MDEV-6492: MariaDB Galera Cluster cant use rsync sst

A donor node does a flush tables and then tries to
freeze innodb writes before proceeding with SST.
However, innodb_disallow_writes was missing in xtradb.
Merged 'InnodbFreeze' patch from maria-5.5-galera's to
xtradb. Also, merged some changes missing in innobase's
os0file.cc.

Added a basic test case for innodb_disallow_writes system
variable.

mysql-test/include/mtr_warnings.sql

@@ -234,8 +234,6 @@ INSERT INTO global_suppressions VALUES
  ("WSREP: last inactive check more than .* skipping check"),
  ("WSREP: Gap in state sequence. Need state transfer."),
  ("WSREP: Failed to prepare for incremental state transfer: .*"),
- ("WSREP: error executing 'SET GLOBAL innodb_disallow_writes=.*"),
- ("WSREP: Failed to disallow InnoDB writes"),
  ("THE_LAST_SUPPRESSION")||
 
 

mysql-test/suite/sys_vars/r/innodb_disallow_writes_basic.result

@@ -0,0 +1,45 @@
+#
+# innodb_disallow_writes
+#
+# save the initial value
+SET @innodb_disallow_writes_global_saved = @@global.innodb_disallow_writes;
+# default
+SELECT @@global.innodb_disallow_writes;
+@@global.innodb_disallow_writes
+0
+
+# scope
+SELECT @@session.innodb_disallow_writes;
+ERROR HY000: Variable 'innodb_disallow_writes' is a GLOBAL variable
+SET @@global.innodb_disallow_writes=OFF;
+SELECT @@global.innodb_disallow_writes;
+@@global.innodb_disallow_writes
+0
+SET @@global.innodb_disallow_writes=ON;
+SELECT @@global.innodb_disallow_writes;
+@@global.innodb_disallow_writes
+1
+
+# valid values
+SET @@global.innodb_disallow_writes='OFF';
+SELECT @@global.innodb_disallow_writes;
+@@global.innodb_disallow_writes
+0
+SET @@global.innodb_disallow_writes=ON;
+SELECT @@global.innodb_disallow_writes;
+@@global.innodb_disallow_writes
+1
+SET @@global.innodb_disallow_writes=default;
+SELECT @@global.innodb_disallow_writes;
+@@global.innodb_disallow_writes
+0
+
+# invalid values
+SET @@global.innodb_disallow_writes=NULL;
+ERROR 42000: Variable 'innodb_disallow_writes' can't be set to the value of 'NULL'
+SET @@global.innodb_disallow_writes='junk';
+ERROR 42000: Variable 'innodb_disallow_writes' can't be set to the value of 'junk'
+
+# restore the initial value
+SET @@global.innodb_disallow_writes = @innodb_disallow_writes_global_saved;
+# End of test

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

@@ -0,0 +1,42 @@
+--source include/have_innodb_disallow_writes.inc
+
+--echo #
+--echo # innodb_disallow_writes
+--echo #
+
+--echo # save the initial value
+SET @innodb_disallow_writes_global_saved = @@global.innodb_disallow_writes;
+
+--echo # default
+SELECT @@global.innodb_disallow_writes;
+
+--echo
+--echo # scope
+--error ER_INCORRECT_GLOBAL_LOCAL_VAR
+SELECT @@session.innodb_disallow_writes;
+SET @@global.innodb_disallow_writes=OFF;
+SELECT @@global.innodb_disallow_writes;
+SET @@global.innodb_disallow_writes=ON;
+SELECT @@global.innodb_disallow_writes;
+
+--echo
+--echo # valid values
+SET @@global.innodb_disallow_writes='OFF';
+SELECT @@global.innodb_disallow_writes;
+SET @@global.innodb_disallow_writes=ON;
+SELECT @@global.innodb_disallow_writes;
+SET @@global.innodb_disallow_writes=default;
+SELECT @@global.innodb_disallow_writes;
+
+--echo
+--echo # invalid values
+--error ER_WRONG_VALUE_FOR_VAR
+SET @@global.innodb_disallow_writes=NULL;
+--error ER_WRONG_VALUE_FOR_VAR
+SET @@global.innodb_disallow_writes='junk';
+
+--echo
+--echo # restore the initial value
+SET @@global.innodb_disallow_writes = @innodb_disallow_writes_global_saved;
+
+--echo # End of test

storage/innobase/os/os0file.cc

@@ -87,6 +87,12 @@ UNIV_INTERN os_ib_mutex_t	os_file_seek_mutexes[OS_FILE_N_SEEK_MUTEXES];
 /* In simulated aio, merge at most this many consecutive i/os */
 #define OS_AIO_MERGE_N_CONSECUTIVE	64
 
+#ifdef WITH_INNODB_DISALLOW_WRITES
+#define WAIT_ALLOW_WRITES() os_event_wait(srv_allow_writes_event)
+#else
+#define WAIT_ALLOW_WRITES() do { } while (0)
+#endif /* WITH_INNODB_DISALLOW_WRITES */
+
 /**********************************************************************
 
 InnoDB AIO Implementation:
@@ -762,7 +768,9 @@ os_file_create_tmpfile(void)
 /*========================*/
 {
 	FILE*	file	= NULL;
-	int	fd	= innobase_mysql_tmpfile();
+	int	fd;
+	WAIT_ALLOW_WRITES();
+	fd	= innobase_mysql_tmpfile();
 
 	ut_ad(!srv_read_only_mode);
 
@@ -1088,6 +1096,7 @@ os_file_create_directory(
 	return(TRUE);
 #else
 	int	rcode;
+	WAIT_ALLOW_WRITES();
 
 	rcode = mkdir(pathname, 0770);
 
@@ -1214,6 +1223,8 @@ os_file_create_simple_func(
 
 #else /* __WIN__ */
 	int		create_flag;
+	if (create_mode != OS_FILE_OPEN && create_mode != OS_FILE_OPEN_RAW)
+		WAIT_ALLOW_WRITES();
 
 	ut_a(!(create_mode & OS_FILE_ON_ERROR_SILENT));
 	ut_a(!(create_mode & OS_FILE_ON_ERROR_NO_EXIT));
@@ -1381,6 +1392,8 @@ os_file_create_simple_no_error_handling_func(
 	int		create_flag;
 
 	ut_a(name);
+	if (create_mode != OS_FILE_OPEN && create_mode != OS_FILE_OPEN_RAW)
+		WAIT_ALLOW_WRITES();
 
 	ut_a(!(create_mode & OS_FILE_ON_ERROR_SILENT));
 	ut_a(!(create_mode & OS_FILE_ON_ERROR_NO_EXIT));
@@ -1672,6 +1685,8 @@ os_file_create_func(
 #else /* __WIN__ */
 	int		create_flag;
 	const char*	mode_str	= NULL;
+	if (create_mode != OS_FILE_OPEN && create_mode != OS_FILE_OPEN_RAW)
+		WAIT_ALLOW_WRITES();
 
 	on_error_no_exit = create_mode & OS_FILE_ON_ERROR_NO_EXIT
 		? TRUE : FALSE;
@@ -1851,6 +1866,7 @@ loop:
 	goto loop;
 #else
 	int	ret;
+	WAIT_ALLOW_WRITES();
 
 	ret = unlink(name);
 
@@ -1915,6 +1931,7 @@ loop:
 	goto loop;
 #else
 	int	ret;
+	WAIT_ALLOW_WRITES();
 
 	ret = unlink(name);
 
@@ -1968,6 +1985,7 @@ os_file_rename_func(
 	return(FALSE);
 #else
 	int	ret;
+	WAIT_ALLOW_WRITES();
 
 	ret = rename(oldpath, newpath);
 
@@ -2196,6 +2214,7 @@ os_file_set_eof(
 	HANDLE h = (HANDLE) _get_osfhandle(fileno(file));
 	return(SetEndOfFile(h));
 #else /* __WIN__ */
+	WAIT_ALLOW_WRITES();
 	return(!ftruncate(fileno(file), ftell(file)));
 #endif /* __WIN__ */
 }
@@ -2290,6 +2309,7 @@ os_file_flush_func(
 	return(FALSE);
 #else
 	int	ret;
+	WAIT_ALLOW_WRITES();
 
 #if defined(HAVE_DARWIN_THREADS)
 # ifndef F_FULLFSYNC
@@ -2971,6 +2991,7 @@ retry:
 	return(FALSE);
 #else
 	ssize_t	ret;
+	WAIT_ALLOW_WRITES();
 
 	ret = os_file_pwrite(file, buf, n, offset);
 

storage/xtradb/handler/ha_innodb.cc

@@ -2135,6 +2135,9 @@ int
 innobase_mysql_tmpfile(void)
 /*========================*/
 {
+#ifdef WITH_INNODB_DISALLOW_WRITES
+	os_event_wait(srv_allow_writes_event);
+#endif /* WITH_INNODB_DISALLOW_WRITES */
 	int	fd2 = -1;
 	File	fd;
 
@@ -19194,6 +19197,40 @@ static MYSQL_SYSVAR_BOOL(disable_background_merge,
   NULL, NULL, FALSE);
 #endif /* UNIV_DEBUG || UNIV_IBUF_DEBUG */
 
+#ifdef WITH_INNODB_DISALLOW_WRITES
+/*******************************************************
+ *    innobase_disallow_writes variable definition     *
+ *******************************************************/
+ 
+/* Must always init to FALSE. */
+static my_bool	innobase_disallow_writes	= FALSE;
+
+/**************************************************************************
+An "update" method for innobase_disallow_writes variable. */
+static
+void
+innobase_disallow_writes_update(
+/*============================*/
+	THD*			thd,		/* in: thread handle */
+	st_mysql_sys_var*	var,		/* in: pointer to system
+						variable */
+	void*			var_ptr,	/* out: pointer to dynamic
+						variable */
+	const void*		save)		/* in: temporary storage */
+{
+	*(my_bool*)var_ptr = *(my_bool*)save;
+	ut_a(srv_allow_writes_event);
+	if (*(my_bool*)var_ptr)
+		os_event_reset(srv_allow_writes_event);
+	else
+		os_event_set(srv_allow_writes_event);
+}
+
+static MYSQL_SYSVAR_BOOL(disallow_writes, innobase_disallow_writes,
+  PLUGIN_VAR_NOCMDOPT,
+  "Tell InnoDB to stop any writes to disk",
+  NULL, innobase_disallow_writes_update, FALSE);
+#endif /* WITH_INNODB_DISALLOW_WRITES */
 static MYSQL_SYSVAR_BOOL(random_read_ahead, srv_random_read_ahead,
   PLUGIN_VAR_NOCMDARG,
   "Whether to use read ahead for random access within an extent.",
@@ -19458,6 +19495,9 @@ static struct st_mysql_sys_var* innobase_system_variables[]= {
   MYSQL_SYSVAR(change_buffering_debug),
   MYSQL_SYSVAR(disable_background_merge),
 #endif /* UNIV_DEBUG || UNIV_IBUF_DEBUG */
+#ifdef WITH_INNODB_DISALLOW_WRITES
+  MYSQL_SYSVAR(disallow_writes),
+#endif /* WITH_INNODB_DISALLOW_WRITES */
   MYSQL_SYSVAR(random_read_ahead),
   MYSQL_SYSVAR(read_ahead_threshold),
   MYSQL_SYSVAR(read_only),

storage/xtradb/include/srv0srv.h

@@ -279,6 +279,10 @@ extern uint	srv_flush_log_at_timeout;
 extern char	srv_use_global_flush_log_at_trx_commit;
 extern char	srv_adaptive_flushing;
 
+#ifdef WITH_INNODB_DISALLOW_WRITES
+/* When this event is reset we do not allow any file writes to take place. */
+extern os_event_t	srv_allow_writes_event;
+#endif /* WITH_INNODB_DISALLOW_WRITES */
 /* If this flag is TRUE, then we will load the indexes' (and tables') metadata
 even if they are marked as "corrupted". Mostly it is for DBA to process
 corrupted index and table */

storage/xtradb/os/os0file.cc

@@ -100,6 +100,12 @@ UNIV_INTERN os_ib_mutex_t	os_file_seek_mutexes[OS_FILE_N_SEEK_MUTEXES];
 /* In simulated aio, merge at most this many consecutive i/os */
 #define OS_AIO_MERGE_N_CONSECUTIVE	64
 
+#ifdef WITH_INNODB_DISALLOW_WRITES
+#define WAIT_ALLOW_WRITES() os_event_wait(srv_allow_writes_event)
+#else
+#define WAIT_ALLOW_WRITES() do { } while (0)
+#endif /* WITH_INNODB_DISALLOW_WRITES */
+
 /**********************************************************************
 
 InnoDB AIO Implementation:
@@ -863,7 +869,9 @@ os_file_create_tmpfile(void)
 /*========================*/
 {
 	FILE*	file	= NULL;
-	int	fd	= innobase_mysql_tmpfile();
+	int	fd;
+	WAIT_ALLOW_WRITES();
+	fd	= innobase_mysql_tmpfile();
 
 	ut_ad(!srv_read_only_mode);
 
@@ -1189,6 +1197,7 @@ os_file_create_directory(
 	return(TRUE);
 #else
 	int	rcode;
+	WAIT_ALLOW_WRITES();
 
 	rcode = mkdir(pathname, 0770);
 
@@ -1315,6 +1324,8 @@ os_file_create_simple_func(
 
 #else /* __WIN__ */
 	int		create_flag;
+	if (create_mode != OS_FILE_OPEN && create_mode != OS_FILE_OPEN_RAW)
+		WAIT_ALLOW_WRITES();
 
 	ut_a(!(create_mode & OS_FILE_ON_ERROR_SILENT));
 	ut_a(!(create_mode & OS_FILE_ON_ERROR_NO_EXIT));
@@ -1482,6 +1493,8 @@ os_file_create_simple_no_error_handling_func(
 	int		create_flag;
 
 	ut_a(name);
+	if (create_mode != OS_FILE_OPEN && create_mode != OS_FILE_OPEN_RAW)
+		WAIT_ALLOW_WRITES();
 
 	ut_a(!(create_mode & OS_FILE_ON_ERROR_SILENT));
 	ut_a(!(create_mode & OS_FILE_ON_ERROR_NO_EXIT));
@@ -1828,6 +1841,8 @@ os_file_create_func(
 #else /* __WIN__ */
 	int		create_flag;
 	const char*	mode_str	= NULL;
+	if (create_mode != OS_FILE_OPEN && create_mode != OS_FILE_OPEN_RAW)
+		WAIT_ALLOW_WRITES();
 
 	on_error_no_exit = create_mode & OS_FILE_ON_ERROR_NO_EXIT
 		? TRUE : FALSE;
@@ -2017,6 +2032,7 @@ loop:
 	goto loop;
 #else
 	int	ret;
+	WAIT_ALLOW_WRITES();
 
 	ret = unlink(name);
 
@@ -2081,6 +2097,7 @@ loop:
 	goto loop;
 #else
 	int	ret;
+	WAIT_ALLOW_WRITES();
 
 	ret = unlink(name);
 
@@ -2134,6 +2151,7 @@ os_file_rename_func(
 	return(FALSE);
 #else
 	int	ret;
+	WAIT_ALLOW_WRITES();
 
 	ret = rename(oldpath, newpath);
 
@@ -2360,6 +2378,7 @@ os_file_set_eof(
 	HANDLE h = (HANDLE) _get_osfhandle(fileno(file));
 	return(SetEndOfFile(h));
 #else /* __WIN__ */
+	WAIT_ALLOW_WRITES();
 	return(!ftruncate(fileno(file), ftell(file)));
 #endif /* __WIN__ */
 }
@@ -2379,6 +2398,7 @@ os_file_set_eof_at(
 	return(SetFilePointerEx(file, li, &li2,FILE_BEGIN)
 	       && SetEndOfFile(file));
 #else
+	WAIT_ALLOW_WRITES();
 	/* TODO: works only with -D_FILE_OFFSET_BITS=64 ? */
 	return(!ftruncate(file, new_len));
 #endif
@@ -2478,6 +2498,7 @@ os_file_flush_func(
 	return(FALSE);
 #else
 	int	ret;
+	WAIT_ALLOW_WRITES();
 
 #if defined(HAVE_DARWIN_THREADS)
 # ifndef F_FULLFSYNC
@@ -3142,6 +3163,7 @@ retry:
 	return(FALSE);
 #else
 	ssize_t	ret;
+	WAIT_ALLOW_WRITES();
 
 	ret = os_file_pwrite(file, buf, n, offset);
 

storage/xtradb/srv/srv0srv.cc

@@ -245,6 +245,10 @@ srv_printf_innodb_monitor() will request mutex acquisition
 with mutex_enter(), which will wait until it gets the mutex. */
 #define MUTEX_NOWAIT(mutex_skipped)	((mutex_skipped) < MAX_MUTEX_NOWAIT)
 
+#ifdef WITH_INNODB_DISALLOW_WRITES
+UNIV_INTERN os_event_t	srv_allow_writes_event;
+#endif /* WITH_INNODB_DISALLOW_WRITES */
+
 /** The sort order table of the MySQL latin1_swedish_ci character set
 collation */
 UNIV_INTERN const byte*	srv_latin1_ordering;
@@ -1132,6 +1136,14 @@ srv_init(void)
 	dict_ind_init();
 
 	srv_conc_init();
+#ifdef WITH_INNODB_DISALLOW_WRITES
+	/* Writes have to be enabled on init or else we hang. Thus, we
+	always set the event here regardless of innobase_disallow_writes.
+	That flag will always be 0 at this point because it isn't settable
+	via my.cnf or command line arg. */
+	srv_allow_writes_event = os_event_create();
+	os_event_set(srv_allow_writes_event);
+#endif /* WITH_INNODB_DISALLOW_WRITES */
 
 	/* Initialize some INFORMATION SCHEMA internal structures */
 	trx_i_s_cache_init(trx_i_s_cache);
@@ -2130,7 +2142,20 @@ loop:
 
 	if (sync_array_print_long_waits(&waiter, &sema)
 	    && sema == old_sema && os_thread_eq(waiter, old_waiter)) {
+#if defined(WITH_WSREP) && defined(WITH_INNODB_DISALLOW_WRITES)
+	  if (srv_allow_writes_event->is_set) {
+#endif /* WITH_WSREP */
 		fatal_cnt++;
+#if defined(WITH_WSREP) && defined(WITH_INNODB_DISALLOW_WRITES)
+	  } else {
+		fprintf(stderr,
+			"WSREP: avoiding InnoDB self crash due to long "
+			"semaphore wait of  > %lu seconds\n"
+			"Server is processing SST donor operation, "
+			"fatal_cnt now: %lu",
+			(ulong) srv_fatal_semaphore_wait_threshold, fatal_cnt);
+	  }
+#endif /* WITH_WSREP */
 		if (fatal_cnt > 10) {
 
 			fprintf(stderr,