On Fri, 12 Aug 2011, Curt Wohlgemuth wrote:
> I don't know much about data=journal, but I've been running xfstests
> with it, and it's a disaster, given that data=journal doesn't support
> O_DIRECT. What kind of testing do people do on data=journal?
Short answer is probably none :). Even though that it seems like an
radical answer I believe that it is mostly true, because simply said
almost no-one care. But I think that Ted mentioned that he actually do
some tests with that mode, but I am not sure about that.
>
> And worse, I consistently crash my machine running xfstests 074 on a
> data=journal partition. I just repeated this to make sure, on
> 3.1.0-rc1; I've also seen it with 3.0.0. There's a BUG_ON firing in
> jbd2_journal_dirty_metadata():
>
> [ 2174.697692] ------------[ cut here ]------------
> [ 2174.698627] kernel BUG at fs/jbd2/transaction.c:1112!
> [ 2174.698627] invalid opcode: 0000 [#1] SMP
> [ 2174.698627] CPU 11
> ...
> [ 2174.698627] Call Trace:
> [ 2174.698627] [<ffffffff8127f9a3>] __ext4_handle_dirty_metadata+0x83/0x120
> [ 2174.698627] [<ffffffff8127fd1e>] ? __ext4_journal_get_write_access+0x3e/0x80
> [ 2174.698627] [<ffffffff81253a78>] __ext4_journalled_writepage+0x338/0x3e0
> [ 2174.698627] [<ffffffff81254244>] ext4_writepage+0x234/0x2f0
> [ 2174.698627] [<ffffffff81158327>] __writepage+0x17/0x40
> [ 2174.698627] [<ffffffff811597ae>] write_cache_pages+0x1fe/0x650
>
> This is at the J_ASSERT_JH below:
>
> /*
> * Metadata already on the current transaction list doesn't
> * need to be filed. Metadata on another transaction's list must
> * be committing, and will be refiled once the commit completes:
> * leave it alone for now.
> */
> if (jh->b_transaction != transaction) {
> JBUFFER_TRACE(jh, "already on other transaction");
> J_ASSERT_JH(jh, jh->b_transaction ==
> journal->j_committing_transaction); <===============
> J_ASSERT_JH(jh, jh->b_next_transaction == transaction);
> /* And this case is illegal: we can't reuse another
> * transaction's data buffer, ever. */
> goto out_unlock_bh;
> }
Wow, that backtrace seems like a flash-back to me I believe that I have
seen it several times, probably in different modes and probably already
fixed. Do no know about this one though.
Thanks!
-Lukas
>
> Curt
>
> On Fri, Aug 12, 2011 at 1:25 AM, Ric Wheeler <rwheeler@xxxxxxxxxx> wrote:
> > On 08/12/2011 09:16 AM, Lukas Czerner wrote:
> >>
> >> On Thu, 11 Aug 2011, Andreas Dilger wrote:
> >>
> >>> On 2011-08-11, at 9:01 AM, Lukas Czerner wrote:
> >>>>
> >>>> On Tue, 28 Jun 2011, Lukas Czerner wrote:
> >>>>>
> >>>>> Data journalling mode (data=journal) is known to be neglected by
> >>>>> developers and only minority of people is actually using it. This
> >>>>> mode is also less tested than the other two modes by the developers.
> >>>>>
> >>>>> This creates a dangerous combination, because the option which seems
> >>>>> *safer* is actually less safe the others. So this commit adds a warning
> >>>>> message in case that data=journal mode is used, so the user is informed
> >>>>> that the mode might be removed in the future.
> >>>>
> >>>> Any comments on this ? Is that feasible to remove is someday ?
> >>>
> >>> I'm less in favour of removing data=journal. Jan made some fixes to
> >>> data=journal mode in the last few weeks, which means that people are
> >>> still using this.
> >>
> >> I think that Jan was actually the one who was in favour of this change
> >> IIRC. But you're right that there are still some (very little possibly?)
> >> users of this. But this change does not remove it, but just let the
> >> users know that it might be removed someday, hence discouraging them from
> >> using it.
> >>
> >> Also we were discussing that several times, so I think that letting
> >> users know that we are considering it is a good thing.
> >>
> >> Thanks!
> >> -Lukas
> >
> > I think that this will be very useful to have - users should definitely
> > chime in when they see this warning if they are using data journal mode.
> >
> > The only work load that I know that benefits from a performance point of
> > view is one which involves an fsync() heavy, small file creation workload.
> > Any workload with larger files tends to lose roughly 50% of the write
> > bandwidth under streaming writes since we end up writing everything twice.
> >
> > Regards,
> >
> > Ric
> >
> >
> >>
> >>>>> Signed-off-by: Lukas Czerner<lczerner@xxxxxxxxxx>
> >>>>> ---
> >>>>> fs/ext4/super.c | 5 +++++
> >>>>> 1 files changed, 5 insertions(+), 0 deletions(-)
> >>>>>
> >>>>> diff --git a/fs/ext4/super.c b/fs/ext4/super.c
> >>>>> index 9ea71aa..9d189cf 100644
> >>>>> --- a/fs/ext4/super.c
> >>>>> +++ b/fs/ext4/super.c
> >>>>> @@ -1631,6 +1631,11 @@ static int parse_options(char *options, struct
> >>>>> super_block *sb,
> >>>>> sbi->s_min_batch_time = option;
> >>>>> break;
> >>>>> case Opt_data_journal:
> >>>>> + ext4_msg(sb, KERN_WARNING,
> >>>>> + "Using data=journal may be removed in
> >>>>> the "
> >>>>> + "future. Please, contact "
> >>>>> + "linux-ext4@xxxxxxxxxxxxxxx if you are
> >>>>> "
> >>>>> + "using this feature.");
> >>>>> data_opt = EXT4_MOUNT_JOURNAL_DATA;
> >>>>> goto datacheck;
> >>>>> case Opt_data_ordered:
> >>>>>
> >>>> --
> >>>> To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
> >>>> the body of a message to majordomo@xxxxxxxxxxxxxxx
> >>>> More majordomo info at http://vger.kernel.org/majordomo-info.html
> >>>
> >>> Cheers, Andreas
> >>>
> >>>
> >>>
> >>>
> >>>
> >>>
> >
> > --
> > To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
> > the body of a message to majordomo@xxxxxxxxxxxxxxx
> > More majordomo info at http://vger.kernel.org/majordomo-info.html
> >
>
--
