tcp: Don't change unlocked socket state in tcp_v4_err().
David S. Miller [Fri, 12 Nov 2010 21:35:00 +0000 (13:35 -0800)]
Alexey Kuznetsov noticed a regression introduced by
commit f1ecd5d9e7366609d640ff4040304ea197fbc618
("Revert Backoff [v3]: Revert RTO on ICMP destination unreachable")

The RTO and timer modification code added to tcp_v4_err()
doesn't check sock_owned_by_user(), which if true means we
don't have exclusive access to the socket and therefore cannot
modify it's critical state.

Just skip this new code block if sock_owned_by_user() is true
and eliminate the now superfluous sock_owned_by_user() code
block contained within.

Reported-by: Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
Signed-off-by: David S. Miller <davem@davemloft.net>
CC: Damian Lukowski <damian@tvk.rwth-aachen.de>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>

net/ipv4/tcp_ipv4.c

index 8f8527d..69ccbc1 100644 (file)
@@ -415,6 +415,9 @@ void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
                    !icsk->icsk_backoff)
                        break;
 
+               if (sock_owned_by_user(sk))
+                       break;
+
                icsk->icsk_backoff--;
                inet_csk(sk)->icsk_rto = __tcp_set_rto(tp) <<
                                         icsk->icsk_backoff;
@@ -429,11 +432,6 @@ void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
                if (remaining) {
                        inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
                                                  remaining, TCP_RTO_MAX);
-               } else if (sock_owned_by_user(sk)) {
-                       /* RTO revert clocked out retransmission,
-                        * but socket is locked. Will defer. */
-                       inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
-                                                 HZ/20, TCP_RTO_MAX);
                } else {
                        /* RTO revert clocked out retransmission.
                         * Will retransmit now */