[LTP] [PATCH v3 2/2] Refactor exit_group01 using new API

[Richard Palethorpe]

Hello,

Cyril Hrubis <chrubis@suse.cz> writes:

> Hi!
>> This could fail due to PID reuse or maybe because state == 'Z'.
>> 
>> Perhaps instead we could check the threads are put into the Z state?
>> 
>> I guess that the child threads are reparented to init and it reaps
>> them. So you will have to prevent that by setting PR_SET_CHILD_SUBREAPER
>> on the main test process with prctl.
>> 
>> The man pages are not clear on this though. I guess this is how it works
>> from reading the kernel code.
>
> What about we:
>
> 1) allocate a piece of shared memory
> 2) each thread would run a loop that would increment a counter in that
>    memory
> 3) the test process will wait() the main child
> 4) then monitor the shared memory to make sure that counters are not
>    incremented anymore

+1

>
> Also child threads shouldn't be reparented to init, as long as at least
> one of the threads runs the child process is technically alive and I
> guess that wait() will not return until last thread in the group is
> dead.

OK, I looked again and it appears that indeed the thread group leader is
not reaped if there are threads in the group that are still alive
(wait_consider_task). Also it is only the thread group leader that can
enter a zombie state (exit_notify).

So this just leaves PID recycling as a potential issue IIRC.

-- 
Thank you,
Richard.