Re: [PATCH v3 3/3] cgroup: fix a race between cgroup_mount() and cgroup_kill_sb()

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

Hello, Li.

I applied the three patches with some updates (mostly comments).
Dynamic root management is turning out to be pretty ugly but I think
we can clean it up a bit.

On Mon, Jun 30, 2014 at 11:50:59AM +0800, Li Zefan wrote:
> @@ -1790,6 +1795,17 @@ out_free:
>  	dentry = kernfs_mount(fs_type, flags, root->kf_root, &new_sb);
>  	if (IS_ERR(dentry) || !new_sb)
>  		cgroup_put(&root->cgrp);
> +
> +	if (sb) {
> +		/*
> +		 * On success kernfs_mount() returns with sb->s_umount held,
> +		 * but kernfs_mount() also increases the superblock's refcnt,
> +		 * so calling deactivate_super() to drop the refcnt we got when
> +		 * looking up cgroup root won't acquire sb->s_umount again.
> +		 */
> +		WARN_ON(new_sb);
> +		deactivate_super(sb);
> +	}

So, @new_sb and @sb indicate the same conditions now.  When we reuse
an existing root, we always reuse its sb too which means that we at
least no longer need the @new_sb ugliness.  I actually kinda prefer it
as this means that the ugliness is confined to one ugly function -
kernfs_pin_sb() - instead of the generic kernfs_mount().  Can you
please prepare patches to clean these up?  I'll pull in for-3.16-fixes
into for-3.17 and apply the cleanup on top.

Thanks!

-- 
tejun