ipv4: tcp: unicast_sock should not land outside of TCP stack

commit be9f4a4 (ipv4: tcp: remove per net tcp_sock) added a
selinux regression, reported and bisected by John Stultz

selinux_ip_postroute_compat() expect to find a valid sk->sk_security
pointer, but this field is NULL for unicast_sock

It turns out that unicast_sock are really temporary stuff to be able
to reuse  part of IP stack (ip_append_data()/ip_push_pending_frames())

Fact is that frames sent by ip_send_unicast_reply() should be orphaned
to not fool LSM.

Note IPv6 never had this problem, as tcp_v6_send_response() doesnt use a
fake socket at all. I'll probably implement tcp_v4_send_response() to
remove these unicast_sock in linux-3.7

Reported-by: John Stultz <[email protected]>
Bisected-by: John Stultz <[email protected]>
Signed-off-by: Eric Dumazet <[email protected]>
Cc: Paul Moore <[email protected]>
Cc: Eric Paris <[email protected]>
Cc: "Serge E. Hallyn" <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

Author: Eric Dumazet

net/ipv4/ip_output.c

@@ -1536,6 +1536,7 @@ void ip_send_unicast_reply(struct net *net, struct sk_buff *skb, __be32 daddr,
 			  arg->csumoffset) = csum_fold(csum_add(nskb->csum,
 								arg->csum));
 		nskb->ip_summed = CHECKSUM_NONE;
+		skb_orphan(nskb);
 		skb_set_queue_mapping(nskb, skb_get_queue_mapping(skb));
 		ip_push_pending_frames(sk, &fl4);
 	}