fs: sync_filesystem should not discard sync_fs return code
PATCH 1/1 fs: sync_filesystem should not discard sync_fs return code
sync_filesystem is called from several kernel subsystems to write out a superblock. After writing out the inodes, it calls the superblock sync_fs operation, which is an int return operation. That error code is then discarded. It then calls __sync_blockdev and returns an error if that fails.
Most filesystems that have a sync_fs operation do most of the work inside that operation and then return an error if it fails. With those filesystems, the sync_blockdev is effectively a no-op and may very well not return an error even though writeback failed.
Have __sync_filesystem preserve the error from sync_fs and return that if it fails, or the return from __sync_blockdev if it succeeds.
--- fs/sync.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-)
fs/sync.c +6 -2
@@ -29,14 +29,18 @@*/ static int __sync_filesystem (struct super_block *sb,int wait) { +int fs_ret =0 , bd_ret; +if (wait)sync_inodes_sb (sb);else writeback_inodes_sb (sb,WB_REASON_SYNC );if (sb->s_op->sync_fs) - sb->s_op->sync_fs (sb, wait); -return __sync_blockdev (sb->s_bdev, wait); + fs_ret = sb->s_op->sync_fs (sb, wait); + bd_ret =__sync_blockdev (sb->s_bdev, wait); + +return fs_ret ? fs_ret : bd_ret; }/*
Signature · 1 line
2.9.4
Quoted 11 lines · Jeff Layton
sync_filesystemis called from several kernel subsystems to write out a superblock. After writing out the inodes, it calls the superblocksync_fsoperation, which is an int return operation. That error code is then discarded. It then calls__sync_blockdevand returns an error if that fails.Most filesystems that have a
sync_fsoperation do most of the work inside that operation and then return an error if it fails. With those filesystems, thesync_blockdevis effectively a no-op and may very well not return an error even though writeback failed.
Quoted 4 lines
Have
__sync_filesystempreserve the error fromsync_fsand return that if it fails, or the return from__sync_blockdevif it succeeds.Signed-off-by Jeff Layton <[email protected]>
Makes sense. Although this will likely have only a small impact since most of the errors are likely to happen in the inode writeback where they just get ignored. Anyway you can add:
Honza
Quoted 32 lines
--- fs/sync.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-)fs/sync.c +6
-2@@ -29,14 +29,18 @@*/ static int __sync_filesystem (struct super_block *sb,int wait) { +int fs_ret =0 , bd_ret; +if (wait)sync_inodes_sb (sb);else writeback_inodes_sb (sb,WB_REASON_SYNC );if (sb->s_op->sync_fs) - sb->s_op->sync_fs (sb, wait); -return __sync_blockdev (sb->s_bdev, wait); + fs_ret = sb->s_op->sync_fs (sb, wait); + bd_ret =__sync_blockdev (sb->s_bdev, wait); + +return fs_ret ? fs_ret : bd_ret; }/* Signature · 1 line
2.9.4
Signature · 2 lines
Jan Kara <[email protected]> SUSE Labs, CR