Re: Forced XFS shutdown

[Eric Sandeen <sandeen@sgi.com>]

On Thu, 21 Mar 2002, Toralf Lund wrote:

> One of our customers recently had a problem with an XFS file system on a
> Linux server installation - the file system was simply missing all of a
> sudden. Inspection of the system log revealed the following messages:
>
> Mar 19 09:41:01 zagreb kernel: xfs_force_shutdown(sd(8,16),0x8) called
> from line 4079 of file xfs_bmap.c.  Return address = 0xc0190be2

This means some sort of i/o error has occured.  Bad cable, bad disk,
in-memory corruption problems, it's hard to tell.  :(

> When someone tried to reboot the machine in an attempt to re-mount the
> file system, the following message occurred:
>
> Mar 19 10:00:58 zagreb kernel: XFS mounting filesystem sd(8,16)
> Mar 19 10:00:58 zagreb kernel: Starting XFS recovery on filesystem:
> sd(8,16) (dev: 8/16)
> Mar 19 10:00:58 zagreb kernel: XFS: xlog_recover_process_data: bad clientid
> Mar 19 10:00:58 zagreb kernel: XFS: log mount/recovery failed
> Mar 19 10:00:58 zagreb kernel: XFS: log mount failed
>
> Any idea what happened here? Is there any reason to suspect that the file
> system has become unstable so that the problem could reappear, and if so,
> is there any way of rectifying it?

There was a bug in some xfs versions that would overwrite the first part
of the disk after a forced shutdown, which is why it would not mount
afterwards.  This bug is fixed in recent CVS kernels and will also be
fixed in some upcoming prerelease kernels, due out soon...

So the real problem is the forced shutdown; the subsequent mount problem
is due to a bug that's been fixed.  So the shutdown could happen again,
but we can avoid the mount problem now.

-Eric