1 files changed, 12 insertions, 3 deletions

diff --git a/kernel/cgroup.c b/kernel/cgroup.c
index c6bd67b3fcf6..548d8d4e86d0 100644
--- a/kernel/cgroup.c
+++ b/kernel/cgroup.c
@@ -4595,10 +4595,19 @@ void cgroup_post_fork(struct task_struct *child)
 {
         if (use_task_css_set_links) {
                 write_lock(&css_set_lock);
-                task_lock(child);
-                if (list_empty(&child->cg_list))
+                if (list_empty(&child->cg_list)) {
+                        /*
+                         * It's safe to use child->cgroups without task_lock()
+                         * here because we are protected through
+                         * threadgroup_change_begin() against concurrent
+                         * css_set change in cgroup_task_migrate(). Also
+                         * the task can't exit at that point until
+                         * wake_up_new_task() is called, so we are protected
+                         * against cgroup_exit() setting child->cgroup to
+                         * init_css_set.
+                         */
                         list_add(&child->cg_list, &child->cgroups->tasks);
-                task_unlock(child);
+                }
                 write_unlock(&css_set_lock);
         }
 }