Merge 10.7 into 10.8

debian/autobake-deb.sh

@@ -26,9 +26,9 @@ then
   sed '/Add support for verbose builds/,/^$/d' -i debian/rules
 elif [ -d storage/columnstore/columnstore/debian ]
 then
-  # ColumnStore is explicitly disabled in the native Debian build, so allow it
+  # ColumnStore is explicitly disabled in the native Debian build. Enable it
   # now when build is triggered by autobake-deb.sh (MariaDB.org) and when the
-  # build is not running on Travis or Gitlab-CI
+  # build is not running on Gitlab-CI.
   sed '/-DPLUGIN_COLUMNSTORE=NO/d' -i debian/rules
   # Take the files and part of control from MCS directory
   if [ ! -f debian/mariadb-plugin-columnstore.install ]
@@ -59,13 +59,13 @@ replace_uring_with_aio()
 {
   sed 's/liburing-dev/libaio-dev/g' -i debian/control
   sed -e '/-DIGNORE_AIO_CHECK=YES/d' \
-      -e '/-DWITH_URING=yes/d' -i debian/rules
+      -e '/-DWITH_URING=YES/d' -i debian/rules
 }
 
 disable_pmem()
 {
   sed '/libpmem-dev/d' -i debian/control
-  sed '/-DWITH_PMEM=yes/d' -i debian/rules
+  sed '/-DWITH_PMEM=YES/d' -i debian/rules
 }
 
 disable_libfmt()
@@ -76,15 +76,34 @@ disable_libfmt()
 
 architecture=$(dpkg-architecture -q DEB_BUILD_ARCH)
 
+# Parse release name and number from Linux standard base release
+# Example:
+#   $ lsb_release -a
+#   No LSB modules are available.
+#   Distributor ID:	Debian
+#   Description:	Debian GNU/Linux bookworm/sid
+#   Release:	n/a
+#   Codename:	n/a
 LSBID="$(lsb_release -si  | tr '[:upper:]' '[:lower:]')"
 LSBVERSION="$(lsb_release -sr | sed -e "s#\.##g")"
 LSBNAME="$(lsb_release -sc)"
 
+# If 'n/a', assume 'sid'
+if [ "${LSBVERSION}" == "n/a" ] || [ "${LSBNAME}" == "n/a" ]
+then
+  LSBVERSION="sid"
+  LSBNAME="sid"
+fi
+
+# If not known, use 'unknown' in .deb version identifier
 if [ -z "${LSBID}" ]
 then
     LSBID="unknown"
 fi
-case "${LSBNAME}" in
+
+case "${LSBNAME}"
+in
+  # Debian
   stretch)
     # MDEV-16525 libzstd-dev-1.1.3 minimum version
     sed -e '/libzstd-dev/d' \
@@ -113,10 +132,10 @@ case "${LSBNAME}" in
     fi
     ;&
   sid)
-    # should always be empty here.
+    # The default packaging should always target Debian Sid, so in this case
-    # need to match here to avoid the default Error however
+    # there is intentionally no customizations whatsoever.
     ;;
-    # UBUNTU
+  # Ubuntu
   bionic)
     remove_rocksdb_tools
     [ "$architecture" != amd64 ] && disable_pmem
@@ -143,7 +162,7 @@ case "${LSBNAME}" in
     fi
     ;;
   *)
-    echo "Error - unknown release codename $LSBNAME" >&2
+    echo "Error: Unknown release '$LSBNAME'" >&2
     exit 1
 esac
 
@@ -162,7 +181,7 @@ LOGSTRING="MariaDB build"
 EPOCH="1:"
 VERSION="${EPOCH}${UPSTREAM}${PATCHLEVEL}~${LSBID:0:3}${LSBVERSION}"
 
-dch -b -D ${LSBNAME} -v "${VERSION}" "Automatic build with ${LOGSTRING}." --controlmaint
+dch -b -D "${LSBNAME}" -v "${VERSION}" "Automatic build with ${LOGSTRING}." --controlmaint
 
 echo "Creating package version ${VERSION} ... "
 

debian/mariadb-server-10.8.mariadb.init

@@ -21,9 +21,18 @@ test -x /usr/sbin/mariadbd || exit 0
 
 . /lib/lsb/init-functions
 
-SELF=$(cd $(dirname $0); pwd -P)/$(basename $0)
+SELF=$(cd "$(dirname $0)"; pwd -P)/$(basename $0)
 
-MYADMIN="/usr/bin/mysqladmin --defaults-file=/etc/mysql/debian.cnf"
+if [ -f /usr/bin/mariadb-admin ]
+then
+  MYADMIN="/usr/bin/mariadb-admin --defaults-file=/etc/mysql/debian.cnf"
+elif [ -f /usr/bin/mysqladmin ]
+then
+  MYADMIN="/usr/bin/mysqladmin --defaults-file=/etc/mysql/debian.cnf"
+else
+  log_failure_msg "Command mariadb-admin/mysqladmin not found! This SysV init script depends on it."
+  exit -1
+fi
 
 # priority can be overridden and "-s" adds output to stderr
 ERR_LOGGER="logger -p daemon.err -t /etc/init.d/mariadb -i"

debian/rules

@@ -5,7 +5,13 @@ export DEB_BUILD_HARDENING=1
 
 # enable Debian Hardening
 # see: https://wiki.debian.org/Hardening
-export DEB_BUILD_MAINT_OPTIONS = hardening=+all optimize=-lto
+export DEB_BUILD_MAINT_OPTIONS = hardening=+all
+
+# Disable LTO on Ubuntu, see LP: #1970634 and https://jira.mariadb.org/browse/MDEV-25633
+ifeq ($(shell dpkg-vendor --derives-from Ubuntu && echo yes), yes)
+    export DEB_BUILD_MAINT_OPTIONS += optimize=-lto
+endif
+
 DPKG_EXPORT_BUILDFLAGS = 1
 # Include all defaults, including buildflags.mk
 include /usr/share/dpkg/default.mk
@@ -46,11 +52,6 @@ ifeq (32,$(DEB_HOST_ARCH_BITS))
     CMAKEFLAGS += -DPLUGIN_ROCKSDB=NO
 endif
 
-# ColumnStore can build only on amd64 and arm64
-ifneq (,$(filter $(DEB_HOST_ARCH_CPU),amd64 arm64))
-    CMAKEFLAGS += -DPLUGIN_COLUMNSTORE=NO
-endif
-
 # Cross building requires stack direction instruction
 ifneq ($(DEB_BUILD_ARCH),$(DEB_HOST_ARCH))
     ifneq (,$(filter $(DEB_HOST_ARCH_CPU),alpha amd64 arm arm64 i386 ia64 m68k mips64el mipsel powerpc ppc64 ppc64el riscv64 s390x sh4 sparc64))
@@ -64,7 +65,7 @@ endif
 # Only attempt to build with PMEM on archs that have package libpmem-dev available
 # See https://packages.debian.org/search?searchon=names&keywords=libpmem-dev
 ifneq (,$(filter $(DEB_HOST_ARCH_CPU),amd64 arm64 ppc64el riscv64))
-		CMAKEFLAGS += -DWITH_PMEM=yes
+    CMAKEFLAGS += -DWITH_PMEM=YES
 endif
 
 # Add support for verbose builds
@@ -107,7 +108,7 @@ endif
 	    -DPLUGIN_AWS_KEY_MANAGEMENT=NO \
 	    -DPLUGIN_COLUMNSTORE=NO \
 	    -DIGNORE_AIO_CHECK=YES \
-	    -DWITH_URING=yes \
+	    -DWITH_URING=YES \
 	    -DDEB=$(DEB_VENDOR)
 
 # This is needed, otherwise 'make test' will run before binaries have been built
@@ -120,6 +121,7 @@ override_dh_auto_build:
 override_dh_auto_test:
 	@echo "RULES.$@"
 	dh_testdir
+	# Ensure at least an empty file exists
 	touch mysql-test/unstable-tests
 	[ ! -f debian/unstable-tests.$(DEB_HOST_ARCH) ] || cat debian/unstable-tests.$(DEB_HOST_ARCH) >> mysql-test/unstable-tests
 	# Run testsuite
@@ -203,6 +205,7 @@ override_dh_installinit-arch:
 	dh_installinit --name=mariadb --no-start -- defaults 19 21
 	dh_systemd_start --restart-after-upgrade
 
+# Use custom server version string variable
 override_dh_gencontrol:
 	dh_gencontrol -- -Tdebian/substvars
 

debian/salsa-ci.yml

@@ -16,12 +16,17 @@ variables:
   GIT_SUBMODULE_STRATEGY: recursive
   SALSA_CI_GBP_BUILDPACKAGE_ARGS: "--git-submodules" # did not apply to extract-sources
 
+# Extend Salsa-CI build jobs to have longer timeout as the default GitLab
+# timeout (1h) is often not enough
+.build-package:
+  timeout: 3h
+
 stages:
   - provisioning
   - build
   - test
   - upgrade in Sid
-  - upgrade from Bullseye/Buster
+  - upgrade from Bullseye
   - upgrade extras
   - test extras
   - publish # Stage referenced by Salsa-CI template aptly stanza, so must exist even though not used
@@ -134,6 +139,17 @@ blhc:
   apt-get update
   apt-get install -y apt
 
+.test-enable-buster-backports-repos: &test-enable-buster-backports-repos |
+  # Enable buster-backports (assumes environment already Debian Buster)
+  echo 'deb http://deb.debian.org/debian buster-backports main' > /etc/apt/sources.list.d/buster-backports.list
+  # Increase default backports priority policy from '100' to '500' so it can actually be used
+  cat << EOF > /etc/apt/preferences.d/enable-backports-to-satisfy-dependencies
+  Package: *
+  Pin: release n=buster-*
+  Pin-Priority: 500
+  EOF
+  apt-get update
+
 .test-install: &test-install |
   # Install MariaDB built in this commit
   apt-get install -y ./*.deb
@@ -213,7 +229,7 @@ fresh install:
   script:
     - *test-prepare-container
     - *test-install
-    - service mariadb status # There is no init.d/mysql in MariaDB 10.8
+    - service mariadb status # There is no init.d/mysql in MariaDB 10.5+
     - *test-verify-final
   variables:
     GIT_STRATEGY: none
@@ -234,7 +250,7 @@ mariadb-10.8 Sid upgrade:
   script:
     - *test-prepare-container
     - *test-install
-    - service mariadb status # There is no init.d/mysql in MariaDB 10.8
+    - service mariadb status # There is no init.d/mysql in MariaDB 10.5+
     - *test-verify-final
   variables:
     GIT_STRATEGY: none
@@ -243,7 +259,7 @@ mariadb-10.8 Sid upgrade:
       - $CI_COMMIT_TAG != null && $SALSA_CI_ENABLE_PIPELINE_ON_TAGS !~ /^(1|yes|true)$/
 
 mariadb-10.5 Bullseye to mariadb-10.8 upgrade:
-  stage: upgrade from Bullseye/Buster
+  stage: upgrade from Bullseye
   needs:
     - job: build
   image: debian:bullseye
@@ -268,10 +284,12 @@ mariadb-10.5 Bullseye to mariadb-10.8 upgrade:
     variables:
       - $CI_COMMIT_TAG != null && $SALSA_CI_ENABLE_PIPELINE_ON_TAGS !~ /^(1|yes|true)$/
 
-mariadb-10.3 Buster to mariadb-10.8 upgrade:
+# Upgrade of libcrypt.so.1 no longer possible from Buster to Sid,
-  stage: upgrade from Bullseye/Buster
+# so test upgrade only inside Buster (https://bugs.debian.org/993755)
+mariadb-10.3 to mariadb-10.8 upgrade in Buster:
+  stage: upgrade extras
   needs:
-    - job: build
+    - job: build buster-backports
   image: debian:buster
   artifacts:
     when: always
@@ -284,7 +302,7 @@ mariadb-10.3 Buster to mariadb-10.8 upgrade:
     - apt-get install -y 'default-mysql*' 'mariadb-*' 'libmariadb*'
     # Verify installation of MariaDB from Buster
     - *test-verify-initial
-    - *test-enable-sid-repos
+    - *test-enable-buster-backports-repos
     - *test-install
     - service mysql status
     - *test-verify-final
@@ -307,7 +325,7 @@ test basic features:
   script:
     - *test-prepare-container
     - *test-install
-    - service mariadb status # There is no init.d/mysql in MariaDB 10.8
+    - service mariadb status # There is no init.d/mysql in MariaDB 10.5+
     - *test-verify-final
     - |
       # Print info about server
@@ -452,7 +470,7 @@ default-libmysqlclient-dev Sid upgrade:
       - $CI_COMMIT_TAG != null && $SALSA_CI_ENABLE_PIPELINE_ON_TAGS !~ /^(1|yes|true)$/
 
 default-libmysqlclient-dev Bullseye upgrade:
-  stage: upgrade from Bullseye/Buster
+  stage: upgrade from Bullseye
   needs:
     - job: build
   image: debian:bullseye
@@ -472,10 +490,12 @@ default-libmysqlclient-dev Bullseye upgrade:
     variables:
       - $CI_COMMIT_TAG != null && $SALSA_CI_ENABLE_PIPELINE_ON_TAGS !~ /^(1|yes|true)$/
 
-default-libmysqlclient-dev Buster upgrade:
+# Upgrade of libcrypt.so.1 no longer possible from Buster to Sid,
-  stage: upgrade from Bullseye/Buster
+# so test upgrade only inside Buster (https://bugs.debian.org/993755)
+default-libmysqlclient-dev upgrade in Buster:
+  stage: upgrade extras
   needs:
-    - job: build
+    - job: build buster-backports
   image: debian:buster
   artifacts:
     when: always
@@ -486,7 +506,7 @@ default-libmysqlclient-dev Buster upgrade:
     - *test-prepare-container
     - apt-get install -y pkg-config default-libmysqlclient-dev
     - pkg-config --list-all
-    - *test-enable-sid-repos
+    - *test-enable-buster-backports-repos
     - *test-install-all-libs
     - *test-verify-libs
   except:
@@ -511,8 +531,20 @@ mysql-8.0 Sid to mariadb-10.8 upgrade:
     - apt-get install -y procps mysql-server 'libmysqlc*'
     - *test-verify-initial
     - *test-install
-    - service mysql status
+    # The Debian version of MariaDB 10.6 still maintains compatibility and there
+    # running 'service mysql status' in Salsa-CI job 'mysql-8.0 Sid to
+    # mariadb-10.6 upgrade' still works.
+    #
+    # However, due to debian/control changes, the upstream MariaDB 10.6 when
+    # installed on a system with a previous installation of MySQL 8.0 will first
+    # fully remove MySQL, including the /etc/init.d/mysql file, so previous
+    # techniques in mariadb-server-10.6.postinst to maintain backwards
+    # compatibility with 'service mysql status' after installing MariaDB on top
+    # MySQL no longer works, and thus the step to test it now intentionally has
+    # a fallback to use the service name 'mariadb' instead, and the fallback is
+    # always used.
     - sleep 5 # Give the mysql_upgrade a bit of time to complete before querying the server
+    - service mysql status || service mariadb status
     - *test-verify-final
   variables:
     GIT_STRATEGY: none
@@ -522,7 +554,7 @@ mysql-8.0 Sid to mariadb-10.8 upgrade:
 
 # Upgrading from MySQL 8.0 with datadir in place is not possible. Users need to do a data dump.
 # The Debian maintainer scripts detect this situation and simply moves old datadir aside and start fresh.
-mysql-8.0 Focal to mariadb-10.8 upgrade:
+mysql-8.0 Focal to mariadb-10.8 upgrade in Buster:
   stage: upgrade extras
   needs:
     - job: build buster-backports
@@ -622,18 +654,13 @@ mariadb.org-10.6 to mariadb-10.8 upgrade:
     # Verify installation of MariaDB built in this commit
     - dpkg -l | grep -iE 'maria|mysql|galera' || true # List installed
     - mariadb --version # Client version
-    - service mariadb status # There is no init.d/mysql in MariaDB 10.8
+    - service mariadb status # There is no init.d/mysql in MariaDB 10.5+
     - *test-verify-final
   variables:
     GIT_STRATEGY: none
   except:
     variables:
       - $CI_COMMIT_TAG != null && $SALSA_CI_ENABLE_PIPELINE_ON_TAGS !~ /^(1|yes|true)$/
-  allow_failure: true
-  # Installation on Sid fails on missing liburing1 because upstream 10.8
-  # MariaDB.org buildbot has not run 'apt upgrade' for a long time.
-  # Remove this allow_failure once buildbot has built a new 10.8
-  # release using latest liburing-dev in Debian Sid.
 
 mariadb.org-10.5 to mariadb-10.8 upgrade:
   stage: upgrade extras
@@ -657,7 +684,7 @@ mariadb.org-10.5 to mariadb-10.8 upgrade:
     # Verify installation of MariaDB built in this commit
     - dpkg -l | grep -iE 'maria|mysql|galera' || true # List installed
     - mariadb --version # Client version
-    - service mariadb status # There is no init.d/mysql in MariaDB 10.5
+    - service mariadb status # There is no init.d/mysql in MariaDB 10.5+
     - *test-verify-final
   variables:
     GIT_STRATEGY: none
@@ -768,7 +795,7 @@ mariadb.org-10.2 to mariadb-10.8 upgrade:
     variables:
       - $CI_COMMIT_TAG != null && $SALSA_CI_ENABLE_PIPELINE_ON_TAGS !~ /^(1|yes|true)$/
 
-mysql.com-5.7 to mariadb-10.8 upgrade:
+mysql.com-5.7 to mariadb-10.8 upgrade in Buster:
   stage: upgrade extras
   needs:
     - job: build buster-backports
@@ -799,7 +826,7 @@ mysql.com-5.7 to mariadb-10.8 upgrade:
     variables:
       - $CI_COMMIT_TAG != null && $SALSA_CI_ENABLE_PIPELINE_ON_TAGS !~ /^(1|yes|true)$/
 
-percona-xtradb-5.7 to mariadb-10.8 upgrade (MDEV-22679):
+percona-xtradb-5.7 to mariadb-10.8 upgrade in Buster (MDEV-22679):
   stage: upgrade extras
   needs:
     - job: build buster-backports

debian/source/lintian-overrides

@@ -48,7 +48,7 @@ very-long-line-length-in-source-file mysql-test/std_data/init_file_longline_3816
 very-long-line-length-in-source-file scripts/fill_help_tables.sql *
 very-long-line-length-in-source-file scripts/mysql_system_tables.sql *
 very-long-line-length-in-source-file scripts/mysql_test_data_timezone.sql *
-# Machine formated HTML
+# Machine formatted HTML
 very-long-line-length-in-source-file sql/share/charsets/languages.html *
 very-long-line-length-in-source-file sql/share/errmsg-utf8.txt *
 # Very long test string

debian/tests/upstream

@@ -10,6 +10,8 @@ echo "Running test 'testsuite'"
 set -e
 
 SKIP_TEST_LST="/tmp/skip-test.lst"
+ARCH=$(dpkg --print-architecture)
+
 WORKDIR=$(mktemp -d)
 trap 'rm -rf $WORKDIR $SKIP_TEST_LST' 0 INT QUIT ABRT PIPE TERM
 cd "$WORKDIR"
@@ -22,16 +24,15 @@ echo "using tmpdir: $WORKDIR/tmp"
 
 echo "Setting up skip-tests-list"
 
-touch $SKIP_TEST_LST
+# Use unstable-tests list as base to skip all tests  considered unstable
+# or create an empty file if that upstream file does not exists on this branch
+cp /usr/share/mysql/mysql-test/unstable-tests $SKIP_TEST_LST || touch $SKIP_TEST_LST
 
-# Also use arch specific skiplists if such files exist
+# Also use the arch specific skiplists if exist
-for filename in /usr/share/mysql/mysql-test/unstable-tests.*
+if [ -f /usr/share/mysql/mysql-test/unstable-tests.$ARCH ]
-do
+then
-  # Check for case that no files matched and glob is returned
+  cat /usr/share/mysql/mysql-test/unstable-tests.$ARCH >> $SKIP_TEST_LST
-  [ -e "$filename" ] || continue
+fi
-  # Append file to the main skip test list file
-  cat "$filename" >> $SKIP_TEST_LST
-done
 
 # Skip tests that cannot run properly on ci.debian.net / autopkgtests.ubuntu.com
 cat >> $SKIP_TEST_LST << EOF
@@ -48,7 +49,6 @@ main.mysqld--help : For unknown reason table-cache is 4000 instead of default 42
 EOF
 fi
 
-ARCH=$(dpkg --print-architecture)
 if [ "$ARCH" = "s390x" ]
 then
   echo "main.func_regexp_pcre : recursion fails on s390x https://bugs.launchpad.net/ubuntu/+source/mariadb-10.1/+bug/1723947" >> $SKIP_TEST_LST
@@ -57,6 +57,10 @@ then
   echo "main.failed_auth_unixsocket : Test returns wrong exit code on armhf and i386 (but only in debci) https://jira.mariadb.org/browse/MDEV-23933" >> $SKIP_TEST_LST
 fi
 
+# Store skipped test list in artifacts so it can be viewed while debugging
+# failed autopkgtest runs
+cp -v $SKIP_TEST_LST $AUTOPKGTEST_ARTIFACTS
+
 cd /usr/share/mysql/mysql-test
 echo "starting mysql-test-tun.pl..."
 eatmydata perl -I. ./mysql-test-run.pl --suite=main \

mysql-test/suite/binlog/r/binlog_empty_xa_prepared.result

@@ -1,3 +1,4 @@
+RESET MASTER;
 CREATE TEMPORARY SEQUENCE seq_1;
 XA START '3';
 CREATE TEMPORARY TABLE tmp_1(c INT);

mysql-test/suite/binlog/r/binlog_table_map_optional_metadata_binary.result

@@ -1,3 +1,4 @@
+RESET MASTER;
 #
 # Verify that SET string values and character sets can be printed correctly
 #

mysql-test/suite/binlog/r/binlog_table_map_optional_metadata_ucs2.result

@@ -1,3 +1,4 @@
+RESET MASTER;
 #
 # Verify that SET string values and character sets can be printed correctly
 #

mysql-test/suite/binlog/r/binlog_table_map_optional_metadata_utf32.result

@@ -1,3 +1,4 @@
+RESET MASTER;
 #
 # Verify that SET string values and character sets can be printed correctly
 #

mysql-test/suite/binlog/t/binlog_empty_xa_prepared.test

@@ -4,7 +4,7 @@
 --source include/have_binlog_format_mixed.inc
 --source include/have_innodb.inc
 
-
+RESET MASTER; # clear binlogs
 # MDEV-22420 DDL on temporary object is prohibited when XA is in prepare state
 
 # Temporary sequnce may not be created within a transaction

mysql-test/suite/binlog/t/binlog_table_map_optional_metadata_binary.test

@@ -20,6 +20,7 @@
 --let $MYSQLD_DATADIR= `select @@datadir`
 --let $binlog_file= $MYSQLD_DATADIR/master-bin.000001
 
+RESET MASTER;
 --echo #
 --echo # Verify that SET string values and character sets can be printed correctly
 --echo #

mysql-test/suite/binlog/t/binlog_table_map_optional_metadata_ucs2.test

@@ -21,6 +21,7 @@
 --let $MYSQLD_DATADIR= `select @@datadir`
 --let $binlog_file= $MYSQLD_DATADIR/master-bin.000001
 
+RESET MASTER;
 --echo #
 --echo # Verify that SET string values and character sets can be printed correctly
 --echo #

mysql-test/suite/binlog/t/binlog_table_map_optional_metadata_utf32.test

@@ -21,6 +21,7 @@
 --let $MYSQLD_DATADIR= `select @@datadir`
 --let $binlog_file= $MYSQLD_DATADIR/master-bin.000001
 
+RESET MASTER;
 --echo #
 --echo # Verify that SET string values and character sets can be printed correctly
 --echo #

mysql-test/suite/compat/oracle/r/sp-package.result

@@ -3268,3 +3268,78 @@ a
 This is db1.pkg1.p1
 DROP DATABASE db1;
 DROP DATABASE db2;
+#
+# MDEV-29370 Functions in packages are slow and seems to ignore deterministic
+#
+SET SQL_MODE=ORACLE;
+CREATE TABLE t1 (c1 CHAR(1));
+CREATE FUNCTION f1_deterministic()
+RETURN CHAR(1)
+DETERMINISTIC
+IS
+BEGIN
+RETURN 'X';
+END;
+//
+CREATE FUNCTION f2_not_deterministic()
+RETURN CHAR(1)
+IS
+BEGIN
+RETURN 'X';
+END;
+//
+CREATE PACKAGE pkg1
+IS
+PROCEDURE t1_populate(numrows INTEGER);
+FUNCTION f3_deterministic() RETURN CHAR(1) DETERMINISTIC;
+FUNCTION f4_not_deterministic() RETURN CHAR(1);
+END;
+//
+CREATE PACKAGE BODY pkg1
+IS
+PROCEDURE t1_populate(numrounds INTEGER)
+IS
+i INTEGER;
+BEGIN
+INSERT INTO t1 VALUES('Y');
+FOR i IN 1..numrounds LOOP
+INSERT INTO t1 SELECT * FROM t1;
+END LOOP;
+END;
+FUNCTION f3_deterministic() RETURN CHAR(1) DETERMINISTIC COMMENT 'xxx'
+  IS
+BEGIN
+RETURN 'X';
+END;
+FUNCTION f4_not_deterministic() RETURN CHAR(1)
+IS
+BEGIN
+RETURN 'X';
+END;
+END;
+//
+CALL pkg1.t1_populate(3);
+EXPLAIN EXTENDED SELECT 'Deterministic function', COUNT(*) FROM t1 WHERE c1 = f1_deterministic();
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	filtered	Extra
+1	SIMPLE	t1	ALL	NULL	NULL	NULL	NULL	8	100.00	Using where
+Warnings:
+Note	1003	select 'Deterministic function' AS "Deterministic function",count(0) AS "COUNT(*)" from "test"."t1" where "test"."t1"."c1" = <cache>("f1_deterministic"())
+EXPLAIN EXTENDED SELECT 'Non-deterministic function', COUNT(*) FROM t1 WHERE c1 = f2_not_deterministic();
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	filtered	Extra
+1	SIMPLE	t1	ALL	NULL	NULL	NULL	NULL	8	100.00	Using where
+Warnings:
+Note	1003	select 'Non-deterministic function' AS "Non-deterministic function",count(0) AS "COUNT(*)" from "test"."t1" where "test"."t1"."c1" = "f2_not_deterministic"()
+EXPLAIN EXTENDED SELECT 'Deterministic package function', COUNT(*) FROM t1 WHERE c1 = pkg1.f3_deterministic();
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	filtered	Extra
+1	SIMPLE	t1	ALL	NULL	NULL	NULL	NULL	8	100.00	Using where
+Warnings:
+Note	1003	select 'Deterministic package function' AS "Deterministic package function",count(0) AS "COUNT(*)" from "test"."t1" where "test"."t1"."c1" = <cache>("test"."pkg1"."f3_deterministic"())
+EXPLAIN EXTENDED SELECT 'Non-deterministic package function', COUNT(*) FROM t1 WHERE c1 = pkg1.f4_not_deterministic();
+id	select_type	table	type	possible_keys	key	key_len	ref	rows	filtered	Extra
+1	SIMPLE	t1	ALL	NULL	NULL	NULL	NULL	8	100.00	Using where
+Warnings:
+Note	1003	select 'Non-deterministic package function' AS "Non-deterministic package function",count(0) AS "COUNT(*)" from "test"."t1" where "test"."t1"."c1" = "test"."pkg1"."f4_not_deterministic"()
+DROP TABLE t1;
+DROP FUNCTION f1_deterministic;
+DROP FUNCTION f2_not_deterministic;
+DROP PACKAGE pkg1;

mysql-test/suite/compat/oracle/t/sp-package.test

@@ -3016,3 +3016,75 @@ CALL db2.pkg1.p2_db1_pkg1_p1;
 
 DROP DATABASE db1;
 DROP DATABASE db2;
+
+
+--echo #
+--echo # MDEV-29370 Functions in packages are slow and seems to ignore deterministic
+--echo #
+
+SET SQL_MODE=ORACLE;
+
+CREATE TABLE t1 (c1 CHAR(1));
+
+DELIMITER //;
+CREATE FUNCTION f1_deterministic()
+RETURN CHAR(1)
+DETERMINISTIC
+IS
+BEGIN
+   RETURN 'X';
+END;
+//
+
+CREATE FUNCTION f2_not_deterministic()
+RETURN CHAR(1)
+IS
+BEGIN
+   RETURN 'X';
+END;
+//
+
+CREATE PACKAGE pkg1
+IS
+  PROCEDURE t1_populate(numrows INTEGER);
+  FUNCTION f3_deterministic() RETURN CHAR(1) DETERMINISTIC;
+  FUNCTION f4_not_deterministic() RETURN CHAR(1);
+END;
+//
+
+CREATE PACKAGE BODY pkg1
+IS
+  PROCEDURE t1_populate(numrounds INTEGER)
+  IS
+    i INTEGER;
+  BEGIN
+    INSERT INTO t1 VALUES('Y');
+    FOR i IN 1..numrounds LOOP
+      INSERT INTO t1 SELECT * FROM t1;
+    END LOOP;
+  END;
+  FUNCTION f3_deterministic() RETURN CHAR(1) DETERMINISTIC COMMENT 'xxx'
+  IS
+  BEGIN
+     RETURN 'X';
+  END;
+  FUNCTION f4_not_deterministic() RETURN CHAR(1)
+  IS
+  BEGIN
+     RETURN 'X';
+  END;
+END;
+//
+DELIMITER ;//
+
+CALL pkg1.t1_populate(3);
+
+EXPLAIN EXTENDED SELECT 'Deterministic function', COUNT(*) FROM t1 WHERE c1 = f1_deterministic();
+EXPLAIN EXTENDED SELECT 'Non-deterministic function', COUNT(*) FROM t1 WHERE c1 = f2_not_deterministic();
+EXPLAIN EXTENDED SELECT 'Deterministic package function', COUNT(*) FROM t1 WHERE c1 = pkg1.f3_deterministic();
+EXPLAIN EXTENDED SELECT 'Non-deterministic package function', COUNT(*) FROM t1 WHERE c1 = pkg1.f4_not_deterministic();
+
+DROP TABLE t1;
+DROP FUNCTION f1_deterministic;
+DROP FUNCTION f2_not_deterministic;
+DROP PACKAGE pkg1;

mysql-test/suite/encryption/t/innochecksum.test

@@ -9,6 +9,7 @@
 -- source include/have_file_key_management_plugin.inc
 -- source include/innodb_page_size_small.inc
 -- source include/innodb_checksum_algorithm.inc
+-- source include/maybe_debug.inc
 
 if (!$INNOCHECKSUM) {
   --echo Need innochecksum binary
@@ -18,6 +19,10 @@ if (!$INNOCHECKSUM) {
 --disable_query_log
 # This may be triggered on a slow system or one that lacks native AIO.
 call mtr.add_suppression("InnoDB: Trying to delete tablespace.*pending operations");
+if ($have_debug) {
+SET GLOBAL DEBUG_DBUG='+d,ib_log_checkpoint_avoid_hard';
+call mtr.add_suppression("InnoDB: Crash recovery is broken due to insufficient innodb_log_file_size");
+}
 --enable_query_log
 
 let $checksum_algorithm = `SELECT @@innodb_checksum_algorithm`;
@@ -259,6 +264,15 @@ print FILE pack("H*", "c00lcafedeadb017");
 close FILE or die "close";
 EOF
 
+if (0 && $have_debug) { # these messages sometimes fail to appear
+--let SEARCH_FILE= $MYSQLTEST_VARDIR/log/mysqld.1.err
+--let SEARCH_PATTERN= InnoDB: Crash recovery is broken due to insufficient innodb_log_file_size; last checkpoint LSN=\\d+, current LSN=\\d+\\. Shutdown is in progress\\..*InnoDB: Crash recovery was broken.*
+--echo # FOUND 1 is expected for both.
+--source include/search_pattern_in_file.inc
+--let SEARCH_PATTERN= InnoDB: Crash recovery was broken
+--source include/search_pattern_in_file.inc
+}
+
 -- disable_result_log
 --error 1
 --exec $INNOCHECKSUM $t1_IBD

mysql-test/suite/innodb/r/insert_into_empty,32k.rdiff

@@ -0,0 +1,9 @@
+@@ -377,8 +377,6 @@
+ c09 text, c10 text, c11 text, c12 text) ENGINE=InnoDB;
+ SET GLOBAL INNODB_DEFAULT_ROW_FORMAT= COMPACT;
+ ALTER TABLE t1 FORCE;
+-Warnings:
+-Warning	139	Row size too large (> 8126). Changing some columns to TEXT or BLOB or using ROW_FORMAT=DYNAMIC or ROW_FORMAT=COMPRESSED may help. In current row format, BLOB prefix of 768 bytes is stored inline.
+ INSERT IGNORE INTO t1 VALUES
+ (1, REPEAT('x',4805), REPEAT('t',2211), REPEAT('u',974), REPEAT('e',871), REPEAT('z',224), REPEAT('j',978), REPEAT('n',190), REPEAT('t',888), REPEAT('x',32768), REPEAT('e',968), REPEAT('b',913), REPEAT('x',12107)),
+ (2, REPEAT('x',4805), REPEAT('t',2211), REPEAT('u',974), REPEAT('e',871), REPEAT('z',224), REPEAT('j',978), REPEAT('n',190), REPEAT('t',888), REPEAT('x',32768), REPEAT('e',968), REPEAT('b',913), REPEAT('x',12107));

mysql-test/suite/innodb/r/insert_into_empty,64k.rdiff

@@ -0,0 +1,9 @@
+@@ -377,8 +377,6 @@
+ c09 text, c10 text, c11 text, c12 text) ENGINE=InnoDB;
+ SET GLOBAL INNODB_DEFAULT_ROW_FORMAT= COMPACT;
+ ALTER TABLE t1 FORCE;
+-Warnings:
+-Warning	139	Row size too large (> 8126). Changing some columns to TEXT or BLOB or using ROW_FORMAT=DYNAMIC or ROW_FORMAT=COMPRESSED may help. In current row format, BLOB prefix of 768 bytes is stored inline.
+ INSERT IGNORE INTO t1 VALUES
+ (1, REPEAT('x',4805), REPEAT('t',2211), REPEAT('u',974), REPEAT('e',871), REPEAT('z',224), REPEAT('j',978), REPEAT('n',190), REPEAT('t',888), REPEAT('x',32768), REPEAT('e',968), REPEAT('b',913), REPEAT('x',12107)),
+ (2, REPEAT('x',4805), REPEAT('t',2211), REPEAT('u',974), REPEAT('e',871), REPEAT('z',224), REPEAT('j',978), REPEAT('n',190), REPEAT('t',888), REPEAT('x',32768), REPEAT('e',968), REPEAT('b',913), REPEAT('x',12107));

mysql-test/suite/innodb/r/insert_into_empty.result

@@ -186,12 +186,31 @@ DROP TABLE t;
 # MDEV-28327 InnoDB persistent statistics fail to update
 #		after bulk insert
 #
-CREATE TABLE t1 (a INT PRIMARY KEY)ENGINE=InnoDB;
+CREATE TABLE t1 (a INT PRIMARY KEY)ENGINE=InnoDB
+STATS_PERSISTENT=1 STATS_AUTO_RECALC=1;
 INSERT INTO t1 SELECT * FROM seq_1_to_4096;
 # Wait till statistics update after bulk insert operation
-SELECT n_rows FROM mysql.innodb_table_stats WHERE TABLE_NAME="t1";
+SELECT n_rows>=4000 FROM mysql.innodb_table_stats WHERE TABLE_NAME="t1";
-n_rows
+n_rows>=4000
-4096
+1
+DROP TABLE t1;
+#
+# MDEV-27214 Import with disabled keys corrupts meta-data like rows, indexes, ...
+#
+SET UNIQUE_CHECKS=0;
+SET FOREIGN_KEY_CHECKS=0;
+CREATE TABLE `t1` (
+`id` int(11) NOT NULL,
+`a` int(11) DEFAULT NULL,
+PRIMARY KEY (`id`)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1
+STATS_PERSISTENT=1 STATS_AUTO_RECALC=1;
+INSERT INTO `t1` VALUES (1,2),(2,3),(3,4);
+# Wait till statistics update after bulk insert operation
+SELECT TABLE_ROWS, AVG_ROW_LENGTH>0 FROM INFORMATION_SCHEMA.TABLES
+WHERE TABLE_NAME='t1' AND TABLE_SCHEMA='test';
+TABLE_ROWS	AVG_ROW_LENGTH>0
+3	1
 DROP TABLE t1;
 # End of 10.6 tests
 #
@@ -351,7 +370,7 @@ DROP TABLE t1, t2;
 #   MDEV-29801  Inconsistent ER_TOO_BIG_ROWSIZE during bulk
 #			insert operation
 #
-call mtr.add_suppression("InnoDB: Cannot add field `c11` in table");
+call mtr.add_suppression("InnoDB: Cannot add field `(c0[36]|c11)` in table");
 SET @format= @@innodb_default_row_format;
 CREATE TABLE t1 (pk int primary key, c01 text, c02 text, c03 text,
 c04 text, c05 text, c06 text, c07 text, c08 text,

mysql-test/suite/innodb/r/insert_into_empty_notembedded.result

@@ -0,0 +1,30 @@
+#
+# Start of 10.6 tests
+#
+#
+# MDEV-27214 Import with disabled keys corrupts meta-data like rows, indexes, ...
+#
+CREATE DATABASE db1;
+CREATE TABLE db1.t1 (id int, a int,PRIMARY KEY (id)) ENGINE=InnoDB
+STATS_PERSISTENT=1 STATS_AUTO_RECALC=1;
+INSERT INTO db1.t1 VALUES (1,2),(2,3),(3,4);
+DROP DATABASE IF EXISTS db1;
+CREATE DATABASE db1;
+# Wait till statistics update after bulk insert operation
+SELECT TABLE_ROWS, AVG_ROW_LENGTH>0 FROM INFORMATION_SCHEMA.TABLES
+WHERE TABLE_NAME='t1' AND TABLE_SCHEMA='db1';
+TABLE_ROWS	AVG_ROW_LENGTH>0
+3	1
+OPTIMIZE TABLE db1.t1;
+Table	Op	Msg_type	Msg_text
+db1.t1	optimize	note	Table does not support optimize, doing recreate + analyze instead
+db1.t1	optimize	status	OK
+# Wait till statistics update after bulk insert operation
+SELECT TABLE_ROWS, AVG_ROW_LENGTH>0 FROM INFORMATION_SCHEMA.TABLES
+WHERE TABLE_NAME='t1' AND TABLE_SCHEMA='db1';
+TABLE_ROWS	AVG_ROW_LENGTH>0
+3	1
+DROP DATABASE db1;
+#
+# End of 10.6 tests
+#

mysql-test/suite/innodb/r/temporary_table.result

@@ -789,4 +789,16 @@ CHECK TABLE t EXTENDED;
 Table	Op	Msg_type	Msg_text
 test.t	check	status	OK
 DROP TEMPORARY TABLE t;
+#
+# MDEV-29978 Corruption errors upon CHECK on temporary InnoDB table
+#
+CREATE TEMPORARY TABLE t (f INT UNIQUE) ENGINE=InnoDB;
+INSERT INTO t (f) VALUES (1),(2);
+CHECK TABLE t;
+Table	Op	Msg_type	Msg_text
+test.t	check	status	OK
+CHECK TABLE t EXTENDED;
+Table	Op	Msg_type	Msg_text
+test.t	check	status	OK
+DROP TEMPORARY TABLE t;
 # End of 10.6 tests

mysql-test/suite/innodb/t/insert_into_empty.test

@@ -1,5 +1,5 @@
 --source include/have_innodb.inc
-
+--source include/innodb_page_size.inc
 --source include/have_sequence.inc
 --source include/maybe_debug.inc
 --source include/have_partition.inc
@@ -199,15 +199,41 @@ DROP TABLE t;
 --echo # MDEV-28327 InnoDB persistent statistics fail to update
 --echo #		after bulk insert
 --echo #
-CREATE TABLE t1 (a INT PRIMARY KEY)ENGINE=InnoDB;
+CREATE TABLE t1 (a INT PRIMARY KEY)ENGINE=InnoDB
+STATS_PERSISTENT=1 STATS_AUTO_RECALC=1;
 INSERT INTO t1 SELECT * FROM seq_1_to_4096;
 --echo # Wait till statistics update after bulk insert operation
 let $wait_condition= select n_rows > 100 from mysql.innodb_table_stats
 where table_name="t1";
 source include/wait_condition.inc;
-SELECT n_rows FROM mysql.innodb_table_stats WHERE TABLE_NAME="t1";
+# At innodb_page_size=4k this will be only 4075, not 4096. Add some slack.
+# This is related to MDEV-24621 and possibly MDEV-26740.
+SELECT n_rows>=4000 FROM mysql.innodb_table_stats WHERE TABLE_NAME="t1";
 DROP TABLE t1;
 
+
+--echo #
+--echo # MDEV-27214 Import with disabled keys corrupts meta-data like rows, indexes, ...
+--echo #
+
+SET UNIQUE_CHECKS=0;
+SET FOREIGN_KEY_CHECKS=0;
+CREATE TABLE `t1` (
+  `id` int(11) NOT NULL,
+  `a` int(11) DEFAULT NULL,
+  PRIMARY KEY (`id`)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1
+STATS_PERSISTENT=1 STATS_AUTO_RECALC=1;
+INSERT INTO `t1` VALUES (1,2),(2,3),(3,4);
+--echo # Wait till statistics update after bulk insert operation
+let $wait_condition= select n_rows > 0 from mysql.innodb_table_stats
+where database_name='test' and table_name='t1';
+source include/wait_condition.inc;
+SELECT TABLE_ROWS, AVG_ROW_LENGTH>0 FROM INFORMATION_SCHEMA.TABLES
+WHERE TABLE_NAME='t1' AND TABLE_SCHEMA='test';
+DROP TABLE t1;
+
+
 --echo # End of 10.6 tests
 
 --echo #
@@ -363,13 +389,14 @@ DROP TABLE t1, t2;
 --echo #   MDEV-29801  Inconsistent ER_TOO_BIG_ROWSIZE during bulk
 --echo #			insert operation
 --echo #
-call mtr.add_suppression("InnoDB: Cannot add field `c11` in table");
+call mtr.add_suppression("InnoDB: Cannot add field `(c0[36]|c11)` in table");
 
 SET @format= @@innodb_default_row_format;
 CREATE TABLE t1 (pk int primary key, c01 text, c02 text, c03 text,
                  c04 text, c05 text, c06 text, c07 text, c08 text,
                  c09 text, c10 text, c11 text, c12 text) ENGINE=InnoDB;
 SET GLOBAL INNODB_DEFAULT_ROW_FORMAT= COMPACT;
+--replace_result 1982 8126 4030 8126
 ALTER TABLE t1 FORCE;
 INSERT IGNORE INTO t1 VALUES
 (1, REPEAT('x',4805), REPEAT('t',2211), REPEAT('u',974), REPEAT('e',871), REPEAT('z',224), REPEAT('j',978), REPEAT('n',190), REPEAT('t',888), REPEAT('x',32768), REPEAT('e',968), REPEAT('b',913), REPEAT('x',12107)),

mysql-test/suite/innodb/t/insert_into_empty_notembedded.test

@@ -0,0 +1,43 @@
+--source include/not_embedded.inc
+--source include/have_innodb.inc
+--source include/innodb_page_size.inc
+--source include/maybe_debug.inc
+
+--echo #
+--echo # Start of 10.6 tests
+--echo #
+
+--echo #
+--echo # MDEV-27214 Import with disabled keys corrupts meta-data like rows, indexes, ...
+--echo #
+
+CREATE DATABASE db1;
+CREATE TABLE db1.t1 (id int, a int,PRIMARY KEY (id)) ENGINE=InnoDB
+STATS_PERSISTENT=1 STATS_AUTO_RECALC=1;
+INSERT INTO db1.t1 VALUES (1,2),(2,3),(3,4);
+--let $file = $MYSQLTEST_VARDIR/tmp/dump.sql
+--exec $MYSQL_DUMP db1 t1 >$file
+DROP DATABASE IF EXISTS db1;
+
+CREATE DATABASE db1;
+--exec $MYSQL db1  < $file
+--remove_file $file
+--echo # Wait till statistics update after bulk insert operation
+let $wait_condition= select n_rows > 0 from mysql.innodb_table_stats
+where database_name='db1' and table_name='t1';
+source include/wait_condition.inc;
+SELECT TABLE_ROWS, AVG_ROW_LENGTH>0 FROM INFORMATION_SCHEMA.TABLES
+WHERE TABLE_NAME='t1' AND TABLE_SCHEMA='db1';
+
+OPTIMIZE TABLE db1.t1;
+--echo # Wait till statistics update after bulk insert operation
+let $wait_condition= select n_rows > 0 from mysql.innodb_table_stats
+where database_name='db1' and table_name='t1';
+source include/wait_condition.inc;
+SELECT TABLE_ROWS, AVG_ROW_LENGTH>0 FROM INFORMATION_SCHEMA.TABLES
+WHERE TABLE_NAME='t1' AND TABLE_SCHEMA='db1';
+DROP DATABASE db1;
+
+--echo #
+--echo # End of 10.6 tests
+--echo #

mysql-test/suite/innodb/t/temporary_table.test

@@ -624,4 +624,15 @@ UPDATE t SET a=2;
 CHECK TABLE t;
 CHECK TABLE t EXTENDED;
 DROP TEMPORARY TABLE t;
+
+--echo #
+--echo # MDEV-29978 Corruption errors upon CHECK on temporary InnoDB table
+--echo #
+
+CREATE TEMPORARY TABLE t (f INT UNIQUE) ENGINE=InnoDB;
+INSERT INTO t (f) VALUES (1),(2);
+CHECK TABLE t;
+CHECK TABLE t EXTENDED;
+DROP TEMPORARY TABLE t;
+
 --echo # End of 10.6 tests

mysql-test/suite/maria/rollback.result

@@ -1,3 +1,4 @@
+reset master;
 call mtr.add_suppression("Table was marked as crashed");
 call mtr.add_suppression("Checking table: .*");
 create table t1 (a int primary key auto_increment, b int) engine=aria transactional= 1;

mysql-test/suite/maria/rollback.test

@@ -3,6 +3,8 @@
 # no-protocol doesn't print warnings about repaired tables
 --source include/no_protocol.inc
 
+reset master; # clear binlogs
+
 call mtr.add_suppression("Table was marked as crashed");
 call mtr.add_suppression("Checking table: .*");
 

mysql-test/suite/mariabackup/defer_space.result

@@ -1,6 +1,7 @@
 call mtr.add_suppression("InnoDB: Expected tablespace id .*");
 # Mariabackup --backup with page0 INIT_PAGE redo record
 # and there is no FILE_CREATE for the tablespace t1
+SET @save_dbug = @@SESSION.debug_dbug;
 SET DEBUG_DBUG="+d,checkpoint_after_file_create";
 CREATE TABLE t1(f1 INT NOT NULL)ENGINE=InnoDB;
 INSERT INTO t1 VALUES(1);
@@ -14,7 +15,7 @@ SELECT * FROM t1;
 f1
 1
 DROP TABLE t1;
-SET DEBUG_DBUG="-d,checkpoint_after_file_create";
+SET @@SESSION.DEBUG_DBUG= @save_debug;
 # Mariabackup fails after corrupting the page0 in disk
 # and there is no INIT_PAGE for page0
 CREATE TABLE t1(c INT) ENGINE=INNODB;

mysql-test/suite/mariabackup/defer_space.test

@@ -6,6 +6,7 @@
 call mtr.add_suppression("InnoDB: Expected tablespace id .*");
 --echo # Mariabackup --backup with page0 INIT_PAGE redo record
 --echo # and there is no FILE_CREATE for the tablespace t1
+SET @save_dbug = @@SESSION.debug_dbug;
 SET DEBUG_DBUG="+d,checkpoint_after_file_create";
 CREATE TABLE t1(f1 INT NOT NULL)ENGINE=InnoDB;
 INSERT INTO t1 VALUES(1);
@@ -26,7 +27,7 @@ exec $XTRABACKUP  --prepare --target-dir=$targetdir;
 SELECT * FROM t1;
 DROP TABLE t1;
 rmdir $targetdir;
-SET DEBUG_DBUG="-d,checkpoint_after_file_create";
+SET @@SESSION.DEBUG_DBUG= @save_debug;
 
 --echo # Mariabackup fails after corrupting the page0 in disk
 --echo # and there is no INIT_PAGE for page0
@@ -39,16 +40,13 @@ let INNODB_PAGE_SIZE=`select @@innodb_page_size`;
 --echo # Corrupt the table
 
 perl;
-use strict;
+my $ps = $ENV{INNODB_PAGE_SIZE};
-use warnings;
-use Fcntl qw(:DEFAULT :seek);
-my $page_size = $ENV{INNODB_PAGE_SIZE};
 
-sysopen FILE, "$ENV{MYSQLD_DATADIR}/test/t1.ibd", O_RDWR
+my $file = "$ENV{MYSQLD_DATADIR}/test/t1.ibd";
-|| die "Cannot open t1.ibd\n";
+open(FILE, "+<$file") || die "Unable to open $file";
-sysseek(FILE, 0, SEEK_SET) || die "Cannot seek t1.ibd\n";
+binmode FILE;
-my $page=chr(0) x $page_size;
+seek (FILE, 0, SEEK_SET) or die "seek";
-syswrite(FILE, $page, $page_size)==$page_size;
+print FILE chr(0x00) x $ps;
 close FILE or die "close";
 EOF
 

mysql-test/suite/sys_vars/t/wsrep_on_without_provider.test

@@ -1,4 +1,5 @@
 --source include/not_embedded.inc
+--source include/have_wsrep.inc
 
 #
 # @@global.wsrep_on is not allowed if there

mysql-test/suite/versioning/r/partition.result

@@ -96,6 +96,16 @@ with system versioning
 partition by system_time (
 partition p0 history,
 partition pn current);
+create or replace table t1 (a int)
+partition by range (a) (
+partition p0 history,
+partition p1 current);
+ERROR HY000: Wrong partition type `SYSTEM_TIME` for partitioning by `RANGE`
+create or replace table t1 (b int)
+partition by range (a) (
+partition p0 current,
+partition p1 history);
+ERROR HY000: Wrong partition type `SYSTEM_TIME` for partitioning by `RANGE`
 ## ALTER TABLE
 alter table t1 add partition (
 partition p1 current);
@@ -150,7 +160,7 @@ partition by system_time limit 1;
 alter table t1 change x big int;
 create or replace table t1 (i int) engine myisam partition by hash(i) partitions 2;
 alter table t1 add partition (partition px history);
-ERROR HY000: Wrong partitioning type, expected type: `SYSTEM_TIME`
+ERROR HY000: Wrong partition type `SYSTEM_TIME` for partitioning by `HASH`
 ## INSERT, UPDATE, DELETE
 create or replace table t1 (x int)
 with system versioning
@@ -1105,7 +1115,7 @@ drop table t1;
 create table t1 (a int) with system versioning partition by system_time
 (partition p1 history, partition pn current);
 alter table t1 add partition (partition p2);
-ERROR HY000: Wrong partitioning type, expected type: `SYSTEM_TIME`
+ERROR HY000: Wrong partition type `HASH` for partitioning by `SYSTEM_TIME`
 # MDEV-17891 Assertion failures in select_insert::abort_result_set and
 #            mysql_load upon attempt to replace into a full table
 set @@max_heap_table_size= 1024*1024;

mysql-test/suite/versioning/t/partition.test

@@ -106,6 +106,18 @@ partition by system_time (
     partition p0 history,
     partition pn current);
 
+--error ER_PARTITION_WRONG_TYPE
+create or replace table t1 (a int)
+partition by range (a) (
+  partition p0 history,
+  partition p1 current);
+
+--error ER_PARTITION_WRONG_TYPE
+create or replace table t1 (b int)
+partition by range (a) (
+  partition p0 current,
+  partition p1 history);
+
 
 --echo ## ALTER TABLE
 

mysys/my_gethwaddr.c

@@ -116,7 +116,7 @@ my_bool my_gethwaddr(uchar *to)
     uint i;
     for (i= 0; res && i < ifc.ifc_len / sizeof(ifr[0]); i++)
     {
-#if !defined(_AIX) || !defined(__linux__)
+#if defined(_AIX) || defined(__linux__)
 #if defined(__linux__)
 #define HWADDR_DATA ifr[i].ifr_hwaddr.sa_data
 #else

sql/partition_info.h

@@ -429,8 +429,13 @@ public:
     return NULL;
   }
   uint next_part_no(uint new_parts) const;
+
+  int gen_part_type(THD *thd, String *str) const;
 };
 
+void part_type_error(THD *thd, partition_info *work_part_info,
+                     const char *part_type, partition_info *tab_part_info);
+
 uint32 get_next_partition_id_range(struct st_partition_iter* part_iter);
 bool check_partition_dirs(partition_info *part_info);
 

sql/share/errmsg-utf8.txt

@@ -9773,9 +9773,9 @@ ER_UNUSED_23
         spa "Nunca debería vd de ver esto"
 
 ER_PARTITION_WRONG_TYPE
-        chi "错误的分区类型，预期类型：%`s"
+        chi "错误的分区类型，预期类型：%`s for partitioning by %`s"
-        eng "Wrong partitioning type, expected type: %`s"
+        eng "Wrong partition type %`s for partitioning by %`s"
-        spa "Tipo de partición equivocada, tipo esperado: %`s"
+        spa "Tipo de partición equivocada, tipo esperado: %`s for partitioning by %`s"
 
 WARN_VERS_PART_FULL
         chi "版本化表%`s.%`s：partition%`s已满，添加更多历史分区(out of %s)"

sql/sql_lex.cc

@@ -9754,6 +9754,7 @@ bool Lex_ident_sys_st::to_size_number(ulonglong *to) const
 }
 
 
+#ifdef WITH_PARTITION_STORAGE_ENGINE
 bool LEX::part_values_current(THD *thd)
 {
   partition_element *elem= part_info->curr_part_elem;
@@ -9761,7 +9762,7 @@ bool LEX::part_values_current(THD *thd)
   {
     if (unlikely(part_info->part_type != VERSIONING_PARTITION))
     {
-      my_error(ER_PARTITION_WRONG_TYPE, MYF(0), "SYSTEM_TIME");
+      part_type_error(thd, NULL, "SYSTEM_TIME", part_info);
       return true;
     }
   }
@@ -9788,7 +9789,7 @@ bool LEX::part_values_history(THD *thd)
   {
     if (unlikely(part_info->part_type != VERSIONING_PARTITION))
     {
-      my_error(ER_PARTITION_WRONG_TYPE, MYF(0), "SYSTEM_TIME");
+      part_type_error(thd, NULL, "SYSTEM_TIME", part_info);
       return true;
     }
   }
@@ -9809,6 +9810,7 @@ bool LEX::part_values_history(THD *thd)
   elem->type= partition_element::HISTORY;
   return false;
 }
+#endif /* WITH_PARTITION_STORAGE_ENGINE */
 
 
 bool LEX::last_field_generated_always_as_row_start_or_end(Lex_ident *p,

sql/sql_lex.h

@@ -3776,8 +3776,10 @@ public:
   bool table_or_sp_used();
 
   bool is_partition_management() const;
+#ifdef WITH_PARTITION_STORAGE_ENGINE
   bool part_values_current(THD *thd);
   bool part_values_history(THD *thd);
+#endif
 
   /**
     @brief check if the statement is a single-level join

sql/sql_partition.cc

@@ -2470,7 +2470,7 @@ end:
     @retval != 0 Failure
 */
 
-static int add_key_with_algorithm(String *str, partition_info *part_info)
+static int add_key_with_algorithm(String *str, const partition_info *part_info)
 {
   int err= 0;
   err+= str->append(STRING_WITH_LEN("KEY "));
@@ -2499,6 +2499,78 @@ char *generate_partition_syntax_for_frm(THD *thd, partition_info *part_info,
   return res;
 }
 
+
+/*
+  Generate the partition type syntax from the partition data structure.
+
+  @return Operation status.
+    @retval 0    Success
+    @retval > 0  Failure
+    @retval -1   Fatal error
+*/
+
+int partition_info::gen_part_type(THD *thd, String *str) const
+{
+  int err= 0;
+  switch (part_type)
+  {
+    case RANGE_PARTITION:
+      err+= str->append(STRING_WITH_LEN("RANGE "));
+      break;
+    case LIST_PARTITION:
+      err+= str->append(STRING_WITH_LEN("LIST "));
+      break;
+    case HASH_PARTITION:
+      if (linear_hash_ind)
+        err+= str->append(STRING_WITH_LEN("LINEAR "));
+      if (list_of_part_fields)
+      {
+        err+= add_key_with_algorithm(str, this);
+        err+= add_part_field_list(thd, str, part_field_list);
+      }
+      else
+        err+= str->append(STRING_WITH_LEN("HASH "));
+      break;
+    case VERSIONING_PARTITION:
+      err+= str->append(STRING_WITH_LEN("SYSTEM_TIME "));
+      break;
+    default:
+      DBUG_ASSERT(0);
+      /* We really shouldn't get here, no use in continuing from here */
+      my_error(ER_OUT_OF_RESOURCES, MYF(ME_FATAL));
+      return -1;
+  }
+  return err;
+}
+
+
+void part_type_error(THD *thd, partition_info *work_part_info,
+                     const char *part_type,
+                     partition_info *tab_part_info)
+{
+  StringBuffer<256> tab_part_type;
+  if (tab_part_info->gen_part_type(thd, &tab_part_type) < 0)
+    return;
+  tab_part_type.length(tab_part_type.length() - 1);
+  if (work_part_info)
+  {
+    DBUG_ASSERT(!part_type);
+    StringBuffer<256> work_part_type;
+    if (work_part_info->gen_part_type(thd, &work_part_type) < 0)
+      return;
+    work_part_type.length(work_part_type.length() - 1);
+    my_error(ER_PARTITION_WRONG_TYPE, MYF(0), work_part_type.c_ptr(),
+             tab_part_type.c_ptr());
+  }
+  else
+  {
+    DBUG_ASSERT(part_type);
+    my_error(ER_PARTITION_WRONG_TYPE, MYF(0), part_type,
+             tab_part_type.c_ptr());
+  }
+}
+
+
 /*
   Generate the partition syntax from the partition data structure.
   Useful for support of generating defaults, SHOW CREATE TABLES
@@ -2542,34 +2614,10 @@ char *generate_partition_syntax(THD *thd, partition_info *part_info,
   DBUG_ENTER("generate_partition_syntax");
 
   err+= str.append(STRING_WITH_LEN(" PARTITION BY "));
-  switch (part_info->part_type)
+  int err2= part_info->gen_part_type(thd, &str);
-  {
+  if (err2 < 0)
-    case RANGE_PARTITION:
-      err+= str.append(STRING_WITH_LEN("RANGE "));
-      break;
-    case LIST_PARTITION:
-      err+= str.append(STRING_WITH_LEN("LIST "));
-      break;
-    case HASH_PARTITION:
-      if (part_info->linear_hash_ind)
-        err+= str.append(STRING_WITH_LEN("LINEAR "));
-      if (part_info->list_of_part_fields)
-      {
-        err+= add_key_with_algorithm(&str, part_info);
-        err+= add_part_field_list(thd, &str, part_info->part_field_list);
-      }
-      else
-        err+= str.append(STRING_WITH_LEN("HASH "));
-      break;
-    case VERSIONING_PARTITION:
-      err+= str.append(STRING_WITH_LEN("SYSTEM_TIME "));
-      break;
-    default:
-      DBUG_ASSERT(0);
-      /* We really shouldn't get here, no use in continuing from here */
-      my_error(ER_OUT_OF_RESOURCES, MYF(ME_FATAL));
     DBUG_RETURN(NULL);
-  }
+  err+= err2;
   if (part_info->part_type == VERSIONING_PARTITION)
   {
     Vers_part_info *vers_info= part_info->vers_info;
@@ -5069,7 +5117,7 @@ uint prep_alter_part_table(THD *thd, TABLE *table, Alter_info *alter_info,
           else if (thd->work_part_info->part_type == VERSIONING_PARTITION ||
                    tab_part_info->part_type == VERSIONING_PARTITION)
           {
-            my_error(ER_PARTITION_WRONG_TYPE, MYF(0), "SYSTEM_TIME");
+            part_type_error(thd, thd->work_part_info, NULL, tab_part_info);
           }
           else
           {

sql/sql_table.cc

@@ -7901,6 +7901,7 @@ void append_drop_column(THD *thd, String *str, Field *field)
 }
 
 
+#ifdef WITH_PARTITION_STORAGE_ENGINE
 static inline
 void rename_field_in_list(Create_field *field, List<const char> *field_list)
 {
@@ -7913,6 +7914,7 @@ void rename_field_in_list(Create_field *field, List<const char> *field_list)
     it.replace(field->field_name.str);
   }
 }
+#endif
 
 
 /**

sql/sql_yacc.yy

@@ -4803,13 +4803,17 @@ opt_part_values:
           part_values_in {}
         | CURRENT_SYM
           {
+#ifdef WITH_PARTITION_STORAGE_ENGINE
             if (Lex->part_values_current(thd))
               MYSQL_YYABORT;
+#endif
           }
         | HISTORY_SYM
           {
+#ifdef WITH_PARTITION_STORAGE_ENGINE
             if (Lex->part_values_history(thd))
               MYSQL_YYABORT;
+#endif
           }
         | DEFAULT
          {
@@ -18874,6 +18878,7 @@ package_implementation_function_body:
             sp_head *sp= pkg->m_current_routine->sphead;
             thd->lex= pkg->m_current_routine;
             sp->reset_thd_mem_root(thd);
+            sp->set_c_chistics(thd->lex->sp_chistics);
             sp->set_body_start(thd, YYLIP->get_cpp_tok_start());
           }
           sp_body opt_package_routine_end_name
@@ -18892,6 +18897,7 @@ package_implementation_procedure_body:
             sp_head *sp= pkg->m_current_routine->sphead;
             thd->lex= pkg->m_current_routine;
             sp->reset_thd_mem_root(thd);
+            sp->set_c_chistics(thd->lex->sp_chistics);
             sp->set_body_start(thd, YYLIP->get_cpp_tok_start());
           }
           sp_body opt_package_routine_end_name

storage/innobase/CMakeLists.txt

@@ -270,8 +270,7 @@ SET(INNOBASE_SOURCES
 	include/handler0alter.h
 	include/hash0hash.h
 	include/ibuf0ibuf.h
-	include/ibuf0ibuf.inl/
+	include/ibuf0ibuf.inl
-	include/ibuf0types.h
 	include/lock0iter.h
 	include/lock0lock.h
 	include/lock0lock.inl
@@ -289,7 +288,6 @@ SET(INNOBASE_SOURCES
 	include/mem0mem.inl
 	include/mtr0log.h
 	include/mtr0mtr.h
-	include/mtr0mtr.inl
 	include/mtr0types.h
 	include/os0file.h
 	include/os0file.inl

storage/innobase/btr/btr0bulk.cc

@@ -827,7 +827,6 @@ PageBulk::storeExt(
 	btr_pcur_t	btr_pcur;
 	btr_pcur.pos_state = BTR_PCUR_IS_POSITIONED;
 	btr_pcur.latch_mode = BTR_MODIFY_LEAF;
-	btr_pcur.btr_cur.index = m_index;
 	btr_pcur.btr_cur.page_cur.index = m_index;
 	btr_pcur.btr_cur.page_cur.rec = m_cur_rec;
 	btr_pcur.btr_cur.page_cur.offsets = offsets;

storage/innobase/btr/btr0defragment.cc

@@ -169,7 +169,7 @@ btr_defragment_find_index(
 @return whether the operation was interrupted */
 bool btr_defragment_add_index(btr_pcur_t *pcur, THD *thd)
 {
-  dict_stats_empty_defrag_summary(pcur->btr_cur.index);
+  dict_stats_empty_defrag_summary(pcur->index());
   pthread_cond_t cond;
   pthread_cond_init(&cond, nullptr);
   btr_defragment_item_t item(pcur, &cond);
@@ -209,7 +209,7 @@ btr_defragment_remove_table(
   mysql_mutex_lock(&btr_defragment_mutex);
   for (auto item : btr_defragment_wq)
   {
-    if (item->cond && table == item->pcur->btr_cur.index->table)
+    if (item->cond && table == item->pcur->index()->table)
     {
       pthread_cond_signal(item->cond);
       item->cond= nullptr;
@@ -405,7 +405,10 @@ btr_defragment_merge_pages(
 		}
 	}
 	btr_cur_t parent;
-	if (!btr_page_get_father(index, from_block, mtr, &parent)) {
+	parent.page_cur.index = index;
+	parent.page_cur.block = from_block;
+
+	if (!btr_page_get_father(mtr, &parent)) {
 		to_block = nullptr;
 	} else if (n_recs_to_move == n_recs) {
 		/* The whole page is merged with the previous page,
@@ -690,7 +693,7 @@ processed:
 		}
 		log_free_check();
 		mtr_start(&mtr);
-		dict_index_t *index = item->pcur->btr_cur.index;
+		dict_index_t *index = item->pcur->index();
 		index->set_modified(mtr);
 		/* To follow the latching order defined in WL#6326,
 		acquire index->lock X-latch.  This entitles us to

storage/innobase/btr/btr0pcur.cc

@@ -29,25 +29,6 @@ Created 2/23/1996 Heikki Tuuri
 #include "rem0cmp.h"
 #include "trx0trx.h"
 
-/**************************************************************//**
-Allocates memory for a persistent cursor object and initializes the cursor.
-@return own: persistent cursor */
-btr_pcur_t*
-btr_pcur_create_for_mysql(void)
-/*============================*/
-{
-	btr_pcur_t*	pcur;
-	DBUG_ENTER("btr_pcur_create_for_mysql");
-
-	pcur = (btr_pcur_t*) ut_malloc_nokey(sizeof(btr_pcur_t));
-
-	pcur->btr_cur.index = NULL;
-	btr_pcur_init(pcur);
-
-	DBUG_PRINT("btr_pcur_create_for_mysql", ("pcur: %p", pcur));
-	DBUG_RETURN(pcur);
-}
-
 /**************************************************************//**
 Resets a persistent cursor object, freeing ::old_rec_buf if it is
 allocated and resetting the other members to their initial values. */
@@ -56,34 +37,17 @@ btr_pcur_reset(
 /*===========*/
 	btr_pcur_t*	cursor)	/*!< in, out: persistent cursor */
 {
-	btr_pcur_free(cursor);
+	ut_free(cursor->old_rec_buf);
+	memset(&cursor->btr_cur.page_cur, 0, sizeof(page_cur_t));
 	cursor->old_rec_buf = NULL;
-	cursor->btr_cur.index = NULL;
-	cursor->btr_cur.page_cur.rec = NULL;
 	cursor->old_rec = NULL;
 	cursor->old_n_core_fields = 0;
 	cursor->old_n_fields = 0;
-	cursor->old_stored = false;
 
 	cursor->latch_mode = BTR_NO_LATCHES;
 	cursor->pos_state = BTR_PCUR_NOT_POSITIONED;
 }
 
-/**************************************************************//**
-Frees the memory for a persistent cursor object. */
-void
-btr_pcur_free_for_mysql(
-/*====================*/
-	btr_pcur_t*	cursor)	/*!< in, own: persistent cursor */
-{
-	DBUG_ENTER("btr_pcur_free_for_mysql");
-	DBUG_PRINT("btr_pcur_free_for_mysql", ("pcur: %p", cursor));
-
-	btr_pcur_free(cursor);
-	ut_free(cursor);
-	DBUG_VOID_RETURN;
-}
-
 /**************************************************************//**
 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
@@ -125,8 +89,6 @@ btr_pcur_store_position(
 		  && mtr->memo_contains_flagged(&index->lock, MTR_MEMO_X_LOCK
 						| MTR_MEMO_SX_LOCK)));
 
-	cursor->old_stored = true;
-
 	if (page_is_empty(block->page.frame)) {
 		/* It must be an empty index tree; NOTE that in this case
 		we do not store the modify_clock, but always do a search
@@ -256,11 +218,12 @@ otherwise. */
 struct optimistic_latch_leaves
 {
   btr_pcur_t *const cursor;
-  ulint *latch_mode;
+  btr_latch_mode *latch_mode;
   mtr_t *const mtr;
 
-  optimistic_latch_leaves(btr_pcur_t *cursor, ulint *latch_mode, mtr_t *mtr)
+  optimistic_latch_leaves(btr_pcur_t *cursor, btr_latch_mode *latch_mode,
-  :cursor(cursor), latch_mode(latch_mode), mtr(mtr) {}
+                          mtr_t *mtr)
+    : cursor(cursor), latch_mode(latch_mode), mtr(mtr) {}
 
   bool operator() (buf_block_t *hint) const
   {
@@ -292,7 +255,7 @@ record with the same unique field values as in the stored record,
 btr_pcur_t::NOT_SAME cursor position is not on user rec or points on
 the record with not the samebuniq field values as in the stored */
 btr_pcur_t::restore_status
-btr_pcur_t::restore_position(ulint restore_latch_mode, mtr_t *mtr)
+btr_pcur_t::restore_position(btr_latch_mode restore_latch_mode, mtr_t *mtr)
 {
 	dict_index_t*	index;
 	dtuple_t*	tuple;
@@ -301,7 +264,6 @@ btr_pcur_t::restore_position(ulint restore_latch_mode, mtr_t *mtr)
 	mem_heap_t*	heap;
 
 	ut_ad(mtr->is_active());
-	//ut_ad(cursor->old_stored);
 	ut_ad(pos_state == BTR_PCUR_WAS_POSITIONED
 	      || pos_state == BTR_PCUR_IS_POSITIONED);
 
@@ -313,10 +275,9 @@ btr_pcur_t::restore_position(ulint restore_latch_mode, mtr_t *mtr)
 		/* In these cases we do not try an optimistic restoration,
 		but always do a search */
 
-		if (btr_cur_open_at_index_side(
+		if (btr_cur.open_leaf(rel_pos == BTR_PCUR_BEFORE_FIRST_IN_TREE,
-			rel_pos == BTR_PCUR_BEFORE_FIRST_IN_TREE,
+				      index, restore_latch_mode, mtr)
-			index, restore_latch_mode,
+		    != DB_SUCCESS) {
-			&btr_cur, 0, mtr) != DB_SUCCESS) {
 			return restore_status::CORRUPTED;
 		}
 
@@ -333,9 +294,10 @@ btr_pcur_t::restore_position(ulint restore_latch_mode, mtr_t *mtr)
 	ut_a(old_n_core_fields <= index->n_core_fields);
 	ut_a(old_n_fields);
 
-	switch (restore_latch_mode) {
+	static_assert(BTR_SEARCH_PREV == (4 | BTR_SEARCH_LEAF), "");
-	case BTR_SEARCH_LEAF:
+	static_assert(BTR_MODIFY_PREV == (4 | BTR_MODIFY_LEAF), "");
-	case BTR_MODIFY_LEAF:
+
+	switch (restore_latch_mode | 4) {
 	case BTR_SEARCH_PREV:
 	case BTR_MODIFY_PREV:
 		/* Try optimistic restoration. */
@@ -420,7 +382,7 @@ btr_pcur_t::restore_position(ulint restore_latch_mode, mtr_t *mtr)
 		mode = PAGE_CUR_UNSUPP;
 	}
 
-	if (btr_pcur_open_with_no_init(index, tuple, mode, restore_latch_mode,
+	if (btr_pcur_open_with_no_init(tuple, mode, restore_latch_mode,
 				       this, mtr) != DB_SUCCESS) {
 		mem_heap_free(heap);
 		return restore_status::CORRUPTED;
@@ -450,7 +412,6 @@ btr_pcur_t::restore_position(ulint restore_latch_mode, mtr_t *mtr)
 			block_when_stored.store(btr_pcur_get_block(this));
 			modify_clock= buf_block_get_modify_clock(
 			    block_when_stored.block());
-			old_stored= true;
 
 			mem_heap_free(heap);
 
@@ -487,7 +448,7 @@ btr_pcur_move_to_next_page(
 	ut_ad(cursor->latch_mode != BTR_NO_LATCHES);
 	ut_ad(btr_pcur_is_after_last_on_page(cursor));
 
-	cursor->old_stored = false;
+	cursor->old_rec = nullptr;
 
 	const page_t* page = btr_pcur_get_page(cursor);
 	const uint32_t next_page_no = btr_page_get_next(page);
@@ -515,7 +476,7 @@ btr_pcur_move_to_next_page(
 
 	dberr_t err;
 	buf_block_t* next_block = btr_block_get(
-		*btr_pcur_get_btr_cur(cursor)->index, next_page_no, mode,
+		*cursor->index(), next_page_no, mode,
 		page_is_leaf(page), mtr, &err);
 
 	if (UNIV_UNLIKELY(!next_block)) {
@@ -529,12 +490,13 @@ btr_pcur_move_to_next_page(
 		return DB_CORRUPTION;
 	}
 
-	btr_leaf_page_release(btr_pcur_get_block(cursor), mode, mtr);
-
 	page_cur_set_before_first(next_block, btr_pcur_get_page_cur(cursor));
 
 	ut_d(page_check_dir(next_page));
-	return err;
+
+	const auto s = mtr->get_savepoint();
+	mtr->rollback_to_savepoint(s - 2, s - 1);
+	return DB_SUCCESS;
 }
 
 MY_ATTRIBUTE((nonnull,warn_unused_result))
@@ -558,7 +520,7 @@ btr_pcur_move_backward_from_page(
 	ut_ad(btr_pcur_is_before_first_on_page(cursor));
 	ut_ad(!btr_pcur_is_before_first_in_tree(cursor));
 
-	const ulint latch_mode = cursor->latch_mode;
+	const auto latch_mode = cursor->latch_mode;
 	ut_ad(latch_mode == BTR_SEARCH_LEAF || latch_mode == BTR_MODIFY_LEAF);
 
 	btr_pcur_store_position(cursor, mtr);
@@ -570,31 +532,32 @@ btr_pcur_move_backward_from_page(
 	static_assert(BTR_SEARCH_PREV == (4 | BTR_SEARCH_LEAF), "");
 	static_assert(BTR_MODIFY_PREV == (4 | BTR_MODIFY_LEAF), "");
 
-	if (UNIV_UNLIKELY(cursor->restore_position(4 | latch_mode, mtr)
+	if (UNIV_UNLIKELY(cursor->restore_position(
+				  btr_latch_mode(4 | latch_mode), mtr)
 			  == btr_pcur_t::CORRUPTED)) {
 		return true;
 	}
 
-	buf_block_t* prev_block = btr_pcur_get_btr_cur(cursor)->left_block;
+	buf_block_t* release_block = nullptr;
 
 	if (!page_has_prev(btr_pcur_get_page(cursor))) {
 	} else if (btr_pcur_is_before_first_on_page(cursor)) {
-		btr_leaf_page_release(btr_pcur_get_block(cursor),
+		release_block = btr_pcur_get_block(cursor);
-				      latch_mode, mtr);
+		page_cur_set_after_last(cursor->btr_cur.left_block,
-
-		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;
+		release_block = cursor->btr_cur.left_block;
-		btr_leaf_page_release(prev_block, latch_mode, mtr);
 	}
 
 	cursor->latch_mode = latch_mode;
-	cursor->old_stored = false;
+	cursor->old_rec = nullptr;
+	if (release_block) {
+		mtr->release(*release_block);
+	}
 	return false;
 }
 
@@ -612,7 +575,7 @@ btr_pcur_move_to_prev(
 	ut_ad(cursor->pos_state == BTR_PCUR_IS_POSITIONED);
 	ut_ad(cursor->latch_mode != BTR_NO_LATCHES);
 
-	cursor->old_stored = false;
+	cursor->old_rec = nullptr;
 
 	if (btr_pcur_is_before_first_on_page(cursor)) {
 		return (!btr_pcur_is_before_first_in_tree(cursor)

storage/innobase/btr/btr0sea.cc

@@ -302,7 +302,7 @@ are consistent.
 @param[in]	cursor	cursor which was just positioned */
 static void btr_search_info_update_hash(btr_search_t *info, btr_cur_t *cursor)
 {
-	dict_index_t*	index = cursor->index;
+	dict_index_t*	index = cursor->index();
 	int		cmp;
 
 	if (dict_index_is_ibuf(index)) {
@@ -704,14 +704,14 @@ btr_search_update_hash_ref(
 		return;
 	}
 
-	if (index != cursor->index) {
+	if (index != cursor->index()) {
-		ut_ad(index->id == cursor->index->id);
+		ut_ad(index->id == cursor->index()->id);
 		btr_search_drop_page_hash_index(block);
 		return;
 	}
 
 	ut_ad(block->page.id().space() == index->table->space_id);
-	ut_ad(index == cursor->index);
+	ut_ad(index == cursor->index());
 	ut_ad(!dict_index_is_ibuf(index));
 	auto part = btr_search_sys.get_part(*index);
 	part->latch.wr_lock(SRW_LOCK_CALL);
@@ -784,7 +784,7 @@ btr_search_check_guess(
 	bool		success		= false;
 	rec_offs_init(offsets_);
 
-	n_unique = dict_index_get_n_unique_in_tree(cursor->index);
+	n_unique = dict_index_get_n_unique_in_tree(cursor->index());
 
 	rec = btr_cur_get_rec(cursor);
 
@@ -792,7 +792,7 @@ btr_search_check_guess(
 			  || !page_rec_is_leaf(rec))) {
 		ut_ad("corrupted index" == 0);
 		return false;
-	} else if (cursor->index->table->not_redundant()) {
+	} else if (cursor->index()->table->not_redundant()) {
 		switch (rec_get_status(rec)) {
 		case REC_STATUS_INSTANT:
 		case REC_STATUS_ORDINARY:
@@ -805,10 +805,10 @@ btr_search_check_guess(
 
 	match = 0;
 
-	offsets = rec_get_offsets(rec, cursor->index, offsets,
+	offsets = rec_get_offsets(rec, cursor->index(), offsets,
-				  cursor->index->n_core_fields,
+				  cursor->index()->n_core_fields,
 				  n_unique, &heap);
-	cmp = cmp_dtuple_rec_with_match(tuple, rec, cursor->index, offsets,
+	cmp = cmp_dtuple_rec_with_match(tuple, rec, cursor->index(), offsets,
 					&match);
 
 	if (mode == PAGE_CUR_GE) {
@@ -860,7 +860,7 @@ btr_search_check_guess(
 			goto exit_func;
 		}
 
-		if (cursor->index->table->not_redundant()) {
+		if (cursor->index()->table->not_redundant()) {
 			switch (rec_get_status(prev_rec)) {
 			case REC_STATUS_INSTANT:
 			case REC_STATUS_ORDINARY:
@@ -871,11 +871,12 @@ btr_search_check_guess(
 			}
 		}
 
-		offsets = rec_get_offsets(prev_rec, cursor->index, offsets,
+		offsets = rec_get_offsets(prev_rec, cursor->index(), offsets,
-					  cursor->index->n_core_fields,
+					  cursor->index()->n_core_fields,
 					  n_unique, &heap);
-		cmp = cmp_dtuple_rec_with_match(
+		cmp = cmp_dtuple_rec_with_match(tuple, prev_rec,
-			tuple, prev_rec, cursor->index, offsets, &match);
+						cursor->index(), offsets,
+						&match);
 		if (mode == PAGE_CUR_GE) {
 			success = cmp > 0;
 		} else {
@@ -900,7 +901,7 @@ btr_search_check_guess(
 			goto exit_func;
 		}
 
-		if (cursor->index->table->not_redundant()) {
+		if (cursor->index()->table->not_redundant()) {
 			switch (rec_get_status(next_rec)) {
 			case REC_STATUS_INSTANT:
 			case REC_STATUS_ORDINARY:
@@ -911,11 +912,11 @@ btr_search_check_guess(
 			}
 		}
 
-		offsets = rec_get_offsets(next_rec, cursor->index, offsets,
+		offsets = rec_get_offsets(next_rec, cursor->index(), offsets,
-					  cursor->index->n_core_fields,
+					  cursor->index()->n_core_fields,
 					  n_unique, &heap);
 		cmp = cmp_dtuple_rec_with_match(
-			tuple, next_rec, cursor->index, offsets, &match);
+			tuple, next_rec, cursor->index(), offsets, &match);
 		if (mode == PAGE_CUR_LE) {
 			success = cmp < 0;
 			cursor->up_match = match;
@@ -1099,16 +1100,16 @@ btr_search_guess_on_hash(
 	part->latch.rd_lock(SRW_LOCK_CALL);
 
 	if (!btr_search_enabled) {
-		goto fail;
+		goto ahi_release_and_fail;
 	}
 
 	rec = static_cast<const rec_t*>(
 		ha_search_and_get_data(&part->table, fold));
 
 	if (!rec) {
-fail:
+ahi_release_and_fail:
 		part->latch.rd_unlock();
-
+fail:
 		btr_search_failure(info, cursor);
 		return false;
 	}
@@ -1117,66 +1118,52 @@ fail:
 
 	buf_pool_t::hash_chain& chain = buf_pool.page_hash.cell_get(
 		block->page.id().fold());
-	bool fail, got_latch;
+	bool got_latch;
 	{
 		transactional_shared_lock_guard<page_hash_latch> g{
 			buf_pool.page_hash.lock_get(chain)};
-
-		const auto state = block->page.state();
-		if (state == buf_page_t::REMOVE_HASH) {
-			/* Another thread is just freeing the block
-			from the LRU list of the buffer pool: do not
-			try to access this page. */
-			goto fail;
-		}
-		if (UNIV_UNLIKELY(state < buf_page_t::UNFIXED)) {
-#ifndef NO_ELISION
-			xend();
-#endif
-			ut_error;
-		}
-
-		fail = index != block->index && index_id == block->index->id;
 		got_latch = (latch_mode == BTR_SEARCH_LEAF)
 			? block->page.lock.s_lock_try()
 			: block->page.lock.x_lock_try();
 	}
 
-	ut_a(!fail || block->index->freed());
 	if (!got_latch) {
-		goto fail;
+		goto ahi_release_and_fail;
+	}
+
+	const auto state = block->page.state();
+	if (UNIV_UNLIKELY(state < buf_page_t::UNFIXED)) {
+		ut_ad(state == buf_page_t::REMOVE_HASH);
+block_and_ahi_release_and_fail:
+		if (latch_mode == BTR_SEARCH_LEAF) {
+			block->page.lock.s_unlock();
+		} else {
+			block->page.lock.x_unlock();
+		}
+		goto ahi_release_and_fail;
+	}
+
+	ut_ad(state < buf_page_t::READ_FIX || state >= buf_page_t::WRITE_FIX);
+	ut_ad(state < buf_page_t::READ_FIX || latch_mode == BTR_SEARCH_LEAF);
+
+	if (index != block->index && index_id == block->index->id) {
+		ut_a(block->index->freed());
+		goto block_and_ahi_release_and_fail;
 	}
 
 	block->page.fix();
 	block->page.set_accessed();
 	buf_page_make_young_if_needed(&block->page);
-	ut_ad(!block->page.is_read_fixed());
-	ut_ad(latch_mode == BTR_SEARCH_LEAF || !block->page.is_io_fixed());
 	static_assert(ulint{MTR_MEMO_PAGE_S_FIX} == ulint{BTR_SEARCH_LEAF},
 		      "");
 	static_assert(ulint{MTR_MEMO_PAGE_X_FIX} == ulint{BTR_MODIFY_LEAF},
 		      "");
-	mtr->memo_push(block, mtr_memo_type_t(latch_mode));
-
-	++buf_pool.stat.n_page_gets;
 
 	part->latch.rd_unlock();
 
-	if (UNIV_UNLIKELY(fail)) {
+	++buf_pool.stat.n_page_gets;
-		goto fail_and_release_page;
-	}
 
-	DBUG_ASSERT(!block->page.is_freed());
+	mtr->memo_push(block, mtr_memo_type_t(latch_mode));
-
-	if (!block->page.in_file()) {
-		ut_ad(block->page.state() == buf_page_t::REMOVE_HASH);
-
-fail_and_release_page:
-		btr_leaf_page_release(block, latch_mode, mtr);
-
-		btr_search_failure(info, cursor);
-		return false;
-	}
 
 	ut_ad(page_rec_is_user_rec(rec));
 
@@ -1191,7 +1178,8 @@ fail_and_release_page:
 	right. */
 	if (index_id != btr_page_get_index_id(block->page.frame)
 	    || !btr_search_check_guess(cursor, false, tuple, mode)) {
-		goto fail_and_release_page;
+		mtr->release_last_page();
+		goto fail;
 	}
 
 	if (info->n_hash_potential < BTR_SEARCH_BUILD_LIMIT + 5) {
@@ -1199,41 +1187,6 @@ fail_and_release_page:
 		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(!ahi_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
@@ -1691,7 +1644,7 @@ exit_func:
 @param[in,out]	cursor	cursor which was just positioned */
 void btr_search_info_update_slow(btr_search_t *info, btr_cur_t *cursor)
 {
-	srw_spin_lock*	ahi_latch = &btr_search_sys.get_part(*cursor->index)
+	srw_spin_lock*	ahi_latch = &btr_search_sys.get_part(*cursor->index())
 		->latch;
 	buf_block_t*	block = btr_cur_get_block(cursor);
 
@@ -1706,7 +1659,7 @@ void btr_search_info_update_slow(btr_search_t *info, btr_cur_t *cursor)
 
 	if (build_index || (cursor->flag == BTR_CUR_HASH_FAIL)) {
 
-		btr_search_check_free_space_in_heap(cursor->index);
+		btr_search_check_free_space_in_heap(cursor->index());
 	}
 
 	if (cursor->flag == BTR_CUR_HASH_FAIL) {
@@ -1723,7 +1676,7 @@ void btr_search_info_update_slow(btr_search_t *info, btr_cur_t *cursor)
 		/* 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. */
-		btr_search_build_page_hash_index(cursor->index, block,
+		btr_search_build_page_hash_index(cursor->index(), block,
 						 ahi_latch,
 						 block->n_fields,
 						 block->n_bytes,
@@ -1835,15 +1788,15 @@ void btr_search_update_hash_on_delete(btr_cur_t *cursor)
 		return;
 	}
 
-	ut_ad(!cursor->index->table->is_temporary());
+	ut_ad(!cursor->index()->table->is_temporary());
 
-	if (index != cursor->index) {
+	if (index != cursor->index()) {
 		btr_search_drop_page_hash_index(block);
 		return;
 	}
 
 	ut_ad(block->page.id().space() == index->table->space_id);
-	ut_a(index == cursor->index);
+	ut_a(index == cursor->index());
 	ut_a(block->curr_n_fields > 0 || block->curr_n_bytes > 0);
 	ut_ad(!dict_index_is_ibuf(index));
 
@@ -1890,7 +1843,7 @@ void btr_search_update_hash_node_on_insert(btr_cur_t *cursor,
 	dict_index_t*	index;
 	rec_t*		rec;
 
-	ut_ad(ahi_latch == &btr_search_sys.get_part(*cursor->index)->latch);
+	ut_ad(ahi_latch == &btr_search_sys.get_part(*cursor->index())->latch);
 
 	if (!btr_search_enabled) {
 		return;
@@ -1909,15 +1862,15 @@ void btr_search_update_hash_node_on_insert(btr_cur_t *cursor,
 		return;
 	}
 
-	ut_ad(!cursor->index->table->is_temporary());
+	ut_ad(!cursor->index()->table->is_temporary());
 
-	if (index != cursor->index) {
+	if (index != cursor->index()) {
-		ut_ad(index->id == cursor->index->id);
+		ut_ad(index->id == cursor->index()->id);
 		btr_search_drop_page_hash_index(block);
 		return;
 	}
 
-	ut_a(cursor->index == index);
+	ut_a(cursor->index() == index);
 	ut_ad(!dict_index_is_ibuf(index));
 	ahi_latch->wr_lock(SRW_LOCK_CALL);
 
@@ -1934,7 +1887,7 @@ void btr_search_update_hash_node_on_insert(btr_cur_t *cursor,
 	    && !block->curr_left_side) {
 		if (const rec_t *new_rec = page_rec_get_next_const(rec)) {
 			if (ha_search_and_update_if_found(
-				&btr_search_sys.get_part(*cursor->index)
+				&btr_search_sys.get_part(*cursor->index())
 				->table,
 				cursor->fold, rec, block, new_rec)) {
 				MONITOR_INC(MONITOR_ADAPTIVE_HASH_ROW_UPDATED);
@@ -1977,7 +1930,7 @@ void btr_search_update_hash_on_insert(btr_cur_t *cursor,
 	rec_offs*	offsets		= offsets_;
 	rec_offs_init(offsets_);
 
-	ut_ad(ahi_latch == &btr_search_sys.get_part(*cursor->index)->latch);
+	ut_ad(ahi_latch == &btr_search_sys.get_part(*cursor->index())->latch);
 	ut_ad(page_is_leaf(btr_cur_get_page(cursor)));
 
 	if (!btr_search_enabled) {
@@ -2001,16 +1954,16 @@ void btr_search_update_hash_on_insert(btr_cur_t *cursor,
 
 	rec = btr_cur_get_rec(cursor);
 
-	ut_ad(!cursor->index->table->is_temporary());
+	ut_ad(!cursor->index()->table->is_temporary());
 
-	if (index != cursor->index) {
+	if (index != cursor->index()) {
-		ut_ad(index->id == cursor->index->id);
+		ut_ad(index->id == cursor->index()->id);
 drop:
 		btr_search_drop_page_hash_index(block);
 		return;
 	}
 
-	ut_a(index == cursor->index);
+	ut_a(index == cursor->index());
 	ut_ad(!dict_index_is_ibuf(index));
 
 	n_fields = block->curr_n_fields;

storage/innobase/buf/buf0flu.cc

@@ -1712,6 +1712,14 @@ inline void log_t::write_checkpoint(lsn_t end_lsn) noexcept
 
   DBUG_PRINT("ib_log", ("checkpoint ended at " LSN_PF ", flushed to " LSN_PF,
                         next_checkpoint_lsn, get_flushed_lsn()));
+  if (overwrite_warned)
+  {
+    sql_print_information("InnoDB: Crash recovery was broken "
+                          "between LSN=" LSN_PF
+                          " and checkpoint LSN=" LSN_PF ".",
+                          overwrite_warned, next_checkpoint_lsn);
+    overwrite_warned= 0;
+  }
 
   latch.wr_unlock();
 }
@@ -1884,6 +1892,7 @@ ATTRIBUTE_COLD void buf_flush_wait_flushed(lsn_t sync_lsn)
     record from a previous fil_names_clear() call, which we must
     write out before we can advance the checkpoint. */
     log_write_up_to(sync_lsn, true);
+    DBUG_EXECUTE_IF("ib_log_checkpoint_avoid_hard", return;);
     log_checkpoint();
   }
 }
@@ -1898,6 +1907,8 @@ ATTRIBUTE_COLD void buf_flush_ahead(lsn_t lsn, bool furious)
   if (recv_recovery_is_on())
     recv_sys.apply(true);
 
+  DBUG_EXECUTE_IF("ib_log_checkpoint_avoid_hard", return;);
+
   Atomic_relaxed<lsn_t> &limit= furious
     ? buf_flush_sync_lsn : buf_flush_async_lsn;
 
@@ -2252,6 +2263,7 @@ unemployed:
       buf_pool.page_cleaner_set_idle(true);
 
       DBUG_EXECUTE_IF("ib_log_checkpoint_avoid", continue;);
+      DBUG_EXECUTE_IF("ib_log_checkpoint_avoid_hard", continue;);
 
       mysql_mutex_unlock(&buf_pool.flush_list_mutex);
 
@@ -2335,6 +2347,7 @@ do_checkpoint:
         here should not affect correctness, because log_free_check()
         should still be invoking checkpoints when needed. */
         DBUG_EXECUTE_IF("ib_log_checkpoint_avoid", goto next;);
+        DBUG_EXECUTE_IF("ib_log_checkpoint_avoid_hard", goto next;);
 
         if (!recv_recovery_is_on() && srv_operation == SRV_OPERATION_NORMAL)
           log_checkpoint();

storage/innobase/dict/dict0crea.cc

@@ -755,11 +755,12 @@ dict_create_index_tree_step(
 
 	search_tuple = dict_create_search_tuple(node->ind_row, node->heap);
 	node->page_no = FIL_NULL;
+	pcur.btr_cur.page_cur.index =
+		UT_LIST_GET_FIRST(dict_sys.sys_indexes->indexes);
 
 	dberr_t err =
-		btr_pcur_open(UT_LIST_GET_FIRST(dict_sys.sys_indexes->indexes),
+		btr_pcur_open(search_tuple, PAGE_CUR_L, BTR_MODIFY_LEAF,
-			      search_tuple, PAGE_CUR_L, BTR_MODIFY_LEAF,
+			      &pcur, 0, &mtr);
-			      &pcur, &mtr);
 
 	if (err != DB_SUCCESS) {
 func_exit:

storage/innobase/dict/dict0dict.cc

@@ -4147,8 +4147,9 @@ void dict_set_corrupted(dict_index_t *index, const char *ctx)
 	dfield_set_data(dfield, buf, 8);
 
 	dict_index_copy_types(tuple, sys_index, 2);
+	cursor.page_cur.index = sys_index;
 
-	if (btr_cur_search_to_nth_level(sys_index, 0, tuple, PAGE_CUR_LE,
+	if (btr_cur_search_to_nth_level(0, tuple, PAGE_CUR_LE,
 					BTR_MODIFY_LEAF, &cursor, &mtr)
 	    != DB_SUCCESS) {
 		goto fail;
@@ -4222,8 +4223,9 @@ dict_index_set_merge_threshold(
 	dfield_set_data(dfield, buf, 8);
 
 	dict_index_copy_types(tuple, sys_index, 2);
+	cursor.page_cur.index = sys_index;
 
-	if (btr_cur_search_to_nth_level(sys_index, 0, tuple, PAGE_CUR_GE,
+	if (btr_cur_search_to_nth_level(0, tuple, PAGE_CUR_GE,
 					BTR_MODIFY_LEAF, &cursor, &mtr)
 	    != DB_SUCCESS) {
 		goto func_exit;

storage/innobase/dict/dict0load.cc

@@ -215,8 +215,9 @@ dict_startscan_system(
 	mtr_t*		mtr,		/*!< in: the mini-transaction */
 	dict_table_t*	table)		/*!< in: system table */
 {
-  if (btr_pcur_open_at_index_side(true, table->indexes.start, BTR_SEARCH_LEAF,
+  btr_pcur_init(pcur);
-                                  pcur, true, 0, mtr) != DB_SUCCESS)
+  if (pcur->open_leaf(true, table->indexes.start, BTR_SEARCH_LEAF, mtr) !=
+      DB_SUCCESS)
     return nullptr;
   const rec_t *rec;
   do
@@ -1316,8 +1317,9 @@ static dberr_t dict_load_columns(dict_table_t *table, unsigned use_uncommitted,
 	mach_write_to_8(table_id, table->id);
 	dfield_set_data(&dfield, table_id, 8);
 	dict_index_copy_types(&tuple, sys_index, 1);
+	pcur.btr_cur.page_cur.index = sys_index;
 
-	dberr_t err = btr_pcur_open_on_user_rec(sys_index, &tuple, PAGE_CUR_GE,
+	dberr_t err = btr_pcur_open_on_user_rec(&tuple, PAGE_CUR_GE,
 						BTR_SEARCH_LEAF, &pcur, &mtr);
 	if (err != DB_SUCCESS) {
 		goto func_exit;
@@ -1446,9 +1448,9 @@ dict_load_virtual_col(dict_table_t *table, bool uncommitted, ulint nth_v_col)
 	dfield_set_data(&dfield[1], vcol_pos, 4);
 
 	dict_index_copy_types(&tuple, sys_virtual_index, 2);
+	pcur.btr_cur.page_cur.index = sys_virtual_index;
 
-	dberr_t err = btr_pcur_open_on_user_rec(sys_virtual_index, &tuple,
+	dberr_t err = btr_pcur_open_on_user_rec(&tuple, PAGE_CUR_GE,
-						PAGE_CUR_GE,
 						BTR_SEARCH_LEAF, &pcur, &mtr);
 	if (err != DB_SUCCESS) {
 		goto func_exit;
@@ -1686,8 +1688,9 @@ static dberr_t dict_load_fields(dict_index_t *index, bool uncommitted,
 	mach_write_to_8(index_id, index->id);
 	dfield_set_data(&dfield, index_id, 8);
 	dict_index_copy_types(&tuple, sys_index, 1);
+	pcur.btr_cur.page_cur.index = sys_index;
 
-	dberr_t error = btr_pcur_open_on_user_rec(sys_index, &tuple,
+	dberr_t error = btr_pcur_open_on_user_rec(&tuple,
 						  PAGE_CUR_GE, BTR_SEARCH_LEAF,
 						  &pcur, &mtr);
 	if (error != DB_SUCCESS) {
@@ -1944,8 +1947,9 @@ dberr_t dict_load_indexes(dict_table_t *table, bool uncommitted,
 	mach_write_to_8(table_id, table->id);
 	dfield_set_data(&dfield, table_id, 8);
 	dict_index_copy_types(&tuple, sys_index, 1);
+	pcur.btr_cur.page_cur.index = sys_index;
 
-	dberr_t error = btr_pcur_open_on_user_rec(sys_index, &tuple,
+	dberr_t error = btr_pcur_open_on_user_rec(&tuple,
 						  PAGE_CUR_GE, BTR_SEARCH_LEAF,
 						  &pcur, &mtr);
 	if (error != DB_SUCCESS) {
@@ -2338,11 +2342,12 @@ static dict_table_t *dict_load_table_one(const span<const char> &name,
 	};
 	dfield_set_data(&dfield, name.data(), name.size());
 	dict_index_copy_types(&tuple, sys_index, 1);
+	pcur.btr_cur.page_cur.index = sys_index;
 
 	bool uncommitted = false;
 reload:
 	mtr.start();
-	dberr_t err = btr_pcur_open_on_user_rec(sys_index, &tuple, PAGE_CUR_GE,
+	dberr_t err = btr_pcur_open_on_user_rec(&tuple, PAGE_CUR_GE,
 						BTR_SEARCH_LEAF, &pcur, &mtr);
 
 	if (err != DB_SUCCESS || !btr_pcur_is_on_user_rec(&pcur)) {
@@ -2596,10 +2601,11 @@ dict_load_table_on_id(
 	mach_write_to_8(id_buf, table_id);
 	dfield_set_data(&dfield, id_buf, 8);
 	dict_index_copy_types(&tuple, sys_table_ids, 1);
+	pcur.btr_cur.page_cur.index = sys_table_ids;
 
 	dict_table_t* table = nullptr;
 
-	if (btr_pcur_open_on_user_rec(sys_table_ids, &tuple, PAGE_CUR_GE,
+	if (btr_pcur_open_on_user_rec(&tuple, PAGE_CUR_GE,
 				      BTR_SEARCH_LEAF, &pcur, &mtr)
 	    == DB_SUCCESS
 	    && btr_pcur_is_on_user_rec(&pcur)) {
@@ -2703,9 +2709,10 @@ static dberr_t dict_load_foreign_cols(dict_foreign_t *foreign, trx_id_t trx_id)
 
 	dfield_set_data(&dfield, foreign->id, id_len);
 	dict_index_copy_types(&tuple, sys_index, 1);
+	pcur.btr_cur.page_cur.index = sys_index;
 
 	mem_heap_t* heap = nullptr;
-	dberr_t err = btr_pcur_open_on_user_rec(sys_index, &tuple, PAGE_CUR_GE,
+	dberr_t err = btr_pcur_open_on_user_rec(&tuple, PAGE_CUR_GE,
 						BTR_SEARCH_LEAF, &pcur, &mtr);
 	if (err != DB_SUCCESS) {
 		goto func_exit;
@@ -2877,11 +2884,12 @@ dict_load_foreign(
 	};
 	dfield_set_data(&dfield, id.data(), id.size());
 	dict_index_copy_types(&tuple, sys_index, 1);
+	pcur.btr_cur.page_cur.index = sys_index;
 
 	mtr.start();
 
 	mem_heap_t* heap = nullptr;
-	dberr_t err = btr_pcur_open_on_user_rec(sys_index, &tuple, PAGE_CUR_GE,
+	dberr_t err = btr_pcur_open_on_user_rec(&tuple, PAGE_CUR_GE,
 						BTR_SEARCH_LEAF, &pcur, &mtr);
 	if (err != DB_SUCCESS) {
 		goto err_exit;
@@ -3090,8 +3098,9 @@ start_load:
 	mtr.start();
 	dfield_set_data(&dfield, table_name, strlen(table_name));
 	dict_index_copy_types(&tuple, sec_index, 1);
+	pcur.btr_cur.page_cur.index = sec_index;
 
-	dberr_t err = btr_pcur_open_on_user_rec(sec_index, &tuple, PAGE_CUR_GE,
+	dberr_t err = btr_pcur_open_on_user_rec(&tuple, PAGE_CUR_GE,
 						BTR_SEARCH_LEAF, &pcur, &mtr);
 	if (err != DB_SUCCESS) {
 		DBUG_RETURN(err);
@@ -3192,7 +3201,6 @@ next_rec:
 
 load_next_index:
 	mtr.commit();
-	ut_free(pcur.old_rec_buf);
 
 	if ((sec_index = dict_table_get_next_index(sec_index))) {
 		/* Switch to scan index on REF_NAME, fk_max_recusive_level
@@ -3202,5 +3210,6 @@ load_next_index:
 		goto start_load;
 	}
 
+	ut_free(pcur.old_rec_buf);
 	DBUG_RETURN(DB_SUCCESS);
 }

storage/innobase/dict/dict0stats.cc

@@ -1560,6 +1560,96 @@ empty_table:
 	return err;
 }
 
+/** Open a cursor at the first page in a tree level.
+@param page_cur  cursor
+@param level     level to search for (0=leaf)
+@param mtr       mini-transaction */
+static dberr_t page_cur_open_level(page_cur_t *page_cur, ulint level,
+                                   mtr_t *mtr)
+{
+  mem_heap_t *heap= nullptr;
+  rec_offs offsets_[REC_OFFS_NORMAL_SIZE];
+  rec_offs *offsets= offsets_;
+  dberr_t err;
+
+  dict_index_t *const index= page_cur->index;
+
+  rec_offs_init(offsets_);
+  ut_ad(level != ULINT_UNDEFINED);
+  ut_ad(mtr->memo_contains_flagged(&index->lock, MTR_MEMO_SX_LOCK));
+  ut_ad(mtr->get_savepoint() == 1);
+
+  uint32_t page= index->page;
+
+  for (ulint height = ULINT_UNDEFINED;; height--)
+  {
+    buf_block_t* block=
+      btr_block_get(*index, page, RW_S_LATCH,
+                    !height && !index->is_clust(), mtr, &err);
+    if (!block)
+      break;
+
+    const uint32_t l= btr_page_get_level(block->page.frame);
+
+    if (height == ULINT_UNDEFINED)
+    {
+      ut_ad(!heap);
+      /* We are in the root node */
+      height= l;
+      if (UNIV_UNLIKELY(height < level))
+        return DB_CORRUPTION;
+    }
+    else if (UNIV_UNLIKELY(height != l) || page_has_prev(block->page.frame))
+    {
+      err= DB_CORRUPTION;
+      break;
+    }
+
+    page_cur_set_before_first(block, page_cur);
+
+    if (height == level)
+      break;
+
+    ut_ad(height);
+
+    if (!page_cur_move_to_next(page_cur))
+    {
+      err= DB_CORRUPTION;
+      break;
+    }
+
+    offsets= rec_get_offsets(page_cur->rec, index, offsets, 0, ULINT_UNDEFINED,
+                             &heap);
+    page= btr_node_ptr_get_child_page_no(page_cur->rec, offsets);
+  }
+
+  if (UNIV_LIKELY_NULL(heap))
+    mem_heap_free(heap);
+
+  /* Release all page latches except the one on the desired page. */
+  const auto end= mtr->get_savepoint();
+  if (end > 1)
+    mtr->rollback_to_savepoint(1, end - 1);
+
+  return err;
+}
+
+/** Open a cursor at the first page in a tree level.
+@param page_cur  cursor
+@param level     level to search for (0=leaf)
+@param mtr       mini-transaction
+@param index     index tree */
+static dberr_t btr_pcur_open_level(btr_pcur_t *pcur, ulint level, mtr_t *mtr,
+                                   dict_index_t *index)
+{
+  pcur->latch_mode= BTR_SEARCH_TREE;
+  pcur->search_mode= PAGE_CUR_G;
+  pcur->pos_state= BTR_PCUR_IS_POSITIONED;
+  pcur->btr_cur.page_cur.index= index;
+  return page_cur_open_level(&pcur->btr_cur.page_cur, level, mtr);
+}
+
+
 /* @{ Pseudo code about the relation between the following functions
 
 let N = N_SAMPLE_PAGES(index)
@@ -1616,7 +1706,8 @@ dict_stats_analyze_index_level(
 	DEBUG_PRINTF("    %s(table=%s, index=%s, level=" ULINTPF ")\n",
 		     __func__, index->table->name, index->name, level);
 
-	ut_ad(mtr->memo_contains(index->lock, MTR_MEMO_SX_LOCK));
+	*total_recs = 0;
+	*total_pages = 0;
 
 	n_uniq = dict_index_get_n_unique(index);
 
@@ -1649,9 +1740,7 @@ dict_stats_analyze_index_level(
 	/* Position pcur on the leftmost record on the leftmost page
 	on the desired level. */
 
-	if (btr_pcur_open_at_index_side(
+	if (btr_pcur_open_level(&pcur, level, mtr, index) != DB_SUCCESS
-		    true, index, BTR_SEARCH_TREE_ALREADY_S_LATCHED,
-		    &pcur, true, level, mtr) != DB_SUCCESS
 	    || !btr_pcur_move_to_next_on_page(&pcur)) {
 		goto func_exit;
 	}
@@ -1661,21 +1750,10 @@ dict_stats_analyze_index_level(
 	/* The page must not be empty, except when
 	it is the root page (and the whole index is empty). */
 	ut_ad(btr_pcur_is_on_user_rec(&pcur) || page_is_leaf(page));
-	ut_ad(btr_pcur_get_rec(&pcur)
-	      == page_rec_get_next_const(page_get_infimum_rec(page)));
 
 	prev_rec = NULL;
 	prev_rec_is_copied = false;
 
-	/* no records by default */
-	*total_recs = 0;
-
-	*total_pages = 0;
-
-	if (page_has_prev(page) || btr_page_get_level(page) != level) {
-		goto func_exit;
-	}
-
 	if (REC_INFO_MIN_REC_FLAG & rec_get_info_bits(
 		    btr_pcur_get_rec(&pcur), page_is_comp(page))) {
 		ut_ad(btr_pcur_is_on_user_rec(&pcur));
@@ -1728,10 +1806,7 @@ dict_stats_analyze_index_level(
 
 		if (level == 0
 		    && !srv_stats_include_delete_marked
-		    && rec_get_deleted_flag(
+		    && rec_get_deleted_flag(rec, page_rec_is_comp(rec))) {
-			    rec,
-			    page_is_comp(btr_pcur_get_page(&pcur)))) {
-
 			if (rec_is_last_on_page
 			    && !prev_rec_is_copied
 			    && prev_rec != NULL) {
@@ -1811,7 +1886,7 @@ dict_stats_analyze_index_level(
 			records on this level at some point we will jump from
 			one page to the next and then rec and prev_rec will
 			be on different pages and
-			btr_pcur_move_to_next_user_rec() will release the
+			btr_cur_move_to_next_user_rec() will release the
 			latch on the page that prev_rec is on */
 			prev_rec = rec_copy_prefix_to_buf(
 				rec, index, n_uniq,
@@ -1820,7 +1895,7 @@ dict_stats_analyze_index_level(
 
 		} else {
 			/* still on the same page, the next call to
-			btr_pcur_move_to_next_user_rec() will not jump
+			btr_cur_move_to_next_user_rec() will not jump
 			on the next page, we can simply assign pointers
 			instead of copying the records like above */
 
@@ -1891,7 +1966,6 @@ dict_stats_analyze_index_level(
 	}
 #endif /* UNIV_STATS_DEBUG */
 
-	btr_leaf_page_release(btr_pcur_get_block(&pcur), BTR_SEARCH_LEAF, mtr);
 func_exit:
 	ut_free(prev_rec_buf);
 	mem_heap_free(heap);
@@ -2280,7 +2354,6 @@ dict_stats_analyze_index_for_n_prefix(
 		     n_prefix, n_diff_data->n_diff_on_level);
 #endif
 
-	ut_ad(mtr->memo_contains(index->lock, MTR_MEMO_SX_LOCK));
 	ut_ad(n_diff_data->level);
 
 	/* Position pcur on the leftmost record on the leftmost page
@@ -2289,9 +2362,7 @@ dict_stats_analyze_index_for_n_prefix(
 	n_diff_data->n_diff_all_analyzed_pages = 0;
 	n_diff_data->n_external_pages_sum = 0;
 
-	if (btr_pcur_open_at_index_side(true, index,
+	if (btr_pcur_open_level(&pcur, n_diff_data->level, mtr, index)
-					BTR_SEARCH_TREE_ALREADY_S_LATCHED,
-					&pcur, true, n_diff_data->level, mtr)
 	    != DB_SUCCESS
 	    || !btr_pcur_move_to_next_on_page(&pcur)) {
 		return;
@@ -2680,6 +2751,7 @@ empty_index:
 		mtr.commit();
 		mtr.start();
 		mtr_sx_lock_index(index, &mtr);
+		ut_ad(mtr.get_savepoint() == 1);
 		buf_block_t *root = btr_root_block_get(index, RW_S_LATCH,
 						       &mtr, &err);
 		if (!root || root_level != btr_page_get_level(root->page.frame)
@@ -2695,7 +2767,7 @@ empty_index:
 			break;
 		}
 
-		mtr.memo_release(root, MTR_MEMO_PAGE_S_FIX);
+		mtr.rollback_to_savepoint(1);
 
 		/* check whether we should pick the current level;
 		we pick level 1 even if it does not have enough
@@ -2757,6 +2829,7 @@ empty_index:
 				break;
 			}
 
+			mtr.rollback_to_savepoint(1);
 			dict_stats_analyze_index_level(index,
 						       level,
 						       n_diff_on_level,
@@ -2764,7 +2837,7 @@ empty_index:
 						       &total_pages,
 						       n_diff_boundaries,
 						       &mtr);
-
+			mtr.rollback_to_savepoint(1);
 			level_is_analyzed = true;
 
 			if (level == 1

storage/innobase/fil/fil0crypt.cc

@@ -1831,7 +1831,6 @@ fil_crypt_rotate_page(
 	} else {
 		/* If block read failed mtr memo and log should be empty. */
 		ut_ad(!mtr.has_modifications());
-		ut_ad(!mtr.is_dirty());
 		ut_ad(mtr.is_empty());
 		mtr.commit();
 	}

storage/innobase/fts/fts0fts.cc

@@ -3347,7 +3347,6 @@ fts_add_doc_by_id(
 	is_id_cluster = (clust_index == fts_id_index);
 
 	mtr_start(&mtr);
-	btr_pcur_init(&pcur);
 
 	/* Search based on Doc ID. Here, we'll need to consider the case
 	when there is no primary index on Doc ID */
@@ -3358,9 +3357,10 @@ fts_add_doc_by_id(
 
 	mach_write_to_8((byte*) &temp_doc_id, doc_id);
 	dfield_set_data(dfield, &temp_doc_id, sizeof(temp_doc_id));
+	pcur.btr_cur.page_cur.index = fts_id_index;
 
 	/* If we have a match, add the data to doc structure */
-	if (btr_pcur_open_with_no_init(fts_id_index, tuple, PAGE_CUR_LE,
+	if (btr_pcur_open_with_no_init(tuple, PAGE_CUR_LE,
 				       BTR_SEARCH_LEAF, &pcur, &mtr)
 	    == DB_SUCCESS
 	    && btr_pcur_get_low_match(&pcur) == 1) {
@@ -3387,7 +3387,6 @@ fts_add_doc_by_id(
 			dtuple_t*	clust_ref;
 			ulint		n_fields;
 
-			btr_pcur_init(&clust_pcur);
 			n_fields = dict_index_get_n_unique(clust_index);
 
 			clust_ref = dtuple_create(heap, n_fields);
@@ -3395,8 +3394,9 @@ fts_add_doc_by_id(
 
 			row_build_row_ref_in_tuple(
 				clust_ref, rec, fts_id_index, NULL);
+			clust_pcur.btr_cur.page_cur.index = clust_index;
 
-			if (btr_pcur_open_with_no_init(clust_index, clust_ref,
+			if (btr_pcur_open_with_no_init(clust_ref,
 						       PAGE_CUR_LE,
 						       BTR_SEARCH_LEAF,
 						       &clust_pcur, &mtr)
@@ -3552,12 +3552,11 @@ fts_get_max_doc_id(
 	ut_ad(innobase_strcasecmp(FTS_DOC_ID_COL_NAME, dfield->name) == 0);
 #endif
 
-	mtr_start(&mtr);
+	mtr.start();
 
 	/* fetch the largest indexes value */
-	if (btr_pcur_open_at_index_side(false, index, BTR_SEARCH_LEAF, &pcur,
+	if (pcur.open_leaf(false, index, BTR_SEARCH_LEAF, &mtr) == DB_SUCCESS
-					true, 0, &mtr) != DB_SUCCESS) {
+	    && !page_is_empty(btr_pcur_get_page(&pcur))) {
-	} else if (!page_is_empty(btr_pcur_get_page(&pcur))) {
 		const rec_t*    rec = NULL;
 
 		do {
@@ -3576,7 +3575,7 @@ fts_get_max_doc_id(
 	}
 
 func_exit:
-	mtr_commit(&mtr);
+	mtr.commit();
 	return(doc_id);
 }
 

storage/innobase/gis/gis0rtree.cc

@@ -70,7 +70,7 @@ rtr_page_split_initialize_nodes(
 
 	block = btr_cur_get_block(cursor);
 	page = buf_block_get_frame(block);
-	n_uniq = dict_index_get_n_unique_in_tree(cursor->index);
+	n_uniq = dict_index_get_n_unique_in_tree(cursor->index());
 
 	n_recs = ulint(page_get_n_recs(page)) + 1;
 
@@ -88,8 +88,8 @@ rtr_page_split_initialize_nodes(
 
 	rec = page_rec_get_next(page_get_infimum_rec(page));
 	const ulint n_core = page_is_leaf(page)
-		? cursor->index->n_core_fields : 0;
+		? cursor->index()->n_core_fields : 0;
-	*offsets = rec_get_offsets(rec, cursor->index, *offsets, n_core,
+	*offsets = rec_get_offsets(rec, cursor->index(), *offsets, n_core,
 				   n_uniq, &heap);
 
 	source_cur = rec_get_nth_field(rec, *offsets, 0, &len);
@@ -101,7 +101,7 @@ rtr_page_split_initialize_nodes(
 		memcpy(cur->coords, source_cur, DATA_MBR_LEN);
 
 		rec = page_rec_get_next(rec);
-		*offsets = rec_get_offsets(rec, cursor->index, *offsets,
+		*offsets = rec_get_offsets(rec, cursor->index(), *offsets,
 					   n_core, n_uniq, &heap);
 		source_cur = rec_get_nth_field(rec, *offsets, 0, &len);
 	}
@@ -111,9 +111,9 @@ rtr_page_split_initialize_nodes(
 		dtuple_get_nth_field(tuple, 0)));
 	cur->coords = reserve_coords(buf_pos, SPDIMS);
 	rec = (byte*) mem_heap_alloc(
-		heap, rec_get_converted_size(cursor->index, tuple, 0));
+		heap, rec_get_converted_size(cursor->index(), tuple, 0));
 
-	rec = rec_convert_dtuple_to_rec(rec, cursor->index, tuple, 0);
+	rec = rec_convert_dtuple_to_rec(rec, cursor->index(), tuple, 0);
 	cur->key = rec;
 
 	memcpy(cur->coords, source_cur, DATA_MBR_LEN);
@@ -200,7 +200,7 @@ rtr_update_mbr_field(
 	rec_t*		new_rec,	/*!< in: rec to use */
 	mtr_t*		mtr)		/*!< in: mtr */
 {
-	dict_index_t*	index = cursor->index;
+	dict_index_t*	index = cursor->index();
 	mem_heap_t*	heap;
 	page_t*		page;
 	rec_t*		rec;
@@ -245,6 +245,7 @@ rtr_update_mbr_field(
 	/* We need to remember the child page no of cursor2, since page could be
 	reorganized or insert a new rec before it. */
 	if (cursor2) {
+		ut_ad(cursor2->index() == index);
 		rec_t*	del_rec = btr_cur_get_rec(cursor2);
 		offsets2 = rec_get_offsets(btr_cur_get_rec(cursor2),
 					   index, NULL, 0,
@@ -268,7 +269,7 @@ rtr_update_mbr_field(
 			if (!btr_cur_update_alloc_zip(
 					page_zip,
 					btr_cur_get_page_cur(cursor),
-					index, offsets,
+					offsets,
 					rec_offs_size(offsets),
 					false, mtr)) {
 
@@ -321,7 +322,7 @@ rtr_update_mbr_field(
 							offsets2));
 
 			page_cur_delete_rec(btr_cur_get_page_cur(cursor2),
-					    index, offsets2, mtr);
+					    offsets2, mtr);
 		}
 	} else if (page_get_n_recs(page) == 1) {
 		/* When there's only one rec in the page, we do insert/delete to
@@ -352,9 +353,10 @@ rtr_update_mbr_field(
 		ut_ad(old_rec != insert_rec);
 
 		page_cur_position(old_rec, block, &page_cur);
+		page_cur.index = index;
 		offsets2 = rec_get_offsets(old_rec, index, NULL, n_core,
 					   ULINT_UNDEFINED, &heap);
-		page_cur_delete_rec(&page_cur, index, offsets2, mtr);
+		page_cur_delete_rec(&page_cur, offsets2, mtr);
 
 	} else {
 update_mbr:
@@ -366,8 +368,7 @@ update_mbr:
 
 		/* Delete the rec which cursor point to. */
 		next_rec = page_rec_get_next(rec);
-		page_cur_delete_rec(btr_cur_get_page_cur(cursor),
+		page_cur_delete_rec(&cursor->page_cur, offsets, mtr);
-				    index, offsets, mtr);
 		if (!ins_suc) {
 			ut_ad(rec_info & REC_INFO_MIN_REC_FLAG);
 
@@ -400,13 +401,12 @@ update_mbr:
 			      == btr_node_ptr_get_child_page_no(cur2_rec,
 								offsets2));
 			page_cur_delete_rec(btr_cur_get_page_cur(cursor2),
-					    index, offsets2, mtr);
+					    offsets2, mtr);
 			cursor2 = NULL;
 		}
 
 		/* Insert the new rec. */
-		if (page_cur_search_with_match(block, index, node_ptr,
+		if (page_cur_search_with_match(node_ptr, PAGE_CUR_LE,
-					       PAGE_CUR_LE,
 					       &up_match, &low_match,
 					       btr_cur_get_page_cur(cursor),
 					       NULL)) {
@@ -424,7 +424,7 @@ update_mbr:
 		} else if (ins_suc) {
 			ut_ad(err == DB_FAIL);
 			err = btr_page_reorganize(btr_cur_get_page_cur(cursor),
-						  index, mtr);
+						  mtr);
 			if (err == DB_SUCCESS) {
 				err = btr_cur_optimistic_insert(
 					flags, cursor, &insert_offsets, &heap,
@@ -505,7 +505,7 @@ update_mbr:
 			ut_ad(cur2_pno == del_page_no && cur2_rec != insert_rec);
 
 			page_cur_delete_rec(btr_cur_get_page_cur(cursor2),
-					    index, offsets2, mtr);
+					    offsets2, mtr);
 		}
 
 		if (!ins_suc) {
@@ -556,7 +556,6 @@ rtr_adjust_upper_level(
 {
 	ulint		page_no;
 	ulint		new_page_no;
-	dict_index_t*	index = sea_cur->index;
 	btr_cur_t	cursor;
 	rec_offs*	offsets;
 	mem_heap_t*	heap;
@@ -570,9 +569,10 @@ rtr_adjust_upper_level(
 
 	/* Create a memory heap where the data tuple is stored */
 	heap = mem_heap_create(1024);
-	cursor.init();
 
 	cursor.thr = sea_cur->thr;
+	cursor.page_cur.index = sea_cur->index();
+	cursor.page_cur.block = block;
 
 	/* Get the level of the split pages */
 	level = btr_page_get_level(buf_block_get_frame(block));
@@ -584,8 +584,7 @@ rtr_adjust_upper_level(
 
 	/* Set new mbr for the old page on the upper level. */
 	/* Look up the index for the node pointer to page */
-	offsets = rtr_page_get_father_block(
+	offsets = rtr_page_get_father_block(NULL, heap, mtr, sea_cur, &cursor);
-		NULL, heap, index, block, mtr, sea_cur, &cursor);
 
 	page_cursor = btr_cur_get_page_cur(&cursor);
 
@@ -607,10 +606,9 @@ rtr_adjust_upper_level(
 		    page_get_infimum_rec(new_block->page.frame))) {
 		/* Insert the node for the new page. */
 		node_ptr_upper = rtr_index_build_node_ptr(
-			index, new_mbr, first, new_page_no, heap);
+			sea_cur->index(), new_mbr, first, new_page_no, heap);
 		ulint	up_match = 0, low_match = 0;
-		err = page_cur_search_with_match(btr_cur_get_block(&cursor),
+		err = page_cur_search_with_match(node_ptr_upper,
-						 index, node_ptr_upper,
 						 PAGE_CUR_LE,
 						 &up_match, &low_match,
 						 btr_cur_get_page_cur(&cursor),
@@ -660,7 +658,7 @@ rtr_adjust_upper_level(
 
 	mem_heap_free(heap);
 
-	ut_ad(block->zip_size() == index->table->space->zip_size());
+	ut_ad(block->zip_size() == sea_cur->index()->table->space->zip_size());
 
 	if (err != DB_SUCCESS) {
 		return err;
@@ -670,7 +668,7 @@ rtr_adjust_upper_level(
 
 	if (next_page_no == FIL_NULL) {
 	} else if (buf_block_t*	next_block =
-		   btr_block_get(*index, next_page_no, RW_X_LATCH,
+		   btr_block_get(*sea_cur->index(), next_page_no, RW_X_LATCH,
 				 false, mtr, &err)) {
 		if (UNIV_UNLIKELY(memcmp_aligned<4>(next_block->page.frame
 						    + FIL_PAGE_PREV,
@@ -740,6 +738,7 @@ rtr_split_page_move_rec_list(
 
 	page_cur_set_before_first(block, &page_cursor);
 	page_cur_set_before_first(new_block, &new_page_cursor);
+	page_cursor.index = new_page_cursor.index = index;
 
 	page = buf_block_get_frame(block);
 	new_page = buf_block_get_frame(new_block);
@@ -774,7 +773,7 @@ rtr_split_page_move_rec_list(
 
 			rec = page_cur_insert_rec_low(
 				&new_page_cursor,
-				index, cur_split_node->key, offsets, mtr);
+				cur_split_node->key, offsets, mtr);
 
 			if (UNIV_UNLIKELY
 			    (!rec
@@ -841,8 +840,7 @@ rtr_split_page_move_rec_list(
 				page_cur_get_rec(&page_cursor), index,
 				offsets, n_core, ULINT_UNDEFINED,
 				&heap);
-			page_cur_delete_rec(&page_cursor,
+			page_cur_delete_rec(&page_cursor, offsets, mtr);
-				index, offsets, mtr);
 		}
 	}
 
@@ -877,7 +875,6 @@ rtr_page_split_and_insert(
 	buf_block_t*		new_block;
 	page_zip_des_t*		page_zip;
 	page_zip_des_t*		new_page_zip;
-	buf_block_t*		insert_block;
 	page_cur_t*		page_cursor;
 	rec_t*			rec = 0;
 	ulint			n_recs;
@@ -906,12 +903,10 @@ func_start:
 	mem_heap_empty(*heap);
 	*offsets = NULL;
 
-	ut_ad(mtr->memo_contains_flagged(&cursor->index->lock, MTR_MEMO_X_LOCK
+	ut_ad(mtr->memo_contains_flagged(&cursor->index()->lock,
-					 | MTR_MEMO_SX_LOCK));
+					 MTR_MEMO_X_LOCK | MTR_MEMO_SX_LOCK));
-	ut_ad(!dict_index_is_online_ddl(cursor->index)
+	ut_ad(!dict_index_is_online_ddl(cursor->index()));
-	      || (flags & BTR_CREATE_FLAG)
+	ut_ad(cursor->index()->lock.have_u_or_x());
-	      || dict_index_is_clust(cursor->index));
-	ut_ad(cursor->index->lock.have_u_or_x());
 
 	block = btr_cur_get_block(cursor);
 	page = buf_block_get_frame(block);
@@ -954,7 +949,7 @@ corrupted:
 	}
 #endif
 
-	insert_size = rec_get_converted_size(cursor->index, tuple, n_ext);
+	insert_size = rec_get_converted_size(cursor->index(), tuple, n_ext);
 	total_data = page_get_data_size(page) + insert_size;
 	first_rec_group = split_rtree_node(rtr_split_node_array,
 					   static_cast<int>(n_recs),
@@ -965,7 +960,7 @@ corrupted:
 
 	/* Allocate a new page to the index */
 	const uint16_t page_level = btr_page_get_level(page);
-	new_block = btr_page_alloc(cursor->index, page_id.page_no() + 1,
+	new_block = btr_page_alloc(cursor->index(), page_id.page_no() + 1,
 				   FSP_UP, page_level, mtr, mtr, err);
 	if (UNIV_UNLIKELY(!new_block)) {
 		return nullptr;
@@ -977,7 +972,7 @@ corrupted:
 		to contain FIL_NULL in FIL_PAGE_PREV at this stage. */
 		memset_aligned<4>(new_block->page.frame + FIL_PAGE_PREV, 0, 4);
 	}
-	btr_page_create(new_block, new_page_zip, cursor->index,
+	btr_page_create(new_block, new_page_zip, cursor->index(),
 			page_level, mtr);
 
 	new_page = buf_block_get_frame(new_block);
@@ -985,7 +980,7 @@ corrupted:
 
 	/* Set new ssn to the new page and page. */
 	page_set_ssn_id(new_block, new_page_zip, current_ssn, mtr);
-	next_ssn = rtr_get_new_ssn_id(cursor->index);
+	next_ssn = rtr_get_new_ssn_id(cursor->index());
 
 	page_set_ssn_id(block, page_zip, next_ssn, mtr);
 
@@ -998,7 +993,7 @@ corrupted:
 	    || (*err = rtr_split_page_move_rec_list(rtr_split_node_array,
 						    first_rec_group,
 						    new_block, block,
-						    first_rec, cursor->index,
+						    first_rec, cursor->index(),
 						    *heap, mtr))) {
 		if (*err != DB_FAIL) {
 			return nullptr;
@@ -1021,7 +1016,7 @@ corrupted:
 		ut_a(new_page_zip);
 
 		page_zip_copy_recs(new_block,
-				   page_zip, page, cursor->index, mtr);
+				   page_zip, page, cursor->index(), mtr);
 
 		page_cursor = btr_cur_get_page_cur(cursor);
 
@@ -1056,7 +1051,7 @@ corrupted:
 		lock_rtr_move_rec_list(new_block, block, rec_move, moved);
 
 		const ulint n_core = page_level
-			? 0 : cursor->index->n_core_fields;
+			? 0 : cursor->index()->n_core_fields;
 
 		/* Delete recs in first group from the new page. */
 		for (cur_split_node = rtr_split_node_array;
@@ -1076,11 +1071,11 @@ corrupted:
 
 				*offsets = rec_get_offsets(
 					page_cur_get_rec(page_cursor),
-					cursor->index, *offsets, n_core,
+					cursor->index(), *offsets, n_core,
 					ULINT_UNDEFINED, heap);
 
 				page_cur_delete_rec(page_cursor,
-					cursor->index, *offsets, mtr);
+						    *offsets, mtr);
 				n++;
 			}
 		}
@@ -1093,32 +1088,30 @@ corrupted:
 						  block, page_cursor);
 				*offsets = rec_get_offsets(
 					page_cur_get_rec(page_cursor),
-					cursor->index, *offsets, n_core,
+					page_cursor->index, *offsets, n_core,
 					ULINT_UNDEFINED, heap);
-				page_cur_delete_rec(page_cursor,
+				page_cur_delete_rec(page_cursor, *offsets,
-					cursor->index, *offsets, mtr);
+						    mtr);
 			}
 		}
 
 #ifdef UNIV_GIS_DEBUG
-		ut_ad(page_validate(new_page, cursor->index));
+		ut_ad(page_validate(new_page, cursor->index()));
-		ut_ad(page_validate(page, cursor->index));
+		ut_ad(page_validate(page, cursor->index()));
 #endif
 	}
 
 	/* Insert the new rec to the proper page. */
 	cur_split_node = end_split_node - 1;
-	if (cur_split_node->n_node != first_rec_group) {
-		insert_block = new_block;
-	} else {
-		insert_block = block;
-	}
 
 	/* Reposition the cursor for insert and try insertion */
 	page_cursor = btr_cur_get_page_cur(cursor);
+	page_cursor->block = cur_split_node->n_node != first_rec_group
+		? new_block : block;
+
 	ulint up_match = 0, low_match = 0;
 
-	if (page_cur_search_with_match(insert_block, cursor->index, tuple,
+	if (page_cur_search_with_match(tuple,
 				       PAGE_CUR_LE, &up_match, &low_match,
 				       page_cursor, nullptr)) {
 		goto corrupted;
@@ -1133,7 +1126,7 @@ corrupted:
 				goto after_insert; }
 	);
 
-	rec = page_cur_tuple_insert(page_cursor, tuple, cursor->index,
+	rec = page_cur_tuple_insert(page_cursor, tuple,
 				    offsets, heap, n_ext, mtr);
 
 	/* If insert did not fit, try page reorganization.
@@ -1141,14 +1134,13 @@ corrupted:
 	attempted this already. */
 	if (rec == NULL) {
 		if (!is_page_cur_get_page_zip(page_cursor)
-		    && btr_page_reorganize(page_cursor, cursor->index, mtr)) {
+		    && btr_page_reorganize(page_cursor, mtr)) {
 			rec = page_cur_tuple_insert(page_cursor, tuple,
-						    cursor->index, offsets,
+						    offsets,
 						    heap, n_ext, mtr);
 
 		}
-		/* If insert fail, we will try to split the insert_block
+		/* If insert fail, we will try to split the block again. */
-		again. */
 	}
 
 #ifdef UNIV_DEBUG
@@ -1156,8 +1148,8 @@ after_insert:
 #endif
 	/* Calculate the mbr on the upper half-page, and the mbr on
 	original page. */
-	rtr_page_cal_mbr(cursor->index, block, &mbr, *heap);
+	rtr_page_cal_mbr(cursor->index(), block, &mbr, *heap);
-	rtr_page_cal_mbr(cursor->index, new_block, &new_mbr, *heap);
+	rtr_page_cal_mbr(cursor->index(), new_block, &new_mbr, *heap);
 	prdt.data = &mbr;
 	new_prdt.data = &new_mbr;
 
@@ -1175,7 +1167,8 @@ after_insert:
 	/* Save the new ssn to the root page, since we need to reinit
 	the first ssn value from it after restart server. */
 
-	root_block = btr_root_block_get(cursor->index, RW_SX_LATCH, mtr, err);
+	root_block = btr_root_block_get(cursor->index(), RW_SX_LATCH,
+					mtr, err);
 	if (UNIV_UNLIKELY(!root_block)) {
 		return nullptr;
 	}
@@ -1187,8 +1180,8 @@ after_insert:
 	 again. */
 	if (!rec) {
 		/* We play safe and reset the free bits for new_page */
-		if (!dict_index_is_clust(cursor->index)
+		if (!dict_index_is_clust(cursor->index())
-		    && !cursor->index->table->is_temporary()) {
+		    && !cursor->index()->table->is_temporary()) {
 			ibuf_reset_free_bits(new_block);
 			ibuf_reset_free_bits(block);
 		}
@@ -1205,16 +1198,16 @@ after_insert:
 		if (UNIV_UNLIKELY(!i_rec)) {
 			goto corrupted;
 		}
-		btr_cur_position(cursor->index, i_rec, block, cursor);
+		btr_cur_position(cursor->index(), i_rec, block, cursor);
 
 		goto func_start;
 	}
 
 #ifdef UNIV_GIS_DEBUG
-	ut_ad(page_validate(buf_block_get_frame(block), cursor->index));
+	ut_ad(page_validate(buf_block_get_frame(block), cursor->index()));
-	ut_ad(page_validate(buf_block_get_frame(new_block), cursor->index));
+	ut_ad(page_validate(buf_block_get_frame(new_block), cursor->index()));
 
-	ut_ad(!rec || rec_offs_validate(rec, cursor->index, *offsets));
+	ut_ad(!rec || rec_offs_validate(rec, cursor->index(), *offsets));
 #endif
 	MONITOR_INC(MONITOR_INDEX_SPLIT);
 
@@ -1234,14 +1227,13 @@ rtr_ins_enlarge_mbr(
 	rtr_mbr_t		new_mbr;
 	buf_block_t*		block;
 	mem_heap_t*		heap;
-	dict_index_t*		index = btr_cur->index;
 	page_cur_t*		page_cursor;
 	rec_offs*		offsets;
 	node_visit_t*		node_visit;
 	btr_cur_t		cursor;
 	page_t*			page;
 
-	ut_ad(dict_index_is_spatial(index));
+	ut_ad(btr_cur->index()->is_spatial());
 
 	/* If no rtr_info or rtree is one level tree, return. */
 	if (!btr_cur->rtr_info || btr_cur->tree_height == 1) {
@@ -1269,20 +1261,20 @@ rtr_ins_enlarge_mbr(
 		}
 
 		/* Calculate the mbr of the child page. */
-		rtr_page_cal_mbr(index, block, &new_mbr, heap);
+		rtr_page_cal_mbr(page_cursor->index, block, &new_mbr, heap);
 
 		/* Get father block. */
-		cursor.init();
+		cursor.page_cur.index = page_cursor->index;
+		cursor.page_cur.block = block;
 		offsets = rtr_page_get_father_block(
-			NULL, heap, index, block, mtr, btr_cur, &cursor);
+			NULL, heap, mtr, btr_cur, &cursor);
 
 		page = buf_block_get_frame(block);
 
 		/* Update the mbr field of the rec. */
 		rtr_update_mbr_field(&cursor, offsets, NULL, page,
 				     &new_mbr, NULL, mtr);
-		page_cursor = btr_cur_get_page_cur(&cursor);
+		block = btr_cur_get_block(&cursor);
-		block = page_cur_get_block(page_cursor);
 	}
 
 	mem_heap_free(heap);
@@ -1338,6 +1330,7 @@ rtr_page_copy_rec_list_end_no_locks(
 		return DB_CORRUPTION;
 	}
 	page_cur_position(cur_rec, new_block, &page_cur);
+	page_cur.index = index;
 
 	/* Copy records from the original page to the new page */
 	while (!page_cur_is_after_last(&cur1)) {
@@ -1399,7 +1392,7 @@ move_to_prev:
 		offsets1 = rec_get_offsets(cur1_rec, index, offsets1, n_core,
 					   ULINT_UNDEFINED, &heap);
 
-		ins_rec = page_cur_insert_rec_low(&page_cur, index,
+		ins_rec = page_cur_insert_rec_low(&page_cur,
 						  cur1_rec, offsets1, mtr);
 		if (UNIV_UNLIKELY(!ins_rec || moved >= max_move)) {
 			return DB_CORRUPTION;
@@ -1461,6 +1454,7 @@ rtr_page_copy_rec_list_start_no_locks(
 		return DB_CORRUPTION;
 	}
 	page_cur_position(cur_rec, new_block, &page_cur);
+	page_cur.index = index;
 
 	while (page_cur_get_rec(&cur1) != rec) {
 		rec_t*	cur1_rec = page_cur_get_rec(&cur1);
@@ -1522,7 +1516,7 @@ move_to_prev:
 		offsets1 = rec_get_offsets(cur1_rec, index, offsets1, n_core,
 					   ULINT_UNDEFINED, &heap);
 
-		ins_rec = page_cur_insert_rec_low(&page_cur, index,
+		ins_rec = page_cur_insert_rec_low(&page_cur,
 						  cur1_rec, offsets1, mtr);
 		if (UNIV_UNLIKELY(!ins_rec || moved >= max_move)) {
 			return DB_CORRUPTION;
@@ -1560,7 +1554,7 @@ rtr_merge_mbr_changed(
 	ulint		len;
 	bool		changed = false;
 
-	ut_ad(dict_index_is_spatial(cursor->index));
+	ut_ad(cursor->index()->is_spatial());
 
 	rec = btr_cur_get_rec(cursor);
 
@@ -1640,11 +1634,11 @@ rtr_check_same_block(
 	btr_cur_t*	cursor,	/*!< in/out: position at the parent entry
 				pointing to the child if successful */
 	buf_block_t*	parentb,/*!< in: parent page to check */
-	buf_block_t*	childb,	/*!< in: child Page */
 	mem_heap_t*	heap)	/*!< in: memory heap */
 
 {
-	ulint		page_no = childb->page.id().page_no();
+	const uint32_t	page_no =
+		btr_cur_get_block(cursor)->page.id().page_no();
 	rec_offs*	offsets;
 	rec_t*		rec = page_get_infimum_rec(parentb->page.frame);
 

storage/innobase/gis/gis0sea.cc

@@ -95,14 +95,13 @@ rtr_pcur_getnext_from_path(
 				/*!< in: index tree locked */
 	mtr_t*		mtr)	/*!< in: mtr */
 {
-	dict_index_t*	index = btr_cur->index;
+	dict_index_t*	index = btr_cur->index();
 	bool		found = false;
 	page_cur_t*	page_cursor;
 	ulint		level = 0;
 	node_visit_t	next_rec;
 	rtr_info_t*	rtr_info = btr_cur->rtr_info;
 	node_seq_t	page_ssn;
-	ulint		my_latch_mode;
 	ulint		skip_parent = false;
 	bool		new_split = false;
 	bool		for_delete = false;
@@ -115,7 +114,7 @@ rtr_pcur_getnext_from_path(
 
 	ut_ad(dtuple_get_n_fields_cmp(tuple));
 
-	my_latch_mode = BTR_LATCH_MODE_WITHOUT_FLAGS(latch_mode);
+	const auto my_latch_mode = BTR_LATCH_MODE_WITHOUT_FLAGS(latch_mode);
 
 	for_delete = latch_mode & BTR_RTREE_DELETE_MARK;
 	for_undo_ins = latch_mode & BTR_RTREE_UNDO_INS;
@@ -299,6 +298,7 @@ rtr_pcur_getnext_from_path(
 
 		page_cursor = btr_cur_get_page_cur(btr_cur);
 		page_cursor->rec = NULL;
+		page_cursor->block = block;
 
 		if (mode == PAGE_CUR_RTREE_LOCATE) {
 			if (target_level == 0 && level == 0) {
@@ -307,7 +307,7 @@ rtr_pcur_getnext_from_path(
 				found = false;
 
 				if (!page_cur_search_with_match(
-					block, index, tuple, PAGE_CUR_LE,
+					tuple, PAGE_CUR_LE,
 					&up_match, &low_match,
 					btr_cur_get_page_cur(btr_cur), nullptr)
 				    && low_match
@@ -351,17 +351,12 @@ rtr_pcur_getnext_from_path(
 						 BTR_PCUR_IS_POSITIONED;
 					r_cursor->latch_mode = my_latch_mode;
 					btr_pcur_store_position(r_cursor, mtr);
-#ifdef UNIV_DEBUG
+					ut_d(ulint num_stored =)
-					ulint num_stored =
 					rtr_store_parent_path(
 						block, btr_cur,
-							rw_latch, level, mtr);
+						btr_latch_mode(rw_latch),
-					ut_ad(num_stored > 0);
-#else
-					rtr_store_parent_path(
-						block, btr_cur, rw_latch,
 						level, mtr);
-#endif /* UNIV_DEBUG */
+					ut_ad(num_stored > 0);
 				}
 			}
 		} else {
@@ -443,11 +438,11 @@ rtr_pcur_getnext_from_path(
 
 	const rec_t* rec = btr_cur_get_rec(btr_cur);
 
-	if (page_rec_is_infimum(rec) || page_rec_is_supremum(rec)) {
+	if (!page_rec_is_user_rec(rec)) {
-		mtr_commit(mtr);
+		mtr->commit();
-		mtr_start(mtr);
+		mtr->start();
 	} else if (!index_locked) {
-		mtr_memo_release(mtr, &index->lock, MTR_MEMO_X_LOCK);
+		mtr->release(index->lock);
 	}
 
 	return(found);
@@ -521,7 +516,7 @@ bool
 rtr_pcur_open(
 	dict_index_t*	index,	/*!< in: index */
 	const dtuple_t*	tuple,	/*!< in: tuple on which search done */
-	ulint		latch_mode,/*!< in: BTR_SEARCH_LEAF, ... */
+	btr_latch_mode	latch_mode,/*!< in: BTR_SEARCH_LEAF, ... */
 	btr_pcur_t*	cursor, /*!< in: memory buffer for persistent cursor */
 	mtr_t*		mtr)	/*!< in: mtr */
 {
@@ -539,6 +534,7 @@ rtr_pcur_open(
 	/* Search with the tree cursor */
 
 	btr_cur_t* btr_cursor = btr_pcur_get_btr_cur(cursor);
+	btr_cursor->page_cur.index = index;
 
 	btr_cursor->rtr_info = rtr_create_rtr_info(false, false,
 						   btr_cursor, index);
@@ -554,7 +550,7 @@ rtr_pcur_open(
 		mtr->lock_upgrade(index->lock);
 	}
 
-	if (btr_cur_search_to_nth_level(index, 0, tuple, PAGE_CUR_RTREE_LOCATE,
+	if (btr_cur_search_to_nth_level(0, tuple, PAGE_CUR_RTREE_LOCATE,
                                         latch_mode,
 					btr_cursor, mtr) != DB_SUCCESS) {
 		return true;
@@ -606,24 +602,16 @@ rtr_pcur_open(
 }
 
 /* Get the rtree page father.
-@param[in]	index		rtree index
-@param[in]	block		child page in the index
 @param[in,out]	mtr		mtr
 @param[in]	sea_cur		search cursor, contains information
 				about parent nodes in search
 @param[out]	cursor		cursor on node pointer record,
 				its page x-latched
 @return whether the cursor was successfully positioned */
-bool
+bool rtr_page_get_father(mtr_t *mtr, btr_cur_t *sea_cur, btr_cur_t *cursor)
-rtr_page_get_father(
-	dict_index_t*	index,
-	buf_block_t*	block,
-	mtr_t*		mtr,
-	btr_cur_t*	sea_cur,
-	btr_cur_t*	cursor)
 {
   mem_heap_t *heap = mem_heap_create(100);
-  rec_offs *offsets= rtr_page_get_father_block(nullptr, heap, index, block,
+  rec_offs *offsets= rtr_page_get_father_block(nullptr, heap,
                                                mtr, sea_cur, cursor);
   mem_heap_free(heap);
   return offsets != nullptr;
@@ -634,7 +622,6 @@ MY_ATTRIBUTE((warn_unused_result))
 Returns the upper level node pointer to a R-Tree page. It is assumed
 that mtr holds an x-latch on the tree. */
 static const rec_t* rtr_get_father_node(
-	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
@@ -647,6 +634,7 @@ static const rec_t* rtr_get_father_node(
 {
 	const rec_t* rec = nullptr;
 	auto had_rtr = btr_cur->rtr_info;
+	dict_index_t* const index = btr_cur->index();
 
 	/* Try to optimally locate the parent node. Level should always
 	less than sea_cur->tree_height unless the root is splitting */
@@ -680,7 +668,7 @@ static const rec_t* rtr_get_father_node(
 
 	btr_cur->rtr_info = rtr_create_rtr_info(false, false, btr_cur, index);
 
-	if (btr_cur_search_to_nth_level(index, level, tuple,
+	if (btr_cur_search_to_nth_level(level, tuple,
 					PAGE_CUR_RTREE_LOCATE,
 					BTR_CONT_MODIFY_TREE, btr_cur, mtr)
 	    != DB_SUCCESS) {
@@ -764,7 +752,7 @@ rtr_page_get_father_node_ptr(
 		sea_cur = NULL;
 	}
 
-	const rec_t* node_ptr = rtr_get_father_node(index, level + 1, tuple,
+	const rec_t* node_ptr = rtr_get_father_node(level + 1, tuple,
 						    sea_cur, cursor,
 						    page_no, mtr);
 	if (!node_ptr) {
@@ -792,23 +780,18 @@ rtr_page_get_father_block(
 /*======================*/
 	rec_offs*	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*	sea_cur,/*!< in: search cursor, contains information
 				about parent nodes in search */
 	btr_cur_t*	cursor)	/*!< out: cursor on node pointer record,
 				its page x-latched */
 {
-	rec_t*  rec = page_rec_get_next(
+  rec_t *rec=
-		page_get_infimum_rec(buf_block_get_frame(block)));
+    page_rec_get_next(page_get_infimum_rec(cursor->block()->page.frame));
-	if (!rec) {
+  if (!rec)
     return nullptr;
-	}
+  cursor->page_cur.rec= rec;
-	btr_cur_position(index, rec, block, cursor);
+  return rtr_page_get_father_node_ptr(offsets, heap, sea_cur, cursor, mtr);
-
-	return(rtr_page_get_father_node_ptr(offsets, heap, sea_cur,
-					    cursor, mtr));
 }
 
 /*******************************************************************//**
@@ -826,7 +809,7 @@ rtr_create_rtr_info(
 {
 	rtr_info_t*	rtr_info;
 
-	index = index ? index : cursor->index;
+	index = index ? index : cursor->index();
 	ut_ad(index);
 
 	rtr_info = static_cast<rtr_info_t*>(ut_zalloc_nokey(sizeof(*rtr_info)));
@@ -1163,6 +1146,7 @@ rtr_cur_restore_position(
 	ut_ad(mtr->is_active());
 
 	index = btr_cur_get_index(btr_cur);
+	ut_ad(r_cursor->index() == btr_cur->index());
 
 	if (r_cursor->rel_pos == BTR_PCUR_AFTER_LAST_IN_TREE
 	    || r_cursor->rel_pos == BTR_PCUR_BEFORE_FIRST_IN_TREE) {
@@ -1222,7 +1206,6 @@ rtr_cur_restore_position(
 
 	/* Page has changed, for R-Tree, the page cannot be shrunk away,
 	so we search the page and its right siblings */
-	buf_block_t*	block;
 	node_seq_t	page_ssn;
 	const page_t*	page;
 	page_cur_t*	page_cursor;
@@ -1242,21 +1225,21 @@ rtr_cur_restore_position(
 search_again:
 	ulint up_match = 0, low_match = 0;
 
-	block = buf_page_get_gen(
+	page_cursor->block = buf_page_get_gen(
 		page_id_t(index->table->space_id, page_no),
 		zip_size, RW_X_LATCH, NULL, BUF_GET, mtr);
 
-	if (!block) {
+	if (!page_cursor->block) {
 corrupted:
 		ret = false;
 		goto func_exit;
 	}
 
 	/* Get the page SSN */
-	page = buf_block_get_frame(block);
+	page = buf_block_get_frame(page_cursor->block);
 	page_ssn = page_get_ssn_id(page);
 
-	if (page_cur_search_with_match(block, index, tuple, PAGE_CUR_LE,
+	if (page_cur_search_with_match(tuple, PAGE_CUR_LE,
 				       &up_match, &low_match, page_cursor,
 				       nullptr)) {
 		goto corrupted;
@@ -1352,7 +1335,7 @@ rtr_store_parent_path(
 /*==================*/
 	const buf_block_t*	block,	/*!< in: block of the page */
 	btr_cur_t*		btr_cur,/*!< in/out: persistent cursor */
-	ulint			latch_mode,
+	btr_latch_mode		latch_mode,
 					/*!< in: latch_mode */
 	ulint			level,	/*!< in: index level */
 	mtr_t*			mtr)	/*!< in: mtr */
@@ -1411,7 +1394,7 @@ rtr_non_leaf_insert_stack_push(
 
 	page_cur_position(rec, block, btr_pcur_get_page_cur(my_cursor));
 
-	(btr_pcur_get_btr_cur(my_cursor))->index = index;
+	btr_pcur_get_page_cur(my_cursor)->index = index;
 
 	new_seq = rtr_get_current_ssn_id(index);
 	rtr_non_leaf_stack_push(path, block->page.id().page_no(),

storage/innobase/handler/ha_innodb.cc

@@ -1542,7 +1542,8 @@ static void innodb_drop_database(handlerton*, char *path)
     std::vector<pfs_os_file_t> to_close;
     mtr_t mtr;
     mtr.start();
-    err= btr_pcur_open_on_user_rec(sys_index, &tuple, PAGE_CUR_GE,
+    pcur.btr_cur.page_cur.index = sys_index;
+    err= btr_pcur_open_on_user_rec(&tuple, PAGE_CUR_GE,
                                    BTR_SEARCH_LEAF, &pcur, &mtr);
     if (err != DB_SUCCESS)
       goto err_exit;
@@ -2000,9 +2001,8 @@ static void drop_garbage_tables_after_restore()
   ut_d(purge_sys.stop_FTS());
 
   mtr.start();
-  if (btr_pcur_open_at_index_side(true, dict_sys.sys_tables->indexes.start,
+  if (pcur.open_leaf(true, dict_sys.sys_tables->indexes.start, BTR_SEARCH_LEAF,
-                                  BTR_SEARCH_LEAF, &pcur, true, 0, &mtr) !=
+                     &mtr) != DB_SUCCESS)
-      DB_SUCCESS)
     goto all_fail;
   for (;;)
   {
@@ -14997,14 +14997,10 @@ inline int ha_innobase::defragment_table()
     }
 
     btr_pcur_t pcur;
-    pcur.btr_cur.index = nullptr;
-    btr_pcur_init(&pcur);
 
     mtr_t mtr;
     mtr.start();
-    if (dberr_t err= btr_pcur_open_at_index_side(true, index,
+    if (dberr_t err= pcur.open_leaf(true, index, BTR_SEARCH_LEAF, &mtr))
-                                                 BTR_SEARCH_LEAF, &pcur,
-                                                 true, 0, &mtr))
     {
       mtr.commit();
       return convert_error_code_to_mysql(err, 0, m_user_thd);
@@ -15018,9 +15014,9 @@ inline int ha_innobase::defragment_table()
     btr_pcur_move_to_next(&pcur, &mtr);
     btr_pcur_store_position(&pcur, &mtr);
     mtr.commit();
-    ut_ad(pcur.btr_cur.index == index);
+    ut_ad(pcur.index() == index);
     const bool interrupted= btr_defragment_add_index(&pcur, m_user_thd);
-    btr_pcur_free(&pcur);
+    ut_free(pcur.old_rec_buf);
     if (interrupted)
       return ER_QUERY_INTERRUPTED;
   }

storage/innobase/handler/handler0alter.cc

@@ -2128,8 +2128,7 @@ static bool innobase_table_is_empty(const dict_table_t *table,
   bool next_page= false;
 
   mtr.start();
-  if (btr_pcur_open_at_index_side(true, clust_index, BTR_SEARCH_LEAF,
+  if (pcur.open_leaf(true, clust_index, BTR_SEARCH_LEAF, &mtr) != DB_SUCCESS)
-                                  &pcur, true, 0, &mtr) != DB_SUCCESS)
   {
 non_empty:
     mtr.commit();
@@ -2159,10 +2158,11 @@ next_page:
                          &mtr);
     if (!block)
       goto non_empty;
-    btr_leaf_page_release(page_cur_get_block(cur), BTR_SEARCH_LEAF, &mtr);
     page_cur_set_before_first(block, cur);
     if (UNIV_UNLIKELY(!page_cur_move_to_next(cur)))
       goto non_empty;
+    const auto s= mtr.get_savepoint();
+    mtr.rollback_to_savepoint(s - 2, s - 1);
   }
 
   rec= page_cur_get_rec(cur);
@@ -6038,8 +6038,7 @@ add_all_virtual:
 	mtr.start();
 	index->set_modified(mtr);
 	btr_pcur_t pcur;
-	dberr_t err = btr_pcur_open_at_index_side(true, index, BTR_MODIFY_TREE,
+	dberr_t err= pcur.open_leaf(true, index, BTR_MODIFY_TREE, &mtr);
-						  &pcur, true, 0, &mtr);
 	if (err != DB_SUCCESS) {
 func_exit:
 		mtr.commit();

storage/innobase/ibuf/ibuf0ibuf.cc

@@ -466,9 +466,8 @@ err_exit:
 	ib::info() << "Dumping the change buffer";
 	ibuf_mtr_start(&mtr);
 	btr_pcur_t pcur;
-	if (DB_SUCCESS == btr_pcur_open_at_index_side(
+	if (DB_SUCCESS
-		    true, ibuf.index, BTR_SEARCH_LEAF, &pcur,
+	    == pcur.open_leaf(true, ibuf.index, BTR_SEARCH_LEAF, &mtr)) {
-		    true, 0, &mtr)) {
 		while (btr_pcur_move_to_next_user_rec(&pcur, &mtr)) {
 			rec_print_old(stderr, btr_pcur_get_rec(&pcur));
 		}
@@ -2337,9 +2336,11 @@ work_around:
 			space_id, page_nos[i], heap);
 loop:
 		btr_pcur_t pcur;
+		pcur.btr_cur.page_cur.index = ibuf.index;
+
 		ibuf_mtr_start(&mtr);
-		if (btr_pcur_open(ibuf.index, tuple, PAGE_CUR_GE,
+		if (btr_pcur_open(tuple, PAGE_CUR_GE,
-				  BTR_MODIFY_LEAF, &pcur, &mtr)
+				  BTR_MODIFY_LEAF, &pcur, 0, &mtr)
 		    != DB_SUCCESS) {
 			goto done;
 		}
@@ -2386,16 +2387,11 @@ done:
 #endif
 }
 
-/*********************************************************************//**
+/** Contract the change buffer by reading pages to the buffer pool.
-Contracts insert buffer trees by reading pages to the buffer pool.
 @return 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
+will be merged from ibuf trees to the pages read
-empty */
+@retval 0 if ibuf.empty */
-static
+ulint ibuf_contract()
-ulint
-ibuf_merge_pages(
-/*=============*/
-	ulint*	n_pages)	/*!< out: number of pages to which merged */
 {
 	mtr_t		mtr;
 	btr_pcur_t	pcur;
@@ -2403,15 +2399,13 @@ ibuf_merge_pages(
 	uint32_t	page_nos[IBUF_MAX_N_PAGES_MERGED];
 	uint32_t	space_ids[IBUF_MAX_N_PAGES_MERGED];
 
-	*n_pages = 0;
-
 	ibuf_mtr_start(&mtr);
 
 	/* Open a cursor to a randomly chosen leaf of the tree, at a random
 	position within the leaf */
 	pcur.pos_state = BTR_PCUR_IS_POSITIONED;
-	pcur.old_stored = false;
+	pcur.old_rec = nullptr;
-	pcur.trx_if_known = NULL;
+	pcur.trx_if_known = nullptr;
 	pcur.search_mode = PAGE_CUR_G;
 	pcur.latch_mode = BTR_SEARCH_LEAF;
 
@@ -2437,13 +2431,14 @@ ibuf_merge_pages(
 		return(0);
 	}
 
+	ulint n_pages = 0;
 	sum_sizes = ibuf_get_merge_page_nos(TRUE,
 					    btr_pcur_get_rec(&pcur), &mtr,
 					    space_ids,
-					    page_nos, n_pages);
+					    page_nos, &n_pages);
 	ibuf_mtr_commit(&mtr);
 
-	ibuf_read_merge_pages(space_ids, page_nos, *n_pages);
+	ibuf_read_merge_pages(space_ids, page_nos, n_pages);
 
 	return(sum_sizes + 1);
 }
@@ -2470,8 +2465,9 @@ ibuf_merge_space(
 
 	/* Position the cursor on the first matching record. */
 
-	dberr_t err = btr_pcur_open(ibuf.index, tuple, PAGE_CUR_GE,
+	pcur.btr_cur.page_cur.index = ibuf.index;
-				    BTR_SEARCH_LEAF, &pcur, &mtr);
+	dberr_t err = btr_pcur_open(tuple, PAGE_CUR_GE,
+				    BTR_SEARCH_LEAF, &pcur, 0, &mtr);
 	ut_ad(err != DB_SUCCESS || page_validate(btr_pcur_get_page(&pcur),
 						 ibuf.index));
 
@@ -2515,73 +2511,6 @@ ibuf_merge_space(
 	return(n_pages);
 }
 
-/** Contract the change buffer by reading pages to the buffer pool.
-@param[out]	n_pages		number of pages merged
-@param[in]	sync		whether the caller waits for
-the issued reads to complete
-@return 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 */
-MY_ATTRIBUTE((warn_unused_result))
-static ulint ibuf_merge(ulint* n_pages)
-{
-	*n_pages = 0;
-
-	/* We perform a dirty read of ibuf.empty, without latching
-	the insert buffer root page. We trust this dirty read except
-	when a slow shutdown is being executed. During a slow
-	shutdown, the insert buffer merge must be completed. */
-
-	if (ibuf.empty && srv_shutdown_state <= SRV_SHUTDOWN_INITIATED) {
-		return(0);
-#if defined UNIV_DEBUG || defined UNIV_IBUF_DEBUG
-	} else if (ibuf_debug) {
-		return(0);
-#endif /* UNIV_DEBUG || UNIV_IBUF_DEBUG */
-	} else {
-		return ibuf_merge_pages(n_pages);
-	}
-}
-
-/** Contract the change buffer by reading pages to the buffer pool.
-@return 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 */
-static ulint ibuf_contract()
-{
-	ulint n_pages;
-	return ibuf_merge_pages(&n_pages);
-}
-
-/** Contract the change buffer by reading pages to the buffer pool.
-@return 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 ibuf_merge_all()
-{
-#if defined UNIV_DEBUG || defined UNIV_IBUF_DEBUG
-	if (ibuf_debug) {
-		return(0);
-	}
-#endif /* UNIV_DEBUG || UNIV_IBUF_DEBUG */
-
-	ulint	sum_bytes	= 0;
-	ulint	n_pages = srv_io_capacity;
-
-	for (ulint sum_pages = 0; sum_pages < n_pages; ) {
-		log_free_check();
-		ulint n_pag2;
-		ulint n_bytes = ibuf_merge(&n_pag2);
-
-		if (n_bytes == 0) {
-			break;
-		}
-
-		sum_bytes += n_bytes;
-	}
-
-	return sum_bytes;
-}
-
 /*********************************************************************//**
 Contract insert buffer trees after insert if they are too big. */
 UNIV_INLINE
@@ -2591,13 +2520,7 @@ ibuf_contract_after_insert(
 	ulint	entry_size)	/*!< in: size of a record which was inserted
 				into an ibuf tree */
 {
-	/* Perform dirty reads of ibuf.size and ibuf.max_size, to
+	/* dirty comparison, to avoid contention on ibuf_mutex */
-	reduce ibuf_mutex contention. ibuf.max_size remains constant
-	after ibuf_init_at_db_start(), but ibuf.size should be
-	protected by ibuf_mutex. Given that ibuf.size fits in a
-	machine word, this should be OK; at worst we are doing some
-	excessive ibuf_contract() or occasionally skipping a
-	ibuf_contract(). */
 	if (ibuf.size < ibuf.max_size) {
 		return;
 	}
@@ -2991,8 +2914,8 @@ ibuf_update_max_tablespace_id(void)
 
 	ibuf_mtr_start(&mtr);
 
-	if (btr_pcur_open_at_index_side(false, ibuf.index, BTR_SEARCH_LEAF,
+	if (pcur.open_leaf(false, ibuf.index, BTR_SEARCH_LEAF, &mtr)
-					&pcur, true, 0, &mtr) != DB_SUCCESS) {
+	    != DB_SUCCESS) {
 func_exit:
 		ibuf_mtr_commit(&mtr);
 		return;
@@ -3170,7 +3093,7 @@ ibuf_index_page_calc_free_from_bits(ulint physical_size, ulint bits)
 
 /** Buffer an operation in the insert/delete buffer, instead of doing it
 directly to the disk page, if this is possible.
-@param[in]	mode		BTR_MODIFY_PREV or BTR_MODIFY_TREE
+@param[in]	mode		BTR_MODIFY_PREV or BTR_INSERT_TREE
 @param[in]	op		operation type
 @param[in]	no_counter	TRUE=use 5.0.3 format; FALSE=allow delete
 buffering
@@ -3185,7 +3108,7 @@ or clustered
 static TRANSACTIONAL_TARGET MY_ATTRIBUTE((warn_unused_result))
 dberr_t
 ibuf_insert_low(
-	ulint			mode,
+	btr_latch_mode		mode,
 	ibuf_op_t		op,
 	ibool			no_counter,
 	const dtuple_t*		entry,
@@ -3225,16 +3148,17 @@ ibuf_insert_low(
 
 	do_merge = FALSE;
 
-	/* Perform dirty reads of ibuf.size and ibuf.max_size, to
+	/* Perform dirty comparison of ibuf.max_size and ibuf.size to
-	reduce ibuf_mutex contention. Given that ibuf.max_size and
+	reduce ibuf_mutex contention. This should be OK; at worst we
-	ibuf.size fit in a machine word, this should be OK; at worst
+	are doing some excessive ibuf_contract() or occasionally
-	we are doing some excessive ibuf_contract() or occasionally
 	skipping an ibuf_contract(). */
-	if (ibuf.max_size == 0) {
+	const ulint max_size = ibuf.max_size;
+
+	if (max_size == 0) {
 		return(DB_STRONG_FAIL);
 	}
 
-	if (ibuf.size >= ibuf.max_size + IBUF_CONTRACT_DO_NOT_INSERT) {
+	if (ibuf.size >= max_size + IBUF_CONTRACT_DO_NOT_INSERT) {
 		/* Insert buffer is now too big, contract it but do not try
 		to insert */
 
@@ -3265,7 +3189,7 @@ ibuf_insert_low(
 	the new entry to it without exceeding the free space limit for the
 	page. */
 
-	if (BTR_LATCH_MODE_WITHOUT_INTENTION(mode) == BTR_MODIFY_TREE) {
+	if (mode == BTR_INSERT_TREE) {
 		for (;;) {
 			mysql_mutex_lock(&ibuf_pessimistic_insert_mutex);
 			mysql_mutex_lock(&ibuf_mutex);
@@ -3287,18 +3211,16 @@ ibuf_insert_low(
 	}
 
 	ibuf_mtr_start(&mtr);
+	pcur.btr_cur.page_cur.index = ibuf.index;
 
-	err = btr_pcur_open(ibuf.index, ibuf_entry, PAGE_CUR_LE, mode, &pcur,
+	err = btr_pcur_open(ibuf_entry, PAGE_CUR_LE, mode, &pcur, 0, &mtr);
-			    &mtr);
 	if (err != DB_SUCCESS) {
 func_exit:
 		ibuf_mtr_commit(&mtr);
 		ut_free(pcur.old_rec_buf);
 		mem_heap_free(heap);
 
-		if (err == DB_SUCCESS
+		if (err == DB_SUCCESS && mode == BTR_INSERT_TREE) {
-		    && BTR_LATCH_MODE_WITHOUT_INTENTION(mode)
-		    == BTR_MODIFY_TREE) {
 			ibuf_contract_after_insert(entry_size);
 		}
 
@@ -3345,7 +3267,7 @@ func_exit:
 		until after the IBUF_OP_DELETE has been buffered. */
 
 fail_exit:
-		if (BTR_LATCH_MODE_WITHOUT_INTENTION(mode) == BTR_MODIFY_TREE) {
+		if (mode == BTR_INSERT_TREE) {
 			mysql_mutex_unlock(&ibuf_mutex);
 			mysql_mutex_unlock(&ibuf_pessimistic_insert_mutex);
 		}
@@ -3461,8 +3383,7 @@ commit_exit:
 			ibuf.empty = page_is_empty(root);
 		}
 	} else {
-		ut_ad(BTR_LATCH_MODE_WITHOUT_INTENTION(mode)
+		ut_ad(mode == BTR_INSERT_TREE);
-		      == BTR_MODIFY_TREE);
 
 		/* We acquire an sx-latch to the root page before the insert,
 		because a pessimistic insert releases the tree x-latch,
@@ -3637,7 +3558,7 @@ skip_watch:
 			      entry, entry_size,
 			      index, page_id, zip_size, thr);
 	if (err == DB_FAIL) {
-		err = ibuf_insert_low(BTR_MODIFY_TREE | BTR_LATCH_FOR_INSERT,
+		err = ibuf_insert_low(BTR_INSERT_TREE,
 				      op, no_counter, entry, entry_size,
 				      index, page_id, zip_size, thr);
 	}
@@ -3658,30 +3579,27 @@ dberr_t
 ibuf_insert_to_index_page_low(
 /*==========================*/
 	const 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 */
 	rec_offs**	offsets,/*!< out: offsets on *rec */
 	mem_heap_t*	heap,	/*!< in/out: memory heap */
 	mtr_t*		mtr,	/*!< in/out: mtr */
 	page_cur_t*	page_cur)/*!< in/out: cursor positioned on the record
 				after which to insert the buffered entry */
 {
-  if (page_cur_tuple_insert(page_cur, entry, index, offsets, &heap, 0, mtr))
+  if (page_cur_tuple_insert(page_cur, entry, offsets, &heap, 0, mtr))
     return DB_SUCCESS;
 
   /* Page reorganization or recompression should already have been
   attempted by page_cur_tuple_insert(). Besides, per
   ibuf_index_page_calc_free_zip() the page should not have been
   recompressed or reorganized. */
-  ut_ad(!is_buf_block_get_page_zip(block));
+  ut_ad(!is_buf_block_get_page_zip(page_cur->block));
 
   /* If the record did not fit, reorganize */
-  if (dberr_t err= btr_page_reorganize(page_cur, index, mtr))
+  if (dberr_t err= btr_page_reorganize(page_cur, mtr))
     return err;
 
   /* This time the record must fit */
-  if (page_cur_tuple_insert(page_cur, entry, index, offsets, &heap, 0, mtr))
+  if (page_cur_tuple_insert(page_cur, entry, offsets, &heap, 0, mtr))
     return DB_SUCCESS;
 
   return DB_CORRUPTION;
@@ -3738,8 +3656,10 @@ ibuf_insert_to_index_page(
 	}
 
 	ulint up_match = 0, low_match = 0;
+	page_cur.index = index;
+	page_cur.block = block;
 
-	if (page_cur_search_with_match(block, index, entry, PAGE_CUR_LE,
+	if (page_cur_search_with_match(entry, PAGE_CUR_LE,
 				       &up_match, &low_match, &page_cur,
 				       nullptr)) {
 		return DB_CORRUPTION;
@@ -3786,7 +3706,7 @@ ibuf_insert_to_index_page(
 		if (!row_upd_changes_field_size_or_external(index, offsets,
 							    update)
 		    && (!page_zip || btr_cur_update_alloc_zip(
-				page_zip, &page_cur, index, offsets,
+				page_zip, &page_cur, offsets,
 				rec_offs_size(offsets), false, mtr))) {
 			/* This is the easy case. Do something similar
 			to btr_cur_update_in_place(). */
@@ -3827,7 +3747,7 @@ ibuf_insert_to_index_page(
 		/* Delete the different-length record, and insert the
 		buffered one. */
 
-		page_cur_delete_rec(&page_cur, index, offsets, mtr);
+		page_cur_delete_rec(&page_cur, offsets, mtr);
 		if (!(page_cur_move_to_prev(&page_cur))) {
 			err = DB_CORRUPTION;
 			goto updated_in_place;
@@ -3836,8 +3756,8 @@ ibuf_insert_to_index_page(
 		offsets = NULL;
 	}
 
-	err = ibuf_insert_to_index_page_low(entry, block, index,
+	err = ibuf_insert_to_index_page_low(entry, &offsets, heap, mtr,
-					    &offsets, heap, mtr, &page_cur);
+                                            &page_cur);
 updated_in_place:
 	mem_heap_free(heap);
 
@@ -3853,16 +3773,18 @@ ibuf_set_del_mark(
 /*==============*/
 	const dtuple_t*		entry,	/*!< in: entry */
 	buf_block_t*		block,	/*!< in/out: block */
-	const dict_index_t*	index,	/*!< in: record descriptor */
+	dict_index_t*		index,	/*!< in: record descriptor */
 	mtr_t*			mtr)	/*!< in: mtr */
 {
 	page_cur_t	page_cur;
+	page_cur.block = block;
+	page_cur.index = index;
 	ulint		up_match = 0, low_match = 0;
 
 	ut_ad(ibuf_inside(mtr));
 	ut_ad(dtuple_check_typed(entry));
 
-	if (!page_cur_search_with_match(block, index, entry, PAGE_CUR_LE,
+	if (!page_cur_search_with_match(entry, PAGE_CUR_LE,
 					&up_match, &low_match, &page_cur,
 					nullptr)
 	    && low_match == dtuple_get_n_fields(entry)) {
@@ -3914,6 +3836,8 @@ ibuf_delete(
 				before latching any further pages */
 {
 	page_cur_t	page_cur;
+	page_cur.block = block;
+	page_cur.index = index;
 	ulint		up_match = 0, low_match = 0;
 
 	ut_ad(ibuf_inside(mtr));
@@ -3921,7 +3845,7 @@ ibuf_delete(
 	ut_ad(!index->is_spatial());
 	ut_ad(!index->is_clust());
 
-	if (!page_cur_search_with_match(block, index, entry, PAGE_CUR_LE,
+	if (!page_cur_search_with_match(entry, PAGE_CUR_LE,
 					&up_match, &low_match, &page_cur,
 					nullptr)
 	    && low_match == dtuple_get_n_fields(entry)) {
@@ -3974,7 +3898,7 @@ ibuf_delete(
 #ifdef UNIV_ZIP_DEBUG
 		ut_a(!page_zip || page_zip_validate(page_zip, page, index));
 #endif /* UNIV_ZIP_DEBUG */
-		page_cur_delete_rec(&page_cur, index, offsets, mtr);
+		page_cur_delete_rec(&page_cur, offsets, mtr);
 #ifdef UNIV_ZIP_DEBUG
 		ut_a(!page_zip || page_zip_validate(page_zip, page, index));
 #endif /* UNIV_ZIP_DEBUG */
@@ -4001,13 +3925,12 @@ ibuf_restore_pos(
 	const page_id_t	page_id,/*!< in: page identifier */
 	const dtuple_t*	search_tuple,
 				/*!< in: search tuple for entries of page_no */
-	ulint		mode,	/*!< in: BTR_MODIFY_LEAF or BTR_MODIFY_TREE */
+	btr_latch_mode	mode,	/*!< in: BTR_MODIFY_LEAF or BTR_PURGE_TREE */
 	btr_pcur_t*	pcur,	/*!< in/out: persistent cursor whose
 				position is to be restored */
 	mtr_t*		mtr)	/*!< in/out: mini-transaction */
 {
-	ut_ad(mode == BTR_MODIFY_LEAF
+	ut_ad(mode == BTR_MODIFY_LEAF || mode == BTR_PURGE_TREE);
-	      || BTR_LATCH_MODE_WITHOUT_INTENTION(mode) == BTR_MODIFY_TREE);
 
 	if (UNIV_LIKELY(pcur->restore_position(mode, mtr) ==
 	      btr_pcur_t::SAME_ALL)) {
@@ -4090,8 +4013,7 @@ bool ibuf_delete_rec(const page_id_t page_id, btr_pcur_t* pcur,
 	ibuf_mtr_start(mtr);
 	mysql_mutex_lock(&ibuf_mutex);
 
-	if (!ibuf_restore_pos(page_id, search_tuple,
+	if (!ibuf_restore_pos(page_id, search_tuple, BTR_PURGE_TREE,
-			      BTR_MODIFY_TREE | BTR_LATCH_FOR_DELETE,
 			      pcur, mtr)) {
 
 		mysql_mutex_unlock(&ibuf_mutex);
@@ -4278,13 +4200,14 @@ dberr_t ibuf_merge_or_delete_for_page(buf_block_t *block,
 
 	memset(mops, 0, sizeof(mops));
 	memset(dops, 0, sizeof(dops));
+	pcur.btr_cur.page_cur.index = ibuf.index;
 
 loop:
 	ibuf_mtr_start(&mtr);
 
 	/* Position pcur in the insert buffer at the first entry for this
 	index page */
-	if (btr_pcur_open_on_user_rec(ibuf.index, search_tuple, PAGE_CUR_GE,
+	if (btr_pcur_open_on_user_rec(search_tuple, PAGE_CUR_GE,
 				      BTR_MODIFY_LEAF, &pcur, &mtr)
 	    != DB_SUCCESS) {
 		err = DB_CORRUPTION;
@@ -4437,7 +4360,6 @@ loop:
 			goto loop;
 		} else if (btr_pcur_is_after_last_on_page(&pcur)) {
 			ibuf_mtr_commit(&mtr);
-			ut_free(pcur.old_rec_buf);
 			goto loop;
 		}
 	}
@@ -4485,13 +4407,15 @@ void ibuf_delete_for_discarded_space(uint32_t space)
 	search_tuple = ibuf_search_tuple_build(space, 0, heap);
 
 	memset(dops, 0, sizeof(dops));
+	pcur.btr_cur.page_cur.index = ibuf.index;
+
 loop:
 	log_free_check();
 	ibuf_mtr_start(&mtr);
 
 	/* Position pcur in the insert buffer at the first entry for the
 	space */
-	if (btr_pcur_open_on_user_rec(ibuf.index, search_tuple, PAGE_CUR_GE,
+	if (btr_pcur_open_on_user_rec(search_tuple, PAGE_CUR_GE,
 				      BTR_MODIFY_LEAF, &pcur, &mtr)
 	    != DB_SUCCESS) {
 		goto leave_loop;
@@ -4577,7 +4501,7 @@ ibuf_print(
 	fprintf(file,
 		"Ibuf: size " ULINTPF ", free list len " ULINTPF ","
 		" seg size " ULINTPF ", " ULINTPF " merges\n",
-		ibuf.size,
+		ulint{ibuf.size},
 		ibuf.free_list_len,
 		ibuf.seg_size,
 		ulint{ibuf.n_merges});

storage/innobase/include/btr0btr.h

@@ -55,103 +55,8 @@ 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(). */
-enum btr_latch_mode {
-	/** Search a record on a leaf page and S-latch it. */
-	BTR_SEARCH_LEAF = RW_S_LATCH,
-	/** (Prepare to) modify a record on a leaf page and X-latch it. */
-	BTR_MODIFY_LEAF	= RW_X_LATCH,
-	/** Obtain no latches. */
-	BTR_NO_LATCHES = RW_NO_LATCH,
-	/** Search the previous record. */
-	BTR_SEARCH_PREV = 4 | BTR_SEARCH_LEAF,
-	/** Modify the previous record. */
-	BTR_MODIFY_PREV = 4 | BTR_MODIFY_LEAF,
-	/** Start searching the entire B-tree. */
-	BTR_SEARCH_TREE = 8 | BTR_SEARCH_LEAF,
-	/** Start modifying1 the entire B-tree. */
-	BTR_MODIFY_TREE = 8 | BTR_MODIFY_LEAF,
-	/** Continue searching the entire B-tree. */
-	BTR_CONT_SEARCH_TREE = 4 | BTR_SEARCH_TREE,
-	/** Continue modifying the entire B-tree. */
-	BTR_CONT_MODIFY_TREE = 4 | BTR_MODIFY_TREE,
-
-	/* BTR_INSERT, BTR_DELETE and BTR_DELETE_MARK are mutually
-	exclusive. */
-	/** The search tuple will be inserted to the secondary index
-	at the searched position.  When the leaf page is not in the
-	buffer pool, try to use the change buffer. */
-	BTR_INSERT = 64,
-
-	/** Try to delete mark a secondary index leaf page record at
-	the searched position using the change buffer when the page is
-	not in the buffer pool. */
-	BTR_DELETE_MARK	= 128,
-
-	/** Try to purge the record using the change buffer when the
-	secondary index leaf page is not in the buffer pool. */
-	BTR_DELETE = BTR_INSERT | BTR_DELETE_MARK,
-
-	/** The caller is already holding dict_index_t::lock S-latch. */
-	BTR_ALREADY_S_LATCHED = 256,
-	/** Search and S-latch a leaf page, assuming that the
-	dict_index_t::lock S-latch is being held. */
-	BTR_SEARCH_LEAF_ALREADY_S_LATCHED = BTR_SEARCH_LEAF
-	| BTR_ALREADY_S_LATCHED,
-	/** Search the entire index tree, assuming that the
-	dict_index_t::lock S-latch is being held. */
-	BTR_SEARCH_TREE_ALREADY_S_LATCHED = BTR_SEARCH_TREE
-	| BTR_ALREADY_S_LATCHED,
-	/** Search and X-latch a leaf page, assuming that the
-	dict_index_t::lock S-latch is being held. */
-	BTR_MODIFY_LEAF_ALREADY_S_LATCHED = BTR_MODIFY_LEAF
-	| BTR_ALREADY_S_LATCHED,
-
-	/** Attempt to delete-mark a secondary index record. */
-	BTR_DELETE_MARK_LEAF = BTR_MODIFY_LEAF | BTR_DELETE_MARK,
-	/** Attempt to delete-mark a secondary index record
-	while holding the dict_index_t::lock S-latch. */
-	BTR_DELETE_MARK_LEAF_ALREADY_S_LATCHED = BTR_DELETE_MARK_LEAF
-	| BTR_ALREADY_S_LATCHED,
-	/** Attempt to purge a secondary index record. */
-	BTR_PURGE_LEAF = BTR_MODIFY_LEAF | BTR_DELETE,
-	/** Attempt to purge a secondary index record
-	while holding the dict_index_t::lock S-latch. */
-	BTR_PURGE_LEAF_ALREADY_S_LATCHED = BTR_PURGE_LEAF
-	| BTR_ALREADY_S_LATCHED,
-
-	/** In the case of BTR_MODIFY_TREE, the caller specifies
-	the intention to delete record only. It is used to optimize
-	block->lock range.*/
-	BTR_LATCH_FOR_DELETE = 512,
-
-	/** Attempt to purge a secondary index record in the tree. */
-	BTR_PURGE_TREE = BTR_MODIFY_TREE | BTR_LATCH_FOR_DELETE
-};
-
-/** This flag ORed to BTR_INSERT 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	2048U
-
-/** In the case of BTR_MODIFY_TREE, the caller specifies the intention
-to insert record only. It is used to optimize block->lock range.*/
-#define BTR_LATCH_FOR_INSERT	4096U
-
-/** This flag is for undo insert of rtree. For rtree, we need this flag
-to find proper rec to undo insert.*/
-#define BTR_RTREE_UNDO_INS	8192U
-
-/** In the case of BTR_MODIFY_LEAF, the caller intends to allocate or
-free the pages of externally stored fields. */
-#define BTR_MODIFY_EXTERNAL	16384U
-
-/** Try to delete mark the record at the searched position when the
-record is in spatial index */
-#define BTR_RTREE_DELETE_MARK	32768U
-
 #define BTR_LATCH_MODE_WITHOUT_FLAGS(latch_mode)		\
-	((latch_mode) & ulint(~(BTR_INSERT			\
+	btr_latch_mode((latch_mode) & ~(BTR_INSERT	\
 				| BTR_DELETE_MARK		\
 				| BTR_RTREE_UNDO_INS		\
 				| BTR_RTREE_DELETE_MARK		\
@@ -159,13 +64,11 @@ record is in spatial index */
 				| BTR_IGNORE_SEC_UNIQUE		\
 				| BTR_ALREADY_S_LATCHED		\
 				| BTR_LATCH_FOR_INSERT		\
-				| BTR_LATCH_FOR_DELETE		\
+				| BTR_LATCH_FOR_DELETE))
-				| BTR_MODIFY_EXTERNAL)))
 
 #define BTR_LATCH_MODE_WITHOUT_INTENTION(latch_mode)			\
-	((latch_mode) & ulint(~(BTR_LATCH_FOR_INSERT		\
+	btr_latch_mode((latch_mode)					\
-				| BTR_LATCH_FOR_DELETE		\
+		       & ~(BTR_LATCH_FOR_INSERT | BTR_LATCH_FOR_DELETE))
-				| BTR_MODIFY_EXTERNAL)))
 
 /**************************************************************//**
 Checks and adjusts the root node of a tree during IMPORT TABLESPACE.
@@ -228,17 +131,6 @@ inline uint32_t btr_page_get_prev(const page_t* page)
   return mach_read_from_4(my_assume_aligned<4>(page + FIL_PAGE_PREV));
 }
 
-/**************************************************************//**
-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 */
-	MY_ATTRIBUTE((nonnull));
 /**************************************************************//**
 Gets the child node file address in a node pointer.
 NOTE: the offsets array must contain all offsets for the record since
@@ -359,14 +251,11 @@ be done either within the same mini-transaction, or by invoking
 ibuf_reset_free_bits() before mtr_commit(). On uncompressed pages,
 IBUF_BITMAP_FREE is unaffected by reorganization.
 
+@param cursor  page cursor
+@param mtr     mini-transaction
 @return error code
 @retval DB_FAIL if reorganizing a ROW_FORMAT=COMPRESSED page failed */
-dberr_t
+dberr_t btr_page_reorganize(page_cur_t *cursor, mtr_t *mtr)
-btr_page_reorganize(
-/*================*/
-	page_cur_t*	cursor,	/*!< in/out: page cursor */
-	dict_index_t*	index,	/*!< in: the index tree of the page */
-	mtr_t*		mtr)	/*!< in/out: mini-transaction */
   MY_ATTRIBUTE((nonnull, warn_unused_result));
 /** Decide if the page should be split at the convergence point of inserts
 converging to the left.
@@ -445,13 +334,10 @@ inline void btr_set_min_rec_mark(rec_t *rec, const buf_block_t &block,
 }
 
 /** Seek to the parent page of a B-tree page.
-@param[in,out]	index	b-tree
-@param[in]	block	child page
 @param[in,out]	mtr	mini-transaction
-@param[out]	cursor	cursor pointing to the x-latched parent page
+@param[in,out]	cursor	cursor pointing to the x-latched parent page
 @return whether the cursor was successfully positioned */
-bool btr_page_get_father(dict_index_t* index, buf_block_t* block, mtr_t* mtr,
+bool btr_page_get_father(mtr_t* mtr, btr_cur_t* cursor)
-			 btr_cur_t* cursor)
 	MY_ATTRIBUTE((nonnull,warn_unused_result));
 #ifdef UNIV_DEBUG
 /************************************************************//**

storage/innobase/include/btr0btr.inl

@@ -1,7 +1,7 @@
 /*****************************************************************************
 
 Copyright (c) 1994, 2016, Oracle and/or its affiliates. All Rights Reserved.
-Copyright (c) 2015, 2021, MariaDB Corporation.
+Copyright (c) 2015, 2022, MariaDB Corporation.
 
 This program is free software; you can redistribute it and/or modify it under
 the terms of the GNU General Public License as published by the Free Software
@@ -24,10 +24,7 @@ The B-tree
 Created 6/2/1994 Heikki Tuuri
 *******************************************************/
 
-#include "mach0data.h"
-#include "mtr0mtr.h"
 #include "mtr0log.h"
-#include "page0zip.h"
 
 /**************************************************************//**
 Gets the index id field of a page.
@@ -112,38 +109,3 @@ btr_node_ptr_get_child_page_no(
 
 	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
-	      || latch_mode == BTR_NO_LATCHES);
-
-	ut_ad(!mtr->memo_contains_flagged(block, MTR_MEMO_MODIFY));
-
-	mtr_memo_type_t mode;
-	switch (latch_mode) {
-		case BTR_SEARCH_LEAF:
-			mode = MTR_MEMO_PAGE_S_FIX;
-			break;
-		case BTR_MODIFY_LEAF:
-			mode = MTR_MEMO_PAGE_X_FIX;
-			break;
-		case BTR_NO_LATCHES:
-			mode = MTR_MEMO_BUF_FIX;
-			break;
-		default:
-			ut_a(0);
-	}
-
-	mtr->memo_release(block, mode);
-}

storage/innobase/include/btr0cur.h

@@ -96,7 +96,7 @@ btr_cur_get_page(
 Returns the index of a cursor.
 @param cursor b-tree cursor
 @return index */
-#define btr_cur_get_index(cursor) ((cursor)->index)
+#define btr_cur_get_index(cursor) ((cursor)->index())
 /*********************************************************//**
 Positions a tree cursor at a given record. */
 UNIV_INLINE
@@ -137,7 +137,7 @@ bool
 btr_cur_optimistic_latch_leaves(
 	buf_block_t*	block,
 	ib_uint64_t	modify_clock,
-	ulint*		latch_mode,
+	btr_latch_mode*	latch_mode,
 	btr_cur_t*	cursor,
 	mtr_t*		mtr);
 
@@ -148,7 +148,6 @@ 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.
-@param index      index
 @param level      the tree level of search
 @param tuple      data tuple; NOTE: n_fields_cmp in tuple must be set so that
                   it cannot get compared to the node ptr page number field!
@@ -166,27 +165,13 @@ If mode is PAGE_CUR_GE, then up_match will a have a sensible value.
 @param mtr        mini-transaction
 @param autoinc    PAGE_ROOT_AUTO_INC to be written (0 if none)
 @return DB_SUCCESS on success or error code otherwise */
-dberr_t btr_cur_search_to_nth_level(dict_index_t *index, ulint level,
+dberr_t btr_cur_search_to_nth_level(ulint level,
                                     const dtuple_t *tuple,
-                                    page_cur_mode_t mode, ulint latch_mode,
+                                    page_cur_mode_t mode,
+                                    btr_latch_mode latch_mode,
                                     btr_cur_t *cursor, mtr_t *mtr,
                                     ib_uint64_t autoinc= 0);
 
-/*****************************************************************//**
-Opens a cursor at either end of an index.
-@return DB_SUCCESS or error code */
-dberr_t
-btr_cur_open_at_index_side(
-	bool		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/out: cursor */
-	ulint		level,		/*!< in: level to search for
-					(0=leaf) */
-	mtr_t*		mtr)		/*!< in/out: mini-transaction */
-	MY_ATTRIBUTE((nonnull, warn_unused_result));
-
 /**********************************************************************//**
 Positions a cursor at a randomly chosen position within a B-tree.
 @return true if the index is available and we have put the cursor, false
@@ -194,7 +179,7 @@ if the index is unavailable */
 bool
 btr_cur_open_at_rnd_pos(
 	dict_index_t*	index,		/*!< in: index */
-	ulint		latch_mode,	/*!< in: BTR_SEARCH_LEAF, ... */
+	btr_latch_mode	latch_mode,	/*!< in: BTR_SEARCH_LEAF, ... */
 	btr_cur_t*	cursor,		/*!< in/out: B-tree cursor */
 	mtr_t*		mtr)		/*!< in: mtr */
 	MY_ATTRIBUTE((nonnull,warn_unused_result));
@@ -281,7 +266,6 @@ btr_cur_update_alloc_zip_func(
 /*==========================*/
 	page_zip_des_t*	page_zip,/*!< in/out: compressed page */
 	page_cur_t*	cursor,	/*!< in/out: B-tree page cursor */
-	dict_index_t*	index,	/*!< in: the index corresponding to cursor */
 #ifdef UNIV_DEBUG
 	rec_offs*	offsets,/*!< in/out: offsets of the cursor record */
 #endif /* UNIV_DEBUG */
@@ -291,11 +275,11 @@ btr_cur_update_alloc_zip_func(
 	mtr_t*		mtr)	/*!< in/out: mini-transaction */
 	MY_ATTRIBUTE((nonnull, warn_unused_result));
 #ifdef UNIV_DEBUG
-# define btr_cur_update_alloc_zip(page_zip,cursor,index,offsets,len,cr,mtr) \
+# define btr_cur_update_alloc_zip(page_zip,cursor,offsets,len,cr,mtr) \
-	btr_cur_update_alloc_zip_func(page_zip,cursor,index,offsets,len,cr,mtr)
+	btr_cur_update_alloc_zip_func(page_zip,cursor,offsets,len,cr,mtr)
 #else /* UNIV_DEBUG */
-# define btr_cur_update_alloc_zip(page_zip,cursor,index,offsets,len,cr,mtr) \
+# define btr_cur_update_alloc_zip(page_zip,cursor,offsets,len,cr,mtr) \
-	btr_cur_update_alloc_zip_func(page_zip,cursor,index,len,cr,mtr)
+	btr_cur_update_alloc_zip_func(page_zip,cursor,len,cr,mtr)
 #endif /* UNIV_DEBUG */
 
 /** Apply an update vector to a record. No field size changes are allowed.
@@ -689,7 +673,7 @@ btr_rec_copy_externally_stored_field(
 void
 btr_cur_latch_leaves(
 	buf_block_t*		block,
-	ulint			latch_mode,
+	btr_latch_mode		latch_mode,
 	btr_cur_t*		cursor,
 	mtr_t*			mtr,
 	btr_latch_leaves_t*	latch_leaves = nullptr);
@@ -752,7 +736,6 @@ enum btr_cur_method {
 /** The tree cursor: the definition appears here only for the compiler
 to know struct size! */
 struct btr_cur_t {
-	dict_index_t*	index;		/*!< index where positioned */
 	page_cur_t	page_cur;	/*!< page cursor */
 	purge_node_t*	purge_node;	/*!< purge node, for BTR_DELETE */
 	buf_block_t*	left_block;	/*!< this field is used to store
@@ -817,28 +800,19 @@ struct btr_cur_t {
 					information of the path through
 					the tree */
 	rtr_info_t*	rtr_info;	/*!< rtree search info */
-	btr_cur_t():thr(NULL), rtr_info(NULL) {}
+  btr_cur_t() { memset((void*) this, 0, sizeof *this); }
-					/* default values */
+
-	/** Zero-initialize all fields */
+  dict_index_t *index() const { return page_cur.index; }
-	void init()
+  buf_block_t *block() const { return page_cur.block; }
-	{
+
-		index = NULL;
+  /** Open the cursor on the first or last record.
-		memset(&page_cur, 0, sizeof page_cur);
+  @param first         true=first record, false=last record
-		purge_node = NULL;
+  @param index         B-tree
-		left_block = NULL;
+  @param latch_mode    which latches to acquire
-		thr = NULL;
+  @param mtr           mini-transaction
-		flag = btr_cur_method(0);
+  @return error code */
-		tree_height = 0;
+  dberr_t open_leaf(bool first, dict_index_t *index, btr_latch_mode latch_mode,
-		up_match = 0;
+                    mtr_t *mtr);
-		up_bytes = 0;
-		low_match = 0;
-		low_bytes = 0;
-		n_fields = 0;
-		n_bytes = 0;
-		fold = 0;
-		path_arr = NULL;
-		rtr_info = NULL;
-	}
 };
 
 /** Modify the delete-mark flag of a record.

storage/innobase/include/btr0cur.inl

@@ -1,7 +1,7 @@
 /*****************************************************************************
 
 Copyright (c) 1994, 2015, Oracle and/or its affiliates. All Rights Reserved.
-Copyright (c) 2018, 2021, MariaDB Corporation.
+Copyright (c) 2018, 2022, MariaDB Corporation.
 
 This program is free software; you can redistribute it and/or modify it under
 the terms of the GNU General Public License as published by the Free Software
@@ -72,7 +72,7 @@ btr_cur_position(
 	btr_cur_t*	cursor)	/*!< out: cursor */
 {
 	page_cur_position(rec, block, btr_cur_get_page_cur(cursor));
-	cursor->index = index;
+	cursor->page_cur.index = index;
 }
 
 /*********************************************************************//**
@@ -98,14 +98,14 @@ btr_cur_compress_recommendation(
 
 	if (!page_has_siblings(page)
 	    || page_get_data_size(page)
-	    < BTR_CUR_PAGE_COMPRESS_LIMIT(cursor->index)) {
+	    < BTR_CUR_PAGE_COMPRESS_LIMIT(cursor->index())) {
 
 		/* 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 cursor->index->page
+		return cursor->index()->page
 			!= btr_cur_get_block(cursor)->page.id().page_no();
 	}
 
@@ -133,14 +133,14 @@ btr_cur_can_delete_without_compress(
 
 	if (!page_has_siblings(page) || page_get_n_recs(page) < 2
 	    || page_get_data_size(page) - rec_size
-	    < BTR_CUR_PAGE_COMPRESS_LIMIT(cursor->index)) {
+	    < BTR_CUR_PAGE_COMPRESS_LIMIT(cursor->index())) {
 
 		/* 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 cursor->index->page
+		return cursor->index()->page
 			== btr_cur_get_block(cursor)->page.id().page_no();
 	}
 

storage/innobase/include/btr0pcur.h

@@ -45,13 +45,6 @@ of a scroll cursor easier */
 	BTR_PCUR_AFTER_LAST_IN_TREE	= 5	/* in an empty tree */
 };
 
-/**************************************************************//**
-Allocates memory for a persistent cursor object and initializes the cursor.
-@return own: persistent cursor */
-btr_pcur_t*
-btr_pcur_create_for_mysql(void);
-/*============================*/
-
 /**************************************************************//**
 Resets a persistent cursor object, freeing ::old_rec_buf if it is
 allocated and resetting the other members to their initial values. */
@@ -61,12 +54,6 @@ btr_pcur_reset(
 	btr_pcur_t*	cursor);/*!< in, out: persistent cursor */
 
 /**************************************************************//**
-Frees the memory for a persistent cursor object. */
-void
-btr_pcur_free_for_mysql(
-/*====================*/
-	btr_pcur_t*	cursor);	/*!< in, own: persistent cursor */
-/**************************************************************//**
 Copies the stored position of a pcur to another pcur. */
 void
 btr_pcur_copy_stored_position(
@@ -83,21 +70,11 @@ btr_pcur_init(
 /*==========*/
 	btr_pcur_t*	pcur);	/*!< in: persistent cursor */
 
-/** Free old_rec_buf.
-@param[in]	pcur	Persistent cursor holding old_rec to be freed. */
-UNIV_INLINE
-void
-btr_pcur_free(
-	btr_pcur_t*	pcur);
-
 /**************************************************************//**
 Initializes and opens a persistent cursor to an index tree. */
 inline
 dberr_t
-btr_pcur_open_low(
+btr_pcur_open(
-/*==============*/
-	dict_index_t*	index,	/*!< in: index */
-	ulint		level,	/*!< in: level in the btree */
 	const dtuple_t*	tuple,	/*!< in: tuple on which search done */
 	page_cur_mode_t	mode,	/*!< in: PAGE_CUR_L, ...;
 				NOTE that if the search is made using a unique
@@ -105,17 +82,14 @@ btr_pcur_open_low(
 				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_latch_mode	latch_mode,/*!< in: BTR_SEARCH_LEAF, ... */
 	btr_pcur_t*	cursor, /*!< in: memory buffer for persistent cursor */
 	ib_uint64_t	autoinc,/*!< in: PAGE_ROOT_AUTO_INC to be written
 				(0 if none) */
 	mtr_t*		mtr)	/*!< in: mtr */
 	MY_ATTRIBUTE((nonnull, warn_unused_result));
-#define btr_pcur_open(i,t,md,l,c,m)				\
-	btr_pcur_open_low(i,0,t,md,l,c,0,m)
 /** Opens an persistent cursor to an index tree without initializing the
 cursor.
-@param index      index
 @param tuple      tuple on which search done
 @param mode       PAGE_CUR_L, ...; NOTE that if the search is made using a
                   unique prefix of a record, mode should be PAGE_CUR_LE, not
@@ -126,26 +100,11 @@ cursor.
 @param mtr        mini-transaction
 @return DB_SUCCESS on success or error code otherwise. */
 inline
-dberr_t btr_pcur_open_with_no_init(dict_index_t *index, const dtuple_t *tuple,
+dberr_t btr_pcur_open_with_no_init(const dtuple_t *tuple,
-                                   page_cur_mode_t mode, ulint latch_mode,
+                                   page_cur_mode_t mode,
+                                   btr_latch_mode latch_mode,
                                    btr_pcur_t *cursor, mtr_t *mtr);
 
-/*****************************************************************//**
-Opens a persistent cursor at either end of an index. */
-UNIV_INLINE
-dberr_t
-btr_pcur_open_at_index_side(
-/*========================*/
-	bool		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/out: cursor */
-	bool		init_pcur,	/*!< in: whether to initialize pcur */
-	ulint		level,		/*!< in: level to search for
-					(0=leaf) */
-	mtr_t*		mtr)		/*!< in/out: mini-transaction */
-	MY_ATTRIBUTE((nonnull,warn_unused_result));
 /**************************************************************//**
 Gets the up_match value for a pcur after a search.
 @return number of matched fields at the cursor or to the right if
@@ -356,7 +315,8 @@ enum pcur_pos_t {
 /* The persistent B-tree cursor structure. This is used mainly for SQL
 selects, updates, and deletes. */
 
-struct btr_pcur_t{
+struct btr_pcur_t
+{
   /** Return value of restore_position() */
   enum restore_status {
     /** cursor position on user rec and points on the record with
@@ -373,61 +333,42 @@ struct btr_pcur_t{
   };
   /** a B-tree cursor */
   btr_cur_t btr_cur;
-	/** see TODO note below!
+  /** @see BTR_PCUR_WAS_POSITIONED
   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; BTR_NO_LATCHES 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		latch_mode;
+  btr_latch_mode latch_mode= BTR_NO_LATCHES;
-	/** true if old_rec is stored */
-	bool		old_stored;
   /** if cursor position is stored, contains an initial segment of the
   latest record cursor was positioned either on, before or after */
-	rec_t*		old_rec;
+  rec_t *old_rec= nullptr;
-	/** btr_cur.index->n_core_fields when old_rec was copied */
+  /** btr_cur.index()->n_core_fields when old_rec was copied */
-	uint16		old_n_core_fields;
+  uint16 old_n_core_fields= 0;
   /** number of fields in old_rec */
-	uint16		old_n_fields;
+  uint16 old_n_fields= 0;
   /** BTR_PCUR_ON, BTR_PCUR_BEFORE, or BTR_PCUR_AFTER, depending on
   whether cursor was on, before, or after the old_rec record */
-	enum btr_pcur_pos_t	rel_pos;
+  btr_pcur_pos_t rel_pos= btr_pcur_pos_t(0);
   /** buffer block when the position was stored */
   buf::Block_hint block_when_stored;
   /** the modify clock value of the buffer block when the cursor position
   was stored */
-	ib_uint64_t	modify_clock;
+  ib_uint64_t modify_clock= 0;
   /** btr_pcur_store_position() and btr_pcur_restore_position() state. */
-	enum pcur_pos_t	pos_state;
+  enum pcur_pos_t pos_state= BTR_PCUR_NOT_POSITIONED;
-	/** PAGE_CUR_G, ... */
+  page_cur_mode_t search_mode= PAGE_CUR_UNSUPP;
-	page_cur_mode_t	search_mode;
   /** the transaction, if we know it; otherwise this field is not defined;
   can ONLY BE USED in error prints in fatal assertion failures! */
-	trx_t*		trx_if_known;
+  trx_t *trx_if_known= nullptr;
-	/*-----------------------------*/
+  /** a dynamically allocated buffer for old_rec */
-	/* NOTE that the following fields may possess dynamically allocated
+  byte *old_rec_buf= nullptr;
-	memory which should be freed if not needed anymore! */
-
-	/** NULL, or a dynamically allocated buffer for old_rec */
-	byte*		old_rec_buf;
   /** old_rec_buf size if old_rec_buf is not NULL */
-	ulint		buf_size;
+  ulint buf_size= 0;
-
-	btr_pcur_t() :
-		btr_cur(), latch_mode(RW_NO_LATCH),
-		old_stored(false), old_rec(NULL),
-		old_n_fields(0), rel_pos(btr_pcur_pos_t(0)),
-		block_when_stored(),
-		modify_clock(0), pos_state(BTR_PCUR_NOT_POSITIONED),
-		search_mode(PAGE_CUR_UNSUPP), trx_if_known(NULL),
-		old_rec_buf(NULL), buf_size(0)
-	{
-		btr_cur.init();
-	}
 
   /** Return the index of this persistent cursor */
-	dict_index_t*	index() const { return(btr_cur.index); }
+  dict_index_t *index() const { return(btr_cur.index()); }
   MY_ATTRIBUTE((nonnull, warn_unused_result))
   /** Restores the stored position of a persistent cursor bufferfixing
   the page and obtaining the specified latches. If the cursor position
@@ -451,7 +392,26 @@ struct btr_pcur_t{
   @retval NOT_SAME cursor position is not on user rec or points on
   the record with not the same uniq field values as in the stored
   @retval CORRUPTED if the index is corrupted */
-	restore_status restore_position(ulint latch_mode, mtr_t *mtr);
+  restore_status restore_position(btr_latch_mode latch_mode, mtr_t *mtr);
+
+  /** Open the cursor on the first or last record.
+  @param first         true=first record, false=last record
+  @param index         B-tree
+  @param latch_mode    which latches to acquire
+  @param mtr           mini-transaction
+  @return error code */
+  dberr_t open_leaf(bool first, dict_index_t *index, btr_latch_mode latch_mode,
+                    mtr_t *mtr)
+
+  {
+    this->latch_mode= BTR_LATCH_MODE_WITHOUT_FLAGS(latch_mode);
+    search_mode= first ? PAGE_CUR_G : PAGE_CUR_L;
+    pos_state= BTR_PCUR_IS_POSITIONED;
+    old_rec= nullptr;
+
+    return btr_cur.open_leaf(first, index,
+                             BTR_LATCH_MODE_WITHOUT_FLAGS(latch_mode), mtr);
+  }
 };
 
 inline buf_block_t *btr_pcur_get_block(btr_pcur_t *cursor)
@@ -479,10 +439,9 @@ MY_ATTRIBUTE((nonnull, warn_unused_result))
 inline
 dberr_t
 btr_pcur_open_on_user_rec(
-	dict_index_t*	index,		/*!< in: index */
 	const dtuple_t*	tuple,		/*!< in: tuple on which search done */
 	page_cur_mode_t	mode,		/*!< in: PAGE_CUR_L, ... */
-	ulint		latch_mode,	/*!< in: BTR_SEARCH_LEAF or
+	btr_latch_mode	latch_mode,	/*!< in: BTR_SEARCH_LEAF or
 					BTR_MODIFY_LEAF */
 	btr_pcur_t*	cursor,		/*!< in: memory buffer for persistent
 					cursor */
@@ -490,7 +449,7 @@ btr_pcur_open_on_user_rec(
 {
   ut_ad(mode == PAGE_CUR_GE || mode == PAGE_CUR_G);
   ut_ad(latch_mode == BTR_SEARCH_LEAF || latch_mode == BTR_MODIFY_LEAF);
-  if (dberr_t err= btr_pcur_open(index, tuple, mode, latch_mode, cursor, mtr))
+  if (dberr_t err= btr_pcur_open(tuple, mode, latch_mode, cursor, 0, mtr))
     return err;
   if (!btr_pcur_is_after_last_on_page(cursor) ||
       btr_pcur_is_after_last_in_tree(cursor))

storage/innobase/include/btr0pcur.inl

@@ -36,7 +36,6 @@ btr_pcur_get_rel_pos(
 {
 	ut_ad(cursor);
 	ut_ad(cursor->old_rec);
-	ut_ad(cursor->old_stored);
 	ut_ad(cursor->pos_state == BTR_PCUR_WAS_POSITIONED
 	      || cursor->pos_state == BTR_PCUR_IS_POSITIONED);
 
@@ -163,7 +162,7 @@ btr_pcur_move_to_next_on_page(
 	ut_ad(cursor->pos_state == BTR_PCUR_IS_POSITIONED);
 	ut_ad(cursor->latch_mode != BTR_NO_LATCHES);
 
-	cursor->old_stored = false;
+	cursor->old_rec = nullptr;
 	return page_cur_move_to_next(btr_pcur_get_page_cur(cursor));
 }
 
@@ -177,7 +176,7 @@ btr_pcur_move_to_prev_on_page(
 {
 	ut_ad(cursor->pos_state == BTR_PCUR_IS_POSITIONED);
 	ut_ad(cursor->latch_mode != BTR_NO_LATCHES);
-	cursor->old_stored = false;
+	cursor->old_rec = nullptr;
 
 	return page_cur_move_to_prev(btr_pcur_get_page_cur(cursor));
 }
@@ -196,7 +195,7 @@ btr_pcur_move_to_next_user_rec(
 {
 	ut_ad(cursor->pos_state == BTR_PCUR_IS_POSITIONED);
 	ut_ad(cursor->latch_mode != BTR_NO_LATCHES);
-	cursor->old_stored = false;
+	cursor->old_rec = nullptr;
 loop:
 	if (btr_pcur_is_after_last_on_page(cursor)) {
 		if (btr_pcur_is_after_last_in_tree(cursor)
@@ -230,7 +229,7 @@ btr_pcur_move_to_next(
   ut_ad(cursor->pos_state == BTR_PCUR_IS_POSITIONED);
   ut_ad(cursor->latch_mode != BTR_NO_LATCHES);
 
-  cursor->old_stored= false;
+  cursor->old_rec= nullptr;
 
   if (btr_pcur_is_after_last_on_page(cursor))
     return !btr_pcur_is_after_last_in_tree(cursor) &&
@@ -294,31 +293,17 @@ btr_pcur_init(
 /*==========*/
 	btr_pcur_t*	pcur)	/*!< in: persistent cursor */
 {
-	pcur->old_stored = false;
 	pcur->old_rec_buf = NULL;
 	pcur->old_rec = NULL;
 
 	pcur->btr_cur.rtr_info = NULL;
 }
 
-/** Free old_rec_buf.
-@param[in]	pcur	Persistent cursor holding old_rec to be freed. */
-UNIV_INLINE
-void
-btr_pcur_free(
-	btr_pcur_t*	pcur)
-{
-	ut_free(pcur->old_rec_buf);
-}
-
 /**************************************************************//**
 Initializes and opens a persistent cursor to an index tree. */
 inline
 dberr_t
-btr_pcur_open_low(
+btr_pcur_open(
-/*==============*/
-	dict_index_t*	index,	/*!< in: index */
-	ulint		level,	/*!< in: level in the btree */
 	const dtuple_t*	tuple,	/*!< in: tuple on which search done */
 	page_cur_mode_t	mode,	/*!< in: PAGE_CUR_L, ...;
 				NOTE that if the search is made using a unique
@@ -326,26 +311,24 @@ btr_pcur_open_low(
 				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_latch_mode	latch_mode,/*!< in: BTR_SEARCH_LEAF, ... */
 	btr_pcur_t*	cursor, /*!< in: memory buffer for persistent cursor */
 	ib_uint64_t	autoinc,/*!< in: PAGE_ROOT_AUTO_INC to be written
 				(0 if none) */
 	mtr_t*		mtr)	/*!< in: mtr */
 {
-  ut_ad(!index->is_spatial());
+  ut_ad(!cursor->index()->is_spatial());
-  btr_pcur_init(cursor);
   cursor->latch_mode= BTR_LATCH_MODE_WITHOUT_FLAGS(latch_mode);
   cursor->search_mode= mode;
   cursor->pos_state= BTR_PCUR_IS_POSITIONED;
   cursor->trx_if_known= nullptr;
-  return btr_cur_search_to_nth_level(index, level, tuple, mode, latch_mode,
+  return btr_cur_search_to_nth_level(0, tuple, mode, latch_mode,
                                      btr_pcur_get_btr_cur(cursor),
                                      mtr, autoinc);
 }
 
 /** Opens an persistent cursor to an index tree without initializing the
 cursor.
-@param index      index
 @param tuple      tuple on which search done
 @param mode       PAGE_CUR_L, ...; NOTE that if the search is made using a
                   unique prefix of a record, mode should be PAGE_CUR_LE, not
@@ -356,59 +339,21 @@ cursor.
 @param mtr        mini-transaction
 @return DB_SUCCESS on success or error code otherwise. */
 inline
-dberr_t btr_pcur_open_with_no_init(dict_index_t *index, const dtuple_t *tuple,
+dberr_t btr_pcur_open_with_no_init(const dtuple_t *tuple,
-                                   page_cur_mode_t mode, ulint latch_mode,
+                                   page_cur_mode_t mode,
+                                   btr_latch_mode latch_mode,
                                    btr_pcur_t *cursor, mtr_t *mtr)
 {
   cursor->latch_mode= BTR_LATCH_MODE_WITHOUT_INTENTION(latch_mode);
   cursor->search_mode= mode;
   cursor->pos_state= BTR_PCUR_IS_POSITIONED;
-  cursor->old_stored= false;
   cursor->trx_if_known= nullptr;
 
   /* Search with the tree cursor */
-  return btr_cur_search_to_nth_level(index, 0, tuple, mode, latch_mode,
+  return btr_cur_search_to_nth_level(0, tuple, mode, latch_mode,
                                      btr_pcur_get_btr_cur(cursor), mtr);
 }
 
-/*****************************************************************//**
-Opens a persistent cursor at either end of an index. */
-UNIV_INLINE
-dberr_t
-btr_pcur_open_at_index_side(
-/*========================*/
-	bool		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/out: cursor */
-	bool		init_pcur,	/*!< in: whether to initialize pcur */
-	ulint		level,		/*!< in: level to search for
-					(0=leaf) */
-	mtr_t*		mtr)		/*!< in/out: mini-transaction */
-{
-	dberr_t		err = DB_SUCCESS;
-
-	pcur->latch_mode = BTR_LATCH_MODE_WITHOUT_FLAGS(latch_mode);
-
-	pcur->search_mode = from_left ? PAGE_CUR_G : PAGE_CUR_L;
-
-	if (init_pcur) {
-		btr_pcur_init(pcur);
-	}
-
-	err = btr_cur_open_at_index_side(
-		from_left, index, latch_mode,
-		btr_pcur_get_btr_cur(pcur), level, mtr);
-	pcur->pos_state = BTR_PCUR_IS_POSITIONED;
-
-	pcur->old_stored = false;
-
-	pcur->trx_if_known = NULL;
-
-	return (err);
-}
-
 /**************************************************************//**
 Frees the possible memory heap of a persistent cursor and sets the latch
 mode of the persistent cursor to BTR_NO_LATCHES.
@@ -427,23 +372,19 @@ btr_pcur_close(
 {
   ut_free(cursor->old_rec_buf);
 
-	if (cursor->btr_cur.rtr_info) {
+  if (cursor->btr_cur.rtr_info)
     rtr_clean_rtr_info(cursor->btr_cur.rtr_info, true);
-		cursor->btr_cur.rtr_info = NULL;
-	}
 
-	cursor->old_rec = NULL;
+  cursor->btr_cur.rtr_info= nullptr;
-	cursor->old_rec_buf = NULL;
+  cursor->old_rec = nullptr;
-	cursor->btr_cur.page_cur.rec = NULL;
+  cursor->old_rec_buf = nullptr;
-	cursor->btr_cur.page_cur.block = NULL;
+  cursor->btr_cur.page_cur.rec = nullptr;
-
+  cursor->btr_cur.page_cur.block = nullptr;
-	cursor->old_rec = NULL;
-	cursor->old_stored = false;
 
   cursor->latch_mode = BTR_NO_LATCHES;
   cursor->pos_state = BTR_PCUR_NOT_POSITIONED;
 
-	cursor->trx_if_known = NULL;
+  cursor->trx_if_known = nullptr;
 }
 
 /*********************************************************//**
@@ -459,5 +400,5 @@ btr_pcur_move_before_first_on_page(
 	page_cur_set_before_first(btr_pcur_get_block(cursor),
 		btr_pcur_get_page_cur(cursor));
 
-	cursor->old_stored = false;
+	cursor->old_rec = nullptr;
 }

storage/innobase/include/btr0types.h

@@ -1,7 +1,7 @@
 /*****************************************************************************
 
 Copyright (c) 1996, 2015, Oracle and/or its affiliates. All Rights Reserved.
-Copyright (c) 2018, 2019, MariaDB Corporation.
+Copyright (c) 2018, 2022, MariaDB Corporation.
 
 This program is free software; you can redistribute it and/or modify it under
 the terms of the GNU General Public License as published by the Free Software
@@ -24,8 +24,7 @@ The index tree general types
 Created 2/17/1996 Heikki Tuuri
 *************************************************************************/
 
-#ifndef btr0types_h
+#pragma once
-#define btr0types_h
 
 #include "page0types.h"
 #include "rem0types.h"
@@ -56,4 +55,93 @@ in the index record. */
 #define BTR_EXTERN_LOCAL_STORED_MAX_SIZE	\
 	(BTR_EXTERN_FIELD_REF_SIZE * 2)
 
-#endif
+/** Latching modes for btr_cur_search_to_nth_level(). */
+enum btr_latch_mode {
+	/** Search a record on a leaf page and S-latch it. */
+	BTR_SEARCH_LEAF = RW_S_LATCH,
+	/** (Prepare to) modify a record on a leaf page and X-latch it. */
+	BTR_MODIFY_LEAF	= RW_X_LATCH,
+	/** Obtain no latches. */
+	BTR_NO_LATCHES = RW_NO_LATCH,
+	/** Search the previous record. */
+	BTR_SEARCH_PREV = 4 | BTR_SEARCH_LEAF,
+	/** Modify the previous record. */
+	BTR_MODIFY_PREV = 4 | BTR_MODIFY_LEAF,
+	/** Start searching the entire B-tree. */
+	BTR_SEARCH_TREE = 8 | BTR_SEARCH_LEAF,
+	/** Start modifying1 the entire B-tree. */
+	BTR_MODIFY_TREE = 8 | BTR_MODIFY_LEAF,
+	/** Continue searching the entire B-tree. */
+	BTR_CONT_SEARCH_TREE = 4 | BTR_SEARCH_TREE,
+	/** Continue modifying the entire B-tree. */
+	BTR_CONT_MODIFY_TREE = 4 | BTR_MODIFY_TREE,
+
+	/* BTR_INSERT, BTR_DELETE and BTR_DELETE_MARK are mutually
+	exclusive. */
+	/** The search tuple will be inserted to the secondary index
+	at the searched position.  When the leaf page is not in the
+	buffer pool, try to use the change buffer. */
+	BTR_INSERT = 64,
+
+	/** Try to delete mark a secondary index leaf page record at
+	the searched position using the change buffer when the page is
+	not in the buffer pool. */
+	BTR_DELETE_MARK	= 128,
+
+	/** Try to purge the record using the change buffer when the
+	secondary index leaf page is not in the buffer pool. */
+	BTR_DELETE = BTR_INSERT | BTR_DELETE_MARK,
+
+	/** The caller is already holding dict_index_t::lock S-latch. */
+	BTR_ALREADY_S_LATCHED = 256,
+	/** Search and S-latch a leaf page, assuming that the
+	dict_index_t::lock S-latch is being held. */
+	BTR_SEARCH_LEAF_ALREADY_S_LATCHED = BTR_SEARCH_LEAF
+	| BTR_ALREADY_S_LATCHED,
+	/** Search the entire index tree, assuming that the
+	dict_index_t::lock S-latch is being held. */
+	BTR_SEARCH_TREE_ALREADY_S_LATCHED = BTR_SEARCH_TREE
+	| BTR_ALREADY_S_LATCHED,
+	/** Search and X-latch a leaf page, assuming that the
+	dict_index_t::lock is being held in non-exclusive mode. */
+	BTR_MODIFY_LEAF_ALREADY_LATCHED = BTR_MODIFY_LEAF
+	| BTR_ALREADY_S_LATCHED,
+
+	/** Attempt to delete-mark a secondary index record. */
+	BTR_DELETE_MARK_LEAF = BTR_MODIFY_LEAF | BTR_DELETE_MARK,
+	/** Attempt to delete-mark a secondary index record
+	while holding the dict_index_t::lock S-latch. */
+	BTR_DELETE_MARK_LEAF_ALREADY_S_LATCHED = BTR_DELETE_MARK_LEAF
+	| BTR_ALREADY_S_LATCHED,
+	/** Attempt to purge a secondary index record. */
+	BTR_PURGE_LEAF = BTR_MODIFY_LEAF | BTR_DELETE,
+	/** Attempt to purge a secondary index record
+	while holding the dict_index_t::lock S-latch. */
+	BTR_PURGE_LEAF_ALREADY_S_LATCHED = BTR_PURGE_LEAF
+	| BTR_ALREADY_S_LATCHED,
+
+	/** In the case of BTR_MODIFY_TREE, the caller specifies
+	the intention to delete record only. It is used to optimize
+	block->lock range.*/
+	BTR_LATCH_FOR_DELETE = 512,
+
+	/** In the case of BTR_MODIFY_TREE, the caller specifies
+	the intention to delete record only. It is used to optimize
+	block->lock range.*/
+	BTR_LATCH_FOR_INSERT = 1024,
+
+	/** Attempt to delete a record in the tree. */
+	BTR_PURGE_TREE = BTR_MODIFY_TREE | BTR_LATCH_FOR_DELETE,
+
+	/** Attempt to insert a record into the tree. */
+	BTR_INSERT_TREE = BTR_MODIFY_TREE | BTR_LATCH_FOR_INSERT,
+
+	/** This flag ORed to BTR_INSERT 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 */
+	BTR_IGNORE_SEC_UNIQUE = 2048,
+	/** Rollback in spatial index */
+	BTR_RTREE_UNDO_INS = 4096,
+	/** Try to delete mark a spatial index record */
+	BTR_RTREE_DELETE_MARK = 8192
+};

storage/innobase/include/dyn0buf.h

@@ -1,7 +1,7 @@
 /*****************************************************************************
 
 Copyright (c) 2013, 2016, Oracle and/or its affiliates. All Rights Reserved.
-Copyright (c) 2018, 2020, MariaDB Corporation.
+Copyright (c) 2018, 2022, MariaDB Corporation.
 
 This program is free software; you can redistribute it and/or modify it under
 the terms of the GNU General Public License as published by the Free Software
@@ -313,67 +313,13 @@ public:
 		return(m_size);
 	}
 
-	/**
-	Iterate over each block and call the functor.
-	@return	false if iteration was terminated. */
-	template <typename Functor>
-	bool for_each_block(Functor& functor) const
-	{
-		for (list_t::iterator it = m_list.begin(), end = m_list.end();
-		     it != end; ++it) {
-
-			if (!functor(&*it)) {
-				return false;
-			}
-		}
-
-		return(true);
-	}
-
 	/**
 	Iterate over each block and call the functor.
 	@return	false if iteration was terminated. */
 	template <typename Functor>
 	bool for_each_block(const Functor& functor) const
 	{
-		for (typename list_t::iterator it = m_list.begin(),
+		for (list_t::iterator it = m_list.begin(), end = m_list.end();
-					       end = m_list.end();
-		     it != end; ++it) {
-
-			if (!functor(&*it)) {
-				return false;
-			}
-		}
-
-		return(true);
-	}
-
-	/**
-	Iterate over all the blocks in reverse and call the iterator
-	@return	false if iteration was terminated. */
-	template <typename Functor>
-	bool for_each_block_in_reverse(Functor& functor) const
-	{
-		for (list_t::reverse_iterator it = m_list.rbegin(),
-					      end = m_list.rend();
-		     it != end; ++it) {
-
-			if (!functor(&*it)) {
-				return false;
-			}
-		}
-
-		return(true);
-	}
-
-	/**
-	Iterate over all the blocks in reverse and call the iterator
-	@return	false if iteration was terminated. */
-	template <typename Functor>
-	bool for_each_block_in_reverse(const Functor& functor) const
-	{
-		for (list_t::reverse_iterator it = m_list.rbegin(),
-					      end = m_list.rend();
 		     it != end; ++it) {
 
 			if (!functor(&*it)) {

storage/innobase/include/gis0rtree.h

@@ -256,23 +256,14 @@ rtr_get_mbr_from_tuple(
 	rtr_mbr*	mbr);	/*!< out: mbr to fill */
 
 /* Get the rtree page father.
-@param[in]	offsets		work area for the return value
-@param[in]	index		rtree index
-@param[in]	block		child page in the index
 @param[in,out]	mtr		mtr
 @param[in]	sea_cur		search cursor, contains information
 				about parent nodes in search
-@param[out]	cursor		cursor on node pointer record,
+@param[in,out]	cursor		cursor on node pointer record,
 				its page x-latched
 @return whether the cursor was successfully positioned */
-bool
+bool rtr_page_get_father(mtr_t *mtr, btr_cur_t *sea_cur, btr_cur_t *cursor)
-rtr_page_get_father(
+  MY_ATTRIBUTE((nonnull(1,3), warn_unused_result));
-	dict_index_t*	index,
-	buf_block_t*	block,
-	mtr_t*		mtr,
-	btr_cur_t*	sea_cur,
-	btr_cur_t*	cursor)
-	MY_ATTRIBUTE((nonnull(1,2,3,5), warn_unused_result));
 
 /************************************************************//**
 Returns the father block to a page. It is assumed that mtr holds
@@ -283,8 +274,6 @@ rtr_page_get_father_block(
 /*======================*/
 	rec_offs*	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*	sea_cur,/*!< in: search cursor, contains information
 				about parent nodes in search */
@@ -298,7 +287,7 @@ rtr_store_parent_path(
 /*==================*/
 	const buf_block_t*	block,	/*!< in: block of the page */
 	btr_cur_t*		btr_cur,/*!< in/out: persistent cursor */
-	ulint			latch_mode,
+	btr_latch_mode		latch_mode,
 					/*!< in: latch_mode */
 	ulint			level,	/*!< in: index level */
 	mtr_t*			mtr);	/*!< in: mtr */
@@ -310,7 +299,7 @@ bool
 rtr_pcur_open(
 	dict_index_t*	index,	/*!< in: index */
 	const dtuple_t*	tuple,	/*!< in: tuple on which search done */
-	ulint		latch_mode,/*!< in: BTR_SEARCH_LEAF, ... */
+	btr_latch_mode	latch_mode,/*!< in: BTR_SEARCH_LEAF, ... */
 	btr_pcur_t*	cursor,	/*!< in: memory buffer for persistent cursor */
 	mtr_t*		mtr)	/*!< in: mtr */
 	MY_ATTRIBUTE((warn_unused_result));
@@ -432,7 +421,6 @@ rtr_check_same_block(
 	btr_cur_t*	cur,	/*!< in/out: position at the parent entry
 				pointing to the child if successful */
 	buf_block_t*	parentb,/*!< in: parent page to check */
-	buf_block_t*	childb, /*!< in: child Page */
 	mem_heap_t*	heap);	/*!< in: memory heap */
 
 /*********************************************************************//**

storage/innobase/include/ibuf0ibuf.h

@@ -30,7 +30,6 @@ Created 7/19/1997 Heikki Tuuri
 #include "mtr0mtr.h"
 #include "dict0mem.h"
 #include "fsp0fsp.h"
-#include "ibuf0types.h"
 
 /** Default value for maximum on-disk size of change buffer in terms
 of percentage of the buffer pool. */
@@ -61,6 +60,37 @@ enum ibuf_use_t {
 /** Operations that can currently be buffered. */
 extern ulong		innodb_change_buffering;
 
+/** Insert buffer struct */
+struct ibuf_t{
+	Atomic_relaxed<ulint> size;	/*!< current size of the ibuf index
+					tree, in pages */
+	Atomic_relaxed<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 */
+	bool		empty;		/*!< Protected by the page
+					latch of the root page of the
+					insert buffer tree
+					(FSP_IBUF_TREE_ROOT_PAGE_NO). true
+					if and only if the insert
+					buffer tree is empty. */
+	ulint		free_list_len;	/*!< length of the free list */
+	ulint		height;		/*!< tree height */
+	dict_index_t*	index;		/*!< insert buffer index */
+
+	/** number of pages merged */
+	Atomic_counter<ulint> n_merges;
+	Atomic_counter<ulint> n_merged_ops[IBUF_OP_COUNT];
+					/*!< number of operations of each type
+					merged to index pages */
+	Atomic_counter<ulint> n_discarded_ops[IBUF_OP_COUNT];
+					/*!< number of operations of each type
+					discarded without merging due to the
+					tablespace being deleted or the
+					index being dropped */
+};
+
 /** The insert buffer control structure */
 extern ibuf_t		ibuf;
 
@@ -339,9 +369,9 @@ void ibuf_delete_for_discarded_space(uint32_t space);
 
 /** Contract the change buffer by reading pages to the buffer pool.
 @return 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
+will be merged from ibuf trees to the pages read
-empty */
+@retval 0 if ibuf.empty */
-ulint ibuf_merge_all();
+ulint ibuf_contract();
 
 /** Contracts insert buffer trees by reading pages referring to space_id
 to the buffer pool.

storage/innobase/include/ibuf0ibuf.inl

@@ -65,37 +65,6 @@ ibuf_mtr_commit(
 	mtr_commit(mtr);
 }
 
-/** Insert buffer struct */
-struct ibuf_t{
-	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 */
-	bool		empty;		/*!< Protected by the page
-					latch of the root page of the
-					insert buffer tree
-					(FSP_IBUF_TREE_ROOT_PAGE_NO). true
-					if and only if the insert
-					buffer tree is empty. */
-	ulint		free_list_len;	/*!< length of the free list */
-	ulint		height;		/*!< tree height */
-	dict_index_t*	index;		/*!< insert buffer index */
-
-	/** number of pages merged */
-	Atomic_counter<ulint> n_merges;
-	Atomic_counter<ulint> n_merged_ops[IBUF_OP_COUNT];
-					/*!< number of operations of each type
-					merged to index pages */
-	Atomic_counter<ulint> n_discarded_ops[IBUF_OP_COUNT];
-					/*!< number of operations of each type
-					discarded without merging due to the
-					tablespace being deleted or the
-					index being dropped */
-};
-
 /************************************************************************//**
 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

storage/innobase/include/ibuf0types.h

@@ -1,31 +0,0 @@
-/*****************************************************************************
-
-Copyright (c) 1997, 2009, Oracle and/or its affiliates. All Rights Reserved.
-
-This program is free software; you can redistribute it and/or modify it under
-the terms of the GNU General Public License as published by the Free Software
-Foundation; version 2 of the License.
-
-This program is distributed in the hope that it will be useful, but WITHOUT
-ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
-FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
-
-You 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 Street, Fifth Floor, Boston, MA 02110-1335 USA
-
-*****************************************************************************/
-
-/**************************************************//**
-@file include/ibuf0types.h
-Insert buffer global types
-
-Created 7/29/1997 Heikki Tuuri
-*******************************************************/
-
-#ifndef ibuf0types_h
-#define ibuf0types_h
-
-struct ibuf_t;
-
-#endif

storage/innobase/include/log0log.h

@@ -214,6 +214,11 @@ public:
   /** number of std::swap(buf, flush_buf) and writes from buf to log;
   protected by latch.wr_lock() */
   ulint write_to_log;
+
+  /** Log sequence number when a log file overwrite (broken crash recovery)
+  was noticed. Protected by latch.wr_lock(). */
+  lsn_t overwrite_warned;
+
   /** innodb_log_buffer_size (size of buf and flush_buf, in bytes) */
   size_t buf_size;
 

storage/innobase/include/mtr0mtr.h

@@ -28,6 +28,8 @@ Created 11/26/1995 Heikki Tuuri
 
 #include "fil0fil.h"
 #include "dyn0buf.h"
+#include "buf0buf.h"
+#include <vector>
 
 /** Start a mini-transaction. */
 #define mtr_start(m)		(m)->start()
@@ -48,11 +50,6 @@ savepoint. */
 @return	old mode */
 #define mtr_set_log_mode(m, d)	(m)->set_log_mode((d))
 
-/** Release an object in the memo stack.
-@return true if released */
-#define mtr_memo_release(m, o, t)					\
-				(m)->memo_release((o), (t))
-
 #ifdef UNIV_PFS_RWLOCK
 # define mtr_s_lock_index(i,m)	(m)->s_lock(__FILE__, __LINE__, &(i)->lock)
 # define mtr_x_lock_index(i,m)	(m)->x_lock(__FILE__, __LINE__, &(i)->lock)
@@ -66,19 +63,16 @@ savepoint. */
 #define mtr_release_block_at_savepoint(m, s, b)				\
 				(m)->release_block_at_savepoint((s), (b))
 
-#define mtr_block_sx_latch_at_savepoint(m, s, b)			\
-				(m)->sx_latch_at_savepoint((s), (b))
-
-#define mtr_block_x_latch_at_savepoint(m, s, b)				\
-				(m)->x_latch_at_savepoint((s), (b))
-
 /** Mini-transaction memo stack slot. */
-struct mtr_memo_slot_t {
+struct mtr_memo_slot_t
-	/** pointer to the object */
+{
-	void*		object;
+  /** pointer to the object, or nullptr if released */
-
+  void *object;
   /** type of the stored object */
   mtr_memo_type_t type;
+
+  /** Release the object */
+  void release() const;
 };
 
 /** Mini-transaction handle and buffer */
@@ -89,14 +83,19 @@ struct mtr_t {
   /** Commit the mini-transaction. */
   void commit();
 
-  /** Release latches till savepoint. To simplify the code only
+  /** Release latches of unmodified buffer pages.
-  MTR_MEMO_S_LOCK and MTR_MEMO_PAGE_S_FIX slot types are allowed to be
+  @param begin   first slot to release
-  released, otherwise it would be neccesary to add one more argument in the
+  @param end     last slot to release, or get_savepoint() */
-  function to point out what slot types are allowed for rollback, and this
+  void rollback_to_savepoint(ulint begin, ulint end);
-  would be overengineering as currently the function is used only in one place
+
-  in the code.
+  /** Release latches of unmodified buffer pages.
-  @param savepoint   savepoint, can be obtained with get_savepoint */
+  @param begin   first slot to release */
-  void rollback_to_savepoint(ulint savepoint);
+  void rollback_to_savepoint(ulint begin)
+  { rollback_to_savepoint(begin, m_memo->size()); }
+
+  /** Release the last acquired buffer page latch. */
+  void release_last_page()
+  { auto s= m_memo->size(); rollback_to_savepoint(s - 1, s); }
 
   /** Commit a mini-transaction that is shrinking a tablespace.
   @param space   tablespace that is being shrunk */
@@ -118,26 +117,89 @@ struct mtr_t {
   lsn_t commit_files(lsn_t checkpoint_lsn= 0);
 
   /** @return mini-transaction savepoint (current size of m_memo) */
-  ulint get_savepoint() const { ut_ad(is_active()); return m_memo.size(); }
+  ulint get_savepoint() const
-
+  {
-	/** Release the (index tree) s-latch stored in an mtr memo after a
+    ut_ad(is_active());
-	savepoint.
+    return m_memo ? m_memo->size() : 0;
-	@param savepoint	value returned by @see set_savepoint.
+  }
-	@param lock		latch to release */
-	inline void release_s_latch_at_savepoint(
-		ulint		savepoint,
-		index_lock*	lock);
 
+  /** Release the (index tree) s-latch stored in an mtr memo after a savepoint.
+  @param savepoint value returned by get_savepoint()
+  @param lock   index latch to release */
+  void release_s_latch_at_savepoint(ulint savepoint, index_lock *lock)
+  {
+    ut_ad(is_active());
+    mtr_memo_slot_t &slot= m_memo->at(savepoint);
+    ut_ad(slot.object == lock);
+    ut_ad(slot.type == MTR_MEMO_S_LOCK);
+    slot.object= nullptr;
+    lock->s_unlock();
+  }
   /** Release the block in an mtr memo after a savepoint. */
-	inline void release_block_at_savepoint(
+  void release_block_at_savepoint(ulint savepoint, buf_block_t *block)
-		ulint		savepoint,
+  {
-		buf_block_t*	block);
+    ut_ad(is_active());
+    mtr_memo_slot_t &slot= m_memo->at(savepoint);
+    ut_ad(slot.object == block);
+    ut_ad(!(slot.type & MTR_MEMO_MODIFY));
+    slot.object= nullptr;
+    block->page.unfix();
 
-	/** SX-latch a not yet latched block after a savepoint. */
+    switch (slot.type) {
-	inline void sx_latch_at_savepoint(ulint savepoint, buf_block_t* block);
+    case MTR_MEMO_PAGE_S_FIX:
+      block->page.lock.s_unlock();
+      break;
+    case MTR_MEMO_PAGE_SX_FIX:
+    case MTR_MEMO_PAGE_X_FIX:
+      block->page.lock.u_or_x_unlock(slot.type == MTR_MEMO_PAGE_SX_FIX);
+      break;
+    default:
+      break;
+    }
+  }
+
+  /** @return if we are about to make a clean buffer block dirty */
+  static bool is_block_dirtied(const buf_page_t &b)
+  {
+    ut_ad(b.in_file());
+    ut_ad(b.frame);
+    ut_ad(b.buf_fix_count());
+    return b.oldest_modification() <= 1 && b.id().space() < SRV_TMP_SPACE_ID;
+  }
 
   /** X-latch a not yet latched block after a savepoint. */
-	inline void x_latch_at_savepoint(ulint savepoint, buf_block_t*	block);
+  void x_latch_at_savepoint(ulint savepoint, buf_block_t *block)
+  {
+    ut_ad(is_active());
+    ut_ad(!memo_contains_flagged(block, MTR_MEMO_PAGE_S_FIX |
+                                 MTR_MEMO_PAGE_X_FIX | MTR_MEMO_PAGE_SX_FIX));
+    mtr_memo_slot_t &slot= m_memo->at(savepoint);
+    ut_ad(slot.object == block);
+    ut_ad(slot.type == MTR_MEMO_BUF_FIX);
+    slot.type= MTR_MEMO_PAGE_X_FIX;
+    block->page.lock.x_lock();
+    ut_ad(!block->page.is_io_fixed());
+
+    if (!m_made_dirty)
+      m_made_dirty= is_block_dirtied(block->page);
+  }
+
+  /** U-latch a not yet latched block after a savepoint. */
+  void sx_latch_at_savepoint(ulint savepoint, buf_block_t *block)
+  {
+    ut_ad(is_active());
+    ut_ad(!memo_contains_flagged(block, MTR_MEMO_PAGE_S_FIX |
+                                 MTR_MEMO_PAGE_X_FIX | MTR_MEMO_PAGE_SX_FIX));
+    mtr_memo_slot_t &slot= m_memo->at(savepoint);
+    ut_ad(slot.object == block);
+    ut_ad(slot.type == MTR_MEMO_BUF_FIX);
+    slot.type= MTR_MEMO_PAGE_SX_FIX;
+    block->page.lock.u_lock();
+    ut_ad(!block->page.is_io_fixed());
+
+    if (!m_made_dirty)
+      m_made_dirty= is_block_dirtied(block->page);
+  }
 
   /** @return the logging mode */
   mtr_log_t get_log_mode() const
@@ -291,19 +353,17 @@ struct mtr_t {
   /** Acquire an exclusive tablespace latch.
   @param space  tablespace */
   void x_lock_space(fil_space_t *space);
-  /** Release an object in the memo stack.
+
-  @param object	object
+  /** Release an index latch. */
-  @param type	object type
+  void release(const index_lock &lock) { release(&lock); }
-  @return bool if lock released */
+  /** Release a latch to an unmodified page. */
-  bool memo_release(const void *object, ulint type);
+  void release(const buf_block_t &block) { release(&block); }
-  /** Release a page latch.
-  @param[in]	ptr	pointer to within a page frame
-  @param[in]	type	object type: MTR_MEMO_PAGE_X_FIX, ... */
-  void release_page(const void *ptr, mtr_memo_type_t type);
 
   /** Note that the mini-transaction will modify data. */
   void flag_modified() { m_modifications = true; }
 private:
+  /** Release an unmodified object. */
+  void release(const void *object);
   /** Mark the given latched page as modified.
   @param block   page that will be modified */
   void modify(const buf_block_t& block);
@@ -338,73 +398,93 @@ public:
   @param rw_latch RW_S_LATCH, RW_SX_LATCH, RW_X_LATCH, RW_NO_LATCH */
   void page_lock(buf_block_t *block, ulint rw_latch);
 
+  /** Acquire a latch on a buffer-fixed buffer pool block.
+  @param savepoint   savepoint location of the buffer-fixed block
+  @param rw_latch    latch to acquire */
+  void upgrade_buffer_fix(ulint savepoint, rw_lock_type_t rw_latch);
+
   /** Register a page latch on a buffer-fixed block was buffer-fixed.
   @param latch   latch type */
   void u_lock_register(ulint savepoint)
   {
-    mtr_memo_slot_t *slot= m_memo.at<mtr_memo_slot_t*>(savepoint);
+    mtr_memo_slot_t &slot= m_memo->at(savepoint);
-    ut_ad(slot->type == MTR_MEMO_BUF_FIX);
+    ut_ad(slot.type == MTR_MEMO_BUF_FIX);
-    slot->type= MTR_MEMO_PAGE_SX_FIX;
+    slot.type= MTR_MEMO_PAGE_SX_FIX;
+  }
+
+  /** Register a page latch on a buffer-fixed block was buffer-fixed.
+  @param latch   latch type */
+  void s_lock_register(ulint savepoint)
+  {
+    mtr_memo_slot_t &slot= m_memo->at(savepoint);
+    ut_ad(slot.type == MTR_MEMO_BUF_FIX);
+    slot.type= MTR_MEMO_PAGE_S_FIX;
   }
 
   /** Upgrade U locks on a block to X */
   void page_lock_upgrade(const buf_block_t &block);
-  /** Upgrade X lock to X */
+  /** Upgrade U lock to X */
   void lock_upgrade(const index_lock &lock);
 
   /** Check if we are holding tablespace latch
   @param space  tablespace to search for
   @param shared whether to look for shared latch, instead of exclusive
   @return whether space.latch is being held */
-  bool memo_contains(const fil_space_t& space, bool shared= false)
+  bool memo_contains(const fil_space_t& space, bool shared= false) const
     MY_ATTRIBUTE((warn_unused_result));
 #ifdef UNIV_DEBUG
   /** Check if we are holding an rw-latch in this mini-transaction
   @param lock   latch to search for
   @param type   held latch type
   @return whether (lock,type) is contained */
-  bool memo_contains(const index_lock &lock, mtr_memo_type_t type)
+  bool memo_contains(const index_lock &lock, mtr_memo_type_t type) const
     MY_ATTRIBUTE((warn_unused_result));
 
-	/** Check if memo contains the given item.
+  /** Check if memo contains an index or buffer block latch.
   @param object    object to search
-	@param flags		specify types of object (can be ORred) of
+  @param flags     specify types of object latches
-				MTR_MEMO_PAGE_S_FIX ... values
   @return true if contains */
-	bool memo_contains_flagged(const void* ptr, ulint flags) const;
+  bool memo_contains_flagged(const void *object, ulint flags) const
+    MY_ATTRIBUTE((warn_unused_result, nonnull));
 
   /** Check if memo contains the given page.
-	@param[in]	ptr	pointer to within buffer frame
+  @param ptr   pointer to within page frame
-	@param[in]	flags	specify types of object with OR of
+  @param flags types latch to look for
-				MTR_MEMO_PAGE_S_FIX... values
   @return the block
-	@retval	NULL	if not found */
+  @retval nullptr    if not found */
-	buf_block_t* memo_contains_page_flagged(
+  buf_block_t *memo_contains_page_flagged(const byte *ptr, ulint flags) const;
-		const byte*	ptr,
-		ulint		flags) const;
 
   /** @return true if mini-transaction contains modifications. */
   bool has_modifications() const { return m_modifications; }
 #endif /* UNIV_DEBUG */
 
-	/** @return true if a record was added to the mini-transaction */
-	bool is_dirty() const { return m_made_dirty; }
-
   /** Push an object to an mtr memo stack.
   @param object	object
   @param type	object type: MTR_MEMO_S_LOCK, ... */
-	inline void memo_push(void* object, mtr_memo_type_t type);
+  void memo_push(void *object, mtr_memo_type_t type) __attribute__((nonnull))
+  {
+    ut_ad(is_active());
+    /* If this mtr has U or X latched a clean page then we set
+    the m_made_dirty flag. This tells us if we need to
+    grab log_sys.flush_order_mutex at mtr_t::commit() so that we
+    can insert the dirtied page into the buf_pool.flush_list.
 
-	/** Check if this mini-transaction is dirtying a clean page.
+    FIXME: Do this only when the MTR_MEMO_MODIFY flag is set! */
-	@param block	block being x-fixed
+    if (!m_made_dirty &&
-	@return true if the mtr is dirtying a clean page. */
+        (type & (MTR_MEMO_PAGE_X_FIX | MTR_MEMO_PAGE_SX_FIX)))
-	static inline bool is_block_dirtied(const buf_block_t* block)
+      m_made_dirty=
-		MY_ATTRIBUTE((warn_unused_result));
+        is_block_dirtied(*static_cast<const buf_page_t*>(object));
+
+    if (!m_memo)
+      m_memo= new std::vector<mtr_memo_slot_t>(1, {object, type});
+    else
+      m_memo->emplace_back(mtr_memo_slot_t{object, type});
+  }
 
   /** @return the size of the log is empty */
   size_t get_log_size() const { return m_log.size(); }
   /** @return whether the log and memo are empty */
-  bool is_empty() const { return m_memo.size() == 0 && m_log.size() == 0; }
+  bool is_empty() const { return !get_savepoint() && !get_log_size(); }
 
   /** Write an OPT_PAGE_CHECKSUM record. */
   inline void page_checksum(const buf_page_t &bpage);
@@ -670,6 +750,8 @@ private:
   @return {start_lsn,flush_ahead} */
   std::pair<lsn_t,page_flush_ahead> finish_write(size_t len);
 
+  /** Release all latches. */
+  void release();
   /** Release the resources */
   inline void release_resources();
 
@@ -727,7 +809,7 @@ private:
 #endif /* UNIV_DEBUG */
 
   /** acquired dict_index_t::lock, fil_space_t::latch, buf_block_t */
-  mtr_buf_t m_memo;
+  std::vector<mtr_memo_slot_t> *m_memo= nullptr;
 
   /** mini-transaction log */
   mtr_buf_t m_log;
@@ -743,5 +825,3 @@ private:
   /** set of freed page ids */
   range_set *m_freed_pages= nullptr;
 };
-
-#include "mtr0mtr.inl"

storage/innobase/include/mtr0mtr.inl

@@ -1,179 +0,0 @@
-/*****************************************************************************
-
-Copyright (c) 1995, 2014, Oracle and/or its affiliates. All Rights Reserved.
-Copyright (c) 2017, 2022, MariaDB Corporation.
-
-This program is free software; you can redistribute it and/or modify it under
-the terms of the GNU General Public License as published by the Free Software
-Foundation; version 2 of the License.
-
-This program is distributed in the hope that it will be useful, but WITHOUT
-ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
-FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
-
-You 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 Street, Fifth Floor, Boston, MA 02110-1335 USA
-
-*****************************************************************************/
-
-/**************************************************//**
-@file include/mtr0mtr.ic
-Mini-transaction buffer
-
-Created 11/26/1995 Heikki Tuuri
-*******************************************************/
-
-#include "buf0buf.h"
-
-/** Check if a mini-transaction is dirtying a clean page.
-@return true if the mtr is dirtying a clean page. */
-inline bool mtr_t::is_block_dirtied(const buf_block_t *block)
-{
-  ut_ad(block->page.in_file());
-  ut_ad(block->page.frame);
-  ut_ad(block->page.buf_fix_count());
-  return block->page.oldest_modification() <= 1 &&
-    block->page.id().space() < SRV_TMP_SPACE_ID;
-}
-
-/**
-Pushes an object to an mtr memo stack. */
-void
-mtr_t::memo_push(void* object, mtr_memo_type_t type)
-{
-	ut_ad(is_active());
-	ut_ad(object != NULL);
-	ut_ad(type >= MTR_MEMO_PAGE_S_FIX);
-	ut_ad(type <= MTR_MEMO_SPACE_S_LOCK);
-	ut_ad(type == MTR_MEMO_PAGE_X_MODIFY || ut_is_2pow(type));
-
-	/* If this mtr has U or X latched a clean page then we set
-	the m_made_dirty flag. This tells us if we need to
-	grab log_sys.flush_order_mutex at mtr_t::commit() so that we
-	can insert the dirtied page into the buf_pool.flush_list. */
-
-	if (!m_made_dirty
-	    && (type & (MTR_MEMO_PAGE_X_FIX | MTR_MEMO_PAGE_SX_FIX))) {
-
-		m_made_dirty = is_block_dirtied(
-			reinterpret_cast<const buf_block_t*>(object));
-	}
-
-	mtr_memo_slot_t* slot = m_memo.push<mtr_memo_slot_t*>(sizeof(*slot));
-
-	slot->type = type;
-	slot->object = object;
-}
-
-/**
-Releases the (index tree) s-latch stored in an mtr memo after a
-savepoint. */
-void
-mtr_t::release_s_latch_at_savepoint(
-	ulint		savepoint,
-	index_lock*	lock)
-{
-	ut_ad(is_active());
-	ut_ad(m_memo.size() > savepoint);
-
-	mtr_memo_slot_t* slot = m_memo.at<mtr_memo_slot_t*>(savepoint);
-
-	ut_ad(slot->object == lock);
-	ut_ad(slot->type == MTR_MEMO_S_LOCK);
-
-	lock->s_unlock();
-
-	slot->object = NULL;
-}
-
-/**
-SX-latches the not yet latched block after a savepoint. */
-
-void
-mtr_t::sx_latch_at_savepoint(
-	ulint		savepoint,
-	buf_block_t*	block)
-{
-	ut_ad(is_active());
-	ut_ad(m_memo.size() > savepoint);
-
-	ut_ad(!memo_contains_flagged(
-			block,
-			MTR_MEMO_PAGE_S_FIX
-			| MTR_MEMO_PAGE_X_FIX
-			| MTR_MEMO_PAGE_SX_FIX));
-
-	mtr_memo_slot_t* slot = m_memo.at<mtr_memo_slot_t*>(savepoint);
-
-	ut_ad(slot->object == block);
-	ut_ad(slot->type == MTR_MEMO_BUF_FIX); /* == RW_NO_LATCH */
-	slot->type = MTR_MEMO_PAGE_SX_FIX;
-
-	block->page.lock.u_lock();
-	ut_ad(!block->page.is_io_fixed());
-
-	if (!m_made_dirty) {
-		m_made_dirty = is_block_dirtied(block);
-	}
-}
-
-/**
-X-latches the not yet latched block after a savepoint. */
-
-void
-mtr_t::x_latch_at_savepoint(
-	ulint		savepoint,
-	buf_block_t*	block)
-{
-	ut_ad(is_active());
-	ut_ad(m_memo.size() > savepoint);
-
-	ut_ad(!memo_contains_flagged(
-			block,
-			MTR_MEMO_PAGE_S_FIX
-			| MTR_MEMO_PAGE_X_FIX
-			| MTR_MEMO_PAGE_SX_FIX));
-
-	mtr_memo_slot_t* slot = m_memo.at<mtr_memo_slot_t*>(savepoint);
-
-	ut_ad(slot->object == block);
-	ut_ad(slot->type == MTR_MEMO_BUF_FIX); /* == RW_NO_LATCH */
-	slot->type = MTR_MEMO_PAGE_X_FIX;
-
-	block->page.lock.x_lock();
-	ut_ad(!block->page.is_io_fixed());
-
-	if (!m_made_dirty) {
-		m_made_dirty = is_block_dirtied(block);
-	}
-}
-
-/**
-Releases the block in an mtr memo after a savepoint. */
-
-void
-mtr_t::release_block_at_savepoint(
-	ulint		savepoint,
-	buf_block_t*	block)
-{
-  ut_ad(is_active());
-
-  mtr_memo_slot_t *slot = m_memo.at<mtr_memo_slot_t*>(savepoint);
-
-  ut_a(slot->object == block);
-  slot->object= nullptr;
-  block->page.unfix();
-
-  switch (slot->type) {
-  case MTR_MEMO_PAGE_S_FIX:
-    block->page.lock.s_unlock();
-    break;
-  case MTR_MEMO_PAGE_SX_FIX:
-  case MTR_MEMO_PAGE_X_FIX:
-    block->page.lock.u_or_x_unlock(slot->type == MTR_MEMO_PAGE_SX_FIX);
-    break;
-  default:
-    break;
-  }
-}

storage/innobase/include/page0cur.h

@@ -129,7 +129,6 @@ page_cur_tuple_insert(
 /*==================*/
 	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 */
 	rec_offs**	offsets,/*!< out: offsets on *rec */
 	mem_heap_t**	heap,	/*!< in/out: pointer to memory heap, or NULL */
 	ulint		n_ext,	/*!< in: number of externally stored columns */
@@ -143,7 +142,6 @@ rec_t*
 page_cur_insert_rec_low(
 /*====================*/
 	const page_cur_t*cur,	/*!< in: page cursor */
-	dict_index_t*	index,	/*!< in: record descriptor */
 	const rec_t*	rec,	/*!< in: record to insert after cur */
 	rec_offs*	offsets,/*!< in/out: rec_get_offsets(rec, index) */
 	mtr_t*		mtr)	/*!< in/out: mini-transaction */
@@ -165,7 +163,6 @@ page_cur_insert_rec_zip(
 /*====================*/
 	page_cur_t*	cursor,	/*!< in/out: page cursor,
 				logical position unchanged  */
-	dict_index_t*	index,	/*!< in: record descriptor */
 	const rec_t*	rec,	/*!< in: pointer to a physical record */
 	rec_offs*	offsets,/*!< in/out: rec_get_offsets(rec, index) */
 	mtr_t*		mtr)	/*!< in/out: mini-transaction */
@@ -177,7 +174,6 @@ void
 page_cur_delete_rec(
 /*================*/
 	page_cur_t*		cursor,	/*!< in/out: a page cursor */
-	const dict_index_t*	index,	/*!< in: record descriptor */
 	const rec_offs*		offsets,/*!< in: rec_get_offsets(
 					cursor->rec, index) */
 	mtr_t*			mtr)	/*!< in/out: mini-transaction */
@@ -239,8 +235,6 @@ Searches the right position for a page cursor. */
 bool
 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 */
 	page_cur_mode_t		mode,	/*!< in: PAGE_CUR_L,
 					PAGE_CUR_LE, PAGE_CUR_G, or
@@ -251,13 +245,11 @@ page_cur_search_with_match(
 	ulint*			ilow_matched_fields,
 					/*!< in/out: already matched
 					fields in lower limit record */
-	page_cur_t*		cursor,	/*!< out: page cursor */
+	page_cur_t*		cursor,	/*!< in/out: page cursor */
 	rtr_info_t*		rtr_info);/*!< in/out: rtree search stack */
 #ifdef BTR_CUR_HASH_ADAPT
 MY_ATTRIBUTE((warn_unused_result))
 /** Search the right position for a page cursor.
-@param[in]	block			buffer block
-@param[in]	index			index tree
 @param[in]	tuple			key to be searched for
 @param[in]	mode			search mode
 @param[in,out]	iup_matched_fields	already matched fields in the
@@ -268,11 +260,9 @@ first partially matched field in the upper limit record
 lower limit record
 @param[in,out]	ilow_matched_bytes	already matched bytes in the
 first partially matched field in the lower limit record
-@param[out]	cursor			page cursor */
+@param[in,out]	cursor			page cursor */
 bool
 page_cur_search_with_match_bytes(
-	const buf_block_t*	block,
-	const dict_index_t*	index,
 	const dtuple_t*		tuple,
 	page_cur_mode_t		mode,
 	ulint*			iup_matched_fields,
@@ -284,16 +274,12 @@ page_cur_search_with_match_bytes(
 /***********************************************************//**
 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. */
-void
+void page_cur_open_on_rnd_user_rec(page_cur_t *cursor);
-page_cur_open_on_rnd_user_rec(
-/*==========================*/
-	buf_block_t*	block,	/*!< in: page */
-	page_cur_t*	cursor);/*!< out: page cursor */
 
 /** Index page cursor */
 
 struct page_cur_t{
-	const dict_index_t*	index;
+	dict_index_t*	index;
 	rec_t*		rec;	/*!< pointer to a record on page */
 	rec_offs*	offsets;
 	buf_block_t*	block;	/*!< pointer to the block containing rec */

storage/innobase/include/page0cur.inl

@@ -167,14 +167,12 @@ page_cur_tuple_insert(
 /*==================*/
 	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 */
 	rec_offs**	offsets,/*!< out: offsets on *rec */
 	mem_heap_t**	heap,	/*!< in/out: pointer to memory heap, or NULL */
 	ulint		n_ext,	/*!< in: number of externally stored columns */
 	mtr_t*		mtr)	/*!< in/out: mini-transaction */
 {
-	rec_t*		rec;
+	ulint size = rec_get_converted_size(cursor->index, tuple, n_ext);
-	ulint		size = rec_get_converted_size(index, tuple, n_ext);
 
 	if (!*heap) {
 		*heap = mem_heap_create(size
@@ -183,24 +181,23 @@ page_cur_tuple_insert(
 					* sizeof **offsets);
 	}
 
-	rec = rec_convert_dtuple_to_rec((byte*) mem_heap_alloc(*heap, size),
+	rec_t* rec = rec_convert_dtuple_to_rec(
-					index, tuple, n_ext);
+		static_cast<byte*>(mem_heap_alloc(*heap, size)),
+		cursor->index, tuple, n_ext);
 
-	*offsets = rec_get_offsets(rec, index, *offsets,
+	*offsets = rec_get_offsets(rec, cursor->index, *offsets,
 				   page_is_leaf(cursor->block->page.frame)
-				   ? index->n_core_fields : 0,
+				   ? cursor->index->n_core_fields : 0,
 				   ULINT_UNDEFINED, heap);
 	ut_ad(size == rec_offs_size(*offsets));
 
 	if (is_buf_block_get_page_zip(cursor->block)) {
-		rec = page_cur_insert_rec_zip(
+		rec = page_cur_insert_rec_zip(cursor, rec, *offsets, mtr);
-			cursor, index, rec, *offsets, mtr);
 	} else {
-		rec = page_cur_insert_rec_low(cursor,
+		rec = page_cur_insert_rec_low(cursor, rec, *offsets, mtr);
-					      index, rec, *offsets, mtr);
 	}
 
-	ut_ad(!rec || !cmp_dtuple_rec(tuple, rec, index, *offsets));
+	ut_ad(!rec || !cmp_dtuple_rec(tuple, rec, cursor->index, *offsets));
 	return(rec);
 }
 

storage/innobase/include/row0ins.h

@@ -1,7 +1,7 @@
 /*****************************************************************************
 
 Copyright (c) 1996, 2016, Oracle and/or its affiliates. All Rights Reserved.
-Copyright (c) 2017, 2021, MariaDB Corporation.
+Copyright (c) 2017, 2022, MariaDB Corporation.
 
 This program is free software; you can redistribute it and/or modify it under
 the terms of the GNU General Public License as published by the Free Software
@@ -78,7 +78,7 @@ dberr_t
 row_ins_clust_index_entry_low(
 /*==========================*/
 	ulint		flags,	/*!< in: undo logging and locking flags */
-	ulint		mode,	/*!< in: BTR_MODIFY_LEAF or BTR_MODIFY_TREE,
+	btr_latch_mode	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: clustered index */
@@ -94,13 +94,13 @@ same fields is found, the other record is necessarily marked deleted.
 It is then unmarked. Otherwise, the entry is just inserted to the index.
 @retval DB_SUCCESS on success
 @retval DB_LOCK_WAIT on lock wait when !(flags & BTR_NO_LOCKING_FLAG)
-@retval DB_FAIL if retry with BTR_MODIFY_TREE is needed
+@retval DB_FAIL if retry with BTR_INSERT_TREE is needed
 @return error code */
 dberr_t
 row_ins_sec_index_entry_low(
 /*========================*/
 	ulint		flags,	/*!< in: undo logging and locking flags */
-	ulint		mode,	/*!< in: BTR_MODIFY_LEAF or BTR_MODIFY_TREE,
+	btr_latch_mode	mode,	/*!< in: BTR_MODIFY_LEAF or BTR_INSERT_TREE,
 				depending on whether we wish optimistic or
 				pessimistic descent down the index tree */
 	dict_index_t*	index,	/*!< in: secondary index */

storage/innobase/include/row0row.h

@@ -1,7 +1,7 @@
 /*****************************************************************************
 
 Copyright (c) 1996, 2016, Oracle and/or its affiliates. All Rights Reserved.
-Copyright (c) 2016, 2020, MariaDB Corporation.
+Copyright (c) 2016, 2022, MariaDB Corporation.
 
 This program is free software; you can redistribute it and/or modify it under
 the terms of the GNU General Public License as published by the Free Software
@@ -303,13 +303,13 @@ row_build_row_ref_fast(
 /***************************************************************//**
 Searches the clustered index record for a row, if we have the row
 reference.
-@return TRUE if found */
+@return true if found */
-ibool
+bool
 row_search_on_row_ref(
 /*==================*/
 	btr_pcur_t*		pcur,	/*!< out: persistent cursor, which must
 					be closed by the caller */
-	ulint			mode,	/*!< in: BTR_MODIFY_LEAF, ... */
+	btr_latch_mode		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 */
@@ -321,7 +321,7 @@ on the secondary index record are preserved.
 rec_t*
 row_get_clust_rec(
 /*==============*/
-	ulint		mode,	/*!< in: BTR_MODIFY_LEAF, ... */
+	btr_latch_mode	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 */
@@ -363,9 +363,8 @@ Searches an index record.
 enum row_search_result
 row_search_index_entry(
 /*===================*/
-	dict_index_t*	index,	/*!< in: index */
 	const dtuple_t*	entry,	/*!< in: index entry */
-	ulint		mode,	/*!< in: BTR_MODIFY_LEAF, ... */
+	btr_latch_mode	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 */

storage/innobase/log/log0log.cc

@@ -812,10 +812,14 @@ func_exit:
 
     if (lsn <= sync_lsn)
     {
+#ifndef DBUG_OFF
+    skip_checkpoint:
+#endif
       log_sys.set_check_flush_or_checkpoint(false);
       goto func_exit;
     }
 
+    DBUG_EXECUTE_IF("ib_log_checkpoint_avoid_hard", goto skip_checkpoint;);
     log_sys.latch.rd_unlock();
 
     /* We must wait to prevent the tail of the log overwriting the head. */
@@ -912,13 +916,9 @@ loop:
 	}
 
 	/* We need these threads to stop early in shutdown. */
-	const char* thread_name;
+	const char* thread_name = srv_fast_shutdown != 2
-
+		&& trx_rollback_is_active
-   if (srv_fast_shutdown != 2 && trx_rollback_is_active) {
+		? "rollback of recovered transactions" : nullptr;
-		thread_name = "rollback of recovered transactions";
-	} else {
-		thread_name = NULL;
-	}
 
 	if (thread_name) {
 		ut_ad(!srv_read_only_mode);

storage/innobase/log/log0recv.cc

@@ -1148,6 +1148,7 @@ public:
 		}
 
 		mtr.commit();
+		clear();
 	}
 
 	/** Clear the data structure */
@@ -1174,14 +1175,6 @@ inline void recv_sys_t::trim(const page_id_t page_id, lsn_t lsn)
 			pages.erase(r);
 		}
 	}
-	if (fil_space_t* space = fil_space_get(page_id.space())) {
-		ut_ad(UT_LIST_GET_LEN(space->chain) == 1);
-		fil_node_t* file = UT_LIST_GET_FIRST(space->chain);
-		ut_ad(file->is_open());
-		os_file_truncate(file->name, file->handle,
-				 os_offset_t{page_id.page_no()}
-				 << srv_page_size_shift, true);
-	}
 	DBUG_VOID_RETURN;
 }
 
@@ -3485,7 +3478,17 @@ void recv_sys_t::apply(bool last_batch)
     {
       const trunc& t= truncated_undo_spaces[id];
       if (t.lsn)
-        trim(page_id_t(id + srv_undo_space_id_start, t.pages), t.lsn);
+      {
+        trim(page_id_t(id + srv_undo_space_id_start, 0), t.lsn);
+        if (fil_space_t *space = fil_space_get(id + srv_undo_space_id_start))
+        {
+          ut_ad(UT_LIST_GET_LEN(space->chain) == 1);
+          fil_node_t *file= UT_LIST_GET_FIRST(space->chain);
+          ut_ad(file->is_open());
+          os_file_truncate(file->name, file->handle,
+                           os_offset_t{t.pages} << srv_page_size_shift, true);
+        }
+      }
     }
 
     fil_system.extend_to_recv_size();

storage/innobase/page/page0cur.cc

@@ -277,8 +277,6 @@ Searches the right position for a page cursor. */
 bool
 page_cur_search_with_match(
 /*=======================*/
-	const buf_block_t*	block,	/*!< in: buffer block */
-	const dict_index_t*	index,	/*!< in/out: record descriptor */
 	const dtuple_t*		tuple,	/*!< in: data tuple */
 	page_cur_mode_t		mode,	/*!< in: PAGE_CUR_L,
 					PAGE_CUR_LE, PAGE_CUR_G, or
@@ -303,6 +301,8 @@ page_cur_search_with_match(
 	ulint		low_matched_fields;
 	ulint		cur_matched_fields;
 	int		cmp;
+	const dict_index_t* const index = cursor->index;
+	const buf_block_t* const block = cursor->block;
 #ifdef UNIV_ZIP_DEBUG
 	const page_zip_des_t*	page_zip = buf_block_get_page_zip(block);
 #endif /* UNIV_ZIP_DEBUG */
@@ -552,8 +552,6 @@ first partially matched field in the lower limit record
 @param[out]	cursor			page cursor */
 bool
 page_cur_search_with_match_bytes(
-	const buf_block_t*	block,
-	const dict_index_t*	index,
 	const dtuple_t*		tuple,
 	page_cur_mode_t		mode,
 	ulint*			iup_matched_fields,
@@ -575,6 +573,8 @@ page_cur_search_with_match_bytes(
 	ulint		cur_matched_fields;
 	ulint		cur_matched_bytes;
 	int		cmp;
+	const dict_index_t* const index = cursor->index;
+	const buf_block_t* const block = cursor->block;
 #ifdef UNIV_ZIP_DEBUG
 	const page_zip_des_t*	page_zip = buf_block_get_page_zip(block);
 #endif /* UNIV_ZIP_DEBUG */
@@ -802,18 +802,13 @@ up_rec_match:
 /***********************************************************//**
 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. */
-void
+void page_cur_open_on_rnd_user_rec(page_cur_t *cursor)
-page_cur_open_on_rnd_user_rec(
-/*==========================*/
-	buf_block_t*	block,	/*!< in: page */
-	page_cur_t*	cursor)	/*!< out: page cursor */
 {
-  cursor->block= block;
+  if (const ulint n_recs= page_get_n_recs(cursor->block->page.frame))
-  if (const ulint n_recs= page_get_n_recs(block->page.frame))
+    if ((cursor->rec= page_rec_get_nth(cursor->block->page.frame,
-    if ((cursor->rec= page_rec_get_nth(block->page.frame,
                                        ut_rnd_interval(n_recs) + 1)))
       return;
-  cursor->rec= page_get_infimum_rec(block->page.frame);
+  cursor->rec= page_get_infimum_rec(cursor->block->page.frame);
 }
 
 /**
@@ -1361,12 +1356,12 @@ rec_t*
 page_cur_insert_rec_low(
 /*====================*/
 	const page_cur_t*cur,	/*!< in: page cursor */
-	dict_index_t*	index,	/*!< in: record descriptor */
 	const rec_t*	rec,	/*!< in: record to insert after cur */
 	rec_offs*	offsets,/*!< in/out: rec_get_offsets(rec, index) */
 	mtr_t*		mtr)	/*!< in/out: mini-transaction */
 {
-  buf_block_t* block= cur->block;
+  buf_block_t *block= cur->block;
+  dict_index_t * const index= cur->index;
 
   ut_ad(rec_offs_validate(rec, index, offsets));
   ut_ad(rec_offs_n_fields(offsets) > 0);
@@ -1786,13 +1781,13 @@ page_cur_insert_rec_zip(
 /*====================*/
 	page_cur_t*	cursor,	/*!< in/out: page cursor,
 				logical position unchanged  */
-	dict_index_t*	index,	/*!< in: record descriptor */
 	const rec_t*	rec,	/*!< in: pointer to a physical record */
 	rec_offs*	offsets,/*!< in/out: rec_get_offsets(rec, index) */
 	mtr_t*		mtr)	/*!< in/out: mini-transaction */
 {
   page_zip_des_t * const page_zip= page_cur_get_page_zip(cursor);
   page_t * const page= cursor->block->page.frame;
+  dict_index_t * const index = cursor->index;
 
   ut_ad(page_zip);
   ut_ad(rec_offs_validate(rec, index, offsets));
@@ -1894,8 +1889,7 @@ page_cur_insert_rec_zip(
 
     /* Try compressing the whole page afterwards. */
     const mtr_log_t log_mode= mtr->set_log_mode(MTR_LOG_NONE);
-    rec_t *insert_rec= page_cur_insert_rec_low(cursor, index, rec, offsets,
+    rec_t *insert_rec= page_cur_insert_rec_low(cursor, rec, offsets, mtr);
-                                               mtr);
     mtr->set_log_mode(log_mode);
 
     if (insert_rec)
@@ -2243,7 +2237,6 @@ void
 page_cur_delete_rec(
 /*================*/
 	page_cur_t*		cursor,	/*!< in/out: a page cursor */
-	const dict_index_t*	index,	/*!< in: record descriptor */
 	const rec_offs*		offsets,/*!< in: rec_get_offsets(
 					cursor->rec, index) */
 	mtr_t*			mtr)	/*!< in/out: mini-transaction */
@@ -2265,6 +2258,7 @@ page_cur_delete_rec(
 	in the smallest user record, it cannot be used here either. */
 
 	current_rec = cursor->rec;
+	const dict_index_t* const index = cursor->index;
 	buf_block_t* const block = cursor->block;
 	ut_ad(rec_offs_validate(current_rec, index, offsets));
 	ut_ad(!!page_is_comp(block->page.frame)

storage/innobase/page/page0page.cc

@@ -458,6 +458,7 @@ page_copy_rec_list_end_no_locks(
 	rec_offs*	offsets		= offsets_;
 	rec_offs_init(offsets_);
 
+	cur1.index = cur2.index = index;
 	page_cur_position(rec, block, &cur1);
 
 	if (page_cur_is_before_first(&cur1) && !page_cur_move_to_next(&cur1)) {
@@ -483,8 +484,8 @@ page_copy_rec_list_end_no_locks(
 		rec_t*	ins_rec;
 		offsets = rec_get_offsets(cur1.rec, index, offsets, n_core,
 					  ULINT_UNDEFINED, &heap);
-		ins_rec = page_cur_insert_rec_low(&cur2, index,
+		ins_rec = page_cur_insert_rec_low(&cur2, cur1.rec, offsets,
-						  cur1.rec, offsets, mtr);
+						  mtr);
 		if (UNIV_UNLIKELY(!ins_rec || !page_cur_move_to_next(&cur1))) {
 			err = DB_CORRUPTION;
 			break;
@@ -733,6 +734,7 @@ corrupted:
 		log_mode = mtr_set_log_mode(mtr, MTR_LOG_NONE);
 	}
 
+	cur2.index = index;
 	page_cur_position(ret, new_block, &cur2);
 
 	const ulint n_core = page_rec_is_leaf(rec) ? index->n_core_fields : 0;
@@ -763,9 +765,8 @@ corrupted:
 			offsets = rec_get_offsets(cur1.rec, index, offsets,
 						  n_core,
 						  ULINT_UNDEFINED, &heap);
-			cur2.rec = page_cur_insert_rec_low(&cur2, index,
+			cur2.rec = page_cur_insert_rec_low(&cur2, cur1.rec,
-							   cur1.rec, offsets,
+							   offsets, mtr);
-							   mtr);
 			if (UNIV_UNLIKELY(!cur2.rec
 					  || !page_cur_move_to_next(&cur1))) {
 				*err = DB_CORRUPTION;
@@ -931,13 +932,14 @@ page_delete_rec_list_end(
     {
       page_cur_t cur;
       page_cur_position(rec, block, &cur);
+      cur.index= index;
       offsets= rec_get_offsets(rec, index, offsets, n_core,
 			       ULINT_UNDEFINED, &heap);
       rec= const_cast<rec_t*>(page_rec_get_next_low(rec, true));
 #ifdef UNIV_ZIP_DEBUG
       ut_a(page_zip_validate(&block->page.zip, page, index));
 #endif /* UNIV_ZIP_DEBUG */
-      page_cur_delete_rec(&cur, index, offsets, mtr);
+      page_cur_delete_rec(&cur, offsets, mtr);
     }
     while (page_offset(rec) != PAGE_NEW_SUPREMUM);
 
@@ -1134,6 +1136,7 @@ page_delete_rec_list_start(
 		return;
 	}
 
+	cur1.index = index;
 	page_cur_set_before_first(block, &cur1);
 	if (UNIV_UNLIKELY(!page_cur_move_to_next(&cur1))) {
 		ut_ad("corrupted page" == 0);
@@ -1147,7 +1150,7 @@ page_delete_rec_list_start(
 		offsets = rec_get_offsets(page_cur_get_rec(&cur1), index,
 					  offsets, n_core,
 					  ULINT_UNDEFINED, &heap);
-		page_cur_delete_rec(&cur1, index, offsets, mtr);
+		page_cur_delete_rec(&cur1, offsets, mtr);
 	}
 
 	if (UNIV_LIKELY_NULL(heap)) {

storage/innobase/que/que0que.cc

@@ -322,13 +322,9 @@ que_graph_free_recursive(
 	case QUE_NODE_UPDATE:
 		upd = static_cast<upd_node_t*>(node);
 
-		if (upd->in_mysql_interface) {
-
-			btr_pcur_free_for_mysql(upd->pcur);
-			upd->in_mysql_interface = false;
-		}
-
 		que_graph_free_recursive(upd->cascade_node);
+		ut_free(upd->pcur->old_rec_buf);
+		upd->pcur->old_rec_buf = NULL;
 
 		if (upd->cascade_heap) {
 			mem_heap_free(upd->cascade_heap);

storage/innobase/rem/rem0rec.cc

@@ -611,7 +611,7 @@ rec_init_offsets(
 	ulint	i	= 0;
 	rec_offs	offs;
 
-	/* This assertion was relaxed for the btr_cur_open_at_index_side()
+	/* This assertion was relaxed for the btr_cur_t::open_leaf()
 	call in btr_cur_instant_init_low(). We cannot invoke
 	index->is_instant(), because the same assertion would fail there
 	until btr_cur_instant_init_low() has invoked
@@ -839,7 +839,7 @@ rec_get_offsets_func(
 	bool	alter_metadata = false;
 
 	ut_ad(index->n_core_fields >= n_core);
-	/* This assertion was relaxed for the btr_cur_open_at_index_side()
+	/* This assertion was relaxed for the btr_cur_t::open_leaf()
 	call in btr_cur_instant_init_low(). We cannot invoke
 	index->is_instant(), because the same assertion would fail there
 	until btr_cur_instant_init_low() has invoked

storage/innobase/row/row0import.cc

@@ -253,9 +253,10 @@ public:
 	}
 
 	/** Position the cursor on the first user record. */
-	rec_t* open(buf_block_t* block) noexcept
+	rec_t* open(buf_block_t* block, const dict_index_t* index) noexcept
 		MY_ATTRIBUTE((warn_unused_result))
 	{
+		m_cur.index = const_cast<dict_index_t*>(index);
 		page_cur_set_before_first(block, &m_cur);
 		return next();
 	}
@@ -285,10 +286,9 @@ public:
 
 	/** Remove the current record
 	@return true on success */
-	bool remove(
+	bool remove(rec_offs* offsets) UNIV_NOTHROW
-		const dict_index_t*	index,
-		rec_offs*		offsets) UNIV_NOTHROW
 	{
+		const dict_index_t* const index = m_cur.index;
 		ut_ad(page_is_leaf(m_cur.block->page.frame));
 		/* We can't end up with an empty page unless it is root. */
 		if (page_get_n_recs(m_cur.block->page.frame) <= 1) {
@@ -311,7 +311,7 @@ public:
 			     page_zip, m_cur.block->page.frame, index));
 #endif /* UNIV_ZIP_DEBUG */
 
-		page_cur_delete_rec(&m_cur, index, offsets, &m_mtr);
+		page_cur_delete_rec(&m_cur, offsets, &m_mtr);
 
 #ifdef UNIV_ZIP_DEBUG
 		ut_a(!page_zip || page_zip_validate(
@@ -1513,8 +1513,9 @@ inline bool IndexPurge::open() noexcept
   m_mtr.start();
   m_mtr.set_log_mode(MTR_LOG_NO_REDO);
 
-  if (btr_pcur_open_at_index_side(true, m_index, BTR_MODIFY_LEAF,
+  btr_pcur_init(&m_pcur);
-                                  &m_pcur, true, 0, &m_mtr) != DB_SUCCESS)
+
+  if (m_pcur.open_leaf(true, m_index, BTR_MODIFY_LEAF, &m_mtr) != DB_SUCCESS)
     return false;
 
   rec_t *rec= page_rec_get_next(btr_pcur_get_rec(&m_pcur));
@@ -1559,7 +1560,7 @@ dberr_t IndexPurge::next() noexcept
 		return DB_CORRUPTION;
 	}
 	/* The following is based on btr_pcur_move_to_next_user_rec(). */
-	m_pcur.old_stored = false;
+	m_pcur.old_rec = nullptr;
 	ut_ad(m_pcur.latch_mode == BTR_MODIFY_LEAF);
 	do {
 		if (btr_pcur_is_after_last_on_page(&m_pcur)) {
@@ -1586,8 +1587,7 @@ tree structure may be changed during a pessimistic delete. */
 inline dberr_t IndexPurge::purge_pessimistic_delete() noexcept
 {
   dberr_t err;
-  if (m_pcur.restore_position(BTR_MODIFY_TREE | BTR_LATCH_FOR_DELETE,
+  if (m_pcur.restore_position(BTR_PURGE_TREE, &m_mtr) != btr_pcur_t::CORRUPTED)
-                              &m_mtr) != btr_pcur_t::CORRUPTED)
   {
     ut_ad(rec_get_deleted_flag(btr_pcur_get_rec(&m_pcur),
                                m_index->table->not_redundant()));
@@ -1722,10 +1722,8 @@ re-organising the B+tree.
 @return true if purge succeeded */
 inline bool PageConverter::purge() UNIV_NOTHROW
 {
-	const dict_index_t*	index = m_index->m_srv_index;
-
 	/* We can't have a page that is empty and not root. */
-	if (m_rec_iter.remove(index, m_offsets)) {
+	if (m_rec_iter.remove(m_offsets)) {
 
 		++m_index->m_stats.m_n_purged;
 
@@ -1789,7 +1787,7 @@ PageConverter::update_records(
 
 	/* This will also position the cursor on the first user record. */
 
-	if (!m_rec_iter.open(block)) {
+	if (!m_rec_iter.open(block, m_index->m_srv_index)) {
 		return DB_CORRUPTION;
 	}
 
@@ -2289,8 +2287,8 @@ row_import_set_sys_max_row_id(
 
 	mtr_set_log_mode(&mtr, MTR_LOG_NO_REDO);
 
-	if (btr_pcur_open_at_index_side(false, index, BTR_SEARCH_LEAF,
+	if (pcur.open_leaf(false, index, BTR_SEARCH_LEAF, &mtr)
-					&pcur, true, 0, &mtr) == DB_SUCCESS) {
+	    == DB_SUCCESS) {
 		rec = btr_pcur_move_to_prev_on_page(&pcur);
 
 		if (!rec) {

storage/innobase/row/row0ins.cc

@@ -174,7 +174,7 @@ dberr_t
 row_ins_sec_index_entry_by_modify(
 /*==============================*/
 	ulint		flags,	/*!< in: undo logging and locking flags */
-	ulint		mode,	/*!< in: BTR_MODIFY_LEAF or BTR_MODIFY_TREE,
+	ulint		mode,	/*!< in: BTR_MODIFY_LEAF or BTR_INSERT_TREE,
 				depending on whether mtr holds just a leaf
 				latch or also a tree latch */
 	btr_cur_t*	cursor,	/*!< in: B-tree cursor */
@@ -194,8 +194,8 @@ row_ins_sec_index_entry_by_modify(
 
 	rec = btr_cur_get_rec(cursor);
 
-	ut_ad(!dict_index_is_clust(cursor->index));
+	ut_ad(!cursor->index()->is_clust());
-	ut_ad(rec_offs_validate(rec, cursor->index, *offsets));
+	ut_ad(rec_offs_validate(rec, cursor->index(), *offsets));
 	ut_ad(!entry->info_bits);
 
 	/* We know that in the alphabetical ordering, entry and rec are
@@ -204,7 +204,7 @@ row_ins_sec_index_entry_by_modify(
 	difference. */
 
 	update = row_upd_build_sec_rec_difference_binary(
-		rec, cursor->index, *offsets, entry, heap);
+		rec, cursor->index(), *offsets, entry, heap);
 
 	if (!rec_get_deleted_flag(rec, rec_offs_comp(*offsets))) {
 		/* We should never insert in place of a record that
@@ -218,8 +218,8 @@ row_ins_sec_index_entry_by_modify(
 		returns. After that point, set_committed(true)
 		would be invoked in commit_inplace_alter_table(). */
 		ut_a(update->n_fields == 0);
-		ut_a(!cursor->index->is_committed());
+		ut_a(!cursor->index()->is_committed());
-		ut_ad(!dict_index_is_online_ddl(cursor->index));
+		ut_ad(!dict_index_is_online_ddl(cursor->index()));
 		return(DB_SUCCESS);
 	}
 
@@ -241,7 +241,7 @@ row_ins_sec_index_entry_by_modify(
 			break;
 		}
 	} else {
-		ut_a(mode == BTR_MODIFY_TREE);
+		ut_ad(mode == BTR_INSERT_TREE);
 		if (buf_pool.running_out()) {
 
 			return(DB_LOCK_TABLE_FULL);
@@ -288,15 +288,15 @@ row_ins_clust_index_entry_by_modify(
 	dberr_t		err = DB_SUCCESS;
 	btr_cur_t*	cursor	= btr_pcur_get_btr_cur(pcur);
 	TABLE*		mysql_table = NULL;
-	ut_ad(dict_index_is_clust(cursor->index));
+	ut_ad(cursor->index()->is_clust());
 
 	rec = btr_cur_get_rec(cursor);
 
 	ut_ad(rec_get_deleted_flag(rec,
-				   dict_table_is_comp(cursor->index->table)));
+				   cursor->index()->table->not_redundant()));
 	/* In delete-marked records, DB_TRX_ID must
 	always refer to an existing undo log record. */
-	ut_ad(rec_get_trx_id(rec, cursor->index));
+	ut_ad(rec_get_trx_id(rec, cursor->index()));
 
 	/* Build an update vector containing all the fields to be modified;
 	NOTE that this vector may NOT contain system columns trx_id or
@@ -307,7 +307,7 @@ row_ins_clust_index_entry_by_modify(
 	}
 
 	update = row_upd_build_difference_binary(
-		cursor->index, entry, rec, NULL, true, true,
+		cursor->index(), entry, rec, NULL, true, true,
 		thr_get_trx(thr), heap, mysql_table, &err);
 	if (err != DB_SUCCESS) {
 		return(err);
@@ -1115,7 +1115,7 @@ row_ins_foreign_check_on_constraint(
 		goto nonstandard_exit_func;
 	}
 
-	index = btr_pcur_get_btr_cur(pcur)->index;
+	index = pcur->index();
 
 	ut_a(index == foreign->foreign_index);
 
@@ -1138,7 +1138,9 @@ row_ins_foreign_check_on_constraint(
 
 		ref = row_build_row_ref(ROW_COPY_POINTERS, index, rec,
 					tmp_heap);
-		err = btr_pcur_open_with_no_init(clust_index, ref,
+		cascade->pcur->old_rec = nullptr;
+		cascade->pcur->btr_cur.page_cur.index = clust_index;
+		err = btr_pcur_open_with_no_init(ref,
 						 PAGE_CUR_LE, BTR_SEARCH_LEAF,
 						 cascade->pcur, mtr);
 		if (UNIV_UNLIKELY(err != DB_SUCCESS)) {
@@ -1622,9 +1624,9 @@ row_ins_check_foreign_constraint(
 	n_fields_cmp = dtuple_get_n_fields_cmp(entry);
 
 	dtuple_set_n_fields_cmp(entry, foreign->n_fields);
-
+	pcur.btr_cur.page_cur.index = check_index;
-	err = btr_pcur_open(check_index, entry, PAGE_CUR_GE,
+	err = btr_pcur_open(entry, PAGE_CUR_GE, BTR_SEARCH_LEAF, &pcur, 0,
-			    BTR_SEARCH_LEAF, &pcur, &mtr);
+			    &mtr);
 	if (UNIV_UNLIKELY(err != DB_SUCCESS)) {
 		goto end_scan;
 	}
@@ -2064,9 +2066,9 @@ row_ins_scan_sec_index_for_duplicate(
 
 	dtuple_set_n_fields_cmp(entry, n_unique);
 	const auto allow_duplicates = thr_get_trx(thr)->duplicates;
-
+	pcur.btr_cur.page_cur.index = index;
-	dberr_t err = btr_pcur_open(index, entry, PAGE_CUR_GE, BTR_SEARCH_LEAF,
+	dberr_t err = btr_pcur_open(entry, PAGE_CUR_GE, BTR_SEARCH_LEAF,
-				    &pcur, mtr);
+				    &pcur, 0, mtr);
 	if (err != DB_SUCCESS) {
 		goto end_scan;
 	}
@@ -2220,14 +2222,14 @@ row_ins_duplicate_error_in_clust_online(
 	dberr_t		err	= DB_SUCCESS;
 	const rec_t*	rec	= btr_cur_get_rec(cursor);
 
-	ut_ad(!cursor->index->is_instant());
+	ut_ad(!cursor->index()->is_instant());
 
 	if (cursor->low_match >= n_uniq && !page_rec_is_infimum(rec)) {
-		*offsets = rec_get_offsets(rec, cursor->index, *offsets,
+		*offsets = rec_get_offsets(rec, cursor->index(), *offsets,
-					   cursor->index->n_fields,
+					   cursor->index()->n_fields,
 					   ULINT_UNDEFINED, heap);
 		err = row_ins_duplicate_online(n_uniq, entry,
-					       rec, cursor->index, *offsets);
+					       rec, cursor->index(), *offsets);
 		if (err != DB_SUCCESS) {
 			return(err);
 		}
@@ -2238,11 +2240,11 @@ row_ins_duplicate_error_in_clust_online(
 	}
 
 	if (cursor->up_match >= n_uniq && !page_rec_is_supremum(rec)) {
-		*offsets = rec_get_offsets(rec, cursor->index, *offsets,
+		*offsets = rec_get_offsets(rec, cursor->index(), *offsets,
-					   cursor->index->n_fields,
+					   cursor->index()->n_fields,
 					   ULINT_UNDEFINED, heap);
 		err = row_ins_duplicate_online(n_uniq, entry,
-					       rec, cursor->index, *offsets);
+					       rec, cursor->index(), *offsets);
 	}
 
 	return(err);
@@ -2273,7 +2275,7 @@ row_ins_duplicate_error_in_clust(
 	rec_offs* offsets		= offsets_;
 	rec_offs_init(offsets_);
 
-	ut_ad(dict_index_is_clust(cursor->index));
+	ut_ad(cursor->index()->is_clust());
 
 	/* NOTE: For unique non-clustered indexes there may be any number
 	of delete marked records with the same value for the non-clustered
@@ -2288,15 +2290,17 @@ row_ins_duplicate_error_in_clust(
 	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);
+	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,
+			offsets = rec_get_offsets(rec, cursor->index(),
-						  cursor->index->n_core_fields,
+						  offsets,
+						  cursor->index()
+						  ->n_core_fields,
 						  ULINT_UNDEFINED, &heap);
 
 			/* We set a lock on the possible duplicate: this
@@ -2317,13 +2321,13 @@ row_ins_duplicate_error_in_clust(
 				err = row_ins_set_exclusive_rec_lock(
 					LOCK_REC_NOT_GAP,
 					btr_cur_get_block(cursor),
-					rec, cursor->index, offsets, thr);
+					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);
+					cursor->index(), offsets, thr);
 			}
 
 			switch (err) {
@@ -2335,11 +2339,11 @@ row_ins_duplicate_error_in_clust(
 			}
 
 			if (row_ins_dupl_error_with_rec(
-				    rec, entry, cursor->index, offsets)) {
+				    rec, entry, cursor->index(), offsets)) {
 duplicate:
-				trx->error_info = cursor->index;
+				trx->error_info = cursor->index();
 				err = DB_DUPLICATE_KEY;
-				if (cursor->index->table->versioned()
+				if (cursor->index()->table->versioned()
 				    && entry->vers_history_row())
 				{
 					ulint trx_id_len;
@@ -2363,8 +2367,10 @@ duplicate:
 		rec = page_rec_get_next(btr_cur_get_rec(cursor));
 
 		if (rec && !page_rec_is_supremum(rec)) {
-			offsets = rec_get_offsets(rec, cursor->index, offsets,
+			offsets = rec_get_offsets(rec, cursor->index(),
-						  cursor->index->n_core_fields,
+						  offsets,
+						  cursor->index()
+						  ->n_core_fields,
 						  ULINT_UNDEFINED, &heap);
 
 			if (trx->duplicates) {
@@ -2377,13 +2383,13 @@ duplicate:
 				err = row_ins_set_exclusive_rec_lock(
 					LOCK_REC_NOT_GAP,
 					btr_cur_get_block(cursor),
-					rec, cursor->index, offsets, thr);
+					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);
+					rec, cursor->index(), offsets, thr);
 			}
 
 			switch (err) {
@@ -2394,7 +2400,7 @@ duplicate:
 				/* fall through */
 			case DB_SUCCESS:
 				if (row_ins_dupl_error_with_rec(
-					    rec, entry, cursor->index,
+					    rec, entry, cursor->index(),
 					    offsets)) {
 					goto duplicate;
 				}
@@ -2436,7 +2442,7 @@ row_ins_must_modify_rec(
 	and a secondary index node pointer contains all index fields. */
 
 	return(cursor->low_match
-	       >= dict_index_get_n_unique_in_tree(cursor->index)
+	       >= dict_index_get_n_unique_in_tree(cursor->index())
 	       && !page_rec_is_infimum(btr_cur_get_rec(cursor)));
 }
 
@@ -2465,6 +2471,7 @@ row_ins_index_entry_big_rec(
 	btr_pcur_t	pcur;
 	rec_t*		rec;
 
+	pcur.btr_cur.page_cur.index = index;
 	ut_ad(index->is_primary());
 
 	DEBUG_SYNC_C_IF_THD(thd, "before_row_ins_extern_latch");
@@ -2476,8 +2483,8 @@ row_ins_index_entry_big_rec(
 		index->set_modified(mtr);
 	}
 
-	dberr_t error = btr_pcur_open(index, entry, PAGE_CUR_LE,
+	dberr_t error = btr_pcur_open(entry, PAGE_CUR_LE,
-				      BTR_MODIFY_TREE, &pcur, &mtr);
+				      BTR_MODIFY_TREE, &pcur, 0, &mtr);
 	if (error != DB_SUCCESS) {
 		return error;
 	}
@@ -2525,7 +2532,7 @@ dberr_t
 row_ins_clust_index_entry_low(
 /*==========================*/
 	ulint		flags,	/*!< in: undo logging and locking flags */
-	ulint		mode,	/*!< in: BTR_MODIFY_LEAF or BTR_MODIFY_TREE,
+	btr_latch_mode	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: clustered index */
@@ -2578,7 +2585,7 @@ row_ins_clust_index_entry_low(
 		} else {
 			if (mode == BTR_MODIFY_LEAF
 			    && dict_index_is_online_ddl(index)) {
-				mode = BTR_MODIFY_LEAF_ALREADY_S_LATCHED;
+				mode = BTR_MODIFY_LEAF_ALREADY_LATCHED;
 				mtr_s_lock_index(index, &mtr);
 			}
 
@@ -2603,8 +2610,8 @@ row_ins_clust_index_entry_low(
 	/* 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 */
- 	err = btr_pcur_open_low(index, 0, entry, PAGE_CUR_LE, mode, &pcur,
+	pcur.btr_cur.page_cur.index = index;
-				auto_inc, &mtr);
+	err = btr_pcur_open(entry, PAGE_CUR_LE, mode, &pcur, auto_inc, &mtr);
 	if (err != DB_SUCCESS) {
 		index->table->file_unreadable = true;
 commit_exit:
@@ -2732,7 +2739,7 @@ skip_bulk_insert:
 				/* fall through */
 			case DB_SUCCESS_LOCKED_REC:
 			case DB_DUPLICATE_KEY:
-				trx->error_info = cursor->index;
+				trx->error_info = cursor->index();
 			}
 		} else {
 			/* Note that the following may return also
@@ -2769,8 +2776,8 @@ do_insert:
 		rec_t*	insert_rec;
 
 		if (mode != BTR_MODIFY_TREE) {
-			ut_ad((mode & ulint(~BTR_ALREADY_S_LATCHED))
+			ut_ad(mode == BTR_MODIFY_LEAF ||
-			      == BTR_MODIFY_LEAF);
+			      mode == BTR_MODIFY_LEAF_ALREADY_LATCHED);
 			err = btr_cur_optimistic_insert(
 				flags, cursor, &offsets, &offsets_heap,
 				entry, &insert_rec, &big_rec,
@@ -2846,13 +2853,13 @@ same fields is found, the other record is necessarily marked deleted.
 It is then unmarked. Otherwise, the entry is just inserted to the index.
 @retval DB_SUCCESS on success
 @retval DB_LOCK_WAIT on lock wait when !(flags & BTR_NO_LOCKING_FLAG)
-@retval DB_FAIL if retry with BTR_MODIFY_TREE is needed
+@retval DB_FAIL if retry with BTR_INSERT_TREE is needed
 @return error code */
 dberr_t
 row_ins_sec_index_entry_low(
 /*========================*/
 	ulint		flags,	/*!< in: undo logging and locking flags */
-	ulint		mode,	/*!< in: BTR_MODIFY_LEAF or BTR_MODIFY_TREE,
+	btr_latch_mode	mode,	/*!< in: BTR_MODIFY_LEAF or BTR_INSERT_TREE,
 				depending on whether we wish optimistic or
 				pessimistic descent down the index tree */
 	dict_index_t*	index,	/*!< in: secondary index */
@@ -2867,7 +2874,7 @@ row_ins_sec_index_entry_low(
 	DBUG_ENTER("row_ins_sec_index_entry_low");
 
 	btr_cur_t	cursor;
-	ulint		search_mode	= mode;
+	btr_latch_mode	search_mode	= mode;
 	dberr_t		err;
 	ulint		n_unique;
 	mtr_t		mtr;
@@ -2877,10 +2884,11 @@ row_ins_sec_index_entry_low(
 	rtr_info_t	rtr_info;
 
 	ut_ad(!dict_index_is_clust(index));
-	ut_ad(mode == BTR_MODIFY_LEAF || mode == BTR_MODIFY_TREE);
+	ut_ad(mode == BTR_MODIFY_LEAF || mode == BTR_INSERT_TREE);
 
 	cursor.thr = thr;
 	cursor.rtr_info = NULL;
+	cursor.page_cur.index = index;
 	ut_ad(thr_get_trx(thr)->id != 0);
 
 	mtr.start();
@@ -2892,24 +2900,19 @@ row_ins_sec_index_entry_low(
 		mtr.set_log_mode(MTR_LOG_NO_REDO);
 	} else {
 		index->set_modified(mtr);
-		if (!dict_index_is_spatial(index)) {
-			search_mode |= BTR_INSERT;
-		}
 	}
 
 	/* 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 (dict_index_is_spatial(index)) {
+	if (index->is_spatial()) {
-		cursor.index = index;
 		rtr_init_rtr_info(&rtr_info, false, &cursor, index, false);
 		rtr_info_update_btr(&cursor, &rtr_info);
 
-		err = btr_cur_search_to_nth_level(
+		err = btr_cur_search_to_nth_level(0, entry,
-			index, 0, entry, PAGE_CUR_RTREE_INSERT,
+						  PAGE_CUR_RTREE_INSERT,
-			search_mode,
+						  search_mode, &cursor, &mtr);
-			&cursor, &mtr);
 
 		if (err == DB_SUCCESS && search_mode == BTR_MODIFY_LEAF
 		    && rtr_info.mbr_adj) {
@@ -2926,9 +2929,8 @@ row_ins_sec_index_entry_low(
 				index->set_modified(mtr);
 			}
 			err = btr_cur_search_to_nth_level(
-				index, 0, entry, PAGE_CUR_RTREE_INSERT,
+				0, entry, PAGE_CUR_RTREE_INSERT,
-				search_mode,
+				search_mode, &cursor, &mtr);
-				&cursor, &mtr);
 		}
 
 		DBUG_EXECUTE_IF(
@@ -2936,14 +2938,16 @@ row_ins_sec_index_entry_low(
 			goto func_exit;});
 
 	} else {
-		if (!thr_get_trx(thr)->check_unique_secondary) {
+		if (!index->table->is_temporary()) {
-			search_mode |= BTR_IGNORE_SEC_UNIQUE;
+			search_mode = btr_latch_mode(
+				search_mode
+				| (thr_get_trx(thr)->check_unique_secondary
+				   ? BTR_INSERT | BTR_IGNORE_SEC_UNIQUE
+				   : BTR_INSERT));
 		}
 
-		err = btr_cur_search_to_nth_level(
+		err = btr_cur_search_to_nth_level(0, entry, PAGE_CUR_LE,
-			index, 0, entry, PAGE_CUR_LE,
+						  search_mode, &cursor, &mtr);
-			search_mode,
-			&cursor, &mtr);
 	}
 
 	if (err != DB_SUCCESS) {
@@ -3018,11 +3022,12 @@ row_ins_sec_index_entry_low(
 		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. */
+		continue the insertion (bypassing the change buffer). */
 		err = btr_cur_search_to_nth_level(
-			index, 0, entry, PAGE_CUR_LE,
+			0, entry, PAGE_CUR_LE,
-			(search_mode
+			btr_latch_mode(search_mode
-			 & ~(BTR_INSERT | BTR_IGNORE_SEC_UNIQUE)),
+				       & ~(BTR_INSERT
+					   | BTR_IGNORE_SEC_UNIQUE)),
 			&cursor, &mtr);
 		if (err != DB_SUCCESS) {
 			goto func_exit;
@@ -3061,7 +3066,6 @@ row_ins_sec_index_entry_low(
 				err = rtr_ins_enlarge_mbr(&cursor, &mtr);
 			}
 		} else {
-			ut_ad(mode == BTR_MODIFY_TREE);
 			if (buf_pool.running_out()) {
 				err = DB_LOCK_TABLE_FULL;
 				goto func_exit;
@@ -3264,7 +3268,7 @@ row_ins_sec_index_entry(
 		log_free_check();
 
 		err = row_ins_sec_index_entry_low(
-			flags, BTR_MODIFY_TREE, index,
+			flags, BTR_INSERT_TREE, index,
 			offsets_heap, heap, entry, 0, thr);
 	}
 

storage/innobase/row/row0log.cc

@@ -1583,7 +1583,7 @@ row_log_table_apply_insert_low(
 
 		entry = row_build_index_entry(row, NULL, index, heap);
 		error = row_ins_sec_index_entry_low(
-			flags, BTR_MODIFY_TREE,
+			flags, BTR_INSERT_TREE,
 			index, offsets_heap, heap, entry,
 			thr_get_trx(thr)->id, thr);
 
@@ -1658,7 +1658,7 @@ row_log_table_apply_delete_low(
 	dberr_t		error;
 	row_ext_t*	ext;
 	dtuple_t*	row;
-	dict_index_t*	index	= btr_pcur_get_btr_cur(pcur)->index;
+	dict_index_t*	index	= pcur->index();
 
 	ut_ad(dict_index_is_clust(index));
 
@@ -1695,9 +1695,9 @@ err_exit:
 			row, ext, index, heap);
 		mtr->start();
 		index->set_modified(*mtr);
-		error = btr_pcur_open(index, entry, PAGE_CUR_LE,
+		pcur->btr_cur.page_cur.index = index;
-				      BTR_MODIFY_TREE | BTR_LATCH_FOR_DELETE,
+		error = btr_pcur_open(entry, PAGE_CUR_LE,
-				      pcur, mtr);
+				      BTR_PURGE_TREE, pcur, 0, mtr);
 		if (error) {
 			goto err_exit;
 		}
@@ -1761,6 +1761,7 @@ row_log_table_apply_delete(
 	btr_pcur_t	pcur;
 	rec_offs*	offsets;
 
+	pcur.btr_cur.page_cur.index = index;
 	ut_ad(rec_offs_n_fields(moffsets) == index->first_user_field());
 	ut_ad(!rec_offs_any_extern(moffsets));
 
@@ -1779,9 +1780,8 @@ row_log_table_apply_delete(
 
 	mtr_start(&mtr);
 	index->set_modified(mtr);
-	dberr_t err = btr_pcur_open(index, old_pk, PAGE_CUR_LE,
+	dberr_t err = btr_pcur_open(old_pk, PAGE_CUR_LE,
-				    BTR_MODIFY_TREE | BTR_LATCH_FOR_DELETE,
+				    BTR_PURGE_TREE, &pcur, 0, &mtr);
-				    &pcur, &mtr);
 	if (err != DB_SUCCESS) {
 		goto all_done;
 	}
@@ -1893,6 +1893,8 @@ row_log_table_apply_update(
 	dberr_t		error;
 	ulint		n_index = 0;
 
+	pcur.btr_cur.page_cur.index = index;
+
 	ut_ad(dtuple_get_n_fields_cmp(old_pk)
 	      == dict_index_get_n_unique(index));
 	ut_ad(dtuple_get_n_fields(old_pk) - (log->same_pk ? 0 : 2)
@@ -1915,8 +1917,8 @@ row_log_table_apply_update(
 
 	mtr.start();
 	index->set_modified(mtr);
-	error = btr_pcur_open(index, old_pk, PAGE_CUR_LE,
+	error = btr_pcur_open(old_pk, PAGE_CUR_LE,
-			      BTR_MODIFY_TREE, &pcur, &mtr);
+			      BTR_MODIFY_TREE, &pcur, 0, &mtr);
 	if (error != DB_SUCCESS) {
 func_exit:
 		mtr.commit();
@@ -2061,7 +2063,7 @@ func_exit_committed:
 
 	for (n_index += index->type != DICT_CLUSTERED;
 	     (index = dict_table_get_next_index(index)); n_index++) {
-		if (index->type & DICT_FTS) {
+		if (!index->is_btree()) {
 			continue;
 		}
 
@@ -2089,9 +2091,10 @@ func_exit_committed:
 
 		mtr.start();
 		index->set_modified(mtr);
+		pcur.btr_cur.page_cur.index = index;
 
 		if (ROW_FOUND != row_search_index_entry(
-			    index, entry, BTR_MODIFY_TREE, &pcur, &mtr)) {
+			    entry, BTR_MODIFY_TREE, &pcur, &mtr)) {
 			ut_ad(0);
 			error = DB_CORRUPTION;
 			break;
@@ -2111,7 +2114,7 @@ func_exit_committed:
 		error = row_ins_sec_index_entry_low(
 			BTR_CREATE_FLAG | BTR_NO_LOCKING_FLAG
 			| BTR_NO_UNDO_LOG_FLAG | BTR_KEEP_SYS_FLAG,
-			BTR_MODIFY_TREE, index, offsets_heap, heap,
+			BTR_INSERT_TREE, index, offsets_heap, heap,
 			entry, thr_get_trx(thr)->id, thr);
 
 		/* Report correct index name for duplicate key error. */
@@ -3071,13 +3074,14 @@ row_log_apply_op_low(
 
 	mtr_start(&mtr);
 	index->set_modified(mtr);
+	cursor.page_cur.index = index;
 
 	/* We perform the pessimistic variant of the operations if we
 	already hold index->lock exclusively. First, search the
 	record. The operation may already have been performed,
 	depending on when the row in the clustered index was
 	scanned. */
-	*error = btr_cur_search_to_nth_level(index, 0, entry, PAGE_CUR_LE,
+	*error = btr_cur_search_to_nth_level(0, entry, PAGE_CUR_LE,
 					     has_index_lock
 					     ? BTR_MODIFY_TREE
 					     : BTR_MODIFY_LEAF,
@@ -3132,7 +3136,7 @@ row_log_apply_op_low(
 				mtr_start(&mtr);
 				index->set_modified(mtr);
 				*error = btr_cur_search_to_nth_level(
-					index, 0, entry, PAGE_CUR_LE,
+					0, entry, PAGE_CUR_LE,
 					BTR_MODIFY_TREE, &cursor, &mtr);
 				if (UNIV_UNLIKELY(*error != DB_SUCCESS)) {
 					goto func_exit;
@@ -3236,7 +3240,7 @@ insert_the_rec:
 				mtr_start(&mtr);
 				index->set_modified(mtr);
 				*error = btr_cur_search_to_nth_level(
-					index, 0, entry, PAGE_CUR_LE,
+					0, entry, PAGE_CUR_LE,
 					BTR_MODIFY_TREE, &cursor, &mtr);
 				if (*error != DB_SUCCESS) {
 					break;
@@ -3951,8 +3955,8 @@ void UndorecApplier::log_insert(const dtuple_t &tuple,
     }
 
     bool success= true;
-    dict_index_t *index= dict_table_get_next_index(clust_index);
+    for (dict_index_t *index= clust_index;
-    while (index)
+         (index= dict_table_get_next_index(index)) != nullptr; )
     {
       index->lock.s_lock(SRW_LOCK_CALL);
       if (index->online_log &&
@@ -3971,7 +3975,6 @@ void UndorecApplier::log_insert(const dtuple_t &tuple,
         row_log_mark_other_online_index_abort(index->table);
         return;
       }
-      index= dict_table_get_next_index(index);
     }
   }
 }

storage/innobase/row/row0merge.cc

@@ -145,13 +145,13 @@ public:
 			mtr.start();
 			index->set_modified(mtr);
 
-			ins_cur.index = index;
+			ins_cur.page_cur.index = index;
 			rtr_init_rtr_info(&rtr_info, false, &ins_cur, index,
 					  false);
 			rtr_info_update_btr(&ins_cur, &rtr_info);
 
 			error = btr_cur_search_to_nth_level(
-				index, 0, dtuple, PAGE_CUR_RTREE_INSERT,
+				0, dtuple, PAGE_CUR_RTREE_INSERT,
 				BTR_MODIFY_LEAF, &ins_cur, &mtr);
 
 			/* It need to update MBR in parent entry,
@@ -165,7 +165,7 @@ public:
 				mtr.start();
 				index->set_modified(mtr);
 				error = btr_cur_search_to_nth_level(
-					index, 0, dtuple,
+					0, dtuple,
 					PAGE_CUR_RTREE_INSERT,
 					BTR_MODIFY_TREE, &ins_cur, &mtr);
 			}
@@ -190,7 +190,7 @@ public:
 
 				rtr_info_update_btr(&ins_cur, &rtr_info);
 				error = btr_cur_search_to_nth_level(
-					index, 0, dtuple,
+					0, dtuple,
 					PAGE_CUR_RTREE_INSERT,
 					BTR_MODIFY_TREE,
 					&ins_cur, &mtr);
@@ -1999,8 +1999,7 @@ row_merge_read_clustered_index(
 		? col_map[old_trx_id_col] : old_trx_id_col;
 	uint64_t n_rows = 0;
 
-	err = btr_pcur_open_at_index_side(true, clust_index, BTR_SEARCH_LEAF,
+	err = pcur.open_leaf(true, clust_index, BTR_SEARCH_LEAF, &mtr);
-					  &pcur, true, 0, &mtr);
 	if (err != DB_SUCCESS) {
 err_exit:
 		trx->error_key_num = 0;
@@ -2232,13 +2231,15 @@ end_of_index:
 				if (!block) {
 					goto err_exit;
 				}
-				btr_leaf_page_release(page_cur_get_block(cur),
+
-						      BTR_SEARCH_LEAF, &mtr);
 				page_cur_set_before_first(block, cur);
 				if (!page_cur_move_to_next(cur)
 				    || page_cur_is_after_last(cur)) {
 					goto corrupted_rec;
 				}
+
+				const auto s = mtr.get_savepoint();
+				mtr.rollback_to_savepoint(s - 2, s - 1);
 			}
 		} else {
 			mem_heap_empty(row_heap);

storage/innobase/row/row0mysql.cc

@@ -1456,11 +1456,8 @@ row_create_update_node_for_mysql(
 
 	node->in_mysql_interface = true;
 	node->is_delete = NO_DELETE;
-	node->searched_update = FALSE;
+	node->pcur = new (mem_heap_alloc(heap, sizeof(btr_pcur_t)))
-	node->select = NULL;
+		btr_pcur_t();
-	node->pcur = btr_pcur_create_for_mysql();
-
-	DBUG_PRINT("info", ("node: %p, pcur: %p", node, node->pcur));
 
 	node->table = table;
 
@@ -1472,10 +1469,6 @@ row_create_update_node_for_mysql(
 	UT_LIST_INIT(node->columns, &sym_node_t::col_var_list);
 
 	node->has_clust_rec_x_lock = TRUE;
-	node->cmpl_info = 0;
-
-	node->table_sym = NULL;
-	node->col_assign_list = NULL;
 
 	DBUG_RETURN(node);
 }
@@ -1650,8 +1643,7 @@ row_update_for_mysql(row_prebuilt_t* prebuilt)
 	clust_index = dict_table_get_first_index(table);
 
 	btr_pcur_copy_stored_position(node->pcur,
-				      prebuilt->pcur->btr_cur.index
+				      prebuilt->pcur->index() == clust_index
-				      == clust_index
 				      ? prebuilt->pcur
 				      : prebuilt->clust_pcur);
 
@@ -1804,7 +1796,7 @@ row_unlock_for_mysql(
 		}
 
 		rec = btr_pcur_get_rec(pcur);
-		index = btr_pcur_get_btr_cur(pcur)->index;
+		index = pcur->index();
 
 		/* If the record has been modified by this
 		transaction, do not unlock it. */

storage/innobase/row/row0purge.cc

@@ -67,7 +67,7 @@ static
 ibool
 row_purge_reposition_pcur(
 /*======================*/
-	ulint		mode,	/*!< in: latching mode */
+	btr_latch_mode	mode,	/*!< in: latching mode */
 	purge_node_t*	node,	/*!< in: row purge node */
 	mtr_t*		mtr)	/*!< in: mtr */
 {
@@ -104,7 +104,7 @@ bool
 row_purge_remove_clust_if_poss_low(
 /*===============================*/
 	purge_node_t*	node,	/*!< in/out: row purge node */
-	ulint		mode)	/*!< in: BTR_MODIFY_LEAF or BTR_MODIFY_TREE */
+	btr_latch_mode	mode)	/*!< in: BTR_MODIFY_LEAF or BTR_MODIFY_TREE */
 {
 	dict_index_t* index = dict_table_get_first_index(node->table);
 	table_id_t table_id = 0;
@@ -216,7 +216,7 @@ close_and_exit:
 			btr_pcur_get_btr_cur(&node->pcur), 0, &mtr);
 	} else {
 		dberr_t	err;
-		ut_ad(mode == (BTR_MODIFY_TREE | BTR_LATCH_FOR_DELETE));
+		ut_ad(mode == BTR_PURGE_TREE);
 		btr_cur_pessimistic_delete(
 			&err, FALSE, btr_pcur_get_btr_cur(&node->pcur), 0,
 			false, &mtr);
@@ -257,8 +257,7 @@ row_purge_remove_clust_if_poss(
 	for (ulint n_tries = 0;
 	     n_tries < BTR_CUR_RETRY_DELETE_N_TIMES;
 	     n_tries++) {
-		if (row_purge_remove_clust_if_poss_low(
+		if (row_purge_remove_clust_if_poss_low(node, BTR_PURGE_TREE)) {
-			    node, BTR_MODIFY_TREE | BTR_LATCH_FOR_DELETE)) {
 			return(true);
 		}
 
@@ -348,10 +347,9 @@ row_purge_remove_sec_if_poss_tree(
 	log_free_check();
 	mtr.start();
 	index->set_modified(mtr);
+	pcur.btr_cur.page_cur.index = index;
 
-	search_result = row_search_index_entry(
+	search_result = row_search_index_entry(entry, BTR_PURGE_TREE,
-				index, entry,
-				BTR_MODIFY_TREE | BTR_LATCH_FOR_DELETE,
 					       &pcur, &mtr);
 
 	switch (search_result) {
@@ -455,6 +453,7 @@ row_purge_remove_sec_if_poss_leaf(
 	virtual index. */
 	mode = (index->type & (DICT_SPATIAL | DICT_VIRTUAL))
 		? BTR_MODIFY_LEAF : BTR_PURGE_LEAF;
+	pcur.btr_cur.page_cur.index = index;
 
 	/* Set the purge node for the call to row_purge_poss_sec(). */
 	pcur.btr_cur.purge_node = node;
@@ -462,7 +461,7 @@ row_purge_remove_sec_if_poss_leaf(
 		pcur.btr_cur.thr = NULL;
 		index->lock.u_lock(SRW_LOCK_CALL);
 		search_result = row_search_index_entry(
-			index, entry, mode, &pcur, &mtr);
+			entry, mode, &pcur, &mtr);
 		index->lock.u_unlock();
 	} else {
 		/* Set the query thread, so that ibuf_insert_low() will be
@@ -470,7 +469,7 @@ row_purge_remove_sec_if_poss_leaf(
 		pcur.btr_cur.thr = static_cast<que_thr_t*>(
 			que_node_get_parent(node));
 		search_result = row_search_index_entry(
-			index, entry, mode, &pcur, &mtr);
+			entry, mode, &pcur, &mtr);
 	}
 
 	switch (search_result) {
@@ -1343,11 +1342,11 @@ purge_node_t::validate_pcur()
 		return(true);
 	}
 
-	if (!pcur.old_stored) {
+	if (!pcur.old_rec) {
 		return(true);
 	}
 
-	dict_index_t*	clust_index = pcur.btr_cur.index;
+	dict_index_t* clust_index = pcur.index();
 
 	rec_offs* offsets = rec_get_offsets(
 		pcur.old_rec, clust_index, NULL, pcur.old_n_core_fields,

storage/innobase/row/row0row.cc

@@ -1183,32 +1183,28 @@ row_build_row_ref_in_tuple(
 /***************************************************************//**
 Searches the clustered index record for a row, if we have the row reference.
 @return TRUE if found */
-ibool
+bool
 row_search_on_row_ref(
 /*==================*/
 	btr_pcur_t*		pcur,	/*!< out: persistent cursor, which must
 					be closed by the caller */
-	ulint			mode,	/*!< in: BTR_MODIFY_LEAF, ... */
+	btr_latch_mode		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);
+	dict_index_t *index = dict_table_get_first_index(table);
+	btr_pcur_init(pcur);
+	pcur->btr_cur.page_cur.index = index;
 
 	if (UNIV_UNLIKELY(ref->info_bits != 0)) {
 		ut_ad(ref->is_metadata());
 		ut_ad(ref->n_fields <= index->n_uniq);
-		if (btr_pcur_open_at_index_side(
+		if (pcur->open_leaf(true, index, mode, mtr) != DB_SUCCESS
-			    true, index, mode, pcur, true, 0, mtr)
-		    != DB_SUCCESS
 		    || !btr_pcur_move_to_next_user_rec(pcur, mtr)) {
-			return FALSE;
+			return false;
 		}
 		/* We do not necessarily have index->is_instant() here,
 		because we could be executing a rollback of an
@@ -1220,27 +1216,14 @@ row_search_on_row_ref(
 			& REC_INFO_MIN_REC_FLAG;
 	} else {
 		ut_a(ref->n_fields == index->n_uniq);
-		if (btr_pcur_open(index, ref, PAGE_CUR_LE, mode, pcur, mtr)
+		if (btr_pcur_open(ref, PAGE_CUR_LE, mode, pcur, 0, mtr)
 		    != DB_SUCCESS) {
-			return FALSE;
+			return false;
 		}
 	}
 
-	low_match = btr_pcur_get_low_match(pcur);
+	return !page_rec_is_infimum(btr_pcur_get_rec(pcur))
-
+		&& btr_pcur_get_low_match(pcur) == dtuple_get_n_fields(ref);
-	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);
 }
 
 /*********************************************************************//**
@@ -1250,7 +1233,7 @@ on the secondary index record are preserved.
 rec_t*
 row_get_clust_rec(
 /*==============*/
-	ulint		mode,	/*!< in: BTR_MODIFY_LEAF, ... */
+	btr_latch_mode	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 */
@@ -1283,9 +1266,8 @@ Searches an index record.
 enum row_search_result
 row_search_index_entry(
 /*===================*/
-	dict_index_t*	index,	/*!< in: index */
 	const dtuple_t*	entry,	/*!< in: index entry */
-	ulint		mode,	/*!< in: BTR_MODIFY_LEAF, ... */
+	btr_latch_mode	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 */
@@ -1296,12 +1278,12 @@ row_search_index_entry(
 
 	ut_ad(dtuple_check_typed(entry));
 
-	if (index->is_spatial()) {
+	if (pcur->index()->is_spatial()) {
-		if (rtr_pcur_open(index, entry, mode, pcur, mtr)) {
+		if (rtr_pcur_open(pcur->index(), entry, mode, pcur, mtr)) {
 			return ROW_NOT_FOUND;
 		}
 	} else {
-		if (btr_pcur_open(index, entry, PAGE_CUR_LE, mode, pcur, mtr)
+		if (btr_pcur_open(entry, PAGE_CUR_LE, mode, pcur, 0, mtr)
 		    != DB_SUCCESS) {
 			return ROW_NOT_FOUND;
 		}
@@ -1310,7 +1292,7 @@ row_search_index_entry(
 	switch (btr_pcur_get_btr_cur(pcur)->flag) {
 	case BTR_CUR_DELETE_REF:
 		ut_ad(!(~mode & BTR_DELETE));
-		ut_ad(!index->is_spatial());
+		ut_ad(!pcur->index()->is_spatial());
 		return(ROW_NOT_DELETED_REF);
 
 	case BTR_CUR_DEL_MARK_IBUF:

storage/innobase/row/row0sel.cc

@@ -1008,17 +1008,16 @@ row_sel_get_clust_rec(
 
 	*out_rec = NULL;
 
-	offsets = rec_get_offsets(rec,
+	offsets = rec_get_offsets(rec, plan->pcur.index(), offsets,
-				  btr_pcur_get_btr_cur(&plan->pcur)->index,
+				  plan->pcur.index()->n_core_fields,
-				  offsets,
+				  ULINT_UNDEFINED, &heap);
-				  btr_pcur_get_btr_cur(&plan->pcur)->index
-				  ->n_core_fields, ULINT_UNDEFINED, &heap);
 
 	row_build_row_ref_fast(plan->clust_ref, plan->clust_map, rec, offsets);
 
 	index = dict_table_get_first_index(plan->table);
-
+	plan->clust_pcur.old_rec = nullptr;
-	dberr_t err = btr_pcur_open_with_no_init(index, plan->clust_ref,
+	plan->clust_pcur.btr_cur.page_cur.index = index;
+	dberr_t err = btr_pcur_open_with_no_init(plan->clust_ref,
 						 PAGE_CUR_LE, BTR_SEARCH_LEAF,
 						 &plan->clust_pcur, mtr);
 	if (UNIV_UNLIKELY(err != DB_SUCCESS)) {
@@ -1410,6 +1409,9 @@ row_sel_open_pcur(
 		cond = UT_LIST_GET_NEXT(cond_list, cond);
 	}
 
+	plan->pcur.old_rec = nullptr;
+	plan->pcur.btr_cur.page_cur.index = index;
+
 	dberr_t err;
 
 	if (plan->tuple) {
@@ -1429,13 +1431,12 @@ row_sel_open_pcur(
 					 que_node_get_val(exp));
 		}
 
-		err = btr_pcur_open_with_no_init(index, plan->tuple,
+		err = btr_pcur_open_with_no_init(plan->tuple,
 						 plan->mode, BTR_SEARCH_LEAF,
 						 &plan->pcur, mtr);
 	} else {
-		err = btr_pcur_open_at_index_side(plan->asc, index,
+		err = plan->pcur.open_leaf(plan->asc, index, BTR_SEARCH_LEAF,
-						  BTR_SEARCH_LEAF, &plan->pcur,
+					   mtr);
-						  false, 0, mtr);
 	}
 
 	plan->pcur_is_open = err == DB_SUCCESS;
@@ -3370,6 +3371,7 @@ Row_sel_get_clust_rec_for_mysql::operator()(
 	rec_t*		old_vers;
 	trx_t*		trx;
 
+	prebuilt->clust_pcur->old_rec = nullptr;
 	*out_rec = NULL;
 	trx = thr_get_trx(thr);
 
@@ -3380,9 +3382,9 @@ Row_sel_get_clust_rec_for_mysql::operator()(
 				   sec_index, *offsets);
 
 	clust_index = dict_table_get_first_index(sec_index->table);
+	prebuilt->clust_pcur->btr_cur.page_cur.index = clust_index;
 
-	dberr_t err = btr_pcur_open_with_no_init(clust_index,
+	dberr_t err = btr_pcur_open_with_no_init(prebuilt->clust_ref,
-						 prebuilt->clust_ref,
 						 PAGE_CUR_LE, BTR_SEARCH_LEAF,
 						 prebuilt->clust_pcur, mtr);
 	if (UNIV_UNLIKELY(err != DB_SUCCESS)) {
@@ -3447,9 +3449,10 @@ Row_sel_get_clust_rec_for_mysql::operator()(
 				rec, sec_index, true,
 				sec_index->n_fields, heap);
 			page_cur_t     page_cursor;
+			page_cursor.block = block;
+			page_cursor.index = sec_index;
 			ulint up_match = 0, low_match = 0;
-			ut_ad(!page_cur_search_with_match(block, sec_index,
+			ut_ad(!page_cur_search_with_match(tuple, PAGE_CUR_LE,
-							  tuple, PAGE_CUR_LE,
 							  &up_match,
 							  &low_match,
 							  &page_cursor,
@@ -3639,7 +3642,8 @@ record with the same ordering prefix in in the B-tree index
 @return 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) */
 static bool sel_restore_position_for_mysql(bool *same_user_rec,
-                                           ulint latch_mode, btr_pcur_t *pcur,
+                                           btr_latch_mode latch_mode,
+                                           btr_pcur_t *pcur,
                                            bool moves_up, mtr_t *mtr)
 {
 	auto status = pcur->restore_position(latch_mode, mtr);
@@ -3668,7 +3672,7 @@ static bool sel_restore_position_for_mysql(bool *same_user_rec,
 next:
 			if (btr_pcur_move_to_next(pcur, mtr)
 			    && rec_is_metadata(btr_pcur_get_rec(pcur),
-					       *pcur->btr_cur.index)) {
+					       *pcur->index())) {
 				btr_pcur_move_to_next(pcur, mtr);
 			}
 
@@ -3684,7 +3688,7 @@ next:
 prev:
 		if (btr_pcur_is_on_user_rec(pcur) && !moves_up
 		    && !rec_is_metadata(btr_pcur_get_rec(pcur),
-					*pcur->btr_cur.index)) {
+					*pcur->index())) {
 			if (!btr_pcur_move_to_prev(pcur, mtr)) {
 				return true;
 			}
@@ -3964,8 +3968,9 @@ row_sel_try_search_shortcut_for_mysql(
 	ut_ad(!index->table->is_temporary());
 	ut_ad(!prebuilt->templ_contains_blob);
 	ut_ad(trx->read_view.is_open());
+	pcur->old_rec = nullptr;
 
-	if (btr_pcur_open_with_no_init(index, search_tuple, PAGE_CUR_GE,
+	if (btr_pcur_open_with_no_init(search_tuple, PAGE_CUR_GE,
 				       BTR_SEARCH_LEAF, pcur, mtr)
 	    != DB_SUCCESS) {
 		return SEL_RETRY;
@@ -4395,6 +4400,8 @@ row_search_mvcc(
 		DBUG_RETURN(DB_CORRUPTION);
 	}
 
+	pcur->btr_cur.page_cur.index = index;
+
 	/* We need to get the virtual column values stored in secondary
 	index key, if this is covered index scan or virtual key read is
 	requested. */
@@ -4766,6 +4773,7 @@ wait_table_again:
 
 	} else if (dtuple_get_n_fields(search_tuple) > 0) {
 		pcur->btr_cur.thr = thr;
+		pcur->old_rec = nullptr;
 
 		if (dict_index_is_spatial(index)) {
 			if (!prebuilt->rtr_info) {
@@ -4785,7 +4793,7 @@ wait_table_again:
 			}
 		}
 
-		err = btr_pcur_open_with_no_init(index, search_tuple, mode,
+		err = btr_pcur_open_with_no_init(search_tuple, mode,
 						 BTR_SEARCH_LEAF, pcur, &mtr);
 
 		if (err != DB_SUCCESS) {
@@ -4831,9 +4839,8 @@ page_corrupted:
 			}
 		}
 	} else if (mode == PAGE_CUR_G || mode == PAGE_CUR_L) {
-		err = btr_pcur_open_at_index_side(
+		err = pcur->open_leaf(mode == PAGE_CUR_G, index,
-			mode == PAGE_CUR_G, index, BTR_SEARCH_LEAF,
+				      BTR_SEARCH_LEAF, &mtr);
-			pcur, false, 0, &mtr);
 
 		if (err != DB_SUCCESS) {
 			if (err == DB_DECRYPTION_FAILED) {
@@ -5019,7 +5026,7 @@ wrong_offs:
 
 			page_cur_set_after_last(btr_pcur_get_block(pcur),
 						btr_pcur_get_page_cur(pcur));
-			pcur->old_stored = false;
+			pcur->old_rec = nullptr;
 			goto next_rec;
 		}
 	}
@@ -5786,7 +5793,7 @@ next_rec_after_check:
 			/* This is based on btr_pcur_move_to_next() */
 			ut_ad(pcur->pos_state == BTR_PCUR_IS_POSITIONED);
 			ut_ad(pcur->latch_mode != BTR_NO_LATCHES);
-			pcur->old_stored = false;
+			pcur->old_rec = nullptr;
 			if (btr_pcur_is_after_last_on_page(pcur)) {
 				if (btr_pcur_is_after_last_in_tree(pcur)) {
 					goto not_moved;
@@ -6198,8 +6205,6 @@ dberr_t row_check_index(row_prebuilt_t *prebuilt, ulint *n_rows)
   *n_rows= 0;
   dict_index_t *const index= prebuilt->index;
 
-  prebuilt->fetch_direction= ROW_SEL_NEXT;
-
   if (!index->is_btree())
     return DB_CORRUPTION;
 
@@ -6210,9 +6215,8 @@ dberr_t row_check_index(row_prebuilt_t *prebuilt, ulint *n_rows)
   mtr.start();
 
   dict_index_t *clust_index= dict_table_get_first_index(prebuilt->table);
-
+  prebuilt->clust_pcur->btr_cur.page_cur.index = clust_index;
-  dberr_t err= btr_pcur_open_at_index_side(true, index, BTR_SEARCH_LEAF,
+  dberr_t err= prebuilt->pcur->open_leaf(true, index, BTR_SEARCH_LEAF, &mtr);
-                                           prebuilt->pcur, false, 0, &mtr);
   if (UNIV_UNLIKELY(err != DB_SUCCESS))
   {
 func_exit:
@@ -6300,14 +6304,16 @@ rec_loop:
     goto next_rec;
   }
 
-  if (index->is_clust())
+  if (prebuilt->table->is_temporary())
   {
-    if (prebuilt->trx->isolation_level == TRX_ISO_READ_UNCOMMITTED)
+  count_or_not:
-    {
+    if (rec_deleted)
-      if (!rec_deleted)
-        goto count_row;
       goto next_rec;
   }
+  else if (index->is_clust())
+  {
+    if (prebuilt->trx->isolation_level == TRX_ISO_READ_UNCOMMITTED)
+      goto count_or_not;
 
     trx_id_t rec_trx_id= row_get_rec_trx_id(rec, index, offsets);
 
@@ -6371,10 +6377,7 @@ rec_loop:
                           ER_NOT_KEYFILE, "InnoDB: %s", w.m_oss.str().c_str());
     }
 
-    if (!rec_deleted)
+    goto count_or_not;
-      goto count_row;
-
-    goto next_rec;
   }
   else if (const trx_id_t page_trx_id= page_get_max_trx_id(page_align(rec)))
   {
@@ -6388,7 +6391,7 @@ rec_loop:
       const auto savepoint= mtr.get_savepoint();
 
       row_build_row_ref_in_tuple(prebuilt->clust_ref, rec, index, offsets);
-      err= btr_pcur_open_with_no_init(clust_index, prebuilt->clust_ref,
+      err= btr_pcur_open_with_no_init(prebuilt->clust_ref,
                                       PAGE_CUR_LE, BTR_SEARCH_LEAF,
                                       prebuilt->clust_pcur, &mtr);
       if (err != DB_SUCCESS)
@@ -6887,8 +6890,7 @@ row_search_get_max_rec(
 	const rec_t*	rec;
 	const bool	desc	= index->fields[0].descending;
 
-	if (btr_pcur_open_at_index_side(desc, index, BTR_SEARCH_LEAF, &pcur,
+	if (pcur.open_leaf(desc, index, BTR_SEARCH_LEAF, mtr) != DB_SUCCESS) {
-					true, 0, mtr) != DB_SUCCESS) {
 		return nullptr;
 	}
 

storage/innobase/row/row0uins.cc

@@ -68,7 +68,7 @@ row_undo_ins_remove_clust_rec(
 	dberr_t		err;
 	ulint		n_tries	= 0;
 	mtr_t		mtr;
-	dict_index_t*	index	= node->pcur.btr_cur.index;
+	dict_index_t*	index	= node->pcur.index();
 	table_id_t table_id = 0;
 	const bool dict_locked = node->trx->dict_operation_lock_mode;
 restart:
@@ -207,9 +207,8 @@ retry:
 	} else {
 		index->set_modified(mtr);
 	}
-	ut_a(
+	ut_a(node->pcur.restore_position(BTR_PURGE_TREE, &mtr)
-	    node->pcur.restore_position(BTR_MODIFY_TREE | BTR_LATCH_FOR_DELETE,
+	     == btr_pcur_t::SAME_ALL);
-	      &mtr) == btr_pcur_t::SAME_ALL);
 
 	btr_cur_pessimistic_delete(&err, FALSE, &node->pcur.btr_cur, 0, true,
 				   &mtr);
@@ -254,7 +253,7 @@ static MY_ATTRIBUTE((nonnull, warn_unused_result))
 dberr_t
 row_undo_ins_remove_sec_low(
 /*========================*/
-	ulint		mode,	/*!< in: BTR_MODIFY_LEAF or BTR_MODIFY_TREE,
+	btr_latch_mode	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 */
@@ -266,25 +265,27 @@ row_undo_ins_remove_sec_low(
 	mtr_t			mtr;
 	const bool		modify_leaf = mode == BTR_MODIFY_LEAF;
 
+	pcur.btr_cur.page_cur.index = index;
 	row_mtr_start(&mtr, index, !modify_leaf);
 
 	if (modify_leaf) {
-		mode = BTR_MODIFY_LEAF | BTR_ALREADY_S_LATCHED;
+		mode = BTR_MODIFY_LEAF_ALREADY_LATCHED;
 		mtr_s_lock_index(index, &mtr);
 	} else {
-		ut_ad(mode == (BTR_MODIFY_TREE | BTR_LATCH_FOR_DELETE));
+		ut_ad(mode == BTR_PURGE_TREE);
 		mtr_sx_lock_index(index, &mtr);
 	}
 
-	if (dict_index_is_spatial(index)) {
+	if (index->is_spatial()) {
-		if (modify_leaf) {
+		mode = modify_leaf
-			mode |= BTR_RTREE_DELETE_MARK;
+			? btr_latch_mode(BTR_MODIFY_LEAF_ALREADY_LATCHED
-		}
+					 | BTR_RTREE_DELETE_MARK
+					 | BTR_RTREE_UNDO_INS)
+			: btr_latch_mode(BTR_PURGE_TREE | BTR_RTREE_UNDO_INS);
 		btr_pcur_get_btr_cur(&pcur)->thr = thr;
-		mode |= BTR_RTREE_UNDO_INS;
 	}
 
-	switch (row_search_index_entry(index, entry, mode, &pcur, &mtr)) {
+	switch (row_search_index_entry(entry, mode, &pcur, &mtr)) {
 	case ROW_BUFFERED:
 	case ROW_NOT_DELETED_REF:
 		/* These are invalid outcomes, because the mode passed
@@ -349,9 +350,7 @@ row_undo_ins_remove_sec(
 
 	/* Try then pessimistic descent to the B-tree */
 retry:
-	err = row_undo_ins_remove_sec_low(
+	err = row_undo_ins_remove_sec_low(BTR_PURGE_TREE, index, entry, thr);
-		BTR_MODIFY_TREE | BTR_LATCH_FOR_DELETE,
-		index, entry, thr);
 
 	/* The delete operation may fail if we have little
 	file space left: TODO: easiest to crash the database

storage/innobase/row/row0umod.cc

@@ -84,7 +84,7 @@ row_undo_mod_clust_low(
 	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_latch_mode	mode)	/*!< in: BTR_MODIFY_LEAF or BTR_MODIFY_TREE */
 {
 	btr_pcur_t*	pcur;
 	btr_cur_t*	btr_cur;
@@ -106,8 +106,8 @@ row_undo_mod_clust_low(
 	      || node->update->info_bits == REC_INFO_METADATA_ALTER);
 
 	if (mode != BTR_MODIFY_TREE) {
-		ut_ad((mode & ulint(~BTR_ALREADY_S_LATCHED))
+		ut_ad(mode == BTR_MODIFY_LEAF
-		      == BTR_MODIFY_LEAF);
+		      || mode == BTR_MODIFY_LEAF_ALREADY_LATCHED);
 
 		err = btr_cur_optimistic_update(
 			BTR_NO_LOCKING_FLAG | BTR_NO_UNDO_LOG_FLAG
@@ -224,14 +224,14 @@ static bool row_undo_mod_must_purge(const undo_node_t &node)
   ut_ad(!node.table->is_temporary());
 
   const btr_cur_t &btr_cur= node.pcur.btr_cur;
-  ut_ad(btr_cur.index->is_primary());
+  ut_ad(btr_cur.index()->is_primary());
   DEBUG_SYNC_C("rollback_purge_clust");
 
   if (!purge_sys.is_purgeable(node.new_trx_id))
     return false;
 
   const rec_t *rec= btr_cur_get_rec(&btr_cur);
-  return trx_read_trx_id(rec + row_trx_id_offset(rec, btr_cur.index)) ==
+  return trx_read_trx_id(rec + row_trx_id_offset(rec, btr_cur.index())) ==
     node.new_trx_id;
 }
 
@@ -356,8 +356,7 @@ row_undo_mod_clust(
 		}
 
 		mtr.start();
-		if (pcur->restore_position(
+		if (pcur->restore_position(BTR_PURGE_TREE, &mtr) !=
-			    BTR_MODIFY_TREE | BTR_LATCH_FOR_DELETE, &mtr) !=
 		    btr_pcur_t::SAME_ALL) {
 			goto mtr_commit_exit;
 		}
@@ -483,7 +482,7 @@ row_undo_mod_del_mark_or_remove_sec_low(
 	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_latch_mode	mode)	/*!< in: latch mode BTR_MODIFY_LEAF or
 				BTR_MODIFY_TREE */
 {
 	btr_pcur_t		pcur;
@@ -496,15 +495,25 @@ row_undo_mod_del_mark_or_remove_sec_low(
 
 	row_mtr_start(&mtr, index, !modify_leaf);
 
-	if (!index->is_committed()) {
+	pcur.btr_cur.page_cur.index = index;
+	btr_cur = btr_pcur_get_btr_cur(&pcur);
+
+	if (index->is_spatial()) {
+		mode = modify_leaf
+			? btr_latch_mode(BTR_MODIFY_LEAF
+					 | BTR_RTREE_DELETE_MARK
+					 | BTR_RTREE_UNDO_INS)
+			: btr_latch_mode(BTR_PURGE_TREE | BTR_RTREE_UNDO_INS);
+		btr_cur->thr = thr;
+	} else if (!index->is_committed()) {
 		/* The index->online_status may change if the index is
 		or was being created online, but not committed yet. It
 		is protected by index->lock. */
 		if (modify_leaf) {
-			mode = BTR_MODIFY_LEAF | BTR_ALREADY_S_LATCHED;
+			mode = BTR_MODIFY_LEAF_ALREADY_LATCHED;
 			mtr_s_lock_index(index, &mtr);
 		} else {
-			ut_ad(mode == (BTR_MODIFY_TREE | BTR_LATCH_FOR_DELETE));
+			ut_ad(mode == BTR_PURGE_TREE);
 			mtr_sx_lock_index(index, &mtr);
 		}
 	} else {
@@ -514,18 +523,7 @@ row_undo_mod_del_mark_or_remove_sec_low(
 		ut_ad(!dict_index_is_online_ddl(index));
 	}
 
-	btr_cur = btr_pcur_get_btr_cur(&pcur);
+	search_result = row_search_index_entry(entry, mode, &pcur, &mtr);
-
-	if (dict_index_is_spatial(index)) {
-		if (modify_leaf) {
-			btr_cur->thr = thr;
-			mode |= BTR_RTREE_DELETE_MARK;
-		}
-		mode |= BTR_RTREE_UNDO_INS;
-	}
-
-	search_result = row_search_index_entry(index, entry, mode,
-					       &pcur, &mtr);
 
 	switch (UNIV_EXPECT(search_result, ROW_FOUND)) {
 	case ROW_NOT_FOUND:
@@ -634,7 +632,7 @@ row_undo_mod_del_mark_or_remove_sec(
 	}
 
 	err = row_undo_mod_del_mark_or_remove_sec_low(node, thr, index,
-		entry, BTR_MODIFY_TREE | BTR_LATCH_FOR_DELETE);
+		entry, BTR_PURGE_TREE);
 	return(err);
 }
 
@@ -652,7 +650,7 @@ static MY_ATTRIBUTE((nonnull, warn_unused_result))
 dberr_t
 row_undo_mod_del_unmark_sec_and_undo_update(
 /*========================================*/
-	ulint		mode,	/*!< in: search mode: BTR_MODIFY_LEAF or
+	btr_latch_mode	mode,	/*!< in: search mode: BTR_MODIFY_LEAF or
 				BTR_MODIFY_TREE */
 	que_thr_t*	thr,	/*!< in: query thread */
 	dict_index_t*	index,	/*!< in: index */
@@ -668,8 +666,9 @@ row_undo_mod_del_unmark_sec_and_undo_update(
 	const ulint		flags
 		= BTR_KEEP_SYS_FLAG | BTR_NO_LOCKING_FLAG;
 	row_search_result	search_result;
-	ulint			orig_mode = mode;
+	const auto		orig_mode = mode;
 
+	pcur.btr_cur.page_cur.index = index;
 	ut_ad(trx->id != 0);
 
 	if (dict_index_is_spatial(index)) {
@@ -679,7 +678,7 @@ row_undo_mod_del_unmark_sec_and_undo_update(
 		secondary index updates to avoid this. */
 		static_assert(BTR_MODIFY_TREE == (8 | BTR_MODIFY_LEAF), "");
 		ut_ad(!(mode & 8));
-		mode |= BTR_RTREE_DELETE_MARK;
+		mode = btr_latch_mode(mode | BTR_RTREE_DELETE_MARK);
 	}
 
 try_again:
@@ -687,8 +686,7 @@ try_again:
 
 	btr_cur->thr = thr;
 
-	search_result = row_search_index_entry(index, entry, mode,
+	search_result = row_search_index_entry(entry, mode, &pcur, &mtr);
-					       &pcur, &mtr);
 
 	switch (search_result) {
 		mem_heap_t*	heap;

storage/innobase/row/row0upd.cc

@@ -1840,7 +1840,7 @@ row_upd_sec_index_entry(
 	btr_cur_t*		btr_cur;
 	dberr_t			err	= DB_SUCCESS;
 	trx_t*			trx	= thr_get_trx(thr);
-	ulint			mode;
+	btr_latch_mode		mode;
 	ulint			flags;
 	enum row_search_result	search_result;
 
@@ -1870,14 +1870,16 @@ row_upd_sec_index_entry(
 			    "before_row_upd_sec_index_entry");
 
 	mtr.start();
+	mode = BTR_MODIFY_LEAF;
 
 	switch (index->table->space_id) {
 	case SRV_TMP_SPACE_ID:
 		mtr.set_log_mode(MTR_LOG_NO_REDO);
 		flags = BTR_NO_LOCKING_FLAG;
-		mode = index->is_spatial()
+		if (index->is_spatial()) {
-			? ulint(BTR_MODIFY_LEAF | BTR_RTREE_DELETE_MARK)
+			mode = btr_latch_mode(BTR_MODIFY_LEAF
-			: ulint(BTR_MODIFY_LEAF);
+					      | BTR_RTREE_DELETE_MARK);
+		}
 		break;
 	default:
 		index->set_modified(mtr);
@@ -1887,18 +1889,19 @@ row_upd_sec_index_entry(
 		/* We can only buffer delete-mark operations if there
 		are no foreign key constraints referring to the index. */
 		mode = index->is_spatial()
-			? ulint(BTR_MODIFY_LEAF | BTR_RTREE_DELETE_MARK)
+			? btr_latch_mode(BTR_MODIFY_LEAF
+					 | BTR_RTREE_DELETE_MARK)
 			: referenced
-			? ulint(BTR_MODIFY_LEAF) : ulint(BTR_DELETE_MARK_LEAF);
+			? BTR_MODIFY_LEAF : BTR_DELETE_MARK_LEAF;
 		break;
 	}
 
 	/* Set the query thread, so that ibuf_insert_low() will be
 	able to invoke thd_get_trx(). */
 	btr_pcur_get_btr_cur(&pcur)->thr = thr;
+	pcur.btr_cur.page_cur.index = index;
 
-	search_result = row_search_index_entry(index, entry, mode,
+	search_result = row_search_index_entry(entry, mode, &pcur, &mtr);
-					       &pcur, &mtr);
 
 	btr_cur = btr_pcur_get_btr_cur(&pcur);
 
@@ -2590,13 +2593,13 @@ row_upd_clust_step(
 
 	ut_a(pcur->rel_pos == BTR_PCUR_ON);
 
-	ulint	mode;
+	btr_latch_mode mode;
 
 	DEBUG_SYNC_C_IF_THD(trx->mysql_thd, "innodb_row_upd_clust_step_enter");
 
 	if (dict_index_is_online_ddl(index)) {
 		ut_ad(node->table->id != DICT_INDEXES_ID);
-		mode = BTR_MODIFY_LEAF | BTR_ALREADY_S_LATCHED;
+		mode = BTR_MODIFY_LEAF_ALREADY_LATCHED;
 		mtr_s_lock_index(index, &mtr);
 	} else {
 		mode = BTR_MODIFY_LEAF;

storage/innobase/srv/srv0srv.cc

@@ -392,9 +392,6 @@ mysql_mutex_t srv_misc_tmpfile_mutex;
 /** Temporary file for miscellanous diagnostic output */
 FILE*	srv_misc_tmpfile;
 
-static ulint	srv_main_thread_process_no;
-static ulint	srv_main_thread_id;
-
 /* The following counts are used by the srv_master_callback. */
 
 /** Iterations of the loop bounded by 'srv_active' label. */
@@ -887,12 +884,7 @@ srv_printf_innodb_monitor(
 			n_reserved);
 	}
 
-	fprintf(file,
+	fprintf(file, "Process ID=0, Main thread ID=0, state: %s\n",
-		"Process ID=" ULINTPF
-		", Main thread ID=" ULINTPF
-		", state: %s\n",
-		srv_main_thread_process_no,
-		srv_main_thread_id,
 		srv_main_thread_op_info);
 	fprintf(file,
 		"Number of rows inserted " ULINTPF
@@ -1456,30 +1448,28 @@ static void srv_sync_log_buffer_in_background()
 	}
 }
 
-/*********************************************************************//**
+/** Report progress during shutdown.
-This function prints progress message every 60 seconds during server
+@param last   time of last output
-shutdown, for any activities that master thread is pending on. */
+@param n_read number of page reads initiated for change buffer merge */
-static
+static void srv_shutdown_print(time_t &last, ulint n_read)
-void
-srv_shutdown_print_master_pending(
-/*==============================*/
-	time_t*		last_print_time,	/*!< last time the function
-						print the message */
-	ulint		n_bytes_merged)		/*!< number of change buffer
-						just merged */
 {
-	time_t current_time = time(NULL);
+  time_t now= time(nullptr);
+  if (now - last >= 15)
+  {
+    last= now;
 
-	if (difftime(current_time, *last_print_time) > 60) {
+    const ulint ibuf_size= ibuf.size;
-		*last_print_time = current_time;
+    sql_print_information("Completing change buffer merge;"
-
+                          " %zu page reads initiated;"
-		/* Check change buffer merge, we only wait for change buffer
+                          " %zu change buffer pages remain",
-		merge if it is a slow shutdown */
+                          n_read, ibuf_size);
-		if (!srv_fast_shutdown && n_bytes_merged) {
+#if defined HAVE_SYSTEMD && !defined EMBEDDED_LIBRARY
-			ib::info() << "Waiting for change buffer merge to"
+    service_manager_extend_timeout(INNODB_EXTEND_TIMEOUT_INTERVAL,
-				" complete number of bytes of change buffer"
+                                   "Completing change buffer merge;"
-				" just merged: " << n_bytes_merged;
+                                   " %zu page reads initiated;"
-		}
+                                   " %zu change buffer pages remain",
+                                   n_read, ibuf_size);
+#endif
   }
 }
 
@@ -1525,7 +1515,7 @@ Complete the shutdown tasks such as background DROP TABLE,
 and optionally change buffer merge (on innodb_fast_shutdown=0). */
 void srv_shutdown(bool ibuf_merge)
 {
-	ulint		n_bytes_merged	= 0;
+	ulint		n_read = 0;
 	time_t		now = time(NULL);
 
 	do {
@@ -1534,21 +1524,12 @@ void srv_shutdown(bool ibuf_merge)
 		++srv_main_shutdown_loops;
 
 		if (ibuf_merge) {
-			srv_main_thread_op_info = "checking free log space";
-			log_free_check();
 			srv_main_thread_op_info = "doing insert buffer merge";
-			n_bytes_merged = ibuf_merge_all();
+			log_free_check();
-
+			n_read = ibuf_contract();
-			/* Flush logs if needed */
+			srv_shutdown_print(now, n_read);
-			srv_sync_log_buffer_in_background();
 		}
-
+	} while (n_read);
-		/* Print progress message every 60 seconds during shutdown */
-		if (srv_print_verbose_log) {
-			srv_shutdown_print_master_pending(&now,
-							  n_bytes_merged);
-		}
-	} while (n_bytes_merged);
 }
 
 /** The periodic master task controlling the server. */

storage/innobase/trx/trx0trx.cc

@@ -594,7 +594,7 @@ static dberr_t trx_resurrect_table_locks(trx_t *trx, const trx_undo_t &undo)
 
       if (undo_block != block)
       {
-        mtr.memo_release(undo_block, MTR_MEMO_PAGE_S_FIX);
+        mtr.release(*undo_block);
         undo_block= block;
       }
       trx_undo_rec_get_pars(undo_rec, &type, &cmpl_info,