On Dec 5, 2011, at 1:12 PM, Jan Stary wrote: >> On Dec 5, 2011, at 11:20 AM, Jan Stary wrote: >> >>> On Dec 05 08:28:15, Michael Durket wrote: >>>> I'm running an old copy of OpenBSD (4.1) on a production server and it >>>> mounts a single NFS volume from another (Solaris) machine using NFS 3 tcp. >>>> This mount worked initially, but then for a short period of time the remote >>>> server had a problem, and I would see messages indicating that the server was >>>> not responding (and an nfsstat command would indicate increasing values in the >>>> TimedOut section of the Rpc Info section of the output). I can do a showmount >>>> to the remote NFS server and it works fine (and shows that my host has mounted >>>> the volume on the remote server), but all attempts to access the volume fail >>>> as do any unmount attempts. I've asked a few OpenBSD experts to take a look >>>> but they can't seem to find what the problem is - it's almost as if once >>>> OpenBSD reports server timeouts that it never forgets (like setting a latch >>>> that I can't reset). I'd like to unmount this volume (and just forget about >>>> using NFS on OpenBSD) but I can't think of way to do it without taking down >>>> the production server for a reboot. >>> >>> Even 'umount -f' doesn't unmount it? >>> > > On Dec 05 13:06:14, Michael Durket wrote: >> Yes - that fails as well. > > What kind of problem was there at the server?
I don't remember at this point - it was a long time ago - our Solaris NFS servers are very flaky. > (Could the connections to that server die?) > > Is the initial tcp connection still alive? No. > (pfctl -ss) > > Can you provide a full tcpdump of an umount attempt? No - because a umount generates no network traffic. That's why I said that I thought it was like some kind of latch - no traffic gets generated and OpenBSD immediately responds to any mount / umount command with "nfs server not responding". I think the problem sounds more like what an earlier responder suggested - that this was a problem at some point with TCP NFS in OpenBSD that has since been fixed, so I either need to upgrade, or just give up on using TCP NFS. > (From both ends preferably?)