pcolladosoto/wg-ebpf

A repository containing the sample deployment used for the XVII REDIMadrid conference

Security, access control and traffic shaping though L3 VPNs

This repository contains several goodies we leveraged to illustrate the concepts set forth in the our talk at the XVII REDIMadrid conference.

We decided to take a deep look at how we can make use of WireGuard, a L3 VPN, as a solution offering security and access control for large scale networks in an efficient and escalable fashion.

Given how WireGuard integrates with the Linux kernel, we also decided to investigate how we could apply QoS techniques on the traffic being "tunelled" through the VPN with the help of tc(8).

We finished by delving into eBPF and how it can also play nicely with WireGuard to offer a completely programmable packet handling framework taking advantage of the latest networking technology the Linux kernel has to offer.

Let's take a (brief) look at these topics!

First things first: Vagrant

All the examples are built with a virtual setup in mind. The setup itself consists of 3 VirtualBox Virtual Machines (VMs) managed through Vagrant. The configuration is embedded into the Vagrantfile you'll find at the repository's root, and making use of it is as simple as running:

vagrant up

The Vagrantfile should take care of everything from starting the machines to installing necessary dependencies. It'll even deploy the WireGuard configuration we've included in the repository!

Bear in mind we have tested this deployment on a machine running macOS. According to our experience everything should be okay on Linux too, but we can't say the same about Windows...

If you find any issues let us know by, well, opening an issue :P

For help on how to install VirtualBox and Vagrant refer to this and that site, respectively.

Checking everything works

At this point you should be able to ssh into a machine. Do so and try to ping the others with address both belonging and not belonging to the VPN. You can use the following as a guide to avoid loosing track of what you've checked and what not:

# Log into the client-a machine
$ vagrant ssh client-a

# Try to ping the host machine
vagrant@client-a:~$ ping -c 1 10.0.123.1

# Try to ping vpn-core through the VM subnet
vagrant@client-a:~$ ping -c 1 10.0.123.1

# Try to ping client-b through the VM subnet
vagrant@client-a:~$ ping -c 1 10.0.123.3

# Try to ping vpn-core through the VPN
vagrant@client-a:~$ ping -c 1 192.168.4.1

# Try to ping client-b through the VPN
vagrant@client-a:~$ ping -c 1 192.168.4.3

If all the above return a line along the lines of the following, you're all set! Bear in mind <pinged-ip> is to be substituted by whatever IPv4 address you pinged with each command...

64 bytes from <pinged-ip>: icmp_seq=1 ttl=63 time=1004 ms

Playtime!

Now it's time to try out and break stuff. You can experiment with different iptables(8) rules and see whether the behaviour is the one you expected, for example.

You can also give our examples a try! Be sure to take a look at the PDF file in the repository's root to get some ideas on what you can do. We are including the tc(8) commands we invoked to modify the delay of packets traversing the VPN, for instance.

Adding delays with tc

In an effort to showcase how one can easily integrate tc(8) wit WireGuard, we decided to leverage the netem (i.e. tc-netem(8)) qdisc to add a steady $500\ ms$ delay to packets traversing the wg0 interface on vpn-core.

We achieved that with:

# We begin by displaying the configured qdisc on wg0.
    # The noqueue qdisc implies none has been set up.
vagrant@vpn-core:~$ tc -g -d -s qdisc show dev wg0
qdisc noqueue 0: root refcnt 2 
 Sent 0 bytes 0 pkt (dropped 0, overlimits 0 requeues 0) 
 backlog 0b 0p requeues 0

# We can then add a 500000 us == 500 ms delay to packets by
    # configuring netem as the root qdisc for wg0
vagrant@vpn-core:~$ sudo tc qdisc add dev wg0 root netem delay 500000

# We can see the changes did indeed take effect. At this point you
    # can check packets traversing the VPN do indeed suffer an
    # additional 500 ms delay.
vagrant@vpn-core:~$ tc -g -d -s qdisc show dev wg0
qdisc netem 8003: root refcnt 2 limit 1000 delay 500ms
 Sent 1260 bytes 15 pkt (dropped 0, overlimits 0 requeues 0) 
 backlog 0b 0p requeues 0

# When we're done, we can undo the changes by deleting the
    # netem qdisc. This will configure the noqueue qdisc again.
vagrant@vpn-core:~$ sudo tc qdisc del dev wg0 root

Inspecting and filtering packets with eBPF

We have included a small eBPF program in the wg_ebpf directory on the repository's root. The directory contains a README.md file with more information on the topic, but we just want to point out how we can easily integrate WireGuard with the eBPF ecosystem, thus opening up the door to a myriad of opportunities and applications!

Taking a look at traffic captures

We have taken some traffic captures at vpn-core generated by sending a single ping from client-a to client-b. We have included them in the traffic_captures directory. You should be able to open them up with WireShark without too much hassle. If you can, feel free to open an issue on the topic!

Conclusion

That's pretty much it from us...

If you find anything is wrong or you have any suggestions or feedback, please feel free to drop us an email at the address shown on our profile. You can also open an issue in the repository if that's your jam.

We hope you found our work interesting or, at least, amusing :)