Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20220211170522.348999-1-agruenba@redhat.com> <CAHk-=wgURPnzwmWtgT-WSrpfFX9_wx_ZsY1XWp1MehrUuBuNiQ@mail.gmail.com>
In-Reply-To: <CAHk-=wgURPnzwmWtgT-WSrpfFX9_wx_ZsY1XWp1MehrUuBuNiQ@mail.gmail.com>
From:   Andreas Gruenbacher <agruenba@redhat.com>
Date:   Fri, 11 Feb 2022 22:40:41 +0100
Message-ID: <CAHc6FU7rMvu=Wry=VOXRv1191xooo+r8Cnv4XwYQysAr1hRFYw@mail.gmail.com>
Subject: Re: [GIT PULL] gfs2 fixes
To:     Linus Torvalds <torvalds@linux-foundation.org>
Cc:     cluster-devel <cluster-devel@redhat.com>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Fri, Feb 11, 2022 at 8:48 PM Linus Torvalds
<torvalds@linux-foundation.org> wrote:
> On Fri, Feb 11, 2022 at 9:05 AM Andreas Gruenbacher <agruenba@redhat.com> wrote:
> >
> > * Revert debug commit that causes unexpected data corruption.
>
> Well, apparently not just unexpected, but unexplained too.
>
> That's a bit worrisome. It sounds like the corruption cause is still
> there, just hidden by the lack of __cond_resched()?

Yes, that's what it looks like. My initial suspicion was that we're
somewhere using gfs2_glock_dq() in non-sleepable context when we know
that we're not dropping the last reference and so gfs2_glock_dq()
won't sleep, but there's no such instance in the code, and testing
would also have revealed such cases. The corruption we've seen always
affects whole pages/blocks. Maybe it's an ordering / memory barrier
issue.

Thanks,
Andreas