1

After upgrading the local file server to Ubuntu server 10.04 LTS, I can no longer run mutt or svn. The lock up happens when the apps attempt to access the NFS mounted drives. I did an ltrace and found that mutt enters the D state when calling the fileno() system call.

The server is running the unfs3 userland server. Is this a known issue? There does not seem to be anything on google about it.

UPDATE:

The problem may be on the server itself. The partitions exported are xfs and a simple copy on that partition on the server resulted in the cp command going into the D state. Should the xfs partition be reformatted when upgrading?

Final update:

The UPS on the server failed. It may have been feeding the server bad power. The switch to the kernel NFS server solved the problems with NFs on the client machine. Many thanks.

NFS user
  • 43
  • 1
  • 6

1 Answers1

4

Switch to the in-kernel NFS server, that is known to work. I don't think the userspace NFS server has been maintained for a looong time.

EDIT To answer your updated question, no, you should not need to reformat the XFS partition after an upgrade. However, the fact that local IO operations on the partition hang suggests that the problem is not with the NFS server but rather with a corrupt filesystem or, say, hardware that is about to fail. The original advice to use the in-kernel NFS server instead of the user space server still holds, though, once you get that far.

janneb
  • 3,841
  • 19
  • 22
  • 2
    And, for the sake of everything holy, make sure you use soft mounts. Otherwise things can hang in the kernel if your NFS server goes down. It's much better to crash an application than have it hang in the kernel. – pehrs Nov 12 '10 at 16:07
  • @pehr: Yeah, well, soft mounts are quite convenient, at least until you lose data. ;-) – janneb Nov 15 '10 at 09:58
  • If the options are "fail a file access and possibly lose data" or "hang in kernel, forcing a reboot and lose data" I think I prefer the first one. ;) – pehrs Nov 15 '10 at 12:47
  • @pehrs: That's not the relevant choice, though. In reality, it's "lose data and don't get any indication which writes failed" vs. "wait until the server comes back online, or explicitly kill the app (with intr), or reboot the client" – janneb Nov 15 '10 at 13:33
  • The kernel NFS server works fine, thanks for the advice. – NFS user Nov 15 '10 at 18:14