On Wed, Sep 30, 2020 at 06:50:11AM +1000, Dave Chinner wrote: > On Tue, Sep 29, 2020 at 10:12:27AM -0400, Brian Foster wrote: > > Implement a quick hack to abuse the superblock freeze mechanism to > > freeze the XFS transaction subsystem. > > > > XXX: to be replaced > > What was wrong with the per-cpu counter that I used in the prototype > I sent? Why re-invent the wheel? > We discussed this in the original thread. See [1] (the tail end of my mail is where we switch from general relogging discussion to the quotaoff prototype) and your immediate reply for reference. The synopsis is that I think a percpu rwsem around transaction allocation (what I've replaced this patch with) is far more straightforward to audit, test and maintain than annotating quota modifying transactions purely for the purpose of quotaoff. > Also, can we call this a pause/resume operation so it doesn't get > confused with filesystem freezing? Freezing as operation name is way > too overloaded already... > Sure, pause/resume seems fine to me. Brian [1] https://lore.kernel.org/linux-xfs/20200702185209.GA58137@bfoster/ > Cheers, > > Dave. > -- > Dave Chinner > david@xxxxxxxxxxxxx >
