Re: Port to 6.14-stable - ublk: fix race between io_uring_cmd_complete_in_task and ublk_cancel_cmd

[Jens Axboe <axboe@kernel.dk>]

On 5/6/25 7:40 AM, Ming Lei wrote:
> On Tue, May 06, 2025 at 04:16:25PM +0300, Jared Holzman wrote:
>> On 06/05/2025 5:13, Ming Lei wrote:
>>> On Mon, May 05, 2025 at 07:06:37PM +0300, Jared Holzman wrote:
>>>> Hi Ming,
>>>>
>>>> I'm attempting to back port the fix for this issue to the 6.14-stable branch.
>>>>
>>>> Greg Kroah-Hartman has already applied d6aa0c178bf8 - "ublk: call ublk_dispatch_req() for handling UBLK_U_IO_NEED_GET_DATA",
>>>> but was unable to apply f40139fde527 cleanly.
>>>>
>>>> I created the patch below. It applies and compiles, but when I rerun the scenario I get several hung tasks waiting on the ub->mutex
>>>> which is being held by the following task:
>>>
>>> Hi Jared,
>>>
>>> You need to pull in the following patchset too:
>>>
>>> https://lore.kernel.org/linux-block/20250416035444.99569-1-ming.lei@redhat.com/
>>>
>>> which avoids ub->mutex in error handling code path.
>>>
>>> I just picked them in the following tree:
>>>
>>> https://github.com/ming1/linux/commits/linux-6.14.y/
>>>
>>> Please test and see if they work for you.
>>>
>>>
>>> Thanks,
>>> Ming
>>>
>>
>> Hi Ming,
>>
>> Tested. It works great!
>>
>> Will you be sending a pull request to Greg or should I send him the patches?
> 
> Hi Jared,
> 
> Please make a PR and send to Greg since I didn't test & verify it on 6.14-y
> tree yet.
> 
> And please Cc me, I will give one double review.

Don't do a PR for stable, send a patch series instead.

-- 
Jens Axboe