Re: [PATCH 2/2] jbd2: flush filesystem device before updating tail sequence

[Kemeng Shi]

on 12/3/2024 3:24 PM, Zhang Yi wrote:
> On 2024/12/3 14:53, Kemeng Shi wrote:
>>
>>
>> on 12/3/2024 9:44 AM, Zhang Yi wrote:
>>> From: Zhang Yi <yi.zhang@xxxxxxxxxx>
>>>
>>> When committing transaction in jbd2_journal_commit_transaction(), the
>>> disk caches for the filesystem device should be flushed before updating
>>> the journal tail sequence. However, this step is missed if the journal
>>> is not located on the filesystem device. As a result, the filesystem may
>>> become inconsistent following a power failure or system crash. Fix it by
>>> ensuring that the filesystem device is flushed appropriately.
>>>
>>> Fixes: 3339578f0578 ("jbd2: cleanup journal tail after transaction commit")
>>> Signed-off-by: Zhang Yi <yi.zhang@xxxxxxxxxx>
>>> ---
>>>  fs/jbd2/commit.c | 4 ++--
>>>  1 file changed, 2 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/fs/jbd2/commit.c b/fs/jbd2/commit.c
>>> index 4305a1ac808a..f95cf272a1b5 100644
>>> --- a/fs/jbd2/commit.c
>>> +++ b/fs/jbd2/commit.c
>>> @@ -776,9 +776,9 @@ void jbd2_journal_commit_transaction(journal_t *journal)
>>>  	/*
>>>  	 * If the journal is not located on the file system device,
>>>  	 * then we must flush the file system device before we issue
>>> -	 * the commit record
>>> +	 * the commit record and update the journal tail sequence.
>>>  	 */
>>> -	if (commit_transaction->t_need_data_flush &&
>>> +	if ((commit_transaction->t_need_data_flush || update_tail) &&
>>>  	    (journal->j_fs_dev != journal->j_dev) &&
>>>  	    (journal->j_flags & JBD2_BARRIER))
>>>  		blkdev_issue_flush(journal->j_fs_dev);
>>>
>> In journal_submit_commit_record(), we will submit commit block with REQ_PREFLUSH
>> which is supposed to ensure disk cache is flushed before writing commit block.
>> So I think the current code is fine.
>> Please correct me if I miss anything.
>>
> 
> The commit I/O with REQ_PREFLUSH only flushes 'journal->j_dev', not
> 'journal->j_fs_dev'. We need to flush journal->j_fs_dev to ensure that all
> written metadata has been persisted to the filesystem disk, Until then, we
> cannot update the tail sequence.
My bad...
Look good to me. Feel free to add:

Reviewed-by: Kemeng Shi <shikemeng@xxxxxxxxxxxxxxx>
> 
> Thanks,
> Yi.
> 
>