Re: [PATCH 3/4] jbd: abort when failed to log metadata buffers (rebased)

[Hidehiro Kawai <hidehiro.kawai.ez@hitachi.com>]

Hi,

Jan Kara wrote:
> 
> On Fri 16-05-08 19:26:57, Hidehiro Kawai wrote:
> 
>>Jan Kara wrote:
>>
>>
>>>On Wed 14-05-08 13:49:51, Hidehiro Kawai wrote:
>>>
>>>
>>>>Subject: [PATCH 3/4] jbd: abort when failed to log metadata buffers
>>>>
>>>>If we failed to write metadata buffers to the journal space and
>>>>succeeded to write the commit record, stale data can be written
>>>>back to the filesystem as metadata in the recovery phase.
>>>>
>>>>To avoid this, when we failed to write out metadata buffers,
>>>>abort the journal before writing the commit record.
>>>>
>>>>Signed-off-by: Hidehiro Kawai <hidehiro.kawai.ez@hitachi.com>
>>>>---
>>>>fs/jbd/commit.c |    3 +++
>>>>1 file changed, 3 insertions(+)
>>>>
>>>>Index: linux-2.6.26-rc2/fs/jbd/commit.c
>>>>===================================================================
>>>>--- linux-2.6.26-rc2.orig/fs/jbd/commit.c
>>>>+++ linux-2.6.26-rc2/fs/jbd/commit.c
>>>>@@ -703,6 +703,9 @@ wait_for_iobuf:
>>>>		__brelse(bh);
>>>>	}
>>>>
>>>>+	if (err)
>>>>+		journal_abort(journal, err);
>>>>+
>>>>	J_ASSERT (commit_transaction->t_shadow_list == NULL);
>>>
>>>  Shouldn't this rather be further just before
>>>journal_write_commit_record()? We should abort also if writing revoke
>>>records etc. failed, shouldn't we?
>>
>>Unlike metadata blocks, each revoke block has a descriptor with the
>>sequence number of the commiting transaction.  If we failed to write
>>a revoke block, there should be an old control block, metadata block,
>>or zero-filled block where we tried to write the revoke block.
>>In the recovery process, this old invalid block is detected by
>>checking its magic number and sequence number, then the transaction
>>is ignored even if we have succeeded to write the commit record.
>>So I think we don't need to check for errors just after writing
>>revoke records.
> 
>   Yes, I agree that not doing such check will not cause data corruption but
> still I think that in case we fail to properly commit a transaction, we
> should detect the error and abort the journal...

I see.  I'll move the aborting point to just before
journal_write_commit_record() in the next version.

Thanks,
-- 
Hidehiro Kawai
Hitachi, Systems Development Laboratory
Linux Technology Center