net: Fix sock_wfree() race

Commit 2b85a34
(net: No more expensive sock_hold()/sock_put() on each tx)
opens a window in sock_wfree() where another cpu
might free the socket we are working on.

A fix is to call sk->sk_write_space(sk) while still
holding a reference on sk.

Reported-by: Jike Song <albcamus@gmail.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net/core/sock.c

@@ -1228,17 +1228,22 @@ void __init sk_init(void)
 void sock_wfree(struct sk_buff *skb)
 {
 	struct sock *sk = skb->sk;
-	int res;
+	unsigned int len = skb->truesize;
 
-	/* In case it might be waiting for more memory. */
-	res = atomic_sub_return(skb->truesize, &sk->sk_wmem_alloc);
-	if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
+	if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
+		/*
+		 * Keep a reference on sk_wmem_alloc, this will be released
+		 * after sk_write_space() call
+		 */
+		atomic_sub(len - 1, &sk->sk_wmem_alloc);
 		sk->sk_write_space(sk);
+		len = 1;
+	}
 	/*
-	 * if sk_wmem_alloc reached 0, we are last user and should
-	 * free this sock, as sk_free() call could not do it.
+	 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
+	 * could not do because of in-flight packets
 	 */
-	if (res == 0)
+	if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
 		__sk_free(sk);
 }
 EXPORT_SYMBOL(sock_wfree);