nfs server busy lockup on 5.3

[Simon Truss]

Hi,

I've been experiencing a series of lockups of nfs recently which
normally require a reboot to clear.

Im about to submit a PR so any pointers on what I could check, trace or
dump would be helpful.

linux 2.6.8.1 nfs client copy or move of large files onto server causes
nfsd to lock. Any access from client of the mount causes those processes
to lock, and any access on the server to the directory containing the
destination file causes a lockup of that program.

I noted that systat -vmstat shows 25% disc usage and
75% cpu usage. The lockup usually occurs within a few 10's or 100's
of MBs of the first file. note files are 2-5GB.

A workaround and possible hint at the cause is to disable soft-deps,
that at least has enabled me to run the system in a stable manner.
The system has been stable for 2 weeks now with around 150+GB of
data transfered without problem.

system details:
nfs client is linux, slackware 10.0, 2.6.8.1 kernel with rpc workqueue
and rpc queue lock patches, 
http://client.linux-nfs.org/Linux-2.6.x/2.6.8.1/ (49 and 50 only)

Server FreeBSD 5.3 GENERIC, problem also present with recent cvsup
of 5.3. nfs exports are from a FFS1 file system on pst RAID array.
I also note that this system was running without this problem with 5.1.
The root partition is not on the pst array.

Simon