I've been struggling with this for a couple of days, so I wanted to
share my solution...
* The story:
If one mounts a NFS directory on a Solaris box, then Solaris just
_assumes_ that there's an rpc.lockd sitting on the other side.
Linux 2.0 NFS implementation doesn't come with the NLM protocol
(lockd) (2.1 has it). Sooo, whenever a Solaris program locks a portion
of a file on a Linux-exported NFS share, the program hangs waiting for
an answer from a lockd on the Linux side which isn't there.
Just for the pleasure of ranting against Sun, this assumes you can get
NFS to work reliably between Linux and Solaris. Sun doesn't seem to be
able to (or want to) have a proper NFS implementation.
Hint: when mounting Linux on Solaris, use -o proto=udp,vers=2.
when mounting Solaris on Linux, use -o udp,nfsvers=2,noac.
* The solutions:
1) Use kernel 2.1.
2) Don't do this.
3) Kill lockd on Solaris (then you loose Solaris-Solaris locking)
4) Complain to Sun (good luck)
5) Use an old (and very buggy) server-side only lockd (from
ftp://ftp.mathematik.th-darmstadt.de/pub/linux/okir/dontuse/OLD/lockd-0.4a.tar.gz).
6) Have a dummy rpc.lockd on Linux.
* The software:
Well, on ftp.fifi.org:/pub/phil/rpc.fakelockd.tgz you'll find a dummy
lockd daemon that will basically grant any lock on any file.
It honors all _synchronous_ NLM requests.
It declines all asynchronous requests (and logs them crudely to
syslog).
It doesn't support NLM v3 requests (and logs them too).
It doesn't come with a dummy SM (statd) daemon either (but I don't
think it's necessary).
That should save you some hair if you have to work with Solaris.
Phil.
