Merge aivanov@bk-internal.mysql.com:/home/bk/mysql-5.0

into  mysql.com:/home/alexi/innodb-ss/mysql-5.0-ss115

innobase/include/data0type.h

@@ -13,6 +13,7 @@ Created 1/16/1996 Heikki Tuuri
 
 extern ulint	data_mysql_default_charset_coll;
 #define DATA_MYSQL_LATIN1_SWEDISH_CHARSET_COLL 8
+#define DATA_MYSQL_BINARY_CHARSET_COLL 63
 
 /* SQL data type struct */
 typedef struct dtype_struct		dtype_t;
@@ -311,7 +312,7 @@ dtype_get_pad_char(
 /*===============*/
 				/* out: padding character code, or
 				ULINT_UNDEFINED if no padding specified */
-	dtype_t*	type);	/* in: type */
+	const dtype_t*	type);	/* in: type */
 /***************************************************************************
 Returns the size of a fixed size data type, 0 if not a fixed size type. */
 UNIV_INLINE

innobase/include/data0type.ic

@@ -188,26 +188,35 @@ dtype_get_pad_char(
 /*===============*/
 				/* out: padding character code, or
 				ULINT_UNDEFINED if no padding specified */
-	dtype_t*	type)	/* in: type */
+	const dtype_t*	type)	/* in: type */
 {
-	if (type->mtype == DATA_CHAR
-	    || type->mtype == DATA_VARCHAR
-	    || type->mtype == DATA_BINARY
-	    || type->mtype == DATA_FIXBINARY
-	    || type->mtype == DATA_MYSQL
-	    || type->mtype == DATA_VARMYSQL
-	    || (type->mtype == DATA_BLOB
-		&& (type->prtype & DATA_BINARY_TYPE) == 0)) {
-
+	switch (type->mtype) {
+	case DATA_FIXBINARY:
+	case DATA_BINARY:
+		if (UNIV_UNLIKELY(dtype_get_charset_coll(type->prtype)
+		    == DATA_MYSQL_BINARY_CHARSET_COLL)) {
+			/* Starting from 5.0.18, do not pad
+			VARBINARY or BINARY columns. */
+			return(ULINT_UNDEFINED);
+		}
+		/* Fall through */
+	case DATA_CHAR:
+	case DATA_VARCHAR:
+	case DATA_MYSQL:
+	case DATA_VARMYSQL:
 		/* Space is the padding character for all char and binary
 	        strings, and starting from 5.0.3, also for TEXT strings. */
 
-		return((ulint)' ');
+		return(0x20);
+	case DATA_BLOB:
+		if ((type->prtype & DATA_BINARY_TYPE) == 0) {
+			return(0x20);
 		}
-
+		/* Fall through */
+	default:
 		/* No padding specified */
-
 		return(ULINT_UNDEFINED);
+	}
 }
 
 /**************************************************************************

innobase/include/lock0lock.h

@@ -64,14 +64,6 @@ lock_clust_rec_some_has_impl(
 	dict_index_t*	index,	/* in: clustered index */
 	const ulint*	offsets);/* in: rec_get_offsets(rec, index) */
 /*****************************************************************
-Resets the lock bits for a single record. Releases transactions
-waiting for lock requests here. */
-
-void
-lock_rec_reset_and_release_wait(
-/*============================*/
-	rec_t*	rec);	/* in: record whose locks bits should be reset */
-/*****************************************************************
 Makes a record to inherit the locks of another record as gap type
 locks, but does not reset the lock bits of the other record. Also
 waiting lock requests on rec are inherited as GRANTED gap locks. */
@@ -427,6 +419,18 @@ lock_is_on_table(
 /*=============*/
 				/* out: TRUE if there are lock(s) */
 	dict_table_t*	table);	/* in: database table in dictionary cache */
+/*****************************************************************
+Removes a granted record lock of a transaction from the queue and grants
+locks to other transactions waiting in the queue if they now are entitled
+to a lock. */
+
+void
+lock_rec_unlock(
+/*============*/
+	trx_t*	trx,  		/* in: transaction that has set a record
+				lock */
+	rec_t*	rec,		/* in: record */
+	ulint	lock_mode);	/* in: LOCK_S or LOCK_X */
 /*************************************************************************
 Releases a table lock.
 Releases possible other transactions waiting for this lock. */

innobase/include/os0file.h

@@ -91,9 +91,10 @@ log. */
 #define	OS_FILE_NOT_FOUND		71
 #define	OS_FILE_DISK_FULL		72
 #define	OS_FILE_ALREADY_EXISTS		73
-#define OS_FILE_AIO_RESOURCES_RESERVED	74	/* wait for OS aio resources
+#define	OS_FILE_PATH_ERROR		74
+#define	OS_FILE_AIO_RESOURCES_RESERVED	75	/* wait for OS aio resources
 						to become available again */
-#define	OS_FILE_ERROR_NOT_SPECIFIED	75
+#define	OS_FILE_ERROR_NOT_SPECIFIED	76
 
 /* Types for aio operations */
 #define OS_FILE_READ	10

innobase/include/row0mysql.h

@@ -250,7 +250,8 @@ trx_register_new_rec_lock() to store the information which new record locks
 really were set. This function removes a newly set lock under prebuilt->pcur,
 and also under prebuilt->clust_pcur. Currently, this is only used and tested
 in the case of an UPDATE or a DELETE statement, where the row lock is of the
-LOCK_X type.
+LOCK_X or LOCK_S type. 
+
 Thus, this implements a 'mini-rollback' that releases the latest record 
 locks we set. */
 

innobase/lock/lock0lock.c

@@ -2392,7 +2392,7 @@ lock_rec_free_all_from_discard_page(
 /*****************************************************************
 Resets the lock bits for a single record. Releases transactions waiting for
 lock requests here. */
-
+static
 void
 lock_rec_reset_and_release_wait(
 /*============================*/
@@ -3760,6 +3760,75 @@ lock_table_dequeue(
 
 /*=========================== LOCK RELEASE ==============================*/
 
+/*****************************************************************
+Removes a granted record lock of a transaction from the queue and grants
+locks to other transactions waiting in the queue if they now are entitled
+to a lock. */
+
+void
+lock_rec_unlock(
+/*============*/
+	trx_t*	trx,  		/* in: transaction that has set a record
+				lock */
+	rec_t*	rec,		/* in: record */
+	ulint	lock_mode)	/* in: LOCK_S or LOCK_X */
+{
+	lock_t*	lock;
+	lock_t*	release_lock	= NULL;
+	ulint	heap_no;
+
+	ut_ad(trx && rec);
+
+	mutex_enter(&kernel_mutex);
+
+	heap_no = rec_get_heap_no(rec, page_rec_is_comp(rec));
+
+	lock = lock_rec_get_first(rec);
+
+	/* Find the last lock with the same lock_mode and transaction
+	from the record. */
+
+	while (lock != NULL) {
+		if (lock->trx == trx && lock_get_mode(lock) == lock_mode) {
+			release_lock = lock;
+			ut_a(!lock_get_wait(lock));
+		}
+
+		lock = lock_rec_get_next(rec, lock);
+	}
+
+	/* If a record lock is found, release the record lock */
+
+	if (UNIV_LIKELY(release_lock != NULL)) {
+		lock_rec_reset_nth_bit(release_lock, heap_no);
+	} else {
+		mutex_exit(&kernel_mutex);
+		ut_print_timestamp(stderr);
+		fprintf(stderr,
+"  InnoDB: Error: unlock row could not find a %lu mode lock on the record\n",
+			(ulong)lock_mode);
+
+		return;
+	}
+
+	/* Check if we can now grant waiting lock requests */
+
+	lock = lock_rec_get_first(rec);
+
+	while (lock != NULL) {
+		if (lock_get_wait(lock)
+			&& !lock_rec_has_to_wait_in_queue(lock)) {
+
+			/* Grant the lock */
+			lock_grant(lock);
+		}
+
+		lock = lock_rec_get_next(rec, lock);
+	}
+
+	mutex_exit(&kernel_mutex);
+} 
+
 /*************************************************************************
 Releases a table lock.
 Releases possible other transactions waiting for this lock. */

innobase/os/os0file.c

@@ -160,13 +160,10 @@ time_t	os_last_printout;
 
 ibool	os_has_said_disk_full	= FALSE;
 
-/* The mutex protecting the following counts of pending pread and pwrite
-operations */
+/* The mutex protecting the following counts of pending I/O operations */
 static os_mutex_t os_file_count_mutex;
 ulint	os_file_n_pending_preads  = 0;
 ulint	os_file_n_pending_pwrites = 0;
-
-/* These are not protected by any mutex */
 ulint	os_n_pending_writes = 0;
 ulint	os_n_pending_reads = 0;
 
@@ -314,6 +311,8 @@ os_file_get_last_error(
 		return(OS_FILE_NOT_FOUND);
 	} else if (err == EEXIST) {
 		return(OS_FILE_ALREADY_EXISTS);
+	} else if (err == EXDEV || err == ENOTDIR || err == EISDIR) {
+		return(OS_FILE_PATH_ERROR);
 	} else {
 		return(100 + err);
 	}
@@ -363,7 +362,8 @@ os_file_handle_error(
 
 		return(TRUE);
 
-	} else if (err == OS_FILE_ALREADY_EXISTS) {
+	} else if (err == OS_FILE_ALREADY_EXISTS
+			|| err == OS_FILE_PATH_ERROR) {
 
 		return(FALSE);
 	} else {
@@ -467,7 +467,8 @@ os_file_handle_error_no_exit(
 
 		return(TRUE);
 
-	} else if (err == OS_FILE_ALREADY_EXISTS) {
+	} else if (err == OS_FILE_ALREADY_EXISTS
+			|| err == OS_FILE_PATH_ERROR) {
 
 		return(FALSE);
 	} else {
@@ -1905,12 +1906,14 @@ os_file_pread(
 #if defined(HAVE_PREAD) && !defined(HAVE_BROKEN_PREAD)
         os_mutex_enter(os_file_count_mutex);
 	os_file_n_pending_preads++;
+	os_n_pending_reads++;
         os_mutex_exit(os_file_count_mutex);
 
         n_bytes = pread(file, buf, (ssize_t)n, offs);
 
         os_mutex_enter(os_file_count_mutex);
 	os_file_n_pending_preads--;
+	os_n_pending_reads--;
         os_mutex_exit(os_file_count_mutex);
 
 	return(n_bytes);
@@ -1920,6 +1923,10 @@ os_file_pread(
 	ssize_t	ret;
 	ulint	i;
 
+        os_mutex_enter(os_file_count_mutex);
+	os_n_pending_reads++;
+        os_mutex_exit(os_file_count_mutex);
+
 	/* Protect the seek / read operation with a mutex */
 	i = ((ulint) file) % OS_FILE_N_SEEK_MUTEXES;
 	
@@ -1928,15 +1935,17 @@ os_file_pread(
 	ret_offset = lseek(file, offs, SEEK_SET);
 
 	if (ret_offset < 0) {
-		os_mutex_exit(os_file_seek_mutexes[i]);
-
-		return(-1);
-	}
-	
+		ret = -1;
+	} else {
 		ret = read(file, buf, (ssize_t)n);
+	}
 
 	os_mutex_exit(os_file_seek_mutexes[i]);
 
+        os_mutex_enter(os_file_count_mutex);
+	os_n_pending_reads--;
+        os_mutex_exit(os_file_count_mutex);
+
 	return(ret);
 	}
 #endif
@@ -1981,12 +1990,14 @@ os_file_pwrite(
 #if defined(HAVE_PWRITE) && !defined(HAVE_BROKEN_PREAD)
         os_mutex_enter(os_file_count_mutex);
 	os_file_n_pending_pwrites++;
+	os_n_pending_writes++;
         os_mutex_exit(os_file_count_mutex);
 
 	ret = pwrite(file, buf, (ssize_t)n, offs);
 
         os_mutex_enter(os_file_count_mutex);
 	os_file_n_pending_pwrites--;
+	os_n_pending_writes--;
         os_mutex_exit(os_file_count_mutex);
 
 # ifdef UNIV_DO_FLUSH
@@ -2008,6 +2019,10 @@ os_file_pwrite(
 	off_t	ret_offset;
 	ulint	i;
 
+	os_mutex_enter(os_file_count_mutex);
+	os_n_pending_writes++;
+	os_mutex_exit(os_file_count_mutex);
+
 	/* Protect the seek / write operation with a mutex */
 	i = ((ulint) file) % OS_FILE_N_SEEK_MUTEXES;
 	
@@ -2016,9 +2031,9 @@ os_file_pwrite(
 	ret_offset = lseek(file, offs, SEEK_SET);
 
 	if (ret_offset < 0) {
-		os_mutex_exit(os_file_seek_mutexes[i]);
+		ret = -1;
 
-		return(-1);
+		goto func_exit;
 	}
 	
 	ret = write(file, buf, (ssize_t)n);
@@ -2036,8 +2051,13 @@ os_file_pwrite(
 	}
 # endif /* UNIV_DO_FLUSH */
 
+func_exit:
 	os_mutex_exit(os_file_seek_mutexes[i]);
 
+	os_mutex_enter(os_file_count_mutex);
+	os_n_pending_writes--;
+	os_mutex_exit(os_file_count_mutex);
+
 	return(ret);
 	}
 #endif
@@ -2082,6 +2102,10 @@ os_file_read(
 	low = (DWORD) offset;
 	high = (DWORD) offset_high;
 
+	os_mutex_enter(os_file_count_mutex);
+	os_n_pending_reads++;
+	os_mutex_exit(os_file_count_mutex);
+
 	/* Protect the seek / read operation with a mutex */
 	i = ((ulint) file) % OS_FILE_N_SEEK_MUTEXES;
 
@@ -2093,17 +2117,21 @@ os_file_read(
 
 		os_mutex_exit(os_file_seek_mutexes[i]);
 
+		os_mutex_enter(os_file_count_mutex);
+		os_n_pending_reads--;
+		os_mutex_exit(os_file_count_mutex);
+
 		goto error_handling;
 	} 
 	
-        os_n_pending_reads++;
-        
 	ret = ReadFile(file, buf, (DWORD) n, &len, NULL);
 
-        os_n_pending_reads--;
-
 	os_mutex_exit(os_file_seek_mutexes[i]);
 	
+	os_mutex_enter(os_file_count_mutex);
+	os_n_pending_reads--;
+	os_mutex_exit(os_file_count_mutex);
+
 	if (ret && len == n) {
 		return(TRUE);
 	}		
@@ -2114,12 +2142,8 @@ os_file_read(
 	os_bytes_read_since_printout += n;
 
 try_again:
-        os_n_pending_reads++;
-        
 	ret = os_file_pread(file, buf, n, offset, offset_high);
 
-        os_n_pending_reads--;
-
 	if ((ulint)ret == n) {
 
 		return(TRUE);
@@ -2193,6 +2217,10 @@ os_file_read_no_error_handling(
 	low = (DWORD) offset;
 	high = (DWORD) offset_high;
 
+	os_mutex_enter(os_file_count_mutex);
+	os_n_pending_reads++;
+	os_mutex_exit(os_file_count_mutex);
+
 	/* Protect the seek / read operation with a mutex */
 	i = ((ulint) file) % OS_FILE_N_SEEK_MUTEXES;
 	
@@ -2204,17 +2232,21 @@ os_file_read_no_error_handling(
 
 		os_mutex_exit(os_file_seek_mutexes[i]);
 
+		os_mutex_enter(os_file_count_mutex);
+		os_n_pending_reads--;
+		os_mutex_exit(os_file_count_mutex);
+
 		goto error_handling;
 	} 
 	
-        os_n_pending_reads++;
-        
 	ret = ReadFile(file, buf, (DWORD) n, &len, NULL);
 
-        os_n_pending_reads--;
-        
 	os_mutex_exit(os_file_seek_mutexes[i]);
 	
+	os_mutex_enter(os_file_count_mutex);
+	os_n_pending_reads--;
+	os_mutex_exit(os_file_count_mutex);
+
 	if (ret && len == n) {
 		return(TRUE);
 	}		
@@ -2225,12 +2257,8 @@ os_file_read_no_error_handling(
 	os_bytes_read_since_printout += n;
 
 try_again:
-        os_n_pending_reads++;
-
 	ret = os_file_pread(file, buf, n, offset, offset_high);
 
-        os_n_pending_reads--;
-
 	if ((ulint)ret == n) {
 
 		return(TRUE);
@@ -2310,6 +2338,10 @@ os_file_write(
 	low = (DWORD) offset;
 	high = (DWORD) offset_high;
 	
+	os_mutex_enter(os_file_count_mutex);
+	os_n_pending_writes++;
+	os_mutex_exit(os_file_count_mutex);
+
 	/* Protect the seek / write operation with a mutex */
 	i = ((ulint) file) % OS_FILE_N_SEEK_MUTEXES;
 	
@@ -2321,6 +2353,10 @@ os_file_write(
 
 		os_mutex_exit(os_file_seek_mutexes[i]);
 		
+		os_mutex_enter(os_file_count_mutex);
+		os_n_pending_writes--;
+		os_mutex_exit(os_file_count_mutex);
+
 		ut_print_timestamp(stderr);
 
 		fprintf(stderr,
@@ -2335,12 +2371,8 @@ os_file_write(
 		return(FALSE);
 	} 
 
-        os_n_pending_writes++;
-        
 	ret = WriteFile(file, buf, (DWORD) n, &len, NULL);
 
-        os_n_pending_writes--;
-	
 	/* Always do fsync to reduce the probability that when the OS crashes,
 	a database page is only partially physically written to disk. */
 
@@ -2352,6 +2384,10 @@ os_file_write(
 
 	os_mutex_exit(os_file_seek_mutexes[i]);
 
+	os_mutex_enter(os_file_count_mutex);
+	os_n_pending_writes--;
+	os_mutex_exit(os_file_count_mutex);
+
 	if (ret && len == n) {
 
 		return(TRUE);
@@ -2402,12 +2438,8 @@ os_file_write(
 #else
 	ssize_t	ret;
 	
-        os_n_pending_writes++;
-        
 	ret = os_file_pwrite(file, buf, n, offset, offset_high);
         
-        os_n_pending_writes--;
-        
 	if ((ulint)ret == n) {
 
 		return(TRUE);

innobase/row/row0ins.c

@@ -549,6 +549,15 @@ row_ins_cascade_calc_update_vec(
 					default:
 						ut_error;
 					case 1:
+						if (UNIV_UNLIKELY(
+							dtype_get_charset_coll(
+							dtype_get_prtype(type))
+					== DATA_MYSQL_BINARY_CHARSET_COLL)) {
+						    /* Do not pad BINARY
+						    columns. */
+						    return(ULINT_UNDEFINED);
+						}
+
 						/* space=0x20 */
 						memset(pad_start, 0x20,
 							pad_end - pad_start);

innobase/row/row0mysql.c

@@ -1436,7 +1436,8 @@ trx_register_new_rec_lock() to store the information which new record locks
 really were set. This function removes a newly set lock under prebuilt->pcur,
 and also under prebuilt->clust_pcur. Currently, this is only used and tested
 in the case of an UPDATE or a DELETE statement, where the row lock is of the
-LOCK_X type.
+LOCK_X or LOCK_S type. 
+
 Thus, this implements a 'mini-rollback' that releases the latest record 
 locks we set. */
 
@@ -1474,7 +1475,14 @@ row_unlock_for_mysql(
 
 	index = btr_pcur_get_btr_cur(pcur)->index;
 
-	if (index != NULL && trx_new_rec_locks_contain(trx, index)) {
+	if (UNIV_UNLIKELY(index == NULL)) {
+		fprintf(stderr,
+"InnoDB: Error: Index is not set for persistent cursor.\n");
+		ut_print_buf(stderr, (const byte*)pcur, sizeof(btr_pcur_t));
+		ut_error;
+	}
+
+	if (trx_new_rec_locks_contain(trx, index)) {
 
 		mtr_start(&mtr);
 			
@@ -1486,11 +1494,7 @@ row_unlock_for_mysql(
 
 		rec = btr_pcur_get_rec(pcur);
 
-		mutex_enter(&kernel_mutex);
-
-		lock_rec_reset_and_release_wait(rec);
-
-		mutex_exit(&kernel_mutex);
+		lock_rec_unlock(trx, rec, prebuilt->select_lock_type);
 
 		mtr_commit(&mtr);
 
@@ -1520,11 +1524,7 @@ row_unlock_for_mysql(
 
 		rec = btr_pcur_get_rec(clust_pcur);
 
-		mutex_enter(&kernel_mutex);
-
-		lock_rec_reset_and_release_wait(rec);
-
-		mutex_exit(&kernel_mutex);
+		lock_rec_unlock(trx, rec, prebuilt->select_lock_type);
 
 		mtr_commit(&mtr);
 	}

innobase/row/row0sel.c

@@ -2771,8 +2771,8 @@ row_sel_get_clust_rec_for_mysql(
 func_exit:
 	*out_rec = clust_rec;
 
-	if (prebuilt->select_lock_type == LOCK_X) {
-		/* We may use the cursor in update: store its position */
+	if (prebuilt->select_lock_type != LOCK_NONE) {
+		/* We may use the cursor in unlock: store its position */
 		
 		btr_pcur_store_position(prebuilt->clust_pcur, mtr);
 	}
@@ -3972,12 +3972,12 @@ cursor lock count is done correctly. See bugs #12263 and #12456!
 	/* We have an optimization to save CPU time: if this is a consistent
 	read on a unique condition on the clustered index, then we do not
 	store the pcur position, because any fetch next or prev will anyway
-	return 'end of file'. An exception is the MySQL HANDLER command
-	where the user can move the cursor with PREV or NEXT even after
-	a unique search. */
+	return 'end of file'. Exceptions are locking reads and the MySQL 
+	HANDLER command where the user can move the cursor with PREV or NEXT 
+	even after a unique search. */
 
 	if (!unique_search_from_clust_index
-	    || prebuilt->select_lock_type == LOCK_X
+	    || prebuilt->select_lock_type != LOCK_NONE
 	    || prebuilt->used_in_HANDLER) {
 
 		/* Inside an update always store the cursor position */

innobase/trx/trx0trx.c

@@ -794,7 +794,8 @@ trx_commit_off_kernel(
 		in trx sys header if MySQL binlogging is on or the database
 		server is a MySQL replication slave */
 
-		if (trx->mysql_log_file_name) {
+		if (trx->mysql_log_file_name
+				&& trx->mysql_log_file_name[0] != '\0') {
 			trx_sys_update_mysql_binlog_offset(
 					trx->mysql_log_file_name,
 					trx->mysql_log_offset,

mysql-test/r/innodb_unsafe_binlog.result

+drop table if exists t1,t2;
+create table t1 (id int not null, f_id int not null, f int not null,
+primary key(f_id, id)) engine=innodb;
+create table t2 (id int not null,s_id int not null,s varchar(200),
+primary key(id)) engine=innodb;
+INSERT INTO t1 VALUES (8, 1, 3);
+INSERT INTO t1 VALUES (1, 2, 1);
+INSERT INTO t2 VALUES (1, 0, '');
+INSERT INTO t2 VALUES (8, 1, '');
+commit;
+DELETE ml.* FROM t1 AS ml LEFT JOIN t2 AS mm ON (mm.id=ml.id)
+WHERE mm.id IS NULL;
+select ml.* from t1 as ml left join t2 as mm on (mm.id=ml.id)
+where mm.id is null lock in share mode;
+id	f_id	f
+drop table t1,t2;
+create table t1 (id int not null, f_id int not null, f int not null,
+primary key(id),key(f_id)) engine=innodb;
+create table t2 (id int not null,s_id int not null,s varchar(200),
+primary key(id),key(s_id)) engine=innodb;
+INSERT INTO t1 VALUES (8, 1, 3);
+INSERT INTO t1 VALUES (1, 2, 1);
+INSERT INTO t2 VALUES (1, 0, '');
+INSERT INTO t2 VALUES (8, 1, '');
+commit;
+delete ml.* from t1 as ml left join t2 as mm on (mm.s_id=ml.f_id) where mm.s is null;
+select ml.* from t1 as ml left join t2 as mm on (mm.s_id=ml.f_id) where mm.s is null lock in share mode;
+id	f_id	f
+drop table t1,t2;
+create table t1(a int not null, b int, primary key(a)) engine=innodb;
+insert into t1 values(1,1),(2,2),(3,1),(4,2),(5,1),(6,2);
+commit;
+set autocommit = 0;
+select * from t1 lock in share mode;
+a	b
+1	1
+2	2
+3	1
+4	2
+5	1
+6	2
+update t1 set b = 5 where b = 1;
+set autocommit = 0;
+select * from t1 where a = 2 and b = 2 for update;
+ERROR HY000: Lock wait timeout exceeded; try restarting transaction
+commit;
+commit;
+drop table t1;

mysql-test/t/innodb_unsafe_binlog-master.opt

+--innodb_locks_unsafe_for_binlog=true

mysql-test/t/innodb_unsafe_binlog.test

+-- source include/have_innodb.inc
+#
+# Note that these tests uses a innodb_locks_unsafe_for_binlog option.
+#
+# Test cases for a bug #15650 DELETE with LEFT JOIN crashes server
+#
+
+--disable_warnings
+drop table if exists t1,t2;
+--enable_warnings
+create table t1 (id int not null, f_id int not null, f int not null,
+primary key(f_id, id)) engine=innodb;
+create table t2 (id int not null,s_id int not null,s varchar(200),
+primary key(id)) engine=innodb;
+INSERT INTO t1 VALUES (8, 1, 3);
+INSERT INTO t1 VALUES (1, 2, 1);
+INSERT INTO t2 VALUES (1, 0, '');
+INSERT INTO t2 VALUES (8, 1, '');
+commit;
+DELETE ml.* FROM t1 AS ml LEFT JOIN t2 AS mm ON (mm.id=ml.id)
+WHERE mm.id IS NULL;
+select ml.* from t1 as ml left join t2 as mm on (mm.id=ml.id)
+where mm.id is null lock in share mode;
+drop table t1,t2;
+
+create table t1 (id int not null, f_id int not null, f int not null,
+primary key(id),key(f_id)) engine=innodb;
+create table t2 (id int not null,s_id int not null,s varchar(200),
+primary key(id),key(s_id)) engine=innodb;
+INSERT INTO t1 VALUES (8, 1, 3);
+INSERT INTO t1 VALUES (1, 2, 1);
+INSERT INTO t2 VALUES (1, 0, '');
+INSERT INTO t2 VALUES (8, 1, '');
+commit;
+delete ml.* from t1 as ml left join t2 as mm on (mm.s_id=ml.f_id) where mm.s is null;
+select ml.* from t1 as ml left join t2 as mm on (mm.s_id=ml.f_id) where mm.s is null lock in share mode;
+drop table t1,t2;
+
+#
+# Test case for unlock row bug where unlock releases all locks granted for
+# a row. Only the latest lock should be released.
+#
+
+connect (a,localhost,root,,);
+connect (b,localhost,root,,);
+connection a;
+create table t1(a int not null, b int, primary key(a)) engine=innodb;
+insert into t1 values(1,1),(2,2),(3,1),(4,2),(5,1),(6,2);
+commit;
+set autocommit = 0; 
+select * from t1 lock in share mode;
+update t1 set b = 5 where b = 1;
+connection b;
+set autocommit = 0;
+#
+# S-lock to records (2,2),(4,2), and (6,2) should not be released in a update
+#
+--error 1205
+select * from t1 where a = 2 and b = 2 for update;
+connection a;
+commit;
+connection b;
+commit;
+drop table t1;
+disconnect a;
+disconnect b;
+

sql/ha_innodb.cc

@@ -1218,7 +1218,7 @@ innobase_init(void)
 				"innobase_buffer_pool_size can't be over 4GB"
 				" on 32-bit systems");
 
-			goto error;
+			DBUG_RETURN(0);
 		}
 
 		if (innobase_log_file_size > UINT_MAX32) {
@@ -1226,7 +1226,7 @@ innobase_init(void)
 				"innobase_log_file_size can't be over 4GB"
 				" on 32-bit systems");
 
-			goto error;
+			DBUG_RETURN(0);
 		}
 	}
 
@@ -1391,6 +1391,7 @@ innobase_init(void)
 
 	ut_a(DATA_MYSQL_LATIN1_SWEDISH_CHARSET_COLL ==
 					my_charset_latin1.number);
+	ut_a(DATA_MYSQL_BINARY_CHARSET_COLL == my_charset_bin.number);
 
 	/* Store the latin1_swedish_ci character ordering table to InnoDB. For
 	non-latin1_swedish_ci charsets we use the MySQL comparison functions,
@@ -2847,8 +2848,6 @@ ha_innobase::store_key_val_for_row(
 	char*		buff_start	= buff;
 	enum_field_types mysql_type;
 	Field*		field;
-	ulint		blob_len;
-	byte*		blob_data;
 	ibool		is_null;
 
   	DBUG_ENTER("store_key_val_for_row");
@@ -2903,14 +2902,18 @@ ha_innobase::store_key_val_for_row(
 			ulint	len;
 			byte*	data;
 			ulint	key_len;
+			ulint	true_len;
 			CHARSET_INFO*	cs;
 			int	error=0;
 
+			key_len = key_part->length;
+
 			if (is_null) {
-				buff += key_part->length + 2;
+				buff += key_len + 2;
 				
 				continue;
 			}
+			cs = field->charset();
 
 			lenlen = (ulint)
 				(((Field_varstring*)field)->length_bytes);
@@ -2920,32 +2923,33 @@ ha_innobase::store_key_val_for_row(
 				+ (ulint)get_field_offset(table, field)),
 				lenlen);
 
-			/* In a column prefix index, we may need to truncate
-			the stored value: */
+			true_len = len;
 
-			cs = key_part->field->charset();
+			/* For multi byte character sets we need to calculate
+			the true length of the key */
 
-			if (cs->mbmaxlen > 1 && key_part->length > 0) {
-				key_len = (ulint) cs->cset->well_formed_len(cs, 
+			if (len > 0 && cs->mbmaxlen > 1) {
+				true_len = (ulint) cs->cset->well_formed_len(cs,
 						(const char *) data,
-					(const char *) data + key_part->length,
-					key_part->length / cs->mbmaxlen, 
+						(const char *) data + len,
+						key_len / cs->mbmaxlen, 
 						&error);
-			} else {
-				key_len = key_part->length;
 			}
 
-			if (len > key_len) {
-				len = key_len;
+			/* In a column prefix index, we may need to truncate
+			the stored value: */
+
+			if (true_len > key_len) {
+				true_len = key_len;
 			}
 
 			/* The length in a key value is always stored in 2
 			bytes */
 
-			row_mysql_store_true_var_len((byte*)buff, len, 2);
+			row_mysql_store_true_var_len((byte*)buff, true_len, 2);
 			buff += 2;
 
-			memcpy(buff, data, len);
+			memcpy(buff, data, true_len);
 
 			/* Note that we always reserve the maximum possible
 			length of the true VARCHAR in the key value, though
@@ -2953,7 +2957,7 @@ ha_innobase::store_key_val_for_row(
 			actual data. The rest of the space was reset to zero
 			in the bzero() call above. */
 
-			buff += key_part->length;
+			buff += key_len;
 
 		} else if (mysql_type == FIELD_TYPE_TINY_BLOB
 		    || mysql_type == FIELD_TYPE_MEDIUM_BLOB
@@ -2963,58 +2967,66 @@ ha_innobase::store_key_val_for_row(
 			CHARSET_INFO*	cs;
 			ulint		key_len;
 			ulint		len;
+			ulint		true_len;
 			int		error=0;
+			ulint		blob_len;
+			byte*		blob_data;
 
 			ut_a(key_part->key_part_flag & HA_PART_KEY_SEG);
 
+			key_len = key_part->length;
+
 		        if (is_null) {
-				buff += key_part->length + 2;
+				buff += key_len + 2;
 				 
 				continue;
 			}
 		    
+			cs = field->charset();
+
 		        blob_data = row_mysql_read_blob_ref(&blob_len,
 				(byte*) (record
 				+ (ulint)get_field_offset(table, field)),
 					(ulint) field->pack_length());
 
+			true_len = blob_len;
+
 			ut_a(get_field_offset(table, field)
 						     == key_part->offset);
 
-			/* All indexes on BLOB and TEXT are column prefix
-			indexes, and we may need to truncate the data to be
-			stored in the key value: */
-
-			cs = key_part->field->charset();
+			/* For multi byte character sets we need to calculate
+			the true length of the key */
 			
-			if (cs->mbmaxlen > 1 && key_part->length > 0) {
-				key_len = (ulint) cs->cset->well_formed_len(cs, 
+			if (blob_len > 0 && cs->mbmaxlen > 1) {
+				true_len = (ulint) cs->cset->well_formed_len(cs,
 						(const char *) blob_data,
 						(const char *) blob_data 
-						+ key_part->length,
-					key_part->length / cs->mbmaxlen,
+							+ blob_len,
+						key_len / cs->mbmaxlen,
 						&error);
-			} else {
-				key_len = key_part->length;
 			}
 
-			if (blob_len > key_len) {
-				blob_len = key_len;
+			/* All indexes on BLOB and TEXT are column prefix
+			indexes, and we may need to truncate the data to be
+			stored in the key value: */
+
+			if (true_len > key_len) {
+				true_len = key_len;
 			}
 
 			/* MySQL reserves 2 bytes for the length and the
 			storage of the number is little-endian */
 
 			innobase_write_to_2_little_endian(
-					(byte*)buff, (ulint)blob_len);
+					(byte*)buff, true_len);
 			buff += 2;
 
-			memcpy(buff, blob_data, blob_len);
+			memcpy(buff, blob_data, true_len);
 
 			/* Note that we always reserve the maximum possible
 			length of the BLOB prefix in the key value. */
 
-			buff += key_part->length;
+			buff += key_len;
 		} else {
 			/* Here we handle all other data types except the
 			true VARCHAR, BLOB and TEXT. Note that the column
@@ -3022,38 +3034,64 @@ ha_innobase::store_key_val_for_row(
 			index. */
 
 			CHARSET_INFO*		cs;
-			ulint			len;
+			ulint			true_len;
+			ulint			key_len;
 			const mysql_byte*	src_start;
 			int			error=0;
+			enum_field_types	real_type;
+
+			key_len = key_part->length;
 
 		        if (is_null) {
-				 buff += key_part->length;
+				 buff += key_len;
 				 
 				 continue;
 			}
 
-			cs = key_part->field->charset();
 			src_start = record + key_part->offset;
+			real_type = field->real_type();
+			true_len = key_len;
+
+			/* Character set for the field is defined only
+			to fields whose type is string and real field
+			type is not enum or set. For these fields check
+			if character set is multi byte. */
+
+			if (real_type != FIELD_TYPE_ENUM 
+				&& real_type != FIELD_TYPE_SET
+				&& ( mysql_type == MYSQL_TYPE_VAR_STRING
+					|| mysql_type == MYSQL_TYPE_STRING)) {
+
+				cs = field->charset();
+
+				/* For multi byte character sets we need to 
+				calculate the true length of the key */
+
+				if (key_len > 0 && cs->mbmaxlen > 1) {
 
-			if (key_part->length > 0 && cs->mbmaxlen > 1) {
-				len = (ulint) cs->cset->well_formed_len(cs, 
-					(const char *) src_start,
-					(const char *) src_start + key_part->length,
-					key_part->length / cs->mbmaxlen, 
+					true_len = (ulint) 
+						cs->cset->well_formed_len(cs,
+							(const char *)src_start,
+							(const char *)src_start 
+								+ key_len,
+							key_len / cs->mbmaxlen,
 							&error);
-			} else {
-				len = key_part->length;
+				}
 			}
 
-			memcpy(buff, src_start, len);
-			buff+=len;
+			memcpy(buff, src_start, true_len);
+			buff += true_len;
 
-			/* Pad the unused space with spaces */
+			/* Pad the unused space with spaces. Note that no 
+			padding is ever needed for UCS-2 because in MySQL, 
+			all UCS2 characters are 2 bytes, as MySQL does not 
+			support surrogate pairs, which are needed to represent
+			characters in the range U+10000 to U+10FFFF. */
 
-			if (len < key_part->length) {
-				len = key_part->length - len;
-				memset(buff, ' ', len);
-				buff+=len;
+			if (true_len < key_len) {
+				ulint pad_len = key_len - true_len;
+				memset(buff, ' ', pad_len);
+				buff += pad_len;
 			}
 		}
   	}
@@ -3770,9 +3808,8 @@ ha_innobase::delete_row(
 }
 
 /**************************************************************************
-Removes a new lock set on a row. This can be called after a row has been read
-in the processing of an UPDATE or a DELETE query, if the option
-innodb_locks_unsafe_for_binlog is set. */
+Removes a new lock set on a row. This method does nothing unless the 
+option innodb_locks_unsafe_for_binlog is set.*/
 
 void
 ha_innobase::unlock_row(void)
@@ -3794,6 +3831,9 @@ ha_innobase::unlock_row(void)
 	if (srv_locks_unsafe_for_binlog) {
 		row_unlock_for_mysql(prebuilt, FALSE);
 	}
+
+	DBUG_VOID_RETURN;
+
 }
 
 /**********************************************************************