2008-07-14 14:10:58

by Mike Snitzer

[permalink] [raw]
Subject: Re: - jbd-strictly-check-for-write-errors-on-data-buffers.patch removed from -mm tree

gah, I had html enabled... resend.

On Mon, Jul 14, 2008 at 10:08 AM, Mike Snitzer <[email protected]> wrote:
>
>
> On Tue, Jun 10, 2008 at 5:17 AM, Andrew Morton <[email protected]> wrote:
>>
>> On Tue, 10 Jun 2008 17:51:35 +0900 Hidehiro Kawai <[email protected]> wrote:
>>
>> > Hello Andrew,
>> >
>> > [email protected] wrote:
>> >
>> > > The patch titled
>> > > jbd: strictly check for write errors on data buffers
>> > > has been removed from the -mm tree. Its filename was
>> > > jbd-strictly-check-for-write-errors-on-data-buffers.patch
>> > >
>> > > This patch was dropped because I don't think we want to go read-only on file data write errors
>> > >
>> > > The current -mm tree may be found at http://userweb.kernel.org/~akpm/mmotm/
>> > >
>> > > ------------------------------------------------------
>> > > Subject: jbd: strictly check for write errors on data buffers
>> > > From: Hidehiro Kawai <[email protected]>
>> >
>> > This patch series doesn't change the behavior on file data write
>> > errors as I stated before, but we found that the current behavior has
>> > been made accidentally. So yesterday I sent an additional patch(*)
>> > which removes the invocation of journal_abort() and thus stop making
>> > the fs read-only on file data write errors, but it seems to be late
>> > for the -mm release preparation.
>> >
>> > Patch(*) can be found at:
>> > http://marc.info/?l=linux-kernel&m=121300618614453&w=2
>> >
>> > Anyway, as this patch series was dropped from -mm, I'm going to
>> > send a revised version.
>> >
>> > I plan to separate these pathces into three patche set.
>> > The first patch (set) corrects the current behavior in ordered
>> > writes, it means it removes the invocation of journal_abort() on file
>> > data write errors. It is the almost same as the patch(*).
>> > The second patch set fixes error handlings for metadata writes and
>> > checkpointing. It should be applied independently of the first
>> > patch set, and it is the same as PATCH 3/5 to 5/5.
>> > The third patch set makes "abort the journal on file data write errors"
>> > tunable for mission critical users. Of course, this feature depends
>> > on the first patch set.
>> >
>>
>> That sounds like a good plan, thanks.
>
> Hidehiro and Andrew,
>
> The first patch(set) has been in -mm with the following patches:
> jbd-dont-abort-if-flushing-file-data-failed.patch
> jbd-dont-abort-if-flushing-file-data-failed-fix.patch
>
> "PATCH 3/5 to 5/5" haven't made their way into -mm; nor has the tunable "abort the journal on file data write errors". Where do things stand on this work?
>
> Given the potential for corruption and the fact that -mm's series file justifiably has a place-holder comment of "jbd write-error stuff: scary" I'm wondering: how soon will all associated fixes be included in -mm?
>
> regards,
> Mike