Bug#389232: linux-image-2.6-amd64: mounting xfs filesystem causes kernel oops
Dne Sun, Sep 24, 2006 at 09:23:15PM +0200, Oleg Verych napsal:
> Let's add XFS team in CC.
There isn't
> On 2006-09-24, Dan Ohnesorg <bugs@ohnesorg.cz> wrote:
> > Package: linux-image-2.6-amd64
> > Version: 2.6.17+2
> > Severity: critical
> > Justification: breaks the whole system
>
> More details, please.
>
> What happend before this boot?
>
> Why ext3 is recovering?
The system was installed as 32 bit system, fresh install of debian etch,
some week ago. (I remember, that beta3 of installer was only one day old).
Runned good, but I have decided to migrate to 64 bit. All drives except xfs
one are reformatted, only xfs has been kept untouched. Even if the xfs was
probably clean unmounted, installer had some problem with mounting xfs into
/target. So I deciced not to mount xfs automatically. After first boot,
while mounting xfs manually the kernel oopsed and was not able to shutdown
correctly, so the next boot was unclean and the ext3 (/) has to be
recovered. I have made the bugreport after second unsecssful boot.
The instalation was made by nightly snapshoot of installer from 24-Sep-2006
02:33. I didn't use beta3, becouse beta 3 is missing xfs libs.
> > raid1: raid set md1 active with 4 out of 4 mirrors
> > EXT3-fs: INFO: recovery required on readonly filesystem.
> > EXT3-fs: write access will be enabled during recovery.
> > kjournald starting. Commit interval 5 seconds
>
> And why XFS does ?
Don't know exactly, as written above.
> In upstream there are very hidden bugs with XFS, which are very hard
> to find, because they are rare and are showed on big machines:
> huge RAM, many CPUs. Maybe you can help.
>
> (s)ata + xfs bug: <http://permalink.gmane.org/gmane.linux.kernel/446612>
> devmaper + xfs bug: <http://permalink.gmane.org/gmane.linux.kernel/444139>
The problem was solved by using my own kernel, 2.6.18.
gmame is currently down:
"""
Couldn't contact deer:8010
The web interface is down for maintenance.
"""
This machine has 2GB RAM and 1,1TB xfs filesystem, it will be bacula sd
machine. Main problem I see here is, that testing is using in freeze phase
kernel, which oopsed instead of just returning some error message from xfs
layer. It can occur, that my xfs filesystem was broken in some way, but xfs
should say, bye bye to my data and no oops whole kernel.
cheers
dan
Reply to: