> At a minimum you need to add a skb->sk == NULL warn-on and abort path > here, otherwise we will corrupt socket accounting and just explode > somewhere else. Adding this patch as-is will just introduce a new I ran the patch for a few days now and nothing exploded, no messages. I can add a skb->sk check and see if it triggers. > There are cases where the Tx path of the wireless loops back packets > back to the Rx path, and in such cases we certainly could see sockets > attached to the SKB. > > And Johannes is right, He keeps talking about cases like monitoring sockets that don't apply, which makes me somewhat suspicious of his analysis. > absolutely cannot modify ->truesize blindly. You can't change the > truesize value if a socket is attached. But pskb_expand_head modifies the size so obviously truesize needs to change too. So you're saying pskb_expand_head() is illegal when there might be a socket attached? Somehow I suspect a lot of the pskb_expand_head() callers are failing that requirement. Ok I guess we could call skb_orphan() unconditionally. Or if you guys have a better patch I can test. -Andi -- ak@xxxxxxxxxxxxxxx -- To unsubscribe from this list: send the line "unsubscribe linux-wireless" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
