From: dawid-g7DOKF3l0ejVItvQsEIGlw@public.gmane.org
Subject: [NFS] Strange NFS behaviour
Date: Fri, 30 Nov 2007 09:30:00 +0100 (CET)
Message-ID: <31796.194.237.142.6.1196411400.squirrel@www.active24.pl>
Mime-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
To: linux-nfs@vger.kernel.org
Sender: linux-nfs-owner@vger.kernel.org

Hi!

I have an NFS resource mounted from a dedicated machine NetApp FAS3050
using NFS v3 over TCP.
The NFS client works on SuSe Linux Enterprise Server 10 SP1, kernel ver
2.6.16.46-0.14-smp.

When the NFS is idle for few hours, the NFS resources sometimes become
inaccessible. Every process accessing such resource hangs.

I have sniffed the NFS connection and found an interesting behaviour of
the NFS client/server:

time          src    dest    proto port numbers content

16692.276474  client server  TCP   1023 > 2049  [SYN]
16692.277171  server client  TCP   2049 > 1023  [SYN, ACK]
16692.277189  client server  TCP   1023 > 2049  [ACK]
16692.277197  client server  NFS   1023 > 2049  V3 ACCESS Call, FH:0x30ed5a16
16692.378613  server client  TCP   2049 > 1023  [ACK]
16701.285748  server client  TCP   2049 > 1023  [FIN, ACK]
16701.285764  client server  TCP   1023 > 2049  [FIN, ACK]
16701.285790  client server  TCP   1022 > 2049  [SYN]
16701.286497  server client  TCP   2049 > 1022  [SYN, ACK]
16701.286506  client server  TCP   1022 > 2049  [ACK]
16701.286508  server client  TCP   2049 > 1023  [ACK]
16701.286512  client server  TCP   1022 > 2049  V3 ACCESS Call, FH:0x30ed5a16
16701.287247  server client  TCP   2049 > 1022  [FIN, ACK]
16701.287257  client server  TCP   1022 > 2049  [FIN, ACK]
16701.287271  client server  TCP   1021 > 2049  [SYN]


Since there was no active TCP connection to the NFS server, it has been
established. Then, an NFS ACCESS command was sent to the server. The
server did not answer the NFS query, but initialised the TCP connection
close instead.

After receiving [FIN, ACK] the client reconnected immediately, but this
time on port 1002 (since 1023 is still in use at this point). Again, the
server closed the TCP connection and another attempt was made on a new
port number.

This is not on the log, but the NFS client keeps repeating this on each of
the port numbers in range <1023,664>. When port number 664 is reached, the
NFS server stops answering the TCP SYN packets.

This started to happen ater ugrading linux from SuSe Linux 10.1
2.6.16.21-0.25-smp. Another thing is that there are also VxWorks clients
using that NFS server which work fine.

Do you have any idea of what is the reason for such behaviour?

Thanks in advance.

Regards,
Dawid Pawlata