cpuset: allow to move tasks to empty cpusets

Currently some cpuset behaviors are not friendly when cpuset is co-mounted
with other cgroup controllers.

Now with this patchset if cpuset is mounted with sane_behavior option,
it behaves differently:

- Tasks will be kept in empty cpusets when hotplug happens and take
  masks of ancestors with non-empty cpus/mems, instead of being moved to
  an ancestor.

- A task can be moved into an empty cpuset, and again it takes masks of
  ancestors, so the user can drop a task into a newly created cgroup without
  having to do anything for it.

As tasks can reside in empy cpusets, here're some rules:

- They can be moved to another cpuset, regardless it's empty or not.

- Though it takes masks from ancestors, it takes other configs from the
  empty cpuset.

- If the ancestors' masks are changed, those tasks will also be updated
  to take new masks.

v2: add documentation in include/linux/cgroup.h

Signed-off-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>

include/linux/cgroup.h

@@ -281,6 +281,9 @@ enum {
 	 *   and take masks of ancestors with non-empty cpus/mems, instead of
 	 *   being moved to an ancestor.
 	 *
+	 * - cpuset: a task can be moved into an empty cpuset, and again it
+	 *   takes masks of ancestors.
+	 *
 	 * - memcg: use_hierarchy is on by default and the cgroup file for
 	 *   the flag is not created.
 	 *

kernel/cpuset.c

@@ -479,7 +479,7 @@ static int validate_change(const struct cpuset *cur, const struct cpuset *trial)
 	 */
 	ret = -ENOSPC;
 	if ((cgroup_task_count(cur->css.cgroup) || cur->attach_in_progress) &&
-	    (cpumask_empty(trial->cpus_allowed) ||
+	    (cpumask_empty(trial->cpus_allowed) &&
 	     nodes_empty(trial->mems_allowed)))
 		goto out;
 
@@ -1466,8 +1466,13 @@ static int cpuset_can_attach(struct cgroup *cgrp, struct cgroup_taskset *tset)
 
 	mutex_lock(&cpuset_mutex);
 
+	/*
+	 * We allow to move tasks into an empty cpuset if sane_behavior
+	 * flag is set.
+	 */
 	ret = -ENOSPC;
-	if (cpumask_empty(cs->cpus_allowed) || nodes_empty(cs->mems_allowed))
+	if (!cgroup_sane_behavior(cgrp) &&
+	    (cpumask_empty(cs->cpus_allowed) || nodes_empty(cs->mems_allowed)))
 		goto out_unlock;
 
 	cgroup_taskset_for_each(task, cgrp, tset) {