Re: [PATCH] sched: fix another race when reading /proc/sched_debug

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

KAMEZAWA Hiroyuki wrote:
> On Mon, 15 Dec 2008 09:39:18 +0800
> Li Zefan <lizf@cn.fujitsu.com> wrote:
> 
>> Paul Menage wrote:
>>> I sent out some patches last week (search for hierarchy_mutex) that would
>>> mean that you'd only need to take a subsystem-local lock to keep a cgroup
>>> alive. People seemed to like them so I'll tweak them based on feedback and
>>> send them on to Andrew.
>>>
>> Unfortunately, AFAICS the proposed hierarchy_mutex can't solve this bug. :(
>>
> Hmm ? how about this way if cgroup->dentry is problem ?
> 

Sounds feasable, though the resulting code might a bit tricky.

> at creation:
>   - cpu_cgroup_populate() should record "tg" that "this cgroup has valid dentry"

This avoids /proc/sched_debug accessing invalid cgrp dentry while a cgroup is
being created.

> at deletion
>   - css_tryget() will be useful to avoid the race.

And this avoids accessing a dead task_group.

Since css_tryget() is a 2.6.29 material, I think we can have this bug fixed for
2.6.28 by holding cgroup_lock, and try to remove the global lock for 2.6.29 ?