diff mbox

[net-next] tcp: avoid tcp loop connection on lo device

Message ID 50616772.8040704@gmail.com
State Rejected, archived
Delegated to: David Miller
Headers show

Commit Message

solomon Sept. 25, 2012, 8:12 a.m. UTC
Tcp supports simultaneous Connection, but we meat odd phenomenon that tcp client can receive what 
send by itself. tcp client and tcp server communicate through loop device. tcp server selects
port 40000 to listen which is in local port range. But after tcp server program is killed, tcp client
still can connect successfully. 


  Client                    Server
connect  <-----OK--------->   listen:127.0.0.1,port:40000
send(d1) ----------------->   recv
recv(d2) <-----------------   send(d2)  
 

         -----------------    killed(40000 not listened)

connect  <-----OK--------->   
send(d1) ----------------->  
recv(d1) <-----------------  

The simultaneous connection has no meaning for lo device, and for this case,
tcp client don't know whether server is listen on port 40000.
Just fix it sending reset to keep consistent state machine.  


Reproduced step:
1. while true ; do  nc 127.0.0.1 40001 ;done
2.  ss -nt dst 127.0.0.1
    State      Recv-Q Send-Q                                  Local Address:Port                                    Peer Address:Port 
    ESTAB      0      0                                           127.0.0.1:40001                                      127.0.0.1:40001 


Signed-off-by: Shan Wei <davidshan@tencent.com>
---
 net/ipv4/tcp_input.c |    8 +++++++-
 1 files changed, 7 insertions(+), 1 deletions(-)

Comments

David Miller Sept. 25, 2012, 5:04 p.m. UTC | #1
From: Shan Wei <shanwei88@gmail.com>
Date: Tue, 25 Sep 2012 16:12:34 +0800

> Tcp supports simultaneous Connection, but we meat odd phenomenon
> that tcp client can receive what send by itself. tcp client and tcp
> server communicate through loop device. tcp server selects port
> 40000 to listen which is in local port range. But after tcp server
> program is killed, tcp client still can connect successfully.

This is expected behavior and we're not special casing this after
it being this way for 20+ years.
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
diff mbox

Patch

diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index e037697..a2f5a10 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -5659,6 +5659,7 @@  static int tcp_rcv_synsent_state_process(struct sock *sk, struct sk_buff *skb,
 	struct tcp_cookie_values *cvp = tp->cookie_values;
 	struct tcp_fastopen_cookie foc = { .len = -1 };
 	int saved_clamp = tp->rx_opt.mss_clamp;
+	struct inet_sock *isk = inet_sk(sk);
 
 	tcp_parse_options(skb, &tp->rx_opt, &hash_location, 0, &foc);
 
@@ -5832,8 +5833,13 @@  discard:
 	if (th->syn) {
 		/* We see SYN without ACK. It is attempt of
 		 * simultaneous connect with crossed SYNs.
-		 * Particularly, it can be connect to self.
+		 * But, avoid tcp loop connection on single socket.
 		 */
+
+		if (isk->inet_dport == isk->inet_sport &&
+		    isk->inet_saddr == isk->inet_daddr)
+			goto reset_and_undo;
+
 		tcp_set_state(sk, TCP_SYN_RECV);
 
 		if (tp->rx_opt.saw_tstamp) {