Note that my fixes for union and nullfs are not sufficient to fix all
problems, but they do make a big difference.

I still see an occasional locking protocol screwup (deadlock between
kernel make, which uses a lot of ../.. processing, and another process).

I also see examples of files that get unlinked from the upper layer but
never get properly released.  I cannot generate this on demand, however.
The symptom is a filesystem which passes fsck clean tests, yet has
unreferenced allocated inodes--if you run fsck by hand, lots of stuff
ends up in lost+found.  It occasionally burps in a more serious fashion,
with valid files moving from a directory with some whiteout entries into
lost+found.  I can't reproduce this one (I suspect something
screwy with whiteout handling of directory entries, although it could
also be a problem with the union code).

My last understanding on this subject was that the union & null FS
authors (JSP & Kirk) were going to integrate my suggested fixes and/or
corrections to them into 4.4BSD-Lite^2.  Maybe they'll fix remaining
problems too?  At least I can hope :)

==John