[dm-devel] convert dm-thin to use dm-bufio

Mike Snitzer snitzer at redhat.com
Fri Aug 19 13:49:21 UTC 2011


On Fri, Aug 19 2011 at  6:22am -0400,
Joe Thornber <thornber at redhat.com> wrote:

> On Fri, Aug 19, 2011 at 05:46:04AM -0400, Mike Snitzer wrote:
> > On Fri, Aug 19 2011 at  5:11am -0400,
> > Joe Thornber <thornber at redhat.com> wrote:
> > 
> > > On Fri, Aug 19, 2011 at 03:04:36AM -0400, Mike Snitzer wrote:
> > > > Question for Joe:
> > > > You're making conflicting changes quick enough that I wonder if you
> > > > and Mikulas will ever converge (e.g. why do multiple block managers need
> > > > to have access to the same metadata device!?).
> > > 
> > > They don't; my issue is with getting an oops if they do through user
> > > error.  I clearly said in the commit message that this was a hack to
> > > get round issues introduced by agk's move to a kmemcache.  Cook
> > > something cleaner up between yourselves, or wait for me to look at it
> > > again once I've got through some more pressing issues.
> > 
> > OK, so this kmemcache problem will go away once you switch over to
> > bufio.
> 
> Assuming agk doesn't decide to switch over to a kmemcache.  Remember
> bm used to allocate via kmalloc or pages like bufio.

Indeed.
 
> > As for stacking, can't we just read the superblock to check if a device
> > is already in use (would work metadata anyway)?  No idea if that'd be
> > too costly -- probably not: read superblock and check if
> > THIN_SUPERBLOCK_MAGIC is set.
> 
> So you've already thought of a scenario where it's worth having two bm
> instances (one read only).

I didn't realize it but excellent point.




More information about the dm-devel mailing list