soreuseport: Fix TCP listener hash collision
authorCraig Gallek <kraig@google.com>
Thu, 28 Apr 2016 23:24:32 +0000 (19:24 -0400)
committerDavid S. Miller <davem@davemloft.net>
Sun, 1 May 2016 23:36:54 +0000 (19:36 -0400)
I forgot to include a check for listener port equality when deciding
if two sockets should belong to the same reuseport group.  This was
not caught previously because it's only necessary when two listening
sockets for the same user happen to hash to the same listener bucket.
The same error does not exist in the UDP path.

Fixes: c125e80b8868("soreuseport: fast reuseport TCP socket selection")
Signed-off-by: Craig Gallek <kraig@google.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/ipv4/inet_hashtables.c

index bc68eced0105716dca6f04abbdd491a1cfb9e908..0d9e9d7bb029373e37e4de6ed4003397d41f82cd 100644 (file)
@@ -470,6 +470,7 @@ static int inet_reuseport_add_sock(struct sock *sk,
                                                     const struct sock *sk2,
                                                     bool match_wildcard))
 {
+       struct inet_bind_bucket *tb = inet_csk(sk)->icsk_bind_hash;
        struct sock *sk2;
        struct hlist_nulls_node *node;
        kuid_t uid = sock_i_uid(sk);
@@ -479,6 +480,7 @@ static int inet_reuseport_add_sock(struct sock *sk,
                    sk2->sk_family == sk->sk_family &&
                    ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
                    sk2->sk_bound_dev_if == sk->sk_bound_dev_if &&
+                   inet_csk(sk2)->icsk_bind_hash == tb &&
                    sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
                    saddr_same(sk, sk2, false))
                        return reuseport_add_sock(sk, sk2);