Re: Making forced unmounts work

[Taylor R Campbell <campbell+netbsd-tech-kern%mumble.net@localhost>]

   Date: Thu, 6 Dec 2012 08:23:07 +0100
   From: "J. Hannken-Illjes" <hannken%eis.cs.tu-bs.de@localhost>

   Looks like this thread is dead.  No one beside David Holland is
   interested and David objects.  I take back my proposal.

I'm interested, but I haven't fit enough of the vnode life cycle or
the fstrans mechanism into my head to assess how well it is likely to
work, and I don't think it's a good strategy in general to put more
boilerplate into every vop implementation -- there is far too much
boilerplate already that lots of vop implementations get wrong.

As a step toward fixing the first problem, could you perhaps write
down the rules about how fstrans currently composes with vnode locks
and operations and life cycle?  I seem to recall there was some issue
a year or so ago about the ordering of vput and fstrans_done in
ufs_rmdir, and the nature of the issue wasn't clear to me from the man
pages or a cursory examination of the source, which suggested to me
that I don't understand fstrans well enough to reason about patches
like this.