Taming the Wild Netfilter

For those of you who have taken the plunge and upgraded from kernel 2.2.X (or even 2.0.X) to 2.4.X, congratulations. If, like a number of folks, you're running some form of firewall using either ipchains or ipfwadm, your scripts may work fine. But sooner or later you're probably going to want to upgrade.

In the 2.4.X kernels, Rusty Russell, the Linux packet-filter guru, and his crew of coders have implemented Netfilter into the kernel. Netfilter is the replacement for ipchains or ipfwadm. Fortunately, Netfilter permits you to keep using ipchains or ipfwadm until you can come to grips with iptables by adding a compatibility layer via a kernel module that permits these older packet filters to run. But Netfilter has so many exciting new additions, you'll want to convert those rules as soon as possible. One word of caution, though, if you load the ipchains or ipfwadm modules, you can't load ip_tables (and vice versa). So it's all or nothing. After reading this article, however, making the change should be easy.

For those new to packet filtering, ignore the ipchains translations and use the iptables examples. While not all ipchains commands and options will be translated to iptables, this text should provide a good idea about how to construct a packet-filter firewall by translating ipchains commands into iptables commands.

The reason you'll want to upgrade to Netfilter is because it, unlike ipchains or ipfwadm, is stateful. What this means is it can track connections and permit incoming responses to outgoing requests without creating gaping holes in the firewall. The connection tracking opens a specific, temporary hole for responses and only from the contacted server. We'll see how this works later. The drawback is that with connection tracking in use, Netfilter will need to use a little more memory because the connections are tracked in RAM. So your 4MB 386-16 may no longer be up to the job, depending on your filtering requirements.

The actual Netfilter implementation is broken into two parts, the kernel portion known as Netfilter and the userland tool that interfaces with Netfilter and creates the rulesets, iptables. Both are required to implement your packet-filtering firewall.

First, we'll concentrate on the kernel portion. Netfilter includes support for IPv4 and IPv6. It does not, however, filter any other protocols, so your firewall should not run IPX, AppleTalk or any other protocol that might be used to circumvent iptables rules. Similarly, you must not enable the kernel fast-switching option. This item is one of the last ones in the network options section of the kernel's configuration menu. The code permits fast switching at a low level in the IP stack. The Netfilter code resides at a much higher level, so fast switching effectively bypasses Netfilter completely.

In order to get started using Netfilter, you'll need to have your kernel compiled for Netfilter support. Most distributions include this support by default, so a quick test is in order. If you can insert the module ip_tables, then you won't need to worry about this section. As root, run the command

modprobe ip_tables

Then run

lsmod | grep ip_tables

Enabling Netfilter in Your Kernel

If you've built and installed all the modules, all modules will auto-install when a rule is entered except ip_tables, ip_nat_ftp and ip_conntrack_ftp. These can be loaded either manually or as part of your iptables startup script.

A full build and installation of Netfilter produces a large number of modules, but most firewalls will only use a few. The modules that aren't loaded aren't taking up memory, so don't worry about what you don't use.

Your distribution may have installed iptables, and it almost certainly did if your kernel has Netfilter support. But if you want the very latest, you'll probably have to get it from the Netfilter site. Netfilter is available at netfilter.filewatcher.org. Download and compile it according to the instructions in the INSTALL file. The following instructions assume the kernel sources are in /usr/src/linux. If not, adjust the following instructions appropriately. If you need to run

make pending-patches KERNEL_DIR=/usr/src/linux

or

make patch-o-matic KERNEL_DIR=/usr/src/linux

After running

make KERNEL_DIR=/usr/src/linux

run

make install KERNEL_DIR=/usr/src/linux