Merge XtraDB from Percona-Server-5.1.47-11 into MariaDB.

2 files changed, 360 insertions, 138 deletions

diff --git a/storage/xtradb/srv/srv0srv.c b/storage/xtradb/srv/srv0srv.c
index e655c4e844d..86bf309bac1 100644
--- a/storage/xtradb/srv/srv0srv.c
+++ b/storage/xtradb/srv/srv0srv.c
@@ -1,7 +1,8 @@
 /*****************************************************************************
 
-Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+Copyright (c) 1995, 2010, Innobase Oy. All Rights Reserved.
 Copyright (c) 2008, 2009 Google Inc.
+Copyright (c) 2009, Percona Inc.
 
 Portions of this file contain modifications contributed and copyrighted by
 Google, Inc. Those modifications are gratefully acknowledged and are described
@@ -9,6 +10,13 @@ briefly in the InnoDB documentation. The contributions by Google are
 incorporated with their permission, and subject to the conditions contained in
 the file COPYING.Google.
 
+Portions of this file contain modifications contributed and copyrighted
+by Percona Inc.. Those modifications are
+gratefully acknowledged and are described briefly in the InnoDB
+documentation. The contributions by Percona Inc. are incorporated with
+their permission, and subject to the conditions contained in the file
+COPYING.Percona.
+
 This program is free software; you can redistribute it and/or modify it under
 the terms of the GNU General Public License as published by the Free Software
 Foundation; version 2 of the License.
@@ -22,32 +30,6 @@ this program; if not, write to the Free Software Foundation, Inc., 59 Temple
 Place, Suite 330, Boston, MA 02111-1307 USA
 
 *****************************************************************************/
-/***********************************************************************
-
-Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
-Copyright (c) 2009, Percona Inc.
-
-Portions of this file contain modifications contributed and copyrighted
-by Percona Inc.. Those modifications are
-gratefully acknowledged and are described briefly in the InnoDB
-documentation. The contributions by Percona Inc. are incorporated with
-their permission, and subject to the conditions contained in the file
-COPYING.Percona.
-
-This program is free software; you can redistribute it and/or modify it
-under the terms of the GNU General Public License as published by the
-Free Software Foundation; version 2 of the License.
-
-This program is distributed in the hope that it will be useful, but
-WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General
-Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
-
-***********************************************************************/
 
 /**************************************************//**
 @file srv/srv0srv.c
@@ -122,7 +104,8 @@ UNIV_INTERN ulint	srv_fatal_semaphore_wait_threshold = 600;
 in microseconds, in order to reduce the lagging of the purge thread. */
 UNIV_INTERN ulint	srv_dml_needed_delay = 0;
 
-UNIV_INTERN ibool	srv_lock_timeout_and_monitor_active = FALSE;
+UNIV_INTERN ibool	srv_lock_timeout_active = FALSE;
+UNIV_INTERN ibool	srv_monitor_active = FALSE;
 UNIV_INTERN ibool	srv_error_monitor_active = FALSE;
 
 UNIV_INTERN const char*	srv_main_thread_op_info = "";
@@ -162,9 +145,10 @@ UNIV_INTERN char**	srv_data_file_names = NULL;
 /* size in database pages */
 UNIV_INTERN ulint*	srv_data_file_sizes = NULL;
 
+UNIV_INTERN char*	srv_doublewrite_file = NULL;
+
 UNIV_INTERN ibool	srv_extra_undoslots = FALSE;
 
-UNIV_INTERN ibool	srv_fast_recovery = FALSE;
 UNIV_INTERN ibool	srv_recovery_stats = FALSE;
 
 UNIV_INTERN ulint	srv_use_purge_thread = 0;
@@ -198,11 +182,20 @@ UNIV_INTERN ulong	srv_flush_log_at_trx_commit = 1;
 the checkpoints. */
 UNIV_INTERN char	srv_adaptive_flushing	= TRUE;
 
-UNIV_INTERN ulong  srv_show_locks_held     = 10;
-UNIV_INTERN ulong  srv_show_verbose_locks  = 0;
+UNIV_INTERN ulong	srv_show_locks_held	= 10;
+UNIV_INTERN ulong	srv_show_verbose_locks	= 0;
 
+/** Maximum number of times allowed to conditionally acquire
+mutex before switching to blocking wait on the mutex */
+#define MAX_MUTEX_NOWAIT	20
 
-/* The sort order table of the MySQL latin1_swedish_ci character set
+/** Check whether the number of failed nonblocking mutex
+acquisition attempts exceeds maximum allowed value. If so,
+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)
+
+/** The sort order table of the MySQL latin1_swedish_ci character set
 collation */
 UNIV_INTERN const byte*	srv_latin1_ordering;
 
@@ -410,7 +403,6 @@ UNIV_INTERN ulong	srv_read_ahead = 3; /* 1: random  2: linear  3: Both */
 UNIV_INTERN ulong	srv_adaptive_checkpoint = 0; /* 0: none  1: reflex  2: estimate */
 
 UNIV_INTERN ulong	srv_expand_import = 0; /* 0:disable 1:enable */
-UNIV_INTERN ulint	srv_relax_table_creation = 0; /* 0:disable 1:enable */
 UNIV_INTERN ulint	srv_pass_corrupt_table = 0; /* 0:disable 1:enable */
 
 UNIV_INTERN ulong	srv_extra_rsegments = 0; /* extra rseg for users */
@@ -439,7 +431,7 @@ static ulint	srv_n_rows_inserted_old		= 0;
 static ulint	srv_n_rows_updated_old		= 0;
 static ulint	srv_n_rows_deleted_old		= 0;
 static ulint	srv_n_rows_read_old		= 0;
-
+UNIV_INTERN ulint               srv_n_lock_deadlock_count       = 0;
 UNIV_INTERN ulint		srv_n_lock_wait_count		= 0;
 UNIV_INTERN ulint		srv_n_lock_wait_current_count	= 0;
 UNIV_INTERN ib_int64_t	srv_n_lock_wait_time		= 0;
@@ -1766,8 +1758,9 @@ srv_suspend_mysql_thread(
 	innodb_lock_wait_timeout, because trx->mysql_thd == NULL. */
 	lock_wait_timeout = thd_lock_wait_timeout(trx->mysql_thd);
 
-	if (lock_wait_timeout < 100000000
-	    && wait_time > (double) lock_wait_timeout) {
+	if (trx_is_interrupted(trx)
+	    || (lock_wait_timeout < 100000000
+		&& wait_time > (double) lock_wait_timeout)) {
 
 		trx->error_state = DB_LOCK_WAIT_TIMEOUT;
 	}
@@ -1833,12 +1826,15 @@ srv_refresh_innodb_monitor_stats(void)
 }
 
 /******************************************************************//**
-Outputs to a file the output of the InnoDB Monitor. */
+Outputs to a file the output of the InnoDB Monitor.
+@return FALSE if not all information printed
+due to failure to obtain necessary mutex */
 UNIV_INTERN
-void
+ibool
 srv_printf_innodb_monitor(
 /*======================*/
 	FILE*	file,		/*!< in: output stream */
+	ibool	nowait,		/*!< in: whether to wait for kernel mutex */
 	ulint*	trx_start,	/*!< out: file position of the start of
 				the list of active transactions */
 	ulint*	trx_end)	/*!< out: file position of the end of
@@ -1847,6 +1843,7 @@ srv_printf_innodb_monitor(
 	double	time_elapsed;
 	time_t	current_time;
 	ulint	n_reserved;
+	ibool	ret;
 
 	ulint	btr_search_sys_subtotal;
 	ulint	lock_sys_subtotal;
@@ -1877,9 +1874,9 @@ srv_printf_innodb_monitor(
 		"Per second averages calculated from the last %lu seconds\n",
 		(ulong)time_elapsed);
 
-	fputs("----------\n"
-		"BACKGROUND THREAD\n"
-		"----------\n", file);
+	fputs("-----------------\n"
+	      "BACKGROUND THREAD\n"
+	      "-----------------\n", file);
 	srv_print_master_thread_info(file);
 
 	fputs("----------\n"
@@ -2069,22 +2066,28 @@ srv_printf_innodb_monitor(
 	srv_n_rows_deleted_old = srv_n_rows_deleted;
 	srv_n_rows_read_old = srv_n_rows_read;
 
-	lock_print_info_summary(file);
-	if (trx_start) {
-		long	t = ftell(file);
-		if (t < 0) {
-			*trx_start = ULINT_UNDEFINED;
-		} else {
-			*trx_start = (ulint) t;
+	/* Only if lock_print_info_summary proceeds correctly,
+	before we call the lock_print_info_all_transactions
+	to print all the lock information. */
+	ret = lock_print_info_summary(file, nowait);
+
+	if (ret) {
+		if (trx_start) {
+			long	t = ftell(file);
+			if (t < 0) {
+				*trx_start = ULINT_UNDEFINED;
+			} else {
+				*trx_start = (ulint) t;
+			}
 		}
-	}
-	lock_print_info_all_transactions(file);
-	if (trx_end) {
-		long	t = ftell(file);
-		if (t < 0) {
-			*trx_end = ULINT_UNDEFINED;
-		} else {
-			*trx_end = (ulint) t;
+		lock_print_info_all_transactions(file);
+		if (trx_end) {
+			long	t = ftell(file);
+			if (t < 0) {
+				*trx_end = ULINT_UNDEFINED;
+			} else {
+				*trx_end = (ulint) t;
+			}
 		}
 	}
 
@@ -2093,6 +2096,8 @@ srv_printf_innodb_monitor(
 	      "============================\n", file);
 	mutex_exit(&srv_innodb_monitor_mutex);
 	fflush(file);
+
+	return(ret);
 }
 
 /******************************************************************//**
@@ -2133,6 +2138,8 @@ srv_export_innodb_status(void)
 		= UT_LIST_GET_LEN(buf_pool->flush_list);
 	export_vars.innodb_buffer_pool_pages_free
 		= UT_LIST_GET_LEN(buf_pool->free);
+	export_vars.innodb_deadlocks
+	        = srv_n_lock_deadlock_count;
 #ifdef UNIV_DEBUG
 	export_vars.innodb_buffer_pool_pages_latched
 		= buf_get_latched_pages_number();
@@ -2181,26 +2188,23 @@ srv_export_innodb_status(void)
 }
 
 /*********************************************************************//**
-A thread which wakes up threads whose lock wait may have lasted too long.
-This also prints the info output by various InnoDB monitors.
+A thread which prints the info output by various InnoDB monitors.
 @return	a dummy parameter */
 UNIV_INTERN
 os_thread_ret_t
-srv_lock_timeout_and_monitor_thread(
-/*================================*/
+srv_monitor_thread(
+/*===============*/
 	void*	arg __attribute__((unused)))
 			/*!< in: a dummy parameter required by
 			os_thread_create */
 {
-	srv_slot_t*	slot;
 	double		time_elapsed;
 	time_t		current_time;
 	time_t		last_table_monitor_time;
 	time_t		last_tablespace_monitor_time;
 	time_t		last_monitor_time;
-	ibool		some_waits;
-	double		wait_time;
-	ulint		i;
+	ulint		mutex_skipped;
+	ibool		last_srv_print_monitor;
 
 #ifdef UNIV_DEBUG_THREAD_CREATION
 	fprintf(stderr, "Lock timeout thread starts, id %lu\n",
@@ -2211,13 +2215,15 @@ srv_lock_timeout_and_monitor_thread(
 	last_table_monitor_time = time(NULL);
 	last_tablespace_monitor_time = time(NULL);
 	last_monitor_time = time(NULL);
+	mutex_skipped = 0;
+	last_srv_print_monitor = srv_print_innodb_monitor;
 loop:
-	srv_lock_timeout_and_monitor_active = TRUE;
+	srv_monitor_active = TRUE;
 
-	/* When someone is waiting for a lock, we wake up every second
-	and check if a timeout has passed for a lock wait */
+	/* Wake up every 5 seconds to see if we need to print
+	monitor information. */
 
-	os_thread_sleep(1000000);
+	os_thread_sleep(5000000);
 
 	current_time = time(NULL);
 
@@ -2227,14 +2233,40 @@ loop:
 		last_monitor_time = time(NULL);
 
 		if (srv_print_innodb_monitor) {
-			srv_printf_innodb_monitor(stderr, NULL, NULL);
+			/* Reset mutex_skipped counter everytime
+			srv_print_innodb_monitor changes. This is to
+			ensure we will not be blocked by kernel_mutex
+			for short duration information printing,
+			such as requested by sync_array_print_long_waits() */
+			if (!last_srv_print_monitor) {
+				mutex_skipped = 0;
+				last_srv_print_monitor = TRUE;
+			}
+
+			if (!srv_printf_innodb_monitor(stderr,
+						MUTEX_NOWAIT(mutex_skipped),
+						NULL, NULL)) {
+				mutex_skipped++;
+			} else {
+				/* Reset the counter */
+				mutex_skipped = 0;
+			}
+		} else {
+			last_srv_print_monitor = FALSE;
 		}
 
+
 		if (srv_innodb_status) {
 			mutex_enter(&srv_monitor_file_mutex);
 			rewind(srv_monitor_file);
-			srv_printf_innodb_monitor(srv_monitor_file, NULL,
-						  NULL);
+			if (!srv_printf_innodb_monitor(srv_monitor_file,
+						MUTEX_NOWAIT(mutex_skipped),
+						NULL, NULL)) {
+				mutex_skipped++;
+			} else {
+				mutex_skipped = 0;
+			}
+
 			os_file_set_eof(srv_monitor_file);
 			mutex_exit(&srv_monitor_file_mutex);
 		}
@@ -2287,6 +2319,56 @@ loop:
 		}
 	}
 
+	if (srv_shutdown_state >= SRV_SHUTDOWN_CLEANUP) {
+		goto exit_func;
+	}
+
+	if (srv_print_innodb_monitor
+	    || srv_print_innodb_lock_monitor
+	    || srv_print_innodb_tablespace_monitor
+	    || srv_print_innodb_table_monitor) {
+		goto loop;
+	}
+
+	srv_monitor_active = FALSE;
+
+	goto loop;
+
+exit_func:
+	srv_monitor_active = FALSE;
+
+	/* We count the number of threads in os_thread_exit(). A created
+	thread should always use that to exit and not use return() to exit. */
+
+	os_thread_exit(NULL);
+
+	OS_THREAD_DUMMY_RETURN;
+}
+
+/*********************************************************************//**
+A thread which wakes up threads whose lock wait may have lasted too long.
+@return	a dummy parameter */
+UNIV_INTERN
+os_thread_ret_t
+srv_lock_timeout_thread(
+/*====================*/
+	void*	arg __attribute__((unused)))
+			/* in: a dummy parameter required by
+			os_thread_create */
+{
+	srv_slot_t*	slot;
+	ibool		some_waits;
+	double		wait_time;
+	ulint		i;
+
+loop:
+	/* When someone is waiting for a lock, we wake up every second
+	and check if a timeout has passed for a lock wait */
+
+	os_thread_sleep(1000000);
+
+	srv_lock_timeout_active = TRUE;
+
 	mutex_enter(&kernel_mutex);
 
 	some_waits = FALSE;
@@ -2310,9 +2392,10 @@ loop:
 			lock_wait_timeout = thd_lock_wait_timeout(
 				trx->mysql_thd);
 
-			if (lock_wait_timeout < 100000000
-			    && (wait_time > (double) lock_wait_timeout
-				|| wait_time < 0)) {
+			if (trx_is_interrupted(trx)
+			    || (lock_wait_timeout < 100000000
+				&& (wait_time > (double) lock_wait_timeout
+				    || wait_time < 0))) {
 
 				/* Timeout exceeded or a wrap-around in system
 				time counter: cancel the lock request queued
@@ -2337,17 +2420,11 @@ loop:
 		goto exit_func;
 	}
 
-	if (some_waits || srv_print_innodb_monitor
-	    || srv_print_innodb_lock_monitor
-	    || srv_print_innodb_tablespace_monitor
-	    || srv_print_innodb_table_monitor) {
+	if (some_waits) {
 		goto loop;
 	}
 
-	/* No one was waiting for a lock and no monitor was active:
-	suspend this thread */
-
-	srv_lock_timeout_and_monitor_active = FALSE;
+	srv_lock_timeout_active = FALSE;
 
 #if 0
 	/* The following synchronisation is disabled, since
@@ -2357,7 +2434,7 @@ loop:
 	goto loop;
 
 exit_func:
-	srv_lock_timeout_and_monitor_active = FALSE;
+	srv_lock_timeout_active = FALSE;
 
 	/* We count the number of threads in os_thread_exit(). A created
 	thread should always use that to exit and not use return() to exit. */
@@ -2706,7 +2783,10 @@ loop:
 						BUF_FLUSH_LIST,
 						n_flush,
 						IB_ULONGLONG_MAX);
-				skip_sleep = TRUE;
+
+				if (n_flush == PCT_IO(100)) {
+					skip_sleep = TRUE;
+				}
 			}
 
 			mutex_enter(&(log_sys->mutex));
diff --git a/storage/xtradb/srv/srv0start.c b/storage/xtradb/srv/srv0start.c
index dcc13ea17b6..7b5581a24f0 100644
--- a/storage/xtradb/srv/srv0start.c
+++ b/storage/xtradb/srv/srv0start.c
@@ -1,7 +1,8 @@
 /*****************************************************************************
 
-Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+Copyright (c) 1996, 2010, Innobase Oy. All Rights Reserved.
 Copyright (c) 2008, Google Inc.
+Copyright (c) 2009, Percona Inc.
 
 Portions of this file contain modifications contributed and copyrighted by
 Google, Inc. Those modifications are gratefully acknowledged and are described
@@ -9,6 +10,13 @@ briefly in the InnoDB documentation. The contributions by Google are
 incorporated with their permission, and subject to the conditions contained in
 the file COPYING.Google.
 
+Portions of this file contain modifications contributed and copyrighted
+by Percona Inc.. Those modifications are
+gratefully acknowledged and are described briefly in the InnoDB
+documentation. The contributions by Percona Inc. are incorporated with
+their permission, and subject to the conditions contained in the file
+COPYING.Percona.
+
 This program is free software; you can redistribute it and/or modify it under
 the terms of the GNU General Public License as published by the Free Software
 Foundation; version 2 of the License.
@@ -22,32 +30,6 @@ this program; if not, write to the Free Software Foundation, Inc., 59 Temple
 Place, Suite 330, Boston, MA 02111-1307 USA
 
 *****************************************************************************/
-/***********************************************************************
-
-Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
-Copyright (c) 2009, Percona Inc.
-
-Portions of this file contain modifications contributed and copyrighted
-by Percona Inc.. Those modifications are
-gratefully acknowledged and are described briefly in the InnoDB
-documentation. The contributions by Percona Inc. are incorporated with
-their permission, and subject to the conditions contained in the file
-COPYING.Percona.
-
-This program is free software; you can redistribute it and/or modify it
-under the terms of the GNU General Public License as published by the
-Free Software Foundation; version 2 of the License.
-
-This program is distributed in the hope that it will be useful, but
-WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General
-Public License for more details.
-
-You should have received a copy of the GNU General Public License along
-with this program; if not, write to the Free Software Foundation, Inc.,
-59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
-
-***********************************************************************/
 
 /********************************************************************//**
 @file srv/srv0start.c
@@ -105,6 +87,7 @@ Created 2/16/1996 Heikki Tuuri
 # include "btr0pcur.h"
 # include "thr0loc.h"
 # include "os0sync.h" /* for INNODB_RW_LOCKS_USE_ATOMICS */
+# include "zlib.h" /* for ZLIB_VERSION */
 
 /** Log sequence number immediately after startup */
 UNIV_INTERN ib_uint64_t	srv_start_lsn;
@@ -143,9 +126,9 @@ static mutex_t		ios_mutex;
 static ulint		ios;
 
 /** io_handler_thread parameters for thread identification */
-static ulint		n[SRV_MAX_N_IO_THREADS + 5 + 64];
+static ulint		n[SRV_MAX_N_IO_THREADS + 6 + 64];
 /** io_handler_thread identifiers */
-static os_thread_id_t	thread_ids[SRV_MAX_N_IO_THREADS + 5 + 64];
+static os_thread_id_t	thread_ids[SRV_MAX_N_IO_THREADS + 6 + 64];
 
 /** We use this mutex to test the return value of pthread_mutex_trylock
    on successful locking. HP-UX does NOT return 0, though Linux et al do. */
@@ -728,6 +711,7 @@ open_or_create_data_files(
 /*======================*/
 	ibool*		create_new_db,	/*!< out: TRUE if new database should be
 					created */
+	ibool*		create_new_doublewrite_file,
 #ifdef UNIV_LOG_ARCHIVE
 	ulint*		min_arch_log_no,/*!< out: min of archived log
 					numbers in data files */
@@ -760,6 +744,7 @@ open_or_create_data_files(
 	*sum_of_new_sizes = 0;
 
 	*create_new_db = FALSE;
+	*create_new_doublewrite_file = FALSE;
 
 	srv_normalize_path_for_win(srv_data_home);
 
@@ -992,6 +977,142 @@ skip_size_check:
 				srv_data_file_is_raw_partition[i] != 0);
 	}
 
+	/* special file for doublewrite buffer */
+	if (srv_doublewrite_file)
+	{
+		srv_normalize_path_for_win(srv_doublewrite_file);
+
+		fprintf(stderr,
+			"InnoDB: Notice: innodb_doublewrite_file is specified.\n"
+			"InnoDB: This is for expert only. Don't use if you don't understand what is it 'WELL'.\n"
+			"InnoDB: ### Don't specify older file than the last checkpoint ###\n"
+			"InnoDB: otherwise the older doublewrite buffer will break your data during recovery!\n");
+
+		strcpy(name, srv_doublewrite_file);
+
+		/* First we try to create the file: if it already
+		exists, ret will get value FALSE */
+
+		files[i] = os_file_create(name, OS_FILE_CREATE,
+					  OS_FILE_NORMAL,
+					  OS_DATA_FILE, &ret);
+
+		if (ret == FALSE && os_file_get_last_error(FALSE)
+		    != OS_FILE_ALREADY_EXISTS
+#ifdef UNIV_AIX
+		    /* AIX 5.1 after security patch ML7 may have
+		    errno set to 0 here, which causes our function
+		    to return 100; work around that AIX problem */
+		    && os_file_get_last_error(FALSE) != 100
+#endif
+		    ) {
+			fprintf(stderr,
+				"InnoDB: Error in creating"
+				" or opening %s\n",
+				name);
+
+			return(DB_ERROR);
+		}
+
+		if (ret == FALSE) {
+			/* We open the data file */
+
+			files[i] = os_file_create(
+				name, OS_FILE_OPEN, OS_FILE_NORMAL,
+				OS_DATA_FILE, &ret);
+
+			if (!ret) {
+				fprintf(stderr,
+					"InnoDB: Error in opening %s\n", name);
+				os_file_get_last_error(TRUE);
+
+				return(DB_ERROR);
+			}
+
+			ret = os_file_get_size(files[i], &size, &size_high);
+			ut_a(ret);
+			/* Round size downward to megabytes */
+
+			rounded_size_pages
+				= (size / (1024 * 1024) + 4096 * size_high)
+					<< (20 - UNIV_PAGE_SIZE_SHIFT);
+
+			if (rounded_size_pages != TRX_SYS_DOUBLEWRITE_BLOCK_SIZE * 9) {
+
+				fprintf(stderr,
+					"InnoDB: Warning: doublewrite buffer file %s"
+					" is of a different size\n"
+					"InnoDB: %lu pages"
+					" (rounded down to MB)\n"
+					"InnoDB: than intended size"
+					" %lu pages...\n",
+					name,
+					(ulong) rounded_size_pages,
+					(ulong) TRX_SYS_DOUBLEWRITE_BLOCK_SIZE * 9);
+			}
+
+			fil_read_flushed_lsn_and_arch_log_no(
+				files[i], one_opened,
+#ifdef UNIV_LOG_ARCHIVE
+				min_arch_log_no, max_arch_log_no,
+#endif /* UNIV_LOG_ARCHIVE */
+				min_flushed_lsn, max_flushed_lsn);
+			one_opened = TRUE;
+		} else {
+			/* We created the data file and now write it full of
+			zeros */
+
+			*create_new_doublewrite_file = TRUE;
+
+			ut_print_timestamp(stderr);
+			fprintf(stderr,
+				"  InnoDB: Doublewrite buffer file %s did not"
+				" exist: new to be created\n",
+				name);
+
+			if (*create_new_db == FALSE) {
+				fprintf(stderr,
+					"InnoDB: Warning: Previous version's ibdata files may cause crash.\n"
+					"        If you use that, please use the ibdata files of this version.\n");
+			}
+
+			ut_print_timestamp(stderr);
+			fprintf(stderr,
+				"  InnoDB: Setting file %s size to %lu MB\n",
+				name,
+				(ulong) ((TRX_SYS_DOUBLEWRITE_BLOCK_SIZE * 9)
+					 >> (20 - UNIV_PAGE_SIZE_SHIFT)));
+
+			fprintf(stderr,
+				"InnoDB: Database physically writes the"
+				" file full: wait...\n");
+
+			ret = os_file_set_size(
+				name, files[i],
+				srv_calc_low32(TRX_SYS_DOUBLEWRITE_BLOCK_SIZE * 9),
+				srv_calc_high32(TRX_SYS_DOUBLEWRITE_BLOCK_SIZE * 9));
+
+			if (!ret) {
+				fprintf(stderr,
+					"InnoDB: Error in creating %s:"
+					" probably out of disk space\n", name);
+
+				return(DB_ERROR);
+			}
+		}
+
+		ret = os_file_close(files[i]);
+		ut_a(ret);
+
+		fil_space_create(name, TRX_DOUBLEWRITE_SPACE, 0, FIL_TABLESPACE);
+
+		ut_a(fil_validate());
+
+		fil_node_create(name, TRX_SYS_DOUBLEWRITE_BLOCK_SIZE * 9, TRX_DOUBLEWRITE_SPACE, FALSE);
+
+		i++;
+	}
+
 	ios = 0;
 
 	mutex_create(&ios_mutex, SYNC_NO_ORDER_CHECK);
@@ -1010,6 +1131,7 @@ innobase_start_or_create_for_mysql(void)
 {
 	buf_pool_t*	ret;
 	ibool		create_new_db;
+	ibool		create_new_doublewrite_file;
 	ibool		log_file_created;
 	ibool		log_created	= FALSE;
 	ibool		log_opened	= FALSE;
@@ -1074,7 +1196,11 @@ innobase_start_or_create_for_mysql(void)
 #ifdef UNIV_IBUF_DEBUG
 	fprintf(stderr,
 		"InnoDB: !!!!!!!! UNIV_IBUF_DEBUG switched on !!!!!!!!!\n"
-		"InnoDB: Crash recovery will fail with UNIV_IBUF_DEBUG\n");
+# ifdef UNIV_IBUF_COUNT_DEBUG
+		"InnoDB: !!!!!!!! UNIV_IBUF_COUNT_DEBUG switched on !!!!!!!!!\n"
+		"InnoDB: Crash recovery will fail with UNIV_IBUF_COUNT_DEBUG\n"
+# endif
+		);
 #endif
 
 #ifdef UNIV_SYNC_DEBUG
@@ -1101,7 +1227,15 @@ innobase_start_or_create_for_mysql(void)
 			"InnoDB: The InnoDB memory heap is disabled\n");
 	}
 
-	fprintf(stderr, "InnoDB: %s\n", IB_ATOMICS_STARTUP_MSG);
+	fputs("InnoDB: " IB_ATOMICS_STARTUP_MSG
+	      "\nInnoDB: Compressed tables use zlib " ZLIB_VERSION
+#ifdef UNIV_ZIP_DEBUG
+	      " with validation"
+#endif /* UNIV_ZIP_DEBUG */
+#ifdef UNIV_ZIP_COPY
+	      " and extra copying"
+#endif /* UNIV_ZIP_COPY */
+	      "\n" , stderr);
 
 	/* Since InnoDB does not currently clean up all its internal data
 	structures in MySQL Embedded Server Library server_end(), we
@@ -1388,6 +1522,7 @@ innobase_start_or_create_for_mysql(void)
 	}
 
 	err = open_or_create_data_files(&create_new_db,
+					&create_new_doublewrite_file,
 #ifdef UNIV_LOG_ARCHIVE
 					&min_arch_log_no, &max_arch_log_no,
 #endif /* UNIV_LOG_ARCHIVE */
@@ -1504,6 +1639,14 @@ innobase_start_or_create_for_mysql(void)
 
 	trx_sys_file_format_init();
 
+	if (create_new_doublewrite_file) {
+		mtr_start(&mtr);
+		fsp_header_init(TRX_DOUBLEWRITE_SPACE, TRX_SYS_DOUBLEWRITE_BLOCK_SIZE * 9, &mtr);
+		mtr_commit(&mtr);
+
+		trx_sys_dummy_create(TRX_DOUBLEWRITE_SPACE);
+	}
+
 	if (create_new_db) {
 		mtr_start(&mtr);
 		fsp_header_init(0, sum_of_new_sizes, &mtr);
@@ -1596,6 +1739,14 @@ innobase_start_or_create_for_mysql(void)
 		dict_boot();
 		trx_sys_init_at_db_start();
 
+		/* Initialize the fsp free limit global variable in the log
+		system */
+		fsp_header_get_free_limit();
+
+		/* recv_recovery_from_checkpoint_finish needs trx lists which
+		are initialized in trx_sys_init_at_db_start(). */
+
+		recv_recovery_from_checkpoint_finish();
 		if (srv_force_recovery < SRV_FORCE_NO_IBUF_MERGE) {
 			/* The following call is necessary for the insert
 			buffer to work with multiple tablespaces. We must
@@ -1611,26 +1762,14 @@ innobase_start_or_create_for_mysql(void)
 			every table in the InnoDB data dictionary that has
 			an .ibd file.
 
-			We also determine the maximum tablespace id used.
-
-			TODO: We may have incomplete transactions in the
-			data dictionary tables. Does that harm the scanning of
-			the data dictionary below? */
+			We also determine the maximum tablespace id used. */
 
 			dict_check_tablespaces_and_store_max_id(
 				recv_needed_recovery);
 		}
 
 		srv_startup_is_before_trx_rollback_phase = FALSE;
-
-		/* Initialize the fsp free limit global variable in the log
-		system */
-		fsp_header_get_free_limit();
-
-		/* recv_recovery_from_checkpoint_finish needs trx lists which
-		are initialized in trx_sys_init_at_db_start(). */
-
-		recv_recovery_from_checkpoint_finish();
+		recv_recovery_rollback_active();
 
 		/* It is possible that file_format tag has never
 		been set. In this case we initialize it to minimum
@@ -1679,15 +1818,18 @@ innobase_start_or_create_for_mysql(void)
 	/* fprintf(stderr, "Max allowed record size %lu\n",
 	page_get_free_space_of_empty() / 2); */
 
-	/* Create the thread which watches the timeouts for lock waits
-	and prints InnoDB monitor info */
-
-	os_thread_create(&srv_lock_timeout_and_monitor_thread, NULL,
+	/* Create the thread which watches the timeouts for lock waits */
+	os_thread_create(&srv_lock_timeout_thread, NULL,
 			 thread_ids + 2 + SRV_MAX_N_IO_THREADS);
 
 	/* Create the thread which warns of long semaphore waits */
 	os_thread_create(&srv_error_monitor_thread, NULL,
 			 thread_ids + 3 + SRV_MAX_N_IO_THREADS);
+
+	/* Create the thread which prints InnoDB monitor info */
+	os_thread_create(&srv_monitor_thread, NULL,
+			 thread_ids + 4 + SRV_MAX_N_IO_THREADS);
+
 	srv_is_being_started = FALSE;
 
 	if (trx_doublewrite == NULL) {
@@ -1712,13 +1854,13 @@ innobase_start_or_create_for_mysql(void)
 		ulint i;
 
 		os_thread_create(&srv_purge_thread, NULL, thread_ids
-				 + (4 + SRV_MAX_N_IO_THREADS));
+				 + (5 + SRV_MAX_N_IO_THREADS));
 
 		for (i = 0; i < srv_use_purge_thread - 1; i++) {
-			n[5 + i + SRV_MAX_N_IO_THREADS] = i; /* using as index for arrays in purge_sys */
+			n[6 + i + SRV_MAX_N_IO_THREADS] = i; /* using as index for arrays in purge_sys */
 			os_thread_create(&srv_purge_worker_thread,
-					 n + (5 + i + SRV_MAX_N_IO_THREADS),
-					 thread_ids + (5 + i + SRV_MAX_N_IO_THREADS));
+					 n + (6 + i + SRV_MAX_N_IO_THREADS),
+					 thread_ids + (6 + i + SRV_MAX_N_IO_THREADS));
 		}
 	}
 #ifdef UNIV_DEBUG
@@ -1821,7 +1963,7 @@ innobase_start_or_create_for_mysql(void)
 	if (srv_print_verbose_log) {
 		ut_print_timestamp(stderr);
 		fprintf(stderr,
-			" InnoDB Plugin %s started; "
+			" Percona XtraDB (http://www.percona.com) %s started; "
 			"log sequence number %llu\n",
 			INNODB_VERSION_STR, srv_start_lsn);
 	}