Route-based VPN using VTI

Route-based VPNs using libreswan

VPN tunnels are normally set up based on an IPsec policy. This is called a policy-based VPN. In libreswan, these policies are specified with leftsubnet= and rightsubnet= and optionally also with leftprotoport= and rightprotport=. Libreswan allow you to setup a route-based VPN. This is basically a policy-based VPN with leftsubnet=0.0.0.0/0 and rightsubnet=0.0.0.0/0. Then a special Virtual Tunnel Interface ("VTI") device is created that is attached to the IPsec policy. Now, whenever a packet is routed into this VTI device, it will be encrypted. If the packet's route misses the interface, the packet leaves in the clear. While this type of setup is much less secure, it is easier to manage because you just need to update the routing table instead of adding or modifying IPsec policies. While libreswan supported this with KLIPS using the ipsec0 interface, when using XFRM/NETKEY this was not supported. As of libreswan-3.18, this is now supported using the Linux VTI interface and network MARKing.

An additional advantage of using libreswan with VTI is that you have a real interface (unlike the nflogXX interface) that supports firewalling with iptables, running tcpdump, etc. It even fixes tcpdump on the non-VTI device, so it really functions as the old KLIPS ipsec0 interface.

New ipsec.conf keywords for VTI support

• mark=XX/0xffffffff (which mark to use for the IPsec SA policy)
• markin= / markout= (same as above but set different for inbound/outbound)
• vti-interface=<name> (set the interface name of the VTI device to use)
• vti-routing=yes|no (whether routes should automatically be created into the VTI device)
• vti-shared=yes|no (if not shared, and not a template connection with %any, delete the VTI device when tunnel goes down)

Creating a virtual ethernet connection

An example configuration is shown below. It is a regular VPN connection with the additional options to turn it into a route-based VPN.

conn routed-vpn
    left=192.1.2.23
    right=192.1.2.45
    authby=rsasigkey
    leftsubnet=0.0.0.0/0
    rightsubnet=0.0.0.0/0
    auto=start
    # route-based VPN requires marking and an interface
    mark=5/0xffffffff
    vti-interface=vti01
    # do not setup routing because we don't want to send 0.0.0.0/0 over the tunnel
    vti-routing=no

This will create the vti01 interface. If you want to encrypt all traffic to 10.0.0.0/8 using this VPN, simply run:

ip route add 10.0.0.0/8 dev vti01

You can also use more complicated routing using "ip rule", shorewall or anything else.

If you want to see the pre-encrypt and post-decrypt traffic, run:

tcpdump -i vti01 -n

And you can add firewall rules if you want:

# do not allow IRC traffic on port 6666
iptables -I INPUT -j DROP -p tcp --dport 6666 -i vti01

To see the post-encrypt and pre-decrypt (assuming your external interface is en0), run:

tcpdump -i en0 -n esp or udp port 4500

You can see information about vti tunnels using:

ip tunnel show
ip -s tunnel show
ifconfig vti01

Create a single VTI device for all VPN clients

If you run a VPN server, it is difficult to monitor all VPN connections using tcpdump because it mixes up encrypted and unencrypted traffic, and doesn't show all packets due to the way XFRM/NETKEY steals the packet for encryption. However, if you use a VTI device, all pre-encrypt and post-decrypt traffic appears on the VTI interface. All post-encrypt and pre-decrypt traffic appears on the regular physical interface.

conn roadwarriors
    # Regular certificate based VPN server
    left=1.2.3.4
    leftsubnet=0.0.0.0/0
    right=%any
    rightaddresspool=10.0.1.0/24
    authby=rsasig
    leftcert=mycert
    leftid=%fromcert
    auto=add
    rekey=no
    # Create route-based VPN using VTI
    mark=12/0xffffff
    vti-interface=vti02
    vti-routing=yes

Now you can monitor all connected VPN clients traffic by running:

tcpdump -i vti02 -n

One minor issue is that the vti02 device will not be created until the first client connects. And so you cannot preload it with firewall rules. You can manually create the VTI device to ensure it exists before you start your firewall and libreswan by running:

# the local IP and key mark must match the above configuration
ip tunnel add vti02 local 1.2.3.4 remote 0.0.0.0 key 12
ip link set vti02 up

Setting up a route-based VPN with Amazon

[ to be filled in ]

VTI related issues

This functionality is all very new, so not all issues or features have been resolved yet. This is a list of known interesting things.

updown script

The standard updown script (_updown.netkey) tries to autodetect what scenario is being deployed and will reconfigure the VTI interface and network options accordingly. As this feature is still new, we expect it to not be perfect yet. If your scenario is not covered, please contact the developers and explain your use case.

interface options

Each interface has a standard set of options associated with it. These can be found in /proc/sys/net/ipv4/conf/vtiXXX/*. The default updown script sets some of these (rp_filter, forwarding, policy_disable) but specific use cases might require different settings. Setting disable_xfrm will cause the interface to completely fail, so do not do that.

MTU

We noticed different kernels create different MTU sizes for new VTI devices. Currently the MTU is not set by libreswan.

combing different IPsec gateways into one VTI interface

You cannot yet have a vti tunnel that uses local 0.0.0.0 and remote 0.0.0.0. Kernel patches for these are pending. This means that you cannot yet combine two different gateways on the same VTI device, eg:

# This configuration DOES NOT work
conn west
     left=1.2.3.4
     right=6.7.8.9
     [...]
     mark=10/0xffffff
     vti-interface=vti01
     vti-routing=yes

conn east
     left=10.11.12.13
     right=9.8.7.6
     [...]
     mark=10/0xffffff
     vti-interface=vti01
     vti-routing=yes

because this would require a different setting for the "local" and "remote" option in the "ip tunnel" command, which is currently not possible. You can enter 0.0.0.0 for both but that is currently not working as expected pending some kernel patches Once the kernel has been made to work with this, we will update this page.