Re: [PATCH 1/2] fuse: Fix race in fuse_writepage_in_flight()

[Miklos Szeredi]

On Tue, Jan 15, 2019 at 4:55 PM Kirill Tkhai <ktkhai@xxxxxxxxxxxxx> wrote:
>
> On 15.01.2019 18:37, Miklos Szeredi wrote:
> > On Thu, Jan 10, 2019 at 12:03 PM Kirill Tkhai <ktkhai@xxxxxxxxxxxxx> wrote:
> >>
> >> On 10.01.2019 14:00, Miklos Szeredi wrote:
> >>> On Thu, Jan 10, 2019 at 11:48 AM Kirill Tkhai <ktkhai@xxxxxxxxxxxxx> wrote:
> >>>>
> >>>> Hi, Miklos,
> >>>>
> >>>> any comments about this?
> >>>
> >>> Is there a reproducer?  ISTR that fsx-linux with mmaps enabled was
> >>> good for stressing the writeback_cache code.
> >>
> >> There is no a reproducer, since I found that by eyes during preparation of another patchset.
> >
> > That's good.  It would even better to have a reproducer, but it
> > doesn't look easy...
> >
> > Completely redid this and reordered the patchset so this change is
> > made before the locking changes actually introduce the bug.
>
> Hm, I meant that I found this during preparation of the patchset,
> but not that fi->lock patchset introduces the bug. I don't think
> the patchset is involved:
>
> 1)before we had race, because different locks fc->lock and fiq->waitq.lock
> are taken in fuse_dev_read() and fuse_writepage_in_flight();
> 2)after we have the same race, and the locks are fi->lock and fiq->waitq.lock.

Ah, so the race was introduced earlier, when fiq->waitq.lock was split
out from fc->lock.

Thanks,
Miklos