NFS client kernel issue?
Andy Whitcroft
apw at canonical.com
Tue Jun 30 12:46:25 UTC 2009
On Mon, Jun 29, 2009 at 04:29:25PM -0700, Keith Jackson wrote:
> Hi,
> I've run into a strange problem, and I'm not sure where to ask about
> it. It appears to be kernel related, so hopefully someone here can
> shed some light on it for me. I have a over a dozen ubuntu 9 server
> machines that share an NFS filesystem served up on a FreeBSD machine.
> We recently upgraded the FreeBSD box to FreeBSD 7.2. The ubuntu
> machines are running the standard 2.6.28-11-server kernel.
>
> At this point, we started having problems with NFS on all of the
> ubuntu machines. They would come up, mount the NFS shares and then
> 5-10 minutes later NFS would be hung. I ran wireshark on one of them,
> while doing "ls -l ~username". When NFS was working everything looked
> fine. Once it hung, I would see the normal YP MATCH RPC call looking
> for the entry for "username". The NIS server would return the password
> entry for "username" as I'd expect. When things were working, I'd then
> see the standard NFS traffic. When NFS was hosed, the YP MATCH reply
> would be the last net traffic I would see to the NFS server. Nothing
> NFS related would ever happen. An strace would show that the ls was
> hung in a stat64 call.
>
> A reboot of the system would cause things to start working for a
> while, but 5-10 minutes later NFS would hang. I eventually updated to
> a 2.6.30 kernel from: http://kernel.ubuntu.com/~kernel-ppa/mainline/v2.6.30/
> This has fixed the problem.
>
> Any thoughts on what the problem might be, or how I might go about
> debugging it better?
I am unsure whether the YP stuff is done in the kernel, but there do
seem to be a number of reverted patches in the v2.6.28 to v2.6.30
period. As an example:
nfsd: Revert "svcrpc: take advantage of tcp autotuning"
It would be worth testing v2.6.29 as the midpoint to try and split the
difference and help reduce the gap. Also get a bug filed in launchpad
and let us know its number.
-apw
More information about the kernel-team
mailing list