In article <Pine.NEB.3.96.980104124506.16734B-100000@like.duh.org> you write:
> It seems to me that the unionfs might do a better job if it had a stateless
> file access mechanism.  Hrm, who knows.

Many locking problems could be solved by using a message-based protocol
in the kernel, and putting each mountpoint in its own thread.  This is
what AmigaOS did, and it worked pretty well (modulo a few problems
specific to that implementation, like poor provisions for *un*-mounting
things). It'd also make it easier to do userland VFS servers.

If we ever get SMP, that could have otehr benefits as well. 'Course,
it'd be a huge amount of work. I suspect things like unionfs and nullfs
would wind up being rewritten from scratch, since pretty much everything
they do is tightly wrapped around the existing in-kernel VFS API.