Merge branch 'merge/merge-xtradb-5.5' into 5.5

5.5.46-37.6

storage/xtradb/CMakeLists.txt

@@ -1,4 +1,4 @@
-# Copyright (c) 2006, 2011, Oracle and/or its affiliates. All rights reserved.
+# Copyright (c) 2006, 2015, Oracle and/or its affiliates. All rights reserved.
 # 
 # This program is free software; you can redistribute it and/or modify
 # it under the terms of the GNU General Public License as published by
@@ -136,6 +136,18 @@ IF(NOT CMAKE_CROSSCOMPILING)
   }"
   HAVE_IB_GCC_ATOMIC_THREAD_FENCE
   )
+  CHECK_C_SOURCE_RUNS(
+  "#include<stdint.h>
+  int main()
+  {
+    unsigned char	c;
+
+    __atomic_test_and_set(&c, __ATOMIC_ACQUIRE);
+    __atomic_clear(&c, __ATOMIC_RELEASE);
+    return(0);
+  }"
+  HAVE_IB_GCC_ATOMIC_TEST_AND_SET
+  )
 ENDIF()
 
 IF(HAVE_IB_GCC_ATOMIC_BUILTINS)
@@ -145,6 +157,9 @@ ENDIF()
 IF(HAVE_IB_GCC_ATOMIC_BUILTINS_64)
  ADD_DEFINITIONS(-DHAVE_IB_GCC_ATOMIC_BUILTINS_64=1)
 ENDIF()
+IF(HAVE_IB_GCC_ATOMIC_TEST_AND_SET)
+ ADD_DEFINITIONS(-DHAVE_IB_GCC_ATOMIC_TEST_AND_SET=1)
+ENDIF()
 
 IF(HAVE_IB_GCC_SYNC_SYNCHRONISE)
  ADD_DEFINITIONS(-DHAVE_IB_GCC_SYNC_SYNCHRONISE=1)

storage/xtradb/btr/btr0cur.c

@@ -1,6 +1,6 @@
 /*****************************************************************************
 
-Copyright (c) 1994, 2013, Oracle and/or its affiliates. All Rights Reserved.
+Copyright (c) 1994, 2015, Oracle and/or its affiliates. All Rights Reserved.
 Copyright (c) 2008, Google Inc.
 
 Portions of this file contain modifications contributed and copyrighted by
@@ -318,7 +318,12 @@ btr_cur_latch_leaves(
 
 				ut_a(page_is_comp(get_block->frame)
 					== page_is_comp(page));
-				ut_a(btr_page_get_next(get_block->frame, mtr)
+
+				/* For fake_change mode we avoid a detailed validation
+				as it operate in tweaked format where-in validation
+				may fail. */
+				ut_a(sibling_mode == RW_NO_LATCH
+				     || btr_page_get_next(get_block->frame, mtr)
 					== page_get_page_no(page));
 			}
 #endif /* UNIV_BTR_DEBUG */
@@ -2228,6 +2233,7 @@ btr_cur_optimistic_update(
 	ulint		max_size;
 	ulint		new_rec_size;
 	ulint		old_rec_size;
+	ulint		max_ins_size = 0;
 	dtuple_t*	new_entry;
 	roll_ptr_t	roll_ptr;
 	trx_t*		trx;
@@ -2339,6 +2345,11 @@ btr_cur_optimistic_update(
 		: (old_rec_size
 		   + page_get_max_insert_size_after_reorganize(page, 1));
 
+	if (!page_zip) {
+		max_ins_size = page_get_max_insert_size_after_reorganize(
+					page, 1);
+	}
+
 	if (!(((max_size >= BTR_CUR_PAGE_REORGANIZE_LIMIT)
 	       && (max_size >= new_rec_size))
 	      || (page_get_n_recs(page) <= 1))) {
@@ -2395,10 +2406,14 @@ btr_cur_optimistic_update(
 	rec = btr_cur_insert_if_possible(cursor, new_entry, 0/*n_ext*/, mtr);
 	ut_a(rec); /* <- We calculated above the insert would fit */
 
-	if (page_zip && !dict_index_is_clust(index)
+	if (!dict_index_is_clust(index)
 	    && page_is_leaf(page)) {
 		/* Update the free bits in the insert buffer. */
-		ibuf_update_free_bits_zip(block, mtr);
+		if (page_zip) {
+			ibuf_update_free_bits_zip(block, mtr);
+		} else {
+			ibuf_update_free_bits_low(block, max_ins_size, mtr);
+		}
 	}
 
 	/* Restore the old explicit lock state on the record */
@@ -2507,6 +2522,7 @@ btr_cur_pessimistic_update(
 	ulint		n_reserved;
 	ulint		n_ext;
 	ulint*		offsets		= NULL;
+	ulint		max_ins_size	= 0;
 
 	*big_rec = NULL;
 
@@ -2659,6 +2675,11 @@ btr_cur_pessimistic_update(
 		/* skip CHANGE, LOG */
 		err = DB_SUCCESS;
 		goto return_after_reservations;
+  }
+
+  if (!page_zip) {
+		max_ins_size = page_get_max_insert_size_after_reorganize(
+					page, 1);
 	}
 
 	/* Store state of explicit locks on rec on the page infimum record,
@@ -2706,10 +2727,15 @@ btr_cur_pessimistic_update(
 			big_rec_vec != NULL && (flags & BTR_KEEP_POS_FLAG),
 			mtr);
 
-		if (page_zip && !dict_index_is_clust(index)
+		if (!dict_index_is_clust(index)
 		    && page_is_leaf(page)) {
 			/* Update the free bits in the insert buffer. */
-			ibuf_update_free_bits_zip(block, mtr);
+			if (page_zip) {
+				ibuf_update_free_bits_zip(block, mtr);
+			} else {
+				ibuf_update_free_bits_low(block, max_ins_size,
+							  mtr);
+			}
 		}
 
 		err = DB_SUCCESS;
@@ -4016,7 +4042,14 @@ btr_estimate_number_of_different_key_vals(
 
 		page = btr_cur_get_page(&cursor);
 
-		SRV_CORRUPT_TABLE_CHECK(page, goto exit_loop;);
+		DBUG_EXECUTE_IF("ib_corrupt_page_while_stats_calc",
+				page = NULL;);
+
+		SRV_CORRUPT_TABLE_CHECK(page,
+		{
+			mtr_commit(&mtr);
+			goto exit_loop;
+		});
 
 		rec = page_rec_get_next(page_get_infimum_rec(page));
 

storage/xtradb/dict/dict0crea.c

@@ -1255,7 +1255,7 @@ dict_create_index_step(
 			>= DICT_TF_FORMAT_ZIP);
 
 		node->index = dict_index_get_if_in_cache_low(index_id);
-		ut_a(err == DB_SUCCESS ? node->index != NULL : node->index == NULL);
+		ut_a((node->index == 0) == (err != DB_SUCCESS));
 
 		if (err != DB_SUCCESS) {
 

storage/xtradb/handler/ha_innodb.cc

@@ -1034,6 +1034,19 @@ thd_supports_xa(
 	return(THDVAR((THD*) thd, support_xa));
 }
 
+/******************************************************************//**
+Check the status of fake changes mode (innodb_fake_changes)
+@return	true	if fake change mode is enabled. */
+extern "C" UNIV_INTERN
+ibool
+thd_fake_changes(
+/*=============*/
+	void*	thd)	/*!< in: thread handle, or NULL to query
+			the global innodb_supports_xa */
+{
+	return(THDVAR((THD*) thd, fake_changes));
+}
+
 /******************************************************************//**
 Returns the lock wait timeout for the current connection.
 @return	the lock wait timeout, in seconds */
@@ -1800,7 +1813,15 @@ innobase_trx_init(
 	trx->check_unique_secondary = !thd_test_options(
 		thd, OPTION_RELAXED_UNIQUE_CHECKS);
 
-	trx->fake_changes = THDVAR(thd, fake_changes);
+	/* Transaction on start caches the fake_changes state and uses it for
+	complete transaction lifetime.
+	There are some APIs that doesn't need an active transaction object
+	but transaction object are just use as a cache object/data carrier.
+	Before using transaction object for such APIs refresh the state of
+	fake_changes. */
+	if (trx->state == TRX_NOT_STARTED) {
+		trx->fake_changes = thd_fake_changes(thd);
+	}
 
 #ifdef EXTENDED_SLOWLOG
 	if (thd_log_slow_verbosity(thd) & (1ULL << SLOG_V_INNODB)) {
@@ -3620,12 +3641,26 @@ innobase_commit(
 	/* No-op in XtraDB */
 	trx_search_latch_release_if_reserved(trx);
 
+	/* If fake-changes mode = ON then allow
+	SELECT (they are read-only) and
+	CREATE ... SELECT * from table (Well this doesn't open up DDL for InnoDB
+	as ha_innobase::create will return appropriate error if fake-change = ON
+	but if create is trying to use other SE and SELECT is executing on
+	InnoDB table then we allow SELECT to proceed.
+	Ideally, statement like this should be marked CREATE_SELECT like
+	INSERT_SELECT but unfortunately it doesn't). */
 	if (UNIV_UNLIKELY(trx->fake_changes
+			  && (thd_sql_command(thd) != SQLCOM_SELECT
+			      && thd_sql_command(thd) != SQLCOM_CREATE_TABLE)
 			  && (all || (!thd_test_options(thd,
 				OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN))))) {
 
-		innobase_rollback(hton, thd, all); /* rollback implicitly */
-		thd->stmt_da->reset_diagnostics_area(); /* because debug assertion code complains, if something left */
+		/* rollback implicitly */
+		innobase_rollback(hton, thd, all);
+
+		/* because debug assertion code complains, if something left */
+		thd->stmt_da->reset_diagnostics_area();
+
 		DBUG_RETURN(HA_ERR_WRONG_COMMAND);
 	}
 	/* Transaction is deregistered only in a commit or a rollback. If
@@ -12714,6 +12749,46 @@ innodb_change_buffering_update(
 		 *static_cast<const char*const*>(save);
 }
 
+#ifdef UNIV_DEBUG
+/*************************************************************//**
+Check if it is a valid value of innodb_track_changed_pages.
+Changed pages tracking is not working correctly without initialization
+procedure on server startup. The function allows to temporary
+disable tracking, but only if the feature was enabled on startup.
+This function is registered as a callback with MySQL.
+@return	0 for valid innodb_track_changed_pages */
+static
+int
+innodb_track_changed_pages_validate(
+	THD*				thd,	/*!< in: thread handle */
+	struct st_mysql_sys_var*	var,	/*!< in: pointer to system
+						variable */
+	void*				save,	/*!< out: immediate result
+						for update function */
+	struct st_mysql_value*		value)	/*!< in: incoming bool */
+{
+	static bool     enabled_on_startup = false;
+	long long	intbuf = 0;
+
+	if (value->val_int(value, &intbuf)) {
+		/* The value is NULL. That is invalid. */
+		return 1;
+	}
+
+	if (srv_track_changed_pages || enabled_on_startup) {
+		enabled_on_startup = true;
+		*reinterpret_cast<ulong*>(save)
+			= static_cast<ulong>(intbuf);
+		return 0;
+	}
+
+	if (intbuf == srv_track_changed_pages)
+		return 0;
+
+	return 1;
+}
+#endif
+
 #ifndef DBUG_OFF
 static char* srv_buffer_pool_evict;
 
@@ -13424,7 +13499,12 @@ provide a testcase sync facility */
 static MYSQL_SYSVAR_BOOL(track_changed_pages, srv_track_changed_pages,
   track_changed_pages_flags,
   "Track the redo log for changed pages and output a changed page bitmap",
-  NULL, NULL, FALSE);
+#ifdef UNIV_DEBUG
+  innodb_track_changed_pages_validate,
+#else
+  NULL,
+#endif
+  NULL, FALSE);
 
 static MYSQL_SYSVAR_ULONGLONG(max_bitmap_file_size, srv_max_bitmap_file_size,
     PLUGIN_VAR_RQCMDARG,

storage/xtradb/include/ha_prototypes.h

@@ -275,6 +275,16 @@ thd_supports_xa(
 	void*	thd);	/*!< in: thread handle (THD*), or NULL to query
 			the global innodb_supports_xa */
 
+/******************************************************************//**
+Check the status of fake changes mode (innodb_fake_changes)
+@return	true	if fake change mode is enabled. */
+
+ibool
+thd_fake_changes(
+/*=============*/
+	void*	thd);	/*!< in: thread handle, or NULL to query
+			the global innodb_supports_xa */
+
 /******************************************************************//**
 Returns the lock wait timeout for the current connection.
 @return	the lock wait timeout, in seconds */

storage/xtradb/include/os0file.h

@@ -119,6 +119,10 @@ log. */
 #define OS_FILE_READ_ONLY		333
 #define	OS_FILE_READ_WRITE		444
 #define	OS_FILE_READ_ALLOW_DELETE	555	/* for ibbackup */
+#define OS_FILE_READ_WRITE_CACHED	666	/* OS_FILE_READ_WRITE but never
+					O_DIRECT. Only for
+					os_file_create_simple_no_error_handling
+				        currently. */
 
 /* Options for file_create */
 #define	OS_FILE_AIO			61
@@ -505,9 +509,10 @@ os_file_create_simple_no_error_handling_func(
 				OS_FILE_CREATE if a new file is created
 				(if exists, error) */
 	ulint		access_type,/*!< in: OS_FILE_READ_ONLY,
-				OS_FILE_READ_WRITE, or
-				OS_FILE_READ_ALLOW_DELETE; the last option is
-				used by a backup program reading the file */
+				OS_FILE_READ_WRITE, OS_FILE_READ_ALLOW_DELETE
+				(used by a backup program reading the file), or
+				OS_FILE_READ_WRITE_CACHED (disable O_DIRECT if
+				it would be enabled otherwise). */
 	ibool*		success);/*!< out: TRUE if succeed, FALSE if error */
 /****************************************************************//**
 Tries to disable OS caching on an opened file descriptor. */