tcp: fix tcp_mtu_probe() vs highest_sack
authorEric Dumazet <>
Tue, 31 Oct 2017 06:08:20 +0000 (23:08 -0700)
committerBen Hutchings <>
Mon, 1 Jan 2018 20:51:00 +0000 (20:51 +0000)
commit 2b7cda9c35d3b940eb9ce74b30bbd5eb30db493d upstream.

Based on SNMP values provided by Roman, Yuchung made the observation
that some crashes in tcp_sacktag_walk() might be caused by MTU probing.

Looking at tcp_mtu_probe(), I found that when a new skb was placed
in front of the write queue, we were not updating tcp highest sack.

If one skb is freed because all its content was copied to the new skb
(for MTU probing), then tp->highest_sack could point to a now freed skb.

Bad things would then happen, including infinite loops.

This patch renames tcp_highest_sack_combine() and uses it
from tcp_mtu_probe() to fix the bug.

Note that I also removed one test against tp->sacked_out,
since we want to replace tp->highest_sack regardless of whatever
condition, since keeping a stale pointer to freed skb is a recipe
for disaster.

Fixes: a47e5a988a57 ("[TCP]: Convert highest_sack to sk_buff to allow direct access")
Signed-off-by: Eric Dumazet <>
Reported-by: Alexei Starovoitov <>
Reported-by: Roman Gushchin <>
Reported-by: Oleksandr Natalenko <>
Acked-by: Alexei Starovoitov <>
Acked-by: Neal Cardwell <>
Acked-by: Yuchung Cheng <>
Signed-off-by: David S. Miller <>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <>

index 95c4211..3ff241e 100644 (file)
@@ -1394,12 +1394,12 @@ static inline void tcp_highest_sack_reset(struct sock *sk)
        tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
-/* Called when old skb is about to be deleted (to be combined with new skb) */
-static inline void tcp_highest_sack_combine(struct sock *sk,
+/* Called when old skb is about to be deleted and replaced by new skb */
+static inline void tcp_highest_sack_replace(struct sock *sk,
                                            struct sk_buff *old,
                                            struct sk_buff *new)
-       if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
+       if (old == tcp_highest_sack(sk))
                tcp_sk(sk)->highest_sack = new;
index 1c96e9b..da11cd7 100644 (file)
@@ -1676,6 +1676,7 @@ static int tcp_mtu_probe(struct sock *sk)
        nskb->ip_summed = skb->ip_summed;
        tcp_insert_write_queue_before(nskb, skb, sk);
+       tcp_highest_sack_replace(sk, skb, nskb);
        len = 0;
        tcp_for_write_queue_from_safe(skb, next, sk) {
@@ -1987,7 +1988,7 @@ static void tcp_collapse_retrans(struct sock *sk, struct sk_buff *skb)
        BUG_ON(tcp_skb_pcount(skb) != 1 || tcp_skb_pcount(next_skb) != 1);
-       tcp_highest_sack_combine(sk, next_skb, skb);
+       tcp_highest_sack_replace(sk, next_skb, skb);
        tcp_unlink_write_queue(next_skb, sk);