Many files:

  Implement ON DELETE CASCADE and facilitate switching off of UNIQUE constraints and foreign keys

innobase/btr/btr0cur.c

@@ -228,6 +228,7 @@ btr_cur_search_to_nth_level(
 	ulint		insert_planned;
 	ulint		buf_mode;
 	ulint		estimate;
+	ulint		ignore_sec_unique;
 	ulint		root_height;
 #ifdef BTR_CUR_ADAPT
 	btr_search_t*	info;
@@ -246,7 +247,9 @@ btr_cur_search_to_nth_level(
 #endif	
 	insert_planned = latch_mode & BTR_INSERT;
 	estimate = latch_mode & BTR_ESTIMATE;
-	latch_mode = latch_mode & ~(BTR_INSERT | BTR_ESTIMATE);
+	ignore_sec_unique = latch_mode & BTR_IGNORE_SEC_UNIQUE;
+	latch_mode = latch_mode & ~(BTR_INSERT | BTR_ESTIMATE
+					| BTR_IGNORE_SEC_UNIQUE);
 
 	ut_ad(!insert_planned || (mode == PAGE_CUR_LE));
 	
@@ -343,7 +346,8 @@ btr_cur_search_to_nth_level(
 
 			rw_latch = latch_mode;
 
-			if (insert_planned && ibuf_should_try(index)) {
+			if (insert_planned && ibuf_should_try(index,
+							ignore_sec_unique)) {
 				
 				/* Try insert to the insert buffer if the
 				page is not in the buffer pool */
@@ -356,7 +360,6 @@ retry_page_get:
 					buf_mode,
 					IB__FILE__, __LINE__,
 					mtr);
-
 		if (page == NULL) {
 			/* This must be a search to perform an insert;
 			try insert to the insert buffer */
@@ -365,7 +368,7 @@ retry_page_get:
 			ut_ad(insert_planned);
 			ut_ad(cursor->thr);
 
-			if (ibuf_should_try(index) &&
+			if (ibuf_should_try(index, ignore_sec_unique) &&
 				ibuf_insert(tuple, index, space, page_no,
 							cursor->thr)) {
 				/* Insertion to the insert buffer succeeded */

innobase/dict/dict0crea.c

@@ -1201,7 +1201,8 @@ loop:
 					ut_dulint_get_low(id),
 					table->name,
 					foreign->referenced_table_name,
-					foreign->n_fields);
+					foreign->n_fields
+					+ (foreign->type << 24));
 
 	for (i = 0; i < foreign->n_fields; i++) {
 

innobase/dict/dict0dict.c

@@ -1648,7 +1648,7 @@ dict_foreign_find_index(
 							->col->name;
 				if (ut_strlen(columns[i]) !=
 						ut_strlen(col_name)
-				    || 0 != ut_memcmp(columns[i],
+				    || 0 != ut_cmp_in_lower_case(columns[i],
 				    			col_name,
 				  			ut_strlen(col_name))) {
 				  	break;
@@ -1853,8 +1853,9 @@ dict_scan_col(
 	ibool*		success,/* out: TRUE if success */
 	dict_table_t*	table,	/* in: table in which the column is */
 	dict_col_t**	column,	/* out: pointer to column if success */
-	char**		column_name)/* out: pointer to column->name if
+	char**		column_name,/* out: pointer to column->name if
 				success */
+	ulint*		column_name_len)/* out: column name length */
 {
 	dict_col_t*	col;
 	char*		old_ptr;
@@ -1882,20 +1883,28 @@ dict_scan_col(
 		ptr++;
 	}
 
-	for (i = 0; i < dict_table_get_n_cols(table); i++) {
+	*column_name_len = (ulint)(ptr - old_ptr);
+	
+	if (table == NULL) {
+		*success = TRUE;
+		*column = NULL;
+		*column_name = old_ptr;
+	} else {
+	    	for (i = 0; i < dict_table_get_n_cols(table); i++) {
 
-		col = dict_table_get_nth_col(table, i);
+			col = dict_table_get_nth_col(table, i);
 
-		if (ut_strlen(col->name) == (ulint)(ptr - old_ptr)
-		    && 0 == ut_cmp_in_lower_case(col->name, old_ptr,
+			if (ut_strlen(col->name) == (ulint)(ptr - old_ptr)
+			    && 0 == ut_cmp_in_lower_case(col->name, old_ptr,
 						(ulint)(ptr - old_ptr))) {
-		    	/* Found */
+		    		/* Found */
 
-		    	*success = TRUE;
-		    	*column = col;
-		    	*column_name = col->name;
+		    		*success = TRUE;
+		    		*column = col;
+		    		*column_name = col->name;
 
-		    	break;
+		    		break;
+			}
 		}
 	}
 	
@@ -1914,14 +1923,18 @@ dict_scan_table_name(
 /*=================*/
 				/* out: scanned to */
 	char*		ptr,	/* in: scanned to */
-	dict_table_t**	table,	/* out: table object or NULL if error */
-	char*		name)	/* in: foreign key table name */
+	dict_table_t**	table,	/* out: table object or NULL */
+	char*		name,	/* in: foreign key table name */
+	ibool*		success,/* out: TRUE if ok name found */
+	char*		second_table_name)/* in/out: buffer where to store
+				the referenced table name; must be at least
+				2500 bytes */
 {
 	char*	dot_ptr			= NULL;
 	char*	old_ptr;
 	ulint	i;
-	char	second_table_name[10000];
 	
+	*success = FALSE;
 	*table = NULL;
 
 	while (isspace(*ptr)) {
@@ -1947,7 +1960,7 @@ dict_scan_table_name(
 		ptr++;
 	}
 
-	if (ptr - old_ptr > 9000) {
+	if (ptr - old_ptr > 2000) {
 		return(old_ptr);
 	}
 	
@@ -1978,6 +1991,8 @@ dict_scan_table_name(
 		second_table_name[ptr - old_ptr] = '\0';
 	}
 
+	*success = TRUE;
+
 	*table = dict_table_get_low(second_table_name);
 
 	if (*ptr == '`') {
@@ -2043,9 +2058,12 @@ dict_create_foreign_constraints(
 	ibool		success;
 	ulint		error;
 	ulint		i;
-	dict_col_t*	columns[1000];
-	char*		column_names[1000];
-
+	ulint		j;
+	dict_col_t*	columns[500];
+	char*		column_names[500];
+	ulint		column_name_lens[500];
+	char		referenced_table_name[2500];
+	
 	ut_ad(mutex_own(&(dict_sys->mutex)));
 
 	table = dict_table_get_low(name);
@@ -2090,7 +2108,7 @@ loop:
 	/* Scan the columns in the first list */
 col_loop1:
 	ptr = dict_scan_col(ptr, &success, table, columns + i,
-							column_names + i);
+				column_names + i, column_name_lens + i);
 	if (!success) {
 		return(DB_CANNOT_ADD_CONSTRAINT);
 	}
@@ -2141,9 +2159,13 @@ col_loop1:
 					1 + ut_strlen(columns[i]->name));
 	}
 	
-	ptr = dict_scan_table_name(ptr, &referenced_table, name);
+	ptr = dict_scan_table_name(ptr, &referenced_table, name,
+					&success, referenced_table_name);
 
-	if (!referenced_table) {
+	/* Note that referenced_table can be NULL if the user has suppressed
+	checking of foreign key constraints! */
+
+	if (!success || (!referenced_table && trx->check_foreigns)) {
 		dict_foreign_free(foreign);
 
 		return(DB_CANNOT_ADD_CONSTRAINT);
@@ -2161,7 +2183,7 @@ col_loop1:
 
 col_loop2:
 	ptr = dict_scan_col(ptr, &success, referenced_table, columns + i,
-							column_names + i);
+				column_names + i, column_name_lens + i);
 	i++;
 	
 	if (!success) {
@@ -2183,43 +2205,104 @@ col_loop2:
 		return(DB_CANNOT_ADD_CONSTRAINT);
 	}
 
+	ptr = dict_accept(ptr, "ON", &success);
+
+	if (!success) {
+
+		goto try_find_index;
+	}
+
+	ptr = dict_accept(ptr, "DELETE", &success);
+
+	if (!success) {
+
+		goto try_find_index;
+	}
+
+	ptr = dict_accept(ptr, "CASCADE", &success);
+
+	if (success) {
+
+		foreign->type = DICT_FOREIGN_ON_DELETE_CASCADE;
+
+		goto try_find_index;
+	}
+
+	ptr = dict_accept(ptr, "SET", &success);
+
+	if (!success) {
+
+		goto try_find_index;
+	}
+
+	ptr = dict_accept(ptr, "NULL", &success);
+
+	if (success) {
+		for (j = 0; j < foreign->n_fields; j++) {
+			if ((dict_index_get_nth_type(
+				foreign->foreign_index, j)->prtype)
+				& DATA_NOT_NULL) {
+
+				/* It is not sensible to define SET NULL
+				if the column is not allowed to be NULL! */
+
+				dict_foreign_free(foreign);
+				return(DB_CANNOT_ADD_CONSTRAINT);
+			}
+		}
+
+		foreign->type = DICT_FOREIGN_ON_DELETE_SET_NULL;
+
+		goto try_find_index;
+	}
+	
+try_find_index:
 	/* Try to find an index which contains the columns as the first fields
 	and in the right order, and the types are the same as in
 	foreign->foreign_index */
 
-	index = dict_foreign_find_index(referenced_table, column_names, i,
+	if (referenced_table) {
+		index = dict_foreign_find_index(referenced_table,
+						column_names, i,
 						foreign->foreign_index);
-
-	if (!index) {
-		dict_foreign_free(foreign);
-		return(DB_CANNOT_ADD_CONSTRAINT);
+		if (!index) {
+			dict_foreign_free(foreign);
+			return(DB_CANNOT_ADD_CONSTRAINT);
+		}
+	} else {
+		ut_a(trx->check_foreigns == FALSE);
+		index = NULL;
 	}
 
 	foreign->referenced_index = index;
 	foreign->referenced_table = referenced_table;
 
 	foreign->referenced_table_name = mem_heap_alloc(foreign->heap,
-					1 + ut_strlen(referenced_table->name));
+					1 + ut_strlen(referenced_table_name));
 
-	ut_memcpy(foreign->referenced_table_name, referenced_table->name,
-					1 + ut_strlen(referenced_table->name));
+	ut_memcpy(foreign->referenced_table_name, referenced_table_name,
+					1 + ut_strlen(referenced_table_name));
 					
 	foreign->referenced_col_names = mem_heap_alloc(foreign->heap,
 							i * sizeof(void*));
 	for (i = 0; i < foreign->n_fields; i++) {
 		foreign->referenced_col_names[i]
 				= mem_heap_alloc(foreign->heap,
-					1 + ut_strlen(columns[i]->name));
-		ut_memcpy(
-			foreign->referenced_col_names[i], columns[i]->name,
-					1 + ut_strlen(columns[i]->name));
+					1 + column_name_lens[i]);
+		ut_memcpy(foreign->referenced_col_names[i], column_names[i],
+					column_name_lens[i]);
+		(foreign->referenced_col_names[i])[column_name_lens[i]] = '\0';
 	}
 
 	/* We found an ok constraint definition: add to the lists */
 	
 	UT_LIST_ADD_LAST(foreign_list, table->foreign_list, foreign);
-	UT_LIST_ADD_LAST(referenced_list, referenced_table->referenced_list,
+
+	if (referenced_table) {
+		UT_LIST_ADD_LAST(referenced_list,
+					referenced_table->referenced_list,
 								foreign);
+	}
 	goto loop;
 }
 
@@ -3034,6 +3117,14 @@ dict_print_info_on_foreign_keys_in_create_format(
 
 		buf2 += sprintf(buf2, ")");
 
+		if (foreign->type == DICT_FOREIGN_ON_DELETE_CASCADE) {
+			buf2 += sprintf(buf2, " ON DELETE CASCADE");
+		}
+	
+		if (foreign->type == DICT_FOREIGN_ON_DELETE_SET_NULL) {
+			buf2 += sprintf(buf2, " ON DELETE SET NULL");
+		}
+
 		foreign = UT_LIST_GET_NEXT(foreign_list, foreign);
 	}
 

innobase/dict/dict0load.c

@@ -946,6 +946,11 @@ dict_load_foreign(
 	foreign->n_fields = mach_read_from_4(rec_get_nth_field(rec, 5, &len));
 
 	ut_a(len == 4);
+
+	/* We store the type to the bits 24-31 of n_fields */
+	
+	foreign->type = foreign->n_fields >> 24;
+	foreign->n_fields = foreign->n_fields & 0xFFFFFF;
 	
 	foreign->id = mem_heap_alloc(foreign->heap, ut_strlen(id) + 1);
 				

innobase/dict/dict0mem.c

@@ -61,7 +61,8 @@ dict_mem_table_create(
 	table->mem_fix = 0;
 
 	table->n_mysql_handles_opened = 0;
-
+	table->n_foreign_key_checks_running = 0;
+		
 	table->cached = FALSE;
 	
 	table->cols = mem_heap_alloc(heap, (n_cols + DATA_N_SYS_COLS)
@@ -235,6 +236,7 @@ dict_mem_foreign_create(void)
 
 	foreign->id = NULL;
 
+	foreign->type = 0;
 	foreign->foreign_table_name = NULL;
 	foreign->foreign_table = NULL;
 	foreign->foreign_col_names = NULL;

innobase/include/btr0btr.h

@@ -49,6 +49,12 @@ inserted to the index, at the searched position */
 /* This flag ORed to latch mode says that we do the search in query
 optimization */
 #define BTR_ESTIMATE		1024
+
+/* This flag ORed to latch mode says that we can ignore possible
+UNIQUE definition on secondary indexes when we decide if we can use the
+insert buffer to speed up inserts */
+#define BTR_IGNORE_SEC_UNIQUE	2048	
+
 /******************************************************************
 Gets a buffer page and declares its latching order level. */
 UNIV_INLINE

innobase/include/dict0mem.h

@@ -249,6 +249,8 @@ struct dict_foreign_struct{
 					this memory heap */
 	char*		id;		/* id of the constraint as a
 					null-terminated string */
+	ulint		type;		/* 0 or DICT_FOREIGN_ON_DELETE_CASCADE
+					or DICT_FOREIGN_ON_DELETE_SET_NULL */
 	char*		foreign_table_name;/* foreign table name */
 	dict_table_t*	foreign_table;	/* table where the foreign key is */
 	char**		foreign_col_names;/* names of the columns in the
@@ -278,6 +280,9 @@ struct dict_foreign_struct{
 					table */
 };
 
+#define DICT_FOREIGN_ON_DELETE_CASCADE	1
+#define DICT_FOREIGN_ON_DELETE_SET_NULL	2
+
 #define	DICT_INDEX_MAGIC_N	76789786
 
 /* Data structure for a database table */
@@ -313,6 +318,12 @@ struct dict_table_struct{
 				NOT allowed until this count gets to zero;
 				MySQL does NOT itself check the number of
 				open handles at drop */
+	ulint		n_foreign_key_checks_running;
+				/* count of how many foreign key check
+				operations are currently being performed
+				on the table: we cannot drop the table while
+				there are foreign key checks running on
+				it! */
 	ibool		cached;	/* TRUE if the table object has been added
 				to the dictionary cache */
 	lock_t*		auto_inc_lock;/* a buffer for an auto-inc lock
@@ -359,17 +370,16 @@ struct dict_table_struct{
 			        after database startup or table creation */
 	ulint		stat_modified_counter;
 				/* when a row is inserted, updated, or deleted,
-				we add the row length to this number; we
-				calculate new estimates for the stat_...
-				values for the table and the indexes at an
-				interval of 2 GB or when about 1 / 16 of table
-				has been modified; also
-				when the estimate operation is called
-				for MySQL SHOW TABLE STATUS; the counter is
-				reset to zero at statistics calculation;
-				this counter
-				is not protected by any latch, because this
-				is only used for heuristics */
+				we add 1 to this number; we calculate new
+				estimates for the stat_... values for the
+				table and the indexes at an interval of 2 GB
+				or when about 1 / 16 of table has been
+				modified; also when the estimate operation is
+				called for MySQL SHOW TABLE STATUS; the
+				counter is reset to zero at statistics
+				calculation; this counter is not protected by
+				any latch, because this is only used for
+				heuristics */
 	/*----------------------*/
 	mutex_t		autoinc_mutex;
 				/* mutex protecting the autoincrement

innobase/include/ibuf0ibuf.h

@@ -127,7 +127,11 @@ UNIV_INLINE
 ibool
 ibuf_should_try(
 /*============*/
-	dict_index_t*	index);	/* in: index where to insert */
+	dict_index_t*	index,			/* in: index where to insert */
+	ulint		ignore_sec_unique);	/* in: if != 0, we should
+						ignore UNIQUE constraint on
+						a secondary index when we
+						decide */
 /**********************************************************************
 Returns TRUE if the current OS thread is performing an insert buffer
 routine. */

innobase/include/ibuf0ibuf.ic

@@ -81,10 +81,16 @@ UNIV_INLINE
 ibool
 ibuf_should_try(
 /*============*/
-	dict_index_t*	index)	/* in: index where to insert */
+	dict_index_t*	index,			/* in: index where to insert */
+	ulint		ignore_sec_unique)	/* in: if != 0, we should
+						ignore UNIQUE constraint on
+						a secondary index when we
+						decide */
 {
-	if (!(index->type & (DICT_CLUSTERED | DICT_UNIQUE))
-					&& ibuf->meter > IBUF_THRESHOLD) {
+	if (!(index->type & DICT_CLUSTERED)
+	    && (ignore_sec_unique || !(index->type & DICT_UNIQUE))
+	    && ibuf->meter > IBUF_THRESHOLD) {
+
 		ibuf_flush_count++;
 
 		if (ibuf_flush_count % 8 == 0) {

innobase/include/os0file.h

@@ -341,6 +341,8 @@ os_aio_windows_handle(
 	void**	message2,
 	ulint*	type);		/* out: OS_FILE_WRITE or ..._READ */
 #endif
+
+/* Currently we do not use Posix async i/o */
 #ifdef POSIX_ASYNC_IO
 /**************************************************************************
 This function is only used in Posix asynchronous i/o. Waits for an aio

innobase/include/os0sync.h

@@ -149,9 +149,9 @@ void
 os_mutex_free(
 /*==========*/
 	os_mutex_t	mutex);	/* in: mutex to free */
-#ifndef _WIN32
 /**************************************************************
-Acquires ownership of a fast mutex. */
+Acquires ownership of a fast mutex. Currently in Windows this is the same
+as os_fast_mutex_lock! */
 UNIV_INLINE
 ulint
 os_fast_mutex_trylock(
@@ -160,7 +160,6 @@ os_fast_mutex_trylock(
 						was reserved by another
 						thread */
 	os_fast_mutex_t*	fast_mutex);	/* in: mutex to acquire */
-#endif
 /**************************************************************
 Releases ownership of a fast mutex. */
 

innobase/include/os0sync.ic

@@ -10,9 +10,9 @@ Created 9/6/1995 Heikki Tuuri
 #include <winbase.h>
 #endif
 
-#ifndef _WIN32
 /**************************************************************
-Acquires ownership of a fast mutex. */
+Acquires ownership of a fast mutex. Currently in Windows this is the same
+as os_fast_mutex_lock! */
 UNIV_INLINE
 ulint
 os_fast_mutex_trylock(
@@ -23,20 +23,11 @@ os_fast_mutex_trylock(
 	os_fast_mutex_t*	fast_mutex)	/* in: mutex to acquire */
 {
 #ifdef __WIN__	
-	int	ret;
+	EnterCriticalSection(fast_mutex);
 
-	/* TODO: TryEnterCriticalSection is probably not found from
-	NT versions < 4! */
-	ret = TryEnterCriticalSection(fast_mutex);
-
-	if (ret) {
-		return(0);
-	}
-
-	return(1);
+	return(0);
 #else
 	return((ulint) pthread_mutex_trylock(fast_mutex));
 #endif
 }
-#endif
 

innobase/include/row0mysql.h

@@ -209,6 +209,27 @@ row_update_for_mysql(
 	row_prebuilt_t*	prebuilt);	/* in: prebuilt struct in MySQL
 					handle */
 /*************************************************************************
+Creates an query graph node of 'update' type to be used in the MySQL
+interface. */
+
+upd_node_t*
+row_create_update_node_for_mysql(
+/*=============================*/
+				/* out, own: update node */
+	dict_table_t*	table,	/* in: table to update */
+	mem_heap_t*	heap);	/* in: mem heap from which allocated */
+/**************************************************************************
+Does a cascaded delete or set null in a foreign key operation. */
+
+ulint
+row_update_cascade_for_mysql(
+/*=========================*/
+				/* out: error code or DB_SUCCESS */
+	que_thr_t*	thr,	/* in: query thread */
+	upd_node_t*	node,	/* in: update node used in the cascade
+				or set null operation */
+	dict_table_t*	table);	/* in: table where we do the operation */
+/*************************************************************************
 Does a table creation operation for MySQL. If the name of the created
 table ends to characters INNODB_MONITOR, then this also starts
 printing of monitor output by the master thread. */

innobase/include/row0upd.h

@@ -312,6 +312,11 @@ struct upd_node_struct{
 	ibool		in_mysql_interface;
 				/* TRUE if the update node was created
 				for the MySQL interface */
+	upd_node_t*	cascade_node;/* NULL or an update node template which
+				is used to implement ON DELETE CASCADE
+				or ... SET NULL for foreign keys */
+	mem_heap_t*	cascade_heap;/* NULL or a mem heap where the cascade
+				node is created */
 	sel_node_t*	select;	/* query graph subtree implementing a base
 				table cursor: the rows returned will be
 				updated */
@@ -322,6 +327,11 @@ struct upd_node_struct{
 				of the MySQL interface */
 	dict_table_t*	table;	/* table where updated */
 	upd_t*		update;	/* update vector for the row */
+	ulint		update_n_fields;
+				/* when this struct is used to implement
+				a cascade operation for foreign keys, we store
+				here the size of the buffer allocated for use
+				as the update vector */
 	sym_node_list_t	columns;/* symbol table nodes for the columns
 				to retrieve from the table */
 	ibool		has_clust_rec_x_lock;

innobase/include/sync0sync.h

@@ -359,12 +359,17 @@ V
 Memory pool mutex */
 
 /* Latching order levels */
+
+/* User transaction locks are higher than any of the latch levels below:
+no latches are allowed when a thread goes to wait for a normal table
+or row lock! */
+#define SYNC_USER_TRX_LOCK	9999
 #define SYNC_NO_ORDER_CHECK	3000	/* this can be used to suppress
 					latching order checking */
 #define	SYNC_LEVEL_NONE		2000	/* default: level not defined */
+#define	SYNC_FOREIGN_KEY_CHECK	1001
 #define SYNC_DICT		1000
 #define SYNC_DICT_AUTOINC_MUTEX	999
-#define	SYNC_FOREIGN_KEY_CHECK	998
 #define	SYNC_PURGE_IS_RUNNING	997
 #define SYNC_DICT_HEADER	995
 #define SYNC_IBUF_HEADER	914
@@ -429,7 +434,7 @@ implementation of a mutual exclusion semaphore. */
 struct mutex_struct {
 	ulint	lock_word;	/* This ulint is the target of the atomic
 				test-and-set instruction in Win32 */
-#ifndef _WIN32
+#if !defined(_WIN32) || !defined(UNIV_CAN_USE_X86_ASSEMBLER)
 	os_fast_mutex_t
 		os_fast_mutex;	/* In other systems we use this OS mutex
 				in place of lock_word */

innobase/include/sync0sync.ic

@@ -53,7 +53,7 @@ mutex_test_and_set(
 				1 */
 	mutex_t*	mutex)	/* in: mutex */
 {
-#ifdef _WIN32
+#if defined(_WIN32) && defined(UNIV_CAN_USE_X86_ASSEMBLER)
 	ulint	res;
 	ulint*	lw;		/* assembler code is used to ensure that
 				lock_word is loaded from memory */
@@ -120,7 +120,7 @@ mutex_reset_lock_word(
 /*==================*/
 	mutex_t*	mutex)	/* in: mutex */
 {
-#ifdef _WIN32
+#if defined(_WIN32) && defined(UNIV_CAN_USE_X86_ASSEMBLER)
 	ulint*	lw;		/* assembler code is used to ensure that
 				lock_word is loaded from memory */
 	ut_ad(mutex);

innobase/include/trx0sys.h

@@ -259,7 +259,7 @@ therefore 256 */
 /* The offset of the transaction system header on the page */
 #define	TRX_SYS		FSEG_PAGE_DATA
 
-/* Transaction system header; protected by trx_sys->mutex */
+/* Transaction system header */
 /*-------------------------------------------------------------*/
 #define	TRX_SYS_TRX_ID_STORE	0	/* the maximum trx id or trx number
 					modulo TRX_SYS_TRX_ID_UPDATE_MARGIN

innobase/include/trx0sys.ic

@@ -93,7 +93,6 @@ trx_sysf_get(
 {
 	trx_sysf_t*	header;
 
-	ut_ad(mutex_own(&(kernel_mutex)));
 	ut_ad(mtr);
 	
 	header = TRX_SYS + buf_page_get(TRX_SYS_SPACE, TRX_SYS_PAGE_NO,

innobase/include/trx0trx.h

@@ -298,6 +298,17 @@ struct trx_struct{
 					of view of concurrency control:
 					TRX_ACTIVE, TRX_COMMITTED_IN_MEMORY,
 					... */
+	ibool		check_foreigns;	/* normally TRUE, but if the user
+					wants to suppress foreign key checks,
+					(in table imports, for example) we
+					set this FALSE */
+	ibool		check_unique_secondary;
+					/* normally TRUE, but if the user
+					wants to speed up inserts by
+					suppressing unique key checks
+					for secondary indexes when we decide
+					if we can use the insert buffer for
+					them, we set this FALSE */
 	dulint		id;		/* transaction id */
 	dulint		no;		/* transaction serialization number ==
 					max trx id when the transaction is 
@@ -328,6 +339,9 @@ struct trx_struct{
                                         /* how many tables the current SQL
 					statement uses, except those
 					in consistent read */
+	ibool		has_dict_foreign_key_check_lock;
+					/* TRUE if the trx currently holds
+					an s-lock on dict_foreign_... */
         ibool           has_search_latch;
 			                /* TRUE if this trx has latched the
 			                search system latch in S-mode */

innobase/lock/lock0lock.c

@@ -1534,6 +1534,8 @@ lock_rec_enqueue_waiting(
 
 	if (que_thr_stop(thr)) {
 
+		ut_a(0);
+
 		return(DB_QUE_THR_SUSPENDED);
 	}
 		
@@ -2918,6 +2920,7 @@ lock_table_enqueue_waiting(
 	stopped anyway */
 
 	if (que_thr_stop(thr)) {
+		ut_a(0);
 
 		return(DB_QUE_THR_SUSPENDED);
 	}

innobase/pars/pars0opt.c

@@ -526,7 +526,8 @@ opt_search_plan_for_table(
 	dict_index_t*	best_index;
 	ulint		n_fields;
 	ulint		goodness;
-	ulint		last_op;
+	ulint		last_op		= 75946965;	/* Eliminate a Purify
+							warning */
 	ulint		best_goodness;
 	ulint		best_last_op;
 	ulint		mix_id_pos;

innobase/que/que0que.c

@@ -555,6 +555,12 @@ que_graph_free_recursive(
 			btr_pcur_free_for_mysql(upd->pcur);
 		}
 
+		que_graph_free_recursive(upd->cascade_node);		
+
+		if (upd->cascade_heap) {
+			mem_heap_free(upd->cascade_heap);
+		}
+		
 		que_graph_free_recursive(upd->select);
 
 		mem_heap_free(upd->heap);
@@ -1110,9 +1116,6 @@ que_thr_move_to_run_state_for_mysql(
 		trx->n_active_thrs++;
 
 		thr->is_active = TRUE;
-
-		ut_ad((thr->graph)->n_active_thrs == 1);
-		ut_ad(trx->n_active_thrs == 1);
 	}
 	
 	thr->state = QUE_THR_RUNNING;

innobase/row/row0ins.c

@@ -355,6 +355,223 @@ row_ins_dupl_error_with_rec(
 	return(FALSE);
 }	
 
+/*************************************************************************
+Either deletes or sets the referencing columns SQL NULL in a child row.
+Used in ON DELETE ... clause for foreign keys when a parent row is
+deleted. */
+static
+ulint
+row_ins_foreign_delete_or_set_null(
+/*===============================*/
+					/* out: DB_SUCCESS, DB_LOCK_WAIT,
+					or error code */
+	que_thr_t*	thr,		/* in: query thread whose run_node
+					is an update node */
+	dict_foreign_t*	foreign,	/* in: foreign key constraint whose
+					type is != 0 */
+	btr_pcur_t*	pcur,		/* in: cursor placed on a matching
+					index record in the child table */
+	mtr_t*		mtr)		/* in: mtr holding the latch of pcur
+					page */
+{
+	upd_node_t*	node;
+	upd_node_t*	cascade;
+	dict_table_t*	table		= foreign->foreign_table;
+	dict_index_t*	index;
+	dict_index_t*	clust_index;
+	dtuple_t*	ref;
+	mem_heap_t*	tmp_heap;
+	rec_t*		rec;
+	rec_t*		clust_rec;
+	upd_t*		update;
+	ulint		err;
+	ulint		i;
+	char		err_buf[1000];
+	
+	ut_a(thr && foreign && pcur && mtr);
+
+	node = thr->run_node;
+
+	if (node->cascade_node == NULL) {
+		/* Extend our query graph by creating a child to current
+		update node. The child is used in the cascade or set null
+		operation. */
+
+		node->cascade_heap = mem_heap_create(128);
+		node->cascade_node = row_create_update_node_for_mysql(
+						table, node->cascade_heap);
+		que_node_set_parent(node->cascade_node, node);
+	}
+
+	/* Initialize cascade_node to do the operation we want. Note that we
+	use the SAME cascade node to do all foreign key operations of the
+	SQL DELETE: the table of the cascade node may change if there are
+	several child tables to the table where the delete is done! */
+
+	cascade = node->cascade_node;
+	
+	cascade->table = table;
+
+	if (foreign->type == DICT_FOREIGN_ON_DELETE_CASCADE ) {
+		cascade->is_delete = TRUE;
+	} else {
+		cascade->is_delete = FALSE;
+
+		if (foreign->n_fields > cascade->update_n_fields) {
+			/* We have to make the update vector longer */
+
+			cascade->update = upd_create(foreign->n_fields,
+							node->cascade_heap);
+			cascade->update_n_fields = foreign->n_fields;
+		}
+	}
+
+	index = btr_pcur_get_btr_cur(pcur)->index;
+
+	rec = btr_pcur_get_rec(pcur);
+
+	if (index->type & DICT_CLUSTERED) {
+		/* pcur is already positioned in the clustered index of
+		the child table */
+	
+		clust_index = index;
+		clust_rec = rec;
+	} else {
+		/* We have to look for the record in the clustered index
+		in the child table */
+
+		clust_index = dict_table_get_first_index(table);
+
+		tmp_heap = mem_heap_create(256);
+		
+		ref = row_build_row_ref(ROW_COPY_POINTERS, index, rec,
+								tmp_heap);
+		btr_pcur_open_with_no_init(clust_index, ref,
+			PAGE_CUR_LE, BTR_SEARCH_LEAF,
+			cascade->pcur, 0, mtr);
+
+		mem_heap_free(tmp_heap);
+
+		clust_rec = btr_pcur_get_rec(cascade->pcur);
+	}
+
+	if (!page_rec_is_user_rec(clust_rec)) {
+	  	fprintf(stderr, "InnoDB: error in cascade of a foreign key op\n"
+		  	"InnoDB: index %s table %s\n", index->name,
+		  	index->table->name);
+
+	  	rec_sprintf(err_buf, 900, rec);
+	  	fprintf(stderr, "InnoDB: record %s\n", err_buf);
+
+	  	rec_sprintf(err_buf, 900, clust_rec);
+	  	fprintf(stderr, "InnoDB: clustered record %s\n", err_buf);
+
+	  	fprintf(stderr,
+			"InnoDB: Make a detailed bug report and send it\n");
+	  	fprintf(stderr, "InnoDB: to mysql@lists.mysql.com\n");
+
+		err = DB_SUCCESS;
+
+		goto nonstandard_exit_func;
+	}
+
+	/* Set an X-lock on the row to delete or update in the child table */
+
+	err = lock_table(0, table, LOCK_IX, thr);
+
+	if (err == DB_SUCCESS) {
+		err = lock_clust_rec_read_check_and_lock(0, clust_rec,
+						clust_index, LOCK_X, thr);
+	}
+	
+	if (err != DB_SUCCESS) {
+
+		goto nonstandard_exit_func;
+	}
+
+	if (rec_get_deleted_flag(clust_rec)) {
+		/* This should never happen since we already have an S-lock
+		on non-delete-marked clust_rec or secondary index record! */
+
+	  	fprintf(stderr,
+			"InnoDB: error 2 in cascade of a foreign key op\n"
+		  	"InnoDB: index %s table %s\n", index->name,
+		  	index->table->name);
+
+	  	rec_sprintf(err_buf, 900, rec);
+	  	fprintf(stderr, "InnoDB: record %s\n", err_buf);
+
+	  	rec_sprintf(err_buf, 900, clust_rec);
+	  	fprintf(stderr, "InnoDB: clustered record %s\n", err_buf);
+
+	  	fprintf(stderr,
+			"InnoDB: Make a detailed bug report and send it\n");
+	  	fprintf(stderr, "InnoDB: to mysql@lists.mysql.com\n");
+
+		ut_a(0);
+
+		goto nonstandard_exit_func;
+	}
+
+	if (foreign->type == DICT_FOREIGN_ON_DELETE_SET_NULL) {
+		/* Build the appropriate update vector which sets
+		foreign->n_fields first fields in rec to SQL NULL */
+
+		update = cascade->update;
+
+		update->info_bits = 0;
+		update->n_fields = foreign->n_fields;
+		
+		for (i = 0; i < foreign->n_fields; i++) {
+			(update->fields + i)->field_no
+				= dict_table_get_nth_col_pos(table,
+					dict_index_get_nth_col_no(index, i));
+			(update->fields + i)->exp = NULL;
+			(update->fields + i)->new_val.len = UNIV_SQL_NULL;
+			(update->fields + i)->new_val.data = NULL;
+			(update->fields + i)->extern_storage = FALSE;
+		}
+	}
+
+	/* Store pcur position and initialize or store the cascade node
+	pcur stored position */
+	
+	btr_pcur_store_position(pcur, mtr);
+	
+	if (index == clust_index) {
+		btr_pcur_copy_stored_position(cascade->pcur, pcur);
+	} else {
+		btr_pcur_store_position(cascade->pcur, mtr);
+	}
+		
+	mtr_commit(mtr);
+
+	ut_a(cascade->pcur->rel_pos == BTR_PCUR_ON);
+
+	cascade->state = UPD_NODE_UPDATE_CLUSTERED;
+	
+	err = row_update_cascade_for_mysql(thr, cascade,
+						foreign->foreign_table);
+	mtr_start(mtr);
+
+	/* Restore pcur position */
+	
+	btr_pcur_restore_position(BTR_SEARCH_LEAF, pcur, mtr);
+
+	return(err);
+
+nonstandard_exit_func:
+
+	btr_pcur_store_position(pcur, mtr);
+
+	mtr_commit(mtr);
+	mtr_start(mtr);
+
+	btr_pcur_restore_position(BTR_SEARCH_LEAF, pcur, mtr);
+
+	return(err);
+}
+
 /*************************************************************************
 Sets a shared lock on a record. Used in locking possible duplicate key
 records. */
@@ -416,6 +633,13 @@ row_ins_check_foreign_constraint(
 
 	ut_ad(rw_lock_own(&dict_foreign_key_check_lock, RW_LOCK_SHARED));
 
+	if (thr_get_trx(thr)->check_foreigns == FALSE) {
+		/* The user has suppressed foreign key checks currently for
+		this session */
+
+		return(DB_SUCCESS);
+	}
+
 	/* If any of the foreign key fields in entry is SQL NULL, we
 	suppress the foreign key check: this is compatible with Oracle,
 	for example */
@@ -478,8 +702,8 @@ row_ins_check_foreign_constraint(
 
 			goto next_rec;
 		}
-				
-		/* Try to place a lock on the index record */	
+		
+		/* Try to place a lock on the index record */
 
 		err = row_ins_set_shared_rec_lock(rec, check_index, thr);
 
@@ -501,11 +725,21 @@ row_ins_check_foreign_constraint(
 
 				if (check_ref) {			
 					err = DB_SUCCESS;
+
+					break;
+				} else if (foreign->type != 0) {
+					err =
+					  row_ins_foreign_delete_or_set_null(
+						thr, foreign, &pcur, &mtr);
+
+					if (err != DB_SUCCESS) {
+
+						break;
+					}
 				} else {
 					err = DB_ROW_IS_REFERENCED;
+					break;
 				}
-			
-				break;
 			}
 		}
 
@@ -534,6 +768,8 @@ next_rec:
 		}
 	}
 
+	btr_pcur_close(&pcur);
+
 	mtr_commit(&mtr);
 
 	/* Restore old value */
@@ -561,6 +797,10 @@ row_ins_check_foreign_constraints(
 {
 	dict_foreign_t*	foreign;
 	ulint		err;
+	trx_t*		trx;
+	ibool		got_s_lock	= FALSE;
+
+	trx = thr_get_trx(thr);
 
 	foreign = UT_LIST_GET_FIRST(table->foreign_list);
 
@@ -569,16 +809,26 @@ row_ins_check_foreign_constraints(
 
 			if (foreign->referenced_table == NULL) {
 				dict_table_get(foreign->referenced_table_name,
-						thr_get_trx(thr));
+									trx);
 			}
 
-			rw_lock_s_lock(&dict_foreign_key_check_lock);	
+			if (!trx->has_dict_foreign_key_check_lock) {
+				got_s_lock = TRUE;
+
+				rw_lock_s_lock(&dict_foreign_key_check_lock);
+
+				trx->has_dict_foreign_key_check_lock = TRUE;
+			}
 
 			err = row_ins_check_foreign_constraint(TRUE, foreign,
 						table, index, entry, thr);
+			if (got_s_lock) {
 
-			rw_lock_s_unlock(&dict_foreign_key_check_lock);	
+				rw_lock_s_unlock(&dict_foreign_key_check_lock);	
 
+				trx->has_dict_foreign_key_check_lock = FALSE;
+			}
+				
 			if (err != DB_SUCCESS) {
 				return(err);
 			}
@@ -868,13 +1118,14 @@ row_ins_index_entry_low(
 	ulint		n_ext_vec,/* in: number of fields in ext_vec */
 	que_thr_t*	thr)	/* in: query thread */
 {
-	btr_cur_t	cursor;		
+	btr_cur_t	cursor;
+	ulint		ignore_sec_unique	= 0;
 	ulint		modify;
 	rec_t*		insert_rec;
 	rec_t*		rec;
 	ulint		err;
 	ulint		n_unique;
-	big_rec_t*	big_rec		= NULL;
+	big_rec_t*	big_rec			= NULL;
 	mtr_t		mtr;
 	
 	log_free_check();
@@ -887,8 +1138,13 @@ row_ins_index_entry_low(
 	the function will return in both low_match and up_match of the
 	cursor sensible values */
 	
+	if (!(thr_get_trx(thr)->check_unique_secondary)) {
+		ignore_sec_unique = BTR_IGNORE_SEC_UNIQUE;
+	}
+
 	btr_cur_search_to_nth_level(index, 0, entry, PAGE_CUR_LE,
-					mode | BTR_INSERT, &cursor, 0, &mtr);
+				mode | BTR_INSERT | ignore_sec_unique,
+				&cursor, 0, &mtr);
 
 	if (cursor.flag == BTR_CUR_INSERT_TO_IBUF) {
 		/* The insertion was made to the insert buffer already during

innobase/row/row0mysql.c

@@ -499,29 +499,24 @@ UNIV_INLINE
 void
 row_update_statistics_if_needed(
 /*============================*/
-	row_prebuilt_t*	prebuilt)	/* in: prebuilt struct */
+	dict_table_t*	table)	/* in: table */
 {
 	ulint	counter;
 	
-	counter = prebuilt->table->stat_modified_counter;
-
-	/* Since the physical size of an InnoDB row is bigger than the
-	MySQL row len, we put a safety factor 2 below */
-
-	counter += 2 * prebuilt->mysql_row_len;
+	counter = table->stat_modified_counter;
 
-	prebuilt->table->stat_modified_counter = counter;
+	table->stat_modified_counter = counter + 1;
 
 	/* Calculate new statistics if 1 / 16 of table has been modified
 	since the last time a statistics batch was run, or if
-	stat_modified_counter > 2 000 000 000 (to avoid wrap-around) */
+	stat_modified_counter > 2 000 000 000 (to avoid wrap-around).
+	We calculate statistics at most every 16th round, since we may have
+	a counter table which is very small and updated very often. */
 
 	if (counter > 2000000000
-	    || ((ib_longlong)counter >
-		(UNIV_PAGE_SIZE * prebuilt->table->stat_clustered_index_size)
-		/ 16)) {
+	    || ((ib_longlong)counter > 16 + table->stat_n_rows / 16)) {
 
-		dict_update_statistics(prebuilt->table);
+		dict_update_statistics(table);
 	}	
 }
 		  	
@@ -712,7 +707,7 @@ run_again:
 		prebuilt->table->stat_n_rows--;
 	}	
 
-	row_update_statistics_if_needed(prebuilt);
+	row_update_statistics_if_needed(prebuilt->table);
 	trx->op_info = "";
 
 	return((int) err);
@@ -745,6 +740,43 @@ row_prebuild_sel_graph(
 	}
 }
 
+/*************************************************************************
+Creates an query graph node of 'update' type to be used in the MySQL
+interface. */
+
+upd_node_t*
+row_create_update_node_for_mysql(
+/*=============================*/
+				/* out, own: update node */
+	dict_table_t*	table,	/* in: table to update */
+	mem_heap_t*	heap)	/* in: mem heap from which allocated */
+{
+	upd_node_t*	node;
+
+	node = upd_node_create(heap);
+		
+	node->in_mysql_interface = TRUE;
+	node->is_delete = FALSE;
+	node->searched_update = FALSE;
+	node->select_will_do_update = FALSE;
+	node->select = NULL;
+	node->pcur = btr_pcur_create_for_mysql();
+	node->table = table;
+
+	node->update = upd_create(dict_table_get_n_cols(table), heap);
+
+	node->update_n_fields = dict_table_get_n_cols(table);
+	
+	UT_LIST_INIT(node->columns);
+	node->has_clust_rec_x_lock = TRUE;
+	node->cmpl_info = 0;
+
+	node->table_sym = NULL;
+	node->col_assign_list = NULL;
+
+	return(node);
+}
+
 /*************************************************************************
 Gets pointer to a prebuilt update vector used in updates. If the update
 graph has not yet been built in the prebuilt struct, then this function
@@ -767,26 +799,9 @@ row_get_prebuilt_update_vector(
 		/* Not called before for this handle: create an update node
 		and query graph to the prebuilt struct */
 
-		node = upd_node_create(prebuilt->heap);
-		
-		prebuilt->upd_node = node;
+		node = row_create_update_node_for_mysql(table, prebuilt->heap);
 
-		node->in_mysql_interface = TRUE;
-		node->is_delete = FALSE;
-		node->searched_update = FALSE;
-		node->select_will_do_update = FALSE;
-		node->select = NULL;
-		node->pcur = btr_pcur_create_for_mysql();
-		node->table = table;
-
-		node->update = upd_create(dict_table_get_n_cols(table),
-							prebuilt->heap);
-		UT_LIST_INIT(node->columns);
-		node->has_clust_rec_x_lock = TRUE;
-		node->cmpl_info = 0;
-
-		node->table_sym = NULL;
-		node->col_assign_list = NULL;
+		prebuilt->upd_node = node;
 		
 		prebuilt->upd_graph =
 			que_node_get_parent(
@@ -914,7 +929,7 @@ run_again:
 
 	que_thr_stop_for_mysql_no_error(thr, trx);
 
-	if (prebuilt->upd_node->is_delete) {
+	if (node->is_delete) {
 		if (prebuilt->table->stat_n_rows > 0) {
 			prebuilt->table->stat_n_rows--;
 		}
@@ -924,13 +939,66 @@ run_again:
 		srv_n_rows_updated++;
 	}
 
-	row_update_statistics_if_needed(prebuilt);
+	row_update_statistics_if_needed(prebuilt->table);
 
 	trx->op_info = "";
 
 	return((int) err);
 }
 
+/**************************************************************************
+Does a cascaded delete or set null in a foreign key operation. */
+
+ulint
+row_update_cascade_for_mysql(
+/*=========================*/
+				/* out: error code or DB_SUCCESS */
+	que_thr_t*	thr,	/* in: query thread */
+	upd_node_t*	node,	/* in: update node used in the cascade
+				or set null operation */
+	dict_table_t*	table)	/* in: table where we do the operation */
+{
+	ulint		err;
+	trx_t*		trx;
+
+	trx = thr_get_trx(thr);
+
+run_again:
+	thr->run_node = node;
+	thr->prev_node = node;
+
+	row_upd_step(thr);
+
+	err = trx->error_state;
+
+	if (err == DB_LOCK_WAIT) {
+		que_thr_stop_for_mysql(thr);
+	
+		row_mysql_handle_errors(&err, trx, thr, NULL);
+
+		goto run_again;
+	}
+
+	if (err != DB_SUCCESS) {
+
+		return(err);
+	}
+
+	if (node->is_delete) {
+		if (table->stat_n_rows > 0) {
+			table->stat_n_rows--;
+		}
+
+		srv_n_rows_deleted++;
+	} else {
+		srv_n_rows_updated++;
+	}
+
+	row_update_statistics_if_needed(table);
+
+	return(err);
+}
+
 /*************************************************************************
 Checks if a table is such that we automatically created a clustered
 index on it (on row id). */
@@ -1169,6 +1237,7 @@ row_create_table_for_mysql(
 	/* Serialize data dictionary operations with dictionary mutex:
 	no deadlocks can occur then in these operations */
 
+	rw_lock_x_lock(&(dict_foreign_key_check_lock));
 	mutex_enter(&(dict_sys->mutex));
 
 	heap = mem_heap_create(512);
@@ -1221,6 +1290,8 @@ row_create_table_for_mysql(
 	}
 
 	mutex_exit(&(dict_sys->mutex));
+	rw_lock_x_unlock(&(dict_foreign_key_check_lock));
+
 	que_graph_free((que_t*) que_node_get_parent(thr));
 
 	trx->op_info = "";
@@ -1268,6 +1339,7 @@ row_create_index_for_mysql(
 	/* Serialize data dictionary operations with dictionary mutex:
 	no deadlocks can occur then in these operations */
 
+	rw_lock_x_lock(&(dict_foreign_key_check_lock));
 	mutex_enter(&(dict_sys->mutex));
 
 	heap = mem_heap_create(512);
@@ -1298,6 +1370,7 @@ row_create_index_for_mysql(
 	}
 
 	mutex_exit(&(dict_sys->mutex));
+	rw_lock_x_unlock(&(dict_foreign_key_check_lock));
 
 	que_graph_free((que_t*) que_node_get_parent(thr));
 	
@@ -1353,6 +1426,7 @@ row_table_add_foreign_constraints(
 	/* Serialize data dictionary operations with dictionary mutex:
 	no deadlocks can occur then in these operations */
 
+	rw_lock_x_lock(&(dict_foreign_key_check_lock));
 	mutex_enter(&(dict_sys->mutex));
 
 	trx->dict_operation = TRUE;
@@ -1377,6 +1451,7 @@ row_table_add_foreign_constraints(
 	}
 
 	mutex_exit(&(dict_sys->mutex));
+	rw_lock_x_unlock(&(dict_foreign_key_check_lock));
 
 	return((int) err);
 }
@@ -1471,7 +1546,8 @@ loop:
 	        goto already_dropped;
 	}
 
-	if (table->n_mysql_handles_opened > 0) {
+	if (table->n_mysql_handles_opened > 0
+				|| table->n_foreign_key_checks_running > 0) {
 
 		return(n_tables + n_tables_dropped);
 	}
@@ -1717,6 +1793,9 @@ row_drop_table_for_mysql(
 	no deadlocks can occur then in these operations */
 
 	if (!has_dict_mutex) {
+		/* Prevent foreign key checks while we are dropping the table */
+		rw_lock_x_lock(&(dict_foreign_key_check_lock));
+
 		mutex_enter(&(dict_sys->mutex));
 	}
 
@@ -1729,9 +1808,6 @@ row_drop_table_for_mysql(
 
 	graph->fork_type = QUE_FORK_MYSQL_INTERFACE;
 
-	/* Prevent foreign key checks while we are dropping the table */
-	rw_lock_x_lock(&(dict_foreign_key_check_lock));
-
 	/* Prevent purge from running while we are dropping the table */
 	rw_lock_s_lock(&(purge_sys->purge_is_running));
 
@@ -1766,6 +1842,22 @@ row_drop_table_for_mysql(
 		goto funct_exit;
 	}
 
+	if (table->n_foreign_key_checks_running > 0) {
+		
+	        ut_print_timestamp(stderr);
+	        fprintf(stderr,
+		  "  InnoDB: You are trying to drop table %s\n"
+		  "InnoDB: though there are foreign key check running on it.\n"
+		  "InnoDB: Adding the table to the background drop queue.\n",
+		  table->name);
+
+		row_add_table_to_background_drop_list(table);
+
+		err = DB_SUCCESS;
+
+		goto funct_exit;
+	}
+	
 	/* Remove any locks there are on the table or its records */
 	
 	lock_reset_all_on_table(table);
@@ -1793,10 +1885,9 @@ row_drop_table_for_mysql(
 funct_exit:	
 	rw_lock_s_unlock(&(purge_sys->purge_is_running));
 
-	rw_lock_x_unlock(&(dict_foreign_key_check_lock));
-
 	if (!has_dict_mutex) {
 		mutex_exit(&(dict_sys->mutex));
+		rw_lock_x_unlock(&(dict_foreign_key_check_lock));
 	}
 
 	que_graph_free(graph);
@@ -1832,6 +1923,7 @@ row_drop_database_for_mysql(
 	
 	trx_start_if_not_started(trx);
 loop:
+	rw_lock_x_lock(&(dict_foreign_key_check_lock));
 	mutex_enter(&(dict_sys->mutex));
 
 	while (table_name = dict_get_first_table_name_in_db(name)) {
@@ -1873,6 +1965,7 @@ loop:
 	}
 
 	mutex_exit(&(dict_sys->mutex));
+	rw_lock_x_unlock(&(dict_foreign_key_check_lock));
 	
 	trx_commit_for_mysql(trx);
 
@@ -2009,6 +2102,7 @@ row_rename_table_for_mysql(
 	/* Serialize data dictionary operations with dictionary mutex:
 	no deadlocks can occur then in these operations */
 
+	rw_lock_x_lock(&(dict_foreign_key_check_lock));
 	mutex_enter(&(dict_sys->mutex));
 
 	table = dict_table_get_low(old_name);
@@ -2090,6 +2184,7 @@ row_rename_table_for_mysql(
 	}
 funct_exit:	
 	mutex_exit(&(dict_sys->mutex));
+	rw_lock_x_unlock(&(dict_foreign_key_check_lock));
 
 	que_graph_free(graph);
 	

innobase/row/row0sel.c

@@ -2115,8 +2115,14 @@ row_sel_store_mysql_rec(
 				extern_field_heap = NULL;
  			}
 		} else {
-			mysql_rec[templ->mysql_null_byte_offset] |=
+			if (!templ->mysql_null_bit_mask) {
+				fprintf(stderr,
+"InnoDB: Error: trying to return an SQL NULL field in a non-null\n"
+"innoDB: column! Table name %s\n", prebuilt->table->name);
+			} else {
+				mysql_rec[templ->mysql_null_byte_offset] |=
 					(byte) (templ->mysql_null_bit_mask);
+			}
 		}
 	} 
 }

innobase/row/row0upd.c

@@ -73,8 +73,7 @@ steps of query graph execution. */
 
 /*************************************************************************
 Checks if index currently is mentioned as a referenced index in a foreign
-key constraint. This function also loads into the dictionary cache the
-possible referencing table. */
+key constraint. */
 static
 ibool
 row_upd_index_is_referenced(
@@ -85,44 +84,28 @@ row_upd_index_is_referenced(
 				the referencing table has been dropped when
 				we leave this function: this function is only
 				for heuristic use! */
-	dict_index_t*	index)	/* in: index */
+	dict_index_t*	index,	/* in: index */
+	trx_t*		trx)	/* in: transaction */
 {
-	dict_table_t*	table	= index->table;
+	dict_table_t*	table		= index->table;
 	dict_foreign_t*	foreign;
-	ulint		phase	= 1;
 
-try_again:	
 	if (!UT_LIST_GET_FIRST(table->referenced_list)) {
 
 		return(FALSE);
 	}
 
-	if (phase == 2) {
-		mutex_enter(&(dict_sys->mutex));
+	if (!trx->has_dict_foreign_key_check_lock) {
+		rw_lock_s_lock(&dict_foreign_key_check_lock);
 	}
 
-	rw_lock_s_lock(&dict_foreign_key_check_lock);
-
 	foreign = UT_LIST_GET_FIRST(table->referenced_list);
 
 	while (foreign) {
 		if (foreign->referenced_index == index) {
-			if (foreign->foreign_table == NULL) {
-				if (phase == 2) {
-					dict_table_get_low(foreign->
-							foreign_table_name);
-				} else {
-					phase = 2;
-					rw_lock_s_unlock(
-						&dict_foreign_key_check_lock);
-					goto try_again;
-				}
-			}
-
-			rw_lock_s_unlock(&dict_foreign_key_check_lock);
 
-			if (phase == 2) {
-				mutex_exit(&(dict_sys->mutex));
+			if (!trx->has_dict_foreign_key_check_lock) {
+				rw_lock_s_unlock(&dict_foreign_key_check_lock);
 			}
 
 			return(TRUE);
@@ -131,10 +114,8 @@ try_again:
 		foreign = UT_LIST_GET_NEXT(referenced_list, foreign);
 	}
 	
-	rw_lock_s_unlock(&dict_foreign_key_check_lock);
-
-	if (phase == 2) {
-		mutex_exit(&(dict_sys->mutex));
+	if (!trx->has_dict_foreign_key_check_lock) {
+		rw_lock_s_unlock(&dict_foreign_key_check_lock);
 	}
 
 	return(FALSE);
@@ -160,8 +141,17 @@ row_upd_check_references_constraints(
 	dict_foreign_t*	foreign;
 	mem_heap_t*	heap;
 	dtuple_t*	entry;
+	trx_t*		trx;
 	rec_t*		rec;
 	ulint		err;
+	ibool		got_s_lock	= FALSE;
+
+	if (UT_LIST_GET_FIRST(table->referenced_list) == NULL) {
+
+		return(DB_SUCCESS);
+	}
+
+	trx = thr_get_trx(thr);
 
 	rec = btr_pcur_get_rec(pcur);
 
@@ -173,17 +163,61 @@ row_upd_check_references_constraints(
 
 	mtr_start(mtr);	
 	
-	rw_lock_s_lock(&dict_foreign_key_check_lock);	
+	if (!trx->has_dict_foreign_key_check_lock) {
+		got_s_lock = TRUE;
 
+		rw_lock_s_lock(&dict_foreign_key_check_lock);
+
+		trx->has_dict_foreign_key_check_lock = TRUE;
+	}
+		
 	foreign = UT_LIST_GET_FIRST(table->referenced_list);
 
 	while (foreign) {
 		if (foreign->referenced_index == index) {
+			if (foreign->foreign_table == NULL) {
+				dict_table_get(foreign->foreign_table_name,
+									trx);
+			}
 
+			if (foreign->foreign_table) {
+				mutex_enter(&(dict_sys->mutex));
+
+				(foreign->foreign_table
+				->n_foreign_key_checks_running)++;
+
+				mutex_exit(&(dict_sys->mutex));
+			}
+
+			/* NOTE that if the thread ends up waiting for a lock
+			we will release dict_foreign_key_check_lock
+			temporarily! But the counter on the table
+			protects 'foreign' from being dropped while the check
+			is running. */
+			
 			err = row_ins_check_foreign_constraint(FALSE, foreign,
 						table, index, entry, thr);
+
+			if (foreign->foreign_table) {
+				mutex_enter(&(dict_sys->mutex));
+
+				ut_a(foreign->foreign_table
+				->n_foreign_key_checks_running > 0);
+
+				(foreign->foreign_table
+				->n_foreign_key_checks_running)--;
+
+				mutex_exit(&(dict_sys->mutex));
+			}
+
 			if (err != DB_SUCCESS) {
-				rw_lock_s_unlock(&dict_foreign_key_check_lock);	
+				if (got_s_lock) {
+					rw_lock_s_unlock(
+						&dict_foreign_key_check_lock);	
+					trx->has_dict_foreign_key_check_lock
+								= FALSE;
+				}
+
 				mem_heap_free(heap);
 
 				return(err);
@@ -193,7 +227,11 @@ row_upd_check_references_constraints(
 		foreign = UT_LIST_GET_NEXT(referenced_list, foreign);
 	}
 
-	rw_lock_s_unlock(&dict_foreign_key_check_lock);	
+	if (got_s_lock) {
+		rw_lock_s_unlock(&dict_foreign_key_check_lock);	
+		trx->has_dict_foreign_key_check_lock = FALSE;
+	}
+
 	mem_heap_free(heap);
 	
 	return(DB_SUCCESS);
@@ -222,6 +260,9 @@ upd_node_create(
 	node->index = NULL;
 	node->update = NULL;
 	
+	node->cascade_heap = NULL;
+	node->cascade_node = NULL;
+	
 	node->select = NULL;
 	
 	node->heap = mem_heap_create(128);
@@ -1027,7 +1068,7 @@ row_upd_sec_index_entry(
 	
 	index = node->index;
 	
-	check_ref = row_upd_index_is_referenced(index);
+	check_ref = row_upd_index_is_referenced(index, thr_get_trx(thr));
 
 	heap = mem_heap_create(1024);
 
@@ -1391,7 +1432,7 @@ row_upd_clust_step(
 	
 	index = dict_table_get_first_index(node->table);
 
-	check_ref = row_upd_index_is_referenced(index);
+	check_ref = row_upd_index_is_referenced(index, thr_get_trx(thr));
 
 	pcur = node->pcur;
 

innobase/srv/srv0srv.c

@@ -2017,15 +2017,18 @@ srv_suspend_mysql_thread(
 /*=====================*/
 				/* out: TRUE if the lock wait timeout was
 				exceeded */
-	que_thr_t*	thr)	/* in: query thread associated with
-				the MySQL OS thread */
+	que_thr_t*	thr)	/* in: query thread associated with the MySQL
+				OS thread */
 {
 	srv_slot_t*	slot;
 	os_event_t	event;
 	double		wait_time;
-
+	trx_t*		trx;
+	
 	ut_ad(!mutex_own(&kernel_mutex));
 
+	trx = thr_get_trx(thr);
+	
 	os_event_set(srv_lock_timeout_thread_event);
 
 	mutex_enter(&kernel_mutex);
@@ -2061,10 +2064,21 @@ srv_suspend_mysql_thread(
 	
 	srv_conc_force_exit_innodb(thr_get_trx(thr));
 
+	/* Release possible foreign key check latch */
+	if (trx->has_dict_foreign_key_check_lock) {
+
+		rw_lock_s_unlock(&dict_foreign_key_check_lock);
+	}
+
 	/* Wait for the release */
 	
 	os_event_wait(event);
 
+	if (trx->has_dict_foreign_key_check_lock) {
+
+		rw_lock_s_lock(&dict_foreign_key_check_lock);
+	}
+
 	/* Return back inside InnoDB */
 	
 	srv_conc_force_enter_innodb(thr_get_trx(thr));

innobase/sync/sync0sync.c

@@ -220,7 +220,7 @@ mutex_create_func(
 	char*		cfile_name,	/* in: file name where created */
 	ulint		cline)		/* in: file line where created */
 {
-#ifdef _WIN32
+#if defined(_WIN32) && defined(UNIV_CAN_USE_X86_ASSEMBLER) 
 	mutex_reset_lock_word(mutex);
 #else	
 	os_fast_mutex_init(&(mutex->os_fast_mutex));
@@ -273,7 +273,7 @@ mutex_free(
 
 	mutex_exit(&mutex_list_mutex);
 
-#ifndef _WIN32
+#if !defined(_WIN32) || !defined(UNIV_CAN_USE_X86_ASSEMBLER) 
 	os_fast_mutex_free(&(mutex->os_fast_mutex));
 #endif
 	/* If we free the mutex protecting the mutex list (freeing is
@@ -1009,7 +1009,7 @@ sync_thread_add_level(
 	} else if (level == SYNC_ANY_LATCH) {
 		ut_a(sync_thread_levels_g(array, SYNC_ANY_LATCH));
 	} else if (level == SYNC_TRX_SYS_HEADER) {
-		ut_a(sync_thread_levels_contain(array, SYNC_KERNEL));
+		ut_a(sync_thread_levels_g(array, SYNC_TRX_SYS_HEADER));
 	} else if (level == SYNC_DOUBLEWRITE) {
 		ut_a(sync_thread_levels_g(array, SYNC_DOUBLEWRITE));
 	} else if (level == SYNC_BUF_BLOCK) {

innobase/trx/trx0sys.c

@@ -438,7 +438,6 @@ trx_sys_update_mysql_binlog_offset(
 	trx_sysf_t*	sys_header;
 	char		namebuf[TRX_SYS_MYSQL_LOG_NAME_LEN];
 	
-	ut_ad(mutex_own(&kernel_mutex));
 	ut_ad(trx->mysql_log_file_name);
 
 	memset(namebuf, ' ', TRX_SYS_MYSQL_LOG_NAME_LEN - 1);
@@ -524,7 +523,7 @@ trx_sys_print_mysql_binlog_offset(void)
 	}
 
 	fprintf(stderr,
-	"InnoDB: Last MySQL binlog file offset %lu %lu, file name %s\n",
+	"InnoDB: Last MySQL binlog file position %lu %lu, file name %s\n",
 		mach_read_from_4(sys_header + TRX_SYS_MYSQL_LOG_INFO
 					+ TRX_SYS_MYSQL_LOG_OFFSET_HIGH),
 		mach_read_from_4(sys_header + TRX_SYS_MYSQL_LOG_INFO

innobase/trx/trx0trx.c

@@ -71,6 +71,9 @@ trx_create(
 	trx->type = TRX_USER;
 	trx->conc_state = TRX_NOT_STARTED;
 
+	trx->check_foreigns = TRUE;
+	trx->check_unique_secondary = TRUE;
+
 	trx->dict_operation = FALSE;
 
 	trx->mysql_thd = NULL;
@@ -113,6 +116,7 @@ trx_create(
 	trx->lock_heap = mem_heap_create_in_buffer(256);
 	UT_LIST_INIT(trx->trx_locks);
 
+	trx->has_dict_foreign_key_check_lock = FALSE;
 	trx->has_search_latch = FALSE;
 	trx->search_latch_timeout = BTR_SEA_TIMEOUT;
 
@@ -703,8 +707,7 @@ trx_commit_off_kernel(
 
 		/*-------------------------------------*/
 
-		/* Only in some performance tests the variable srv_flush..
-		will be set to FALSE: */
+		/* Most MySQL users run with srv_flush.. set to FALSE: */
 
 		if (srv_flush_log_at_trx_commit) {