Re: possible deadlock in free_ioctx_users

[Dmitry Vyukov]

On Mon, Sep 10, 2018 at 11:50 AM, Miklos Szeredi <miklos@xxxxxxxxxx> wrote:
> On Mon, Sep 10, 2018 at 11:43 AM, Dmitry Vyukov <dvyukov@xxxxxxxxxx> wrote:
>> On Mon, Sep 10, 2018 at 11:28 AM, Miklos Szeredi <miklos@xxxxxxxxxx> wrote:
>>> On Sun, Sep 9, 2018 at 8:41 PM, syzbot
>>> <syzbot+d86c4426a01f60feddc7@xxxxxxxxxxxxxxxxxxxxxxxxx> wrote:
>>>> Hello,
>>>>
>>>> syzbot found the following crash on:
>>>>
>>>> HEAD commit:    f8f65382c98a Merge tag 'for-linus' of git://git.kernel.org..
>>>> git tree:       upstream
>>>> console output: https://syzkaller.appspot.com/x/log.txt?x=113260ae400000
>>>> kernel config:  https://syzkaller.appspot.com/x/.config?x=8f59875069d721b6
>>>> dashboard link: https://syzkaller.appspot.com/bug?extid=d86c4426a01f60feddc7
>>>> compiler:       gcc (GCC) 8.0.1 20180413 (experimental)
>>>> syz repro:      https://syzkaller.appspot.com/x/repro.syz?x=120baa9e400000
>>>> C reproducer:   https://syzkaller.appspot.com/x/repro.c?x=13979cbe400000
>>>>
>>>> IMPORTANT: if you fix the bug, please add the following tag to the commit:
>>>> Reported-by: syzbot+d86c4426a01f60feddc7@xxxxxxxxxxxxxxxxxxxxxxxxx
>>>>
>>>> random: sshd: uninitialized urandom read (32 bytes read)
>>>> random: sshd: uninitialized urandom read (32 bytes read)
>>>> random: sshd: uninitialized urandom read (32 bytes read)
>>>>
>>>> ========================================================
>>>> WARNING: possible irq lock inversion dependency detected
>>>> 4.19.0-rc2+ #229 Not tainted
>>>> --------------------------------------------------------
>>>> swapper/0/0 just changed the state of lock:
>>>> 00000000c02bddef (&(&ctx->ctx_lock)->rlock){..-.}, at: spin_lock_irq
>>>> include/linux/spinlock.h:354 [inline]
>>>> 00000000c02bddef (&(&ctx->ctx_lock)->rlock){..-.}, at:
>>>> free_ioctx_users+0xbc/0x710 fs/aio.c:603
>>>> but this lock took another, SOFTIRQ-unsafe lock in the past:
>>>>  (&fiq->waitq){+.+.}
>>>>
>>>>
>>>> and interrupts could create inverse lock ordering between them.
>>>>
>>>>
>>>> other info that might help us debug this:
>>>>  Possible interrupt unsafe locking scenario:
>>>>
>>>>        CPU0                    CPU1
>>>>        ----                    ----
>>>>   lock(&fiq->waitq);
>>>>                                local_irq_disable();
>>>>                                lock(&(&ctx->ctx_lock)->rlock);
>>>>                                lock(&fiq->waitq);
>>>>   <Interrupt>
>>>>     lock(&(&ctx->ctx_lock)->rlock);
>>>
>>> Fuse device doesn't support AIO ops.  So false positive, AFAICS.
>>
>> Hi Miklos,
>>
>> We still need to annotate this. How?
>
> Good question.
>
> Isn't lockdep assuming too much here?  It hasn't shown that that
> ctx_lock instance was actually called from interrupt context, has it?

+lockdep maintainers for lockdep false positive and how to annotate it

Full reports are available here:
https://syzkaller.appspot.com/bug?extid=d86c4426a01f60feddc7