From: Theodore Ts'o tytso@mit.edu
[ Upstream commit b5776e7524afbd4569978ff790864755c438bba7 ]
In the case where we need to do an interior node split, and immediately afterwards, we are unable to allocate a new directory leaf block due to ENOSPC, the directory index checksum's will not be filled in correctly (and indeed, will not be correctly journalled).
This looks like a bug that was introduced when we added largedir support. The original code doesn't make any sense (and should have been caught in code review), but it was hidden because most of the time, the index node checksum will be set by do_split(). But if do_split bails out due to ENOSPC, then ext4_handle_dirty_dx_node() won't get called, and so the directory index checksum field will not get set, leading to:
EXT4-fs error (device sdb): dx_probe:858: inode #6635543: block 4022: comm nfsd: Directory index failed checksum
Google-Bug-Id: 176345532 Fixes: e08ac99fa2a2 ("ext4: add largedir feature") Cc: Artem Blagodarenko artem.blagodarenko@gmail.com Signed-off-by: Theodore Ts'o tytso@mit.edu Signed-off-by: Sasha Levin sashal@kernel.org --- fs/ext4/namei.c | 7 +++---- 1 file changed, 3 insertions(+), 4 deletions(-)
diff --git a/fs/ext4/namei.c b/fs/ext4/namei.c index f05ec9bfbf4fd..7f22487d502b5 100644 --- a/fs/ext4/namei.c +++ b/fs/ext4/namei.c @@ -2405,11 +2405,10 @@ again: (frame - 1)->bh); if (err) goto journal_error; - if (restart) { - err = ext4_handle_dirty_dx_node(handle, dir, - frame->bh); + err = ext4_handle_dirty_dx_node(handle, dir, + frame->bh); + if (err) goto journal_error; - } } else { struct dx_root *dxroot; memcpy((char *) entries2, (char *) entries,