Strange interaction between dump and NFS

[John Levine]

Every day, in my logs I see this on an 8.2-RELEASE server:

Mar 14 03:11:14 leila mountd[937]: can't export / 
Mar 14 03:11:14 leila mountd[937]: bad exports list line /      -maproot

and, indeed, showmout shows that the root isn't exported and NFS
mounts of the exported root filesystem stop working.  But if I do a
kill -HUP of mountd, the export comes back and resumes working.  (The
/etc/exports file is fine, and does not change.)

What happens at 3:11 AM is that dump starts running on the root file
system.  I use a -L flag, so that's when it's doing the snapshot,
which somehow seems to be confusing the mount daemon.  But I also
dump /var and /usr a little later, and they don't have this problem.

The only difference I can see is that /var and /usr have soft updates,
while the root doesn't.  Any insight into what the heck is going on?

To recap:

* All three filesystems are NFS exported and are working normally
* Dump starts, nothing should change in NFS, but / export stops working
* Poke mountd, / export resumes workng.

R's,
John