9.3. Breaking a network in two with proxy ARP

Proxy ARP is a technique for splitting an IP network into two separate segments. Hosts on one segment can only reach hosts in the other segment through the router performing proxy ARP. If a router sits between two parts of an IP network and is not running bridging software, then routes to hosts in each segment and proxy ARP are required on the router to allow each half of the network to communicate with the other half.

Occasionally, this technique is incorrectly called proxy ARP bridging. An Ethernet bridge operates on frames and a router operates on packets. The proxy ARP router should have routes to all hosts on both segments. Once the router can reach all locally connected destinations via the correct interfaces, you can begin to configure the proxy ARP functionality.

Although proxy ARP complicates a network, a great advantage of proxy ARP technique is the greater control over IP connections between hosts.

There are two primary proxy ARP techniques. With the 2.4 kernel, it is possible to use the sysctl net/ipv4/conf/all/proxy_arp to perform proxy ARP. Alternatively, manual population of the ARP table reaches the same end.

The key part of the correct functioning of proxy ARP in a network is that the host breaking a network into two parts has correct routes for all destinations in both halves of the network. If the host which has interfaces in both networks does not have an accurate routing table, IP packets will get dropped on the routing device.

One common method of breaking a network in two involves making a very small stub subnet at one end or the other of the IP range. This small subnet (maybe as small as a /30 network, with two usable IPs) makes an excellent sequestered location for a host which requires more protection or even, a generally untrusted host which shouldn't have complete access to the Ethernet to which the other machines connect.

For a practical example of this, see the relationship between the service-router, masq-gw and isolde in the network map. isolde and service-router share the same IP network, 192.168.100.0/24. If either has a packet for the other, it will generate an ARP request which should be answered by masq-gw. Naturally, masq-gw has its routes configured in such a way that both hosts are reachable from it. Thus, the packet will successfully pass through masq-gw.

Let's examine what the sequence of events is by which the packet will reach service-router from isolde. In this example, isolde will send an echo request packet to service-router. Please also refer to Section B.1, “arp for examples and command lines to create a proxy ARP configuration.

Where possible, a simplified network is easier to maintain, but occasionally, this sort of trickery is necessary. This is an excellent way to insert a firewall into the middle of a network. The firewall, naturally, has to have its routes set properly, and proxy ARP entries will be required for routers.

Now, here's a short script and configuration file which can be run as a SysVInit style script. This script provides a great deal of control over the ARP table directly so may be preferable in some cases to an alternate solution outlined below. This proxy-arp script reads the following configuration file. Each is commented heavily so it should be clear how to use them.

This chapter discussed how to break a network in twain with proxy ARP techniques. For another explanation of the same concepts, read the Proxy ARP Subnet mini-HOWTO. Available in most (all?) 2.4 kernels is built-in capability for Proxy ARP. This is documented in deeper detail above. Consider familiarizing yourself with the methods of suppressing and controling ARP through Julian Anastasov's work.