Applied InnoDB snapshot innodb-5.1-ss2298

Fixes the following bugs:

- Bug #33349: possible race condition revolving around data dictionary and repartitioning
  Introduce retry/sleep logic as a workaround for a transient bug
  where ::open fails for partitioned tables randomly if we are using
  one file per table.

- Bug #34053: normal users can enable innodb_monitor logging
  In CREATE TABLE and DROP TABLE check whether the table in question is one
  of the magic innodb_monitor tables and whether the user has enough rights
  to mess with it before doing anything else.

- Bug #22868: 'Thread thrashing' with > 50 concurrent conns under an upd-intensive workloadw
- Bug #29560: InnoDB >= 5.0.30 hangs on adaptive hash rw-lock 'waiting for an X-lock'
  This is a combination of changes that forward port the scalability fix applied to 5.0
  through r1001.
  It reverts changes r149 and r122 (these were 5.1 specific changes made in lieu of
  scalability fix of 5.0)
  Then it applies r1001 to 5.0 which is the original scalability fix.
  Finally it applies r2082 which fixes an issue with the original fix.

- Bug #30930: Add auxiliary function to retrieve THD::thread_id
  Add thd_get_thread_id() function.  Also make check_global_access() function
  visible to InnoDB under INNODB_COMPATIBILITY_HOOKS #define.

include/mysql/plugin.h

@@ -714,6 +714,16 @@ int mysql_tmpfile(const char *prefix);
 */
 int thd_killed(const MYSQL_THD thd);
 
+
+/**
+  Return the thread id of a user thread
+
+  @param thd  user thread connection handle
+  @return  thread id
+*/
+unsigned long thd_get_thread_id(const MYSQL_THD thd);
+
+
 /**
   Allocate memory in the connection's local memory pool
 

mysql-test/r/innodb.result

@@ -1062,6 +1062,19 @@ select count(*) from t2 /* must be 2 as restored after rollback caused by the er
 count(*)
 2
 drop table t1, t2;
+drop table if exists t1, t2;
+CREATE TABLE t1 (a int, PRIMARY KEY (a));
+CREATE TABLE t2 (a int, PRIMARY KEY (a)) ENGINE=InnoDB;
+create trigger trg_del_t2 after  delete on t2 for each row
+insert into t1 values (1);
+insert into t1 values (1);
+insert into t2 values (1),(2);
+delete t2 from t2;
+ERROR 23000: Duplicate entry '1' for key 'PRIMARY'
+select count(*) from t2 /* must be 2 as restored after rollback caused by the error */;
+count(*)
+2
+drop table t1, t2;
 create table t1 (a int, b int) engine=innodb;
 insert into t1 values(20,null);
 select t2.b, ifnull(t2.b,"this is null") from t1 as t2 left join t1 as t3 on
@@ -1727,10 +1740,10 @@ Variable_name	Value
 Innodb_page_size	16384
 show status like "Innodb_rows_deleted";
 Variable_name	Value
-Innodb_rows_deleted	70
+Innodb_rows_deleted	71
 show status like "Innodb_rows_inserted";
 Variable_name	Value
-Innodb_rows_inserted	1082
+Innodb_rows_inserted	1084
 show status like "Innodb_rows_updated";
 Variable_name	Value
 Innodb_rows_updated	885

mysql-test/r/innodb_bug34053.result

+SET storage_engine=InnoDB;

mysql-test/t/innodb.test

@@ -752,6 +752,38 @@ select count(*) from t2 /* must be 2 as restored after rollback caused by the er
 drop table t1, t2;
 
 
+#
+# Bug #29136  	erred multi-delete on trans table does not rollback 
+#
+
+# prepare
+--disable_warnings
+drop table if exists t1, t2;
+--enable_warnings
+CREATE TABLE t1 (a int, PRIMARY KEY (a));
+CREATE TABLE t2 (a int, PRIMARY KEY (a)) ENGINE=InnoDB;
+create trigger trg_del_t2 after  delete on t2 for each row
+       insert into t1 values (1);
+insert into t1 values (1);
+insert into t2 values (1),(2);
+
+
+# exec cases A, B - see multi_update.test
+
+# A. send_error() w/o send_eof() branch
+
+--error ER_DUP_ENTRY
+delete t2 from t2;
+
+# check
+
+select count(*) from t2 /* must be 2 as restored after rollback caused by the error */;
+
+# cleanup bug#29136
+
+drop table t1, t2;
+
+
 #
 # Testing of IFNULL
 #

mysql-test/t/innodb_bug34053.test

+#
+# Make sure http://bugs.mysql.com/34053 remains fixed.
+#
+
+-- source include/have_innodb.inc
+
+SET storage_engine=InnoDB;
+
+# we do not really care about what gets printed, we are only
+# interested in getting success or failure according to our
+# expectations
+-- disable_query_log
+-- disable_result_log
+
+GRANT USAGE ON *.* TO 'shane'@'localhost' IDENTIFIED BY '12345';
+FLUSH PRIVILEGES;
+
+-- connect (con1,localhost,shane,12345,)
+
+-- connection con1
+-- error ER_SPECIFIC_ACCESS_DENIED_ERROR
+CREATE TABLE innodb_monitor (a INT) ENGINE=INNODB;
+-- error ER_SPECIFIC_ACCESS_DENIED_ERROR
+CREATE TABLE innodb_mem_validate (a INT) ENGINE=INNODB;
+CREATE TABLE innodb_monitorx (a INT) ENGINE=INNODB;
+DROP TABLE innodb_monitorx;
+CREATE TABLE innodb_monito (a INT) ENGINE=INNODB;
+DROP TABLE innodb_monito;
+CREATE TABLE xinnodb_monitor (a INT) ENGINE=INNODB;
+DROP TABLE xinnodb_monitor;
+CREATE TABLE nnodb_monitor (a INT) ENGINE=INNODB;
+DROP TABLE nnodb_monitor;
+
+-- connection default
+CREATE TABLE innodb_monitor (a INT) ENGINE=INNODB;
+CREATE TABLE innodb_mem_validate (a INT) ENGINE=INNODB;
+
+-- connection con1
+-- error ER_SPECIFIC_ACCESS_DENIED_ERROR
+DROP TABLE innodb_monitor;
+-- error ER_SPECIFIC_ACCESS_DENIED_ERROR
+DROP TABLE innodb_mem_validate;
+
+-- connection default
+DROP TABLE innodb_monitor;
+DROP TABLE innodb_mem_validate;
+DROP USER 'shane'@'localhost';
+
+-- disconnect con1

sql/mysql_priv.h

@@ -1035,7 +1035,6 @@ bool check_access(THD *thd, ulong access, const char *db, ulong *save_priv,
 		  bool no_grant, bool no_errors, bool schema_db);
 bool check_table_access(THD *thd, ulong want_access, TABLE_LIST *tables,
 			uint number, bool no_errors);
-bool check_global_access(THD *thd, ulong want_access);
 #else
 inline bool check_access(THD *thd, ulong access, const char *db,
                          ulong *save_priv, bool no_grant, bool no_errors,
@@ -1048,9 +1047,18 @@ inline bool check_access(THD *thd, ulong access, const char *db,
 inline bool check_table_access(THD *thd, ulong want_access, TABLE_LIST *tables,
 			uint number, bool no_errors)
 { return false; }
+#endif /*NO_EMBEDDED_ACCESS_CHECKS*/
+
+#endif /* MYSQL_SERVER */
+#if defined MYSQL_SERVER || defined INNODB_COMPATIBILITY_HOOKS
+#ifndef NO_EMBEDDED_ACCESS_CHECKS
+bool check_global_access(THD *thd, ulong want_access);
+#else
 inline bool check_global_access(THD *thd, ulong want_access)
 { return false; }
 #endif /*NO_EMBEDDED_ACCESS_CHECKS*/
+#endif /* MYSQL_SERVER || INNODB_COMPATIBILITY_HOOKS */
+#ifdef MYSQL_SERVER
 
 /*
   Support routine for SQL parser on partitioning syntax

sql/sql_class.cc

@@ -682,6 +682,7 @@ void *thd_memdup(MYSQL_THD thd, const void* str, unsigned int size)
   return thd->memdup(str, size);
 }
 
+extern "C"
 void thd_get_xid(const MYSQL_THD thd, MYSQL_XID *xid)
 {
   *xid = *(MYSQL_XID *) &thd->transaction.xid_state.xid;
@@ -2760,6 +2761,17 @@ extern "C" int thd_killed(const MYSQL_THD thd)
   return(thd->killed);
 }
 
+/**
+  Return the thread id of a user thread
+  @param thd user thread
+  @return thread id
+*/
+extern "C" unsigned long thd_get_thread_id(const MYSQL_THD thd)
+{
+  return((unsigned long)thd->thread_id);
+}
+
+
 #ifdef INNODB_COMPATIBILITY_HOOKS
 extern "C" struct charset_info_st *thd_charset(MYSQL_THD thd)
 {

storage/innobase/btr/btr0cur.c

@@ -52,7 +52,7 @@ can be released by page reorganize, then it is reorganized */
 
 #define BTR_CUR_PAGE_REORGANIZE_LIMIT	(UNIV_PAGE_SIZE / 32)
 
-/* When estimating number of different kay values in an index sample
+/* When estimating number of different key values in an index, sample
 this many index pages */
 #define BTR_KEY_VAL_ESTIMATE_N_PAGES	8
 

storage/innobase/handler/ha_innodb.cc

@@ -129,7 +129,7 @@ static my_bool	innobase_locks_unsafe_for_binlog	= FALSE;
 static my_bool	innobase_rollback_on_timeout		= FALSE;
 static my_bool	innobase_create_status_file		= FALSE;
 static my_bool innobase_stats_on_metadata		= TRUE;
-static my_bool	innobase_use_adaptive_hash_indexes	= TRUE;
+static my_bool	innobase_adaptive_hash_index	= TRUE;
 
 static char*	internal_innobase_data_file_path	= NULL;
 
@@ -1627,7 +1627,7 @@ innobase_init(
 	srv_stats_on_metadata = (ibool) innobase_stats_on_metadata;
 
 	srv_use_adaptive_hash_indexes =
-		(ibool) innobase_use_adaptive_hash_indexes;
+		(ibool) innobase_adaptive_hash_index;
 
 	srv_print_verbose_log = mysqld_embedded ? 0 : 1;
 
@@ -2275,6 +2275,8 @@ ha_innobase::open(
 	dict_table_t*	ib_table;
 	char		norm_name[1000];
 	THD*		thd;
+	ulint		retries = 0;
+	char*		is_part = NULL;
 
 	DBUG_ENTER("ha_innobase::open");
 
@@ -2308,11 +2310,29 @@ ha_innobase::open(
 		DBUG_RETURN(1);
 	}
 
+	/* We look for pattern #P# to see if the table is partitioned
+	MySQL table. The retry logic for partitioned tables is a
+	workaround for http://bugs.mysql.com/bug.php?id=33349. Look
+	at support issue https://support.mysql.com/view.php?id=21080
+	for more details. */
+	is_part = strstr(norm_name, "#P#");
+retry:
 	/* Get pointer to a table object in InnoDB dictionary cache */
-
 	ib_table = dict_table_get(norm_name, TRUE);
 	
 	if (NULL == ib_table) {
+		if (is_part && retries < 10) {
+			++retries;
+			os_thread_sleep(100000);
+			goto retry;
+		}
+
+		if (is_part) {
+			sql_print_error("Failed to open table %s after "
+					"%lu attemtps.\n", norm_name,
+					retries);
+		}
+
 		sql_print_error("Cannot find or open table %s from\n"
 				"the internal data dictionary of InnoDB "
 				"though the .frm file for the\n"
@@ -4625,6 +4645,12 @@ innodb_check_for_record_too_big_error(
 	}
 }
 
+/* limit innodb monitor access to users with PROCESS privilege.
+See http://bugs.mysql.com/32710 for expl. why we choose PROCESS. */
+#define IS_MAGIC_TABLE_AND_USER_DENIED_ACCESS(table_name, thd) \
+	(row_is_magic_monitor_table(table_name) \
+	 && check_global_access(thd, PROCESS_ACL))
+
 /*********************************************************************
 Creates a table definition to an InnoDB database. */
 static
@@ -4661,6 +4687,12 @@ create_table_def(
 	DBUG_ENTER("create_table_def");
 	DBUG_PRINT("enter", ("table_name: %s", table_name));
 
+	ut_a(trx->mysql_thd != NULL);
+	if (IS_MAGIC_TABLE_AND_USER_DENIED_ACCESS(table_name,
+						  (THD*) trx->mysql_thd)) {
+		DBUG_RETURN(HA_ERR_GENERIC);
+	}
+
 	n_cols = form->s->fields;
 
 	/* We pass 0 as the space id, and determine at a lower level the space
@@ -5201,6 +5233,14 @@ ha_innobase::delete_table(
 
 	DBUG_ENTER("ha_innobase::delete_table");
 
+	/* Strangely, MySQL passes the table name without the '.frm'
+	extension, in contrast to ::create */
+	normalize_table_name(norm_name, name);
+
+	if (IS_MAGIC_TABLE_AND_USER_DENIED_ACCESS(norm_name, thd)) {
+		DBUG_RETURN(HA_ERR_GENERIC);
+	}
+
 	/* Get the transaction associated with the current thd, or create one
 	if not yet created */
 
@@ -5234,11 +5274,6 @@ ha_innobase::delete_table(
 
 	assert(name_len < 1000);
 
-	/* Strangely, MySQL passes the table name without the '.frm'
-	extension, in contrast to ::create */
-
-	normalize_table_name(norm_name, name);
-
 	/* Drop the table in InnoDB */
 
 	error = row_drop_table_for_mysql(norm_name, trx,
@@ -7988,9 +8023,10 @@ static MYSQL_SYSVAR_BOOL(stats_on_metadata, innobase_stats_on_metadata,
   "Enable statistics gathering for metadata commands such as SHOW TABLE STATUS (on by default)",
   NULL, NULL, TRUE);
 
-static MYSQL_SYSVAR_BOOL(use_adaptive_hash_indexes, innobase_use_adaptive_hash_indexes,
+static MYSQL_SYSVAR_BOOL(adaptive_hash_index, innobase_adaptive_hash_index,
   PLUGIN_VAR_OPCMDARG | PLUGIN_VAR_READONLY,
-  "Enable the InnoDB adaptive hash indexes (enabled by default)",
+  "Enable InnoDB adaptive hash index (enabled by default).  "
+  "Disable with --skip-innodb-adaptive-hash-index.",
   NULL, NULL, TRUE);
 
 static MYSQL_SYSVAR_LONG(additional_mem_pool_size, innobase_additional_mem_pool_size,
@@ -8080,10 +8116,11 @@ static MYSQL_SYSVAR_STR(data_file_path, innobase_data_file_path,
 
 static MYSQL_SYSVAR_LONG(autoinc_lock_mode, innobase_autoinc_lock_mode,
   PLUGIN_VAR_RQCMDARG | PLUGIN_VAR_READONLY,
-  "The AUTOINC lock modes supported by InnoDB:\n"
-  "  0 => Old style AUTOINC locking (for backward compatibility)\n"
-  "  1 => New style AUTOINC locking\n"
-  "  2 => No AUTOINC locking (unsafe for SBR)",
+  "The AUTOINC lock modes supported by InnoDB:               "
+  "0 => Old style AUTOINC locking (for backward"
+  " compatibility)                                           "
+  "1 => New style AUTOINC locking                            "
+  "2 => No AUTOINC locking (unsafe for SBR)",
   NULL, NULL,
   AUTOINC_NEW_STYLE_LOCKING,	/* Default setting */
   AUTOINC_OLD_STYLE_LOCKING,	/* Minimum value */
@@ -8121,7 +8158,7 @@ static struct st_mysql_sys_var* innobase_system_variables[]= {
   MYSQL_SYSVAR(open_files),
   MYSQL_SYSVAR(rollback_on_timeout),
   MYSQL_SYSVAR(stats_on_metadata),
-  MYSQL_SYSVAR(use_adaptive_hash_indexes),
+  MYSQL_SYSVAR(adaptive_hash_index),
   MYSQL_SYSVAR(status_file),
   MYSQL_SYSVAR(support_xa),
   MYSQL_SYSVAR(sync_spin_loops),

storage/innobase/include/os0sync.h

@@ -112,9 +112,13 @@ os_event_set(
 	os_event_t	event);	/* in: event to set */
 /**************************************************************
 Resets an event semaphore to the nonsignaled state. Waiting threads will
-stop to wait for the event. */
+stop to wait for the event.
+The return value should be passed to os_even_wait_low() if it is desired
+that this thread should not wait in case of an intervening call to
+os_event_set() between this os_event_reset() and the
+os_event_wait_low() call. See comments for os_event_wait_low(). */
 
-void
+ib_longlong
 os_event_reset(
 /*===========*/
 	os_event_t	event);	/* in: event to reset */
@@ -125,16 +129,38 @@ void
 os_event_free(
 /*==========*/
 	os_event_t	event);	/* in: event to free */
+
 /**************************************************************
 Waits for an event object until it is in the signaled state. If
 srv_shutdown_state == SRV_SHUTDOWN_EXIT_THREADS this also exits the
 waiting thread when the event becomes signaled (or immediately if the
-event is already in the signaled state). */
+event is already in the signaled state).
+
+Typically, if the event has been signalled after the os_event_reset()
+we'll return immediately because event->is_set == TRUE.
+There are, however, situations (e.g.: sync_array code) where we may
+lose this information. For example:
+
+thread A calls os_event_reset()
+thread B calls os_event_set()   [event->is_set == TRUE]
+thread C calls os_event_reset() [event->is_set == FALSE]
+thread A calls os_event_wait()  [infinite wait!]
+thread C calls os_event_wait()  [infinite wait!]
+
+Where such a scenario is possible, to avoid infinite wait, the
+value returned by os_event_reset() should be passed in as
+reset_sig_count. */
+
+#define os_event_wait(event) os_event_wait_low((event), 0)
 
 void
-os_event_wait(
-/*==========*/
-	os_event_t	event);	/* in: event to wait */
+os_event_wait_low(
+/*==============*/
+	os_event_t	event,		/* in: event to wait */
+	ib_longlong	reset_sig_count);/* in: zero or the value
+					returned by previous call of
+					os_event_reset(). */
+
 /**************************************************************
 Waits for an event object until it is in the signaled state or
 a timeout is exceeded. In Unix the timeout is always infinite. */

storage/innobase/include/read0read.h

@@ -111,10 +111,6 @@ struct read_view_struct{
 	dulint	undo_no;	/* (0, 0) or if type is VIEW_HIGH_GRANULARITY
 				transaction undo_no when this high-granularity
 				consistent read view was created */
-	ibool	can_be_too_old;	/* TRUE if the system has had to purge old
-				versions which this read view should be able
-				to access: the read view can bump into the
-				DB_MISSING_HISTORY error */
 	dulint	low_limit_no;	/* The view does not need to see the undo
 				logs for transactions whose transaction number
 				is strictly smaller (<) than this value: they

storage/innobase/include/row0mysql.h

@@ -319,7 +319,7 @@ row_mysql_unfreeze_data_dictionary(
 /*===============================*/
 	trx_t*	trx);	/* in: transaction */
 /*************************************************************************
-Drops a table for MySQL. If the name of the table ends in
+Creates a table for MySQL. If the name of the table ends in
 one of "innodb_monitor", "innodb_lock_monitor", "innodb_tablespace_monitor",
 "innodb_table_monitor", then this will also start the printing of monitor
 output by the master thread. If the table name ends in "innodb_mem_validate",
@@ -464,6 +464,16 @@ row_check_table_for_mysql(
 	row_prebuilt_t*	prebuilt);	/* in: prebuilt struct in MySQL
 					handle */
 
+/*************************************************************************
+Determines if a table is a magic monitor table. */
+
+ibool
+row_is_magic_monitor_table(
+/*=======================*/
+					/* out: TRUE if monitor table */
+	const char*	table_name);	/* in: name of the table, in the
+					form database/table_name */
+
 /* A struct describing a place for an individual column in the MySQL
 row format which is presented to the table handler in ha_innobase.
 This template struct is used to speed up row transformations between

storage/innobase/include/sync0arr.h

@@ -66,26 +66,21 @@ sync_array_wait_event(
 	sync_array_t*	arr,	/* in: wait array */
 	ulint		index);	 /* in: index of the reserved cell */
 /**********************************************************************
-Frees the cell safely by reserving the sync array mutex and decrementing
-n_reserved if necessary. Should only be called from mutex_spin_wait. */
+Frees the cell. NOTE! sync_array_wait_event frees the cell
+automatically! */
 
 void
-sync_array_free_cell_protected(
-/*===========================*/
+sync_array_free_cell(
+/*=================*/
 	sync_array_t*	arr,	/* in: wait array */
 	ulint		index);	/* in: index of the cell in array */
 /**************************************************************************
-Looks for the cells in the wait array which refer
-to the wait object specified,
-and sets their corresponding events to the signaled state. In this
-way releases the threads waiting for the object to contend for the object.
-It is possible that no such cell is found, in which case does nothing. */
+Note that one of the wait objects was signalled. */
 
 void
-sync_array_signal_object(
-/*=====================*/
-	sync_array_t*	arr,	/* in: wait array */
-	void*		object);/* in: wait object */
+sync_array_object_signalled(
+/*========================*/
+	sync_array_t*	arr);	/* in: wait array */
 /**************************************************************************
 If the wakeup algorithm does not work perfectly at semaphore relases,
 this function will do the waking (see the comment in mutex_exit). This

storage/innobase/include/sync0rw.h

@@ -421,6 +421,18 @@ blocked by readers, a writer may queue for the lock by setting the writer
 field. Then no new readers are allowed in. */
 
 struct rw_lock_struct {
+	os_event_t	event;	/* Used by sync0arr.c for thread queueing */
+
+#ifdef __WIN__
+	os_event_t	wait_ex_event;	/* This windows specific event is
+				used by the thread which has set the
+				lock state to RW_LOCK_WAIT_EX. The
+				rw_lock design guarantees that this
+				thread will be the next one to proceed
+				once the current the event gets
+				signalled. See LEMMA 2 in sync0sync.c */
+#endif
+
 	ulint	reader_count;	/* Number of readers who have locked this
 				lock in the shared mode */
 	ulint	writer;		/* This field is set to RW_LOCK_EX if there

storage/innobase/include/sync0rw.ic

@@ -381,7 +381,11 @@ rw_lock_s_unlock_func(
 	mutex_exit(mutex);
 
 	if (UNIV_UNLIKELY(sg)) {
-		sync_array_signal_object(sync_primary_wait_array, lock);
+#ifdef __WIN__
+		os_event_set(lock->wait_ex_event);
+#endif
+		os_event_set(lock->event);
+		sync_array_object_signalled(sync_primary_wait_array);
 	}
 
 	ut_ad(rw_lock_validate(lock));
@@ -461,7 +465,11 @@ rw_lock_x_unlock_func(
 	mutex_exit(&(lock->mutex));
 
 	if (UNIV_UNLIKELY(sg)) {
-		sync_array_signal_object(sync_primary_wait_array, lock);
+#ifdef __WIN__
+		os_event_set(lock->wait_ex_event);
+#endif
+		os_event_set(lock->event);
+		sync_array_object_signalled(sync_primary_wait_array);
 	}
 
 	ut_ad(rw_lock_validate(lock));

storage/innobase/include/sync0sync.h

@@ -470,6 +470,7 @@ Do not use its fields directly! The structure used in the spin lock
 implementation of a mutual exclusion semaphore. */
 
 struct mutex_struct {
+	os_event_t	event;	/* Used by sync0arr.c for the wait queue */
 	ulint	lock_word;	/* This ulint is the target of the atomic
 				test-and-set instruction in Win32 */
 #if !defined(_WIN32) || !defined(UNIV_CAN_USE_X86_ASSEMBLER)

storage/innobase/include/sync0sync.ic

@@ -211,7 +211,7 @@ mutex_exit(
 	perform the read first, which could leave a waiting
 	thread hanging indefinitely.
 
-	Our current solution call every 10 seconds
+	Our current solution call every second
 	sync_arr_wake_threads_if_sema_free()
 	to wake up possible hanging threads if
 	they are missed in mutex_signal_object. */

storage/innobase/os/os0sync.c

@@ -21,6 +21,7 @@ Created 9/6/1995 Heikki Tuuri
 
 /* Type definition for an operating system mutex struct */
 struct os_mutex_struct{
+	os_event_t	event;	/* Used by sync0arr.c for queing threads */
 	void*		handle;	/* OS handle to mutex */
 	ulint		count;	/* we use this counter to check
 				that the same thread does not
@@ -35,6 +36,7 @@ struct os_mutex_struct{
 /* Mutex protecting counts and the lists of OS mutexes and events */
 os_mutex_t	os_sync_mutex;
 ibool		os_sync_mutex_inited	= FALSE;
+ibool		os_sync_free_called	= FALSE;
 
 /* This is incremented by 1 in os_thread_create and decremented by 1 in
 os_thread_exit */
@@ -50,6 +52,10 @@ ulint	os_event_count		= 0;
 ulint	os_mutex_count		= 0;
 ulint	os_fast_mutex_count	= 0;
 
+/* Because a mutex is embedded inside an event and there is an
+event embedded inside a mutex, on free, this generates a recursive call.
+This version of the free event function doesn't acquire the global lock */
+static void os_event_free_internal(os_event_t	event);
 
 /*************************************************************
 Initializes global event and OS 'slow' mutex lists. */
@@ -76,6 +82,7 @@ os_sync_free(void)
 	os_event_t	event;
 	os_mutex_t	mutex;
 
+	os_sync_free_called = TRUE;
 	event = UT_LIST_GET_FIRST(os_event_list);
 
 	while (event) {
@@ -99,6 +106,7 @@ os_sync_free(void)
 
 		mutex = UT_LIST_GET_FIRST(os_mutex_list);
 	}
+	os_sync_free_called = FALSE;
 }
 
 /*************************************************************
@@ -144,17 +152,31 @@ os_event_create(
 	ut_a(0 == pthread_cond_init(&(event->cond_var), NULL));
 #endif
 	event->is_set = FALSE;
-	event->signal_count = 0;
+
+	/* We return this value in os_event_reset(), which can then be
+	be used to pass to the os_event_wait_low(). The value of zero
+	is reserved in os_event_wait_low() for the case when the
+	caller does not want to pass any signal_count value. To
+	distinguish between the two cases we initialize signal_count
+	to 1 here. */
+	event->signal_count = 1;
 #endif /* __WIN__ */
 
-	/* Put to the list of events */
+	/* The os_sync_mutex can be NULL because during startup an event
+	can be created [ because it's embedded in the mutex/rwlock ] before
+	this module has been initialized */
+	if (os_sync_mutex != NULL) {
 		os_mutex_enter(os_sync_mutex);
+	}
 
+	/* Put to the list of events */
 	UT_LIST_ADD_FIRST(os_event_list, os_event_list, event);
 
 	os_event_count++;
 
+	if (os_sync_mutex != NULL) {
 		os_mutex_exit(os_sync_mutex);
+	}
 
 	return(event);
 }
@@ -231,13 +253,20 @@ os_event_set(
 
 /**************************************************************
 Resets an event semaphore to the nonsignaled state. Waiting threads will
-stop to wait for the event. */
+stop to wait for the event.
+The return value should be passed to os_even_wait_low() if it is desired
+that this thread should not wait in case of an intervening call to
+os_event_set() between this os_event_reset() and the
+os_event_wait_low() call. See comments for os_event_wait_low(). */
 
-void
+ib_longlong
 os_event_reset(
 /*===========*/
+				/* out: current signal_count. */
 	os_event_t	event)	/* in: event to reset */
 {
+	ib_longlong	ret = 0;
+
 #ifdef __WIN__
 	ut_a(event);
 
@@ -252,9 +281,40 @@ os_event_reset(
 	} else {
 		event->is_set = FALSE;
 	}
+	ret = event->signal_count;
 
 	os_fast_mutex_unlock(&(event->os_mutex));
 #endif
+	return(ret);
+}
+
+/**************************************************************
+Frees an event object, without acquiring the global lock. */
+static
+void
+os_event_free_internal(
+/*===================*/
+	os_event_t	event)	/* in: event to free */
+{
+#ifdef __WIN__
+	ut_a(event);
+
+	ut_a(CloseHandle(event->handle));
+#else
+	ut_a(event);
+
+	/* This is to avoid freeing the mutex twice */
+	os_fast_mutex_free(&(event->os_mutex));
+
+	ut_a(0 == pthread_cond_destroy(&(event->cond_var)));
+#endif
+	/* Remove from the list of events */
+
+	UT_LIST_REMOVE(os_event_list, os_event_list, event);
+
+	os_event_count--;
+
+	ut_free(event);
 }
 
 /**************************************************************
@@ -293,18 +353,38 @@ os_event_free(
 Waits for an event object until it is in the signaled state. If
 srv_shutdown_state == SRV_SHUTDOWN_EXIT_THREADS this also exits the
 waiting thread when the event becomes signaled (or immediately if the
-event is already in the signaled state). */
+event is already in the signaled state).
+
+Typically, if the event has been signalled after the os_event_reset()
+we'll return immediately because event->is_set == TRUE.
+There are, however, situations (e.g.: sync_array code) where we may
+lose this information. For example:
+
+thread A calls os_event_reset()
+thread B calls os_event_set()   [event->is_set == TRUE]
+thread C calls os_event_reset() [event->is_set == FALSE]
+thread A calls os_event_wait()  [infinite wait!]
+thread C calls os_event_wait()  [infinite wait!]
+
+Where such a scenario is possible, to avoid infinite wait, the
+value returned by os_event_reset() should be passed in as
+reset_sig_count. */
 
 void
-os_event_wait(
-/*==========*/
-	os_event_t	event)	/* in: event to wait */
+os_event_wait_low(
+/*==============*/
+	os_event_t	event,		/* in: event to wait */
+	ib_longlong	reset_sig_count)/* in: zero or the value
+					returned by previous call of
+					os_event_reset(). */
 {
 #ifdef __WIN__
 	DWORD	err;
 
 	ut_a(event);
 
+	UT_NOT_USED(reset_sig_count);
+
 	/* Specify an infinite time limit for waiting */
 	err = WaitForSingleObject(event->handle, INFINITE);
 
@@ -318,7 +398,11 @@ os_event_wait(
 
 	os_fast_mutex_lock(&(event->os_mutex));
 
+	if (reset_sig_count) {
+		old_signal_count = reset_sig_count;
+	} else {
 		old_signal_count = event->signal_count;
+	}
 
 	for (;;) {
 		if (event->is_set == TRUE
@@ -458,6 +542,7 @@ os_mutex_create(
 
 	mutex_str->handle = mutex;
 	mutex_str->count = 0;
+	mutex_str->event = os_event_create(NULL);
 
 	if (os_sync_mutex_inited) {
 		/* When creating os_sync_mutex itself we cannot reserve it */
@@ -534,6 +619,10 @@ os_mutex_free(
 {
 	ut_a(mutex);
 
+	if (!os_sync_free_called) {
+		os_event_free_internal(mutex->event);
+	}
+
 	if (os_sync_mutex_inited) {
 		os_mutex_enter(os_sync_mutex);
 	}

storage/innobase/read/read0read.c

@@ -212,7 +212,6 @@ read_view_oldest_copy_or_open_new(
 	view_copy->low_limit_no = old_view->low_limit_no;
 	view_copy->low_limit_id = old_view->low_limit_id;
 
-	view_copy->can_be_too_old = FALSE;
 
 	if (n > 0) {
 		/* The last active transaction has the smallest id: */
@@ -258,8 +257,6 @@ read_view_open_now(
 	view->low_limit_no = trx_sys->max_trx_id;
 	view->low_limit_id = view->low_limit_no;
 
-	view->can_be_too_old = FALSE;
-
 	n = 0;
 	trx = UT_LIST_GET_FIRST(trx_sys->trx_list);
 
@@ -432,8 +429,6 @@ read_cursor_view_create_for_mysql(
 	view->low_limit_no = trx_sys->max_trx_id;
 	view->low_limit_id = view->low_limit_no;
 
-	view->can_be_too_old = FALSE;
-
 	n = 0;
 	trx = UT_LIST_GET_FIRST(trx_sys->trx_list);
 

storage/innobase/row/row0mysql.c

@@ -57,6 +57,12 @@ static const char S_innodb_tablespace_monitor[] = "innodb_tablespace_monitor";
 static const char S_innodb_table_monitor[] = "innodb_table_monitor";
 static const char S_innodb_mem_validate[] = "innodb_mem_validate";
 
+/* Evaluates to true if str1 equals str2_onstack, used for comparing
+the above strings. */
+#define STR_EQ(str1, str1_len, str2_onstack) \
+	((str1_len) == sizeof(str2_onstack) \
+	 && memcmp(str1, str2_onstack, sizeof(str2_onstack)) == 0)
+
 /***********************************************************************
 Determine if the given name is a name reserved for MySQL system tables. */
 static
@@ -1728,7 +1734,7 @@ row_mysql_unlock_data_dictionary(
 }
 
 /*************************************************************************
-Drops a table for MySQL. If the name of the table ends in
+Creates a table for MySQL. If the name of the table ends in
 one of "innodb_monitor", "innodb_lock_monitor", "innodb_tablespace_monitor",
 "innodb_table_monitor", then this will also start the printing of monitor
 output by the master thread. If the table name ends in "innodb_mem_validate",
@@ -1809,9 +1815,7 @@ row_create_table_for_mysql(
 	table_name++;
 	table_name_len = strlen(table_name) + 1;
 
-	if (table_name_len == sizeof S_innodb_monitor
-	    && !memcmp(table_name, S_innodb_monitor,
-		       sizeof S_innodb_monitor)) {
+	if (STR_EQ(table_name, table_name_len, S_innodb_monitor)) {
 
 		/* Table equals "innodb_monitor":
 		start monitor prints */
@@ -1822,28 +1826,24 @@ row_create_table_for_mysql(
 		of InnoDB monitor prints */
 
 		os_event_set(srv_lock_timeout_thread_event);
-	} else if (table_name_len == sizeof S_innodb_lock_monitor
-		   && !memcmp(table_name, S_innodb_lock_monitor,
-			      sizeof S_innodb_lock_monitor)) {
+	} else if (STR_EQ(table_name, table_name_len,
+			  S_innodb_lock_monitor)) {
 
 		srv_print_innodb_monitor = TRUE;
 		srv_print_innodb_lock_monitor = TRUE;
 		os_event_set(srv_lock_timeout_thread_event);
-	} else if (table_name_len == sizeof S_innodb_tablespace_monitor
-		   && !memcmp(table_name, S_innodb_tablespace_monitor,
-			      sizeof S_innodb_tablespace_monitor)) {
+	} else if (STR_EQ(table_name, table_name_len,
+			  S_innodb_tablespace_monitor)) {
 
 		srv_print_innodb_tablespace_monitor = TRUE;
 		os_event_set(srv_lock_timeout_thread_event);
-	} else if (table_name_len == sizeof S_innodb_table_monitor
-		   && !memcmp(table_name, S_innodb_table_monitor,
-			      sizeof S_innodb_table_monitor)) {
+	} else if (STR_EQ(table_name, table_name_len,
+			  S_innodb_table_monitor)) {
 
 		srv_print_innodb_table_monitor = TRUE;
 		os_event_set(srv_lock_timeout_thread_event);
-	} else if (table_name_len == sizeof S_innodb_mem_validate
-		   && !memcmp(table_name, S_innodb_mem_validate,
-			      sizeof S_innodb_mem_validate)) {
+	} else if (STR_EQ(table_name, table_name_len,
+			  S_innodb_mem_validate)) {
 		/* We define here a debugging feature intended for
 		developers */
 
@@ -4130,3 +4130,33 @@ row_check_table_for_mysql(
 
 	return(ret);
 }
+
+/*************************************************************************
+Determines if a table is a magic monitor table. */
+
+ibool
+row_is_magic_monitor_table(
+/*=======================*/
+					/* out: TRUE if monitor table */
+	const char*	table_name)	/* in: name of the table, in the
+					form database/table_name */
+{
+	const char*	name; /* table_name without database/ */
+	ulint		len;
+
+	name = strchr(table_name, '/');
+	ut_a(name != NULL);
+	name++;
+	len = strlen(name) + 1;
+
+	if (STR_EQ(name, len, S_innodb_monitor)
+	    || STR_EQ(name, len, S_innodb_lock_monitor)
+	    || STR_EQ(name, len, S_innodb_tablespace_monitor)
+	    || STR_EQ(name, len, S_innodb_table_monitor)
+	    || STR_EQ(name, len, S_innodb_mem_validate)) {
+
+		return(TRUE);
+	}
+
+	return(FALSE);
+}

storage/innobase/srv/srv0srv.c

@@ -1904,12 +1904,6 @@ loop:
 
 	os_thread_sleep(1000000);
 
-	/* In case mutex_exit is not a memory barrier, it is
-	theoretically possible some threads are left waiting though
-	the semaphore is already released. Wake up those threads: */
-
-	sync_arr_wake_threads_if_sema_free();
-
 	current_time = time(NULL);
 
 	time_elapsed = difftime(current_time, last_monitor_time);
@@ -2106,9 +2100,15 @@ loop:
 		srv_refresh_innodb_monitor_stats();
 	}
 
+	/* In case mutex_exit is not a memory barrier, it is
+	theoretically possible some threads are left waiting though
+	the semaphore is already released. Wake up those threads: */
+	
+	sync_arr_wake_threads_if_sema_free();
+
 	if (sync_array_print_long_waits()) {
 		fatal_cnt++;
-		if (fatal_cnt > 5) {
+		if (fatal_cnt > 10) {
 
 			fprintf(stderr,
 				"InnoDB: Error: semaphore wait has lasted"
@@ -2128,7 +2128,7 @@ loop:
 
 	fflush(stderr);
 
-	os_thread_sleep(2000000);
+	os_thread_sleep(1000000);
 
 	if (srv_shutdown_state < SRV_SHUTDOWN_CLEANUP) {
 

storage/innobase/sync/sync0arr.c

@@ -40,24 +40,23 @@ because we can do with a very small number of OS events,
 say 200. In NT 3.51, allocating events seems to be a quadratic
 algorithm, because 10 000 events are created fast, but
 100 000 events takes a couple of minutes to create.
-*/
+
+As of 5.0.30 the above mentioned design is changed. Since now
+OS can handle millions of wait events efficiently, we no longer
+have this concept of each cell of wait array having one event.
+Instead, now the event that a thread wants to wait on is embedded
+in the wait object (mutex or rw_lock). We still keep the global
+wait array for the sake of diagnostics and also to avoid infinite
+wait The error_monitor thread scans the global wait array to signal
+any waiting threads who have missed the signal. */
 
 /* A cell where an individual thread may wait suspended
 until a resource is released. The suspending is implemented
 using an operating system event semaphore. */
 struct sync_cell_struct {
-	/* State of the cell. SC_WAKING_UP means
-	sync_array_struct->n_reserved has been decremented, but the thread
-	in this cell has not waken up yet. When it does, it will set the
-	state to SC_FREE. Note that this is done without the protection of
-	any mutex. */
-	enum { SC_FREE, SC_RESERVED, SC_WAKING_UP } state;
-
 	void*		wait_object;	/* pointer to the object the
-					thread is waiting for; this is not
-					reseted to NULL when a cell is
-					freed. */
-
+					thread is waiting for; if NULL
+					the cell is free for use */
 	mutex_t*	old_wait_mutex;	/* the latest wait mutex in cell */
 	rw_lock_t*	old_wait_rw_lock;/* the latest wait rw-lock in cell */
 	ulint		request_type;	/* lock type requested on the
@@ -71,13 +70,23 @@ struct sync_cell_struct {
 	ibool		waiting;	/* TRUE if the thread has already
 					called sync_array_event_wait
 					on this cell */
-	ibool		event_set;	/* TRUE if the event is set */
-	os_event_t	event;		/* operating system event
-					semaphore handle */
+	ib_longlong	signal_count;	/* We capture the signal_count
+					of the wait_object when we
+					reset the event. This value is
+					then passed on to os_event_wait
+					and we wait only if the event
+					has not been signalled in the
+					period between the reset and
+					wait call. */
 	time_t		reservation_time;/* time when the thread reserved
 					the wait cell */
 };
 
+/* NOTE: It is allowed for a thread to wait
+for an event allocated for the array without owning the
+protecting mutex (depending on the case: OS or database mutex), but
+all changes (set or reset) to the state of the event must be made
+while owning the mutex. */
 struct sync_array_struct {
 	ulint		n_reserved;	/* number of currently reserved
 					cells in the wait array */
@@ -220,12 +229,9 @@ sync_array_create(
 
 	for (i = 0; i < n_cells; i++) {
 		cell = sync_array_get_nth_cell(arr, i);
-		cell->state = SC_FREE;
 	cell->wait_object = NULL;
-
-		/* Create an operating system event semaphore with no name */
-		cell->event = os_event_create(NULL);
-		cell->event_set = FALSE; /* it is created in reset state */
+		cell->waiting = FALSE;
+		cell->signal_count = 0;
 	}
 
 	return(arr);
@@ -239,19 +245,12 @@ sync_array_free(
 /*============*/
 	sync_array_t*	arr)	/* in, own: sync wait array */
 {
-	ulint		i;
-	sync_cell_t*	cell;
 	ulint		protection;
 
 	ut_a(arr->n_reserved == 0);
 
 	sync_array_validate(arr);
 
-	for (i = 0; i < arr->n_cells; i++) {
-		cell = sync_array_get_nth_cell(arr, i);
-		os_event_free(cell->event);
-	}
-
 	protection = arr->protection;
 
 	/* Release the mutex protecting the wait array complex */
@@ -285,8 +284,7 @@ sync_array_validate(
 
 	for (i = 0; i < arr->n_cells; i++) {
 		cell = sync_array_get_nth_cell(arr, i);
-
-		if (cell->state == SC_RESERVED) {
+		if (cell->wait_object != NULL) {
 			count++;
 		}
 	}
@@ -296,6 +294,29 @@ sync_array_validate(
 	sync_array_exit(arr);
 }
 
+/***********************************************************************
+Puts the cell event in reset state. */
+static
+ib_longlong
+sync_cell_event_reset(
+/*==================*/
+				/* out: value of signal_count
+				at the time of reset. */
+	ulint		type,	/* in: lock type mutex/rw_lock */
+	void*		object) /* in: the rw_lock/mutex object */
+{
+	if (type == SYNC_MUTEX) {
+		return(os_event_reset(((mutex_t *) object)->event));
+#ifdef __WIN__
+	} else if (type == RW_LOCK_WAIT_EX) {
+		return(os_event_reset(
+		       ((rw_lock_t *) object)->wait_ex_event));
+#endif
+	} else {
+		return(os_event_reset(((rw_lock_t *) object)->event));
+	}
+}
+
 /**********************************************************************
 Reserves a wait array cell for waiting for an object.
 The event of the cell is reset to nonsignalled state. */
@@ -324,21 +345,9 @@ sync_array_reserve_cell(
 	for (i = 0; i < arr->n_cells; i++) {
 		cell = sync_array_get_nth_cell(arr, i);
 
-		if (cell->state == SC_FREE) {
-
-			/* We do not check cell->event_set because it is
-			set outside the protection of the sync array mutex
-			and we had a bug regarding it, and since resetting
-			an event when it is not needed does no harm it is
-			safer always to do it. */
-
-			cell->event_set = FALSE;
-			os_event_reset(cell->event);
-
-			cell->state = SC_RESERVED;
-			cell->reservation_time = time(NULL);
-			cell->thread = os_thread_get_curr_id();
+		if (cell->wait_object == NULL) {
 
+			cell->waiting = FALSE;
 			cell->wait_object = object;
 
 			if (type == SYNC_MUTEX) {
@@ -348,7 +357,6 @@ sync_array_reserve_cell(
 			}
 
 			cell->request_type = type;
-			cell->waiting = FALSE;
 
 			cell->file = file;
 			cell->line = line;
@@ -359,6 +367,16 @@ sync_array_reserve_cell(
 
 			sync_array_exit(arr);
 
+			/* Make sure the event is reset and also store
+			the value of signal_count at which the event
+			was reset. */
+			cell->signal_count = sync_cell_event_reset(type,
+								object);
+
+			cell->reservation_time = time(NULL);
+
+			cell->thread = os_thread_get_curr_id();
+
 			return;
 		}
 	}
@@ -368,68 +386,6 @@ sync_array_reserve_cell(
 	return;
 }
 
-/**********************************************************************
-Frees the cell. Note that we don't have any mutex reserved when calling
-this. */
-static
-void
-sync_array_free_cell(
-/*=================*/
-	sync_array_t*	arr,	/* in: wait array */
-	ulint		index)	/* in: index of the cell in array */
-{
-	sync_cell_t*	cell;
-
-	cell = sync_array_get_nth_cell(arr, index);
-
-	ut_a(cell->state == SC_WAKING_UP);
-	ut_a(cell->wait_object != NULL);
-
-	cell->state = SC_FREE;
-}
-
-/**********************************************************************
-Frees the cell safely by reserving the sync array mutex and decrementing
-n_reserved if necessary. Should only be called from mutex_spin_wait. */
-
-void
-sync_array_free_cell_protected(
-/*===========================*/
-	sync_array_t*	arr,	/* in: wait array */
-	ulint		index)	/* in: index of the cell in array */
-{
-	sync_cell_t*	cell;
-
-	sync_array_enter(arr);
-
-	cell = sync_array_get_nth_cell(arr, index);
-
-	ut_a(cell->state != SC_FREE);
-	ut_a(cell->wait_object != NULL);
-
-	/* We only need to decrement n_reserved if it has not already been
-	done by sync_array_signal_object. */
-	if (cell->state == SC_RESERVED) {
-		ut_a(arr->n_reserved > 0);
-		arr->n_reserved--;
-	} else if (cell->state == SC_WAKING_UP) {
-		/* This is tricky; if we don't wait for the event to be
-		signaled, signal_object can set the state of a cell to
-		SC_WAKING_UP, mutex_spin_wait can call this and set the
-		state to SC_FREE, and then signal_object gets around to
-		calling os_set_event for the cell but since it's already
-		been freed things break horribly. */
-
-		sync_array_exit(arr);
-		os_event_wait(cell->event);
-		sync_array_enter(arr);
-	}
-
-	cell->state = SC_FREE;
-
-	sync_array_exit(arr);
-}
-
 /**********************************************************************
 This function should be called when a thread starts to wait on
 a wait array cell. In the debug version this function checks
@@ -447,14 +403,27 @@ sync_array_wait_event(
 
 	ut_a(arr);
 
+	sync_array_enter(arr);
+
 	cell = sync_array_get_nth_cell(arr, index);
 
-	ut_a((cell->state == SC_RESERVED) || (cell->state == SC_WAKING_UP));
 	ut_a(cell->wait_object);
 	ut_a(!cell->waiting);
 	ut_ad(os_thread_get_curr_id() == cell->thread);
 
-	event = cell->event;
+	if (cell->request_type == SYNC_MUTEX) {
+		event = ((mutex_t*) cell->wait_object)->event;
+#ifdef __WIN__
+	/* On windows if the thread about to wait is the one which
+	has set the state of the rw_lock to RW_LOCK_WAIT_EX, then
+	it waits on a special event i.e.: wait_ex_event. */
+	} else if (cell->request_type == RW_LOCK_WAIT_EX) {
+		event = ((rw_lock_t*) cell->wait_object)->wait_ex_event;
+#endif
+	} else {
+		event = ((rw_lock_t*) cell->wait_object)->event;
+	}
+
 		cell->waiting = TRUE;
 
 #ifdef UNIV_SYNC_DEBUG
@@ -464,7 +433,6 @@ sync_array_wait_event(
 	recursively sync_array routines, leading to trouble.
 	rw_lock_debug_mutex freezes the debug lists. */
 
-	sync_array_enter(arr);
 	rw_lock_debug_mutex_enter();
 
 	if (TRUE == sync_array_detect_deadlock(arr, cell, cell, 0)) {
@@ -474,16 +442,16 @@ sync_array_wait_event(
 	}
 
 	rw_lock_debug_mutex_exit();
-	sync_array_exit(arr);
 #endif
-	os_event_wait(event);
+	sync_array_exit(arr);
+
+	os_event_wait_low(event, cell->signal_count);
 
 	sync_array_free_cell(arr, index);
 }
 
 /**********************************************************************
-Reports info of a wait array cell. Note: sync_array_print_long_waits()
-calls this without mutex protection. */
+Reports info of a wait array cell. */
 static
 void
 sync_array_cell_print(
@@ -503,17 +471,8 @@ sync_array_cell_print(
 		(ulong) os_thread_pf(cell->thread), cell->file,
 		(ulong) cell->line,
 		difftime(time(NULL), cell->reservation_time));
-	fprintf(file, "Wait array cell state %lu\n", (ulong)cell->state);
-
-	/* If the memory area pointed to by old_wait_mutex /
-	old_wait_rw_lock has been freed, this can crash. */
-
-	if (cell->state != SC_RESERVED) {
-		/* If cell has this state, then even if we are holding the sync
-		array mutex, the wait object may get freed meanwhile. Do not
-		print the wait object then. */
 
-	} else if (type == SYNC_MUTEX) {
+	if (type == SYNC_MUTEX) {
 		/* We use old_wait_mutex in case the cell has already
 		been freed meanwhile */
 		mutex = cell->old_wait_mutex;
@@ -531,7 +490,11 @@ sync_array_cell_print(
 #endif /* UNIV_SYNC_DEBUG */
 			(ulong) mutex->waiters);
 
-	} else if (type == RW_LOCK_EX || type == RW_LOCK_SHARED) {
+	} else if (type == RW_LOCK_EX
+#ifdef __WIN__
+		   || type == RW_LOCK_WAIT_EX
+#endif
+		   || type == RW_LOCK_SHARED) {
 
 		fputs(type == RW_LOCK_EX ? "X-lock on" : "S-lock on", file);
 
@@ -565,8 +528,8 @@ sync_array_cell_print(
 		ut_error;
 	}
 
-	if (cell->event_set) {
-		fputs("wait is ending\n", file);
+	if (!cell->waiting) {
+		fputs("wait has ended\n", file);
 	}
 }
 
@@ -589,7 +552,7 @@ sync_array_find_thread(
 
 		cell = sync_array_get_nth_cell(arr, i);
 
-		if ((cell->state == SC_RESERVED)
+		if (cell->wait_object != NULL
 		    && os_thread_eq(cell->thread, thread)) {
 
 			return(cell);	/* Found */
@@ -679,7 +642,7 @@ sync_array_detect_deadlock(
 
 	depth++;
 
-	if (cell->event_set || !cell->waiting) {
+	if (!cell->waiting) {
 
 		return(FALSE); /* No deadlock here */
 	}
@@ -704,10 +667,8 @@ sync_array_detect_deadlock(
 						       depth);
 			if (ret) {
 				fprintf(stderr,
-					"Mutex %p owned by thread %lu"
-					" file %s line %lu\n",
-					(void*) mutex,
-					(ulong) os_thread_pf(mutex->thread_id),
+			"Mutex %p owned by thread %lu file %s line %lu\n",
+					mutex, (ulong) os_thread_pf(mutex->thread_id),
 					mutex->file_name, (ulong) mutex->line);
 				sync_array_cell_print(stderr, cell);
 
@@ -717,7 +678,8 @@ sync_array_detect_deadlock(
 
 		return(FALSE); /* No deadlock */
 
-	} else if (cell->request_type == RW_LOCK_EX) {
+	} else if (cell->request_type == RW_LOCK_EX
+		   || cell->request_type == RW_LOCK_WAIT_EX) {
 
 		lock = cell->wait_object;
 
@@ -816,7 +778,8 @@ sync_arr_cell_can_wake_up(
 			return(TRUE);
 		}
 
-	} else if (cell->request_type == RW_LOCK_EX) {
+	} else if (cell->request_type == RW_LOCK_EX
+		   || cell->request_type == RW_LOCK_WAIT_EX) {
 
 		lock = cell->wait_object;
 
@@ -845,101 +808,47 @@ sync_arr_cell_can_wake_up(
 	return(FALSE);
 }
 
-/**************************************************************************
-Looks for the cells in the wait array which refer to the wait object
-specified, and sets their corresponding events to the signaled state. In this
-way releases the threads waiting for the object to contend for the object.
-It is possible that no such cell is found, in which case does nothing. */
+/**********************************************************************
+Frees the cell. NOTE! sync_array_wait_event frees the cell
+automatically! */
 
 void
-sync_array_signal_object(
-/*=====================*/
+sync_array_free_cell(
+/*=================*/
 	sync_array_t*	arr,	/* in: wait array */
-	void*		object)	/* in: wait object */
+	ulint		index)  /* in: index of the cell in array */
 {
 	sync_cell_t*	cell;
-	ulint		count;
-	ulint		i;
-	ulint		res_count;
-
-	/* We store the addresses of cells we need to signal and signal
-	them only after we have released the sync array's mutex (for
-	performance reasons). cell_count is the number of such cells, and
-	cell_ptr points to the first one. If there are less than
-	UT_ARR_SIZE(cells) of them, cell_ptr == &cells[0], otherwise
-	cell_ptr points to malloc'd memory that we must free. */
-
-	sync_cell_t*	cells[100];
-	sync_cell_t**	cell_ptr = &cells[0];
-	ulint		cell_count = 0;
-	ulint		cell_max_count = UT_ARR_SIZE(cells);
-
-	ut_a(100 == cell_max_count);
 
 	sync_array_enter(arr);
 
-	arr->sg_count++;
-
-	i = 0;
-	count = 0;
-
-	/* We need to store this to a local variable because it is modified
-	inside the loop */
-	res_count = arr->n_reserved;
-
-	while (count < res_count) {
-
-		cell = sync_array_get_nth_cell(arr, i);
+	cell = sync_array_get_nth_cell(arr, index);
 
-		if (cell->state == SC_RESERVED) {
+	ut_a(cell->wait_object != NULL);
 
-			count++;
-			if (cell->wait_object == object) {
-				cell->state = SC_WAKING_UP;
+	cell->waiting = FALSE;
+	cell->wait_object =  NULL;
+	cell->signal_count = 0;
 
 	ut_a(arr->n_reserved > 0);
 	arr->n_reserved--;
 
-				if (cell_count == cell_max_count) {
-					sync_cell_t** old_cell_ptr = cell_ptr;
-					size_t old_size, new_size;
-
-					old_size = cell_max_count
-						* sizeof(sync_cell_t*);
-					cell_max_count *= 2;
-					new_size = cell_max_count
-						* sizeof(sync_cell_t*);
-
-					cell_ptr = malloc(new_size);
-					ut_a(cell_ptr);
-					memcpy(cell_ptr, old_cell_ptr,
-					       old_size);
+	sync_array_exit(arr);
+}
 
-					if (old_cell_ptr != &cells[0]) {
-						free(old_cell_ptr);
-					}
-				}
+/**************************************************************************
+Increments the signalled count. */
 
-				cell_ptr[cell_count] = cell;
-				cell_count++;
-			}
-		}
+void
+sync_array_object_signalled(
+/*========================*/
+	sync_array_t*	arr)	/* in: wait array */
+{
+	sync_array_enter(arr);
 
-		i++;
-	}
+	arr->sg_count++;
 
 	sync_array_exit(arr);
-
-	for (i = 0; i < cell_count; i++) {
-		cell = cell_ptr[i];
-
-		cell->event_set = TRUE;
-		os_event_set(cell->event);
-	}
-
-	if (cell_ptr != &cells[0]) {
-		free(cell_ptr);
-	}
 }
 
 /**************************************************************************
@@ -959,33 +868,41 @@ sync_arr_wake_threads_if_sema_free(void)
 	sync_cell_t*	cell;
 	ulint		count;
 	ulint		i;
-	ulint		res_count;
 
 	sync_array_enter(arr);
 
 	i = 0;
 	count = 0;
 
-	/* We need to store this to a local variable because it is modified
-	inside the loop */
-
-	res_count = arr->n_reserved;
-
-	while (count < res_count) {
+	while (count < arr->n_reserved) {
 
 		cell = sync_array_get_nth_cell(arr, i);
 
-		if (cell->state == SC_RESERVED) {
+		if (cell->wait_object != NULL) {
 
 			count++;
 
 			if (sync_arr_cell_can_wake_up(cell)) {
-				cell->state = SC_WAKING_UP;
-				cell->event_set = TRUE;
-				os_event_set(cell->event);
 
-				ut_a(arr->n_reserved > 0);
-				arr->n_reserved--;
+				if (cell->request_type == SYNC_MUTEX) {
+					mutex_t*	mutex;
+
+					mutex = cell->wait_object;
+					os_event_set(mutex->event);
+#ifdef __WIN__
+				} else if (cell->request_type
+					   == RW_LOCK_WAIT_EX) {
+					rw_lock_t*	lock;
+
+					lock = cell->wait_object;
+					os_event_set(lock->wait_ex_event);
+#endif
+				} else {
+					rw_lock_t*	lock;
+
+					lock = cell->wait_object;
+					os_event_set(lock->event);
+				}
 			}
 		}
 
@@ -1015,7 +932,7 @@ sync_array_print_long_waits(void)
 
 		cell = sync_array_get_nth_cell(sync_primary_wait_array, i);
 
-		if ((cell->state != SC_FREE)
+		if (cell->wait_object != NULL && cell->waiting
 		    && difftime(time(NULL), cell->reservation_time) > 240) {
 			fputs("InnoDB: Warning: a long semaphore wait:\n",
 			      stderr);
@@ -1023,7 +940,7 @@ sync_array_print_long_waits(void)
 			noticed = TRUE;
 		}
 
-		if ((cell->state != SC_FREE)
+		if (cell->wait_object != NULL && cell->waiting
 		    && difftime(time(NULL), cell->reservation_time)
 		    > fatal_timeout) {
 			fatal = TRUE;
@@ -1072,20 +989,25 @@ sync_array_output_info(
 				mutex */
 {
 	sync_cell_t*	cell;
+	ulint		count;
 	ulint		i;
 
 	fprintf(file,
-		"OS WAIT ARRAY INFO: reservation count %ld,"
-		" signal count %ld\n",
-		(long) arr->res_count,
-		(long) arr->sg_count);
-	for (i = 0; i < arr->n_cells; i++) {
+		"OS WAIT ARRAY INFO: reservation count %ld, signal count %ld\n",
+						(long) arr->res_count, (long) arr->sg_count);
+	i = 0;
+	count = 0;
+
+	while (count < arr->n_reserved) {
 
 		cell = sync_array_get_nth_cell(arr, i);
 
-		if (cell->state != SC_FREE) {
+	if (cell->wait_object != NULL) {
+		count++;
 			sync_array_cell_print(file, cell);
 		}
+
+		i++;
 	}
 }
 

storage/innobase/sync/sync0rw.c

@@ -151,6 +151,11 @@ rw_lock_create_func(
 	lock->last_x_file_name = "not yet reserved";
 	lock->last_s_line = 0;
 	lock->last_x_line = 0;
+	lock->event = os_event_create(NULL);
+
+#ifdef __WIN__
+	lock->wait_ex_event = os_event_create(NULL);
+#endif
 
 	mutex_enter(&rw_lock_list_mutex);
 
@@ -184,6 +189,11 @@ rw_lock_free(
 	mutex_free(rw_lock_get_mutex(lock));
 
 	mutex_enter(&rw_lock_list_mutex);
+	os_event_free(lock->event);
+
+#ifdef __WIN__
+	os_event_free(lock->wait_ex_event);
+#endif
 
 	if (UT_LIST_GET_PREV(list, lock)) {
 		ut_a(UT_LIST_GET_PREV(list, lock)->magic_n == RW_LOCK_MAGIC_N);
@@ -544,7 +554,15 @@ lock_loop:
 	rw_x_system_call_count++;
 
 	sync_array_reserve_cell(sync_primary_wait_array,
-				lock, RW_LOCK_EX,
+				lock,
+#ifdef __WIN__
+				/* On windows RW_LOCK_WAIT_EX signifies
+				that this thread should wait on the
+				special wait_ex_event. */
+				(state == RW_LOCK_WAIT_EX)
+				 ? RW_LOCK_WAIT_EX :
+#endif
+				RW_LOCK_EX,
 				file_name, line,
 				&index);
 

storage/innobase/sync/sync0sync.c

@@ -95,17 +95,47 @@ have happened that the thread which was holding the mutex has just released
 it and did not see the waiters byte set to 1, a case which would lead the
 other thread to an infinite wait.
 
-LEMMA 1: After a thread resets the event of the cell it reserves for waiting
-========
-for a mutex, some thread will eventually call sync_array_signal_object with
-the mutex as an argument. Thus no infinite wait is possible.
+LEMMA 1: After a thread resets the event of a mutex (or rw_lock), some
+=======
+thread will eventually call os_event_set() on that particular event.
+Thus no infinite wait is possible in this case.
 
 Proof:	After making the reservation the thread sets the waiters field in the
 mutex to 1. Then it checks that the mutex is still reserved by some thread,
 or it reserves the mutex for itself. In any case, some thread (which may be
 also some earlier thread, not necessarily the one currently holding the mutex)
 will set the waiters field to 0 in mutex_exit, and then call
-sync_array_signal_object with the mutex as an argument.
+os_event_set() with the mutex as an argument.
+Q.E.D.
+
+LEMMA 2: If an os_event_set() call is made after some thread has called
+=======
+the os_event_reset() and before it starts wait on that event, the call
+will not be lost to the second thread. This is true even if there is an
+intervening call to os_event_reset() by another thread.
+Thus no infinite wait is possible in this case.
+
+Proof (non-windows platforms): os_event_reset() returns a monotonically
+increasing value of signal_count. This value is increased at every
+call of os_event_set() If thread A has called os_event_reset() followed
+by thread B calling os_event_set() and then some other thread C calling
+os_event_reset(), the is_set flag of the event will be set to FALSE;
+but now if thread A calls os_event_wait_low() with the signal_count
+value returned from the earlier call of os_event_reset(), it will
+return immediately without waiting.
+Q.E.D.
+
+Proof (windows): If there is a writer thread which is forced to wait for
+the lock, it may be able to set the state of rw_lock to RW_LOCK_WAIT_EX
+The design of rw_lock ensures that there is one and only one thread
+that is able to change the state to RW_LOCK_WAIT_EX and this thread is
+guaranteed to acquire the lock after it is released by the current
+holders and before any other waiter gets the lock.
+On windows this thread waits on a separate event i.e.: wait_ex_event.
+Since only one thread can wait on this event there is no chance
+of this event getting reset before the writer starts wait on it.
+Therefore, this thread is guaranteed to catch the os_set_event()
+signalled unconditionally at the release of the lock.
 Q.E.D. */
 
 /* The number of system calls made in this module. Intended for performance
@@ -217,6 +247,7 @@ mutex_create_func(
 	os_fast_mutex_init(&(mutex->os_fast_mutex));
 	mutex->lock_word = 0;
 #endif
+	mutex->event = os_event_create(NULL);
 	mutex_set_waiters(mutex, 0);
 #ifdef UNIV_DEBUG
 	mutex->magic_n = MUTEX_MAGIC_N;
@@ -300,6 +331,8 @@ mutex_free(
 		mutex_exit(&mutex_list_mutex);
 	}
 
+	os_event_free(mutex->event);
+
 #if !defined(_WIN32) || !defined(UNIV_CAN_USE_X86_ASSEMBLER)
 	os_fast_mutex_free(&(mutex->os_fast_mutex));
 #endif
@@ -509,8 +542,7 @@ spin_loop:
 		if (mutex_test_and_set(mutex) == 0) {
 			/* Succeeded! Free the reserved wait cell */
 
-			sync_array_free_cell_protected(sync_primary_wait_array,
-						       index);
+			sync_array_free_cell(sync_primary_wait_array, index);
 
 			ut_d(mutex->thread_id = os_thread_get_curr_id());
 #ifdef UNIV_SYNC_DEBUG
@@ -591,8 +623,8 @@ mutex_signal_object(
 
 	/* The memory order of resetting the waiters field and
 	signaling the object is important. See LEMMA 1 above. */
-
-	sync_array_signal_object(sync_primary_wait_array, mutex);
+	os_event_set(mutex->event);
+	sync_array_object_signalled(sync_primary_wait_array);
 }
 
 #ifdef UNIV_SYNC_DEBUG
@@ -1130,6 +1162,7 @@ sync_thread_add_level(
 		break;
 	case SYNC_TREE_NODE:
 		ut_a(sync_thread_levels_contain(array, SYNC_INDEX_TREE)
+		     || sync_thread_levels_contain(array, SYNC_DICT_OPERATION)
 		     || sync_thread_levels_g(array, SYNC_TREE_NODE - 1));
 		break;
 	case SYNC_TREE_NODE_NEW: