Re: [PATCH 2/4] freezer_cg: remove redundant check in freezer_can_attach()

[Cedric Le Goater <clg@fr.ibm.com>]

Li Zefan wrote:
> Cedric Le Goater wrote:
>> Li Zefan wrote:
>>> It is sufficient to check if @task is frozen, and no need to check if
>>> the original freezer is frozen.
>> hmm, a task being frozen does not mean that its freezer cgroup is 
>> frozen.
> 
> If a task has being frozen, then can_attach() returns -EBUSY at once.
> If a task isn't frozen, then we have orig_freezer->state == THAWED.
> 
> So we need to check the state of the task only.
> 
>> So the extra check seems valid but looking at the comment :
>>
>> 	/*
>> 	 * The call to cgroup_lock() in the freezer.state write method prevents
>> 	 * a write to that file racing against an attach, and hence the
>> 	 * can_attach() result will remain valid until the attach completes.
>> 	 */
>> 	static int freezer_can_attach(struct cgroup_subsys *ss,
>>
>> how do we know that the task_freezer(task), which is not protected by
>> the cgroup_lock(), is not going to change its state to CGROUP_FROZEN 
>> it looks racy.
>>
> 
> Since freezer_can_attach() is called by cgroup core with cgroup_lock held, there is
> no race with task attaching and state changing.

ok I see. cgroup_mutex is global, I thought it had changed and that we 
were only locking the cgroup the task was being attached to. 

Acked-by: Cedric Le Goater <clg@fr.ibm.com>

thanks,

C.