On Thu, Dec 22, 2011 at 01:50:46PM +0800, Li Zefan wrote:

quoted

@@ -2091,6 +2010,10 @@ int cgroup_attach_proc(struct cgroup *cgrp, struct task_struct *leader)
 	 * rcu or tasklist locked. instead, build an array of all threads in the
 	 * group - group_rwsem prevents new threads from appearing, and if
 	 * threads exit, this will just be an over-estimate.
+	 *
+	 * While creating the list, also make sure css_sets exist for all
+	 * threads to be migrated. we use find_css_set, which allocates a new
+	 * one if necessary.
 	 */
 	group_size = get_nr_threads(leader);
 	/* flex_array supports very large thread-groups better than kmalloc. */

@@ -2137,6 +2060,12 @@ int cgroup_attach_proc(struct cgroup *cgrp, struct task_struct *leader)
 		/* nothing to do if this task is already in the cgroup */
 		if (ent.cgrp == cgrp)
 			continue;
+		ent.cg = find_css_set(tsk->cgroups, cgrp);

unfortunately This won't work, because we are holding tasklist_lock.

I believe we can remove tasklist_lock now (in a seperate patch).

It was there in order to protect while_each_thread() against exec but
now we have threadgroup_lock().

I think we only need to use rcu_read_lock() to protect against concurrent
removal in exit.
 

quoted

+		if (!ent.cg) {
+			retval = -ENOMEM;
+			group_size = i;
+			goto out_list_teardown;
+		}
 		retval = flex_array_put(group, i, &ent, GFP_ATOMIC);
 		BUG_ON(retval != 0);
 		i++;