net: fix divide by zero in tcp algorithm illinois
authorJesper Dangaard Brouer <brouer@redhat.com>
Wed, 31 Oct 2012 02:45:32 +0000 (02:45 +0000)
committerDavid S. Miller <davem@davemloft.net>
Thu, 1 Nov 2012 15:55:59 +0000 (11:55 -0400)
Reading TCP stats when using TCP Illinois congestion control algorithm
can cause a divide by zero kernel oops.

The division by zero occur in tcp_illinois_info() at:
 do_div(t, ca->cnt_rtt);
where ca->cnt_rtt can become zero (when rtt_reset is called)

Steps to Reproduce:
 1. Register tcp_illinois:
     # sysctl -w net.ipv4.tcp_congestion_control=illinois
 2. Monitor internal TCP information via command "ss -i"
     # watch -d ss -i
 3. Establish new TCP conn to machine

Either it fails at the initial conn, or else it needs to wait
for a loss or a reset.

This is only related to reading stats.  The function avg_delay() also
performs the same divide, but is guarded with a (ca->cnt_rtt > 0) at its
calling point in update_params().  Thus, simply fix tcp_illinois_info().

Function tcp_illinois_info() / get_info() is called without
socket lock.  Thus, eliminate any race condition on ca->cnt_rtt
by using a local stack variable.  Simply reuse info.tcpv_rttcnt,
as its already set to ca->cnt_rtt.
Function avg_delay() is not affected by this race condition, as
its called with the socket lock.

Cc: Petr Matousek <pmatouse@redhat.com>
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Acked-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/ipv4/tcp_illinois.c

index 813b43a76fece71e2614878a5830962d55f35a22..834857f3c8713e8d8b80ab7b4f413725ec0ca9b4 100644 (file)
@@ -313,11 +313,13 @@ static void tcp_illinois_info(struct sock *sk, u32 ext,
                        .tcpv_rttcnt = ca->cnt_rtt,
                        .tcpv_minrtt = ca->base_rtt,
                };
-               u64 t = ca->sum_rtt;
 
-               do_div(t, ca->cnt_rtt);
-               info.tcpv_rtt = t;
+               if (info.tcpv_rttcnt > 0) {
+                       u64 t = ca->sum_rtt;
 
+                       do_div(t, info.tcpv_rttcnt);
+                       info.tcpv_rtt = t;
+               }
                nla_put(skb, INET_DIAG_VEGASINFO, sizeof(info), &info);
        }
 }