Re: [PATCH] freezer,sched: Rewrite core freezer logic

[Oleg Nesterov <oleg@redhat.com>]

Hi Peter, sorry for delay,

On 06/11, Peter Zijlstra wrote:
>
> +/* Recursion relies on tail-call optimization to not blow away the stack */
> +static bool __frozen(struct task_struct *p)
> +{
> +	if (p->state == TASK_FROZEN)
> +		return true;
> +
> +	/*
> +	 * If stuck in TRACED, and the ptracer is FROZEN, we're frozen too.
> +	 */
> +	if (task_is_traced(p))
> +		return frozen(rcu_dereference(p->parent));

Why does it use frozen(), not __frozen() ?

This looks racy, p->parent can resume this task and then enter
__refrigerator().

Plus this task can be SIGKILL'ed even if it is traced.


> +	/*
> +	 * If stuck in STOPPED and the parent is FROZEN, we're frozen too.
> +	 */
> +	if (task_is_stopped(p))
> +		return frozen(rcu_dereference(p->real_parent));

(you could use ->parent in this case too and unify this check with the
"traced" case above)

I don't understand. How this connects to ->parent or ->real_parent?
SIGCONT can come from anywhere and wake this stopped task up?


I guess you do this to avoid freezable_schedule() in ptrace/signal_stop,
and we can't use TASK_STOPPED|TASK_FREEZABLE, it should not run after
thaw()... But see above, we can't rely on __frozen(parent).

Oleg.