On Fri, 24 Aug 2012 10:56:16 -0700
"Paul E. McKenney" [off-list ref] wrote:

quoted

+			return -EPERM;
+
+		rcu_read_lock();
+		if ((p = br_get_port(br, args[1])) == NULL) {
+			rcu_read_unlock();
+			return -EINVAL;
+		}
+		rcu_read_unlock();  

Why is it safe to pass "p" out of the RCU read-side critical section?
I don't see that br_get_port() does anything to make this safe, at least
not in v3.5.

It would be accidentally protected by rtnl?