Date:        Thu, 8 Oct 1998 09:01:14 -0400
    From:        Mason Loring Bliss <mason@acheron.middleboro.ma.us>
    Message-ID:  <19981008090114.B4117@acheron.middleboro.ma.us>

  | PS: I'm using a kernel build four days ago, and a userland built three days
  | ago, today being 1998.10.08.

Oh yes, I should have said which versions I'm using - I have seen this with
1.3 1.3.2 and 1.3H kernels on the clients (though none were compiled using
UVM - all are sparcs).   The server in my case is an alpha running Digital
Unix (aka OSF1), not NetBSD   (its a multi-processor with lots of net
interfaces which does not much more than NFS serving, but unfortunately has
had periods of unreliability recently - I think (guess) caused by Digital's
(Compaq's now I suppose) ADVFS thing).   I may have seen the same effect when
a NetBSD 1.3 server went down and rebooted, but that's comparatively a little
used server (there are 10-15 mount points from the DUnix server, with people's
home directories on them - just one from the NetBSD server, and it just has a
few occasionally referenced system type files)

I doubt that the server version should be relevant though - NFS servers are
supposed to be stateless, and generally are, if a request would have been
answered before the server went down (and for me that seems to work OK,
though I'm not sure that NFS is actually being stressed much from the NetBSD
clients) it ought be answered just fine after the server restarts - if the
server ever gets the request.   But even if the server never came back,
soft mounts should eventually time out, and return errors.

kre