Re: [PATCH 1/2] memcg: mm_update_next_owner: kill the "retry" logic

[Oleg Nesterov]

Michal, thanks for looking at this,

On 06/27, Michal Hocko wrote:
>
> On Wed 26-06-24 17:29:24, Oleg Nesterov wrote:
> > @@ -446,7 +463,6 @@ void mm_update_next_owner(struct mm_struct *mm)
> >  {
> >  	struct task_struct *c, *g, *p = current;
> >
> > -retry:
> >  	/*
> >  	 * If the exiting or execing task is not the owner, it's
> >  	 * someone else's problem.
> > @@ -468,16 +484,16 @@ void mm_update_next_owner(struct mm_struct *mm)
> >  	 * Search in the children
> >  	 */
> >  	list_for_each_entry(c, &p->children, sibling) {
> > -		if (c->mm == mm)
> > -			goto assign_new_owner;
> > +		if (c->mm == mm && try_to_set_owner(c, mm))
> > +			goto ret;
>
> You need to unlock tasklist_lock, right? Same for other goto ret.

No. From the patch

	+/* drops tasklist_lock if succeeds */
	+static bool try_to_set_owner(struct task_struct *tsk, struct mm_struct *mm)
	+{
	+       bool ret = false;
	+
	+       task_lock(tsk);
	+       if (likely(tsk->mm == mm)) {
	+               /* tsk can't pass exit_mm/exec_mmap and exit */
	+               read_unlock(&tasklist_lock);
	                ^^^^^^^^^^^^^^^^^^^^^^^^^^^

try_to_set_owner() drops tasklist right after it verifies that
tsk->mm == mm under task_lock().

> It should still die but it can do so in a better shape.

Agreed!

Oleg.