NFS: Processes marked "in disk"

[Robert Watson]

On Tue, 20 Jan 2004, Namik Dala wrote:

> I've upgraded to 5.2-RELEASE last days. Now NFS seems to be
> broken. There is an NFS server and 2 NFS clients. The /home
> partition is shared. If I start mutt on an NFS client it wants to
> read the default mailbox (over NFS) and hangs:
> 
> $ ps -auxww | fgrep mutt
> namik 71553  0.0  1.0  4221 2312  p7  D+    1:50AM   0:00.75 mutt
> 
> Is this a known problem? And is there a workaround? 

Questions:

(1) Are you using rpc.lockd on the client?  Server?  Are both clinet and
    server running 5.2-R?  Are you using NFS over UDP, over TCP; NFSv2 or
    NFSv3? 

(2) Could you show the output of "ps -axl" for hung processes?  Might also
    be interesting to have you recompile the kernel with "options DDB" 
    and "options BREAK_TO_DEBUGGER".  When a process hangs on the client,
    hit ctrl-alt-escape (or serial break on serial console), then type in
    'trace pid', substituting the pid of the process in question.

There are some known issues involving NFS, most of them involving
rpc.lockd.  There are some other reports of NFS problems, but they've been
hard to track down, as they appear to be very sensitive to the local
configuration. 

Robert N M Watson             FreeBSD Core Team, TrustedBSD Projects
robert at fledge.watson.org      Senior Research Scientist, McAfee Research