On Sun 08-08-10 07:07:26, Jens Axboe wrote: > On 08/08/2010 03:29 AM, Wu Fengguang wrote: > > On Sun, Aug 08, 2010 at 12:12:30PM +0800, Dave Chinner wrote: > >> On Thu, Aug 05, 2010 at 04:45:35PM -0700, Andrew Morton wrote: > >>> On Thu, 5 Aug 2010 20:53:17 +0200 > >>> Jan Kara <jack@xxxxxxx> wrote: > >>>> --- > >>>> fs/fs-writeback.c | 8 ++++++++ > >>>> 1 files changed, 8 insertions(+), 0 deletions(-) > >>>> > >>>> diff --git a/fs/fs-writeback.c b/fs/fs-writeback.c > >>>> index d5be169..542471e 100644 > >>>> --- a/fs/fs-writeback.c > >>>> +++ b/fs/fs-writeback.c > >>>> @@ -633,6 +633,14 @@ static long wb_writeback(struct bdi_writeback *wb, > >>>> break; > >>>> > >>>> /* > >>>> + * Background writeout and kupdate-style writeback are > >>>> + * easily livelockable. Stop them if there is other work > >>>> + * to do so that e.g. sync can proceed. > >>>> + */ > >>>> + if ((work->for_background || work->for_kupdate) && > >>>> + !list_empty(&wb->bdi->work_list)) > >>>> + break; > >>>> + /* > >>> > >>> So what happens if an application sits in a loop doing write&fsync to a > >>> file? The vm's call for help gets ignored and your data doesn't get > >>> written back for three days?? > >> > >> To avoid the possibility of any such occurrence, perhaps requeuing > >> the work rather than cancelling it would be better? i.e. stop, put > >> it behind whatever work just came in and so when the new work > >> completes, we restart the background/expiry based writeback? > > > > That would be better. Ie. add a flag BDI_background_writeback to > > indicate the background work should be started after finishing with > > the queued works. bdi_start_background_writeback() can be modified to > > simply set the bit and wake up the flusher thread. > > I think that is better, we ideally want background writeout to > interleave smoothly with any other write activity. But why not just > mark the 'background writeout was interrupted' bit here when breaking, > and have the thread check-clear that when finishing some other piece > of work (or when going idle)? OK, although what would probably work as well would be to extend wb_check_old_data_flush() to also start background writeback (not only kupdate) if needed. Which probably makes sense regardless of any interruption of background writeback we might do... Maybe let's talk about this in some of the writeback sessions. Honza -- Jan Kara <jack@xxxxxxx> SUSE Labs, CR -- To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
