Content merge.

sql/ha_innodb.cc:
  Auto merged
sql/ha_innodb.h:
  Auto merged
sql/handler.h:
  Auto merged
sql/mysql_priv.h:
  Auto merged
sql/sql_parse.cc:
  Auto merged
sql/sql_yacc.yy:
  Auto merged
innobase/trx/trx0roll.c:
  Auto merged.

15 files changed, 312 insertions, 94 deletions

diff --git a/innobase/buf/buf0buf.c b/innobase/buf/buf0buf.c
index 3930ea93889..89f851709db 100644
--- a/innobase/buf/buf0buf.c
+++ b/innobase/buf/buf0buf.c
@@ -331,33 +331,43 @@ buf_page_is_corrupted(
 		}
 	}
 #endif
-	old_checksum = buf_calc_page_old_checksum(read_buf);
-
-	old_checksum_field = mach_read_from_4(read_buf + UNIV_PAGE_SIZE
+  
+  /* If we use checksums validation, make additional check before returning
+  TRUE to ensure that the checksum is not equal to BUF_NO_CHECKSUM_MAGIC which
+  might be stored by InnoDB with checksums disabled.
+     Otherwise, skip checksum calculation and return FALSE */
+  
+  if (srv_use_checksums) {
+    old_checksum = buf_calc_page_old_checksum(read_buf); 
+
+    old_checksum_field = mach_read_from_4(read_buf + UNIV_PAGE_SIZE
 					- FIL_PAGE_END_LSN_OLD_CHKSUM);
 
-	/* There are 2 valid formulas for old_checksum_field:
-	1. Very old versions of InnoDB only stored 8 byte lsn to the start
-	and the end of the page.
-	2. Newer InnoDB versions store the old formula checksum there. */
+    /* There are 2 valid formulas for old_checksum_field:
+	  1. Very old versions of InnoDB only stored 8 byte lsn to the start
+	  and the end of the page.
+	  2. Newer InnoDB versions store the old formula checksum there. */
 	
-	if (old_checksum_field != mach_read_from_4(read_buf + FIL_PAGE_LSN)
-	    && old_checksum_field != old_checksum) {
+    if (old_checksum_field != mach_read_from_4(read_buf + FIL_PAGE_LSN)
+        && old_checksum_field != old_checksum
+        && old_checksum_field != BUF_NO_CHECKSUM_MAGIC) {
 
-		return(TRUE);
-	}
+      return(TRUE);
+    }
 
-	checksum = buf_calc_page_new_checksum(read_buf);
-	checksum_field = mach_read_from_4(read_buf + FIL_PAGE_SPACE_OR_CHKSUM);
+    checksum = buf_calc_page_new_checksum(read_buf);
+    checksum_field = mach_read_from_4(read_buf + FIL_PAGE_SPACE_OR_CHKSUM);
 
-	/* InnoDB versions < 4.0.14 and < 4.1.1 stored the space id
-	(always equal to 0), to FIL_PAGE_SPACE_SPACE_OR_CHKSUM */
+    /* InnoDB versions < 4.0.14 and < 4.1.1 stored the space id
+	  (always equal to 0), to FIL_PAGE_SPACE_SPACE_OR_CHKSUM */
 
-	if (checksum_field != 0 && checksum_field != checksum) {
-
-	        return(TRUE);
-	}
+    if (checksum_field != 0 && checksum_field != checksum
+        && checksum_field != BUF_NO_CHECKSUM_MAGIC) {
 
+      return(TRUE);
+    }
+  }
+  
 	return(FALSE);
 }
 
@@ -379,8 +389,10 @@ buf_page_print(
 	ut_print_buf(stderr, read_buf, UNIV_PAGE_SIZE);
 	fputs("InnoDB: End of page dump\n", stderr);
 
-	checksum = buf_calc_page_new_checksum(read_buf);
-	old_checksum = buf_calc_page_old_checksum(read_buf);
+	checksum = srv_use_checksums ?
+    buf_calc_page_new_checksum(read_buf) : BUF_NO_CHECKSUM_MAGIC;
+	old_checksum = srv_use_checksums ?
+    buf_calc_page_old_checksum(read_buf) : BUF_NO_CHECKSUM_MAGIC;
 
 	ut_print_timestamp(stderr);
 	fprintf(stderr, 
@@ -548,7 +560,7 @@ buf_pool_init(
 		}
 		/*----------------------------------------*/
 	} else {
-		buf_pool->frame_mem = ut_malloc_low(
+		buf_pool->frame_mem = os_mem_alloc_large(
 					UNIV_PAGE_SIZE * (n_frames + 1),
 					TRUE, FALSE);
 	}
diff --git a/innobase/buf/buf0flu.c b/innobase/buf/buf0flu.c
index aff4fe92a71..a0ca614d9b3 100644
--- a/innobase/buf/buf0flu.c
+++ b/innobase/buf/buf0flu.c
@@ -448,7 +448,8 @@ buf_flush_init_for_writing(
 	/* Store the new formula checksum */
 
 	mach_write_to_4(page + FIL_PAGE_SPACE_OR_CHKSUM,
-					buf_calc_page_new_checksum(page));
+					srv_use_checksums ?
+                  buf_calc_page_new_checksum(page) : BUF_NO_CHECKSUM_MAGIC);
 
 	/* We overwrite the first 4 bytes of the end lsn field to store
 	the old formula checksum. Since it depends also on the field
@@ -456,7 +457,8 @@ buf_flush_init_for_writing(
 	new formula checksum. */
 
 	mach_write_to_4(page + UNIV_PAGE_SIZE - FIL_PAGE_END_LSN_OLD_CHKSUM,
-					buf_calc_page_old_checksum(page));
+					srv_use_checksums ?
+                  buf_calc_page_old_checksum(page) : BUF_NO_CHECKSUM_MAGIC);
 }
 
 /************************************************************************
diff --git a/innobase/include/Makefile.am b/innobase/include/Makefile.am
index 102d25566da..eb1e3b72877 100644
--- a/innobase/include/Makefile.am
+++ b/innobase/include/Makefile.am
@@ -49,7 +49,7 @@ noinst_HEADERS = btr0btr.h btr0btr.ic btr0cur.h btr0cur.ic \
         thr0loc.h thr0loc.ic trx0purge.h trx0purge.ic trx0rec.h \
         trx0rec.ic trx0roll.h trx0roll.ic trx0rseg.h trx0rseg.ic \
         trx0sys.h trx0sys.ic trx0trx.h trx0trx.ic trx0types.h \
-        trx0undo.h trx0undo.ic univ.i \
+        trx0undo.h trx0undo.ic trx0xa.h univ.i \
         usr0sess.h usr0sess.ic usr0types.h ut0byte.h ut0byte.ic \
         ut0dbg.h ut0lst.h ut0mem.h ut0mem.ic ut0rnd.h ut0rnd.ic \
         ut0sort.h ut0ut.h ut0ut.ic
diff --git a/innobase/include/buf0buf.h b/innobase/include/buf0buf.h
index 8df1211327f..5ee323f1b1e 100644
--- a/innobase/include/buf0buf.h
+++ b/innobase/include/buf0buf.h
@@ -52,6 +52,8 @@ Created 11/5/1995 Heikki Tuuri
 /* Modes for buf_page_get_known_nowait */
 #define BUF_MAKE_YOUNG	51
 #define BUF_KEEP_OLD	52
+/* Magic value to use instead of checksums when they are disabled */
+#define BUF_NO_CHECKSUM_MAGIC 0xDEADBEEFUL
 
 extern buf_pool_t* 	buf_pool; 	/* The buffer pool of the database */
 extern ibool		buf_debug_prints;/* If this is set TRUE, the program
diff --git a/innobase/include/os0proc.h b/innobase/include/os0proc.h
index d0d3cf82e38..b0b72e18675 100644
--- a/innobase/include/os0proc.h
+++ b/innobase/include/os0proc.h
@@ -12,6 +12,11 @@ Created 9/30/1995 Heikki Tuuri
 
 #include "univ.i"
 
+#ifdef UNIV_LINUX
+#include <sys/ipc.h>
+#include <sys/shm.h>
+#endif
+
 typedef void*			os_process_t;
 typedef unsigned long int	os_process_id_t;
 
@@ -27,6 +32,10 @@ page size of an Intel x86 processor. We cannot use AWE with 2 MB or 4 MB
 pages. */
 #define	OS_AWE_X86_PAGE_SIZE	4096
 
+extern ibool os_use_large_pages;
+/* Large page size. This may be a boot-time option on some platforms */
+extern ulint os_large_page_size;
+
 /********************************************************************
 Windows AWE support. Tries to enable the "lock pages in memory" privilege for
 the current process so that the current process can allocate memory-locked
@@ -103,6 +112,25 @@ os_mem_alloc_nocache(
 			/* out: allocated memory */
 	ulint	n);	/* in: number of bytes */
 /********************************************************************
+Allocates large pages memory. */
+
+void*
+os_mem_alloc_large(
+/*=================*/
+      /* out: allocated memory */
+	ulint   n,     /* in: number of bytes */
+  ibool   set_to_zero, /* in: TRUE if allocated memory should be set
+          to zero if UNIV_SET_MEM_TO_ZERO is defined */
+  ibool	  assert_on_error); /* in: if TRUE, we crash mysqld if the memory
+          cannot be allocated */
+/********************************************************************
+Frees large pages memory. */
+
+void
+os_mem_free_large(
+/*=================*/
+void    *ptr);  /* in: number of bytes */
+/********************************************************************
 Sets the priority boost for threads released from waiting within the current
 process. */
 
diff --git a/innobase/include/page0page.ic b/innobase/include/page0page.ic
index 1d5ea337031..a63b5ca4238 100644
--- a/innobase/include/page0page.ic
+++ b/innobase/include/page0page.ic
@@ -153,6 +153,19 @@ page_header_reset_last_insert(
 }
 
 /****************************************************************
+Determine whether the page is in new-style compact format. */
+UNIV_INLINE
+ibool
+page_is_comp(
+/*=========*/
+			/* out: TRUE if the page is in compact format
+			FALSE if it is in old-style format */
+	page_t* page)	/* in: index page */
+{
+	return(!!(page_header_get_field(page, PAGE_N_HEAP) & 0x8000));
+}
+
+/****************************************************************
 Gets the first record on the page. */
 UNIV_INLINE
 rec_t*
@@ -514,19 +527,6 @@ page_dir_calc_reserved_space(
 } 
 
 /****************************************************************
-Determine whether the page is in new-style compact format. */
-UNIV_INLINE
-ibool
-page_is_comp(
-/*=========*/
-			/* out: TRUE if the page is in compact format
-			FALSE if it is in old-style format */
-	page_t* page)	/* in: index page */
-{
-	return(!!(page_header_get_field(page, PAGE_N_HEAP) & 0x8000));
-}
-
-/****************************************************************
 Gets the pointer to the next record on the page. */
 UNIV_INLINE
 rec_t*
diff --git a/innobase/include/rem0rec.ic b/innobase/include/rem0rec.ic
index db938aa9fa5..6c3dabf04a2 100644
--- a/innobase/include/rem0rec.ic
+++ b/innobase/include/rem0rec.ic
@@ -40,8 +40,18 @@ most significant bytes and bits are written below less significant.
 
 	(1) byte offset		(2) bit usage within byte
 	downward from
-	origin ->	1	8 bits pointer to next record (relative)
-			2	8 bits pointer to next record (relative)
+	origin ->	1	8 bits relative offset of next record
+			2	8 bits relative offset of next record
+				  the relative offset is an unsigned 16-bit
+				  integer:
+				  (offset_of_next_record
+				   - offset_of_this_record) mod 64Ki,
+				  where mod is the modulo as a non-negative
+				  number;
+				  we can calculate the the offset of the next
+				  record with the formula:
+				  relative_offset + offset_of_this_record
+				  mod UNIV_PAGE_SIZE
 			3	3 bits status:
 					000=conventional record
 					001=node pointer record (inside B-tree)
@@ -252,26 +262,37 @@ UNIV_INLINE
 ulint 
 rec_get_next_offs(
 /*==============*/
-			/* out: the page offset of the next chained record */
+			/* out: the page offset of the next chained record, or
+			0 if none */
 	rec_t*	rec,	/* in: physical record */
 	ibool	comp)	/* in: TRUE=compact page format */
 {	
+	ulint	field_value;
+		
+	ut_ad(REC_NEXT_MASK == 0xFFFFUL);
+	ut_ad(REC_NEXT_SHIFT == 0);
+
+	field_value = mach_read_from_2(rec - REC_NEXT);
+
 	if (comp) {
-		lint ret = (int16_t) rec_get_bit_field_2(rec, REC_NEXT,
-						REC_NEXT_MASK, REC_NEXT_SHIFT);
 #if UNIV_PAGE_SIZE <= 32768
-		/* with 64 KiB page size, the pointer will "wrap around",
-		and the following assertions are invalid */
-		ut_ad(ret + ut_align_offset(rec, UNIV_PAGE_SIZE) <
-							UNIV_PAGE_SIZE);
+		/* Note that for 64 KiB pages, field_value can 'wrap around'
+		and the debug assertion is not valid */
+
+		ut_ad((int16_t)field_value
+		       + ut_align_offset(rec, UNIV_PAGE_SIZE)
+		      < UNIV_PAGE_SIZE);
 #endif
-		return(ret ? ut_align_offset(rec + ret, UNIV_PAGE_SIZE) : 0);
-	}
-	else {
-		ulint ret = rec_get_bit_field_2(rec, REC_NEXT,
-						REC_NEXT_MASK, REC_NEXT_SHIFT);
-		ut_ad(ret < UNIV_PAGE_SIZE);
-		return(ret);
+		if (field_value == 0) {
+
+			return(0);
+		}
+		
+		return(ut_align_offset(rec + field_value, UNIV_PAGE_SIZE));
+	} else {
+		ut_ad(field_value < UNIV_PAGE_SIZE);
+
+		return(field_value);
 	}
 }
 
@@ -284,21 +305,31 @@ rec_set_next_offs(
 /*==============*/
 	rec_t*	rec,	/* in: physical record */
 	ibool	comp,	/* in: TRUE=compact page format */
-	ulint	next)	/* in: offset of the next record */
+	ulint	next)	/* in: offset of the next record, or 0 if none */
 {
 	ut_ad(rec);
 	ut_ad(UNIV_PAGE_SIZE > next);
+	ut_ad(REC_NEXT_MASK == 0xFFFFUL);
+	ut_ad(REC_NEXT_SHIFT == 0);
 
 	if (comp) {
-		rec_set_bit_field_2(rec, next
-				? (next - ut_align_offset(rec, UNIV_PAGE_SIZE))
-#ifdef UNIV_DEBUG /* avoid an assertion failure */
-				& (REC_NEXT_MASK >> REC_NEXT_SHIFT)
-#endif
-				: 0, REC_NEXT, REC_NEXT_MASK, REC_NEXT_SHIFT);
+		ulint field_value;
+
+		if (next) {
+			/* The following two statements calculate
+			next - offset_of_rec mod 64Ki, where mod is the modulo
+			as a non-negative number */
+		
+			field_value = (ulint)((lint)next 
+				- (lint)ut_align_offset(rec, UNIV_PAGE_SIZE));
+			field_value &= REC_NEXT_MASK;
+		} else {
+			field_value = 0;
+		}
+
+		mach_write_to_2(rec - REC_NEXT, field_value);
 	} else {
-		rec_set_bit_field_2(rec, next,
-				REC_NEXT, REC_NEXT_MASK, REC_NEXT_SHIFT);
+		mach_write_to_2(rec - REC_NEXT, next);
 	}
 }
 
diff --git a/innobase/include/srv0srv.h b/innobase/include/srv0srv.h
index d4cc7d8222f..84b7d14ca00 100644
--- a/innobase/include/srv0srv.h
+++ b/innobase/include/srv0srv.h
@@ -107,6 +107,7 @@ extern ibool	srv_very_fast_shutdown;  /* if this TRUE, do not flush the
 extern ibool	srv_innodb_status;
 
 extern ibool	srv_use_doublewrite_buf;
+extern ibool	srv_use_checksums;
 
 extern ibool    srv_set_thread_priorities;
 extern int      srv_query_thread_priority;
diff --git a/innobase/include/trx0roll.h b/innobase/include/trx0roll.h
index 893e5af6c01..9d025da4a5f 100644
--- a/innobase/include/trx0roll.h
+++ b/innobase/include/trx0roll.h
@@ -105,11 +105,19 @@ trx_rollback(
 Rollback or clean up transactions which have no user session. If the
 transaction already was committed, then we clean up a possible insert
 undo log. If the transaction was not yet committed, then we roll it back. 
-Note: this is done in a background thread */
+Note: this is done in a background thread. */
 
-void *
-trx_rollback_or_clean_all_without_sess(void *);
-/*============================================*/
+#ifndef __WIN__
+void*
+#else
+ulint
+#endif
+trx_rollback_or_clean_all_without_sess(
+/*===================================*/
+                        /* out: a dummy parameter */
+        void*   arg __attribute__((unused)));
+                        /* in: a dummy parameter required by
+                        os_thread_create */
 /********************************************************************
 Finishes a transaction rollback. */
 
diff --git a/innobase/log/log0recv.c b/innobase/log/log0recv.c
index f42f0eb8c72..5eefd32c8a6 100644
--- a/innobase/log/log0recv.c
+++ b/innobase/log/log0recv.c
@@ -2937,7 +2937,6 @@ recv_recovery_from_checkpoint_finish(void)
 #ifndef UNIV_LOG_DEBUG
 	recv_sys_free();
 #endif
-
 	if (srv_force_recovery < SRV_FORCE_NO_TRX_UNDO) {
 		os_thread_create(trx_rollback_or_clean_all_without_sess,
 				(void *)&i, &recovery_thread_id);
diff --git a/innobase/os/os0proc.c b/innobase/os/os0proc.c
index 2f155788420..167aed93de7 100644
--- a/innobase/os/os0proc.c
+++ b/innobase/os/os0proc.c
@@ -69,6 +69,10 @@ byte*		os_awe_window;
 ulint		os_awe_window_size;
 #endif
 
+ibool os_use_large_pages;
+/* Large page size. This may be a boot-time option on some platforms */
+ulint os_large_page_size;
+
 /********************************************************************
 Windows AWE support. Tries to enable the "lock pages in memory" privilege for
 the current process so that the current process can allocate memory-locked
@@ -516,6 +520,89 @@ os_mem_alloc_nocache(
 }
 
 /********************************************************************
+Allocates large pages memory. */
+
+void*
+os_mem_alloc_large(
+/*=================*/
+      /* out: allocated memory */
+  ulint	n, /* in: number of bytes */
+	ibool set_to_zero, /* in: TRUE if allocated memory should be set
+        to zero if UNIV_SET_MEM_TO_ZERO is defined */
+	ibool	assert_on_error) /* in: if TRUE, we crash mysqld if the memory
+				cannot be allocated */
+{
+#ifdef HAVE_LARGE_PAGES
+  ulint size;
+  int shmid;
+  void *ptr = NULL;
+  struct shmid_ds buf;
+  
+  if (!os_use_large_pages || !os_large_page_size) {
+    goto skip;
+  }
+
+#ifdef UNIV_LINUX
+  /* Align block size to os_large_page_size */
+  size = ((n - 1) & ~(os_large_page_size - 1)) + os_large_page_size;
+  
+  shmid = shmget(IPC_PRIVATE, (size_t)size, SHM_HUGETLB | SHM_R | SHM_W);
+  if (shmid < 0) {
+    fprintf(stderr, "InnoDB: HugeTLB: Warning: Failed to allocate %lu bytes. "
+            "errno %d\n", n, errno);
+  } else {
+    ptr = shmat(shmid, NULL, 0);
+    if (ptr == (void *)-1) {
+      fprintf(stderr, "InnoDB: HugeTLB: Warning: Failed to attach shared memory "
+              "segment, errno %d\n", errno);
+    }
+    /*
+      Remove the shared memory segment so that it will be automatically freed
+      after memory is detached or process exits
+    */
+    shmctl(shmid, IPC_RMID, &buf);
+  }
+#endif
+  
+  if (ptr) {
+    if (set_to_zero) {
+#ifdef UNIV_SET_MEM_TO_ZERO
+      memset(ptr, '\0', size);
+#endif
+    }
+
+    return(ptr);
+  }
+
+  fprintf(stderr, "InnoDB HugeTLB: Warning: Using conventional memory pool\n");
+skip:
+#endif /* HAVE_LARGE_PAGES */
+  
+	return(ut_malloc_low(n, set_to_zero, assert_on_error));
+}
+
+/********************************************************************
+Frees large pages memory. */
+
+void
+os_mem_free_large(
+/*=================*/
+	void	*ptr)	/* in: number of bytes */
+{
+#ifdef HAVE_LARGE_PAGES
+  if (os_use_large_pages && os_large_page_size
+#ifdef UNIV_LINUX
+      && !shmdt(ptr)
+#endif
+      ) {
+    return;
+  }
+#endif
+
+  ut_free(ptr);
+}
+
+/********************************************************************
 Sets the priority boost for threads released from waiting within the current
 process. */
 
diff --git a/innobase/srv/srv0srv.c b/innobase/srv/srv0srv.c
index 40befae424e..83d4fb4d39d 100644
--- a/innobase/srv/srv0srv.c
+++ b/innobase/srv/srv0srv.c
@@ -313,6 +313,7 @@ ibool	srv_very_fast_shutdown	= FALSE; /* if this TRUE, do not flush the
 ibool	srv_innodb_status	= FALSE;
 
 ibool	srv_use_doublewrite_buf	= TRUE;
+ibool	srv_use_checksums = TRUE;
 
 ibool   srv_set_thread_priorities = TRUE;
 int     srv_query_thread_priority = 0;
diff --git a/innobase/srv/srv0start.c b/innobase/srv/srv0start.c
index 69341a1d7d1..a0e763d7a44 100644
--- a/innobase/srv/srv0start.c
+++ b/innobase/srv/srv0start.c
@@ -1403,15 +1403,13 @@ NetWare. */
 		fsp_header_inc_size(0, sum_of_new_sizes, &mtr);		
 
 		mtr_commit(&mtr);
-	}
 
-	if (recv_needed_recovery) {
-	    	ut_print_timestamp(stderr);
-		fprintf(stderr,
-	        "  InnoDB: Flushing modified pages from the buffer pool...\n");
-	}
+		/* Immediately write the log record about increased tablespace
+		size to disk, so that it is durable even if mysqld would crash
+		quickly */
 
-	log_make_checkpoint_at(ut_dulint_max, TRUE);
+		log_buffer_flush_to_disk();
+	}
 
 #ifdef UNIV_LOG_ARCHIVE
 	/* Archiving is always off under MySQL */
diff --git a/innobase/trx/trx0roll.c b/innobase/trx/trx0roll.c
index 105fa97080f..e5cffd2a4f3 100644
--- a/innobase/trx/trx0roll.c
+++ b/innobase/trx/trx0roll.c
@@ -30,9 +30,13 @@ Created 3/26/1996 Heikki Tuuri
 /* This many pages must be undone before a truncate is tried within rollback */
 #define TRX_ROLL_TRUNC_THRESHOLD	1
 
+/* In crash recovery, the current trx to be rolled back */
+trx_t*		trx_roll_crash_recv_trx	= NULL;
+
 /* In crash recovery we set this to the undo n:o of the current trx to be
 rolled back. Then we can print how many % the rollback has progressed. */
 ib_longlong	trx_roll_max_undo_no;
+
 /* Auxiliary variable which tells the previous progress % we printed */
 ulint		trx_roll_progress_printed_pct;
 
@@ -332,13 +336,19 @@ trx_savept_take(
 Rollback or clean up transactions which have no user session. If the
 transaction already was committed, then we clean up a possible insert
 undo log. If the transaction was not yet committed, then we roll it back. 
-Note: this is done in a background thread */
+Note: this is done in a background thread. */
 
-void *
+#ifndef __WIN__
+void*
+#else
+ulint
+#endif
 trx_rollback_or_clean_all_without_sess(
 /*===================================*/
-			/* out: arguments */
-	void	*i)	/* in: arguments (unused) */
+                        /* out: a dummy parameter */
+        void*   arg __attribute__((unused)))
+                        /* in: a dummy parameter required by
+                        os_thread_create */
 {
 	mem_heap_t*	heap;
 	que_fork_t*	fork;
@@ -363,9 +373,9 @@ trx_rollback_or_clean_all_without_sess(
 	if (UT_LIST_GET_FIRST(trx_sys->trx_list)) {
 
 		fprintf(stderr,
-		"InnoDB: Starting rollback of uncommitted transactions\n");
+"InnoDB: Starting in background the rollback of uncommitted transactions\n");
 	} else {		
-		os_thread_exit(i);
+		goto leave_function;
 	}
 loop:
 	heap = mem_heap_create(512);
@@ -375,7 +385,6 @@ loop:
 	trx = UT_LIST_GET_FIRST(trx_sys->trx_list);
 
 	while (trx) {
-
 		if ((trx->sess || (trx->conc_state == TRX_NOT_STARTED))) {
 			trx = UT_LIST_GET_NEXT(trx_list, trx);
 		} else if (trx->conc_state == TRX_PREPARED) {
@@ -388,17 +397,17 @@ loop:
 	mutex_exit(&kernel_mutex);
 
 	if (trx == NULL) {
+		ut_print_timestamp(stderr);
 		fprintf(stderr,
-		"InnoDB: Rollback of uncommitted transactions completed\n");
+		"  InnoDB: Rollback of uncommitted transactions completed\n");
 
  		mem_heap_free(heap);
-		
-		os_thread_exit(i);
+
+		goto leave_function;
 	}
 
 	trx->sess = trx_dummy_sess;
 
-	
 	if (trx->conc_state == TRX_COMMITTED_IN_MEMORY) {	
 		fprintf(stderr, "InnoDB: Cleaning up trx with id %lu %lu\n",
 					(ulong) ut_dulint_get_high(trx->id),
@@ -427,21 +436,28 @@ loop:
 
 	ut_a(thr == que_fork_start_command(fork));
 	
+	trx_roll_crash_recv_trx	= trx;
 	trx_roll_max_undo_no = ut_conv_dulint_to_longlong(trx->undo_no);
 	trx_roll_progress_printed_pct = 0;
 	rows_to_undo = trx_roll_max_undo_no;
+
 	if (rows_to_undo > 1000000000) {
 		rows_to_undo = rows_to_undo / 1000000;
 		unit = "M";
 	}
 
+	ut_print_timestamp(stderr);
 	fprintf(stderr,
-"InnoDB: Rolling back trx with id %lu %lu, %lu%s rows to undo",
+"  InnoDB: Rolling back trx with id %lu %lu, %lu%s rows to undo\n",
 					(ulong) ut_dulint_get_high(trx->id),
 					(ulong) ut_dulint_get_low(trx->id),
 					(ulong) rows_to_undo, unit);
 	mutex_exit(&kernel_mutex);
 
+	trx->mysql_thread_id = os_thread_get_curr_id();
+
+	trx->mysql_process_no = os_proc_get_number();
+
 	if (trx->dict_operation) {
 		row_mysql_lock_data_dictionary(trx);
 	}
@@ -456,7 +472,7 @@ loop:
 
 		fprintf(stderr,
 		"InnoDB: Waiting for rollback of trx id %lu to end\n",
-						(ulong) ut_dulint_get_low(trx->id));
+					(ulong) ut_dulint_get_low(trx->id));
 		os_thread_sleep(100000);
 
 		mutex_enter(&kernel_mutex);
@@ -495,10 +511,23 @@ loop:
 					(ulong) ut_dulint_get_low(trx->id));
 	mem_heap_free(heap);
 
+	trx_roll_crash_recv_trx	= NULL;
+
 	goto loop;
 
-	os_thread_exit(i); /* not reached */
-	return(i);
+leave_function:
+	/* 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);
+
+	/* The following is dummy code to keep the compiler happy: */
+
+#ifndef __WIN__
+        return(NULL);
+#else
+        return(0);
+#endif
 }
 	
 /***********************************************************************
@@ -859,16 +888,17 @@ try_again:
 	ut_ad(ut_dulint_cmp(ut_dulint_add(undo_no, 1), trx->undo_no) == 0);
 
 	/* We print rollback progress info if we are in a crash recovery
-	and the transaction has at least 1000 row operations to undo */
+	and the transaction has at least 1000 row operations to undo. */
+
+	if (trx == trx_roll_crash_recv_trx && trx_roll_max_undo_no > 1000) {
 
-	if (srv_is_being_started && trx_roll_max_undo_no > 1000) {
-	  progress_pct = 100 - (ulint)
+	  	progress_pct = 100 - (ulint)
 				((ut_conv_dulint_to_longlong(undo_no) * 100)
 				/ trx_roll_max_undo_no);
 		if (progress_pct != trx_roll_progress_printed_pct) {
 			if (trx_roll_progress_printed_pct == 0) {
 				fprintf(stderr,
-			"\nInnoDB: Progress in percents: %lu", (ulong) progress_pct);
+"\nInnoDB: Progress in percents: %lu\n", (ulong) progress_pct);
 			} else {
 				fprintf(stderr,
 				" %lu", (ulong) progress_pct);
diff --git a/innobase/trx/trx0sys.c b/innobase/trx/trx0sys.c
index 35e18064329..57166e98f45 100644
--- a/innobase/trx/trx0sys.c
+++ b/innobase/trx/trx0sys.c
@@ -125,6 +125,22 @@ trx_doublewrite_init(
 }
 
 /********************************************************************
+Frees the doublewrite buffer. */
+static
+void
+trx_doublewrite_free(void)
+/*======================*/
+{
+  mutex_free(&(trx_doublewrite->mutex));
+
+  mem_free(trx_doublewrite->buf_block_arr);
+  ut_free(trx_doublewrite->write_buf_unaligned);
+
+  mem_free(trx_doublewrite);
+  trx_doublewrite = NULL;
+}
+
+/********************************************************************
 Marks the trx sys header when we have successfully upgraded to the >= 4.1.x
 multiple tablespace format. */
 
@@ -512,6 +528,9 @@ trx_sys_doublewrite_init_or_restore_pages(
 
 	fil_flush_file_spaces(FIL_TABLESPACE);
 	
+  if (!srv_use_doublewrite_buf)
+    trx_doublewrite_free();
+
 leave_func:
 	ut_free(unaligned_read_buf);
 }