Merge from mysql-trunk-innodb to local repository

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

@@ -85,7 +85,8 @@ SELECT table_schema, table_name, row_format
 FROM information_schema.tables WHERE engine='innodb';
 drop table t1,t2;
 
-# The following should fail even in non-strict mode.
+# The following should fail in non-strict mode too.
+# (The fix of Bug #50945 only affects REDUNDANT and COMPACT tables.)
 SET SESSION innodb_strict_mode = off;
 --error ER_TOO_BIG_ROWSIZE
 CREATE TABLE t1(

mysql-test/suite/innodb/t/innodb_bug36169.test

@@ -24,6 +24,7 @@ SET GLOBAL innodb_file_per_table=ON;
 # Generating 10 tables
 # Creating a table with 94 columns and 24 indexes
 DROP TABLE IF EXISTS `table0`;
+set innodb_strict_mode=on;
 --error ER_TOO_BIG_ROWSIZE
 CREATE TABLE IF NOT EXISTS `table0`
 (`col0` BOOL,

storage/innobase/CMakeLists.txt

@@ -232,7 +232,7 @@ SET(INNOBASE_SOURCES	btr/btr0btr.c btr/btr0cur.c btr/btr0pcur.c btr/btr0sea.c
 			rem/rem0cmp.c rem/rem0rec.c
 			row/row0ext.c row/row0ins.c row/row0merge.c row/row0mysql.c row/row0purge.c row/row0row.c
 			row/row0sel.c row/row0uins.c row/row0umod.c row/row0undo.c row/row0upd.c row/row0vers.c
-			srv/srv0que.c srv/srv0srv.c srv/srv0start.c
+			srv/srv0srv.c srv/srv0start.c
 			sync/sync0arr.c sync/sync0rw.c sync/sync0sync.c
 			thr/thr0loc.c
 			trx/trx0i_s.c trx/trx0purge.c trx/trx0rec.c trx/trx0roll.c trx/trx0rseg.c

storage/innobase/Makefile.am

@@ -174,7 +174,6 @@ noinst_HEADERS=		\
 			include/row0upd.ic	\
 			include/row0vers.h	\
 			include/row0vers.ic	\
-			include/srv0que.h	\
 			include/srv0srv.h	\
 			include/srv0srv.ic	\
 			include/srv0start.h	\
@@ -299,7 +298,6 @@ libinnobase_a_SOURCES=	\
 			row/row0undo.c			\
 			row/row0upd.c			\
 			row/row0vers.c			\
-			srv/srv0que.c			\
 			srv/srv0srv.c			\
 			srv/srv0start.c			\
 			sync/sync0arr.c			\

storage/innobase/btr/btr0btr.c

@@ -1455,11 +1455,11 @@ Calculates a split record such that the tuple will certainly fit on
 its half-page when the split is performed. We assume in this function
 only that the cursor page has at least one user record.
 @return split record, or NULL if tuple will be the first record on
-upper half-page */
+the lower or upper half-page (determined by btr_page_tuple_smaller()) */
 static
 rec_t*
-btr_page_get_sure_split_rec(
-/*========================*/
+btr_page_get_split_rec(
+/*===================*/
 	btr_cur_t*	cursor,	/*!< in: cursor at which insert should be made */
 	const dtuple_t*	tuple,	/*!< in: tuple to insert */
 	ulint		n_ext)	/*!< in: number of externally stored columns */
@@ -1835,6 +1835,37 @@ btr_attach_half_pages(
 	btr_page_set_next(upper_page, upper_page_zip, next_page_no, mtr);
 }
 
+/*************************************************************//**
+Determine if a tuple is smaller than any record on the page.
+@return TRUE if smaller */
+static
+ibool
+btr_page_tuple_smaller(
+/*===================*/
+	btr_cur_t*	cursor,	/*!< in: b-tree cursor */
+	const dtuple_t*	tuple,	/*!< in: tuple to consider */
+	ulint*		offsets,/*!< in/out: temporary storage */
+	ulint		n_uniq,	/*!< in: number of unique fields
+				in the index page records */
+	mem_heap_t**	heap)	/*!< in/out: heap for offsets */
+{
+	buf_block_t*	block;
+	const rec_t*	first_rec;
+	page_cur_t	pcur;
+
+	/* Read the first user record in the page. */
+	block = btr_cur_get_block(cursor);
+	page_cur_set_before_first(block, &pcur);
+	page_cur_move_to_next(&pcur);
+	first_rec = page_cur_get_rec(&pcur);
+
+	offsets = rec_get_offsets(
+		first_rec, cursor->index, offsets,
+		n_uniq, heap);
+
+	return(cmp_dtuple_rec(tuple, first_rec, offsets) < 0);
+}
+
 /*************************************************************//**
 Splits an index page to halves and inserts the tuple. It is assumed
 that mtr holds an x-latch to the index tree. NOTE: the tree x-latch is
@@ -1909,49 +1940,45 @@ btr_page_split_and_insert(
 	if (n_iterations > 0) {
 		direction = FSP_UP;
 		hint_page_no = page_no + 1;
-		split_rec = btr_page_get_sure_split_rec(cursor, tuple, n_ext);
+		split_rec = btr_page_get_split_rec(cursor, tuple, n_ext);
 
+		if (UNIV_UNLIKELY(split_rec == NULL)) {
+			insert_left = btr_page_tuple_smaller(
+				cursor, tuple, offsets, n_uniq, &heap);
+		}
 	} else if (btr_page_get_split_rec_to_right(cursor, &split_rec)) {
 		direction = FSP_UP;
 		hint_page_no = page_no + 1;
+		insert_left = FALSE;
 
 	} else if (btr_page_get_split_rec_to_left(cursor, &split_rec)) {
 		direction = FSP_DOWN;
 		hint_page_no = page_no - 1;
+		ut_ad(split_rec);
 	} else {
 		direction = FSP_UP;
 		hint_page_no = page_no + 1;
 
-		if (page_get_n_recs(page) == 1) {
-			page_cur_t	pcur;
-
-			/* There is only one record in the index page
-			therefore we can't split the node in the middle
-			by default. We need to determine whether the
-			new record will be inserted to the left or right. */
+		/* If there is only one record in the index page, we
+		can't split the node in the middle by default. We need
+		to determine whether the new record will be inserted
+		to the left or right. */
 
-			/* Read the first (and only) record in the page. */
-			page_cur_set_before_first(block, &pcur);
-			page_cur_move_to_next(&pcur);
-			first_rec = page_cur_get_rec(&pcur);
-
-			offsets = rec_get_offsets(
-				first_rec, cursor->index, offsets,
-				n_uniq, &heap);
-
-			/* If the new record is less than the existing record
-			the split in the middle will copy the existing
-			record to the new node. */
-			if (cmp_dtuple_rec(tuple, first_rec, offsets) < 0) {
+		if (page_get_n_recs(page) > 1) {
 			split_rec = page_get_middle_rec(page);
+		} else if (btr_page_tuple_smaller(cursor, tuple,
+						  offsets, n_uniq, &heap)) {
+			split_rec = page_rec_get_next(
+				page_get_infimum_rec(page));
 		} else {
 			split_rec = NULL;
-			}
-		} else {
-			split_rec = page_get_middle_rec(page);
+			insert_left = FALSE;
 		}
 	}
 
+	/* At this point, insert_left is initialized if split_rec == NULL
+	and may be uninitialized otherwise. */
+
 	/* 2. Allocate a new page to the index */
 	new_block = btr_page_alloc(cursor->index, hint_page_no, direction,
 				   btr_page_get_level(page, mtr), mtr);
@@ -1978,11 +2005,11 @@ btr_page_split_and_insert(
 			avoid further splits by inserting the record
 			to an empty page. */
 			split_rec = NULL;
-			goto insert_right;
+			goto insert_empty;
 		}
 	} else {
-insert_right:
-		insert_left = FALSE;
+insert_empty:
+		ut_ad(!split_rec);
 		buf = mem_alloc(rec_get_converted_size(cursor->index,
 						       tuple, n_ext));
 
@@ -2019,7 +2046,17 @@ btr_page_split_and_insert(
 	}
 
 	/* 5. Move then the records to the new page */
-	if (direction == FSP_DOWN) {
+	if (direction == FSP_DOWN
+#ifdef UNIV_BTR_AVOID_COPY
+	    && page_rec_is_supremum(move_limit)) {
+		/* Instead of moving all records, make the new page
+		the empty page. */
+
+		left_block = block;
+		right_block = new_block;
+	} else if (direction == FSP_DOWN
+#endif /* UNIV_BTR_AVOID_COPY */
+		   ) {
 		/*		fputs("Split left\n", stderr); */
 
 		if (0
@@ -2062,6 +2099,14 @@ btr_page_split_and_insert(
 		right_block = block;
 
 		lock_update_split_left(right_block, left_block);
+#ifdef UNIV_BTR_AVOID_COPY
+	} else if (!split_rec) {
+		/* Instead of moving all records, make the new page
+		the empty page. */
+
+		left_block = new_block;
+		right_block = block;
+#endif /* UNIV_BTR_AVOID_COPY */
 	} else {
 		/*		fputs("Split right\n", stderr); */
 

storage/innobase/btr/btr0cur.c

@@ -1184,7 +1184,6 @@ btr_cur_optimistic_insert(
 	ibool		inherit;
 	ulint		zip_size;
 	ulint		rec_size;
-	mem_heap_t*	heap		= NULL;
 	ulint		err;
 
 	*big_rec = NULL;
@@ -1264,10 +1263,6 @@ btr_cur_optimistic_insert(
 					index, entry, big_rec_vec);
 			}
 
-			if (heap) {
-				mem_heap_free(heap);
-			}
-
 			return(DB_TOO_BIG_RECORD);
 		}
 	}
@@ -1290,10 +1285,6 @@ btr_cur_optimistic_insert(
 			dtuple_convert_back_big_rec(index, entry, big_rec_vec);
 		}
 
-		if (UNIV_LIKELY_NULL(heap)) {
-			mem_heap_free(heap);
-		}
-
 		return(err);
 	}
 
@@ -1364,10 +1355,6 @@ btr_cur_optimistic_insert(
 		}
 	}
 
-	if (UNIV_LIKELY_NULL(heap)) {
-		mem_heap_free(heap);
-	}
-
 #ifdef BTR_CUR_HASH_ADAPT
 	if (!reorg && leaf && (cursor->flag == BTR_CUR_HASH)) {
 		btr_search_update_hash_node_on_insert(cursor);

storage/innobase/data/data0data.c

@@ -666,6 +666,21 @@ dtuple_convert_big_rec(
 				goto skip_field;
 			}
 
+			/* In DYNAMIC and COMPRESSED format, store
+			locally any non-BLOB columns whose maximum
+			length does not exceed 256 bytes.  This is
+			because there is no room for the "external
+			storage" flag when the maximum length is 255
+			bytes or less. This restriction trivially
+			holds in REDUNDANT and COMPACT format, because
+			there we always store locally columns whose
+			length is up to local_len == 788 bytes.
+			@see rec_init_offsets_comp_ordinary */
+			if (ifield->col->mtype != DATA_BLOB
+			    && ifield->col->len < 256) {
+				goto skip_field;
+			}
+
 			longest_i = i;
 			longest = savings;
 

storage/innobase/dict/dict0boot.c

@@ -368,8 +368,8 @@ dict_boot(void)
 #if DICT_SYS_INDEXES_TYPE_FIELD != 4 + 2
 #error "DICT_SYS_INDEXES_TYPE_FIELD != 4 + 2"
 #endif
-#if DICT_SYS_INDEXES_NAME_FIELD != 1 + 2
-#error "DICT_SYS_INDEXES_NAME_FIELD != 1 + 2"
+#if DICT_SYS_INDEXES_NAME_FIELD != 2 + 2
+#error "DICT_SYS_INDEXES_NAME_FIELD != 2 + 2"
 #endif
 
 	table->id = DICT_INDEXES_ID;

storage/innobase/dict/dict0crea.c

@@ -1105,8 +1105,11 @@ dict_create_index_step(
 
 		dulint	index_id = node->index->id;
 
-		err = dict_index_add_to_cache(node->table, node->index,
-					      FIL_NULL, TRUE);
+		err = dict_index_add_to_cache(
+			node->table, node->index, FIL_NULL,
+			trx_is_strict(trx)
+			|| dict_table_get_format(node->table)
+			>= DICT_TF_FORMAT_ZIP);
 
 		node->index = dict_index_get_if_in_cache_low(index_id);
 		ut_a(!node->index == (err != DB_SUCCESS));

storage/innobase/handler/ha_innodb.cc

@@ -1971,6 +1971,19 @@ trx_is_interrupted(
 	return(trx && trx->mysql_thd && thd_killed((THD*) trx->mysql_thd));
 }
 
+/**********************************************************************//**
+Determines if the currently running transaction is in strict mode.
+@return	TRUE if strict */
+extern "C" UNIV_INTERN
+ibool
+trx_is_strict(
+/*==========*/
+	trx_t*	trx)	/*!< in: transaction */
+{
+	return(trx && trx->mysql_thd
+	       && THDVAR((THD*) trx->mysql_thd, strict_mode));
+}
+
 /**************************************************************//**
 Resets some fields of a prebuilt struct. The template is used in fast
 retrieval of just those column values MySQL needs in its processing. */
@@ -2283,7 +2296,7 @@ innobase_init(
 		}
 
 		sql_print_error("InnoDB: invalid value "
-				"innodb_file_format_check=%s",
+				"innodb_change_buffering=%s",
 				innobase_change_buffering);
 		goto mem_free_and_error;
 	}

storage/innobase/handler/handler0alter.cc

@@ -922,9 +922,8 @@ ha_innobase::add_index(
 		trx_commit_for_mysql(prebuilt->trx);
 	}
 
-	ut_d(dict_table_check_for_dup_indexes(innodb_table, FALSE));
-
 	if (dict_locked) {
+		ut_d(dict_table_check_for_dup_indexes(innodb_table, FALSE));
 		row_mysql_unlock_data_dictionary(trx);
 	}
 

storage/innobase/include/dict0boot.h

@@ -137,7 +137,7 @@ clustered index */
 #define DICT_SYS_INDEXES_PAGE_NO_FIELD	 8
 #define DICT_SYS_INDEXES_SPACE_NO_FIELD	 7
 #define DICT_SYS_INDEXES_TYPE_FIELD	 6
-#define DICT_SYS_INDEXES_NAME_FIELD	 3
+#define DICT_SYS_INDEXES_NAME_FIELD	 4
 
 /* When a row id which is zero modulo this number (which must be a power of
 two) is assigned, the field DICT_HDR_ROW_ID on the dictionary header page is

storage/innobase/include/que0que.h

@@ -381,6 +381,9 @@ struct que_thr_struct{
 					thus far */
 	ulint		lock_state;	/*!< lock state of thread (table or
 					row) */
+	struct srv_slot_struct*
+			slot;		/* The thread slot in the wait
+					array in srv_sys_t */
 };
 
 #define QUE_THR_MAGIC_N		8476583

storage/innobase/include/srv0que.h

-/*****************************************************************************
-
-Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
-
-This program is free software; you can redistribute it and/or modify it under
-the terms of the GNU General Public License as published by the Free Software
-Foundation; version 2 of the License.
-
-This program is distributed in the hope that it will be useful, but WITHOUT
-ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
-FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along with
-this program; if not, write to the Free Software Foundation, Inc., 59 Temple
-Place, Suite 330, Boston, MA 02111-1307 USA
-
-*****************************************************************************/
-
-/**************************************************//**
-@file include/srv0que.h
-Server query execution
-
-Created 6/5/1996 Heikki Tuuri
-*******************************************************/
-
-#ifndef srv0que_h
-#define srv0que_h
-
-#include "univ.i"
-#include "que0types.h"
-
-/**********************************************************************//**
-Enqueues a task to server task queue and releases a worker thread, if there
-is a suspended one. */
-UNIV_INTERN
-void
-srv_que_task_enqueue_low(
-/*=====================*/
-	que_thr_t*	thr);	/*!< in: query thread */
-
-#endif
-

storage/innobase/include/srv0srv.h

@@ -239,7 +239,6 @@ extern	ibool	srv_print_latch_waits;
 # define srv_print_latch_waits		FALSE
 #endif /* UNIV_DEBUG */
 
-extern ulint	srv_activity_count;
 extern ulint	srv_fatal_semaphore_wait_threshold;
 extern ulint	srv_dml_needed_delay;
 
@@ -314,12 +313,6 @@ typedef struct export_var_struct export_struc;
 /** Status variables to be passed to MySQL */
 extern export_struc export_vars;
 
-/** The server system */
-typedef struct srv_sys_struct	srv_sys_t;
-
-/** The server system */
-extern srv_sys_t*	srv_sys;
-
 # ifdef UNIV_PFS_THREAD
 /* Keys to register InnoDB threads with performance schema */
 extern mysql_pfs_key_t	trx_rollback_clean_thread_key;
@@ -421,6 +414,8 @@ enum srv_thread_type {
 			be biggest) */
 };
 
+struct srv_slot_struct;
+
 /*********************************************************************//**
 Boots Innobase server.
 @return	DB_SUCCESS or error code */
@@ -471,17 +466,6 @@ srv_set_io_thread_op_info(
 	const char*	str);	/*!< in: constant char string describing the
 				state */
 /*********************************************************************//**
-Releases threads of the type given from suspension in the thread table.
-NOTE! The server mutex has to be reserved by the caller!
-@return number of threads released: this may be less than n if not
-enough threads were suspended at the moment */
-UNIV_INTERN
-ulint
-srv_release_threads(
-/*================*/
-	enum srv_thread_type	type,	/*!< in: thread type */
-	ulint			n);	/*!< in: number of threads to release */
-/*********************************************************************//**
 The master thread controlling the server.
 @return	a dummy parameter */
 UNIV_INTERN
@@ -628,6 +612,13 @@ void
 srv_export_innodb_status(void);
 /*==========================*/
 
+/******************************************************************//**
+Increment the server activity counter. */
+UNIV_INTERN
+void
+srv_inc_activity_count(void);
+/*=========================*/
+
 /*********************************************************************//**
 Asynchronous purge thread.
 @return	a dummy parameter */
@@ -637,11 +628,23 @@ srv_purge_thread(
 /*=============*/
 	void*	arg __attribute__((unused))); /*!< in: a dummy parameter
 					      required by os_thread_create */
-/** Thread slot in the thread table */
-typedef struct srv_slot_struct	srv_slot_t;
 
-/** Thread table is an array of slots */
-typedef srv_slot_t	srv_table_t;
+/**********************************************************************//**
+Enqueues a task to server task queue and releases a worker thread, if there
+is a suspended one. */
+UNIV_INTERN
+void
+srv_que_task_enqueue_low(
+/*=====================*/
+	que_thr_t*	thr);	/*!< in: query thread */
+
+/**********************************************************************//**
+Check whether the master thread is active.
+@return FALSE is it is not active. */
+UNIV_INTERN
+ibool
+srv_is_master_thread_active(void);
+/*==============================*/
 
 /** Status variables to be passed to MySQL */
 struct export_var_struct{
@@ -697,14 +700,6 @@ struct export_var_struct{
 	ulint innodb_rows_deleted;		/*!< srv_n_rows_deleted */
 };
 
-/** The server system struct */
-struct srv_sys_struct{
-	srv_table_t*	threads;	/*!< server thread table */
-	UT_LIST_BASE_NODE_T(que_thr_t)
-			tasks;		/*!< task queue */
-};
-
-extern ulint	srv_n_threads_active[];
 #else /* !UNIV_HOTBACKUP */
 # define srv_use_adaptive_hash_indexes		FALSE
 # define srv_use_checksums			TRUE

storage/innobase/include/sync0sync.h

@@ -105,6 +105,7 @@ extern mysql_pfs_key_t	rw_lock_mutex_key;
 extern mysql_pfs_key_t	srv_dict_tmpfile_mutex_key;
 extern mysql_pfs_key_t	srv_innodb_monitor_mutex_key;
 extern mysql_pfs_key_t	srv_misc_tmpfile_mutex_key;
+extern mysql_pfs_key_t	srv_threads_mutex_key;
 extern mysql_pfs_key_t	srv_monitor_file_mutex_key;
 extern mysql_pfs_key_t	syn_arr_mutex_key;
 # ifdef UNIV_SYNC_DEBUG
@@ -587,6 +588,9 @@ Kernel mutex				If a kernel operation needs a file
 |					fsp x-latch before acquiring the kernel
 |					mutex.
 V
+Threads mutex				Thread scheduling mutex
+|
+V
 Search system mutex
 |
 V
@@ -657,8 +661,9 @@ or row lock! */
 /*------------------------------------- MySQL binlog mutex */
 /*-------------------------------*/
 #define	SYNC_KERNEL		300
-#define SYNC_REC_LOCK		299
-#define	SYNC_TRX_LOCK_HEAP	298
+#define SYNC_THREADS		299
+#define SYNC_REC_LOCK		298
+#define	SYNC_TRX_LOCK_HEAP	297
 #define SYNC_TRX_SYS_HEADER	290
 #define SYNC_LOG		170
 #define SYNC_LOG_FLUSH_ORDER	147

storage/innobase/include/trx0trx.h

@@ -391,6 +391,14 @@ ibool
 trx_is_interrupted(
 /*===============*/
 	trx_t*	trx);	/*!< in: transaction */
+/**********************************************************************//**
+Determines if the currently running transaction is in strict mode.
+@return	TRUE if strict */
+UNIV_INTERN
+ibool
+trx_is_strict(
+/*==========*/
+	trx_t*	trx);	/*!< in: transaction */
 #else /* !UNIV_HOTBACKUP */
 #define trx_is_interrupted(trx) FALSE
 #endif /* !UNIV_HOTBACKUP */

storage/innobase/include/univ.i

@@ -46,7 +46,7 @@ Created 1/20/1994 Heikki Tuuri
 
 #define INNODB_VERSION_MAJOR	1
 #define INNODB_VERSION_MINOR	1
-#define INNODB_VERSION_BUGFIX	0
+#define INNODB_VERSION_BUGFIX	1
 
 /* The following is the InnoDB version as shown in
 SELECT plugin_version FROM information_schema.plugins;
@@ -182,6 +182,9 @@ command. Not tested on Windows. */
 #define UNIV_COMPILE_TEST_FUNCS
 */
 
+#ifdef HAVE_purify
+# define UNIV_DEBUG_VALGRIND
+#endif /* HAVE_purify */
 #if 0
 #define UNIV_DEBUG_VALGRIND			/* Enable extra
 						Valgrind instrumentation */
@@ -219,6 +222,10 @@ operations (very slow); also UNIV_DEBUG must be defined */
 						adaptive hash index */
 #define UNIV_SRV_PRINT_LATCH_WAITS		/* enable diagnostic output
 						in sync0sync.c */
+#define UNIV_BTR_AVOID_COPY			/* when splitting B-tree nodes,
+						do not move any records when
+						all the records would
+						be moved */
 #define UNIV_BTR_PRINT				/* enable functions for
 						printing B-trees */
 #define UNIV_ZIP_DEBUG				/* extensive consistency checks

storage/innobase/log/log0log.c

@@ -3131,17 +3131,14 @@ logs_empty_and_mark_files_at_shutdown(void)
 		return; /* We SKIP ALL THE REST !! */
 	}
 
-	/* Check that the master thread is suspended */
-
-	if (srv_n_threads_active[SRV_MASTER] != 0) {
-
 	mutex_exit(&kernel_mutex);
 
+	/* Check that the master thread is suspended */
+
+	if (srv_is_master_thread_active()) {
 		goto loop;
 	}
 
-	mutex_exit(&kernel_mutex);
-
 	mutex_enter(&(log_sys->mutex));
 
 	if (log_sys->n_pending_checkpoint_writes
@@ -3199,18 +3196,14 @@ logs_empty_and_mark_files_at_shutdown(void)
 
 	mutex_exit(&(log_sys->mutex));
 
-	mutex_enter(&kernel_mutex);
 	/* Check that the master thread has stayed suspended */
-	if (srv_n_threads_active[SRV_MASTER] != 0) {
+	if (srv_is_master_thread_active()) {
 		fprintf(stderr,
 			"InnoDB: Warning: the master thread woke up"
 			" during shutdown\n");
 
-		mutex_exit(&kernel_mutex);
-
 		goto loop;
 	}
-	mutex_exit(&kernel_mutex);
 
 	fil_flush_file_spaces(FIL_TABLESPACE);
 	fil_flush_file_spaces(FIL_LOG);
@@ -3228,7 +3221,8 @@ logs_empty_and_mark_files_at_shutdown(void)
 	srv_shutdown_state = SRV_SHUTDOWN_LAST_PHASE;
 
 	/* Make some checks that the server really is quiet */
-	ut_a(srv_n_threads_active[SRV_MASTER] == 0);
+	ut_a(!srv_is_master_thread_active());
+
 	ut_a(buf_all_freed());
 	ut_a(lsn == log_sys->lsn);
 
@@ -3249,7 +3243,8 @@ logs_empty_and_mark_files_at_shutdown(void)
 	fil_close_all_files();
 
 	/* Make some checks that the server really is quiet */
-	ut_a(srv_n_threads_active[SRV_MASTER] == 0);
+	ut_a(!srv_is_master_thread_active());
+
 	ut_a(buf_all_freed());
 	ut_a(lsn == log_sys->lsn);
 }

storage/innobase/que/que0que.c

@@ -29,7 +29,6 @@ Created 5/27/1996 Heikki Tuuri
 #include "que0que.ic"
 #endif
 
-#include "srv0que.h"
 #include "usr0sess.h"
 #include "trx0trx.h"
 #include "trx0roll.h"
@@ -311,7 +310,9 @@ que_thr_end_wait_no_next_thr(
 	/* In MySQL we let the OS thread (not just the query thread) to wait
 	for the lock to be released: */
 
+	if (thr != NULL) {
 		srv_release_mysql_thread_if_suspended(thr);
+	}
 
 	/* srv_que_task_enqueue_low(thr); */
 }

storage/innobase/rem/rem0rec.c

@@ -212,6 +212,13 @@ rec_get_n_extern_new(
 			const dict_col_t*	col
 				= dict_field_get_col(field);
 			len = *lens--;
+			/* If the maximum length of the field is up
+			to 255 bytes, the actual length is always
+			stored in one byte. If the maximum length is
+			more than 255 bytes, the actual length is
+			stored in one byte for 0..127.  The length
+			will be encoded in two bytes when it is 128 or
+			more, or when the field is stored externally. */
 			if (UNIV_UNLIKELY(col->len > 255)
 			    || UNIV_UNLIKELY(col->mtype == DATA_BLOB)) {
 				if (len & 0x80) {
@@ -294,6 +301,13 @@ rec_init_offsets_comp_ordinary(
 			const dict_col_t*	col
 				= dict_field_get_col(field);
 			len = *lens--;
+			/* If the maximum length of the field is up
+			to 255 bytes, the actual length is always
+			stored in one byte. If the maximum length is
+			more than 255 bytes, the actual length is
+			stored in one byte for 0..127.  The length
+			will be encoded in two bytes when it is 128 or
+			more, or when the field is stored externally. */
 			if (UNIV_UNLIKELY(col->len > 255)
 			    || UNIV_UNLIKELY(col->mtype
 					     == DATA_BLOB)) {
@@ -425,6 +439,15 @@ rec_init_offsets(
 				const dict_col_t*	col
 					= dict_field_get_col(field);
 				len = *lens--;
+				/* If the maximum length of the field
+				is up to 255 bytes, the actual length
+				is always stored in one byte. If the
+				maximum length is more than 255 bytes,
+				the actual length is stored in one
+				byte for 0..127.  The length will be
+				encoded in two bytes when it is 128 or
+				more, or when the field is stored
+				externally. */
 				if (UNIV_UNLIKELY(col->len > 255)
 				    || UNIV_UNLIKELY(col->mtype
 						     == DATA_BLOB)) {
@@ -647,6 +670,13 @@ rec_get_offsets_reverse(
 			const dict_col_t*	col
 				= dict_field_get_col(field);
 			len = *lens++;
+			/* If the maximum length of the field is up
+			to 255 bytes, the actual length is always
+			stored in one byte. If the maximum length is
+			more than 255 bytes, the actual length is
+			stored in one byte for 0..127.  The length
+			will be encoded in two bytes when it is 128 or
+			more, or when the field is stored externally. */
 			if (UNIV_UNLIKELY(col->len > 255)
 			    || UNIV_UNLIKELY(col->mtype == DATA_BLOB)) {
 				if (len & 0x80) {
@@ -781,12 +811,20 @@ rec_get_converted_size_comp_prefix(
 
 		ut_ad(len <= col->len || col->mtype == DATA_BLOB);
 
+		/* If the maximum length of a variable-length field
+		is up to 255 bytes, the actual length is always stored
+		in one byte. If the maximum length is more than 255
+		bytes, the actual length is stored in one byte for
+		0..127.  The length will be encoded in two bytes when
+		it is 128 or more, or when the field is stored externally. */
+
 		if (field->fixed_len) {
 			ut_ad(len == field->fixed_len);
 			/* dict_index_add_col() should guarantee this */
 			ut_ad(!field->prefix_len
 			      || field->fixed_len == field->prefix_len);
 		} else if (dfield_is_ext(&fields[i])) {
+			ut_ad(col->len >= 256 || col->mtype == DATA_BLOB);
 			extra_size += 2;
 		} else if (len < 128
 			   || (col->len < 256 && col->mtype != DATA_BLOB)) {
@@ -1086,6 +1124,8 @@ rec_convert_dtuple_to_rec_comp(
 	/* Store the data and the offsets */
 
 	for (i = 0, field = fields; i < n_fields; i++, field++) {
+		const dict_field_t*	ifield;
+
 		type = dfield_get_type(field);
 		len = dfield_get_len(field);
 
@@ -1120,12 +1160,20 @@ rec_convert_dtuple_to_rec_comp(
 		/* only nullable fields can be null */
 		ut_ad(!dfield_is_null(field));
 
-		fixed_len = dict_index_get_nth_field(index, i)->fixed_len;
-
+		ifield = dict_index_get_nth_field(index, i);
+		fixed_len = ifield->fixed_len;
+		/* If the maximum length of a variable-length field
+		is up to 255 bytes, the actual length is always stored
+		in one byte. If the maximum length is more than 255
+		bytes, the actual length is stored in one byte for
+		0..127.  The length will be encoded in two bytes when
+		it is 128 or more, or when the field is stored externally. */
 		if (fixed_len) {
 			ut_ad(len == fixed_len);
 			ut_ad(!dfield_is_ext(field));
 		} else if (dfield_is_ext(field)) {
+			ut_ad(ifield->col->len >= 256
+			      || ifield->col->mtype == DATA_BLOB);
 			ut_ad(len <= REC_MAX_INDEX_COL_LEN
 			      + BTR_EXTERN_FIELD_REF_SIZE);
 			*lens-- = (byte) (len >> 8) | 0xc0;
@@ -1215,11 +1263,20 @@ rec_convert_dtuple_to_rec(
 		mem_heap_t*	heap	= NULL;
 		ulint		offsets_[REC_OFFS_NORMAL_SIZE];
 		const ulint*	offsets;
+		ulint		i;
 		rec_offs_init(offsets_);
 
 		offsets = rec_get_offsets(rec, index,
 					  offsets_, ULINT_UNDEFINED, &heap);
 		ut_ad(rec_validate(rec, offsets));
+		ut_ad(dtuple_get_n_fields(dtuple)
+		      == rec_offs_n_fields(offsets));
+
+		for (i = 0; i < rec_offs_n_fields(offsets); i++) {
+			ut_ad(!dfield_is_ext(dtuple_get_nth_field(dtuple, i))
+			      == !rec_offs_nth_extern(offsets, i));
+		}
+
 		if (UNIV_LIKELY_NULL(heap)) {
 			mem_heap_free(heap);
 		}
@@ -1402,6 +1459,13 @@ rec_copy_prefix_to_buf(
 			prefix_len += field->fixed_len;
 		} else {
 			ulint	len = *lens--;
+			/* If the maximum length of the column is up
+			to 255 bytes, the actual length is always
+			stored in one byte. If the maximum length is
+			more than 255 bytes, the actual length is
+			stored in one byte for 0..127.  The length
+			will be encoded in two bytes when it is 128 or
+			more, or when the column is stored externally. */
 			if (col->len > 255 || col->mtype == DATA_BLOB) {
 				if (len & 0x80) {
 					/* 1exxxxxx */

storage/innobase/row/row0undo.c

@@ -341,7 +341,7 @@ row_undo_step(
 
 	ut_ad(thr);
 
-	srv_activity_count++;
+	srv_inc_activity_count();
 
 	trx = thr_get_trx(thr);
 

storage/innobase/srv/srv0que.c

-/*****************************************************************************
-
-Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
-
-This program is free software; you can redistribute it and/or modify it under
-the terms of the GNU General Public License as published by the Free Software
-Foundation; version 2 of the License.
-
-This program is distributed in the hope that it will be useful, but WITHOUT
-ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
-FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
-
-You should have received a copy of the GNU General Public License along with
-this program; if not, write to the Free Software Foundation, Inc., 59 Temple
-Place, Suite 330, Boston, MA 02111-1307 USA
-
-*****************************************************************************/
-
-/**************************************************//**
-@file srv/srv0que.c
-Server query execution
-
-Created 6/5/1996 Heikki Tuuri
-*******************************************************/
-
-#include "srv0que.h"
-
-#include "srv0srv.h"
-#include "sync0sync.h"
-#include "os0thread.h"
-#include "usr0sess.h"
-#include "que0que.h"
-
-/**********************************************************************//**
-Enqueues a task to server task queue and releases a worker thread, if there
-is a suspended one. */
-UNIV_INTERN
-void
-srv_que_task_enqueue_low(
-/*=====================*/
-	que_thr_t*	thr)	/*!< in: query thread */
-{
-	ut_ad(thr);
-	ut_ad(mutex_own(&kernel_mutex));
-
-	UT_LIST_ADD_LAST(queue, srv_sys->tasks, thr);
-
-	srv_release_threads(SRV_WORKER, 1);
-}

storage/innobase/srv/srv0srv.c

@@ -68,7 +68,6 @@ Created 10/8/1995 Heikki Tuuri
 #include "sync0sync.h"
 #include "thr0loc.h"
 #include "que0que.h"
-#include "srv0que.h"
 #include "log0recv.h"
 #include "pars0pars.h"
 #include "usr0sess.h"
@@ -90,10 +89,6 @@ Created 10/8/1995 Heikki Tuuri
 affects only FOREIGN KEY definition parsing */
 UNIV_INTERN ibool	srv_lower_case_table_names	= FALSE;
 
-/* The following counter is incremented whenever there is some user activity
-in the server */
-UNIV_INTERN ulint	srv_activity_count	= 0;
-
 /* The following is the maximum allowed duration of a lock wait. */
 UNIV_INTERN ulint	srv_fatal_semaphore_wait_threshold = 600;
 
@@ -324,16 +319,18 @@ concurrency check. */
 
 UNIV_INTERN ulong	srv_thread_concurrency	= 0;
 
-/* this mutex protects srv_conc data structures */
-UNIV_INTERN os_fast_mutex_t	srv_conc_mutex;
 /* number of transactions that have declared_to_be_inside_innodb set.
 It used to be a non-error for this value to drop below zero temporarily.
 This is no longer true. We'll, however, keep the lint datatype to add
 assertions to catch any corner cases that we may have missed. */
 UNIV_INTERN lint	srv_conc_n_threads	= 0;
+
+/* this mutex protects srv_conc data structures */
+static os_fast_mutex_t	srv_conc_mutex;
+
 /* number of OS threads waiting in the FIFO for a permission to enter
 InnoDB */
-UNIV_INTERN ulint	srv_conc_n_waiting_threads = 0;
+static ulint	srv_conc_n_waiting_threads = 0;
 
 typedef struct srv_conc_slot_struct	srv_conc_slot_t;
 struct srv_conc_slot_struct{
@@ -351,9 +348,9 @@ struct srv_conc_slot_struct{
 };
 
 /* queue of threads waiting to get in */
-UNIV_INTERN UT_LIST_BASE_NODE_T(srv_conc_slot_t)	srv_conc_queue;
+static UT_LIST_BASE_NODE_T(srv_conc_slot_t)	srv_conc_queue;
 /* array of wait slots */
-UNIV_INTERN srv_conc_slot_t* srv_conc_slots;
+static srv_conc_slot_t* srv_conc_slots;
 
 /* Number of times a thread is allowed to enter InnoDB within the same
 SQL query after it has once got the ticket at srv_conc_enter_innodb */
@@ -443,6 +440,8 @@ UNIV_INTERN mysql_pfs_key_t	srv_monitor_file_mutex_key;
 UNIV_INTERN mysql_pfs_key_t	srv_dict_tmpfile_mutex_key;
 /* Key to register the mutex with performance schema */
 UNIV_INTERN mysql_pfs_key_t	srv_misc_tmpfile_mutex_key;
+/* Key to register srv_sys_t::mutex with performance schema */
+UNIV_INTERN mysql_pfs_key_t	srv_srv_sys_mutex_key;
 #endif /* UNIV_PFS_MUTEX */
 
 /* Temporary file for innodb monitor output */
@@ -491,6 +490,19 @@ intervals. Following macros define thresholds for these conditions. */
 #define SRV_RECENT_IO_ACTIVITY	(PCT_IO(5))
 #define SRV_PAST_IO_ACTIVITY	(PCT_IO(200))
 
+/** Acquire the system_mutex. */
+#define srv_sys_mutex_enter() do {		\
+	mutex_enter(&srv_sys->mutex);		\
+} while (0)
+
+/** Test if the system mutex is owned. */
+#define srv_sys_mutex_own() mutex_own(&srv_sys->mutex)
+
+/** Release the system mutex. */
+#define srv_sys_mutex_exit() do {		\
+	mutex_exit(&srv_sys->mutex);		\
+} while (0)
+
 /*
 	IMPLEMENTATION OF THE SERVER MAIN PROGRAM
 	=========================================
@@ -659,7 +671,7 @@ boosted at least to normal. This priority requirement can be seen similar to
 the privileged mode used when processing the kernel calls in traditional
 Unix.*/
 
-/* Thread slot in the thread table */
+/** Thread slot in the thread table.  */
 struct srv_slot_struct{
 	os_thread_id_t	id;		/*!< thread id */
 	os_thread_t	handle;		/*!< thread handle */
@@ -675,12 +687,43 @@ struct srv_slot_struct{
 					used for MySQL threads) */
 };
 
-/* Table for MySQL threads where they will be suspended to wait for locks */
-UNIV_INTERN srv_slot_t*	srv_mysql_table = NULL;
+/** Thread slot in the thread table */
+typedef struct srv_slot_struct	srv_slot_t;
+
+/** Thread table is an array of slots */
+typedef srv_slot_t	srv_table_t;
+
+/** The server system */
+typedef struct srv_sys_struct	srv_sys_t;
+
+/** The server system struct */
+struct srv_sys_struct{
+	mutex_t		mutex;			/*!< variable protecting the
+						fields in this structure. */
+	srv_table_t*	sys_threads;		/*!< server thread table */
+
+	UT_LIST_BASE_NODE_T(que_thr_t)
+			tasks;			/*!< task queue */
+
+	ulint		n_threads[SRV_MASTER + 1];
+						/*!< number of system threads
+						in a thread class */
+
+	ulint		n_threads_active[SRV_MASTER + 1];
+						/*!< number of threads active
+						in a thread class */
+
+	srv_slot_t*	waiting_threads;	/*!< Array  of user threads
+						suspended while waiting for
+					       	locks within InnoDB */
+	srv_slot_t*	last_slot;		/*!< highest slot ever used
+						in the waiting_threads array */
+	ulint		activity_count;		/*!< For tracking server
+						activity */
+};
 
 UNIV_INTERN os_event_t	srv_lock_timeout_thread_event;
 
-UNIV_INTERN srv_sys_t*	srv_sys	= NULL;
 
 /* padding to prevent other memory update hotspots from residing on
 the same memory cache line */
@@ -691,6 +734,8 @@ UNIV_INTERN mutex_t*	kernel_mutex_temp;
 the same memory cache line */
 UNIV_INTERN byte	srv_pad2[64];
 
+static srv_sys_t*	srv_sys	= NULL;
+
 #if 0
 /* The following three values measure the urgency of the jobs of
 buffer, version, and insert threads. They may vary from 0 - 1000.
@@ -705,13 +750,6 @@ static ulint	srv_meter_high_water2[SRV_MASTER + 1];
 static ulint	srv_meter_foreground[SRV_MASTER + 1];
 #endif
 
-/* The following values give info about the activity going on in
-the database. They are protected by the server mutex. The arrays
-are indexed by the type of the thread. */
-
-UNIV_INTERN ulint	srv_n_threads_active[SRV_MASTER + 1];
-UNIV_INTERN ulint	srv_n_threads[SRV_MASTER + 1];
-
 /*********************************************************************//**
 Asynchronous purge thread.
 @return	a dummy parameter */
@@ -764,14 +802,15 @@ srv_table_get_nth_slot(
 /*===================*/
 	ulint	index)		/*!< in: index of the slot */
 {
+	ut_ad(srv_sys_mutex_own());
 	ut_a(index < OS_THREAD_MAX_N);
 
-	return(srv_sys->threads + index);
+	return(srv_sys->sys_threads + index);
 }
 
 /*********************************************************************//**
 Gets the number of threads in the system.
-@return	sum of srv_n_threads[] */
+@return	sum of srv_sys_t::n_threads[] */
 UNIV_INTERN
 ulint
 srv_get_n_threads(void)
@@ -780,14 +819,14 @@ srv_get_n_threads(void)
 	ulint	i;
 	ulint	n_threads	= 0;
 
-	mutex_enter(&kernel_mutex);
+	srv_sys_mutex_enter();
 
 	for (i = SRV_COM; i < SRV_MASTER + 1; i++) {
 
-		n_threads += srv_n_threads[i];
+		n_threads += srv_sys->n_threads[i];
 	}
 
-	mutex_exit(&kernel_mutex);
+	srv_sys_mutex_exit();
 
 	return(n_threads);
 }
@@ -806,6 +845,8 @@ srv_table_reserve_slot(
 	srv_slot_t*	slot;
 	ulint		i;
 
+	ut_ad(srv_sys_mutex_own());
+
 	ut_a(type > 0);
 	ut_a(type <= SRV_MASTER);
 
@@ -846,7 +887,7 @@ srv_suspend_thread(void)
 	ulint			slot_no;
 	enum srv_thread_type	type;
 
-	ut_ad(mutex_own(&kernel_mutex));
+	srv_sys_mutex_enter();
 
 	slot_no = thr_local_get_slot_no(os_thread_get_curr_id());
 
@@ -867,12 +908,14 @@ srv_suspend_thread(void)
 
 	slot->suspended = TRUE;
 
-	ut_ad(srv_n_threads_active[type] > 0);
+	ut_ad(srv_sys->n_threads_active[type] > 0);
 
-	srv_n_threads_active[type]--;
+	srv_sys->n_threads_active[type]--;
 
 	os_event_reset(event);
 
+	srv_sys_mutex_exit();
+
 	return(event);
 }
 
@@ -881,23 +924,24 @@ Releases threads of the type given from suspension in the thread table.
 NOTE! The server mutex has to be reserved by the caller!
 @return number of threads released: this may be less than n if not
 enough threads were suspended at the moment */
-UNIV_INTERN
+static
 ulint
 srv_release_threads(
 /*================*/
 	enum srv_thread_type	type,	/*!< in: thread type */
 	ulint			n)	/*!< in: number of threads to release */
 {
-	srv_slot_t*	slot;
 	ulint		i;
 	ulint		count	= 0;
 
 	ut_ad(type >= SRV_WORKER);
 	ut_ad(type <= SRV_MASTER);
 	ut_ad(n > 0);
-	ut_ad(mutex_own(&kernel_mutex));
+
+	srv_sys_mutex_enter();
 
 	for (i = 0; i < OS_THREAD_MAX_N; i++) {
+		srv_slot_t*	slot;
 
 		slot = srv_table_get_nth_slot(i);
 
@@ -905,7 +949,7 @@ srv_release_threads(
 
 			slot->suspended = FALSE;
 
-			srv_n_threads_active[type]++;
+			srv_sys->n_threads_active[type]++;
 
 			os_event_set(slot->event);
 
@@ -925,6 +969,8 @@ srv_release_threads(
 		}
 	}
 
+	srv_sys_mutex_exit();
+
 	return(count);
 }
 
@@ -940,7 +986,7 @@ srv_get_thread_type(void)
 	srv_slot_t*		slot;
 	enum srv_thread_type	type;
 
-	mutex_enter(&kernel_mutex);
+	srv_sys_mutex_enter();
 
 	slot_no = thr_local_get_slot_no(os_thread_get_curr_id());
 
@@ -951,7 +997,7 @@ srv_get_thread_type(void)
 	ut_ad(type >= SRV_WORKER);
 	ut_ad(type <= SRV_MASTER);
 
-	mutex_exit(&kernel_mutex);
+	srv_sys_mutex_exit();
 
 	return(type);
 }
@@ -963,11 +1009,14 @@ void
 srv_init(void)
 /*==========*/
 {
-	srv_conc_slot_t*	conc_slot;
-	srv_slot_t*		slot;
 	ulint			i;
+	srv_conc_slot_t*	conc_slot;
+	ulint			srv_sys_sz;
 
-	srv_sys = mem_alloc(sizeof(srv_sys_t));
+	srv_sys_sz = sizeof(*srv_sys)
+	       	   + (OS_THREAD_MAX_N * sizeof(srv_slot_t) * 2);
+
+	srv_sys = mem_zalloc(srv_sys_sz);
 
 	kernel_mutex_temp = mem_alloc(sizeof(mutex_t));
 	mutex_create(kernel_mutex_key, &kernel_mutex, SYNC_KERNEL);
@@ -975,42 +1024,30 @@ srv_init(void)
 	mutex_create(srv_innodb_monitor_mutex_key,
 		     &srv_innodb_monitor_mutex, SYNC_NO_ORDER_CHECK);
 
-	srv_sys->threads = mem_alloc(OS_THREAD_MAX_N * sizeof(srv_slot_t));
+	mutex_create(srv_srv_sys_mutex_key, &srv_sys->mutex, SYNC_THREADS);
 
-	for (i = 0; i < OS_THREAD_MAX_N; i++) {
-		slot = srv_table_get_nth_slot(i);
-		slot->in_use = FALSE;
-		slot->type=0;	/* Avoid purify errors */
-		slot->event = os_event_create(NULL);
-		ut_a(slot->event);
-	}
+	srv_sys_mutex_enter();
 
-	srv_mysql_table = mem_alloc(OS_THREAD_MAX_N * sizeof(srv_slot_t));
+	srv_sys->sys_threads = (srv_slot_t*) &srv_sys[1];
+	srv_sys->waiting_threads = srv_sys->sys_threads + OS_THREAD_MAX_N;
+	srv_sys->last_slot = srv_sys->waiting_threads;
 
 	for (i = 0; i < OS_THREAD_MAX_N; i++) {
-		slot = srv_mysql_table + i;
-		slot->in_use = FALSE;
-		slot->type = 0;
+		srv_slot_t*	slot;
+
+		slot = srv_table_get_nth_slot(i);
+
 		slot->event = os_event_create(NULL);
+
 		ut_a(slot->event);
 	}
 
 	srv_lock_timeout_thread_event = os_event_create(NULL);
 
-	for (i = 0; i < SRV_MASTER + 1; i++) {
-		srv_n_threads_active[i] = 0;
-		srv_n_threads[i] = 0;
-#if 0
-		srv_meter[i] = 30;
-		srv_meter_low_water[i] = 50;
-		srv_meter_high_water[i] = 100;
-		srv_meter_high_water2[i] = 200;
-		srv_meter_foreground[i] = 250;
-#endif
-	}
-
 	UT_LIST_INIT(srv_sys->tasks);
 
+	srv_sys_mutex_exit();
+
 	/* Create dummy indexes for infimum and supremum records */
 
 	dict_ind_init();
@@ -1045,14 +1082,11 @@ srv_free(void)
 	mem_free(srv_conc_slots);
 	srv_conc_slots = NULL;
 
-	mem_free(srv_sys->threads);
 	mem_free(srv_sys);
 	srv_sys = NULL;
 
 	mem_free(kernel_mutex_temp);
 	kernel_mutex_temp = NULL;
-	mem_free(srv_mysql_table);
-	srv_mysql_table = NULL;
 
 	trx_i_s_cache_free(trx_i_s_cache);
 }
@@ -1408,27 +1442,107 @@ srv_boot(void)
 	return(DB_SUCCESS);
 }
 
+/*********************************************************************//**
+Print the contents of the srv_sys_t::waiting_threads array. */
+static
+void
+srv_print_mysql_threads(void)
+/*=========================*/
+{
+	ulint	i;
+
+	for (i = 0; i < OS_THREAD_MAX_N; i++) {
+		srv_slot_t*	slot;
+
+		slot = srv_sys->waiting_threads + i;
+
+		fprintf(stderr,
+			"Slot %lu: thread id %lu, type %lu,"
+			" in use %lu, susp %lu, time %lu\n",
+			(ulong) i,
+			(ulong) os_thread_pf(slot->id),
+			(ulong) slot->type,
+			(ulong) slot->in_use,
+			(ulong) slot->suspended,
+			(ulong) difftime(ut_time(), slot->suspend_time));
+	}
+}
+
+/*********************************************************************//**
+Release a slot in the srv_sys_t::waiting_threads. Adjust the array last pointer
+if there are empty slots towards the end of the table. */
+static
+void
+srv_table_release_slot_for_mysql(
+/*=============================*/
+	srv_slot_t*	slot)		/*!< in: slot to release */
+{
+#ifdef UNIV_DEBUG
+	srv_slot_t*	upper = srv_sys->waiting_threads + OS_THREAD_MAX_N;
+#endif /* UNIV_DEBUG */
+
+	srv_sys_mutex_enter();
+
+	ut_a(slot->in_use);
+	ut_a(slot->thr != NULL);
+	ut_a(slot->thr->slot != NULL);
+	ut_a(slot->thr->slot == slot);
+
+	/* Must be within the array boundaries. */
+	ut_ad(slot >= srv_sys->waiting_threads);
+	ut_ad(slot < upper);
+
+	slot->thr->slot = NULL;
+	slot->thr = NULL;
+	slot->in_use = FALSE;
+
+	/* Scan backwards and adjust the last free slot pointer. */
+	for (slot = srv_sys->last_slot;
+	     slot > srv_sys->waiting_threads && !slot->in_use;
+	     --slot) {
+		/* No op */
+	}
+
+	/* Either the array is empty or the last scanned slot is in use. */
+	ut_ad(slot->in_use || slot == srv_sys->waiting_threads);
+
+	srv_sys->last_slot = slot + 1;
+
+	/* The last slot is either outside of the array boundry or it's
+	on an empty slot. */
+	ut_ad(srv_sys->last_slot == upper || !srv_sys->last_slot->in_use);
+
+	ut_ad(srv_sys->last_slot >= srv_sys->waiting_threads);
+	ut_ad(srv_sys->last_slot <= upper);
+
+	srv_sys_mutex_exit();
+}
+
 /*********************************************************************//**
 Reserves a slot in the thread table for the current MySQL OS thread.
-NOTE! The kernel mutex has to be reserved by the caller!
 @return	reserved slot */
 static
 srv_slot_t*
-srv_table_reserve_slot_for_mysql(void)
-/*==================================*/
+srv_table_reserve_slot_for_mysql(
+/*=============================*/
+	que_thr_t*	thr)		/*!< in: query thread associated
+					with the MySQL OS thread */
 {
-	srv_slot_t*	slot;
 	ulint		i;
+	srv_slot_t*	slot;
 
-	ut_ad(mutex_own(&kernel_mutex));
+	srv_sys_mutex_enter();
 
-	i = 0;
-	slot = srv_mysql_table + i;
+	slot = srv_sys->waiting_threads;
 
-	while (slot->in_use) {
-		i++;
+	for (i = 0; i < OS_THREAD_MAX_N; ++i, ++slot) {
+		if (!slot->in_use) {
+			break;
+		}
+	}
 
-		if (i >= OS_THREAD_MAX_N) {
+	/* Check if we have run out of slots. */
+	if (slot == srv_sys->waiting_threads+ OS_THREAD_MAX_N) {
 
 		ut_print_timestamp(stderr);
 
@@ -1442,33 +1556,36 @@ srv_table_reserve_slot_for_mysql(void)
 			"InnoDB: on Linux. But first we print"
 			" a list of waiting threads.\n", (ulong) i);
 
-			for (i = 0; i < OS_THREAD_MAX_N; i++) {
+		srv_print_mysql_threads();
 
-				slot = srv_mysql_table + i;
+		ut_error;
+	} else {
 
-				fprintf(stderr,
-					"Slot %lu: thread id %lu, type %lu,"
-					" in use %lu, susp %lu, time %lu\n",
-					(ulong) i,
-					(ulong) os_thread_pf(slot->id),
-					(ulong) slot->type,
-					(ulong) slot->in_use,
-					(ulong) slot->suspended,
-					(ulong) difftime(ut_time(),
-							 slot->suspend_time));
+		ut_a(slot->in_use == FALSE);
+
+		slot->in_use = TRUE;
+		slot->thr = thr;
+		slot->thr->slot = slot;
+		slot->id = os_thread_get_curr_id();
+		slot->handle = os_thread_get_curr();
+
+		if (slot->event == NULL) {
+			slot->event = os_event_create(NULL);
+			ut_a(slot->event);
 		}
 
-			ut_error;
+		os_event_reset(slot->event);
+		slot->suspended = TRUE;
+		slot->suspend_time = ut_time();
 	}
 
-		slot = srv_mysql_table + i;
+	if (slot == srv_sys->last_slot) {
+		++srv_sys->last_slot;
 	}
 
-	ut_a(slot->in_use == FALSE);
+	ut_ad(srv_sys->last_slot <= srv_sys->waiting_threads+ OS_THREAD_MAX_N);
 
-	slot->in_use = TRUE;
-	slot->id = os_thread_get_curr_id();
-	slot->handle = os_thread_get_curr();
+	srv_sys_mutex_exit();
 
 	return(slot);
 }
@@ -1487,7 +1604,6 @@ srv_suspend_mysql_thread(
 				OS thread */
 {
 	srv_slot_t*	slot;
-	os_event_t	event;
 	double		wait_time;
 	trx_t*		trx;
 	ulint		had_dict_lock;
@@ -1529,15 +1645,7 @@ srv_suspend_mysql_thread(
 
 	ut_ad(thr->is_active == FALSE);
 
-	slot = srv_table_reserve_slot_for_mysql();
-
-	event = slot->event;
-
-	slot->thr = thr;
-
-	os_event_reset(event);
-
-	slot->suspend_time = ut_time();
+	slot = srv_table_reserve_slot_for_mysql(thr);
 
 	if (thr->lock_state == QUE_THR_LOCK_ROW) {
 		srv_n_lock_wait_count++;
@@ -1583,7 +1691,7 @@ srv_suspend_mysql_thread(
 
 	/* Suspend this thread and wait for the event. */
 
-	os_event_wait(event);
+	os_event_wait(slot->event);
 
 	/* After resuming, reacquire the data dictionary latch if
 	necessary. */
@@ -1604,13 +1712,13 @@ srv_suspend_mysql_thread(
 		srv_conc_force_enter_innodb(trx);
 	}
 
+	wait_time = ut_difftime(ut_time(), slot->suspend_time);
+
 	mutex_enter(&kernel_mutex);
 
 	/* Release the slot for others to use */
 
-	slot->in_use = FALSE;
-
-	wait_time = ut_difftime(ut_time(), slot->suspend_time);
+	srv_table_release_slot_for_mysql(slot);
 
 	if (thr->lock_state == QUE_THR_LOCK_ROW) {
 		if (ut_usectime(&sec, &ms) == -1) {
@@ -1663,25 +1771,13 @@ srv_release_mysql_thread_if_suspended(
 	que_thr_t*	thr)	/*!< in: query thread associated with the
 				MySQL OS thread	 */
 {
-	srv_slot_t*	slot;
-	ulint		i;
-
 	ut_ad(mutex_own(&kernel_mutex));
 
-	for (i = 0; i < OS_THREAD_MAX_N; i++) {
-
-		slot = srv_mysql_table + i;
+	if (thr->slot != NULL) {
+		ut_a(thr->slot->in_use);
 
-		if (slot->in_use && slot->thr == thr) {
-			/* Found */
-
-			os_event_set(slot->event);
-
-			return;
-		}
+		os_event_set(thr->slot->event);
 	}
-
-	/* not found */
 }
 
 /******************************************************************//**
@@ -2156,6 +2252,83 @@ srv_monitor_thread(
 	OS_THREAD_DUMMY_RETURN;
 }
 
+/*********************************************************************//**
+Check if the thread lock wait has timed out. Release its locks if the
+wait has actually timed out. */
+UNIV_INTERN
+void
+srv_lock_check_wait(
+/*================*/
+	srv_slot_t*	slot)
+{
+	trx_t*		trx;
+	double		wait_time;
+	ulong		lock_wait_timeout;
+	ib_time_t	suspend_time = slot->suspend_time;
+
+	ut_ad(srv_sys_mutex_own());
+
+	wait_time = ut_difftime(ut_time(), suspend_time);
+
+	trx = thr_get_trx(slot->thr);
+
+	lock_wait_timeout = thd_lock_wait_timeout(trx->mysql_thd);
+
+	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
+		by the transaction and release possible
+		other transactions waiting behind; it is
+		possible that the lock has already been
+		granted: in that case do nothing */
+
+		if (trx->wait_lock) {
+			trx_t*	slot_trx;
+
+			/* Release the srv_sys_t->mutex to preserve the
+			latch order only. */
+			srv_sys_mutex_exit();
+
+			/* It is possible that the thread has already
+			freed its slot and released its locks and another
+			thread is now using this slot. We need to
+			check whether the slot is still in use by the
+			same thread before cancelling the wait and releasing
+		       	the locks. */
+
+			mutex_enter(&kernel_mutex);
+
+			srv_sys_mutex_enter();
+
+			slot_trx = thr_get_trx(slot->thr);
+
+			/* We can't compare the pointers here because the
+			memory can be recycled. Transaction ids are not
+			recyled and therefore safe to use. We also check if
+			the transaction suspend time is the same that we
+			used for calculating the wait earlier. If the
+		       	transaction has already released its locks there
+		       	is nothing more we can do. */
+			if (slot->in_use
+			    && suspend_time == slot->suspend_time
+			    && ut_dulint_cmp(trx->id, slot_trx->id) == 0
+			    && trx->wait_lock != NULL) {
+
+				ut_a(trx->que_state == TRX_QUE_LOCK_WAIT);
+
+				lock_cancel_waiting_and_release(
+					trx->wait_lock);
+			}
+
+			mutex_exit(&kernel_mutex);
+		}
+	}
+}
+
 /*********************************************************************//**
 A thread which wakes up threads whose lock wait may have lasted too long.
 @return	a dummy parameter */
@@ -2169,8 +2342,6 @@ srv_lock_timeout_thread(
 {
 	srv_slot_t*	slot;
 	ibool		some_waits;
-	double		wait_time;
-	ulint		i;
 
 #ifdef UNIV_PFS_THREAD
 	pfs_register_thread(srv_lock_timeout_thread_key);
@@ -2184,52 +2355,26 @@ srv_lock_timeout_thread(
 
 	srv_lock_timeout_active = TRUE;
 
-	mutex_enter(&kernel_mutex);
+	srv_sys_mutex_enter();
 
 	some_waits = FALSE;
 
-	/* Check of all slots if a thread is waiting there, and if it
-	has exceeded the time limit */
+	/* Check all slots for user threads that are waiting on locks, and
+	if they have exceeded the time limit. */
 
-	for (i = 0; i < OS_THREAD_MAX_N; i++) {
-
-		slot = srv_mysql_table + i;
+	for (slot = srv_sys->waiting_threads;
+	     slot < srv_sys->last_slot;
+	     ++slot) {
 
 		if (slot->in_use) {
-			trx_t*	trx;
-			ulong	lock_wait_timeout;
-
 			some_waits = TRUE;
-
-			wait_time = ut_difftime(ut_time(), slot->suspend_time);
-
-			trx = thr_get_trx(slot->thr);
-			lock_wait_timeout = thd_lock_wait_timeout(
-				trx->mysql_thd);
-
-			if (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
-				by the transaction and release possible
-				other transactions waiting behind; it is
-				possible that the lock has already been
-				granted: in that case do nothing */
-
-				if (trx->wait_lock) {
-					lock_cancel_waiting_and_release(
-						trx->wait_lock);
-				}
-			}
+			srv_lock_check_wait(slot);
 		}
 	}
 
 	os_event_reset(srv_lock_timeout_thread_event);
 
-	mutex_exit(&kernel_mutex);
+	srv_sys_mutex_exit();
 
 	if (srv_shutdown_state >= SRV_SHUTDOWN_CLEANUP) {
 		goto exit_func;
@@ -2367,26 +2512,68 @@ srv_error_monitor_thread(
 	OS_THREAD_DUMMY_RETURN;
 }
 
+/******************************************************************//**
+Increment the server activity count. */
+UNIV_INLINE
+void
+srv_inc_activity_count_low(void)
+/*============================*/
+{
+	srv_sys_mutex_enter();
+
+	++srv_sys->activity_count;
+
+	srv_sys_mutex_exit();
+}
+
+/******************************************************************//**
+Increment the server activity count. */
+UNIV_INTERN
+void
+srv_inc_activity_count(void)
+/*========================*/
+{
+	srv_inc_activity_count_low();
+}
+
+/**********************************************************************//**
+Check whether the master thread is active.
+@return FALSE is it is not active. */
+UNIV_INTERN
+ibool
+srv_is_master_thread_active(void)
+/*=============================*/
+{
+	ibool	ret;
+
+	srv_sys_mutex_enter();
+
+	ret = srv_sys->n_threads_active[SRV_MASTER] != 0;
+
+	srv_sys_mutex_exit();
+
+	return(ret);
+}
+
 /*******************************************************************//**
 Tells the InnoDB server that there has been activity in the database
 and wakes up the master thread if it is suspended (not sleeping). Used
 in the MySQL interface. Note that there is a small chance that the master
-thread stays suspended (we do not protect our operation with the kernel
-mutex, for performace reasons). */
+thread stays suspended (we do not protect our operation with the
+srv_sys_t->mutex, for performance reasons). */
 UNIV_INTERN
 void
 srv_active_wake_master_thread(void)
 /*===============================*/
 {
-	srv_activity_count++;
+	ut_ad(!mutex_own(&kernel_mutex));
+	ut_ad(!srv_sys_mutex_own());
 
-	if (srv_n_threads_active[SRV_MASTER] == 0) {
+	srv_inc_activity_count_low();
 
-		mutex_enter(&kernel_mutex);
+	if (srv_sys->n_threads_active[SRV_MASTER] == 0) {
 
 		srv_release_threads(SRV_MASTER, 1);
-
-		mutex_exit(&kernel_mutex);
 	}
 }
 
@@ -2394,23 +2581,20 @@ srv_active_wake_master_thread(void)
 Tells the purge thread that there has been activity in the database
 and wakes up the purge thread if it is suspended (not sleeping).  Note
 that there is a small chance that the purge thread stays suspended
-(we do not protect our operation with the kernel mutex, for
-performace reasons). */
+(we do not protect our operation with the srv_sys_t:mutex, for
+performance reasons). */
 UNIV_INTERN
 void
 srv_wake_purge_thread_if_not_active(void)
 /*=====================================*/
 {
 	ut_ad(!mutex_own(&kernel_mutex));
+	ut_ad(!srv_sys_mutex_own());
 
 	if (srv_n_purge_threads > 0
-	    && srv_n_threads_active[SRV_WORKER] == 0) {
-
-		mutex_enter(&kernel_mutex);
+	    && srv_sys->n_threads_active[SRV_WORKER] == 0) {
 
 		srv_release_threads(SRV_WORKER, 1);
-
-		mutex_exit(&kernel_mutex);
 	}
 }
 
@@ -2421,13 +2605,12 @@ void
 srv_wake_master_thread(void)
 /*========================*/
 {
-	srv_activity_count++;
+	ut_ad(!mutex_own(&kernel_mutex));
+	ut_ad(!srv_sys_mutex_own());
 
-	mutex_enter(&kernel_mutex);
+	srv_inc_activity_count_low();
 
 	srv_release_threads(SRV_MASTER, 1);
-
-	mutex_exit(&kernel_mutex);
 }
 
 /*******************************************************************//**
@@ -2438,17 +2621,34 @@ srv_wake_purge_thread(void)
 /*=======================*/
 {
 	ut_ad(!mutex_own(&kernel_mutex));
+	ut_ad(!srv_sys_mutex_own());
 
 	if (srv_n_purge_threads > 0) {
 
-		mutex_enter(&kernel_mutex);
-
 		srv_release_threads(SRV_WORKER, 1);
-
-		mutex_exit(&kernel_mutex);
 	}
 }
 
+/*******************************************************************//**
+Check if there has been any activity.
+@return FALSE if no hange in activity counter. */
+UNIV_INLINE
+ibool
+srv_check_activity(
+/*===============*/
+	ulint		old_activity_count)	/*!< old activity count */
+{
+	ibool		ret;
+
+	srv_sys_mutex_enter();
+
+	ret = srv_sys->activity_count != old_activity_count;
+
+	srv_sys_mutex_exit();
+
+	return(ret);
+}
+
 /**********************************************************************
 The master thread is tasked to ensure that flush of log file happens
 once every second in the background. This is to ensure that not more
@@ -2535,13 +2735,13 @@ srv_master_thread(
 	srv_main_thread_process_no = os_proc_get_number();
 	srv_main_thread_id = os_thread_pf(os_thread_get_curr_id());
 
-	srv_table_reserve_slot(SRV_MASTER);
+	srv_sys_mutex_enter();
 
-	mutex_enter(&kernel_mutex);
+	srv_table_reserve_slot(SRV_MASTER);
 
-	srv_n_threads_active[SRV_MASTER]++;
+	srv_sys->n_threads_active[SRV_MASTER]++;
 
-	mutex_exit(&kernel_mutex);
+	srv_sys_mutex_exit();
 
 loop:
 	/*****************************************************************/
@@ -2553,12 +2753,13 @@ srv_master_thread(
 	buf_get_total_stat(&buf_stat);
 	n_ios_very_old = log_sys->n_log_ios + buf_stat.n_pages_read
 		+ buf_stat.n_pages_written;
-	mutex_enter(&kernel_mutex);
+
+	srv_sys_mutex_enter();
 
 	/* Store the user activity counter at the start of this loop */
-	old_activity_count = srv_activity_count;
+	old_activity_count = srv_sys->activity_count;
 
-	mutex_exit(&kernel_mutex);
+	srv_sys_mutex_exit();
 
 	if (srv_force_recovery >= SRV_FORCE_NO_BACKGROUND) {
 
@@ -2664,7 +2865,7 @@ srv_master_thread(
 			}
 		}
 
-		if (srv_activity_count == old_activity_count) {
+		if (srv_sys->activity_count == old_activity_count) {
 
 			/* There is no user activity at the moment, go to
 			the background loop */
@@ -2755,18 +2956,13 @@ srv_master_thread(
 
 	srv_main_thread_op_info = "reserving kernel mutex";
 
-	mutex_enter(&kernel_mutex);
-
 	/* ---- When there is database activity, we jump from here back to
 	the start of loop */
 
-	if (srv_activity_count != old_activity_count) {
-		mutex_exit(&kernel_mutex);
+	if (srv_check_activity(old_activity_count)) {
 		goto loop;
 	}
 
-	mutex_exit(&kernel_mutex);
-
 	/* If the database is quiet, we enter the background loop */
 
 	/*****************************************************************/
@@ -2799,12 +2995,9 @@ srv_master_thread(
 
 	srv_main_thread_op_info = "reserving kernel mutex";
 
-	mutex_enter(&kernel_mutex);
-	if (srv_activity_count != old_activity_count) {
-		mutex_exit(&kernel_mutex);
+	if (srv_check_activity(old_activity_count)) {
 		goto loop;
 	}
-	mutex_exit(&kernel_mutex);
 
 	srv_main_thread_op_info = "doing insert buffer merge";
 
@@ -2821,12 +3014,9 @@ srv_master_thread(
 
 	srv_main_thread_op_info = "reserving kernel mutex";
 
-	mutex_enter(&kernel_mutex);
-	if (srv_activity_count != old_activity_count) {
-		mutex_exit(&kernel_mutex);
+	if (srv_check_activity(old_activity_count)) {
 		goto loop;
 	}
-	mutex_exit(&kernel_mutex);
 
 flush_loop:
 	srv_main_thread_op_info = "flushing buffer pool pages";
@@ -2843,12 +3033,9 @@ srv_master_thread(
 
 	srv_main_thread_op_info = "reserving kernel mutex";
 
-	mutex_enter(&kernel_mutex);
-	if (srv_activity_count != old_activity_count) {
-		mutex_exit(&kernel_mutex);
+	if (srv_check_activity(old_activity_count)) {
 		goto loop;
 	}
-	mutex_exit(&kernel_mutex);
 
 	srv_main_thread_op_info = "waiting for buffer pool flush to end";
 	buf_flush_wait_batch_end(NULL, BUF_FLUSH_LIST);
@@ -2870,12 +3057,10 @@ srv_master_thread(
 
 	srv_main_thread_op_info = "reserving kernel mutex";
 
-	mutex_enter(&kernel_mutex);
-	if (srv_activity_count != old_activity_count) {
-		mutex_exit(&kernel_mutex);
+	if (srv_check_activity(old_activity_count)) {
 		goto loop;
 	}
-	mutex_exit(&kernel_mutex);
+
 	/*
 	srv_main_thread_op_info = "archiving log (if log archive is on)";
 
@@ -2921,10 +3106,10 @@ srv_master_thread(
 		goto loop;
 	}
 
-	event = srv_suspend_thread();
-
 	mutex_exit(&kernel_mutex);
 
+	event = srv_suspend_thread();
+
 	/* DO NOT CHANGE THIS STRING. innobase_start_or_create_for_mysql()
 	waits for database activity to die down when converting < 4.1.x
 	databases, and relies on this string being exactly as it is. InnoDB
@@ -2974,13 +3159,13 @@ srv_purge_thread(
 		os_thread_pf(os_thread_get_curr_id()));
 #endif /* UNIV_DEBUG_THREAD_CREATION */
 
-	mutex_enter(&kernel_mutex);
+	srv_sys_mutex_enter();
 
 	slot_no = srv_table_reserve_slot(SRV_WORKER);
 
-	++srv_n_threads_active[SRV_WORKER];
+	++srv_sys->n_threads_active[SRV_WORKER];
 
-	mutex_exit(&kernel_mutex);
+	srv_sys_mutex_exit();
 
 	while (srv_shutdown_state != SRV_SHUTDOWN_EXIT_THREADS) {
 
@@ -2996,12 +3181,8 @@ srv_purge_thread(
 
 			os_event_t	event;
 
-			mutex_enter(&kernel_mutex);
-
 			event = srv_suspend_thread();
 
-			mutex_exit(&kernel_mutex);
-
 			os_event_wait(event);
 		}
 
@@ -3030,13 +3211,13 @@ srv_purge_thread(
 	/* Free the thread local memory. */
 	thr_local_free(os_thread_get_curr_id());
 
-	mutex_enter(&kernel_mutex);
+	srv_sys_mutex_enter();
 
 	/* Free the slot for reuse. */
 	slot = srv_table_get_nth_slot(slot_no);
 	slot->in_use = FALSE;
 
-	mutex_exit(&kernel_mutex);
+	srv_sys_mutex_exit();
 
 #ifdef UNIV_DEBUG_THREAD_CREATION
 	fprintf(stderr, "InnoDB: Purge thread exiting, id %lu\n",
@@ -3049,3 +3230,24 @@ srv_purge_thread(
 
 	OS_THREAD_DUMMY_RETURN;	/* Not reached, avoid compiler warning */
 }
+
+/**********************************************************************//**
+Enqueues a task to server task queue and releases a worker thread, if there
+is a suspended one. */
+UNIV_INTERN
+void
+srv_que_task_enqueue_low(
+/*=====================*/
+	que_thr_t*	thr)	/*!< in: query thread */
+{
+	ut_ad(thr);
+
+	srv_sys_mutex_enter();
+
+	UT_LIST_ADD_LAST(queue, srv_sys->tasks, thr);
+
+	srv_sys_mutex_exit();
+
+	srv_release_threads(SRV_WORKER, 1);
+}
+

storage/innobase/sync/sync0sync.c

@@ -1167,6 +1167,7 @@ sync_thread_add_level(
 	case SYNC_SEARCH_SYS_CONF:
 	case SYNC_TRX_LOCK_HEAP:
 	case SYNC_KERNEL:
+	case SYNC_THREADS:
 	case SYNC_IBUF_BITMAP_MUTEX:
 	case SYNC_RSEG:
 	case SYNC_TRX_UNDO:

storage/innobase/trx/trx0rec.c

@@ -350,8 +350,13 @@ trx_undo_rec_get_col_val(
 
 		ut_ad(*orig_len >= BTR_EXTERN_FIELD_REF_SIZE);
 		ut_ad(*len > *orig_len);
-		ut_ad(*len >= REC_MAX_INDEX_COL_LEN
+		/* @see dtuple_convert_big_rec() */
+		ut_ad(*len >= BTR_EXTERN_FIELD_REF_SIZE * 2);
+		/* we do not have access to index->table here
+		ut_ad(dict_table_get_format(index->table) >= DICT_TF_FORMAT_ZIP
+		      || *len >= REC_MAX_INDEX_COL_LEN
 		      + BTR_EXTERN_FIELD_REF_SIZE);
+		*/
 
 		*len += UNIV_EXTERN_STORAGE_FIELD;
 		break;
@@ -1075,13 +1080,17 @@ trx_undo_rec_get_partial_row(
 			/* If the prefix of this column is indexed,
 			ensure that enough prefix is stored in the
 			undo log record. */
-			ut_a(ignore_prefix
-			     || !col->ord_part
+			if (!ignore_prefix && col->ord_part) {
+				ut_a(dfield_get_len(dfield)
+				     >= 2 * BTR_EXTERN_FIELD_REF_SIZE);
+				ut_a(dict_table_get_format(index->table)
+				     >= DICT_TF_FORMAT_ZIP
 				     || dfield_get_len(dfield)
 				     >= REC_MAX_INDEX_COL_LEN
 				     + BTR_EXTERN_FIELD_REF_SIZE);
 			}
 		}
+	}
 
 	return(ptr);
 }

storage/innobase/trx/trx0roll.c

@@ -37,7 +37,6 @@ Created 3/26/1996 Heikki Tuuri
 #include "trx0rec.h"
 #include "que0que.h"
 #include "usr0sess.h"
-#include "srv0que.h"
 #include "srv0start.h"
 #include "row0undo.h"
 #include "row0mysql.h"