xfs
[Top] [All Lists]

Re: kernel panic with nfsd and xfs filesystem

To: Vladimir Vukicevic <vladimir@xxxxxxxxxx>
Subject: Re: kernel panic with nfsd and xfs filesystem
From: Andi Kleen <ak@xxxxxxx>
Date: Fri, 9 Mar 2001 21:37:37 +0100
Cc: Steve Lord <lord@xxxxxxx>, linux-xfs@xxxxxxxxxxx
In-reply-to: <E14bQXC-0007hT-00@xxxxxxxxxxxxxxx>; from vladimir@xxxxxxxxxx on Fri, Mar 09, 2001 at 12:18:34PM -0500
References: <E14bQ1Y-0007cM-00@xxxxxxxxxxxxxxx> <E14bQXC-0007hT-00@xxxxxxxxxxxxxxx>
Sender: owner-linux-xfs@xxxxxxxxxxx
User-agent: Mutt/1.2.5i
On Fri, Mar 09, 2001 at 12:18:34PM -0500, Vladimir Vukicevic wrote:
> On 09 Mar 2001 11:45:51 -0500, Vladimir Vukicevic wrote:
> > 
> > 
> > Hmm.  I now have a repeatable case of this I/O error, but as the other
> > end is running the kernel nfsd server, I'm not exactly sure how to debug
> > it or turn on debugging on the other end... A 'cat foo.ogg /dev/null' on
> > the mounted partition repeatably gives an I/O error.
> > 
> > Any thoughts on how to diagnose this?  I'll keep poking..
> 
> 
> Doh. Forgot about tcpdump. :-)  So, this is what I'm seeing:
> 
> 11:57:46.159530 rain.ximian.priv.4040915765 > ogg.nfs: 140 lookup fh
> Unknown/1 "01-letters_from_the_wasteland.ogg" (DF)
> 11:57:46.163988 ogg.nfs > rain.ximian.priv.4040915765: reply ok 128
> lookup fh Unknown/1 (DF)
> 
> So, the lookup goes okay. Then the weirdness starts.

You're probably overflowing some of the too small default buffers and 
therefore get packet loss. You can check using netstat -s. If you see
IpReasmFails increase /proc/sys/net/ipv4/ipfrag_high_thresh
If you see UdpInErrs increase the socket buffer  (net/core/rmem_{default,max})
This will probably need a reload of the nfs modules to take effect.
If you see IpInErrs or device errors then you're out of luck in 2.4 currently.

Note that when you overflow the socket buffer you'll likely also not see
it in ethertap, because it tends to use the same too small socket buffer sizes
on its packet sniffing socket. The single fragments you saw probably point
to this.


-Andi


<Prev in Thread] Current Thread [Next in Thread>