Description of problem:
We have an nfs server exporting 2 filesystems, which should not be interrupted 
(hence we cannot simply restart the nfs service).

I have configured further ext3 filesystems to export, for the moment read only, 
and exportfs presents these to clients as expected.  However, after mounting an 
exported filesystem from a client, then unmounting it, exportfs -u does not 
completely remove export entries.  References are removed from:-
/var/lib/nfs/rmtab
/var/lib/nfs/etab

but do not go from:-
/var/lib/nfs/xtab
/proc/fs/nfs/exports

Because the kernel entry is still there (is that because of the xtab entries 
still there, or causing the xtab entries to still be there?), we are unable to 
umount it.  Restarting nfs does fix the problem, but doing so interferes with 
an important application.

Version-Release number of selected component (if applicable):
nfs-utils-1.0.6-42EL with kernel-smp-2.4.21-37.EL

How reproducible:
Every time.

Steps to Reproduce:
1. (server): exportfs -o mp,ro,no_root_squash \*:/stage/test
2. (client): mount server:/stage/test /mnt/test ; umount /mnt/test
3. (server): exportfs -u \*:/stage/test
4. umount /stage/test
  
Actual results:
# fuser -v /stage/test
                     USER        PID ACCESS COMMAND
/stage/test
                     root     kernel mount  /stage/test

# umount /stage/test
umount: /stage/test: device is busy

This remains the case until nfs is restarted on the server.


Expected results:
Filesystem should have umounted properly.  I would also expect xtab references 
for the unexported share to disappear.


Additional info:
If the filesystem is simply exported and then unexported, without being mounted 
at all then there is no problem.  But once the entries appear 
in /var/lib/nfs/xtab and /proc/fs/nfs/exports then those entries will not go 
until nfs is restarted on the server.