mptcp: tfo: record 'deny join id0' info

When TFO is used, the check to see if the 'C' flag (deny join id0) was
set was bypassed.

This flag can be set when TFO is used, so the check should also be done
when TFO is used.

Note that the set_fully_established label is also used when a 4th ACK is
received. In this case, deny_join_id0 will not be set.

Fixes: dfc8d06 ("mptcp: implement delayed seq generation for passive fastopen")
Reviewed-by: Mat Martineau <martineau@kernel.org>
Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
Link: https://patch.msgid.link/20250912-net-mptcp-pm-uspace-deny_join_id0-v1-4-40171884ade8@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Author: matttbe

net/mptcp/options.c

@@ -985,13 +985,13 @@ static bool check_fully_established(struct mptcp_sock *msk, struct sock *ssk,
 		return false;
 	}
 
-	if (mp_opt->deny_join_id0)
-		WRITE_ONCE(msk->pm.remote_deny_join_id0, true);
-
 	if (unlikely(!READ_ONCE(msk->pm.server_side)))
 		pr_warn_once("bogus mpc option on established client sk");
 
 set_fully_established:
+	if (mp_opt->deny_join_id0)
+		WRITE_ONCE(msk->pm.remote_deny_join_id0, true);
+
 	mptcp_data_lock((struct sock *)msk);
 	__mptcp_subflow_fully_established(msk, subflow, mp_opt);
 	mptcp_data_unlock((struct sock *)msk);