tcp: md5: fix IPv4-mapped support

After the blamed commit, IPv4 SYN packets handled
by a dual stack IPv6 socket are dropped, even if
perfectly valid.

$ nstat | grep MD5
TcpExtTCPMD5Failure             5                  0.0

For a dual stack listener, an incoming IPv4 SYN packet
would call tcp_inbound_md5_hash() with @family == AF_INET,
while tp->af_specific is pointing to tcp_sock_ipv6_specific.

Only later when an IPv4-mapped child is created, tp->af_specific
is changed to tcp_sock_ipv6_mapped_specific.

Fixes: 7bbb765 ("net/tcp: Merge TCP-MD5 inbound callbacks")
Reported-by: Brian Vazquez <[email protected]>
Signed-off-by: Eric Dumazet <[email protected]>
Reviewed-by: David Ahern <[email protected]>
Reviewed-by: Dmitry Safonov <[email protected]>
Tested-by: Leonard Crestez <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Jakub Kicinski <[email protected]>

Author: edumazet

net/ipv4/tcp.c

@@ -4459,9 +4459,18 @@ tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb,
 		return SKB_DROP_REASON_TCP_MD5UNEXPECTED;
 	}
 
-	/* check the signature */
-	genhash = tp->af_specific->calc_md5_hash(newhash, hash_expected,
-						 NULL, skb);
+	/* Check the signature.
+	 * To support dual stack listeners, we need to handle
+	 * IPv4-mapped case.
+	 */
+	if (family == AF_INET)
+		genhash = tcp_v4_md5_hash_skb(newhash,
+					      hash_expected,
+					      NULL, skb);
+	else
+		genhash = tp->af_specific->calc_md5_hash(newhash,
+							 hash_expected,
+							 NULL, skb);
 
 	if (genhash || memcmp(hash_location, newhash, 16) != 0) {
 		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5FAILURE);