Merge patch series "iomap: fix zero padding data issue in concurrent append writes"

Long Li <leo.lilong@huawei.com> says:

This patch series fixes zero padding data issues in concurrent append
write scenarios. A detailed problem description and solution can be
found in patch 2. Patch 1 is introduced as preparation for the fix in
patch 2, eliminating the need to resample inode size for io_size
trimming and avoiding issues caused by inode size changes during
concurrent writeback and truncate operations.

* patches from https://lore.kernel.org/r/20241209114241.3725722-1-leo.lilong@huawei.com:
  iomap: fix zero padding data issue in concurrent append writes
  iomap: pass byte granular end position to iomap_add_to_ioend

Link: https://lore.kernel.org/r/20241209114241.3725722-1-leo.lilong@huawei.com
Signed-off-by: Christian Brauner <brauner@kernel.org>

Author: brauner

fs/iomap/buffered-io.c

@@ -1774,7 +1774,8 @@ static bool iomap_can_add_to_ioend(struct iomap_writepage_ctx *wpc, loff_t pos)
  */
 static int iomap_add_to_ioend(struct iomap_writepage_ctx *wpc,
 		struct writeback_control *wbc, struct folio *folio,
-		struct inode *inode, loff_t pos, unsigned len)
+		struct inode *inode, loff_t pos, loff_t end_pos,
+		unsigned len)
 {
 	struct iomap_folio_state *ifs = folio->private;
 	size_t poff = offset_in_folio(folio, pos);
@@ -1793,15 +1794,60 @@ static int iomap_add_to_ioend(struct iomap_writepage_ctx *wpc,
 
 	if (ifs)
 		atomic_add(len, &ifs->write_bytes_pending);
+
+	/*
+	 * Clamp io_offset and io_size to the incore EOF so that ondisk
+	 * file size updates in the ioend completion are byte-accurate.
+	 * This avoids recovering files with zeroed tail regions when
+	 * writeback races with appending writes:
+	 *
+	 *    Thread 1:                  Thread 2:
+	 *    ------------               -----------
+	 *    write [A, A+B]
+	 *    update inode size to A+B
+	 *    submit I/O [A, A+BS]
+	 *                               write [A+B, A+B+C]
+	 *                               update inode size to A+B+C
+	 *    <I/O completes, updates disk size to min(A+B+C, A+BS)>
+	 *    <power failure>
+	 *
+	 *  After reboot:
+	 *    1) with A+B+C < A+BS, the file has zero padding in range
+	 *       [A+B, A+B+C]
+	 *
+	 *    |<     Block Size (BS)   >|
+	 *    |DDDDDDDDDDDD0000000000000|
+	 *    ^           ^        ^
+	 *    A          A+B     A+B+C
+	 *                       (EOF)
+	 *
+	 *    2) with A+B+C > A+BS, the file has zero padding in range
+	 *       [A+B, A+BS]
+	 *
+	 *    |<     Block Size (BS)   >|<     Block Size (BS)    >|
+	 *    |DDDDDDDDDDDD0000000000000|00000000000000000000000000|
+	 *    ^           ^             ^           ^
+	 *    A          A+B           A+BS       A+B+C
+	 *                             (EOF)
+	 *
+	 *    D = Valid Data
+	 *    0 = Zero Padding
+	 *
+	 * Note that this defeats the ability to chain the ioends of
+	 * appending writes.
+	 */
 	wpc->ioend->io_size += len;
+	if (wpc->ioend->io_offset + wpc->ioend->io_size > end_pos)
+		wpc->ioend->io_size = end_pos - wpc->ioend->io_offset;
+
 	wbc_account_cgroup_owner(wbc, folio, len);
 	return 0;
 }
 
 static int iomap_writepage_map_blocks(struct iomap_writepage_ctx *wpc,
 		struct writeback_control *wbc, struct folio *folio,
-		struct inode *inode, u64 pos, unsigned dirty_len,
-		unsigned *count)
+		struct inode *inode, u64 pos, u64 end_pos,
+		unsigned dirty_len, unsigned *count)
 {
 	int error;
 
@@ -1826,7 +1872,7 @@ static int iomap_writepage_map_blocks(struct iomap_writepage_ctx *wpc,
 			break;
 		default:
 			error = iomap_add_to_ioend(wpc, wbc, folio, inode, pos,
-					map_len);
+					end_pos, map_len);
 			if (!error)
 				(*count)++;
 			break;
@@ -1897,11 +1943,11 @@ static bool iomap_writepage_handle_eof(struct folio *folio, struct inode *inode,
 		 *    remaining memory is zeroed when mapped, and writes to that
 		 *    region are not written out to the file.
 		 *
-		 * Also adjust the writeback range to skip all blocks entirely
-		 * beyond i_size.
+		 * Also adjust the end_pos to the end of file and skip writeback
+		 * for all blocks entirely beyond i_size.
 		 */
 		folio_zero_segment(folio, poff, folio_size(folio));
-		*end_pos = round_up(isize, i_blocksize(inode));
+		*end_pos = isize;
 	}
 
 	return true;
@@ -1914,6 +1960,7 @@ static int iomap_writepage_map(struct iomap_writepage_ctx *wpc,
 	struct inode *inode = folio->mapping->host;
 	u64 pos = folio_pos(folio);
 	u64 end_pos = pos + folio_size(folio);
+	u64 end_aligned = 0;
 	unsigned count = 0;
 	int error = 0;
 	u32 rlen;
@@ -1955,9 +2002,10 @@ static int iomap_writepage_map(struct iomap_writepage_ctx *wpc,
 	/*
 	 * Walk through the folio to find dirty areas to write back.
 	 */
-	while ((rlen = iomap_find_dirty_range(folio, &pos, end_pos))) {
+	end_aligned = round_up(end_pos, i_blocksize(inode));
+	while ((rlen = iomap_find_dirty_range(folio, &pos, end_aligned))) {
 		error = iomap_writepage_map_blocks(wpc, wbc, folio, inode,
-				pos, rlen, &count);
+				pos, end_pos, rlen, &count);
 		if (error)
 			break;
 		pos += rlen;

include/linux/iomap.h

@@ -335,7 +335,7 @@ struct iomap_ioend {
 	u16			io_type;
 	u16			io_flags;	/* IOMAP_F_* */
 	struct inode		*io_inode;	/* file being written to */
-	size_t			io_size;	/* size of the extent */
+	size_t			io_size;	/* size of data within eof */
 	loff_t			io_offset;	/* offset in the file */
 	sector_t		io_sector;	/* start sector of ioend */
 	struct bio		io_bio;		/* MUST BE LAST! */