Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
MIME-Version: 1.0
References: <00000000000043fe9c058720a5d3@google.com> <ad884b18-0e03-d6be-02b0-4e5273f068ce@kernel.dk>
In-Reply-To: <ad884b18-0e03-d6be-02b0-4e5273f068ce@kernel.dk>
From:   Linus Torvalds <torvalds@linux-foundation.org>
Date:   Mon, 22 Apr 2019 09:27:58 -0700
Message-ID: <CAHk-=wgG=RqK8T5kWiZfH7da6rBuqp3kAesP=ETgceb6RzvZng@mail.gmail.com>
Subject: Re: WARNING in percpu_ref_kill_and_confirm
To:     Jens Axboe <axboe@kernel.dk>
Cc:     syzbot <syzbot+10d25e23199614b7721f@syzkaller.appspotmail.com>,
        Arnd Bergmann <arnd@arndb.de>, Borislav Petkov <bp@alien8.de>,
        "Darrick J. Wong" <darrick.wong@oracle.com>,
        Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        Peter Anvin <hpa@zytor.com>,
        Linux API <linux-api@vger.kernel.org>,
        linux-arch <linux-arch@vger.kernel.org>,
        linux-block <linux-block@vger.kernel.org>,
        linux-fsdevel <linux-fsdevel@vger.kernel.org>,
        Linux List Kernel Mailing <linux-kernel@vger.kernel.org>,
        Andrew Lutomirski <luto@kernel.org>,
        Mathieu Desnoyers <mathieu.desnoyers@efficios.com>,
        Ingo Molnar <mingo@redhat.com>,
        Michael Ellerman <mpe@ellerman.id.au>,
        syzkaller-bugs <syzkaller-bugs@googlegroups.com>,
        Thomas Gleixner <tglx@linutronix.de>,
        Al Viro <viro@zeniv.linux.org.uk>,
        "the arch/x86 maintainers" <x86@kernel.org>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

[ Crossed emails ]

On Mon, Apr 22, 2019 at 9:23 AM Jens Axboe <axboe@kernel.dk> wrote:
>
> I think the below should fix this. Very early versions of io_uring didn't
> have this issue, since we did the percpu ref tryget for io_uring_register().

Ok, so I like your patch better than mine, but note how syzbot
bisected this to the initial merge of the io_uring code.

I agree that code shouldn't have had this particular issue, but it
looks like it does.

Is there some way to race with io_ring_ctx_wait_and_kill(), which
_also_ does that ref_kill() thing? I'm not seeing how that could
happen, but maybe if the file ref counts get screwed up you have
->release() called early..

                  Linus