Re: [PATCH] cgroup: don't queue css_release_work if one already pending

[Tejun Heo <tj-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org>]

Hello,

On Thu, Apr 14, 2022 at 10:51:18AM -0700, Tadeusz Struk wrote:
> What happened was, the write triggered:
> cgroup_subtree_control_write()->cgroup_apply_control()->cgroup_apply_control_enable()->css_create()
> 
> which, allocates and initializes the css, then fails in cgroup_idr_alloc(),
> bails out and calls queue_rcu_work(cgroup_destroy_wq, &css->destroy_rwork);

Yes, but this css hasn't been installed yet.

> then cgroup_subtree_control_write() bails out to out_unlock:, which then goes:
> 
> cgroup_kn_unlock()->cgroup_put()->css_put()->percpu_ref_put(&css->refcnt)->percpu_ref_put_many(ref)

And this is a different css. cgroup->self which isn't connected to the half
built css which got destroyed in css_create().

So, I have a bit of difficulty following this scenario. The way that the
current code uses destroy_work is definitely nasty and it'd probably be a
good idea to separate out the different use cases, but let's first
understand what's failing.

Thanks.

-- 
tejun