All of lore.kernel.org
 help / color / mirror / Atom feed
* Huge timeout with loose=1 pickup tcp connections
@ 2013-06-13 10:51 Florian Westphal
  2013-06-13 12:07 ` Pablo Neira Ayuso
  2013-06-13 13:49 ` Jozsef Kadlecsik
  0 siblings, 2 replies; 3+ messages in thread
From: Florian Westphal @ 2013-06-13 10:51 UTC (permalink / raw)
  To: netfilter-devel

Hi.

With nf_conntrack_tcp_loose = 1, stray packets will create
conntrack with 5 days-timeout.  Is this necessary?

I ask, because I've got following tcp trace:


192.168.x is initiator/origin, 10.184. is responder/reply, connection is establised by usual 3whs,
 data was transmitted in both directions)

A 55:07 (id 12053, [DF], TCP, len 40) 192.168.x.52792 > 10.184.y.80: F, 426:426(0) ack 9237 win 255
B 55:07 (id 6129, [DF], TCP, len 40) 10.184.y.80 > 192.168.x.52792: ., ack 427 win 123
C 56:08 (id 6130, [DF], TCP, len 40) 10.184.y.80 > 192.168.x.52792: F, 9237:9237(0) ack 427 win 123
D 56:08 (id 12159, [DF], TCP, len 40) 192.168.x.52792 > 10.184.y.80: ., ack 9238 win 255

D is last packet seen for that connection.  Notice 61 second pause between B and C.

>From conntrack POV, we have following state transitions:
   [NEW] tcp   6 60 SYN_RECV ..
 [UPDATE] tcp  6 432000 ESTABLISHED ..
 [UPDATE] tcp  6 120 FIN_WAIT .. (triggered by packet A)
 [UPDATE] tcp  6 60 CLOSE_WAIT .. (triggered by packet B)
[DESTROY] tcp  6 (triggered by 60-second CLOSE_WAIT timeout)
 [NEW] tcp  6 432034 ESTABLISHED .. [UNREPLIED] (triggered by packet D)

Result is that, after a couple of hours, conntrack table starts to fill up
with UNREPLIED crud.

Follwing changes fix it:
- set loose=0 (avoids bogus NEW connection), OR,
- set net.ipv4.netfilter.ip_conntrack_tcp_timeout_close_wait=65
  (avoids NEW connection, new state transitions are:
  [UPDATE] tcp  6 60 CLOSE_WAIT .. (triggered by packet B)
  [UPDATE] tcp  6 30 LAST_ACK .. (triggered by C)
  [UPDATE] tcp  6 120 TIME_WAIT .. (triggered by D)

However, I think conntrack shouldn't insert a just-picked
up connection with default ESTABLISHED timeout value.  Suggestion:

diff --git a/net/netfilter/nf_conntrack_proto_tcp.c b/net/netfilter/nf_conntrack_proto_tcp.c
index 4d4d8f1..33ca4f3 100644
--- a/net/netfilter/nf_conntrack_proto_tcp.c
+++ b/net/netfilter/nf_conntrack_proto_tcp.c
@@ -1043,6 +1043,11 @@ static int tcp_packet(struct nf_conn *ct,
 			nf_ct_kill_acct(ct, ctinfo, skb);
 			return NF_ACCEPT;
 		}
+		/* loose=1 picked up existing connection, avoid large timeout */
+		if (old_state == TCP_CONNTRACK_NONE &&
+		    new_state == TCP_CONNTRACK_ESTABLISHED &&
+		    timeout > timeouts[TCP_CONNTRACK_UNACK])
+			timeout = timeouts[TCP_CONNTRACK_UNACK];
 	} else if (!test_bit(IPS_ASSURED_BIT, &ct->status)
 		   && (old_state == TCP_CONNTRACK_SYN_RECV
 		       || old_state == TCP_CONNTRACK_ESTABLISHED)

Jozsef, If you aggree I would make a formal submission.

Thanks,
Florian

^ permalink raw reply related	[flat|nested] 3+ messages in thread

end of thread, other threads:[~2013-06-13 13:49 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2013-06-13 10:51 Huge timeout with loose=1 pickup tcp connections Florian Westphal
2013-06-13 12:07 ` Pablo Neira Ayuso
2013-06-13 13:49 ` Jozsef Kadlecsik

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.