subflow: always init 'rel_write_seq'

Currently we do not init the subflow write sequence for
MP_JOIN subflows. This will cause bad mapping being
generated as soon as we will use non backup subflow.

Reviewed-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Tested-by: Christoph Paasch <cpaasch@apple.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net/mptcp/protocol.c

@@ -1814,7 +1814,6 @@ void mptcp_finish_connect(struct sock *ssk)
 	ack_seq++;
 	subflow->map_seq = ack_seq;
 	subflow->map_subflow_seq = 1;
-	subflow->rel_write_seq = 1;
 
 	/* the socket is not connected yet, no msk/subflow ops can access/race
 	 * accessing the field below

net/mptcp/subflow.c

@@ -200,6 +200,7 @@ static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
 	if (subflow->conn_finished)
 		return;
 
+	subflow->rel_write_seq = 1;
 	subflow->conn_finished = 1;
 	subflow->ssn_offset = TCP_SKB_CB(skb)->seq;
 	pr_debug("subflow=%p synack seq=%x", subflow, subflow->ssn_offset);