ipv4: tcp: set unicast_sock uc_ttl to -1

Set unicast_sock uc_ttl to -1 so that we select the right ttl,
instead of sending packets with a 0 ttl.

Bug added in commit be9f4a4 (ipv4: tcp: remove per net tcp_sock)

Signed-off-by: Hiroaki SHIMODA <shimoda.hiroaki@gmail.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net/ipv4/ip_output.c

@@ -1476,7 +1476,8 @@ static DEFINE_PER_CPU(struct inet_sock, unicast_sock) = {
 		.sk_allocation	= GFP_ATOMIC,
 		.sk_flags	= (1UL << SOCK_USE_WRITE_QUEUE),
 	},
-	.pmtudisc = IP_PMTUDISC_WANT,
+	.pmtudisc	= IP_PMTUDISC_WANT,
+	.uc_ttl		= -1,
 };
 
 void ip_send_unicast_reply(struct net *net, struct sk_buff *skb, __be32 daddr,