cgroup: Fix potential race between cgroup_exit and cpuset_attach

There is a potential race between cgroup_exit() and the taskset migration path. This race happens when all tasks associated with the cg_list entries in mg_tasks, detach themselves before the tasks can be attached to the destination cpuset in cpuset_attach(). Below is the sequence where race is observed:

cpuset_hotplug_workfn()
cgroup_transfer_tasks()
cgroup_migrate()
cgroup_migrate_execute()
<TASK EXIT> list_del_init(&task->cg_list)
cpuset_attach()
cgroup_taskset_first(tset, &css) // css is not set guarantee_online_mems(cs, ...) // data abort

Fix this by adding a checking to verify that css is set from cgroup_taskset_first(), before proceeding.

---
Hi,

We observed this issue for cgroup code corresponding to stable v4.4.85 snapshot 3144d81 ("cgroup, kthread: close race window where new kthreads can be migrated to non-root cgroups"). Can you please tell us, if there are any patches in latest code, which fixes these issue?

 kernel/cgroup/cpuset.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

index 87a1213..7e245f26 100644
@@ -1510,11 +1510,17 @@ static void cpuset_attach(struct cgroup_taskset *tset)
 	static nodemask_t cpuset_attach_nodemask_to;
 	struct task_struct *task;
 	struct task_struct *leader;
-	struct cgroup_subsys_state *css;
+	struct cgroup_subsys_state *css = NULL;
 	struct cpuset *cs;
 	struct cpuset *oldcs = cpuset_attach_old_cs;
 
 	cgroup_taskset_first(tset, &css);
+	/* If all mg_tasks detached (from cgroup_exit())
+	 * before we started scanning the mg_tasks in
+	 * cgroup_taskset_next().
+	 */
+	if (css == NULL)
+		return;
 	cs = css_cs(css);
 
 	mutex_lock(&cpuset_mutex);

QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a
member of the Code Aurora Forum, hosted by The Linux Foundation

Hello,

I don't think this can happen anymore. We lock out task exits during migration and don't call migration callbacks if there's no tasks to migrate.

Thanks.

tejun