Re: [PATCH -next] jbd2: discard last transaction when commit block checksum broken in v2v3
From: Jan Kara <jack@suse.cz>
Date: 2021-11-22 18:22:11
Also in:
lkml
On Wed 29-09-21 11:55:28, Ye Bin wrote:
Now, we meet an issue that commit block has broken checksum when cold reboot
device, that lead to mount failed.
The reason maybe only some sector store on disk, and then device power off.
But we calculate checksum with whole logic block.The data stored on disk can
only ensure the atomicity of sector level.
Actually, we already replay previous transactions. We can just discard last
transaction. As now, descriptor/revocation/commit/superblock has it's own
checksum.
Fixes:80b3767fbe15("jbd2: don't wipe the journal on a failed journal checksum")
Signed-off-by: Ye Bin <redacted>Thanks for the patch. It seems to have fallen through the cracks. Sorry for that.
quoted hunk ↗ jump to hunk
--- fs/jbd2/journal.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)diff --git a/fs/jbd2/journal.c b/fs/jbd2/journal.c index 35302bc192eb..a3dd7b757b3d 100644 --- a/fs/jbd2/journal.c +++ b/fs/jbd2/journal.c@@ -2080,7 +2080,7 @@ int jbd2_journal_load(journal_t *journal) if (jbd2_journal_recover(journal)) goto recovery_error; - if (journal->j_failed_commit) { + if (journal->j_failed_commit && !jbd2_journal_has_csum_v2or3(journal)) {
I guess this decision somewhat questionable. If the failed commit was indeed the last one, I guess loosing the last transaction as you suggest is a sensible thing to do. However if the checksum failed somewhere in the middle of the journal because of a bitflip or something like that, we probably don't want to loose that many transactions and rather want to do fsck and try to recover as much data as possible... What do others think? Honza
printk(KERN_ERR "JBD2: journal transaction %u on %s " "is corrupt.\n", journal->j_failed_commit, journal->j_devname); -- 2.31.1
-- Jan Kara [off-list ref] SUSE Labs, CR