Merge mysql-5.5-innodb -> mysql-5.5

10 files changed, 131 insertions, 56 deletions

diff --git a/storage/innobase/dict/dict0crea.c b/storage/innobase/dict/dict0crea.c
index d0344e72703..9a528d679a7 100644
--- a/storage/innobase/dict/dict0crea.c
+++ b/storage/innobase/dict/dict0crea.c
@@ -659,9 +659,9 @@ dict_create_index_tree_step(
 	/* printf("Created a new index tree in space %lu root page %lu\n",
 	index->space, index->page_no); */
 
-	page_rec_write_index_page_no(btr_pcur_get_rec(&pcur),
-				     DICT_SYS_INDEXES_PAGE_NO_FIELD,
-				     node->page_no, &mtr);
+	page_rec_write_field(btr_pcur_get_rec(&pcur),
+			     DICT_SYS_INDEXES_PAGE_NO_FIELD,
+			     node->page_no, &mtr);
 	btr_pcur_close(&pcur);
 	mtr_commit(&mtr);
 
@@ -731,9 +731,8 @@ dict_drop_index_tree(
 	root_page_no); */
 	btr_free_root(space, zip_size, root_page_no, mtr);
 
-	page_rec_write_index_page_no(rec,
-				     DICT_SYS_INDEXES_PAGE_NO_FIELD,
-				     FIL_NULL, mtr);
+	page_rec_write_field(rec, DICT_SYS_INDEXES_PAGE_NO_FIELD,
+			     FIL_NULL, mtr);
 }
 
 /*******************************************************************//**
@@ -836,8 +835,8 @@ create:
 	in SYS_INDEXES, so that the database will not get into an
 	inconsistent state in case it crashes between the mtr_commit()
 	below and the following mtr_commit() call. */
-	page_rec_write_index_page_no(rec, DICT_SYS_INDEXES_PAGE_NO_FIELD,
-				     FIL_NULL, mtr);
+	page_rec_write_field(rec, DICT_SYS_INDEXES_PAGE_NO_FIELD,
+			     FIL_NULL, mtr);
 
 	/* We will need to commit the mini-transaction in order to avoid
 	deadlocks in the btr_create() call, because otherwise we would
diff --git a/storage/innobase/dict/dict0load.c b/storage/innobase/dict/dict0load.c
index aad3145f7a4..7d531c26157 100644
--- a/storage/innobase/dict/dict0load.c
+++ b/storage/innobase/dict/dict0load.c
@@ -429,7 +429,7 @@ dict_process_sys_fields_rec(
 	mach_write_to_8(last_index_id, last_id);
 
 	err_msg = dict_load_field_low(buf, NULL, sys_field,
-				      pos, last_index_id, heap, rec);
+				      pos, last_index_id, heap, rec, NULL, 0);
 
 	*index_id = mach_read_from_8(buf);
 
@@ -994,6 +994,9 @@ dict_load_columns(
 /** Error message for a delete-marked record in dict_load_field_low() */
 static const char* dict_load_field_del = "delete-marked record in SYS_FIELDS";
 
+static const char* dict_load_field_too_big = "column prefix exceeds maximum"
+					     " limit";
+
 /********************************************************************//**
 Loads an index field definition from a SYS_FIELDS record to
 dict_index_t.
@@ -1015,7 +1018,12 @@ dict_load_field_low(
 	byte*		last_index_id,	/*!< in: last index id */
 	mem_heap_t*	heap,		/*!< in/out: memory heap
 					for temporary storage */
-	const rec_t*	rec)		/*!< in: SYS_FIELDS record */
+	const rec_t*	rec,		/*!< in: SYS_FIELDS record */
+	char*		addition_err_str,/*!< out: additional error message
+					that requires information to be
+					filled, or NULL */
+	ulint		err_str_len)	/*!< in: length of addition_err_str
+					in bytes */
 {
 	const byte*	field;
 	ulint		len;
@@ -1095,6 +1103,19 @@ err_len:
 		goto err_len;
 	}
 
+	if (prefix_len >= DICT_MAX_INDEX_COL_LEN) {
+		if (addition_err_str) {
+			ut_snprintf(addition_err_str, err_str_len,
+				    "index field '%s' has a prefix length"
+				    " of %lu bytes",
+				    mem_heap_strdupl(
+						heap, (const char*) field, len),
+				    (ulong) prefix_len);
+		}
+
+		return(dict_load_field_too_big);
+	}
+
 	if (index) {
 		dict_mem_index_add_field(
 			index, mem_heap_strdupl(heap, (const char*) field, len),
@@ -1154,14 +1175,16 @@ dict_load_fields(
 	btr_pcur_open_on_user_rec(sys_index, tuple, PAGE_CUR_GE,
 				  BTR_SEARCH_LEAF, &pcur, &mtr);
 	for (i = 0; i < index->n_fields; i++) {
-		const char* err_msg;
+		const char*	err_msg;
+		char		addition_err_str[1024];
 
 		rec = btr_pcur_get_rec(&pcur);
 
 		ut_a(btr_pcur_is_on_user_rec(&pcur));
 
 		err_msg = dict_load_field_low(buf, index, NULL, NULL, NULL,
-					      heap, rec);
+					      heap, rec, addition_err_str,
+					      sizeof(addition_err_str));
 
 		if (err_msg == dict_load_field_del) {
 			/* There could be delete marked records in
@@ -1170,7 +1193,24 @@ dict_load_fields(
 
 			goto next_rec;
 		} else if (err_msg) {
-			fprintf(stderr, "InnoDB: %s\n", err_msg);
+			if (err_msg == dict_load_field_too_big) {
+				fprintf(stderr, "InnoDB: Error: load index"
+					" '%s' failed.\n"
+					"InnoDB: %s,\n"
+					"InnoDB: which exceeds the"
+					" maximum limit of %lu bytes.\n"
+					"InnoDB: Please use server that"
+					" supports long index prefix\n"
+					"InnoDB: or turn on"
+					" innodb_force_recovery to load"
+					" the table\n",
+					index->name, addition_err_str,
+					(ulong) (DICT_MAX_INDEX_COL_LEN - 1));
+
+			} else {
+				fprintf(stderr, "InnoDB: %s\n", err_msg);
+			}
+
 			error = DB_CORRUPTION;
 			goto func_exit;
 		}
@@ -1446,7 +1486,26 @@ corrupted:
 			of the database server */
 			dict_mem_index_free(index);
 		} else {
-			dict_load_fields(index, heap);
+			error = dict_load_fields(index, heap);
+
+			if (error != DB_SUCCESS) {
+
+				fprintf(stderr, "InnoDB: Error: load index '%s'"
+					" for table '%s' failed\n",
+					index->name, table->name);
+
+				/* If the force recovery flag is set, and
+				if the failed index is not the primary index, we
+				will continue and open other indexes */
+				if (srv_force_recovery
+				    && !dict_index_is_clust(index)) {
+					error = DB_SUCCESS;
+					goto next_rec;
+				} else {
+					goto func_exit;
+				}
+			}
+
 			error = dict_index_add_to_cache(table, index,
 							index->page, FALSE);
 			/* The data dictionary tables should never contain
@@ -1771,9 +1830,18 @@ err_exit:
 		} else {
 			table->fk_max_recusive_level = 0;
 		}
-	} else if (!srv_force_recovery) {
-		dict_table_remove_from_cache(table);
-		table = NULL;
+	} else {
+		dict_index_t*	index;
+
+		/* Make sure that at least the clustered index was loaded.
+		Otherwise refuse to load the table */
+		index = dict_table_get_first_index(table);
+
+		if (!srv_force_recovery || !index
+		     || !dict_index_is_clust(index)) {
+			dict_table_remove_from_cache(table);
+			table = NULL;
+		}
 	}
 #if 0
 	if (err != DB_SUCCESS && table != NULL) {
diff --git a/storage/innobase/fil/fil0fil.c b/storage/innobase/fil/fil0fil.c
index 0d9846fdbf8..196f4bd3f42 100644
--- a/storage/innobase/fil/fil0fil.c
+++ b/storage/innobase/fil/fil0fil.c
@@ -856,7 +856,8 @@ fil_node_close_file(
 	ut_a(node->open);
 	ut_a(node->n_pending == 0);
 	ut_a(node->n_pending_flushes == 0);
-	ut_a(node->modification_counter == node->flush_counter);
+	ut_a(node->modification_counter == node->flush_counter
+	     || srv_fast_shutdown == 2);
 
 	ret = os_file_close(node->handle);
 	ut_a(ret);
diff --git a/storage/innobase/include/dict0load.h b/storage/innobase/include/dict0load.h
index 51d07f43446..16177ade713 100644
--- a/storage/innobase/include/dict0load.h
+++ b/storage/innobase/include/dict0load.h
@@ -155,7 +155,12 @@ dict_load_field_low(
 	byte*		last_index_id,	/*!< in: last index id */
 	mem_heap_t*	heap,		/*!< in/out: memory heap
 					for temporary storage */
-	const rec_t*	rec);		/*!< in: SYS_FIELDS record */
+	const rec_t*	rec,		/*!< in: SYS_FIELDS record */
+	char*		addition_err_str,/*!< out: additional error message
+					that requires information to be
+					filled, or NULL */
+	ulint		err_str_len);	/*!< in: length of addition_err_str
+					in bytes */
 /********************************************************************//**
 Loads a table definition and also all its index definitions, and also
 the cluster definition if the table is a member in a cluster. Also loads
diff --git a/storage/innobase/include/page0page.h b/storage/innobase/include/page0page.h
index 0a15dfbf2a1..fc082e9b189 100644
--- a/storage/innobase/include/page0page.h
+++ b/storage/innobase/include/page0page.h
@@ -619,17 +619,16 @@ page_rec_find_owner_rec(
 /*====================*/
 	rec_t*	rec);	/*!< in: the physical record */
 /***********************************************************************//**
-This is a low-level operation which is used in a database index creation
-to update the page number of a created B-tree to a data dictionary
-record. */
-UNIV_INTERN
+Write a 32-bit field in a data dictionary record. */
+UNIV_INLINE
 void
-page_rec_write_index_page_no(
-/*=========================*/
-	rec_t*	rec,	/*!< in: record to update */
+page_rec_write_field(
+/*=================*/
+	rec_t*	rec,	/*!< in/out: record to update */
 	ulint	i,	/*!< in: index of the field to update */
 	ulint	page_no,/*!< in: value to write */
-	mtr_t*	mtr);	/*!< in: mtr */
+	mtr_t*	mtr)	/*!< in/out: mini-transaction */
+	__attribute__((nonnull));
 /************************************************************//**
 Returns the maximum combined size of records which can be inserted on top
 of record heap.
diff --git a/storage/innobase/include/page0page.ic b/storage/innobase/include/page0page.ic
index e9624c2360f..75e562a158d 100644
--- a/storage/innobase/include/page0page.ic
+++ b/storage/innobase/include/page0page.ic
@@ -959,6 +959,27 @@ page_get_free_space_of_empty(
 		       - 2 * PAGE_DIR_SLOT_SIZE));
 }
 
+/***********************************************************************//**
+Write a 32-bit field in a data dictionary record. */
+UNIV_INLINE
+void
+page_rec_write_field(
+/*=================*/
+	rec_t*	rec,	/*!< in/out: record to update */
+	ulint	i,	/*!< in: index of the field to update */
+	ulint	val,	/*!< in: value to write */
+	mtr_t*	mtr)	/*!< in/out: mini-transaction */
+{
+	byte*	data;
+	ulint	len;
+
+	data = rec_get_nth_field_old(rec, i, &len);
+
+	ut_ad(len == 4);
+
+	mlog_write_ulint(data, val, MLOG_4BYTES, mtr);
+}
+
 /************************************************************//**
 Each user record on a page, and also the deleted user records in the heap
 takes its size plus the fraction of the dir cell size /
diff --git a/storage/innobase/os/os0file.c b/storage/innobase/os/os0file.c
index 50607e07076..bd48d0e4b6e 100644
--- a/storage/innobase/os/os0file.c
+++ b/storage/innobase/os/os0file.c
@@ -4051,16 +4051,23 @@ os_aio_func(
 		Windows async i/o, Windows does not allow us to use
 		ordinary synchronous os_file_read etc. on the same file,
 		therefore we have built a special mechanism for synchronous
-		wait in the Windows case. */
+		wait in the Windows case.
+		Also note that the Performance Schema instrumentation has
+		been performed by current os_aio_func()'s wrapper function
+		pfs_os_aio_func(). So we would no longer need to call
+		Performance Schema instrumented os_file_read() and
+		os_file_write(). Instead, we should use os_file_read_func()
+		and os_file_write_func() */
 
 		if (type == OS_FILE_READ) {
-			return(os_file_read(file, buf, offset,
+			return(os_file_read_func(file, buf, offset,
 					    offset_high, n));
 		}
 
 		ut_a(type == OS_FILE_WRITE);
 
-		return(os_file_write(name, file, buf, offset, offset_high, n));
+		return(os_file_write_func(name, file, buf, offset,
+					  offset_high, n));
 	}
 
 try_again:
diff --git a/storage/innobase/page/page0page.c b/storage/innobase/page/page0page.c
index 964d0e2abef..6064d028ae1 100644
--- a/storage/innobase/page/page0page.c
+++ b/storage/innobase/page/page0page.c
@@ -1253,28 +1253,6 @@ page_move_rec_list_start(
 
 	return(TRUE);
 }
-
-/***********************************************************************//**
-This is a low-level operation which is used in a database index creation
-to update the page number of a created B-tree to a data dictionary record. */
-UNIV_INTERN
-void
-page_rec_write_index_page_no(
-/*=========================*/
-	rec_t*	rec,	/*!< in: record to update */
-	ulint	i,	/*!< in: index of the field to update */
-	ulint	page_no,/*!< in: value to write */
-	mtr_t*	mtr)	/*!< in: mtr */
-{
-	byte*	data;
-	ulint	len;
-
-	data = rec_get_nth_field_old(rec, i, &len);
-
-	ut_ad(len == 4);
-
-	mlog_write_ulint(data, page_no, MLOG_4BYTES, mtr);
-}
 #endif /* !UNIV_HOTBACKUP */
 
 /**************************************************************//**
diff --git a/storage/innobase/row/row0mysql.c b/storage/innobase/row/row0mysql.c
index f7e5c5fdceb..aef4595f5fe 100644
--- a/storage/innobase/row/row0mysql.c
+++ b/storage/innobase/row/row0mysql.c
@@ -2941,7 +2941,7 @@ row_truncate_table_for_mysql(
 		rec = btr_pcur_get_rec(&pcur);
 
 		if (root_page_no != FIL_NULL) {
-			page_rec_write_index_page_no(
+			page_rec_write_field(
 				rec, DICT_SYS_INDEXES_PAGE_NO_FIELD,
 				root_page_no, &mtr);
 			/* We will need to commit and restart the
diff --git a/storage/innobase/trx/trx0undo.c b/storage/innobase/trx/trx0undo.c
index 070d6332a4f..4cb4b7b79c5 100644
--- a/storage/innobase/trx/trx0undo.c
+++ b/storage/innobase/trx/trx0undo.c
@@ -1985,8 +1985,6 @@ trx_undo_free_prepared(
 /*===================*/
 	trx_t*	trx)	/*!< in/out: PREPARED transaction */
 {
-	mutex_enter(&trx->rseg->mutex);
-
 	ut_ad(srv_shutdown_state == SRV_SHUTDOWN_EXIT_THREADS);
 
 	if (trx->update_undo) {
@@ -2001,6 +1999,5 @@ trx_undo_free_prepared(
 			       trx->insert_undo);
 		trx_undo_mem_free(trx->insert_undo);
 	}
-	mutex_exit(&trx->rseg->mutex);
 }
 #endif /* !UNIV_HOTBACKUP */