On 11/9/06, Jason Thorpe <thorpej@shagadelic.org> wrote:
>
> On Nov 9, 2006, at 5:51 AM, Julio M. Merino Vidal wrote:
>
> >
> > Module Name:  src
> > Committed By: jmmv
> > Date:         Thu Nov  9 13:51:49 UTC 2006
> >
> > Modified Files:
> >       src/regress/sys/fs/tmpfs: h_funcs.subr h_tools.c t_create t_link
> >           t_mkdir t_mknod t_read_write t_remove t_rename t_rmdir t_setattr
> >           t_sockets t_symlink
> >
> > Log Message:
> > Add several checks to see if tmpfs raises the proper kqueue events
> > during
> > operation.  These all match the behavior exposed by MFS (except for a
> > corner case that is described in t_link).
> >
> > Fixes to tmpfs itself to make these tests pass will come soon.
>
> Instead of changing tmpfs, how about moving the kqueue notifications
> into the VFS layer?  FreeBSD has already done this.

That'd certainly be great; I wondered why this wasn't like that at the
moment.  (The fixes for tmpfs are in already though.)

The thing is that these notifications are sooo delicate -- and there
is no documentation that details exactly when each event has to be
raised -- that I don't dare to touch them without a complete suite of
tests that catches all possibilities.  So a first step, I think, could
be to extract all the new tests added to tmpfs, generalize them (easy)
and later ensure they are complete enough.  At last, do the changes in
the kernel with some certainty that no regressions are introduced.

-- 
Julio M. Merino Vidal <jmmv84@gmail.com>
The Julipedia - http://julipedia.blogspot.com/