I believe I have finally figured out what was behind those weird
VM/filesystem incoherecy problems I was reporting.

The problem _is_ nullfs, but is _not_ locking-related.  Briefly, the
ufs/ffs/nfs code is making the VM subsystem callbacks, but with
ufs/ffs/nfs vnodes.  And it's nullfs vnodes that are being mmap()ped,
so the VM subsystem ignores the callbacks.  See PR kern/1309.

					der Mouse

			    mouse@collatz.mcrcim.mcgill.edu