Skip to contentSkip to conversation

tcp connection tracking 2.4.19

PATCH In review Martin Renold 3 participants 6 messages Oct 8, 2002 – Oct 10, 2002

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 tcp connection tracking 2.4.19

hi,

There is a bug in the stable 2.4.19 kernel in the ip_conntrack code that allows the final ACK of a SYN - SYN/ACK - ACK tcp handshake to establish an ASSURED connection even if it has a wrong sequence number. The current code only checks the ACK number.

This allows a DoS attack that will make it impossible to establish *real* connections for some days, once the maximum is reached. Somebody sent me an exploit:

http://old.homeip.net/martin/cdos.tgz

So I wrote a simple patch against 2.4.19, but I must admit that I do not really understand the code around it, especially why it does not mark such a packet as invalid (I'm new to most things here).

include/linux/netfilter_ipv4/ip_conntrack_tcp.h +3 -2
diff -urN -X dontdiff kernel-source-2.4.19.origin/include/linux/netfilter_ipv4/ip_conntrack_tcp.h kernel-source-2.4.19.patch/include/linux/netfilter_ipv4/ip_conntrack_tcp.h
@@ -24,8 +24,9 @@
 {
 	enum tcp_conntrack state;
 
-	/* Poor man's window tracking: sequence number of valid ACK
-           handshake completion packet */
+	/* Poor man's window tracking: expected sequence and acknowledge 
+	   number of valid ACK handshake completion packet */
+	u_int32_t handshake_seq;
 	u_int32_t handshake_ack;
 };
 
net/ipv4/netfilter/ip_conntrack_proto_tcp.c +3 -0
diff -urN -X dontdiff kernel-source-2.4.19.origin/net/ipv4/netfilter/ip_conntrack_proto_tcp.c kernel-source-2.4.19.patch/net/ipv4/netfilter/ip_conntrack_proto_tcp.c
@@ -180,6 +180,8 @@
 	if (oldtcpstate == TCP_CONNTRACK_SYN_SENT
 	    && CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY
 	    && tcph->syn && tcph->ack)
+		conntrack->proto.tcp.handshake_seq
+			= tcph->ack_seq;
 		conntrack->proto.tcp.handshake_ack
 			= htonl(ntohl(tcph->seq) + 1);
 	WRITE_UNLOCK(&tcp_lock);
@@ -196,6 +198,7 @@
 		if (oldtcpstate == TCP_CONNTRACK_SYN_RECV
 		    && CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL
 		    && tcph->ack && !tcph->syn
+		    && tcph->seq == conntrack->proto.tcp.handshake_seq
 		    && tcph->ack_seq == conntrack->proto.tcp.handshake_ack)
 			set_bit(IPS_ASSURED_BIT, &conntrack->status);
 
Signature · 2 lines
Thunder's just a noise, boys, lightnin' does the work
-- (Chad Brock/John Hadley/Kelly Garrett)
#1 lore raw

Hello Martin,

Quoted 4 lines

There is a bug in the stable 2.4.19 kernel in the ip_conntrack code that allows the final ACK of a SYN - SYN/ACK - ACK tcp handshake to establish an ASSURED connection even if it has a wrong sequence number. The current code only checks the ACK number.

Yes, and more than that. You can remove ESTABLISHED entries in the conntrack table by sending packets with the RST flag set and matching the template <srcIP, srcPORT, dstIP, dstPORT>.

Quoted 3 lines

This allows a DoS attack that will make it impossible to establish *real* connections for some days, once the maximum is reached. Somebody sent me an exploit:

:) You should probably send stuff like that to the netfilter-devel ml. Besides that it isn't really an exploit.

Quoted 5 lines

http://old.homeip.net/martin/cdos.tgz

So I wrote a simple patch against 2.4.19, but I must admit that I do not really understand the code around it, especially why it does not mark such a packet as invalid (I'm new to most things here).

You might want to take a look at the TCP window tracking patch which comes with the latest pom. It's part of the extra patches. This will solve the problems for you.

Quoted 13 lines

diff -urN -X dontdiff kernel-source-2.4.19.origin/net/ipv4/netfilter/ip_conntrack_proto_tcp.c kernel-source-2.4.19.patch/net/ipv4/netfilter/ip_conntrack_proto_tcp.c

net/ipv4/netfilter/ip_conntrack_proto_tcp.c +2 -0
@@ -180,6 +180,8 @@
 	if (oldtcpstate == TCP_CONNTRACK_SYN_SENT
 	    && CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY
 	    && tcph->syn && tcph->ack)
+		conntrack->proto.tcp.handshake_seq
+			= tcph->ack_seq;
 		conntrack->proto.tcp.handshake_ack
 			= htonl(ntohl(tcph->seq) + 1);
 	WRITE_UNLOCK(&tcp_lock);

net/ipv4/netfilter/ip_conntrack_proto_tcp.c line 204

Quoted 7 lines
net/ipv4/netfilter/ip_conntrack_proto_tcp.c +1 -0
@@ -196,6 +198,7 @@
 		if (oldtcpstate == TCP_CONNTRACK_SYN_RECV
 		    && CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL
 		    && tcph->ack && !tcph->syn
+		    && tcph->seq == conntrack->proto.tcp.handshake_seq
 		    && tcph->ack_seq == conntrack->proto.tcp.handshake_ack)
 			set_bit(IPS_ASSURED_BIT, &conntrack->status);
 

Welcome to the world of almost-stateful packet filtering. Hey, other than that, the 3wahas 'exploit' is old. Also don't I understand why they claim that SYN cookies prevent syn flooding. Next time you meet someone of the guys, tell them about the backlog queue.

Cheers,
Roberto Nibali, ratz

Signature · 1 line
echo '[q]sa[ln0=aln256%Pln256/snlbx]sb3135071790101768542287578439snlbxq'|dc
#2 lore raw
Quoted 5 lines · Roberto Nibali

Welcome to the world of almost-stateful packet filtering. Hey, other than that, the 3wahas 'exploit' is old. Also don't I understand why they claim that SYN cookies prevent syn flooding. Next time you meet someone of the guys, tell them about the backlog queue.

"When syncookies are enabled the packets are still answered and this value [tcp_max_syn_backlog] is effectively ignored." -- From tcp(7) manpage.

The whole point of syncookies is to negate the need for a backlog queue.

Or did I miss your point?

Signature · 3 lines
// Gianni Tedesco (gianni at ecsc dot co dot uk)
lynx --source www.scaramanga.co.uk/gianni-at-ecsc.asc | gpg --import
8646BE7D: 6D9F 2287 870E A2C9 8F60 3A3C 91B5 7669 8646 BE7D
#3 lore raw

Hi,

Quoted 3 lines

"When syncookies are enabled the packets are still answered and this value [tcp_max_syn_backlog] is effectively ignored." -- From tcp(7) manpage.

Fair enough. I thought that last time I checked with the code the SYN cookie functionality would only kick in _after_ the backlog queue is full.

Quoted 1 line

The whole point of syncookies is to negate the need for a backlog queue.

Well, after a successful match of the MSS encoded part or the cookie, you add it back to the SYN queue. But yes, the backlog queue is indeed omited.

Quoted 1 line

Or did I miss your point?

Well, my point should have been stated more clearly. It is simply that SYN cookies do not prevent you from being SYN flooded. They provide you, from a user perspective view, a mean to still be able to log in onto your server under a SYN flood because you will send legitimate ACKs and because your connection will not be dropped.

It doesn't prevent SYN flooding, although I just checked back with ../Documentation/networking/ip-sysctrl.txt:

tcp_syncookies - BOOLEAN
Only valid when the kernel was compiled with CONFIG_SYNCOOKIES Send out syncookies when the syn backlog queue of a socket overflows. This is to prevent against the common 'syn flood attack'
Default: FALSE

Note, that syncookies is fallback facility.
It MUST NOT be used to help highly loaded servers to stand against legal connection rate. If you see synflood warnings in your logs, but investigation shows that they occur because of overload with legal connections, you should tune another parameters until this warning disappear.
See: tcp_max_syn_backlog, tcp_synack_retries,
tcp_abort_on_overflow.

syncookies seriously violate TCP protocol, do not allow to use TCP extensions, can result in serious degradation of some services (f.e. SMTP relaying), visible not by you, but your clients and relays, contacting you. While you see synflood warnings in logs not being really flooded, your server is seriously misconfigured.

The best thing is to go back and check with the actual implementation. I'm just checking on ../net/ipv4/tcp_ipv4.c:tcp_v4_conn_request():

[...]

        if (tcp_synq_is_full(sk) && !isn) {
#ifdef CONFIG_SYN_COOKIES
                if (sysctl_tcp_syncookies) {
                        want_cookie = 1;
                } else
#endif
                goto drop;
        }

        /* Accept backlog is full. If we have already queued enough
         * of warm entries in syn queue, drop request. It is better than
         * clogging syn queue with openreqs with exponentially increasing
         * timeout.
         */
        if (tcp_acceptq_is_full(sk) && tcp_synq_young(sk) > 1)
                goto drop;

[...]

Best regards and sorry for the confusion,
Roberto Nibali, ratz

Signature · 1 line
echo '[q]sa[ln0=aln256%Pln256/snlbx]sb3135071790101768542287578439snlbxq'|dc
#4 lore raw
Quoted 8 lines · Roberto Nibali

Hi,

Quoted 3 lines

"When syncookies are enabled the packets are still answered and this value [tcp_max_syn_backlog] is effectively ignored." -- From tcp(7) manpage.

Fair enough. I thought that last time I checked with the code the SYN cookie functionality would only kick in _after_ the backlog queue is full.

It does. When using syn cookies you cant use some of the new advanced features of tcp. Linux uses the backlog queue when not under attack. When the queue overflows it just uses cookies - but can still accept connections.

Quoted 14 lines
Quoted 1 line

The whole point of syncookies is to negate the need for a backlog queue.

Well, after a successful match of the MSS encoded part or the cookie, you add it back to the SYN queue. But yes, the backlog queue is indeed omited.

Quoted 1 line

Or did I miss your point?

Well, my point should have been stated more clearly. It is simply that SYN cookies do not prevent you from being SYN flooded. They provide you, from a user perspective view, a mean to still be able to log in onto your server under a SYN flood because you will send legitimate ACKs and because your connection will not be dropped.

Quoted 10 lines

It doesn't prevent SYN flooding, although I just checked back with ../Documentation/networking/ip-sysctrl.txt:
[snip]
tcp_abort_on_overflow.
syncookies seriously violate TCP protocol, do not allow to use TCP extensions, can result in serious degradation of some services (f.e. SMTP relaying), visible not by you, but your clients and relays, contacting you. While you see synflood warnings in logs not being really flooded, your server is seriously misconfigured.

I don't think these statements are entirely true. While it is true that you can't use things like window scaling or SACK - syncookies 100% successfully stop syn flood attacks.

The attack is that if you fill the syn backlog queue with bogus requests then legitimate clients can no longer connect. The syn flood attack isn't "your legitimate connections wont be able to use window scaling".

Signature · 3 lines
// Gianni Tedesco (gianni at ecsc dot co dot uk)
lynx --source www.scaramanga.co.uk/gianni-at-ecsc.asc | gpg --import
8646BE7D: 6D9F 2287 870E A2C9 8F60 3A3C 91B5 7669 8646 BE7D
#5 lore raw
Quoted 5 lines
Quoted 2 lines

Fair enough. I thought that last time I checked with the code the SYN cookie functionality would only kick in _after_ the backlog queue is full. It does. When using syn cookies you cant use some of the new advanced

features of tcp. Linux uses the backlog queue when not under attack. When the queue overflows it just uses cookies - but can still accept connections.

That was my understanding so far. So then I haven't gone crazy, thank god.

Quoted 3 lines

I don't think these statements are entirely true. While it is true that you can't use things like window scaling or SACK - syncookies 100% successfully stop syn flood attacks.

Someone needs to adjust the text then.

Quoted 3 lines

The attack is that if you fill the syn backlog queue with bogus requests then legitimate clients can no longer connect. The syn flood attack isn't "your legitimate connections wont be able to use window scaling".

I completely agree with this definition of SYN flooding and then I will also say that you can 'stop' SYN flooding, well, at least you give legitimate clients a real chance to still successfully connect to your service, while it is under a SYN flood. I think we agree now. It should only be remarked that the line is still flooded, thus the wording "100 stop SYN flood" is simply inappropriate in my eyes. It's all a matter of definition.

Regards,
Roberto Nibali, ratz

Signature · 1 line
echo '[q]sa[ln0=aln256%Pln256/snlbx]sb3135071790101768542287578439snlbxq'|dc