In order for this to work, your system needs to both answer and perform multicast DNS queries. The two pieces you need for this are:

• mDNSResponder: From Apple, and covered under the APSL, but included in [Porchdog Software]‘s Howl package.
• nss-mdns: Written by Lennart Poettering, provides a plugin for the GNU Name Service Switch (NSS) functionality of the GNU C Library (glibc) providing host name resolution via Multicast DNS

As of 0.3, nss-mdns is already Debian-ready, simply build a package executing fakeroot dpkg-buildpackage from the unpacked tarball. There is an mDNSResponder package in Sarge which is rumored to be removed shortly due to license problems with APSL, so grab a copy soon.

Once these are installed, simply add mdns to the hosts line of /etc/nsswitch.conf and you’re done.

You will now be able to resolve “.local” hosts from your Debian system and also resolve this system’s hostname on other local hosts. By editing /etc/mdnsresponder/mDNSResponder.conf, you can advertise any service you want.

The Shoreline Firewall has full support for declaratively defining IPsec tunnels and hosts under the new IPsec support. To do this however, you will need the policy match netfilter extension from the Patch-o-Matic NG and also 4 other patches:

• ipsec-01-output-hooks.patch
• ipsec-02-input-hooks.patch
• ipsec-03-policy-lookup.patch
• ipsec-04-policy-checks.patch

While these patches are in the POM-NG, they will not cleanly apply to linux 2.6.9. Ronald Moesbergen was nice enough to port the patches are are available from the list archive.

To apply the patches, simply run patch -p1 < ipsec-* from the top of our kernel source tree. Also copy the policy match extension’s files from the POM-NG directory into your kernel tree and insert the contents of the two .ladd files into the relevant Makefile and Kconfig file. Then reconfigure, enabling the policy match, and rebuild. Then follow the documentation.

If I have time, I’ll try and add some more information here.

Since I was unable to find a clear description of setting up an IPSec-based VPN using the native IPSec stack from USAGI and KAME userland tools that are part of Linux 2.6.x, I thought I would post a short explanation.

Software

• Linux Kernel 2.6.8 from Debian Sarge
• ipsec-tools 0.3.3 from Debian Sarge
• racoon 0.3.3 from Debian Sarge (This package includes racoon-tool.)

Network Topology

I am going to explain a very straightforward topology often found in the field. You have two networks, network A (10.0.0.0/24) and network B (10.0.1.0/24). Each network has router/gateway/firewall system, gateway A (10.0.0.1) and gateway B (10.0.1.1). Each of these gateway systems as an external, public, IP address: 3.3.3.3 & 2.2.2.2.

In this topology, our end goal is to allow a hosts on network A, say 10.0.0.123, to securely contact a host on network B, say 10.0.1.158, via an IPSec tunnel.

Note: In many cases, I will provide only a single example. You must ensure the other gateway is configured similarly, often by reversing the configuration. This is left as an exercise to the reader.

IPSec

Setting up IPSec involves two steps: keying and policy. Normally, you use racoon to provide keying and setkey to establish policy. I will instead use racoon-tool as convenient shortcut. This tool was written by the Debian Maintainer of the racoon package in order to emulate some of the nice configuration syntaxt of the FreeS/WAN, an older IPSec implementation for Linux. It dynamically generates a racoon.conf (/var/lib/racoon/racoon.con) and also sets up the policy based on a single configuration file.

First, we simply state that racoon should use the notify priority when sending messages to syslogd:

global:
        log: notify

It is helpful to create a %default peer and connection in order to avoid duplication of configuration directives:

peer(%default):
        verify_identifier: on
        hash_algorithm[0]: sha1
        encryption_algorithm[0]: aes
connection(%default):
        src_ip: 3.3.3.3

The above src_ip directive simply says that our connections will be using our public IP.

Next, we define our peer, gatewayB, identifying it by address:

peer(2.2.2.2):
        peers_identifier: address

Finally, we define a policy so that packets from networkA to networkB are encrypted via gatewayB:

connection(to-gatewayB):
        dst_ip: 2.2.2.2
        src_range: 10.0.0.0/24
        dst_range: 10.0.1.0/24
        admin_status: enabled

Unfortunately, I have no idea what the admin_status: enabled directive does, or even whether it is required.

You must also add a key to /etc/racoon/psk.txt for the remote gateway:

# Entry for gatewayB
2.2.2.2 0x2eba016ffc2314869ae9f9a3b8901a173242f0c8

A randomly generated key is best, and can be created with the following command:

$ dd if=/dev/random count=20 bs=1 | xxd -ps

(The xxd command is part of the vim package.)

Make sure that you can the racoon package configured to use racoon-tool by either editing /etc/defaults/racoon or reconfiguring the package. Also, you should probably reload the tool by executing /etc/init.d/racoon reload. You can then check that your policy is in effect by running setkey -DP. You should get something like the following, followed by a bunch of default policies:

10.0.1.0/24[any] 10.0.0.0/24[any] any
        in ipsec
        esp/tunnel/2.2.2.2-3.3.3.3/unique#16385
        created: Nov 18 23:01:24 2004  lastused:
        lifetime: 0(s) validtime: 0(s)
        spid=1512 seq=9 pid=9800
        refcnt=1
10.0.0.0/24[any] 10.0.1.0/24[any] any
        out ipsec
        esp/tunnel/3.3.3.3-2.2.2.2/unique#16384
        created: Nov 18 23:01:24 2004  lastused: Nov 18 23:05:31 2004
        lifetime: 0(s) validtime: 0(s)
        spid=1505 seq=8 pid=9800
        refcnt=1

Now that the policy is in effect, the kernel will ask the racoon daemon for a security association (SA), when it needs to deal with a packet matching the policy. Racoon will negotiate an SA with the remote gateway on the fly. (This will not happen until after we’ve setup the routing, below, but you can see it by using the setkey -D command.)

Routing

In order for a packet to match the policy we have setup, and also be forwarded to the remove gateway, we must adjust the routing table using the ip command (Provided by the Debian iproute2 package.).

The policy we have setup, says that packets coming from 10.0.0.0/24 and going to 10.0.1.0/24 must be encrypted and authenticated. This is achieved by adding a slightly weird entry to the routing table on gatewayA:

$ ip route add 10.0.1.0/24 via 2.2.2.2 src 10.0.0.1

And on gatewayB:

$ ip route add 10.0.0.0/24 via 3.3.3.3 src 10.0.1.1

Once you have added the new entries to the routing tables on both gateways, we test the tunnel. If you ping 10.0.1.0 from gatewayA, and then execute setkey -D, you will see the new SA that has been automatically created.

This is the routing table you should have on gateway “A”:

root@gatewayA:/tmp# ip route show
3.3.3.0/24 dev eth1  proto kernel  scope link  src 3.3.3.3
10.0.1.0/24 via 3.3.3.1 dev eth1  src 10.0.0.1
10.0.0.0/24 dev eth0  proto kernel  scope link  src 10.0.0.1
default via 3.3.3.3 dev eth1

Followed by the routing table present on gateway “B”:

root@gatewayB:/tmp# ip route show
2.2.2.0/24 dev eth1  proto kernel  scope link  src 2.2.2.2
10.0.1.0/24 dev eth0  proto kernel  scope link  src 10.0.1.1
10.0.0.0/24 via 2.2.2.1 dev eth1  src 10.0.0.1.1
default via 2.2.2.1 dev eth1

In the above examples, 2.2.2.1 & 3.3.3.1 are the default gateways that connect your public networks. (The routers usually provided by your ISP.)

Note: if you have any firewall rules enabled, you are going to have hassles. Notice that the IPSec tunnel doesn’t have it’s own device: the encrypted packets are going through a public interface. This will confuse any firewall rules that expect to be able to categorize packets by interface. I intend to figure out a good solution and post it at a later date.

Update: I now have a fully working IPsec gateway/router/firewall. See this other post.

Update: It seems that when the external addresses of the gateways on are different subnets, which was not the case during my testing, you must add a route like ip route add OTHER_NETWORK via LOCAL_DEFAULT_GW src INTERNAL_IP. Otherwise, you will get an error: RTNETLINK answers: Network is unreachable. Please leave a comment if you can clarify this.