I've been taking a look at what would be required to export the
NULLFS-redived filesystems (Umapfs in particular, but equally applied to
all of them) via NFS.  I have diffs that will do this, which I will
make into change-request in the reasonably near future...but right now
they have a gross dependency on the underlying filesystem being an FFS (a
three-high stack won't work).

What I think is needed is to break the check for staleness out of
VOP_FHTOVP; there would instead be a VOP_STALE would take the place of

  ip->i_ffs_mode == 0 || ip->i_ffs_gen != ((struct ufid *)fidp)->ufid_gen

for non-UFS filesystems.  Then the bypass routine could pass VOP_STALE
on down the line until it hit a filesystem that could actually evaluate
staleness.

What does anyone think?  (Is there an even more deep-seated reason why
exporting nullfs and friends is a Bad Idea?)

-------------------------------------------------------------------------
Konrad Schroder           http://www.hitl.washington.edu/people/perseant/
System Administrator                         perseant@hitl.washington.edu
Human Interface Technology Lab                      Voice: (206) 616-1478
Box 352142, University of Washington, 98195, USA      FAX: (206) 543-5380