Re: [RFC PATCH] cgroup: Remove task_lock() from cgroup_post_fork()

[Li Zefan <lizf-BthXqXjhjHXQFUHtdCDX3A@public.gmane.org>]

Frederic Weisbecker wrote:
> cgroup_post_fork() is protected between threadgroup_change_begin()
> and threadgroup_change_end() against concurrent changes of the
> child's css_set in cgroup_task_migrate(). Also the child can't
> exit and call cgroup_exit() at this stage, this means it's css_set
> can't be changed with init_css_set concurrently.
> 
> For these reasons, we don't need to hold task_lock() on the child
> because it's css_set can only remain stable in this place.
> 
> Let's remove the lock there.
> 
> NOTE: We could do something else: move threadgroup_change_end()
> before cgroup_post_fork() and keep the task_lock() which, combined
> with the css_set_lock, would be enough to synchronize against
> cgroup_task_migrate()'s change on child->cgroup and its cglist.
> Because outside that, the threadgroup lock doesn't appear to be
> needed on cgroup_post_fork().
> 

To narrow the scope of the threadgroup lock? I think it's preferable to keep
cgroup_post_fork() inside the lock, to make things simpler and we have
the same lock rule for both cgroup_fork() and cgroup_post_fork().

> Let's debate!
> 
> Signed-off-by: Frederic Weisbecker <fweisbec-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
> Cc: Li Zefan <lizf-BthXqXjhjHXQFUHtdCDX3A@public.gmane.org>
> Cc: Tejun Heo <tj-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org>
> Cc: Containers <containers-cunTk1MwBs9QetFLy7KEm3xJsTq8ys+cHZ5vskTnxNA@public.gmane.org>
> Cc: Cgroups <cgroups-u79uwXL29TY76Z2rM5mHXA@public.gmane.org>
> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu-+CUm20s59erQFUHtdCDX3A@public.gmane.org>
> Cc: Oleg Nesterov <oleg-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
> Cc: Andrew Morton <akpm-de/tnXTf+JLsfHDXvbKv3WD2FQJk+8+b@public.gmane.org>
> Cc: Paul Menage <paul-inf54ven1CmVyaH7bEyXVA@public.gmane.org>
> Cc: Mandeep Singh Baines <msb-F7+t8E8rja9g9hUCZPvPmw@public.gmane.org>
> ---
>  kernel/cgroup.c |   11 ++++++++---
>  1 files changed, 8 insertions(+), 3 deletions(-)
> 
> diff --git a/kernel/cgroup.c b/kernel/cgroup.c
> index 4936d88..d0dbf72 100644
> --- a/kernel/cgroup.c
> +++ b/kernel/cgroup.c
> @@ -4622,10 +4622,15 @@ void cgroup_post_fork(struct task_struct *child)
>  {
>  	if (use_task_css_set_links) {
>  		write_lock(&css_set_lock);
> -		task_lock(child);
> -		if (list_empty(&child->cg_list))
> +		if (list_empty(&child->cg_list)) {
> +			/*
> +			 * It's safe to use child->cgroups without task_lock()
> +			 * here because we are protected through
> +			 * threadgroup_change_begin() against concurrent
> +			 * css_set change in cgroup_task_migrate()
> +			 */

Also explain why it won't race with cgroup_exit()? You were not quite confident
about that before Oleg's explanation. ;)

>  			list_add(&child->cg_list, &child->cgroups->tasks);
> -		task_unlock(child);
> +		}
>  		write_unlock(&css_set_lock);
>  	}
>  }