On 9/21/23 12:20, Paolo Abeni wrote:

On Mon, 2023-09-18 at 20:00 +0100, Dmitry Safonov wrote:

quoted

@@ -615,19 +616,43 @@ static void bpf_skops_write_hdr_opt(struct sock *sk, struct sk_buff *skb,
  * (but it may well be that other scenarios fail similarly).
  */
 static void tcp_options_write(struct tcphdr *th, struct tcp_sock *tp,
-			      struct tcp_out_options *opts)
+			      struct tcp_out_options *opts,
+			      struct tcp_key *key)
 {
 	__be32 *ptr = (__be32 *)(th + 1);
 	u16 options = opts->options;	/* mungable copy */
 
-	if (unlikely(OPTION_MD5 & options)) {
+	if (tcp_key_is_md5(key)) {
 		*ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
 			       (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
 		/* overload cookie hash location */
 		opts->hash_location = (__u8 *)ptr;
 		ptr += 4;
-	}
+	} else if (tcp_key_is_ao(key)) {
+#ifdef CONFIG_TCP_AO

+		struct tcp_ao_key *rnext_key;
+		struct tcp_ao_info *ao_info;
+		u8 maclen;
 
+		ao_info = rcu_dereference_check(tp->ao_info,
+				lockdep_sock_is_held(&tp->inet_conn.icsk_inet.sk));
+		rnext_key = READ_ONCE(ao_info->rnext_key);
+		if (WARN_ON_ONCE(!rnext_key))
+			goto out_ao;
+		maclen = tcp_ao_maclen(key->ao_key);

It looks like only TCP_AO really accesses 'key', and TCP_AO can easily
fetch it from ao_info. Can the AO key change in between
tcp_get_current_key() and here?

Yes, you read it right: current_key can be changed at any moment, when
the peer asks to start a rotation (tcp_inbound_ao_hash() on RX does
that). So, here we have to provide the fetched key as ao_key->maclen
(the length of MAC, the authentication/hash may be different between
different keys).

Otherwise I think it would be better omitting the 'key' argument here
and use the 'options' flag to pick TCP_AO vs MD5.

And:

	if (unlikely(OPTION_MD5 & options)) {

could possibly be moved under a CONFIG_MD5 compiler conditional.

Thanks,
          Dmitry