Messages in this thread | | | From | "Rafael J. Wysocki" <> | Subject | Re: [PATCH] Freeze bdevs when freezing processes. | Date | Thu, 26 Oct 2006 11:08:22 +0200 |
| |
On Thursday, 26 October 2006 10:18, Nigel Cunningham wrote: > Hi Dave. > > On Thu, 2006-10-26 at 17:30 +1000, David Chinner wrote: > > On Wed, Oct 25, 2006 at 09:05:56PM +0200, Rafael J. Wysocki wrote: > > > On Wednesday, 25 October 2006 15:23, Nigel Cunningham wrote: > > > > > > > > > > Well, my impression is that this is exactly what happens here: Something > > > > > in the XFS code causes metadata to be written to disk _after_ the atomic > > > > > snapshot. > > > > > > > > > > That's why I asked if the dirty XFS metadata were flushed by a kernel thread. > > > > > > > > When I first added bdev freezing it was because there was an XFS timer > > > > doing writes. > > > > > > Yes, I noticed you said that, but I'd like someone from the XFS team to either > > > confirm or deny it. > > > > We have daemons running in the background that can definitely do stuff > > after a sync. hmm - one does try_to_freeze() after a wakeup, the > > other does: > > > > if (unlikely(freezing(current))) { > > set_bit(XBT_FORCE_SLEEP, &target->bt_flags); > > refrigerator(); > > } else { > > clear_bit(XBT_FORCE_SLEEP, &target->bt_flags); > > } > > > > before it goes to sleep. So that one (xfsbufd - metadata buffer flushing) > > can definitely wake up after the sync and do work, and the other could if > > the kernel thread freeze occurs after the sync. > > > > Another good question at this point - exactly how should we be putting > > these thread to to sleep? Are both these valid methods for freezing them? > > And should we be freezing when we wake up instead of before we go to > > sleep? i.e. what are teh rules we are supposed to be following? > > As you have them at the moment, the threads seem to be freezing fine. > The issue I've seen in the past related not to threads but to timer > based activity. Admittedly it was 2.6.14 when I last looked at it, but > there used to be a possibility for XFS to submit I/O from a timer when > the threads are frozen but the bdev isn't frozen.
Also there may be a problem if a workqueue is used for that, because worker_threads run with PF_NOFREEZE set.
Greetings, Rafael
-- You never change things by fighting the existing reality. R. Buckminster Fuller - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
| |