About me: My name is Solène Rapenne, pronouns she/her. I like learning and sharing knowledge. Hobbies: '(BSD OpenBSD Qubes OS Lisp cmdline gaming security QubesOS internet-stuff). I love percent and lambda characters. OpenBSD developer solene@. No AI is involved in this blog.
Contact me: solene at dataswamp dot org or @solene@bsd.network (mastodon).
You can sponsor my work financially if you want to help me writing this blog and contributing to Free Software as my daily job.
Written by Solène, on 09 October 2021.
Tags:
#openbsd
#wireguard
#vpn
We want all our network traffic to go through a WireGuard VPN tunnel automatically, both WireGuard client and server are running OpenBSD, how to do that? While I thought it was simple at first, it soon became clear that the "default" part of the problem was not easy to solve, fortunately there are solutions.
This guide should work from OpenBSD 6.9.
ifconfig man page, WireGuard section
For this setup I assume we have a server running OpenBSD with a public IP address (1.2.3.4 for the example) and an OpenBSD computer with Internet connectivity.
Because you want to use the WireGuard tunnel as the default route, you can't define a default route through WireGuard as this, that would prevent our interface to reach the WireGuard endpoint to make the tunnel working. We could play with the routing table by deleting the default route found on the interface, create a new route to reach the WireGuard server and then create a default route through WireGuard, but the whole process is fragile and there is no right place to trigger a script doing this.
Instead, you can assign the network interface used to access the Internet to the rdomain 1, configure WireGuard to reach its remote peer through rdomain 1 and create a default route through WireGuard on the rdomain 0. Quick explanation about rdomain: they are different routing tables, default is rdomain 0 but you can create new routing tables and run commands using a specific routing table with "route -T 1 exec ping perso.pw" to make a ping through rdomain 1.
+-------------+
| server | wg0: 192.168.10.1
| |---------------+
+-------------+ |
| public IP |
| 1.2.3.4 |
| |
| |
/\/\/\/\/\/\/\ |WireGuard
| internet | |VPN
\/\/\/\/\/\/\/ |
| |
| |
|rdomain 1 |
+-------------+ |
| computer |---------------+
+-------------+ wg0: 192.168.10.2
rdomain 0 (default)
The configuration process will be done in this order:
Our WireGuard server will accept connections on address 1.2.3.4 at the UDP port 4433, we will use the network 192.168.10.0/24 for the VPN, the server IP on WireGuard will be 192.168.10.1 and this will be our future default route.
We will make a simple script to generate the configuration file, you can easily understand what is being done. Replace "1.2.3.4 4433" by your IP and UDP port to match your setup.
PRIVKEY=$(openssl rand -base64 32)
cat <<EOF > /etc/hostname.wg0
wgkey $PRIVKEY
wgpeer wgendpoint 1.2.3.4 4433 wgaip 0.0.0.0/0
inet 192.168.10.2/24
up
EOF
# start interface so you can get the public key
# we should have an error here, this is normal
sh /etc/netstart wg0
PUBKEY=$(ifconfig wg0 | grep 'wgpubkey' | cut -d ' ' -f 2)
echo "You need $PUBKEY to setup the remote peer"
Like we did on the computer, we will use a script to configure the server. It's important to get the PUBKEY displayed in the previous step.
PUBKEY=PASTE_PUBKEY_HERE
PRIVKEY=$(openssl rand -base64 32)
cat <<EOF > /etc/hostname.wg0
wgkey $PRIVKEY
wgpeer $PUBKEY wgaip 192.168.10.0/24
inet 192.168.10.1/24
wgport 4433
up
EOF
# start interface so you can get the public key
# we should have an error here, this is normal
sh /etc/netstart wg0
PUBKEY=$(ifconfig wg0 | grep 'wgpubkey' | cut -d ' ' -f 2)
echo "You need $PUBKEY to setup the local peer"
Keep the public key for next step.
You want to enable NAT so you can reach the Internet through the server using WireGuard, edit /etc/pf.conf to add the following line (after the skip lines):
pass out quick on egress from wg0:network to any nat-to (egress)
Reload with "pfctl -f /etc/pf.conf".
NOTE: if you block all incoming traffic by default, you need to open UDP port 4433. You will also need to either skip firewall on wg0 or configure PF to open what you need. This is beyond the scope of this guide.
We need to enable IP forwarding because we will pass packets from an interface to another, this is done with "sysctl net.inet.ip.forwarding=1" as root. To make it persistent across reboot, add "net.inet.ip.forwarding=1" to /etc/sysctl.conf (you may have to create the file).
From now, the server should be ready.
Edit /etc/hostname.wg0 and paste the public key between "wgpeer" and "wgaip", the public key is wgpeer's parameter. Then run "sh /etc/netstart wg0" to reconfigure your wg0 tunnel.
After this step, you should be able to ping 192.168.10.1 from your computer (and 192.168.10.2 from the server). If not, please double check the WireGuard and PF configurations on both side.
This simple setup for the default route will truly make WireGuard your default route. You have to understand services listening on all interfaces will only attach to WireGuard interface because it's the only address in rdomain 0, if needed you can use a specific routing table for a service as explained in rc.d man page.
Replace the line "up" with the following:
wgrtable 1
up
!route add -net default 192.168.10.1
Your configuration file should look like this:
wgkey YOUR_KEY
wgpeer YOUR_PUBKEY wgendpoint REMOTE_IP 4433 wgaip 0.0.0.0/0
inet 192.168.10.2/24
wgrtable 1
up
!route add -net default 192.168.10.1
Now, add "rdomain 1" to your network interface used to reach the Internet, in my setup it's /etc/hostname.iwn0 and it looks like this.
join network wpakey superprivatekey
join home wpakey notsuperprivatekey
rdomain 1
up
autoconf
Now, you can restart network with "sh /etc/netstart" and all the network should pass through the WireGuard tunnel.
Because you may use a nameserver in /etc/resolv.conf that was provided by your local network, it's not reachable anymore. I highly recommend to use unwind (in every case anyway) to have a local resolver, or modify /etc/resolv.conf to use a public resolver.
unwind can be enabled with "rcctl enable unwind" and "rcctl start unwind", from OpenBSD 7.0 you should have resolvd running by default that will rewrite /etc/resolv.conf if unwind is started, otherwise you need to write "nameserver 127.0.0.1" in /etc/resolv.conf
If you need for some reason to run a program and not route its traffic through the VPN, it is possible. The following command will run firefox using the routing table 1, however depending on the content of your /etc/resolv.conf you may have issues resolving names (because 127.0.0.1 is only reachable on rdomain 0!). So a simple fix would be to use a public resolver if you really need to do so often.
route -T 1 exec firefox
route man page about exec command
If you are behind a NAT you may need to use the KeepAlive option on your WireGuard tunnel to keep it working. Just add "wgpka 20" to enable a KeepAlive packet every 20 seconds in /etc/hostname.wg0 like this:
wgpeer YOUR_PUBKEY wgendpoint REMOTE_IP 4433 wgaip 0.0.0.0/0 wgpka 20
[....]
ifconfig man page explaining wgpka parameter
WireGuard is easy to deploy but making it a default network interface adds some complexity. This is usually simpler for protocols like OpenVPN because the OpenVPN daemon can automatically do the magic to rewrite the routes (and it doesn't do it very well) and won't prevent non-VPN access until the VPN is connected.