Fixed database corruption caused by innodb-use-trim

8 files changed, 36 insertions, 42 deletions

diff --git a/storage/innobase/fil/fil0fil.cc b/storage/innobase/fil/fil0fil.cc
index 3d583a6df15..db304203a18 100644
--- a/storage/innobase/fil/fil0fil.cc
+++ b/storage/innobase/fil/fil0fil.cc
@@ -5225,7 +5225,7 @@ retry:
 #else
 		success = os_aio(OS_FILE_WRITE, 0, OS_AIO_SYNC,
 				 node->name, node->handle, buf,
-				 offset, page_size * n_pages,
+			         offset, page_size * n_pages, page_size,
 			         node, NULL, 0);
 #endif /* UNIV_HOTBACKUP */
 
@@ -5879,6 +5879,7 @@ fil_io(
 		buf,
 		offset,
 		len,
+		zip_size ? zip_size : UNIV_PAGE_SIZE,
 		node,
 		message,
 		write_size);
diff --git a/storage/innobase/include/os0file.h b/storage/innobase/include/os0file.h
index 005b44108db..c630e91e815 100644
--- a/storage/innobase/include/os0file.h
+++ b/storage/innobase/include/os0file.h
@@ -312,9 +312,9 @@ The wrapper functions have the prefix of "innodb_". */
 	pfs_os_file_close_func(file, __FILE__, __LINE__)
 
 # define os_aio(type, is_log, mode, name, file, buf, offset,		\
-	n, message1, message2, write_size)				\
+	n, page_size, message1, message2, write_size)			\
 	pfs_os_aio_func(type, is_log, mode, name, file, buf, offset,	\
-			n, message1, message2, write_size,              \
+		        n, page_size, message1, message2, write_size,	\
 		        __FILE__, __LINE__)
 
 
@@ -357,10 +357,10 @@ to original un-instrumented file I/O APIs */
 
 # define os_file_close(file)	os_file_close_func(file)
 
-# define os_aio(type, is_log, mode, name, file, buf, offset, n, message1, \
+# define os_aio(type, is_log, mode, name, file, buf, offset, n, page_size, message1, \
 	message2, write_size)						\
 	os_aio_func(type, is_log, mode, name, file, buf, offset, n,	\
-		message1, message2, write_size)
+		page_size, message1, message2, write_size)
 
 # define os_file_read(file, buf, offset, n)				\
 	os_file_read_func(file, buf, offset, n)
@@ -758,6 +758,7 @@ pfs_os_aio_func(
 				to write */
 	os_offset_t	offset,	/*!< in: file offset where to read or write */
 	ulint		n,	/*!< in: number of bytes to read or write */
+	ulint           page_size, /*!< in: page size in bytes */
 	fil_node_t*	message1,/*!< in: message for the aio handler
 				(can be used to identify a completed
 				aio operation); ignored if mode is
@@ -1129,6 +1130,7 @@ os_aio_func(
 				to write */
 	os_offset_t	offset,	/*!< in: file offset where to read or write */
 	ulint		n,	/*!< in: number of bytes to read or write */
+	ulint           page_size, /*!< in: page size in bytes */
 	fil_node_t*	message1,/*!< in: message for the aio handler
 				(can be used to identify a completed
 				aio operation); ignored if mode is
diff --git a/storage/innobase/include/os0file.ic b/storage/innobase/include/os0file.ic
index bafa3976763..6ca8b371093 100644
--- a/storage/innobase/include/os0file.ic
+++ b/storage/innobase/include/os0file.ic
@@ -208,6 +208,7 @@ pfs_os_aio_func(
 				to write */
 	os_offset_t	offset,	/*!< in: file offset where to read or write */
 	ulint		n,	/*!< in: number of bytes to read or write */
+	ulint           page_size, /*!< in: page size in bytes */
 	fil_node_t*	message1,/*!< in: message for the aio handler
 				(can be used to identify a completed
 				aio operation); ignored if mode is
@@ -236,7 +237,7 @@ pfs_os_aio_func(
 				   src_file, src_line);
 
 	result = os_aio_func(type, is_log, mode, name, file, buf, offset,
-		n, message1, message2, write_size);
+			     n, page_size, message1, message2, write_size);
 
 	register_pfs_file_io_end(locker, n);
 
diff --git a/storage/innobase/os/os0file.cc b/storage/innobase/os/os0file.cc
index 606437486e3..03beeca6591 100644
--- a/storage/innobase/os/os0file.cc
+++ b/storage/innobase/os/os0file.cc
@@ -204,7 +204,8 @@ struct os_aio_slot_t{
 					write */
 	byte*		buf;		/*!< buffer used in i/o */
 	ulint		type;		/*!< OS_FILE_READ or OS_FILE_WRITE */
-	ulint		is_log;		/*!< in: 1 is OS_FILE_LOG or 0 */
+	ulint		is_log;		/*!< 1 if OS_FILE_LOG or 0 */
+	ulint		page_size;      /*!< UNIV_PAGE_SIZE or zip_size */
 
 	os_offset_t	offset;		/*!< file offset in bytes */
 	os_file_t	file;		/*!< file where to read or write */
@@ -4484,6 +4485,7 @@ os_aio_array_reserve_slot(
 				to write */
 	os_offset_t	offset,	/*!< in: file offset */
 	ulint		len,	/*!< in: length of the block to read or write */
+	ulint           page_size, /*!< in: page size in bytes */
 	ulint*		write_size)/*!< in/out: Actual write size initialized
 			       after fist successfull trim
 			       operation for this page and if
@@ -4579,6 +4581,7 @@ found:
 	slot->io_already_done = FALSE;
 	slot->write_size = write_size;
 	slot->is_log   = is_log;
+	slot->page_size = page_size;
 
 	if (message1) {
 		slot->file_block_size = fil_node_get_block_size(message1);
@@ -4856,6 +4859,7 @@ os_aio_func(
 				to write */
 	os_offset_t	offset,	/*!< in: file offset where to read or write */
 	ulint		n,	/*!< in: number of bytes to read or write */
+	ulint           page_size, /*!< in: page size in bytes */
 	fil_node_t*	message1,/*!< in: message for the aio handler
 				(can be used to identify a completed
 				aio operation); ignored if mode is
@@ -4983,7 +4987,7 @@ try_again:
 	}
 
 	slot = os_aio_array_reserve_slot(type, is_log, array, message1, message2, file,
-		name, buf, offset, n, write_size);
+					 name, buf, offset, n, page_size, write_size);
 
 	if (type == OS_FILE_READ) {
 		if (srv_use_native_aio) {
@@ -6226,19 +6230,13 @@ os_file_trim(
 {
 
 	size_t len = slot->len;
-	size_t trim_len = UNIV_PAGE_SIZE - len;
+	size_t trim_len = slot->page_size - len;
 	os_offset_t off = slot->offset + len;
 	size_t bsize = slot->file_block_size;
 
-	// len here should be alligned to sector size
-	ut_ad((trim_len % bsize) == 0);
-	ut_ad((len % bsize) == 0);
-	ut_ad(bsize != 0);
-	ut_ad((off % bsize) == 0);
-
 #ifdef UNIV_TRIM_DEBUG
 	fprintf(stderr, "Note: TRIM: write_size %lu trim_len %lu len %lu off %lu bz %lu\n",
-		*slot->write_size, trim_len, len, off, bsize);
+		slot->write_size ? *slot->write_size : 0, trim_len, len, off, bsize);
 #endif
 
 	// Nothing to do if trim length is zero or if actual write
@@ -6253,11 +6251,6 @@ os_file_trim(
 		    *slot->write_size > 0 &&
 		    len >= *slot->write_size)) {
 
-#ifdef UNIV_PAGECOMPRESS_DEBUG
-		fprintf(stderr, "Note: TRIM: write_size %lu trim_len %lu len %lu\n",
-			*slot->write_size, trim_len, len);
-#endif
-
 		if (slot->write_size) {
 			if (*slot->write_size > 0 && len >= *slot->write_size) {
 				srv_stats.page_compressed_trim_op_saved.inc();
diff --git a/storage/xtradb/fil/fil0fil.cc b/storage/xtradb/fil/fil0fil.cc
index 2f0cebf0522..70f0ee14ab5 100644
--- a/storage/xtradb/fil/fil0fil.cc
+++ b/storage/xtradb/fil/fil0fil.cc
@@ -5255,7 +5255,7 @@ retry:
 #else
 		success = os_aio(OS_FILE_WRITE, 0, OS_AIO_SYNC,
 				 node->name, node->handle, buf,
-				 offset, page_size * n_pages,
+				 offset, page_size * n_pages, page_size,
 				 node, NULL, space_id, NULL, 0);
 #endif /* UNIV_HOTBACKUP */
 
@@ -5925,6 +5925,7 @@ _fil_io(
 		buf,
 		offset,
 		len,
+		zip_size ? zip_size : UNIV_PAGE_SIZE,
 		node,
 		message,
 		space_id,
diff --git a/storage/xtradb/include/os0file.h b/storage/xtradb/include/os0file.h
index b16738e28fa..b1058f07c8e 100644
--- a/storage/xtradb/include/os0file.h
+++ b/storage/xtradb/include/os0file.h
@@ -322,10 +322,10 @@ The wrapper functions have the prefix of "innodb_". */
 	pfs_os_file_close_func(file, __FILE__, __LINE__)
 
 # define os_aio(type, is_log, mode, name, file, buf, offset,		\
-	n, message1, message2, space_id, 				\
+	n, page_size, message1, message2, space_id,			\
 	trx, write_size) 					 	\
 	pfs_os_aio_func(type, is_log, mode, name, file, buf, offset,	\
-		n, message1, message2, space_id, trx, write_size,	\
+		n, page_size, message1, message2, space_id, trx, write_size, \
 		__FILE__, __LINE__)
 
 # define os_file_read(file, buf, offset, n)				\
@@ -372,10 +372,10 @@ to original un-instrumented file I/O APIs */
 
 # define os_file_close(file)	os_file_close_func(file)
 
-# define os_aio(type, is_log, mode, name, file, buf, offset, n, message1, \
+# define os_aio(type, is_log, mode, name, file, buf, offset, n, page_size, message1, \
 	message2, space_id, trx, write_size)				\
 	os_aio_func(type, is_log, mode, name, file, buf, offset, n,	\
-		message1, message2, space_id, trx, write_size)
+		page_size, message1, message2, space_id, trx, write_size)
 
 # define os_file_read(file, buf, offset, n)				\
 	os_file_read_func(file, buf, offset, n, NULL)
@@ -781,6 +781,7 @@ pfs_os_aio_func(
 				to write */
 	os_offset_t	offset,	/*!< in: file offset where to read or write */
 	ulint		n,	/*!< in: number of bytes to read or write */
+	ulint		page_size,/*!< in: page size in bytes */
 	fil_node_t*	message1,/*!< in: message for the aio handler
 				(can be used to identify a completed
 				aio operation); ignored if mode is
@@ -1161,6 +1162,7 @@ os_aio_func(
 				to write */
 	os_offset_t	offset,	/*!< in: file offset where to read or write */
 	ulint		n,	/*!< in: number of bytes to read or write */
+	ulint		page_size, /*!< in: page size in bytes */
 	fil_node_t*	message1,/*!< in: message for the aio handler
 				(can be used to identify a completed
 				aio operation); ignored if mode is
diff --git a/storage/xtradb/include/os0file.ic b/storage/xtradb/include/os0file.ic
index cfce273882c..b284d7ea9ac 100644
--- a/storage/xtradb/include/os0file.ic
+++ b/storage/xtradb/include/os0file.ic
@@ -211,6 +211,7 @@ pfs_os_aio_func(
 				to write */
 	os_offset_t	offset,	/*!< in: file offset where to read or write */
 	ulint		n,	/*!< in: number of bytes to read or write */
+	ulint		page_size, /*!< in: page size in bytes */
 	fil_node_t*	message1,/*!< in: message for the aio handler
 				(can be used to identify a completed
 				aio operation); ignored if mode is
@@ -241,7 +242,7 @@ pfs_os_aio_func(
 				   src_file, src_line);
 
 	result = os_aio_func(type, is_log, mode, name, file, buf, offset,
-			     n, message1, message2, space_id, trx,
+			     n, page_size, message1, message2, space_id, trx,
 			     write_size);
 
 	register_pfs_file_io_end(locker, n);
diff --git a/storage/xtradb/os/os0file.cc b/storage/xtradb/os/os0file.cc
index 26accb84e40..e87ce888e10 100644
--- a/storage/xtradb/os/os0file.cc
+++ b/storage/xtradb/os/os0file.cc
@@ -216,7 +216,8 @@ struct os_aio_slot_t{
 					write */
 	byte*		buf;		/*!< buffer used in i/o */
 	ulint		type;		/*!< OS_FILE_READ or OS_FILE_WRITE */
-	ulint		is_log;		/*!< in: 1 is OS_FILE_LOG or 0 */
+	ulint		is_log;		/*!< 1 is OS_FILE_LOG or 0 */
+	ulint		page_size;	/*!< UNIV_PAGE_SIZE or zip_size */
 
 	os_offset_t	offset;		/*!< file offset in bytes */
 	os_file_t	file;		/*!< file where to read or write */
@@ -4583,6 +4584,7 @@ os_aio_array_reserve_slot(
 				to write */
 	os_offset_t	offset,	/*!< in: file offset */
 	ulint		len,	/*!< in: length of the block to read or write */
+	ulint		page_size, /*!< in: page size in bytes */
 	ulint		space_id,
 	ulint*		write_size)/*!< in/out: Actual write size initialized
 			       after first successfull trim
@@ -4680,6 +4682,7 @@ found:
 	slot->io_already_done = FALSE;
 	slot->space_id = space_id;
 	slot->is_log   = is_log;
+	slot->page_size = page_size;
 
 	if (message1) {
 		slot->file_block_size = fil_node_get_block_size(message1);
@@ -4954,6 +4957,7 @@ os_aio_func(
 				to write */
 	os_offset_t	offset,	/*!< in: file offset where to read or write */
 	ulint		n,	/*!< in: number of bytes to read or write */
+	ulint		page_size, /*!< in: page size in bytes */
 	fil_node_t*	message1,/*!< in: message for the aio handler
 				(can be used to identify a completed
 				aio operation); ignored if mode is
@@ -5073,7 +5077,7 @@ try_again:
 	}
 
 	slot = os_aio_array_reserve_slot(type, is_log, array, message1, message2, file,
-					 name, buf, offset, n, space_id,
+		                         name, buf, offset, n, page_size, space_id,
 		                         write_size);
 
 	if (type == OS_FILE_READ) {
@@ -6305,19 +6309,13 @@ os_file_trim(
 	os_aio_slot_t*	slot) /*!< in: slot structure     */
 {
 	size_t len = slot->len;
-	size_t trim_len = UNIV_PAGE_SIZE - slot->len;
+	size_t trim_len = slot->page_size - slot->len;
 	os_offset_t off __attribute__((unused)) = slot->offset + len;
 	size_t bsize = slot->file_block_size;
 
-	// len here should be alligned to sector size
-	ut_ad((trim_len % bsize) == 0);
-	ut_ad((len % bsize) == 0);
-	ut_ad(bsize != 0);
-	ut_ad((off % bsize) == 0);
-
 #ifdef UNIV_TRIM_DEBUG
 	fprintf(stderr, "Note: TRIM: write_size %lu trim_len %lu len %lu off %lu bz %lu\n",
-		*slot->write_size, trim_len, len, off, bsize);
+		slot->write_size ? *slot->write_size : 0, trim_len, len, off, bsize);
 #endif
 
 	// Nothing to do if trim length is zero or if actual write
@@ -6332,11 +6330,6 @@ os_file_trim(
 		    *slot->write_size > 0 &&
 		    len >= *slot->write_size)) {
 
-#ifdef UNIV_PAGECOMPRESS_DEBUG
-		fprintf(stderr, "Note: TRIM: write_size %lu trim_len %lu len %lu\n",
-			*slot->write_size, trim_len, len);
-#endif
-
 		if (slot->write_size) {
 			if (*slot->write_size > 0 && len >= *slot->write_size) {
 				srv_stats.page_compressed_trim_op_saved.inc();