1 ext4: Fix WARN_ON_ONCE in ext4_commit_super()
3 From: "Pranay Kr. Srivastava" <pranjas@gmail.com>
5 If there are racing calls to ext4_commit_super() it's possible for
6 another writeback of the superblock to result in the buffer being
7 marked with an error after we check if the buffer is marked as having
8 a write error and the buffer up-to-date flag is set again. If that
9 happens mark_buffer_dirty() can end up throwing a WARN_ON_ONCE.
11 Fix this by moving this check to write before we call
12 write_buffer_dirty(), and keeping the buffer locked during this whole
15 Signed-off-by: Pranay Kr. Srivastava <pranjas@gmail.com>
16 Signed-off-by: Theodore Ts'o <tytso@mit.edu>
18 fs/ext4/super.c | 30 ++++++++++++++++--------------
19 1 file changed, 16 insertions(+), 14 deletions(-)
21 diff --git a/fs/ext4/super.c b/fs/ext4/super.c
22 index 6e2f9d6..5664ee6 100644
25 @@ -4327,20 +4327,6 @@ static int ext4_commit_super(struct super_block *sb, int sync)
27 if (!sbh || block_device_ejected(sb))
29 - if (buffer_write_io_error(sbh)) {
31 - * Oh, dear. A previous attempt to write the
32 - * superblock failed. This could happen because the
33 - * USB device was yanked out. Or it could happen to
34 - * be a transient write error and maybe the block will
35 - * be remapped. Nothing we can do but to retry the
36 - * write and hope for the best.
38 - ext4_msg(sb, KERN_ERR, "previous I/O error to "
39 - "superblock detected");
40 - clear_buffer_write_io_error(sbh);
41 - set_buffer_uptodate(sbh);
44 * If the file system is mounted read-only, don't update the
45 * superblock write time. This avoids updating the superblock
46 @@ -4371,7 +4357,23 @@ static int ext4_commit_super(struct super_block *sb, int sync)
47 &EXT4_SB(sb)->s_freeinodes_counter));
48 BUFFER_TRACE(sbh, "marking dirty");
49 ext4_superblock_csum_set(sb);
51 + if (buffer_write_io_error(sbh)) {
53 + * Oh, dear. A previous attempt to write the
54 + * superblock failed. This could happen because the
55 + * USB device was yanked out. Or it could happen to
56 + * be a transient write error and maybe the block will
57 + * be remapped. Nothing we can do but to retry the
58 + * write and hope for the best.
60 + ext4_msg(sb, KERN_ERR, "previous I/O error to "
61 + "superblock detected");
62 + clear_buffer_write_io_error(sbh);
63 + set_buffer_uptodate(sbh);
65 mark_buffer_dirty(sbh);
68 error = __sync_dirty_buffer(sbh,
69 test_opt(sb, BARRIER) ? WRITE_FUA : WRITE_SYNC);