Skip site navigation (1)Skip section navigation (2)
Date:      Sat, 23 Jan 1999 09:33:44 +1030
From:      Greg Lehey <grog@lemis.com>
To:        Scott Mitchell <scott@dcs.qmw.ac.uk>
Cc:        freebsd-questions@FreeBSD.ORG
Subject:   Re: NFS mount failures -- HELP!
Message-ID:  <19990123093344.G19921@freebie.lemis.com>
In-Reply-To: <19990113135237.C993@dcs.qmw.ac.uk>; from Scott Mitchell on Wed, Jan 13, 1999 at 01:52:37PM %2B0000
References:  <19990113135237.C993@dcs.qmw.ac.uk>

next in thread | previous in thread | raw e-mail | index | archive | help
On Wednesday, 13 January 1999 at 13:52:37 +0000, Scott Mitchell wrote:
> OK, all you gurus, this one has everybody around here stumped:
>
> The story so far -- one small FreeBSD server, running 2.2.7R and configured
> much like everything else around here (NIS, everything mounted by amd using
> some *very* hairy maps).  Working just great until last Friday, when it
> suddently started refusing to mount anything from one particular file
> server.  The timing co-incides with the installation of a new switch with
> something of a nanny syndrome; it likes to throw away over-long UDP packets
> and those with bad checksums.
>
> In general our amd maps mount things using UDP with (I think) a 8K block
> size.  I now can't mount anything from the offending server over UDP, even
> wfter reducing the block size to 1K.  TCP mounts work fine, but I can't
> force those to be used without messing with the (shared) amd maps or
> hacking on amd itself.
>
> The problem appears to be limited to servers that route through the new
> switch to get to my client, but I'm not 100% sure on that point.  Other
> machines (Linux, IRIX) on the same network as the client have no problems.
> I haven't changed any configuration on the client for several months, and
> it was working perfectly until this started.
>
> The obvious culprit is the switch, but that doesn't explain why only this
> pair of machines is affected (the server runs FreeBSD also, BTW).  I've
> attached the packet trace result from one such failed mount attempt (chorus
> is the client, hotpoint the server)
>
> Nobody here can figure out exactly what's causing this, so any suggestions
> will be most welcome.
>
> chorus# tcpdump host hotpoint and host chorus
> tcpdump: listening on ed0
> 11:18:44.104319 chorus.9dd92009 > hotpoint.dcs.qmw.ac.uk.nfs: 92 getattr [|nfs]
> 11:18:44.105054 hotpoint.dcs.qmw.ac.uk.nfs > chorus.9dd92009: reply ok 96
> 11:18:44.105595 chorus > hotpoint.dcs.qmw.ac.uk: icmp: chorus udp port 1035 unreachable

This looks very much like the problem I reported in kern/9612.  I
suspect you would have seen the problem if you had used the -n option
to tcpdump.  The important detail is that hotpoint has two IP
addresses:

Name:    hotpoint.dcs.qmw.ac.uk
Addresses:  138.37.94.24, 138.37.88.162

I'd guess that the request from chorus to hotpoint went to one
address, and the reply came from the other address.  I haven't fixed
the problem yes, but for the time being, you can work around the
problem by using the address 138.37.88.162 instead of the name
hotpoint.

Greg
--
See complete headers for address, home page and phone numbers
finger grog@lemis.com for PGP public key

To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-questions" in the body of the message



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?19990123093344.G19921>