Expanation needed for Connection Tracking with NAT One-Way
I am trying to test some research ideas and am not able to perform NAT translation on incoming packets only. I am trying to explain why, and am looking for a confirmation (or correction) of what I think is happening. I have two Routers (B and C) in my network. I am trying to send packets directly from A to D through B, but route the replies through C and B using NAT translation.
D
/ \
/ \
A --- B ---- CThe gateway route of Host D is to Router C. Router C has the following iptables entries in the NAT PREROUTING and NAT POSTROUTING table.
iptables -t nat -A PREROUTING -s D -d A -j DNAT --to B iptables -t nat -A POSTROUTING -s D -d B -j SNAT --to C
Router B has the following entries in its iptables NAT PREROUTING and POSTROUTING tables.
iptables -t nat -A PREROUTING -s C -d B -j LOG iptables -t nat -A PREROUTING -s C -d B -j DNAT --to A iptables -t nat -A POSTROUTING -s C -d A -j LOG iptables -t nat -A POSTROUTING -s C -d A -j SNAT --to D
I am able to see the first TCP SYN packet travel from A through B to D. At this point Router B has an ip_conntrack entry from A to D. I then see the reply travel from D to C. C successfully performs NAT translation, and the packet is sent to Router B with a source of C and a destination of B. I added logging to the iptables entries in Router B, and I see the packet get translated in both the PREROUTING and the POSTROUTING tables, but the packet is never sent. I never see a new conntrack entry for this packet.
I think that Router B is dropping the packet through connection tracking since it cannot add a new conntrack entry for this packet because it already has an entry for packets from A to D that do not require NAT translation. If it added a new entry, then it would not know whether to translate the next packet from A to D or send it directly to D.
Is this analysis correct? Is there anyway to send packets directly to D and router them back to A through C and B using NAT translation the way I have described using iptables or otherwise? Any explanation on what criteria connection tracking uses to drop packets in this case would be a great help.
Thanks,
Tish Best
Quoted 1 line · Tish Best
I am able to see the first TCP SYN packet travel from A through B to D. At this point Router B has an
ip_conntrackentry from A to D. I then see the reply travel from D to C. C successfully performs NAT translation, and the packet is sent to Router B with a source of C and a destination of B. I added logging to the iptables entries in Router B, and I see the packet get translated in both the PREROUTING and the POSTROUTING tables, but the packet is never sent. I never see a new conntrack entry for this packet.
Sounds logical to me that you don't see new conntrack entry for return packet. B never saw SYN sent with IP src B and dst C, so it can't relate the return packet with IP src C and dst B to anything. I don't think connection tracking works at all with asymentric routing. My guess is that return packet would end up in INVALID state (try logging "-m state --state INVALID", I guess you'll see it there).
Theoretically, connection tracking could work for asymentric routing, but it would require B and C exchanging information about states of connections (which is not possible with iptables, and I don't know of any product that has this functionality), and they would have to have insight of each others configuration (which they don't).
Signature · 3 lines
Aleksandar Milivojevic <[email protected]> Pollard Banknote Limited Systems Administrator 1499 Buffalo Place Tel: (204) 474-2323 ext 276 Winnipeg, MB R3T 1L7