MDEV-9678: Data Directory bug

Problem was that link file (.isl) is also opened using O_DIRECT mode and if this fails the whole create table fails on internal error. Fixed by not using O_DIRECT on link files as they are used only on create table and startup and do not contain real data. O_DIRECT failures are successfully ignored for data files if O_DIRECT is not supported by file system on used data directory.
MariaDB · Mar 22, 2016 · f71c45c · f71c45c
1 parent 37f915c
commit f71c45c
Show file tree

Hide file tree

Showing 6 changed files with 69 additions and 35 deletions.
diff --git a/storage/innobase/fil/fil0fil.cc b/storage/innobase/fil/fil0fil.cc
@@ -1,7 +1,7 @@
 /*****************************************************************************
 
 Copyright (c) 1995, 2015, Oracle and/or its affiliates.
-Copyright (c) 2013, 2015, MariaDB Corporation.
+Copyright (c) 2013, 2016, MariaDB Corporation.
 
 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 the Free Software
@@ -3207,10 +3207,12 @@ fil_create_link_file(
 	}
 
 	link_filepath = fil_make_isl_name(tablename);
-
+	/* Note that OS_FILE_READ_WRITE_CACHED used here to avoid
+	unnecessary errors on O_DIRECT, link files are not really
+	a data files. */
 	file = os_file_create_simple_no_error_handling(
 		innodb_file_data_key, link_filepath,
-		OS_FILE_CREATE, OS_FILE_READ_WRITE, &success, 0);
+		OS_FILE_CREATE, OS_FILE_READ_WRITE_CACHED, &success, 0);
 
 	if (!success) {
 		/* The following call will print an error message */

diff --git a/storage/innobase/include/os0file.h b/storage/innobase/include/os0file.h
@@ -128,7 +128,10 @@ enum os_file_create_t {
 #define OS_FILE_READ_ONLY		333
 #define	OS_FILE_READ_WRITE		444
 #define	OS_FILE_READ_ALLOW_DELETE	555	/* for mysqlbackup */
-
+#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
 #define	OS_FILE_NORMAL			62
@@ -542,7 +545,7 @@ UNIV_INTERN
 void
 os_file_set_nocache(
 /*================*/
-	int		fd,		/*!< in: file descriptor to alter */
+	os_file_t	fd,		/*!< in: file descriptor to alter */
 	const char*	file_name,	/*!< in: file name, used in the
 					diagnostic message */
 	const char*	operation_name);/*!< in: "open" or "create"; used in the

diff --git a/storage/innobase/os/os0file.cc b/storage/innobase/os/os0file.cc
@@ -2,7 +2,7 @@
 
 Copyright (c) 1995, 2015, Oracle and/or its affiliates. All Rights Reserved.
 Copyright (c) 2009, Percona Inc.
-Copyright (c) 2013, 2015, MariaDB Corporation.
+Copyright (c) 2013, 2016, MariaDB Corporation.
 
 Portions of this file contain modifications contributed and copyrighted
 by Percona Inc.. Those modifications are
@@ -1350,7 +1350,8 @@ os_file_create_simple_func(
 
 		access = GENERIC_READ;
 
-	} else if (access_type == OS_FILE_READ_WRITE) {
+	} else if (access_type == OS_FILE_READ_WRITE
+		   || access_type == OS_FILE_READ_WRITE_CACHED) {
 		access = GENERIC_READ | GENERIC_WRITE;
 	} else {
 		ib_logf(IB_LOG_LEVEL_ERROR,
@@ -1454,7 +1455,8 @@ os_file_create_simple_func(
 #ifdef USE_FILE_LOCK
 	if (!srv_read_only_mode
 	    && *success
-	    && access_type == OS_FILE_READ_WRITE
+	    && (access_type == OS_FILE_READ_WRITE
+		|| access_type == OS_FILE_READ_WRITE_CACHED)
 	    && os_file_lock(file, name)) {
 
 		*success = FALSE;
@@ -1468,6 +1470,31 @@ os_file_create_simple_func(
 	return(file);
 }
 
+/** Disable OS I/O caching on the file if the file type and server
+configuration requires it.
+@param file handle to the file
+@param name name of the file, for diagnostics
+@param mode_str operation on the file, for diagnostics
+@param type OS_LOG_FILE or OS_DATA_FILE
+@param access_type if OS_FILE_READ_WRITE_CACHED, then caching will be disabled
+unconditionally, ignored otherwise */
+static
+void
+os_file_set_nocache_if_needed(os_file_t file, const char* name,
+                             const char *mode_str, ulint type,
+                             ulint access_type)
+{
+       if (srv_read_only_mode || access_type == OS_FILE_READ_WRITE_CACHED) {
+               return;
+       }
+
+       if (type == OS_DATA_FILE
+           && (srv_unix_file_flush_method == SRV_UNIX_O_DIRECT
+               || (srv_unix_file_flush_method == SRV_UNIX_O_DIRECT_NO_FSYNC))) {
+               os_file_set_nocache(file, name, mode_str);
+       }
+}
+
 /****************************************************************//**
 NOTE! Use the corresponding macro
 os_file_create_simple_no_error_handling(), not directly this function!
@@ -1523,7 +1550,8 @@ os_file_create_simple_no_error_handling_func(
 		access = GENERIC_READ;
 	} else if (srv_read_only_mode) {
 		access = GENERIC_READ;
-	} else if (access_type == OS_FILE_READ_WRITE) {
+	} else if (access_type == OS_FILE_READ_WRITE
+		   || access_type == OS_FILE_READ_WRITE_CACHED) {
 		access = GENERIC_READ | GENERIC_WRITE;
 	} else if (access_type == OS_FILE_READ_ALLOW_DELETE) {
 
@@ -1595,7 +1623,8 @@ os_file_create_simple_no_error_handling_func(
 		} else {
 
 			ut_a(access_type == OS_FILE_READ_WRITE
-			     || access_type == OS_FILE_READ_ALLOW_DELETE);
+				|| access_type == OS_FILE_READ_ALLOW_DELETE
+				|| access_type == OS_FILE_READ_WRITE_CACHED);
 
 			create_flag = O_RDWR;
 		}
@@ -1627,18 +1656,16 @@ os_file_create_simple_no_error_handling_func(
 	/* This function is always called for data files, we should disable
 	OS caching (O_DIRECT) here as we do in os_file_create_func(), so
 	we open the same file in the same mode, see man page of open(2). */
-	if (!srv_read_only_mode
-	    && *success
-	    && (srv_unix_file_flush_method == SRV_UNIX_O_DIRECT
-		|| srv_unix_file_flush_method == SRV_UNIX_O_DIRECT_NO_FSYNC)) {
-
-		os_file_set_nocache(file, name, mode_str);
+	if (*success) {
+		os_file_set_nocache_if_needed(file, name, mode_str,
+			OS_DATA_FILE, access_type);
 	}
 
 #ifdef USE_FILE_LOCK
 	if (!srv_read_only_mode
 	    && *success
-	    && access_type == OS_FILE_READ_WRITE
+	    && (access_type == OS_FILE_READ_WRITE
+		|| access_type == OS_FILE_READ_WRITE_CACHED)
 	    && os_file_lock(file, name)) {
 
 		*success = FALSE;
@@ -1677,7 +1704,7 @@ UNIV_INTERN
 void
 os_file_set_nocache(
 /*================*/
-	int		fd		/*!< in: file descriptor to alter */
+	os_file_t	fd		/*!< in: file descriptor to alter */
 					__attribute__((unused)),
 	const char*	file_name	/*!< in: used in the diagnostic
 					message */
@@ -2007,13 +2034,8 @@ os_file_create_func(
 
 	/* We disable OS caching (O_DIRECT) only on data files */
 
-	if (!srv_read_only_mode
-	    && *success
-	    && type != OS_LOG_FILE
-	    && (srv_unix_file_flush_method == SRV_UNIX_O_DIRECT
-		|| srv_unix_file_flush_method == SRV_UNIX_O_DIRECT_NO_FSYNC)) {
-
-		os_file_set_nocache(file, name, mode_str);
+	if (*success) {
+		os_file_set_nocache_if_needed(file, name, mode_str, type, 0);
 	}
 
 #ifdef USE_FILE_LOCK

diff --git a/storage/innobase/row/row0merge.cc b/storage/innobase/row/row0merge.cc
@@ -3454,7 +3454,7 @@ row_merge_file_create(
 
 	if (merge_file->fd >= 0) {
 		if (srv_disable_sort_file_cache) {
-			os_file_set_nocache(merge_file->fd,
+			os_file_set_nocache((os_file_t)merge_file->fd,
 				"row0merge.cc", "sort");
 		}
 	}

diff --git a/storage/xtradb/fil/fil0fil.cc b/storage/xtradb/fil/fil0fil.cc
@@ -3240,9 +3240,12 @@ fil_create_link_file(
 
 	link_filepath = fil_make_isl_name(tablename);
 
+	/* Note that OS_FILE_READ_WRITE_CACHED used here to avoid
+	unnecessary errors on O_DIRECT, link files are not really
+	a data files. */
 	file = os_file_create_simple_no_error_handling(
 		innodb_file_data_key, link_filepath,
-		OS_FILE_CREATE, OS_FILE_READ_WRITE, &success, 0);
+		OS_FILE_CREATE, OS_FILE_READ_WRITE_CACHED, &success, 0);
 
 	if (!success) {
 		/* The following call will print an error message */

diff --git a/storage/xtradb/os/os0file.cc b/storage/xtradb/os/os0file.cc
@@ -2,7 +2,7 @@
 
 Copyright (c) 1995, 2015, Oracle and/or its affiliates. All Rights Reserved.
 Copyright (c) 2009, Percona Inc.
-Copyright (c) 2013, 2015, MariaDB Corporation.
+Copyright (c) 2013, 2016, MariaDB Corporation.
 
 Portions of this file contain modifications contributed and copyrighted
 by Percona Inc.. Those modifications are
@@ -1422,7 +1422,8 @@ os_file_create_simple_func(
 
 		access = GENERIC_READ;
 
-	} else if (access_type == OS_FILE_READ_WRITE) {
+	} else if (access_type == OS_FILE_READ_WRITE
+		   || access_type == OS_FILE_READ_WRITE_CACHED) {
 		access = GENERIC_READ | GENERIC_WRITE;
 	} else {
 		ib_logf(IB_LOG_LEVEL_ERROR,
@@ -1526,7 +1527,8 @@ os_file_create_simple_func(
 #ifdef USE_FILE_LOCK
 	if (!srv_read_only_mode
 	    && *success
-	    && access_type == OS_FILE_READ_WRITE
+	    && (access_type == OS_FILE_READ_WRITE
+		|| access_type == OS_FILE_READ_WRITE_CACHED)
 	    && os_file_lock(file, name)) {
 
 		*success = FALSE;
@@ -1554,15 +1556,16 @@ os_file_set_nocache_if_needed(os_file_t file, const char* name,
 			      const char *mode_str, ulint type,
 			      ulint access_type)
 {
-	if (srv_read_only_mode || access_type == OS_FILE_READ_WRITE_CACHED)
+	if (srv_read_only_mode || access_type == OS_FILE_READ_WRITE_CACHED) {
 		return;
+	}
 
 	if (srv_unix_file_flush_method == SRV_UNIX_ALL_O_DIRECT
-	    || (type != OS_LOG_FILE
+	    || (type == OS_LOG_FILE
 		&& (srv_unix_file_flush_method == SRV_UNIX_O_DIRECT
-		    || (srv_unix_file_flush_method
-			== SRV_UNIX_O_DIRECT_NO_FSYNC))))
+		    || (srv_unix_file_flush_method == SRV_UNIX_O_DIRECT_NO_FSYNC)))) {
 		os_file_set_nocache(file, name, mode_str);
+	}
 }
 
 /****************************************************************//**
@@ -2154,8 +2157,9 @@ os_file_create_func(
 
 	} while (retry);
 
-	if (*success) {
+	/* We disable OS caching (O_DIRECT) only on data files */
 
+	if (*success) {
 		os_file_set_nocache_if_needed(file, name, mode_str, type, 0);
 	}