xfs filesystem corruption under -ck

[Felix X]

Hi,

I'm not sure if this is actually a -ck problem, but it does only seem
to happen under -ck (and it's reproducible). Running xfs_fsr (the
defrag utility for XFS) while there are files open on the filesystem
(sometimes?) results in some files getting null-byte areas in them.
(Yes, xfs_fsr is meant to be run on mounted filesystems.)

The full bug report I've filed for the distribution I'm using is here:

http://bugs.archlinux.org/task/7778

I was running the ck patched kernel for a few months on my old
machine, and never ran xfs_fsr there and never had any problems. Now
that I've switched back to the standard 2.6.22 kernel (as a
precaution), I am noticing that performance was better under ck and I
miss it.

Quantifiable differences:

- the Flash plugin freezes under Opera and needs to be killed a lot
more than it did formerly (I think Adobe needs programming help).

- file I/O -- my disks were about 20% faster on large file copies under ck.

- mount/unmounting of filesystems was more reliable, especially when
mounting encrypted luks volumes. occasionally, trying to mount a luks
volume under the standard kernel would cause the mount command to lock
up and refuse to die even with kill -9. This has not happened once
under -ck. The same is true for NTFS mounts (with the read-only
driver, not ntfs-3g) from USB drives.

~Felix.
_______________________________________________
http://ck.kolivas.org/faqs/replying-to-mailing-list.txt
ck mailing list - mailto: ck@...
http://vds.kolivas.org/mailman/listinfo/ck

[Frederik-2]

On 8/12/07, Felix X <01001101@...> wrote:
> Hi,
>
> I'm not sure if this is actually a -ck problem, but it does only seem
> to happen under -ck (and it's reproducible). Running xfs_fsr (the
> defrag utility for XFS) while there are files open on the filesystem
> (sometimes?) results in some files getting null-byte areas in them.
> (Yes, xfs_fsr is meant to be run on mounted filesystems.)

I think I've had the same problem about one month ago. I also had
corrupted files, also probably after I ran xfs_fsr. I attributed it to
a problem in the nvidia sata driver because it was hanging my system
at bootup for some time and giving exceptions at boot. But maybe this
was unrelated, because I discovered later on these exceptions were
caused/triggered by smartmontools.

Anyway, I've currently disabled smartmontools and I am running 2.6.22
with CFS and swap prefetch, and haven't seen this ever since. I guess
now would be the right time to start switching over to a non-ck based
kernel, as ck is unmaintained anyway unfortunately...

--
Frederik
_______________________________________________
http://ck.kolivas.org/faqs/replying-to-mailing-list.txt
ck mailing list - mailto: ck@...
http://vds.kolivas.org/mailman/listinfo/ck

[Martin Steigerwald]

Am Sonntag 12 August 2007 schrieb Felix X:

> Hi,
>
> I'm not sure if this is actually a -ck problem, but it does only seem
> to happen under -ck (and it's reproducible). Running xfs_fsr (the
> defrag utility for XFS) while there are files open on the filesystem
> (sometimes?) results in some files getting null-byte areas in them.
> (Yes, xfs_fsr is meant to be run on mounted filesystems.)
>
> The full bug report I've filed for the distribution I'm using is here:
>
> http://bugs.archlinux.org/task/7778

Hello Felix!

Good to hear about this. I faced something similar:

http://oss.sgi.com/archives/xfs/2007-07/msg00038.html

(this posts includes information on how I have fixed most corrupted files
from the backup I had without overwriting newer files using rsync with
and without checksumming ;-)

I did not come about that it may have been related to ck.  I did not know
exactly anymore which kernel I was using back then whether it was still
the ck2 one or the CFS one, and I didn't think that this may be relevant.
Seems it is and seems I have been using ck patchset, cause

I even went as far as running XFSQA test, but then with a kernel using
only CFS scheduler instead of ck2 patchset. Well and it did work so I was
not able to reproduce the problem:

http://oss.sgi.com/archives/xfs/2007-07/msg00320.html

Could to see that this problem actually is for really. I am still quite
hesitant to try xfs_fsr again in real world usage, since that kind of
file corruption was quite difficult to detect and fix (well once I knew
about rsync -c it was easy, but first I had to notice that corruption at
all).

I will add a link to your post and your bug report to my thread in xfs
mailinglist to let people there know.

If you happen to find anyhting new I would like to know about it.

> I was running the ck patched kernel for a few months on my old
> machine, and never ran xfs_fsr there and never had any problems. Now
> that I've switched back to the standard 2.6.22 kernel (as a
> precaution), I am noticing that performance was better under ck and I
> miss it.

I am quite happy with CFS. Although I do not have all the other ck
goodies. Since my laptops have way enough RAM I am not dependant on swap
prefetech... so I think I will keep it with just CFS for now.

Regards,
--
Martin 'Helios' Steigerwald - http://www.Lichtvoll.de
GPG: 03B0 0D6C 0040 0710 4AFA  B82F 991B EAAC A599 84C7


_______________________________________________
http://ck.kolivas.org/faqs/replying-to-mailing-list.txt
ck mailing list - mailto: ck@...
http://vds.kolivas.org/mailman/listinfo/ck