netfilter: conntrack: handle tcp challenge acks during connection reuse
authorFlorian Westphal <fw@strlen.de>
Wed, 11 Jan 2023 13:42:32 +0000 (14:42 +0100)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 1 Feb 2023 07:34:21 +0000 (08:34 +0100)
[ Upstream commit c410cb974f2ba562920ecb8492ee66945dcf88af ]

When a connection is re-used, following can happen:
[ connection starts to close, fin sent in either direction ]
 > syn   # initator quickly reuses connection
 < ack   # peer sends a challenge ack
 > rst   # rst, sequence number == ack_seq of previous challenge ack
 > syn   # this syn is expected to pass

Problem is that the rst will fail window validation, so it gets
tagged as invalid.

If ruleset drops such packets, we get repeated syn-retransmits until
initator gives up or peer starts responding with syn/ack.

Before the commit indicated in the "Fixes" tag below this used to work:

The challenge-ack made conntrack re-init state based on the challenge
ack itself, so the following rst would pass window validation.

Add challenge-ack support: If we get ack for syn, record the ack_seq,
and then check if the rst sequence number matches the last ack number
seen in reverse direction.

Fixes: c7aab4f17021 ("netfilter: nf_conntrack_tcp: re-init for syn packets only")
Reported-by: Michal Tesar <mtesar@redhat.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
net/netfilter/nf_conntrack_proto_tcp.c

index 6566310..3ac1af6 100644 (file)
@@ -1068,6 +1068,13 @@ int nf_conntrack_tcp_packet(struct nf_conn *ct,
                                ct->proto.tcp.last_flags |=
                                        IP_CT_EXP_CHALLENGE_ACK;
                }
+
+               /* possible challenge ack reply to syn */
+               if (old_state == TCP_CONNTRACK_SYN_SENT &&
+                   index == TCP_ACK_SET &&
+                   dir == IP_CT_DIR_REPLY)
+                       ct->proto.tcp.last_ack = ntohl(th->ack_seq);
+
                spin_unlock_bh(&ct->lock);
                nf_ct_l4proto_log_invalid(skb, ct, state,
                                          "packet (index %d) in dir %d ignored, state %s",
@@ -1193,6 +1200,14 @@ int nf_conntrack_tcp_packet(struct nf_conn *ct,
                         * segments we ignored. */
                        goto in_window;
                }
+
+               /* Reset in response to a challenge-ack we let through earlier */
+               if (old_state == TCP_CONNTRACK_SYN_SENT &&
+                   ct->proto.tcp.last_index == TCP_ACK_SET &&
+                   ct->proto.tcp.last_dir == IP_CT_DIR_REPLY &&
+                   ntohl(th->seq) == ct->proto.tcp.last_ack)
+                       goto in_window;
+
                break;
        default:
                /* Keep compilers happy. */