On Thu, Dec 07, 2000 at 05:17:20PM +0100, Frank van der Linden wrote:
> On Thu, Dec 07, 2000 at 05:16:26PM +0100, Manuel Bouyer wrote:
> > On NetBSD we switched the default to UDP because of the problem with TCP
> > mounts: if the TCP connection is broken (link down for too long, server
> > reboot, etc) then the client doesn't find the NFS server and needs to be
> > rebooted.
> 
> This problem has been reported for long server downtimes (> a few hours),
> I think. I have never been able to reproduce it. Despite pulling
> cables everywhere, clients always went on their merry way after the
> link was back up.

I've been able to reproduce it (client i386 1.5H, server sparc 1.5_ALPHA2).
I mounted the server on /mnt (-o -T), halted the server and did a
'ls -l /mnt'. After a while the client prints the expected
nfs server horn:/home: not responding
I have a tcpdump running on the interface; the last network packets where:
20:50:55.120308 rochebonne.antioche.eu.org.4270104586 > horn.antioche.eu.org.nfs: 112 getattr fh 7,4/2
20:51:59.120081 rochebonne.antioche.eu.org.4270104586 > horn.antioche.eu.org.nfs: 112 getattr fh 7,4/2

It's now 22:15:38 and no more traffic than this has happened. I can't see
how the client can wakeup when the server is alive again if it doesn't ping
it regulary.
Anyway the server is now running again but the client is still hung.

--
Manuel Bouyer <bouyer@antioche.eu.org>
--