Re: UVM/genfs review

[Joerg Sonnenberger <joerg%britannica.bec.de@localhost>]

On Fri, Jan 09, 2009 at 01:09:40PM +0000, Andrew Doran wrote:
> On Thu, Jan 08, 2009 at 09:12:55PM +0100, Juergen Hannken-Illjes wrote:
> 
> > On Thu, Jan 08, 2009 at 08:01:56PM +0100, Joerg Sonnenberger wrote:
> > > Hi all,
> > > given the state of netbsd-5, I would like to get the WAPBL patch
> > > http://mail-index.netbsd.org/current-users/2008/10/31/msg005591.html
> > > into HEAD and pull it up. For that I would really like to have review
> > > from someone who understand UVM and/or genfs. The patch is likely to
> > > break persistent snapshots with fss(4), but I don't think it will break
> > > non-persistent ones (e.g. as used for backups).
> > > 
> > > Joerg
> > 
> > Just for the records, still strong objection from me.  As I told you before
> > this would bring back a deadlock (see Rev. 1.12 of genfs_io.c).  We have
> > to take the wapbl LOCK before we lock pages.  Therefore it will break
> > ALL file system internal snapshots, persistent or not.
> 
> If this change introduces another deadlock or problem, even if it fixes
> something else, don't submit it for pull-up as-is.

In the current form WAPBL can be trivially killed by either doing lots
of parallel IO or simply by using something like rtorrent. The latter
can result in the infamous tstile issue in less than 30sec. That is
completely inacceptable. The patch solves most of the performance issues
as well. I have to look again at fss, but frankly, if snapshots are
broken with WAPBL it is still much less an issue than the current state.

Joerg