Re: [PATCH 2/4] xfs: improve handling of busy extents in the low-level allocator

[Brian Foster <bfoster@redhat.com>]

On Sat, Feb 04, 2017 at 10:56:06AM +0100, Christoph Hellwig wrote:
> On Fri, Feb 03, 2017 at 11:22:43AM -0500, Brian Foster wrote:
> > I didn't catch this until looking through everything after the next
> > patch, but I think there's a problem with the wakeup here as well. If we
> > have a busy extent with XFS_EXTENT_BUSY_SKIP_DISCARD set, we immediately
> > issue a wake from the first xfs_extent_busy_clear() in the cil committed
> > handler, regardless of whether !SKIP_DISCARD extents exist as well
> > under the current gen value. I think that means we'd get a premature
> > wake any time a busy_list has at least one of each type..?
> 
> We'll need to wake as soon as a previously busy extent in the AG
> becomes available, and that includes  XFS_EXTENT_BUSY_SKIP_DISCARD
> ones.  Otherwise e.g. a transaction only containing
> XFS_EXTENT_BUSY_SKIP_DISCARD will never wake at all.
> 

Hmm, Ok. I suppose that isn't a problem so long as we don't ever wait
for a particular extent based on a particular generation number. The
current code is just retrying allocations and whatnot, so is probably
safe.

That said, that limitation should be noted somewhere. Can we add a
comment in xfs_extent_busy_clear() right above the hunk where we do the
SKIP_DISCARD wake? E.g., something that points out the gen number for
any particular extent could be bumped in such a situation.. (or
something along those lines)?

Brian

> --
> To unsubscribe from this list: send the line "unsubscribe linux-xfs" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html