Re: [PATCH][RFC] Always unlock 'tasklist_lock' in kernel/exit.c::do_wait()

[Oleg Nesterov <oleg@redhat.com>]

On 12/20, Jesper Juhl wrote:
>
> Hi,
>
> The Coverity checker spotted this. I think it has a point, but I'm not
> intimate with this code so there could be somethingI'm missing.
>
> It seems that kernel/exit.c::do_wait() does not always release
> 'tasklist_lock'. There are multiple ways the code could be changed to make
> sure it's always released, I just picked the most straight forward one.
>
> Does this look right to everyone else or is it just me?
> I've only compile tested the patch so far.
>
>
> Signed-off-by: Jesper Juhl <jj@chaosbits.net>
> ---
>  exit.c |    8 ++++++--
>  1 file changed, 6 insertions(+), 2 deletions(-)
>
> diff --git a/kernel/exit.c b/kernel/exit.c
> index 676149a..456b13d 100644
> --- a/kernel/exit.c
> +++ b/kernel/exit.c
> @@ -1647,12 +1647,16 @@ repeat:
>  	tsk = current;
>  	do {
>  		retval = do_wait_thread(wo, tsk);
> -		if (retval)
> +		if (retval) {
> +			read_unlock(&tasklist_lock);

No, this is not right.

If wait_consider_task() returns nonzero (pid or errcode), tasklist_lock
should be already released.

That is why, for example, wait_task_stopped() does BUG_ON(!retval) after
it drops tasklist.

Oleg.