MDEV-11520 Extending an InnoDB data file unnecessarily allocates

a large memory buffer on Windows

fil_extend_space_to_desired_size(), os_file_set_size(): Use calloc()
for memory allocation, and handle failures. Properly check the return
status of posix_fallocate().

On Windows, instead of extending the file by at most 1 megabyte at a time,
write a zero-filled page at the end of the file.
According to the Microsoft blog post
https://blogs.msdn.microsoft.com/oldnewthing/20110922-00/?p=9573
this will physically extend the file by writing zero bytes.
(InnoDB never uses DeviceIoControl() to set the file sparse.)

For innodb_plugin, port the XtraDB fix for MySQL Bug#56433
(introducing fil_system->file_extend_mutex). The bug was
fixed differently in MySQL 5.6 (and MariaDB Server 10.0).

storage/innobase/fil/fil0fil.c

 /*****************************************************************************
 
 Copyright (c) 1995, 2013, Oracle and/or its affiliates. All Rights Reserved.
+Copyright (c) 2014, 2017, MariaDB Corporation. 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 the Free Software
@@ -248,6 +249,7 @@ the ib_logfiles form a 'space' and it is handled here */
 struct fil_system_struct {
 #ifndef UNIV_HOTBACKUP
 	mutex_t		mutex;		/*!< The mutex protecting the cache */
+	mutex_t		file_extend_mutex;
 #endif /* !UNIV_HOTBACKUP */
 	hash_table_t*	spaces;		/*!< The hash table of spaces in the
 					system; they are hashed on the space
@@ -1658,6 +1660,8 @@ fil_init(
 
 	mutex_create(fil_system_mutex_key,
 		     &fil_system->mutex, SYNC_ANY_LATCH);
+	mutex_create(fil_system_mutex_key,
+		     &fil_system->file_extend_mutex, SYNC_OUTER_ANY_LATCH);
 
 	fil_system->spaces = hash_create(hash_size);
 	fil_system->name_hash = hash_create(hash_size);
@@ -4096,6 +4100,10 @@ fil_extend_space_to_desired_size(
 	ulint		page_size;
 	ibool		success		= TRUE;
 
+	/* fil_system->file_extend_mutex is for http://bugs.mysql.com/56433
+	to prevent concurrent fil_extend_space_to_desired_size()
+	while fil_system->mutex is temporarily released */
+	mutex_enter(&fil_system->file_extend_mutex);
 	fil_mutex_enter_and_prepare_for_io(space_id);
 
 	space = fil_space_get_by_id(space_id);
@@ -4107,6 +4115,7 @@ fil_extend_space_to_desired_size(
 		*actual_size = space->size;
 
 		mutex_exit(&fil_system->mutex);
+		mutex_exit(&fil_system->file_extend_mutex);
 
 		return(TRUE);
 	}
@@ -4123,22 +4132,24 @@ fil_extend_space_to_desired_size(
 	start_page_no = space->size;
 	file_start_page_no = space->size - node->size;
 
+	mutex_exit(&fil_system->mutex);
+
 #ifdef HAVE_POSIX_FALLOCATE
 	if (srv_use_posix_fallocate) {
 		ib_int64_t start_offset = start_page_no * page_size;
 		ib_int64_t end_offset   = (size_after_extend - start_page_no) * page_size;
 		ib_int64_t desired_size = size_after_extend*page_size;
+		int err = posix_fallocate(
+			node->handle, start_offset, end_offset);
 
-		mutex_exit(&fil_system->mutex);
+		success = !err;
 
-		if (posix_fallocate(node->handle, start_offset, end_offset) == -1) {
-			fprintf(stderr, "InnoDB: Error: preallocating file "
-				"space for file \'%s\' failed. Current size "
-				" %lld, len %lld, desired size %lld\n",
-				node->name, start_offset, end_offset, desired_size);
-			success = FALSE;
-		} else {
-			success = TRUE;
+		if (!success) {
+			fprintf(stderr,
+				"InnoDB: Error: extending file %s"
+				" from %lld to %lld bytes"
+				" failed with error %d\n",
+				node->name, start_offset, end_offset, err);
 		}
 
 		mutex_enter(&fil_system->mutex);
@@ -4154,14 +4165,25 @@ fil_extend_space_to_desired_size(
 	}
 #endif
 
+#ifdef _WIN32
+	/* Write 1 page of zeroes at the desired end. */
+	start_page_no = size_after_extend - 1;
+	buf_size = page_size;
+#else
 	/* Extend at most 64 pages at a time */
 	buf_size = ut_min(64, size_after_extend - start_page_no) * page_size;
-	buf2 = mem_alloc(buf_size + page_size);
+#endif
+	buf2 = calloc(1, buf_size + page_size);
+	if (!buf2) {
+		fprintf(stderr, "InnoDB: Cannot allocate " ULINTPF
+			" bytes to extend file\n",
+			buf_size + page_size);
+		mutex_exit(&fil_system->file_extend_mutex);
+		return(FALSE);
+	}
 	buf = ut_align(buf2, page_size);
 
-	memset(buf, 0, buf_size);
-
-	while (start_page_no < size_after_extend) {
+	for (;;) {
 		ulint	n_pages = ut_min(buf_size / page_size,
 					 size_after_extend - start_page_no);
 
@@ -4170,6 +4192,7 @@ fil_extend_space_to_desired_size(
 		offset_low  = ((start_page_no - file_start_page_no)
 			       % (4096 * ((1024 * 1024) / page_size)))
 			* page_size;
+
 #ifdef UNIV_HOTBACKUP
 		success = os_file_write(node->name, node->handle, buf,
 					offset_low, offset_high,
@@ -4181,34 +4204,37 @@ fil_extend_space_to_desired_size(
 				 page_size * n_pages,
 				 NULL, NULL);
 #endif
-		if (success) {
-			node->size += n_pages;
-			space->size += n_pages;
 
-			os_has_said_disk_full = FALSE;
-		} else {
-			/* Let us measure the size of the file to determine
-			how much we were able to extend it */
+		/* Let us measure the size of the file to determine
+		how much we were able to extend it */
 
-			n_pages = ((ulint)
-				   (os_file_get_size_as_iblonglong(
-					   node->handle)
-				    / page_size)) - node->size;
+		n_pages = (ulint) (os_file_get_size_as_iblonglong(node->handle)
+				   / page_size);
 
-			node->size += n_pages;
-			space->size += n_pages;
+		mutex_enter(&fil_system->mutex);
+		ut_a(n_pages >= node->size);
+
+		start_page_no += n_pages - node->size;
+		space->size += n_pages - node->size;
+		node->size = n_pages;
 
+		if (success) {
+			os_has_said_disk_full = FALSE;
+		}
+
+		if (!success || start_page_no >= size_after_extend) {
 			break;
 		}
 
-		start_page_no += n_pages;
+		mutex_exit(&fil_system->mutex);
 	}
 
-	mem_free(buf2);
-
+	free(buf2);
 	fil_node_complete_io(node, fil_system, OS_FILE_WRITE);
 
+#ifdef HAVE_POSIX_FALLOCATE
 complete_io:
+#endif /* HAVE_POSIX_FALLOCATE */
 
 	*actual_size = space->size;
 
@@ -4228,6 +4254,7 @@ fil_extend_space_to_desired_size(
 	printf("Extended %s to %lu, actual size %lu pages\n", space->name,
 	size_after_extend, *actual_size); */
 	mutex_exit(&fil_system->mutex);
+	mutex_exit(&fil_system->file_extend_mutex);
 
 	fil_flush(space_id);
 

storage/innobase/include/sync0sync.h

@@ -675,6 +675,7 @@ or row lock! */
 #define	SYNC_BUF_BLOCK		146	/* Block mutex */
 #define	SYNC_BUF_FLUSH_LIST	145	/* Buffer flush list mutex */
 #define SYNC_DOUBLEWRITE	140
+#define	SYNC_OUTER_ANY_LATCH	136
 #define	SYNC_ANY_LATCH		135
 #define	SYNC_MEM_HASH		131
 #define	SYNC_MEM_POOL		130

storage/innobase/os/os0file.c

@@ -2,6 +2,7 @@
 
 Copyright (c) 1995, 2015, Oracle and/or its affiliates. All Rights Reserved.
 Copyright (c) 2009, Percona Inc.
+Copyright (c) 2012, 2017, MariaDB Corporation. All Rights Reserved.
 
 Portions of this file contain modifications contributed and copyrighted
 by Percona Inc.. Those modifications are
@@ -2027,48 +2028,44 @@ os_file_set_size(
 
 	ut_a(size == (size & 0xFFFFFFFF));
 
-	current_size = 0;
 	desired_size = (ib_int64_t)size + (((ib_int64_t)size_high) << 32);
 
 #ifdef HAVE_POSIX_FALLOCATE
-        if (srv_use_posix_fallocate) {
-		if (posix_fallocate(file, current_size, desired_size) == -1) {
+	if (srv_use_posix_fallocate) {
+		int err = posix_fallocate(file, 0, desired_size);
+		if (err) {
 			fprintf(stderr,
-		 	"InnoDB: Error: preallocating data for"
-			" file %s failed at\n"
-			"InnoDB: offset 0 size %lld %lld. Operating system"
-			" error number %d.\n"
-			"InnoDB: Check that the disk is not full"
-			" or a disk quota exceeded.\n"
-			"InnoDB: Some operating system error numbers"
-			" are described at\n"
-			"InnoDB: "
-			REFMAN "operating-system-error-codes.html\n",
-			name,  (long long)size_high,  (long long)size, errno);
-
-			return (FALSE);
+				"InnoDB: Error: preallocating %lld bytes for"
+				" file %s failed with error %d.\n",
+				desired_size, name, err);
 		}
-		return (TRUE);
+		return(!err);
 	}
 #endif
 
+#ifdef _WIN32
+	/* Write 1 page of zeroes at the desired end. */
+	buf_size = UNIV_PAGE_SIZE;
+	current_size = desired_size - buf_size;
+#else
 	/* Write up to 1 megabyte at a time. */
 	buf_size = ut_min(64, (ulint) (desired_size / UNIV_PAGE_SIZE))
 		* UNIV_PAGE_SIZE;
-	buf2 = ut_malloc(buf_size + UNIV_PAGE_SIZE);
+	current_size = 0;
+#endif
+	buf2 = calloc(1, buf_size + UNIV_PAGE_SIZE);
+
+	if (!buf2) {
+		fprintf(stderr, "InnoDB: Cannot allocate " ULINTPF
+			" bytes to extend file\n",
+			buf_size + UNIV_PAGE_SIZE);
+		return(FALSE);
+	}
 
 	/* Align the buffer for possible raw i/o */
 	buf = ut_align(buf2, UNIV_PAGE_SIZE);
 
-	/* Write buffer full of zeros */
-	memset(buf, 0, buf_size);
-
-	if (desired_size >= (ib_int64_t)(100 * 1024 * 1024)) {
-
-		fprintf(stderr, "InnoDB: Progress in MB:");
-	}
-
-	while (current_size < desired_size) {
+	do {
 		ulint	n_bytes;
 
 		if (desired_size - current_size < (ib_int64_t) buf_size) {
@@ -2082,37 +2079,14 @@ os_file_set_size(
 				    (ulint)(current_size >> 32),
 				    n_bytes);
 		if (!ret) {
-			ut_free(buf2);
-			goto error_handling;
-		}
-
-		/* Print about progress for each 100 MB written */
-		if ((ib_int64_t) (current_size + n_bytes) / (ib_int64_t)(100 * 1024 * 1024)
-		    != current_size / (ib_int64_t)(100 * 1024 * 1024)) {
-
-			fprintf(stderr, " %lu00",
-				(ulong) ((current_size + n_bytes)
-					 / (ib_int64_t)(100 * 1024 * 1024)));
+			break;
 		}
 
 		current_size += n_bytes;
-	}
-
-	if (desired_size >= (ib_int64_t)(100 * 1024 * 1024)) {
-
-		fprintf(stderr, "\n");
-	}
+	} while (current_size < desired_size);
 
-	ut_free(buf2);
-
-	ret = os_file_flush(file);
-
-	if (ret) {
-		return(TRUE);
-	}
-
-error_handling:
-	return(FALSE);
+	free(buf2);
+	return(ret && os_file_flush(file));
 }
 
 /***********************************************************************//**

storage/xtradb/fil/fil0fil.c

 /*****************************************************************************
 
 Copyright (c) 1995, 2013, Oracle and/or its affiliates. All Rights Reserved.
+Copyright (c) 2014, 2017, MariaDB Corporation. 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 the Free Software
@@ -4934,9 +4935,9 @@ fil_extend_space_to_desired_size(
 	ulint		page_size;
 	ibool		success		= TRUE;
 
-	/* file_extend_mutex is for http://bugs.mysql.com/56433 */
-	/* to protect from the other fil_extend_space_to_desired_size() */
-	/* during temprary releasing &fil_system->mutex */
+	/* fil_system->file_extend_mutex is for http://bugs.mysql.com/56433
+	to prevent concurrent fil_extend_space_to_desired_size()
+	while fil_system->mutex is temporarily released */
 	mutex_enter(&fil_system->file_extend_mutex);
 	fil_mutex_enter_and_prepare_for_io(space_id);
 
@@ -4966,6 +4967,8 @@ fil_extend_space_to_desired_size(
 	start_page_no = space->size;
 	file_start_page_no = space->size - node->size;
 
+	mutex_exit(&fil_system->mutex);
+
 #ifdef HAVE_POSIX_FALLOCATE
 	if (srv_use_posix_fallocate) {
 
@@ -4973,19 +4976,19 @@ fil_extend_space_to_desired_size(
 			= file_start_page_no * page_size;
 		ib_int64_t	end_offset
 			= (size_after_extend - file_start_page_no) * page_size;
+		int err = posix_fallocate(
+			node->handle, start_offset, end_offset);
 
-		mutex_exit(&fil_system->mutex);
-		success = (posix_fallocate(node->handle, start_offset,
-					   end_offset) == 0);
-		if (!success)
-		{
+		success = !err;
+
+		if (!success) {
 			fprintf(stderr,
-				"InnoDB: Error: preallocating file space for "
-				"file \'%s\' failed.  Current size %lld, "
-				"len %lld, desired size %lld\n", node->name,
-				start_offset, end_offset,
-				start_offset + end_offset);
+				"InnoDB: Error: extending file %s"
+				" from %lld to %lld bytes"
+				" failed with error %d\n",
+				node->name, start_offset, end_offset, err);
 		}
+
 		mutex_enter(&fil_system->mutex);
 
 		if (success) {
@@ -4999,14 +5002,25 @@ fil_extend_space_to_desired_size(
 	}
 #endif
 
+#ifdef _WIN32
+	/* Write 1 page of zeroes at the desired end. */
+	start_page_no = size_after_extend - 1;
+	buf_size = page_size;
+#else
 	/* Extend at most 64 pages at a time */
 	buf_size = ut_min(64, size_after_extend - start_page_no) * page_size;
-	buf2 = mem_alloc(buf_size + page_size);
+#endif
+	buf2 = calloc(1, buf_size + page_size);
+	if (!buf2) {
+		fprintf(stderr, "InnoDB: Cannot allocate " ULINTPF
+			" bytes to extend file\n",
+			buf_size + page_size);
+		mutex_exit(&fil_system->file_extend_mutex);
+		return(FALSE);
+	}
 	buf = ut_align(buf2, page_size);
 
-	memset(buf, 0, buf_size);
-
-	while (start_page_no < size_after_extend) {
+	for (;;) {
 		ulint	n_pages = ut_min(buf_size / page_size,
 					 size_after_extend - start_page_no);
 
@@ -5016,7 +5030,6 @@ fil_extend_space_to_desired_size(
 			       % (4096 * ((1024 * 1024) / page_size)))
 			* page_size;
 
-		mutex_exit(&fil_system->mutex);
 #ifdef UNIV_HOTBACKUP
 		success = os_file_write(node->name, node->handle, buf,
 					offset_low, offset_high,
@@ -5028,36 +5041,37 @@ fil_extend_space_to_desired_size(
 				 page_size * n_pages,
 				 NULL, NULL, space_id, NULL);
 #endif
-		mutex_enter(&fil_system->mutex);
 
-		if (success) {
-			node->size += n_pages;
-			space->size += n_pages;
+		/* Let us measure the size of the file to determine
+		how much we were able to extend it */
 
-			os_has_said_disk_full = FALSE;
-		} else {
-			/* Let us measure the size of the file to determine
-			how much we were able to extend it */
+		n_pages = (ulint) (os_file_get_size_as_iblonglong(node->handle)
+				   / page_size);
 
-			n_pages = ((ulint)
-				   (os_file_get_size_as_iblonglong(
-					   node->handle)
-				    / page_size)) - node->size;
+		mutex_enter(&fil_system->mutex);
+		ut_a(n_pages >= node->size);
 
-			node->size += n_pages;
-			space->size += n_pages;
+		start_page_no += n_pages - node->size;
+		space->size += n_pages - node->size;
+		node->size = n_pages;
 
+		if (success) {
+			os_has_said_disk_full = FALSE;
+		}
+
+		if (!success || start_page_no >= size_after_extend) {
 			break;
 		}
 
-		start_page_no += n_pages;
+		mutex_exit(&fil_system->mutex);
 	}
 
-	mem_free(buf2);
-
+	free(buf2);
 	fil_node_complete_io(node, fil_system, OS_FILE_WRITE);
 
+#ifdef HAVE_POSIX_FALLOCATE
 complete_io:
+#endif /* HAVE_POSIX_FALLOCATE */
 
 	*actual_size = space->size;
 

storage/xtradb/os/os0file.c

@@ -2,6 +2,7 @@
 
 Copyright (c) 1995, 2015, Oracle and/or its affiliates. All Rights Reserved.
 Copyright (c) 2009, Percona Inc.
+Copyright (c) 2011, 2017, MariaDB Corporation. All Rights Reserved.
 
 Portions of this file contain modifications contributed and copyrighted
 by Percona Inc.. Those modifications are
@@ -2184,42 +2185,44 @@ os_file_set_size(
 
 	ut_a(size == (size & 0xFFFFFFFF));
 
-	current_size = 0;
 	desired_size = (ib_int64_t)size + (((ib_int64_t)size_high) << 32);
 
 #ifdef HAVE_POSIX_FALLOCATE
 	if (srv_use_posix_fallocate) {
-
-		if (posix_fallocate(file, current_size, desired_size) == -1) {
-
-			fprintf(stderr, "InnoDB: Error: preallocating file "
-				"space for file \'%s\' failed.  Current size "
-				"%lld, desired size %lld\n",
-				name, current_size, desired_size);
-			os_file_handle_error_no_exit(name, "posix_fallocate");
-			return(FALSE);
+		int err = posix_fallocate(file, 0, desired_size);
+		if (err) {
+			fprintf(stderr,
+				"InnoDB: Error: preallocating %lld bytes for"
+				" file %s failed with error %d.\n",
+				desired_size, name, err);
 		}
-		return(TRUE);
+		return(!err);
 	}
 #endif
 
+#ifdef _WIN32
+	/* Write 1 page of zeroes at the desired end. */
+	buf_size = UNIV_PAGE_SIZE;
+	current_size = desired_size - buf_size;
+#else
 	/* Write up to 1 megabyte at a time. */
 	buf_size = ut_min(64, (ulint) (desired_size / UNIV_PAGE_SIZE))
 		* UNIV_PAGE_SIZE;
-	buf2 = ut_malloc(buf_size + UNIV_PAGE_SIZE);
+	current_size = 0;
+#endif
+	buf2 = calloc(1, buf_size + UNIV_PAGE_SIZE);
+
+	if (!buf2) {
+		fprintf(stderr, "InnoDB: Cannot allocate " ULINTPF
+			" bytes to extend file\n",
+			buf_size + UNIV_PAGE_SIZE);
+		return(FALSE);
+	}
 
 	/* Align the buffer for possible raw i/o */
 	buf = ut_align(buf2, UNIV_PAGE_SIZE);
 
-	/* Write buffer full of zeros */
-	memset(buf, 0, buf_size);
-
-	if (desired_size >= (ib_int64_t)(100 * 1024 * 1024)) {
-
-		fprintf(stderr, "InnoDB: Progress in MB:");
-	}
-
-	while (current_size < desired_size) {
+	do {
 		ulint	n_bytes;
 
 		if (desired_size - current_size < (ib_int64_t) buf_size) {
@@ -2233,37 +2236,14 @@ os_file_set_size(
 				    (ulint)(current_size >> 32),
 				    n_bytes);
 		if (!ret) {
-			ut_free(buf2);
-			goto error_handling;
-		}
-
-		/* Print about progress for each 100 MB written */
-		if ((ib_int64_t) (current_size + n_bytes) / (ib_int64_t)(100 * 1024 * 1024)
-		    != current_size / (ib_int64_t)(100 * 1024 * 1024)) {
-
-			fprintf(stderr, " %lu00",
-				(ulong) ((current_size + n_bytes)
-					 / (ib_int64_t)(100 * 1024 * 1024)));
+			break;
 		}
 
 		current_size += n_bytes;
-	}
-
-	if (desired_size >= (ib_int64_t)(100 * 1024 * 1024)) {
-
-		fprintf(stderr, "\n");
-	}
-
-	ut_free(buf2);
-
-	ret = os_file_flush(file, TRUE);
-
-	if (ret) {
-		return(TRUE);
-	}
+	} while (current_size < desired_size);
 
-error_handling:
-	return(FALSE);
+	free(buf2);
+	return(ret && os_file_flush(file, TRUE));
 }
 
 /***********************************************************************//**