net: datagram: fix unbounded loop in __skb_try_recv_datagram()
authorPaolo Abeni <pabeni@redhat.com>
Mon, 25 Mar 2019 13:18:06 +0000 (14:18 +0100)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 3 Apr 2019 04:26:16 +0000 (06:26 +0200)
[ Upstream commit 0b91bce1ebfc797ff3de60c8f4a1e6219a8a3187 ]

Christoph reported a stall while peeking datagram with an offset when
busy polling is enabled. __skb_try_recv_datagram() uses as the loop
termination condition 'queue empty'. When peeking, the socket
queue can be not empty, even when no additional packets are received.

Address the issue explicitly checking for receive queue changes,
as currently done by __skb_wait_for_more_packets().

Fixes: 2b5cd0dfa384 ("net: Change return type of sk_busy_loop from bool to void")
Reported-and-tested-by: Christoph Paasch <cpaasch@apple.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
net/core/datagram.c

index 57f3a6f..a487df5 100644 (file)
@@ -279,7 +279,7 @@ struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned int flags,
                        break;
 
                sk_busy_loop(sk, flags & MSG_DONTWAIT);
-       } while (!skb_queue_empty(&sk->sk_receive_queue));
+       } while (sk->sk_receive_queue.prev != *last);
 
        error = -EAGAIN;