	thr)	/* in: query thread associated with the MySQL
+				OS thread */
+{
+#ifndef UNIV_HOTBACKUP
+	srv_slot_t*	slot;
+	os_event_t	event;
+	double		wait_time;
+	trx_t*		trx;
+	ulint		had_dict_lock;
+	ibool		was_declared_inside_innodb	= FALSE;
+	ib_int64_t	start_time			= 0;
+	ib_int64_t	finish_time;
+	ulint		diff_time;
+	ulint		sec;
+	ulint		ms;
+	ulong		lock_wait_timeout;
+
+	ut_ad(!mutex_own(&kernel_mutex));
+
+	trx = thr_get_trx(thr);
+
+	os_event_set(srv_lock_timeout_thread_event);
+
+	mutex_enter(&kernel_mutex);
+
+	trx->error_state = DB_SUCCESS;
+
+	if (thr->state == QUE_THR_RUNNING) {
+
+		ut_ad(thr->is_active == TRUE);
+
+		/* The lock has already been released or this transaction
+		was chosen as a deadlock victim: no need to suspend */
+
+		if (trx->was_chosen_as_deadlock_victim) {
+
+			trx->error_state = DB_DEADLOCK;
+			trx->was_chosen_as_deadlock_victim = FALSE;
+		}
+
+		mutex_exit(&kernel_mutex);
+
+		return;
+	}
+
+	ut_ad(thr->is_active == FALSE);
+
+	slot = srv_table_reserve_slot_for_mysql();
+
+	event = slot->event;
+
+	slot->thr = thr;
+
+	os_event_reset(event);
+
+	slot->suspend_time = ut_time();
+
+	if (thr->lock_state == QUE_THR_LOCK_ROW) {
+		srv_n_lock_wait_count++;
+		srv_n_lock_wait_current_count++;
+
+		if (ut_usectime(&sec, &ms) == -1) {
+			start_time = -1;
+		} else {
+			start_time = (ib_int64_t) sec * 1000000 + ms;
+		}
+	}
+	/* Wake the lock timeout monitor thread, if it is suspended */
+
+	os_event_set(srv_lock_timeout_thread_event);
+
+	mutex_exit(&kernel_mutex);
+
+	if (trx->declared_to_be_inside_innodb) {
+
+		was_declared_inside_innodb = TRUE;
+
+		/* We must declare this OS thread to exit InnoDB, since a
+		possible other thread holding a lock which this thread waits
+		for must be allowed to enter, sooner or later */
+
+		srv_conc_force_exit_innodb(trx);
+	}
+
+	had_dict_lock = trx->dict_operation_lock_mode;
+
+	switch (had_dict_lock) {
+	case RW_S_LATCH:
+		/* Release foreign key check latch */
+		row_mysql_unfreeze_data_dictionary(trx);
+		break;
+	case RW_X_LATCH:
+		/* Release fast index creation latch */
+		row_mysql_unlock_data_dictionary(trx);
+		break;
+	}
+
+	ut_a(trx->dict_operation_lock_mode == 0);
+
+	/* Suspend this thread and wait for the event. */
+
+	os_event_wait(event);
+
+	/* After resuming, reacquire the data dictionary latch if
+	necessary. */
+
+	switch (had_dict_lock) {
+	case RW_S_LATCH:
+		row_mysql_freeze_data_dictionary(trx);
+		break;
+	case RW_X_LATCH:
+		row_mysql_lock_data_dictionary(trx);
+		break;
+	}
+
+	if (was_declared_inside_innodb) {
+
+		/* Return back inside InnoDB */
+
+		srv_conc_force_enter_innodb(trx);
+	}
+
+	mutex_enter(&kernel_mutex);
+
+	/* Release the slot for others to use */
+
+	slot->in_use = FALSE;
+
+	wait_time = ut_difftime(ut_time(), slot->suspend_time);
+
+	if (thr->lock_state == QUE_THR_LOCK_ROW) {
+		if (ut_usectime(&sec, &ms) == -1) {
+			finish_time = -1;
+		} else {
+			finish_time = (ib_int64_t) sec * 1000000 + ms;
+		}
+
+		diff_time = (ulint) (finish_time - start_time);
+
+		srv_n_lock_wait_current_count--;
+		srv_n_lock_wait_time = srv_n_lock_wait_time + diff_time;
+		if (diff_time > srv_n_lock_max_wait_time &&
+		    /* only update the variable if we successfully
+		    retrieved the start and finish times. See Bug#36819. */
+		    start_time != -1 && finish_time != -1) {
+			srv_n_lock_max_wait_time = diff_time;
+		}
+	}
+
+	if (trx->was_chosen_as_deadlock_victim) {
+
+		trx->error_state = DB_DEADLOCK;
+		trx->was_chosen_as_deadlock_victim = FALSE;
+	}
+
+	mutex_exit(&kernel_mutex);
+
+	/* InnoDB system transactions (such as the purge, and
+	incomplete transactions that are being rolled back after crash
+	recovery) will use the global value of
+	innodb_lock_wait_timeout, because trx->mysql_thd == NULL. */
+	lock_wait_timeout = thd_lock_wait_timeout(trx->mysql_thd);
+
+	if (lock_wait_timeout < 100000000
+	    && wait_time > (double) lock_wait_timeout) {
+
+		trx->error_state = DB_LOCK_WAIT_TIMEOUT;
+	}
+#else /* UNIV_HOTBACKUP */
+	/* This function depends on MySQL code that is not included in
+	InnoDB Hot Backup builds.  Besides, this function should never
+	be called in InnoDB Hot Backup. */
+	ut_error;
+#endif /* UNIV_HOTBACKUP */
+}
+
+/************************************************************************
+Releases a MySQL OS thread waiting for a lock to be released, if the
+thread is already suspended. */
+UNIV_INTERN
+void
+srv_release_mysql_thread_if_suspended(
+/*==================================*/
+	que_thr_t*	thr)	/* in: query thread associated with the
+				MySQL OS thread	 */
+{
+#ifndef UNIV_HOTBACKUP
+	srv_slot_t*	slot;
+	ulint		i;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	for (i = 0; i < OS_THREAD_MAX_N; i++) {
+
+		slot = srv_mysql_table + i;
+
+		if (slot->in_use && slot->thr == thr) {
+			/* Found */
+
+			os_event_set(slot->event);
+
+			return;
+		}
+	}
+
+	/* not found */
+#else /* UNIV_HOTBACKUP */
+	/* This function depends on MySQL code that is not included in
+	InnoDB Hot Backup builds.  Besides, this function should never
+	be called in InnoDB Hot Backup. */
+	ut_error;
+#endif /* UNIV_HOTBACKUP */
+}
+
+#ifndef UNIV_HOTBACKUP
+/**********************************************************************
+Refreshes the values used to calculate per-second averages. */
+static
+void
+srv_refresh_innodb_monitor_stats(void)
+/*==================================*/
+{
+	mutex_enter(&srv_innodb_monitor_mutex);
+
+	srv_last_monitor_time = time(NULL);
+
+	os_aio_refresh_stats();
+
+	btr_cur_n_sea_old = btr_cur_n_sea;
+	btr_cur_n_non_sea_old = btr_cur_n_non_sea;
+
+	log_refresh_stats();
+
+	buf_refresh_io_stats();
+
+	srv_n_rows_inserted_old = srv_n_rows_inserted;
+	srv_n_rows_updated_old = srv_n_rows_updated;
+	srv_n_rows_deleted_old = srv_n_rows_deleted;
+	srv_n_rows_read_old = srv_n_rows_read;
+
+	mutex_exit(&srv_innodb_monitor_mutex);
+}
+
+/**********************************************************************
+Outputs to a file the output of the InnoDB Monitor. */
+UNIV_INTERN
+void
+srv_printf_innodb_monitor(
+/*======================*/
+	FILE*	file,		/* in: output stream */
+	ulint*	trx_start,	/* out: file position of the start of
+				the list of active transactions */
+	ulint*	trx_end)	/* out: file position of the end of
+				the list of active transactions */
+{
+	double	time_elapsed;
+	time_t	current_time;
+	ulint	n_reserved;
+
+	ulint	btr_search_sys_subtotal;
+	ulint	lock_sys_subtotal;
+	ulint	recv_sys_subtotal;
+	ulint	io_counter_subtotal;
+
+	ulint	i;
+	trx_t*	trx;
+
+	mutex_enter(&srv_innodb_monitor_mutex);
+
+	current_time = time(NULL);
+
+	/* We add 0.001 seconds to time_elapsed to prevent division
+	by zero if two users happen to call SHOW INNODB STATUS at the same
+	time */
+
+	time_elapsed = difftime(current_time, srv_last_monitor_time)
+		+ 0.001;
+
+	srv_last_monitor_time = time(NULL);
+
+	fputs("\n=====================================\n", file);
+
+	ut_print_timestamp(file);
+	fprintf(file,
+		" INNODB MONITOR OUTPUT\n"
+		"=====================================\n"
+		"Per second averages calculated from the last %lu seconds\n",
+		(ulong)time_elapsed);
+
+	fputs("----------\n"
+	      "SEMAPHORES\n"
+	      "----------\n", file);
+	sync_print(file);
+
+	/* Conceptually, srv_innodb_monitor_mutex has a very high latching
+	order level in sync0sync.h, while dict_foreign_err_mutex has a very
+	low level 135. Therefore we can reserve the latter mutex here without
+	a danger of a deadlock of threads. */
+
+	mutex_enter(&dict_foreign_err_mutex);
+
+	if (ftell(dict_foreign_err_file) != 0L) {
+		fputs("------------------------\n"
+		      "LATEST FOREIGN KEY ERROR\n"
+		      "------------------------\n", file);
+		ut_copy_file(file, dict_foreign_err_file);
+	}
+
+	mutex_exit(&dict_foreign_err_mutex);
+
+	fputs("--------\n"
+	      "FILE I/O\n"
+	      "--------\n", file);
+	os_aio_print(file);
+
+	fputs("-------------------------------------\n"
+	      "INSERT BUFFER AND ADAPTIVE HASH INDEX\n"
+	      "-------------------------------------\n", file);
+	ibuf_print(file);
+
+	ha_print_info(file, btr_search_sys->hash_index);
+
+	fprintf(file,
+		"%.2f hash searches/s, %.2f non-hash searches/s\n",
+		(btr_cur_n_sea - btr_cur_n_sea_old)
+		/ time_elapsed,
+		(btr_cur_n_non_sea - btr_cur_n_non_sea_old)
+		/ time_elapsed);
+	btr_cur_n_sea_old = btr_cur_n_sea;
+	btr_cur_n_non_sea_old = btr_cur_n_non_sea;
+
+	fputs("---\n"
+	      "LOG\n"
+	      "---\n", file);
+	log_print(file);
+
+	fputs("----------------------\n"
+	      "BUFFER POOL AND MEMORY\n"
+	      "----------------------\n", file);
+	fprintf(file,
+			"Total memory allocated " ULINTPF
+			"; in additional pool allocated " ULINTPF "\n",
+			ut_total_allocated_memory,
+			mem_pool_get_reserved(mem_comm_pool));
+	/* Calcurate reserved memories */
+	if (btr_search_sys && btr_search_sys->hash_index->heap) {
+		btr_search_sys_subtotal = mem_heap_get_size(btr_search_sys->hash_index->heap);
+	} else {
+		btr_search_sys_subtotal = 0;
+		for (i=0; i < btr_search_sys->hash_index->n_mutexes; i++) {
+			btr_search_sys_subtotal += mem_heap_get_size(btr_search_sys->hash_index->heaps[i]);
+		}
+	}
+
+	lock_sys_subtotal = 0;
+	if (trx_sys) {
+		mutex_enter(&kernel_mutex);
+		trx = UT_LIST_GET_FIRST(trx_sys->mysql_trx_list);
+		while (trx) {
+			lock_sys_subtotal += ((trx->lock_heap) ? mem_heap_get_size(trx->lock_heap) : 0);
+			trx = UT_LIST_GET_NEXT(mysql_trx_list, trx);
+		}
+		mutex_exit(&kernel_mutex);
+	}
+
+	recv_sys_subtotal = ((recv_sys && recv_sys->addr_hash)
+			? mem_heap_get_size(recv_sys->heap) : 0);
+
+	fprintf(file,
+			"Internal hash tables (constant factor + variable factor)\n"
+			"    Adaptive hash index %lu \t(%lu + %lu)\n"
+			"    Page hash           %lu\n"
+			"    Dictionary cache    %lu \t(%lu + %lu)\n"
+			"    File system         %lu \t(%lu + %lu)\n"
+			"    Lock system         %lu \t(%lu + %lu)\n"
+			"    Recovery system     %lu \t(%lu + %lu)\n"
+			"    Threads             %lu \t(%lu + %lu)\n",
+
+			(ulong) (btr_search_sys
+				? (btr_search_sys->hash_index->n_cells * sizeof(hash_cell_t)) : 0)
+			+ btr_search_sys_subtotal,
+			(ulong) (btr_search_sys
+				? (btr_search_sys->hash_index->n_cells * sizeof(hash_cell_t)) : 0),
+			(ulong) btr_search_sys_subtotal,
+
+			(ulong) (buf_pool->page_hash->n_cells * sizeof(hash_cell_t)),
+
+			(ulong) (dict_sys ? ((dict_sys->table_hash->n_cells
+						+ dict_sys->table_id_hash->n_cells
+						) * sizeof(hash_cell_t)
+					+ dict_sys->size) : 0),
+			(ulong) (dict_sys ? ((dict_sys->table_hash->n_cells
+							+ dict_sys->table_id_hash->n_cells
+							) * sizeof(hash_cell_t)) : 0),
+			(ulong) (dict_sys ? (dict_sys->size) : 0),
+
+			(ulong) (fil_system_hash_cells() * sizeof(hash_cell_t)
+					+ fil_system_hash_nodes()),
+			(ulong) (fil_system_hash_cells() * sizeof(hash_cell_t)),
+			(ulong) fil_system_hash_nodes(),
+
+			(ulong) ((lock_sys ? (lock_sys->rec_hash->n_cells * sizeof(hash_cell_t)) : 0)
+					+ lock_sys_subtotal),
+			(ulong) (lock_sys ? (lock_sys->rec_hash->n_cells * sizeof(hash_cell_t)) : 0),
+			(ulong) lock_sys_subtotal,
+
+			(ulong) (((recv_sys && recv_sys->addr_hash)
+						? (recv_sys->addr_hash->n_cells * sizeof(hash_cell_t)) : 0)
+					+ recv_sys_subtotal),
+			(ulong) ((recv_sys && recv_sys->addr_hash)
+					? (recv_sys->addr_hash->n_cells * sizeof(hash_cell_t)) : 0),
+			(ulong) recv_sys_subtotal,
+
+			(ulong) (thr_local_hash_cells() * sizeof(hash_cell_t)
+					+ thr_local_hash_nodes()),
+			(ulong) (thr_local_hash_cells() * sizeof(hash_cell_t)),
+			(ulong) thr_local_hash_nodes());
+
+	fprintf(file, "Dictionary memory allocated " ULINTPF "\n",
+		dict_sys->size);
+
+	buf_print_io(file);
+
+	fputs("--------------\n"
+	      "ROW OPERATIONS\n"
+	      "--------------\n", file);
+	fprintf(file, "%ld queries inside InnoDB, %lu queries in queue\n",
+		(long) srv_conc_n_threads,
+		(ulong) srv_conc_n_waiting_threads);
+
+	fprintf(file, "%lu read views open inside InnoDB\n",
+		UT_LIST_GET_LEN(trx_sys->view_list));
+
+	n_reserved = fil_space_get_n_reserved_extents(0);
+	if (n_reserved > 0) {
+		fprintf(file,
+			"%lu tablespace extents now reserved for"
+			" B-tree split operations\n",
+			(ulong) n_reserved);
+	}
+
+#ifdef UNIV_LINUX
+	fprintf(file, "Main thread process no. %lu, id %lu, state: %s\n",
+		(ulong) srv_main_thread_process_no,
+		(ulong) srv_main_thread_id,
+		srv_main_thread_op_info);
+#else
+	fprintf(file, "Main thread id %lu, state: %s\n",
+		(ulong) srv_main_thread_id,
+		srv_main_thread_op_info);
+#endif
+	fprintf(file,
+		"Number of rows inserted " ULINTPF
+		", updated " ULINTPF ", deleted " ULINTPF
+		", read " ULINTPF "\n",
+		srv_n_rows_inserted,
+		srv_n_rows_updated,
+		srv_n_rows_deleted,
+		srv_n_rows_read);
+	fprintf(file,
+		"%.2f inserts/s, %.2f updates/s,"
+		" %.2f deletes/s, %.2f reads/s\n",
+		(srv_n_rows_inserted - srv_n_rows_inserted_old)
+		/ time_elapsed,
+		(srv_n_rows_updated - srv_n_rows_updated_old)
+		/ time_elapsed,
+		(srv_n_rows_deleted - srv_n_rows_deleted_old)
+		/ time_elapsed,
+		(srv_n_rows_read - srv_n_rows_read_old)
+		/ time_elapsed);
+
+	srv_n_rows_inserted_old = srv_n_rows_inserted;
+	srv_n_rows_updated_old = srv_n_rows_updated;
+	srv_n_rows_deleted_old = srv_n_rows_deleted;
+	srv_n_rows_read_old = srv_n_rows_read;
+
+	lock_print_info_summary(file);
+	if (trx_start) {
+		long	t = ftell(file);
+		if (t < 0) {
+			*trx_start = ULINT_UNDEFINED;
+		} else {
+			*trx_start = (ulint) t;
+		}
+	}
+	lock_print_info_all_transactions(file);
+	if (trx_end) {
+		long	t = ftell(file);
+		if (t < 0) {
+			*trx_end = ULINT_UNDEFINED;
+		} else {
+			*trx_end = (ulint) t;
+		}
+	}
+
+	fputs("----------------------------\n"
+	      "END OF INNODB MONITOR OUTPUT\n"
+	      "============================\n", file);
+	mutex_exit(&srv_innodb_monitor_mutex);
+	fflush(file);
+}
+
+/**********************************************************************
+Function to pass InnoDB status variables to MySQL */
+UNIV_INTERN
+void
+srv_export_innodb_status(void)
+{
+	mutex_enter(&srv_innodb_monitor_mutex);
+
+	export_vars.innodb_data_pending_reads
+		= os_n_pending_reads;
+	export_vars.innodb_data_pending_writes
+		= os_n_pending_writes;
+	export_vars.innodb_data_pending_fsyncs
+		= fil_n_pending_log_flushes
+		+ fil_n_pending_tablespace_flushes;
+	export_vars.innodb_data_fsyncs = os_n_fsyncs;
+	export_vars.innodb_data_read = srv_data_read;
+	export_vars.innodb_data_reads = os_n_file_reads;
+	export_vars.innodb_data_writes = os_n_file_writes;
+	export_vars.innodb_data_written = srv_data_written;
+	export_vars.innodb_buffer_pool_read_requests = buf_pool->n_page_gets;
+	export_vars.innodb_buffer_pool_write_requests
+		= srv_buf_pool_write_requests;
+	export_vars.innodb_buffer_pool_wait_free = srv_buf_pool_wait_free;
+	export_vars.innodb_buffer_pool_pages_flushed = srv_buf_pool_flushed;
+	export_vars.innodb_buffer_pool_reads = srv_buf_pool_reads;
+	export_vars.innodb_buffer_pool_read_ahead_rnd = srv_read_ahead_rnd;
+	export_vars.innodb_buffer_pool_read_ahead_seq = srv_read_ahead_seq;
+	export_vars.innodb_buffer_pool_pages_data
+		= UT_LIST_GET_LEN(buf_pool->LRU);
+	export_vars.innodb_buffer_pool_pages_dirty
+		= UT_LIST_GET_LEN(buf_pool->flush_list);
+	export_vars.innodb_buffer_pool_pages_free
+		= UT_LIST_GET_LEN(buf_pool->free);
+#ifdef UNIV_DEBUG
+	export_vars.innodb_buffer_pool_pages_latched
+		= buf_get_latched_pages_number();
+#endif /* UNIV_DEBUG */
+	export_vars.innodb_buffer_pool_pages_total = buf_pool->curr_size;
+
+	export_vars.innodb_buffer_pool_pages_misc = buf_pool->curr_size
+		- UT_LIST_GET_LEN(buf_pool->LRU)
+		- UT_LIST_GET_LEN(buf_pool->free);
+#ifdef HAVE_GCC_ATOMIC_BUILTINS
+	export_vars.innodb_have_atomic_builtins = 1;
+#else
+	export_vars.innodb_have_atomic_builtins = 0;
+#endif
+	export_vars.innodb_page_size = UNIV_PAGE_SIZE;
+	export_vars.innodb_log_waits = srv_log_waits;
+	export_vars.innodb_os_log_written = srv_os_log_written;
+	export_vars.innodb_os_log_fsyncs = fil_n_log_flushes;
+	export_vars.innodb_os_log_pending_fsyncs = fil_n_pending_log_flushes;
+	export_vars.innodb_os_log_pending_writes = srv_os_log_pending_writes;
+	export_vars.innodb_log_write_requests = srv_log_write_requests;
+	export_vars.innodb_log_writes = srv_log_writes;
+	export_vars.innodb_dblwr_pages_written = srv_dblwr_pages_written;
+	export_vars.innodb_dblwr_writes = srv_dblwr_writes;
+	export_vars.innodb_pages_created = buf_pool->n_pages_created;
+	export_vars.innodb_pages_read = buf_pool->n_pages_read;
+	export_vars.innodb_pages_written = buf_pool->n_pages_written;
+	export_vars.innodb_row_lock_waits = srv_n_lock_wait_count;
+	export_vars.innodb_row_lock_current_waits
+		= srv_n_lock_wait_current_count;
+	export_vars.innodb_row_lock_time = srv_n_lock_wait_time / 1000;
+	if (srv_n_lock_wait_count > 0) {
+		export_vars.innodb_row_lock_time_avg = (ulint)
+			(srv_n_lock_wait_time / 1000 / srv_n_lock_wait_count);
+	} else {
+		export_vars.innodb_row_lock_time_avg = 0;
+	}
+	export_vars.innodb_row_lock_time_max
+		= srv_n_lock_max_wait_time / 1000;
+	export_vars.innodb_rows_read = srv_n_rows_read;
+	export_vars.innodb_rows_inserted = srv_n_rows_inserted;
+	export_vars.innodb_rows_updated = srv_n_rows_updated;
+	export_vars.innodb_rows_deleted = srv_n_rows_deleted;
+
+	mutex_exit(&srv_innodb_monitor_mutex);
+}
+
+/*************************************************************************
+A thread which wakes up threads whose lock wait may have lasted too long.
+This also prints the info output by various InnoDB monitors. */
+UNIV_INTERN
+os_thread_ret_t
+srv_lock_timeout_and_monitor_thread(
+/*================================*/
+			/* out: a dummy parameter */
+	void*	arg __attribute__((unused)))
+			/* in: a dummy parameter required by
+			os_thread_create */
+{
+	srv_slot_t*	slot;
+	double		time_elapsed;
+	time_t		current_time;
+	time_t		last_table_monitor_time;
+	time_t		last_tablespace_monitor_time;
+	time_t		last_monitor_time;
+	ibool		some_waits;
+	double		wait_time;
+	ulint		i;
+
+#ifdef UNIV_DEBUG_THREAD_CREATION
+	fprintf(stderr, "Lock timeout thread starts, id %lu\n",
+		os_thread_pf(os_thread_get_curr_id()));
+#endif
+	UT_NOT_USED(arg);
+	srv_last_monitor_time = time(NULL);
+	last_table_monitor_time = time(NULL);
+	last_tablespace_monitor_time = time(NULL);
+	last_monitor_time = time(NULL);
+loop:
+	srv_lock_timeout_and_monitor_active = TRUE;
+
+	/* When someone is waiting for a lock, we wake up every second
+	and check if a timeout has passed for a lock wait */
+
+	os_thread_sleep(1000000);
+
+	current_time = time(NULL);
+
+	time_elapsed = difftime(current_time, last_monitor_time);
+
+	if (time_elapsed > 15) {
+		last_monitor_time = time(NULL);
+
+		if (srv_print_innodb_monitor) {
+			srv_printf_innodb_monitor(stderr, NULL, NULL);
+		}
+
+		if (srv_innodb_status) {
+			mutex_enter(&srv_monitor_file_mutex);
+			rewind(srv_monitor_file);
+			srv_printf_innodb_monitor(srv_monitor_file, NULL,
+						  NULL);
+			os_file_set_eof(srv_monitor_file);
+			mutex_exit(&srv_monitor_file_mutex);
+		}
+
+		if (srv_print_innodb_tablespace_monitor
+		    && difftime(current_time,
+				last_tablespace_monitor_time) > 60) {
+			last_tablespace_monitor_time = time(NULL);
+
+			fputs("========================"
+			      "========================\n",
+			      stderr);
+
+			ut_print_timestamp(stderr);
+
+			fputs(" INNODB TABLESPACE MONITOR OUTPUT\n"
+			      "========================"
+			      "========================\n",
+			      stderr);
+
+			fsp_print(0);
+			fputs("Validating tablespace\n", stderr);
+			fsp_validate(0);
+			fputs("Validation ok\n"
+			      "---------------------------------------\n"
+			      "END OF INNODB TABLESPACE MONITOR OUTPUT\n"
+			      "=======================================\n",
+			      stderr);
+		}
+
+		if (srv_print_innodb_table_monitor
+		    && difftime(current_time, last_table_monitor_time) > 60) {
+
+			last_table_monitor_time = time(NULL);
+
+			fputs("===========================================\n",
+			      stderr);
+
+			ut_print_timestamp(stderr);
+
+			fputs(" INNODB TABLE MONITOR OUTPUT\n"
+			      "===========================================\n",
+			      stderr);
+			dict_print();
+
+			fputs("-----------------------------------\n"
+			      "END OF INNODB TABLE MONITOR OUTPUT\n"
+			      "==================================\n",
+			      stderr);
+		}
+	}
+
+	mutex_enter(&kernel_mutex);
+
+	some_waits = FALSE;
+
+	/* Check of all slots if a thread is waiting there, and if it
+	has exceeded the time limit */
+
+	for (i = 0; i < OS_THREAD_MAX_N; i++) {
+
+		slot = srv_mysql_table + i;
+
+		if (slot->in_use) {
+			trx_t*	trx;
+			ulong	lock_wait_timeout;
+
+			some_waits = TRUE;
+
+			wait_time = ut_difftime(ut_time(), slot->suspend_time);
+
+			trx = thr_get_trx(slot->thr);
+			lock_wait_timeout = thd_lock_wait_timeout(
+				trx->mysql_thd);
+
+			if (lock_wait_timeout < 100000000
+			    && (wait_time > (double) lock_wait_timeout
+				|| wait_time < 0)) {
+
+				/* Timeout exceeded or a wrap-around in system
+				time counter: cancel the lock request queued
+				by the transaction and release possible
+				other transactions waiting behind; it is
+				possible that the lock has already been
+				granted: in that case do nothing */
+
+				if (trx->wait_lock) {
+					lock_cancel_waiting_and_release(
+						trx->wait_lock);
+				}
+			}
+		}
+	}
+
+	os_event_reset(srv_lock_timeout_thread_event);
+
+	mutex_exit(&kernel_mutex);
+
+	if (srv_shutdown_state >= SRV_SHUTDOWN_CLEANUP) {
+		goto exit_func;
+	}
+
+	if (some_waits || srv_print_innodb_monitor
+	    || srv_print_innodb_lock_monitor
+	    || srv_print_innodb_tablespace_monitor
+	    || srv_print_innodb_table_monitor) {
+		goto loop;
+	}
+
+	/* No one was waiting for a lock and no monitor was active:
+	suspend this thread */
+
+	srv_lock_timeout_and_monitor_active = FALSE;
+
+#if 0
+	/* The following synchronisation is disabled, since
+	the InnoDB monitor output is to be updated every 15 seconds. */
+	os_event_wait(srv_lock_timeout_thread_event);
+#endif
+	goto loop;
+
+exit_func:
+	srv_lock_timeout_and_monitor_active = FALSE;
+
+	/* We count the number of threads in os_thread_exit(). A created
+	thread should always use that to exit and not use return() to exit. */
+
+	os_thread_exit(NULL);
+
+	OS_THREAD_DUMMY_RETURN;
+}
+
+/*************************************************************************
+A thread which prints warnings about semaphore waits which have lasted
+too long. These can be used to track bugs which cause hangs. */
+UNIV_INTERN
+os_thread_ret_t
+srv_error_monitor_thread(
+/*=====================*/
+			/* out: a dummy parameter */
+	void*	arg __attribute__((unused)))
+			/* in: a dummy parameter required by
+			os_thread_create */
+{
+	/* number of successive fatal timeouts observed */
+	ulint		fatal_cnt	= 0;
+	ib_uint64_t	old_lsn;
+	ib_uint64_t	new_lsn;
+
+	old_lsn = srv_start_lsn;
+
+#ifdef UNIV_DEBUG_THREAD_CREATION
+	fprintf(stderr, "Error monitor thread starts, id %lu\n",
+		os_thread_pf(os_thread_get_curr_id()));
+#endif
+loop:
+	srv_error_monitor_active = TRUE;
+
+	/* Try to track a strange bug reported by Harald Fuchs and others,
+	where the lsn seems to decrease at times */
+
+	new_lsn = log_get_lsn();
+
+	if (new_lsn < old_lsn) {
+		ut_print_timestamp(stderr);
+		fprintf(stderr,
+			"  InnoDB: Error: old log sequence number %llu"
+			" was greater\n"
+			"InnoDB: than the new log sequence number %llu!\n"
+			"InnoDB: Please submit a bug report"
+			" to http://bugs.mysql.com\n",
+			old_lsn, new_lsn);
+	}
+
+	old_lsn = new_lsn;
+
+	if (difftime(time(NULL), srv_last_monitor_time) > 60) {
+		/* We referesh InnoDB Monitor values so that averages are
+		printed from at most 60 last seconds */
+
+		srv_refresh_innodb_monitor_stats();
+	}
+
+	/* Update the statistics collected for deciding LRU
+ 	eviction policy. */
+	buf_LRU_stat_update();
+
+	/* In case mutex_exit is not a memory barrier, it is
+	theoretically possible some threads are left waiting though
+	the semaphore is already released. Wake up those threads: */
+	
+	sync_arr_wake_threads_if_sema_free();
+
+	if (sync_array_print_long_waits()) {
+		fatal_cnt++;
+		if (fatal_cnt > 10) {
+
+			fprintf(stderr,
+				"InnoDB: Error: semaphore wait has lasted"
+				" > %lu seconds\n"
+				"InnoDB: We intentionally crash the server,"
+				" because it appears to be hung.\n",
+				(ulong) srv_fatal_semaphore_wait_threshold);
+
+			ut_error;
+		}
+	} else {
+		fatal_cnt = 0;
+	}
+
+	/* Flush stderr so that a database user gets the output
+	to possible MySQL error file */
+
+	fflush(stderr);
+
+	os_thread_sleep(1000000);
+
+	if (srv_shutdown_state < SRV_SHUTDOWN_CLEANUP) {
+
+		goto loop;
+	}
+
+	srv_error_monitor_active = FALSE;
+
+	/* We count the number of threads in os_thread_exit(). A created
+	thread should always use that to exit and not use return() to exit. */
+
+	os_thread_exit(NULL);
+
+	OS_THREAD_DUMMY_RETURN;
+}
+
+/***********************************************************************
+Tells the InnoDB server that there has been activity in the database
+and wakes up the master thread if it is suspended (not sleeping). Used
+in the MySQL interface. Note that there is a small chance that the master
+thread stays suspended (we do not protect our operation with the kernel
+mutex, for performace reasons). */
+UNIV_INTERN
+void
+srv_active_wake_master_thread(void)
+/*===============================*/
+{
+	srv_activity_count++;
+
+	if (srv_n_threads_active[SRV_MASTER] == 0) {
+
+		mutex_enter(&kernel_mutex);
+
+		srv_release_threads(SRV_MASTER, 1);
+
+		mutex_exit(&kernel_mutex);
+	}
+}
+
+/***********************************************************************
+Wakes up the master thread if it is suspended or being suspended. */
+UNIV_INTERN
+void
+srv_wake_master_thread(void)
+/*========================*/
+{
+	srv_activity_count++;
+
+	mutex_enter(&kernel_mutex);
+
+	srv_release_threads(SRV_MASTER, 1);
+
+	mutex_exit(&kernel_mutex);
+}
+
+/*************************************************************************
+The master thread controlling the server. */
+UNIV_INTERN
+os_thread_ret_t
+srv_master_thread(
+/*==============*/
+			/* out: a dummy parameter */
+	void*	arg __attribute__((unused)))
+			/* in: a dummy parameter required by
+			os_thread_create */
+{
+	os_event_t	event;
+	time_t		last_flush_time;
+	time_t		current_time;
+	ulint		old_activity_count;
+	ulint		n_pages_purged	= 0;
+	ulint		n_bytes_merged;
+	ulint		n_pages_flushed;
+	ulint		n_bytes_archived;
+	ulint		n_tables_to_drop;
+	ulint		n_ios;
+	ulint		n_ios_old;
+	ulint		n_ios_very_old;
+	ulint		n_pend_ios;
+	ibool		skip_sleep	= FALSE;
+	ulint		i;
+
+	ib_uint64_t	oldest_lsn;
+	
+#ifdef UNIV_DEBUG_THREAD_CREATION
+	fprintf(stderr, "Master thread starts, id %lu\n",
+		os_thread_pf(os_thread_get_curr_id()));
+#endif
+	srv_main_thread_process_no = os_proc_get_number();
+	srv_main_thread_id = os_thread_pf(os_thread_get_curr_id());
+
+	srv_table_reserve_slot(SRV_MASTER);
+
+	mutex_enter(&kernel_mutex);
+
+	srv_n_threads_active[SRV_MASTER]++;
+
+	mutex_exit(&kernel_mutex);
+
+loop:
+	/*****************************************************************/
+	/* ---- When there is database activity by users, we cycle in this
+	loop */
+
+	srv_main_thread_op_info = "reserving kernel mutex";
+
+	n_ios_very_old = log_sys->n_log_ios + buf_pool->n_pages_read
+		+ buf_pool->n_pages_written;
+	mutex_enter(&kernel_mutex);
+
+	/* Store the user activity counter at the start of this loop */
+	old_activity_count = srv_activity_count;
+
+	mutex_exit(&kernel_mutex);
+
+	if (srv_force_recovery >= SRV_FORCE_NO_BACKGROUND) {
+
+		goto suspend_thread;
+	}
+
+	/* ---- We run the following loop approximately once per second
+	when there is database activity */
+
+	skip_sleep = FALSE;
+
+	for (i = 0; i < 10; i++) {
+		n_ios_old = log_sys->n_log_ios + buf_pool->n_pages_read
+			+ buf_pool->n_pages_written;
+		srv_main_thread_op_info = "sleeping";
+
+		if (!skip_sleep) {
+
+			os_thread_sleep(1000000);
+		}
+
+		skip_sleep = FALSE;
+
+		/* ALTER TABLE in MySQL requires on Unix that the table handler
+		can drop tables lazily after there no longer are SELECT
+		queries to them. */
+
+		srv_main_thread_op_info = "doing background drop tables";
+
+		row_drop_tables_for_mysql_in_background();
+
+		srv_main_thread_op_info = "";
+
+		if (srv_fast_shutdown && srv_shutdown_state > 0) {
+
+			goto background_loop;
+		}
+
+		/* We flush the log once in a second even if no commit
+		is issued or the we have specified in my.cnf no flush
+		at transaction commit */
+
+		srv_main_thread_op_info = "flushing log";
+		log_buffer_flush_to_disk();
+
+		srv_main_thread_op_info = "making checkpoint";
+		log_free_check();
+
+		/* If there were less than 5 i/os during the
+		one second sleep, we assume that there is free
+		disk i/o capacity available, and it makes sense to
+		do an insert buffer merge. */
+
+		n_pend_ios = buf_get_n_pending_ios()
+			+ log_sys->n_pending_writes;
+		n_ios = log_sys->n_log_ios + buf_pool->n_pages_read
+			+ buf_pool->n_pages_written;
+		if (n_pend_ios < 3 && (n_ios - n_ios_old < PCT_IO(5))) {
+			srv_main_thread_op_info = "doing insert buffer merge";
+			ibuf_contract_for_n_pages(
+				TRUE, PCT_IBUF_IO((srv_insert_buffer_batch_size / 4)));
+
+			srv_main_thread_op_info = "flushing log";
+
+			log_buffer_flush_to_disk();
+		}
+
+		if (UNIV_UNLIKELY(buf_get_modified_ratio_pct()
+				  > srv_max_buf_pool_modified_pct)) {
+
+			/* Try to keep the number of modified pages in the
+			buffer pool under the limit wished by the user */
+
+ 			n_pages_flushed = buf_flush_batch(BUF_FLUSH_LIST, PCT_IO(100),
+							  IB_ULONGLONG_MAX);
+
+			/* If we had to do the flush, it may have taken
+			even more than 1 second, and also, there may be more
+			to flush. Do not sleep 1 second during the next
+			iteration of this loop. */
+
+			skip_sleep = TRUE;
+		} else if (srv_adaptive_checkpoint) {
+
+			/* Try to keep modified age not to exceed
+			max_checkpoint_age * 7/8 line */
+
+			mutex_enter(&(log_sys->mutex));
+
+			oldest_lsn = buf_pool_get_oldest_modification();
+			if (oldest_lsn == 0) {
+
+				mutex_exit(&(log_sys->mutex));
+
+			} else {
+				if ((log_sys->lsn - oldest_lsn)
+				    > (log_sys->max_checkpoint_age) - ((log_sys->max_checkpoint_age) / 8)) {
+					/* LOG_POOL_PREFLUSH_RATIO_ASYNC is exceeded. */
+					/* We should not flush from here. */
+					mutex_exit(&(log_sys->mutex));
+				} else if ((log_sys->lsn - oldest_lsn)
+				    > (log_sys->max_checkpoint_age) - ((log_sys->max_checkpoint_age) / 4)) {
+
+					/* 2nd defence line (max_checkpoint_age * 3/4) */
+
+					mutex_exit(&(log_sys->mutex));
+
+					n_pages_flushed = buf_flush_batch(BUF_FLUSH_LIST, PCT_IO(100),
+									  IB_ULONGLONG_MAX);
+					skip_sleep = TRUE;
+				} else if ((log_sys->lsn - oldest_lsn)
+					   > (log_sys->max_checkpoint_age)/2 ) {
+
+					/* 1st defence line (max_checkpoint_age * 1/2) */
+
+					mutex_exit(&(log_sys->mutex));
+
+					n_pages_flushed = buf_flush_batch(BUF_FLUSH_LIST, PCT_IO(10),
+									  IB_ULONGLONG_MAX);
+					skip_sleep = TRUE;
+				} else {
+					mutex_exit(&(log_sys->mutex));
+				}
+			}
+
+		}
+
+		if (srv_activity_count == old_activity_count) {
+
+			/* There is no user activity at the moment, go to
+			the background loop */
+
+			goto background_loop;
+		}
+	}
+
+	/* ---- We perform the following code approximately once per
+	10 seconds when there is database activity */
+
+#ifdef MEM_PERIODIC_CHECK
+	/* Check magic numbers of every allocated mem block once in 10
+	seconds */
+	mem_validate_all_blocks();
+#endif
+	/* If there were less than 200 i/os during the 10 second period,
+	we assume that there is free disk i/o capacity available, and it
+	makes sense to flush 100 pages. */
+
+	n_pend_ios = buf_get_n_pending_ios() + log_sys->n_pending_writes;
+	n_ios = log_sys->n_log_ios + buf_pool->n_pages_read
+		+ buf_pool->n_pages_written;
+ 	if (n_pend_ios < 3 && (n_ios - n_ios_very_old < PCT_IO(200))) {
+  
+  		srv_main_thread_op_info = "flushing buffer pool pages";
+ 		buf_flush_batch(BUF_FLUSH_LIST, PCT_IO(100), IB_ULONGLONG_MAX);
+
+		srv_main_thread_op_info = "flushing log";
+		log_buffer_flush_to_disk();
+	}
+
+	/* We run a batch of insert buffer merge every 10 seconds,
+	even if the server were active */
+
+	srv_main_thread_op_info = "doing insert buffer merge";
+	ibuf_contract_for_n_pages(TRUE, PCT_IBUF_IO((srv_insert_buffer_batch_size / 4)));
+
+	srv_main_thread_op_info = "flushing log";
+	log_buffer_flush_to_disk();
+
+	/* We run a full purge every 10 seconds, even if the server
+	were active */
+
+	last_flush_time = time(NULL);
+
+	do {
+
+		if (srv_fast_shutdown && srv_shutdown_state > 0) {
+
+			goto background_loop;
+		}
+
+		srv_main_thread_op_info = "purging";
+		n_pages_purged = trx_purge();
+
+		current_time = time(NULL);
+
+		if (difftime(current_time, last_flush_time) > 1) {
+			srv_main_thread_op_info = "flushing log";
+
+			log_buffer_flush_to_disk();
+			last_flush_time = current_time;
+		}
+	} while (n_pages_purged);
+
+	srv_main_thread_op_info = "flushing buffer pool pages";
+
+	/* Flush a few oldest pages to make a new checkpoint younger */
+
+	if (buf_get_modified_ratio_pct() > 70) {
+
+		/* If there are lots of modified pages in the buffer pool
+		(> 70 %), we assume we can afford reserving the disk(s) for
+		the time it requires to flush 100 pages */
+
+ 	        n_pages_flushed = buf_flush_batch(BUF_FLUSH_LIST, PCT_IO(100),
+						  IB_ULONGLONG_MAX);
+	} else {
+		/* Otherwise, we only flush a small number of pages so that
+		we do not unnecessarily use much disk i/o capacity from
+		other work */
+
+ 	        n_pages_flushed = buf_flush_batch(BUF_FLUSH_LIST, PCT_IO(10),
+						  IB_ULONGLONG_MAX);
+	}
+
+	srv_main_thread_op_info = "making checkpoint";
+
+	/* Make a new checkpoint about once in 10 seconds */
+
+	log_checkpoint(TRUE, FALSE);
+
+	srv_main_thread_op_info = "reserving kernel mutex";
+
+	mutex_enter(&kernel_mutex);
+
+	/* ---- When there is database activity, we jump from here back to
+	the start of loop */
+
+	if (srv_activity_count != old_activity_count) {
+		mutex_exit(&kernel_mutex);
+		goto loop;
+	}
+
+	mutex_exit(&kernel_mutex);
+
+	/* If the database is quiet, we enter the background loop */
+
+	/*****************************************************************/
+background_loop:
+	/* ---- In this loop we run background operations when the server
+	is quiet from user activity. Also in the case of a shutdown, we
+	loop here, flushing the buffer pool to the data files. */
+
+	/* The server has been quiet for a while: start running background
+	operations */
+
+	srv_main_thread_op_info = "doing background drop tables";
+
+	n_tables_to_drop = row_drop_tables_for_mysql_in_background();
+
+	if (n_tables_to_drop > 0) {
+		/* Do not monopolize the CPU even if there are tables waiting
+		in the background drop queue. (It is essentially a bug if
+		MySQL tries to drop a table while there are still open handles
+		to it and we had to put it to the background drop queue.) */
+
+		os_thread_sleep(100000);
+	}
+
+	srv_main_thread_op_info = "purging";
+
+	/* Run a full purge */
+
+	last_flush_time = time(NULL);
+
+	do {
+		if (srv_fast_shutdown && srv_shutdown_state > 0) {
+
+			break;
+		}
+
+		srv_main_thread_op_info = "purging";
+		n_pages_purged = trx_purge();
+
+		current_time = time(NULL);
+
+		if (difftime(current_time, last_flush_time) > 1) {
+			srv_main_thread_op_info = "flushing log";
+
+			log_buffer_flush_to_disk();
+			last_flush_time = current_time;
+		}
+	} while (n_pages_purged);
+
+	srv_main_thread_op_info = "reserving kernel mutex";
+
+	mutex_enter(&kernel_mutex);
+	if (srv_activity_count != old_activity_count) {
+		mutex_exit(&kernel_mutex);
+		goto loop;
+	}
+	mutex_exit(&kernel_mutex);
+
+	srv_main_thread_op_info = "doing insert buffer merge";
+
+	if (srv_fast_shutdown && srv_shutdown_state > 0) {
+		n_bytes_merged = 0;
+	} else {
+		n_bytes_merged = ibuf_contract_for_n_pages(
+			TRUE, PCT_IBUF_IO((srv_insert_buffer_batch_size * 5)));
+	}
+
+	srv_main_thread_op_info = "reserving kernel mutex";
+
+	mutex_enter(&kernel_mutex);
+	if (srv_activity_count != old_activity_count) {
+		mutex_exit(&kernel_mutex);
+		goto loop;
+	}
+	mutex_exit(&kernel_mutex);
+
+flush_loop:
+	srv_main_thread_op_info = "flushing buffer pool pages";
+
+	if (srv_fast_shutdown < 2) {
+ 	        n_pages_flushed = buf_flush_batch(BUF_FLUSH_LIST, PCT_IO(100),
+						  IB_ULONGLONG_MAX);
+	} else {
+		/* In the fastest shutdown we do not flush the buffer pool
+		to data files: we set n_pages_flushed to 0 artificially. */
+
+		n_pages_flushed = 0;
+	}
+
+	srv_main_thread_op_info = "reserving kernel mutex";
+
+	mutex_enter(&kernel_mutex);
+	if (srv_activity_count != old_activity_count) {
+		mutex_exit(&kernel_mutex);
+		goto loop;
+	}
+	mutex_exit(&kernel_mutex);
+
+	srv_main_thread_op_info = "waiting for buffer pool flush to end";
+	buf_flush_wait_batch_end(BUF_FLUSH_LIST);
+
+	srv_main_thread_op_info = "flushing log";
+
+	log_buffer_flush_to_disk();
+
+	srv_main_thread_op_info = "making checkpoint";
+
+	log_checkpoint(TRUE, FALSE);
+
+	if (buf_get_modified_ratio_pct() > srv_max_buf_pool_modified_pct) {
+
+		/* Try to keep the number of modified pages in the
+		buffer pool under the limit wished by the user */
+
+		goto flush_loop;
+	}
+
+	srv_main_thread_op_info = "reserving kernel mutex";
+
+	mutex_enter(&kernel_mutex);
+	if (srv_activity_count != old_activity_count) {
+		mutex_exit(&kernel_mutex);
+		goto loop;
+	}
+	mutex_exit(&kernel_mutex);
+	/*
+	srv_main_thread_op_info = "archiving log (if log archive is on)";
+
+	log_archive_do(FALSE, &n_bytes_archived);
+	*/
+	n_bytes_archived = 0;
+
+	/* Keep looping in the background loop if still work to do */
+
+	if (srv_fast_shutdown && srv_shutdown_state > 0) {
+		if (n_tables_to_drop + n_pages_flushed
+		    + n_bytes_archived != 0) {
+
+			/* If we are doing a fast shutdown (= the default)
+			we do not do purge or insert buffer merge. But we
+			flush the buffer pool completely to disk.
+			In a 'very fast' shutdown we do not flush the buffer
+			pool to data files: we have set n_pages_flushed to
+			0 artificially. */
+
+			goto background_loop;
+		}
+	} else if (n_tables_to_drop
+		   + n_pages_purged + n_bytes_merged + n_pages_flushed
+		   + n_bytes_archived != 0) {
+		/* In a 'slow' shutdown we run purge and the insert buffer
+		merge to completion */
+
+		goto background_loop;
+	}
+
+	/* There is no work for background operations either: suspend
+	master thread to wait for more server activity */
+
+suspend_thread:
+	srv_main_thread_op_info = "suspending";
+
+	mutex_enter(&kernel_mutex);
+
+	if (row_get_background_drop_list_len_low() > 0) {
+		mutex_exit(&kernel_mutex);
+
+		goto loop;
+	}
+
+	event = srv_suspend_thread();
+
+	mutex_exit(&kernel_mutex);
+
+	/* DO NOT CHANGE THIS STRING. innobase_start_or_create_for_mysql()
+	waits for database activity to die down when converting < 4.1.x
+	databases, and relies on this string being exactly as it is. InnoDB
+	manual also mentions this string in several places. */
+	srv_main_thread_op_info = "waiting for server activity";
+
+	os_event_wait(event);
+
+	if (srv_shutdown_state == SRV_SHUTDOWN_EXIT_THREADS) {
+		/* This is only extra safety, the thread should exit
+		already when the event wait ends */
+
+		os_thread_exit(NULL);
+	}
+
+	/* When there is user activity, InnoDB will set the event and the
+	main thread goes back to loop. */
+
+	goto loop;
+
+	OS_THREAD_DUMMY_RETURN;	/* Not reached, avoid compiler warning */
+}
+#endif /* !UNIV_HOTBACKUP */
diff --git a/storage/xtradb/srv/srv0start.c b/storage/xtradb/srv/srv0start.c
new file mode 100644
index 00000000000..2933e020b62
--- /dev/null
+++ b/storage/xtradb/srv/srv0start.c
@@ -0,0 +1,2052 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+Copyright (c) 2008, Google Inc.
+
+Portions of this file contain modifications contributed and copyrighted by
+Google, Inc. Those modifications are gratefully acknowledged and are described
+briefly in the InnoDB documentation. The contributions by Google are
+incorporated with their permission, and subject to the conditions contained in
+the file COPYING.Google.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/************************************************************************
+Starts the InnoDB database server
+
+Created 2/16/1996 Heikki Tuuri
+*************************************************************************/
+
+#include "os0proc.h"
+#include "sync0sync.h"
+#include "ut0mem.h"
+#include "mem0mem.h"
+#include "data0data.h"
+#include "data0type.h"
+#include "dict0dict.h"
+#include "buf0buf.h"
+#include "buf0flu.h"
+#include "buf0rea.h"
+#include "os0file.h"
+#include "os0thread.h"
+#include "fil0fil.h"
+#include "fsp0fsp.h"
+#include "rem0rec.h"
+#include "rem0cmp.h"
+#include "mtr0mtr.h"
+#include "log0log.h"
+#include "log0recv.h"
+#include "page0page.h"
+#include "page0cur.h"
+#include "trx0trx.h"
+#include "dict0boot.h"
+#include "dict0load.h"
+#include "trx0sys.h"
+#include "dict0crea.h"
+#include "btr0btr.h"
+#include "btr0pcur.h"
+#include "btr0cur.h"
+#include "btr0sea.h"
+#include "rem0rec.h"
+#include "srv0srv.h"
+#include "que0que.h"
+#include "usr0sess.h"
+#include "lock0lock.h"
+#include "trx0roll.h"
+#include "trx0purge.h"
+#include "row0ins.h"
+#include "row0sel.h"
+#include "row0upd.h"
+#include "row0row.h"
+#include "row0mysql.h"
+#include "lock0lock.h"
+#include "ibuf0ibuf.h"
+#include "pars0pars.h"
+#include "btr0sea.h"
+#include "srv0start.h"
+#include "que0que.h"
+
+/* Log sequence number immediately after startup */
+UNIV_INTERN ib_uint64_t	srv_start_lsn;
+/* Log sequence number at shutdown */
+UNIV_INTERN ib_uint64_t	srv_shutdown_lsn;
+
+#ifdef HAVE_DARWIN_THREADS
+# include <sys/utsname.h>
+UNIV_INTERN ibool	srv_have_fullfsync = FALSE;
+#endif
+
+UNIV_INTERN ibool	srv_start_raw_disk_in_use = FALSE;
+
+UNIV_INTERN ibool	srv_startup_is_before_trx_rollback_phase = FALSE;
+UNIV_INTERN ibool	srv_is_being_started = FALSE;
+UNIV_INTERN ibool	srv_was_started = FALSE;
+#ifndef UNIV_HOTBACKUP
+static ibool	srv_start_has_been_called = FALSE;
+#endif /* !UNIV_HOTBACKUP */
+
+/* At a shutdown the value first climbs to SRV_SHUTDOWN_CLEANUP
+and then to SRV_SHUTDOWN_LAST_PHASE */
+UNIV_INTERN ulint		srv_shutdown_state = 0;
+
+#ifndef UNIV_HOTBACKUP
+static os_file_t	files[1000];
+
+static mutex_t		ios_mutex;
+static ulint		ios;
+
+static ulint		n[SRV_MAX_N_IO_THREADS + 5];
+static os_thread_id_t	thread_ids[SRV_MAX_N_IO_THREADS + 5];
+
+/* We use this mutex to test the return value of pthread_mutex_trylock
+   on successful locking. HP-UX does NOT return 0, though Linux et al do. */
+static os_fast_mutex_t	srv_os_test_mutex;
+
+/* Name of srv_monitor_file */
+static char*	srv_monitor_file_name;
+#endif /* !UNIV_HOTBACKUP */
+
+#define SRV_N_PENDING_IOS_PER_THREAD	OS_AIO_N_PENDING_IOS_PER_THREAD
+#define SRV_MAX_N_PENDING_SYNC_IOS	100
+
+
+/* Avoid warnings when using purify */
+
+#ifdef HAVE_valgrind
+static int inno_bcmp(register const char *s1, register const char *s2,
+	register uint len)
+{
+	while ((len-- != 0) && (*s1++ == *s2++))
+		;
+
+	return(len + 1);
+}
+#define memcmp(A,B,C) inno_bcmp((A),(B),(C))
+#endif
+
+static
+char*
+srv_parse_megabytes(
+/*================*/
+			/* out: next character in string */
+	char*	str,	/* in: string containing a quantity in bytes */
+	ulint*	megs)	/* out: the number in megabytes */
+{
+	char*	endp;
+	ulint	size;
+
+	size = strtoul(str, &endp, 10);
+
+	str = endp;
+
+	switch (*str) {
+	case 'G': case 'g':
+		size *= 1024;
+		/* fall through */
+	case 'M': case 'm':
+		str++;
+		break;
+	default:
+		size /= 1024 * 1024;
+		break;
+	}
+
+	*megs = size;
+	return(str);
+}
+
+/*************************************************************************
+Reads the data files and their sizes from a character string given in
+the .cnf file. */
+UNIV_INTERN
+ibool
+srv_parse_data_file_paths_and_sizes(
+/*================================*/
+			/* out: TRUE if ok, FALSE on parse error */
+	char*	str)	/* in/out: the data file path string */
+{
+	char*	input_str;
+	char*	path;
+	ulint	size;
+	ulint	i	= 0;
+
+	srv_auto_extend_last_data_file = FALSE;
+	srv_last_file_size_max = 0;
+	srv_data_file_names = NULL;
+	srv_data_file_sizes = NULL;
+	srv_data_file_is_raw_partition = NULL;
+
+	input_str = str;
+
+	/* First calculate the number of data files and check syntax:
+	path:size[M | G];path:size[M | G]... . Note that a Windows path may
+	contain a drive name and a ':'. */
+
+	while (*str != '\0') {
+		path = str;
+
+		while ((*str != ':' && *str != '\0')
+		       || (*str == ':'
+			   && (*(str + 1) == '\\' || *(str + 1) == '/'
+			       || *(str + 1) == ':'))) {
+			str++;
+		}
+
+		if (*str == '\0') {
+			return(FALSE);
+		}
+
+		str++;
+
+		str = srv_parse_megabytes(str, &size);
+
+		if (0 == strncmp(str, ":autoextend",
+				 (sizeof ":autoextend") - 1)) {
+
+			str += (sizeof ":autoextend") - 1;
+
+			if (0 == strncmp(str, ":max:",
+					 (sizeof ":max:") - 1)) {
+
+				str += (sizeof ":max:") - 1;
+
+				str = srv_parse_megabytes(str, &size);
+			}
+
+			if (*str != '\0') {
+
+				return(FALSE);
+			}
+		}
+
+		if (strlen(str) >= 6
+		    && *str == 'n'
+		    && *(str + 1) == 'e'
+		    && *(str + 2) == 'w') {
+			str += 3;
+		}
+
+		if (*str == 'r' && *(str + 1) == 'a' && *(str + 2) == 'w') {
+			str += 3;
+		}
+
+		if (size == 0) {
+			return(FALSE);
+		}
+
+		i++;
+
+		if (*str == ';') {
+			str++;
+		} else if (*str != '\0') {
+
+			return(FALSE);
+		}
+	}
+
+	if (i == 0) {
+		/* If innodb_data_file_path was defined it must contain
+		at least one data file definition */
+
+		return(FALSE);
+	}
+
+	srv_data_file_names = malloc(i * sizeof *srv_data_file_names);
+	srv_data_file_sizes = malloc(i * sizeof *srv_data_file_sizes);
+	srv_data_file_is_raw_partition = malloc(
+		i * sizeof *srv_data_file_is_raw_partition);
+
+	srv_n_data_files = i;
+
+	/* Then store the actual values to our arrays */
+
+	str = input_str;
+	i = 0;
+
+	while (*str != '\0') {
+		path = str;
+
+		/* Note that we must step over the ':' in a Windows path;
+		a Windows path normally looks like C:\ibdata\ibdata1:1G, but
+		a Windows raw partition may have a specification like
+		\\.\C::1Gnewraw or \\.\PHYSICALDRIVE2:1Gnewraw */
+
+		while ((*str != ':' && *str != '\0')
+		       || (*str == ':'
+			   && (*(str + 1) == '\\' || *(str + 1) == '/'
+			       || *(str + 1) == ':'))) {
+			str++;
+		}
+
+		if (*str == ':') {
+			/* Make path a null-terminated string */
+			*str = '\0';
+			str++;
+		}
+
+		str = srv_parse_megabytes(str, &size);
+
+		srv_data_file_names[i] = path;
+		srv_data_file_sizes[i] = size;
+
+		if (0 == strncmp(str, ":autoextend",
+				 (sizeof ":autoextend") - 1)) {
+
+			srv_auto_extend_last_data_file = TRUE;
+
+			str += (sizeof ":autoextend") - 1;
+
+			if (0 == strncmp(str, ":max:",
+					 (sizeof ":max:") - 1)) {
+
+				str += (sizeof ":max:") - 1;
+
+				str = srv_parse_megabytes(
+					str, &srv_last_file_size_max);
+			}
+
+			if (*str != '\0') {
+
+				return(FALSE);
+			}
+		}
+
+		(srv_data_file_is_raw_partition)[i] = 0;
+
+		if (strlen(str) >= 6
+		    && *str == 'n'
+		    && *(str + 1) == 'e'
+		    && *(str + 2) == 'w') {
+			str += 3;
+			(srv_data_file_is_raw_partition)[i] = SRV_NEW_RAW;
+		}
+
+		if (*str == 'r' && *(str + 1) == 'a' && *(str + 2) == 'w') {
+			str += 3;
+
+			if ((srv_data_file_is_raw_partition)[i] == 0) {
+				(srv_data_file_is_raw_partition)[i] = SRV_OLD_RAW;
+			}
+		}
+
+		i++;
+
+		if (*str == ';') {
+			str++;
+		}
+	}
+
+	return(TRUE);
+}
+
+/*************************************************************************
+Reads log group home directories from a character string given in
+the .cnf file. */
+UNIV_INTERN
+ibool
+srv_parse_log_group_home_dirs(
+/*==========================*/
+			/* out: TRUE if ok, FALSE on parse error */
+	char*	str)	/* in/out: character string */
+{
+	char*	input_str;
+	char*	path;
+	ulint	i	= 0;
+
+	srv_log_group_home_dirs = NULL;
+
+	input_str = str;
+
+	/* First calculate the number of directories and check syntax:
+	path;path;... */
+
+	while (*str != '\0') {
+		path = str;
+
+		while (*str != ';' && *str != '\0') {
+			str++;
+		}
+
+		i++;
+
+		if (*str == ';') {
+			str++;
+		} else if (*str != '\0') {
+
+			return(FALSE);
+		}
+	}
+
+	if (i != 1) {
+		/* If innodb_log_group_home_dir was defined it must
+		contain exactly one path definition under current MySQL */
+
+		return(FALSE);
+	}
+
+	srv_log_group_home_dirs = malloc(i * sizeof *srv_log_group_home_dirs);
+
+	/* Then store the actual values to our array */
+
+	str = input_str;
+	i = 0;
+
+	while (*str != '\0') {
+		path = str;
+
+		while (*str != ';' && *str != '\0') {
+			str++;
+		}
+
+		if (*str == ';') {
+			*str = '\0';
+			str++;
+		}
+
+		srv_log_group_home_dirs[i] = path;
+
+		i++;
+	}
+
+	return(TRUE);
+}
+
+/*************************************************************************
+Frees the memory allocated by srv_parse_data_file_paths_and_sizes()
+and srv_parse_log_group_home_dirs(). */
+UNIV_INTERN
+void
+srv_free_paths_and_sizes(void)
+/*==========================*/
+{
+	free(srv_data_file_names);
+	srv_data_file_names = NULL;
+	free(srv_data_file_sizes);
+	srv_data_file_sizes = NULL;
+	free(srv_data_file_is_raw_partition);
+	srv_data_file_is_raw_partition = NULL;
+	free(srv_log_group_home_dirs);
+	srv_log_group_home_dirs = NULL;
+}
+
+#ifndef UNIV_HOTBACKUP
+/************************************************************************
+I/o-handler thread function. */
+static
+
+os_thread_ret_t
+io_handler_thread(
+/*==============*/
+	void*	arg)
+{
+	ulint	segment;
+	ulint	i;
+
+	segment = *((ulint*)arg);
+
+#ifdef UNIV_DEBUG_THREAD_CREATION
+	fprintf(stderr, "Io handler thread %lu starts, id %lu\n", segment,
+		os_thread_pf(os_thread_get_curr_id()));
+#endif
+	for (i = 0;; i++) {
+		fil_aio_wait(segment);
+
+		mutex_enter(&ios_mutex);
+		ios++;
+		mutex_exit(&ios_mutex);
+	}
+
+	/* We count the number of threads in os_thread_exit(). A created
+	thread should always use that to exit and not use return() to exit.
+	The thread actually never comes here because it is exited in an
+	os_event_wait(). */
+
+	os_thread_exit(NULL);
+
+	OS_THREAD_DUMMY_RETURN;
+}
+#endif /* !UNIV_HOTBACKUP */
+
+#ifdef __WIN__
+#define SRV_PATH_SEPARATOR	'\\'
+#else
+#define SRV_PATH_SEPARATOR	'/'
+#endif
+
+/*************************************************************************
+Normalizes a directory path for Windows: converts slashes to backslashes. */
+UNIV_INTERN
+void
+srv_normalize_path_for_win(
+/*=======================*/
+	char*	str __attribute__((unused)))	/* in/out: null-terminated
+						character string */
+{
+#ifdef __WIN__
+	for (; *str; str++) {
+
+		if (*str == '/') {
+			*str = '\\';
+		}
+	}
+#endif
+}
+
+/*************************************************************************
+Adds a slash or a backslash to the end of a string if it is missing
+and the string is not empty. */
+UNIV_INTERN
+char*
+srv_add_path_separator_if_needed(
+/*=============================*/
+			/* out: string which has the separator if the
+			string is not empty */
+	char*	str)	/* in: null-terminated character string */
+{
+	char*	out_str;
+	ulint	len	= ut_strlen(str);
+
+	if (len == 0 || str[len - 1] == SRV_PATH_SEPARATOR) {
+
+		return(str);
+	}
+
+	out_str = ut_malloc(len + 2);
+	memcpy(out_str, str, len);
+	out_str[len] = SRV_PATH_SEPARATOR;
+	out_str[len + 1] = 0;
+
+	return(out_str);
+}
+
+#ifndef UNIV_HOTBACKUP
+/*************************************************************************
+Calculates the low 32 bits when a file size which is given as a number
+database pages is converted to the number of bytes. */
+static
+ulint
+srv_calc_low32(
+/*===========*/
+				/* out: low 32 bytes of file size when
+				expressed in bytes */
+	ulint	file_size)	/* in: file size in database pages */
+{
+	return(0xFFFFFFFFUL & (file_size << UNIV_PAGE_SIZE_SHIFT));
+}
+
+/*************************************************************************
+Calculates the high 32 bits when a file size which is given as a number
+database pages is converted to the number of bytes. */
+static
+ulint
+srv_calc_high32(
+/*============*/
+				/* out: high 32 bytes of file size when
+				expressed in bytes */
+	ulint	file_size)	/* in: file size in database pages */
+{
+	return(file_size >> (32 - UNIV_PAGE_SIZE_SHIFT));
+}
+
+/*************************************************************************
+Creates or opens the log files and closes them. */
+static
+ulint
+open_or_create_log_file(
+/*====================*/
+					/* out: DB_SUCCESS or error code */
+	ibool	create_new_db,		/* in: TRUE if we should create a
+					new database */
+	ibool*	log_file_created,	/* out: TRUE if new log file
+					created */
+	ibool	log_file_has_been_opened,/* in: TRUE if a log file has been
+					opened before: then it is an error
+					to try to create another log file */
+	ulint	k,			/* in: log group number */
+	ulint	i)			/* in: log file number in group */
+{
+	ibool	ret;
+	ulint	size;
+	ulint	size_high;
+	char	name[10000];
+
+	UT_NOT_USED(create_new_db);
+
+	*log_file_created = FALSE;
+
+	srv_normalize_path_for_win(srv_log_group_home_dirs[k]);
+	srv_log_group_home_dirs[k] = srv_add_path_separator_if_needed(
+		srv_log_group_home_dirs[k]);
+
+	ut_a(strlen(srv_log_group_home_dirs[k])
+	     < (sizeof name) - 10 - sizeof "ib_logfile");
+	sprintf(name, "%s%s%lu", srv_log_group_home_dirs[k],
+		"ib_logfile", (ulong) i);
+
+	files[i] = os_file_create(name, OS_FILE_CREATE, OS_FILE_NORMAL,
+				  OS_LOG_FILE, &ret);
+	if (ret == FALSE) {
+		if (os_file_get_last_error(FALSE) != OS_FILE_ALREADY_EXISTS
+#ifdef UNIV_AIX
+		    /* AIX 5.1 after security patch ML7 may have errno set
+		    to 0 here, which causes our function to return 100;
+		    work around that AIX problem */
+		    && os_file_get_last_error(FALSE) != 100
+#endif
+		    ) {
+			fprintf(stderr,
+				"InnoDB: Error in creating"
+				" or opening %s\n", name);
+
+			return(DB_ERROR);
+		}
+
+		files[i] = os_file_create(name, OS_FILE_OPEN, OS_FILE_AIO,
+					  OS_LOG_FILE, &ret);
+		if (!ret) {
+			fprintf(stderr,
+				"InnoDB: Error in opening %s\n", name);
+
+			return(DB_ERROR);
+		}
+
+		ret = os_file_get_size(files[i], &size, &size_high);
+		ut_a(ret);
+
+		if (size != srv_calc_low32(srv_log_file_size)
+		    || size_high != srv_calc_high32(srv_log_file_size)) {
+
+			fprintf(stderr,
+				"InnoDB: Error: log file %s is"
+				" of different size %lu %lu bytes\n"
+				"InnoDB: than specified in the .cnf"
+				" file %lu %lu bytes!\n",
+				name, (ulong) size_high, (ulong) size,
+				(ulong) srv_calc_high32(srv_log_file_size),
+				(ulong) srv_calc_low32(srv_log_file_size));
+
+			return(DB_ERROR);
+		}
+	} else {
+		*log_file_created = TRUE;
+
+		ut_print_timestamp(stderr);
+
+		fprintf(stderr,
+			"  InnoDB: Log file %s did not exist:"
+			" new to be created\n",
+			name);
+		if (log_file_has_been_opened) {
+
+			return(DB_ERROR);
+		}
+
+		fprintf(stderr, "InnoDB: Setting log file %s size to %lu MB\n",
+			name, (ulong) srv_log_file_size
+			>> (20 - UNIV_PAGE_SIZE_SHIFT));
+
+		fprintf(stderr,
+			"InnoDB: Database physically writes the file"
+			" full: wait...\n");
+
+		ret = os_file_set_size(name, files[i],
+				       srv_calc_low32(srv_log_file_size),
+				       srv_calc_high32(srv_log_file_size));
+		if (!ret) {
+			fprintf(stderr,
+				"InnoDB: Error in creating %s:"
+				" probably out of disk space\n",
+				name);
+
+			return(DB_ERROR);
+		}
+	}
+
+	ret = os_file_close(files[i]);
+	ut_a(ret);
+
+	if (i == 0) {
+		/* Create in memory the file space object
+		which is for this log group */
+
+		fil_space_create(name,
+				 2 * k + SRV_LOG_SPACE_FIRST_ID, 0, FIL_LOG);
+	}
+
+	ut_a(fil_validate());
+
+	fil_node_create(name, srv_log_file_size,
+			2 * k + SRV_LOG_SPACE_FIRST_ID, FALSE);
+#ifdef UNIV_LOG_ARCHIVE
+	/* If this is the first log group, create the file space object
+	for archived logs.
+	Under MySQL, no archiving ever done. */
+
+	if (k == 0 && i == 0) {
+		arch_space_id = 2 * k + 1 + SRV_LOG_SPACE_FIRST_ID;
+
+		fil_space_create("arch_log_space", arch_space_id, 0, FIL_LOG);
+	} else {
+		arch_space_id = ULINT_UNDEFINED;
+	}
+#endif /* UNIV_LOG_ARCHIVE */
+	if (i == 0) {
+		log_group_init(k, srv_n_log_files,
+			       srv_log_file_size * UNIV_PAGE_SIZE,
+			       2 * k + SRV_LOG_SPACE_FIRST_ID,
+			       SRV_LOG_SPACE_FIRST_ID + 1); /* dummy arch
+							    space id */
+	}
+
+	return(DB_SUCCESS);
+}
+
+/*************************************************************************
+Creates or opens database data files and closes them. */
+static
+ulint
+open_or_create_data_files(
+/*======================*/
+					/* out: DB_SUCCESS or error code */
+	ibool*		create_new_db,	/* out: TRUE if new database should be
+					created */
+#ifdef UNIV_LOG_ARCHIVE
+	ulint*		min_arch_log_no,/* out: min of archived log
+					numbers in data files */
+	ulint*		max_arch_log_no,/* out: max of archived log
+					numbers in data files */
+#endif /* UNIV_LOG_ARCHIVE */
+	ib_uint64_t*	min_flushed_lsn,/* out: min of flushed lsn
+					values in data files */
+	ib_uint64_t*	max_flushed_lsn,/* out: max of flushed lsn
+					values in data files */
+	ulint*		sum_of_new_sizes)/* out: sum of sizes of the
+					new files added */
+{
+	ibool	ret;
+	ulint	i;
+	ibool	one_opened	= FALSE;
+	ibool	one_created	= FALSE;
+	ulint	size;
+	ulint	size_high;
+	ulint	rounded_size_pages;
+	char	name[10000];
+
+	if (srv_n_data_files >= 1000) {
+		fprintf(stderr, "InnoDB: can only have < 1000 data files\n"
+			"InnoDB: you have defined %lu\n",
+			(ulong) srv_n_data_files);
+		return(DB_ERROR);
+	}
+
+	*sum_of_new_sizes = 0;
+
+	*create_new_db = FALSE;
+
+	srv_normalize_path_for_win(srv_data_home);
+	srv_data_home = srv_add_path_separator_if_needed(srv_data_home);
+
+	for (i = 0; i < srv_n_data_files; i++) {
+		srv_normalize_path_for_win(srv_data_file_names[i]);
+
+		ut_a(strlen(srv_data_home) + strlen(srv_data_file_names[i])
+		     < (sizeof name) - 1);
+		sprintf(name, "%s%s", srv_data_home, srv_data_file_names[i]);
+
+		if (srv_data_file_is_raw_partition[i] == 0) {
+
+			/* First we try to create the file: if it already
+			exists, ret will get value FALSE */
+
+			files[i] = os_file_create(name, OS_FILE_CREATE,
+						  OS_FILE_NORMAL,
+						  OS_DATA_FILE, &ret);
+
+			if (ret == FALSE && os_file_get_last_error(FALSE)
+			    != OS_FILE_ALREADY_EXISTS
+#ifdef UNIV_AIX
+			    /* AIX 5.1 after security patch ML7 may have
+			    errno set to 0 here, which causes our function
+			    to return 100; work around that AIX problem */
+			    && os_file_get_last_error(FALSE) != 100
+#endif
+			    ) {
+				fprintf(stderr,
+					"InnoDB: Error in creating"
+					" or opening %s\n",
+					name);
+
+				return(DB_ERROR);
+			}
+		} else if (srv_data_file_is_raw_partition[i] == SRV_NEW_RAW) {
+			/* The partition is opened, not created; then it is
+			written over */
+
+			srv_start_raw_disk_in_use = TRUE;
+			srv_created_new_raw = TRUE;
+
+			files[i] = os_file_create(name, OS_FILE_OPEN_RAW,
+						  OS_FILE_NORMAL,
+						  OS_DATA_FILE, &ret);
+			if (!ret) {
+				fprintf(stderr,
+					"InnoDB: Error in opening %s\n", name);
+
+				return(DB_ERROR);
+			}
+		} else if (srv_data_file_is_raw_partition[i] == SRV_OLD_RAW) {
+			srv_start_raw_disk_in_use = TRUE;
+
+			ret = FALSE;
+		} else {
+			ut_a(0);
+		}
+
+		if (ret == FALSE) {
+			/* We open the data file */
+
+			if (one_created) {
+				fprintf(stderr,
+					"InnoDB: Error: data files can only"
+					" be added at the end\n");
+				fprintf(stderr,
+					"InnoDB: of a tablespace, but"
+					" data file %s existed beforehand.\n",
+					name);
+				return(DB_ERROR);
+			}
+
+			if (srv_data_file_is_raw_partition[i] == SRV_OLD_RAW) {
+				files[i] = os_file_create(
+					name, OS_FILE_OPEN_RAW,
+					OS_FILE_NORMAL, OS_DATA_FILE, &ret);
+			} else if (i == 0) {
+				files[i] = os_file_create(
+					name, OS_FILE_OPEN_RETRY,
+					OS_FILE_NORMAL, OS_DATA_FILE, &ret);
+			} else {
+				files[i] = os_file_create(
+					name, OS_FILE_OPEN, OS_FILE_NORMAL,
+					OS_DATA_FILE, &ret);
+			}
+
+			if (!ret) {
+				fprintf(stderr,
+					"InnoDB: Error in opening %s\n", name);
+				os_file_get_last_error(TRUE);
+
+				return(DB_ERROR);
+			}
+
+			if (srv_data_file_is_raw_partition[i] == SRV_OLD_RAW) {
+
+				goto skip_size_check;
+			}
+
+			ret = os_file_get_size(files[i], &size, &size_high);
+			ut_a(ret);
+			/* Round size downward to megabytes */
+
+			rounded_size_pages
+				= (size / (1024 * 1024) + 4096 * size_high)
+					<< (20 - UNIV_PAGE_SIZE_SHIFT);
+
+			if (i == srv_n_data_files - 1
+			    && srv_auto_extend_last_data_file) {
+
+				if (srv_data_file_sizes[i] > rounded_size_pages
+				    || (srv_last_file_size_max > 0
+					&& srv_last_file_size_max
+					< rounded_size_pages)) {
+
+					fprintf(stderr,
+						"InnoDB: Error: auto-extending"
+						" data file %s is"
+						" of a different size\n"
+						"InnoDB: %lu pages (rounded"
+						" down to MB) than specified"
+						" in the .cnf file:\n"
+						"InnoDB: initial %lu pages,"
+						" max %lu (relevant if"
+						" non-zero) pages!\n",
+						name,
+						(ulong) rounded_size_pages,
+						(ulong) srv_data_file_sizes[i],
+						(ulong)
+						srv_last_file_size_max);
+
+					return(DB_ERROR);
+				}
+
+				srv_data_file_sizes[i] = rounded_size_pages;
+			}
+
+			if (rounded_size_pages != srv_data_file_sizes[i]) {
+
+				fprintf(stderr,
+					"InnoDB: Error: data file %s"
+					" is of a different size\n"
+					"InnoDB: %lu pages"
+					" (rounded down to MB)\n"
+					"InnoDB: than specified"
+					" in the .cnf file %lu pages!\n",
+					name,
+					(ulong) rounded_size_pages,
+					(ulong) srv_data_file_sizes[i]);
+
+				return(DB_ERROR);
+			}
+skip_size_check:
+			fil_read_flushed_lsn_and_arch_log_no(
+				files[i], one_opened,
+#ifdef UNIV_LOG_ARCHIVE
+				min_arch_log_no, max_arch_log_no,
+#endif /* UNIV_LOG_ARCHIVE */
+				min_flushed_lsn, max_flushed_lsn);
+			one_opened = TRUE;
+		} else {
+			/* We created the data file and now write it full of
+			zeros */
+
+			one_created = TRUE;
+
+			if (i > 0) {
+				ut_print_timestamp(stderr);
+				fprintf(stderr,
+					"  InnoDB: Data file %s did not"
+					" exist: new to be created\n",
+					name);
+			} else {
+				fprintf(stderr,
+					"InnoDB: The first specified"
+					" data file %s did not exist:\n"
+					"InnoDB: a new database"
+					" to be created!\n", name);
+				*create_new_db = TRUE;
+			}
+
+			ut_print_timestamp(stderr);
+			fprintf(stderr,
+				"  InnoDB: Setting file %s size to %lu MB\n",
+				name,
+				(ulong) (srv_data_file_sizes[i]
+					 >> (20 - UNIV_PAGE_SIZE_SHIFT)));
+
+			fprintf(stderr,
+				"InnoDB: Database physically writes the"
+				" file full: wait...\n");
+
+			ret = os_file_set_size(
+				name, files[i],
+				srv_calc_low32(srv_data_file_sizes[i]),
+				srv_calc_high32(srv_data_file_sizes[i]));
+
+			if (!ret) {
+				fprintf(stderr,
+					"InnoDB: Error in creating %s:"
+					" probably out of disk space\n", name);
+
+				return(DB_ERROR);
+			}
+
+			*sum_of_new_sizes = *sum_of_new_sizes
+				+ srv_data_file_sizes[i];
+		}
+
+		ret = os_file_close(files[i]);
+		ut_a(ret);
+
+		if (i == 0) {
+			fil_space_create(name, 0, 0, FIL_TABLESPACE);
+		}
+
+		ut_a(fil_validate());
+
+		fil_node_create(name, srv_data_file_sizes[i], 0,
+				srv_data_file_is_raw_partition[i] != 0);
+	}
+
+	ios = 0;
+
+	mutex_create(&ios_mutex, SYNC_NO_ORDER_CHECK);
+
+	return(DB_SUCCESS);
+}
+
+/********************************************************************
+Starts InnoDB and creates a new database if database files
+are not found and the user wants. */
+UNIV_INTERN
+int
+innobase_start_or_create_for_mysql(void)
+/*====================================*/
+				/* out: DB_SUCCESS or error code */
+{
+	buf_pool_t*	ret;
+	ibool		create_new_db;
+	ibool		log_file_created;
+	ibool		log_created	= FALSE;
+	ibool		log_opened	= FALSE;
+	ib_uint64_t	min_flushed_lsn;
+	ib_uint64_t	max_flushed_lsn;
+#ifdef UNIV_LOG_ARCHIVE
+	ulint		min_arch_log_no;
+	ulint		max_arch_log_no;
+#endif /* UNIV_LOG_ARCHIVE */
+	ulint		sum_of_new_sizes;
+	ulint		sum_of_data_file_sizes;
+	ulint		tablespace_size_in_header;
+	ulint		err;
+	ulint		i;
+	my_bool		srv_file_per_table_original_value
+		= srv_file_per_table;
+	mtr_t		mtr;
+#ifdef HAVE_DARWIN_THREADS
+# ifdef F_FULLFSYNC
+	/* This executable has been compiled on Mac OS X 10.3 or later.
+	Assume that F_FULLFSYNC is available at run-time. */
+	srv_have_fullfsync = TRUE;
+# else /* F_FULLFSYNC */
+	/* This executable has been compiled on Mac OS X 10.2
+	or earlier.  Determine if the executable is running
+	on Mac OS X 10.3 or later. */
+	struct utsname utsname;
+	if (uname(&utsname)) {
+		fputs("InnoDB: cannot determine Mac OS X version!\n", stderr);
+	} else {
+		srv_have_fullfsync = strcmp(utsname.release, "7.") >= 0;
+	}
+	if (!srv_have_fullfsync) {
+		fputs("InnoDB: On Mac OS X, fsync() may be"
+		      " broken on internal drives,\n"
+		      "InnoDB: making transactions unsafe!\n", stderr);
+	}
+# endif /* F_FULLFSYNC */
+#endif /* HAVE_DARWIN_THREADS */
+
+	if (sizeof(ulint) != sizeof(void*)) {
+		fprintf(stderr,
+			"InnoDB: Error: size of InnoDB's ulint is %lu,"
+			" but size of void* is %lu.\n"
+			"InnoDB: The sizes should be the same"
+			" so that on a 64-bit platform you can\n"
+			"InnoDB: allocate more than 4 GB of memory.",
+			(ulong)sizeof(ulint), (ulong)sizeof(void*));
+	}
+
+	/* System tables are created in tablespace 0.  Thus, we must
+	temporarily clear srv_file_per_table.  This is ok, because the
+	server will not accept connections (which could modify
+	innodb_file_per_table) until this function has returned. */
+	srv_file_per_table = FALSE;
+#ifdef UNIV_DEBUG
+	fprintf(stderr,
+		"InnoDB: !!!!!!!! UNIV_DEBUG switched on !!!!!!!!!\n");
+#endif
+
+#ifdef UNIV_IBUF_DEBUG
+	fprintf(stderr,
+		"InnoDB: !!!!!!!! UNIV_IBUF_DEBUG switched on !!!!!!!!!\n"
+		"InnoDB: Crash recovery will fail with UNIV_IBUF_DEBUG\n");
+#endif
+
+#ifdef UNIV_SYNC_DEBUG
+	fprintf(stderr,
+		"InnoDB: !!!!!!!! UNIV_SYNC_DEBUG switched on !!!!!!!!!\n");
+#endif
+
+#ifdef UNIV_SEARCH_DEBUG
+	fprintf(stderr,
+		"InnoDB: !!!!!!!! UNIV_SEARCH_DEBUG switched on !!!!!!!!!\n");
+#endif
+
+#ifdef UNIV_MEM_DEBUG
+	fprintf(stderr,
+		"InnoDB: !!!!!!!! UNIV_MEM_DEBUG switched on !!!!!!!!!\n");
+#endif
+
+	if (UNIV_LIKELY(srv_use_sys_malloc)) {
+		fprintf(stderr,
+			"InnoDB: The InnoDB memory heap is disabled\n");
+	}
+
+#ifdef HAVE_GCC_ATOMIC_BUILTINS
+#ifdef INNODB_RW_LOCKS_USE_ATOMICS
+	fprintf(stderr,
+		"InnoDB: Mutexes and rw_locks use GCC atomic builtins.\n");
+#else /* INNODB_RW_LOCKS_USE_ATOMICS */
+	fprintf(stderr,
+		"InnoDB: Mutexes use GCC atomic builtins, rw_locks do not.\n");
+#endif /* INNODB_RW_LOCKS_USE_ATOMICS */
+#else /* HAVE_GCC_ATOMIC_BUILTINS */
+	fprintf(stderr,
+		"InnoDB: Neither mutexes nor rw_locks use GCC atomic builtins.\n");
+#endif /* HAVE_GCC_ATOMIC_BUILTINS */
+
+	/* Since InnoDB does not currently clean up all its internal data
+	structures in MySQL Embedded Server Library server_end(), we
+	print an error message if someone tries to start up InnoDB a
+	second time during the process lifetime. */
+
+	if (srv_start_has_been_called) {
+		fprintf(stderr,
+			"InnoDB: Error:startup called second time"
+			" during the process lifetime.\n"
+			"InnoDB: In the MySQL Embedded Server Library"
+			" you cannot call server_init()\n"
+			"InnoDB: more than once during"
+			" the process lifetime.\n");
+	}
+
+	srv_start_has_been_called = TRUE;
+
+#ifdef UNIV_DEBUG
+	log_do_write = TRUE;
+#endif /* UNIV_DEBUG */
+	/*	yydebug = TRUE; */
+
+	srv_is_being_started = TRUE;
+	srv_startup_is_before_trx_rollback_phase = TRUE;
+	os_aio_use_native_aio = FALSE;
+
+#ifdef __WIN__
+	if (os_get_os_version() == OS_WIN95
+	    || os_get_os_version() == OS_WIN31
+	    || os_get_os_version() == OS_WINNT) {
+
+		/* On Win 95, 98, ME, Win32 subsystem for Windows 3.1,
+		and NT use simulated aio. In NT Windows provides async i/o,
+		but when run in conjunction with InnoDB Hot Backup, it seemed
+		to corrupt the data files. */
+
+		os_aio_use_native_aio = FALSE;
+	} else {
+		/* On Win 2000 and XP use async i/o */
+		os_aio_use_native_aio = TRUE;
+	}
+#endif
+	if (srv_file_flush_method_str == NULL) {
+		/* These are the default options */
+
+		srv_unix_file_flush_method = SRV_UNIX_FSYNC;
+
+		srv_win_file_flush_method = SRV_WIN_IO_UNBUFFERED;
+#ifndef __WIN__
+	} else if (0 == ut_strcmp(srv_file_flush_method_str, "fsync")) {
+		srv_unix_file_flush_method = SRV_UNIX_FSYNC;
+
+	} else if (0 == ut_strcmp(srv_file_flush_method_str, "O_DSYNC")) {
+		srv_unix_file_flush_method = SRV_UNIX_O_DSYNC;
+
+	} else if (0 == ut_strcmp(srv_file_flush_method_str, "O_DIRECT")) {
+		srv_unix_file_flush_method = SRV_UNIX_O_DIRECT;
+
+	} else if (0 == ut_strcmp(srv_file_flush_method_str, "littlesync")) {
+		srv_unix_file_flush_method = SRV_UNIX_LITTLESYNC;
+
+	} else if (0 == ut_strcmp(srv_file_flush_method_str, "nosync")) {
+		srv_unix_file_flush_method = SRV_UNIX_NOSYNC;
+#else
+	} else if (0 == ut_strcmp(srv_file_flush_method_str, "normal")) {
+		srv_win_file_flush_method = SRV_WIN_IO_NORMAL;
+		os_aio_use_native_aio = FALSE;
+
+	} else if (0 == ut_strcmp(srv_file_flush_method_str, "unbuffered")) {
+		srv_win_file_flush_method = SRV_WIN_IO_UNBUFFERED;
+		os_aio_use_native_aio = FALSE;
+
+	} else if (0 == ut_strcmp(srv_file_flush_method_str,
+				  "async_unbuffered")) {
+		srv_win_file_flush_method = SRV_WIN_IO_UNBUFFERED;
+#endif
+	} else {
+		fprintf(stderr,
+			"InnoDB: Unrecognized value %s for"
+			" innodb_flush_method\n",
+			srv_file_flush_method_str);
+		return(DB_ERROR);
+	}
+
+	/* Note that the call srv_boot() also changes the values of
+	some variables to the units used by InnoDB internally */
+
+	/* Set the maximum number of threads which can wait for a semaphore
+	inside InnoDB: this is the 'sync wait array' size, as well as the
+	maximum number of threads that can wait in the 'srv_conc array' for
+	their time to enter InnoDB. */
+
+#if defined(__NETWARE__)
+
+	/* Create less event semaphores because Win 98/ME had
+	difficulty creating 40000 event semaphores.  Comment from
+	Novell, Inc.: also, these just take a lot of memory on
+	NetWare. */
+	srv_max_n_threads = 1000;
+#else
+	if (srv_buf_pool_size >= 1000 * 1024 * 1024) {
+		/* If buffer pool is less than 1000 MB,
+		assume fewer threads. */
+		srv_max_n_threads = 50000;
+
+	} else if (srv_buf_pool_size >= 8 * 1024 * 1024) {
+
+		srv_max_n_threads = 10000;
+	} else {
+		srv_max_n_threads = 1000;	/* saves several MB of memory,
+						especially in 64-bit
+						computers */
+	}
+#endif
+	err = srv_boot();
+
+	if (err != DB_SUCCESS) {
+
+		return((int) err);
+	}
+
+	mutex_create(&srv_monitor_file_mutex, SYNC_NO_ORDER_CHECK);
+
+	if (srv_innodb_status) {
+		srv_monitor_file_name = mem_alloc(
+			strlen(fil_path_to_mysql_datadir)
+			+ 20 + sizeof "/innodb_status.");
+		sprintf(srv_monitor_file_name, "%s/innodb_status.%lu",
+			fil_path_to_mysql_datadir, os_proc_get_number());
+		srv_monitor_file = fopen(srv_monitor_file_name, "w+");
+		if (!srv_monitor_file) {
+			fprintf(stderr, "InnoDB: unable to create %s: %s\n",
+				srv_monitor_file_name, strerror(errno));
+			return(DB_ERROR);
+		}
+	} else {
+		srv_monitor_file_name = NULL;
+		srv_monitor_file = os_file_create_tmpfile();
+		if (!srv_monitor_file) {
+			return(DB_ERROR);
+		}
+	}
+
+	mutex_create(&srv_dict_tmpfile_mutex, SYNC_DICT_OPERATION);
+
+	srv_dict_tmpfile = os_file_create_tmpfile();
+	if (!srv_dict_tmpfile) {
+		return(DB_ERROR);
+	}
+
+	mutex_create(&srv_misc_tmpfile_mutex, SYNC_ANY_LATCH);
+
+	srv_misc_tmpfile = os_file_create_tmpfile();
+	if (!srv_misc_tmpfile) {
+		return(DB_ERROR);
+	}
+
+	/* over write innodb_file_io_threads */
+	srv_n_file_io_threads = 2 + srv_n_read_io_threads + srv_n_write_io_threads;
+
+	/* Restrict the maximum number of file i/o threads */
+	if (srv_n_file_io_threads > SRV_MAX_N_IO_THREADS) {
+
+		srv_n_file_io_threads = SRV_MAX_N_IO_THREADS;
+		srv_n_read_io_threads = srv_n_write_io_threads = (SRV_MAX_N_IO_THREADS - 2) / 2;
+	}
+
+	if (!os_aio_use_native_aio) {
+		/* In simulated aio we currently have use only for 4 threads */
+		/*srv_n_file_io_threads = 4;*/
+
+		os_aio_init(8 * SRV_N_PENDING_IOS_PER_THREAD
+			    * srv_n_file_io_threads,
+			    srv_n_read_io_threads, srv_n_write_io_threads,
+			    SRV_MAX_N_PENDING_SYNC_IOS * 8);
+	} else {
+		os_aio_init(SRV_N_PENDING_IOS_PER_THREAD
+			    * srv_n_file_io_threads,
+			    srv_n_read_io_threads, srv_n_write_io_threads,
+			    SRV_MAX_N_PENDING_SYNC_IOS);
+	}
+
+	fil_init(srv_max_n_open_files);
+
+	ret = buf_pool_init();
+
+	if (ret == NULL) {
+		fprintf(stderr,
+			"InnoDB: Fatal error: cannot allocate the memory"
+			" for the buffer pool\n");
+
+		return(DB_ERROR);
+	}
+
+#ifdef UNIV_DEBUG
+	/* We have observed deadlocks with a 5MB buffer pool but
+	the actual lower limit could very well be a little higher. */
+
+	if (srv_buf_pool_size <= 5 * 1024 * 1024) {
+
+		fprintf(stderr, "InnoDB: Warning: Small buffer pool size "
+			"(%luM), the flst_validate() debug function "
+			"can cause a deadlock if the buffer pool fills up.\n",
+			srv_buf_pool_size / 1024 / 1024);
+	}
+#endif
+
+	fsp_init();
+	log_init();
+
+	lock_sys_create(srv_lock_table_size);
+
+	/* Create i/o-handler threads: */
+
+	for (i = 0; i < srv_n_file_io_threads; i++) {
+		n[i] = i;
+
+		os_thread_create(io_handler_thread, n + i, thread_ids + i);
+	}
+
+#ifdef UNIV_LOG_ARCHIVE
+	if (0 != ut_strcmp(srv_log_group_home_dirs[0], srv_arch_dir)) {
+		fprintf(stderr,
+			"InnoDB: Error: you must set the log group"
+			" home dir in my.cnf the\n"
+			"InnoDB: same as log arch dir.\n");
+
+		return(DB_ERROR);
+	}
+#endif /* UNIV_LOG_ARCHIVE */
+
+	if (srv_n_log_files * srv_log_file_size >= 262144) {
+		fprintf(stderr,
+			"InnoDB: Error: combined size of log files"
+			" must be < 4 GB\n");
+
+		return(DB_ERROR);
+	}
+
+	sum_of_new_sizes = 0;
+
+	for (i = 0; i < srv_n_data_files; i++) {
+#ifndef __WIN__
+		if (sizeof(off_t) < 5 && srv_data_file_sizes[i] >= 262144) {
+			fprintf(stderr,
+				"InnoDB: Error: file size must be < 4 GB"
+				" with this MySQL binary\n"
+				"InnoDB: and operating system combination,"
+				" in some OS's < 2 GB\n");
+
+			return(DB_ERROR);
+		}
+#endif
+		sum_of_new_sizes += srv_data_file_sizes[i];
+	}
+
+	if (sum_of_new_sizes < 640) {
+		fprintf(stderr,
+			"InnoDB: Error: tablespace size must be"
+			" at least 10 MB\n");
+
+		return(DB_ERROR);
+	}
+
+	err = open_or_create_data_files(&create_new_db,
+#ifdef UNIV_LOG_ARCHIVE
+					&min_arch_log_no, &max_arch_log_no,
+#endif /* UNIV_LOG_ARCHIVE */
+					&min_flushed_lsn, &max_flushed_lsn,
+					&sum_of_new_sizes);
+	if (err != DB_SUCCESS) {
+		fprintf(stderr,
+			"InnoDB: Could not open or create data files.\n"
+			"InnoDB: If you tried to add new data files,"
+			" and it failed here,\n"
+			"InnoDB: you should now edit innodb_data_file_path"
+			" in my.cnf back\n"
+			"InnoDB: to what it was, and remove the"
+			" new ibdata files InnoDB created\n"
+			"InnoDB: in this failed attempt. InnoDB only wrote"
+			" those files full of\n"
+			"InnoDB: zeros, but did not yet use them in any way."
+			" But be careful: do not\n"
+			"InnoDB: remove old data files"
+			" which contain your precious data!\n");
+
+		return((int) err);
+	}
+
+#ifdef UNIV_LOG_ARCHIVE
+	srv_normalize_path_for_win(srv_arch_dir);
+	srv_arch_dir = srv_add_path_separator_if_needed(srv_arch_dir);
+#endif /* UNIV_LOG_ARCHIVE */
+
+	for (i = 0; i < srv_n_log_files; i++) {
+		err = open_or_create_log_file(create_new_db, &log_file_created,
+					      log_opened, 0, i);
+		if (err != DB_SUCCESS) {
+
+			return((int) err);
+		}
+
+		if (log_file_created) {
+			log_created = TRUE;
+		} else {
+			log_opened = TRUE;
+		}
+		if ((log_opened && create_new_db)
+		    || (log_opened && log_created)) {
+			fprintf(stderr,
+				"InnoDB: Error: all log files must be"
+				" created at the same time.\n"
+				"InnoDB: All log files must be"
+				" created also in database creation.\n"
+				"InnoDB: If you want bigger or smaller"
+				" log files, shut down the\n"
+				"InnoDB: database and make sure there"
+				" were no errors in shutdown.\n"
+				"InnoDB: Then delete the existing log files."
+				" Edit the .cnf file\n"
+				"InnoDB: and start the database again.\n");
+
+			return(DB_ERROR);
+		}
+	}
+
+	/* Open all log files and data files in the system tablespace: we
+	keep them open until database shutdown */
+
+	fil_open_log_and_system_tablespace_files();
+
+	if (log_created && !create_new_db
+#ifdef UNIV_LOG_ARCHIVE
+	    && !srv_archive_recovery
+#endif /* UNIV_LOG_ARCHIVE */
+	    ) {
+		if (max_flushed_lsn != min_flushed_lsn
+#ifdef UNIV_LOG_ARCHIVE
+		    || max_arch_log_no != min_arch_log_no
+#endif /* UNIV_LOG_ARCHIVE */
+		    ) {
+			fprintf(stderr,
+				"InnoDB: Cannot initialize created"
+				" log files because\n"
+				"InnoDB: data files were not in sync"
+				" with each other\n"
+				"InnoDB: or the data files are corrupt.\n");
+
+			return(DB_ERROR);
+		}
+
+		if (max_flushed_lsn < (ib_uint64_t) 1000) {
+			fprintf(stderr,
+				"InnoDB: Cannot initialize created"
+				" log files because\n"
+				"InnoDB: data files are corrupt,"
+				" or new data files were\n"
+				"InnoDB: created when the database"
+				" was started previous\n"
+				"InnoDB: time but the database"
+				" was not shut down\n"
+				"InnoDB: normally after that.\n");
+
+			return(DB_ERROR);
+		}
+
+		mutex_enter(&(log_sys->mutex));
+
+#ifdef UNIV_LOG_ARCHIVE
+		/* Do not + 1 arch_log_no because we do not use log
+		archiving */
+		recv_reset_logs(max_flushed_lsn, max_arch_log_no, TRUE);
+#else
+		recv_reset_logs(max_flushed_lsn, TRUE);
+#endif /* UNIV_LOG_ARCHIVE */
+
+		mutex_exit(&(log_sys->mutex));
+	}
+
+	trx_sys_file_format_init();
+
+	if (create_new_db) {
+		mtr_start(&mtr);
+		fsp_header_init(0, sum_of_new_sizes, &mtr);
+
+		mtr_commit(&mtr);
+
+		trx_sys_create();
+		dict_create();
+		srv_startup_is_before_trx_rollback_phase = FALSE;
+
+		if (srv_extra_rsegments)
+			trx_sys_create_extra_rseg(srv_extra_rsegments);
+#ifdef UNIV_LOG_ARCHIVE
+	} else if (srv_archive_recovery) {
+		fprintf(stderr,
+			"InnoDB: Starting archive"
+			" recovery from a backup...\n");
+		err = recv_recovery_from_archive_start(
+			min_flushed_lsn, srv_archive_recovery_limit_lsn,
+			min_arch_log_no);
+		if (err != DB_SUCCESS) {
+
+			return(DB_ERROR);
+		}
+		/* Since ibuf init is in dict_boot, and ibuf is needed
+		in any disk i/o, first call dict_boot */
+
+		dict_boot();
+		trx_sys_init_at_db_start();
+		srv_startup_is_before_trx_rollback_phase = FALSE;
+
+		/* Initialize the fsp free limit global variable in the log
+		system */
+		fsp_header_get_free_limit();
+
+		recv_recovery_from_archive_finish();
+#endif /* UNIV_LOG_ARCHIVE */
+	} else {
+
+		/* Check if we support the max format that is stamped
+		on the system tablespace. 
+		Note:  We are NOT allowed to make any modifications to
+		the TRX_SYS_PAGE_NO page before recovery  because this
+		page also contains the max_trx_id etc. important system
+		variables that are required for recovery.  We need to
+		ensure that we return the system to a state where normal
+		recovery is guaranteed to work. We do this by
+		invalidating the buffer cache, this will force the
+		reread of the page and restoration to its last known
+		consistent state, this is REQUIRED for the recovery
+		process to work. */
+		err = trx_sys_file_format_max_check(
+			srv_check_file_format_at_startup);
+
+		if (err != DB_SUCCESS) {
+			return(err);
+		}
+
+		/* Invalidate the buffer pool to ensure that we reread
+		the page that we read above, during recovery.
+		Note that this is not as heavy weight as it seems. At
+		this point there will be only ONE page in the buf_LRU
+		and there must be no page in the buf_flush list. */
+		buf_pool_invalidate();
+
+		/* We always try to do a recovery, even if the database had
+		been shut down normally: this is the normal startup path */
+
+		err = recv_recovery_from_checkpoint_start(LOG_CHECKPOINT,
+							  IB_ULONGLONG_MAX,
+							  min_flushed_lsn,
+							  max_flushed_lsn);
+		if (err != DB_SUCCESS) {
+
+			return(DB_ERROR);
+		}
+
+		/* Since the insert buffer init is in dict_boot, and the
+		insert buffer is needed in any disk i/o, first we call
+		dict_boot(). Note that trx_sys_init_at_db_start() only needs
+		to access space 0, and the insert buffer at this stage already
+		works for space 0. */
+
+		dict_boot();
+		trx_sys_init_at_db_start();
+
+		if (srv_force_recovery < SRV_FORCE_NO_IBUF_MERGE) {
+			/* The following call is necessary for the insert
+			buffer to work with multiple tablespaces. We must
+			know the mapping between space id's and .ibd file
+			names.
+
+			In a crash recovery, we check that the info in data
+			dictionary is consistent with what we already know
+			about space id's from the call of
+			fil_load_single_table_tablespaces().
+
+			In a normal startup, we create the space objects for
+			every table in the InnoDB data dictionary that has
+			an .ibd file.
+
+			We also determine the maximum tablespace id used.
+
+			TODO: We may have incomplete transactions in the
+			data dictionary tables. Does that harm the scanning of
+			the data dictionary below? */
+
+			dict_check_tablespaces_and_store_max_id(
+				recv_needed_recovery);
+		}
+
+		srv_startup_is_before_trx_rollback_phase = FALSE;
+
+		/* Initialize the fsp free limit global variable in the log
+		system */
+		fsp_header_get_free_limit();
+
+		/* recv_recovery_from_checkpoint_finish needs trx lists which
+		are initialized in trx_sys_init_at_db_start(). */
+
+		recv_recovery_from_checkpoint_finish();
+
+		/* It is possible that file_format tag has never
+		been set. In this case we initialize it to minimum
+		value.  Important to note that we can do it ONLY after
+		we have finished the recovery process so that the
+		image of TRX_SYS_PAGE_NO is not stale. */
+		trx_sys_file_format_tag_init();
+	}
+
+	if (!create_new_db && sum_of_new_sizes > 0) {
+		/* New data file(s) were added */
+		mtr_start(&mtr);
+
+		fsp_header_inc_size(0, sum_of_new_sizes, &mtr);
+
+		mtr_commit(&mtr);
+
+		/* Immediately write the log record about increased tablespace
+		size to disk, so that it is durable even if mysqld would crash
+		quickly */
+
+		log_buffer_flush_to_disk();
+	}
+
+#ifdef UNIV_LOG_ARCHIVE
+	/* Archiving is always off under MySQL */
+	if (!srv_log_archive_on) {
+		ut_a(DB_SUCCESS == log_archive_noarchivelog());
+	} else {
+		mutex_enter(&(log_sys->mutex));
+
+		start_archive = FALSE;
+
+		if (log_sys->archiving_state == LOG_ARCH_OFF) {
+			start_archive = TRUE;
+		}
+
+		mutex_exit(&(log_sys->mutex));
+
+		if (start_archive) {
+			ut_a(DB_SUCCESS == log_archive_archivelog());
+		}
+	}
+#endif /* UNIV_LOG_ARCHIVE */
+
+	/* fprintf(stderr, "Max allowed record size %lu\n",
+	page_get_free_space_of_empty() / 2); */
+
+	/* Create the thread which watches the timeouts for lock waits
+	and prints InnoDB monitor info */
+
+	os_thread_create(&srv_lock_timeout_and_monitor_thread, NULL,
+			 thread_ids + 2 + SRV_MAX_N_IO_THREADS);
+
+	/* Create the thread which warns of long semaphore waits */
+	os_thread_create(&srv_error_monitor_thread, NULL,
+			 thread_ids + 3 + SRV_MAX_N_IO_THREADS);
+	srv_is_being_started = FALSE;
+
+	if (trx_doublewrite == NULL) {
+		/* Create the doublewrite buffer to a new tablespace */
+
+		trx_sys_create_doublewrite_buf();
+	}
+
+	err = dict_create_or_check_foreign_constraint_tables();
+
+	if (err != DB_SUCCESS) {
+		return((int)DB_ERROR);
+	}
+
+	/* Create the master thread which does purge and other utility
+	operations */
+
+	os_thread_create(&srv_master_thread, NULL, thread_ids
+			 + (1 + SRV_MAX_N_IO_THREADS));
+#ifdef UNIV_DEBUG
+	/* buf_debug_prints = TRUE; */
+#endif /* UNIV_DEBUG */
+	sum_of_data_file_sizes = 0;
+
+	for (i = 0; i < srv_n_data_files; i++) {
+		sum_of_data_file_sizes += srv_data_file_sizes[i];
+	}
+
+	tablespace_size_in_header = fsp_header_get_tablespace_size();
+
+	if (!srv_auto_extend_last_data_file
+	    && sum_of_data_file_sizes != tablespace_size_in_header) {
+
+		fprintf(stderr,
+			"InnoDB: Error: tablespace size"
+			" stored in header is %lu pages, but\n"
+			"InnoDB: the sum of data file sizes is %lu pages\n",
+			(ulong) tablespace_size_in_header,
+			(ulong) sum_of_data_file_sizes);
+
+		if (srv_force_recovery == 0
+		    && sum_of_data_file_sizes < tablespace_size_in_header) {
+			/* This is a fatal error, the tail of a tablespace is
+			missing */
+
+			fprintf(stderr,
+				"InnoDB: Cannot start InnoDB."
+				" The tail of the system tablespace is\n"
+				"InnoDB: missing. Have you edited"
+				" innodb_data_file_path in my.cnf in an\n"
+				"InnoDB: inappropriate way, removing"
+				" ibdata files from there?\n"
+				"InnoDB: You can set innodb_force_recovery=1"
+				" in my.cnf to force\n"
+				"InnoDB: a startup if you are trying"
+				" to recover a badly corrupt database.\n");
+
+			return(DB_ERROR);
+		}
+	}
+
+	if (srv_auto_extend_last_data_file
+	    && sum_of_data_file_sizes < tablespace_size_in_header) {
+
+		fprintf(stderr,
+			"InnoDB: Error: tablespace size stored in header"
+			" is %lu pages, but\n"
+			"InnoDB: the sum of data file sizes"
+			" is only %lu pages\n",
+			(ulong) tablespace_size_in_header,
+			(ulong) sum_of_data_file_sizes);
+
+		if (srv_force_recovery == 0) {
+
+			fprintf(stderr,
+				"InnoDB: Cannot start InnoDB. The tail of"
+				" the system tablespace is\n"
+				"InnoDB: missing. Have you edited"
+				" innodb_data_file_path in my.cnf in an\n"
+				"InnoDB: inappropriate way, removing"
+				" ibdata files from there?\n"
+				"InnoDB: You can set innodb_force_recovery=1"
+				" in my.cnf to force\n"
+				"InnoDB: a startup if you are trying to"
+				" recover a badly corrupt database.\n");
+
+			return(DB_ERROR);
+		}
+	}
+
+	/* Check that os_fast_mutexes work as expected */
+	os_fast_mutex_init(&srv_os_test_mutex);
+
+	if (0 != os_fast_mutex_trylock(&srv_os_test_mutex)) {
+		fprintf(stderr,
+			"InnoDB: Error: pthread_mutex_trylock returns"
+			" an unexpected value on\n"
+			"InnoDB: success! Cannot continue.\n");
+		exit(1);
+	}
+
+	os_fast_mutex_unlock(&srv_os_test_mutex);
+
+	os_fast_mutex_lock(&srv_os_test_mutex);
+
+	os_fast_mutex_unlock(&srv_os_test_mutex);
+
+	os_fast_mutex_free(&srv_os_test_mutex);
+
+	if (srv_print_verbose_log) {
+		ut_print_timestamp(stderr);
+		fprintf(stderr,
+			" InnoDB Plugin %s started; "
+			"log sequence number %llu\n",
+			INNODB_VERSION_STR, srv_start_lsn);
+	}
+
+	if (srv_force_recovery > 0) {
+		fprintf(stderr,
+			"InnoDB: !!! innodb_force_recovery"
+			" is set to %lu !!!\n",
+			(ulong) srv_force_recovery);
+	}
+
+	fflush(stderr);
+
+	if (trx_doublewrite_must_reset_space_ids) {
+		/* Actually, we did not change the undo log format between
+		4.0 and 4.1.1, and we would not need to run purge to
+		completion. Note also that the purge algorithm in 4.1.1
+		can process the the history list again even after a full
+		purge, because our algorithm does not cut the end of the
+		history list in all cases so that it would become empty
+		after a full purge. That mean that we may purge 4.0 type
+		undo log even after this phase.
+
+		The insert buffer record format changed between 4.0 and
+		4.1.1. It is essential that the insert buffer is emptied
+		here! */
+
+		fprintf(stderr,
+			"InnoDB: You are upgrading to an"
+			" InnoDB version which allows multiple\n"
+			"InnoDB: tablespaces. Wait that purge"
+			" and insert buffer merge run to\n"
+			"InnoDB: completion...\n");
+		for (;;) {
+			os_thread_sleep(1000000);
+
+			if (0 == strcmp(srv_main_thread_op_info,
+					"waiting for server activity")) {
+
+				ut_a(ibuf_is_empty());
+
+				break;
+			}
+		}
+		fprintf(stderr,
+			"InnoDB: Full purge and insert buffer merge"
+			" completed.\n");
+
+		trx_sys_mark_upgraded_to_multiple_tablespaces();
+
+		fprintf(stderr,
+			"InnoDB: You have now successfully upgraded"
+			" to the multiple tablespaces\n"
+			"InnoDB: format. You should NOT DOWNGRADE"
+			" to an earlier version of\n"
+			"InnoDB: InnoDB! But if you absolutely need to"
+			" downgrade, see\n"
+			"InnoDB: http://dev.mysql.com/doc/refman/5.1/en/"
+			"multiple-tablespaces.html\n"
+			"InnoDB: for instructions.\n");
+	}
+
+	if (srv_force_recovery == 0) {
+		/* In the insert buffer we may have even bigger tablespace
+		id's, because we may have dropped those tablespaces, but
+		insert buffer merge has not had time to clean the records from
+		the ibuf tree. */
+
+		ibuf_update_max_tablespace_id();
+	}
+
+	srv_file_per_table = srv_file_per_table_original_value;
+
+	srv_was_started = TRUE;
+
+	return((int) DB_SUCCESS);
+}
+
+/********************************************************************
+Shuts down the InnoDB database. */
+UNIV_INTERN
+int
+innobase_shutdown_for_mysql(void)
+/*=============================*/
+				/* out: DB_SUCCESS or error code */
+{
+	ulint	i;
+#ifdef __NETWARE__
+	extern ibool panic_shutdown;
+#endif
+	if (!srv_was_started) {
+		if (srv_is_being_started) {
+			ut_print_timestamp(stderr);
+			fprintf(stderr,
+				"  InnoDB: Warning: shutting down"
+				" a not properly started\n"
+				"InnoDB: or created database!\n");
+		}
+
+		return(DB_SUCCESS);
+	}
+
+	/* 1. Flush the buffer pool to disk, write the current lsn to
+	the tablespace header(s), and copy all log data to archive.
+	The step 1 is the real InnoDB shutdown. The remaining steps 2 - ...
+	just free data structures after the shutdown. */
+
+
+	if (srv_fast_shutdown == 2) {
+		ut_print_timestamp(stderr);
+		fprintf(stderr,
+			"  InnoDB: MySQL has requested a very fast shutdown"
+			" without flushing "
+			"the InnoDB buffer pool to data files."
+			" At the next mysqld startup "
+			"InnoDB will do a crash recovery!\n");
+	}
+
+#ifdef __NETWARE__
+	if(!panic_shutdown)
+#endif
+		logs_empty_and_mark_files_at_shutdown();
+
+	if (srv_conc_n_threads != 0) {
+		fprintf(stderr,
+			"InnoDB: Warning: query counter shows %ld queries"
+			" still\n"
+			"InnoDB: inside InnoDB at shutdown\n",
+			srv_conc_n_threads);
+	}
+
+	/* 2. Make all threads created by InnoDB to exit */
+
+	srv_shutdown_state = SRV_SHUTDOWN_EXIT_THREADS;
+
+	/* In a 'very fast' shutdown, we do not need to wait for these threads
+	to die; all which counts is that we flushed the log; a 'very fast'
+	shutdown is essentially a crash. */
+
+	if (srv_fast_shutdown == 2) {
+		return(DB_SUCCESS);
+	}
+
+	/* All threads end up waiting for certain events. Put those events
+	to the signaled state. Then the threads will exit themselves in
+	os_thread_event_wait(). */
+
+	for (i = 0; i < 1000; i++) {
+		/* NOTE: IF YOU CREATE THREADS IN INNODB, YOU MUST EXIT THEM
+		HERE OR EARLIER */
+
+		/* a. Let the lock timeout thread exit */
+		os_event_set(srv_lock_timeout_thread_event);
+
+		/* b. srv error monitor thread exits automatically, no need
+		to do anything here */
+
+		/* c. We wake the master thread so that it exits */
+		srv_wake_master_thread();
+
+		/* d. Exit the i/o threads */
+
+		os_aio_wake_all_threads_at_shutdown();
+
+		os_mutex_enter(os_sync_mutex);
+
+		if (os_thread_count == 0) {
+			/* All the threads have exited or are just exiting;
+			NOTE that the threads may not have completed their
+			exit yet. Should we use pthread_join() to make sure
+			they have exited? Now we just sleep 0.1 seconds and
+			hope that is enough! */
+
+			os_mutex_exit(os_sync_mutex);
+
+			os_thread_sleep(100000);
+
+			break;
+		}
+
+		os_mutex_exit(os_sync_mutex);
+
+		os_thread_sleep(100000);
+	}
+
+	if (i == 1000) {
+		fprintf(stderr,
+			"InnoDB: Warning: %lu threads created by InnoDB"
+			" had not exited at shutdown!\n",
+			(ulong) os_thread_count);
+	}
+
+	if (srv_monitor_file) {
+		fclose(srv_monitor_file);
+		srv_monitor_file = 0;
+		if (srv_monitor_file_name) {
+			unlink(srv_monitor_file_name);
+			mem_free(srv_monitor_file_name);
+		}
+	}
+	if (srv_dict_tmpfile) {
+		fclose(srv_dict_tmpfile);
+		srv_dict_tmpfile = 0;
+	}
+
+	if (srv_misc_tmpfile) {
+		fclose(srv_misc_tmpfile);
+		srv_misc_tmpfile = 0;
+	}
+
+	trx_sys_file_format_close();
+
+	mutex_free(&srv_monitor_file_mutex);
+	mutex_free(&srv_dict_tmpfile_mutex);
+	mutex_free(&srv_misc_tmpfile_mutex);
+
+	/* 3. Free all InnoDB's own mutexes and the os_fast_mutexes inside
+	them */
+	sync_close();
+
+	/* 4. Free the os_conc_mutex and all os_events and os_mutexes */
+
+	srv_free();
+	os_sync_free();
+
+	/* Check that all read views are closed except read view owned
+	by a purge. */
+
+	if (UT_LIST_GET_LEN(trx_sys->view_list) > 1) {
+		fprintf(stderr,
+			"InnoDB: Error: all read views were not closed"
+			" before shutdown:\n"
+			"InnoDB: %lu read views open \n",
+			UT_LIST_GET_LEN(trx_sys->view_list) - 1);
+	}
+
+	/* 5. Free all allocated memory and the os_fast_mutex created in
+	ut0mem.c */
+
+	buf_pool_free();
+	ut_free_all_mem();
+
+	if (os_thread_count != 0
+	    || os_event_count != 0
+	    || os_mutex_count != 0
+	    || os_fast_mutex_count != 0) {
+		fprintf(stderr,
+			"InnoDB: Warning: some resources were not"
+			" cleaned up in shutdown:\n"
+			"InnoDB: threads %lu, events %lu,"
+			" os_mutexes %lu, os_fast_mutexes %lu\n",
+			(ulong) os_thread_count, (ulong) os_event_count,
+			(ulong) os_mutex_count, (ulong) os_fast_mutex_count);
+	}
+
+	if (dict_foreign_err_file) {
+		fclose(dict_foreign_err_file);
+	}
+	if (lock_latest_err_file) {
+		fclose(lock_latest_err_file);
+	}
+
+	if (srv_print_verbose_log) {
+		ut_print_timestamp(stderr);
+		fprintf(stderr,
+			"  InnoDB: Shutdown completed;"
+			" log sequence number %llu\n",
+			srv_shutdown_lsn);
+	}
+
+	srv_was_started = FALSE;
+
+	return((int) DB_SUCCESS);
+}
+
+#ifdef __NETWARE__
+void set_panic_flag_for_netware()
+{
+	extern ibool panic_shutdown;
+	panic_shutdown = TRUE;
+}
+#endif /* __NETWARE__ */
+#endif /* !UNIV_HOTBACKUP */
diff --git a/storage/xtradb/sync/sync0arr.c b/storage/xtradb/sync/sync0arr.c
new file mode 100644
index 00000000000..62165eefd46
--- /dev/null
+++ b/storage/xtradb/sync/sync0arr.c
@@ -0,0 +1,1029 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+Copyright (c) 2008, Google Inc.
+
+Portions of this file contain modifications contributed and copyrighted by
+Google, Inc. Those modifications are gratefully acknowledged and are described
+briefly in the InnoDB documentation. The contributions by Google are
+incorporated with their permission, and subject to the conditions contained in
+the file COPYING.Google.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The wait array used in synchronization primitives
+
+Created 9/5/1995 Heikki Tuuri
+*******************************************************/
+
+#include "sync0arr.h"
+#ifdef UNIV_NONINL
+#include "sync0arr.ic"
+#endif
+
+#include "sync0sync.h"
+#include "sync0rw.h"
+#include "os0sync.h"
+#include "os0file.h"
+#include "srv0srv.h"
+
+/*
+			WAIT ARRAY
+			==========
+
+The wait array consists of cells each of which has an
+an operating system event object created for it. The threads
+waiting for a mutex, for example, can reserve a cell
+in the array and suspend themselves to wait for the event
+to become signaled. When using the wait array, remember to make
+sure that some thread holding the synchronization object
+will eventually know that there is a waiter in the array and
+signal the object, to prevent infinite wait.
+Why we chose to implement a wait array? First, to make
+mutexes fast, we had to code our own implementation of them,
+which only in usually uncommon cases resorts to using
+slow operating system primitives. Then we had the choice of
+assigning a unique OS event for each mutex, which would
+be simpler, or using a global wait array. In some operating systems,
+the global wait array solution is more efficient and flexible,
+because we can do with a very small number of OS events,
+say 200. In NT 3.51, allocating events seems to be a quadratic
+algorithm, because 10 000 events are created fast, but
+100 000 events takes a couple of minutes to create.
+
+As of 5.0.30 the above mentioned design is changed. Since now
+OS can handle millions of wait events efficiently, we no longer
+have this concept of each cell of wait array having one event.
+Instead, now the event that a thread wants to wait on is embedded
+in the wait object (mutex or rw_lock). We still keep the global
+wait array for the sake of diagnostics and also to avoid infinite
+wait The error_monitor thread scans the global wait array to signal
+any waiting threads who have missed the signal. */
+
+/* A cell where an individual thread may wait suspended
+until a resource is released. The suspending is implemented
+using an operating system event semaphore. */
+struct sync_cell_struct {
+	void*		wait_object;	/* pointer to the object the
+					thread is waiting for; if NULL
+					the cell is free for use */
+	mutex_t*	old_wait_mutex;	/* the latest wait mutex in cell */
+	rw_lock_t*	old_wait_rw_lock;/* the latest wait rw-lock in cell */
+	ulint		request_type;	/* lock type requested on the
+					object */
+	const char*	file;		/* in debug version file where
+					requested */
+	ulint		line;		/* in debug version line where
+					requested */
+	os_thread_id_t	thread;		/* thread id of this waiting
+					thread */
+	ibool		waiting;	/* TRUE if the thread has already
+					called sync_array_event_wait
+					on this cell */
+	ib_int64_t	signal_count;	/* We capture the signal_count
+					of the wait_object when we
+					reset the event. This value is
+					then passed on to os_event_wait
+					and we wait only if the event
+					has not been signalled in the
+					period between the reset and
+					wait call. */
+	time_t		reservation_time;/* time when the thread reserved
+					the wait cell */
+};
+
+/* NOTE: It is allowed for a thread to wait
+for an event allocated for the array without owning the
+protecting mutex (depending on the case: OS or database mutex), but
+all changes (set or reset) to the state of the event must be made
+while owning the mutex. */
+struct sync_array_struct {
+	ulint		n_reserved;	/* number of currently reserved
+					cells in the wait array */
+	ulint		n_cells;	/* number of cells in the
+					wait array */
+	sync_cell_t*	array;		/* pointer to wait array */
+	ulint		protection;	/* this flag tells which
+					mutex protects the data */
+	mutex_t		mutex;		/* possible database mutex
+					protecting this data structure */
+	os_mutex_t	os_mutex;	/* Possible operating system mutex
+					protecting the data structure.
+					As this data structure is used in
+					constructing the database mutex,
+					to prevent infinite recursion
+					in implementation, we fall back to
+					an OS mutex. */
+	ulint		sg_count;	/* count of how many times an
+					object has been signalled */
+	ulint		res_count;	/* count of cell reservations
+					since creation of the array */
+};
+
+#ifdef UNIV_SYNC_DEBUG
+/**********************************************************************
+This function is called only in the debug version. Detects a deadlock
+of one or more threads because of waits of semaphores. */
+static
+ibool
+sync_array_detect_deadlock(
+/*=======================*/
+				/* out: TRUE if deadlock detected */
+	sync_array_t*	arr,	/* in: wait array; NOTE! the caller must
+				own the mutex to array */
+	sync_cell_t*	start,	/* in: cell where recursive search started */
+	sync_cell_t*	cell,	/* in: cell to search */
+	ulint		depth);	/* in: recursion depth */
+#endif /* UNIV_SYNC_DEBUG */
+
+/*********************************************************************
+Gets the nth cell in array. */
+static
+sync_cell_t*
+sync_array_get_nth_cell(
+/*====================*/
+				/* out: cell */
+	sync_array_t*	arr,	/* in: sync array */
+	ulint		n)	/* in: index */
+{
+	ut_a(arr);
+	ut_a(n < arr->n_cells);
+
+	return(arr->array + n);
+}
+
+/**********************************************************************
+Reserves the mutex semaphore protecting a sync array. */
+static
+void
+sync_array_enter(
+/*=============*/
+	sync_array_t*	arr)	/* in: sync wait array */
+{
+	ulint	protection;
+
+	protection = arr->protection;
+
+	if (protection == SYNC_ARRAY_OS_MUTEX) {
+		os_mutex_enter(arr->os_mutex);
+	} else if (protection == SYNC_ARRAY_MUTEX) {
+		mutex_enter(&(arr->mutex));
+	} else {
+		ut_error;
+	}
+}
+
+/**********************************************************************
+Releases the mutex semaphore protecting a sync array. */
+static
+void
+sync_array_exit(
+/*============*/
+	sync_array_t*	arr)	/* in: sync wait array */
+{
+	ulint	protection;
+
+	protection = arr->protection;
+
+	if (protection == SYNC_ARRAY_OS_MUTEX) {
+		os_mutex_exit(arr->os_mutex);
+	} else if (protection == SYNC_ARRAY_MUTEX) {
+		mutex_exit(&(arr->mutex));
+	} else {
+		ut_error;
+	}
+}
+
+/***********************************************************************
+Creates a synchronization wait array. It is protected by a mutex
+which is automatically reserved when the functions operating on it
+are called. */
+UNIV_INTERN
+sync_array_t*
+sync_array_create(
+/*==============*/
+				/* out, own: created wait array */
+	ulint	n_cells,	/* in: number of cells in the array
+				to create */
+	ulint	protection)	/* in: either SYNC_ARRAY_OS_MUTEX or
+				SYNC_ARRAY_MUTEX: determines the type
+				of mutex protecting the data structure */
+{
+	sync_array_t*	arr;
+	sync_cell_t*	cell_array;
+	sync_cell_t*	cell;
+	ulint		i;
+
+	ut_a(n_cells > 0);
+
+	/* Allocate memory for the data structures */
+	arr = ut_malloc(sizeof(sync_array_t));
+
+	cell_array = ut_malloc(sizeof(sync_cell_t) * n_cells);
+
+	arr->n_cells = n_cells;
+	arr->n_reserved = 0;
+	arr->array = cell_array;
+	arr->protection = protection;
+	arr->sg_count = 0;
+	arr->res_count = 0;
+
+	/* Then create the mutex to protect the wait array complex */
+	if (protection == SYNC_ARRAY_OS_MUTEX) {
+		arr->os_mutex = os_mutex_create(NULL);
+	} else if (protection == SYNC_ARRAY_MUTEX) {
+		mutex_create(&arr->mutex, SYNC_NO_ORDER_CHECK);
+	} else {
+		ut_error;
+	}
+
+	for (i = 0; i < n_cells; i++) {
+		cell = sync_array_get_nth_cell(arr, i);
+	cell->wait_object = NULL;
+		cell->waiting = FALSE;
+		cell->signal_count = 0;
+	}
+
+	return(arr);
+}
+
+/**********************************************************************
+Frees the resources in a wait array. */
+UNIV_INTERN
+void
+sync_array_free(
+/*============*/
+	sync_array_t*	arr)	/* in, own: sync wait array */
+{
+	ulint		protection;
+
+	ut_a(arr->n_reserved == 0);
+
+	sync_array_validate(arr);
+
+	protection = arr->protection;
+
+	/* Release the mutex protecting the wait array complex */
+
+	if (protection == SYNC_ARRAY_OS_MUTEX) {
+		os_mutex_free(arr->os_mutex);
+	} else if (protection == SYNC_ARRAY_MUTEX) {
+		mutex_free(&(arr->mutex));
+	} else {
+		ut_error;
+	}
+
+	ut_free(arr->array);
+	ut_free(arr);
+}
+
+/************************************************************************
+Validates the integrity of the wait array. Checks
+that the number of reserved cells equals the count variable. */
+UNIV_INTERN
+void
+sync_array_validate(
+/*================*/
+	sync_array_t*	arr)	/* in: sync wait array */
+{
+	ulint		i;
+	sync_cell_t*	cell;
+	ulint		count		= 0;
+
+	sync_array_enter(arr);
+
+	for (i = 0; i < arr->n_cells; i++) {
+		cell = sync_array_get_nth_cell(arr, i);
+		if (cell->wait_object != NULL) {
+			count++;
+		}
+	}
+
+	ut_a(count == arr->n_reserved);
+
+	sync_array_exit(arr);
+}
+
+/***********************************************************************
+Returns the event that the thread owning the cell waits for. */
+static
+os_event_t
+sync_cell_get_event(
+/*================*/
+	sync_cell_t*	cell) /* in: non-empty sync array cell */
+{
+	ulint type = cell->request_type;
+
+	if (type == SYNC_MUTEX) {
+		return(((mutex_t *) cell->wait_object)->event);
+	} else if (type == RW_LOCK_WAIT_EX) {
+		return(((rw_lock_t *) cell->wait_object)->wait_ex_event);
+	} else { /* RW_LOCK_SHARED and RW_LOCK_EX wait on the same event */
+		return(((rw_lock_t *) cell->wait_object)->event);
+	}
+}
+
+/**********************************************************************
+Reserves a wait array cell for waiting for an object.
+The event of the cell is reset to nonsignalled state. */
+UNIV_INTERN
+void
+sync_array_reserve_cell(
+/*====================*/
+	sync_array_t*	arr,	/* in: wait array */
+	void*		object, /* in: pointer to the object to wait for */
+	ulint		type,	/* in: lock request type */
+	const char*	file,	/* in: file where requested */
+	ulint		line,	/* in: line where requested */
+	ulint*		index)	/* out: index of the reserved cell */
+{
+	sync_cell_t*	cell;
+	os_event_t      event;
+	ulint		i;
+
+	ut_a(object);
+	ut_a(index);
+
+	sync_array_enter(arr);
+
+	arr->res_count++;
+
+	/* Reserve a new cell. */
+	for (i = 0; i < arr->n_cells; i++) {
+		cell = sync_array_get_nth_cell(arr, i);
+
+		if (cell->wait_object == NULL) {
+
+			cell->waiting = FALSE;
+			cell->wait_object = object;
+
+			if (type == SYNC_MUTEX) {
+				cell->old_wait_mutex = object;
+			} else {
+				cell->old_wait_rw_lock = object;
+			}
+
+			cell->request_type = type;
+
+			cell->file = file;
+			cell->line = line;
+
+			arr->n_reserved++;
+
+			*index = i;
+
+			sync_array_exit(arr);
+
+			/* Make sure the event is reset and also store
+			the value of signal_count at which the event
+			was reset. */
+                        event = sync_cell_get_event(cell);
+			cell->signal_count = os_event_reset(event);
+
+			cell->reservation_time = time(NULL);
+
+			cell->thread = os_thread_get_curr_id();
+
+			return;
+		}
+	}
+
+	ut_error; /* No free cell found */
+
+	return;
+}
+
+/**********************************************************************
+This function should be called when a thread starts to wait on
+a wait array cell. In the debug version this function checks
+if the wait for a semaphore will result in a deadlock, in which
+case prints info and asserts. */
+UNIV_INTERN
+void
+sync_array_wait_event(
+/*==================*/
+	sync_array_t*	arr,	/* in: wait array */
+	ulint		index)	/* in: index of the reserved cell */
+{
+	sync_cell_t*	cell;
+	os_event_t	event;
+
+	ut_a(arr);
+
+	sync_array_enter(arr);
+
+	cell = sync_array_get_nth_cell(arr, index);
+
+	ut_a(cell->wait_object);
+	ut_a(!cell->waiting);
+	ut_ad(os_thread_get_curr_id() == cell->thread);
+
+	event = sync_cell_get_event(cell);
+		cell->waiting = TRUE;
+
+#ifdef UNIV_SYNC_DEBUG
+
+	/* We use simple enter to the mutex below, because if
+	we cannot acquire it at once, mutex_enter would call
+	recursively sync_array routines, leading to trouble.
+	rw_lock_debug_mutex freezes the debug lists. */
+
+	rw_lock_debug_mutex_enter();
+
+	if (TRUE == sync_array_detect_deadlock(arr, cell, cell, 0)) {
+
+		fputs("########################################\n", stderr);
+		ut_error;
+	}
+
+	rw_lock_debug_mutex_exit();
+#endif
+	sync_array_exit(arr);
+
+	os_event_wait_low(event, cell->signal_count);
+
+	sync_array_free_cell(arr, index);
+}
+
+/**********************************************************************
+Reports info of a wait array cell. */
+static
+void
+sync_array_cell_print(
+/*==================*/
+	FILE*		file,	/* in: file where to print */
+	sync_cell_t*	cell)	/* in: sync cell */
+{
+	mutex_t*	mutex;
+	rw_lock_t*	rwlock;
+	ulint		type;
+	ulint		writer;
+
+	type = cell->request_type;
+
+	fprintf(file,
+		"--Thread %lu has waited at %s line %lu"
+		" for %#.5g seconds the semaphore:\n",
+		(ulong) os_thread_pf(cell->thread), cell->file,
+		(ulong) cell->line,
+		difftime(time(NULL), cell->reservation_time));
+
+	if (type == SYNC_MUTEX) {
+		/* We use old_wait_mutex in case the cell has already
+		been freed meanwhile */
+		mutex = cell->old_wait_mutex;
+
+		fprintf(file,
+			"Mutex at %p created file %s line %lu, lock var %lu\n"
+#ifdef UNIV_SYNC_DEBUG
+			"Last time reserved in file %s line %lu, "
+#endif /* UNIV_SYNC_DEBUG */
+			"waiters flag %lu\n",
+			(void*) mutex, mutex->cfile_name, (ulong) mutex->cline,
+			(ulong) mutex->lock_word,
+#ifdef UNIV_SYNC_DEBUG
+			mutex->file_name, (ulong) mutex->line,
+#endif /* UNIV_SYNC_DEBUG */
+			(ulong) mutex->waiters);
+
+	} else if (type == RW_LOCK_EX
+		   || type == RW_LOCK_WAIT_EX
+		   || type == RW_LOCK_SHARED) {
+
+		fputs(type == RW_LOCK_EX ? "X-lock on" : "S-lock on", file);
+
+		rwlock = cell->old_wait_rw_lock;
+
+		fprintf(file,
+			" RW-latch at %p created in file %s line %lu\n",
+			(void*) rwlock, rwlock->cfile_name,
+			(ulong) rwlock->cline);
+		writer = rw_lock_get_writer(rwlock);
+		if (writer != RW_LOCK_NOT_LOCKED) {
+			fprintf(file,
+				"a writer (thread id %lu) has"
+				" reserved it in mode %s",
+				(ulong) os_thread_pf(rwlock->writer_thread),
+				writer == RW_LOCK_EX
+				? " exclusive\n"
+				: " wait exclusive\n");
+		}
+
+		fprintf(file,
+			"number of readers %lu, waiters flag %lu, "
+                        "lock_word: %lx\n"
+			"Last time read locked in file %s line %lu\n"
+			"Last time write locked in file %s line %lu\n",
+			(ulong) rw_lock_get_reader_count(rwlock),
+			(ulong) rwlock->waiters,
+			rwlock->lock_word,
+			rwlock->last_s_file_name,
+			(ulong) rwlock->last_s_line,
+			rwlock->last_x_file_name,
+			(ulong) rwlock->last_x_line);
+	} else {
+		ut_error;
+	}
+
+	if (!cell->waiting) {
+		fputs("wait has ended\n", file);
+	}
+}
+
+#ifdef UNIV_SYNC_DEBUG
+/**********************************************************************
+Looks for a cell with the given thread id. */
+static
+sync_cell_t*
+sync_array_find_thread(
+/*===================*/
+				/* out: pointer to cell or NULL
+				if not found */
+	sync_array_t*	arr,	/* in: wait array */
+	os_thread_id_t	thread)	/* in: thread id */
+{
+	ulint		i;
+	sync_cell_t*	cell;
+
+	for (i = 0; i < arr->n_cells; i++) {
+
+		cell = sync_array_get_nth_cell(arr, i);
+
+		if (cell->wait_object != NULL
+		    && os_thread_eq(cell->thread, thread)) {
+
+			return(cell);	/* Found */
+		}
+	}
+
+	return(NULL);	/* Not found */
+}
+
+/**********************************************************************
+Recursion step for deadlock detection. */
+static
+ibool
+sync_array_deadlock_step(
+/*=====================*/
+				/* out: TRUE if deadlock detected */
+	sync_array_t*	arr,	/* in: wait array; NOTE! the caller must
+				own the mutex to array */
+	sync_cell_t*	start,	/* in: cell where recursive search
+				started */
+	os_thread_id_t	thread,	/* in: thread to look at */
+	ulint		pass,	/* in: pass value */
+	ulint		depth)	/* in: recursion depth */
+{
+	sync_cell_t*	new;
+	ibool		ret;
+
+	depth++;
+
+	if (pass != 0) {
+		/* If pass != 0, then we do not know which threads are
+		responsible of releasing the lock, and no deadlock can
+		be detected. */
+
+		return(FALSE);
+	}
+
+	new = sync_array_find_thread(arr, thread);
+
+	if (new == start) {
+		/* Stop running of other threads */
+
+		ut_dbg_stop_threads = TRUE;
+
+		/* Deadlock */
+		fputs("########################################\n"
+		      "DEADLOCK of threads detected!\n", stderr);
+
+		return(TRUE);
+
+	} else if (new) {
+		ret = sync_array_detect_deadlock(arr, start, new, depth);
+
+		if (ret) {
+			return(TRUE);
+		}
+	}
+	return(FALSE);
+}
+
+/**********************************************************************
+This function is called only in the debug version. Detects a deadlock
+of one or more threads because of waits of semaphores. */
+static
+ibool
+sync_array_detect_deadlock(
+/*=======================*/
+				/* out: TRUE if deadlock detected */
+	sync_array_t*	arr,	/* in: wait array; NOTE! the caller must
+				own the mutex to array */
+	sync_cell_t*	start,	/* in: cell where recursive search started */
+	sync_cell_t*	cell,	/* in: cell to search */
+	ulint		depth)	/* in: recursion depth */
+{
+	mutex_t*	mutex;
+	rw_lock_t*	lock;
+	os_thread_id_t	thread;
+	ibool		ret;
+	rw_lock_debug_t*debug;
+
+	ut_a(arr);
+	ut_a(start);
+	ut_a(cell);
+	ut_ad(cell->wait_object);
+	ut_ad(os_thread_get_curr_id() == start->thread);
+	ut_ad(depth < 100);
+
+	depth++;
+
+	if (!cell->waiting) {
+
+		return(FALSE); /* No deadlock here */
+	}
+
+	if (cell->request_type == SYNC_MUTEX) {
+
+		mutex = cell->wait_object;
+
+		if (mutex_get_lock_word(mutex) != 0) {
+
+			thread = mutex->thread_id;
+
+			/* Note that mutex->thread_id above may be
+			also OS_THREAD_ID_UNDEFINED, because the
+			thread which held the mutex maybe has not
+			yet updated the value, or it has already
+			released the mutex: in this case no deadlock
+			can occur, as the wait array cannot contain
+			a thread with ID_UNDEFINED value. */
+
+			ret = sync_array_deadlock_step(arr, start, thread, 0,
+						       depth);
+			if (ret) {
+				fprintf(stderr,
+			"Mutex %p owned by thread %lu file %s line %lu\n",
+					mutex, (ulong) os_thread_pf(mutex->thread_id),
+					mutex->file_name, (ulong) mutex->line);
+				sync_array_cell_print(stderr, cell);
+
+				return(TRUE);
+			}
+		}
+
+		return(FALSE); /* No deadlock */
+
+	} else if (cell->request_type == RW_LOCK_EX
+		   || cell->request_type == RW_LOCK_WAIT_EX) {
+
+		lock = cell->wait_object;
+
+		debug = UT_LIST_GET_FIRST(lock->debug_list);
+
+		while (debug != NULL) {
+
+			thread = debug->thread_id;
+
+			if (((debug->lock_type == RW_LOCK_EX)
+			     && !os_thread_eq(thread, cell->thread))
+			    || ((debug->lock_type == RW_LOCK_WAIT_EX)
+				&& !os_thread_eq(thread, cell->thread))
+			    || (debug->lock_type == RW_LOCK_SHARED)) {
+
+				/* The (wait) x-lock request can block
+				infinitely only if someone (can be also cell
+				thread) is holding s-lock, or someone
+				(cannot be cell thread) (wait) x-lock, and
+				he is blocked by start thread */
+
+				ret = sync_array_deadlock_step(
+					arr, start, thread, debug->pass,
+					depth);
+				if (ret) {
+print:
+					fprintf(stderr, "rw-lock %p ",
+						(void*) lock);
+					sync_array_cell_print(stderr, cell);
+					rw_lock_debug_print(debug);
+					return(TRUE);
+				}
+			}
+
+			debug = UT_LIST_GET_NEXT(list, debug);
+		}
+
+		return(FALSE);
+
+	} else if (cell->request_type == RW_LOCK_SHARED) {
+
+		lock = cell->wait_object;
+		debug = UT_LIST_GET_FIRST(lock->debug_list);
+
+		while (debug != NULL) {
+
+			thread = debug->thread_id;
+
+			if ((debug->lock_type == RW_LOCK_EX)
+			    || (debug->lock_type == RW_LOCK_WAIT_EX)) {
+
+				/* The s-lock request can block infinitely
+				only if someone (can also be cell thread) is
+				holding (wait) x-lock, and he is blocked by
+				start thread */
+
+				ret = sync_array_deadlock_step(
+					arr, start, thread, debug->pass,
+					depth);
+				if (ret) {
+					goto print;
+				}
+			}
+
+			debug = UT_LIST_GET_NEXT(list, debug);
+		}
+
+		return(FALSE);
+
+	} else {
+		ut_error;
+	}
+
+	return(TRUE);	/* Execution never reaches this line: for compiler
+			fooling only */
+}
+#endif /* UNIV_SYNC_DEBUG */
+
+/**********************************************************************
+Determines if we can wake up the thread waiting for a sempahore. */
+static
+ibool
+sync_arr_cell_can_wake_up(
+/*======================*/
+	sync_cell_t*	cell)	/* in: cell to search */
+{
+	mutex_t*	mutex;
+	rw_lock_t*	lock;
+
+	if (cell->request_type == SYNC_MUTEX) {
+
+		mutex = cell->wait_object;
+
+		if (mutex_get_lock_word(mutex) == 0) {
+
+			return(TRUE);
+		}
+
+	} else if (cell->request_type == RW_LOCK_EX) {
+
+		lock = cell->wait_object;
+
+		if (lock->lock_word > 0) {
+		/* Either unlocked or only read locked. */
+
+			return(TRUE);
+		}
+
+        } else if (cell->request_type == RW_LOCK_WAIT_EX) {
+
+		lock = cell->wait_object;
+
+                /* lock_word == 0 means all readers have left */
+		if (lock->lock_word == 0) {
+
+			return(TRUE);
+		}
+	} else if (cell->request_type == RW_LOCK_SHARED) {
+		lock = cell->wait_object;
+
+                /* lock_word > 0 means no writer or reserved writer */
+		if (lock->lock_word > 0) {
+
+			return(TRUE);
+		}
+	}
+
+	return(FALSE);
+}
+
+/**********************************************************************
+Frees the cell. NOTE! sync_array_wait_event frees the cell
+automatically! */
+UNIV_INTERN
+void
+sync_array_free_cell(
+/*=================*/
+	sync_array_t*	arr,	/* in: wait array */
+	ulint		index)  /* in: index of the cell in array */
+{
+	sync_cell_t*	cell;
+
+	sync_array_enter(arr);
+
+	cell = sync_array_get_nth_cell(arr, index);
+
+	ut_a(cell->wait_object != NULL);
+
+	cell->waiting = FALSE;
+	cell->wait_object =  NULL;
+	cell->signal_count = 0;
+
+	ut_a(arr->n_reserved > 0);
+	arr->n_reserved--;
+
+	sync_array_exit(arr);
+}
+
+/**************************************************************************
+Increments the signalled count. */
+UNIV_INTERN
+void
+sync_array_object_signalled(
+/*========================*/
+	sync_array_t*	arr)	/* in: wait array */
+{
+#ifdef HAVE_GCC_ATOMIC_BUILTINS
+	(void) os_atomic_increment(&arr->sg_count, 1);
+#else
+	sync_array_enter(arr);
+
+	arr->sg_count++;
+
+	sync_array_exit(arr);
+#endif
+}
+
+/**************************************************************************
+If the wakeup algorithm does not work perfectly at semaphore relases,
+this function will do the waking (see the comment in mutex_exit). This
+function should be called about every 1 second in the server.
+
+Note that there's a race condition between this thread and mutex_exit
+changing the lock_word and calling signal_object, so sometimes this finds
+threads to wake up even when nothing has gone wrong. */
+UNIV_INTERN
+void
+sync_arr_wake_threads_if_sema_free(void)
+/*====================================*/
+{
+	sync_array_t*	arr	= sync_primary_wait_array;
+	sync_cell_t*	cell;
+	ulint		count;
+	ulint		i;
+	os_event_t      event;
+
+	sync_array_enter(arr);
+
+	i = 0;
+	count = 0;
+
+	while (count < arr->n_reserved) {
+
+		cell = sync_array_get_nth_cell(arr, i);
+		i++;
+
+		if (cell->wait_object == NULL) {
+			continue;
+		}
+			count++;
+
+			if (sync_arr_cell_can_wake_up(cell)) {
+
+			event = sync_cell_get_event(cell);
+
+			os_event_set(event);
+		}
+
+	}
+
+	sync_array_exit(arr);
+}
+
+/**************************************************************************
+Prints warnings of long semaphore waits to stderr. */
+UNIV_INTERN
+ibool
+sync_array_print_long_waits(void)
+/*=============================*/
+			/* out: TRUE if fatal semaphore wait threshold
+			was exceeded */
+{
+	sync_cell_t*	cell;
+	ibool		old_val;
+	ibool		noticed = FALSE;
+	ulint		i;
+	ulint		fatal_timeout = srv_fatal_semaphore_wait_threshold;
+	ibool		fatal = FALSE;
+
+	for (i = 0; i < sync_primary_wait_array->n_cells; i++) {
+
+		cell = sync_array_get_nth_cell(sync_primary_wait_array, i);
+
+		if (cell->wait_object != NULL && cell->waiting
+		    && difftime(time(NULL), cell->reservation_time) > 240) {
+			fputs("InnoDB: Warning: a long semaphore wait:\n",
+			      stderr);
+			sync_array_cell_print(stderr, cell);
+			noticed = TRUE;
+		}
+
+		if (cell->wait_object != NULL && cell->waiting
+		    && difftime(time(NULL), cell->reservation_time)
+		    > fatal_timeout) {
+			fatal = TRUE;
+		}
+	}
+
+	if (noticed) {
+		fprintf(stderr,
+			"InnoDB: ###### Starts InnoDB Monitor"
+			" for 30 secs to print diagnostic info:\n");
+		old_val = srv_print_innodb_monitor;
+
+		/* If some crucial semaphore is reserved, then also the InnoDB
+		Monitor can hang, and we do not get diagnostics. Since in
+		many cases an InnoDB hang is caused by a pwrite() or a pread()
+		call hanging inside the operating system, let us print right
+		now the values of pending calls of these. */
+
+		fprintf(stderr,
+			"InnoDB: Pending preads %lu, pwrites %lu\n",
+			(ulong)os_file_n_pending_preads,
+			(ulong)os_file_n_pending_pwrites);
+
+		srv_print_innodb_monitor = TRUE;
+		os_event_set(srv_lock_timeout_thread_event);
+
+		os_thread_sleep(30000000);
+
+		srv_print_innodb_monitor = old_val;
+		fprintf(stderr,
+			"InnoDB: ###### Diagnostic info printed"
+			" to the standard error stream\n");
+	}
+
+	return(fatal);
+}
+
+/**************************************************************************
+Prints info of the wait array. */
+static
+void
+sync_array_output_info(
+/*===================*/
+	FILE*		file,	/* in: file where to print */
+	sync_array_t*	arr)	/* in: wait array; NOTE! caller must own the
+				mutex */
+{
+	sync_cell_t*	cell;
+	ulint		count;
+	ulint		i;
+
+	fprintf(file,
+		"OS WAIT ARRAY INFO: reservation count %ld, signal count %ld\n",
+						(long) arr->res_count, (long) arr->sg_count);
+	i = 0;
+	count = 0;
+
+	while (count < arr->n_reserved) {
+
+		cell = sync_array_get_nth_cell(arr, i);
+
+	if (cell->wait_object != NULL) {
+		count++;
+			sync_array_cell_print(file, cell);
+		}
+
+		i++;
+	}
+}
+
+/**************************************************************************
+Prints info of the wait array. */
+UNIV_INTERN
+void
+sync_array_print_info(
+/*==================*/
+	FILE*		file,	/* in: file where to print */
+	sync_array_t*	arr)	/* in: wait array */
+{
+	sync_array_enter(arr);
+
+	sync_array_output_info(file, arr);
+
+	sync_array_exit(arr);
+}
diff --git a/storage/xtradb/sync/sync0rw.c b/storage/xtradb/sync/sync0rw.c
new file mode 100644
index 00000000000..09c732eefc9
--- /dev/null
+++ b/storage/xtradb/sync/sync0rw.c
@@ -0,0 +1,1035 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+Copyright (c) 2008, Google Inc.
+
+Portions of this file contain modifications contributed and copyrighted by
+Google, Inc. Those modifications are gratefully acknowledged and are described
+briefly in the InnoDB documentation. The contributions by Google are
+incorporated with their permission, and subject to the conditions contained in
+the file COPYING.Google.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The read-write lock (for thread synchronization)
+
+Created 9/11/1995 Heikki Tuuri
+*******************************************************/
+
+#include "sync0rw.h"
+#ifdef UNIV_NONINL
+#include "sync0rw.ic"
+#endif
+
+#include "os0thread.h"
+#include "mem0mem.h"
+#include "srv0srv.h"
+
+/*
+	IMPLEMENTATION OF THE RW_LOCK
+	=============================
+The status of a rw_lock is held in lock_word. The initial value of lock_word is
+X_LOCK_DECR. lock_word is decremented by 1 for each s-lock and by X_LOCK_DECR
+for each x-lock. This describes the lock state for each value of lock_word:
+
+lock_word == X_LOCK_DECR:      Unlocked.
+0 < lock_word < X_LOCK_DECR:   Read locked, no waiting writers.
+			       (X_LOCK_DECR - lock_word) is the
+			       number of readers that hold the lock.
+lock_word == 0:		       Write locked
+-X_LOCK_DECR < lock_word < 0:  Read locked, with a waiting writer.
+			       (-lock_word) is the number of readers
+			       that hold the lock.
+lock_word <= -X_LOCK_DECR:     Recursively write locked. lock_word has been
+			       decremented by X_LOCK_DECR once for each lock,
+			       so the number of locks is:
+			       ((-lock_word) / X_LOCK_DECR) + 1
+When lock_word <= -X_LOCK_DECR, we also know that lock_word % X_LOCK_DECR == 0:
+other values of lock_word are invalid.
+
+The lock_word is always read and updated atomically and consistently, so that
+it always represents the state of the lock, and the state of the lock changes
+with a single atomic operation. This lock_word holds all of the information
+that a thread needs in order to determine if it is eligible to gain the lock
+or if it must spin or sleep. The one exception to this is that writer_thread
+must be verified before recursive write locks: to solve this scenario, we make
+writer_thread readable by all threads, but only writeable by the x-lock holder.
+
+The other members of the lock obey the following rules to remain consistent:
+
+recursive:	This and the writer_thread field together control the
+		behaviour of recursive x-locking.
+		lock->recursive must be FALSE in following states:
+			1) The writer_thread contains garbage i.e.: the
+			lock has just been initialized.
+			2) The lock is not x-held and there is no
+			x-waiter waiting on WAIT_EX event.
+			3) The lock is x-held or there is an x-waiter
+			waiting on WAIT_EX event but the 'pass' value
+			is non-zero.
+		lock->recursive is TRUE iff:
+			1) The lock is x-held or there is an x-waiter
+			waiting on WAIT_EX event and the 'pass' value
+			is zero.
+		This flag must be set after the writer_thread field
+		has been updated with a memory ordering barrier.
+		It is unset before the lock_word has been incremented.
+writer_thread:	Is used only in recursive x-locking. Can only be safely
+		read iff lock->recursive flag is TRUE.
+		This field is uninitialized at lock creation time and
+		is updated atomically when x-lock is acquired or when
+		move_ownership is called. A thread is only allowed to
+		set the value of this field to it's thread_id i.e.: a
+		thread cannot set writer_thread to some other thread's
+		id.
+waiters:	May be set to 1 anytime, but to avoid unnecessary wake-up
+		signals, it should only be set to 1 when there are threads
+		waiting on event. Must be 1 when a writer starts waiting to
+		ensure the current x-locking thread sends a wake-up signal
+		during unlock. May only be reset to 0 immediately before a
+		a wake-up signal is sent to event. On most platforms, a
+		memory barrier is required after waiters is set, and before
+		verifying lock_word is still held, to ensure some unlocker
+		really does see the flags new value.
+event:		Threads wait on event for read or writer lock when another
+		thread has an x-lock or an x-lock reservation (wait_ex). A
+		thread may only	wait on event after performing the following
+		actions in order:
+		   (1) Record the counter value of event (with os_event_reset).
+		   (2) Set waiters to 1.
+		   (3) Verify lock_word <= 0.
+		(1) must come before (2) to ensure signal is not missed.
+		(2) must come before (3) to ensure a signal is sent.
+		These restrictions force the above ordering.
+		Immediately before sending the wake-up signal, we should:
+		   (1) Verify lock_word == X_LOCK_DECR (unlocked)
+		   (2) Reset waiters to 0.
+wait_ex_event:	A thread may only wait on the wait_ex_event after it has
+		performed the following actions in order:
+		   (1) Decrement lock_word by X_LOCK_DECR.
+		   (2) Record counter value of wait_ex_event (os_event_reset,
+                       called from sync_array_reserve_cell).
+		   (3) Verify that lock_word < 0.
+		(1) must come first to ensures no other threads become reader
+                or next writer, and notifies unlocker that signal must be sent.
+                (2) must come before (3) to ensure the signal is not missed.
+		These restrictions force the above ordering.
+		Immediately before sending the wake-up signal, we should:
+		   Verify lock_word == 0 (waiting thread holds x_lock)
+*/
+
+
+/* number of spin waits on rw-latches,
+resulted during shared (read) locks */
+UNIV_INTERN ib_int64_t	rw_s_spin_wait_count	= 0;
+UNIV_INTERN ib_int64_t	rw_s_spin_round_count	= 0;
+
+/* number of OS waits on rw-latches,
+resulted during shared (read) locks */
+UNIV_INTERN ib_int64_t	rw_s_os_wait_count	= 0;
+
+/* number of unlocks (that unlock shared locks),
+set only when UNIV_SYNC_PERF_STAT is defined */
+UNIV_INTERN ib_int64_t	rw_s_exit_count		= 0;
+
+/* number of spin waits on rw-latches,
+resulted during exclusive (write) locks */
+UNIV_INTERN ib_int64_t	rw_x_spin_wait_count	= 0;
+UNIV_INTERN ib_int64_t	rw_x_spin_round_count	= 0;
+
+/* number of OS waits on rw-latches,
+resulted during exclusive (write) locks */
+UNIV_INTERN ib_int64_t	rw_x_os_wait_count	= 0;
+
+/* number of unlocks (that unlock exclusive locks),
+set only when UNIV_SYNC_PERF_STAT is defined */
+UNIV_INTERN ib_int64_t	rw_x_exit_count		= 0;
+
+/* The global list of rw-locks */
+UNIV_INTERN rw_lock_list_t	rw_lock_list;
+UNIV_INTERN mutex_t		rw_lock_list_mutex;
+
+#ifdef UNIV_SYNC_DEBUG
+/* The global mutex which protects debug info lists of all rw-locks.
+To modify the debug info list of an rw-lock, this mutex has to be
+acquired in addition to the mutex protecting the lock. */
+
+UNIV_INTERN mutex_t		rw_lock_debug_mutex;
+/* If deadlock detection does not get immediately the mutex,
+it may wait for this event */
+UNIV_INTERN os_event_t		rw_lock_debug_event;
+/* This is set to TRUE, if there may be waiters for the event */
+UNIV_INTERN ibool		rw_lock_debug_waiters;
+
+/**********************************************************************
+Creates a debug info struct. */
+static
+rw_lock_debug_t*
+rw_lock_debug_create(void);
+/*======================*/
+/**********************************************************************
+Frees a debug info struct. */
+static
+void
+rw_lock_debug_free(
+/*===============*/
+	rw_lock_debug_t* info);
+
+/**********************************************************************
+Creates a debug info struct. */
+static
+rw_lock_debug_t*
+rw_lock_debug_create(void)
+/*======================*/
+{
+	return((rw_lock_debug_t*) mem_alloc(sizeof(rw_lock_debug_t)));
+}
+
+/**********************************************************************
+Frees a debug info struct. */
+static
+void
+rw_lock_debug_free(
+/*===============*/
+	rw_lock_debug_t* info)
+{
+	mem_free(info);
+}
+#endif /* UNIV_SYNC_DEBUG */
+
+/**********************************************************************
+Creates, or rather, initializes an rw-lock object in a specified memory
+location (which must be appropriately aligned). The rw-lock is initialized
+to the non-locked state. Explicit freeing of the rw-lock with rw_lock_free
+is necessary only if the memory block containing it is freed. */
+UNIV_INTERN
+void
+rw_lock_create_func(
+/*================*/
+	rw_lock_t*	lock,		/* in: pointer to memory */
+#ifdef UNIV_DEBUG
+# ifdef UNIV_SYNC_DEBUG
+	ulint		level,		/* in: level */
+# endif /* UNIV_SYNC_DEBUG */
+	const char*	cmutex_name, 	/* in: mutex name */
+#endif /* UNIV_DEBUG */
+	const char*	cfile_name,	/* in: file name where created */
+	ulint		cline)		/* in: file line where created */
+{
+	/* If this is the very first time a synchronization object is
+	created, then the following call initializes the sync system. */
+
+#ifndef INNODB_RW_LOCKS_USE_ATOMICS
+	mutex_create(rw_lock_get_mutex(lock), SYNC_NO_ORDER_CHECK);
+
+	lock->mutex.cfile_name = cfile_name;
+	lock->mutex.cline = cline;
+
+# if defined UNIV_DEBUG && !defined UNIV_HOTBACKUP
+	lock->mutex.cmutex_name = cmutex_name;
+	lock->mutex.mutex_type = 1;
+# endif /* UNIV_DEBUG && !UNIV_HOTBACKUP */
+
+#else /* INNODB_RW_LOCKS_USE_ATOMICS */
+# ifdef UNIV_DEBUG
+	UT_NOT_USED(cmutex_name);
+# endif
+#endif /* INNODB_RW_LOCKS_USE_ATOMICS */
+
+	lock->lock_word = X_LOCK_DECR;
+	lock->waiters = 0;
+
+	/* We set this value to signify that lock->writer_thread
+	contains garbage at initialization and cannot be used for
+	recursive x-locking. */
+	lock->recursive = FALSE;
+
+#ifdef UNIV_SYNC_DEBUG
+	UT_LIST_INIT(lock->debug_list);
+
+	lock->level = level;
+#endif /* UNIV_SYNC_DEBUG */
+
+	lock->magic_n = RW_LOCK_MAGIC_N;
+
+	lock->cfile_name = cfile_name;
+	lock->cline = (unsigned int) cline;
+
+	lock->count_os_wait = 0;
+	lock->last_s_file_name = "not yet reserved";
+	lock->last_x_file_name = "not yet reserved";
+	lock->last_s_line = 0;
+	lock->last_x_line = 0;
+	lock->event = os_event_create(NULL);
+	lock->wait_ex_event = os_event_create(NULL);
+
+	mutex_enter(&rw_lock_list_mutex);
+
+	if (UT_LIST_GET_LEN(rw_lock_list) > 0) {
+		ut_a(UT_LIST_GET_FIRST(rw_lock_list)->magic_n
+		     == RW_LOCK_MAGIC_N);
+	}
+
+	UT_LIST_ADD_FIRST(list, rw_lock_list, lock);
+
+	mutex_exit(&rw_lock_list_mutex);
+}
+
+/**********************************************************************
+Calling this function is obligatory only if the memory buffer containing
+the rw-lock is freed. Removes an rw-lock object from the global list. The
+rw-lock is checked to be in the non-locked state. */
+UNIV_INTERN
+void
+rw_lock_free(
+/*=========*/
+	rw_lock_t*	lock)	/* in: rw-lock */
+{
+	ut_ad(rw_lock_validate(lock));
+	ut_a(lock->lock_word == X_LOCK_DECR);
+
+	lock->magic_n = 0;
+
+#ifndef INNODB_RW_LOCKS_USE_ATOMICS
+	mutex_free(rw_lock_get_mutex(lock));
+#endif /* INNODB_RW_LOCKS_USE_ATOMICS */
+
+	mutex_enter(&rw_lock_list_mutex);
+	os_event_free(lock->event);
+
+	os_event_free(lock->wait_ex_event);
+
+	if (UT_LIST_GET_PREV(list, lock)) {
+		ut_a(UT_LIST_GET_PREV(list, lock)->magic_n == RW_LOCK_MAGIC_N);
+	}
+	if (UT_LIST_GET_NEXT(list, lock)) {
+		ut_a(UT_LIST_GET_NEXT(list, lock)->magic_n == RW_LOCK_MAGIC_N);
+	}
+
+	UT_LIST_REMOVE(list, rw_lock_list, lock);
+
+	mutex_exit(&rw_lock_list_mutex);
+}
+
+#ifdef UNIV_DEBUG
+/**********************************************************************
+Checks that the rw-lock has been initialized and that there are no
+simultaneous shared and exclusive locks. */
+UNIV_INTERN
+ibool
+rw_lock_validate(
+/*=============*/
+	rw_lock_t*	lock)
+{
+	ut_a(lock);
+
+	ulint waiters = rw_lock_get_waiters(lock);
+	lint lock_word = lock->lock_word;
+
+	ut_a(lock->magic_n == RW_LOCK_MAGIC_N);
+	ut_a(waiters == 0 || waiters == 1);
+	ut_a(lock_word > -X_LOCK_DECR ||(-lock_word) % X_LOCK_DECR == 0);
+
+	return(TRUE);
+}
+#endif /* UNIV_DEBUG */
+
+/**********************************************************************
+Lock an rw-lock in shared mode for the current thread. If the rw-lock is
+locked in exclusive mode, or there is an exclusive lock request waiting,
+the function spins a preset time (controlled by SYNC_SPIN_ROUNDS), waiting
+for the lock, before suspending the thread. */
+UNIV_INTERN
+void
+rw_lock_s_lock_spin(
+/*================*/
+	rw_lock_t*	lock,	/* in: pointer to rw-lock */
+	ulint		pass,	/* in: pass value; != 0, if the lock
+				will be passed to another thread to unlock */
+	const char*	file_name, /* in: file name where lock requested */
+	ulint		line)	/* in: line where requested */
+{
+	ulint	 index;	/* index of the reserved wait cell */
+	ulint	 i = 0;	/* spin round count */
+
+	ut_ad(rw_lock_validate(lock));
+
+	rw_s_spin_wait_count++;	/* Count calls to this function */
+lock_loop:
+
+	/* Spin waiting for the writer field to become free */
+	while (i < SYNC_SPIN_ROUNDS && lock->lock_word <= 0) {
+		if (srv_spin_wait_delay) {
+			ut_delay(ut_rnd_interval(0, srv_spin_wait_delay));
+		}
+
+		i++;
+	}
+
+	if (i == SYNC_SPIN_ROUNDS) {
+		os_thread_yield();
+	}
+
+	if (srv_print_latch_waits) {
+		fprintf(stderr,
+			"Thread %lu spin wait rw-s-lock at %p"
+			" cfile %s cline %lu rnds %lu\n",
+			(ulong) os_thread_pf(os_thread_get_curr_id()),
+			(void*) lock,
+			lock->cfile_name, (ulong) lock->cline, (ulong) i);
+	}
+
+	/* We try once again to obtain the lock */
+	if (TRUE == rw_lock_s_lock_low(lock, pass, file_name, line)) {
+		rw_s_spin_round_count += i;
+
+		return; /* Success */
+	} else {
+
+		if (i < SYNC_SPIN_ROUNDS) {
+			goto lock_loop;
+		}
+
+		rw_s_spin_round_count += i;
+
+		sync_array_reserve_cell(sync_primary_wait_array,
+					lock, RW_LOCK_SHARED,
+					file_name, line,
+					&index);
+
+		/* Set waiters before checking lock_word to ensure wake-up
+                signal is sent. This may lead to some unnecessary signals. */
+		rw_lock_set_waiter_flag(lock);
+
+		if (TRUE == rw_lock_s_lock_low(lock, pass, file_name, line)) {
+			sync_array_free_cell(sync_primary_wait_array, index);
+			return; /* Success */
+		}
+
+		if (srv_print_latch_waits) {
+			fprintf(stderr,
+				"Thread %lu OS wait rw-s-lock at %p"
+				" cfile %s cline %lu\n",
+				os_thread_pf(os_thread_get_curr_id()),
+				(void*) lock, lock->cfile_name,
+				(ulong) lock->cline);
+		}
+
+		/* these stats may not be accurate */
+		lock->count_os_wait++;
+		rw_s_os_wait_count++;
+
+		sync_array_wait_event(sync_primary_wait_array, index);
+
+		i = 0;
+		goto lock_loop;
+	}
+}
+
+/**********************************************************************
+This function is used in the insert buffer to move the ownership of an
+x-latch on a buffer frame to the current thread. The x-latch was set by
+the buffer read operation and it protected the buffer frame while the
+read was done. The ownership is moved because we want that the current
+thread is able to acquire a second x-latch which is stored in an mtr.
+This, in turn, is needed to pass the debug checks of index page
+operations. */
+UNIV_INTERN
+void
+rw_lock_x_lock_move_ownership(
+/*==========================*/
+	rw_lock_t*	lock)	/* in: lock which was x-locked in the
+				buffer read */
+{
+	ut_ad(rw_lock_is_locked(lock, RW_LOCK_EX));
+
+	rw_lock_set_writer_id_and_recursion_flag(lock, TRUE);
+}
+
+/**********************************************************************
+Function for the next writer to call. Waits for readers to exit.
+The caller must have already decremented lock_word by X_LOCK_DECR.*/
+UNIV_INLINE
+void
+rw_lock_x_lock_wait(
+/*================*/
+	rw_lock_t*	lock,	/* in: pointer to rw-lock */
+#ifdef UNIV_SYNC_DEBUG
+	ulint		pass,	/* in: pass value; != 0, if the lock will
+				be passed to another thread to unlock */
+#endif
+	const char*	file_name,/* in: file name where lock requested */
+	ulint		line)	/* in: line where requested */
+{
+	ulint index;
+	ulint i = 0;
+
+	ut_ad(lock->lock_word <= 0);
+
+	while (lock->lock_word < 0) {
+		if (srv_spin_wait_delay) {
+			ut_delay(ut_rnd_interval(0, srv_spin_wait_delay));
+		}
+		if(i < SYNC_SPIN_ROUNDS) {
+			i++;
+			continue;
+		}
+
+		/* If there is still a reader, then go to sleep.*/
+		rw_x_spin_round_count += i;
+		i = 0;
+		sync_array_reserve_cell(sync_primary_wait_array,
+					lock,
+					RW_LOCK_WAIT_EX,
+					file_name, line,
+					&index);
+		/* Check lock_word to ensure wake-up isn't missed.*/
+		if(lock->lock_word < 0) {
+
+			/* these stats may not be accurate */
+			lock->count_os_wait++;
+			rw_x_os_wait_count++;
+
+                        /* Add debug info as it is needed to detect possible
+                        deadlock. We must add info for WAIT_EX thread for
+                        deadlock detection to work properly. */
+#ifdef UNIV_SYNC_DEBUG
+			rw_lock_add_debug_info(lock, pass, RW_LOCK_WAIT_EX,
+					       file_name, line);
+#endif
+
+			sync_array_wait_event(sync_primary_wait_array,
+					      index);
+#ifdef UNIV_SYNC_DEBUG
+			rw_lock_remove_debug_info(lock, pass,
+					       RW_LOCK_WAIT_EX);
+#endif
+                        /* It is possible to wake when lock_word < 0.
+                        We must pass the while-loop check to proceed.*/
+		} else {
+			sync_array_free_cell(sync_primary_wait_array,
+					     index);
+		}
+	}
+	rw_x_spin_round_count += i;
+}
+
+/**********************************************************************
+Low-level function for acquiring an exclusive lock. */
+UNIV_INLINE
+ibool
+rw_lock_x_lock_low(
+/*===============*/
+				/* out: RW_LOCK_NOT_LOCKED if did
+				not succeed, RW_LOCK_EX if success. */
+	rw_lock_t*	lock,	/* in: pointer to rw-lock */
+	ulint		pass,	/* in: pass value; != 0, if the lock will
+				be passed to another thread to unlock */
+	const char*	file_name,/* in: file name where lock requested */
+	ulint		line)	/* in: line where requested */
+{
+	os_thread_id_t	curr_thread	= os_thread_get_curr_id();
+
+	if (rw_lock_lock_word_decr(lock, X_LOCK_DECR)) {
+
+		/* lock->recursive also tells us if the writer_thread
+		field is stale or active. As we are going to write
+		our own thread id in that field it must be that the
+		current writer_thread value is not active. */
+		ut_a(!lock->recursive);
+
+		/* Decrement occurred: we are writer or next-writer. */
+		rw_lock_set_writer_id_and_recursion_flag(lock,
+						pass ? FALSE : TRUE);
+
+		rw_lock_x_lock_wait(lock,
+#ifdef UNIV_SYNC_DEBUG
+				    pass,
+#endif
+                                    file_name, line);
+
+	} else {
+		/* Decrement failed: relock or failed lock */
+		if (!pass && lock->recursive
+		    && os_thread_eq(lock->writer_thread, curr_thread)) {
+			/* Relock */
+                        lock->lock_word -= X_LOCK_DECR;
+		} else {
+			/* Another thread locked before us */
+			return(FALSE);
+		}
+	}
+#ifdef UNIV_SYNC_DEBUG
+	rw_lock_add_debug_info(lock, pass, RW_LOCK_EX,
+			       file_name, line);
+#endif
+	lock->last_x_file_name = file_name;
+	lock->last_x_line = (unsigned int) line;
+
+	return(TRUE);
+}
+
+/**********************************************************************
+NOTE! Use the corresponding macro, not directly this function! Lock an
+rw-lock in exclusive mode for the current thread. If the rw-lock is locked
+in shared or exclusive mode, or there is an exclusive lock request waiting,
+the function spins a preset time (controlled by SYNC_SPIN_ROUNDS), waiting
+for the lock before suspending the thread. If the same thread has an x-lock
+on the rw-lock, locking succeed, with the following exception: if pass != 0,
+only a single x-lock may be taken on the lock. NOTE: If the same thread has
+an s-lock, locking does not succeed! */
+UNIV_INTERN
+void
+rw_lock_x_lock_func(
+/*================*/
+	rw_lock_t*	lock,	/* in: pointer to rw-lock */
+	ulint		pass,	/* in: pass value; != 0, if the lock will
+				be passed to another thread to unlock */
+	const char*	file_name,/* in: file name where lock requested */
+	ulint		line)	/* in: line where requested */
+{
+	ulint	index;	/* index of the reserved wait cell */
+	ulint	i;	/* spin round count */
+	ibool   spinning = FALSE;
+
+	ut_ad(rw_lock_validate(lock));
+
+	i = 0;
+
+lock_loop:
+
+	if (rw_lock_x_lock_low(lock, pass, file_name, line)) {
+		rw_x_spin_round_count += i;
+
+		return;	/* Locking succeeded */
+
+	} else {
+
+                if (!spinning) {
+                        spinning = TRUE;
+                        rw_x_spin_wait_count++;
+		}
+
+		/* Spin waiting for the lock_word to become free */
+		while (i < SYNC_SPIN_ROUNDS
+		       && lock->lock_word <= 0) {
+			if (srv_spin_wait_delay) {
+				ut_delay(ut_rnd_interval(0,
+							 srv_spin_wait_delay));
+			}
+
+			i++;
+		}
+		if (i == SYNC_SPIN_ROUNDS) {
+			os_thread_yield();
+		} else {
+			goto lock_loop;
+		}
+	}
+
+	rw_x_spin_round_count += i;
+
+	if (srv_print_latch_waits) {
+		fprintf(stderr,
+			"Thread %lu spin wait rw-x-lock at %p"
+			" cfile %s cline %lu rnds %lu\n",
+			os_thread_pf(os_thread_get_curr_id()), (void*) lock,
+			lock->cfile_name, (ulong) lock->cline, (ulong) i);
+	}
+
+	sync_array_reserve_cell(sync_primary_wait_array,
+				lock,
+				RW_LOCK_EX,
+				file_name, line,
+				&index);
+
+	/* Waiters must be set before checking lock_word, to ensure signal
+	is sent. This could lead to a few unnecessary wake-up signals. */
+	rw_lock_set_waiter_flag(lock);
+
+	if (rw_lock_x_lock_low(lock, pass, file_name, line)) {
+		sync_array_free_cell(sync_primary_wait_array, index);
+		return; /* Locking succeeded */
+	}
+
+	if (srv_print_latch_waits) {
+		fprintf(stderr,
+			"Thread %lu OS wait for rw-x-lock at %p"
+			" cfile %s cline %lu\n",
+			os_thread_pf(os_thread_get_curr_id()), (void*) lock,
+			lock->cfile_name, (ulong) lock->cline);
+	}
+
+	/* these stats may not be accurate */
+	lock->count_os_wait++;
+	rw_x_os_wait_count++;
+
+	sync_array_wait_event(sync_primary_wait_array, index);
+
+	i = 0;
+	goto lock_loop;
+}
+
+#ifdef UNIV_SYNC_DEBUG
+/**********************************************************************
+Acquires the debug mutex. We cannot use the mutex defined in sync0sync,
+because the debug mutex is also acquired in sync0arr while holding the OS
+mutex protecting the sync array, and the ordinary mutex_enter might
+recursively call routines in sync0arr, leading to a deadlock on the OS
+mutex. */
+UNIV_INTERN
+void
+rw_lock_debug_mutex_enter(void)
+/*==========================*/
+{
+loop:
+	if (0 == mutex_enter_nowait(&rw_lock_debug_mutex)) {
+		return;
+	}
+
+	os_event_reset(rw_lock_debug_event);
+
+	rw_lock_debug_waiters = TRUE;
+
+	if (0 == mutex_enter_nowait(&rw_lock_debug_mutex)) {
+		return;
+	}
+
+	os_event_wait(rw_lock_debug_event);
+
+	goto loop;
+}
+
+/**********************************************************************
+Releases the debug mutex. */
+UNIV_INTERN
+void
+rw_lock_debug_mutex_exit(void)
+/*==========================*/
+{
+	mutex_exit(&rw_lock_debug_mutex);
+
+	if (rw_lock_debug_waiters) {
+		rw_lock_debug_waiters = FALSE;
+		os_event_set(rw_lock_debug_event);
+	}
+}
+
+/**********************************************************************
+Inserts the debug information for an rw-lock. */
+UNIV_INTERN
+void
+rw_lock_add_debug_info(
+/*===================*/
+	rw_lock_t*	lock,		/* in: rw-lock */
+	ulint		pass,		/* in: pass value */
+	ulint		lock_type,	/* in: lock type */
+	const char*	file_name,	/* in: file where requested */
+	ulint		line)		/* in: line where requested */
+{
+	rw_lock_debug_t*	info;
+
+	ut_ad(lock);
+	ut_ad(file_name);
+
+	info = rw_lock_debug_create();
+
+	rw_lock_debug_mutex_enter();
+
+	info->file_name = file_name;
+	info->line	= line;
+	info->lock_type = lock_type;
+	info->thread_id = os_thread_get_curr_id();
+	info->pass	= pass;
+
+	UT_LIST_ADD_FIRST(list, lock->debug_list, info);
+
+	rw_lock_debug_mutex_exit();
+
+	if ((pass == 0) && (lock_type != RW_LOCK_WAIT_EX)) {
+		sync_thread_add_level(lock, lock->level);
+	}
+}
+
+/**********************************************************************
+Removes a debug information struct for an rw-lock. */
+UNIV_INTERN
+void
+rw_lock_remove_debug_info(
+/*======================*/
+	rw_lock_t*	lock,		/* in: rw-lock */
+	ulint		pass,		/* in: pass value */
+	ulint		lock_type)	/* in: lock type */
+{
+	rw_lock_debug_t*	info;
+
+	ut_ad(lock);
+
+	if ((pass == 0) && (lock_type != RW_LOCK_WAIT_EX)) {
+		sync_thread_reset_level(lock);
+	}
+
+	rw_lock_debug_mutex_enter();
+
+	info = UT_LIST_GET_FIRST(lock->debug_list);
+
+	while (info != NULL) {
+		if ((pass == info->pass)
+		    && ((pass != 0)
+			|| os_thread_eq(info->thread_id,
+					os_thread_get_curr_id()))
+		    && (info->lock_type == lock_type)) {
+
+			/* Found! */
+			UT_LIST_REMOVE(list, lock->debug_list, info);
+			rw_lock_debug_mutex_exit();
+
+			rw_lock_debug_free(info);
+
+			return;
+		}
+
+		info = UT_LIST_GET_NEXT(list, info);
+	}
+
+	ut_error;
+}
+#endif /* UNIV_SYNC_DEBUG */
+
+#ifdef UNIV_SYNC_DEBUG
+/**********************************************************************
+Checks if the thread has locked the rw-lock in the specified mode, with
+the pass value == 0. */
+UNIV_INTERN
+ibool
+rw_lock_own(
+/*========*/
+					/* out: TRUE if locked */
+	rw_lock_t*	lock,		/* in: rw-lock */
+	ulint		lock_type)	/* in: lock type: RW_LOCK_SHARED,
+					RW_LOCK_EX */
+{
+	rw_lock_debug_t*	info;
+
+	ut_ad(lock);
+	ut_ad(rw_lock_validate(lock));
+
+	rw_lock_debug_mutex_enter();
+
+	info = UT_LIST_GET_FIRST(lock->debug_list);
+
+	while (info != NULL) {
+
+		if (os_thread_eq(info->thread_id, os_thread_get_curr_id())
+		    && (info->pass == 0)
+		    && (info->lock_type == lock_type)) {
+
+			rw_lock_debug_mutex_exit();
+			/* Found! */
+
+			return(TRUE);
+		}
+
+		info = UT_LIST_GET_NEXT(list, info);
+	}
+	rw_lock_debug_mutex_exit();
+
+	return(FALSE);
+}
+#endif /* UNIV_SYNC_DEBUG */
+
+/**********************************************************************
+Checks if somebody has locked the rw-lock in the specified mode. */
+UNIV_INTERN
+ibool
+rw_lock_is_locked(
+/*==============*/
+					/* out: TRUE if locked */
+	rw_lock_t*	lock,		/* in: rw-lock */
+	ulint		lock_type)	/* in: lock type: RW_LOCK_SHARED,
+					RW_LOCK_EX */
+{
+	ibool	ret	= FALSE;
+
+	ut_ad(lock);
+	ut_ad(rw_lock_validate(lock));
+
+	if (lock_type == RW_LOCK_SHARED) {
+		if (rw_lock_get_reader_count(lock) > 0) {
+			ret = TRUE;
+		}
+	} else if (lock_type == RW_LOCK_EX) {
+		if (rw_lock_get_writer(lock) == RW_LOCK_EX) {
+			ret = TRUE;
+		}
+	} else {
+		ut_error;
+	}
+
+	return(ret);
+}
+
+#ifdef UNIV_SYNC_DEBUG
+/*******************************************************************
+Prints debug info of currently locked rw-locks. */
+UNIV_INTERN
+void
+rw_lock_list_print_info(
+/*====================*/
+	FILE*	file)		/* in: file where to print */
+{
+	rw_lock_t*	lock;
+	ulint		count		= 0;
+	rw_lock_debug_t* info;
+
+	mutex_enter(&rw_lock_list_mutex);
+
+	fputs("-------------\n"
+	      "RW-LATCH INFO\n"
+	      "-------------\n", file);
+
+	lock = UT_LIST_GET_FIRST(rw_lock_list);
+
+	while (lock != NULL) {
+
+		count++;
+
+#ifndef INNODB_RW_LOCKS_USE_ATOMICS
+		mutex_enter(&(lock->mutex));
+#endif
+		if (lock->lock_word != X_LOCK_DECR) {
+
+			fprintf(file, "RW-LOCK: %p ", (void*) lock);
+
+			if (rw_lock_get_waiters(lock)) {
+				fputs(" Waiters for the lock exist\n", file);
+			} else {
+				putc('\n', file);
+			}
+
+			info = UT_LIST_GET_FIRST(lock->debug_list);
+			while (info != NULL) {
+				rw_lock_debug_print(info);
+				info = UT_LIST_GET_NEXT(list, info);
+			}
+		}
+#ifndef INNODB_RW_LOCKS_USE_ATOMICS
+		mutex_exit(&(lock->mutex));
+#endif
+
+		lock = UT_LIST_GET_NEXT(list, lock);
+	}
+
+	fprintf(file, "Total number of rw-locks %ld\n", count);
+	mutex_exit(&rw_lock_list_mutex);
+}
+
+/*******************************************************************
+Prints debug info of an rw-lock. */
+UNIV_INTERN
+void
+rw_lock_print(
+/*==========*/
+	rw_lock_t*	lock)	/* in: rw-lock */
+{
+	rw_lock_debug_t* info;
+
+	fprintf(stderr,
+		"-------------\n"
+		"RW-LATCH INFO\n"
+		"RW-LATCH: %p ", (void*) lock);
+
+#ifndef INNODB_RW_LOCKS_USE_ATOMICS
+	mutex_enter(&(lock->mutex));
+#endif
+	if (lock->lock_word != X_LOCK_DECR) {
+
+		if (rw_lock_get_waiters(lock)) {
+			fputs(" Waiters for the lock exist\n", stderr);
+		} else {
+			putc('\n', stderr);
+		}
+
+		info = UT_LIST_GET_FIRST(lock->debug_list);
+		while (info != NULL) {
+			rw_lock_debug_print(info);
+			info = UT_LIST_GET_NEXT(list, info);
+		}
+	}
+#ifndef INNODB_RW_LOCKS_USE_ATOMICS
+	mutex_exit(&(lock->mutex));
+#endif
+}
+
+/*************************************************************************
+Prints info of a debug struct. */
+UNIV_INTERN
+void
+rw_lock_debug_print(
+/*================*/
+	rw_lock_debug_t*	info)	/* in: debug struct */
+{
+	ulint	rwt;
+
+	rwt	  = info->lock_type;
+
+	fprintf(stderr, "Locked: thread %ld file %s line %ld  ",
+		(ulong) os_thread_pf(info->thread_id), info->file_name,
+		(ulong) info->line);
+	if (rwt == RW_LOCK_SHARED) {
+		fputs("S-LOCK", stderr);
+	} else if (rwt == RW_LOCK_EX) {
+		fputs("X-LOCK", stderr);
+	} else if (rwt == RW_LOCK_WAIT_EX) {
+		fputs("WAIT X-LOCK", stderr);
+	} else {
+		ut_error;
+	}
+	if (info->pass != 0) {
+		fprintf(stderr, " pass value %lu", (ulong) info->pass);
+	}
+	putc('\n', stderr);
+}
+
+/*******************************************************************
+Returns the number of currently locked rw-locks. Works only in the debug
+version. */
+UNIV_INTERN
+ulint
+rw_lock_n_locked(void)
+/*==================*/
+{
+	rw_lock_t*	lock;
+	ulint		count		= 0;
+
+	mutex_enter(&rw_lock_list_mutex);
+
+	lock = UT_LIST_GET_FIRST(rw_lock_list);
+
+	while (lock != NULL) {
+
+		if (lock->lock_word != X_LOCK_DECR) {
+			count++;
+		}
+
+		lock = UT_LIST_GET_NEXT(list, lock);
+	}
+
+	mutex_exit(&rw_lock_list_mutex);
+
+	return(count);
+}
+#endif /* UNIV_SYNC_DEBUG */
diff --git a/storage/xtradb/sync/sync0sync.c b/storage/xtradb/sync/sync0sync.c
new file mode 100644
index 00000000000..3b2d033aae5
--- /dev/null
+++ b/storage/xtradb/sync/sync0sync.c
@@ -0,0 +1,1411 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+Copyright (c) 2008, Google Inc.
+
+Portions of this file contain modifications contributed and copyrighted by
+Google, Inc. Those modifications are gratefully acknowledged and are described
+briefly in the InnoDB documentation. The contributions by Google are
+incorporated with their permission, and subject to the conditions contained in
+the file COPYING.Google.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Mutex, the basic synchronization primitive
+
+Created 9/5/1995 Heikki Tuuri
+*******************************************************/
+
+#include "sync0sync.h"
+#ifdef UNIV_NONINL
+#include "sync0sync.ic"
+#endif
+
+#include "sync0rw.h"
+#include "buf0buf.h"
+#include "srv0srv.h"
+#include "buf0types.h"
+
+/*
+	REASONS FOR IMPLEMENTING THE SPIN LOCK MUTEX
+	============================================
+
+Semaphore operations in operating systems are slow: Solaris on a 1993 Sparc
+takes 3 microseconds (us) for a lock-unlock pair and Windows NT on a 1995
+Pentium takes 20 microseconds for a lock-unlock pair. Therefore, we have to
+implement our own efficient spin lock mutex. Future operating systems may
+provide efficient spin locks, but we cannot count on that.
+
+Another reason for implementing a spin lock is that on multiprocessor systems
+it can be more efficient for a processor to run a loop waiting for the
+semaphore to be released than to switch to a different thread. A thread switch
+takes 25 us on both platforms mentioned above. See Gray and Reuter's book
+Transaction processing for background.
+
+How long should the spin loop last before suspending the thread? On a
+uniprocessor, spinning does not help at all, because if the thread owning the
+mutex is not executing, it cannot be released. Spinning actually wastes
+resources.
+
+On a multiprocessor, we do not know if the thread owning the mutex is
+executing or not. Thus it would make sense to spin as long as the operation
+guarded by the mutex would typically last assuming that the thread is
+executing. If the mutex is not released by that time, we may assume that the
+thread owning the mutex is not executing and suspend the waiting thread.
+
+A typical operation (where no i/o involved) guarded by a mutex or a read-write
+lock may last 1 - 20 us on the current Pentium platform. The longest
+operations are the binary searches on an index node.
+
+We conclude that the best choice is to set the spin time at 20 us. Then the
+system should work well on a multiprocessor. On a uniprocessor we have to
+make sure that thread swithches due to mutex collisions are not frequent,
+i.e., they do not happen every 100 us or so, because that wastes too much
+resources. If the thread switches are not frequent, the 20 us wasted in spin
+loop is not too much.
+
+Empirical studies on the effect of spin time should be done for different
+platforms.
+
+
+	IMPLEMENTATION OF THE MUTEX
+	===========================
+
+For background, see Curt Schimmel's book on Unix implementation on modern
+architectures. The key points in the implementation are atomicity and
+serialization of memory accesses. The test-and-set instruction (XCHG in
+Pentium) must be atomic. As new processors may have weak memory models, also
+serialization of memory references may be necessary. The successor of Pentium,
+P6, has at least one mode where the memory model is weak. As far as we know,
+in Pentium all memory accesses are serialized in the program order and we do
+not have to worry about the memory model. On other processors there are
+special machine instructions called a fence, memory barrier, or storage
+barrier (STBAR in Sparc), which can be used to serialize the memory accesses
+to happen in program order relative to the fence instruction.
+
+Leslie Lamport has devised a "bakery algorithm" to implement a mutex without
+the atomic test-and-set, but his algorithm should be modified for weak memory
+models. We do not use Lamport's algorithm, because we guess it is slower than
+the atomic test-and-set.
+
+Our mutex implementation works as follows: After that we perform the atomic
+test-and-set instruction on the memory word. If the test returns zero, we
+know we got the lock first. If the test returns not zero, some other thread
+was quicker and got the lock: then we spin in a loop reading the memory word,
+waiting it to become zero. It is wise to just read the word in the loop, not
+perform numerous test-and-set instructions, because they generate memory
+traffic between the cache and the main memory. The read loop can just access
+the cache, saving bus bandwidth.
+
+If we cannot acquire the mutex lock in the specified time, we reserve a cell
+in the wait array, set the waiters byte in the mutex to 1. To avoid a race
+condition, after setting the waiters byte and before suspending the waiting
+thread, we still have to check that the mutex is reserved, because it may
+have happened that the thread which was holding the mutex has just released
+it and did not see the waiters byte set to 1, a case which would lead the
+other thread to an infinite wait.
+
+LEMMA 1: After a thread resets the event of a mutex (or rw_lock), some
+=======
+thread will eventually call os_event_set() on that particular event.
+Thus no infinite wait is possible in this case.
+
+Proof:	After making the reservation the thread sets the waiters field in the
+mutex to 1. Then it checks that the mutex is still reserved by some thread,
+or it reserves the mutex for itself. In any case, some thread (which may be
+also some earlier thread, not necessarily the one currently holding the mutex)
+will set the waiters field to 0 in mutex_exit, and then call
+os_event_set() with the mutex as an argument.
+Q.E.D.
+
+LEMMA 2: If an os_event_set() call is made after some thread has called
+=======
+the os_event_reset() and before it starts wait on that event, the call
+will not be lost to the second thread. This is true even if there is an
+intervening call to os_event_reset() by another thread.
+Thus no infinite wait is possible in this case.
+
+Proof (non-windows platforms): os_event_reset() returns a monotonically
+increasing value of signal_count. This value is increased at every
+call of os_event_set() If thread A has called os_event_reset() followed
+by thread B calling os_event_set() and then some other thread C calling
+os_event_reset(), the is_set flag of the event will be set to FALSE;
+but now if thread A calls os_event_wait_low() with the signal_count
+value returned from the earlier call of os_event_reset(), it will
+return immediately without waiting.
+Q.E.D.
+
+Proof (windows): If there is a writer thread which is forced to wait for
+the lock, it may be able to set the state of rw_lock to RW_LOCK_WAIT_EX
+The design of rw_lock ensures that there is one and only one thread
+that is able to change the state to RW_LOCK_WAIT_EX and this thread is
+guaranteed to acquire the lock after it is released by the current
+holders and before any other waiter gets the lock.
+On windows this thread waits on a separate event i.e.: wait_ex_event.
+Since only one thread can wait on this event there is no chance
+of this event getting reset before the writer starts wait on it.
+Therefore, this thread is guaranteed to catch the os_set_event()
+signalled unconditionally at the release of the lock.
+Q.E.D. */
+
+/* Number of spin waits on mutexes: for performance monitoring */
+
+/* round=one iteration of a spin loop */
+UNIV_INTERN ib_int64_t	mutex_spin_round_count		= 0;
+UNIV_INTERN ib_int64_t	mutex_spin_wait_count		= 0;
+UNIV_INTERN ib_int64_t	mutex_os_wait_count		= 0;
+UNIV_INTERN ib_int64_t	mutex_exit_count		= 0;
+
+/* The global array of wait cells for implementation of the database's own
+mutexes and read-write locks */
+UNIV_INTERN sync_array_t*	sync_primary_wait_array;
+
+/* This variable is set to TRUE when sync_init is called */
+UNIV_INTERN ibool	sync_initialized	= FALSE;
+
+
+typedef struct sync_level_struct	sync_level_t;
+typedef struct sync_thread_struct	sync_thread_t;
+
+#ifdef UNIV_SYNC_DEBUG
+/* The latch levels currently owned by threads are stored in this data
+structure; the size of this array is OS_THREAD_MAX_N */
+
+UNIV_INTERN sync_thread_t*	sync_thread_level_arrays;
+
+/* Mutex protecting sync_thread_level_arrays */
+UNIV_INTERN mutex_t		sync_thread_mutex;
+#endif /* UNIV_SYNC_DEBUG */
+
+/* Global list of database mutexes (not OS mutexes) created. */
+UNIV_INTERN ut_list_base_node_t  mutex_list;
+
+/* Mutex protecting the mutex_list variable */
+UNIV_INTERN mutex_t mutex_list_mutex;
+
+#ifdef UNIV_SYNC_DEBUG
+/* Latching order checks start when this is set TRUE */
+UNIV_INTERN ibool	sync_order_checks_on	= FALSE;
+#endif /* UNIV_SYNC_DEBUG */
+
+struct sync_thread_struct{
+	os_thread_id_t	id;	/* OS thread id */
+	sync_level_t*	levels;	/* level array for this thread; if this is NULL
+				this slot is unused */
+};
+
+/* Number of slots reserved for each OS thread in the sync level array */
+#define SYNC_THREAD_N_LEVELS	10000
+
+struct sync_level_struct{
+	void*	latch;	/* pointer to a mutex or an rw-lock; NULL means that
+			the slot is empty */
+	ulint	level;	/* level of the latch in the latching order */
+};
+
+/**********************************************************************
+Creates, or rather, initializes a mutex object in a specified memory
+location (which must be appropriately aligned). The mutex is initialized
+in the reset state. Explicit freeing of the mutex with mutex_free is
+necessary only if the memory block containing it is freed. */
+UNIV_INTERN
+void
+mutex_create_func(
+/*==============*/
+	mutex_t*	mutex,		/* in: pointer to memory */
+#ifdef UNIV_DEBUG
+	const char*	cmutex_name,	/* in: mutex name */
+# ifdef UNIV_SYNC_DEBUG
+	ulint		level,		/* in: level */
+# endif /* UNIV_SYNC_DEBUG */
+#endif /* UNIV_DEBUG */
+	const char*	cfile_name,	/* in: file name where created */
+	ulint		cline)		/* in: file line where created */
+{
+#if defined(_WIN32) && defined(UNIV_CAN_USE_X86_ASSEMBLER)
+	mutex_reset_lock_word(mutex);
+#elif defined(HAVE_GCC_ATOMIC_BUILTINS)
+	mutex_reset_lock_word(mutex);
+#else
+	os_fast_mutex_init(&(mutex->os_fast_mutex));
+	mutex->lock_word = 0;
+#endif
+	mutex->event = os_event_create(NULL);
+	mutex_set_waiters(mutex, 0);
+#ifdef UNIV_DEBUG
+	mutex->magic_n = MUTEX_MAGIC_N;
+#endif /* UNIV_DEBUG */
+#ifdef UNIV_SYNC_DEBUG
+	mutex->line = 0;
+	mutex->file_name = "not yet reserved";
+	mutex->level = level;
+#endif /* UNIV_SYNC_DEBUG */
+	mutex->cfile_name = cfile_name;
+	mutex->cline = cline;
+#ifndef UNIV_HOTBACKUP
+	mutex->count_os_wait = 0;
+# ifdef UNIV_DEBUG
+	mutex->cmutex_name=	  cmutex_name;
+	mutex->count_using=	  0;
+	mutex->mutex_type=	  0;
+	mutex->lspent_time=	  0;
+	mutex->lmax_spent_time=     0;
+	mutex->count_spin_loop= 0;
+	mutex->count_spin_rounds=   0;
+	mutex->count_os_yield=  0;
+# endif /* UNIV_DEBUG */
+#endif /* !UNIV_HOTBACKUP */
+
+	/* Check that lock_word is aligned; this is important on Intel */
+	ut_ad(((ulint)(&(mutex->lock_word))) % 4 == 0);
+
+	/* NOTE! The very first mutexes are not put to the mutex list */
+
+	if ((mutex == &mutex_list_mutex)
+#ifdef UNIV_SYNC_DEBUG
+	    || (mutex == &sync_thread_mutex)
+#endif /* UNIV_SYNC_DEBUG */
+	    ) {
+
+		return;
+	}
+
+	mutex_enter(&mutex_list_mutex);
+
+	ut_ad(UT_LIST_GET_LEN(mutex_list) == 0
+	      || UT_LIST_GET_FIRST(mutex_list)->magic_n == MUTEX_MAGIC_N);
+
+	UT_LIST_ADD_FIRST(list, mutex_list, mutex);
+
+	mutex_exit(&mutex_list_mutex);
+}
+
+/**********************************************************************
+Calling this function is obligatory only if the memory buffer containing
+the mutex is freed. Removes a mutex object from the mutex list. The mutex
+is checked to be in the reset state. */
+UNIV_INTERN
+void
+mutex_free(
+/*=======*/
+	mutex_t*	mutex)	/* in: mutex */
+{
+	ut_ad(mutex_validate(mutex));
+	ut_a(mutex_get_lock_word(mutex) == 0);
+	ut_a(mutex_get_waiters(mutex) == 0);
+
+	if (mutex != &mutex_list_mutex
+#ifdef UNIV_SYNC_DEBUG
+	    && mutex != &sync_thread_mutex
+#endif /* UNIV_SYNC_DEBUG */
+	    ) {
+
+		mutex_enter(&mutex_list_mutex);
+
+		ut_ad(!UT_LIST_GET_PREV(list, mutex)
+		      || UT_LIST_GET_PREV(list, mutex)->magic_n
+		      == MUTEX_MAGIC_N);
+		ut_ad(!UT_LIST_GET_NEXT(list, mutex)
+		      || UT_LIST_GET_NEXT(list, mutex)->magic_n
+		      == MUTEX_MAGIC_N);
+
+		UT_LIST_REMOVE(list, mutex_list, mutex);
+
+		mutex_exit(&mutex_list_mutex);
+	}
+
+	os_event_free(mutex->event);
+
+#if defined(_WIN32) && defined(UNIV_CAN_USE_X86_ASSEMBLER)
+#elif defined(HAVE_GCC_ATOMIC_BUILTINS)
+#else
+	os_fast_mutex_free(&(mutex->os_fast_mutex));
+#endif
+	/* If we free the mutex protecting the mutex list (freeing is
+	not necessary), we have to reset the magic number AFTER removing
+	it from the list. */
+#ifdef UNIV_DEBUG
+	mutex->magic_n = 0;
+#endif /* UNIV_DEBUG */
+}
+
+/************************************************************************
+NOTE! Use the corresponding macro in the header file, not this function
+directly. Tries to lock the mutex for the current thread. If the lock is not
+acquired immediately, returns with return value 1. */
+UNIV_INTERN
+ulint
+mutex_enter_nowait_func(
+/*====================*/
+					/* out: 0 if succeed, 1 if not */
+	mutex_t*	mutex,		/* in: pointer to mutex */
+	const char*	file_name __attribute__((unused)),
+					/* in: file name where mutex
+					requested */
+	ulint		line __attribute__((unused)))
+					/* in: line where requested */
+{
+	ut_ad(mutex_validate(mutex));
+
+	if (!mutex_test_and_set(mutex)) {
+
+		ut_d(mutex->thread_id = os_thread_get_curr_id());
+#ifdef UNIV_SYNC_DEBUG
+		mutex_set_debug_info(mutex, file_name, line);
+#endif
+
+		return(0);	/* Succeeded! */
+	}
+
+	return(1);
+}
+
+#ifdef UNIV_DEBUG
+/**********************************************************************
+Checks that the mutex has been initialized. */
+UNIV_INTERN
+ibool
+mutex_validate(
+/*===========*/
+	const mutex_t*	mutex)
+{
+	ut_a(mutex);
+	ut_a(mutex->magic_n == MUTEX_MAGIC_N);
+
+	return(TRUE);
+}
+
+/**********************************************************************
+Checks that the current thread owns the mutex. Works only in the debug
+version. */
+UNIV_INTERN
+ibool
+mutex_own(
+/*======*/
+				/* out: TRUE if owns */
+	const mutex_t*	mutex)	/* in: mutex */
+{
+	ut_ad(mutex_validate(mutex));
+
+	return(mutex_get_lock_word(mutex) == 1
+	       && os_thread_eq(mutex->thread_id, os_thread_get_curr_id()));
+}
+#endif /* UNIV_DEBUG */
+
+/**********************************************************************
+Sets the waiters field in a mutex. */
+UNIV_INTERN
+void
+mutex_set_waiters(
+/*==============*/
+	mutex_t*	mutex,	/* in: mutex */
+	ulint		n)	/* in: value to set */
+{
+	volatile ulint*	ptr;		/* declared volatile to ensure that
+					the value is stored to memory */
+	ut_ad(mutex);
+
+	ptr = &(mutex->waiters);
+
+	*ptr = n;		/* Here we assume that the write of a single
+				word in memory is atomic */
+}
+
+/**********************************************************************
+Reserves a mutex for the current thread. If the mutex is reserved, the
+function spins a preset time (controlled by SYNC_SPIN_ROUNDS), waiting
+for the mutex before suspending the thread. */
+UNIV_INTERN
+void
+mutex_spin_wait(
+/*============*/
+	mutex_t*	mutex,		/* in: pointer to mutex */
+	const char*	file_name,	/* in: file name where mutex
+					requested */
+	ulint		line)		/* in: line where requested */
+{
+	ulint	   index; /* index of the reserved wait cell */
+	ulint	   i;	  /* spin round count */
+#if defined UNIV_DEBUG && !defined UNIV_HOTBACKUP
+	ib_int64_t lstart_time = 0, lfinish_time; /* for timing os_wait */
+	ulint ltime_diff;
+	ulint sec;
+	ulint ms;
+	uint timer_started = 0;
+#endif /* UNIV_DEBUG && !UNIV_HOTBACKUP */
+	ut_ad(mutex);
+
+	/* This update is not thread safe, but we don't mind if the count
+	isn't exact. Moved out of ifdef that follows because we are willing
+	to sacrifice the cost of counting this as the data is valuable.
+	Count the number of calls to mutex_spin_wait. */
+	mutex_spin_wait_count++;
+
+mutex_loop:
+
+	i = 0;
+
+	/* Spin waiting for the lock word to become zero. Note that we do
+	not have to assume that the read access to the lock word is atomic,
+	as the actual locking is always committed with atomic test-and-set.
+	In reality, however, all processors probably have an atomic read of
+	a memory word. */
+
+spin_loop:
+#if defined UNIV_DEBUG && !defined UNIV_HOTBACKUP
+	mutex->count_spin_loop++;
+#endif /* UNIV_DEBUG && !UNIV_HOTBACKUP */
+
+	while (mutex_get_lock_word(mutex) != 0 && i < SYNC_SPIN_ROUNDS) {
+		if (srv_spin_wait_delay) {
+			ut_delay(ut_rnd_interval(0, srv_spin_wait_delay));
+		}
+
+		i++;
+	}
+
+	if (i == SYNC_SPIN_ROUNDS) {
+#if defined UNIV_DEBUG && !defined UNIV_HOTBACKUP
+		mutex->count_os_yield++;
+		if (timed_mutexes == 1 && timer_started==0) {
+			ut_usectime(&sec, &ms);
+			lstart_time= (ib_int64_t)sec * 1000000 + ms;
+			timer_started = 1;
+		}
+#endif /* UNIV_DEBUG && !UNIV_HOTBACKUP */
+		os_thread_yield();
+	}
+
+#ifdef UNIV_SRV_PRINT_LATCH_WAITS
+	fprintf(stderr,
+		"Thread %lu spin wait mutex at %p"
+		" cfile %s cline %lu rnds %lu\n",
+		(ulong) os_thread_pf(os_thread_get_curr_id()), (void*) mutex,
+		mutex->cfile_name, (ulong) mutex->cline, (ulong) i);
+#endif
+
+	mutex_spin_round_count += i;
+
+#if defined UNIV_DEBUG && !defined UNIV_HOTBACKUP
+	mutex->count_spin_rounds += i;
+#endif /* UNIV_DEBUG && !UNIV_HOTBACKUP */
+
+	if (mutex_test_and_set(mutex) == 0) {
+		/* Succeeded! */
+
+		ut_d(mutex->thread_id = os_thread_get_curr_id());
+#ifdef UNIV_SYNC_DEBUG
+		mutex_set_debug_info(mutex, file_name, line);
+#endif
+
+		goto finish_timing;
+	}
+
+	/* We may end up with a situation where lock_word is 0 but the OS
+	fast mutex is still reserved. On FreeBSD the OS does not seem to
+	schedule a thread which is constantly calling pthread_mutex_trylock
+	(in mutex_test_and_set implementation). Then we could end up
+	spinning here indefinitely. The following 'i++' stops this infinite
+	spin. */
+
+	i++;
+
+	if (i < SYNC_SPIN_ROUNDS) {
+		goto spin_loop;
+	}
+
+	sync_array_reserve_cell(sync_primary_wait_array, mutex,
+				SYNC_MUTEX, file_name, line, &index);
+
+	/* The memory order of the array reservation and the change in the
+	waiters field is important: when we suspend a thread, we first
+	reserve the cell and then set waiters field to 1. When threads are
+	released in mutex_exit, the waiters field is first set to zero and
+	then the event is set to the signaled state. */
+
+	mutex_set_waiters(mutex, 1);
+
+	/* Try to reserve still a few times */
+	for (i = 0; i < 4; i++) {
+		if (mutex_test_and_set(mutex) == 0) {
+			/* Succeeded! Free the reserved wait cell */
+
+			sync_array_free_cell(sync_primary_wait_array, index);
+
+			ut_d(mutex->thread_id = os_thread_get_curr_id());
+#ifdef UNIV_SYNC_DEBUG
+			mutex_set_debug_info(mutex, file_name, line);
+#endif
+
+#ifdef UNIV_SRV_PRINT_LATCH_WAITS
+			fprintf(stderr, "Thread %lu spin wait succeeds at 2:"
+				" mutex at %p\n",
+				(ulong) os_thread_pf(os_thread_get_curr_id()),
+				(void*) mutex);
+#endif
+
+			goto finish_timing;
+
+			/* Note that in this case we leave the waiters field
+			set to 1. We cannot reset it to zero, as we do not
+			know if there are other waiters. */
+		}
+	}
+
+	/* Now we know that there has been some thread holding the mutex
+	after the change in the wait array and the waiters field was made.
+	Now there is no risk of infinite wait on the event. */
+
+#ifdef UNIV_SRV_PRINT_LATCH_WAITS
+	fprintf(stderr,
+		"Thread %lu OS wait mutex at %p cfile %s cline %lu rnds %lu\n",
+		(ulong) os_thread_pf(os_thread_get_curr_id()), (void*) mutex,
+		mutex->cfile_name, (ulong) mutex->cline, (ulong) i);
+#endif
+
+	mutex_os_wait_count++;
+
+#ifndef UNIV_HOTBACKUP
+	mutex->count_os_wait++;
+# ifdef UNIV_DEBUG
+	/* !!!!! Sometimes os_wait can be called without os_thread_yield */
+
+	if (timed_mutexes == 1 && timer_started==0) {
+		ut_usectime(&sec, &ms);
+		lstart_time= (ib_int64_t)sec * 1000000 + ms;
+		timer_started = 1;
+	}
+# endif /* UNIV_DEBUG */
+#endif /* !UNIV_HOTBACKUP */
+
+	sync_array_wait_event(sync_primary_wait_array, index);
+	goto mutex_loop;
+
+finish_timing:
+#if defined UNIV_DEBUG && !defined UNIV_HOTBACKUP
+	if (timed_mutexes == 1 && timer_started==1) {
+		ut_usectime(&sec, &ms);
+		lfinish_time= (ib_int64_t)sec * 1000000 + ms;
+
+		ltime_diff= (ulint) (lfinish_time - lstart_time);
+		mutex->lspent_time += ltime_diff;
+
+		if (mutex->lmax_spent_time < ltime_diff) {
+			mutex->lmax_spent_time= ltime_diff;
+		}
+	}
+#endif /* UNIV_DEBUG && !UNIV_HOTBACKUP */
+	return;
+}
+
+/**********************************************************************
+Releases the threads waiting in the primary wait array for this mutex. */
+UNIV_INTERN
+void
+mutex_signal_object(
+/*================*/
+	mutex_t*	mutex)	/* in: mutex */
+{
+	mutex_set_waiters(mutex, 0);
+
+	/* The memory order of resetting the waiters field and
+	signaling the object is important. See LEMMA 1 above. */
+	os_event_set(mutex->event);
+	sync_array_object_signalled(sync_primary_wait_array);
+}
+
+#ifdef UNIV_SYNC_DEBUG
+/**********************************************************************
+Sets the debug information for a reserved mutex. */
+UNIV_INTERN
+void
+mutex_set_debug_info(
+/*=================*/
+	mutex_t*	mutex,		/* in: mutex */
+	const char*	file_name,	/* in: file where requested */
+	ulint		line)		/* in: line where requested */
+{
+	ut_ad(mutex);
+	ut_ad(file_name);
+
+	sync_thread_add_level(mutex, mutex->level);
+
+	mutex->file_name = file_name;
+	mutex->line	 = line;
+}
+
+/**********************************************************************
+Gets the debug information for a reserved mutex. */
+UNIV_INTERN
+void
+mutex_get_debug_info(
+/*=================*/
+	mutex_t*	mutex,		/* in: mutex */
+	const char**	file_name,	/* out: file where requested */
+	ulint*		line,		/* out: line where requested */
+	os_thread_id_t* thread_id)	/* out: id of the thread which owns
+					the mutex */
+{
+	ut_ad(mutex);
+
+	*file_name = mutex->file_name;
+	*line	   = mutex->line;
+	*thread_id = mutex->thread_id;
+}
+
+/**********************************************************************
+Prints debug info of currently reserved mutexes. */
+static
+void
+mutex_list_print_info(
+/*==================*/
+	FILE*	file)		/* in: file where to print */
+{
+	mutex_t*	mutex;
+	const char*	file_name;
+	ulint		line;
+	os_thread_id_t	thread_id;
+	ulint		count		= 0;
+
+	fputs("----------\n"
+	      "MUTEX INFO\n"
+	      "----------\n", file);
+
+	mutex_enter(&mutex_list_mutex);
+
+	mutex = UT_LIST_GET_FIRST(mutex_list);
+
+	while (mutex != NULL) {
+		count++;
+
+		if (mutex_get_lock_word(mutex) != 0) {
+			mutex_get_debug_info(mutex, &file_name, &line,
+					     &thread_id);
+			fprintf(file,
+				"Locked mutex: addr %p thread %ld"
+				" file %s line %ld\n",
+				(void*) mutex, os_thread_pf(thread_id),
+				file_name, line);
+		}
+
+		mutex = UT_LIST_GET_NEXT(list, mutex);
+	}
+
+	fprintf(file, "Total number of mutexes %ld\n", count);
+
+	mutex_exit(&mutex_list_mutex);
+}
+
+/**********************************************************************
+Counts currently reserved mutexes. Works only in the debug version. */
+UNIV_INTERN
+ulint
+mutex_n_reserved(void)
+/*==================*/
+{
+	mutex_t*	mutex;
+	ulint		count		= 0;
+
+	mutex_enter(&mutex_list_mutex);
+
+	mutex = UT_LIST_GET_FIRST(mutex_list);
+
+	while (mutex != NULL) {
+		if (mutex_get_lock_word(mutex) != 0) {
+
+			count++;
+		}
+
+		mutex = UT_LIST_GET_NEXT(list, mutex);
+	}
+
+	mutex_exit(&mutex_list_mutex);
+
+	ut_a(count >= 1);
+
+	return(count - 1); /* Subtract one, because this function itself
+			   was holding one mutex (mutex_list_mutex) */
+}
+
+/**********************************************************************
+Returns TRUE if no mutex or rw-lock is currently locked. Works only in
+the debug version. */
+UNIV_INTERN
+ibool
+sync_all_freed(void)
+/*================*/
+{
+	return(mutex_n_reserved() + rw_lock_n_locked() == 0);
+}
+
+/**********************************************************************
+Gets the value in the nth slot in the thread level arrays. */
+static
+sync_thread_t*
+sync_thread_level_arrays_get_nth(
+/*=============================*/
+			/* out: pointer to thread slot */
+	ulint	n)	/* in: slot number */
+{
+	ut_ad(n < OS_THREAD_MAX_N);
+
+	return(sync_thread_level_arrays + n);
+}
+
+/**********************************************************************
+Looks for the thread slot for the calling thread. */
+static
+sync_thread_t*
+sync_thread_level_arrays_find_slot(void)
+/*====================================*/
+			/* out: pointer to thread slot, NULL if not found */
+
+{
+	sync_thread_t*	slot;
+	os_thread_id_t	id;
+	ulint		i;
+
+	id = os_thread_get_curr_id();
+
+	for (i = 0; i < OS_THREAD_MAX_N; i++) {
+
+		slot = sync_thread_level_arrays_get_nth(i);
+
+		if (slot->levels && os_thread_eq(slot->id, id)) {
+
+			return(slot);
+		}
+	}
+
+	return(NULL);
+}
+
+/**********************************************************************
+Looks for an unused thread slot. */
+static
+sync_thread_t*
+sync_thread_level_arrays_find_free(void)
+/*====================================*/
+			/* out: pointer to thread slot */
+
+{
+	sync_thread_t*	slot;
+	ulint		i;
+
+	for (i = 0; i < OS_THREAD_MAX_N; i++) {
+
+		slot = sync_thread_level_arrays_get_nth(i);
+
+		if (slot->levels == NULL) {
+
+			return(slot);
+		}
+	}
+
+	return(NULL);
+}
+
+/**********************************************************************
+Gets the value in the nth slot in the thread level array. */
+static
+sync_level_t*
+sync_thread_levels_get_nth(
+/*=======================*/
+				/* out: pointer to level slot */
+	sync_level_t*	arr,	/* in: pointer to level array for an OS
+				thread */
+	ulint		n)	/* in: slot number */
+{
+	ut_ad(n < SYNC_THREAD_N_LEVELS);
+
+	return(arr + n);
+}
+
+/**********************************************************************
+Checks if all the level values stored in the level array are greater than
+the given limit. */
+static
+ibool
+sync_thread_levels_g(
+/*=================*/
+				/* out: TRUE if all greater */
+	sync_level_t*	arr,	/* in: pointer to level array for an OS
+				thread */
+	ulint		limit)	/* in: level limit */
+{
+	sync_level_t*	slot;
+	rw_lock_t*	lock;
+	mutex_t*	mutex;
+	ulint		i;
+
+	for (i = 0; i < SYNC_THREAD_N_LEVELS; i++) {
+
+		slot = sync_thread_levels_get_nth(arr, i);
+
+		if (slot->latch != NULL) {
+			if (slot->level <= limit) {
+
+				lock = slot->latch;
+				mutex = slot->latch;
+
+				fprintf(stderr,
+					"InnoDB: sync levels should be"
+					" > %lu but a level is %lu\n",
+					(ulong) limit, (ulong) slot->level);
+
+				if (mutex->magic_n == MUTEX_MAGIC_N) {
+					fprintf(stderr,
+						"Mutex created at %s %lu\n",
+						mutex->cfile_name,
+						(ulong) mutex->cline);
+
+					if (mutex_get_lock_word(mutex) != 0) {
+						const char*	file_name;
+						ulint		line;
+						os_thread_id_t	thread_id;
+
+						mutex_get_debug_info(
+							mutex, &file_name,
+							&line, &thread_id);
+
+						fprintf(stderr,
+							"InnoDB: Locked mutex:"
+							" addr %p thread %ld"
+							" file %s line %ld\n",
+							(void*) mutex,
+							os_thread_pf(
+								thread_id),
+							file_name,
+							(ulong) line);
+					} else {
+						fputs("Not locked\n", stderr);
+					}
+				} else {
+					rw_lock_print(lock);
+				}
+
+				return(FALSE);
+			}
+		}
+	}
+
+	return(TRUE);
+}
+
+/**********************************************************************
+Checks if the level value is stored in the level array. */
+static
+ibool
+sync_thread_levels_contain(
+/*=======================*/
+				/* out: TRUE if stored */
+	sync_level_t*	arr,	/* in: pointer to level array for an OS
+				thread */
+	ulint		level)	/* in: level */
+{
+	sync_level_t*	slot;
+	ulint		i;
+
+	for (i = 0; i < SYNC_THREAD_N_LEVELS; i++) {
+
+		slot = sync_thread_levels_get_nth(arr, i);
+
+		if (slot->latch != NULL) {
+			if (slot->level == level) {
+
+				return(TRUE);
+			}
+		}
+	}
+
+	return(FALSE);
+}
+
+/**********************************************************************
+Checks that the level array for the current thread is empty. */
+UNIV_INTERN
+ibool
+sync_thread_levels_empty_gen(
+/*=========================*/
+					/* out: TRUE if empty except the
+					exceptions specified below */
+	ibool	dict_mutex_allowed)	/* in: TRUE if dictionary mutex is
+					allowed to be owned by the thread,
+					also purge_is_running mutex is
+					allowed */
+{
+	sync_level_t*	arr;
+	sync_thread_t*	thread_slot;
+	sync_level_t*	slot;
+	ulint		i;
+
+	if (!sync_order_checks_on) {
+
+		return(TRUE);
+	}
+
+	mutex_enter(&sync_thread_mutex);
+
+	thread_slot = sync_thread_level_arrays_find_slot();
+
+	if (thread_slot == NULL) {
+
+		mutex_exit(&sync_thread_mutex);
+
+		return(TRUE);
+	}
+
+	arr = thread_slot->levels;
+
+	for (i = 0; i < SYNC_THREAD_N_LEVELS; i++) {
+
+		slot = sync_thread_levels_get_nth(arr, i);
+
+		if (slot->latch != NULL
+		    && (!dict_mutex_allowed
+			|| (slot->level != SYNC_DICT
+			    && slot->level != SYNC_DICT_OPERATION))) {
+
+			mutex_exit(&sync_thread_mutex);
+			ut_error;
+
+			return(FALSE);
+		}
+	}
+
+	mutex_exit(&sync_thread_mutex);
+
+	return(TRUE);
+}
+
+/**********************************************************************
+Checks that the level array for the current thread is empty. */
+UNIV_INTERN
+ibool
+sync_thread_levels_empty(void)
+/*==========================*/
+			/* out: TRUE if empty */
+{
+	return(sync_thread_levels_empty_gen(FALSE));
+}
+
+/**********************************************************************
+Adds a latch and its level in the thread level array. Allocates the memory
+for the array if called first time for this OS thread. Makes the checks
+against other latch levels stored in the array for this thread. */
+UNIV_INTERN
+void
+sync_thread_add_level(
+/*==================*/
+	void*	latch,	/* in: pointer to a mutex or an rw-lock */
+	ulint	level)	/* in: level in the latching order; if
+			SYNC_LEVEL_VARYING, nothing is done */
+{
+	sync_level_t*	array;
+	sync_level_t*	slot;
+	sync_thread_t*	thread_slot;
+	ulint		i;
+
+	if (!sync_order_checks_on) {
+
+		return;
+	}
+
+	if ((latch == (void*)&sync_thread_mutex)
+	    || (latch == (void*)&mutex_list_mutex)
+	    || (latch == (void*)&rw_lock_debug_mutex)
+	    || (latch == (void*)&rw_lock_list_mutex)) {
+
+		return;
+	}
+
+	if (level == SYNC_LEVEL_VARYING) {
+
+		return;
+	}
+
+	mutex_enter(&sync_thread_mutex);
+
+	thread_slot = sync_thread_level_arrays_find_slot();
+
+	if (thread_slot == NULL) {
+		/* We have to allocate the level array for a new thread */
+		array = ut_malloc(sizeof(sync_level_t) * SYNC_THREAD_N_LEVELS);
+
+		thread_slot = sync_thread_level_arrays_find_free();
+
+		thread_slot->id = os_thread_get_curr_id();
+		thread_slot->levels = array;
+
+		for (i = 0; i < SYNC_THREAD_N_LEVELS; i++) {
+
+			slot = sync_thread_levels_get_nth(array, i);
+
+			slot->latch = NULL;
+		}
+	}
+
+	array = thread_slot->levels;
+
+	/* NOTE that there is a problem with _NODE and _LEAF levels: if the
+	B-tree height changes, then a leaf can change to an internal node
+	or the other way around. We do not know at present if this can cause
+	unnecessary assertion failures below. */
+
+	switch (level) {
+	case SYNC_NO_ORDER_CHECK:
+	case SYNC_EXTERN_STORAGE:
+	case SYNC_TREE_NODE_FROM_HASH:
+		/* Do no order checking */
+		break;
+	case SYNC_MEM_POOL:
+	case SYNC_MEM_HASH:
+	case SYNC_RECV:
+	case SYNC_WORK_QUEUE:
+	case SYNC_LOG:
+	case SYNC_THR_LOCAL:
+	case SYNC_ANY_LATCH:
+	case SYNC_TRX_SYS_HEADER:
+	case SYNC_FILE_FORMAT_TAG:
+	case SYNC_DOUBLEWRITE:
+	case SYNC_BUF_POOL:
+	case SYNC_SEARCH_SYS:
+	case SYNC_SEARCH_SYS_CONF:
+	case SYNC_TRX_LOCK_HEAP:
+	case SYNC_KERNEL:
+	case SYNC_IBUF_BITMAP_MUTEX:
+	case SYNC_RSEG:
+	case SYNC_TRX_UNDO:
+	case SYNC_PURGE_LATCH:
+	case SYNC_PURGE_SYS:
+	case SYNC_DICT_AUTOINC_MUTEX:
+	case SYNC_DICT_OPERATION:
+	case SYNC_DICT_HEADER:
+	case SYNC_TRX_I_S_RWLOCK:
+	case SYNC_TRX_I_S_LAST_READ:
+		if (!sync_thread_levels_g(array, level)) {
+			fprintf(stderr,
+				"InnoDB: sync_thread_levels_g(array, %lu)"
+				" does not hold!\n", level);
+			ut_error;
+		}
+		break;
+	case SYNC_BUF_BLOCK:
+		/* Either the thread must own the buffer pool mutex
+		(buf_pool_mutex), or it is allowed to latch only ONE
+		buffer block (block->mutex or buf_pool_zip_mutex). */
+		ut_a((sync_thread_levels_contain(array, SYNC_BUF_POOL)
+		      && sync_thread_levels_g(array, SYNC_BUF_BLOCK - 1))
+		     || sync_thread_levels_g(array, SYNC_BUF_BLOCK));
+		break;
+	case SYNC_REC_LOCK:
+		ut_a((sync_thread_levels_contain(array, SYNC_KERNEL)
+		      && sync_thread_levels_g(array, SYNC_REC_LOCK - 1))
+		     || sync_thread_levels_g(array, SYNC_REC_LOCK));
+		break;
+	case SYNC_IBUF_BITMAP:
+		/* Either the thread must own the master mutex to all
+		the bitmap pages, or it is allowed to latch only ONE
+		bitmap page. */
+		ut_a((sync_thread_levels_contain(array, SYNC_IBUF_BITMAP_MUTEX)
+		      && sync_thread_levels_g(array, SYNC_IBUF_BITMAP - 1))
+		     || sync_thread_levels_g(array, SYNC_IBUF_BITMAP));
+		break;
+	case SYNC_FSP_PAGE:
+		ut_a(sync_thread_levels_contain(array, SYNC_FSP));
+		break;
+	case SYNC_FSP:
+		ut_a(sync_thread_levels_contain(array, SYNC_FSP)
+		     || sync_thread_levels_g(array, SYNC_FSP));
+		break;
+	case SYNC_TRX_UNDO_PAGE:
+		ut_a(sync_thread_levels_contain(array, SYNC_TRX_UNDO)
+		     || sync_thread_levels_contain(array, SYNC_RSEG)
+		     || sync_thread_levels_contain(array, SYNC_PURGE_SYS)
+		     || sync_thread_levels_g(array, SYNC_TRX_UNDO_PAGE));
+		break;
+	case SYNC_RSEG_HEADER:
+		ut_a(sync_thread_levels_contain(array, SYNC_RSEG));
+		break;
+	case SYNC_RSEG_HEADER_NEW:
+		ut_a(sync_thread_levels_contain(array, SYNC_KERNEL)
+		     && sync_thread_levels_contain(array, SYNC_FSP_PAGE));
+		break;
+	case SYNC_TREE_NODE:
+		ut_a(sync_thread_levels_contain(array, SYNC_INDEX_TREE)
+		     || sync_thread_levels_contain(array, SYNC_DICT_OPERATION)
+		     || sync_thread_levels_g(array, SYNC_TREE_NODE - 1));
+		break;
+	case SYNC_TREE_NODE_NEW:
+		ut_a(sync_thread_levels_contain(array, SYNC_FSP_PAGE)
+		     || sync_thread_levels_contain(array, SYNC_IBUF_MUTEX));
+		break;
+	case SYNC_INDEX_TREE:
+		ut_a((sync_thread_levels_contain(array, SYNC_IBUF_MUTEX)
+		      && sync_thread_levels_contain(array, SYNC_FSP)
+		      && sync_thread_levels_g(array, SYNC_FSP_PAGE - 1))
+		     || sync_thread_levels_g(array, SYNC_TREE_NODE - 1));
+		break;
+	case SYNC_IBUF_MUTEX:
+		ut_a(sync_thread_levels_g(array, SYNC_FSP_PAGE - 1));
+		break;
+	case SYNC_IBUF_PESS_INSERT_MUTEX:
+		ut_a(sync_thread_levels_g(array, SYNC_FSP - 1)
+		     && !sync_thread_levels_contain(array, SYNC_IBUF_MUTEX));
+		break;
+	case SYNC_IBUF_HEADER:
+		ut_a(sync_thread_levels_g(array, SYNC_FSP - 1)
+		     && !sync_thread_levels_contain(array, SYNC_IBUF_MUTEX)
+		     && !sync_thread_levels_contain(
+			     array, SYNC_IBUF_PESS_INSERT_MUTEX));
+		break;
+	case SYNC_DICT:
+#ifdef UNIV_DEBUG
+		ut_a(buf_debug_prints
+		     || sync_thread_levels_g(array, SYNC_DICT));
+#else /* UNIV_DEBUG */
+		ut_a(sync_thread_levels_g(array, SYNC_DICT));
+#endif /* UNIV_DEBUG */
+		break;
+	default:
+		ut_error;
+	}
+
+	for (i = 0; i < SYNC_THREAD_N_LEVELS; i++) {
+
+		slot = sync_thread_levels_get_nth(array, i);
+
+		if (slot->latch == NULL) {
+			slot->latch = latch;
+			slot->level = level;
+
+			break;
+		}
+	}
+
+	ut_a(i < SYNC_THREAD_N_LEVELS);
+
+	mutex_exit(&sync_thread_mutex);
+}
+
+/**********************************************************************
+Removes a latch from the thread level array if it is found there. */
+UNIV_INTERN
+ibool
+sync_thread_reset_level(
+/*====================*/
+			/* out: TRUE if found from the array; it is an error
+			if the latch is not found */
+	void*	latch)	/* in: pointer to a mutex or an rw-lock */
+{
+	sync_level_t*	array;
+	sync_level_t*	slot;
+	sync_thread_t*	thread_slot;
+	ulint		i;
+
+	if (!sync_order_checks_on) {
+
+		return(FALSE);
+	}
+
+	if ((latch == (void*)&sync_thread_mutex)
+	    || (latch == (void*)&mutex_list_mutex)
+	    || (latch == (void*)&rw_lock_debug_mutex)
+	    || (latch == (void*)&rw_lock_list_mutex)) {
+
+		return(FALSE);
+	}
+
+	mutex_enter(&sync_thread_mutex);
+
+	thread_slot = sync_thread_level_arrays_find_slot();
+
+	if (thread_slot == NULL) {
+
+		ut_error;
+
+		mutex_exit(&sync_thread_mutex);
+		return(FALSE);
+	}
+
+	array = thread_slot->levels;
+
+	for (i = 0; i < SYNC_THREAD_N_LEVELS; i++) {
+
+		slot = sync_thread_levels_get_nth(array, i);
+
+		if (slot->latch == latch) {
+			slot->latch = NULL;
+
+			mutex_exit(&sync_thread_mutex);
+
+			return(TRUE);
+		}
+	}
+
+	if (((mutex_t*) latch)->magic_n != MUTEX_MAGIC_N) {
+		rw_lock_t*	rw_lock;
+
+		rw_lock = (rw_lock_t*) latch;
+
+		if (rw_lock->level == SYNC_LEVEL_VARYING) {
+			mutex_exit(&sync_thread_mutex);
+
+			return(TRUE);
+		}
+	}
+
+	ut_error;
+
+	mutex_exit(&sync_thread_mutex);
+
+	return(FALSE);
+}
+#endif /* UNIV_SYNC_DEBUG */
+
+/**********************************************************************
+Initializes the synchronization data structures. */
+UNIV_INTERN
+void
+sync_init(void)
+/*===========*/
+{
+#ifdef UNIV_SYNC_DEBUG
+	sync_thread_t*	thread_slot;
+	ulint		i;
+#endif /* UNIV_SYNC_DEBUG */
+
+	ut_a(sync_initialized == FALSE);
+
+	sync_initialized = TRUE;
+
+	/* Create the primary system wait array which is protected by an OS
+	mutex */
+
+	sync_primary_wait_array = sync_array_create(OS_THREAD_MAX_N,
+						    SYNC_ARRAY_OS_MUTEX);
+#ifdef UNIV_SYNC_DEBUG
+	/* Create the thread latch level array where the latch levels
+	are stored for each OS thread */
+
+	sync_thread_level_arrays = ut_malloc(OS_THREAD_MAX_N
+					     * sizeof(sync_thread_t));
+	for (i = 0; i < OS_THREAD_MAX_N; i++) {
+
+		thread_slot = sync_thread_level_arrays_get_nth(i);
+		thread_slot->levels = NULL;
+	}
+#endif /* UNIV_SYNC_DEBUG */
+	/* Init the mutex list and create the mutex to protect it. */
+
+	UT_LIST_INIT(mutex_list);
+	mutex_create(&mutex_list_mutex, SYNC_NO_ORDER_CHECK);
+#ifdef UNIV_SYNC_DEBUG
+	mutex_create(&sync_thread_mutex, SYNC_NO_ORDER_CHECK);
+#endif /* UNIV_SYNC_DEBUG */
+
+	/* Init the rw-lock list and create the mutex to protect it. */
+
+	UT_LIST_INIT(rw_lock_list);
+	mutex_create(&rw_lock_list_mutex, SYNC_NO_ORDER_CHECK);
+
+#ifdef UNIV_SYNC_DEBUG
+	mutex_create(&rw_lock_debug_mutex, SYNC_NO_ORDER_CHECK);
+
+	rw_lock_debug_event = os_event_create(NULL);
+	rw_lock_debug_waiters = FALSE;
+#endif /* UNIV_SYNC_DEBUG */
+}
+
+/**********************************************************************
+Frees the resources in InnoDB's own synchronization data structures. Use
+os_sync_free() after calling this. */
+UNIV_INTERN
+void
+sync_close(void)
+/*===========*/
+{
+	mutex_t*	mutex;
+
+	sync_array_free(sync_primary_wait_array);
+
+	mutex = UT_LIST_GET_FIRST(mutex_list);
+
+	while (mutex) {
+		mutex_free(mutex);
+		mutex = UT_LIST_GET_FIRST(mutex_list);
+	}
+
+	mutex_free(&mutex_list_mutex);
+#ifdef UNIV_SYNC_DEBUG
+	mutex_free(&sync_thread_mutex);
+#endif /* UNIV_SYNC_DEBUG */
+}
+
+/***********************************************************************
+Prints wait info of the sync system. */
+UNIV_INTERN
+void
+sync_print_wait_info(
+/*=================*/
+	FILE*	file)		/* in: file where to print */
+{
+#ifdef UNIV_SYNC_DEBUG
+	fprintf(file, "Mutex exits %llu, rws exits %llu, rwx exits %llu\n",
+		mutex_exit_count, rw_s_exit_count, rw_x_exit_count);
+#endif
+
+	fprintf(file,
+		"Mutex spin waits %llu, rounds %llu, OS waits %llu\n"
+		"RW-shared spins %llu, OS waits %llu;"
+		" RW-excl spins %llu, OS waits %llu\n",
+		mutex_spin_wait_count,
+		mutex_spin_round_count,
+		mutex_os_wait_count,
+		rw_s_spin_wait_count,
+		rw_s_os_wait_count,
+		rw_x_spin_wait_count,
+		rw_x_os_wait_count);
+
+	fprintf(file,
+		"Spin rounds per wait: %.2f mutex, %.2f RW-shared, "
+		"%.2f RW-excl\n",
+		(double) mutex_spin_round_count /
+		(mutex_spin_wait_count ? mutex_spin_wait_count : 1),
+		(double) rw_s_spin_round_count /
+		(rw_s_spin_wait_count ? rw_s_spin_wait_count : 1),
+		(double) rw_x_spin_round_count /
+		(rw_x_spin_wait_count ? rw_x_spin_wait_count : 1));
+}
+
+/***********************************************************************
+Prints info of the sync system. */
+UNIV_INTERN
+void
+sync_print(
+/*=======*/
+	FILE*	file)		/* in: file where to print */
+{
+#ifdef UNIV_SYNC_DEBUG
+	mutex_list_print_info(file);
+
+	rw_lock_list_print_info(file);
+#endif /* UNIV_SYNC_DEBUG */
+
+	sync_array_print_info(file, sync_primary_wait_array);
+
+	sync_print_wait_info(file);
+}
diff --git a/storage/xtradb/thr/thr0loc.c b/storage/xtradb/thr/thr0loc.c
new file mode 100644
index 00000000000..aea1992a921
--- /dev/null
+++ b/storage/xtradb/thr/thr0loc.c
@@ -0,0 +1,273 @@
+/*****************************************************************************
+
+Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The thread local storage
+
+Created 10/5/1995 Heikki Tuuri
+*******************************************************/
+
+#include "thr0loc.h"
+#ifdef UNIV_NONINL
+#include "thr0loc.ic"
+#endif
+
+#include "sync0sync.h"
+#include "hash0hash.h"
+#include "mem0mem.h"
+#include "srv0srv.h"
+
+/*
+	IMPLEMENTATION OF THREAD LOCAL STORAGE
+	======================================
+
+The threads sometimes need private data which depends on the thread id.
+This is implemented as a hash table, where the hash value is calculated
+from the thread id, to prepare for a large number of threads. The hash table
+is protected by a mutex. If you need modify the program and put new data to
+the thread local storage, just add it to struct thr_local_struct in the
+header file. */
+
+/* Mutex protecting the local storage hash table */
+static mutex_t		thr_local_mutex;
+
+/* The hash table. The module is not yet initialized when it is NULL. */
+static hash_table_t*	thr_local_hash	= NULL;
+ulint		thr_local_hash_n_nodes = 0;
+
+/* The private data for each thread should be put to
+the structure below and the accessor functions written
+for the field. */
+typedef struct thr_local_struct thr_local_t;
+
+struct thr_local_struct{
+	os_thread_id_t	id;	/* id of the thread which owns this struct */
+	os_thread_t	handle;	/* operating system handle to the thread */
+	ulint		slot_no;/* the index of the slot in the thread table
+				for this thread */
+	ibool		in_ibuf;/* TRUE if the the thread is doing an ibuf
+				operation */
+	hash_node_t	hash;	/* hash chain node */
+	ulint		magic_n;
+};
+
+#define THR_LOCAL_MAGIC_N	1231234
+
+/***********************************************************************
+Returns the local storage struct for a thread. */
+static
+thr_local_t*
+thr_local_get(
+/*==========*/
+				/* out: local storage */
+	os_thread_id_t	id)	/* in: thread id of the thread */
+{
+	thr_local_t*	local;
+
+try_again:
+	ut_ad(thr_local_hash);
+	ut_ad(mutex_own(&thr_local_mutex));
+
+	/* Look for the local struct in the hash table */
+
+	local = NULL;
+
+	HASH_SEARCH(hash, thr_local_hash, os_thread_pf(id),
+		    thr_local_t*, local,, os_thread_eq(local->id, id));
+	if (local == NULL) {
+		mutex_exit(&thr_local_mutex);
+
+		thr_local_create();
+
+		mutex_enter(&thr_local_mutex);
+
+		goto try_again;
+	}
+
+	ut_ad(local->magic_n == THR_LOCAL_MAGIC_N);
+
+	return(local);
+}
+
+/***********************************************************************
+Gets the slot number in the thread table of a thread. */
+UNIV_INTERN
+ulint
+thr_local_get_slot_no(
+/*==================*/
+				/* out: slot number */
+	os_thread_id_t	id)	/* in: thread id of the thread */
+{
+	ulint		slot_no;
+	thr_local_t*	local;
+
+	mutex_enter(&thr_local_mutex);
+
+	local = thr_local_get(id);
+
+	slot_no = local->slot_no;
+
+	mutex_exit(&thr_local_mutex);
+
+	return(slot_no);
+}
+
+/***********************************************************************
+Sets the slot number in the thread table of a thread. */
+UNIV_INTERN
+void
+thr_local_set_slot_no(
+/*==================*/
+	os_thread_id_t	id,	/* in: thread id of the thread */
+	ulint		slot_no)/* in: slot number */
+{
+	thr_local_t*	local;
+
+	mutex_enter(&thr_local_mutex);
+
+	local = thr_local_get(id);
+
+	local->slot_no = slot_no;
+
+	mutex_exit(&thr_local_mutex);
+}
+
+/***********************************************************************
+Returns pointer to the 'in_ibuf' field within the current thread local
+storage. */
+UNIV_INTERN
+ibool*
+thr_local_get_in_ibuf_field(void)
+/*=============================*/
+			/* out: pointer to the in_ibuf field */
+{
+	thr_local_t*	local;
+
+	mutex_enter(&thr_local_mutex);
+
+	local = thr_local_get(os_thread_get_curr_id());
+
+	mutex_exit(&thr_local_mutex);
+
+	return(&(local->in_ibuf));
+}
+
+/***********************************************************************
+Creates a local storage struct for the calling new thread. */
+UNIV_INTERN
+void
+thr_local_create(void)
+/*==================*/
+{
+	thr_local_t*	local;
+
+	if (thr_local_hash == NULL) {
+		thr_local_init();
+	}
+
+	local = mem_alloc(sizeof(thr_local_t));
+
+	local->id = os_thread_get_curr_id();
+	local->handle = os_thread_get_curr();
+	local->magic_n = THR_LOCAL_MAGIC_N;
+
+	local->in_ibuf = FALSE;
+
+	mutex_enter(&thr_local_mutex);
+
+	HASH_INSERT(thr_local_t, hash, thr_local_hash,
+		    os_thread_pf(os_thread_get_curr_id()),
+		    local);
+
+	thr_local_hash_n_nodes++;
+	mutex_exit(&thr_local_mutex);
+}
+
+/***********************************************************************
+Frees the local storage struct for the specified thread. */
+UNIV_INTERN
+void
+thr_local_free(
+/*===========*/
+	os_thread_id_t	id)	/* in: thread id */
+{
+	thr_local_t*	local;
+
+	mutex_enter(&thr_local_mutex);
+
+	/* Look for the local struct in the hash table */
+
+	HASH_SEARCH(hash, thr_local_hash, os_thread_pf(id),
+		    thr_local_t*, local,, os_thread_eq(local->id, id));
+	if (local == NULL) {
+		mutex_exit(&thr_local_mutex);
+
+		return;
+	}
+
+	HASH_DELETE(thr_local_t, hash, thr_local_hash,
+		    os_thread_pf(id), local);
+	thr_local_hash_n_nodes--;
+
+	mutex_exit(&thr_local_mutex);
+
+	ut_a(local->magic_n == THR_LOCAL_MAGIC_N);
+
+	mem_free(local);
+}
+
+/********************************************************************
+Initializes the thread local storage module. */
+UNIV_INTERN
+void
+thr_local_init(void)
+/*================*/
+{
+
+	ut_a(thr_local_hash == NULL);
+
+	thr_local_hash = hash_create(OS_THREAD_MAX_N + 100);
+
+	mutex_create(&thr_local_mutex, SYNC_THR_LOCAL);
+}
+
+/*************************************************************************
+Return local hash table informations. */
+
+ulint
+thr_local_hash_cells(void)
+/*======================*/
+{
+	if (thr_local_hash) {
+		return (thr_local_hash->n_cells);
+	} else {
+		return 0;
+	}
+}
+
+ulint
+thr_local_hash_nodes(void)
+/*======================*/
+{
+	if (thr_local_hash) {
+		return (thr_local_hash_n_nodes
+			* (sizeof(thr_local_t) + MEM_BLOCK_HEADER_SIZE));
+	} else {
+		return 0;
+	}
+}
diff --git a/storage/xtradb/trx/trx0i_s.c b/storage/xtradb/trx/trx0i_s.c
new file mode 100644
index 00000000000..512e38cc17e
--- /dev/null
+++ b/storage/xtradb/trx/trx0i_s.c
@@ -0,0 +1,1431 @@
+/*****************************************************************************
+
+Copyright (c) 2007, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+INFORMATION SCHEMA innodb_trx, innodb_locks and
+innodb_lock_waits tables fetch code.
+
+The code below fetches information needed to fill those
+3 dynamic tables and uploads it into a "transactions
+table cache" for later retrieval.
+
+Created July 17, 2007 Vasil Dimov
+*******************************************************/
+
+#include <mysql/plugin.h>
+
+#include "mysql_addons.h"
+
+#include "univ.i"
+#include "buf0buf.h"
+#include "dict0dict.h"
+#include "ha0storage.h"
+#include "ha_prototypes.h"
+#include "hash0hash.h"
+#include "lock0iter.h"
+#include "lock0lock.h"
+#include "mem0mem.h"
+#include "page0page.h"
+#include "rem0rec.h"
+#include "row0row.h"
+#include "srv0srv.h"
+#include "sync0rw.h"
+#include "sync0sync.h"
+#include "sync0types.h"
+#include "trx0i_s.h"
+#include "trx0sys.h"
+#include "trx0trx.h"
+#include "ut0mem.h"
+#include "ut0ut.h"
+
+#define TABLE_CACHE_INITIAL_ROWSNUM	1024
+
+/* Table cache's rows are stored in a set of chunks. When a new row is
+added a new chunk is allocated if necessary. MEM_CHUNKS_IN_TABLE_CACHE
+specifies the maximum number of chunks.
+Assuming that the first one is 1024 rows (TABLE_CACHE_INITIAL_ROWSNUM)
+and each subsequent is N/2 where N is the number of rows we have
+allocated till now, then 39th chunk would have 1677416425 number of rows
+and all chunks would have 3354832851 number of rows. */
+#define MEM_CHUNKS_IN_TABLE_CACHE	39
+
+/* The following are some testing auxiliary macros. Do not enable them
+in a production environment. */
+
+#if 0
+/* If this is enabled then lock folds will always be different
+resulting in equal rows being put in a different cells of the hash
+table. Checking for duplicates will be flawed because different
+fold will be calculated when a row is searched in the hash table. */
+#define TEST_LOCK_FOLD_ALWAYS_DIFFERENT
+#endif
+
+#if 0
+/* This effectively kills the search-for-duplicate-before-adding-a-row
+function, but searching in the hash is still performed. It will always
+be assumed that lock is not present and insertion will be performed in
+the hash table. */
+#define TEST_NO_LOCKS_ROW_IS_EVER_EQUAL_TO_LOCK_T
+#endif
+
+#if 0
+/* This aggressively repeats adding each row many times. Depending on
+the above settings this may be noop or may result in lots of rows being
+added. */
+#define TEST_ADD_EACH_LOCKS_ROW_MANY_TIMES
+#endif
+
+#if 0
+/* Very similar to TEST_NO_LOCKS_ROW_IS_EVER_EQUAL_TO_LOCK_T but hash
+table search is not performed at all. */
+#define TEST_DO_NOT_CHECK_FOR_DUPLICATE_ROWS
+#endif
+
+#if 0
+/* Do not insert each row into the hash table, duplicates may appear
+if this is enabled, also if this is enabled searching into the hash is
+noop because it will be empty. */
+#define TEST_DO_NOT_INSERT_INTO_THE_HASH_TABLE
+#endif
+
+#define MAX_ALLOWED_FOR_STORAGE(cache)		\
+	(TRX_I_S_MEM_LIMIT			\
+	 - (cache)->mem_allocd)
+
+#define MAX_ALLOWED_FOR_ALLOC(cache)		\
+	(TRX_I_S_MEM_LIMIT			\
+	 - (cache)->mem_allocd			\
+	 - ha_storage_get_size((cache)->storage))
+
+/* Memory for each table in the intermediate buffer is allocated in
+separate chunks. These chunks are considered to be concatenated to
+represent one flat array of rows. */
+typedef struct i_s_mem_chunk_struct {
+	ulint	offset;		/* offset, in number of rows */
+	ulint	rows_allocd;	/* the size of this chunk, in number
+				of rows */
+	void*	base;		/* start of the chunk */
+} i_s_mem_chunk_t;
+
+/* This represents one table's cache. */
+typedef struct i_s_table_cache_struct {
+	ulint		rows_used;	/* number of used rows */
+	ulint		rows_allocd;	/* number of allocated rows */
+	ulint		row_size;	/* size of a single row */
+	i_s_mem_chunk_t	chunks[MEM_CHUNKS_IN_TABLE_CACHE]; /* array of
+					memory chunks that stores the
+					rows */
+} i_s_table_cache_t;
+
+/* This structure describes the intermediate buffer */
+struct trx_i_s_cache_struct {
+	rw_lock_t	rw_lock;	/* read-write lock protecting
+					the rest of this structure */
+	ullint		last_read;	/* last time the cache was read;
+					measured in microseconds since
+					epoch */
+	mutex_t		last_read_mutex;/* mutex protecting the
+					last_read member - it is updated
+					inside a shared lock of the
+					rw_lock member */
+	i_s_table_cache_t innodb_trx;	/* innodb_trx table */
+	i_s_table_cache_t innodb_locks;	/* innodb_locks table */
+	i_s_table_cache_t innodb_lock_waits;/* innodb_lock_waits table */
+/* the hash table size is LOCKS_HASH_CELLS_NUM * sizeof(void*) bytes */
+#define LOCKS_HASH_CELLS_NUM		10000
+	hash_table_t*	locks_hash;	/* hash table used to eliminate
+					duplicate entries in the
+					innodb_locks table */
+#define CACHE_STORAGE_INITIAL_SIZE	1024
+#define CACHE_STORAGE_HASH_CELLS	2048
+	ha_storage_t*	storage;	/* storage for external volatile
+					data that can possibly not be
+					available later, when we release
+					the kernel mutex */
+	ulint		mem_allocd;	/* the amount of memory
+					allocated with mem_alloc*() */
+	ibool		is_truncated;	/* this is TRUE if the memory
+					limit was hit and thus the data
+					in the cache is truncated */
+};
+
+/* This is the intermediate buffer where data needed to fill the
+INFORMATION SCHEMA tables is fetched and later retrieved by the C++
+code in handler/i_s.cc. */
+static trx_i_s_cache_t	trx_i_s_cache_static;
+UNIV_INTERN trx_i_s_cache_t*	trx_i_s_cache = &trx_i_s_cache_static;
+
+/***********************************************************************
+For a record lock that is in waiting state retrieves the only bit that
+is set, for a table lock returns ULINT_UNDEFINED. */
+static
+ulint
+wait_lock_get_heap_no(
+/*==================*/
+				/* out: record number within the heap */
+	const lock_t*	lock)	/* in: lock */
+{
+	ulint	ret;
+
+	switch (lock_get_type(lock)) {
+	case LOCK_REC:
+		ret = lock_rec_find_set_bit(lock);
+		ut_a(ret != ULINT_UNDEFINED);
+		break;
+	case LOCK_TABLE:
+		ret = ULINT_UNDEFINED;
+		break;
+	default:
+		ut_error;
+	}
+
+	return(ret);
+}
+
+/***********************************************************************
+Initializes the members of a table cache. */
+static
+void
+table_cache_init(
+/*=============*/
+	i_s_table_cache_t*	table_cache,	/* out: table cache */
+	size_t			row_size)	/* in: the size of a
+						row */
+{
+	ulint	i;
+
+	table_cache->rows_used = 0;
+	table_cache->rows_allocd = 0;
+	table_cache->row_size = row_size;
+
+	for (i = 0; i < MEM_CHUNKS_IN_TABLE_CACHE; i++) {
+
+		/* the memory is actually allocated in
+		table_cache_create_empty_row() */
+		table_cache->chunks[i].base = NULL;
+	}
+}
+
+/***********************************************************************
+Returns an empty row from a table cache. The row is allocated if no more
+empty rows are available. The number of used rows is incremented.
+If the memory limit is hit then NULL is returned and nothing is
+allocated. */
+static
+void*
+table_cache_create_empty_row(
+/*=========================*/
+						/* out: empty row, or
+						NULL if out of memory */
+	i_s_table_cache_t*	table_cache,	/* in/out: table cache */
+	trx_i_s_cache_t*	cache)		/* in/out: cache to record
+						how many bytes are
+						allocated */
+{
+	ulint	i;
+	void*	row;
+
+	ut_a(table_cache->rows_used <= table_cache->rows_allocd);
+
+	if (table_cache->rows_used == table_cache->rows_allocd) {
+
+		/* rows_used == rows_allocd means that new chunk needs
+		to be allocated: either no more empty rows in the
+		last allocated chunk or nothing has been allocated yet
+		(rows_num == rows_allocd == 0); */
+
+		i_s_mem_chunk_t*	chunk;
+		ulint			req_bytes;
+		ulint			got_bytes;
+		ulint			req_rows;
+		ulint			got_rows;
+
+		/* find the first not allocated chunk */
+		for (i = 0; i < MEM_CHUNKS_IN_TABLE_CACHE; i++) {
+
+			if (table_cache->chunks[i].base == NULL) {
+
+				break;
+			}
+		}
+
+		/* i == MEM_CHUNKS_IN_TABLE_CACHE means that all chunks
+		have been allocated :-X */
+		ut_a(i < MEM_CHUNKS_IN_TABLE_CACHE);
+
+		/* allocate the chunk we just found */
+
+		if (i == 0) {
+
+			/* first chunk, nothing is allocated yet */
+			req_rows = TABLE_CACHE_INITIAL_ROWSNUM;
+		} else {
+
+			/* Memory is increased by the formula
+			new = old + old / 2; We are trying not to be
+			aggressive here (= using the common new = old * 2)
+			because the allocated memory will not be freed
+			until InnoDB exit (it is reused). So it is better
+			to once allocate the memory in more steps, but
+			have less unused/wasted memory than to use less
+			steps in allocation (which is done once in a
+			lifetime) but end up with lots of unused/wasted
+			memory. */
+			req_rows = table_cache->rows_allocd / 2;
+		}
+		req_bytes = req_rows * table_cache->row_size;
+
+		if (req_bytes > MAX_ALLOWED_FOR_ALLOC(cache)) {
+
+			return(NULL);
+		}
+
+		chunk = &table_cache->chunks[i];
+
+		chunk->base = mem_alloc2(req_bytes, &got_bytes);
+
+		got_rows = got_bytes / table_cache->row_size;
+
+		cache->mem_allocd += got_bytes;
+
+#if 0
+		printf("allocating chunk %d req bytes=%lu, got bytes=%lu, "
+		       "row size=%lu, "
+		       "req rows=%lu, got rows=%lu\n",
+		       i, req_bytes, got_bytes,
+		       table_cache->row_size,
+		       req_rows, got_rows);
+#endif
+
+		chunk->rows_allocd = got_rows;
+
+		table_cache->rows_allocd += got_rows;
+
+		/* adjust the offset of the next chunk */
+		if (i < MEM_CHUNKS_IN_TABLE_CACHE - 1) {
+
+			table_cache->chunks[i + 1].offset
+				= chunk->offset + chunk->rows_allocd;
+		}
+
+		/* return the first empty row in the newly allocated
+		chunk */
+		row = chunk->base;
+	} else {
+
+		char*	chunk_start;
+		ulint	offset;
+
+		/* there is an empty row, no need to allocate new
+		chunks */
+
+		/* find the first chunk that contains allocated but
+		empty/unused rows */
+		for (i = 0; i < MEM_CHUNKS_IN_TABLE_CACHE; i++) {
+
+			if (table_cache->chunks[i].offset
+			    + table_cache->chunks[i].rows_allocd
+			    > table_cache->rows_used) {
+
+				break;
+			}
+		}
+
+		/* i == MEM_CHUNKS_IN_TABLE_CACHE means that all chunks
+		are full, but
+		table_cache->rows_used != table_cache->rows_allocd means
+		exactly the opposite - there are allocated but
+		empty/unused rows :-X */
+		ut_a(i < MEM_CHUNKS_IN_TABLE_CACHE);
+
+		chunk_start = (char*) table_cache->chunks[i].base;
+		offset = table_cache->rows_used
+			- table_cache->chunks[i].offset;
+
+		row = chunk_start + offset * table_cache->row_size;
+	}
+
+	table_cache->rows_used++;
+
+	return(row);
+}
+
+/***********************************************************************
+Fills i_s_trx_row_t object.
+If memory can not be allocated then FALSE is returned. */
+static
+ibool
+fill_trx_row(
+/*=========*/
+						/* out: FALSE if
+						allocation fails */
+	i_s_trx_row_t*		row,		/* out: result object
+						that's filled */
+	const trx_t*		trx,		/* in: transaction to
+						get data from */
+	const i_s_locks_row_t*	requested_lock_row,/* in: pointer to the
+						corresponding row in
+						innodb_locks if trx is
+						waiting or NULL if trx
+						is not waiting */
+	trx_i_s_cache_t*	cache)		/* in/out: cache into
+						which to copy volatile
+						strings */
+{
+	row->trx_id = trx_get_id(trx);
+	row->trx_started = (ib_time_t) trx->start_time;
+	row->trx_state = trx_get_que_state_str(trx);
+
+	if (trx->wait_lock != NULL) {
+
+		ut_a(requested_lock_row != NULL);
+
+		row->requested_lock_row = requested_lock_row;
+		row->trx_wait_started = (ib_time_t) trx->wait_started;
+	} else {
+
+		ut_a(requested_lock_row == NULL);
+
+		row->requested_lock_row = NULL;
+		row->trx_wait_started = 0;
+	}
+
+	row->trx_weight = (ullint) ut_conv_dulint_to_longlong(TRX_WEIGHT(trx));
+
+	if (trx->mysql_thd != NULL) {
+		row->trx_mysql_thread_id
+			= thd_get_thread_id(trx->mysql_thd);
+	} else {
+		/* For internal transactions e.g., purge and transactions
+		being recovered at startup there is no associated MySQL
+		thread data structure. */
+		row->trx_mysql_thread_id = 0;
+	}
+
+	if (trx->mysql_query_str != NULL && *trx->mysql_query_str != NULL) {
+
+		if (strlen(*trx->mysql_query_str)
+		    > TRX_I_S_TRX_QUERY_MAX_LEN) {
+
+			char	query[TRX_I_S_TRX_QUERY_MAX_LEN + 1];
+
+			memcpy(query, *trx->mysql_query_str,
+			       TRX_I_S_TRX_QUERY_MAX_LEN);
+			query[TRX_I_S_TRX_QUERY_MAX_LEN] = '\0';
+
+			row->trx_query = ha_storage_put_memlim(
+				cache->storage, query,
+				TRX_I_S_TRX_QUERY_MAX_LEN + 1,
+				MAX_ALLOWED_FOR_STORAGE(cache));
+		} else {
+
+			row->trx_query = ha_storage_put_str_memlim(
+				cache->storage, *trx->mysql_query_str,
+				MAX_ALLOWED_FOR_STORAGE(cache));
+		}
+
+		if (row->trx_query == NULL) {
+
+			return(FALSE);
+		}
+	} else {
+
+		row->trx_query = NULL;
+	}
+
+	return(TRUE);
+}
+
+/***********************************************************************
+Format the nth field of "rec" and put it in "buf". The result is always
+'\0'-terminated. Returns the number of bytes that were written to "buf"
+(including the terminating '\0'). */
+static
+ulint
+put_nth_field(
+/*==========*/
+					/* out: end of the result */
+	char*			buf,	/* out: buffer */
+	ulint			buf_size,/* in: buffer size in bytes */
+	ulint			n,	/* in: number of field */
+	const dict_index_t*	index,	/* in: index */
+	const rec_t*		rec,	/* in: record */
+	const ulint*		offsets)/* in: record offsets, returned
+					by rec_get_offsets() */
+{
+	const byte*	data;
+	ulint		data_len;
+	dict_field_t*	dict_field;
+	ulint		ret;
+
+	ut_ad(rec_offs_validate(rec, NULL, offsets));
+
+	if (buf_size == 0) {
+
+		return(0);
+	}
+
+	ret = 0;
+
+	if (n > 0) {
+		/* we must append ", " before the actual data */
+
+		if (buf_size < 3) {
+
+			buf[0] = '\0';
+			return(1);
+		}
+
+		memcpy(buf, ", ", 3);
+
+		buf += 2;
+		buf_size -= 2;
+		ret += 2;
+	}
+
+	/* now buf_size >= 1 */
+
+	data = rec_get_nth_field(rec, offsets, n, &data_len);
+
+	dict_field = dict_index_get_nth_field(index, n);
+
+	ret += row_raw_format((const char*) data, data_len,
+			      dict_field, buf, buf_size);
+
+	return(ret);
+}
+
+/***********************************************************************
+Fills the "lock_data" member of i_s_locks_row_t object.
+If memory can not be allocated then FALSE is returned. */
+static
+ibool
+fill_lock_data(
+/*===========*/
+					/* out: FALSE if allocation fails */
+	const char**		lock_data,/* out: "lock_data" to fill */
+	const lock_t*		lock,	/* in: lock used to find the data */
+	ulint			heap_no,/* in: rec num used to find the data */
+	trx_i_s_cache_t*	cache)	/* in/out: cache where to store
+					volatile data */
+{
+	mtr_t			mtr;
+
+	const buf_block_t*	block;
+	const page_t*		page;
+	const rec_t*		rec;
+
+	ut_a(lock_get_type(lock) == LOCK_REC);
+
+	mtr_start(&mtr);
+
+	block = buf_page_try_get(lock_rec_get_space_id(lock),
+				 lock_rec_get_page_no(lock),
+				 &mtr);
+
+	if (block == NULL) {
+
+		*lock_data = NULL;
+
+		mtr_commit(&mtr);
+
+		return(TRUE);
+	}
+
+	page = (const page_t*) buf_block_get_frame(block);
+
+	rec = page_find_rec_with_heap_no(page, heap_no);
+
+	if (page_rec_is_infimum(rec)) {
+
+		*lock_data = ha_storage_put_str_memlim(
+			cache->storage, "infimum pseudo-record",
+			MAX_ALLOWED_FOR_STORAGE(cache));
+	} else if (page_rec_is_supremum(rec)) {
+
+		*lock_data = ha_storage_put_str_memlim(
+			cache->storage, "supremum pseudo-record",
+			MAX_ALLOWED_FOR_STORAGE(cache));
+	} else {
+
+		const dict_index_t*	index;
+		ulint			n_fields;
+		mem_heap_t*		heap;
+		ulint			offsets_onstack[REC_OFFS_NORMAL_SIZE];
+		ulint*			offsets;
+		char			buf[TRX_I_S_LOCK_DATA_MAX_LEN];
+		ulint			buf_used;
+		ulint			i;
+
+		rec_offs_init(offsets_onstack);
+		offsets = offsets_onstack;
+
+		index = lock_rec_get_index(lock);
+
+		n_fields = dict_index_get_n_unique(index);
+
+		ut_a(n_fields > 0);
+
+		heap = NULL;
+		offsets = rec_get_offsets(rec, index, offsets, n_fields,
+					  &heap);
+
+		/* format and store the data */
+
+		buf_used = 0;
+		for (i = 0; i < n_fields; i++) {
+
+			buf_used += put_nth_field(
+				buf + buf_used, sizeof(buf) - buf_used,
+				i, index, rec, offsets) - 1;
+		}
+
+		*lock_data = (const char*) ha_storage_put_memlim(
+			cache->storage, buf, buf_used + 1,
+			MAX_ALLOWED_FOR_STORAGE(cache));
+
+		if (UNIV_UNLIKELY(heap != NULL)) {
+
+			/* this means that rec_get_offsets() has created a new
+			heap and has stored offsets in it; check that this is
+			really the case and free the heap */
+			ut_a(offsets != offsets_onstack);
+			mem_heap_free(heap);
+		}
+	}
+
+	mtr_commit(&mtr);
+
+	if (*lock_data == NULL) {
+
+		return(FALSE);
+	}
+
+	return(TRUE);
+}
+
+/***********************************************************************
+Fills i_s_locks_row_t object. Returns its first argument.
+If memory can not be allocated then FALSE is returned. */
+static
+ibool
+fill_locks_row(
+/*===========*/
+				/* out: FALSE if allocation fails */
+	i_s_locks_row_t* row,	/* out: result object that's filled */
+	const lock_t*	lock,	/* in: lock to get data from */
+	ulint		heap_no,/* in: lock's record number
+				or ULINT_UNDEFINED if the lock
+				is a table lock */
+	trx_i_s_cache_t* cache)	/* in/out: cache into which to copy
+				volatile strings */
+{
+	row->lock_trx_id = lock_get_trx_id(lock);
+	row->lock_mode = lock_get_mode_str(lock);
+	row->lock_type = lock_get_type_str(lock);
+
+	row->lock_table = ha_storage_put_str_memlim(
+		cache->storage, lock_get_table_name(lock),
+		MAX_ALLOWED_FOR_STORAGE(cache));
+
+	/* memory could not be allocated */
+	if (row->lock_table == NULL) {
+
+		return(FALSE);
+	}
+
+	switch (lock_get_type(lock)) {
+	case LOCK_REC:
+		row->lock_index = ha_storage_put_str_memlim(
+			cache->storage, lock_rec_get_index_name(lock),
+			MAX_ALLOWED_FOR_STORAGE(cache));
+
+		/* memory could not be allocated */
+		if (row->lock_index == NULL) {
+
+			return(FALSE);
+		}
+
+		row->lock_space = lock_rec_get_space_id(lock);
+		row->lock_page = lock_rec_get_page_no(lock);
+		row->lock_rec = heap_no;
+
+		if (!fill_lock_data(&row->lock_data, lock, heap_no, cache)) {
+
+			/* memory could not be allocated */
+			return(FALSE);
+		}
+
+		break;
+	case LOCK_TABLE:
+		row->lock_index = NULL;
+
+		row->lock_space = ULINT_UNDEFINED;
+		row->lock_page = ULINT_UNDEFINED;
+		row->lock_rec = ULINT_UNDEFINED;
+
+		row->lock_data = NULL;
+
+		break;
+	default:
+		ut_error;
+	}
+
+	row->lock_table_id = lock_get_table_id(lock);
+
+	row->hash_chain.value = row;
+
+	return(TRUE);
+}
+
+/***********************************************************************
+Fills i_s_lock_waits_row_t object. Returns its first argument. */
+static
+i_s_lock_waits_row_t*
+fill_lock_waits_row(
+/*================*/
+						/* out: result object
+						that's filled */
+	i_s_lock_waits_row_t*	row,		/* out: result object
+						that's filled */
+	const i_s_locks_row_t*	requested_lock_row,/* in: pointer to the
+						relevant requested lock
+						row in innodb_locks */
+	const i_s_locks_row_t*	blocking_lock_row)/* in: pointer to the
+						relevant blocking lock
+						row in innodb_locks */
+{
+	row->requested_lock_row = requested_lock_row;
+	row->blocking_lock_row = blocking_lock_row;
+
+	return(row);
+}
+
+/***********************************************************************
+Calculates a hash fold for a lock. For a record lock the fold is
+calculated from 4 elements, which uniquely identify a lock at a given
+point in time: transaction id, space id, page number, record number.
+For a table lock the fold is table's id. */
+static
+ulint
+fold_lock(
+/*======*/
+				/* out: fold */
+	const lock_t*	lock,	/* in: lock object to fold */
+	ulint		heap_no)/* in: lock's record number
+				or ULINT_UNDEFINED if the lock
+				is a table lock */
+{
+#ifdef TEST_LOCK_FOLD_ALWAYS_DIFFERENT
+	static ulint	fold = 0;
+
+	return(fold++);
+#else
+	ulint	ret;
+
+	switch (lock_get_type(lock)) {
+	case LOCK_REC:
+		ut_a(heap_no != ULINT_UNDEFINED);
+
+		ret = ut_fold_ulint_pair((ulint) lock_get_trx_id(lock),
+					 lock_rec_get_space_id(lock));
+
+		ret = ut_fold_ulint_pair(ret,
+					 lock_rec_get_page_no(lock));
+
+		ret = ut_fold_ulint_pair(ret, heap_no);
+
+		break;
+	case LOCK_TABLE:
+		/* this check is actually not necessary for continuing
+		correct operation, but something must have gone wrong if
+		it fails. */
+		ut_a(heap_no == ULINT_UNDEFINED);
+
+		ret = (ulint) lock_get_table_id(lock);
+
+		break;
+	default:
+		ut_error;
+	}
+
+	return(ret);
+#endif
+}
+
+/***********************************************************************
+Checks whether i_s_locks_row_t object represents a lock_t object. */
+static
+ibool
+locks_row_eq_lock(
+/*==============*/
+					/* out: TRUE if they match */
+	const i_s_locks_row_t*	row,	/* in: innodb_locks row */
+	const lock_t*		lock,	/* in: lock object */
+	ulint			heap_no)/* in: lock's record number
+					or ULINT_UNDEFINED if the lock
+					is a table lock */
+{
+#ifdef TEST_NO_LOCKS_ROW_IS_EVER_EQUAL_TO_LOCK_T
+	return(0);
+#else
+	switch (lock_get_type(lock)) {
+	case LOCK_REC:
+		ut_a(heap_no != ULINT_UNDEFINED);
+
+		return(row->lock_trx_id == lock_get_trx_id(lock)
+		       && row->lock_space == lock_rec_get_space_id(lock)
+		       && row->lock_page == lock_rec_get_page_no(lock)
+		       && row->lock_rec == heap_no);
+
+	case LOCK_TABLE:
+		/* this check is actually not necessary for continuing
+		correct operation, but something must have gone wrong if
+		it fails. */
+		ut_a(heap_no == ULINT_UNDEFINED);
+
+		return(row->lock_trx_id == lock_get_trx_id(lock)
+		       && row->lock_table_id == lock_get_table_id(lock));
+
+	default:
+		ut_error;
+		return(FALSE);
+	}
+#endif
+}
+
+/***********************************************************************
+Searches for a row in the innodb_locks cache that has a specified id.
+This happens in O(1) time since a hash table is used. Returns pointer to
+the row or NULL if none is found. */
+static
+i_s_locks_row_t*
+search_innodb_locks(
+/*================*/
+					/* out: row or NULL */
+	trx_i_s_cache_t*	cache,	/* in: cache */
+	const lock_t*		lock,	/* in: lock to search for */
+	ulint			heap_no)/* in: lock's record number
+					or ULINT_UNDEFINED if the lock
+					is a table lock */
+{
+	i_s_hash_chain_t*	hash_chain;
+
+	HASH_SEARCH(
+		/* hash_chain->"next" */
+		next,
+		/* the hash table */
+		cache->locks_hash,
+		/* fold */
+		fold_lock(lock, heap_no),
+		/* the type of the next variable */
+		i_s_hash_chain_t*,
+		/* auxiliary variable */
+		hash_chain,
+		/* assertion on every traversed item */
+		,
+		/* this determines if we have found the lock */
+		locks_row_eq_lock(hash_chain->value, lock, heap_no));
+
+	if (hash_chain == NULL) {
+
+		return(NULL);
+	}
+	/* else */
+
+	return(hash_chain->value);
+}
+
+/***********************************************************************
+Adds new element to the locks cache, enlarging it if necessary.
+Returns a pointer to the added row. If the row is already present then
+no row is added and a pointer to the existing row is returned.
+If row can not be allocated then NULL is returned. */
+static
+i_s_locks_row_t*
+add_lock_to_cache(
+/*==============*/
+					/* out: row */
+	trx_i_s_cache_t*	cache,	/* in/out: cache */
+	const lock_t*		lock,	/* in: the element to add */
+	ulint			heap_no)/* in: lock's record number
+					or ULINT_UNDEFINED if the lock
+					is a table lock */
+{
+	i_s_locks_row_t*	dst_row;
+
+#ifdef TEST_ADD_EACH_LOCKS_ROW_MANY_TIMES
+	ulint	i;
+	for (i = 0; i < 10000; i++) {
+#endif
+#ifndef TEST_DO_NOT_CHECK_FOR_DUPLICATE_ROWS
+	/* quit if this lock is already present */
+	dst_row = search_innodb_locks(cache, lock, heap_no);
+	if (dst_row != NULL) {
+
+		return(dst_row);
+	}
+#endif
+
+	dst_row = (i_s_locks_row_t*)
+		table_cache_create_empty_row(&cache->innodb_locks, cache);
+
+	/* memory could not be allocated */
+	if (dst_row == NULL) {
+
+		return(NULL);
+	}
+
+	if (!fill_locks_row(dst_row, lock, heap_no, cache)) {
+
+		/* memory could not be allocated */
+		cache->innodb_locks.rows_used--;
+		return(NULL);
+	}
+
+#ifndef TEST_DO_NOT_INSERT_INTO_THE_HASH_TABLE
+	HASH_INSERT(
+		/* the type used in the hash chain */
+		i_s_hash_chain_t,
+		/* hash_chain->"next" */
+		next,
+		/* the hash table */
+		cache->locks_hash,
+		/* fold */
+		fold_lock(lock, heap_no),
+		/* add this data to the hash */
+		&dst_row->hash_chain);
+#endif
+#ifdef TEST_ADD_EACH_LOCKS_ROW_MANY_TIMES
+	} /* for()-loop */
+#endif
+
+	return(dst_row);
+}
+
+/***********************************************************************
+Adds new pair of locks to the lock waits cache.
+If memory can not be allocated then FALSE is returned. */
+static
+ibool
+add_lock_wait_to_cache(
+/*===================*/
+						/* out: FALSE if
+						allocation fails */
+	trx_i_s_cache_t*	cache,		/* in/out: cache */
+	const i_s_locks_row_t*	requested_lock_row,/* in: pointer to the
+						relevant requested lock
+						row in innodb_locks */
+	const i_s_locks_row_t*	blocking_lock_row)/* in: pointer to the
+						relevant blocking lock
+						row in innodb_locks */
+{
+	i_s_lock_waits_row_t*	dst_row;
+
+	dst_row = (i_s_lock_waits_row_t*)
+		table_cache_create_empty_row(&cache->innodb_lock_waits,
+					     cache);
+
+	/* memory could not be allocated */
+	if (dst_row == NULL) {
+
+		return(FALSE);
+	}
+
+	fill_lock_waits_row(dst_row, requested_lock_row, blocking_lock_row);
+
+	return(TRUE);
+}
+
+/***********************************************************************
+Adds transaction's relevant (important) locks to cache.
+If the transaction is waiting, then the wait lock is added to
+innodb_locks and a pointer to the added row is returned in
+requested_lock_row, otherwise requested_lock_row is set to NULL.
+If rows can not be allocated then FALSE is returned and the value of
+requested_lock_row is undefined. */
+static
+ibool
+add_trx_relevant_locks_to_cache(
+/*============================*/
+					/* out: FALSE if allocation fails */
+	trx_i_s_cache_t*	cache,	/* in/out: cache */
+	const trx_t*		trx,	/* in: transaction */
+	i_s_locks_row_t**	requested_lock_row)/* out: pointer to the
+					requested lock row, or NULL or
+					undefined */
+{
+	ut_ad(mutex_own(&kernel_mutex));
+
+	/* If transaction is waiting we add the wait lock and all locks
+	from another transactions that are blocking the wait lock. */
+	if (trx->que_state == TRX_QUE_LOCK_WAIT) {
+
+		const lock_t*		curr_lock;
+		ulint			wait_lock_heap_no;
+		i_s_locks_row_t*	blocking_lock_row;
+		lock_queue_iterator_t	iter;
+
+		ut_a(trx->wait_lock != NULL);
+
+		wait_lock_heap_no
+			= wait_lock_get_heap_no(trx->wait_lock);
+
+		/* add the requested lock */
+		*requested_lock_row
+			= add_lock_to_cache(cache, trx->wait_lock,
+					    wait_lock_heap_no);
+
+		/* memory could not be allocated */
+		if (*requested_lock_row == NULL) {
+
+			return(FALSE);
+		}
+
+		/* then iterate over the locks before the wait lock and
+		add the ones that are blocking it */
+
+		lock_queue_iterator_reset(&iter, trx->wait_lock,
+					  ULINT_UNDEFINED);
+
+		curr_lock = lock_queue_iterator_get_prev(&iter);
+		while (curr_lock != NULL) {
+
+			if (lock_has_to_wait(trx->wait_lock,
+					     curr_lock)) {
+
+				/* add the lock that is
+				blocking trx->wait_lock */
+				blocking_lock_row
+					= add_lock_to_cache(
+						cache, curr_lock,
+						/* heap_no is the same
+						for the wait and waited
+						locks */
+						wait_lock_heap_no);
+
+				/* memory could not be allocated */
+				if (blocking_lock_row == NULL) {
+
+					return(FALSE);
+				}
+
+				/* add the relation between both locks
+				to innodb_lock_waits */
+				if (!add_lock_wait_to_cache(
+						cache, *requested_lock_row,
+						blocking_lock_row)) {
+
+					/* memory could not be allocated */
+					return(FALSE);
+				}
+			}
+
+			curr_lock = lock_queue_iterator_get_prev(&iter);
+		}
+	} else {
+
+		*requested_lock_row = NULL;
+	}
+
+	return(TRUE);
+}
+
+/***********************************************************************
+Checks if the cache can safely be updated. */
+static
+ibool
+can_cache_be_updated(
+/*=================*/
+	trx_i_s_cache_t*	cache)	/* in: cache */
+{
+	ullint	now;
+
+/* The minimum time that a cache must not be updated after it has been
+read for the last time; measured in microseconds. We use this technique
+to ensure that SELECTs which join several INFORMATION SCHEMA tables read
+the same version of the cache. */
+#define CACHE_MIN_IDLE_TIME_US	100000 /* 0.1 sec */
+
+	/* Here we read cache->last_read without acquiring its mutex
+	because last_read is only updated when a shared rw lock on the
+	whole cache is being held (see trx_i_s_cache_end_read()) and
+	we are currently holding an exclusive rw lock on the cache.
+	So it is not possible for last_read to be updated while we are
+	reading it. */
+
+#ifdef UNIV_SYNC_DEBUG
+	ut_a(rw_lock_own(&cache->rw_lock, RW_LOCK_EX));
+#endif
+
+	now = ut_time_us(NULL);
+	if (now - cache->last_read > CACHE_MIN_IDLE_TIME_US) {
+
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
+
+/***********************************************************************
+Declare a cache empty, preparing it to be filled up. Not all resources
+are freed because they can be reused. */
+static
+void
+trx_i_s_cache_clear(
+/*================*/
+	trx_i_s_cache_t*	cache)	/* out: cache to clear */
+{
+	cache->innodb_trx.rows_used = 0;
+	cache->innodb_locks.rows_used = 0;
+	cache->innodb_lock_waits.rows_used = 0;
+
+	hash_table_clear(cache->locks_hash);
+
+	ha_storage_empty(&cache->storage);
+}
+
+/***********************************************************************
+Fetches the data needed to fill the 3 INFORMATION SCHEMA tables into the
+table cache buffer. Cache must be locked for write. */
+static
+void
+fetch_data_into_cache(
+/*==================*/
+	trx_i_s_cache_t*	cache)	/* in/out: cache */
+{
+	trx_t*			trx;
+	i_s_trx_row_t*		trx_row;
+	i_s_locks_row_t*	requested_lock_row;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	trx_i_s_cache_clear(cache);
+
+	/* We iterate over the list of all transactions and add each one
+	to innodb_trx's cache. We also add all locks that are relevant
+	to each transaction into innodb_locks' and innodb_lock_waits'
+	caches. */
+
+	for (trx = UT_LIST_GET_FIRST(trx_sys->trx_list);
+	     trx != NULL;
+	     trx = UT_LIST_GET_NEXT(trx_list, trx)) {
+
+		if (!add_trx_relevant_locks_to_cache(cache, trx,
+						     &requested_lock_row)) {
+
+			cache->is_truncated = TRUE;
+			return;
+		}
+
+		trx_row = (i_s_trx_row_t*)
+			table_cache_create_empty_row(&cache->innodb_trx,
+						     cache);
+
+		/* memory could not be allocated */
+		if (trx_row == NULL) {
+
+			cache->is_truncated = TRUE;
+			return;
+		}
+
+		if (!fill_trx_row(trx_row, trx, requested_lock_row, cache)) {
+
+			/* memory could not be allocated */
+			cache->innodb_trx.rows_used--;
+			cache->is_truncated = TRUE;
+			return;
+		}
+	}
+
+	cache->is_truncated = FALSE;
+}
+
+/***********************************************************************
+Update the transactions cache if it has not been read for some time.
+Called from handler/i_s.cc. */
+UNIV_INTERN
+int
+trx_i_s_possibly_fetch_data_into_cache(
+/*===================================*/
+					/* out: 0 - fetched, 1 - not */
+	trx_i_s_cache_t*	cache)	/* in/out: cache */
+{
+	if (!can_cache_be_updated(cache)) {
+
+		return(1);
+	}
+
+	/* We are going to access trx->query in all transactions */
+	innobase_mysql_prepare_print_arbitrary_thd();
+
+	/* We need to read trx_sys and record/table lock queues */
+	mutex_enter(&kernel_mutex);
+
+	fetch_data_into_cache(cache);
+
+	mutex_exit(&kernel_mutex);
+
+	innobase_mysql_end_print_arbitrary_thd();
+
+	return(0);
+}
+
+/***********************************************************************
+Returns TRUE if the data in the cache is truncated due to the memory
+limit posed by TRX_I_S_MEM_LIMIT. */
+UNIV_INTERN
+ibool
+trx_i_s_cache_is_truncated(
+/*=======================*/
+					/* out: TRUE if truncated */
+	trx_i_s_cache_t*	cache)	/* in: cache */
+{
+	return(cache->is_truncated);
+}
+
+/***********************************************************************
+Initialize INFORMATION SCHEMA trx related cache. */
+UNIV_INTERN
+void
+trx_i_s_cache_init(
+/*===============*/
+	trx_i_s_cache_t*	cache)	/* out: cache to init */
+{
+	/* The latching is done in the following order:
+	acquire trx_i_s_cache_t::rw_lock, X
+	acquire kernel_mutex
+	release kernel_mutex
+	release trx_i_s_cache_t::rw_lock
+	acquire trx_i_s_cache_t::rw_lock, S
+	acquire trx_i_s_cache_t::last_read_mutex
+	release trx_i_s_cache_t::last_read_mutex
+	release trx_i_s_cache_t::rw_lock */
+
+	rw_lock_create(&cache->rw_lock, SYNC_TRX_I_S_RWLOCK);
+
+	cache->last_read = 0;
+
+	mutex_create(&cache->last_read_mutex, SYNC_TRX_I_S_LAST_READ);
+
+	table_cache_init(&cache->innodb_trx, sizeof(i_s_trx_row_t));
+	table_cache_init(&cache->innodb_locks, sizeof(i_s_locks_row_t));
+	table_cache_init(&cache->innodb_lock_waits,
+			 sizeof(i_s_lock_waits_row_t));
+
+	cache->locks_hash = hash_create(LOCKS_HASH_CELLS_NUM);
+
+	cache->storage = ha_storage_create(CACHE_STORAGE_INITIAL_SIZE,
+					   CACHE_STORAGE_HASH_CELLS);
+
+	cache->mem_allocd = 0;
+
+	cache->is_truncated = FALSE;
+}
+
+/***********************************************************************
+Issue a shared/read lock on the tables cache. */
+UNIV_INTERN
+void
+trx_i_s_cache_start_read(
+/*=====================*/
+	trx_i_s_cache_t*	cache)	/* in: cache */
+{
+	rw_lock_s_lock(&cache->rw_lock);
+}
+
+/***********************************************************************
+Release a shared/read lock on the tables cache. */
+UNIV_INTERN
+void
+trx_i_s_cache_end_read(
+/*===================*/
+	trx_i_s_cache_t*	cache)	/* in: cache */
+{
+	ullint	now;
+
+#ifdef UNIV_SYNC_DEBUG
+	ut_a(rw_lock_own(&cache->rw_lock, RW_LOCK_SHARED));
+#endif
+
+	/* update cache last read time */
+	now = ut_time_us(NULL);
+	mutex_enter(&cache->last_read_mutex);
+	cache->last_read = now;
+	mutex_exit(&cache->last_read_mutex);
+
+	rw_lock_s_unlock(&cache->rw_lock);
+}
+
+/***********************************************************************
+Issue an exclusive/write lock on the tables cache. */
+UNIV_INTERN
+void
+trx_i_s_cache_start_write(
+/*======================*/
+	trx_i_s_cache_t*	cache)	/* in: cache */
+{
+	rw_lock_x_lock(&cache->rw_lock);
+}
+
+/***********************************************************************
+Release an exclusive/write lock on the tables cache. */
+UNIV_INTERN
+void
+trx_i_s_cache_end_write(
+/*====================*/
+	trx_i_s_cache_t*	cache)	/* in: cache */
+{
+#ifdef UNIV_SYNC_DEBUG
+	ut_a(rw_lock_own(&cache->rw_lock, RW_LOCK_EX));
+#endif
+
+	rw_lock_x_unlock(&cache->rw_lock);
+}
+
+/***********************************************************************
+Selects a INFORMATION SCHEMA table cache from the whole cache. */
+static
+i_s_table_cache_t*
+cache_select_table(
+/*===============*/
+					/* out: table cache */
+	trx_i_s_cache_t*	cache,	/* in: whole cache */
+	enum i_s_table		table)	/* in: which table */
+{
+	i_s_table_cache_t*	table_cache;
+
+#ifdef UNIV_SYNC_DEBUG
+	ut_a(rw_lock_own(&cache->rw_lock, RW_LOCK_SHARED)
+	     || rw_lock_own(&cache->rw_lock, RW_LOCK_EX));
+#endif
+
+	switch (table) {
+	case I_S_INNODB_TRX:
+		table_cache = &cache->innodb_trx;
+		break;
+	case I_S_INNODB_LOCKS:
+		table_cache = &cache->innodb_locks;
+		break;
+	case I_S_INNODB_LOCK_WAITS:
+		table_cache = &cache->innodb_lock_waits;
+		break;
+	default:
+		ut_error;
+	}
+
+	return(table_cache);
+}
+
+/***********************************************************************
+Retrieves the number of used rows in the cache for a given
+INFORMATION SCHEMA table. */
+UNIV_INTERN
+ulint
+trx_i_s_cache_get_rows_used(
+/*========================*/
+					/* out: number of rows */
+	trx_i_s_cache_t*	cache,	/* in: cache */
+	enum i_s_table		table)	/* in: which table */
+{
+	i_s_table_cache_t*	table_cache;
+
+	table_cache = cache_select_table(cache, table);
+
+	return(table_cache->rows_used);
+}
+
+/***********************************************************************
+Retrieves the nth row (zero-based) in the cache for a given
+INFORMATION SCHEMA table. */
+UNIV_INTERN
+void*
+trx_i_s_cache_get_nth_row(
+/*======================*/
+					/* out: row */
+	trx_i_s_cache_t*	cache,	/* in: cache */
+	enum i_s_table		table,	/* in: which table */
+	ulint			n)	/* in: row number */
+{
+	i_s_table_cache_t*	table_cache;
+	ulint			i;
+	void*			row;
+
+	table_cache = cache_select_table(cache, table);
+
+	ut_a(n < table_cache->rows_used);
+
+	row = NULL;
+
+	for (i = 0; i < MEM_CHUNKS_IN_TABLE_CACHE; i++) {
+
+		if (table_cache->chunks[i].offset
+		    + table_cache->chunks[i].rows_allocd > n) {
+
+			row = (char*) table_cache->chunks[i].base
+				+ (n - table_cache->chunks[i].offset)
+				* table_cache->row_size;
+			break;
+		}
+	}
+
+	ut_a(row != NULL);
+
+	return(row);
+}
+
+/***********************************************************************
+Crafts a lock id string from a i_s_locks_row_t object. Returns its
+second argument. This function aborts if there is not enough space in
+lock_id. Be sure to provide at least TRX_I_S_LOCK_ID_MAX_LEN + 1 if you
+want to be 100% sure that it will not abort. */
+UNIV_INTERN
+char*
+trx_i_s_create_lock_id(
+/*===================*/
+					/* out: resulting lock id */
+	const i_s_locks_row_t*	row,	/* in: innodb_locks row */
+	char*			lock_id,/* out: resulting lock_id */
+	ulint			lock_id_size)/* in: size of the lock id
+					buffer */
+{
+	int	res_len;
+
+	/* please adjust TRX_I_S_LOCK_ID_MAX_LEN if you change this */
+
+	if (row->lock_space != ULINT_UNDEFINED) {
+		/* record lock */
+		res_len = ut_snprintf(lock_id, lock_id_size,
+				      TRX_ID_FMT ":%lu:%lu:%lu",
+				      row->lock_trx_id, row->lock_space,
+				      row->lock_page, row->lock_rec);
+	} else {
+		/* table lock */
+		res_len = ut_snprintf(lock_id, lock_id_size,
+				      TRX_ID_FMT ":%llu",
+				      row->lock_trx_id,
+				      row->lock_table_id);
+	}
+
+	/* the typecast is safe because snprintf(3) never returns
+	negative result */
+	ut_a(res_len >= 0);
+	ut_a((ulint) res_len < lock_id_size);
+
+	return(lock_id);
+}
diff --git a/storage/xtradb/trx/trx0purge.c b/storage/xtradb/trx/trx0purge.c
new file mode 100644
index 00000000000..7a2a27a94ff
--- /dev/null
+++ b/storage/xtradb/trx/trx0purge.c
@@ -0,0 +1,1175 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Purge old versions
+
+Created 3/26/1996 Heikki Tuuri
+*******************************************************/
+
+#include "trx0purge.h"
+
+#ifdef UNIV_NONINL
+#include "trx0purge.ic"
+#endif
+
+#include "fsp0fsp.h"
+#include "mach0data.h"
+#include "trx0rseg.h"
+#include "trx0trx.h"
+#include "trx0roll.h"
+#include "read0read.h"
+#include "fut0fut.h"
+#include "que0que.h"
+#include "row0purge.h"
+#include "row0upd.h"
+#include "trx0rec.h"
+#include "srv0que.h"
+#include "os0thread.h"
+
+/* The global data structure coordinating a purge */
+UNIV_INTERN trx_purge_t*	purge_sys = NULL;
+
+/* A dummy undo record used as a return value when we have a whole undo log
+which needs no purge */
+UNIV_INTERN trx_undo_rec_t	trx_purge_dummy_rec;
+
+/*********************************************************************
+Checks if trx_id is >= purge_view: then it is guaranteed that its update
+undo log still exists in the system. */
+UNIV_INTERN
+ibool
+trx_purge_update_undo_must_exist(
+/*=============================*/
+			/* out: TRUE if is sure that it is preserved, also
+			if the function returns FALSE, it is possible that
+			the undo log still exists in the system */
+	dulint	trx_id)	/* in: transaction id */
+{
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(rw_lock_own(&(purge_sys->latch), RW_LOCK_SHARED));
+#endif /* UNIV_SYNC_DEBUG */
+
+	if (!read_view_sees_trx_id(purge_sys->view, trx_id)) {
+
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
+
+/*=================== PURGE RECORD ARRAY =============================*/
+
+/***********************************************************************
+Stores info of an undo log record during a purge. */
+static
+trx_undo_inf_t*
+trx_purge_arr_store_info(
+/*=====================*/
+			/* out: pointer to the storage cell */
+	dulint	trx_no,	/* in: transaction number */
+	dulint	undo_no)/* in: undo number */
+{
+	trx_undo_inf_t*	cell;
+	trx_undo_arr_t*	arr;
+	ulint		i;
+
+	arr = purge_sys->arr;
+
+	for (i = 0;; i++) {
+		cell = trx_undo_arr_get_nth_info(arr, i);
+
+		if (!(cell->in_use)) {
+			/* Not in use, we may store here */
+			cell->undo_no = undo_no;
+			cell->trx_no = trx_no;
+			cell->in_use = TRUE;
+
+			arr->n_used++;
+
+			return(cell);
+		}
+	}
+}
+
+/***********************************************************************
+Removes info of an undo log record during a purge. */
+UNIV_INLINE
+void
+trx_purge_arr_remove_info(
+/*======================*/
+	trx_undo_inf_t*	cell)	/* in: pointer to the storage cell */
+{
+	trx_undo_arr_t*	arr;
+
+	arr = purge_sys->arr;
+
+	cell->in_use = FALSE;
+
+	ut_ad(arr->n_used > 0);
+
+	arr->n_used--;
+}
+
+/***********************************************************************
+Gets the biggest pair of a trx number and an undo number in a purge array. */
+static
+void
+trx_purge_arr_get_biggest(
+/*======================*/
+	trx_undo_arr_t*	arr,	/* in: purge array */
+	dulint*		trx_no,	/* out: transaction number: ut_dulint_zero
+				if array is empty */
+	dulint*		undo_no)/* out: undo number */
+{
+	trx_undo_inf_t*	cell;
+	dulint		pair_trx_no;
+	dulint		pair_undo_no;
+	int		trx_cmp;
+	ulint		n_used;
+	ulint		i;
+	ulint		n;
+
+	n = 0;
+	n_used = arr->n_used;
+	pair_trx_no = ut_dulint_zero;
+	pair_undo_no = ut_dulint_zero;
+
+	for (i = 0;; i++) {
+		cell = trx_undo_arr_get_nth_info(arr, i);
+
+		if (cell->in_use) {
+			n++;
+			trx_cmp = ut_dulint_cmp(cell->trx_no, pair_trx_no);
+
+			if ((trx_cmp > 0)
+			    || ((trx_cmp == 0)
+				&& (ut_dulint_cmp(cell->undo_no,
+						  pair_undo_no) >= 0))) {
+
+				pair_trx_no = cell->trx_no;
+				pair_undo_no = cell->undo_no;
+			}
+		}
+
+		if (n == n_used) {
+			*trx_no = pair_trx_no;
+			*undo_no = pair_undo_no;
+
+			return;
+		}
+	}
+}
+
+/********************************************************************
+Builds a purge 'query' graph. The actual purge is performed by executing
+this query graph. */
+static
+que_t*
+trx_purge_graph_build(void)
+/*=======================*/
+				/* out, own: the query graph */
+{
+	mem_heap_t*	heap;
+	que_fork_t*	fork;
+	que_thr_t*	thr;
+	/*	que_thr_t*	thr2; */
+
+	heap = mem_heap_create(512);
+	fork = que_fork_create(NULL, NULL, QUE_FORK_PURGE, heap);
+	fork->trx = purge_sys->trx;
+
+	thr = que_thr_create(fork, heap);
+
+	thr->child = row_purge_node_create(thr, heap);
+
+	/*	thr2 = que_thr_create(fork, fork, heap);
+
+	thr2->child = row_purge_node_create(fork, thr2, heap);	 */
+
+	return(fork);
+}
+
+/************************************************************************
+Creates the global purge system control structure and inits the history
+mutex. */
+UNIV_INTERN
+void
+trx_purge_sys_create(void)
+/*======================*/
+{
+	ut_ad(mutex_own(&kernel_mutex));
+
+	purge_sys = mem_alloc(sizeof(trx_purge_t));
+
+	purge_sys->state = TRX_STOP_PURGE;
+
+	purge_sys->n_pages_handled = 0;
+
+	purge_sys->purge_trx_no = ut_dulint_zero;
+	purge_sys->purge_undo_no = ut_dulint_zero;
+	purge_sys->next_stored = FALSE;
+
+	rw_lock_create(&purge_sys->latch, SYNC_PURGE_LATCH);
+
+	mutex_create(&purge_sys->mutex, SYNC_PURGE_SYS);
+
+	purge_sys->heap = mem_heap_create(256);
+
+	purge_sys->arr = trx_undo_arr_create();
+
+	purge_sys->sess = sess_open();
+
+	purge_sys->trx = purge_sys->sess->trx;
+
+	purge_sys->trx->is_purge = 1;
+
+	ut_a(trx_start_low(purge_sys->trx, ULINT_UNDEFINED));
+
+	purge_sys->query = trx_purge_graph_build();
+
+	purge_sys->view = read_view_oldest_copy_or_open_new(ut_dulint_zero,
+							    purge_sys->heap);
+}
+
+/*================ UNDO LOG HISTORY LIST =============================*/
+
+/************************************************************************
+Adds the update undo log as the first log in the history list. Removes the
+update undo log segment from the rseg slot if it is too big for reuse. */
+UNIV_INTERN
+void
+trx_purge_add_update_undo_to_history(
+/*=================================*/
+	trx_t*	trx,		/* in: transaction */
+	page_t*	undo_page,	/* in: update undo log header page,
+				x-latched */
+	mtr_t*	mtr)		/* in: mtr */
+{
+	trx_undo_t*	undo;
+	trx_rseg_t*	rseg;
+	trx_rsegf_t*	rseg_header;
+	trx_usegf_t*	seg_header;
+	trx_ulogf_t*	undo_header;
+	trx_upagef_t*	page_header;
+	ulint		hist_size;
+
+	undo = trx->update_undo;
+
+	ut_ad(undo);
+
+	rseg = undo->rseg;
+
+	ut_ad(mutex_own(&(rseg->mutex)));
+
+	rseg_header = trx_rsegf_get(rseg->space, rseg->zip_size,
+				    rseg->page_no, mtr);
+
+	undo_header = undo_page + undo->hdr_offset;
+	seg_header  = undo_page + TRX_UNDO_SEG_HDR;
+	page_header = undo_page + TRX_UNDO_PAGE_HDR;
+
+	if (undo->state != TRX_UNDO_CACHED) {
+		/* The undo log segment will not be reused */
+
+		if (undo->id >= TRX_RSEG_N_SLOTS) {
+			fprintf(stderr,
+				"InnoDB: Error: undo->id is %lu\n",
+				(ulong) undo->id);
+			ut_error;
+		}
+
+		trx_rsegf_set_nth_undo(rseg_header, undo->id, FIL_NULL, mtr);
+
+		hist_size = mtr_read_ulint(rseg_header + TRX_RSEG_HISTORY_SIZE,
+					   MLOG_4BYTES, mtr);
+		ut_ad(undo->size == flst_get_len(
+			      seg_header + TRX_UNDO_PAGE_LIST, mtr));
+
+		mlog_write_ulint(rseg_header + TRX_RSEG_HISTORY_SIZE,
+				 hist_size + undo->size, MLOG_4BYTES, mtr);
+	}
+
+	/* Add the log as the first in the history list */
+	flst_add_first(rseg_header + TRX_RSEG_HISTORY,
+		       undo_header + TRX_UNDO_HISTORY_NODE, mtr);
+	mutex_enter(&kernel_mutex);
+	trx_sys->rseg_history_len++;
+	mutex_exit(&kernel_mutex);
+
+	/* Write the trx number to the undo log header */
+	mlog_write_dulint(undo_header + TRX_UNDO_TRX_NO, trx->no, mtr);
+	/* Write information about delete markings to the undo log header */
+
+	if (!undo->del_marks) {
+		mlog_write_ulint(undo_header + TRX_UNDO_DEL_MARKS, FALSE,
+				 MLOG_2BYTES, mtr);
+	}
+
+	if (rseg->last_page_no == FIL_NULL) {
+
+		rseg->last_page_no = undo->hdr_page_no;
+		rseg->last_offset = undo->hdr_offset;
+		rseg->last_trx_no = trx->no;
+		rseg->last_del_marks = undo->del_marks;
+	}
+}
+
+/**************************************************************************
+Frees an undo log segment which is in the history list. Cuts the end of the
+history list at the youngest undo log in this segment. */
+static
+void
+trx_purge_free_segment(
+/*===================*/
+	trx_rseg_t*	rseg,		/* in: rollback segment */
+	fil_addr_t	hdr_addr,	/* in: the file address of log_hdr */
+	ulint		n_removed_logs)	/* in: count of how many undo logs we
+					will cut off from the end of the
+					history list */
+{
+	page_t*		undo_page;
+	trx_rsegf_t*	rseg_hdr;
+	trx_ulogf_t*	log_hdr;
+	trx_usegf_t*	seg_hdr;
+	ibool		freed;
+	ulint		seg_size;
+	ulint		hist_size;
+	ibool		marked		= FALSE;
+	mtr_t		mtr;
+
+	/*	fputs("Freeing an update undo log segment\n", stderr); */
+
+	ut_ad(mutex_own(&(purge_sys->mutex)));
+loop:
+	mtr_start(&mtr);
+	mutex_enter(&(rseg->mutex));
+
+	rseg_hdr = trx_rsegf_get(rseg->space, rseg->zip_size,
+				 rseg->page_no, &mtr);
+
+	undo_page = trx_undo_page_get(rseg->space, rseg->zip_size,
+				      hdr_addr.page, &mtr);
+	seg_hdr = undo_page + TRX_UNDO_SEG_HDR;
+	log_hdr = undo_page + hdr_addr.boffset;
+
+	/* Mark the last undo log totally purged, so that if the system
+	crashes, the tail of the undo log will not get accessed again. The
+	list of pages in the undo log tail gets inconsistent during the
+	freeing of the segment, and therefore purge should not try to access
+	them again. */
+
+	if (!marked) {
+		mlog_write_ulint(log_hdr + TRX_UNDO_DEL_MARKS, FALSE,
+				 MLOG_2BYTES, &mtr);
+		marked = TRUE;
+	}
+
+	freed = fseg_free_step_not_header(seg_hdr + TRX_UNDO_FSEG_HEADER,
+					  &mtr);
+	if (!freed) {
+		mutex_exit(&(rseg->mutex));
+		mtr_commit(&mtr);
+
+		goto loop;
+	}
+
+	/* The page list may now be inconsistent, but the length field
+	stored in the list base node tells us how big it was before we
+	started the freeing. */
+
+	seg_size = flst_get_len(seg_hdr + TRX_UNDO_PAGE_LIST, &mtr);
+
+	/* We may free the undo log segment header page; it must be freed
+	within the same mtr as the undo log header is removed from the
+	history list: otherwise, in case of a database crash, the segment
+	could become inaccessible garbage in the file space. */
+
+	flst_cut_end(rseg_hdr + TRX_RSEG_HISTORY,
+		     log_hdr + TRX_UNDO_HISTORY_NODE, n_removed_logs, &mtr);
+
+	mutex_enter(&kernel_mutex);
+	ut_ad(trx_sys->rseg_history_len >= n_removed_logs);
+	trx_sys->rseg_history_len -= n_removed_logs;
+	mutex_exit(&kernel_mutex);
+
+	freed = FALSE;
+
+	while (!freed) {
+		/* Here we assume that a file segment with just the header
+		page can be freed in a few steps, so that the buffer pool
+		is not flooded with bufferfixed pages: see the note in
+		fsp0fsp.c. */
+
+		freed = fseg_free_step(seg_hdr + TRX_UNDO_FSEG_HEADER,
+				       &mtr);
+	}
+
+	hist_size = mtr_read_ulint(rseg_hdr + TRX_RSEG_HISTORY_SIZE,
+				   MLOG_4BYTES, &mtr);
+	ut_ad(hist_size >= seg_size);
+
+	mlog_write_ulint(rseg_hdr + TRX_RSEG_HISTORY_SIZE,
+			 hist_size - seg_size, MLOG_4BYTES, &mtr);
+
+	ut_ad(rseg->curr_size >= seg_size);
+
+	rseg->curr_size -= seg_size;
+
+	mutex_exit(&(rseg->mutex));
+
+	mtr_commit(&mtr);
+}
+
+/************************************************************************
+Removes unnecessary history data from a rollback segment. */
+static
+void
+trx_purge_truncate_rseg_history(
+/*============================*/
+	trx_rseg_t*	rseg,		/* in: rollback segment */
+	dulint		limit_trx_no,	/* in: remove update undo logs whose
+					trx number is < limit_trx_no */
+	dulint		limit_undo_no)	/* in: if transaction number is equal
+					to limit_trx_no, truncate undo records
+					with undo number < limit_undo_no */
+{
+	fil_addr_t	hdr_addr;
+	fil_addr_t	prev_hdr_addr;
+	trx_rsegf_t*	rseg_hdr;
+	page_t*		undo_page;
+	trx_ulogf_t*	log_hdr;
+	trx_usegf_t*	seg_hdr;
+	int		cmp;
+	ulint		n_removed_logs	= 0;
+	mtr_t		mtr;
+
+	ut_ad(mutex_own(&(purge_sys->mutex)));
+
+	mtr_start(&mtr);
+	mutex_enter(&(rseg->mutex));
+
+	rseg_hdr = trx_rsegf_get(rseg->space, rseg->zip_size,
+				 rseg->page_no, &mtr);
+
+	hdr_addr = trx_purge_get_log_from_hist(
+		flst_get_last(rseg_hdr + TRX_RSEG_HISTORY, &mtr));
+loop:
+	if (hdr_addr.page == FIL_NULL) {
+
+		mutex_exit(&(rseg->mutex));
+
+		mtr_commit(&mtr);
+
+		return;
+	}
+
+	undo_page = trx_undo_page_get(rseg->space, rseg->zip_size,
+				      hdr_addr.page, &mtr);
+
+	log_hdr = undo_page + hdr_addr.boffset;
+
+	cmp = ut_dulint_cmp(mach_read_from_8(log_hdr + TRX_UNDO_TRX_NO),
+			    limit_trx_no);
+	if (cmp == 0) {
+		trx_undo_truncate_start(rseg, rseg->space, hdr_addr.page,
+					hdr_addr.boffset, limit_undo_no);
+	}
+
+	if (cmp >= 0) {
+		mutex_enter(&kernel_mutex);
+		ut_a(trx_sys->rseg_history_len >= n_removed_logs);
+		trx_sys->rseg_history_len -= n_removed_logs;
+		mutex_exit(&kernel_mutex);
+
+		flst_truncate_end(rseg_hdr + TRX_RSEG_HISTORY,
+				  log_hdr + TRX_UNDO_HISTORY_NODE,
+				  n_removed_logs, &mtr);
+
+		mutex_exit(&(rseg->mutex));
+		mtr_commit(&mtr);
+
+		return;
+	}
+
+	prev_hdr_addr = trx_purge_get_log_from_hist(
+		flst_get_prev_addr(log_hdr + TRX_UNDO_HISTORY_NODE, &mtr));
+	n_removed_logs++;
+
+	seg_hdr = undo_page + TRX_UNDO_SEG_HDR;
+
+	if ((mach_read_from_2(seg_hdr + TRX_UNDO_STATE) == TRX_UNDO_TO_PURGE)
+	    && (mach_read_from_2(log_hdr + TRX_UNDO_NEXT_LOG) == 0)) {
+
+		/* We can free the whole log segment */
+
+		mutex_exit(&(rseg->mutex));
+		mtr_commit(&mtr);
+
+		trx_purge_free_segment(rseg, hdr_addr, n_removed_logs);
+
+		n_removed_logs = 0;
+	} else {
+		mutex_exit(&(rseg->mutex));
+		mtr_commit(&mtr);
+	}
+
+	mtr_start(&mtr);
+	mutex_enter(&(rseg->mutex));
+
+	rseg_hdr = trx_rsegf_get(rseg->space, rseg->zip_size,
+				 rseg->page_no, &mtr);
+
+	hdr_addr = prev_hdr_addr;
+
+	goto loop;
+}
+
+/************************************************************************
+Removes unnecessary history data from rollback segments. NOTE that when this
+function is called, the caller must not have any latches on undo log pages! */
+static
+void
+trx_purge_truncate_history(void)
+/*============================*/
+{
+	trx_rseg_t*	rseg;
+	dulint		limit_trx_no;
+	dulint		limit_undo_no;
+
+	ut_ad(mutex_own(&(purge_sys->mutex)));
+
+	trx_purge_arr_get_biggest(purge_sys->arr, &limit_trx_no,
+				  &limit_undo_no);
+
+	if (ut_dulint_is_zero(limit_trx_no)) {
+
+		limit_trx_no = purge_sys->purge_trx_no;
+		limit_undo_no = purge_sys->purge_undo_no;
+	}
+
+	/* We play safe and set the truncate limit at most to the purge view
+	low_limit number, though this is not necessary */
+
+	if (ut_dulint_cmp(limit_trx_no, purge_sys->view->low_limit_no) >= 0) {
+		limit_trx_no = purge_sys->view->low_limit_no;
+		limit_undo_no = ut_dulint_zero;
+	}
+
+	ut_ad((ut_dulint_cmp(limit_trx_no,
+			     purge_sys->view->low_limit_no) <= 0));
+
+	rseg = UT_LIST_GET_FIRST(trx_sys->rseg_list);
+
+	while (rseg) {
+		trx_purge_truncate_rseg_history(rseg, limit_trx_no,
+						limit_undo_no);
+		rseg = UT_LIST_GET_NEXT(rseg_list, rseg);
+	}
+}
+
+/************************************************************************
+Does a truncate if the purge array is empty. NOTE that when this function is
+called, the caller must not have any latches on undo log pages! */
+UNIV_INLINE
+ibool
+trx_purge_truncate_if_arr_empty(void)
+/*=================================*/
+			/* out: TRUE if array empty */
+{
+	ut_ad(mutex_own(&(purge_sys->mutex)));
+
+	if (purge_sys->arr->n_used == 0) {
+
+		trx_purge_truncate_history();
+
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
+
+/***************************************************************************
+Updates the last not yet purged history log info in rseg when we have purged
+a whole undo log. Advances also purge_sys->purge_trx_no past the purged log. */
+static
+void
+trx_purge_rseg_get_next_history_log(
+/*================================*/
+	trx_rseg_t*	rseg)	/* in: rollback segment */
+{
+	page_t*		undo_page;
+	trx_ulogf_t*	log_hdr;
+	trx_usegf_t*	seg_hdr;
+	fil_addr_t	prev_log_addr;
+	dulint		trx_no;
+	ibool		del_marks;
+	mtr_t		mtr;
+
+	ut_ad(mutex_own(&(purge_sys->mutex)));
+
+	mutex_enter(&(rseg->mutex));
+
+	ut_a(rseg->last_page_no != FIL_NULL);
+
+	purge_sys->purge_trx_no = ut_dulint_add(rseg->last_trx_no, 1);
+	purge_sys->purge_undo_no = ut_dulint_zero;
+	purge_sys->next_stored = FALSE;
+
+	mtr_start(&mtr);
+
+	undo_page = trx_undo_page_get_s_latched(rseg->space, rseg->zip_size,
+						rseg->last_page_no, &mtr);
+	log_hdr = undo_page + rseg->last_offset;
+	seg_hdr = undo_page + TRX_UNDO_SEG_HDR;
+
+	/* Increase the purge page count by one for every handled log */
+
+	purge_sys->n_pages_handled++;
+
+	prev_log_addr = trx_purge_get_log_from_hist(
+		flst_get_prev_addr(log_hdr + TRX_UNDO_HISTORY_NODE, &mtr));
+	if (prev_log_addr.page == FIL_NULL) {
+		/* No logs left in the history list */
+
+		rseg->last_page_no = FIL_NULL;
+
+		mutex_exit(&(rseg->mutex));
+		mtr_commit(&mtr);
+
+		mutex_enter(&kernel_mutex);
+
+		/* Add debug code to track history list corruption reported
+		on the MySQL mailing list on Nov 9, 2004. The fut0lst.c
+		file-based list was corrupt. The prev node pointer was
+		FIL_NULL, even though the list length was over 8 million nodes!
+		We assume that purge truncates the history list in moderate
+		size pieces, and if we here reach the head of the list, the
+		list cannot be longer than 20 000 undo logs now. */
+
+		if (trx_sys->rseg_history_len > 20000) {
+			ut_print_timestamp(stderr);
+			fprintf(stderr,
+				"  InnoDB: Warning: purge reached the"
+				" head of the history list,\n"
+				"InnoDB: but its length is still"
+				" reported as %lu! Make a detailed bug\n"
+				"InnoDB: report, and submit it"
+				" to http://bugs.mysql.com\n",
+				(ulong) trx_sys->rseg_history_len);
+		}
+
+		mutex_exit(&kernel_mutex);
+
+		return;
+	}
+
+	mutex_exit(&(rseg->mutex));
+	mtr_commit(&mtr);
+
+	/* Read the trx number and del marks from the previous log header */
+	mtr_start(&mtr);
+
+	log_hdr = trx_undo_page_get_s_latched(rseg->space, rseg->zip_size,
+					      prev_log_addr.page, &mtr)
+		+ prev_log_addr.boffset;
+
+	trx_no = mach_read_from_8(log_hdr + TRX_UNDO_TRX_NO);
+
+	del_marks = mach_read_from_2(log_hdr + TRX_UNDO_DEL_MARKS);
+
+	mtr_commit(&mtr);
+
+	mutex_enter(&(rseg->mutex));
+
+	rseg->last_page_no = prev_log_addr.page;
+	rseg->last_offset = prev_log_addr.boffset;
+	rseg->last_trx_no = trx_no;
+	rseg->last_del_marks = del_marks;
+
+	mutex_exit(&(rseg->mutex));
+}
+
+/***************************************************************************
+Chooses the next undo log to purge and updates the info in purge_sys. This
+function is used to initialize purge_sys when the next record to purge is
+not known, and also to update the purge system info on the next record when
+purge has handled the whole undo log for a transaction. */
+static
+void
+trx_purge_choose_next_log(void)
+/*===========================*/
+{
+	trx_undo_rec_t*	rec;
+	trx_rseg_t*	rseg;
+	trx_rseg_t*	min_rseg;
+	dulint		min_trx_no;
+	ulint		space = 0;   /* remove warning (??? bug ???) */
+	ulint		zip_size = 0;
+	ulint		page_no = 0; /* remove warning (??? bug ???) */
+	ulint		offset = 0;  /* remove warning (??? bug ???) */
+	mtr_t		mtr;
+
+	ut_ad(mutex_own(&(purge_sys->mutex)));
+	ut_ad(purge_sys->next_stored == FALSE);
+
+	rseg = UT_LIST_GET_FIRST(trx_sys->rseg_list);
+
+	min_trx_no = ut_dulint_max;
+
+	min_rseg = NULL;
+
+	while (rseg) {
+		mutex_enter(&(rseg->mutex));
+
+		if (rseg->last_page_no != FIL_NULL) {
+
+			if ((min_rseg == NULL)
+			    || (ut_dulint_cmp(min_trx_no,
+					      rseg->last_trx_no) > 0)) {
+
+				min_rseg = rseg;
+				min_trx_no = rseg->last_trx_no;
+				space = rseg->space;
+				zip_size = rseg->zip_size;
+				ut_a(space == 0); /* We assume in purge of
+						  externally stored fields
+						  that space id == 0 */
+				page_no = rseg->last_page_no;
+				offset = rseg->last_offset;
+			}
+		}
+
+		mutex_exit(&(rseg->mutex));
+
+		rseg = UT_LIST_GET_NEXT(rseg_list, rseg);
+	}
+
+	if (min_rseg == NULL) {
+
+		return;
+	}
+
+	mtr_start(&mtr);
+
+	if (!min_rseg->last_del_marks) {
+		/* No need to purge this log */
+
+		rec = &trx_purge_dummy_rec;
+	} else {
+		rec = trx_undo_get_first_rec(space, zip_size, page_no, offset,
+					     RW_S_LATCH, &mtr);
+		if (rec == NULL) {
+			/* Undo log empty */
+
+			rec = &trx_purge_dummy_rec;
+		}
+	}
+
+	purge_sys->next_stored = TRUE;
+	purge_sys->rseg = min_rseg;
+
+	purge_sys->hdr_page_no = page_no;
+	purge_sys->hdr_offset = offset;
+
+	purge_sys->purge_trx_no = min_trx_no;
+
+	if (rec == &trx_purge_dummy_rec) {
+
+		purge_sys->purge_undo_no = ut_dulint_zero;
+		purge_sys->page_no = page_no;
+		purge_sys->offset = 0;
+	} else {
+		purge_sys->purge_undo_no = trx_undo_rec_get_undo_no(rec);
+
+		purge_sys->page_no = page_get_page_no(page_align(rec));
+		purge_sys->offset = page_offset(rec);
+	}
+
+	mtr_commit(&mtr);
+}
+
+/***************************************************************************
+Gets the next record to purge and updates the info in the purge system. */
+static
+trx_undo_rec_t*
+trx_purge_get_next_rec(
+/*===================*/
+				/* out: copy of an undo log record or
+				pointer to the dummy undo log record */
+	mem_heap_t*	heap)	/* in: memory heap where copied */
+{
+	trx_undo_rec_t*	rec;
+	trx_undo_rec_t*	rec_copy;
+	trx_undo_rec_t*	rec2;
+	trx_undo_rec_t*	next_rec;
+	page_t*		undo_page;
+	page_t*		page;
+	ulint		offset;
+	ulint		page_no;
+	ulint		space;
+	ulint		zip_size;
+	ulint		type;
+	ulint		cmpl_info;
+	mtr_t		mtr;
+
+	ut_ad(mutex_own(&(purge_sys->mutex)));
+	ut_ad(purge_sys->next_stored);
+
+	space = purge_sys->rseg->space;
+	zip_size = purge_sys->rseg->zip_size;
+	page_no = purge_sys->page_no;
+	offset = purge_sys->offset;
+
+	if (offset == 0) {
+		/* It is the dummy undo log record, which means that there is
+		no need to purge this undo log */
+
+		trx_purge_rseg_get_next_history_log(purge_sys->rseg);
+
+		/* Look for the next undo log and record to purge */
+
+		trx_purge_choose_next_log();
+
+		return(&trx_purge_dummy_rec);
+	}
+
+	mtr_start(&mtr);
+
+	undo_page = trx_undo_page_get_s_latched(space, zip_size,
+						page_no, &mtr);
+	rec = undo_page + offset;
+
+	rec2 = rec;
+
+	for (;;) {
+		/* Try first to find the next record which requires a purge
+		operation from the same page of the same undo log */
+
+		next_rec = trx_undo_page_get_next_rec(rec2,
+						      purge_sys->hdr_page_no,
+						      purge_sys->hdr_offset);
+		if (next_rec == NULL) {
+			rec2 = trx_undo_get_next_rec(
+				rec2, purge_sys->hdr_page_no,
+				purge_sys->hdr_offset, &mtr);
+			break;
+		}
+
+		rec2 = next_rec;
+
+		type = trx_undo_rec_get_type(rec2);
+
+		if (type == TRX_UNDO_DEL_MARK_REC) {
+
+			break;
+		}
+
+		cmpl_info = trx_undo_rec_get_cmpl_info(rec2);
+
+		if (trx_undo_rec_get_extern_storage(rec2)) {
+			break;
+		}
+
+		if ((type == TRX_UNDO_UPD_EXIST_REC)
+		    && !(cmpl_info & UPD_NODE_NO_ORD_CHANGE)) {
+			break;
+		}
+	}
+
+	if (rec2 == NULL) {
+		mtr_commit(&mtr);
+
+		trx_purge_rseg_get_next_history_log(purge_sys->rseg);
+
+		/* Look for the next undo log and record to purge */
+
+		trx_purge_choose_next_log();
+
+		mtr_start(&mtr);
+
+		undo_page = trx_undo_page_get_s_latched(space, zip_size,
+							page_no, &mtr);
+
+		rec = undo_page + offset;
+	} else {
+		page = page_align(rec2);
+
+		purge_sys->purge_undo_no = trx_undo_rec_get_undo_no(rec2);
+		purge_sys->page_no = page_get_page_no(page);
+		purge_sys->offset = rec2 - page;
+
+		if (undo_page != page) {
+			/* We advance to a new page of the undo log: */
+			purge_sys->n_pages_handled++;
+		}
+	}
+
+	rec_copy = trx_undo_rec_copy(rec, heap);
+
+	mtr_commit(&mtr);
+
+	return(rec_copy);
+}
+
+/************************************************************************
+Fetches the next undo log record from the history list to purge. It must be
+released with the corresponding release function. */
+UNIV_INTERN
+trx_undo_rec_t*
+trx_purge_fetch_next_rec(
+/*=====================*/
+				/* out: copy of an undo log record or
+				pointer to the dummy undo log record
+				&trx_purge_dummy_rec, if the whole undo log
+				can skipped in purge; NULL if none left */
+	dulint*		roll_ptr,/* out: roll pointer to undo record */
+	trx_undo_inf_t** cell,	/* out: storage cell for the record in the
+				purge array */
+	mem_heap_t*	heap)	/* in: memory heap where copied */
+{
+	trx_undo_rec_t*	undo_rec;
+
+	mutex_enter(&(purge_sys->mutex));
+
+	if (purge_sys->state == TRX_STOP_PURGE) {
+		trx_purge_truncate_if_arr_empty();
+
+		mutex_exit(&(purge_sys->mutex));
+
+		return(NULL);
+	}
+
+	if (!purge_sys->next_stored) {
+		trx_purge_choose_next_log();
+
+		if (!purge_sys->next_stored) {
+			purge_sys->state = TRX_STOP_PURGE;
+
+			trx_purge_truncate_if_arr_empty();
+
+			if (srv_print_thread_releases) {
+				fprintf(stderr,
+					"Purge: No logs left in the"
+					" history list; pages handled %lu\n",
+					(ulong) purge_sys->n_pages_handled);
+			}
+
+			mutex_exit(&(purge_sys->mutex));
+
+			return(NULL);
+		}
+	}
+
+	if (purge_sys->n_pages_handled >= purge_sys->handle_limit) {
+
+		purge_sys->state = TRX_STOP_PURGE;
+
+		trx_purge_truncate_if_arr_empty();
+
+		mutex_exit(&(purge_sys->mutex));
+
+		return(NULL);
+	}
+
+	if (ut_dulint_cmp(purge_sys->purge_trx_no,
+			  purge_sys->view->low_limit_no) >= 0) {
+		purge_sys->state = TRX_STOP_PURGE;
+
+		trx_purge_truncate_if_arr_empty();
+
+		mutex_exit(&(purge_sys->mutex));
+
+		return(NULL);
+	}
+
+	/*	fprintf(stderr, "Thread %lu purging trx %lu undo record %lu\n",
+	os_thread_get_curr_id(),
+	ut_dulint_get_low(purge_sys->purge_trx_no),
+	ut_dulint_get_low(purge_sys->purge_undo_no)); */
+
+	*roll_ptr = trx_undo_build_roll_ptr(FALSE, (purge_sys->rseg)->id,
+					    purge_sys->page_no,
+					    purge_sys->offset);
+
+	*cell = trx_purge_arr_store_info(purge_sys->purge_trx_no,
+					 purge_sys->purge_undo_no);
+
+	ut_ad(ut_dulint_cmp(purge_sys->purge_trx_no,
+			    (purge_sys->view)->low_limit_no) < 0);
+
+	/* The following call will advance the stored values of purge_trx_no
+	and purge_undo_no, therefore we had to store them first */
+
+	undo_rec = trx_purge_get_next_rec(heap);
+
+	mutex_exit(&(purge_sys->mutex));
+
+	return(undo_rec);
+}
+
+/***********************************************************************
+Releases a reserved purge undo record. */
+UNIV_INTERN
+void
+trx_purge_rec_release(
+/*==================*/
+	trx_undo_inf_t*	cell)	/* in: storage cell */
+{
+	trx_undo_arr_t*	arr;
+
+	mutex_enter(&(purge_sys->mutex));
+
+	arr = purge_sys->arr;
+
+	trx_purge_arr_remove_info(cell);
+
+	mutex_exit(&(purge_sys->mutex));
+}
+
+/***********************************************************************
+This function runs a purge batch. */
+UNIV_INTERN
+ulint
+trx_purge(void)
+/*===========*/
+				/* out: number of undo log pages handled in
+				the batch */
+{
+	que_thr_t*	thr;
+	/*	que_thr_t*	thr2; */
+	ulint		old_pages_handled;
+
+	mutex_enter(&(purge_sys->mutex));
+
+	if (purge_sys->trx->n_active_thrs > 0) {
+
+		mutex_exit(&(purge_sys->mutex));
+
+		/* Should not happen */
+
+		ut_error;
+
+		return(0);
+	}
+
+	rw_lock_x_lock(&(purge_sys->latch));
+
+	mutex_enter(&kernel_mutex);
+
+	/* Close and free the old purge view */
+
+	read_view_close(purge_sys->view);
+	purge_sys->view = NULL;
+	mem_heap_empty(purge_sys->heap);
+
+	/* Determine how much data manipulation language (DML) statements
+	need to be delayed in order to reduce the lagging of the purge
+	thread. */
+	srv_dml_needed_delay = 0; /* in microseconds; default: no delay */
+
+	/* If we cannot advance the 'purge view' because of an old
+	'consistent read view', then the DML statements cannot be delayed.
+	Also, srv_max_purge_lag <= 0 means 'infinity'. */
+	if (srv_max_purge_lag > 0
+	    && !UT_LIST_GET_LAST(trx_sys->view_list)) {
+		float	ratio = (float) trx_sys->rseg_history_len
+			/ srv_max_purge_lag;
+		if (ratio > ULINT_MAX / 10000) {
+			/* Avoid overflow: maximum delay is 4295 seconds */
+			srv_dml_needed_delay = ULINT_MAX;
+		} else if (ratio > 1) {
+			/* If the history list length exceeds the
+			innodb_max_purge_lag, the
+			data manipulation statements are delayed
+			by at least 5000 microseconds. */
+			srv_dml_needed_delay = (ulint) ((ratio - .5) * 10000);
+		}
+	}
+
+	purge_sys->view = read_view_oldest_copy_or_open_new(ut_dulint_zero,
+							    purge_sys->heap);
+	mutex_exit(&kernel_mutex);
+
+	rw_lock_x_unlock(&(purge_sys->latch));
+
+	purge_sys->state = TRX_PURGE_ON;
+
+	/* Handle at most 20 undo log pages in one purge batch */
+
+	purge_sys->handle_limit = purge_sys->n_pages_handled + 20;
+
+	old_pages_handled = purge_sys->n_pages_handled;
+
+	mutex_exit(&(purge_sys->mutex));
+
+	mutex_enter(&kernel_mutex);
+
+	thr = que_fork_start_command(purge_sys->query);
+
+	ut_ad(thr);
+
+	/*	thr2 = que_fork_start_command(purge_sys->query);
+
+	ut_ad(thr2); */
+
+
+	mutex_exit(&kernel_mutex);
+
+	/*	srv_que_task_enqueue(thr2); */
+
+	if (srv_print_thread_releases) {
+
+		fputs("Starting purge\n", stderr);
+	}
+
+	que_run_threads(thr);
+
+	if (srv_print_thread_releases) {
+
+		fprintf(stderr,
+			"Purge ends; pages handled %lu\n",
+			(ulong) purge_sys->n_pages_handled);
+	}
+
+	return(purge_sys->n_pages_handled - old_pages_handled);
+}
+
+/**********************************************************************
+Prints information of the purge system to stderr. */
+UNIV_INTERN
+void
+trx_purge_sys_print(void)
+/*=====================*/
+{
+	fprintf(stderr, "InnoDB: Purge system view:\n");
+	read_view_print(purge_sys->view);
+
+	fprintf(stderr, "InnoDB: Purge trx n:o " TRX_ID_FMT
+		", undo n:o " TRX_ID_FMT "\n",
+		TRX_ID_PREP_PRINTF(purge_sys->purge_trx_no),
+		TRX_ID_PREP_PRINTF(purge_sys->purge_undo_no));
+	fprintf(stderr,
+		"InnoDB: Purge next stored %lu, page_no %lu, offset %lu,\n"
+		"InnoDB: Purge hdr_page_no %lu, hdr_offset %lu\n",
+		(ulong) purge_sys->next_stored,
+		(ulong) purge_sys->page_no,
+		(ulong) purge_sys->offset,
+		(ulong) purge_sys->hdr_page_no,
+		(ulong) purge_sys->hdr_offset);
+}
diff --git a/storage/xtradb/trx/trx0rec.c b/storage/xtradb/trx/trx0rec.c
new file mode 100644
index 00000000000..148f93cdbe7
--- /dev/null
+++ b/storage/xtradb/trx/trx0rec.c
@@ -0,0 +1,1607 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Transaction undo log record
+
+Created 3/26/1996 Heikki Tuuri
+*******************************************************/
+
+#include "trx0rec.h"
+
+#ifdef UNIV_NONINL
+#include "trx0rec.ic"
+#endif
+
+#include "fsp0fsp.h"
+#include "mach0data.h"
+#include "trx0rseg.h"
+#include "trx0trx.h"
+#include "trx0undo.h"
+#include "dict0dict.h"
+#include "ut0mem.h"
+#include "row0ext.h"
+#include "row0upd.h"
+#include "que0que.h"
+#include "trx0purge.h"
+#include "row0row.h"
+
+/*=========== UNDO LOG RECORD CREATION AND DECODING ====================*/
+
+/**************************************************************************
+Writes the mtr log entry of the inserted undo log record on the undo log
+page. */
+UNIV_INLINE
+void
+trx_undof_page_add_undo_rec_log(
+/*============================*/
+	page_t* undo_page,	/* in: undo log page */
+	ulint	old_free,	/* in: start offset of the inserted entry */
+	ulint	new_free,	/* in: end offset of the entry */
+	mtr_t*	mtr)		/* in: mtr */
+{
+	byte*		log_ptr;
+	const byte*	log_end;
+	ulint		len;
+
+	log_ptr = mlog_open(mtr, 11 + 13 + MLOG_BUF_MARGIN);
+
+	if (log_ptr == NULL) {
+
+		return;
+	}
+
+	log_end = &log_ptr[11 + 13 + MLOG_BUF_MARGIN];
+	log_ptr = mlog_write_initial_log_record_fast(
+		undo_page, MLOG_UNDO_INSERT, log_ptr, mtr);
+	len = new_free - old_free - 4;
+
+	mach_write_to_2(log_ptr, len);
+	log_ptr += 2;
+
+	if (log_ptr + len <= log_end) {
+		memcpy(log_ptr, undo_page + old_free + 2, len);
+		mlog_close(mtr, log_ptr + len);
+	} else {
+		mlog_close(mtr, log_ptr);
+		mlog_catenate_string(mtr, undo_page + old_free + 2, len);
+	}
+}
+
+/***************************************************************
+Parses a redo log record of adding an undo log record. */
+UNIV_INTERN
+byte*
+trx_undo_parse_add_undo_rec(
+/*========================*/
+			/* out: end of log record or NULL */
+	byte*	ptr,	/* in: buffer */
+	byte*	end_ptr,/* in: buffer end */
+	page_t*	page)	/* in: page or NULL */
+{
+	ulint	len;
+	byte*	rec;
+	ulint	first_free;
+
+	if (end_ptr < ptr + 2) {
+
+		return(NULL);
+	}
+
+	len = mach_read_from_2(ptr);
+	ptr += 2;
+
+	if (end_ptr < ptr + len) {
+
+		return(NULL);
+	}
+
+	if (page == NULL) {
+
+		return(ptr + len);
+	}
+
+	first_free = mach_read_from_2(page + TRX_UNDO_PAGE_HDR
+				      + TRX_UNDO_PAGE_FREE);
+	rec = page + first_free;
+
+	mach_write_to_2(rec, first_free + 4 + len);
+	mach_write_to_2(rec + 2 + len, first_free);
+
+	mach_write_to_2(page + TRX_UNDO_PAGE_HDR + TRX_UNDO_PAGE_FREE,
+			first_free + 4 + len);
+	ut_memcpy(rec + 2, ptr, len);
+
+	return(ptr + len);
+}
+
+/**************************************************************************
+Calculates the free space left for extending an undo log record. */
+UNIV_INLINE
+ulint
+trx_undo_left(
+/*==========*/
+				/* out: bytes left */
+	const page_t*	page,	/* in: undo log page */
+	const byte*	ptr)	/* in: pointer to page */
+{
+	/* The '- 10' is a safety margin, in case we have some small
+	calculation error below */
+
+	return(UNIV_PAGE_SIZE - (ptr - page) - 10 - FIL_PAGE_DATA_END);
+}
+
+/**************************************************************************
+Set the next and previous pointers in the undo page for the undo record
+that was written to ptr. Update the first free value by the number of bytes
+written for this undo record.*/
+static
+ulint
+trx_undo_page_set_next_prev_and_add(
+/*================================*/
+					/* out: offset of the inserted entry
+					on the page if succeeded, 0 if fail */
+	page_t*		undo_page,	/* in/out: undo log page */
+	byte*		ptr,		/* in: ptr up to where data has been
+					written on this undo page. */
+	mtr_t*		mtr)		/* in: mtr */
+{
+	ulint		first_free;	/* offset within undo_page */
+	ulint		end_of_rec;	/* offset within undo_page */
+	byte*		ptr_to_first_free;
+					/* pointer within undo_page
+					that points to the next free
+					offset value within undo_page.*/
+
+	ut_ad(ptr > undo_page);
+	ut_ad(ptr < undo_page + UNIV_PAGE_SIZE);
+
+	if (UNIV_UNLIKELY(trx_undo_left(undo_page, ptr) < 2)) {
+
+		return(0);
+	}
+
+	ptr_to_first_free = undo_page + TRX_UNDO_PAGE_HDR + TRX_UNDO_PAGE_FREE;
+
+	first_free = mach_read_from_2(ptr_to_first_free);
+
+	/* Write offset of the previous undo log record */
+	mach_write_to_2(ptr, first_free);
+	ptr += 2;
+
+	end_of_rec = ptr - undo_page;
+
+	/* Write offset of the next undo log record */
+	mach_write_to_2(undo_page + first_free, end_of_rec);
+
+	/* Update the offset to first free undo record */
+	mach_write_to_2(ptr_to_first_free, end_of_rec);
+
+	/* Write this log entry to the UNDO log */
+	trx_undof_page_add_undo_rec_log(undo_page, first_free,
+					end_of_rec, mtr);
+
+	return(first_free);
+}
+
+/**************************************************************************
+Reports in the undo log of an insert of a clustered index record. */
+static
+ulint
+trx_undo_page_report_insert(
+/*========================*/
+					/* out: offset of the inserted entry
+					on the page if succeed, 0 if fail */
+	page_t*		undo_page,	/* in: undo log page */
+	trx_t*		trx,		/* in: transaction */
+	dict_index_t*	index,		/* in: clustered index */
+	const dtuple_t*	clust_entry,	/* in: index entry which will be
+					inserted to the clustered index */
+	mtr_t*		mtr)		/* in: mtr */
+{
+	ulint		first_free;
+	byte*		ptr;
+	ulint		i;
+
+	ut_ad(dict_index_is_clust(index));
+	ut_ad(mach_read_from_2(undo_page + TRX_UNDO_PAGE_HDR
+			       + TRX_UNDO_PAGE_TYPE) == TRX_UNDO_INSERT);
+
+	first_free = mach_read_from_2(undo_page + TRX_UNDO_PAGE_HDR
+				      + TRX_UNDO_PAGE_FREE);
+	ptr = undo_page + first_free;
+
+	ut_ad(first_free <= UNIV_PAGE_SIZE);
+
+	if (trx_undo_left(undo_page, ptr) < 2 + 1 + 11 + 11) {
+
+		/* Not enough space for writing the general parameters */
+
+		return(0);
+	}
+
+	/* Reserve 2 bytes for the pointer to the next undo log record */
+	ptr += 2;
+
+	/* Store first some general parameters to the undo log */
+	*ptr++ = TRX_UNDO_INSERT_REC;
+	ptr += mach_dulint_write_much_compressed(ptr, trx->undo_no);
+	ptr += mach_dulint_write_much_compressed(ptr, index->table->id);
+	/*----------------------------------------*/
+	/* Store then the fields required to uniquely determine the record
+	to be inserted in the clustered index */
+
+	for (i = 0; i < dict_index_get_n_unique(index); i++) {
+
+		const dfield_t*	field	= dtuple_get_nth_field(clust_entry, i);
+		ulint		flen	= dfield_get_len(field);
+
+		if (trx_undo_left(undo_page, ptr) < 5) {
+
+			return(0);
+		}
+
+		ptr += mach_write_compressed(ptr, flen);
+
+		if (flen != UNIV_SQL_NULL) {
+			if (trx_undo_left(undo_page, ptr) < flen) {
+
+				return(0);
+			}
+
+			ut_memcpy(ptr, dfield_get_data(field), flen);
+			ptr += flen;
+		}
+	}
+
+	return(trx_undo_page_set_next_prev_and_add(undo_page, ptr, mtr));
+}
+
+/**************************************************************************
+Reads from an undo log record the general parameters. */
+UNIV_INTERN
+byte*
+trx_undo_rec_get_pars(
+/*==================*/
+					/* out: remaining part of undo log
+					record after reading these values */
+	trx_undo_rec_t*	undo_rec,	/* in: undo log record */
+	ulint*		type,		/* out: undo record type:
+					TRX_UNDO_INSERT_REC, ... */
+	ulint*		cmpl_info,	/* out: compiler info, relevant only
+					for update type records */
+	ibool*		updated_extern,	/* out: TRUE if we updated an
+					externally stored fild */
+	dulint*		undo_no,	/* out: undo log record number */
+	dulint*		table_id)	/* out: table id */
+{
+	byte*		ptr;
+	ulint		type_cmpl;
+
+	ptr = undo_rec + 2;
+
+	type_cmpl = mach_read_from_1(ptr);
+	ptr++;
+
+	if (type_cmpl & TRX_UNDO_UPD_EXTERN) {
+		*updated_extern = TRUE;
+		type_cmpl -= TRX_UNDO_UPD_EXTERN;
+	} else {
+		*updated_extern = FALSE;
+	}
+
+	*type = type_cmpl & (TRX_UNDO_CMPL_INFO_MULT - 1);
+	*cmpl_info = type_cmpl / TRX_UNDO_CMPL_INFO_MULT;
+
+	*undo_no = mach_dulint_read_much_compressed(ptr);
+	ptr += mach_dulint_get_much_compressed_size(*undo_no);
+
+	*table_id = mach_dulint_read_much_compressed(ptr);
+	ptr += mach_dulint_get_much_compressed_size(*table_id);
+
+	return(ptr);
+}
+
+/**************************************************************************
+Reads from an undo log record a stored column value. */
+static
+byte*
+trx_undo_rec_get_col_val(
+/*=====================*/
+			/* out: remaining part of undo log record after
+			reading these values */
+	byte*	ptr,	/* in: pointer to remaining part of undo log record */
+	byte**	field,	/* out: pointer to stored field */
+	ulint*	len,	/* out: length of the field, or UNIV_SQL_NULL */
+	ulint*	orig_len)/* out: original length of the locally
+			stored part of an externally stored column, or 0 */
+{
+	*len = mach_read_compressed(ptr);
+	ptr += mach_get_compressed_size(*len);
+
+	*orig_len = 0;
+
+	switch (*len) {
+	case UNIV_SQL_NULL:
+		*field = NULL;
+		break;
+	case UNIV_EXTERN_STORAGE_FIELD:
+		*orig_len = mach_read_compressed(ptr);
+		ptr += mach_get_compressed_size(*orig_len);
+		*len = mach_read_compressed(ptr);
+		ptr += mach_get_compressed_size(*len);
+		*field = ptr;
+		ptr += *len;
+
+		ut_ad(*orig_len >= BTR_EXTERN_FIELD_REF_SIZE);
+		ut_ad(*len > *orig_len);
+		ut_ad(*len >= REC_MAX_INDEX_COL_LEN
+		      + BTR_EXTERN_FIELD_REF_SIZE);
+
+		*len += UNIV_EXTERN_STORAGE_FIELD;
+		break;
+	default:
+		*field = ptr;
+		if (*len >= UNIV_EXTERN_STORAGE_FIELD) {
+			ptr += *len - UNIV_EXTERN_STORAGE_FIELD;
+		} else {
+			ptr += *len;
+		}
+	}
+
+	return(ptr);
+}
+
+/***********************************************************************
+Builds a row reference from an undo log record. */
+UNIV_INTERN
+byte*
+trx_undo_rec_get_row_ref(
+/*=====================*/
+				/* out: pointer to remaining part of undo
+				record */
+	byte*		ptr,	/* in: remaining part of a copy of an undo log
+				record, at the start of the row reference;
+				NOTE that this copy of the undo log record must
+				be preserved as long as the row reference is
+				used, as we do NOT copy the data in the
+				record! */
+	dict_index_t*	index,	/* in: clustered index */
+	dtuple_t**	ref,	/* out, own: row reference */
+	mem_heap_t*	heap)	/* in: memory heap from which the memory
+				needed is allocated */
+{
+	ulint		ref_len;
+	ulint		i;
+
+	ut_ad(index && ptr && ref && heap);
+	ut_a(dict_index_is_clust(index));
+
+	ref_len = dict_index_get_n_unique(index);
+
+	*ref = dtuple_create(heap, ref_len);
+
+	dict_index_copy_types(*ref, index, ref_len);
+
+	for (i = 0; i < ref_len; i++) {
+		dfield_t*	dfield;
+		byte*		field;
+		ulint		len;
+		ulint		orig_len;
+
+		dfield = dtuple_get_nth_field(*ref, i);
+
+		ptr = trx_undo_rec_get_col_val(ptr, &field, &len, &orig_len);
+
+		dfield_set_data(dfield, field, len);
+	}
+
+	return(ptr);
+}
+
+/***********************************************************************
+Skips a row reference from an undo log record. */
+UNIV_INTERN
+byte*
+trx_undo_rec_skip_row_ref(
+/*======================*/
+				/* out: pointer to remaining part of undo
+				record */
+	byte*		ptr,	/* in: remaining part in update undo log
+				record, at the start of the row reference */
+	dict_index_t*	index)	/* in: clustered index */
+{
+	ulint	ref_len;
+	ulint	i;
+
+	ut_ad(index && ptr);
+	ut_a(dict_index_is_clust(index));
+
+	ref_len = dict_index_get_n_unique(index);
+
+	for (i = 0; i < ref_len; i++) {
+		byte*	field;
+		ulint	len;
+		ulint	orig_len;
+
+		ptr = trx_undo_rec_get_col_val(ptr, &field, &len, &orig_len);
+	}
+
+	return(ptr);
+}
+
+/**************************************************************************
+Fetch a prefix of an externally stored column, for writing to the undo log
+of an update or delete marking of a clustered index record. */
+static
+byte*
+trx_undo_page_fetch_ext(
+/*====================*/
+					/* out: ext_buf */
+	byte*		ext_buf,	/* in: a buffer of
+					REC_MAX_INDEX_COL_LEN
+					+ BTR_EXTERN_FIELD_REF_SIZE */
+	ulint		zip_size,	/* compressed page size in bytes,
+					or 0 for uncompressed BLOB  */
+	const byte*	field,		/* in: an externally stored column */
+	ulint*		len)		/* in: length of field;
+					out: used length of ext_buf */
+{
+	/* Fetch the BLOB. */
+	ulint	ext_len = btr_copy_externally_stored_field_prefix(
+		ext_buf, REC_MAX_INDEX_COL_LEN, zip_size, field, *len);
+	/* BLOBs should always be nonempty. */
+	ut_a(ext_len);
+	/* Append the BLOB pointer to the prefix. */
+	memcpy(ext_buf + ext_len,
+	       field + *len - BTR_EXTERN_FIELD_REF_SIZE,
+	       BTR_EXTERN_FIELD_REF_SIZE);
+	*len = ext_len + BTR_EXTERN_FIELD_REF_SIZE;
+	return(ext_buf);
+}
+
+/**************************************************************************
+Writes to the undo log a prefix of an externally stored column. */
+static
+byte*
+trx_undo_page_report_modify_ext(
+/*============================*/
+					/* out: undo log position */
+	byte*		ptr,		/* in: undo log position,
+					at least 15 bytes must be available */
+	byte*		ext_buf,	/* in: a buffer of
+					REC_MAX_INDEX_COL_LEN
+					+ BTR_EXTERN_FIELD_REF_SIZE,
+					or NULL when should not fetch
+					a longer prefix */
+	ulint		zip_size,	/* compressed page size in bytes,
+					or 0 for uncompressed BLOB  */
+	const byte**	field,		/* in/out: the locally stored part of
+					the externally stored column */
+	ulint*		len)		/* in/out: length of field, in bytes */
+{
+	if (ext_buf) {
+		/* If an ordering column is externally stored, we will
+		have to store a longer prefix of the field.  In this
+		case, write to the log a marker followed by the
+		original length and the real length of the field. */
+		ptr += mach_write_compressed(ptr, UNIV_EXTERN_STORAGE_FIELD);
+
+		ptr += mach_write_compressed(ptr, *len);
+
+		*field = trx_undo_page_fetch_ext(ext_buf, zip_size,
+						 *field, len);
+
+		ptr += mach_write_compressed(ptr, *len);
+	} else {
+		ptr += mach_write_compressed(ptr, UNIV_EXTERN_STORAGE_FIELD
+					     + *len);
+	}
+
+	return(ptr);
+}
+
+/**************************************************************************
+Reports in the undo log of an update or delete marking of a clustered index
+record. */
+static
+ulint
+trx_undo_page_report_modify(
+/*========================*/
+					/* out: byte offset of the inserted
+					undo log entry on the page if succeed,
+					0 if fail */
+	page_t*		undo_page,	/* in: undo log page */
+	trx_t*		trx,		/* in: transaction */
+	dict_index_t*	index,		/* in: clustered index where update or
+					delete marking is done */
+	const rec_t*	rec,		/* in: clustered index record which
+					has NOT yet been modified */
+	const ulint*	offsets,	/* in: rec_get_offsets(rec, index) */
+	const upd_t*	update,		/* in: update vector which tells the
+					columns to be updated; in the case of
+					a delete, this should be set to NULL */
+	ulint		cmpl_info,	/* in: compiler info on secondary
+					index updates */
+	mtr_t*		mtr)		/* in: mtr */
+{
+	dict_table_t*	table;
+	ulint		first_free;
+	byte*		ptr;
+	const byte*	field;
+	ulint		flen;
+	ulint		col_no;
+	ulint		type_cmpl;
+	byte*		type_cmpl_ptr;
+	ulint		i;
+	dulint		trx_id;
+	ibool		ignore_prefix = FALSE;
+	byte		ext_buf[REC_MAX_INDEX_COL_LEN
+				+ BTR_EXTERN_FIELD_REF_SIZE];
+
+	ut_a(dict_index_is_clust(index));
+	ut_ad(rec_offs_validate(rec, index, offsets));
+	ut_ad(mach_read_from_2(undo_page + TRX_UNDO_PAGE_HDR
+			       + TRX_UNDO_PAGE_TYPE) == TRX_UNDO_UPDATE);
+	table = index->table;
+
+	first_free = mach_read_from_2(undo_page + TRX_UNDO_PAGE_HDR
+				      + TRX_UNDO_PAGE_FREE);
+	ptr = undo_page + first_free;
+
+	ut_ad(first_free <= UNIV_PAGE_SIZE);
+
+	if (trx_undo_left(undo_page, ptr) < 50) {
+
+		/* NOTE: the value 50 must be big enough so that the general
+		fields written below fit on the undo log page */
+
+		return(0);
+	}
+
+	/* Reserve 2 bytes for the pointer to the next undo log record */
+	ptr += 2;
+
+	/* Store first some general parameters to the undo log */
+
+	if (!update) {
+		type_cmpl = TRX_UNDO_DEL_MARK_REC;
+	} else if (rec_get_deleted_flag(rec, dict_table_is_comp(table))) {
+		type_cmpl = TRX_UNDO_UPD_DEL_REC;
+		/* We are about to update a delete marked record.
+		We don't typically need the prefix in this case unless
+		the delete marking is done by the same transaction
+		(which we check below). */
+		ignore_prefix = TRUE;
+	} else {
+		type_cmpl = TRX_UNDO_UPD_EXIST_REC;
+	}
+
+	type_cmpl |= cmpl_info * TRX_UNDO_CMPL_INFO_MULT;
+	type_cmpl_ptr = ptr;
+
+	*ptr++ = (byte) type_cmpl;
+	ptr += mach_dulint_write_much_compressed(ptr, trx->undo_no);
+
+	ptr += mach_dulint_write_much_compressed(ptr, table->id);
+
+	/*----------------------------------------*/
+	/* Store the state of the info bits */
+
+	*ptr++ = (byte) rec_get_info_bits(rec, dict_table_is_comp(table));
+
+	/* Store the values of the system columns */
+	field = rec_get_nth_field(rec, offsets,
+				  dict_index_get_sys_col_pos(
+					  index, DATA_TRX_ID), &flen);
+	ut_ad(flen == DATA_TRX_ID_LEN);
+
+	trx_id = trx_read_trx_id(field);
+
+	/* If it is an update of a delete marked record, then we are
+	allowed to ignore blob prefixes if the delete marking was done
+	by some other trx as it must have committed by now for us to
+	allow an over-write. */
+	if (ignore_prefix) {
+		ignore_prefix = ut_dulint_cmp(trx_id, trx->id) != 0;
+	}
+	ptr += mach_dulint_write_compressed(ptr, trx_id);
+
+	field = rec_get_nth_field(rec, offsets,
+				  dict_index_get_sys_col_pos(
+					  index, DATA_ROLL_PTR), &flen);
+	ut_ad(flen == DATA_ROLL_PTR_LEN);
+
+	ptr += mach_dulint_write_compressed(ptr, trx_read_roll_ptr(field));
+
+	/*----------------------------------------*/
+	/* Store then the fields required to uniquely determine the
+	record which will be modified in the clustered index */
+
+	for (i = 0; i < dict_index_get_n_unique(index); i++) {
+
+		field = rec_get_nth_field(rec, offsets, i, &flen);
+
+		/* The ordering columns must not be stored externally. */
+		ut_ad(!rec_offs_nth_extern(offsets, i));
+		ut_ad(dict_index_get_nth_col(index, i)->ord_part);
+
+		if (trx_undo_left(undo_page, ptr) < 5) {
+
+			return(0);
+		}
+
+		ptr += mach_write_compressed(ptr, flen);
+
+		if (flen != UNIV_SQL_NULL) {
+			if (trx_undo_left(undo_page, ptr) < flen) {
+
+				return(0);
+			}
+
+			ut_memcpy(ptr, field, flen);
+			ptr += flen;
+		}
+	}
+
+	/*----------------------------------------*/
+	/* Save to the undo log the old values of the columns to be updated. */
+
+	if (update) {
+		if (trx_undo_left(undo_page, ptr) < 5) {
+
+			return(0);
+		}
+
+		ptr += mach_write_compressed(ptr, upd_get_n_fields(update));
+
+		for (i = 0; i < upd_get_n_fields(update); i++) {
+
+			ulint	pos = upd_get_nth_field(update, i)->field_no;
+
+			/* Write field number to undo log */
+			if (trx_undo_left(undo_page, ptr) < 5) {
+
+				return(0);
+			}
+
+			ptr += mach_write_compressed(ptr, pos);
+
+			/* Save the old value of field */
+			field = rec_get_nth_field(rec, offsets, pos, &flen);
+
+			if (trx_undo_left(undo_page, ptr) < 15) {
+
+				return(0);
+			}
+
+			if (rec_offs_nth_extern(offsets, pos)) {
+				ptr = trx_undo_page_report_modify_ext(
+					ptr,
+					dict_index_get_nth_col(index, pos)
+					->ord_part
+					&& !ignore_prefix
+					&& flen < REC_MAX_INDEX_COL_LEN
+					? ext_buf : NULL,
+					dict_table_zip_size(table),
+					&field, &flen);
+
+				/* Notify purge that it eventually has to
+				free the old externally stored field */
+
+				trx->update_undo->del_marks = TRUE;
+
+				*type_cmpl_ptr |= TRX_UNDO_UPD_EXTERN;
+			} else {
+				ptr += mach_write_compressed(ptr, flen);
+			}
+
+			if (flen != UNIV_SQL_NULL) {
+				if (trx_undo_left(undo_page, ptr) < flen) {
+
+					return(0);
+				}
+
+				ut_memcpy(ptr, field, flen);
+				ptr += flen;
+			}
+		}
+	}
+
+	/*----------------------------------------*/
+	/* In the case of a delete marking, and also in the case of an update
+	where any ordering field of any index changes, store the values of all
+	columns which occur as ordering fields in any index. This info is used
+	in the purge of old versions where we use it to build and search the
+	delete marked index records, to look if we can remove them from the
+	index tree. Note that starting from 4.0.14 also externally stored
+	fields can be ordering in some index. Starting from 5.2, we no longer
+	store REC_MAX_INDEX_COL_LEN first bytes to the undo log record,
+	but we can construct the column prefix fields in the index by
+	fetching the first page of the BLOB that is pointed to by the
+	clustered index. This works also in crash recovery, because all pages
+	(including BLOBs) are recovered before anything is rolled back. */
+
+	if (!update || !(cmpl_info & UPD_NODE_NO_ORD_CHANGE)) {
+		byte*	old_ptr = ptr;
+
+		trx->update_undo->del_marks = TRUE;
+
+		if (trx_undo_left(undo_page, ptr) < 5) {
+
+			return(0);
+		}
+
+		/* Reserve 2 bytes to write the number of bytes the stored
+		fields take in this undo record */
+
+		ptr += 2;
+
+		for (col_no = 0; col_no < dict_table_get_n_cols(table);
+		     col_no++) {
+
+			const dict_col_t*	col
+				= dict_table_get_nth_col(table, col_no);
+
+			if (col->ord_part) {
+				ulint	pos;
+
+				/* Write field number to undo log */
+				if (trx_undo_left(undo_page, ptr) < 5 + 15) {
+
+					return(0);
+				}
+
+				pos = dict_index_get_nth_col_pos(index,
+								 col_no);
+				ptr += mach_write_compressed(ptr, pos);
+
+				/* Save the old value of field */
+				field = rec_get_nth_field(rec, offsets, pos,
+							  &flen);
+
+				if (rec_offs_nth_extern(offsets, pos)) {
+					ptr = trx_undo_page_report_modify_ext(
+						ptr,
+						flen < REC_MAX_INDEX_COL_LEN
+						&& !ignore_prefix
+						? ext_buf : NULL,
+						dict_table_zip_size(table),
+						&field, &flen);
+				} else {
+					ptr += mach_write_compressed(
+						ptr, flen);
+				}
+
+				if (flen != UNIV_SQL_NULL) {
+					if (trx_undo_left(undo_page, ptr)
+					    < flen) {
+
+						return(0);
+					}
+
+					ut_memcpy(ptr, field, flen);
+					ptr += flen;
+				}
+			}
+		}
+
+		mach_write_to_2(old_ptr, ptr - old_ptr);
+	}
+
+	/*----------------------------------------*/
+	/* Write pointers to the previous and the next undo log records */
+	if (trx_undo_left(undo_page, ptr) < 2) {
+
+		return(0);
+	}
+
+	mach_write_to_2(ptr, first_free);
+	ptr += 2;
+	mach_write_to_2(undo_page + first_free, ptr - undo_page);
+
+	mach_write_to_2(undo_page + TRX_UNDO_PAGE_HDR + TRX_UNDO_PAGE_FREE,
+			ptr - undo_page);
+
+	/* Write to the REDO log about this change in the UNDO log */
+
+	trx_undof_page_add_undo_rec_log(undo_page, first_free,
+					ptr - undo_page, mtr);
+	return(first_free);
+}
+
+/**************************************************************************
+Reads from an undo log update record the system field values of the old
+version. */
+UNIV_INTERN
+byte*
+trx_undo_update_rec_get_sys_cols(
+/*=============================*/
+				/* out: remaining part of undo log
+				record after reading these values */
+	byte*	ptr,		/* in: remaining part of undo log
+				record after reading general
+				parameters */
+	dulint*	trx_id,		/* out: trx id */
+	dulint*	roll_ptr,	/* out: roll ptr */
+	ulint*	info_bits)	/* out: info bits state */
+{
+	/* Read the state of the info bits */
+	*info_bits = mach_read_from_1(ptr);
+	ptr += 1;
+
+	/* Read the values of the system columns */
+
+	*trx_id = mach_dulint_read_compressed(ptr);
+	ptr += mach_dulint_get_compressed_size(*trx_id);
+
+	*roll_ptr = mach_dulint_read_compressed(ptr);
+	ptr += mach_dulint_get_compressed_size(*roll_ptr);
+
+	return(ptr);
+}
+
+/**************************************************************************
+Reads from an update undo log record the number of updated fields. */
+UNIV_INLINE
+byte*
+trx_undo_update_rec_get_n_upd_fields(
+/*=================================*/
+			/* out: remaining part of undo log record after
+			reading this value */
+	byte*	ptr,	/* in: pointer to remaining part of undo log record */
+	ulint*	n)	/* out: number of fields */
+{
+	*n = mach_read_compressed(ptr);
+	ptr += mach_get_compressed_size(*n);
+
+	return(ptr);
+}
+
+/**************************************************************************
+Reads from an update undo log record a stored field number. */
+UNIV_INLINE
+byte*
+trx_undo_update_rec_get_field_no(
+/*=============================*/
+			/* out: remaining part of undo log record after
+			reading this value */
+	byte*	ptr,	/* in: pointer to remaining part of undo log record */
+	ulint*	field_no)/* out: field number */
+{
+	*field_no = mach_read_compressed(ptr);
+	ptr += mach_get_compressed_size(*field_no);
+
+	return(ptr);
+}
+
+/***********************************************************************
+Builds an update vector based on a remaining part of an undo log record. */
+UNIV_INTERN
+byte*
+trx_undo_update_rec_get_update(
+/*===========================*/
+				/* out: remaining part of the record,
+				NULL if an error detected, which means that
+				the record is corrupted */
+	byte*		ptr,	/* in: remaining part in update undo log
+				record, after reading the row reference
+				NOTE that this copy of the undo log record must
+				be preserved as long as the update vector is
+				used, as we do NOT copy the data in the
+				record! */
+	dict_index_t*	index,	/* in: clustered index */
+	ulint		type,	/* in: TRX_UNDO_UPD_EXIST_REC,
+				TRX_UNDO_UPD_DEL_REC, or
+				TRX_UNDO_DEL_MARK_REC; in the last case,
+				only trx id and roll ptr fields are added to
+				the update vector */
+	dulint		trx_id,	/* in: transaction id from this undo record */
+	dulint		roll_ptr,/* in: roll pointer from this undo record */
+	ulint		info_bits,/* in: info bits from this undo record */
+	trx_t*		trx,	/* in: transaction */
+	mem_heap_t*	heap,	/* in: memory heap from which the memory
+				needed is allocated */
+	upd_t**		upd)	/* out, own: update vector */
+{
+	upd_field_t*	upd_field;
+	upd_t*		update;
+	ulint		n_fields;
+	byte*		buf;
+	ulint		i;
+
+	ut_a(dict_index_is_clust(index));
+
+	if (type != TRX_UNDO_DEL_MARK_REC) {
+		ptr = trx_undo_update_rec_get_n_upd_fields(ptr, &n_fields);
+	} else {
+		n_fields = 0;
+	}
+
+	update = upd_create(n_fields + 2, heap);
+
+	update->info_bits = info_bits;
+
+	/* Store first trx id and roll ptr to update vector */
+
+	upd_field = upd_get_nth_field(update, n_fields);
+	buf = mem_heap_alloc(heap, DATA_TRX_ID_LEN);
+	trx_write_trx_id(buf, trx_id);
+
+	upd_field_set_field_no(upd_field,
+			       dict_index_get_sys_col_pos(index, DATA_TRX_ID),
+			       index, trx);
+	dfield_set_data(&(upd_field->new_val), buf, DATA_TRX_ID_LEN);
+
+	upd_field = upd_get_nth_field(update, n_fields + 1);
+	buf = mem_heap_alloc(heap, DATA_ROLL_PTR_LEN);
+	trx_write_roll_ptr(buf, roll_ptr);
+
+	upd_field_set_field_no(
+		upd_field, dict_index_get_sys_col_pos(index, DATA_ROLL_PTR),
+		index, trx);
+	dfield_set_data(&(upd_field->new_val), buf, DATA_ROLL_PTR_LEN);
+
+	/* Store then the updated ordinary columns to the update vector */
+
+	for (i = 0; i < n_fields; i++) {
+
+		byte*	field;
+		ulint	len;
+		ulint	field_no;
+		ulint	orig_len;
+
+		ptr = trx_undo_update_rec_get_field_no(ptr, &field_no);
+
+		if (field_no >= dict_index_get_n_fields(index)) {
+			fprintf(stderr,
+				"InnoDB: Error: trying to access"
+				" update undo rec field %lu in ",
+				(ulong) field_no);
+			dict_index_name_print(stderr, trx, index);
+			fprintf(stderr, "\n"
+				"InnoDB: but index has only %lu fields\n"
+				"InnoDB: Submit a detailed bug report"
+				" to http://bugs.mysql.com\n"
+				"InnoDB: Run also CHECK TABLE ",
+				(ulong) dict_index_get_n_fields(index));
+			ut_print_name(stderr, trx, TRUE, index->table_name);
+			fprintf(stderr, "\n"
+				"InnoDB: n_fields = %lu, i = %lu, ptr %p\n",
+				(ulong) n_fields, (ulong) i, ptr);
+			return(NULL);
+		}
+
+		upd_field = upd_get_nth_field(update, i);
+
+		upd_field_set_field_no(upd_field, field_no, index, trx);
+
+		ptr = trx_undo_rec_get_col_val(ptr, &field, &len, &orig_len);
+
+		upd_field->orig_len = orig_len;
+
+		if (len == UNIV_SQL_NULL) {
+			dfield_set_null(&upd_field->new_val);
+		} else if (len < UNIV_EXTERN_STORAGE_FIELD) {
+			dfield_set_data(&upd_field->new_val, field, len);
+		} else {
+			len -= UNIV_EXTERN_STORAGE_FIELD;
+
+			dfield_set_data(&upd_field->new_val, field, len);
+			dfield_set_ext(&upd_field->new_val);
+		}
+	}
+
+	*upd = update;
+
+	return(ptr);
+}
+
+/***********************************************************************
+Builds a partial row from an update undo log record. It contains the
+columns which occur as ordering in any index of the table. */
+UNIV_INTERN
+byte*
+trx_undo_rec_get_partial_row(
+/*=========================*/
+				/* out: pointer to remaining part of undo
+				record */
+	byte*		ptr,	/* in: remaining part in update undo log
+				record of a suitable type, at the start of
+				the stored index columns;
+				NOTE that this copy of the undo log record must
+				be preserved as long as the partial row is
+				used, as we do NOT copy the data in the
+				record! */
+	dict_index_t*	index,	/* in: clustered index */
+	dtuple_t**	row,	/* out, own: partial row */
+	ibool		ignore_prefix, /* in: flag to indicate if we
+				expect blob prefixes in undo. Used
+				only in the assertion. */
+	mem_heap_t*	heap)	/* in: memory heap from which the memory
+				needed is allocated */
+{
+	const byte*	end_ptr;
+	ulint		row_len;
+
+	ut_ad(index);
+	ut_ad(ptr);
+	ut_ad(row);
+	ut_ad(heap);
+	ut_ad(dict_index_is_clust(index));
+
+	row_len = dict_table_get_n_cols(index->table);
+
+	*row = dtuple_create(heap, row_len);
+
+	dict_table_copy_types(*row, index->table);
+
+	end_ptr = ptr + mach_read_from_2(ptr);
+	ptr += 2;
+
+	while (ptr != end_ptr) {
+		dfield_t*		dfield;
+		byte*			field;
+		ulint			field_no;
+		const dict_col_t*	col;
+		ulint			col_no;
+		ulint			len;
+		ulint			orig_len;
+
+		ptr = trx_undo_update_rec_get_field_no(ptr, &field_no);
+
+		col = dict_index_get_nth_col(index, field_no);
+		col_no = dict_col_get_no(col);
+
+		ptr = trx_undo_rec_get_col_val(ptr, &field, &len, &orig_len);
+
+		dfield = dtuple_get_nth_field(*row, col_no);
+
+		dfield_set_data(dfield, field, len);
+
+		if (len != UNIV_SQL_NULL
+		    && len >= UNIV_EXTERN_STORAGE_FIELD) {
+			dfield_set_len(dfield,
+				       len - UNIV_EXTERN_STORAGE_FIELD);
+			dfield_set_ext(dfield);
+			/* If the prefix of this column is indexed,
+			ensure that enough prefix is stored in the
+			undo log record. */
+			ut_a(ignore_prefix
+			     || !col->ord_part
+			     || dfield_get_len(dfield)
+			     >= REC_MAX_INDEX_COL_LEN
+			     + BTR_EXTERN_FIELD_REF_SIZE);
+		}
+	}
+
+	return(ptr);
+}
+
+/***************************************************************************
+Erases the unused undo log page end. */
+static
+void
+trx_undo_erase_page_end(
+/*====================*/
+	page_t*	undo_page,	/* in: undo page whose end to erase */
+	mtr_t*	mtr)		/* in: mtr */
+{
+	ulint	first_free;
+
+	first_free = mach_read_from_2(undo_page + TRX_UNDO_PAGE_HDR
+				      + TRX_UNDO_PAGE_FREE);
+	memset(undo_page + first_free, 0xff,
+	       (UNIV_PAGE_SIZE - FIL_PAGE_DATA_END) - first_free);
+
+	mlog_write_initial_log_record(undo_page, MLOG_UNDO_ERASE_END, mtr);
+}
+
+/***************************************************************
+Parses a redo log record of erasing of an undo page end. */
+UNIV_INTERN
+byte*
+trx_undo_parse_erase_page_end(
+/*==========================*/
+			/* out: end of log record or NULL */
+	byte*	ptr,	/* in: buffer */
+	byte*	end_ptr __attribute__((unused)), /* in: buffer end */
+	page_t*	page,	/* in: page or NULL */
+	mtr_t*	mtr)	/* in: mtr or NULL */
+{
+	ut_ad(ptr && end_ptr);
+
+	if (page == NULL) {
+
+		return(ptr);
+	}
+
+	trx_undo_erase_page_end(page, mtr);
+
+	return(ptr);
+}
+
+/***************************************************************************
+Writes information to an undo log about an insert, update, or a delete marking
+of a clustered index record. This information is used in a rollback of the
+transaction and in consistent reads that must look to the history of this
+transaction. */
+UNIV_INTERN
+ulint
+trx_undo_report_row_operation(
+/*==========================*/
+					/* out: DB_SUCCESS or error code */
+	ulint		flags,		/* in: if BTR_NO_UNDO_LOG_FLAG bit is
+					set, does nothing */
+	ulint		op_type,	/* in: TRX_UNDO_INSERT_OP or
+					TRX_UNDO_MODIFY_OP */
+	que_thr_t*	thr,		/* in: query thread */
+	dict_index_t*	index,		/* in: clustered index */
+	const dtuple_t*	clust_entry,	/* in: in the case of an insert,
+					index entry to insert into the
+					clustered index, otherwise NULL */
+	const upd_t*	update,		/* in: in the case of an update,
+					the update vector, otherwise NULL */
+	ulint		cmpl_info,	/* in: compiler info on secondary
+					index updates */
+	const rec_t*	rec,		/* in: in case of an update or delete
+					marking, the record in the clustered
+					index, otherwise NULL */
+	dulint*		roll_ptr)	/* out: rollback pointer to the
+					inserted undo log record,
+					ut_dulint_zero if BTR_NO_UNDO_LOG
+					flag was specified */
+{
+	trx_t*		trx;
+	trx_undo_t*	undo;
+	ulint		page_no;
+	trx_rseg_t*	rseg;
+	mtr_t		mtr;
+	ulint		err		= DB_SUCCESS;
+	mem_heap_t*	heap		= NULL;
+	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
+	ulint*		offsets		= offsets_;
+	rec_offs_init(offsets_);
+
+	ut_a(dict_index_is_clust(index));
+
+	if (flags & BTR_NO_UNDO_LOG_FLAG) {
+
+		*roll_ptr = ut_dulint_zero;
+
+		return(DB_SUCCESS);
+	}
+
+	ut_ad(thr);
+	ut_ad((op_type != TRX_UNDO_INSERT_OP)
+	      || (clust_entry && !update && !rec));
+
+	trx = thr_get_trx(thr);
+	rseg = trx->rseg;
+
+	mutex_enter(&(trx->undo_mutex));
+
+	/* If the undo log is not assigned yet, assign one */
+
+	if (op_type == TRX_UNDO_INSERT_OP) {
+
+		if (trx->insert_undo == NULL) {
+
+			err = trx_undo_assign_undo(trx, TRX_UNDO_INSERT);
+		}
+
+		undo = trx->insert_undo;
+
+		if (UNIV_UNLIKELY(!undo)) {
+			/* Did not succeed */
+			mutex_exit(&(trx->undo_mutex));
+
+			return(err);
+		}
+	} else {
+		ut_ad(op_type == TRX_UNDO_MODIFY_OP);
+
+		if (trx->update_undo == NULL) {
+
+			err = trx_undo_assign_undo(trx, TRX_UNDO_UPDATE);
+
+		}
+
+		undo = trx->update_undo;
+
+		if (UNIV_UNLIKELY(!undo)) {
+			/* Did not succeed */
+			mutex_exit(&(trx->undo_mutex));
+			return(err);
+		}
+
+		offsets = rec_get_offsets(rec, index, offsets,
+					  ULINT_UNDEFINED, &heap);
+	}
+
+	page_no = undo->last_page_no;
+
+	mtr_start(&mtr);
+
+	for (;;) {
+		buf_block_t*	undo_block;
+		page_t*		undo_page;
+		ulint		offset;
+
+		undo_block = buf_page_get_gen(undo->space, undo->zip_size,
+					      page_no, RW_X_LATCH,
+					      undo->guess_block, BUF_GET,
+					      __FILE__, __LINE__, &mtr);
+		buf_block_dbg_add_level(undo_block, SYNC_TRX_UNDO_PAGE);
+
+		undo_page = buf_block_get_frame(undo_block);
+
+		if (op_type == TRX_UNDO_INSERT_OP) {
+			offset = trx_undo_page_report_insert(
+				undo_page, trx, index, clust_entry, &mtr);
+		} else {
+			offset = trx_undo_page_report_modify(
+				undo_page, trx, index, rec, offsets, update,
+				cmpl_info, &mtr);
+		}
+
+		if (UNIV_UNLIKELY(offset == 0)) {
+			/* The record did not fit on the page. We erase the
+			end segment of the undo log page and write a log
+			record of it: this is to ensure that in the debug
+			version the replicate page constructed using the log
+			records stays identical to the original page */
+
+			trx_undo_erase_page_end(undo_page, &mtr);
+			mtr_commit(&mtr);
+		} else {
+			/* Success */
+
+			mtr_commit(&mtr);
+
+			undo->empty = FALSE;
+			undo->top_page_no = page_no;
+			undo->top_offset  = offset;
+			undo->top_undo_no = trx->undo_no;
+			undo->guess_block = undo_block;
+
+			UT_DULINT_INC(trx->undo_no);
+
+			mutex_exit(&trx->undo_mutex);
+
+			*roll_ptr = trx_undo_build_roll_ptr(
+				op_type == TRX_UNDO_INSERT_OP,
+				rseg->id, page_no, offset);
+			if (UNIV_LIKELY_NULL(heap)) {
+				mem_heap_free(heap);
+			}
+			return(DB_SUCCESS);
+		}
+
+		ut_ad(page_no == undo->last_page_no);
+
+		/* We have to extend the undo log by one page */
+
+		mtr_start(&mtr);
+
+		/* When we add a page to an undo log, this is analogous to
+		a pessimistic insert in a B-tree, and we must reserve the
+		counterpart of the tree latch, which is the rseg mutex. */
+
+		mutex_enter(&(rseg->mutex));
+
+		page_no = trx_undo_add_page(trx, undo, &mtr);
+
+		mutex_exit(&(rseg->mutex));
+
+		if (UNIV_UNLIKELY(page_no == FIL_NULL)) {
+			/* Did not succeed: out of space */
+
+			mutex_exit(&(trx->undo_mutex));
+			mtr_commit(&mtr);
+			if (UNIV_LIKELY_NULL(heap)) {
+				mem_heap_free(heap);
+			}
+			return(DB_OUT_OF_FILE_SPACE);
+		}
+	}
+}
+
+/*============== BUILDING PREVIOUS VERSION OF A RECORD ===============*/
+
+/**********************************************************************
+Copies an undo record to heap. This function can be called if we know that
+the undo log record exists. */
+UNIV_INTERN
+trx_undo_rec_t*
+trx_undo_get_undo_rec_low(
+/*======================*/
+					/* out, own: copy of the record */
+	dulint		roll_ptr,	/* in: roll pointer to record */
+	mem_heap_t*	heap)		/* in: memory heap where copied */
+{
+	trx_undo_rec_t*	undo_rec;
+	ulint		rseg_id;
+	ulint		page_no;
+	ulint		offset;
+	page_t*		undo_page;
+	trx_rseg_t*	rseg;
+	ibool		is_insert;
+	mtr_t		mtr;
+
+	trx_undo_decode_roll_ptr(roll_ptr, &is_insert, &rseg_id, &page_no,
+				 &offset);
+	rseg = trx_rseg_get_on_id(rseg_id);
+
+	mtr_start(&mtr);
+
+	undo_page = trx_undo_page_get_s_latched(rseg->space, rseg->zip_size,
+						page_no, &mtr);
+
+	undo_rec = trx_undo_rec_copy(undo_page + offset, heap);
+
+	mtr_commit(&mtr);
+
+	return(undo_rec);
+}
+
+/**********************************************************************
+Copies an undo record to heap. */
+UNIV_INTERN
+ulint
+trx_undo_get_undo_rec(
+/*==================*/
+					/* out: DB_SUCCESS, or
+					DB_MISSING_HISTORY if the undo log
+					has been truncated and we cannot
+					fetch the old version; NOTE: the
+					caller must have latches on the
+					clustered index page and purge_view */
+	dulint		roll_ptr,	/* in: roll pointer to record */
+	dulint		trx_id,		/* in: id of the trx that generated
+					the roll pointer: it points to an
+					undo log of this transaction */
+	trx_undo_rec_t** undo_rec,	/* out, own: copy of the record */
+	mem_heap_t*	heap)		/* in: memory heap where copied */
+{
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(rw_lock_own(&(purge_sys->latch), RW_LOCK_SHARED));
+#endif /* UNIV_SYNC_DEBUG */
+
+	if (!trx_purge_update_undo_must_exist(trx_id)) {
+
+		/* It may be that the necessary undo log has already been
+		deleted */
+
+		return(DB_MISSING_HISTORY);
+	}
+
+	*undo_rec = trx_undo_get_undo_rec_low(roll_ptr, heap);
+
+	return(DB_SUCCESS);
+}
+
+/***********************************************************************
+Build a previous version of a clustered index record. This function checks
+that the caller has a latch on the index page of the clustered index record
+and an s-latch on the purge_view. This guarantees that the stack of versions
+is locked all the way down to the purge_view. */
+UNIV_INTERN
+ulint
+trx_undo_prev_version_build(
+/*========================*/
+				/* out: DB_SUCCESS, or DB_MISSING_HISTORY if
+				the previous version is not >= purge_view,
+				which means that it may have been removed,
+				DB_ERROR if corrupted record */
+	const rec_t*	index_rec,/* in: clustered index record in the
+				index tree */
+	mtr_t*		index_mtr __attribute__((unused)),
+				/* in: mtr which contains the latch to
+				index_rec page and purge_view */
+	const rec_t*	rec,	/* in: version of a clustered index record */
+	dict_index_t*	index,	/* in: clustered index */
+	ulint*		offsets,/* in: rec_get_offsets(rec, index) */
+	mem_heap_t*	heap,	/* in: memory heap from which the memory
+				needed is allocated */
+	rec_t**		old_vers)/* out, own: previous version, or NULL if
+				rec is the first inserted version, or if
+				history data has been deleted (an error),
+				or if the purge COULD have removed the version
+				though it has not yet done so */
+{
+	trx_undo_rec_t*	undo_rec	= NULL;
+	dtuple_t*	entry;
+	dulint		rec_trx_id;
+	ulint		type;
+	dulint		undo_no;
+	dulint		table_id;
+	dulint		trx_id;
+	dulint		roll_ptr;
+	dulint		old_roll_ptr;
+	upd_t*		update;
+	byte*		ptr;
+	ulint		info_bits;
+	ulint		cmpl_info;
+	ibool		dummy_extern;
+	byte*		buf;
+	ulint		err;
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(rw_lock_own(&(purge_sys->latch), RW_LOCK_SHARED));
+#endif /* UNIV_SYNC_DEBUG */
+	ut_ad(mtr_memo_contains_page(index_mtr, index_rec, MTR_MEMO_PAGE_S_FIX)
+	      || mtr_memo_contains_page(index_mtr, index_rec,
+					MTR_MEMO_PAGE_X_FIX));
+	ut_ad(rec_offs_validate(rec, index, offsets));
+
+	if (!dict_index_is_clust(index)) {
+		fprintf(stderr, "InnoDB: Error: trying to access"
+			" update undo rec for non-clustered index %s\n"
+			"InnoDB: Submit a detailed bug report to"
+			" http://bugs.mysql.com\n"
+			"InnoDB: index record ", index->name);
+		rec_print(stderr, index_rec, index);
+		fputs("\n"
+		      "InnoDB: record version ", stderr);
+		rec_print_new(stderr, rec, offsets);
+		putc('\n', stderr);
+		return(DB_ERROR);
+	}
+
+	roll_ptr = row_get_rec_roll_ptr(rec, index, offsets);
+	old_roll_ptr = roll_ptr;
+
+	*old_vers = NULL;
+
+	if (trx_undo_roll_ptr_is_insert(roll_ptr)) {
+
+		/* The record rec is the first inserted version */
+
+		return(DB_SUCCESS);
+	}
+
+	rec_trx_id = row_get_rec_trx_id(rec, index, offsets);
+
+	err = trx_undo_get_undo_rec(roll_ptr, rec_trx_id, &undo_rec, heap);
+
+	if (UNIV_UNLIKELY(err != DB_SUCCESS)) {
+		/* The undo record may already have been purged.
+		This should never happen in InnoDB. */
+
+		return(err);
+	}
+
+	ptr = trx_undo_rec_get_pars(undo_rec, &type, &cmpl_info,
+				    &dummy_extern, &undo_no, &table_id);
+
+	ptr = trx_undo_update_rec_get_sys_cols(ptr, &trx_id, &roll_ptr,
+					       &info_bits);
+
+	/* (a) If a clustered index record version is such that the
+	trx id stamp in it is bigger than purge_sys->view, then the
+	BLOBs in that version are known to exist (the purge has not
+	progressed that far);
+
+	(b) if the version is the first version such that trx id in it
+	is less than purge_sys->view, and it is not delete-marked,
+	then the BLOBs in that version are known to exist (the purge
+	cannot have purged the BLOBs referenced by that version
+	yet).
+
+	This function does not fetch any BLOBs.  The callers might, by
+	possibly invoking row_ext_create() via row_build().  However,
+	they should have all needed information in the *old_vers
+	returned by this function.  This is because *old_vers is based
+	on the transaction undo log records.  The function
+	trx_undo_page_fetch_ext() will write BLOB prefixes to the
+	transaction undo log that are at least as long as the longest
+	possible column prefix in a secondary index.  Thus, secondary
+	index entries for *old_vers can be constructed without
+	dereferencing any BLOB pointers. */
+
+	ptr = trx_undo_rec_skip_row_ref(ptr, index);
+
+	ptr = trx_undo_update_rec_get_update(ptr, index, type, trx_id,
+					     roll_ptr, info_bits,
+					     NULL, heap, &update);
+
+	if (ut_dulint_cmp(table_id, index->table->id) != 0) {
+		ptr = NULL;
+
+		fprintf(stderr,
+			"InnoDB: Error: trying to access update undo rec"
+			" for table %s\n"
+			"InnoDB: but the table id in the"
+			" undo record is wrong\n"
+			"InnoDB: Submit a detailed bug report"
+			" to http://bugs.mysql.com\n"
+			"InnoDB: Run also CHECK TABLE %s\n",
+			index->table_name, index->table_name);
+	}
+
+	if (ptr == NULL) {
+		/* The record was corrupted, return an error; these printfs
+		should catch an elusive bug in row_vers_old_has_index_entry */
+
+		fprintf(stderr,
+			"InnoDB: table %s, index %s, n_uniq %lu\n"
+			"InnoDB: undo rec address %p, type %lu cmpl_info %lu\n"
+			"InnoDB: undo rec table id %lu %lu,"
+			" index table id %lu %lu\n"
+			"InnoDB: dump of 150 bytes in undo rec: ",
+			index->table_name, index->name,
+			(ulong) dict_index_get_n_unique(index),
+			undo_rec, (ulong) type, (ulong) cmpl_info,
+			(ulong) ut_dulint_get_high(table_id),
+			(ulong) ut_dulint_get_low(table_id),
+			(ulong) ut_dulint_get_high(index->table->id),
+			(ulong) ut_dulint_get_low(index->table->id));
+		ut_print_buf(stderr, undo_rec, 150);
+		fputs("\n"
+		      "InnoDB: index record ", stderr);
+		rec_print(stderr, index_rec, index);
+		fputs("\n"
+		      "InnoDB: record version ", stderr);
+		rec_print_new(stderr, rec, offsets);
+		fprintf(stderr, "\n"
+			"InnoDB: Record trx id " TRX_ID_FMT
+			", update rec trx id " TRX_ID_FMT "\n"
+			"InnoDB: Roll ptr in rec %lu %lu, in update rec"
+			" %lu %lu\n",
+			TRX_ID_PREP_PRINTF(rec_trx_id),
+			TRX_ID_PREP_PRINTF(trx_id),
+			(ulong) ut_dulint_get_high(old_roll_ptr),
+			(ulong) ut_dulint_get_low(old_roll_ptr),
+			(ulong) ut_dulint_get_high(roll_ptr),
+			(ulong) ut_dulint_get_low(roll_ptr));
+
+		trx_purge_sys_print();
+		return(DB_ERROR);
+	}
+
+	if (row_upd_changes_field_size_or_external(index, offsets, update)) {
+		ulint	n_ext;
+
+		/* We have to set the appropriate extern storage bits in the
+		old version of the record: the extern bits in rec for those
+		fields that update does NOT update, as well as the the bits for
+		those fields that update updates to become externally stored
+		fields. Store the info: */
+
+		entry = row_rec_to_index_entry(ROW_COPY_DATA, rec, index,
+					       offsets, &n_ext, heap);
+		n_ext += btr_push_update_extern_fields(entry, update, heap);
+		/* The page containing the clustered index record
+		corresponding to entry is latched in mtr.  Thus the
+		following call is safe. */
+		row_upd_index_replace_new_col_vals(entry, index, update, heap);
+
+		buf = mem_heap_alloc(heap, rec_get_converted_size(index, entry,
+								  n_ext));
+
+		*old_vers = rec_convert_dtuple_to_rec(buf, index,
+						      entry, n_ext);
+	} else {
+		buf = mem_heap_alloc(heap, rec_offs_size(offsets));
+		*old_vers = rec_copy(buf, rec, offsets);
+		rec_offs_make_valid(*old_vers, index, offsets);
+		row_upd_rec_in_place(*old_vers, index, offsets, update, NULL);
+	}
+
+	return(DB_SUCCESS);
+}
diff --git a/storage/xtradb/trx/trx0roll.c b/storage/xtradb/trx/trx0roll.c
new file mode 100644
index 00000000000..5f3cb15a254
--- /dev/null
+++ b/storage/xtradb/trx/trx0roll.c
@@ -0,0 +1,1359 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Transaction rollback
+
+Created 3/26/1996 Heikki Tuuri
+*******************************************************/
+
+#include "trx0roll.h"
+
+#ifdef UNIV_NONINL
+#include "trx0roll.ic"
+#endif
+
+#include "fsp0fsp.h"
+#include "mach0data.h"
+#include "trx0rseg.h"
+#include "trx0trx.h"
+#include "trx0undo.h"
+#include "trx0rec.h"
+#include "que0que.h"
+#include "usr0sess.h"
+#include "srv0que.h"
+#include "srv0start.h"
+#include "row0undo.h"
+#include "row0mysql.h"
+#include "lock0lock.h"
+#include "pars0pars.h"
+
+/* This many pages must be undone before a truncate is tried within rollback */
+#define TRX_ROLL_TRUNC_THRESHOLD	1
+
+/* In crash recovery, the current trx to be rolled back */
+static trx_t*		trx_roll_crash_recv_trx	= NULL;
+
+/* In crash recovery we set this to the undo n:o of the current trx to be
+rolled back. Then we can print how many % the rollback has progressed. */
+static ib_int64_t	trx_roll_max_undo_no;
+
+/* Auxiliary variable which tells the previous progress % we printed */
+static ulint		trx_roll_progress_printed_pct;
+
+/***********************************************************************
+Rollback a transaction used in MySQL. */
+UNIV_INTERN
+int
+trx_general_rollback_for_mysql(
+/*===========================*/
+				/* out: error code or DB_SUCCESS */
+	trx_t*		trx,	/* in: transaction handle */
+	ibool		partial,/* in: TRUE if partial rollback requested */
+	trx_savept_t*	savept)	/* in: pointer to savepoint undo number, if
+				partial rollback requested */
+{
+#ifndef UNIV_HOTBACKUP
+	mem_heap_t*	heap;
+	que_thr_t*	thr;
+	roll_node_t*	roll_node;
+
+	/* Tell Innobase server that there might be work for
+	utility threads: */
+
+	srv_active_wake_master_thread();
+
+	trx_start_if_not_started(trx);
+
+	heap = mem_heap_create(512);
+
+	roll_node = roll_node_create(heap);
+
+	roll_node->partial = partial;
+
+	if (partial) {
+		roll_node->savept = *savept;
+	}
+
+	trx->error_state = DB_SUCCESS;
+
+	thr = pars_complete_graph_for_exec(roll_node, trx, heap);
+
+	ut_a(thr == que_fork_start_command(que_node_get_parent(thr)));
+	que_run_threads(thr);
+
+	mutex_enter(&kernel_mutex);
+
+	while (trx->que_state != TRX_QUE_RUNNING) {
+
+		mutex_exit(&kernel_mutex);
+
+		os_thread_sleep(100000);
+
+		mutex_enter(&kernel_mutex);
+	}
+
+	mutex_exit(&kernel_mutex);
+
+	mem_heap_free(heap);
+
+	ut_a(trx->error_state == DB_SUCCESS);
+
+	/* Tell Innobase server that there might be work for
+	utility threads: */
+
+	srv_active_wake_master_thread();
+
+	return((int) trx->error_state);
+#else /* UNIV_HOTBACKUP */
+	/* This function depends on MySQL code that is not included in
+	InnoDB Hot Backup builds.  Besides, this function should never
+	be called in InnoDB Hot Backup. */
+	ut_error;
+	return(DB_FAIL);
+#endif /* UNIV_HOTBACKUP */
+}
+
+/***********************************************************************
+Rollback a transaction used in MySQL. */
+UNIV_INTERN
+int
+trx_rollback_for_mysql(
+/*===================*/
+			/* out: error code or DB_SUCCESS */
+	trx_t*	trx)	/* in: transaction handle */
+{
+	int	err;
+
+	if (trx->conc_state == TRX_NOT_STARTED) {
+
+		return(DB_SUCCESS);
+	}
+
+	trx->op_info = "rollback";
+
+	/* If we are doing the XA recovery of prepared transactions, then
+	the transaction object does not have an InnoDB session object, and we
+	set a dummy session that we use for all MySQL transactions. */
+
+	err = trx_general_rollback_for_mysql(trx, FALSE, NULL);
+
+	trx->op_info = "";
+
+	return(err);
+}
+
+/***********************************************************************
+Rollback the latest SQL statement for MySQL. */
+UNIV_INTERN
+int
+trx_rollback_last_sql_stat_for_mysql(
+/*=================================*/
+			/* out: error code or DB_SUCCESS */
+	trx_t*	trx)	/* in: transaction handle */
+{
+	int	err;
+
+	if (trx->conc_state == TRX_NOT_STARTED) {
+
+		return(DB_SUCCESS);
+	}
+
+	trx->op_info = "rollback of SQL statement";
+
+	err = trx_general_rollback_for_mysql(trx, TRUE,
+					     &(trx->last_sql_stat_start));
+	/* The following call should not be needed, but we play safe: */
+	trx_mark_sql_stat_end(trx);
+
+	trx->op_info = "";
+
+	return(err);
+}
+
+/***********************************************************************
+Frees a single savepoint struct. */
+UNIV_INTERN
+void
+trx_roll_savepoint_free(
+/*=====================*/
+	trx_t*			trx,	/* in: transaction handle */
+	trx_named_savept_t*	savep)	/* in: savepoint to free */
+{
+	ut_a(savep != NULL);
+	ut_a(UT_LIST_GET_LEN(trx->trx_savepoints) > 0);
+
+	UT_LIST_REMOVE(trx_savepoints, trx->trx_savepoints, savep);
+	mem_free(savep->name);
+	mem_free(savep);
+}
+
+/***********************************************************************
+Frees savepoint structs starting from savep, if savep == NULL then
+free all savepoints. */
+
+void
+trx_roll_savepoints_free(
+/*=====================*/
+	trx_t*			trx,	/* in: transaction handle */
+	trx_named_savept_t*	savep)	/* in: free all savepoints > this one;
+					if this is NULL, free all savepoints
+					of trx */
+{
+	trx_named_savept_t*	next_savep;
+
+	if (savep == NULL) {
+		savep = UT_LIST_GET_FIRST(trx->trx_savepoints);
+	} else {
+		savep = UT_LIST_GET_NEXT(trx_savepoints, savep);
+	}
+
+	while (savep != NULL) {
+		next_savep = UT_LIST_GET_NEXT(trx_savepoints, savep);
+
+		trx_roll_savepoint_free(trx, savep);
+
+		savep = next_savep;
+	}
+}
+
+/***********************************************************************
+Rolls back a transaction back to a named savepoint. Modifications after the
+savepoint are undone but InnoDB does NOT release the corresponding locks
+which are stored in memory. If a lock is 'implicit', that is, a new inserted
+row holds a lock where the lock information is carried by the trx id stored in
+the row, these locks are naturally released in the rollback. Savepoints which
+were set after this savepoint are deleted. */
+UNIV_INTERN
+ulint
+trx_rollback_to_savepoint_for_mysql(
+/*================================*/
+						/* out: if no savepoint
+						of the name found then
+						DB_NO_SAVEPOINT,
+						otherwise DB_SUCCESS */
+	trx_t*		trx,			/* in: transaction handle */
+	const char*	savepoint_name,		/* in: savepoint name */
+	ib_int64_t*	mysql_binlog_cache_pos)	/* out: the MySQL binlog cache
+						position corresponding to this
+						savepoint; MySQL needs this
+						information to remove the
+						binlog entries of the queries
+						executed after the savepoint */
+{
+	trx_named_savept_t*	savep;
+	ulint			err;
+
+	savep = UT_LIST_GET_FIRST(trx->trx_savepoints);
+
+	while (savep != NULL) {
+		if (0 == ut_strcmp(savep->name, savepoint_name)) {
+			/* Found */
+			break;
+		}
+		savep = UT_LIST_GET_NEXT(trx_savepoints, savep);
+	}
+
+	if (savep == NULL) {
+
+		return(DB_NO_SAVEPOINT);
+	}
+
+	if (trx->conc_state == TRX_NOT_STARTED) {
+		ut_print_timestamp(stderr);
+		fputs("  InnoDB: Error: transaction has a savepoint ", stderr);
+		ut_print_name(stderr, trx, FALSE, savep->name);
+		fputs(" though it is not started\n", stderr);
+		return(DB_ERROR);
+	}
+
+	/* We can now free all savepoints strictly later than this one */
+
+	trx_roll_savepoints_free(trx, savep);
+
+	*mysql_binlog_cache_pos = savep->mysql_binlog_cache_pos;
+
+	trx->op_info = "rollback to a savepoint";
+
+	err = trx_general_rollback_for_mysql(trx, TRUE, &(savep->savept));
+
+	/* Store the current undo_no of the transaction so that we know where
+	to roll back if we have to roll back the next SQL statement: */
+
+	trx_mark_sql_stat_end(trx);
+
+	trx->op_info = "";
+
+	return(err);
+}
+
+/***********************************************************************
+Creates a named savepoint. If the transaction is not yet started, starts it.
+If there is already a savepoint of the same name, this call erases that old
+savepoint and replaces it with a new. Savepoints are deleted in a transaction
+commit or rollback. */
+UNIV_INTERN
+ulint
+trx_savepoint_for_mysql(
+/*====================*/
+						/* out: always DB_SUCCESS */
+	trx_t*		trx,			/* in: transaction handle */
+	const char*	savepoint_name,		/* in: savepoint name */
+	ib_int64_t	binlog_cache_pos)	/* in: MySQL binlog cache
+						position corresponding to this
+						connection at the time of the
+						savepoint */
+{
+	trx_named_savept_t*	savep;
+
+	ut_a(trx);
+	ut_a(savepoint_name);
+
+	trx_start_if_not_started(trx);
+
+	savep = UT_LIST_GET_FIRST(trx->trx_savepoints);
+
+	while (savep != NULL) {
+		if (0 == ut_strcmp(savep->name, savepoint_name)) {
+			/* Found */
+			break;
+		}
+		savep = UT_LIST_GET_NEXT(trx_savepoints, savep);
+	}
+
+	if (savep) {
+		/* There is a savepoint with the same name: free that */
+
+		UT_LIST_REMOVE(trx_savepoints, trx->trx_savepoints, savep);
+
+		mem_free(savep->name);
+		mem_free(savep);
+	}
+
+	/* Create a new savepoint and add it as the last in the list */
+
+	savep = mem_alloc(sizeof(trx_named_savept_t));
+
+	savep->name = mem_strdup(savepoint_name);
+
+	savep->savept = trx_savept_take(trx);
+
+	savep->mysql_binlog_cache_pos = binlog_cache_pos;
+
+	UT_LIST_ADD_LAST(trx_savepoints, trx->trx_savepoints, savep);
+
+	return(DB_SUCCESS);
+}
+
+/***********************************************************************
+Releases only the named savepoint. Savepoints which were set after this
+savepoint are left as is. */
+UNIV_INTERN
+ulint
+trx_release_savepoint_for_mysql(
+/*============================*/
+						/* out: if no savepoint
+						of the name found then
+						DB_NO_SAVEPOINT,
+						otherwise DB_SUCCESS */
+	trx_t*		trx,			/* in: transaction handle */
+	const char*	savepoint_name)		/* in: savepoint name */
+{
+	trx_named_savept_t*	savep;
+
+	savep = UT_LIST_GET_FIRST(trx->trx_savepoints);
+
+	/* Search for the savepoint by name and free if found. */
+	while (savep != NULL) {
+		if (0 == ut_strcmp(savep->name, savepoint_name)) {
+			trx_roll_savepoint_free(trx, savep);
+			return(DB_SUCCESS);
+		}
+		savep = UT_LIST_GET_NEXT(trx_savepoints, savep);
+	}
+
+	return(DB_NO_SAVEPOINT);
+}
+
+/***********************************************************************
+Determines if this transaction is rolling back an incomplete transaction
+in crash recovery. */
+UNIV_INTERN
+ibool
+trx_is_recv(
+/*========*/
+				/* out: TRUE if trx is an incomplete
+				transaction that is being rolled back
+				in crash recovery */
+	const trx_t*	trx)	/* in: transaction */
+{
+	return(trx == trx_roll_crash_recv_trx);
+}
+
+/***********************************************************************
+Returns a transaction savepoint taken at this point in time. */
+UNIV_INTERN
+trx_savept_t
+trx_savept_take(
+/*============*/
+			/* out: savepoint */
+	trx_t*	trx)	/* in: transaction */
+{
+	trx_savept_t	savept;
+
+	savept.least_undo_no = trx->undo_no;
+
+	return(savept);
+}
+
+/***********************************************************************
+Roll back an active transaction. */
+static
+void
+trx_rollback_active(
+/*================*/
+	trx_t*	trx)	/* in/out: transaction */
+{
+	mem_heap_t*	heap;
+	que_fork_t*	fork;
+	que_thr_t*	thr;
+	roll_node_t*	roll_node;
+	dict_table_t*	table;
+	ib_int64_t	rows_to_undo;
+	const char*	unit		= "";
+	ibool		dictionary_locked = FALSE;
+
+	heap = mem_heap_create(512);
+
+	fork = que_fork_create(NULL, NULL, QUE_FORK_RECOVERY, heap);
+	fork->trx = trx;
+
+	thr = que_thr_create(fork, heap);
+
+	roll_node = roll_node_create(heap);
+
+	thr->child = roll_node;
+	roll_node->common.parent = thr;
+
+	mutex_enter(&kernel_mutex);
+
+	trx->graph = fork;
+
+	ut_a(thr == que_fork_start_command(fork));
+
+	trx_roll_crash_recv_trx	= trx;
+	trx_roll_max_undo_no = ut_conv_dulint_to_longlong(trx->undo_no);
+	trx_roll_progress_printed_pct = 0;
+	rows_to_undo = trx_roll_max_undo_no;
+
+	if (rows_to_undo > 1000000000) {
+		rows_to_undo = rows_to_undo / 1000000;
+		unit = "M";
+	}
+
+	ut_print_timestamp(stderr);
+	fprintf(stderr,
+		"  InnoDB: Rolling back trx with id " TRX_ID_FMT ", %lu%s"
+		" rows to undo\n",
+		TRX_ID_PREP_PRINTF(trx->id),
+		(ulong) rows_to_undo, unit);
+	mutex_exit(&kernel_mutex);
+
+	trx->mysql_thread_id = os_thread_get_curr_id();
+
+	trx->mysql_process_no = os_proc_get_number();
+
+	if (trx_get_dict_operation(trx) != TRX_DICT_OP_NONE) {
+		row_mysql_lock_data_dictionary(trx);
+		dictionary_locked = TRUE;
+	}
+
+	que_run_threads(thr);
+
+	mutex_enter(&kernel_mutex);
+
+	while (trx->que_state != TRX_QUE_RUNNING) {
+
+		mutex_exit(&kernel_mutex);
+
+		fprintf(stderr,
+			"InnoDB: Waiting for rollback of trx id %lu to end\n",
+			(ulong) ut_dulint_get_low(trx->id));
+		os_thread_sleep(100000);
+
+		mutex_enter(&kernel_mutex);
+	}
+
+	mutex_exit(&kernel_mutex);
+
+	if (trx_get_dict_operation(trx) != TRX_DICT_OP_NONE
+	    && !ut_dulint_is_zero(trx->table_id)) {
+
+		/* If the transaction was for a dictionary operation, we
+		drop the relevant table, if it still exists */
+
+		fprintf(stderr,
+			"InnoDB: Dropping table with id %lu %lu"
+			" in recovery if it exists\n",
+			(ulong) ut_dulint_get_high(trx->table_id),
+			(ulong) ut_dulint_get_low(trx->table_id));
+
+		table = dict_table_get_on_id_low(trx->table_id);
+
+		if (table) {
+			ulint	err;
+
+			fputs("InnoDB: Table found: dropping table ", stderr);
+			ut_print_name(stderr, trx, TRUE, table->name);
+			fputs(" in recovery\n", stderr);
+
+			err = row_drop_table_for_mysql(table->name, trx, TRUE);
+			trx_commit_for_mysql(trx);
+
+			ut_a(err == (int) DB_SUCCESS);
+		}
+	}
+
+	if (dictionary_locked) {
+		row_mysql_unlock_data_dictionary(trx);
+	}
+
+	fprintf(stderr, "\nInnoDB: Rolling back of trx id " TRX_ID_FMT
+		" completed\n",
+		TRX_ID_PREP_PRINTF(trx->id));
+	mem_heap_free(heap);
+
+	trx_roll_crash_recv_trx	= NULL;
+}
+
+/***********************************************************************
+Rollback or clean up any incomplete transactions which were
+encountered in crash recovery.  If the transaction already was
+committed, then we clean up a possible insert undo log. If the
+transaction was not yet committed, then we roll it back.
+Note: this is done in a background thread. */
+UNIV_INTERN
+os_thread_ret_t
+trx_rollback_or_clean_all_recovered(
+/*================================*/
+			/* out: a dummy parameter */
+	void*	arg __attribute__((unused)))
+			/* in: a dummy parameter required by
+			os_thread_create */
+{
+	trx_t*	trx;
+
+	mutex_enter(&kernel_mutex);
+
+	if (UT_LIST_GET_FIRST(trx_sys->trx_list)) {
+
+		fprintf(stderr,
+			"InnoDB: Starting in background the rollback"
+			" of uncommitted transactions\n");
+	} else {
+		goto leave_function;
+	}
+
+	mutex_exit(&kernel_mutex);
+
+loop:
+	mutex_enter(&kernel_mutex);
+
+	for (trx = UT_LIST_GET_FIRST(trx_sys->trx_list); trx;
+	     trx = UT_LIST_GET_NEXT(trx_list, trx)) {
+		if (!trx->is_recovered) {
+			continue;
+		}
+
+		switch (trx->conc_state) {
+		case TRX_NOT_STARTED:
+		case TRX_PREPARED:
+			continue;
+
+		case TRX_COMMITTED_IN_MEMORY:
+			mutex_exit(&kernel_mutex);
+			fprintf(stderr,
+				"InnoDB: Cleaning up trx with id "
+				TRX_ID_FMT "\n",
+				TRX_ID_PREP_PRINTF(trx->id));
+			trx_cleanup_at_db_startup(trx);
+			goto loop;
+
+		case TRX_ACTIVE:
+			mutex_exit(&kernel_mutex);
+			trx_rollback_active(trx);
+			goto loop;
+		}
+	}
+
+	ut_print_timestamp(stderr);
+	fprintf(stderr,
+		"  InnoDB: Rollback of non-prepared transactions completed\n");
+
+leave_function:
+	mutex_exit(&kernel_mutex);
+
+	/* We count the number of threads in os_thread_exit(). A created
+	thread should always use that to exit and not use return() to exit. */
+
+	os_thread_exit(NULL);
+
+	OS_THREAD_DUMMY_RETURN;
+}
+
+/***********************************************************************
+Creates an undo number array. */
+UNIV_INTERN
+trx_undo_arr_t*
+trx_undo_arr_create(void)
+/*=====================*/
+{
+	trx_undo_arr_t*	arr;
+	mem_heap_t*	heap;
+	ulint		i;
+
+	heap = mem_heap_create(1024);
+
+	arr = mem_heap_alloc(heap, sizeof(trx_undo_arr_t));
+
+	arr->infos = mem_heap_alloc(heap, sizeof(trx_undo_inf_t)
+				    * UNIV_MAX_PARALLELISM);
+	arr->n_cells = UNIV_MAX_PARALLELISM;
+	arr->n_used = 0;
+
+	arr->heap = heap;
+
+	for (i = 0; i < UNIV_MAX_PARALLELISM; i++) {
+
+		(trx_undo_arr_get_nth_info(arr, i))->in_use = FALSE;
+	}
+
+	return(arr);
+}
+
+/***********************************************************************
+Frees an undo number array. */
+UNIV_INTERN
+void
+trx_undo_arr_free(
+/*==============*/
+	trx_undo_arr_t*	arr)	/* in: undo number array */
+{
+	ut_ad(arr->n_used == 0);
+
+	mem_heap_free(arr->heap);
+}
+
+/***********************************************************************
+Stores info of an undo log record to the array if it is not stored yet. */
+static
+ibool
+trx_undo_arr_store_info(
+/*====================*/
+			/* out: FALSE if the record already existed in the
+			array */
+	trx_t*	trx,	/* in: transaction */
+	dulint	undo_no)/* in: undo number */
+{
+	trx_undo_inf_t*	cell;
+	trx_undo_inf_t*	stored_here;
+	trx_undo_arr_t*	arr;
+	ulint		n_used;
+	ulint		n;
+	ulint		i;
+
+	n = 0;
+	arr = trx->undo_no_arr;
+	n_used = arr->n_used;
+	stored_here = NULL;
+
+	for (i = 0;; i++) {
+		cell = trx_undo_arr_get_nth_info(arr, i);
+
+		if (!cell->in_use) {
+			if (!stored_here) {
+				/* Not in use, we may store here */
+				cell->undo_no = undo_no;
+				cell->in_use = TRUE;
+
+				arr->n_used++;
+
+				stored_here = cell;
+			}
+		} else {
+			n++;
+
+			if (0 == ut_dulint_cmp(cell->undo_no, undo_no)) {
+
+				if (stored_here) {
+					stored_here->in_use = FALSE;
+					ut_ad(arr->n_used > 0);
+					arr->n_used--;
+				}
+
+				ut_ad(arr->n_used == n_used);
+
+				return(FALSE);
+			}
+		}
+
+		if (n == n_used && stored_here) {
+
+			ut_ad(arr->n_used == 1 + n_used);
+
+			return(TRUE);
+		}
+	}
+}
+
+/***********************************************************************
+Removes an undo number from the array. */
+static
+void
+trx_undo_arr_remove_info(
+/*=====================*/
+	trx_undo_arr_t*	arr,	/* in: undo number array */
+	dulint		undo_no)/* in: undo number */
+{
+	trx_undo_inf_t*	cell;
+	ulint		n_used;
+	ulint		n;
+	ulint		i;
+
+	n_used = arr->n_used;
+	n = 0;
+
+	for (i = 0;; i++) {
+		cell = trx_undo_arr_get_nth_info(arr, i);
+
+		if (cell->in_use
+		    && 0 == ut_dulint_cmp(cell->undo_no, undo_no)) {
+
+			cell->in_use = FALSE;
+
+			ut_ad(arr->n_used > 0);
+
+			arr->n_used--;
+
+			return;
+		}
+	}
+}
+
+/***********************************************************************
+Gets the biggest undo number in an array. */
+static
+dulint
+trx_undo_arr_get_biggest(
+/*=====================*/
+				/* out: biggest value, ut_dulint_zero if
+				the array is empty */
+	trx_undo_arr_t*	arr)	/* in: undo number array */
+{
+	trx_undo_inf_t*	cell;
+	ulint		n_used;
+	dulint		biggest;
+	ulint		n;
+	ulint		i;
+
+	n = 0;
+	n_used = arr->n_used;
+	biggest = ut_dulint_zero;
+
+	for (i = 0;; i++) {
+		cell = trx_undo_arr_get_nth_info(arr, i);
+
+		if (cell->in_use) {
+			n++;
+			if (ut_dulint_cmp(cell->undo_no, biggest) > 0) {
+
+				biggest = cell->undo_no;
+			}
+		}
+
+		if (n == n_used) {
+			return(biggest);
+		}
+	}
+}
+
+/***************************************************************************
+Tries truncate the undo logs. */
+UNIV_INTERN
+void
+trx_roll_try_truncate(
+/*==================*/
+	trx_t*	trx)	/* in: transaction */
+{
+	trx_undo_arr_t*	arr;
+	dulint		limit;
+	dulint		biggest;
+
+	ut_ad(mutex_own(&(trx->undo_mutex)));
+	ut_ad(mutex_own(&((trx->rseg)->mutex)));
+
+	trx->pages_undone = 0;
+
+	arr = trx->undo_no_arr;
+
+	limit = trx->undo_no;
+
+	if (arr->n_used > 0) {
+		biggest = trx_undo_arr_get_biggest(arr);
+
+		if (ut_dulint_cmp(biggest, limit) >= 0) {
+
+			limit = ut_dulint_add(biggest, 1);
+		}
+	}
+
+	if (trx->insert_undo) {
+		trx_undo_truncate_end(trx, trx->insert_undo, limit);
+	}
+
+	if (trx->update_undo) {
+		trx_undo_truncate_end(trx, trx->update_undo, limit);
+	}
+}
+
+/***************************************************************************
+Pops the topmost undo log record in a single undo log and updates the info
+about the topmost record in the undo log memory struct. */
+static
+trx_undo_rec_t*
+trx_roll_pop_top_rec(
+/*=================*/
+				/* out: undo log record, the page s-latched */
+	trx_t*		trx,	/* in: transaction */
+	trx_undo_t*	undo,	/* in: undo log */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	page_t*		undo_page;
+	ulint		offset;
+	trx_undo_rec_t*	prev_rec;
+	page_t*		prev_rec_page;
+
+	ut_ad(mutex_own(&(trx->undo_mutex)));
+
+	undo_page = trx_undo_page_get_s_latched(undo->space, undo->zip_size,
+						undo->top_page_no, mtr);
+	offset = undo->top_offset;
+
+	/*	fprintf(stderr, "Thread %lu undoing trx %lu undo record %lu\n",
+	os_thread_get_curr_id(), ut_dulint_get_low(trx->id),
+	ut_dulint_get_low(undo->top_undo_no)); */
+
+	prev_rec = trx_undo_get_prev_rec(undo_page + offset,
+					 undo->hdr_page_no, undo->hdr_offset,
+					 mtr);
+	if (prev_rec == NULL) {
+
+		undo->empty = TRUE;
+	} else {
+		prev_rec_page = page_align(prev_rec);
+
+		if (prev_rec_page != undo_page) {
+
+			trx->pages_undone++;
+		}
+
+		undo->top_page_no = page_get_page_no(prev_rec_page);
+		undo->top_offset  = prev_rec - prev_rec_page;
+		undo->top_undo_no = trx_undo_rec_get_undo_no(prev_rec);
+	}
+
+	return(undo_page + offset);
+}
+
+/************************************************************************
+Pops the topmost record when the two undo logs of a transaction are seen
+as a single stack of records ordered by their undo numbers. Inserts the
+undo number of the popped undo record to the array of currently processed
+undo numbers in the transaction. When the query thread finishes processing
+of this undo record, it must be released with trx_undo_rec_release. */
+UNIV_INTERN
+trx_undo_rec_t*
+trx_roll_pop_top_rec_of_trx(
+/*========================*/
+				/* out: undo log record copied to heap, NULL
+				if none left, or if the undo number of the
+				top record would be less than the limit */
+	trx_t*		trx,	/* in: transaction */
+	dulint		limit,	/* in: least undo number we need */
+	dulint*		roll_ptr,/* out: roll pointer to undo record */
+	mem_heap_t*	heap)	/* in: memory heap where copied */
+{
+	trx_undo_t*	undo;
+	trx_undo_t*	ins_undo;
+	trx_undo_t*	upd_undo;
+	trx_undo_rec_t*	undo_rec;
+	trx_undo_rec_t*	undo_rec_copy;
+	dulint		undo_no;
+	ibool		is_insert;
+	trx_rseg_t*	rseg;
+	ulint		progress_pct;
+	mtr_t		mtr;
+
+	rseg = trx->rseg;
+try_again:
+	mutex_enter(&(trx->undo_mutex));
+
+	if (trx->pages_undone >= TRX_ROLL_TRUNC_THRESHOLD) {
+		mutex_enter(&(rseg->mutex));
+
+		trx_roll_try_truncate(trx);
+
+		mutex_exit(&(rseg->mutex));
+	}
+
+	ins_undo = trx->insert_undo;
+	upd_undo = trx->update_undo;
+
+	if (!ins_undo || ins_undo->empty) {
+		undo = upd_undo;
+	} else if (!upd_undo || upd_undo->empty) {
+		undo = ins_undo;
+	} else if (ut_dulint_cmp(upd_undo->top_undo_no,
+				 ins_undo->top_undo_no) > 0) {
+		undo = upd_undo;
+	} else {
+		undo = ins_undo;
+	}
+
+	if (!undo || undo->empty
+	    || (ut_dulint_cmp(limit, undo->top_undo_no) > 0)) {
+
+		if ((trx->undo_no_arr)->n_used == 0) {
+			/* Rollback is ending */
+
+			mutex_enter(&(rseg->mutex));
+
+			trx_roll_try_truncate(trx);
+
+			mutex_exit(&(rseg->mutex));
+		}
+
+		mutex_exit(&(trx->undo_mutex));
+
+		return(NULL);
+	}
+
+	if (undo == ins_undo) {
+		is_insert = TRUE;
+	} else {
+		is_insert = FALSE;
+	}
+
+	*roll_ptr = trx_undo_build_roll_ptr(is_insert, (undo->rseg)->id,
+					    undo->top_page_no,
+					    undo->top_offset);
+	mtr_start(&mtr);
+
+	undo_rec = trx_roll_pop_top_rec(trx, undo, &mtr);
+
+	undo_no = trx_undo_rec_get_undo_no(undo_rec);
+
+	ut_ad(ut_dulint_cmp(ut_dulint_add(undo_no, 1), trx->undo_no) == 0);
+
+	/* We print rollback progress info if we are in a crash recovery
+	and the transaction has at least 1000 row operations to undo. */
+
+	if (trx == trx_roll_crash_recv_trx && trx_roll_max_undo_no > 1000) {
+
+		progress_pct = 100 - (ulint)
+			((ut_conv_dulint_to_longlong(undo_no) * 100)
+			 / trx_roll_max_undo_no);
+		if (progress_pct != trx_roll_progress_printed_pct) {
+			if (trx_roll_progress_printed_pct == 0) {
+				fprintf(stderr,
+					"\nInnoDB: Progress in percents:"
+					" %lu", (ulong) progress_pct);
+			} else {
+				fprintf(stderr,
+					" %lu", (ulong) progress_pct);
+			}
+			fflush(stderr);
+			trx_roll_progress_printed_pct = progress_pct;
+		}
+	}
+
+	trx->undo_no = undo_no;
+
+	if (!trx_undo_arr_store_info(trx, undo_no)) {
+		/* A query thread is already processing this undo log record */
+
+		mutex_exit(&(trx->undo_mutex));
+
+		mtr_commit(&mtr);
+
+		goto try_again;
+	}
+
+	undo_rec_copy = trx_undo_rec_copy(undo_rec, heap);
+
+	mutex_exit(&(trx->undo_mutex));
+
+	mtr_commit(&mtr);
+
+	return(undo_rec_copy);
+}
+
+/************************************************************************
+Reserves an undo log record for a query thread to undo. This should be
+called if the query thread gets the undo log record not using the pop
+function above. */
+UNIV_INTERN
+ibool
+trx_undo_rec_reserve(
+/*=================*/
+			/* out: TRUE if succeeded */
+	trx_t*	trx,	/* in: transaction */
+	dulint	undo_no)/* in: undo number of the record */
+{
+	ibool	ret;
+
+	mutex_enter(&(trx->undo_mutex));
+
+	ret = trx_undo_arr_store_info(trx, undo_no);
+
+	mutex_exit(&(trx->undo_mutex));
+
+	return(ret);
+}
+
+/***********************************************************************
+Releases a reserved undo record. */
+UNIV_INTERN
+void
+trx_undo_rec_release(
+/*=================*/
+	trx_t*	trx,	/* in: transaction */
+	dulint	undo_no)/* in: undo number */
+{
+	trx_undo_arr_t*	arr;
+
+	mutex_enter(&(trx->undo_mutex));
+
+	arr = trx->undo_no_arr;
+
+	trx_undo_arr_remove_info(arr, undo_no);
+
+	mutex_exit(&(trx->undo_mutex));
+}
+
+/*************************************************************************
+Starts a rollback operation. */
+UNIV_INTERN
+void
+trx_rollback(
+/*=========*/
+	trx_t*		trx,	/* in: transaction */
+	trx_sig_t*	sig,	/* in: signal starting the rollback */
+	que_thr_t**	next_thr)/* in/out: next query thread to run;
+				if the value which is passed in is
+				a pointer to a NULL pointer, then the
+				calling function can start running
+				a new query thread; if the passed value is
+				NULL, the parameter is ignored */
+{
+	que_t*		roll_graph;
+	que_thr_t*	thr;
+	/*	que_thr_t*	thr2; */
+
+	ut_ad(mutex_own(&kernel_mutex));
+	ut_ad((trx->undo_no_arr == NULL) || ((trx->undo_no_arr)->n_used == 0));
+
+	/* Initialize the rollback field in the transaction */
+
+	if (sig->type == TRX_SIG_TOTAL_ROLLBACK) {
+
+		trx->roll_limit = ut_dulint_zero;
+
+	} else if (sig->type == TRX_SIG_ROLLBACK_TO_SAVEPT) {
+
+		trx->roll_limit = (sig->savept).least_undo_no;
+
+	} else if (sig->type == TRX_SIG_ERROR_OCCURRED) {
+
+		trx->roll_limit = trx->last_sql_stat_start.least_undo_no;
+	} else {
+		ut_error;
+	}
+
+	ut_a(ut_dulint_cmp(trx->roll_limit, trx->undo_no) <= 0);
+
+	trx->pages_undone = 0;
+
+	if (trx->undo_no_arr == NULL) {
+		trx->undo_no_arr = trx_undo_arr_create();
+	}
+
+	/* Build a 'query' graph which will perform the undo operations */
+
+	roll_graph = trx_roll_graph_build(trx);
+
+	trx->graph = roll_graph;
+	trx->que_state = TRX_QUE_ROLLING_BACK;
+
+	thr = que_fork_start_command(roll_graph);
+
+	ut_ad(thr);
+
+	/*	thr2 = que_fork_start_command(roll_graph);
+
+	ut_ad(thr2); */
+
+	if (next_thr && (*next_thr == NULL)) {
+		*next_thr = thr;
+		/*		srv_que_task_enqueue_low(thr2); */
+	} else {
+		srv_que_task_enqueue_low(thr);
+		/*		srv_que_task_enqueue_low(thr2); */
+	}
+}
+
+/********************************************************************
+Builds an undo 'query' graph for a transaction. The actual rollback is
+performed by executing this query graph like a query subprocedure call.
+The reply about the completion of the rollback will be sent by this
+graph. */
+UNIV_INTERN
+que_t*
+trx_roll_graph_build(
+/*=================*/
+			/* out, own: the query graph */
+	trx_t*	trx)	/* in: trx handle */
+{
+	mem_heap_t*	heap;
+	que_fork_t*	fork;
+	que_thr_t*	thr;
+	/*	que_thr_t*	thr2; */
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	heap = mem_heap_create(512);
+	fork = que_fork_create(NULL, NULL, QUE_FORK_ROLLBACK, heap);
+	fork->trx = trx;
+
+	thr = que_thr_create(fork, heap);
+	/*	thr2 = que_thr_create(fork, heap); */
+
+	thr->child = row_undo_node_create(trx, thr, heap);
+	/*	thr2->child = row_undo_node_create(trx, thr2, heap); */
+
+	return(fork);
+}
+
+/*************************************************************************
+Finishes error processing after the necessary partial rollback has been
+done. */
+static
+void
+trx_finish_error_processing(
+/*========================*/
+	trx_t*	trx)	/* in: transaction */
+{
+	trx_sig_t*	sig;
+	trx_sig_t*	next_sig;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	sig = UT_LIST_GET_FIRST(trx->signals);
+
+	while (sig != NULL) {
+		next_sig = UT_LIST_GET_NEXT(signals, sig);
+
+		if (sig->type == TRX_SIG_ERROR_OCCURRED) {
+
+			trx_sig_remove(trx, sig);
+		}
+
+		sig = next_sig;
+	}
+
+	trx->que_state = TRX_QUE_RUNNING;
+}
+
+/*************************************************************************
+Finishes a partial rollback operation. */
+static
+void
+trx_finish_partial_rollback_off_kernel(
+/*===================================*/
+	trx_t*		trx,	/* in: transaction */
+	que_thr_t**	next_thr)/* in/out: next query thread to run;
+				if the value which is passed in is a pointer
+				to a NULL pointer, then the calling function
+				can start running a new query thread; if this
+				parameter is NULL, it is ignored */
+{
+	trx_sig_t*	sig;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	sig = UT_LIST_GET_FIRST(trx->signals);
+
+	/* Remove the signal from the signal queue and send reply message
+	to it */
+
+	trx_sig_reply(sig, next_thr);
+	trx_sig_remove(trx, sig);
+
+	trx->que_state = TRX_QUE_RUNNING;
+}
+
+/********************************************************************
+Finishes a transaction rollback. */
+UNIV_INTERN
+void
+trx_finish_rollback_off_kernel(
+/*===========================*/
+	que_t*		graph,	/* in: undo graph which can now be freed */
+	trx_t*		trx,	/* in: transaction */
+	que_thr_t**	next_thr)/* in/out: next query thread to run;
+				if the value which is passed in is
+				a pointer to a NULL pointer, then the
+				calling function can start running
+				a new query thread; if this parameter is
+				NULL, it is ignored */
+{
+	trx_sig_t*	sig;
+	trx_sig_t*	next_sig;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	ut_a(trx->undo_no_arr == NULL || trx->undo_no_arr->n_used == 0);
+
+	/* Free the memory reserved by the undo graph */
+	que_graph_free(graph);
+
+	sig = UT_LIST_GET_FIRST(trx->signals);
+
+	if (sig->type == TRX_SIG_ROLLBACK_TO_SAVEPT) {
+
+		trx_finish_partial_rollback_off_kernel(trx, next_thr);
+
+		return;
+
+	} else if (sig->type == TRX_SIG_ERROR_OCCURRED) {
+
+		trx_finish_error_processing(trx);
+
+		return;
+	}
+
+#ifdef UNIV_DEBUG
+	if (lock_print_waits) {
+		fprintf(stderr, "Trx %lu rollback finished\n",
+			(ulong) ut_dulint_get_low(trx->id));
+	}
+#endif /* UNIV_DEBUG */
+
+	trx_commit_off_kernel(trx);
+
+	/* Remove all TRX_SIG_TOTAL_ROLLBACK signals from the signal queue and
+	send reply messages to them */
+
+	trx->que_state = TRX_QUE_RUNNING;
+
+	while (sig != NULL) {
+		next_sig = UT_LIST_GET_NEXT(signals, sig);
+
+		if (sig->type == TRX_SIG_TOTAL_ROLLBACK) {
+
+			trx_sig_reply(sig, next_thr);
+
+			trx_sig_remove(trx, sig);
+		}
+
+		sig = next_sig;
+	}
+}
+
+/*************************************************************************
+Creates a rollback command node struct. */
+UNIV_INTERN
+roll_node_t*
+roll_node_create(
+/*=============*/
+				/* out, own: rollback node struct */
+	mem_heap_t*	heap)	/* in: mem heap where created */
+{
+	roll_node_t*	node;
+
+	node = mem_heap_alloc(heap, sizeof(roll_node_t));
+	node->common.type = QUE_NODE_ROLLBACK;
+	node->state = ROLL_NODE_SEND;
+
+	node->partial = FALSE;
+
+	return(node);
+}
+
+/***************************************************************
+Performs an execution step for a rollback command node in a query graph. */
+UNIV_INTERN
+que_thr_t*
+trx_rollback_step(
+/*==============*/
+				/* out: query thread to run next, or NULL */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	roll_node_t*	node;
+	ulint		sig_no;
+	trx_savept_t*	savept;
+
+	node = thr->run_node;
+
+	ut_ad(que_node_get_type(node) == QUE_NODE_ROLLBACK);
+
+	if (thr->prev_node == que_node_get_parent(node)) {
+		node->state = ROLL_NODE_SEND;
+	}
+
+	if (node->state == ROLL_NODE_SEND) {
+		mutex_enter(&kernel_mutex);
+
+		node->state = ROLL_NODE_WAIT;
+
+		if (node->partial) {
+			sig_no = TRX_SIG_ROLLBACK_TO_SAVEPT;
+			savept = &(node->savept);
+		} else {
+			sig_no = TRX_SIG_TOTAL_ROLLBACK;
+			savept = NULL;
+		}
+
+		/* Send a rollback signal to the transaction */
+
+		trx_sig_send(thr_get_trx(thr), sig_no, TRX_SIG_SELF, thr,
+			     savept, NULL);
+
+		thr->state = QUE_THR_SIG_REPLY_WAIT;
+
+		mutex_exit(&kernel_mutex);
+
+		return(NULL);
+	}
+
+	ut_ad(node->state == ROLL_NODE_WAIT);
+
+	thr->run_node = que_node_get_parent(node);
+
+	return(thr);
+}
diff --git a/storage/xtradb/trx/trx0rseg.c b/storage/xtradb/trx/trx0rseg.c
new file mode 100644
index 00000000000..db5efd65eb3
--- /dev/null
+++ b/storage/xtradb/trx/trx0rseg.c
@@ -0,0 +1,282 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Rollback segment
+
+Created 3/26/1996 Heikki Tuuri
+*******************************************************/
+
+#include "trx0rseg.h"
+
+#ifdef UNIV_NONINL
+#include "trx0rseg.ic"
+#endif
+
+#include "trx0undo.h"
+#include "fut0lst.h"
+#include "srv0srv.h"
+#include "trx0purge.h"
+
+/**********************************************************************
+Looks for a rollback segment, based on the rollback segment id. */
+UNIV_INTERN
+trx_rseg_t*
+trx_rseg_get_on_id(
+/*===============*/
+			/* out: rollback segment */
+	ulint	id)	/* in: rollback segment id */
+{
+	trx_rseg_t*	rseg;
+
+	rseg = UT_LIST_GET_FIRST(trx_sys->rseg_list);
+	ut_ad(rseg);
+
+	while (rseg->id != id) {
+		rseg = UT_LIST_GET_NEXT(rseg_list, rseg);
+		ut_ad(rseg);
+	}
+
+	return(rseg);
+}
+
+/********************************************************************
+Creates a rollback segment header. This function is called only when
+a new rollback segment is created in the database. */
+UNIV_INTERN
+ulint
+trx_rseg_header_create(
+/*===================*/
+				/* out: page number of the created segment,
+				FIL_NULL if fail */
+	ulint	space,		/* in: space id */
+	ulint	zip_size,	/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	ulint	max_size,	/* in: max size in pages */
+	ulint*	slot_no,	/* out: rseg id == slot number in trx sys */
+	mtr_t*	mtr)		/* in: mtr */
+{
+	ulint		page_no;
+	trx_rsegf_t*	rsegf;
+	trx_sysf_t*	sys_header;
+	ulint		i;
+	buf_block_t*	block;
+
+	ut_ad(mtr);
+	ut_ad(mutex_own(&kernel_mutex));
+	ut_ad(mtr_memo_contains(mtr, fil_space_get_latch(space, NULL),
+				MTR_MEMO_X_LOCK));
+	sys_header = trx_sysf_get(mtr);
+
+	*slot_no = trx_sysf_rseg_find_free(mtr);
+
+	if (*slot_no == ULINT_UNDEFINED) {
+
+		return(FIL_NULL);
+	}
+
+	/* Allocate a new file segment for the rollback segment */
+	block = fseg_create(space, 0,
+			    TRX_RSEG + TRX_RSEG_FSEG_HEADER, mtr);
+
+	if (block == NULL) {
+		/* No space left */
+
+		return(FIL_NULL);
+	}
+
+	buf_block_dbg_add_level(block, SYNC_RSEG_HEADER_NEW);
+
+	page_no = buf_block_get_page_no(block);
+
+	/* Get the rollback segment file page */
+	rsegf = trx_rsegf_get_new(space, zip_size, page_no, mtr);
+
+	/* Initialize max size field */
+	mlog_write_ulint(rsegf + TRX_RSEG_MAX_SIZE, max_size,
+			 MLOG_4BYTES, mtr);
+
+	/* Initialize the history list */
+
+	mlog_write_ulint(rsegf + TRX_RSEG_HISTORY_SIZE, 0, MLOG_4BYTES, mtr);
+	flst_init(rsegf + TRX_RSEG_HISTORY, mtr);
+
+	/* Reset the undo log slots */
+	for (i = 0; i < TRX_RSEG_N_SLOTS; i++) {
+
+		trx_rsegf_set_nth_undo(rsegf, i, FIL_NULL, mtr);
+	}
+
+	/* Add the rollback segment info to the free slot in the trx system
+	header */
+
+	trx_sysf_rseg_set_space(sys_header, *slot_no, space, mtr);
+	trx_sysf_rseg_set_page_no(sys_header, *slot_no, page_no, mtr);
+
+	return(page_no);
+}
+
+/***************************************************************************
+Creates and initializes a rollback segment object. The values for the
+fields are read from the header. The object is inserted to the rseg
+list of the trx system object and a pointer is inserted in the rseg
+array in the trx system object. */
+static
+trx_rseg_t*
+trx_rseg_mem_create(
+/*================*/
+				/* out, own: rollback segment object */
+	ulint	id,		/* in: rollback segment id */
+	ulint	space,		/* in: space where the segment placed */
+	ulint	zip_size,	/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	ulint	page_no,	/* in: page number of the segment header */
+	mtr_t*	mtr)		/* in: mtr */
+{
+	trx_rsegf_t*	rseg_header;
+	trx_rseg_t*	rseg;
+	trx_ulogf_t*	undo_log_hdr;
+	fil_addr_t	node_addr;
+	ulint		sum_of_undo_sizes;
+	ulint		len;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	rseg = mem_alloc(sizeof(trx_rseg_t));
+
+	rseg->id = id;
+	rseg->space = space;
+	rseg->zip_size = zip_size;
+	rseg->page_no = page_no;
+
+	mutex_create(&rseg->mutex, SYNC_RSEG);
+
+	UT_LIST_ADD_LAST(rseg_list, trx_sys->rseg_list, rseg);
+
+	trx_sys_set_nth_rseg(trx_sys, id, rseg);
+
+	rseg_header = trx_rsegf_get_new(space, zip_size, page_no, mtr);
+
+	rseg->max_size = mtr_read_ulint(rseg_header + TRX_RSEG_MAX_SIZE,
+					MLOG_4BYTES, mtr);
+
+	/* Initialize the undo log lists according to the rseg header */
+
+	sum_of_undo_sizes = trx_undo_lists_init(rseg);
+
+	rseg->curr_size = mtr_read_ulint(rseg_header + TRX_RSEG_HISTORY_SIZE,
+					 MLOG_4BYTES, mtr)
+		+ 1 + sum_of_undo_sizes;
+
+	len = flst_get_len(rseg_header + TRX_RSEG_HISTORY, mtr);
+	if (len > 0) {
+		trx_sys->rseg_history_len += len;
+
+		node_addr = trx_purge_get_log_from_hist(
+			flst_get_last(rseg_header + TRX_RSEG_HISTORY, mtr));
+		rseg->last_page_no = node_addr.page;
+		rseg->last_offset = node_addr.boffset;
+
+		undo_log_hdr = trx_undo_page_get(rseg->space, rseg->zip_size,
+						 node_addr.page,
+						 mtr) + node_addr.boffset;
+
+		rseg->last_trx_no = mtr_read_dulint(
+			undo_log_hdr + TRX_UNDO_TRX_NO, mtr);
+		rseg->last_del_marks = mtr_read_ulint(
+			undo_log_hdr + TRX_UNDO_DEL_MARKS, MLOG_2BYTES, mtr);
+	} else {
+		rseg->last_page_no = FIL_NULL;
+	}
+
+	return(rseg);
+}
+
+/*************************************************************************
+Creates the memory copies for rollback segments and initializes the
+rseg list and array in trx_sys at a database startup. */
+UNIV_INTERN
+void
+trx_rseg_list_and_array_init(
+/*=========================*/
+	trx_sysf_t*	sys_header,	/* in: trx system header */
+	mtr_t*		mtr)		/* in: mtr */
+{
+	ulint	i;
+	ulint	page_no;
+	ulint	space;
+
+	UT_LIST_INIT(trx_sys->rseg_list);
+
+	trx_sys->rseg_history_len = 0;
+
+	for (i = 0; i < TRX_SYS_N_RSEGS; i++) {
+
+		page_no = trx_sysf_rseg_get_page_no(sys_header, i, mtr);
+
+		if (page_no == FIL_NULL) {
+
+			trx_sys_set_nth_rseg(trx_sys, i, NULL);
+		} else {
+			ulint	zip_size;
+
+			space = trx_sysf_rseg_get_space(sys_header, i, mtr);
+
+			zip_size = space ? fil_space_get_zip_size(space) : 0;
+
+			trx_rseg_mem_create(i, space, zip_size, page_no, mtr);
+		}
+	}
+}
+
+/********************************************************************
+Creates a new rollback segment to the database. */
+UNIV_INTERN
+trx_rseg_t*
+trx_rseg_create(
+/*============*/
+				/* out: the created segment object, NULL if
+				fail */
+	ulint	space,		/* in: space id */
+	ulint	max_size,	/* in: max size in pages */
+	ulint*	id,		/* out: rseg id */
+	mtr_t*	mtr)		/* in: mtr */
+{
+	ulint		flags;
+	ulint		zip_size;
+	ulint		page_no;
+	trx_rseg_t*	rseg;
+
+	mtr_x_lock(fil_space_get_latch(space, &flags), mtr);
+	zip_size = dict_table_flags_to_zip_size(flags);
+	mutex_enter(&kernel_mutex);
+
+	page_no = trx_rseg_header_create(space, zip_size, max_size, id, mtr);
+
+	if (page_no == FIL_NULL) {
+
+		mutex_exit(&kernel_mutex);
+		return(NULL);
+	}
+
+	rseg = trx_rseg_mem_create(*id, space, zip_size, page_no, mtr);
+
+	mutex_exit(&kernel_mutex);
+
+	return(rseg);
+}
diff --git a/storage/xtradb/trx/trx0sys.c b/storage/xtradb/trx/trx0sys.c
new file mode 100644
index 00000000000..b80a50738c0
--- /dev/null
+++ b/storage/xtradb/trx/trx0sys.c
@@ -0,0 +1,1391 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Transaction system
+
+Created 3/26/1996 Heikki Tuuri
+*******************************************************/
+
+#include "trx0sys.h"
+
+#ifdef UNIV_NONINL
+#include "trx0sys.ic"
+#endif
+
+#include "fsp0fsp.h"
+#include "mtr0mtr.h"
+#include "trx0trx.h"
+#include "trx0rseg.h"
+#include "trx0undo.h"
+#include "srv0srv.h"
+#include "trx0purge.h"
+#include "log0log.h"
+#include "os0file.h"
+
+/* The file format tag structure with id and name. */
+struct file_format_struct {
+	ulint		id;		/* id of the file format */
+	const char*	name;		/* text representation of the
+					file format */
+	mutex_t		mutex;		/* covers changes to the above
+					fields */
+};
+
+typedef struct file_format_struct	file_format_t;
+
+/* The transaction system */
+UNIV_INTERN trx_sys_t*		trx_sys		= NULL;
+UNIV_INTERN trx_doublewrite_t*	trx_doublewrite = NULL;
+
+/* The following is set to TRUE when we are upgrading from the old format data
+files to the new >= 4.1.x format multiple tablespaces format data files */
+
+UNIV_INTERN ibool	trx_doublewrite_must_reset_space_ids	= FALSE;
+
+/* The following is TRUE when we are using the database in the new format,
+i.e., we have successfully upgraded, or have created a new database
+installation */
+
+UNIV_INTERN ibool	trx_sys_multiple_tablespace_format	= FALSE;
+
+/* In a MySQL replication slave, in crash recovery we store the master log
+file name and position here. We have successfully got the updates to InnoDB
+up to this position. If .._pos is -1, it means no crash recovery was needed,
+or there was no master log position info inside InnoDB. */
+
+UNIV_INTERN char	trx_sys_mysql_master_log_name[TRX_SYS_MYSQL_MASTER_LOG_NAME_LEN];
+UNIV_INTERN ib_int64_t	trx_sys_mysql_master_log_pos	= -1;
+
+UNIV_INTERN char	trx_sys_mysql_relay_log_name[TRX_SYS_MYSQL_MASTER_LOG_NAME_LEN];
+UNIV_INTERN ib_int64_t	trx_sys_mysql_relay_log_pos	= -1;
+
+/* If this MySQL server uses binary logging, after InnoDB has been inited
+and if it has done a crash recovery, we store the binlog file name and position
+here. If .._pos is -1, it means there was no binlog position info inside
+InnoDB. */
+
+UNIV_INTERN char	trx_sys_mysql_bin_log_name[TRX_SYS_MYSQL_LOG_NAME_LEN];
+UNIV_INTERN ib_int64_t	trx_sys_mysql_bin_log_pos	= -1;
+
+/* List of animal names representing file format. */
+static const char*	file_format_name_map[] = {
+	"Antelope",
+	"Barracuda",
+	"Cheetah",
+	"Dragon",
+	"Elk",
+	"Fox",
+	"Gazelle",
+	"Hornet",
+	"Impala",
+	"Jaguar",
+	"Kangaroo",
+	"Leopard",
+	"Moose",
+	"Nautilus",
+	"Ocelot",
+	"Porpoise",
+	"Quail",
+	"Rabbit",
+	"Shark",
+	"Tiger",
+	"Urchin",
+	"Viper",
+	"Whale",
+	"Xenops",
+	"Yak",
+	"Zebra"
+};
+
+/* The number of elements in the file format name array. */
+static const ulint	FILE_FORMAT_NAME_N
+	= sizeof(file_format_name_map) / sizeof(file_format_name_map[0]);
+
+/* This is used to track the maximum file format id known to InnoDB. It's
+updated via SET GLOBAL innodb_file_format_check = 'x' or when we open
+or create a table. */
+static	file_format_t	file_format_max;
+
+/********************************************************************
+Determines if a page number is located inside the doublewrite buffer. */
+UNIV_INTERN
+ibool
+trx_doublewrite_page_inside(
+/*========================*/
+				/* out: TRUE if the location is inside
+				the two blocks of the doublewrite buffer */
+	ulint	page_no)	/* in: page number */
+{
+	if (trx_doublewrite == NULL) {
+
+		return(FALSE);
+	}
+
+	if (page_no >= trx_doublewrite->block1
+	    && page_no < trx_doublewrite->block1
+	    + TRX_SYS_DOUBLEWRITE_BLOCK_SIZE) {
+		return(TRUE);
+	}
+
+	if (page_no >= trx_doublewrite->block2
+	    && page_no < trx_doublewrite->block2
+	    + TRX_SYS_DOUBLEWRITE_BLOCK_SIZE) {
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
+
+/********************************************************************
+Creates or initialializes the doublewrite buffer at a database start. */
+static
+void
+trx_doublewrite_init(
+/*=================*/
+	byte*	doublewrite)	/* in: pointer to the doublewrite buf
+				header on trx sys page */
+{
+	trx_doublewrite = mem_alloc(sizeof(trx_doublewrite_t));
+
+	/* Since we now start to use the doublewrite buffer, no need to call
+	fsync() after every write to a data file */
+#ifdef UNIV_DO_FLUSH
+	os_do_not_call_flush_at_each_write = TRUE;
+#endif /* UNIV_DO_FLUSH */
+
+	mutex_create(&trx_doublewrite->mutex, SYNC_DOUBLEWRITE);
+
+	trx_doublewrite->first_free = 0;
+
+	trx_doublewrite->block1 = mach_read_from_4(
+		doublewrite + TRX_SYS_DOUBLEWRITE_BLOCK1);
+	trx_doublewrite->block2 = mach_read_from_4(
+		doublewrite + TRX_SYS_DOUBLEWRITE_BLOCK2);
+	trx_doublewrite->write_buf_unaligned = ut_malloc(
+		(1 + 2 * TRX_SYS_DOUBLEWRITE_BLOCK_SIZE) * UNIV_PAGE_SIZE);
+
+	trx_doublewrite->write_buf = ut_align(
+		trx_doublewrite->write_buf_unaligned, UNIV_PAGE_SIZE);
+	trx_doublewrite->buf_block_arr = mem_alloc(
+		2 * TRX_SYS_DOUBLEWRITE_BLOCK_SIZE * sizeof(void*));
+}
+
+/********************************************************************
+Marks the trx sys header when we have successfully upgraded to the >= 4.1.x
+multiple tablespace format. */
+UNIV_INTERN
+void
+trx_sys_mark_upgraded_to_multiple_tablespaces(void)
+/*===============================================*/
+{
+	buf_block_t*	block;
+	byte*		doublewrite;
+	mtr_t		mtr;
+
+	/* We upgraded to 4.1.x and reset the space id fields in the
+	doublewrite buffer. Let us mark to the trx_sys header that the upgrade
+	has been done. */
+
+	mtr_start(&mtr);
+
+	block = buf_page_get(TRX_SYS_SPACE, 0, TRX_SYS_PAGE_NO,
+			     RW_X_LATCH, &mtr);
+	buf_block_dbg_add_level(block, SYNC_NO_ORDER_CHECK);
+
+	doublewrite = buf_block_get_frame(block) + TRX_SYS_DOUBLEWRITE;
+
+	mlog_write_ulint(doublewrite + TRX_SYS_DOUBLEWRITE_SPACE_ID_STORED,
+			 TRX_SYS_DOUBLEWRITE_SPACE_ID_STORED_N,
+			 MLOG_4BYTES, &mtr);
+	mtr_commit(&mtr);
+
+	/* Flush the modified pages to disk and make a checkpoint */
+	log_make_checkpoint_at(IB_ULONGLONG_MAX, TRUE);
+
+	trx_sys_multiple_tablespace_format = TRUE;
+}
+
+/********************************************************************
+Creates the doublewrite buffer to a new InnoDB installation. The header of the
+doublewrite buffer is placed on the trx system header page. */
+UNIV_INTERN
+void
+trx_sys_create_doublewrite_buf(void)
+/*================================*/
+{
+	buf_block_t*	block;
+	buf_block_t*	block2;
+	buf_block_t*	new_block;
+	byte*	doublewrite;
+	byte*	fseg_header;
+	ulint	page_no;
+	ulint	prev_page_no;
+	ulint	i;
+	mtr_t	mtr;
+
+	if (trx_doublewrite) {
+		/* Already inited */
+
+		return;
+	}
+
+start_again:
+	mtr_start(&mtr);
+
+	block = buf_page_get(TRX_SYS_SPACE, 0, TRX_SYS_PAGE_NO,
+			     RW_X_LATCH, &mtr);
+	buf_block_dbg_add_level(block, SYNC_NO_ORDER_CHECK);
+
+	doublewrite = buf_block_get_frame(block) + TRX_SYS_DOUBLEWRITE;
+
+	if (mach_read_from_4(doublewrite + TRX_SYS_DOUBLEWRITE_MAGIC)
+	    == TRX_SYS_DOUBLEWRITE_MAGIC_N) {
+		/* The doublewrite buffer has already been created:
+		just read in some numbers */
+
+		trx_doublewrite_init(doublewrite);
+
+		mtr_commit(&mtr);
+	} else {
+		fprintf(stderr,
+			"InnoDB: Doublewrite buffer not found:"
+			" creating new\n");
+
+		if (buf_pool_get_curr_size()
+		    < ((2 * TRX_SYS_DOUBLEWRITE_BLOCK_SIZE
+			+ FSP_EXTENT_SIZE / 2 + 100)
+		       * UNIV_PAGE_SIZE)) {
+			fprintf(stderr,
+				"InnoDB: Cannot create doublewrite buffer:"
+				" you must\n"
+				"InnoDB: increase your buffer pool size.\n"
+				"InnoDB: Cannot continue operation.\n");
+
+			exit(1);
+		}
+
+		block2 = fseg_create(TRX_SYS_SPACE, TRX_SYS_PAGE_NO,
+				     TRX_SYS_DOUBLEWRITE
+				     + TRX_SYS_DOUBLEWRITE_FSEG, &mtr);
+
+		/* fseg_create acquires a second latch on the page,
+		therefore we must declare it: */
+
+		buf_block_dbg_add_level(block2, SYNC_NO_ORDER_CHECK);
+
+		if (block2 == NULL) {
+			fprintf(stderr,
+				"InnoDB: Cannot create doublewrite buffer:"
+				" you must\n"
+				"InnoDB: increase your tablespace size.\n"
+				"InnoDB: Cannot continue operation.\n");
+
+			/* We exit without committing the mtr to prevent
+			its modifications to the database getting to disk */
+
+			exit(1);
+		}
+
+		fseg_header = buf_block_get_frame(block)
+			+ TRX_SYS_DOUBLEWRITE + TRX_SYS_DOUBLEWRITE_FSEG;
+		prev_page_no = 0;
+
+		for (i = 0; i < 2 * TRX_SYS_DOUBLEWRITE_BLOCK_SIZE
+			     + FSP_EXTENT_SIZE / 2; i++) {
+			page_no = fseg_alloc_free_page(fseg_header,
+						       prev_page_no + 1,
+						       FSP_UP, &mtr);
+			if (page_no == FIL_NULL) {
+				fprintf(stderr,
+					"InnoDB: Cannot create doublewrite"
+					" buffer: you must\n"
+					"InnoDB: increase your"
+					" tablespace size.\n"
+					"InnoDB: Cannot continue operation.\n"
+					);
+
+				exit(1);
+			}
+
+			/* We read the allocated pages to the buffer pool;
+			when they are written to disk in a flush, the space
+			id and page number fields are also written to the
+			pages. When we at database startup read pages
+			from the doublewrite buffer, we know that if the
+			space id and page number in them are the same as
+			the page position in the tablespace, then the page
+			has not been written to in doublewrite. */
+
+			new_block = buf_page_get(TRX_SYS_SPACE, 0, page_no,
+						 RW_X_LATCH, &mtr);
+			buf_block_dbg_add_level(new_block,
+						SYNC_NO_ORDER_CHECK);
+
+			/* Make a dummy change to the page to ensure it will
+			be written to disk in a flush */
+
+			mlog_write_ulint(buf_block_get_frame(new_block)
+					 + FIL_PAGE_DATA,
+					 TRX_SYS_DOUBLEWRITE_MAGIC_N,
+					 MLOG_4BYTES, &mtr);
+
+			if (i == FSP_EXTENT_SIZE / 2) {
+				mlog_write_ulint(doublewrite
+						 + TRX_SYS_DOUBLEWRITE_BLOCK1,
+						 page_no, MLOG_4BYTES, &mtr);
+				mlog_write_ulint(doublewrite
+						 + TRX_SYS_DOUBLEWRITE_REPEAT
+						 + TRX_SYS_DOUBLEWRITE_BLOCK1,
+						 page_no, MLOG_4BYTES, &mtr);
+			} else if (i == FSP_EXTENT_SIZE / 2
+				   + TRX_SYS_DOUBLEWRITE_BLOCK_SIZE) {
+				mlog_write_ulint(doublewrite
+						 + TRX_SYS_DOUBLEWRITE_BLOCK2,
+						 page_no, MLOG_4BYTES, &mtr);
+				mlog_write_ulint(doublewrite
+						 + TRX_SYS_DOUBLEWRITE_REPEAT
+						 + TRX_SYS_DOUBLEWRITE_BLOCK2,
+						 page_no, MLOG_4BYTES, &mtr);
+			} else if (i > FSP_EXTENT_SIZE / 2) {
+				ut_a(page_no == prev_page_no + 1);
+			}
+
+			prev_page_no = page_no;
+		}
+
+		mlog_write_ulint(doublewrite + TRX_SYS_DOUBLEWRITE_MAGIC,
+				 TRX_SYS_DOUBLEWRITE_MAGIC_N,
+				 MLOG_4BYTES, &mtr);
+		mlog_write_ulint(doublewrite + TRX_SYS_DOUBLEWRITE_MAGIC
+				 + TRX_SYS_DOUBLEWRITE_REPEAT,
+				 TRX_SYS_DOUBLEWRITE_MAGIC_N,
+				 MLOG_4BYTES, &mtr);
+
+		mlog_write_ulint(doublewrite
+				 + TRX_SYS_DOUBLEWRITE_SPACE_ID_STORED,
+				 TRX_SYS_DOUBLEWRITE_SPACE_ID_STORED_N,
+				 MLOG_4BYTES, &mtr);
+		mtr_commit(&mtr);
+
+		/* Flush the modified pages to disk and make a checkpoint */
+		log_make_checkpoint_at(IB_ULONGLONG_MAX, TRUE);
+
+		fprintf(stderr, "InnoDB: Doublewrite buffer created\n");
+
+		trx_sys_multiple_tablespace_format = TRUE;
+
+		goto start_again;
+	}
+}
+
+/********************************************************************
+At a database startup initializes the doublewrite buffer memory structure if
+we already have a doublewrite buffer created in the data files. If we are
+upgrading to an InnoDB version which supports multiple tablespaces, then this
+function performs the necessary update operations. If we are in a crash
+recovery, this function uses a possible doublewrite buffer to restore
+half-written pages in the data files. */
+UNIV_INTERN
+void
+trx_sys_doublewrite_init_or_restore_pages(
+/*======================================*/
+	ibool	restore_corrupt_pages)
+{
+	byte*	buf;
+	byte*	read_buf;
+	byte*	unaligned_read_buf;
+	ulint	block1;
+	ulint	block2;
+	ulint	source_page_no;
+	byte*	page;
+	byte*	doublewrite;
+	ulint	space_id;
+	ulint	page_no;
+	ulint	i;
+
+	/* We do the file i/o past the buffer pool */
+
+	unaligned_read_buf = ut_malloc(2 * UNIV_PAGE_SIZE);
+	read_buf = ut_align(unaligned_read_buf, UNIV_PAGE_SIZE);
+
+	/* Read the trx sys header to check if we are using the doublewrite
+	buffer */
+
+	fil_io(OS_FILE_READ, TRUE, TRX_SYS_SPACE, 0, TRX_SYS_PAGE_NO, 0,
+	       UNIV_PAGE_SIZE, read_buf, NULL);
+	doublewrite = read_buf + TRX_SYS_DOUBLEWRITE;
+
+	if (mach_read_from_4(doublewrite + TRX_SYS_DOUBLEWRITE_MAGIC)
+	    == TRX_SYS_DOUBLEWRITE_MAGIC_N) {
+		/* The doublewrite buffer has been created */
+
+		trx_doublewrite_init(doublewrite);
+
+		block1 = trx_doublewrite->block1;
+		block2 = trx_doublewrite->block2;
+
+		buf = trx_doublewrite->write_buf;
+	} else {
+		goto leave_func;
+	}
+
+	if (mach_read_from_4(doublewrite + TRX_SYS_DOUBLEWRITE_SPACE_ID_STORED)
+	    != TRX_SYS_DOUBLEWRITE_SPACE_ID_STORED_N) {
+
+		/* We are upgrading from a version < 4.1.x to a version where
+		multiple tablespaces are supported. We must reset the space id
+		field in the pages in the doublewrite buffer because starting
+		from this version the space id is stored to
+		FIL_PAGE_ARCH_LOG_NO_OR_SPACE_ID. */
+
+		trx_doublewrite_must_reset_space_ids = TRUE;
+
+		fprintf(stderr,
+			"InnoDB: Resetting space id's in the"
+			" doublewrite buffer\n");
+	} else {
+		trx_sys_multiple_tablespace_format = TRUE;
+	}
+
+	/* Read the pages from the doublewrite buffer to memory */
+
+	fil_io(OS_FILE_READ, TRUE, TRX_SYS_SPACE, 0, block1, 0,
+	       TRX_SYS_DOUBLEWRITE_BLOCK_SIZE * UNIV_PAGE_SIZE,
+	       buf, NULL);
+	fil_io(OS_FILE_READ, TRUE, TRX_SYS_SPACE, 0, block2, 0,
+	       TRX_SYS_DOUBLEWRITE_BLOCK_SIZE * UNIV_PAGE_SIZE,
+	       buf + TRX_SYS_DOUBLEWRITE_BLOCK_SIZE * UNIV_PAGE_SIZE,
+	       NULL);
+	/* Check if any of these pages is half-written in data files, in the
+	intended position */
+
+	page = buf;
+
+	for (i = 0; i < TRX_SYS_DOUBLEWRITE_BLOCK_SIZE * 2; i++) {
+
+		page_no = mach_read_from_4(page + FIL_PAGE_OFFSET);
+
+		if (trx_doublewrite_must_reset_space_ids) {
+
+			space_id = 0;
+			mach_write_to_4(page
+					+ FIL_PAGE_ARCH_LOG_NO_OR_SPACE_ID, 0);
+			/* We do not need to calculate new checksums for the
+			pages because the field .._SPACE_ID does not affect
+			them. Write the page back to where we read it from. */
+
+			if (i < TRX_SYS_DOUBLEWRITE_BLOCK_SIZE) {
+				source_page_no = block1 + i;
+			} else {
+				source_page_no = block2
+					+ i - TRX_SYS_DOUBLEWRITE_BLOCK_SIZE;
+			}
+
+			fil_io(OS_FILE_WRITE, TRUE, 0, 0, source_page_no, 0,
+			       UNIV_PAGE_SIZE, page, NULL);
+			/* printf("Resetting space id in page %lu\n",
+			source_page_no); */
+		} else {
+			space_id = mach_read_from_4(
+				page + FIL_PAGE_ARCH_LOG_NO_OR_SPACE_ID);
+		}
+
+		if (!restore_corrupt_pages) {
+			/* The database was shut down gracefully: no need to
+			restore pages */
+
+		} else if (!fil_tablespace_exists_in_mem(space_id)) {
+			/* Maybe we have dropped the single-table tablespace
+			and this page once belonged to it: do nothing */
+
+		} else if (!fil_check_adress_in_tablespace(space_id,
+							   page_no)) {
+			fprintf(stderr,
+				"InnoDB: Warning: a page in the"
+				" doublewrite buffer is not within space\n"
+				"InnoDB: bounds; space id %lu"
+				" page number %lu, page %lu in"
+				" doublewrite buf.\n",
+				(ulong) space_id, (ulong) page_no, (ulong) i);
+
+		} else if (space_id == TRX_SYS_SPACE
+			   && ((page_no >= block1
+				&& page_no
+				< block1 + TRX_SYS_DOUBLEWRITE_BLOCK_SIZE)
+			       || (page_no >= block2
+				   && page_no
+				   < (block2
+				      + TRX_SYS_DOUBLEWRITE_BLOCK_SIZE)))) {
+
+			/* It is an unwritten doublewrite buffer page:
+			do nothing */
+		} else {
+			ulint	zip_size = fil_space_get_zip_size(space_id);
+
+			/* Read in the actual page from the file */
+			fil_io(OS_FILE_READ, TRUE, space_id, zip_size,
+			       page_no, 0,
+			       zip_size ? zip_size : UNIV_PAGE_SIZE,
+			       read_buf, NULL);
+
+			/* Check if the page is corrupt */
+
+			if (UNIV_UNLIKELY
+			    (buf_page_is_corrupted(read_buf, zip_size))) {
+
+				fprintf(stderr,
+					"InnoDB: Warning: database page"
+					" corruption or a failed\n"
+					"InnoDB: file read of"
+					" space %lu page %lu.\n"
+					"InnoDB: Trying to recover it from"
+					" the doublewrite buffer.\n",
+					(ulong) space_id, (ulong) page_no);
+
+				if (buf_page_is_corrupted(page, zip_size)) {
+					fprintf(stderr,
+						"InnoDB: Dump of the page:\n");
+					buf_page_print(read_buf, zip_size);
+					fprintf(stderr,
+						"InnoDB: Dump of"
+						" corresponding page"
+						" in doublewrite buffer:\n");
+					buf_page_print(page, zip_size);
+
+					fprintf(stderr,
+						"InnoDB: Also the page in the"
+						" doublewrite buffer"
+						" is corrupt.\n"
+						"InnoDB: Cannot continue"
+						" operation.\n"
+						"InnoDB: You can try to"
+						" recover the database"
+						" with the my.cnf\n"
+						"InnoDB: option:\n"
+						"InnoDB: set-variable="
+						"innodb_force_recovery=6\n");
+					exit(1);
+				}
+
+				/* Write the good page from the
+				doublewrite buffer to the intended
+				position */
+
+				fil_io(OS_FILE_WRITE, TRUE, space_id,
+				       zip_size, page_no, 0,
+				       zip_size ? zip_size : UNIV_PAGE_SIZE,
+				       page, NULL);
+				fprintf(stderr,
+					"InnoDB: Recovered the page from"
+					" the doublewrite buffer.\n");
+			}
+		}
+
+		page += UNIV_PAGE_SIZE;
+	}
+
+	fil_flush_file_spaces(FIL_TABLESPACE);
+
+leave_func:
+	ut_free(unaligned_read_buf);
+}
+
+/********************************************************************
+Checks that trx is in the trx list. */
+UNIV_INTERN
+ibool
+trx_in_trx_list(
+/*============*/
+			/* out: TRUE if is in */
+	trx_t*	in_trx)	/* in: trx */
+{
+	trx_t*	trx;
+
+	ut_ad(mutex_own(&(kernel_mutex)));
+
+	trx = UT_LIST_GET_FIRST(trx_sys->trx_list);
+
+	while (trx != NULL) {
+
+		if (trx == in_trx) {
+
+			return(TRUE);
+		}
+
+		trx = UT_LIST_GET_NEXT(trx_list, trx);
+	}
+
+	return(FALSE);
+}
+
+/*********************************************************************
+Writes the value of max_trx_id to the file based trx system header. */
+UNIV_INTERN
+void
+trx_sys_flush_max_trx_id(void)
+/*==========================*/
+{
+	trx_sysf_t*	sys_header;
+	mtr_t		mtr;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	mtr_start(&mtr);
+
+	sys_header = trx_sysf_get(&mtr);
+
+	mlog_write_dulint(sys_header + TRX_SYS_TRX_ID_STORE,
+			  trx_sys->max_trx_id, &mtr);
+	mtr_commit(&mtr);
+}
+
+/*********************************************************************
+Updates the offset information about the end of the MySQL binlog entry
+which corresponds to the transaction just being committed. In a MySQL
+replication slave updates the latest master binlog position up to which
+replication has proceeded. */
+UNIV_INTERN
+void
+trx_sys_update_mysql_binlog_offset(
+/*===============================*/
+	const char*	file_name_in,/* in: MySQL log file name */
+	ib_int64_t	offset,	/* in: position in that log file */
+	ulint		field,	/* in: offset of the MySQL log info field in
+				the trx sys header */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	trx_sysf_t*	sys_header;
+	const char*	file_name;
+
+	if (ut_strlen(file_name_in) >= TRX_SYS_MYSQL_MASTER_LOG_NAME_LEN) {
+
+		/* We cannot fit the name to the 512 bytes we have reserved */
+		/* -> To store relay log file information, file_name must fit to the 480 bytes */
+
+		file_name = "";
+	}
+	else {
+		file_name = file_name_in;
+	}
+
+	sys_header = trx_sysf_get(mtr);
+
+	if (mach_read_from_4(sys_header + field
+			     + TRX_SYS_MYSQL_LOG_MAGIC_N_FLD)
+	    != TRX_SYS_MYSQL_LOG_MAGIC_N) {
+
+		mlog_write_ulint(sys_header + field
+				 + TRX_SYS_MYSQL_LOG_MAGIC_N_FLD,
+				 TRX_SYS_MYSQL_LOG_MAGIC_N,
+				 MLOG_4BYTES, mtr);
+	}
+
+	if (0 != strcmp((char*) (sys_header + field + TRX_SYS_MYSQL_LOG_NAME),
+			file_name)) {
+
+		mlog_write_string(sys_header + field
+				  + TRX_SYS_MYSQL_LOG_NAME,
+				  (byte*) file_name, 1 + ut_strlen(file_name),
+				  mtr);
+	}
+
+	if (mach_read_from_4(sys_header + field
+			     + TRX_SYS_MYSQL_LOG_OFFSET_HIGH) > 0
+	    || (offset >> 32) > 0) {
+
+		mlog_write_ulint(sys_header + field
+				 + TRX_SYS_MYSQL_LOG_OFFSET_HIGH,
+				 (ulint)(offset >> 32),
+				 MLOG_4BYTES, mtr);
+	}
+
+	mlog_write_ulint(sys_header + field
+			 + TRX_SYS_MYSQL_LOG_OFFSET_LOW,
+			 (ulint)(offset & 0xFFFFFFFFUL),
+			 MLOG_4BYTES, mtr);
+}
+
+#ifdef UNIV_HOTBACKUP
+/*********************************************************************
+Prints to stderr the MySQL binlog info in the system header if the
+magic number shows it valid. */
+UNIV_INTERN
+void
+trx_sys_print_mysql_binlog_offset_from_page(
+/*========================================*/
+	const byte*	page)	/* in: buffer containing the trx
+				system header page, i.e., page number
+				TRX_SYS_PAGE_NO in the tablespace */
+{
+	const trx_sysf_t*	sys_header;
+
+	sys_header = page + TRX_SYS;
+
+	if (mach_read_from_4(sys_header + TRX_SYS_MYSQL_LOG_INFO
+			     + TRX_SYS_MYSQL_LOG_MAGIC_N_FLD)
+	    == TRX_SYS_MYSQL_LOG_MAGIC_N) {
+
+		fprintf(stderr,
+			"ibbackup: Last MySQL binlog file position %lu %lu,"
+			" file name %s\n",
+			(ulong) mach_read_from_4(
+				sys_header + TRX_SYS_MYSQL_LOG_INFO
+				+ TRX_SYS_MYSQL_LOG_OFFSET_HIGH),
+			(ulong) mach_read_from_4(
+				sys_header + TRX_SYS_MYSQL_LOG_INFO
+				+ TRX_SYS_MYSQL_LOG_OFFSET_LOW),
+			sys_header + TRX_SYS_MYSQL_LOG_INFO
+			+ TRX_SYS_MYSQL_LOG_NAME);
+	}
+}
+#endif /* UNIV_HOTBACKUP */
+
+/*********************************************************************
+Stores the MySQL binlog offset info in the trx system header if
+the magic number shows it valid, and print the info to stderr */
+UNIV_INTERN
+void
+trx_sys_print_mysql_binlog_offset(void)
+/*===================================*/
+{
+	trx_sysf_t*	sys_header;
+	mtr_t		mtr;
+	ulint		trx_sys_mysql_bin_log_pos_high;
+	ulint		trx_sys_mysql_bin_log_pos_low;
+
+	mtr_start(&mtr);
+
+	sys_header = trx_sysf_get(&mtr);
+
+	if (mach_read_from_4(sys_header + TRX_SYS_MYSQL_LOG_INFO
+			     + TRX_SYS_MYSQL_LOG_MAGIC_N_FLD)
+	    != TRX_SYS_MYSQL_LOG_MAGIC_N) {
+
+		mtr_commit(&mtr);
+
+		return;
+	}
+
+	trx_sys_mysql_bin_log_pos_high = mach_read_from_4(
+		sys_header + TRX_SYS_MYSQL_LOG_INFO
+		+ TRX_SYS_MYSQL_LOG_OFFSET_HIGH);
+	trx_sys_mysql_bin_log_pos_low = mach_read_from_4(
+		sys_header + TRX_SYS_MYSQL_LOG_INFO
+		+ TRX_SYS_MYSQL_LOG_OFFSET_LOW);
+
+	trx_sys_mysql_bin_log_pos
+		= (((ib_int64_t)trx_sys_mysql_bin_log_pos_high) << 32)
+		+ (ib_int64_t)trx_sys_mysql_bin_log_pos_low;
+
+	ut_memcpy(trx_sys_mysql_bin_log_name,
+		  sys_header + TRX_SYS_MYSQL_LOG_INFO
+		  + TRX_SYS_MYSQL_LOG_NAME, TRX_SYS_MYSQL_LOG_NAME_LEN);
+
+	fprintf(stderr,
+		"InnoDB: Last MySQL binlog file position %lu %lu,"
+		" file name %s\n",
+		trx_sys_mysql_bin_log_pos_high, trx_sys_mysql_bin_log_pos_low,
+		trx_sys_mysql_bin_log_name);
+
+	mtr_commit(&mtr);
+}
+
+/*********************************************************************
+Prints to stderr the MySQL master log offset info in the trx system header if
+the magic number shows it valid. */
+UNIV_INTERN
+void
+trx_sys_print_mysql_master_log_pos(void)
+/*====================================*/
+{
+	trx_sysf_t*	sys_header;
+	mtr_t		mtr;
+
+	mtr_start(&mtr);
+
+	sys_header = trx_sysf_get(&mtr);
+
+	if (mach_read_from_4(sys_header + TRX_SYS_MYSQL_MASTER_LOG_INFO
+			     + TRX_SYS_MYSQL_LOG_MAGIC_N_FLD)
+	    != TRX_SYS_MYSQL_LOG_MAGIC_N) {
+
+		mtr_commit(&mtr);
+
+		return;
+	}
+
+	fprintf(stderr,
+		"InnoDB: In a MySQL replication slave the last"
+		" master binlog file\n"
+		"InnoDB: position %lu %lu, file name %s\n",
+		(ulong) mach_read_from_4(sys_header
+					 + TRX_SYS_MYSQL_MASTER_LOG_INFO
+					 + TRX_SYS_MYSQL_LOG_OFFSET_HIGH),
+		(ulong) mach_read_from_4(sys_header
+					 + TRX_SYS_MYSQL_MASTER_LOG_INFO
+					 + TRX_SYS_MYSQL_LOG_OFFSET_LOW),
+		sys_header + TRX_SYS_MYSQL_MASTER_LOG_INFO
+		+ TRX_SYS_MYSQL_LOG_NAME);
+
+	fprintf(stderr,
+		"InnoDB: and relay log file\n"
+		"InnoDB: position %lu %lu, file name %s\n",
+		(ulong) mach_read_from_4(sys_header
+					 + TRX_SYS_MYSQL_RELAY_LOG_INFO
+					 + TRX_SYS_MYSQL_LOG_OFFSET_HIGH),
+		(ulong) mach_read_from_4(sys_header
+					 + TRX_SYS_MYSQL_RELAY_LOG_INFO
+					 + TRX_SYS_MYSQL_LOG_OFFSET_LOW),
+		sys_header + TRX_SYS_MYSQL_RELAY_LOG_INFO
+		+ TRX_SYS_MYSQL_LOG_NAME);
+
+	/* Copy the master log position info to global variables we can
+	use in ha_innobase.cc to initialize glob_mi to right values */
+
+	ut_memcpy(trx_sys_mysql_master_log_name,
+		  sys_header + TRX_SYS_MYSQL_MASTER_LOG_INFO
+		  + TRX_SYS_MYSQL_LOG_NAME,
+		  TRX_SYS_MYSQL_MASTER_LOG_NAME_LEN);
+
+	trx_sys_mysql_master_log_pos
+		= (((ib_int64_t) mach_read_from_4(
+			    sys_header + TRX_SYS_MYSQL_MASTER_LOG_INFO
+			    + TRX_SYS_MYSQL_LOG_OFFSET_HIGH)) << 32)
+		+ ((ib_int64_t) mach_read_from_4(
+			   sys_header + TRX_SYS_MYSQL_MASTER_LOG_INFO
+			   + TRX_SYS_MYSQL_LOG_OFFSET_LOW));
+
+	ut_memcpy(trx_sys_mysql_relay_log_name,
+		  sys_header + TRX_SYS_MYSQL_RELAY_LOG_INFO
+		  + TRX_SYS_MYSQL_LOG_NAME,
+		  TRX_SYS_MYSQL_MASTER_LOG_NAME_LEN);
+
+	trx_sys_mysql_relay_log_pos
+		= (((ib_int64_t) mach_read_from_4(
+			    sys_header + TRX_SYS_MYSQL_RELAY_LOG_INFO
+			    + TRX_SYS_MYSQL_LOG_OFFSET_HIGH)) << 32)
+		+ ((ib_int64_t) mach_read_from_4(
+			   sys_header + TRX_SYS_MYSQL_RELAY_LOG_INFO
+			   + TRX_SYS_MYSQL_LOG_OFFSET_LOW));
+	mtr_commit(&mtr);
+}
+
+/********************************************************************
+Looks for a free slot for a rollback segment in the trx system file copy. */
+UNIV_INTERN
+ulint
+trx_sysf_rseg_find_free(
+/*====================*/
+			/* out: slot index or ULINT_UNDEFINED if not found */
+	mtr_t*	mtr)	/* in: mtr */
+{
+	trx_sysf_t*	sys_header;
+	ulint		page_no;
+	ulint		i;
+
+	ut_ad(mutex_own(&(kernel_mutex)));
+
+	sys_header = trx_sysf_get(mtr);
+
+	for (i = 0; i < TRX_SYS_N_RSEGS; i++) {
+
+		page_no = trx_sysf_rseg_get_page_no(sys_header, i, mtr);
+
+		if (page_no == FIL_NULL) {
+
+			return(i);
+		}
+	}
+
+	return(ULINT_UNDEFINED);
+}
+
+/*********************************************************************
+Creates the file page for the transaction system. This function is called only
+at the database creation, before trx_sys_init. */
+static
+void
+trx_sysf_create(
+/*============*/
+	mtr_t*	mtr)	/* in: mtr */
+{
+	trx_sysf_t*	sys_header;
+	ulint		slot_no;
+	buf_block_t*	block;
+	page_t*		page;
+	ulint		page_no;
+	ulint		i;
+
+	ut_ad(mtr);
+
+	/* Note that below we first reserve the file space x-latch, and
+	then enter the kernel: we must do it in this order to conform
+	to the latching order rules. */
+
+	mtr_x_lock(fil_space_get_latch(TRX_SYS_SPACE, NULL), mtr);
+	mutex_enter(&kernel_mutex);
+
+	/* Create the trx sys file block in a new allocated file segment */
+	block = fseg_create(TRX_SYS_SPACE, 0, TRX_SYS + TRX_SYS_FSEG_HEADER,
+			    mtr);
+	buf_block_dbg_add_level(block, SYNC_TRX_SYS_HEADER);
+
+	ut_a(buf_block_get_page_no(block) == TRX_SYS_PAGE_NO);
+
+	page = buf_block_get_frame(block);
+
+	mlog_write_ulint(page + FIL_PAGE_TYPE, FIL_PAGE_TYPE_TRX_SYS,
+			 MLOG_2BYTES, mtr);
+
+	/* Reset the doublewrite buffer magic number to zero so that we
+	know that the doublewrite buffer has not yet been created (this
+	suppresses a Valgrind warning) */
+
+	mlog_write_ulint(page + TRX_SYS_DOUBLEWRITE
+			 + TRX_SYS_DOUBLEWRITE_MAGIC, 0, MLOG_4BYTES, mtr);
+
+	sys_header = trx_sysf_get(mtr);
+
+	/* Start counting transaction ids from number 1 up */
+	mlog_write_dulint(sys_header + TRX_SYS_TRX_ID_STORE,
+			  ut_dulint_create(0, 1), mtr);
+
+	/* Reset the rollback segment slots */
+	for (i = 0; i < TRX_SYS_N_RSEGS; i++) {
+
+		trx_sysf_rseg_set_space(sys_header, i, ULINT_UNDEFINED, mtr);
+		trx_sysf_rseg_set_page_no(sys_header, i, FIL_NULL, mtr);
+	}
+
+	/* The remaining area (up to the page trailer) is uninitialized.
+	Silence Valgrind warnings about it. */
+	UNIV_MEM_VALID(sys_header + (TRX_SYS_RSEGS
+				     + TRX_SYS_N_RSEGS * TRX_SYS_RSEG_SLOT_SIZE
+				     + TRX_SYS_RSEG_SPACE),
+		       (UNIV_PAGE_SIZE - FIL_PAGE_DATA_END
+			- (TRX_SYS_RSEGS
+			   + TRX_SYS_N_RSEGS * TRX_SYS_RSEG_SLOT_SIZE
+			   + TRX_SYS_RSEG_SPACE))
+		       + page - sys_header);
+
+	/* Create the first rollback segment in the SYSTEM tablespace */
+	page_no = trx_rseg_header_create(TRX_SYS_SPACE, 0, ULINT_MAX, &slot_no,
+					 mtr);
+	ut_a(slot_no == TRX_SYS_SYSTEM_RSEG_ID);
+	ut_a(page_no != FIL_NULL);
+
+	mutex_exit(&kernel_mutex);
+}
+
+/*********************************************************************
+Creates and initializes the central memory structures for the transaction
+system. This is called when the database is started. */
+UNIV_INTERN
+void
+trx_sys_init_at_db_start(void)
+/*==========================*/
+{
+	trx_sysf_t*	sys_header;
+	ib_int64_t	rows_to_undo	= 0;
+	const char*	unit		= "";
+	trx_t*		trx;
+	mtr_t		mtr;
+
+	mtr_start(&mtr);
+
+	ut_ad(trx_sys == NULL);
+
+	mutex_enter(&kernel_mutex);
+
+	trx_sys = mem_alloc(sizeof(trx_sys_t));
+
+	sys_header = trx_sysf_get(&mtr);
+
+	trx_rseg_list_and_array_init(sys_header, &mtr);
+
+	trx_sys->latest_rseg = UT_LIST_GET_FIRST(trx_sys->rseg_list);
+
+	/* VERY important: after the database is started, max_trx_id value is
+	divisible by TRX_SYS_TRX_ID_WRITE_MARGIN, and the 'if' in
+	trx_sys_get_new_trx_id will evaluate to TRUE when the function
+	is first time called, and the value for trx id will be written
+	to the disk-based header! Thus trx id values will not overlap when
+	the database is repeatedly started! */
+
+	trx_sys->max_trx_id = ut_dulint_add(
+		ut_dulint_align_up(mtr_read_dulint(
+					   sys_header
+					   + TRX_SYS_TRX_ID_STORE, &mtr),
+				   TRX_SYS_TRX_ID_WRITE_MARGIN),
+		2 * TRX_SYS_TRX_ID_WRITE_MARGIN);
+
+	UT_LIST_INIT(trx_sys->mysql_trx_list);
+	trx_dummy_sess = sess_open();
+	trx_lists_init_at_db_start();
+
+	if (UT_LIST_GET_LEN(trx_sys->trx_list) > 0) {
+		trx = UT_LIST_GET_FIRST(trx_sys->trx_list);
+
+		for (;;) {
+
+			if ( trx->conc_state != TRX_PREPARED) {
+				rows_to_undo += ut_conv_dulint_to_longlong(
+					trx->undo_no);
+			}
+
+			trx = UT_LIST_GET_NEXT(trx_list, trx);
+
+			if (!trx) {
+				break;
+			}
+		}
+
+		if (rows_to_undo > 1000000000) {
+			unit = "M";
+			rows_to_undo = rows_to_undo / 1000000;
+		}
+
+		fprintf(stderr,
+			"InnoDB: %lu transaction(s) which must be"
+			" rolled back or cleaned up\n"
+			"InnoDB: in total %lu%s row operations to undo\n",
+			(ulong) UT_LIST_GET_LEN(trx_sys->trx_list),
+			(ulong) rows_to_undo, unit);
+
+		fprintf(stderr, "InnoDB: Trx id counter is " TRX_ID_FMT "\n",
+			TRX_ID_PREP_PRINTF(trx_sys->max_trx_id));
+	}
+
+	UT_LIST_INIT(trx_sys->view_list);
+
+	trx_purge_sys_create();
+
+	mutex_exit(&kernel_mutex);
+
+	mtr_commit(&mtr);
+}
+
+/*********************************************************************
+Creates and initializes the transaction system at the database creation. */
+UNIV_INTERN
+void
+trx_sys_create(void)
+/*================*/
+{
+	mtr_t	mtr;
+
+	mtr_start(&mtr);
+
+	trx_sysf_create(&mtr);
+
+	mtr_commit(&mtr);
+
+	trx_sys_init_at_db_start();
+}
+
+/*********************************************************************
+Create extra rollback segments when create_new_db */
+UNIV_INTERN
+void
+trx_sys_create_extra_rseg(
+/*======================*/
+	ulint	num)	/* in: number of extra user rollback segments */
+{
+	mtr_t	mtr;
+	ulint	slot_no;
+	ulint	i;
+
+	/* Craete extra rollback segments */
+	mtr_start(&mtr);
+	for (i = 1; i < num + 1; i++) {
+		if(!trx_rseg_create(TRX_SYS_SPACE, ULINT_MAX, &slot_no, &mtr)) {
+			fprintf(stderr,
+"InnoDB: Warning: Failed to create extra rollback segments.\n");
+			break;
+		}
+		ut_a(slot_no == i);
+	}
+	mtr_commit(&mtr);
+}
+
+/*********************************************************************
+Update the file format tag. */
+static
+ibool
+trx_sys_file_format_max_write(
+/*==========================*/
+					/* out: always TRUE */
+	ulint		format_id,	/* in: file format id */
+	const char**	name)		/* out: max file format name, can
+					be NULL */
+{
+	mtr_t		mtr;
+	byte*		ptr;
+	buf_block_t*	block;
+	ulint		tag_value_low;
+
+	mtr_start(&mtr);
+
+	block = buf_page_get(
+		TRX_SYS_SPACE, 0, TRX_SYS_PAGE_NO, RW_X_LATCH, &mtr);
+
+	file_format_max.id = format_id;
+	file_format_max.name = trx_sys_file_format_id_to_name(format_id);
+
+	ptr = buf_block_get_frame(block) + TRX_SYS_FILE_FORMAT_TAG;
+	tag_value_low = format_id + TRX_SYS_FILE_FORMAT_TAG_MAGIC_N_LOW;
+
+	if (name) {
+		*name = file_format_max.name;
+	}
+
+	mlog_write_dulint(
+		ptr,
+		ut_dulint_create(TRX_SYS_FILE_FORMAT_TAG_MAGIC_N_HIGH,
+				 tag_value_low),
+		&mtr);
+
+	mtr_commit(&mtr);
+
+	return(TRUE);
+}
+
+/*********************************************************************
+Read the file format tag. */
+static
+ulint
+trx_sys_file_format_max_read(void)
+/*==============================*/
+				/* out: the file format or
+				ULINT_UNDEFINED if not set. */
+{
+	mtr_t			mtr;
+	const byte*		ptr;
+	const buf_block_t*	block;
+	ulint			format_id;
+	dulint			file_format_id;
+
+	/* Since this is called during the startup phase it's safe to
+	read the value without a covering mutex. */
+	mtr_start(&mtr);
+
+	block = buf_page_get(
+		TRX_SYS_SPACE, 0, TRX_SYS_PAGE_NO, RW_X_LATCH, &mtr);
+
+	ptr = buf_block_get_frame(block) + TRX_SYS_FILE_FORMAT_TAG;
+	file_format_id = mach_read_from_8(ptr);
+
+	mtr_commit(&mtr);
+
+	format_id = file_format_id.low - TRX_SYS_FILE_FORMAT_TAG_MAGIC_N_LOW;
+
+	if (file_format_id.high != TRX_SYS_FILE_FORMAT_TAG_MAGIC_N_HIGH
+	    || format_id >= FILE_FORMAT_NAME_N) {
+
+		/* Either it has never been tagged, or garbage in it. */
+		return(ULINT_UNDEFINED);
+	}
+
+	return(format_id);
+}
+
+/*********************************************************************
+Get the name representation of the file format from its id. */
+UNIV_INTERN
+const char*
+trx_sys_file_format_id_to_name(
+/*===========================*/
+				/* out: pointer to the name */
+	const ulint	id)	/* in: id of the file format */
+{
+	ut_a(id < FILE_FORMAT_NAME_N);
+
+	return(file_format_name_map[id]);
+}
+
+/*********************************************************************
+Check for the max file format tag stored on disk. Note: If max_format_id
+is == DICT_TF_FORMAT_MAX + 1 then we only print a warning. */
+UNIV_INTERN
+ulint
+trx_sys_file_format_max_check(
+/*==========================*/
+				/* out: DB_SUCCESS or error code */
+	ulint	max_format_id)	/* in: max format id to check */
+{
+	ulint	format_id;
+
+	/* Check the file format in the tablespace. Do not try to
+	recover if the file format is not supported by the engine
+	unless forced by the user. */
+	format_id = trx_sys_file_format_max_read();
+	if (format_id == ULINT_UNDEFINED) {
+		/* Format ID was not set. Set it to minimum possible
+		value. */
+		format_id = DICT_TF_FORMAT_51;
+	}
+
+	ut_print_timestamp(stderr);
+	fprintf(stderr,
+		"  InnoDB: highest supported file format is %s.\n",
+		trx_sys_file_format_id_to_name(DICT_TF_FORMAT_MAX));
+
+	if (format_id > DICT_TF_FORMAT_MAX) {
+
+		ut_a(format_id < FILE_FORMAT_NAME_N);
+
+		ut_print_timestamp(stderr);
+		fprintf(stderr,
+			"  InnoDB: %s: the system tablespace is in a file "
+			"format that this version doesn't support - %s\n",
+			((max_format_id <= DICT_TF_FORMAT_MAX)
+				? "Error" : "Warning"),
+			trx_sys_file_format_id_to_name(format_id));
+
+		if (max_format_id <= DICT_TF_FORMAT_MAX) {
+			return(DB_ERROR);
+		}
+	}
+
+	format_id = (format_id > max_format_id) ? format_id : max_format_id;
+
+	/* We don't need a mutex here, as this function should only
+	be called once at start up. */
+	file_format_max.id = format_id;
+	file_format_max.name = trx_sys_file_format_id_to_name(format_id);
+
+	return(DB_SUCCESS);
+}
+
+/*********************************************************************
+Set the file format id unconditionally except if it's already the
+same value. */
+UNIV_INTERN
+ibool
+trx_sys_file_format_max_set(
+/*========================*/
+					/* out: TRUE if value updated */
+	ulint		format_id,	/* in: file format id */
+	const char**	name)		/* out: max file format name or
+					NULL if not needed. */
+{
+	ibool		ret = FALSE;
+
+	ut_a(format_id <= DICT_TF_FORMAT_MAX);
+
+	mutex_enter(&file_format_max.mutex);
+
+	/* Only update if not already same value. */
+	if (format_id != file_format_max.id) {
+
+		ret = trx_sys_file_format_max_write(format_id, name);
+	}
+
+	mutex_exit(&file_format_max.mutex);
+
+	return(ret);
+}
+
+/************************************************************************
+Tags the system table space with minimum format id if it has not been
+tagged yet.
+WARNING: This function is only called during the startup and AFTER the
+redo log application during recovery has finished. */
+UNIV_INTERN
+void
+trx_sys_file_format_tag_init(void)
+/*==============================*/
+{
+	ulint	format_id;
+
+	format_id = trx_sys_file_format_max_read();
+
+	/* If format_id is not set then set it to the minimum. */
+	if (format_id == ULINT_UNDEFINED) {
+		trx_sys_file_format_max_set(DICT_TF_FORMAT_51, NULL);
+	}
+}
+
+/************************************************************************
+Update the file format tag in the system tablespace only if the given
+format id is greater than the known max id. */
+UNIV_INTERN
+ibool
+trx_sys_file_format_max_upgrade(
+/*============================*/
+					/* out: TRUE if format_id was
+					bigger than the known max id */
+	const char**	name,		/* out: max file format name */
+	ulint		format_id)	/* in: file format identifier */
+{
+	ibool		ret = FALSE;
+
+	ut_a(name);
+	ut_a(file_format_max.name != NULL);
+	ut_a(format_id <= DICT_TF_FORMAT_MAX);
+
+	mutex_enter(&file_format_max.mutex);
+
+	if (format_id > file_format_max.id) {
+
+		ret = trx_sys_file_format_max_write(format_id, name);
+	}
+
+	mutex_exit(&file_format_max.mutex);
+
+	return(ret);
+}
+
+/*********************************************************************
+Get the name representation of the file format from its id. */
+UNIV_INTERN
+const char*
+trx_sys_file_format_max_get(void)
+/*=============================*/
+				/* out: pointer to the max format name */
+{
+	return(file_format_max.name);
+}
+
+/*********************************************************************
+Initializes the tablespace tag system. */
+UNIV_INTERN
+void
+trx_sys_file_format_init(void)
+/*==========================*/
+{
+	mutex_create(&file_format_max.mutex, SYNC_FILE_FORMAT_TAG);
+
+	/* We don't need a mutex here, as this function should only
+	be called once at start up. */
+	file_format_max.id = DICT_TF_FORMAT_51;
+
+	file_format_max.name = trx_sys_file_format_id_to_name(
+		file_format_max.id);
+}
+
+/*********************************************************************
+Closes the tablespace tag system. */
+UNIV_INTERN
+void
+trx_sys_file_format_close(void)
+/*===========================*/
+{
+	/* Does nothing at the moment */
+}
diff --git a/storage/xtradb/trx/trx0trx.c b/storage/xtradb/trx/trx0trx.c
new file mode 100644
index 00000000000..5fb234e3aa7
--- /dev/null
+++ b/storage/xtradb/trx/trx0trx.c
@@ -0,0 +1,2082 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+The transaction
+
+Created 3/26/1996 Heikki Tuuri
+*******************************************************/
+
+#include "trx0trx.h"
+
+#ifdef UNIV_NONINL
+#include "trx0trx.ic"
+#endif
+
+#include "trx0undo.h"
+#include "trx0rseg.h"
+#include "log0log.h"
+#include "que0que.h"
+#include "lock0lock.h"
+#include "trx0roll.h"
+#include "usr0sess.h"
+#include "read0read.h"
+#include "srv0srv.h"
+#include "thr0loc.h"
+#include "btr0sea.h"
+#include "os0proc.h"
+#include "trx0xa.h"
+#include "ha_prototypes.h"
+
+/* Dummy session used currently in MySQL interface */
+UNIV_INTERN sess_t*		trx_dummy_sess = NULL;
+
+/* Number of transactions currently allocated for MySQL: protected by
+the kernel mutex */
+UNIV_INTERN ulint	trx_n_mysql_transactions = 0;
+
+/*****************************************************************
+Set detailed error message for the transaction. */
+UNIV_INTERN
+void
+trx_set_detailed_error(
+/*===================*/
+	trx_t*		trx,	/* in: transaction struct */
+	const char*	msg)	/* in: detailed error message */
+{
+	ut_strlcpy(trx->detailed_error, msg, sizeof(trx->detailed_error));
+}
+
+/*****************************************************************
+Set detailed error message for the transaction from a file. Note that the
+file is rewinded before reading from it. */
+UNIV_INTERN
+void
+trx_set_detailed_error_from_file(
+/*=============================*/
+	trx_t*	trx,	/* in: transaction struct */
+	FILE*	file)	/* in: file to read message from */
+{
+	os_file_read_string(file, trx->detailed_error,
+			    sizeof(trx->detailed_error));
+}
+
+/********************************************************************
+Creates and initializes a transaction object. */
+UNIV_INTERN
+trx_t*
+trx_create(
+/*=======*/
+			/* out, own: the transaction */
+	sess_t*	sess)	/* in: session */
+{
+	trx_t*	trx;
+
+	ut_ad(mutex_own(&kernel_mutex));
+	ut_ad(sess);
+
+	trx = mem_alloc(sizeof(trx_t));
+
+	trx->magic_n = TRX_MAGIC_N;
+
+	trx->op_info = "";
+
+	trx->is_purge = 0;
+	trx->is_recovered = 0;
+	trx->conc_state = TRX_NOT_STARTED;
+	trx->start_time = time(NULL);
+
+	trx->isolation_level = TRX_ISO_REPEATABLE_READ;
+
+	trx->id = ut_dulint_zero;
+	trx->no = ut_dulint_max;
+
+	trx->support_xa = TRUE;
+
+	trx->check_foreigns = TRUE;
+	trx->check_unique_secondary = TRUE;
+
+	trx->flush_log_later = FALSE;
+	trx->must_flush_log_later = FALSE;
+
+	trx->dict_operation = TRX_DICT_OP_NONE;
+	trx->table_id = ut_dulint_zero;
+
+	trx->mysql_thd = NULL;
+	trx->mysql_query_str = NULL;
+	trx->active_trans = 0;
+	trx->duplicates = 0;
+
+	trx->n_mysql_tables_in_use = 0;
+	trx->mysql_n_tables_locked = 0;
+
+	trx->mysql_log_file_name = NULL;
+	trx->mysql_log_offset = 0;
+	trx->mysql_master_log_file_name = "";
+	trx->mysql_master_log_pos = 0;
+	trx->mysql_relay_log_file_name = "";
+	trx->mysql_relay_log_pos = 0;
+
+	mutex_create(&trx->undo_mutex, SYNC_TRX_UNDO);
+
+	trx->rseg = NULL;
+
+	trx->undo_no = ut_dulint_zero;
+	trx->last_sql_stat_start.least_undo_no = ut_dulint_zero;
+	trx->insert_undo = NULL;
+	trx->update_undo = NULL;
+	trx->undo_no_arr = NULL;
+
+	trx->error_state = DB_SUCCESS;
+	trx->error_key_num = 0;
+	trx->detailed_error[0] = '\0';
+
+	trx->sess = sess;
+	trx->que_state = TRX_QUE_RUNNING;
+	trx->n_active_thrs = 0;
+
+	trx->handling_signals = FALSE;
+
+	UT_LIST_INIT(trx->signals);
+	UT_LIST_INIT(trx->reply_signals);
+
+	trx->graph = NULL;
+
+	trx->wait_lock = NULL;
+	trx->was_chosen_as_deadlock_victim = FALSE;
+	UT_LIST_INIT(trx->wait_thrs);
+
+	trx->lock_heap = mem_heap_create_in_buffer(256);
+	UT_LIST_INIT(trx->trx_locks);
+
+	UT_LIST_INIT(trx->trx_savepoints);
+
+	trx->dict_operation_lock_mode = 0;
+	trx->has_search_latch = FALSE;
+	trx->search_latch_timeout = BTR_SEA_TIMEOUT;
+
+	trx->declared_to_be_inside_innodb = FALSE;
+	trx->n_tickets_to_enter_innodb = 0;
+
+	trx->global_read_view_heap = mem_heap_create(256);
+	trx->global_read_view = NULL;
+	trx->read_view = NULL;
+
+	/* Set X/Open XA transaction identification to NULL */
+	memset(&trx->xid, 0, sizeof(trx->xid));
+	trx->xid.formatID = -1;
+
+	trx->n_autoinc_rows = 0;
+
+	/* Remember to free the vector explicitly. */
+	trx->autoinc_locks = ib_vector_create(
+		mem_heap_create(sizeof(ib_vector_t) + sizeof(void*) * 4), 4);
+
+	trx_reset_new_rec_lock_info(trx);
+
+	return(trx);
+}
+
+/************************************************************************
+Creates a transaction object for MySQL. */
+UNIV_INTERN
+trx_t*
+trx_allocate_for_mysql(void)
+/*========================*/
+				/* out, own: transaction object */
+{
+	trx_t*	trx;
+
+	mutex_enter(&kernel_mutex);
+
+	trx = trx_create(trx_dummy_sess);
+
+	trx_n_mysql_transactions++;
+
+	UT_LIST_ADD_FIRST(mysql_trx_list, trx_sys->mysql_trx_list, trx);
+
+	mutex_exit(&kernel_mutex);
+
+	trx->mysql_thread_id = os_thread_get_curr_id();
+
+	trx->mysql_process_no = os_proc_get_number();
+
+	return(trx);
+}
+
+/************************************************************************
+Creates a transaction object for background operations by the master thread. */
+UNIV_INTERN
+trx_t*
+trx_allocate_for_background(void)
+/*=============================*/
+				/* out, own: transaction object */
+{
+	trx_t*	trx;
+
+	mutex_enter(&kernel_mutex);
+
+	trx = trx_create(trx_dummy_sess);
+
+	mutex_exit(&kernel_mutex);
+
+	return(trx);
+}
+
+/************************************************************************
+Releases the search latch if trx has reserved it. */
+UNIV_INTERN
+void
+trx_search_latch_release_if_reserved(
+/*=================================*/
+	trx_t*	   trx) /* in: transaction */
+{
+	if (trx->has_search_latch) {
+		rw_lock_s_unlock(&btr_search_latch);
+
+		trx->has_search_latch = FALSE;
+	}
+}
+
+/************************************************************************
+Frees a transaction object. */
+UNIV_INTERN
+void
+trx_free(
+/*=====*/
+	trx_t*	trx)	/* in, own: trx object */
+{
+	ut_ad(mutex_own(&kernel_mutex));
+
+	if (trx->declared_to_be_inside_innodb) {
+		ut_print_timestamp(stderr);
+		fputs("  InnoDB: Error: Freeing a trx which is declared"
+		      " to be processing\n"
+		      "InnoDB: inside InnoDB.\n", stderr);
+		trx_print(stderr, trx, 600);
+		putc('\n', stderr);
+
+		/* This is an error but not a fatal error. We must keep
+		the counters like srv_conc_n_threads accurate. */
+		srv_conc_force_exit_innodb(trx);
+	}
+
+	if (trx->n_mysql_tables_in_use != 0
+	    || trx->mysql_n_tables_locked != 0) {
+
+		ut_print_timestamp(stderr);
+		fprintf(stderr,
+			"  InnoDB: Error: MySQL is freeing a thd\n"
+			"InnoDB: though trx->n_mysql_tables_in_use is %lu\n"
+			"InnoDB: and trx->mysql_n_tables_locked is %lu.\n",
+			(ulong)trx->n_mysql_tables_in_use,
+			(ulong)trx->mysql_n_tables_locked);
+
+		trx_print(stderr, trx, 600);
+
+		ut_print_buf(stderr, trx, sizeof(trx_t));
+		putc('\n', stderr);
+	}
+
+	ut_a(trx->magic_n == TRX_MAGIC_N);
+
+	trx->magic_n = 11112222;
+
+	ut_a(trx->conc_state == TRX_NOT_STARTED);
+
+	mutex_free(&(trx->undo_mutex));
+
+	ut_a(trx->insert_undo == NULL);
+	ut_a(trx->update_undo == NULL);
+
+	if (trx->undo_no_arr) {
+		trx_undo_arr_free(trx->undo_no_arr);
+	}
+
+	ut_a(UT_LIST_GET_LEN(trx->signals) == 0);
+	ut_a(UT_LIST_GET_LEN(trx->reply_signals) == 0);
+
+	ut_a(trx->wait_lock == NULL);
+	ut_a(UT_LIST_GET_LEN(trx->wait_thrs) == 0);
+
+	ut_a(!trx->has_search_latch);
+
+	ut_a(trx->dict_operation_lock_mode == 0);
+
+	if (trx->lock_heap) {
+		mem_heap_free(trx->lock_heap);
+	}
+
+	ut_a(UT_LIST_GET_LEN(trx->trx_locks) == 0);
+
+	if (trx->global_read_view_heap) {
+		mem_heap_free(trx->global_read_view_heap);
+	}
+
+	trx->global_read_view = NULL;
+
+	ut_a(trx->read_view == NULL);
+
+	ut_a(ib_vector_is_empty(trx->autoinc_locks));
+	/* We allocated a dedicated heap for the vector. */
+	ib_vector_free(trx->autoinc_locks);
+
+	mem_free(trx);
+}
+
+/************************************************************************
+Frees a transaction object for MySQL. */
+UNIV_INTERN
+void
+trx_free_for_mysql(
+/*===============*/
+	trx_t*	trx)	/* in, own: trx object */
+{
+	mutex_enter(&kernel_mutex);
+
+	UT_LIST_REMOVE(mysql_trx_list, trx_sys->mysql_trx_list, trx);
+
+	trx_free(trx);
+
+	ut_a(trx_n_mysql_transactions > 0);
+
+	trx_n_mysql_transactions--;
+
+	mutex_exit(&kernel_mutex);
+}
+
+/************************************************************************
+Frees a transaction object of a background operation of the master thread. */
+UNIV_INTERN
+void
+trx_free_for_background(
+/*====================*/
+	trx_t*	trx)	/* in, own: trx object */
+{
+	mutex_enter(&kernel_mutex);
+
+	trx_free(trx);
+
+	mutex_exit(&kernel_mutex);
+}
+
+/********************************************************************
+Inserts the trx handle in the trx system trx list in the right position.
+The list is sorted on the trx id so that the biggest id is at the list
+start. This function is used at the database startup to insert incomplete
+transactions to the list. */
+static
+void
+trx_list_insert_ordered(
+/*====================*/
+	trx_t*	trx)	/* in: trx handle */
+{
+	trx_t*	trx2;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	trx2 = UT_LIST_GET_FIRST(trx_sys->trx_list);
+
+	while (trx2 != NULL) {
+		if (ut_dulint_cmp(trx->id, trx2->id) >= 0) {
+
+			ut_ad(ut_dulint_cmp(trx->id, trx2->id) == 1);
+			break;
+		}
+		trx2 = UT_LIST_GET_NEXT(trx_list, trx2);
+	}
+
+	if (trx2 != NULL) {
+		trx2 = UT_LIST_GET_PREV(trx_list, trx2);
+
+		if (trx2 == NULL) {
+			UT_LIST_ADD_FIRST(trx_list, trx_sys->trx_list, trx);
+		} else {
+			UT_LIST_INSERT_AFTER(trx_list, trx_sys->trx_list,
+					     trx2, trx);
+		}
+	} else {
+		UT_LIST_ADD_LAST(trx_list, trx_sys->trx_list, trx);
+	}
+}
+
+/********************************************************************
+Creates trx objects for transactions and initializes the trx list of
+trx_sys at database start. Rollback segment and undo log lists must
+already exist when this function is called, because the lists of
+transactions to be rolled back or cleaned up are built based on the
+undo log lists. */
+UNIV_INTERN
+void
+trx_lists_init_at_db_start(void)
+/*============================*/
+{
+	trx_rseg_t*	rseg;
+	trx_undo_t*	undo;
+	trx_t*		trx;
+
+	UT_LIST_INIT(trx_sys->trx_list);
+
+	/* Look from the rollback segments if there exist undo logs for
+	transactions */
+
+	rseg = UT_LIST_GET_FIRST(trx_sys->rseg_list);
+
+	while (rseg != NULL) {
+		undo = UT_LIST_GET_FIRST(rseg->insert_undo_list);
+
+		while (undo != NULL) {
+
+			trx = trx_create(trx_dummy_sess);
+
+			trx->is_recovered = TRUE;
+			trx->id = undo->trx_id;
+			trx->xid = undo->xid;
+			trx->insert_undo = undo;
+			trx->rseg = rseg;
+
+			if (undo->state != TRX_UNDO_ACTIVE) {
+
+				/* Prepared transactions are left in
+				the prepared state waiting for a
+				commit or abort decision from MySQL */
+
+				if (undo->state == TRX_UNDO_PREPARED) {
+
+					fprintf(stderr,
+						"InnoDB: Transaction "
+						TRX_ID_FMT
+						" was in the"
+						" XA prepared state.\n",
+						TRX_ID_PREP_PRINTF(trx->id));
+
+					if (srv_force_recovery == 0) {
+
+						trx->conc_state = TRX_PREPARED;
+					} else {
+						fprintf(stderr,
+							"InnoDB: Since"
+							" innodb_force_recovery"
+							" > 0, we will"
+							" rollback it"
+							" anyway.\n");
+
+						trx->conc_state = TRX_ACTIVE;
+					}
+				} else {
+					trx->conc_state
+						= TRX_COMMITTED_IN_MEMORY;
+				}
+
+				/* We give a dummy value for the trx no;
+				this should have no relevance since purge
+				is not interested in committed transaction
+				numbers, unless they are in the history
+				list, in which case it looks the number
+				from the disk based undo log structure */
+
+				trx->no = trx->id;
+			} else {
+				trx->conc_state = TRX_ACTIVE;
+
+				/* A running transaction always has the number
+				field inited to ut_dulint_max */
+
+				trx->no = ut_dulint_max;
+			}
+
+			if (undo->dict_operation) {
+				trx_set_dict_operation(
+					trx, TRX_DICT_OP_TABLE);
+				trx->table_id = undo->table_id;
+			}
+
+			if (!undo->empty) {
+				trx->undo_no = ut_dulint_add(undo->top_undo_no,
+							     1);
+			}
+
+			trx_list_insert_ordered(trx);
+
+			undo = UT_LIST_GET_NEXT(undo_list, undo);
+		}
+
+		undo = UT_LIST_GET_FIRST(rseg->update_undo_list);
+
+		while (undo != NULL) {
+			trx = trx_get_on_id(undo->trx_id);
+
+			if (NULL == trx) {
+				trx = trx_create(trx_dummy_sess);
+
+				trx->is_recovered = TRUE;
+				trx->id = undo->trx_id;
+				trx->xid = undo->xid;
+
+				if (undo->state != TRX_UNDO_ACTIVE) {
+
+					/* Prepared transactions are left in
+					the prepared state waiting for a
+					commit or abort decision from MySQL */
+
+					if (undo->state == TRX_UNDO_PREPARED) {
+						fprintf(stderr,
+							"InnoDB: Transaction "
+							TRX_ID_FMT " was in the"
+							" XA prepared state.\n",
+							TRX_ID_PREP_PRINTF(
+								trx->id));
+
+						if (srv_force_recovery == 0) {
+
+							trx->conc_state
+								= TRX_PREPARED;
+						} else {
+							fprintf(stderr,
+								"InnoDB: Since"
+								" innodb_force_recovery"
+								" > 0, we will"
+								" rollback it"
+								" anyway.\n");
+
+							trx->conc_state
+								= TRX_ACTIVE;
+						}
+					} else {
+						trx->conc_state
+							= TRX_COMMITTED_IN_MEMORY;
+					}
+
+					/* We give a dummy value for the trx
+					number */
+
+					trx->no = trx->id;
+				} else {
+					trx->conc_state = TRX_ACTIVE;
+
+					/* A running transaction always has
+					the number field inited to
+					ut_dulint_max */
+
+					trx->no = ut_dulint_max;
+				}
+
+				trx->rseg = rseg;
+				trx_list_insert_ordered(trx);
+
+				if (undo->dict_operation) {
+					trx_set_dict_operation(
+						trx, TRX_DICT_OP_TABLE);
+					trx->table_id = undo->table_id;
+				}
+			}
+
+			trx->update_undo = undo;
+
+			if ((!undo->empty)
+			    && (ut_dulint_cmp(undo->top_undo_no,
+					      trx->undo_no) >= 0)) {
+
+				trx->undo_no = ut_dulint_add(undo->top_undo_no,
+							     1);
+			}
+
+			undo = UT_LIST_GET_NEXT(undo_list, undo);
+		}
+
+		rseg = UT_LIST_GET_NEXT(rseg_list, rseg);
+	}
+}
+
+/**********************************************************************
+Assigns a rollback segment to a transaction in a round-robin fashion.
+Skips the SYSTEM rollback segment if another is available. */
+UNIV_INLINE
+ulint
+trx_assign_rseg(void)
+/*=================*/
+			/* out: assigned rollback segment id */
+{
+	trx_rseg_t*	rseg	= trx_sys->latest_rseg;
+
+	ut_ad(mutex_own(&kernel_mutex));
+loop:
+	/* Get next rseg in a round-robin fashion */
+
+	rseg = UT_LIST_GET_NEXT(rseg_list, rseg);
+
+	if (rseg == NULL) {
+		rseg = UT_LIST_GET_FIRST(trx_sys->rseg_list);
+	}
+
+	/* If it is the SYSTEM rollback segment, and there exist others, skip
+	it */
+
+	if ((rseg->id == TRX_SYS_SYSTEM_RSEG_ID)
+	    && (UT_LIST_GET_LEN(trx_sys->rseg_list) > 1)) {
+		goto loop;
+	}
+
+	trx_sys->latest_rseg = rseg;
+
+	return(rseg->id);
+}
+
+/********************************************************************
+Starts a new transaction. */
+UNIV_INTERN
+ibool
+trx_start_low(
+/*==========*/
+			/* out: TRUE */
+	trx_t*	trx,	/* in: transaction */
+	ulint	rseg_id)/* in: rollback segment id; if ULINT_UNDEFINED
+			is passed, the system chooses the rollback segment
+			automatically in a round-robin fashion */
+{
+	trx_rseg_t*	rseg;
+
+	ut_ad(mutex_own(&kernel_mutex));
+	ut_ad(trx->rseg == NULL);
+
+	if (trx->is_purge) {
+		trx->id = ut_dulint_zero;
+		trx->conc_state = TRX_ACTIVE;
+		trx->start_time = time(NULL);
+
+		return(TRUE);
+	}
+
+	ut_ad(trx->conc_state != TRX_ACTIVE);
+
+	if (rseg_id == ULINT_UNDEFINED) {
+
+		rseg_id = trx_assign_rseg();
+	}
+
+	rseg = trx_sys_get_nth_rseg(trx_sys, rseg_id);
+
+	trx->id = trx_sys_get_new_trx_id();
+
+	/* The initial value for trx->no: ut_dulint_max is used in
+	read_view_open_now: */
+
+	trx->no = ut_dulint_max;
+
+	trx->rseg = rseg;
+
+	trx->conc_state = TRX_ACTIVE;
+	trx->start_time = time(NULL);
+
+	UT_LIST_ADD_FIRST(trx_list, trx_sys->trx_list, trx);
+
+	return(TRUE);
+}
+
+/********************************************************************
+Starts a new transaction. */
+UNIV_INTERN
+ibool
+trx_start(
+/*======*/
+			/* out: TRUE */
+	trx_t*	trx,	/* in: transaction */
+	ulint	rseg_id)/* in: rollback segment id; if ULINT_UNDEFINED
+			is passed, the system chooses the rollback segment
+			automatically in a round-robin fashion */
+{
+	ibool	ret;
+
+	/* Update the info whether we should skip XA steps that eat CPU time
+	For the duration of the transaction trx->support_xa is not reread
+	from thd so any changes in the value take effect in the next
+	transaction. This is to avoid a scenario where some undo
+	generated by a transaction, has XA stuff, and other undo,
+	generated by the same transaction, doesn't. */
+	trx->support_xa = thd_supports_xa(trx->mysql_thd);
+
+	mutex_enter(&kernel_mutex);
+
+	ret = trx_start_low(trx, rseg_id);
+
+	mutex_exit(&kernel_mutex);
+
+	return(ret);
+}
+
+/********************************************************************
+Commits a transaction. */
+UNIV_INTERN
+void
+trx_commit_off_kernel(
+/*==================*/
+	trx_t*	trx)	/* in: transaction */
+{
+	page_t*		update_hdr_page;
+	ib_uint64_t	lsn		= 0;
+	trx_rseg_t*	rseg;
+	trx_undo_t*	undo;
+	mtr_t		mtr;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	trx->must_flush_log_later = FALSE;
+
+	rseg = trx->rseg;
+
+	if (trx->insert_undo != NULL || trx->update_undo != NULL) {
+
+		mutex_exit(&kernel_mutex);
+
+		mtr_start(&mtr);
+
+		/* Change the undo log segment states from TRX_UNDO_ACTIVE
+		to some other state: these modifications to the file data
+		structure define the transaction as committed in the file
+		based world, at the serialization point of the log sequence
+		number lsn obtained below. */
+
+		mutex_enter(&(rseg->mutex));
+
+		if (trx->insert_undo != NULL) {
+			trx_undo_set_state_at_finish(
+				rseg, trx, trx->insert_undo, &mtr);
+		}
+
+		undo = trx->update_undo;
+
+		if (undo) {
+			mutex_enter(&kernel_mutex);
+			trx->no = trx_sys_get_new_trx_no();
+
+			mutex_exit(&kernel_mutex);
+
+			/* It is not necessary to obtain trx->undo_mutex here
+			because only a single OS thread is allowed to do the
+			transaction commit for this transaction. */
+
+			update_hdr_page = trx_undo_set_state_at_finish(
+				rseg, trx, undo, &mtr);
+
+			/* We have to do the cleanup for the update log while
+			holding the rseg mutex because update log headers
+			have to be put to the history list in the order of
+			the trx number. */
+
+			trx_undo_update_cleanup(trx, update_hdr_page, &mtr);
+		}
+
+		mutex_exit(&(rseg->mutex));
+
+		/* Update the latest MySQL binlog name and offset info
+		in trx sys header if MySQL binlogging is on or the database
+		server is a MySQL replication slave */
+
+		if (trx->mysql_log_file_name
+		    && trx->mysql_log_file_name[0] != '\0') {
+			trx_sys_update_mysql_binlog_offset(
+				trx->mysql_log_file_name,
+				trx->mysql_log_offset,
+				TRX_SYS_MYSQL_LOG_INFO, &mtr);
+			trx->mysql_log_file_name = NULL;
+		}
+
+		if (trx->mysql_master_log_file_name[0] != '\0') {
+			/* This database server is a MySQL replication slave */
+			trx_sys_update_mysql_binlog_offset(
+				trx->mysql_relay_log_file_name,
+				trx->mysql_relay_log_pos,
+				TRX_SYS_MYSQL_RELAY_LOG_INFO, &mtr);
+			trx_sys_update_mysql_binlog_offset(
+				trx->mysql_master_log_file_name,
+				trx->mysql_master_log_pos,
+				TRX_SYS_MYSQL_MASTER_LOG_INFO, &mtr);
+			trx->mysql_master_log_file_name = "";
+		}
+
+		/* The following call commits the mini-transaction, making the
+		whole transaction committed in the file-based world, at this
+		log sequence number. The transaction becomes 'durable' when
+		we write the log to disk, but in the logical sense the commit
+		in the file-based data structures (undo logs etc.) happens
+		here.
+
+		NOTE that transaction numbers, which are assigned only to
+		transactions with an update undo log, do not necessarily come
+		in exactly the same order as commit lsn's, if the transactions
+		have different rollback segments. To get exactly the same
+		order we should hold the kernel mutex up to this point,
+		adding to to the contention of the kernel mutex. However, if
+		a transaction T2 is able to see modifications made by
+		a transaction T1, T2 will always get a bigger transaction
+		number and a bigger commit lsn than T1. */
+
+		/*--------------*/
+		mtr_commit(&mtr);
+		/*--------------*/
+		lsn = mtr.end_lsn;
+
+		mutex_enter(&kernel_mutex);
+	}
+
+	ut_ad(trx->conc_state == TRX_ACTIVE
+	      || trx->conc_state == TRX_PREPARED);
+	ut_ad(mutex_own(&kernel_mutex));
+
+	/* The following assignment makes the transaction committed in memory
+	and makes its changes to data visible to other transactions.
+	NOTE that there is a small discrepancy from the strict formal
+	visibility rules here: a human user of the database can see
+	modifications made by another transaction T even before the necessary
+	log segment has been flushed to the disk. If the database happens to
+	crash before the flush, the user has seen modifications from T which
+	will never be a committed transaction. However, any transaction T2
+	which sees the modifications of the committing transaction T, and
+	which also itself makes modifications to the database, will get an lsn
+	larger than the committing transaction T. In the case where the log
+	flush fails, and T never gets committed, also T2 will never get
+	committed. */
+
+	/*--------------------------------------*/
+	trx->conc_state = TRX_COMMITTED_IN_MEMORY;
+	/*--------------------------------------*/
+
+	/* If we release kernel_mutex below and we are still doing
+	recovery i.e.: back ground rollback thread is still active
+	then there is a chance that the rollback thread may see
+	this trx as COMMITTED_IN_MEMORY and goes adhead to clean it
+	up calling trx_cleanup_at_db_startup(). This can happen 
+	in the case we are committing a trx here that is left in
+	PREPARED state during the crash. Note that commit of the
+	rollback of a PREPARED trx happens in the recovery thread
+	while the rollback of other transactions happen in the
+	background thread. To avoid this race we unconditionally
+	unset the is_recovered flag from the trx. */
+
+	trx->is_recovered = FALSE;
+
+	lock_release_off_kernel(trx);
+
+	if (trx->global_read_view) {
+		read_view_close(trx->global_read_view);
+		mem_heap_empty(trx->global_read_view_heap);
+		trx->global_read_view = NULL;
+	}
+
+	trx->read_view = NULL;
+
+	if (lsn) {
+
+		mutex_exit(&kernel_mutex);
+
+		if (trx->insert_undo != NULL) {
+
+			trx_undo_insert_cleanup(trx);
+		}
+
+		/* NOTE that we could possibly make a group commit more
+		efficient here: call os_thread_yield here to allow also other
+		trxs to come to commit! */
+
+		/*-------------------------------------*/
+
+		/* Depending on the my.cnf options, we may now write the log
+		buffer to the log files, making the transaction durable if
+		the OS does not crash. We may also flush the log files to
+		disk, making the transaction durable also at an OS crash or a
+		power outage.
+
+		The idea in InnoDB's group commit is that a group of
+		transactions gather behind a trx doing a physical disk write
+		to log files, and when that physical write has been completed,
+		one of those transactions does a write which commits the whole
+		group. Note that this group commit will only bring benefit if
+		there are > 2 users in the database. Then at least 2 users can
+		gather behind one doing the physical log write to disk.
+
+		If we are calling trx_commit() under MySQL's binlog mutex, we
+		will delay possible log write and flush to a separate function
+		trx_commit_complete_for_mysql(), which is only called when the
+		thread has released the binlog mutex. This is to make the
+		group commit algorithm to work. Otherwise, the MySQL binlog
+		mutex would serialize all commits and prevent a group of
+		transactions from gathering. */
+
+		if (trx->flush_log_later) {
+			/* Do nothing yet */
+			trx->must_flush_log_later = TRUE;
+		} else if (srv_flush_log_at_trx_commit == 0) {
+			/* Do nothing */
+		} else if (srv_flush_log_at_trx_commit == 1) {
+			if (srv_unix_file_flush_method == SRV_UNIX_NOSYNC) {
+				/* Write the log but do not flush it to disk */
+
+				log_write_up_to(lsn, LOG_WAIT_ONE_GROUP,
+						FALSE);
+			} else {
+				/* Write the log to the log files AND flush
+				them to disk */
+
+				log_write_up_to(lsn, LOG_WAIT_ONE_GROUP, TRUE);
+			}
+		} else if (srv_flush_log_at_trx_commit == 2) {
+
+			/* Write the log but do not flush it to disk */
+
+			log_write_up_to(lsn, LOG_WAIT_ONE_GROUP, FALSE);
+		} else {
+			ut_error;
+		}
+
+		trx->commit_lsn = lsn;
+
+		/*-------------------------------------*/
+
+		mutex_enter(&kernel_mutex);
+	}
+
+	/* Free all savepoints */
+	trx_roll_free_all_savepoints(trx);
+
+	trx->conc_state = TRX_NOT_STARTED;
+	trx->rseg = NULL;
+	trx->undo_no = ut_dulint_zero;
+	trx->last_sql_stat_start.least_undo_no = ut_dulint_zero;
+	trx->mysql_query_str = NULL;
+
+	ut_ad(UT_LIST_GET_LEN(trx->wait_thrs) == 0);
+	ut_ad(UT_LIST_GET_LEN(trx->trx_locks) == 0);
+
+	UT_LIST_REMOVE(trx_list, trx_sys->trx_list, trx);
+}
+
+/********************************************************************
+Cleans up a transaction at database startup. The cleanup is needed if
+the transaction already got to the middle of a commit when the database
+crashed, andf we cannot roll it back. */
+UNIV_INTERN
+void
+trx_cleanup_at_db_startup(
+/*======================*/
+	trx_t*	trx)	/* in: transaction */
+{
+	if (trx->insert_undo != NULL) {
+
+		trx_undo_insert_cleanup(trx);
+	}
+
+	trx->conc_state = TRX_NOT_STARTED;
+	trx->rseg = NULL;
+	trx->undo_no = ut_dulint_zero;
+	trx->last_sql_stat_start.least_undo_no = ut_dulint_zero;
+
+	UT_LIST_REMOVE(trx_list, trx_sys->trx_list, trx);
+}
+
+/************************************************************************
+Assigns a read view for a consistent read query. All the consistent reads
+within the same transaction will get the same read view, which is created
+when this function is first called for a new started transaction. */
+UNIV_INTERN
+read_view_t*
+trx_assign_read_view(
+/*=================*/
+			/* out: consistent read view */
+	trx_t*	trx)	/* in: active transaction */
+{
+	ut_ad(trx->conc_state == TRX_ACTIVE);
+
+	if (trx->read_view) {
+		return(trx->read_view);
+	}
+
+	mutex_enter(&kernel_mutex);
+
+	if (!trx->read_view) {
+		trx->read_view = read_view_open_now(
+			trx->id, trx->global_read_view_heap);
+		trx->global_read_view = trx->read_view;
+	}
+
+	mutex_exit(&kernel_mutex);
+
+	return(trx->read_view);
+}
+
+/********************************************************************
+Commits a transaction. NOTE that the kernel mutex is temporarily released. */
+static
+void
+trx_handle_commit_sig_off_kernel(
+/*=============================*/
+	trx_t*		trx,		/* in: transaction */
+	que_thr_t**	next_thr)	/* in/out: next query thread to run;
+					if the value which is passed in is
+					a pointer to a NULL pointer, then the
+					calling function can start running
+					a new query thread */
+{
+	trx_sig_t*	sig;
+	trx_sig_t*	next_sig;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	trx->que_state = TRX_QUE_COMMITTING;
+
+	trx_commit_off_kernel(trx);
+
+	ut_ad(UT_LIST_GET_LEN(trx->wait_thrs) == 0);
+
+	/* Remove all TRX_SIG_COMMIT signals from the signal queue and send
+	reply messages to them */
+
+	sig = UT_LIST_GET_FIRST(trx->signals);
+
+	while (sig != NULL) {
+		next_sig = UT_LIST_GET_NEXT(signals, sig);
+
+		if (sig->type == TRX_SIG_COMMIT) {
+
+			trx_sig_reply(sig, next_thr);
+			trx_sig_remove(trx, sig);
+		}
+
+		sig = next_sig;
+	}
+
+	trx->que_state = TRX_QUE_RUNNING;
+}
+
+/***************************************************************
+The transaction must be in the TRX_QUE_LOCK_WAIT state. Puts it to
+the TRX_QUE_RUNNING state and releases query threads which were
+waiting for a lock in the wait_thrs list. */
+UNIV_INTERN
+void
+trx_end_lock_wait(
+/*==============*/
+	trx_t*	trx)	/* in: transaction */
+{
+	que_thr_t*	thr;
+
+	ut_ad(mutex_own(&kernel_mutex));
+	ut_ad(trx->que_state == TRX_QUE_LOCK_WAIT);
+
+	thr = UT_LIST_GET_FIRST(trx->wait_thrs);
+
+	while (thr != NULL) {
+		que_thr_end_wait_no_next_thr(thr);
+
+		UT_LIST_REMOVE(trx_thrs, trx->wait_thrs, thr);
+
+		thr = UT_LIST_GET_FIRST(trx->wait_thrs);
+	}
+
+	trx->que_state = TRX_QUE_RUNNING;
+}
+
+/***************************************************************
+Moves the query threads in the lock wait list to the SUSPENDED state and puts
+the transaction to the TRX_QUE_RUNNING state. */
+static
+void
+trx_lock_wait_to_suspended(
+/*=======================*/
+	trx_t*	trx)	/* in: transaction in the TRX_QUE_LOCK_WAIT state */
+{
+	que_thr_t*	thr;
+
+	ut_ad(mutex_own(&kernel_mutex));
+	ut_ad(trx->que_state == TRX_QUE_LOCK_WAIT);
+
+	thr = UT_LIST_GET_FIRST(trx->wait_thrs);
+
+	while (thr != NULL) {
+		thr->state = QUE_THR_SUSPENDED;
+
+		UT_LIST_REMOVE(trx_thrs, trx->wait_thrs, thr);
+
+		thr = UT_LIST_GET_FIRST(trx->wait_thrs);
+	}
+
+	trx->que_state = TRX_QUE_RUNNING;
+}
+
+/***************************************************************
+Moves the query threads in the sig reply wait list of trx to the SUSPENDED
+state. */
+static
+void
+trx_sig_reply_wait_to_suspended(
+/*============================*/
+	trx_t*	trx)	/* in: transaction */
+{
+	trx_sig_t*	sig;
+	que_thr_t*	thr;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	sig = UT_LIST_GET_FIRST(trx->reply_signals);
+
+	while (sig != NULL) {
+		thr = sig->receiver;
+
+		ut_ad(thr->state == QUE_THR_SIG_REPLY_WAIT);
+
+		thr->state = QUE_THR_SUSPENDED;
+
+		sig->receiver = NULL;
+
+		UT_LIST_REMOVE(reply_signals, trx->reply_signals, sig);
+
+		sig = UT_LIST_GET_FIRST(trx->reply_signals);
+	}
+}
+
+/*********************************************************************
+Checks the compatibility of a new signal with the other signals in the
+queue. */
+static
+ibool
+trx_sig_is_compatible(
+/*==================*/
+			/* out: TRUE if the signal can be queued */
+	trx_t*	trx,	/* in: trx handle */
+	ulint	type,	/* in: signal type */
+	ulint	sender)	/* in: TRX_SIG_SELF or TRX_SIG_OTHER_SESS */
+{
+	trx_sig_t*	sig;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	if (UT_LIST_GET_LEN(trx->signals) == 0) {
+
+		return(TRUE);
+	}
+
+	if (sender == TRX_SIG_SELF) {
+		if (type == TRX_SIG_ERROR_OCCURRED) {
+
+			return(TRUE);
+
+		} else if (type == TRX_SIG_BREAK_EXECUTION) {
+
+			return(TRUE);
+		} else {
+			return(FALSE);
+		}
+	}
+
+	ut_ad(sender == TRX_SIG_OTHER_SESS);
+
+	sig = UT_LIST_GET_FIRST(trx->signals);
+
+	if (type == TRX_SIG_COMMIT) {
+		while (sig != NULL) {
+
+			if (sig->type == TRX_SIG_TOTAL_ROLLBACK) {
+
+				return(FALSE);
+			}
+
+			sig = UT_LIST_GET_NEXT(signals, sig);
+		}
+
+		return(TRUE);
+
+	} else if (type == TRX_SIG_TOTAL_ROLLBACK) {
+		while (sig != NULL) {
+
+			if (sig->type == TRX_SIG_COMMIT) {
+
+				return(FALSE);
+			}
+
+			sig = UT_LIST_GET_NEXT(signals, sig);
+		}
+
+		return(TRUE);
+
+	} else if (type == TRX_SIG_BREAK_EXECUTION) {
+
+		return(TRUE);
+	} else {
+		ut_error;
+
+		return(FALSE);
+	}
+}
+
+/********************************************************************
+Sends a signal to a trx object. */
+UNIV_INTERN
+void
+trx_sig_send(
+/*=========*/
+	trx_t*		trx,		/* in: trx handle */
+	ulint		type,		/* in: signal type */
+	ulint		sender,		/* in: TRX_SIG_SELF or
+					TRX_SIG_OTHER_SESS */
+	que_thr_t*	receiver_thr,	/* in: query thread which wants the
+					reply, or NULL; if type is
+					TRX_SIG_END_WAIT, this must be NULL */
+	trx_savept_t*	savept,		/* in: possible rollback savepoint, or
+					NULL */
+	que_thr_t**	next_thr)	/* in/out: next query thread to run;
+					if the value which is passed in is
+					a pointer to a NULL pointer, then the
+					calling function can start running
+					a new query thread; if the parameter
+					is NULL, it is ignored */
+{
+	trx_sig_t*	sig;
+	trx_t*		receiver_trx;
+
+	ut_ad(trx);
+	ut_ad(mutex_own(&kernel_mutex));
+
+	if (!trx_sig_is_compatible(trx, type, sender)) {
+		/* The signal is not compatible with the other signals in
+		the queue: die */
+
+		ut_error;
+	}
+
+	/* Queue the signal object */
+
+	if (UT_LIST_GET_LEN(trx->signals) == 0) {
+
+		/* The signal list is empty: the 'sig' slot must be unused
+		(we improve performance a bit by avoiding mem_alloc) */
+		sig = &(trx->sig);
+	} else {
+		/* It might be that the 'sig' slot is unused also in this
+		case, but we choose the easy way of using mem_alloc */
+
+		sig = mem_alloc(sizeof(trx_sig_t));
+	}
+
+	UT_LIST_ADD_LAST(signals, trx->signals, sig);
+
+	sig->type = type;
+	sig->sender = sender;
+	sig->receiver = receiver_thr;
+
+	if (savept) {
+		sig->savept = *savept;
+	}
+
+	if (receiver_thr) {
+		receiver_trx = thr_get_trx(receiver_thr);
+
+		UT_LIST_ADD_LAST(reply_signals, receiver_trx->reply_signals,
+				 sig);
+	}
+
+	if (trx->sess->state == SESS_ERROR) {
+
+		trx_sig_reply_wait_to_suspended(trx);
+	}
+
+	if ((sender != TRX_SIG_SELF) || (type == TRX_SIG_BREAK_EXECUTION)) {
+		ut_error;
+	}
+
+	/* If there were no other signals ahead in the queue, try to start
+	handling of the signal */
+
+	if (UT_LIST_GET_FIRST(trx->signals) == sig) {
+
+		trx_sig_start_handle(trx, next_thr);
+	}
+}
+
+/********************************************************************
+Ends signal handling. If the session is in the error state, and
+trx->graph_before_signal_handling != NULL, then returns control to the error
+handling routine of the graph (currently just returns the control to the
+graph root which then will send an error message to the client). */
+UNIV_INTERN
+void
+trx_end_signal_handling(
+/*====================*/
+	trx_t*	trx)	/* in: trx */
+{
+	ut_ad(mutex_own(&kernel_mutex));
+	ut_ad(trx->handling_signals == TRUE);
+
+	trx->handling_signals = FALSE;
+
+	trx->graph = trx->graph_before_signal_handling;
+
+	if (trx->graph && (trx->sess->state == SESS_ERROR)) {
+
+		que_fork_error_handle(trx, trx->graph);
+	}
+}
+
+/********************************************************************
+Starts handling of a trx signal. */
+UNIV_INTERN
+void
+trx_sig_start_handle(
+/*=================*/
+	trx_t*		trx,		/* in: trx handle */
+	que_thr_t**	next_thr)	/* in/out: next query thread to run;
+					if the value which is passed in is
+					a pointer to a NULL pointer, then the
+					calling function can start running
+					a new query thread; if the parameter
+					is NULL, it is ignored */
+{
+	trx_sig_t*	sig;
+	ulint		type;
+loop:
+	/* We loop in this function body as long as there are queued signals
+	we can process immediately */
+
+	ut_ad(trx);
+	ut_ad(mutex_own(&kernel_mutex));
+
+	if (trx->handling_signals && (UT_LIST_GET_LEN(trx->signals) == 0)) {
+
+		trx_end_signal_handling(trx);
+
+		return;
+	}
+
+	if (trx->conc_state == TRX_NOT_STARTED) {
+
+		trx_start_low(trx, ULINT_UNDEFINED);
+	}
+
+	/* If the trx is in a lock wait state, moves the waiting query threads
+	to the suspended state */
+
+	if (trx->que_state == TRX_QUE_LOCK_WAIT) {
+
+		trx_lock_wait_to_suspended(trx);
+	}
+
+	/* If the session is in the error state and this trx has threads
+	waiting for reply from signals, moves these threads to the suspended
+	state, canceling wait reservations; note that if the transaction has
+	sent a commit or rollback signal to itself, and its session is not in
+	the error state, then nothing is done here. */
+
+	if (trx->sess->state == SESS_ERROR) {
+		trx_sig_reply_wait_to_suspended(trx);
+	}
+
+	/* If there are no running query threads, we can start processing of a
+	signal, otherwise we have to wait until all query threads of this
+	transaction are aware of the arrival of the signal. */
+
+	if (trx->n_active_thrs > 0) {
+
+		return;
+	}
+
+	if (trx->handling_signals == FALSE) {
+		trx->graph_before_signal_handling = trx->graph;
+
+		trx->handling_signals = TRUE;
+	}
+
+	sig = UT_LIST_GET_FIRST(trx->signals);
+	type = sig->type;
+
+	if (type == TRX_SIG_COMMIT) {
+
+		trx_handle_commit_sig_off_kernel(trx, next_thr);
+
+	} else if ((type == TRX_SIG_TOTAL_ROLLBACK)
+		   || (type == TRX_SIG_ROLLBACK_TO_SAVEPT)) {
+
+		trx_rollback(trx, sig, next_thr);
+
+		/* No further signals can be handled until the rollback
+		completes, therefore we return */
+
+		return;
+
+	} else if (type == TRX_SIG_ERROR_OCCURRED) {
+
+		trx_rollback(trx, sig, next_thr);
+
+		/* No further signals can be handled until the rollback
+		completes, therefore we return */
+
+		return;
+
+	} else if (type == TRX_SIG_BREAK_EXECUTION) {
+
+		trx_sig_reply(sig, next_thr);
+		trx_sig_remove(trx, sig);
+	} else {
+		ut_error;
+	}
+
+	goto loop;
+}
+
+/********************************************************************
+Send the reply message when a signal in the queue of the trx has been
+handled. */
+UNIV_INTERN
+void
+trx_sig_reply(
+/*==========*/
+	trx_sig_t*	sig,		/* in: signal */
+	que_thr_t**	next_thr)	/* in/out: next query thread to run;
+					if the value which is passed in is
+					a pointer to a NULL pointer, then the
+					calling function can start running
+					a new query thread */
+{
+	trx_t*	receiver_trx;
+
+	ut_ad(sig);
+	ut_ad(mutex_own(&kernel_mutex));
+
+	if (sig->receiver != NULL) {
+		ut_ad((sig->receiver)->state == QUE_THR_SIG_REPLY_WAIT);
+
+		receiver_trx = thr_get_trx(sig->receiver);
+
+		UT_LIST_REMOVE(reply_signals, receiver_trx->reply_signals,
+			       sig);
+		ut_ad(receiver_trx->sess->state != SESS_ERROR);
+
+		que_thr_end_wait(sig->receiver, next_thr);
+
+		sig->receiver = NULL;
+
+	}
+}
+
+/********************************************************************
+Removes a signal object from the trx signal queue. */
+UNIV_INTERN
+void
+trx_sig_remove(
+/*===========*/
+	trx_t*		trx,	/* in: trx handle */
+	trx_sig_t*	sig)	/* in, own: signal */
+{
+	ut_ad(trx && sig);
+	ut_ad(mutex_own(&kernel_mutex));
+
+	ut_ad(sig->receiver == NULL);
+
+	UT_LIST_REMOVE(signals, trx->signals, sig);
+	sig->type = 0;	/* reset the field to catch possible bugs */
+
+	if (sig != &(trx->sig)) {
+		mem_free(sig);
+	}
+}
+
+/*************************************************************************
+Creates a commit command node struct. */
+UNIV_INTERN
+commit_node_t*
+commit_node_create(
+/*===============*/
+				/* out, own: commit node struct */
+	mem_heap_t*	heap)	/* in: mem heap where created */
+{
+	commit_node_t*	node;
+
+	node = mem_heap_alloc(heap, sizeof(commit_node_t));
+	node->common.type  = QUE_NODE_COMMIT;
+	node->state = COMMIT_NODE_SEND;
+
+	return(node);
+}
+
+/***************************************************************
+Performs an execution step for a commit type node in a query graph. */
+UNIV_INTERN
+que_thr_t*
+trx_commit_step(
+/*============*/
+				/* out: query thread to run next, or NULL */
+	que_thr_t*	thr)	/* in: query thread */
+{
+	commit_node_t*	node;
+	que_thr_t*	next_thr;
+
+	node = thr->run_node;
+
+	ut_ad(que_node_get_type(node) == QUE_NODE_COMMIT);
+
+	if (thr->prev_node == que_node_get_parent(node)) {
+		node->state = COMMIT_NODE_SEND;
+	}
+
+	if (node->state == COMMIT_NODE_SEND) {
+		mutex_enter(&kernel_mutex);
+
+		node->state = COMMIT_NODE_WAIT;
+
+		next_thr = NULL;
+
+		thr->state = QUE_THR_SIG_REPLY_WAIT;
+
+		/* Send the commit signal to the transaction */
+
+		trx_sig_send(thr_get_trx(thr), TRX_SIG_COMMIT, TRX_SIG_SELF,
+			     thr, NULL, &next_thr);
+
+		mutex_exit(&kernel_mutex);
+
+		return(next_thr);
+	}
+
+	ut_ad(node->state == COMMIT_NODE_WAIT);
+
+	node->state = COMMIT_NODE_SEND;
+
+	thr->run_node = que_node_get_parent(node);
+
+	return(thr);
+}
+
+/**************************************************************************
+Does the transaction commit for MySQL. */
+UNIV_INTERN
+ulint
+trx_commit_for_mysql(
+/*=================*/
+			/* out: DB_SUCCESS or error number */
+	trx_t*	trx)	/* in: trx handle */
+{
+	/* Because we do not do the commit by sending an Innobase
+	sig to the transaction, we must here make sure that trx has been
+	started. */
+
+	ut_a(trx);
+
+	trx_start_if_not_started(trx);
+
+	trx->op_info = "committing";
+
+	mutex_enter(&kernel_mutex);
+
+	trx_commit_off_kernel(trx);
+
+	mutex_exit(&kernel_mutex);
+
+	trx->op_info = "";
+
+	return(DB_SUCCESS);
+}
+
+/**************************************************************************
+If required, flushes the log to disk if we called trx_commit_for_mysql()
+with trx->flush_log_later == TRUE. */
+UNIV_INTERN
+ulint
+trx_commit_complete_for_mysql(
+/*==========================*/
+			/* out: 0 or error number */
+	trx_t*	trx)	/* in: trx handle */
+{
+	ib_uint64_t	lsn	= trx->commit_lsn;
+
+	ut_a(trx);
+
+	trx->op_info = "flushing log";
+
+	if (!trx->must_flush_log_later) {
+		/* Do nothing */
+	} else if (srv_flush_log_at_trx_commit == 0) {
+		/* Do nothing */
+	} else if (srv_flush_log_at_trx_commit == 1) {
+		if (srv_unix_file_flush_method == SRV_UNIX_NOSYNC) {
+			/* Write the log but do not flush it to disk */
+
+			log_write_up_to(lsn, LOG_WAIT_ONE_GROUP, FALSE);
+		} else {
+			/* Write the log to the log files AND flush them to
+			disk */
+
+			log_write_up_to(lsn, LOG_WAIT_ONE_GROUP, TRUE);
+		}
+	} else if (srv_flush_log_at_trx_commit == 2) {
+
+		/* Write the log but do not flush it to disk */
+
+		log_write_up_to(lsn, LOG_WAIT_ONE_GROUP, FALSE);
+	} else {
+		ut_error;
+	}
+
+	trx->must_flush_log_later = FALSE;
+
+	trx->op_info = "";
+
+	return(0);
+}
+
+/**************************************************************************
+Marks the latest SQL statement ended. */
+UNIV_INTERN
+void
+trx_mark_sql_stat_end(
+/*==================*/
+	trx_t*	trx)	/* in: trx handle */
+{
+	ut_a(trx);
+
+	if (trx->conc_state == TRX_NOT_STARTED) {
+		trx->undo_no = ut_dulint_zero;
+	}
+
+	trx->last_sql_stat_start.least_undo_no = trx->undo_no;
+}
+
+/**************************************************************************
+Prints info about a transaction to the given file. The caller must own the
+kernel mutex and must have called
+innobase_mysql_prepare_print_arbitrary_thd(), unless he knows that MySQL
+or InnoDB cannot meanwhile change the info printed here. */
+UNIV_INTERN
+void
+trx_print(
+/*======*/
+	FILE*	f,		/* in: output stream */
+	trx_t*	trx,		/* in: transaction */
+	ulint	max_query_len)	/* in: max query length to print, or 0 to
+				   use the default max length */
+{
+	ibool	newline;
+
+	fprintf(f, "TRANSACTION " TRX_ID_FMT, TRX_ID_PREP_PRINTF(trx->id));
+
+	switch (trx->conc_state) {
+	case TRX_NOT_STARTED:
+		fputs(", not started", f);
+		break;
+	case TRX_ACTIVE:
+		fprintf(f, ", ACTIVE %lu sec",
+			(ulong)difftime(time(NULL), trx->start_time));
+		break;
+	case TRX_PREPARED:
+		fprintf(f, ", ACTIVE (PREPARED) %lu sec",
+			(ulong)difftime(time(NULL), trx->start_time));
+		break;
+	case TRX_COMMITTED_IN_MEMORY:
+		fputs(", COMMITTED IN MEMORY", f);
+		break;
+	default:
+		fprintf(f, " state %lu", (ulong) trx->conc_state);
+	}
+
+#ifdef UNIV_LINUX
+	fprintf(f, ", process no %lu", trx->mysql_process_no);
+#endif
+	fprintf(f, ", OS thread id %lu",
+		(ulong) os_thread_pf(trx->mysql_thread_id));
+
+	if (*trx->op_info) {
+		putc(' ', f);
+		fputs(trx->op_info, f);
+	}
+
+	if (trx->is_recovered) {
+		fputs(" recovered trx", f);
+	}
+
+	if (trx->is_purge) {
+		fputs(" purge trx", f);
+	}
+
+	if (trx->declared_to_be_inside_innodb) {
+		fprintf(f, ", thread declared inside InnoDB %lu",
+			(ulong) trx->n_tickets_to_enter_innodb);
+	}
+
+	putc('\n', f);
+
+	if (trx->n_mysql_tables_in_use > 0 || trx->mysql_n_tables_locked > 0) {
+		fprintf(f, "mysql tables in use %lu, locked %lu\n",
+			(ulong) trx->n_mysql_tables_in_use,
+			(ulong) trx->mysql_n_tables_locked);
+	}
+
+	newline = TRUE;
+
+	switch (trx->que_state) {
+	case TRX_QUE_RUNNING:
+		newline = FALSE; break;
+	case TRX_QUE_LOCK_WAIT:
+		fputs("LOCK WAIT ", f); break;
+	case TRX_QUE_ROLLING_BACK:
+		fputs("ROLLING BACK ", f); break;
+	case TRX_QUE_COMMITTING:
+		fputs("COMMITTING ", f); break;
+	default:
+		fprintf(f, "que state %lu ", (ulong) trx->que_state);
+	}
+
+	if (0 < UT_LIST_GET_LEN(trx->trx_locks)
+	    || mem_heap_get_size(trx->lock_heap) > 400) {
+		newline = TRUE;
+
+		fprintf(f, "%lu lock struct(s), heap size %lu,"
+			" %lu row lock(s)",
+			(ulong) UT_LIST_GET_LEN(trx->trx_locks),
+			(ulong) mem_heap_get_size(trx->lock_heap),
+			(ulong) lock_number_of_rows_locked(trx));
+	}
+
+	if (trx->has_search_latch) {
+		newline = TRUE;
+		fputs(", holds adaptive hash latch", f);
+	}
+
+	if (!ut_dulint_is_zero(trx->undo_no)) {
+		newline = TRUE;
+		fprintf(f, ", undo log entries %lu",
+			(ulong) ut_dulint_get_low(trx->undo_no));
+	}
+
+	if (newline) {
+		putc('\n', f);
+	}
+
+	if (trx->mysql_thd != NULL) {
+		innobase_mysql_print_thd(f, trx->mysql_thd, max_query_len);
+	}
+}
+
+/***********************************************************************
+Compares the "weight" (or size) of two transactions. Transactions that
+have edited non-transactional tables are considered heavier than ones
+that have not. */
+UNIV_INTERN
+int
+trx_weight_cmp(
+/*===========*/
+				/* out: <0, 0 or >0; similar to strcmp(3) */
+	const trx_t*	a,	/* in: the first transaction to be compared */
+	const trx_t*	b)	/* in: the second transaction to be compared */
+{
+	ibool	a_notrans_edit;
+	ibool	b_notrans_edit;
+
+	/* If mysql_thd is NULL for a transaction we assume that it has
+	not edited non-transactional tables. */
+
+	a_notrans_edit = a->mysql_thd != NULL
+	    && thd_has_edited_nontrans_tables(a->mysql_thd);
+
+	b_notrans_edit = b->mysql_thd != NULL
+	    && thd_has_edited_nontrans_tables(b->mysql_thd);
+
+	if (a_notrans_edit && !b_notrans_edit) {
+
+		return(1);
+	}
+
+	if (!a_notrans_edit && b_notrans_edit) {
+
+		return(-1);
+	}
+
+	/* Either both had edited non-transactional tables or both had
+	not, we fall back to comparing the number of altered/locked
+	rows. */
+
+#if 0
+	fprintf(stderr,
+		"%s TRX_WEIGHT(a): %lld+%lu, TRX_WEIGHT(b): %lld+%lu\n",
+		__func__,
+		ut_conv_dulint_to_longlong(a->undo_no),
+		UT_LIST_GET_LEN(a->trx_locks),
+		ut_conv_dulint_to_longlong(b->undo_no),
+		UT_LIST_GET_LEN(b->trx_locks));
+#endif
+
+	return(ut_dulint_cmp(TRX_WEIGHT(a), TRX_WEIGHT(b)));
+}
+
+/********************************************************************
+Prepares a transaction. */
+UNIV_INTERN
+void
+trx_prepare_off_kernel(
+/*===================*/
+	trx_t*	trx)	/* in: transaction */
+{
+	page_t*		update_hdr_page;
+	trx_rseg_t*	rseg;
+	ib_uint64_t	lsn		= 0;
+	mtr_t		mtr;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	rseg = trx->rseg;
+
+	if (trx->insert_undo != NULL || trx->update_undo != NULL) {
+
+		mutex_exit(&kernel_mutex);
+
+		mtr_start(&mtr);
+
+		/* Change the undo log segment states from TRX_UNDO_ACTIVE
+		to TRX_UNDO_PREPARED: these modifications to the file data
+		structure define the transaction as prepared in the
+		file-based world, at the serialization point of lsn. */
+
+		mutex_enter(&(rseg->mutex));
+
+		if (trx->insert_undo != NULL) {
+
+			/* It is not necessary to obtain trx->undo_mutex here
+			because only a single OS thread is allowed to do the
+			transaction prepare for this transaction. */
+
+			trx_undo_set_state_at_prepare(trx, trx->insert_undo,
+						      &mtr);
+		}
+
+		if (trx->update_undo) {
+			update_hdr_page = trx_undo_set_state_at_prepare(
+				trx, trx->update_undo, &mtr);
+		}
+
+		mutex_exit(&(rseg->mutex));
+
+		/*--------------*/
+		mtr_commit(&mtr);	/* This mtr commit makes the
+					transaction prepared in the file-based
+					world */
+		/*--------------*/
+		lsn = mtr.end_lsn;
+
+		mutex_enter(&kernel_mutex);
+	}
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	/*--------------------------------------*/
+	trx->conc_state = TRX_PREPARED;
+	/*--------------------------------------*/
+
+	if (lsn) {
+		/* Depending on the my.cnf options, we may now write the log
+		buffer to the log files, making the prepared state of the
+		transaction durable if the OS does not crash. We may also
+		flush the log files to disk, making the prepared state of the
+		transaction durable also at an OS crash or a power outage.
+
+		The idea in InnoDB's group prepare is that a group of
+		transactions gather behind a trx doing a physical disk write
+		to log files, and when that physical write has been completed,
+		one of those transactions does a write which prepares the whole
+		group. Note that this group prepare will only bring benefit if
+		there are > 2 users in the database. Then at least 2 users can
+		gather behind one doing the physical log write to disk.
+
+		TODO: find out if MySQL holds some mutex when calling this.
+		That would spoil our group prepare algorithm. */
+
+		mutex_exit(&kernel_mutex);
+
+		if (srv_flush_log_at_trx_commit == 0) {
+			/* Do nothing */
+		} else if (srv_flush_log_at_trx_commit == 1) {
+			if (srv_unix_file_flush_method == SRV_UNIX_NOSYNC) {
+				/* Write the log but do not flush it to disk */
+
+				log_write_up_to(lsn, LOG_WAIT_ONE_GROUP,
+						FALSE);
+			} else {
+				/* Write the log to the log files AND flush
+				them to disk */
+
+				log_write_up_to(lsn, LOG_WAIT_ONE_GROUP, TRUE);
+			}
+		} else if (srv_flush_log_at_trx_commit == 2) {
+
+			/* Write the log but do not flush it to disk */
+
+			log_write_up_to(lsn, LOG_WAIT_ONE_GROUP, FALSE);
+		} else {
+			ut_error;
+		}
+
+		mutex_enter(&kernel_mutex);
+	}
+}
+
+/**************************************************************************
+Does the transaction prepare for MySQL. */
+UNIV_INTERN
+ulint
+trx_prepare_for_mysql(
+/*==================*/
+			/* out: 0 or error number */
+	trx_t*	trx)	/* in: trx handle */
+{
+	/* Because we do not do the prepare by sending an Innobase
+	sig to the transaction, we must here make sure that trx has been
+	started. */
+
+	ut_a(trx);
+
+	trx->op_info = "preparing";
+
+	trx_start_if_not_started(trx);
+
+	mutex_enter(&kernel_mutex);
+
+	trx_prepare_off_kernel(trx);
+
+	mutex_exit(&kernel_mutex);
+
+	trx->op_info = "";
+
+	return(0);
+}
+
+/**************************************************************************
+This function is used to find number of prepared transactions and
+their transaction objects for a recovery. */
+UNIV_INTERN
+int
+trx_recover_for_mysql(
+/*==================*/
+				/* out: number of prepared transactions
+				stored in xid_list */
+	XID*	xid_list,	/* in/out: prepared transactions */
+	ulint	len)		/* in: number of slots in xid_list */
+{
+	trx_t*	trx;
+	ulint	count = 0;
+
+	ut_ad(xid_list);
+	ut_ad(len);
+
+	/* We should set those transactions which are in the prepared state
+	to the xid_list */
+
+	mutex_enter(&kernel_mutex);
+
+	trx = UT_LIST_GET_FIRST(trx_sys->trx_list);
+
+	while (trx) {
+		if (trx->conc_state == TRX_PREPARED) {
+			xid_list[count] = trx->xid;
+
+			if (count == 0) {
+				ut_print_timestamp(stderr);
+				fprintf(stderr,
+					"  InnoDB: Starting recovery for"
+					" XA transactions...\n");
+			}
+
+			ut_print_timestamp(stderr);
+			fprintf(stderr,
+				"  InnoDB: Transaction " TRX_ID_FMT " in"
+				" prepared state after recovery\n",
+				TRX_ID_PREP_PRINTF(trx->id));
+
+			ut_print_timestamp(stderr);
+			fprintf(stderr,
+				"  InnoDB: Transaction contains changes"
+				" to %lu rows\n",
+				(ulong) ut_conv_dulint_to_longlong(
+					trx->undo_no));
+
+			count++;
+
+			if (count == len) {
+				break;
+			}
+		}
+
+		trx = UT_LIST_GET_NEXT(trx_list, trx);
+	}
+
+	mutex_exit(&kernel_mutex);
+
+	if (count > 0){
+		ut_print_timestamp(stderr);
+		fprintf(stderr,
+			"  InnoDB: %lu transactions in prepared state"
+			" after recovery\n",
+			(ulong) count);
+	}
+
+	return ((int) count);
+}
+
+/***********************************************************************
+This function is used to find one X/Open XA distributed transaction
+which is in the prepared state */
+UNIV_INTERN
+trx_t*
+trx_get_trx_by_xid(
+/*===============*/
+			/* out: trx or NULL */
+	XID*	xid)	/* in: X/Open XA transaction identification */
+{
+	trx_t*	trx;
+
+	if (xid == NULL) {
+
+		return (NULL);
+	}
+
+	mutex_enter(&kernel_mutex);
+
+	trx = UT_LIST_GET_FIRST(trx_sys->trx_list);
+
+	while (trx) {
+		/* Compare two X/Open XA transaction id's: their
+		length should be the same and binary comparison
+		of gtrid_lenght+bqual_length bytes should be
+		the same */
+
+		if (xid->gtrid_length == trx->xid.gtrid_length
+		    && xid->bqual_length == trx->xid.bqual_length
+		    && memcmp(xid->data, trx->xid.data,
+			      xid->gtrid_length + xid->bqual_length) == 0) {
+			break;
+		}
+
+		trx = UT_LIST_GET_NEXT(trx_list, trx);
+	}
+
+	mutex_exit(&kernel_mutex);
+
+	if (trx) {
+		if (trx->conc_state != TRX_PREPARED) {
+
+			return(NULL);
+		}
+
+		return(trx);
+	} else {
+		return(NULL);
+	}
+}
diff --git a/storage/xtradb/trx/trx0undo.c b/storage/xtradb/trx/trx0undo.c
new file mode 100644
index 00000000000..bb5710aeba9
--- /dev/null
+++ b/storage/xtradb/trx/trx0undo.c
@@ -0,0 +1,2021 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Transaction undo log
+
+Created 3/26/1996 Heikki Tuuri
+*******************************************************/
+
+#include "trx0undo.h"
+
+#ifdef UNIV_NONINL
+#include "trx0undo.ic"
+#endif
+
+#include "fsp0fsp.h"
+#include "mach0data.h"
+#include "trx0rseg.h"
+#include "trx0trx.h"
+#include "srv0srv.h"
+#include "trx0rec.h"
+#include "trx0purge.h"
+#include "trx0xa.h"
+
+/* How should the old versions in the history list be managed?
+   ----------------------------------------------------------
+If each transaction is given a whole page for its update undo log, file
+space consumption can be 10 times higher than necessary. Therefore,
+partly filled update undo log pages should be reusable. But then there
+is no way individual pages can be ordered so that the ordering agrees
+with the serialization numbers of the transactions on the pages. Thus,
+the history list must be formed of undo logs, not their header pages as
+it was in the old implementation.
+	However, on a single header page the transactions are placed in
+the order of their serialization numbers. As old versions are purged, we
+may free the page when the last transaction on the page has been purged.
+	A problem is that the purge has to go through the transactions
+in the serialization order. This means that we have to look through all
+rollback segments for the one that has the smallest transaction number
+in its history list.
+	When should we do a purge? A purge is necessary when space is
+running out in any of the rollback segments. Then we may have to purge
+also old version which might be needed by some consistent read. How do
+we trigger the start of a purge? When a transaction writes to an undo log,
+it may notice that the space is running out. When a read view is closed,
+it may make some history superfluous. The server can have an utility which
+periodically checks if it can purge some history.
+	In a parallellized purge we have the problem that a query thread
+can remove a delete marked clustered index record before another query
+thread has processed an earlier version of the record, which cannot then
+be done because the row cannot be constructed from the clustered index
+record. To avoid this problem, we will store in the update and delete mark
+undo record also the columns necessary to construct the secondary index
+entries which are modified.
+	We can latch the stack of versions of a single clustered index record
+by taking a latch on the clustered index page. As long as the latch is held,
+no new versions can be added and no versions removed by undo. But, a purge
+can still remove old versions from the bottom of the stack. */
+
+/* How to protect rollback segments, undo logs, and history lists with
+   -------------------------------------------------------------------
+latches?
+-------
+The contention of the kernel mutex should be minimized. When a transaction
+does its first insert or modify in an index, an undo log is assigned for it.
+Then we must have an x-latch to the rollback segment header.
+	When the transaction does more modifys or rolls back, the undo log is
+protected with undo_mutex in the transaction.
+	When the transaction commits, its insert undo log is either reset and
+cached for a fast reuse, or freed. In these cases we must have an x-latch on
+the rollback segment page. The update undo log is put to the history list. If
+it is not suitable for reuse, its slot in the rollback segment is reset. In
+both cases, an x-latch must be acquired on the rollback segment.
+	The purge operation steps through the history list without modifying
+it until a truncate operation occurs, which can remove undo logs from the end
+of the list and release undo log segments. In stepping through the list,
+s-latches on the undo log pages are enough, but in a truncate, x-latches must
+be obtained on the rollback segment and individual pages. */
+
+/************************************************************************
+Initializes the fields in an undo log segment page. */
+static
+void
+trx_undo_page_init(
+/*===============*/
+	page_t* undo_page,	/* in: undo log segment page */
+	ulint	type,		/* in: undo log segment type */
+	mtr_t*	mtr);		/* in: mtr */
+/************************************************************************
+Creates and initializes an undo log memory object. */
+static
+trx_undo_t*
+trx_undo_mem_create(
+/*================*/
+				/* out, own: the undo log memory object */
+	trx_rseg_t*	rseg,	/* in: rollback segment memory object */
+	ulint		id,	/* in: slot index within rseg */
+	ulint		type,	/* in: type of the log: TRX_UNDO_INSERT or
+				TRX_UNDO_UPDATE */
+	dulint		trx_id,	/* in: id of the trx for which the undo log
+				is created */
+	const XID*	xid,	/* in: X/Open XA transaction identification*/
+	ulint		page_no,/* in: undo log header page number */
+	ulint		offset);/* in: undo log header byte offset on page */
+/*******************************************************************
+Initializes a cached insert undo log header page for new use. NOTE that this
+function has its own log record type MLOG_UNDO_HDR_REUSE. You must NOT change
+the operation of this function! */
+static
+ulint
+trx_undo_insert_header_reuse(
+/*=========================*/
+				/* out: undo log header byte offset on page */
+	page_t*	undo_page,	/* in: insert undo log segment header page,
+				x-latched */
+	dulint	trx_id,		/* in: transaction id */
+	mtr_t*	mtr);		/* in: mtr */
+/**************************************************************************
+If an update undo log can be discarded immediately, this function frees the
+space, resetting the page to the proper state for caching. */
+static
+void
+trx_undo_discard_latest_update_undo(
+/*================================*/
+	page_t*	undo_page,	/* in: header page of an undo log of size 1 */
+	mtr_t*	mtr);		/* in: mtr */
+
+
+/***************************************************************************
+Gets the previous record in an undo log from the previous page. */
+static
+trx_undo_rec_t*
+trx_undo_get_prev_rec_from_prev_page(
+/*=================================*/
+				/* out: undo log record, the page s-latched,
+				NULL if none */
+	trx_undo_rec_t*	rec,	/* in: undo record */
+	ulint		page_no,/* in: undo log header page number */
+	ulint		offset,	/* in: undo log header offset on page */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	ulint	space;
+	ulint	zip_size;
+	ulint	prev_page_no;
+	page_t* prev_page;
+	page_t*	undo_page;
+
+	undo_page = page_align(rec);
+
+	prev_page_no = flst_get_prev_addr(undo_page + TRX_UNDO_PAGE_HDR
+					  + TRX_UNDO_PAGE_NODE, mtr)
+		.page;
+
+	if (prev_page_no == FIL_NULL) {
+
+		return(NULL);
+	}
+
+	space = page_get_space_id(undo_page);
+	zip_size = fil_space_get_zip_size(space);
+
+	prev_page = trx_undo_page_get_s_latched(space, zip_size,
+						prev_page_no, mtr);
+
+	return(trx_undo_page_get_last_rec(prev_page, page_no, offset));
+}
+
+/***************************************************************************
+Gets the previous record in an undo log. */
+UNIV_INTERN
+trx_undo_rec_t*
+trx_undo_get_prev_rec(
+/*==================*/
+				/* out: undo log record, the page s-latched,
+				NULL if none */
+	trx_undo_rec_t*	rec,	/* in: undo record */
+	ulint		page_no,/* in: undo log header page number */
+	ulint		offset,	/* in: undo log header offset on page */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	trx_undo_rec_t*	prev_rec;
+
+	prev_rec = trx_undo_page_get_prev_rec(rec, page_no, offset);
+
+	if (prev_rec) {
+
+		return(prev_rec);
+	}
+
+	/* We have to go to the previous undo log page to look for the
+	previous record */
+
+	return(trx_undo_get_prev_rec_from_prev_page(rec, page_no, offset,
+						    mtr));
+}
+
+/***************************************************************************
+Gets the next record in an undo log from the next page. */
+static
+trx_undo_rec_t*
+trx_undo_get_next_rec_from_next_page(
+/*=================================*/
+			/* out: undo log record, the page latched, NULL if
+			none */
+	ulint	space,	/* in: undo log header space */
+	ulint	zip_size,/* in: compressed page size in bytes
+			or 0 for uncompressed pages */
+	page_t*	undo_page, /* in: undo log page */
+	ulint	page_no,/* in: undo log header page number */
+	ulint	offset,	/* in: undo log header offset on page */
+	ulint	mode,	/* in: latch mode: RW_S_LATCH or RW_X_LATCH */
+	mtr_t*	mtr)	/* in: mtr */
+{
+	trx_ulogf_t*	log_hdr;
+	ulint		next_page_no;
+	page_t*		next_page;
+	ulint		next;
+
+	if (page_no == page_get_page_no(undo_page)) {
+
+		log_hdr = undo_page + offset;
+		next = mach_read_from_2(log_hdr + TRX_UNDO_NEXT_LOG);
+
+		if (next != 0) {
+
+			return(NULL);
+		}
+	}
+
+	next_page_no = flst_get_next_addr(undo_page + TRX_UNDO_PAGE_HDR
+					  + TRX_UNDO_PAGE_NODE, mtr)
+		.page;
+	if (next_page_no == FIL_NULL) {
+
+		return(NULL);
+	}
+
+	if (mode == RW_S_LATCH) {
+		next_page = trx_undo_page_get_s_latched(space, zip_size,
+							next_page_no, mtr);
+	} else {
+		ut_ad(mode == RW_X_LATCH);
+		next_page = trx_undo_page_get(space, zip_size,
+					      next_page_no, mtr);
+	}
+
+	return(trx_undo_page_get_first_rec(next_page, page_no, offset));
+}
+
+/***************************************************************************
+Gets the next record in an undo log. */
+UNIV_INTERN
+trx_undo_rec_t*
+trx_undo_get_next_rec(
+/*==================*/
+				/* out: undo log record, the page s-latched,
+				NULL if none */
+	trx_undo_rec_t*	rec,	/* in: undo record */
+	ulint		page_no,/* in: undo log header page number */
+	ulint		offset,	/* in: undo log header offset on page */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	ulint		space;
+	ulint		zip_size;
+	trx_undo_rec_t*	next_rec;
+
+	next_rec = trx_undo_page_get_next_rec(rec, page_no, offset);
+
+	if (next_rec) {
+		return(next_rec);
+	}
+
+	space = page_get_space_id(page_align(rec));
+	zip_size = fil_space_get_zip_size(space);
+
+	return(trx_undo_get_next_rec_from_next_page(space, zip_size,
+						    page_align(rec),
+						    page_no, offset,
+						    RW_S_LATCH, mtr));
+}
+
+/***************************************************************************
+Gets the first record in an undo log. */
+UNIV_INTERN
+trx_undo_rec_t*
+trx_undo_get_first_rec(
+/*===================*/
+			/* out: undo log record, the page latched, NULL if
+			none */
+	ulint	space,	/* in: undo log header space */
+	ulint	zip_size,/* in: compressed page size in bytes
+			or 0 for uncompressed pages */
+	ulint	page_no,/* in: undo log header page number */
+	ulint	offset,	/* in: undo log header offset on page */
+	ulint	mode,	/* in: latching mode: RW_S_LATCH or RW_X_LATCH */
+	mtr_t*	mtr)	/* in: mtr */
+{
+	page_t*		undo_page;
+	trx_undo_rec_t*	rec;
+
+	if (mode == RW_S_LATCH) {
+		undo_page = trx_undo_page_get_s_latched(space, zip_size,
+							page_no, mtr);
+	} else {
+		undo_page = trx_undo_page_get(space, zip_size, page_no, mtr);
+	}
+
+	rec = trx_undo_page_get_first_rec(undo_page, page_no, offset);
+
+	if (rec) {
+		return(rec);
+	}
+
+	return(trx_undo_get_next_rec_from_next_page(space, zip_size,
+						    undo_page, page_no, offset,
+						    mode, mtr));
+}
+
+/*============== UNDO LOG FILE COPY CREATION AND FREEING ==================*/
+
+/**************************************************************************
+Writes the mtr log entry of an undo log page initialization. */
+UNIV_INLINE
+void
+trx_undo_page_init_log(
+/*===================*/
+	page_t* undo_page,	/* in: undo log page */
+	ulint	type,		/* in: undo log type */
+	mtr_t*	mtr)		/* in: mtr */
+{
+	mlog_write_initial_log_record(undo_page, MLOG_UNDO_INIT, mtr);
+
+	mlog_catenate_ulint_compressed(mtr, type);
+}
+
+/***************************************************************
+Parses the redo log entry of an undo log page initialization. */
+UNIV_INTERN
+byte*
+trx_undo_parse_page_init(
+/*=====================*/
+			/* out: end of log record or NULL */
+	byte*	ptr,	/* in: buffer */
+	byte*	end_ptr,/* in: buffer end */
+	page_t*	page,	/* in: page or NULL */
+	mtr_t*	mtr)	/* in: mtr or NULL */
+{
+	ulint	type;
+
+	ptr = mach_parse_compressed(ptr, end_ptr, &type);
+
+	if (ptr == NULL) {
+
+		return(NULL);
+	}
+
+	if (page) {
+		trx_undo_page_init(page, type, mtr);
+	}
+
+	return(ptr);
+}
+
+/************************************************************************
+Initializes the fields in an undo log segment page. */
+static
+void
+trx_undo_page_init(
+/*===============*/
+	page_t* undo_page,	/* in: undo log segment page */
+	ulint	type,		/* in: undo log segment type */
+	mtr_t*	mtr)		/* in: mtr */
+{
+	trx_upagef_t*	page_hdr;
+
+	page_hdr = undo_page + TRX_UNDO_PAGE_HDR;
+
+	mach_write_to_2(page_hdr + TRX_UNDO_PAGE_TYPE, type);
+
+	mach_write_to_2(page_hdr + TRX_UNDO_PAGE_START,
+			TRX_UNDO_PAGE_HDR + TRX_UNDO_PAGE_HDR_SIZE);
+	mach_write_to_2(page_hdr + TRX_UNDO_PAGE_FREE,
+			TRX_UNDO_PAGE_HDR + TRX_UNDO_PAGE_HDR_SIZE);
+
+	fil_page_set_type(undo_page, FIL_PAGE_UNDO_LOG);
+
+	trx_undo_page_init_log(undo_page, type, mtr);
+}
+
+/*******************************************************************
+Creates a new undo log segment in file. */
+static
+ulint
+trx_undo_seg_create(
+/*================*/
+				/* out: DB_SUCCESS if page creation OK
+				possible error codes are:
+				DB_TOO_MANY_CONCURRENT_TRXS
+				DB_OUT_OF_FILE_SPACE */
+	trx_rseg_t*	rseg __attribute__((unused)),/* in: rollback segment */
+	trx_rsegf_t*	rseg_hdr,/* in: rollback segment header, page
+				x-latched */
+	ulint		type,	/* in: type of the segment: TRX_UNDO_INSERT or
+				TRX_UNDO_UPDATE */
+	ulint*		id,	/* out: slot index within rseg header */
+	page_t**	undo_page,
+				/* out: segment header page x-latched, NULL
+				if there was an error */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	ulint		slot_no;
+	ulint		space;
+	buf_block_t*	block;
+	trx_upagef_t*	page_hdr;
+	trx_usegf_t*	seg_hdr;
+	ulint		n_reserved;
+	ibool		success;
+	ulint		err = DB_SUCCESS;
+
+	ut_ad(mtr && id && rseg_hdr);
+	ut_ad(mutex_own(&(rseg->mutex)));
+
+	/*	fputs(type == TRX_UNDO_INSERT
+	? "Creating insert undo log segment\n"
+	: "Creating update undo log segment\n", stderr); */
+	slot_no = trx_rsegf_undo_find_free(rseg_hdr, mtr);
+
+	if (slot_no == ULINT_UNDEFINED) {
+		ut_print_timestamp(stderr);
+		fprintf(stderr,
+			"  InnoDB: Warning: cannot find a free slot for"
+			" an undo log. Do you have too\n"
+			"InnoDB: many active transactions"
+			" running concurrently?\n");
+
+		return(DB_TOO_MANY_CONCURRENT_TRXS);
+	}
+
+	space = page_get_space_id(page_align(rseg_hdr));
+
+	success = fsp_reserve_free_extents(&n_reserved, space, 2, FSP_UNDO,
+					   mtr);
+	if (!success) {
+
+		return(DB_OUT_OF_FILE_SPACE);
+	}
+
+	/* Allocate a new file segment for the undo log */
+	block = fseg_create_general(space, 0,
+				    TRX_UNDO_SEG_HDR
+				    + TRX_UNDO_FSEG_HEADER, TRUE, mtr);
+
+	fil_space_release_free_extents(space, n_reserved);
+
+	if (block == NULL) {
+		/* No space left */
+
+		return(DB_OUT_OF_FILE_SPACE);
+	}
+
+	buf_block_dbg_add_level(block, SYNC_TRX_UNDO_PAGE);
+
+	*undo_page = buf_block_get_frame(block);
+
+	page_hdr = *undo_page + TRX_UNDO_PAGE_HDR;
+	seg_hdr = *undo_page + TRX_UNDO_SEG_HDR;
+
+	trx_undo_page_init(*undo_page, type, mtr);
+
+	mlog_write_ulint(page_hdr + TRX_UNDO_PAGE_FREE,
+			 TRX_UNDO_SEG_HDR + TRX_UNDO_SEG_HDR_SIZE,
+			 MLOG_2BYTES, mtr);
+
+	mlog_write_ulint(seg_hdr + TRX_UNDO_LAST_LOG, 0, MLOG_2BYTES, mtr);
+
+	flst_init(seg_hdr + TRX_UNDO_PAGE_LIST, mtr);
+
+	flst_add_last(seg_hdr + TRX_UNDO_PAGE_LIST,
+		      page_hdr + TRX_UNDO_PAGE_NODE, mtr);
+
+	trx_rsegf_set_nth_undo(rseg_hdr, slot_no,
+			       page_get_page_no(*undo_page), mtr);
+	*id = slot_no;
+
+	return(err);
+}
+
+/**************************************************************************
+Writes the mtr log entry of an undo log header initialization. */
+UNIV_INLINE
+void
+trx_undo_header_create_log(
+/*=======================*/
+	page_t* undo_page,	/* in: undo log header page */
+	dulint	trx_id,		/* in: transaction id */
+	mtr_t*	mtr)		/* in: mtr */
+{
+	mlog_write_initial_log_record(undo_page, MLOG_UNDO_HDR_CREATE, mtr);
+
+	mlog_catenate_dulint_compressed(mtr, trx_id);
+}
+
+/*******************************************************************
+Creates a new undo log header in file. NOTE that this function has its own
+log record type MLOG_UNDO_HDR_CREATE. You must NOT change the operation of
+this function! */
+static
+ulint
+trx_undo_header_create(
+/*===================*/
+				/* out: header byte offset on page */
+	page_t*	undo_page,	/* in: undo log segment header page,
+				x-latched; it is assumed that there are
+				TRX_UNDO_LOG_XA_HDR_SIZE bytes free space
+				on it */
+	dulint	trx_id,		/* in: transaction id */
+	mtr_t*	mtr)		/* in: mtr */
+{
+	trx_upagef_t*	page_hdr;
+	trx_usegf_t*	seg_hdr;
+	trx_ulogf_t*	log_hdr;
+	trx_ulogf_t*	prev_log_hdr;
+	ulint		prev_log;
+	ulint		free;
+	ulint		new_free;
+
+	ut_ad(mtr && undo_page);
+
+	page_hdr = undo_page + TRX_UNDO_PAGE_HDR;
+	seg_hdr = undo_page + TRX_UNDO_SEG_HDR;
+
+	free = mach_read_from_2(page_hdr + TRX_UNDO_PAGE_FREE);
+
+	log_hdr = undo_page + free;
+
+	new_free = free + TRX_UNDO_LOG_OLD_HDR_SIZE;
+
+	ut_a(free + TRX_UNDO_LOG_XA_HDR_SIZE < UNIV_PAGE_SIZE - 100);
+
+	mach_write_to_2(page_hdr + TRX_UNDO_PAGE_START, new_free);
+
+	mach_write_to_2(page_hdr + TRX_UNDO_PAGE_FREE, new_free);
+
+	mach_write_to_2(seg_hdr + TRX_UNDO_STATE, TRX_UNDO_ACTIVE);
+
+	prev_log = mach_read_from_2(seg_hdr + TRX_UNDO_LAST_LOG);
+
+	if (prev_log != 0) {
+		prev_log_hdr = undo_page + prev_log;
+
+		mach_write_to_2(prev_log_hdr + TRX_UNDO_NEXT_LOG, free);
+	}
+
+	mach_write_to_2(seg_hdr + TRX_UNDO_LAST_LOG, free);
+
+	log_hdr = undo_page + free;
+
+	mach_write_to_2(log_hdr + TRX_UNDO_DEL_MARKS, TRUE);
+
+	mach_write_to_8(log_hdr + TRX_UNDO_TRX_ID, trx_id);
+	mach_write_to_2(log_hdr + TRX_UNDO_LOG_START, new_free);
+
+	mach_write_to_1(log_hdr + TRX_UNDO_XID_EXISTS, FALSE);
+	mach_write_to_1(log_hdr + TRX_UNDO_DICT_TRANS, FALSE);
+
+	mach_write_to_2(log_hdr + TRX_UNDO_NEXT_LOG, 0);
+	mach_write_to_2(log_hdr + TRX_UNDO_PREV_LOG, prev_log);
+
+	/* Write the log record about the header creation */
+	trx_undo_header_create_log(undo_page, trx_id, mtr);
+
+	return(free);
+}
+
+/************************************************************************
+Write X/Open XA Transaction Identification (XID) to undo log header */
+static
+void
+trx_undo_write_xid(
+/*===============*/
+	trx_ulogf_t*	log_hdr,/* in: undo log header */
+	const XID*	xid,	/* in: X/Open XA Transaction Identification */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	mlog_write_ulint(log_hdr + TRX_UNDO_XA_FORMAT,
+			 (ulint)xid->formatID, MLOG_4BYTES, mtr);
+
+	mlog_write_ulint(log_hdr + TRX_UNDO_XA_TRID_LEN,
+			 (ulint)xid->gtrid_length, MLOG_4BYTES, mtr);
+
+	mlog_write_ulint(log_hdr + TRX_UNDO_XA_BQUAL_LEN,
+			 (ulint)xid->bqual_length, MLOG_4BYTES, mtr);
+
+	mlog_write_string(log_hdr + TRX_UNDO_XA_XID, (const byte*) xid->data,
+			  XIDDATASIZE, mtr);
+}
+
+/************************************************************************
+Read X/Open XA Transaction Identification (XID) from undo log header */
+static
+void
+trx_undo_read_xid(
+/*==============*/
+	trx_ulogf_t*	log_hdr,/* in: undo log header */
+	XID*		xid)	/* out: X/Open XA Transaction Identification */
+{
+	xid->formatID = (long)mach_read_from_4(log_hdr + TRX_UNDO_XA_FORMAT);
+
+	xid->gtrid_length
+		= (long) mach_read_from_4(log_hdr + TRX_UNDO_XA_TRID_LEN);
+	xid->bqual_length
+		= (long) mach_read_from_4(log_hdr + TRX_UNDO_XA_BQUAL_LEN);
+
+	memcpy(xid->data, log_hdr + TRX_UNDO_XA_XID, XIDDATASIZE);
+}
+
+/*******************************************************************
+Adds space for the XA XID after an undo log old-style header. */
+static
+void
+trx_undo_header_add_space_for_xid(
+/*==============================*/
+	page_t*		undo_page,/* in: undo log segment header page */
+	trx_ulogf_t*	log_hdr,/* in: undo log header */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	trx_upagef_t*	page_hdr;
+	ulint		free;
+	ulint		new_free;
+
+	page_hdr = undo_page + TRX_UNDO_PAGE_HDR;
+
+	free = mach_read_from_2(page_hdr + TRX_UNDO_PAGE_FREE);
+
+	/* free is now the end offset of the old style undo log header */
+
+	ut_a(free == (ulint)(log_hdr - undo_page) + TRX_UNDO_LOG_OLD_HDR_SIZE);
+
+	new_free = free + (TRX_UNDO_LOG_XA_HDR_SIZE
+			   - TRX_UNDO_LOG_OLD_HDR_SIZE);
+
+	/* Add space for a XID after the header, update the free offset
+	fields on the undo log page and in the undo log header */
+
+	mlog_write_ulint(page_hdr + TRX_UNDO_PAGE_START, new_free,
+			 MLOG_2BYTES, mtr);
+
+	mlog_write_ulint(page_hdr + TRX_UNDO_PAGE_FREE, new_free,
+			 MLOG_2BYTES, mtr);
+
+	mlog_write_ulint(log_hdr + TRX_UNDO_LOG_START, new_free,
+			 MLOG_2BYTES, mtr);
+}
+
+/**************************************************************************
+Writes the mtr log entry of an undo log header reuse. */
+UNIV_INLINE
+void
+trx_undo_insert_header_reuse_log(
+/*=============================*/
+	page_t* undo_page,	/* in: undo log header page */
+	dulint	trx_id,		/* in: transaction id */
+	mtr_t*	mtr)		/* in: mtr */
+{
+	mlog_write_initial_log_record(undo_page, MLOG_UNDO_HDR_REUSE, mtr);
+
+	mlog_catenate_dulint_compressed(mtr, trx_id);
+}
+
+/***************************************************************
+Parses the redo log entry of an undo log page header create or reuse. */
+UNIV_INTERN
+byte*
+trx_undo_parse_page_header(
+/*=======================*/
+			/* out: end of log record or NULL */
+	ulint	type,	/* in: MLOG_UNDO_HDR_CREATE or MLOG_UNDO_HDR_REUSE */
+	byte*	ptr,	/* in: buffer */
+	byte*	end_ptr,/* in: buffer end */
+	page_t*	page,	/* in: page or NULL */
+	mtr_t*	mtr)	/* in: mtr or NULL */
+{
+	dulint	trx_id;
+
+	ptr = mach_dulint_parse_compressed(ptr, end_ptr, &trx_id);
+
+	if (ptr == NULL) {
+
+		return(NULL);
+	}
+
+	if (page) {
+		if (type == MLOG_UNDO_HDR_CREATE) {
+			trx_undo_header_create(page, trx_id, mtr);
+		} else {
+			ut_ad(type == MLOG_UNDO_HDR_REUSE);
+			trx_undo_insert_header_reuse(page, trx_id, mtr);
+		}
+	}
+
+	return(ptr);
+}
+
+/*******************************************************************
+Initializes a cached insert undo log header page for new use. NOTE that this
+function has its own log record type MLOG_UNDO_HDR_REUSE. You must NOT change
+the operation of this function! */
+static
+ulint
+trx_undo_insert_header_reuse(
+/*=========================*/
+				/* out: undo log header byte offset on page */
+	page_t*	undo_page,	/* in: insert undo log segment header page,
+				x-latched */
+	dulint	trx_id,		/* in: transaction id */
+	mtr_t*	mtr)		/* in: mtr */
+{
+	trx_upagef_t*	page_hdr;
+	trx_usegf_t*	seg_hdr;
+	trx_ulogf_t*	log_hdr;
+	ulint		free;
+	ulint		new_free;
+
+	ut_ad(mtr && undo_page);
+
+	page_hdr = undo_page + TRX_UNDO_PAGE_HDR;
+	seg_hdr = undo_page + TRX_UNDO_SEG_HDR;
+
+	free = TRX_UNDO_SEG_HDR + TRX_UNDO_SEG_HDR_SIZE;
+
+	ut_a(free + TRX_UNDO_LOG_XA_HDR_SIZE < UNIV_PAGE_SIZE - 100);
+
+	log_hdr = undo_page + free;
+
+	new_free = free + TRX_UNDO_LOG_OLD_HDR_SIZE;
+
+	/* Insert undo data is not needed after commit: we may free all
+	the space on the page */
+
+	ut_a(mach_read_from_2(undo_page + TRX_UNDO_PAGE_HDR
+			      + TRX_UNDO_PAGE_TYPE)
+	     == TRX_UNDO_INSERT);
+
+	mach_write_to_2(page_hdr + TRX_UNDO_PAGE_START, new_free);
+
+	mach_write_to_2(page_hdr + TRX_UNDO_PAGE_FREE, new_free);
+
+	mach_write_to_2(seg_hdr + TRX_UNDO_STATE, TRX_UNDO_ACTIVE);
+
+	log_hdr = undo_page + free;
+
+	mach_write_to_8(log_hdr + TRX_UNDO_TRX_ID, trx_id);
+	mach_write_to_2(log_hdr + TRX_UNDO_LOG_START, new_free);
+
+	mach_write_to_1(log_hdr + TRX_UNDO_XID_EXISTS, FALSE);
+	mach_write_to_1(log_hdr + TRX_UNDO_DICT_TRANS, FALSE);
+
+	/* Write the log record MLOG_UNDO_HDR_REUSE */
+	trx_undo_insert_header_reuse_log(undo_page, trx_id, mtr);
+
+	return(free);
+}
+
+/**************************************************************************
+Writes the redo log entry of an update undo log header discard. */
+UNIV_INLINE
+void
+trx_undo_discard_latest_log(
+/*========================*/
+	page_t* undo_page,	/* in: undo log header page */
+	mtr_t*	mtr)		/* in: mtr */
+{
+	mlog_write_initial_log_record(undo_page, MLOG_UNDO_HDR_DISCARD, mtr);
+}
+
+/***************************************************************
+Parses the redo log entry of an undo log page header discard. */
+UNIV_INTERN
+byte*
+trx_undo_parse_discard_latest(
+/*==========================*/
+			/* out: end of log record or NULL */
+	byte*	ptr,	/* in: buffer */
+	byte*	end_ptr __attribute__((unused)), /* in: buffer end */
+	page_t*	page,	/* in: page or NULL */
+	mtr_t*	mtr)	/* in: mtr or NULL */
+{
+	ut_ad(end_ptr);
+
+	if (page) {
+		trx_undo_discard_latest_update_undo(page, mtr);
+	}
+
+	return(ptr);
+}
+
+/**************************************************************************
+If an update undo log can be discarded immediately, this function frees the
+space, resetting the page to the proper state for caching. */
+static
+void
+trx_undo_discard_latest_update_undo(
+/*================================*/
+	page_t*	undo_page,	/* in: header page of an undo log of size 1 */
+	mtr_t*	mtr)		/* in: mtr */
+{
+	trx_usegf_t*	seg_hdr;
+	trx_upagef_t*	page_hdr;
+	trx_ulogf_t*	log_hdr;
+	trx_ulogf_t*	prev_log_hdr;
+	ulint		free;
+	ulint		prev_hdr_offset;
+
+	seg_hdr = undo_page + TRX_UNDO_SEG_HDR;
+	page_hdr = undo_page + TRX_UNDO_PAGE_HDR;
+
+	free = mach_read_from_2(seg_hdr + TRX_UNDO_LAST_LOG);
+	log_hdr = undo_page + free;
+
+	prev_hdr_offset = mach_read_from_2(log_hdr + TRX_UNDO_PREV_LOG);
+
+	if (prev_hdr_offset != 0) {
+		prev_log_hdr = undo_page + prev_hdr_offset;
+
+		mach_write_to_2(page_hdr + TRX_UNDO_PAGE_START,
+				mach_read_from_2(prev_log_hdr
+						 + TRX_UNDO_LOG_START));
+		mach_write_to_2(prev_log_hdr + TRX_UNDO_NEXT_LOG, 0);
+	}
+
+	mach_write_to_2(page_hdr + TRX_UNDO_PAGE_FREE, free);
+
+	mach_write_to_2(seg_hdr + TRX_UNDO_STATE, TRX_UNDO_CACHED);
+	mach_write_to_2(seg_hdr + TRX_UNDO_LAST_LOG, prev_hdr_offset);
+
+	trx_undo_discard_latest_log(undo_page, mtr);
+}
+
+/************************************************************************
+Tries to add a page to the undo log segment where the undo log is placed. */
+UNIV_INTERN
+ulint
+trx_undo_add_page(
+/*==============*/
+				/* out: page number if success, else
+				FIL_NULL */
+	trx_t*		trx,	/* in: transaction */
+	trx_undo_t*	undo,	/* in: undo log memory object */
+	mtr_t*		mtr)	/* in: mtr which does not have a latch to any
+				undo log page; the caller must have reserved
+				the rollback segment mutex */
+{
+	page_t*		header_page;
+	page_t*		new_page;
+	trx_rseg_t*	rseg;
+	ulint		page_no;
+	ulint		n_reserved;
+	ibool		success;
+
+	ut_ad(mutex_own(&(trx->undo_mutex)));
+	ut_ad(!mutex_own(&kernel_mutex));
+	ut_ad(mutex_own(&(trx->rseg->mutex)));
+
+	rseg = trx->rseg;
+
+	if (rseg->curr_size == rseg->max_size) {
+
+		return(FIL_NULL);
+	}
+
+	header_page = trx_undo_page_get(undo->space, undo->zip_size,
+					undo->hdr_page_no, mtr);
+
+	success = fsp_reserve_free_extents(&n_reserved, undo->space, 1,
+					   FSP_UNDO, mtr);
+	if (!success) {
+
+		return(FIL_NULL);
+	}
+
+	page_no = fseg_alloc_free_page_general(header_page + TRX_UNDO_SEG_HDR
+					       + TRX_UNDO_FSEG_HEADER,
+					       undo->top_page_no + 1, FSP_UP,
+					       TRUE, mtr);
+
+	fil_space_release_free_extents(undo->space, n_reserved);
+
+	if (page_no == FIL_NULL) {
+
+		/* No space left */
+
+		return(FIL_NULL);
+	}
+
+	undo->last_page_no = page_no;
+
+	new_page = trx_undo_page_get(undo->space, undo->zip_size,
+				     page_no, mtr);
+
+	trx_undo_page_init(new_page, undo->type, mtr);
+
+	flst_add_last(header_page + TRX_UNDO_SEG_HDR + TRX_UNDO_PAGE_LIST,
+		      new_page + TRX_UNDO_PAGE_HDR + TRX_UNDO_PAGE_NODE, mtr);
+	undo->size++;
+	rseg->curr_size++;
+
+	return(page_no);
+}
+
+/************************************************************************
+Frees an undo log page that is not the header page. */
+static
+ulint
+trx_undo_free_page(
+/*===============*/
+				/* out: last page number in remaining log */
+	trx_rseg_t* rseg,	/* in: rollback segment */
+	ibool	in_history,	/* in: TRUE if the undo log is in the history
+				list */
+	ulint	space,		/* in: space */
+	ulint	hdr_page_no,	/* in: header page number */
+	ulint	page_no,	/* in: page number to free: must not be the
+				header page */
+	mtr_t*	mtr)		/* in: mtr which does not have a latch to any
+				undo log page; the caller must have reserved
+				the rollback segment mutex */
+{
+	page_t*		header_page;
+	page_t*		undo_page;
+	fil_addr_t	last_addr;
+	trx_rsegf_t*	rseg_header;
+	ulint		hist_size;
+	ulint		zip_size;
+
+	ut_a(hdr_page_no != page_no);
+	ut_ad(!mutex_own(&kernel_mutex));
+	ut_ad(mutex_own(&(rseg->mutex)));
+
+	zip_size = rseg->zip_size;
+
+	undo_page = trx_undo_page_get(space, zip_size, page_no, mtr);
+
+	header_page = trx_undo_page_get(space, zip_size, hdr_page_no, mtr);
+
+	flst_remove(header_page + TRX_UNDO_SEG_HDR + TRX_UNDO_PAGE_LIST,
+		    undo_page + TRX_UNDO_PAGE_HDR + TRX_UNDO_PAGE_NODE, mtr);
+
+	fseg_free_page(header_page + TRX_UNDO_SEG_HDR + TRX_UNDO_FSEG_HEADER,
+		       space, page_no, mtr);
+
+	last_addr = flst_get_last(header_page + TRX_UNDO_SEG_HDR
+				  + TRX_UNDO_PAGE_LIST, mtr);
+	rseg->curr_size--;
+
+	if (in_history) {
+		rseg_header = trx_rsegf_get(space, zip_size,
+					    rseg->page_no, mtr);
+
+		hist_size = mtr_read_ulint(rseg_header + TRX_RSEG_HISTORY_SIZE,
+					   MLOG_4BYTES, mtr);
+		ut_ad(hist_size > 0);
+		mlog_write_ulint(rseg_header + TRX_RSEG_HISTORY_SIZE,
+				 hist_size - 1, MLOG_4BYTES, mtr);
+	}
+
+	return(last_addr.page);
+}
+
+/************************************************************************
+Frees an undo log page when there is also the memory object for the undo
+log. */
+static
+void
+trx_undo_free_page_in_rollback(
+/*===========================*/
+	trx_t*		trx __attribute__((unused)), /* in: transaction */
+	trx_undo_t*	undo,	/* in: undo log memory copy */
+	ulint		page_no,/* in: page number to free: must not be the
+				header page */
+	mtr_t*		mtr)	/* in: mtr which does not have a latch to any
+				undo log page; the caller must have reserved
+				the rollback segment mutex */
+{
+	ulint	last_page_no;
+
+	ut_ad(undo->hdr_page_no != page_no);
+	ut_ad(mutex_own(&(trx->undo_mutex)));
+
+	last_page_no = trx_undo_free_page(undo->rseg, FALSE, undo->space,
+					  undo->hdr_page_no, page_no, mtr);
+
+	undo->last_page_no = last_page_no;
+	undo->size--;
+}
+
+/************************************************************************
+Empties an undo log header page of undo records for that undo log. Other
+undo logs may still have records on that page, if it is an update undo log. */
+static
+void
+trx_undo_empty_header_page(
+/*=======================*/
+	ulint	space,		/* in: space */
+	ulint	zip_size,	/* in: compressed page size in bytes
+				or 0 for uncompressed pages */
+	ulint	hdr_page_no,	/* in: header page number */
+	ulint	hdr_offset,	/* in: header offset */
+	mtr_t*	mtr)		/* in: mtr */
+{
+	page_t*		header_page;
+	trx_ulogf_t*	log_hdr;
+	ulint		end;
+
+	header_page = trx_undo_page_get(space, zip_size, hdr_page_no, mtr);
+
+	log_hdr = header_page + hdr_offset;
+
+	end = trx_undo_page_get_end(header_page, hdr_page_no, hdr_offset);
+
+	mlog_write_ulint(log_hdr + TRX_UNDO_LOG_START, end, MLOG_2BYTES, mtr);
+}
+
+/***************************************************************************
+Truncates an undo log from the end. This function is used during a rollback
+to free space from an undo log. */
+UNIV_INTERN
+void
+trx_undo_truncate_end(
+/*==================*/
+	trx_t*		trx,	/* in: transaction whose undo log it is */
+	trx_undo_t*	undo,	/* in: undo log */
+	dulint		limit)	/* in: all undo records with undo number
+				>= this value should be truncated */
+{
+	page_t*		undo_page;
+	ulint		last_page_no;
+	trx_undo_rec_t* rec;
+	trx_undo_rec_t* trunc_here;
+	trx_rseg_t*	rseg;
+	mtr_t		mtr;
+
+	ut_ad(mutex_own(&(trx->undo_mutex)));
+	ut_ad(mutex_own(&(trx->rseg->mutex)));
+
+	rseg = trx->rseg;
+
+	for (;;) {
+		mtr_start(&mtr);
+
+		trunc_here = NULL;
+
+		last_page_no = undo->last_page_no;
+
+		undo_page = trx_undo_page_get(undo->space, undo->zip_size,
+					      last_page_no, &mtr);
+
+		rec = trx_undo_page_get_last_rec(undo_page, undo->hdr_page_no,
+						 undo->hdr_offset);
+		for (;;) {
+			if (rec == NULL) {
+				if (last_page_no == undo->hdr_page_no) {
+
+					goto function_exit;
+				}
+
+				trx_undo_free_page_in_rollback(
+					trx, undo, last_page_no, &mtr);
+				break;
+			}
+
+			if (ut_dulint_cmp(trx_undo_rec_get_undo_no(rec), limit)
+			    >= 0) {
+				/* Truncate at least this record off, maybe
+				more */
+				trunc_here = rec;
+			} else {
+				goto function_exit;
+			}
+
+			rec = trx_undo_page_get_prev_rec(rec,
+							 undo->hdr_page_no,
+							 undo->hdr_offset);
+		}
+
+		mtr_commit(&mtr);
+	}
+
+function_exit:
+	if (trunc_here) {
+		mlog_write_ulint(undo_page + TRX_UNDO_PAGE_HDR
+				 + TRX_UNDO_PAGE_FREE,
+				 trunc_here - undo_page, MLOG_2BYTES, &mtr);
+	}
+
+	mtr_commit(&mtr);
+}
+
+/***************************************************************************
+Truncates an undo log from the start. This function is used during a purge
+operation. */
+UNIV_INTERN
+void
+trx_undo_truncate_start(
+/*====================*/
+	trx_rseg_t* rseg,	/* in: rollback segment */
+	ulint	space,		/* in: space id of the log */
+	ulint	hdr_page_no,	/* in: header page number */
+	ulint	hdr_offset,	/* in: header offset on the page */
+	dulint	limit)		/* in: all undo pages with undo numbers <
+				this value should be truncated; NOTE that
+				the function only frees whole pages; the
+				header page is not freed, but emptied, if
+				all the records there are < limit */
+{
+	page_t*		undo_page;
+	trx_undo_rec_t* rec;
+	trx_undo_rec_t* last_rec;
+	ulint		page_no;
+	mtr_t		mtr;
+
+	ut_ad(mutex_own(&(rseg->mutex)));
+
+	if (ut_dulint_is_zero(limit)) {
+
+		return;
+	}
+loop:
+	mtr_start(&mtr);
+
+	rec = trx_undo_get_first_rec(space, rseg->zip_size,
+				     hdr_page_no, hdr_offset,
+				     RW_X_LATCH, &mtr);
+	if (rec == NULL) {
+		/* Already empty */
+
+		mtr_commit(&mtr);
+
+		return;
+	}
+
+	undo_page = page_align(rec);
+
+	last_rec = trx_undo_page_get_last_rec(undo_page, hdr_page_no,
+					      hdr_offset);
+	if (ut_dulint_cmp(trx_undo_rec_get_undo_no(last_rec), limit) >= 0) {
+
+		mtr_commit(&mtr);
+
+		return;
+	}
+
+	page_no = page_get_page_no(undo_page);
+
+	if (page_no == hdr_page_no) {
+		trx_undo_empty_header_page(space, rseg->zip_size,
+					   hdr_page_no, hdr_offset,
+					   &mtr);
+	} else {
+		trx_undo_free_page(rseg, TRUE, space, hdr_page_no,
+				   page_no, &mtr);
+	}
+
+	mtr_commit(&mtr);
+
+	goto loop;
+}
+
+/**************************************************************************
+Frees an undo log segment which is not in the history list. */
+static
+void
+trx_undo_seg_free(
+/*==============*/
+	trx_undo_t*	undo)	/* in: undo log */
+{
+	trx_rseg_t*	rseg;
+	fseg_header_t*	file_seg;
+	trx_rsegf_t*	rseg_header;
+	trx_usegf_t*	seg_header;
+	ibool		finished;
+	mtr_t		mtr;
+
+	rseg = undo->rseg;
+
+	do {
+
+		mtr_start(&mtr);
+
+		ut_ad(!mutex_own(&kernel_mutex));
+
+		mutex_enter(&(rseg->mutex));
+
+		seg_header = trx_undo_page_get(undo->space, undo->zip_size,
+					       undo->hdr_page_no,
+					       &mtr) + TRX_UNDO_SEG_HDR;
+
+		file_seg = seg_header + TRX_UNDO_FSEG_HEADER;
+
+		finished = fseg_free_step(file_seg, &mtr);
+
+		if (finished) {
+			/* Update the rseg header */
+			rseg_header = trx_rsegf_get(
+				rseg->space, rseg->zip_size, rseg->page_no,
+				&mtr);
+			trx_rsegf_set_nth_undo(rseg_header, undo->id, FIL_NULL,
+					       &mtr);
+		}
+
+		mutex_exit(&(rseg->mutex));
+		mtr_commit(&mtr);
+	} while (!finished);
+}
+
+/*========== UNDO LOG MEMORY COPY INITIALIZATION =====================*/
+
+/************************************************************************
+Creates and initializes an undo log memory object according to the values
+in the header in file, when the database is started. The memory object is
+inserted in the appropriate list of rseg. */
+static
+trx_undo_t*
+trx_undo_mem_create_at_db_start(
+/*============================*/
+				/* out, own: the undo log memory object */
+	trx_rseg_t*	rseg,	/* in: rollback segment memory object */
+	ulint		id,	/* in: slot index within rseg */
+	ulint		page_no,/* in: undo log segment page number */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	page_t*		undo_page;
+	trx_upagef_t*	page_header;
+	trx_usegf_t*	seg_header;
+	trx_ulogf_t*	undo_header;
+	trx_undo_t*	undo;
+	ulint		type;
+	ulint		state;
+	dulint		trx_id;
+	ulint		offset;
+	fil_addr_t	last_addr;
+	page_t*		last_page;
+	trx_undo_rec_t*	rec;
+	XID		xid;
+	ibool		xid_exists = FALSE;
+
+	if (id >= TRX_RSEG_N_SLOTS) {
+		fprintf(stderr,
+			"InnoDB: Error: undo->id is %lu\n", (ulong) id);
+		ut_error;
+	}
+
+	undo_page = trx_undo_page_get(rseg->space, rseg->zip_size,
+				      page_no, mtr);
+
+	page_header = undo_page + TRX_UNDO_PAGE_HDR;
+
+	type = mtr_read_ulint(page_header + TRX_UNDO_PAGE_TYPE, MLOG_2BYTES,
+			      mtr);
+	seg_header = undo_page + TRX_UNDO_SEG_HDR;
+
+	state = mach_read_from_2(seg_header + TRX_UNDO_STATE);
+
+	offset = mach_read_from_2(seg_header + TRX_UNDO_LAST_LOG);
+
+	undo_header = undo_page + offset;
+
+	trx_id = mtr_read_dulint(undo_header + TRX_UNDO_TRX_ID, mtr);
+
+	xid_exists = mtr_read_ulint(undo_header + TRX_UNDO_XID_EXISTS,
+				    MLOG_1BYTE, mtr);
+
+	/* Read X/Open XA transaction identification if it exists, or
+	set it to NULL. */
+
+	memset(&xid, 0, sizeof(xid));
+	xid.formatID = -1;
+
+	if (xid_exists == TRUE) {
+		trx_undo_read_xid(undo_header, &xid);
+	}
+
+	mutex_enter(&(rseg->mutex));
+
+	undo = trx_undo_mem_create(rseg, id, type, trx_id, &xid,
+				   page_no, offset);
+	mutex_exit(&(rseg->mutex));
+
+	undo->dict_operation =	mtr_read_ulint(
+		undo_header + TRX_UNDO_DICT_TRANS, MLOG_1BYTE, mtr);
+
+	undo->table_id = mtr_read_dulint(undo_header + TRX_UNDO_TABLE_ID, mtr);
+	undo->state = state;
+	undo->size = flst_get_len(seg_header + TRX_UNDO_PAGE_LIST, mtr);
+
+	/* If the log segment is being freed, the page list is inconsistent! */
+	if (state == TRX_UNDO_TO_FREE) {
+
+		goto add_to_list;
+	}
+
+	last_addr = flst_get_last(seg_header + TRX_UNDO_PAGE_LIST, mtr);
+
+	undo->last_page_no = last_addr.page;
+	undo->top_page_no = last_addr.page;
+
+	last_page = trx_undo_page_get(rseg->space, rseg->zip_size,
+				      undo->last_page_no, mtr);
+
+	rec = trx_undo_page_get_last_rec(last_page, page_no, offset);
+
+	if (rec == NULL) {
+		undo->empty = TRUE;
+	} else {
+		undo->empty = FALSE;
+		undo->top_offset = rec - last_page;
+		undo->top_undo_no = trx_undo_rec_get_undo_no(rec);
+	}
+add_to_list:
+	if (type == TRX_UNDO_INSERT) {
+		if (state != TRX_UNDO_CACHED) {
+			UT_LIST_ADD_LAST(undo_list, rseg->insert_undo_list,
+					 undo);
+		} else {
+			UT_LIST_ADD_LAST(undo_list, rseg->insert_undo_cached,
+					 undo);
+		}
+	} else {
+		ut_ad(type == TRX_UNDO_UPDATE);
+		if (state != TRX_UNDO_CACHED) {
+			UT_LIST_ADD_LAST(undo_list, rseg->update_undo_list,
+					 undo);
+		} else {
+			UT_LIST_ADD_LAST(undo_list, rseg->update_undo_cached,
+					 undo);
+		}
+	}
+
+	return(undo);
+}
+
+/************************************************************************
+Initializes the undo log lists for a rollback segment memory copy. This
+function is only called when the database is started or a new rollback
+segment is created. */
+UNIV_INTERN
+ulint
+trx_undo_lists_init(
+/*================*/
+				/* out: the combined size of undo log segments
+				in pages */
+	trx_rseg_t*	rseg)	/* in: rollback segment memory object */
+{
+	ulint		page_no;
+	trx_undo_t*	undo;
+	ulint		size	= 0;
+	trx_rsegf_t*	rseg_header;
+	ulint		i;
+	mtr_t		mtr;
+
+	UT_LIST_INIT(rseg->update_undo_list);
+	UT_LIST_INIT(rseg->update_undo_cached);
+	UT_LIST_INIT(rseg->insert_undo_list);
+	UT_LIST_INIT(rseg->insert_undo_cached);
+
+	mtr_start(&mtr);
+
+	rseg_header = trx_rsegf_get_new(rseg->space, rseg->zip_size,
+					rseg->page_no, &mtr);
+
+	if (!srv_extra_undoslots) {
+		/* uses direct call for avoid "Assertion failure" */
+		//page_no = trx_rsegf_get_nth_undo(rseg_header, TRX_RSEG_N_EXTRA_SLOTS - 1, &mtr);
+		page_no = mtr_read_ulint(rseg_header + TRX_RSEG_UNDO_SLOTS
+					 + (TRX_RSEG_N_EXTRA_SLOTS - 1) * TRX_RSEG_SLOT_SIZE,
+					 MLOG_4BYTES, &mtr);
+		if (page_no != 0) {
+			/* check extended slots are not used */
+			for (i = TRX_RSEG_N_SLOTS; i < TRX_RSEG_N_EXTRA_SLOTS; i++) {
+				/* uses direct call for avoid "Assertion failure" */
+				page_no = mtr_read_ulint(rseg_header + TRX_RSEG_UNDO_SLOTS
+							 + i * TRX_RSEG_SLOT_SIZE,
+							 MLOG_4BYTES, &mtr);
+				if (page_no != FIL_NULL) {
+					srv_extra_undoslots = TRUE;
+					fprintf(stderr,
+"InnoDB: Error: innodb_extra_undoslots option is disabled, but it was enabled before.\n"
+"InnoDB: The datafile is not normal for mysqld and disabled innodb_extra_undoslots.\n"
+"InnoDB: Enable innodb_extra_undoslots if it was enabled before, and\n"
+"InnoDB: ### don't use this datafile with other mysqld or ibbackup! ###\n"
+"InnoDB: Cannot continue operation for the safety. Calling exit(1).\n");
+					exit(1);
+				}
+			}
+			fprintf(stderr,
+"InnoDB: Warning: innodb_extra_undoslots option is disabled, but it was  enabled before.\n"
+"InnoDB: But extended undo slots seem not used, so continue operation.\n");
+		}
+	}
+
+	for (i = 0; i < TRX_RSEG_N_SLOTS; i++) {
+		page_no = trx_rsegf_get_nth_undo(rseg_header, i, &mtr);
+
+		/* If it was not initialized when the datafile created,
+		page_no will be 0 for the extended slots after that */
+
+		if (page_no == 0) {
+			page_no = FIL_NULL;
+			trx_rsegf_set_nth_undo(rseg_header, i, page_no, &mtr);
+		}
+
+		/* In forced recovery: try to avoid operations which look
+		at database pages; undo logs are rapidly changing data, and
+		the probability that they are in an inconsistent state is
+		high */
+
+		if (page_no != FIL_NULL
+		    && srv_force_recovery < SRV_FORCE_NO_UNDO_LOG_SCAN) {
+
+			undo = trx_undo_mem_create_at_db_start(rseg, i,
+							       page_no, &mtr);
+			size += undo->size;
+
+			mtr_commit(&mtr);
+
+			mtr_start(&mtr);
+
+			rseg_header = trx_rsegf_get(
+				rseg->space, rseg->zip_size, rseg->page_no,
+				&mtr);
+		}
+	}
+
+	mtr_commit(&mtr);
+
+	return(size);
+}
+
+/************************************************************************
+Creates and initializes an undo log memory object. */
+static
+trx_undo_t*
+trx_undo_mem_create(
+/*================*/
+				/* out, own: the undo log memory object */
+	trx_rseg_t*	rseg,	/* in: rollback segment memory object */
+	ulint		id,	/* in: slot index within rseg */
+	ulint		type,	/* in: type of the log: TRX_UNDO_INSERT or
+				TRX_UNDO_UPDATE */
+	dulint		trx_id,	/* in: id of the trx for which the undo log
+				is created */
+	const XID*	xid,	/* in: X/Open transaction identification */
+	ulint		page_no,/* in: undo log header page number */
+	ulint		offset)	/* in: undo log header byte offset on page */
+{
+	trx_undo_t*	undo;
+
+	ut_ad(mutex_own(&(rseg->mutex)));
+
+	if (id >= TRX_RSEG_N_SLOTS) {
+		fprintf(stderr,
+			"InnoDB: Error: undo->id is %lu\n", (ulong) id);
+		ut_error;
+	}
+
+	undo = mem_alloc(sizeof(trx_undo_t));
+
+	if (undo == NULL) {
+
+		return NULL;
+	}
+
+	undo->id = id;
+	undo->type = type;
+	undo->state = TRX_UNDO_ACTIVE;
+	undo->del_marks = FALSE;
+	undo->trx_id = trx_id;
+	undo->xid = *xid;
+
+	undo->dict_operation = FALSE;
+
+	undo->rseg = rseg;
+
+	undo->space = rseg->space;
+	undo->zip_size = rseg->zip_size;
+	undo->hdr_page_no = page_no;
+	undo->hdr_offset = offset;
+	undo->last_page_no = page_no;
+	undo->size = 1;
+
+	undo->empty = TRUE;
+	undo->top_page_no = page_no;
+	undo->guess_block = NULL;
+
+	return(undo);
+}
+
+/************************************************************************
+Initializes a cached undo log object for new use. */
+static
+void
+trx_undo_mem_init_for_reuse(
+/*========================*/
+	trx_undo_t*	undo,	/* in: undo log to init */
+	dulint		trx_id,	/* in: id of the trx for which the undo log
+				is created */
+	const XID*	xid,	/* in: X/Open XA transaction identification*/
+	ulint		offset)	/* in: undo log header byte offset on page */
+{
+	ut_ad(mutex_own(&((undo->rseg)->mutex)));
+
+	if (UNIV_UNLIKELY(undo->id >= TRX_RSEG_N_SLOTS)) {
+		fprintf(stderr, "InnoDB: Error: undo->id is %lu\n",
+			(ulong) undo->id);
+
+		mem_analyze_corruption(undo);
+		ut_error;
+	}
+
+	undo->state = TRX_UNDO_ACTIVE;
+	undo->del_marks = FALSE;
+	undo->trx_id = trx_id;
+	undo->xid = *xid;
+
+	undo->dict_operation = FALSE;
+
+	undo->hdr_offset = offset;
+	undo->empty = TRUE;
+}
+
+/************************************************************************
+Frees an undo log memory copy. */
+static
+void
+trx_undo_mem_free(
+/*==============*/
+	trx_undo_t*	undo)	/* in: the undo object to be freed */
+{
+	if (undo->id >= TRX_RSEG_N_SLOTS) {
+		fprintf(stderr,
+			"InnoDB: Error: undo->id is %lu\n", (ulong) undo->id);
+		ut_error;
+	}
+
+	mem_free(undo);
+}
+
+/**************************************************************************
+Creates a new undo log. */
+static
+ulint
+trx_undo_create(
+/*============*/
+				/* out: DB_SUCCESS if successful in creating
+				the new undo lob object, possible error
+				codes are: 
+				DB_TOO_MANY_CONCURRENT_TRXS
+				DB_OUT_OF_FILE_SPACE 
+				DB_OUT_OF_MEMORY*/
+	trx_t*		trx,	/* in: transaction */
+	trx_rseg_t*	rseg,	/* in: rollback segment memory copy */
+	ulint		type,	/* in: type of the log: TRX_UNDO_INSERT or
+				TRX_UNDO_UPDATE */
+	dulint		trx_id,	/* in: id of the trx for which the undo log
+				is created */
+	const XID*	xid,	/* in: X/Open transaction identification*/
+	trx_undo_t**	undo,	/* out: the new undo log object, undefined
+				 * if did not succeed */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	trx_rsegf_t*	rseg_header;
+	ulint		page_no;
+	ulint		offset;
+	ulint		id;
+	page_t*		undo_page;
+	ulint		err;
+
+	ut_ad(mutex_own(&(rseg->mutex)));
+
+	if (rseg->curr_size == rseg->max_size) {
+
+		return(DB_OUT_OF_FILE_SPACE);
+	}
+
+	rseg->curr_size++;
+
+	rseg_header = trx_rsegf_get(rseg->space, rseg->zip_size, rseg->page_no,
+				    mtr);
+
+	err = trx_undo_seg_create(rseg, rseg_header, type, &id,
+				  &undo_page, mtr);
+
+	if (err != DB_SUCCESS) {
+		/* Did not succeed */
+
+		rseg->curr_size--;
+
+		return(err);
+	}
+
+	page_no = page_get_page_no(undo_page);
+
+	offset = trx_undo_header_create(undo_page, trx_id, mtr);
+
+	if (trx->support_xa) {
+		trx_undo_header_add_space_for_xid(undo_page,
+						  undo_page + offset, mtr);
+	}
+
+	*undo = trx_undo_mem_create(rseg, id, type, trx_id, xid,
+				   page_no, offset);
+	if (*undo == NULL) {
+
+		err = DB_OUT_OF_MEMORY;
+	}
+
+	return(err);
+}
+
+/*================ UNDO LOG ASSIGNMENT AND CLEANUP =====================*/
+
+/************************************************************************
+Reuses a cached undo log. */
+static
+trx_undo_t*
+trx_undo_reuse_cached(
+/*==================*/
+				/* out: the undo log memory object, NULL if
+				none cached */
+	trx_t*		trx,	/* in: transaction */
+	trx_rseg_t*	rseg,	/* in: rollback segment memory object */
+	ulint		type,	/* in: type of the log: TRX_UNDO_INSERT or
+				TRX_UNDO_UPDATE */
+	dulint		trx_id,	/* in: id of the trx for which the undo log
+				is used */
+	const XID*	xid,	/* in: X/Open XA transaction identification */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	trx_undo_t*	undo;
+	page_t*		undo_page;
+	ulint		offset;
+
+	ut_ad(mutex_own(&(rseg->mutex)));
+
+	if (type == TRX_UNDO_INSERT) {
+
+		undo = UT_LIST_GET_FIRST(rseg->insert_undo_cached);
+		if (undo == NULL) {
+
+			return(NULL);
+		}
+
+		UT_LIST_REMOVE(undo_list, rseg->insert_undo_cached, undo);
+	} else {
+		ut_ad(type == TRX_UNDO_UPDATE);
+
+		undo = UT_LIST_GET_FIRST(rseg->update_undo_cached);
+		if (undo == NULL) {
+
+			return(NULL);
+		}
+
+		UT_LIST_REMOVE(undo_list, rseg->update_undo_cached, undo);
+	}
+
+	ut_ad(undo->size == 1);
+
+	if (undo->id >= TRX_RSEG_N_SLOTS) {
+		fprintf(stderr, "InnoDB: Error: undo->id is %lu\n",
+			(ulong) undo->id);
+		mem_analyze_corruption(undo);
+		ut_error;
+	}
+
+	undo_page = trx_undo_page_get(undo->space, undo->zip_size,
+				      undo->hdr_page_no, mtr);
+
+	if (type == TRX_UNDO_INSERT) {
+		offset = trx_undo_insert_header_reuse(undo_page, trx_id, mtr);
+
+		if (trx->support_xa) {
+			trx_undo_header_add_space_for_xid(
+				undo_page, undo_page + offset, mtr);
+		}
+	} else {
+		ut_a(mach_read_from_2(undo_page + TRX_UNDO_PAGE_HDR
+				      + TRX_UNDO_PAGE_TYPE)
+		     == TRX_UNDO_UPDATE);
+
+		offset = trx_undo_header_create(undo_page, trx_id, mtr);
+
+		if (trx->support_xa) {
+			trx_undo_header_add_space_for_xid(
+				undo_page, undo_page + offset, mtr);
+		}
+	}
+
+	trx_undo_mem_init_for_reuse(undo, trx_id, xid, offset);
+
+	return(undo);
+}
+
+/**************************************************************************
+Marks an undo log header as a header of a data dictionary operation
+transaction. */
+static
+void
+trx_undo_mark_as_dict_operation(
+/*============================*/
+	trx_t*		trx,	/* in: dict op transaction */
+	trx_undo_t*	undo,	/* in: assigned undo log */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	page_t*	hdr_page;
+
+	hdr_page = trx_undo_page_get(undo->space, undo->zip_size,
+				     undo->hdr_page_no, mtr);
+
+	switch (trx_get_dict_operation(trx)) {
+	case TRX_DICT_OP_NONE:
+		ut_error;
+	case TRX_DICT_OP_INDEX:
+		/* Do not discard the table on recovery. */
+		undo->table_id = ut_dulint_zero;
+		break;
+	case TRX_DICT_OP_TABLE:
+		undo->table_id = trx->table_id;
+		break;
+	}
+
+	mlog_write_ulint(hdr_page + undo->hdr_offset
+			 + TRX_UNDO_DICT_TRANS,
+			 TRUE, MLOG_1BYTE, mtr);
+
+	mlog_write_dulint(hdr_page + undo->hdr_offset + TRX_UNDO_TABLE_ID,
+			  undo->table_id, mtr);
+
+	undo->dict_operation = TRUE;
+}
+
+/**************************************************************************
+Assigns an undo log for a transaction. A new undo log is created or a cached
+undo log reused. */
+UNIV_INTERN
+ulint
+trx_undo_assign_undo(
+/*=================*/
+				/* out: DB_SUCCESS if undo log assign
+				successful, possible error codes are:
+				DB_TOO_MANY_CONCURRENT_TRXS
+				DB_OUT_OF_FILE_SPACE DB_OUT_OF_MEMORY*/
+	trx_t*		trx,	/* in: transaction */
+	ulint		type)	/* in: TRX_UNDO_INSERT or TRX_UNDO_UPDATE */
+{
+	trx_rseg_t*	rseg;
+	trx_undo_t*	undo;
+	mtr_t		mtr;
+	ulint		err = DB_SUCCESS;
+
+	ut_ad(trx);
+	ut_ad(trx->rseg);
+
+	rseg = trx->rseg;
+
+	ut_ad(mutex_own(&(trx->undo_mutex)));
+
+	mtr_start(&mtr);
+
+	ut_ad(!mutex_own(&kernel_mutex));
+
+	mutex_enter(&(rseg->mutex));
+
+	undo = trx_undo_reuse_cached(trx, rseg, type, trx->id, &trx->xid,
+				     &mtr);
+	if (undo == NULL) {
+		err = trx_undo_create(trx, rseg, type, trx->id, &trx->xid,
+								&undo, &mtr);
+		if (err != DB_SUCCESS) {
+
+			goto func_exit;
+		}
+	}
+
+	if (type == TRX_UNDO_INSERT) {
+		UT_LIST_ADD_FIRST(undo_list, rseg->insert_undo_list, undo);
+		ut_ad(trx->insert_undo == NULL);
+		trx->insert_undo = undo;
+	} else {
+		UT_LIST_ADD_FIRST(undo_list, rseg->update_undo_list, undo);
+		ut_ad(trx->update_undo == NULL);
+		trx->update_undo = undo;
+	}
+
+	if (trx_get_dict_operation(trx) != TRX_DICT_OP_NONE) {
+		trx_undo_mark_as_dict_operation(trx, undo, &mtr);
+	}
+
+func_exit:
+	mutex_exit(&(rseg->mutex));
+	mtr_commit(&mtr);
+
+	return err;
+}
+
+/**********************************************************************
+Sets the state of the undo log segment at a transaction finish. */
+UNIV_INTERN
+page_t*
+trx_undo_set_state_at_finish(
+/*=========================*/
+				/* out: undo log segment header page,
+				x-latched */
+	trx_rseg_t*	rseg,	/* in: rollback segment memory object */
+	trx_t*		trx __attribute__((unused)), /* in: transaction */
+	trx_undo_t*	undo,	/* in: undo log memory copy */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	trx_usegf_t*	seg_hdr;
+	trx_upagef_t*	page_hdr;
+	page_t*		undo_page;
+	ulint		state;
+
+	ut_ad(trx);
+	ut_ad(undo);
+	ut_ad(mtr);
+	ut_ad(mutex_own(&rseg->mutex));
+
+	if (undo->id >= TRX_RSEG_N_SLOTS) {
+		fprintf(stderr, "InnoDB: Error: undo->id is %lu\n",
+			(ulong) undo->id);
+		mem_analyze_corruption(undo);
+		ut_error;
+	}
+
+	undo_page = trx_undo_page_get(undo->space, undo->zip_size,
+				      undo->hdr_page_no, mtr);
+
+	seg_hdr = undo_page + TRX_UNDO_SEG_HDR;
+	page_hdr = undo_page + TRX_UNDO_PAGE_HDR;
+
+	if (undo->size == 1
+	    && mach_read_from_2(page_hdr + TRX_UNDO_PAGE_FREE)
+	       < TRX_UNDO_PAGE_REUSE_LIMIT) {
+
+		/* This is a heuristic to avoid the problem of all UNDO
+		slots ending up in one of the UNDO lists. Previously if
+		the server crashed with all the slots in one of the lists,
+		transactions that required the slots of a different type
+		would fail for lack of slots. */
+
+		if (UT_LIST_GET_LEN(rseg->update_undo_list) < 500
+		    && UT_LIST_GET_LEN(rseg->insert_undo_list) < 500) {
+
+			state = TRX_UNDO_CACHED;
+		} else {
+			state = TRX_UNDO_TO_FREE;
+		}
+
+	} else if (undo->type == TRX_UNDO_INSERT) {
+
+		state = TRX_UNDO_TO_FREE;
+	} else {
+		state = TRX_UNDO_TO_PURGE;
+	}
+
+	undo->state = state;
+
+	mlog_write_ulint(seg_hdr + TRX_UNDO_STATE, state, MLOG_2BYTES, mtr);
+
+	return(undo_page);
+}
+
+/**********************************************************************
+Sets the state of the undo log segment at a transaction prepare. */
+UNIV_INTERN
+page_t*
+trx_undo_set_state_at_prepare(
+/*==========================*/
+				/* out: undo log segment header page,
+				x-latched */
+	trx_t*		trx,	/* in: transaction */
+	trx_undo_t*	undo,	/* in: undo log memory copy */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	trx_usegf_t*	seg_hdr;
+	trx_upagef_t*	page_hdr;
+	trx_ulogf_t*	undo_header;
+	page_t*		undo_page;
+	ulint		offset;
+
+	ut_ad(trx && undo && mtr);
+
+	if (undo->id >= TRX_RSEG_N_SLOTS) {
+		fprintf(stderr, "InnoDB: Error: undo->id is %lu\n",
+			(ulong) undo->id);
+		mem_analyze_corruption(undo);
+		ut_error;
+	}
+
+	undo_page = trx_undo_page_get(undo->space, undo->zip_size,
+				      undo->hdr_page_no, mtr);
+
+	seg_hdr = undo_page + TRX_UNDO_SEG_HDR;
+	page_hdr = undo_page + TRX_UNDO_PAGE_HDR;
+
+	/*------------------------------*/
+	undo->state = TRX_UNDO_PREPARED;
+	undo->xid   = trx->xid;
+	/*------------------------------*/
+
+	mlog_write_ulint(seg_hdr + TRX_UNDO_STATE, undo->state,
+			 MLOG_2BYTES, mtr);
+
+	offset = mach_read_from_2(seg_hdr + TRX_UNDO_LAST_LOG);
+	undo_header = undo_page + offset;
+
+	mlog_write_ulint(undo_header + TRX_UNDO_XID_EXISTS,
+			 TRUE, MLOG_1BYTE, mtr);
+
+	trx_undo_write_xid(undo_header, &undo->xid, mtr);
+
+	return(undo_page);
+}
+
+/**************************************************************************
+Adds the update undo log header as the first in the history list, and
+frees the memory object, or puts it to the list of cached update undo log
+segments. */
+UNIV_INTERN
+void
+trx_undo_update_cleanup(
+/*====================*/
+	trx_t*	trx,		/* in: trx owning the update undo log */
+	page_t*	undo_page,	/* in: update undo log header page,
+				x-latched */
+	mtr_t*	mtr)		/* in: mtr */
+{
+	trx_rseg_t*	rseg;
+	trx_undo_t*	undo;
+
+	undo = trx->update_undo;
+	rseg = trx->rseg;
+
+	ut_ad(mutex_own(&(rseg->mutex)));
+
+	trx_purge_add_update_undo_to_history(trx, undo_page, mtr);
+
+	UT_LIST_REMOVE(undo_list, rseg->update_undo_list, undo);
+
+	trx->update_undo = NULL;
+
+	if (undo->state == TRX_UNDO_CACHED) {
+
+		UT_LIST_ADD_FIRST(undo_list, rseg->update_undo_cached, undo);
+	} else {
+		ut_ad(undo->state == TRX_UNDO_TO_PURGE);
+
+		trx_undo_mem_free(undo);
+	}
+}
+
+/**********************************************************************
+Frees or caches an insert undo log after a transaction commit or rollback.
+Knowledge of inserts is not needed after a commit or rollback, therefore
+the data can be discarded. */
+UNIV_INTERN
+void
+trx_undo_insert_cleanup(
+/*====================*/
+	trx_t*	trx)	/* in: transaction handle */
+{
+	trx_undo_t*	undo;
+	trx_rseg_t*	rseg;
+
+	undo = trx->insert_undo;
+	ut_ad(undo);
+
+	rseg = trx->rseg;
+
+	mutex_enter(&(rseg->mutex));
+
+	UT_LIST_REMOVE(undo_list, rseg->insert_undo_list, undo);
+	trx->insert_undo = NULL;
+
+	if (undo->state == TRX_UNDO_CACHED) {
+
+		UT_LIST_ADD_FIRST(undo_list, rseg->insert_undo_cached, undo);
+	} else {
+		ut_ad(undo->state == TRX_UNDO_TO_FREE);
+
+		/* Delete first the undo log segment in the file */
+
+		mutex_exit(&(rseg->mutex));
+
+		trx_undo_seg_free(undo);
+
+		mutex_enter(&(rseg->mutex));
+
+		ut_ad(rseg->curr_size > undo->size);
+
+		rseg->curr_size -= undo->size;
+
+		trx_undo_mem_free(undo);
+	}
+
+	mutex_exit(&(rseg->mutex));
+}
diff --git a/storage/xtradb/usr/usr0sess.c b/storage/xtradb/usr/usr0sess.c
new file mode 100644
index 00000000000..f45c43869ea
--- /dev/null
+++ b/storage/xtradb/usr/usr0sess.c
@@ -0,0 +1,97 @@
+/*****************************************************************************
+
+Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/******************************************************
+Sessions
+
+Created 6/25/1996 Heikki Tuuri
+*******************************************************/
+
+#include "usr0sess.h"
+
+#ifdef UNIV_NONINL
+#include "usr0sess.ic"
+#endif
+
+#include "trx0trx.h"
+
+/*************************************************************************
+Closes a session, freeing the memory occupied by it. */
+static
+void
+sess_close(
+/*=======*/
+	sess_t*		sess);	/* in, own: session object */
+
+/*************************************************************************
+Opens a session. */
+UNIV_INTERN
+sess_t*
+sess_open(void)
+/*===========*/
+					/* out, own: session object */
+{
+	sess_t*	sess;
+
+	ut_ad(mutex_own(&kernel_mutex));
+
+	sess = mem_alloc(sizeof(sess_t));
+
+	sess->state = SESS_ACTIVE;
+
+	sess->trx = trx_create(sess);
+
+	UT_LIST_INIT(sess->graphs);
+
+	return(sess);
+}
+
+/*************************************************************************
+Closes a session, freeing the memory occupied by it. */
+static
+void
+sess_close(
+/*=======*/
+	sess_t*	sess)	/* in, own: session object */
+{
+	ut_ad(mutex_own(&kernel_mutex));
+	ut_ad(sess->trx == NULL);
+
+	mem_free(sess);
+}
+
+/*************************************************************************
+Closes a session, freeing the memory occupied by it, if it is in a state
+where it should be closed. */
+UNIV_INTERN
+ibool
+sess_try_close(
+/*===========*/
+			/* out: TRUE if closed */
+	sess_t*	sess)	/* in, own: session object */
+{
+	ut_ad(mutex_own(&kernel_mutex));
+
+	if (UT_LIST_GET_LEN(sess->graphs) == 0) {
+		sess_close(sess);
+
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
diff --git a/storage/xtradb/ut/ut0auxconf.c b/storage/xtradb/ut/ut0auxconf.c
new file mode 100644
index 00000000000..fd9433d16f6
--- /dev/null
+++ b/storage/xtradb/ut/ut0auxconf.c
@@ -0,0 +1,13 @@
+#include <pthread.h>
+
+int
+main(int argc, char** argv)
+{
+	pthread_t	x1;
+	pthread_t	x2;
+	pthread_t	x3;
+
+	__sync_bool_compare_and_swap(&x1, x2, x3);
+
+	return(0);
+}
diff --git a/storage/xtradb/ut/ut0byte.c b/storage/xtradb/ut/ut0byte.c
new file mode 100644
index 00000000000..5e11e37d0b6
--- /dev/null
+++ b/storage/xtradb/ut/ut0byte.c
@@ -0,0 +1,50 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/*******************************************************************
+Byte utilities
+
+Created 5/11/1994 Heikki Tuuri
+********************************************************************/
+
+#include "ut0byte.h"
+
+#ifdef UNIV_NONINL
+#include "ut0byte.ic"
+#endif
+
+/* Zero value for a dulint */
+UNIV_INTERN const dulint	ut_dulint_zero	= {0, 0};
+
+/* Maximum value for a dulint */
+UNIV_INTERN const dulint	ut_dulint_max	= {0xFFFFFFFFUL, 0xFFFFFFFFUL};
+
+#ifdef notdefined /* unused code */
+#include "ut0sort.h"
+
+/****************************************************************
+Sort function for dulint arrays. */
+UNIV_INTERN
+void
+ut_dulint_sort(dulint* arr, dulint* aux_arr, ulint low, ulint high)
+/*===============================================================*/
+{
+	UT_SORT_FUNCTION_BODY(ut_dulint_sort, arr, aux_arr, low, high,
+			      ut_dulint_cmp);
+}
+#endif /* notdefined */
diff --git a/storage/xtradb/ut/ut0dbg.c b/storage/xtradb/ut/ut0dbg.c
new file mode 100644
index 00000000000..55dd457a177
--- /dev/null
+++ b/storage/xtradb/ut/ut0dbg.c
@@ -0,0 +1,180 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/*********************************************************************
+Debug utilities for Innobase.
+
+Created 1/30/1994 Heikki Tuuri
+**********************************************************************/
+
+#include "univ.i"
+#include "ut0dbg.h"
+
+#if defined(__GNUC__) && (__GNUC__ > 2)
+#else
+/* This is used to eliminate compiler warnings */
+UNIV_INTERN ulint	ut_dbg_zero	= 0;
+#endif
+
+#if defined(UNIV_SYNC_DEBUG) || !defined(UT_DBG_USE_ABORT)
+/* If this is set to TRUE all threads will stop into the next assertion
+and assert */
+UNIV_INTERN ibool	ut_dbg_stop_threads	= FALSE;
+#endif
+#ifdef __NETWARE__
+/* This is set to TRUE when on NetWare there happens an InnoDB
+assertion failure or other fatal error condition that requires an
+immediate shutdown. */
+UNIV_INTERN ibool panic_shutdown = FALSE;
+#elif !defined(UT_DBG_USE_ABORT)
+/* Null pointer used to generate memory trap */
+UNIV_INTERN ulint*	ut_dbg_null_ptr		= NULL;
+#endif
+
+/*****************************************************************
+Report a failed assertion. */
+UNIV_INTERN
+void
+ut_dbg_assertion_failed(
+/*====================*/
+	const char* expr,	/* in: the failed assertion (optional) */
+	const char* file,	/* in: source file containing the assertion */
+	ulint line)		/* in: line number of the assertion */
+{
+	ut_print_timestamp(stderr);
+	fprintf(stderr,
+		"  InnoDB: Assertion failure in thread %lu"
+		" in file %s line %lu\n",
+		os_thread_pf(os_thread_get_curr_id()), file, line);
+	if (expr) {
+		fprintf(stderr,
+			"InnoDB: Failing assertion: %s\n", expr);
+	}
+
+	fputs("InnoDB: We intentionally generate a memory trap.\n"
+	      "InnoDB: Submit a detailed bug report"
+	      " to http://bugs.mysql.com.\n"
+	      "InnoDB: If you get repeated assertion failures"
+	      " or crashes, even\n"
+	      "InnoDB: immediately after the mysqld startup, there may be\n"
+	      "InnoDB: corruption in the InnoDB tablespace. Please refer to\n"
+	      "InnoDB: http://dev.mysql.com/doc/refman/5.1/en/"
+	      "forcing-recovery.html\n"
+	      "InnoDB: about forcing recovery.\n", stderr);
+#if defined(UNIV_SYNC_DEBUG) || !defined(UT_DBG_USE_ABORT)
+	ut_dbg_stop_threads = TRUE;
+#endif
+}
+
+#ifdef __NETWARE__
+/*****************************************************************
+Shut down MySQL/InnoDB after assertion failure. */
+UNIV_INTERN
+void
+ut_dbg_panic(void)
+/*==============*/
+{
+	if (!panic_shutdown) {
+		panic_shutdown = TRUE;
+		innobase_shutdown_for_mysql();
+	}
+	exit(1);
+}
+#else /* __NETWARE__ */
+# if defined(UNIV_SYNC_DEBUG) || !defined(UT_DBG_USE_ABORT)
+/*****************************************************************
+Stop a thread after assertion failure. */
+UNIV_INTERN
+void
+ut_dbg_stop_thread(
+/*===============*/
+	const char*	file,
+	ulint		line)
+{
+	fprintf(stderr, "InnoDB: Thread %lu stopped in file %s line %lu\n",
+		os_thread_pf(os_thread_get_curr_id()), file, line);
+	os_thread_sleep(1000000000);
+}
+# endif
+#endif /* __NETWARE__ */
+
+#ifdef UNIV_COMPILE_TEST_FUNCS
+
+#include <sys/types.h>
+#include <sys/time.h>
+#include <sys/resource.h>
+
+#include <unistd.h>
+
+#ifndef timersub
+#define timersub(a, b, r)						\
+	do {								\
+		(r)->tv_sec = (a)->tv_sec - (b)->tv_sec;		\
+		(r)->tv_usec = (a)->tv_usec - (b)->tv_usec;		\
+		if ((r)->tv_usec < 0) {					\
+			(r)->tv_sec--;					\
+			(r)->tv_usec += 1000000;			\
+		}							\
+	} while (0)
+#endif /* timersub */
+
+/***********************************************************************
+Resets a speedo (records the current time in it). */
+UNIV_INTERN
+void
+speedo_reset(
+/*=========*/
+	speedo_t*	speedo)	/* out: speedo */
+{
+	gettimeofday(&speedo->tv, NULL);
+
+	getrusage(RUSAGE_SELF, &speedo->ru);
+}
+
+/***********************************************************************
+Shows the time elapsed and usage statistics since the last reset of a
+speedo. */
+UNIV_INTERN
+void
+speedo_show(
+/*========*/
+	const speedo_t*	speedo)	/* in: speedo */
+{
+	struct rusage	ru_now;
+	struct timeval	tv_now;
+	struct timeval	tv_diff;
+
+	getrusage(RUSAGE_SELF, &ru_now);
+
+	gettimeofday(&tv_now, NULL);
+
+#define PRINT_TIMEVAL(prefix, tvp)		\
+	fprintf(stderr, "%s% 5ld.%06ld sec\n",	\
+		prefix, (tvp)->tv_sec, (tvp)->tv_usec)
+
+	timersub(&tv_now, &speedo->tv, &tv_diff);
+	PRINT_TIMEVAL("real", &tv_diff);
+
+	timersub(&ru_now.ru_utime, &speedo->ru.ru_utime, &tv_diff);
+	PRINT_TIMEVAL("user", &tv_diff);
+
+	timersub(&ru_now.ru_stime, &speedo->ru.ru_stime, &tv_diff);
+	PRINT_TIMEVAL("sys ", &tv_diff);
+}
+
+#endif /* UNIV_COMPILE_TEST_FUNCS */
diff --git a/storage/xtradb/ut/ut0list.c b/storage/xtradb/ut/ut0list.c
new file mode 100644
index 00000000000..c6250edb6cd
--- /dev/null
+++ b/storage/xtradb/ut/ut0list.c
@@ -0,0 +1,187 @@
+/*****************************************************************************
+
+Copyright (c) 2006, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+#include "ut0list.h"
+#ifdef UNIV_NONINL
+#include "ut0list.ic"
+#endif
+
+/********************************************************************
+Create a new list. */
+UNIV_INTERN
+ib_list_t*
+ib_list_create(void)
+/*=================*/
+			/* out: list */
+{
+	ib_list_t*	list = mem_alloc(sizeof(ib_list_t));
+
+	list->first = NULL;
+	list->last = NULL;
+	list->is_heap_list = FALSE;
+
+	return(list);
+}
+
+/********************************************************************
+Create a new list using the given heap. ib_list_free MUST NOT BE CALLED for
+lists created with this function. */
+UNIV_INTERN
+ib_list_t*
+ib_list_create_heap(
+/*================*/
+				/* out: list */
+	mem_heap_t*	heap)	/* in: memory heap to use */
+{
+	ib_list_t*	list = mem_heap_alloc(heap, sizeof(ib_list_t));
+
+	list->first = NULL;
+	list->last = NULL;
+	list->is_heap_list = TRUE;
+
+	return(list);
+}
+
+/********************************************************************
+Free a list. */
+UNIV_INTERN
+void
+ib_list_free(
+/*=========*/
+	ib_list_t*	list)	/* in: list */
+{
+	ut_a(!list->is_heap_list);
+
+	/* We don't check that the list is empty because it's entirely valid
+	to e.g. have all the nodes allocated from a single heap that is then
+	freed after the list itself is freed. */
+
+	mem_free(list);
+}
+
+/********************************************************************
+Add the data to the start of the list. */
+UNIV_INTERN
+ib_list_node_t*
+ib_list_add_first(
+/*==============*/
+				/* out: new list node*/
+	ib_list_t*	list,	/* in: list */
+	void*		data,	/* in: data */
+	mem_heap_t*	heap)	/* in: memory heap to use */
+{
+	return(ib_list_add_after(list, ib_list_get_first(list), data, heap));
+}
+
+/********************************************************************
+Add the data to the end of the list. */
+UNIV_INTERN
+ib_list_node_t*
+ib_list_add_last(
+/*=============*/
+				/* out: new list node*/
+	ib_list_t*	list,	/* in: list */
+	void*		data,	/* in: data */
+	mem_heap_t*	heap)	/* in: memory heap to use */
+{
+	return(ib_list_add_after(list, ib_list_get_last(list), data, heap));
+}
+
+/********************************************************************
+Add the data after the indicated node. */
+UNIV_INTERN
+ib_list_node_t*
+ib_list_add_after(
+/*==============*/
+					/* out: new list node*/
+	ib_list_t*	list,		/* in: list */
+	ib_list_node_t*	prev_node,	/* in: node preceding new node (can
+					be NULL) */
+	void*		data,		/* in: data */
+	mem_heap_t*	heap)		/* in: memory heap to use */
+{
+	ib_list_node_t*	node = mem_heap_alloc(heap, sizeof(ib_list_node_t));
+
+	node->data = data;
+
+	if (!list->first) {
+		/* Empty list. */
+
+		ut_a(!prev_node);
+
+		node->prev = NULL;
+		node->next = NULL;
+
+		list->first = node;
+		list->last = node;
+	} else if (!prev_node) {
+		/* Start of list. */
+
+		node->prev = NULL;
+		node->next = list->first;
+
+		list->first->prev = node;
+
+		list->first = node;
+	} else {
+		/* Middle or end of list. */
+
+		node->prev = prev_node;
+		node->next = prev_node->next;
+
+		prev_node->next = node;
+
+		if (node->next) {
+			node->next->prev = node;
+		} else {
+			list->last = node;
+		}
+	}
+
+	return(node);
+}
+
+/********************************************************************
+Remove the node from the list. */
+UNIV_INTERN
+void
+ib_list_remove(
+/*===========*/
+	ib_list_t*	list,	/* in: list */
+	ib_list_node_t*	node)	/* in: node to remove */
+{
+	if (node->prev) {
+		node->prev->next = node->next;
+	} else {
+		/* First item in list. */
+
+		ut_ad(list->first == node);
+
+		list->first = node->next;
+	}
+
+	if (node->next) {
+		node->next->prev = node->prev;
+	} else {
+		/* Last item in list. */
+
+		ut_ad(list->last == node);
+
+		list->last = node->prev;
+	}
+}
diff --git a/storage/xtradb/ut/ut0mem.c b/storage/xtradb/ut/ut0mem.c
new file mode 100644
index 00000000000..c87a6a4b57e
--- /dev/null
+++ b/storage/xtradb/ut/ut0mem.c
@@ -0,0 +1,673 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/************************************************************************
+Memory primitives
+
+Created 5/11/1994 Heikki Tuuri
+*************************************************************************/
+
+#include "ut0mem.h"
+
+#ifdef UNIV_NONINL
+#include "ut0mem.ic"
+#endif
+
+#include "mem0mem.h"
+#include "os0thread.h"
+#include "srv0srv.h"
+
+#include <stdlib.h>
+
+/* This struct is placed first in every allocated memory block */
+typedef struct ut_mem_block_struct ut_mem_block_t;
+
+/* The total amount of memory currently allocated from the operating
+system with os_mem_alloc_large() or malloc().  Does not count malloc()
+if srv_use_sys_malloc is set.  Protected by ut_list_mutex. */
+UNIV_INTERN ulint		ut_total_allocated_memory	= 0;
+
+/* Mutex protecting ut_total_allocated_memory and ut_mem_block_list */
+UNIV_INTERN os_fast_mutex_t	ut_list_mutex;
+
+struct ut_mem_block_struct{
+	UT_LIST_NODE_T(ut_mem_block_t) mem_block_list;
+			/* mem block list node */
+	ulint	size;	/* size of allocated memory */
+	ulint	magic_n;
+};
+
+#define UT_MEM_MAGIC_N	1601650166
+
+/* List of all memory blocks allocated from the operating system
+with malloc.  Protected by ut_list_mutex. */
+static UT_LIST_BASE_NODE_T(ut_mem_block_t)   ut_mem_block_list;
+
+static ibool  ut_mem_block_list_inited = FALSE;
+
+static ulint*	ut_mem_null_ptr	= NULL;
+
+/**************************************************************************
+Initializes the mem block list at database startup. */
+UNIV_INTERN
+void
+ut_mem_init(void)
+/*=============*/
+{
+	ut_a(!ut_mem_block_list_inited);
+	os_fast_mutex_init(&ut_list_mutex);
+	UT_LIST_INIT(ut_mem_block_list);
+	ut_mem_block_list_inited = TRUE;
+}
+
+/**************************************************************************
+Allocates memory. Sets it also to zero if UNIV_SET_MEM_TO_ZERO is
+defined and set_to_zero is TRUE. */
+UNIV_INTERN
+void*
+ut_malloc_low(
+/*==========*/
+				/* out, own: allocated memory */
+	ulint	n,		/* in: number of bytes to allocate */
+	ibool	set_to_zero,	/* in: TRUE if allocated memory should be
+				set to zero if UNIV_SET_MEM_TO_ZERO is
+				defined */
+	ibool	assert_on_error)/* in: if TRUE, we crash mysqld if the
+				memory cannot be allocated */
+{
+	ulint	retry_count;
+	void*	ret;
+
+	if (UNIV_LIKELY(srv_use_sys_malloc)) {
+		ret = malloc(n);
+		ut_a(ret || !assert_on_error);
+
+#ifdef UNIV_SET_MEM_TO_ZERO
+		if (set_to_zero) {
+			memset(ret, '\0', n);
+			UNIV_MEM_ALLOC(ret, n);
+		}
+#endif
+		return(ret);
+	}
+
+	ut_ad((sizeof(ut_mem_block_t) % 8) == 0); /* check alignment ok */
+	ut_a(ut_mem_block_list_inited);
+
+	retry_count = 0;
+retry:
+	os_fast_mutex_lock(&ut_list_mutex);
+
+	ret = malloc(n + sizeof(ut_mem_block_t));
+
+	if (ret == NULL && retry_count < 60) {
+		if (retry_count == 0) {
+			ut_print_timestamp(stderr);
+
+			fprintf(stderr,
+				"  InnoDB: Error: cannot allocate"
+				" %lu bytes of\n"
+				"InnoDB: memory with malloc!"
+				" Total allocated memory\n"
+				"InnoDB: by InnoDB %lu bytes."
+				" Operating system errno: %lu\n"
+				"InnoDB: Check if you should"
+				" increase the swap file or\n"
+				"InnoDB: ulimits of your operating system.\n"
+				"InnoDB: On FreeBSD check you"
+				" have compiled the OS with\n"
+				"InnoDB: a big enough maximum process size.\n"
+				"InnoDB: Note that in most 32-bit"
+				" computers the process\n"
+				"InnoDB: memory space is limited"
+				" to 2 GB or 4 GB.\n"
+				"InnoDB: We keep retrying"
+				" the allocation for 60 seconds...\n",
+				(ulong) n, (ulong) ut_total_allocated_memory,
+#ifdef __WIN__
+				(ulong) GetLastError()
+#else
+				(ulong) errno
+#endif
+				);
+		}
+
+		os_fast_mutex_unlock(&ut_list_mutex);
+
+		/* Sleep for a second and retry the allocation; maybe this is
+		just a temporary shortage of memory */
+
+		os_thread_sleep(1000000);
+
+		retry_count++;
+
+		goto retry;
+	}
+
+	if (ret == NULL) {
+		/* Flush stderr to make more probable that the error
+		message gets in the error file before we generate a seg
+		fault */
+
+		fflush(stderr);
+
+		os_fast_mutex_unlock(&ut_list_mutex);
+
+		/* Make an intentional seg fault so that we get a stack
+		trace */
+		/* Intentional segfault on NetWare causes an abend. Avoid this
+		by graceful exit handling in ut_a(). */
+#if (!defined __NETWARE__)
+		if (assert_on_error) {
+			ut_print_timestamp(stderr);
+
+			fprintf(stderr,
+				"  InnoDB: We now intentionally"
+				" generate a seg fault so that\n"
+				"InnoDB: on Linux we get a stack trace.\n");
+
+			if (*ut_mem_null_ptr) ut_mem_null_ptr = 0;
+		} else {
+			return(NULL);
+		}
+#else
+		ut_a(0);
+#endif
+	}
+
+	if (set_to_zero) {
+#ifdef UNIV_SET_MEM_TO_ZERO
+		memset(ret, '\0', n + sizeof(ut_mem_block_t));
+#endif
+	}
+
+	UNIV_MEM_ALLOC(ret, n + sizeof(ut_mem_block_t));
+
+	((ut_mem_block_t*)ret)->size = n + sizeof(ut_mem_block_t);
+	((ut_mem_block_t*)ret)->magic_n = UT_MEM_MAGIC_N;
+
+	ut_total_allocated_memory += n + sizeof(ut_mem_block_t);
+
+	UT_LIST_ADD_FIRST(mem_block_list, ut_mem_block_list,
+			  ((ut_mem_block_t*)ret));
+	os_fast_mutex_unlock(&ut_list_mutex);
+
+	return((void*)((byte*)ret + sizeof(ut_mem_block_t)));
+}
+
+/**************************************************************************
+Allocates memory. Sets it also to zero if UNIV_SET_MEM_TO_ZERO is
+defined. */
+UNIV_INTERN
+void*
+ut_malloc(
+/*======*/
+			/* out, own: allocated memory */
+	ulint	n)	/* in: number of bytes to allocate */
+{
+	return(ut_malloc_low(n, TRUE, TRUE));
+}
+
+/**************************************************************************
+Tests if malloc of n bytes would succeed. ut_malloc() asserts if memory runs
+out. It cannot be used if we want to return an error message. Prints to
+stderr a message if fails. */
+UNIV_INTERN
+ibool
+ut_test_malloc(
+/*===========*/
+			/* out: TRUE if succeeded */
+	ulint	n)	/* in: try to allocate this many bytes */
+{
+	void*	ret;
+
+	ret = malloc(n);
+
+	if (ret == NULL) {
+		ut_print_timestamp(stderr);
+		fprintf(stderr,
+			"  InnoDB: Error: cannot allocate"
+			" %lu bytes of memory for\n"
+			"InnoDB: a BLOB with malloc! Total allocated memory\n"
+			"InnoDB: by InnoDB %lu bytes."
+			" Operating system errno: %d\n"
+			"InnoDB: Check if you should increase"
+			" the swap file or\n"
+			"InnoDB: ulimits of your operating system.\n"
+			"InnoDB: On FreeBSD check you have"
+			" compiled the OS with\n"
+			"InnoDB: a big enough maximum process size.\n",
+			(ulong) n,
+			(ulong) ut_total_allocated_memory,
+			(int) errno);
+		return(FALSE);
+	}
+
+	free(ret);
+
+	return(TRUE);
+}
+
+/**************************************************************************
+Frees a memory block allocated with ut_malloc. */
+UNIV_INTERN
+void
+ut_free(
+/*====*/
+	void* ptr)  /* in, own: memory block */
+{
+	ut_mem_block_t* block;
+
+	if (UNIV_LIKELY(srv_use_sys_malloc)) {
+		free(ptr);
+		return;
+	}
+
+	block = (ut_mem_block_t*)((byte*)ptr - sizeof(ut_mem_block_t));
+
+	os_fast_mutex_lock(&ut_list_mutex);
+
+	ut_a(block->magic_n == UT_MEM_MAGIC_N);
+	ut_a(ut_total_allocated_memory >= block->size);
+
+	ut_total_allocated_memory -= block->size;
+
+	UT_LIST_REMOVE(mem_block_list, ut_mem_block_list, block);
+	free(block);
+
+	os_fast_mutex_unlock(&ut_list_mutex);
+}
+
+/**************************************************************************
+Implements realloc. This is needed by /pars/lexyy.c. Otherwise, you should not
+use this function because the allocation functions in mem0mem.h are the
+recommended ones in InnoDB.
+
+man realloc in Linux, 2004:
+
+       realloc()  changes the size of the memory block pointed to
+       by ptr to size bytes.  The contents will be  unchanged  to
+       the minimum of the old and new sizes; newly allocated mem­
+       ory will be uninitialized.  If ptr is NULL,  the	 call  is
+       equivalent  to malloc(size); if size is equal to zero, the
+       call is equivalent to free(ptr).	 Unless ptr is	NULL,  it
+       must  have  been	 returned by an earlier call to malloc(),
+       calloc() or realloc().
+
+RETURN VALUE
+       realloc() returns a pointer to the newly allocated memory,
+       which is suitably aligned for any kind of variable and may
+       be different from ptr, or NULL if the  request  fails.  If
+       size  was equal to 0, either NULL or a pointer suitable to
+       be passed to free() is returned.	 If realloc()  fails  the
+       original	 block	is  left  untouched  - it is not freed or
+       moved. */
+UNIV_INTERN
+void*
+ut_realloc(
+/*=======*/
+			/* out, own: pointer to new mem block or NULL */
+	void*	ptr,	/* in: pointer to old block or NULL */
+	ulint	size)	/* in: desired size */
+{
+	ut_mem_block_t* block;
+	ulint		old_size;
+	ulint		min_size;
+	void*		new_ptr;
+
+	if (UNIV_LIKELY(srv_use_sys_malloc)) {
+		return(realloc(ptr, size));
+	}
+
+	if (ptr == NULL) {
+
+		return(ut_malloc(size));
+	}
+
+	if (size == 0) {
+		ut_free(ptr);
+
+		return(NULL);
+	}
+
+	block = (ut_mem_block_t*)((byte*)ptr - sizeof(ut_mem_block_t));
+
+	ut_a(block->magic_n == UT_MEM_MAGIC_N);
+
+	old_size = block->size - sizeof(ut_mem_block_t);
+
+	if (size < old_size) {
+		min_size = size;
+	} else {
+		min_size = old_size;
+	}
+
+	new_ptr = ut_malloc(size);
+
+	if (new_ptr == NULL) {
+
+		return(NULL);
+	}
+
+	/* Copy the old data from ptr */
+	ut_memcpy(new_ptr, ptr, min_size);
+
+	ut_free(ptr);
+
+	return(new_ptr);
+}
+
+/**************************************************************************
+Frees in shutdown all allocated memory not freed yet. */
+UNIV_INTERN
+void
+ut_free_all_mem(void)
+/*=================*/
+{
+	ut_mem_block_t* block;
+
+	ut_a(ut_mem_block_list_inited);
+	ut_mem_block_list_inited = FALSE;
+	os_fast_mutex_free(&ut_list_mutex);
+
+	while ((block = UT_LIST_GET_FIRST(ut_mem_block_list))) {
+
+		ut_a(block->magic_n == UT_MEM_MAGIC_N);
+		ut_a(ut_total_allocated_memory >= block->size);
+
+		ut_total_allocated_memory -= block->size;
+
+		UT_LIST_REMOVE(mem_block_list, ut_mem_block_list, block);
+		free(block);
+	}
+
+	if (ut_total_allocated_memory != 0) {
+		fprintf(stderr,
+			"InnoDB: Warning: after shutdown"
+			" total allocated memory is %lu\n",
+			(ulong) ut_total_allocated_memory);
+	}
+}
+
+/**************************************************************************
+Copies up to size - 1 characters from the NUL-terminated string src to
+dst, NUL-terminating the result. Returns strlen(src), so truncation
+occurred if the return value >= size. */
+UNIV_INTERN
+ulint
+ut_strlcpy(
+/*=======*/
+				/* out: strlen(src) */
+	char*		dst,	/* in: destination buffer */
+	const char*	src,	/* in: source buffer */
+	ulint		size)	/* in: size of destination buffer */
+{
+	ulint	src_size = strlen(src);
+
+	if (size != 0) {
+		ulint	n = ut_min(src_size, size - 1);
+
+		memcpy(dst, src, n);
+		dst[n] = '\0';
+	}
+
+	return(src_size);
+}
+
+/**************************************************************************
+Like ut_strlcpy, but if src doesn't fit in dst completely, copies the last
+(size - 1) bytes of src, not the first. */
+UNIV_INTERN
+ulint
+ut_strlcpy_rev(
+/*===========*/
+				/* out: strlen(src) */
+	char*		dst,	/* in: destination buffer */
+	const char*	src,	/* in: source buffer */
+	ulint		size)	/* in: size of destination buffer */
+{
+	ulint	src_size = strlen(src);
+
+	if (size != 0) {
+		ulint	n = ut_min(src_size, size - 1);
+
+		memcpy(dst, src + src_size - n, n + 1);
+	}
+
+	return(src_size);
+}
+
+/**************************************************************************
+Make a quoted copy of a NUL-terminated string.	Leading and trailing
+quotes will not be included; only embedded quotes will be escaped.
+See also ut_strlenq() and ut_memcpyq(). */
+UNIV_INTERN
+char*
+ut_strcpyq(
+/*=======*/
+				/* out: pointer to end of dest */
+	char*		dest,	/* in: output buffer */
+	char		q,	/* in: the quote character */
+	const char*	src)	/* in: null-terminated string */
+{
+	while (*src) {
+		if ((*dest++ = *src++) == q) {
+			*dest++ = q;
+		}
+	}
+
+	return(dest);
+}
+
+/**************************************************************************
+Make a quoted copy of a fixed-length string.  Leading and trailing
+quotes will not be included; only embedded quotes will be escaped.
+See also ut_strlenq() and ut_strcpyq(). */
+UNIV_INTERN
+char*
+ut_memcpyq(
+/*=======*/
+				/* out: pointer to end of dest */
+	char*		dest,	/* in: output buffer */
+	char		q,	/* in: the quote character */
+	const char*	src,	/* in: string to be quoted */
+	ulint		len)	/* in: length of src */
+{
+	const char*	srcend = src + len;
+
+	while (src < srcend) {
+		if ((*dest++ = *src++) == q) {
+			*dest++ = q;
+		}
+	}
+
+	return(dest);
+}
+
+/**************************************************************************
+Return the number of times s2 occurs in s1. Overlapping instances of s2
+are only counted once. */
+UNIV_INTERN
+ulint
+ut_strcount(
+/*========*/
+				/* out: the number of times s2 occurs in s1 */
+	const char*	s1,	/* in: string to search in */
+	const char*	s2)	/* in: string to search for */
+{
+	ulint	count = 0;
+	ulint	len = strlen(s2);
+
+	if (len == 0) {
+
+		return(0);
+	}
+
+	for (;;) {
+		s1 = strstr(s1, s2);
+
+		if (!s1) {
+
+			break;
+		}
+
+		count++;
+		s1 += len;
+	}
+
+	return(count);
+}
+
+/**************************************************************************
+Replace every occurrence of s1 in str with s2. Overlapping instances of s1
+are only replaced once. */
+UNIV_INTERN
+char*
+ut_strreplace(
+/*==========*/
+				/* out, own: modified string, must be
+				freed with mem_free() */
+	const char*	str,	/* in: string to operate on */
+	const char*	s1,	/* in: string to replace */
+	const char*	s2)	/* in: string to replace s1 with */
+{
+	char*		new_str;
+	char*		ptr;
+	const char*	str_end;
+	ulint		str_len = strlen(str);
+	ulint		s1_len = strlen(s1);
+	ulint		s2_len = strlen(s2);
+	ulint		count = 0;
+	int		len_delta = (int)s2_len - (int)s1_len;
+
+	str_end = str + str_len;
+
+	if (len_delta <= 0) {
+		len_delta = 0;
+	} else {
+		count = ut_strcount(str, s1);
+	}
+
+	new_str = mem_alloc(str_len + count * len_delta + 1);
+	ptr = new_str;
+
+	while (str) {
+		const char*	next = strstr(str, s1);
+
+		if (!next) {
+			next = str_end;
+		}
+
+		memcpy(ptr, str, next - str);
+		ptr += next - str;
+
+		if (next == str_end) {
+
+			break;
+		}
+
+		memcpy(ptr, s2, s2_len);
+		ptr += s2_len;
+
+		str = next + s1_len;
+	}
+
+	*ptr = '\0';
+
+	return(new_str);
+}
+
+#ifdef UNIV_COMPILE_TEST_FUNCS
+
+void
+test_ut_str_sql_format()
+{
+	char	buf[128];
+	ulint	ret;
+
+#define CALL_AND_TEST(str, str_len, buf, buf_size, ret_expected, buf_expected)\
+	do {\
+		ibool	ok = TRUE;\
+		memset(buf, 'x', 10);\
+		buf[10] = '\0';\
+		fprintf(stderr, "TESTING \"%s\", %lu, %lu\n",\
+			str, (ulint) str_len, (ulint) buf_size);\
+		ret = ut_str_sql_format(str, str_len, buf, buf_size);\
+		if (ret != ret_expected) {\
+			fprintf(stderr, "expected ret %lu, got %lu\n",\
+				(ulint) ret_expected, ret);\
+			ok = FALSE;\
+		}\
+		if (strcmp((char*) buf, buf_expected) != 0) {\
+			fprintf(stderr, "expected buf \"%s\", got \"%s\"\n",\
+				buf_expected, buf);\
+			ok = FALSE;\
+		}\
+		if (ok) {\
+			fprintf(stderr, "OK: %lu, \"%s\"\n\n",\
+				(ulint) ret, buf);\
+		} else {\
+			return;\
+		}\
+	} while (0)
+
+	CALL_AND_TEST("abcd", 4, buf, 0, 0, "xxxxxxxxxx");
+
+	CALL_AND_TEST("abcd", 4, buf, 1, 1, "");
+
+	CALL_AND_TEST("abcd", 4, buf, 2, 1, "");
+
+	CALL_AND_TEST("abcd", 0, buf, 3, 3, "''");
+	CALL_AND_TEST("abcd", 1, buf, 3, 1, "");
+	CALL_AND_TEST("abcd", 2, buf, 3, 1, "");
+	CALL_AND_TEST("abcd", 3, buf, 3, 1, "");
+	CALL_AND_TEST("abcd", 4, buf, 3, 1, "");
+
+	CALL_AND_TEST("abcd", 0, buf, 4, 3, "''");
+	CALL_AND_TEST("abcd", 1, buf, 4, 4, "'a'");
+	CALL_AND_TEST("abcd", 2, buf, 4, 4, "'a'");
+	CALL_AND_TEST("abcd", 3, buf, 4, 4, "'a'");
+	CALL_AND_TEST("abcd", 4, buf, 4, 4, "'a'");
+	CALL_AND_TEST("abcde", 5, buf, 4, 4, "'a'");
+	CALL_AND_TEST("'", 1, buf, 4, 3, "''");
+	CALL_AND_TEST("''", 2, buf, 4, 3, "''");
+	CALL_AND_TEST("a'", 2, buf, 4, 4, "'a'");
+	CALL_AND_TEST("'a", 2, buf, 4, 3, "''");
+	CALL_AND_TEST("ab", 2, buf, 4, 4, "'a'");
+
+	CALL_AND_TEST("abcdef", 0, buf, 5, 3, "''");
+	CALL_AND_TEST("abcdef", 1, buf, 5, 4, "'a'");
+	CALL_AND_TEST("abcdef", 2, buf, 5, 5, "'ab'");
+	CALL_AND_TEST("abcdef", 3, buf, 5, 5, "'ab'");
+	CALL_AND_TEST("abcdef", 4, buf, 5, 5, "'ab'");
+	CALL_AND_TEST("abcdef", 5, buf, 5, 5, "'ab'");
+	CALL_AND_TEST("abcdef", 6, buf, 5, 5, "'ab'");
+	CALL_AND_TEST("'", 1, buf, 5, 5, "''''");
+	CALL_AND_TEST("''", 2, buf, 5, 5, "''''");
+	CALL_AND_TEST("a'", 2, buf, 5, 4, "'a'");
+	CALL_AND_TEST("'a", 2, buf, 5, 5, "''''");
+	CALL_AND_TEST("ab", 2, buf, 5, 5, "'ab'");
+	CALL_AND_TEST("abc", 3, buf, 5, 5, "'ab'");
+
+	CALL_AND_TEST("ab", 2, buf, 6, 5, "'ab'");
+
+	CALL_AND_TEST("a'b'c", 5, buf, 32, 10, "'a''b''c'");
+	CALL_AND_TEST("a'b'c'", 6, buf, 32, 12, "'a''b''c'''");
+}
+
+#endif /* UNIV_COMPILE_TEST_FUNCS */
diff --git a/storage/xtradb/ut/ut0rnd.c b/storage/xtradb/ut/ut0rnd.c
new file mode 100644
index 00000000000..f5d6cb08b0f
--- /dev/null
+++ b/storage/xtradb/ut/ut0rnd.c
@@ -0,0 +1,94 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/*******************************************************************
+Random numbers and hashing
+
+Created 5/11/1994 Heikki Tuuri
+********************************************************************/
+
+#include "ut0rnd.h"
+
+#ifdef UNIV_NONINL
+#include "ut0rnd.ic"
+#endif
+
+/* These random numbers are used in ut_find_prime */
+#define	UT_RANDOM_1	1.0412321
+#define	UT_RANDOM_2	1.1131347
+#define UT_RANDOM_3	1.0132677
+
+
+UNIV_INTERN ulint	ut_rnd_ulint_counter = 65654363;
+
+/***************************************************************
+Looks for a prime number slightly greater than the given argument.
+The prime is chosen so that it is not near any power of 2. */
+UNIV_INTERN
+ulint
+ut_find_prime(
+/*==========*/
+			/* out: prime */
+	ulint	n)	/* in: positive number > 100 */
+{
+	ulint	pow2;
+	ulint	i;
+
+	n += 100;
+
+	pow2 = 1;
+	while (pow2 * 2 < n) {
+		pow2 = 2 * pow2;
+	}
+
+	if ((double)n < 1.05 * (double)pow2) {
+		n = (ulint) ((double)n * UT_RANDOM_1);
+	}
+
+	pow2 = 2 * pow2;
+
+	if ((double)n > 0.95 * (double)pow2) {
+		n = (ulint) ((double)n * UT_RANDOM_2);
+	}
+
+	if (n > pow2 - 20) {
+		n += 30;
+	}
+
+	/* Now we have n far enough from powers of 2. To make
+	n more random (especially, if it was not near
+	a power of 2), we then multiply it by a random number. */
+
+	n = (ulint) ((double)n * UT_RANDOM_3);
+
+	for (;; n++) {
+		i = 2;
+		while (i * i <= n) {
+			if (n % i == 0) {
+				goto next_n;
+			}
+			i++;
+		}
+
+		/* Found a prime */
+		break;
+next_n:		;
+	}
+
+	return(n);
+}
diff --git a/storage/xtradb/ut/ut0ut.c b/storage/xtradb/ut/ut0ut.c
new file mode 100644
index 00000000000..12500988ec6
--- /dev/null
+++ b/storage/xtradb/ut/ut0ut.c
@@ -0,0 +1,594 @@
+/*****************************************************************************
+
+Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+/*******************************************************************
+Various utilities for Innobase.
+
+Created 5/11/1994 Heikki Tuuri
+********************************************************************/
+
+#include "ut0ut.h"
+
+#ifdef UNIV_NONINL
+#include "ut0ut.ic"
+#endif
+
+#include <stdarg.h>
+#include <string.h>
+#include <ctype.h>
+
+#include "trx0trx.h"
+#include "ha_prototypes.h"
+#ifndef UNIV_HOTBACKUP
+# include "mysql_com.h" /* NAME_LEN */
+#endif /* UNIV_HOTBACKUP */
+
+UNIV_INTERN ibool	ut_always_false	= FALSE;
+
+#ifdef __WIN__
+/*********************************************************************
+NOTE: The Windows epoch starts from 1601/01/01 whereas the Unix
+epoch starts from 1970/1/1. For selection of constant see:
+http://support.microsoft.com/kb/167296/ */
+#define WIN_TO_UNIX_DELTA_USEC  ((ib_int64_t) 11644473600000000ULL)
+
+
+/*********************************************************************
+This is the Windows version of gettimeofday(2).*/
+static
+int
+ut_gettimeofday(
+/*============*/
+			/* out: 0 if all OK else -1 */
+	struct timeval*	tv,	/* out: Values are relative to Unix epoch */
+	void*		tz)	/* in: not used */
+{
+	FILETIME	ft;
+	ib_int64_t	tm;
+
+	if (!tv) {
+		errno = EINVAL;
+		return(-1);
+	}
+
+	GetSystemTimeAsFileTime(&ft);
+
+	tm = (ib_int64_t) ft.dwHighDateTime << 32;
+	tm |= ft.dwLowDateTime;
+
+	ut_a(tm >= 0);	/* If tm wraps over to negative, the quotient / 10
+			does not work */
+
+	tm /= 10;	/* Convert from 100 nsec periods to usec */
+
+	/* If we don't convert to the Unix epoch the value for
+	struct timeval::tv_sec will overflow.*/
+	tm -= WIN_TO_UNIX_DELTA_USEC;
+
+	tv->tv_sec  = (long) (tm / 1000000L);
+	tv->tv_usec = (long) (tm % 1000000L);
+
+	return(0);
+}
+#else
+#define	ut_gettimeofday		gettimeofday
+#endif
+
+/************************************************************
+Gets the high 32 bits in a ulint. That is makes a shift >> 32,
+but since there seem to be compiler bugs in both gcc and Visual C++,
+we do this by a special conversion. */
+UNIV_INTERN
+ulint
+ut_get_high32(
+/*==========*/
+			/* out: a >> 32 */
+	ulint	a)	/* in: ulint */
+{
+	ib_int64_t	i;
+
+	i = (ib_int64_t)a;
+
+	i = i >> 32;
+
+	return((ulint)i);
+}
+
+/**************************************************************
+Returns system time. We do not specify the format of the time returned:
+the only way to manipulate it is to use the function ut_difftime. */
+UNIV_INTERN
+ib_time_t
+ut_time(void)
+/*=========*/
+{
+	return(time(NULL));
+}
+
+/**************************************************************
+Returns system time.
+Upon successful completion, the value 0 is returned; otherwise the
+value -1 is returned and the global variable errno is set to indicate the
+error. */
+UNIV_INTERN
+int
+ut_usectime(
+/*========*/
+			/* out: 0 on success, -1 otherwise */
+	ulint*	sec,	/* out: seconds since the Epoch */
+	ulint*	ms)	/* out: microseconds since the Epoch+*sec */
+{
+	struct timeval	tv;
+	int		ret;
+	int		errno_gettimeofday;
+	int		i;
+
+	for (i = 0; i < 10; i++) {
+
+		ret = ut_gettimeofday(&tv, NULL);
+
+		if (ret == -1) {
+			errno_gettimeofday = errno;
+			ut_print_timestamp(stderr);
+			fprintf(stderr, "  InnoDB: gettimeofday(): %s\n",
+				strerror(errno_gettimeofday));
+			os_thread_sleep(100000);  /* 0.1 sec */
+			errno = errno_gettimeofday;
+		} else {
+			break;
+		}
+	}
+
+	if (ret != -1) {
+		*sec = (ulint) tv.tv_sec;
+		*ms  = (ulint) tv.tv_usec;
+	}
+
+	return(ret);
+}
+
+/**************************************************************
+Returns the number of microseconds since epoch. Similar to
+time(3), the return value is also stored in *tloc, provided
+that tloc is non-NULL. */
+UNIV_INTERN
+ullint
+ut_time_us(
+/*=======*/
+			/* out: us since epoch */
+	ullint*	tloc)	/* out: us since epoch, if non-NULL */
+{
+	struct timeval	tv;
+	ullint		us;
+
+	ut_gettimeofday(&tv, NULL);
+
+	us = (ullint) tv.tv_sec * 1000000 + tv.tv_usec;
+
+	if (tloc != NULL) {
+		*tloc = us;
+	}
+
+	return(us);
+}
+
+/**************************************************************
+Returns the difference of two times in seconds. */
+UNIV_INTERN
+double
+ut_difftime(
+/*========*/
+				/* out: time2 - time1 expressed in seconds */
+	ib_time_t	time2,	/* in: time */
+	ib_time_t	time1)	/* in: time */
+{
+	return(difftime(time2, time1));
+}
+
+/**************************************************************
+Prints a timestamp to a file. */
+UNIV_INTERN
+void
+ut_print_timestamp(
+/*===============*/
+	FILE*  file) /* in: file where to print */
+{
+#ifdef __WIN__
+	SYSTEMTIME cal_tm;
+
+	GetLocalTime(&cal_tm);
+
+	fprintf(file,"%02d%02d%02d %2d:%02d:%02d",
+		(int)cal_tm.wYear % 100,
+		(int)cal_tm.wMonth,
+		(int)cal_tm.wDay,
+		(int)cal_tm.wHour,
+		(int)cal_tm.wMinute,
+		(int)cal_tm.wSecond);
+#else
+	struct tm  cal_tm;
+	struct tm* cal_tm_ptr;
+	time_t	   tm;
+
+	time(&tm);
+
+#ifdef HAVE_LOCALTIME_R
+	localtime_r(&tm, &cal_tm);
+	cal_tm_ptr = &cal_tm;
+#else
+	cal_tm_ptr = localtime(&tm);
+#endif
+	fprintf(file,"%02d%02d%02d %2d:%02d:%02d",
+		cal_tm_ptr->tm_year % 100,
+		cal_tm_ptr->tm_mon + 1,
+		cal_tm_ptr->tm_mday,
+		cal_tm_ptr->tm_hour,
+		cal_tm_ptr->tm_min,
+		cal_tm_ptr->tm_sec);
+#endif
+}
+
+/**************************************************************
+Sprintfs a timestamp to a buffer, 13..14 chars plus terminating NUL. */
+UNIV_INTERN
+void
+ut_sprintf_timestamp(
+/*=================*/
+	char*	buf) /* in: buffer where to sprintf */
+{
+#ifdef __WIN__
+	SYSTEMTIME cal_tm;
+
+	GetLocalTime(&cal_tm);
+
+	sprintf(buf, "%02d%02d%02d %2d:%02d:%02d",
+		(int)cal_tm.wYear % 100,
+		(int)cal_tm.wMonth,
+		(int)cal_tm.wDay,
+		(int)cal_tm.wHour,
+		(int)cal_tm.wMinute,
+		(int)cal_tm.wSecond);
+#else
+	struct tm  cal_tm;
+	struct tm* cal_tm_ptr;
+	time_t	   tm;
+
+	time(&tm);
+
+#ifdef HAVE_LOCALTIME_R
+	localtime_r(&tm, &cal_tm);
+	cal_tm_ptr = &cal_tm;
+#else
+	cal_tm_ptr = localtime(&tm);
+#endif
+	sprintf(buf, "%02d%02d%02d %2d:%02d:%02d",
+		cal_tm_ptr->tm_year % 100,
+		cal_tm_ptr->tm_mon + 1,
+		cal_tm_ptr->tm_mday,
+		cal_tm_ptr->tm_hour,
+		cal_tm_ptr->tm_min,
+		cal_tm_ptr->tm_sec);
+#endif
+}
+
+#ifdef UNIV_HOTBACKUP
+/**************************************************************
+Sprintfs a timestamp to a buffer with no spaces and with ':' characters
+replaced by '_'. */
+UNIV_INTERN
+void
+ut_sprintf_timestamp_without_extra_chars(
+/*=====================================*/
+	char*	buf) /* in: buffer where to sprintf */
+{
+#ifdef __WIN__
+	SYSTEMTIME cal_tm;
+
+	GetLocalTime(&cal_tm);
+
+	sprintf(buf, "%02d%02d%02d_%2d_%02d_%02d",
+		(int)cal_tm.wYear % 100,
+		(int)cal_tm.wMonth,
+		(int)cal_tm.wDay,
+		(int)cal_tm.wHour,
+		(int)cal_tm.wMinute,
+		(int)cal_tm.wSecond);
+#else
+	struct tm  cal_tm;
+	struct tm* cal_tm_ptr;
+	time_t	   tm;
+
+	time(&tm);
+
+#ifdef HAVE_LOCALTIME_R
+	localtime_r(&tm, &cal_tm);
+	cal_tm_ptr = &cal_tm;
+#else
+	cal_tm_ptr = localtime(&tm);
+#endif
+	sprintf(buf, "%02d%02d%02d_%2d_%02d_%02d",
+		cal_tm_ptr->tm_year % 100,
+		cal_tm_ptr->tm_mon + 1,
+		cal_tm_ptr->tm_mday,
+		cal_tm_ptr->tm_hour,
+		cal_tm_ptr->tm_min,
+		cal_tm_ptr->tm_sec);
+#endif
+}
+
+/**************************************************************
+Returns current year, month, day. */
+UNIV_INTERN
+void
+ut_get_year_month_day(
+/*==================*/
+	ulint*	year,	/* out: current year */
+	ulint*	month,	/* out: month */
+	ulint*	day)	/* out: day */
+{
+#ifdef __WIN__
+	SYSTEMTIME cal_tm;
+
+	GetLocalTime(&cal_tm);
+
+	*year = (ulint)cal_tm.wYear;
+	*month = (ulint)cal_tm.wMonth;
+	*day = (ulint)cal_tm.wDay;
+#else
+	struct tm  cal_tm;
+	struct tm* cal_tm_ptr;
+	time_t	   tm;
+
+	time(&tm);
+
+#ifdef HAVE_LOCALTIME_R
+	localtime_r(&tm, &cal_tm);
+	cal_tm_ptr = &cal_tm;
+#else
+	cal_tm_ptr = localtime(&tm);
+#endif
+	*year = (ulint)cal_tm_ptr->tm_year + 1900;
+	*month = (ulint)cal_tm_ptr->tm_mon + 1;
+	*day = (ulint)cal_tm_ptr->tm_mday;
+#endif
+}
+#endif /* UNIV_HOTBACKUP */
+
+/*****************************************************************
+Runs an idle loop on CPU. The argument gives the desired delay
+in microseconds on 100 MHz Pentium + Visual C++. */
+UNIV_INTERN
+ulint
+ut_delay(
+/*=====*/
+			/* out: dummy value */
+	ulint	delay)	/* in: delay in microseconds on 100 MHz Pentium */
+{
+	ulint	i, j;
+
+	j = 0;
+
+	for (i = 0; i < delay * 50; i++) {
+		j += i;
+	}
+
+	if (ut_always_false) {
+		ut_always_false = (ibool) j;
+	}
+
+	return(j);
+}
+
+/*****************************************************************
+Prints the contents of a memory buffer in hex and ascii. */
+UNIV_INTERN
+void
+ut_print_buf(
+/*=========*/
+	FILE*		file,	/* in: file where to print */
+	const void*	buf,	/* in: memory buffer */
+	ulint		len)	/* in: length of the buffer */
+{
+	const byte*	data;
+	ulint		i;
+
+	UNIV_MEM_ASSERT_RW(buf, len);
+
+	fprintf(file, " len %lu; hex ", len);
+
+	for (data = (const byte*)buf, i = 0; i < len; i++) {
+		fprintf(file, "%02lx", (ulong)*data++);
+	}
+
+	fputs("; asc ", file);
+
+	data = (const byte*)buf;
+
+	for (i = 0; i < len; i++) {
+		int	c = (int) *data++;
+		putc(isprint(c) ? c : ' ', file);
+	}
+
+	putc(';', file);
+}
+
+/*****************************************************************
+Calculates fast the number rounded up to the nearest power of 2. */
+UNIV_INTERN
+ulint
+ut_2_power_up(
+/*==========*/
+			/* out: first power of 2 which is >= n */
+	ulint	n)	/* in: number != 0 */
+{
+	ulint	res;
+
+	res = 1;
+
+	ut_ad(n > 0);
+
+	while (res < n) {
+		res = res * 2;
+	}
+
+	return(res);
+}
+
+/**************************************************************************
+Outputs a NUL-terminated file name, quoted with apostrophes. */
+UNIV_INTERN
+void
+ut_print_filename(
+/*==============*/
+	FILE*		f,	/* in: output stream */
+	const char*	name)	/* in: name to print */
+{
+	putc('\'', f);
+	for (;;) {
+		int	c = *name++;
+		switch (c) {
+		case 0:
+			goto done;
+		case '\'':
+			putc(c, f);
+			/* fall through */
+		default:
+			putc(c, f);
+		}
+	}
+done:
+	putc('\'', f);
+}
+
+/**************************************************************************
+Outputs a fixed-length string, quoted as an SQL identifier.
+If the string contains a slash '/', the string will be
+output as two identifiers separated by a period (.),
+as in SQL database_name.identifier. */
+UNIV_INTERN
+void
+ut_print_name(
+/*==========*/
+	FILE*		f,	/* in: output stream */
+	trx_t*		trx,	/* in: transaction */
+	ibool		table_id,/* in: TRUE=print a table name,
+				FALSE=print other identifier */
+	const char*	name)	/* in: name to print */
+{
+	ut_print_namel(f, trx, table_id, name, strlen(name));
+}
+
+/**************************************************************************
+Outputs a fixed-length string, quoted as an SQL identifier.
+If the string contains a slash '/', the string will be
+output as two identifiers separated by a period (.),
+as in SQL database_name.identifier. */
+UNIV_INTERN
+void
+ut_print_namel(
+/*===========*/
+	FILE*		f,	/* in: output stream */
+	trx_t*		trx,	/* in: transaction (NULL=no quotes) */
+	ibool		table_id,/* in: TRUE=print a table name,
+				FALSE=print other identifier */
+	const char*	name,	/* in: name to print */
+	ulint		namelen)/* in: length of name */
+{
+#ifdef UNIV_HOTBACKUP
+	fwrite(name, 1, namelen, f);
+#else
+	/* 2 * NAME_LEN for database and table name,
+	and some slack for the #mysql50# prefix and quotes */
+	char		buf[3 * NAME_LEN];
+	const char*	bufend;
+
+	bufend = innobase_convert_name(buf, sizeof buf,
+				       name, namelen,
+				       trx ? trx->mysql_thd : NULL,
+				       table_id);
+
+	fwrite(buf, 1, bufend - buf, f);
+#endif
+}
+
+/**************************************************************************
+Catenate files. */
+UNIV_INTERN
+void
+ut_copy_file(
+/*=========*/
+	FILE*	dest,	/* in: output file */
+	FILE*	src)	/* in: input file to be appended to output */
+{
+	long	len = ftell(src);
+	char	buf[4096];
+
+	rewind(src);
+	do {
+		size_t	maxs = len < (long) sizeof buf
+			? (size_t) len
+			: sizeof buf;
+		size_t	size = fread(buf, 1, maxs, src);
+		fwrite(buf, 1, size, dest);
+		len -= (long) size;
+		if (size < maxs) {
+			break;
+		}
+	} while (len > 0);
+}
+
+/**************************************************************************
+snprintf(). */
+
+#ifdef __WIN__
+#include <stdarg.h>
+int
+ut_snprintf(
+				/* out: number of characters that would
+				have been printed if the size were
+				unlimited, not including the terminating
+				'\0'. */
+	char*		str,	/* out: string */
+	size_t		size,	/* in: str size */
+	const char*	fmt,	/* in: format */
+	...)			/* in: format values */
+{
+	int	res;
+	va_list	ap1;
+	va_list	ap2;
+
+	va_start(ap1, fmt);
+	va_start(ap2, fmt);
+
+	res = _vscprintf(fmt, ap1);
+	ut_a(res != -1);
+
+	if (size > 0) {
+		_vsnprintf(str, size, fmt, ap2);
+
+		if ((size_t) res >= size) {
+			str[size - 1] = '\0';
+		}
+	}
+
+	va_end(ap1);
+	va_end(ap2);
+
+	return(res);
+}
+#endif /* __WIN__ */
diff --git a/storage/xtradb/ut/ut0vec.c b/storage/xtradb/ut/ut0vec.c
new file mode 100644
index 00000000000..69b7bec701a
--- /dev/null
+++ b/storage/xtradb/ut/ut0vec.c
@@ -0,0 +1,72 @@
+/*****************************************************************************
+
+Copyright (c) 2006, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+#include "ut0vec.h"
+#ifdef UNIV_NONINL
+#include "ut0vec.ic"
+#endif
+#include <string.h>
+
+/********************************************************************
+Create a new vector with the given initial size. */
+UNIV_INTERN
+ib_vector_t*
+ib_vector_create(
+/*=============*/
+				/* out: vector */
+	mem_heap_t*	heap,	/* in: heap */
+	ulint		size)	/* in: initial size */
+{
+	ib_vector_t*	vec;
+
+	ut_a(size > 0);
+
+	vec = mem_heap_alloc(heap, sizeof(*vec));
+
+	vec->heap = heap;
+	vec->data = mem_heap_alloc(heap, sizeof(void*) * size);
+	vec->used = 0;
+	vec->total = size;
+
+	return(vec);
+}
+
+/********************************************************************
+Push a new element to the vector, increasing its size if necessary. */
+UNIV_INTERN
+void
+ib_vector_push(
+/*===========*/
+	ib_vector_t*	vec,	/* in: vector */
+	void*		elem)	/* in: data element */
+{
+	if (vec->used >= vec->total) {
+		void**	new_data;
+		ulint	new_total = vec->total * 2;
+
+		new_data = mem_heap_alloc(vec->heap,
+					  sizeof(void*) * new_total);
+		memcpy(new_data, vec->data, sizeof(void*) * vec->total);
+
+		vec->data = new_data;
+		vec->total = new_total;
+	}
+
+	vec->data[vec->used] = elem;
+	vec->used++;
+}
diff --git a/storage/xtradb/ut/ut0wqueue.c b/storage/xtradb/ut/ut0wqueue.c
new file mode 100644
index 00000000000..a5c14ac8130
--- /dev/null
+++ b/storage/xtradb/ut/ut0wqueue.c
@@ -0,0 +1,110 @@
+/*****************************************************************************
+
+Copyright (c) 2006, 2009, Innobase Oy. All Rights Reserved.
+
+This program is free software; you can redistribute it and/or modify it under
+the terms of the GNU General Public License as published by the Free Software
+Foundation; version 2 of the License.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+Place, Suite 330, Boston, MA 02111-1307 USA
+
+*****************************************************************************/
+
+#include "ut0wqueue.h"
+
+/********************************************************************
+Create a new work queue. */
+UNIV_INTERN
+ib_wqueue_t*
+ib_wqueue_create(void)
+/*===================*/
+			/* out: work queue */
+{
+	ib_wqueue_t*	wq = mem_alloc(sizeof(ib_wqueue_t));
+
+	mutex_create(&wq->mutex, SYNC_WORK_QUEUE);
+
+	wq->items = ib_list_create();
+	wq->event = os_event_create(NULL);
+
+	return(wq);
+}
+
+/********************************************************************
+Free a work queue. */
+UNIV_INTERN
+void
+ib_wqueue_free(
+/*===========*/
+	ib_wqueue_t*	wq)	/* in: work queue */
+{
+	ut_a(!ib_list_get_first(wq->items));
+
+	mutex_free(&wq->mutex);
+	ib_list_free(wq->items);
+	os_event_free(wq->event);
+
+	mem_free(wq);
+}
+
+/********************************************************************
+Add a work item to the queue. */
+UNIV_INTERN
+void
+ib_wqueue_add(
+/*==========*/
+	ib_wqueue_t*	wq,	/* in: work queue */
+	void*		item,	/* in: work item */
+	mem_heap_t*	heap)	/* in: memory heap to use for allocating the
+				list node */
+{
+	mutex_enter(&wq->mutex);
+
+	ib_list_add_last(wq->items, item, heap);
+	os_event_set(wq->event);
+
+	mutex_exit(&wq->mutex);
+}
+
+/********************************************************************
+Wait for a work item to appear in the queue. */
+UNIV_INTERN
+void*
+ib_wqueue_wait(
+				/* out: work item */
+	ib_wqueue_t*	wq)	/* in: work queue */
+{
+	ib_list_node_t*	node;
+
+	for (;;) {
+		os_event_wait(wq->event);
+
+		mutex_enter(&wq->mutex);
+
+		node = ib_list_get_first(wq->items);
+
+		if (node) {
+			ib_list_remove(wq->items, node);
+
+			if (!ib_list_get_first(wq->items)) {
+				/* We must reset the event when the list
+				gets emptied. */
+				os_event_reset(wq->event);
+			}
+
+			break;
+		}
+
+		mutex_exit(&wq->mutex);
+	}
+
+	mutex_exit(&wq->mutex);
+
+	return(node->data);
+}
diff --git a/storage/xtradb/win-plugin/README b/storage/xtradb/win-plugin/README
new file mode 100644
index 00000000000..9182f2c555c
--- /dev/null
+++ b/storage/xtradb/win-plugin/README
@@ -0,0 +1,25 @@
+This directory contains patches that need to be applied to the MySQL
+source tree in order to build the dynamic plugin on Windows --
+HA_INNODB.DLL. Please note the followings when adding the patches:
+
+* The patch must be applied from the mysql top-level source directory.
+  patch -p0 < win-plugin.diff
+* The patch filenames end in ".diff".
+* All patches here are expected to apply cleanly to the latest MySQL 5.1
+  tree when storage/innobase is replaced with this InnoDB branch.
+
+When applying the patch, the following files will be modified:
+
+  * CMakeLists.txt
+  * sql/CMakeLists.txt
+  * win/configure.js
+  * win/build-vs71.bat
+  * win/build-vs8.bat
+  * win/build-vs8_x64.bat
+
+Also, two new files will be added:
+
+  * sql/mysqld.def
+  * sql/mysqld_x64.def
+
+You can get "patch" utility for Windows from http://unxutils.sourceforge.net/
diff --git a/storage/xtradb/win-plugin/win-plugin.diff b/storage/xtradb/win-plugin/win-plugin.diff
new file mode 100644
index 00000000000..46d2e5b2d2d
--- /dev/null
+++ b/storage/xtradb/win-plugin/win-plugin.diff
@@ -0,0 +1,297 @@
+diff -Nur CMakeLists.txt.orig CMakeLists.txt
+--- CMakeLists.txt.orig	2008-10-03 12:25:41 -05:00
++++ CMakeLists.txt	2008-09-26 17:32:51 -05:00
+@@ -244,9 +244,9 @@
+ IF(WITH_FEDERATED_STORAGE_ENGINE)
+   ADD_SUBDIRECTORY(storage/federated)
+ ENDIF(WITH_FEDERATED_STORAGE_ENGINE)
+-IF(WITH_INNOBASE_STORAGE_ENGINE)
++IF(WITH_INNOBASE_STORAGE_ENGINE OR INNODB_DYNAMIC_PLUGIN)
+   ADD_SUBDIRECTORY(storage/innobase)
+-ENDIF(WITH_INNOBASE_STORAGE_ENGINE)
++ENDIF(WITH_INNOBASE_STORAGE_ENGINE OR INNODB_DYNAMIC_PLUGIN)
+ ADD_SUBDIRECTORY(sql)
+ ADD_SUBDIRECTORY(server-tools/instance-manager)
+ ADD_SUBDIRECTORY(libmysql)
+ 
+diff -Nur sql/CMakeLists.txt.orig sql/CMakeLists.txt
+--- sql/CMakeLists.txt.orig	2008-10-03 12:25:41 -05:00
++++ sql/CMakeLists.txt	2008-09-24 03:58:19 -05:00
+@@ -100,6 +100,15 @@
+                       LINK_FLAGS  "/PDB:${CMAKE_CFG_INTDIR}/mysqld${MYSQLD_EXE_SUFFIX}.pdb")
+ ENDIF(cmake_version EQUAL 20406)
+ 
++# Checks for 64-bit version
++IF(CMAKE_SIZEOF_VOID_P MATCHES 8)
++SET_TARGET_PROPERTIES(mysqld PROPERTIES
++                      LINK_FLAGS "/def:\"${PROJECT_SOURCE_DIR}/sql/mysqld_x64.def\"")
++ELSE(CMAKE_SIZEOF_VOID_P MATCHES 8)
++SET_TARGET_PROPERTIES(mysqld PROPERTIES
++                      LINK_FLAGS "/def:\"${PROJECT_SOURCE_DIR}/sql/mysqld.def\"")
++ENDIF(CMAKE_SIZEOF_VOID_P MATCHES 8)
++
+ IF(EMBED_MANIFESTS)
+   MYSQL_EMBED_MANIFEST("mysqld" "asInvoker")
+ ENDIF(EMBED_MANIFESTS)
+ 
+diff -Nur sql/mysqld.def.orig sql/mysqld.def
+--- sql/mysqld.def.orig	1969-12-31 18:00:00 -06:00
++++ sql/mysqld.def	2008-10-31 02:20:32 -05:00
+@@ -0,0 +1,98 @@
++EXPORTS
++	?use_hidden_primary_key@handler@@UAEXXZ
++	?get_dynamic_partition_info@handler@@UAEXPAUPARTITION_INFO@@I@Z
++	?read_first_row@handler@@UAEHPAEI@Z
++	?read_range_next@handler@@UAEHXZ
++	?read_range_first@handler@@UAEHPBUst_key_range@@0_N1@Z
++	?read_multi_range_first@handler@@UAEHPAPAUst_key_multi_range@@PAU2@I_NPAUst_handler_buffer@@@Z
++	?read_multi_range_next@handler@@UAEHPAPAUst_key_multi_range@@@Z
++	?index_read_idx_map@handler@@UAEHPAEIPBEKW4ha_rkey_function@@@Z
++	?print_error@handler@@UAEXHH@Z
++	?clone@handler@@UAEPAV1@PAUst_mem_root@@@Z
++	?get_auto_increment@handler@@UAEX_K00PA_K1@Z
++	?index_next_same@handler@@UAEHPAEPBEI@Z
++	?get_error_message@handler@@UAE_NHPAVString@@@Z
++	?ha_thd@handler@@IBEPAVTHD@@XZ
++	?update_auto_increment@handler@@QAEHXZ
++	?ha_statistic_increment@handler@@IBEXPQsystem_status_var@@K@Z
++	?trans_register_ha@@YAXPAVTHD@@_NPAUhandlerton@@@Z
++	?cmp@Field_blob@@QAEHPBEI0I@Z
++	?set_time@Field_timestamp@@QAEXXZ
++	?sql_print_error@@YAXPBDZZ
++	?sql_print_warning@@YAXPBDZZ
++	?check_global_access@@YA_NPAVTHD@@K@Z
++	?schema_table_store_record@@YA_NPAVTHD@@PAUst_table@@@Z
++	?get_quote_char_for_identifier@@YAHPAVTHD@@PBDI@Z
++	?copy@String@@QAE_NXZ
++	?copy@String@@QAE_NABV1@@Z
++	?copy@String@@QAE_NPBDIPAUcharset_info_st@@@Z
++	?copy_and_convert@@YAIPADIPAUcharset_info_st@@PBDI1PAI@Z
++	?filename_to_tablename@@YAIPBDPADI@Z
++	?strconvert@@YAIPAUcharset_info_st@@PBD0PADIPAI@Z
++	?calculate_key_len@@YAIPAUst_table@@IPBEK@Z
++	?sql_alloc@@YAPAXI@Z
++	?localtime_to_TIME@@YAXPAUst_mysql_time@@PAUtm@@@Z
++	?push_warning@@YAPAVMYSQL_ERROR@@PAVTHD@@W4enum_warning_level@1@IPBD@Z
++	?push_warning_printf@@YAXPAVTHD@@W4enum_warning_level@MYSQL_ERROR@@IPBDZZ
++	?drop_table@handler@@EAEXPBD@Z
++	?column_bitmaps_signal@handler@@UAEXXZ
++	?delete_table@handler@@MAEHPBD@Z
++	?rename_table@handler@@MAEHPBD0@Z
++	?key_map_empty@@3V?$Bitmap@$0EA@@@B
++	?THR_THD@@3PAVTHD@@A
++	?end_of_list@@3Ulist_node@@A
++	?mysql_tmpdir_list@@3Ust_my_tmpdir@@A
++	mysql_query_cache_invalidate4
++	thd_query
++	thd_sql_command
++	thd_get_thread_id
++	thd_get_xid
++	thd_slave_thread
++	thd_non_transactional_update
++	thd_mark_transaction_to_rollback
++	thd_security_context
++	thd_charset
++	thd_test_options
++	thd_ha_data
++	thd_killed
++	thd_tx_isolation
++	thd_tablespace_op
++	thd_sql_command
++	thd_memdup
++	thd_make_lex_string
++	thd_in_lock_tables
++	thd_binlog_format
++	_my_hash_init
++	my_hash_free
++	my_tmpdir
++	check_if_legal_filename
++	my_filename
++	my_sync_dir_by_file
++	alloc_root
++	thr_lock_data_init
++	thr_lock_init
++	thr_lock_delete
++	my_multi_malloc
++	get_charset
++	unpack_filename
++	my_hash_insert
++	my_hash_search
++	my_hash_delete
++	mysql_bin_log_file_pos
++	mysql_bin_log_file_name
++	mysqld_embedded
++	my_thread_name
++	my_malloc
++	my_no_flags_free
++	_sanity
++	_mymalloc
++	_myfree
++	_my_strdup
++	_my_thread_var
++	my_error
++	pthread_cond_init
++	pthread_cond_signal
++	pthread_cond_wait
++	pthread_cond_destroy
++	localtime_r
++	my_strdup
+
+diff -Nur sql/mysqld_x64.def.orig sql/mysqld_x64.def
+--- sql/mysqld_x64.def.orig	1969-12-31 18:00:00 -06:00
++++ sql/mysqld_x64.def		2008-10-31 02:22:04 -05:00
+@@ -0,0 +1,98 @@
++EXPORTS
++	?use_hidden_primary_key@handler@@UEAAXXZ
++	?get_dynamic_partition_info@handler@@UEAAXPEAUPARTITION_INFO@@I@Z
++	?read_first_row@handler@@UEAAHPEAEI@Z
++	?read_range_next@handler@@UEAAHXZ
++	?read_range_first@handler@@UEAAHPEBUst_key_range@@0_N1@Z
++	?read_multi_range_first@handler@@UEAAHPEAPEAUst_key_multi_range@@PEAU2@I_NPEAUst_handler_buffer@@@Z
++	?read_multi_range_next@handler@@UEAAHPEAPEAUst_key_multi_range@@@Z
++	?index_read_idx_map@handler@@UEAAHPEAEIPEBEKW4ha_rkey_function@@@Z
++	?print_error@handler@@UEAAXHH@Z
++	?clone@handler@@UEAAPEAV1@PEAUst_mem_root@@@Z
++	?get_auto_increment@handler@@UEAAX_K00PEA_K1@Z
++	?index_next_same@handler@@UEAAHPEAEPEBEI@Z
++	?get_error_message@handler@@UEAA_NHPEAVString@@@Z
++	?ha_thd@handler@@IEBAPEAVTHD@@XZ
++	?update_auto_increment@handler@@QEAAHXZ
++	?ha_statistic_increment@handler@@IEBAXPEQsystem_status_var@@K@Z
++	?trans_register_ha@@YAXPEAVTHD@@_NPEAUhandlerton@@@Z
++	?cmp@Field_blob@@QEAAHPEBEI0I@Z
++	?set_time@Field_timestamp@@QEAAXXZ
++	?sql_print_error@@YAXPEBDZZ
++	?sql_print_warning@@YAXPEBDZZ
++	?check_global_access@@YA_NPEAVTHD@@K@Z
++	?schema_table_store_record@@YA_NPEAVTHD@@PEAUst_table@@@Z
++	?get_quote_char_for_identifier@@YAHPEAVTHD@@PEBDI@Z
++	?copy@String@@QEAA_NXZ
++	?copy@String@@QEAA_NAEBV1@@Z
++	?copy@String@@QEAA_NPEBDIPEAUcharset_info_st@@@Z
++	?copy_and_convert@@YAIPEADIPEAUcharset_info_st@@PEBDI1PEAI@Z
++	?filename_to_tablename@@YAIPEBDPEADI@Z
++	?strconvert@@YAIPEAUcharset_info_st@@PEBD0PEADIPEAI@Z
++	?calculate_key_len@@YAIPEAUst_table@@IPEBEK@Z
++	?sql_alloc@@YAPEAX_K@Z
++	?localtime_to_TIME@@YAXPEAUst_mysql_time@@PEAUtm@@@Z
++	?push_warning@@YAPEAVMYSQL_ERROR@@PEAVTHD@@W4enum_warning_level@1@IPEBD@Z
++	?push_warning_printf@@YAXPEAVTHD@@W4enum_warning_level@MYSQL_ERROR@@IPEBDZZ
++	?drop_table@handler@@EEAAXPEBD@Z
++	?column_bitmaps_signal@handler@@UEAAXXZ
++	?delete_table@handler@@MEAAHPEBD@Z
++	?rename_table@handler@@MEAAHPEBD0@Z
++	?key_map_empty@@3V?$Bitmap@$0EA@@@B
++	?THR_THD@@3PEAVTHD@@EA
++	?end_of_list@@3Ulist_node@@A
++	?mysql_tmpdir_list@@3Ust_my_tmpdir@@A
++	mysql_query_cache_invalidate4
++	thd_query
++	thd_sql_command
++	thd_get_thread_id
++	thd_get_xid
++	thd_slave_thread
++	thd_non_transactional_update
++	thd_mark_transaction_to_rollback
++	thd_security_context
++	thd_charset
++	thd_test_options
++	thd_ha_data
++	thd_killed
++	thd_tx_isolation
++	thd_tablespace_op
++	thd_sql_command
++	thd_memdup
++	thd_make_lex_string
++	thd_in_lock_tables
++	thd_binlog_format
++	_my_hash_init
++	my_hash_free
++	my_tmpdir
++	check_if_legal_filename
++	my_filename
++	my_sync_dir_by_file
++	alloc_root
++	thr_lock_data_init
++	thr_lock_init
++	thr_lock_delete
++	my_multi_malloc
++	get_charset
++	unpack_filename
++	my_hash_insert
++	my_hash_search
++	my_hash_delete
++	mysql_bin_log_file_pos
++	mysql_bin_log_file_name
++	mysqld_embedded
++	my_thread_name
++	my_malloc
++	my_no_flags_free
++	_sanity
++	_mymalloc
++	_myfree
++	_my_strdup
++	_my_thread_var
++	my_error
++	pthread_cond_init
++	pthread_cond_signal
++	pthread_cond_wait
++	pthread_cond_destroy
++	localtime_r
++	my_strdup
+
+diff -Nur win/configure.js.orig win/configure.js
+--- win/configure.js.orig	2008-09-26 21:18:37 -05:00
++++ win/configure.js	2008-10-01 11:21:27 -05:00
+@@ -49,6 +49,7 @@
+             case "CYBOZU":
+             case "EMBED_MANIFESTS":
+             case "WITH_EMBEDDED_SERVER":
++            case "INNODB_DYNAMIC_PLUGIN":
+                     configfile.WriteLine("SET (" + args.Item(i) + " TRUE)");
+                     break;
+             case "MYSQL_SERVER_SUFFIX":
+
+diff -Nur win/build-vs71.bat.orig win/build-vs71.bat
+--- win/build-vs71.bat.orig	2008-08-20 10:21:59 -05:00
++++ win/build-vs71.bat	2008-10-27 10:52:38 -05:00
+@@ -15,8 +15,10 @@
+ REM along with this program; if not, write to the Free Software
+ REM Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
+ 
++REM CMAKE_BUILD_TYPE can be specified as Release or Debug
++
+ if exist cmakecache.txt del cmakecache.txt
+ copy win\vs71cache.txt cmakecache.txt
+-cmake -G "Visual Studio 7 .NET 2003"
++cmake -G "Visual Studio 7 .NET 2003" -DCMAKE_BUILD_TYPE=%1
+ copy cmakecache.txt win\vs71cache.txt
+ 
+diff -Nur win/build-vs8.bat.orig win/build-vs8.bat
+--- win/build-vs8.bat.orig	2008-08-20 10:21:59 -05:00
++++ win/build-vs8.bat	2008-10-27 10:52:31 -05:00
+@@ -15,7 +15,9 @@
+ REM along with this program; if not, write to the Free Software
+ REM Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
+ 
++REM CMAKE_BUILD_TYPE can be specified as Release or Debug
++
+ if exist cmakecache.txt del cmakecache.txt
+ copy win\vs8cache.txt cmakecache.txt
+-cmake -G "Visual Studio 8 2005"
++cmake -G "Visual Studio 8 2005" -DCMAKE_BUILD_TYPE=%1
+ copy cmakecache.txt win\vs8cache.txt
+diff -Nur win/build-vs8_x64.bat.orig win/build-vs8_x64.bat
+--- win/build-vs8_x64.bat.orig	2008-08-20 10:21:59 -05:00
++++ win/build-vs8_x64.bat	2008-10-27 10:53:11 -05:00
+@@ -15,7 +15,9 @@
+ REM along with this program; if not, write to the Free Software
+ REM Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
+ 
++REM CMAKE_BUILD_TYPE can be specified as Release or Debug
++
+ if exist cmakecache.txt del cmakecache.txt
+ copy win\vs8cache.txt cmakecache.txt
+-cmake -G "Visual Studio 8 2005 Win64"
++cmake -G "Visual Studio 8 2005 Win64" -DCMAKE_BUILD_TYPE=%1
+ copy cmakecache.txt win\vs8cache.txt