Re: [RFC] exit: do exit_task_work() before shooting off mm

[Pavel Begunkov]

On 20/12/2020 13:58, Oleg Nesterov wrote:
> On 12/20, Pavel Begunkov wrote:
>>
>> On 08/12/2020 01:37, Al Viro wrote:
>>> On Thu, Dec 03, 2020 at 02:30:46AM +0000, Pavel Begunkov wrote:
>>>> Handle task works and lock it earlier before it starts killing off
>>>> task's resources like mm. io_uring makes use of it a lot and it'd
>>>> nicer to have all added task_work finding tasks in a consistent state.
> 
> I too do not understand this patch. task_work_add() will fail after
> exit_task_work(). This means that, for example, exit_files() will use
> schedule_delayed_work().

The first one? Between PF_EXITING and exit_task_work() do_exit() will
kill mm/etc., I wanted to not see tasks half dismantled for task_works
run in the exit_task_work(). Anyway, forget about it :)

>> One more moment, after we've set PF_EXITING any task_work_run() would be
>> equivalent to exit_task_work()
> 
> Yes, currently task_work_run() can not be called after exit_signals().
> And shouldn't be called imo ;)
> 
>> io_uring
>> may want (currently doesn't) to run works for cancellation purposes.t
> 
> Please see https://lore.kernel.org/io-uring/20200407163816.GB9655@xxxxxxxxxx/>> Shouldn't it be like below (not tested)? Also simplifies task_work_run().
> 
> I'd prefer the patch from the link above, but your version looks correct too.

I missed the thread, thanks! tbh, splitting into 2 functions looks better
to me, but it's not like that matters

> However, I still think it would be better to not abuse task_work_run() too
> much...

The problem is that io_uring cancels requests in exit_files() and some
of them may be sitting in task_works, and we need to get them out of there
to complete.

Also, I need to double check, but seems new requests may be added during
and by cancellation because we did not yet set it to work_exited by the
time (in exit_files()). 

-- 
Pavel Begunkov