Re: [PATCH next] sched,freezer: prevent tasks from escaping being frozen

[Peter Zijlstra]

On Mon, Jul 07, 2025 at 07:32:59PM +0800, Chen Ridong wrote:

> Below is the race condition scenario:
> 
> get_signal				read freezer.state
> try_to_freeze
> __refrigerator				freezer_read
> 					update_if_frozen
> WRITE_ONCE(current->__state, TASK_FROZEN);			
> 					// The task is set to frozen(now, frozen(task) == ture).
> 					// we suppose other tasks are all frozen.
> 					// set cgroup frozen when all tasks are frozen
> 					freezer->state |= CGROUP_FROZEN;

Ooh, yes, now I see. Somehow I kept missing update_if_frozen().
Sometimes reading is hard :-) Thanks!

> // freezing(current) will return false,
> // since cgroup is frozen(not freezing)
> break out
> __set_current_state(TASK_RUNNING);
> //bug: the task is set to running, but it should be frozen.