[OpenAFS] OpenAFS 1.4.7 (namei, libafs64.o) + Solaris 10 x86 u5 + NFS write = crash & reboot

Derrick Brashear shadow@gmail.com
Wed, 13 Aug 2008 19:37:49 -0400


On Wed, Aug 13, 2008 at 5:56 PM, Lee Damon <nomad@crow.ee.washington.edu> wrote:
> I've been working on a Sun x4500 (thumper) running Solaris 10 x86 (u5) and
> OpenAFS 1.4.7.  I was able to make it crash & reboot reliably with the
> simple expedient of
>
> 1. NFS exporting any filesystem (UFS or ZFS) from the x4500.
>
> 2. NFS (v3) mounting the filesystem to a RHEL 4 (u5 or u6, doesn't matter)
> box.  Note that the problem doesn't happen under NFSv4.
>
> 3. On the RHEL box touch (or cp, edit, rm, or any other write) a file on the
> NFS mounted filesystem.  Boom, instant panic on the x4500 NFS server.
>
> Sun finally got around to looking at the core files (Sun's report is
> attached).  After reading their report I rebooted the x4500 with AFS turned
> off and sure enough, the panic went away.
>
> I then replaced /kernel/fs/afs with libafs64.nonfs.o (on the x4500) and
> rebooted.  No crash upon NFS write attempt.
>
> This problem report brought to you by the letter "hoping to help others
> avoid the same problem" and the number "NFS still sucks!"

There's an open ticket in RT; The problem wasn't readily apparent when
I looked and I haven't had time to go back to it. Sorry.

Derrick