On Mon, Feb 04, 2019 at 12:53:10PM -0800, Christoph Hellwig wrote: > On Mon, Feb 04, 2019 at 02:24:51PM -0500, Brian Foster wrote: > > I'm a little confused where we're at with this one given the discussion > > on the previous version. We've dropped the locking, but left the > > tracking in place. Are we just relying on holding the agi in the > > iunlink/iunlink_remove cases? If so, that seems reasonable to me but the > > commit log should probably have a sentence or two on the serialization > > rules. The commit log could also use an update to describe how this > > value is actually used in this patch (as an unmount time check) as > > opposed to some apparent throttling functionality that isn't a part of > > this series. > > I thought we Darrick was going to drop this tracking from the series, > as it isn't very useful (at least yet), but maybe I misunderstood the > previous thread. I decided to leave the unlinked counter so that we could have a debugging check. I will make it more explicit that anyone accessing the counter needs to hold the AGI buffer lock or otherwise assured that there aren't any other threads. --D
