tcp: fix NULL ref in tail loss probe
authorYuchung Cheng <ycheng@google.com>
Wed, 5 Dec 2018 22:38:38 +0000 (14:38 -0800)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Mon, 17 Dec 2018 08:24:28 +0000 (09:24 +0100)
[ Upstream commit b2b7af861122a0c0f6260155c29a1b2e594cd5b5 ]

TCP loss probe timer may fire when the retranmission queue is empty but
has a non-zero tp->packets_out counter. tcp_send_loss_probe will call
tcp_rearm_rto which triggers NULL pointer reference by fetching the
retranmission queue head in its sub-routines.

Add a more detailed warning to help catch the root cause of the inflight
accounting inconsistency.

Reported-by: Rafael Tinoco <rafael.tinoco@linaro.org>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
net/ipv4/tcp_output.c

index 97e8aea..65632f8 100644 (file)
@@ -2479,15 +2479,18 @@ void tcp_send_loss_probe(struct sock *sk)
                goto rearm_timer;
        }
        skb = skb_rb_last(&sk->tcp_rtx_queue);
+       if (unlikely(!skb)) {
+               WARN_ONCE(tp->packets_out,
+                         "invalid inflight: %u state %u cwnd %u mss %d\n",
+                         tp->packets_out, sk->sk_state, tp->snd_cwnd, mss);
+               inet_csk(sk)->icsk_pending = 0;
+               return;
+       }
 
        /* At most one outstanding TLP retransmission. */
        if (tp->tlp_high_seq)
                goto rearm_timer;
 
-       /* Retransmit last segment. */
-       if (WARN_ON(!skb))
-               goto rearm_timer;
-
        if (skb_still_in_host_queue(sk, skb))
                goto rearm_timer;