cgroup: don't recycle cgroup id until all csses' have been destroyed

Hugh reported this bug:

> CONFIG_MEMCG_SWAP is broken in 3.13-rc.  Try something like this:
>
> mkdir -p /tmp/tmpfs /tmp/memcg
> mount -t tmpfs -o size=1G tmpfs /tmp/tmpfs
> mount -t cgroup -o memory memcg /tmp/memcg
> mkdir /tmp/memcg/old
> echo 512M >/tmp/memcg/old/memory.limit_in_bytes
> echo $$ >/tmp/memcg/old/tasks
> cp /dev/zero /tmp/tmpfs/zero 2>/dev/null
> echo $$ >/tmp/memcg/tasks
> rmdir /tmp/memcg/old
> sleep 1	# let rmdir work complete
> mkdir /tmp/memcg/new
> umount /tmp/tmpfs
> dmesg | grep WARNING
> rmdir /tmp/memcg/new
> umount /tmp/memcg
>
> Shows lots of WARNING: CPU: 1 PID: 1006 at kernel/res_counter.c:91
>                            res_counter_uncharge_locked+0x1f/0x2f()
>
> Breakage comes from 34c00c3 ("memcg: convert to use cgroup id").
>
> The lifetime of a cgroup id is different from the lifetime of the
> css id it replaced: memsw's css_get()s do nothing to hold on to the
> old cgroup id, it soon gets recycled to a new cgroup, which then
> mysteriously inherits the old's swap, without any charge for it.

Instead of removing cgroup id right after all the csses have been
offlined, we should do that after csses have been destroyed.

To make sure an invalid css pointer won't be returned after the css
is destroyed, make sure css_from_id() returns NULL in this case.

tj: Updated comment to note planned changes for cgrp->id.

Reported-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Li Zefan <lizefan@huawei.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Signed-off-by: Tejun Heo <tj@kernel.org>

kernel/cgroup.c

@@ -890,6 +890,16 @@ static void cgroup_diput(struct dentry *dentry, struct inode *inode)
 		struct cgroup *cgrp = dentry->d_fsdata;
 
 		BUG_ON(!(cgroup_is_dead(cgrp)));
+
+		/*
+		 * XXX: cgrp->id is only used to look up css's.  As cgroup
+		 * and css's lifetimes will be decoupled, it should be made
+		 * per-subsystem and moved to css->id so that lookups are
+		 * successful until the target css is released.
+		 */
+		idr_remove(&cgrp->root->cgroup_idr, cgrp->id);
+		cgrp->id = -1;
+
 		call_rcu(&cgrp->rcu_head, cgroup_free_rcu);
 	} else {
 		struct cfent *cfe = __d_cfe(dentry);
@@ -4268,6 +4278,7 @@ static void css_release(struct percpu_ref *ref)
 	struct cgroup_subsys_state *css =
 		container_of(ref, struct cgroup_subsys_state, refcnt);
 
+	rcu_assign_pointer(css->cgroup->subsys[css->ss->subsys_id], NULL);
 	call_rcu(&css->rcu_head, css_free_rcu_fn);
 }
 
@@ -4733,14 +4744,6 @@ static void cgroup_destroy_css_killed(struct cgroup *cgrp)
 	/* delete this cgroup from parent->children */
 	list_del_rcu(&cgrp->sibling);
 
-	/*
-	 * We should remove the cgroup object from idr before its grace
-	 * period starts, so we won't be looking up a cgroup while the
-	 * cgroup is being freed.
-	 */
-	idr_remove(&cgrp->root->cgroup_idr, cgrp->id);
-	cgrp->id = -1;
-
 	dput(d);
 
 	set_bit(CGRP_RELEASABLE, &parent->flags);