device_cgroup: remove direct access to cgroup->children

Currently, devcg::has_children() directly tests cgroup->children for list emptiness. The field is not a published field and scheduled to go away. In addition, the test isn't strictly correct as devcg should only care about children which are visible to userland. This patch converts has_children() to use css_next_child() instead. The subtle incorrectness is noted and will be dealt with later. Signed-off-by: Tejun Heo <[email protected]> Acked-by: Aristeu Rozanski <[email protected]> Acked-by: Serge Hallyn <[email protected]> Acked-by: Li Zefan <[email protected]>
ambv · May 16, 2014 · 5877019 · 5877019
1 parent ea280e7
commit 5877019
Showing 1 changed file with 10 additions and 2 deletions.
diff --git a/security/device_cgroup.c b/security/device_cgroup.c
@@ -589,9 +589,17 @@ static int propagate_exception(struct dev_cgroup *devcg_root,
 
 static inline bool has_children(struct dev_cgroup *devcgroup)
 {
-	struct cgroup *cgrp = devcgroup->css.cgroup;
+	bool ret;
 
-	return !list_empty(&cgrp->children);
+	/*
+	 * FIXME: There may be lingering offline csses and this function
+	 * may return %true when there isn't any userland-visible child
+	 * which is incorrect for our purposes.
+	 */
+	rcu_read_lock();
+	ret = css_next_child(NULL, &devcgroup->css);
+	rcu_read_unlock();
+	return ret;
 }
 
 /*