Re: [Bug] Deadlock between rfkill_fop_write() and nfc_unregister_device()

[Krzysztof Kozlowski <krzk@kernel.org>]

On 17/12/2024 13:09, Johannes Berg wrote:
> On Tue, 2024-12-17 at 20:01 +0800, Sam Sun wrote:
>> On Tue, Dec 17, 2024 at 7:33 PM Johannes Berg <johannes@sipsolutions.net> wrote:
>>>
>>> On Tue, 2024-12-17 at 11:46 +0100, Johannes Berg wrote:
>>>> On Tue, 2024-12-17 at 17:33 +0800, Sam Sun wrote:
>>>>> Dear developers and maintainers,
>>>>>
>>>>> We originally encountered a task hung while using our modified
>>>>> syzkaller. It was tested against the latest upstream kernel. We
>>>>> analyzed the root cause and pinpoint the kernel crash log to the
>>>>> following two tasks.
>>>>>
>>>>
>>>> This issue has been known a very long time and should be fixed in NFC,
>>>> but I guess nobody is around to do it.
>>>>
>>>> https://syzkaller.appspot.com/bug?extid=bb540a4bbfb4ae3b425d
>>>>
>>>
>>> I think this one is also the same:
>>>
>>> https://syzkaller.appspot.com/bug?extid=9ef743bba3a17c756174
>>>
>>> and that's much older still.
>>>
>>
>> Thanks for your quick reply! I am sorry that I didn't double-check the
>> call stack of historical bugs reported by Syzbot. I will be careful
>> next time.
>>
> 
> No worries. Maybe someone who feels responsible for NFC will wake up ;-)

Patches are welcomed. The NFC stack was full of deadlocks, races and
uses-after-free. The only consolation was that all of them were
triggered by the virtual device driver, not real world cases. Many
syzkaller reports were eventually fixed, but I guess many are still open.

If anyone wants to take the responsibility for NFC in terms of actually
developing and fixes the stack, go ahead, because I have time here only
for reviews (and these are still behind netdev timeframe expectations).

Best regards,
Krzysztof