conntrack event missing TCP protoinfo
Subsystems (2)
PATCH 1/1 conntrack event missing TCP protoinfo
Hi all,
Under some conditions it's possible to send a conntrack netlink event to userspace which is missing the TCP connection state (CDA_PROTOINFO_TCP_STATE) even though it is well defined. This one liner should fix the problem. What do you think?
Br,
Rui
net/netfilter/nf_conntrack_proto_tcp.c +1 -0
@@ -1045,6 +1045,7 @@ static int tcp_packet(struct nf_conn *ct,after SYN_RECV or a valid answer for a picked up connection. */ set_bit (IPS_ASSURED_BIT , &ct->status); +nf_conntrack_event_cache (IPCT_PROTOINFO , ct);nf_conntrack_event_cache (IPCT_STATUS , ct); }nf_ct_refresh_acct (ct, ctinfo, skb, timeout);
Pablo Neira Ayuso
maintainerHi,
Quoted 4 lines · Rui Sousa
Hi all,
Under some conditions it's possible to send a conntrack netlink event to userspace which is missing the TCP connection state (
CDA_PROTOINFO_TCP_STATE) even though it is well defined. This one liner should fix the problem. What do you think?
I see, I had to look at the code to find the condition that you mention, that it is the following:
if (!test_bit(IPS_ASSURED_BIT, &ct->status)
&& (old_state == TCP_CONNTRACK_SYN_RECV
|| old_state == TCP_CONNTRACK_ESTABLISHED)
&& new_state == TCP_CONNTRACK_ESTABLISHED)
/* Set ASSURED if we see see valid ack in ESTABLISHED
after SYN_RECV or a valid answer for a picked up
connection. */However, the policy is to include the TCP protocol state iff it has changed.
Quoted 1 line · On Thursday 31 December 2009 19:09:34 Pablo Neira Ayuso wrote:
Hi,
Hi Pablo,
Quoted 12 lines
Quoted 6 lines · Rui Sousa
Hi all,
Under some conditions it's possible to send a conntrack netlink event to userspace which is missing the TCP connection state (
CDA_PROTOINFO_TCP_STATE) even though it is well defined. This one liner should fix the problem. What do you think?I see, I had to look at the code to find the condition that you mention, that it is the following:
if (!test_bit(IPS_ASSURED_BIT,
&ct->status)
Quoted 8 lines
&& (old_state == TCP_CONNTRACK_SYN_RECV
|| old_state == TCP_CONNTRACK_ESTABLISHED)
&& new_state == TCP_CONNTRACK_ESTABLISHED)
/* Set ASSURED if we see see valid ack in ESTABLISHED
after SYN_RECV or a valid answer for a picked up
connection. */I believe so, but I having a hard time understanding what I'm seeing. AFAICT, I'm exercising the state transition sNo->sES->sES, my procedure is:
PC1 -> linux router -> PC2
1. establish TCP connection between two PC's (using iperf). PC1 is the client, PC2 is the server.
2. On the router, ifdown of output interface (the interface on PC2 side)
3. On the router, manual destroy of connection in kernel
4. On the router, ifup of output interface
5. wait for iperf to start sending packets again. The connection between the endpoints is always established.
Between 3 and 4 I receive a NFCT_DESTROY event (from libnetfilter_conntrack). During 5. I get two events, both NFCT_UPDATEs, the first with conntrack status CONFIRMED/SEEN_REPLY, the second with conntrack status ASSURED. Both are missing the TCP protoinfo. In this is my problem, the kernel correctly picked up the on going TCP connection but never sends enough information to userspace about it.
Quoted 2 lines
However, the policy is to include the TCP protocol state iff it has changed.
But in my case it does change and I never get a protocol state indication.
Br,
Rui
Pablo Neira Ayuso
maintainerHi Rui,
Quoted 5 lines · Rui Sousa
I believe so, but I having a hard time understanding what I'm seeing. AFAICT, I'm exercising the state transition
sNo->sES->sES, my procedure is:PC1 -> linux router -> PC2
Quoted 8 lines
1. establish TCP connection between two PC's (using iperf). PC1 is the client, PC2 is the server.
2. On the router, ifdown of output interface (the interface on PC2 side)
3. On the router, manual destroy of connection in kernel
4. On the router, ifup of output interface
5. wait for iperf to start sending packets again. The connection between the endpoints is always established.Between 3 and 4 I receive a
NFCT_DESTROYevent (fromlibnetfilter_conntrack).
During 5. I get two events, bothNFCT_UPDATEs, the first with conntrack status CONFIRMED/SEEN_REPLY, the second with conntrack status ASSURED. Both are missing the TCP protoinfo. In this is my problem, the kernel correctly picked up the on going TCP connection but never sends enough information to userspace about it.
Good analysis. Since Linux kernel 2.6.30, you should see a NFCT_NEW event (for conntracks that have been created via ctnetlink) before you get the two NFCT_UPDATE events. The NFCT_NEW contains the TCP protocol state.
I guess that you're using an older Linux kernel.
Pablo Neira Ayuso
maintainerQuoted 12 lines · Pablo Neira Ayuso
Quoted 11 lines · Rui Sousa
I believe so, but I having a hard time understanding what I'm seeing. AFAICT, I'm exercising the state transition
sNo->sES->sES, my procedure is:PC1 -> linux router -> PC2
1. establish TCP connection between two PC's (using iperf). PC1 is the client, PC2 is the server.
2. On the router, ifdown of output interface (the interface on PC2 side)
3. On the router, manual destroy of connection in kernel
4. On the router, ifup of output interface
5. wait for iperf to start sending packets again. The connection between the endpoints is always established.
Quoted 4 lines
Quoted 2 lines
Between 3 and 4 I receive a
NFCT_DESTROYevent (fromlibnetfilter_conntrack).
During 5. I get two events, bothNFCT_UPDATEs, the first with conntrack status CONFIRMED/SEEN_REPLY, the second with conntrack status ASSURED. Both are missing the TCP protoinfo. In this is my problem, the kernel correctly picked up the on going TCP connection but never sends enough information to userspace about it.Good analysis. Since Linux kernel 2.6.30, you should see a
NFCT_NEWevent (for conntracks that have been created via ctnetlink) before you get the twoNFCT_UPDATEevents. TheNFCT_NEWcontains the TCP protocol state.
Oh, I read your email badly, sorry. Your using the pickup facility of the TCP tracking.
I cannot reproduce the problem following your steps, I'm using conntrack -F, the loopback device and some netcat commands:
[DESTROY] tcp 6 src=127.0.0.1 dst=127.0.0.1 sport=58383 dport=8000 packets=2 bytes=112 src=127.0.0.1 dst=127.0.0.1 sport=8000 dport=58383 packets=1 bytes=60 [ASSURED]
[NEW] tcp 6 432000 ESTABLISHED src=127.0.0.1 dst=127.0.0.1
sport=58383 dport=8000 [UNREPLIED] src=127.0.0.1 dst=127.0.0.1 sport=8000 dport=58383
[UPDATE] tcp 6 432000 src=127.0.0.1 dst=127.0.0.1 sport=58383
dport=8000 src=127.0.0.1 dst=127.0.0.1 sport=8000 dport=58383
[UPDATE] tcp 6 300 src=127.0.0.1 dst=127.0.0.1 sport=58383
dport=8000 src=127.0.0.1 dst=127.0.0.1 sport=8000 dport=58383 [ASSURED]
I'm using a git snapshot from the current Linux kernel development version.