Re: [RFC][PATCH 1/2] CGROUP ID and Hierarchy Code

[Li Zefan <lizf@cn.fujitsu.com>]

KAMEZAWA Hiroyuki wrote:
> On Fri, 28 Nov 2008 10:35:33 +0900
> KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> wrote:
> 
>> On Fri, 28 Nov 2008 09:28:33 +0800
>> Li Zefan <lizf@cn.fujitsu.com> wrote:
> 
>>> I think the safe way is:
>>>
>>> rcu_read_lock();
>>> cgrp = cgroup_get_next()
>>> if (!inc_not_zero(cgrp) {
>>> 	rcu_read_unlock();
>>> 	return NULL;
>>> }
>>> rcu_read_unlock();
>>> return cgrp;
>>>
>>> But it's also safe to use cgrp = list_entry(&parent->children.next) for the above
>>> scenario, seems you don't have to invent this cgroup_get_next().
>>>
>> But list-walk can't provide us view of hierarchy. Up-Down list walk is better ?
>> please see memcg's code. it met some troubles.
>> I'll make kfree(cgrp) to be called by RCU.
>>
> 
> Ah please, I have a question.
> Following is right ?
> 	- until mount, cgroup's subsystem root is tied to dummy? root cgroup.

Yes.

> 	- cgroup cannot be unmounted while there are any children.

No, you can unmount. You can have it a try. :)

> 	- cgroup_root is freed/allocated at umount/mount, then I have to handle

No, kill_sb() is called when deactivate_super() decreases sb->s_active to 0, and
whenever we create a cgroup, sb->s_active is increased, so if there are sub-dirs
the cgroup_root won't be freed at umount.

> 	  this event in cgroup-id.
> 	 (maybe current code has leak? in umount.)
>  

So I think there is no leak. ;)

> Thanks,
> -Kame
> 
> 
>