Skip site navigation (1)Skip section navigation (2)
Date:      Thu, 26 May 2005 12:58:06 +0200
From:      Bernd Walter <ticso@cicely12.cicely.de>
To:        Sten Spans <sten@blinkenlights.nl>
Cc:        current@freebsd.org, Mohan Srinivasan <mohan_srinivasan@yahoo.com>, stable@freebsd.org, ticso@cicely.de, Oliver Lehmann <lehmann@ans-netz.de>
Subject:   Re: problems with nfs+TCP - Resource temporarily unavailable
Message-ID:  <20050526105806.GB80082@cicely12.cicely.de>
In-Reply-To: <Pine.SOC.4.61.0505261243440.18365@tea.blinkenlights.nl>
References:  <20050525223355.56551.qmail@web80601.mail.yahoo.com> <20050526010325.02415410.lehmann@ans-netz.de> <20050526085750.GZ80082@cicely12.cicely.de> <Pine.SOC.4.61.0505261243440.18365@tea.blinkenlights.nl>

next in thread | previous in thread | raw e-mail | index | archive | help
On Thu, May 26, 2005 at 12:45:50PM +0200, Sten Spans wrote:
> On Thu, 26 May 2005, Bernd Walter wrote:
> 
> >On Thu, May 26, 2005 at 01:03:25AM +0200, Oliver Lehmann wrote:
> >>######
> >>
> >>
> >>I tried the same with an other nfs server (using dill as nfs server this
> >>time - system description is in my 1st mail, same mount options like /
> >>mnt/files). And guess what? dill rebooted immediate... dd came never
> >>back, gave no output
> >>
> >>dill's dmesg shows me:
> >>
> >>fatal kernel trap:
> >>
> >>    trap entry     = 0x4 (unaligned access fault)
> >>    faulting va    = 0xfffffc0006b6f44d
> >>    opcode         = 0x28
> >>    register       = 0x5
> >>    pc             = 0xfffffc0000541e08
> >>    ra             = 0xfffffc0000541df4
> >>    sp             = 0xfffffe000a0f9b70
> >>    usp            = 0x11ffea80
> >>    curthread      = 0xfffffc000f91ee10
> >>        pid = 343, comm = nfsd
> >
> >This is absolutely known - TCP/nfs has bugs in realigning packets.
> >Don't use TCP on strong aligned architectures.
> 
> Still a pr with a proper backtrace would be nice.
> Or does one exist already ?

Not that I know.
I did know exactly when this happens years ago.
The backtrace as such will not help you as the panic happens much
later than the cause.
IIRC the basic problem was that the realignment code only fixes
a single missalignment, while theres a chance for more then one.
Verify nfs_realign in nfsserver and nfsclient to get an idea.
If you are interested - I've found a (non-working) patch that I wrote
for it, but the intention of it should be clear.

-- 
B.Walter                   BWCT                http://www.bwct.de
bernd@bwct.de                                  info@bwct.de




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20050526105806.GB80082>