1324 views
--- Title: Setting up a WireGuard® client with routing domains on OpenBSD langs: en Breaks: False --- # Setting up a WireGuard® client with routing domains on OpenBSD _Matthieu Herrb - last edited october 24, 2020._ OpenBSD-current recently got native support for the [Wireguard®](https://wireguard.com/) VPN, thanks to the work of Matt Dunwoodie and Jason A. Donefeld. Their code was merged by David Gwynne. This is great, especially since I've been using WireGuard for various purposes already. A local assocative ISP here in Toulouse is offering WireGuard based VPNs for 5€/month. In this note, I'm going to show how to setup a machine running OpenBSD to connect to this kind of VPN, using routing domains to make the VPN the default interface. This configuration can be used to get a fixed public IPv4 and IPv6 address for a self-hosted server to run any kind of internet services. ## General principle The client machine is going to connect to the internet via any kind of LAN that provides IPv4 DHCP auto-configuration. It can be an ADSL or fiber modem at home or some enterprise network. The only serious requirement is that UDP traffic on the WireGuard (51820) port goes out. This connexion is only going to be used to connect to the VPN, and will be setup in routing domain 1. The Wireguard connexion is going to be the _egress_ connexion in the default routing domain (0). In this setup there is no need to provide complicated routing or packet filter configuration. All traffic will go through the VPN "naturally" as the default routing domain only knows about the VPN interface. ### Routing Domains and routing tables For those not familiar with routing domains and tables on OpenBSD, it is recommended to read the [rdomain(4)](https://man.openbsd.org/rdomain.4) manual page. Routing domains are separated network address spaces in the OpenBSD kernel. Each routing domain has at least one routing table, separated from the routing tables of other domains. Network interfaces belong to one domain and are not usable by the others. Also, processes are running inside a routing domain, using by default the default routing table of their domain. So their network traffic can only use the interfaces in this domain. So routing domains and tables are a great way to isolate network trafic between processes. ### WireGuard® at Tetaneutral.net The [Tetaneutral.net](https://www.tetaneutral.net/) non-profit ISP provides WireGuard based VPN connections with one fixed IPv4 address and a /56 IPv6 prefix. When you subscribe to the service, you provide the public key for your end point, and Tetaneutral gives you back the assigned IP adresses and the IP address, port number and public key of the VPN endpoint. It is likely that many other ISP providing WireGuard based VPNs will work approxymately the same as Tetaneutral, and thus this can be expaned easily. You can even setup your own WireGuard VPN service using OpenBSD, but this is out of the scope of this note. ## Let's go ! ### OpenBSD installation OpenBSD is installed as usual on the machine, setting up the network connection for the network on which it is going to have its internet access. In this example the __re0__ interface is going to be used for this. So with DHCP `/etc/hostname.re0` looks like: dhcp ### Obtaining VPN parameters Install the `wireguard-tools` package, to have access to the [wg(8)](https://git.zx2c4.com/wireguard-tools/about/src/man/wg.8) command: $ pkg_add wireguard-tools Now the parameters for the VPN are needed. For this a private key will be generated using the following commands: $ umask 077 $ wg genkey > /etc/wg0.key The corresponding public key is displayed by the following command: $ wg pubkey < /etc/wg0.key _Note:_ if you don't want to install the `wireguard-tool` package, the [wg(4)](https://man.openbsd.org/wg.4) manual page describes an alternate way to generate the private key and display the public key using tools in the base system. To obtain the remaining of the parameters, contact the VPN provider and send them the public key shown above. They should reply providing: (the values below are fake examples) - the IP and port number of their endpoint: `198.51.100.116 51820` - the public key of their endpoint: `Qay+wLckmRMr08t/9psjhrltJ0T4xb0m8++gQj0FpEA=` - the IPv4 and IPv6 prefix assigned to your endpoint: `203.0.113.3/32` and `2001:db8:8087:300::/56` respectively. - the addresses of the DNS servers that can be used inside the VPN: `198.31.100.10` and `198.51.100.11` Keep those informations somewhere, to use them later. ### Moving the default interface to rdomain 1 Now edit `/etc/hostname.re0` to add a `rdomain` directive to move this interface to rdomain 1: rdomain 1 dhcp That's it. If the machine is rebooted at this point, most processes won't have network access since only routing domain 1 has network access. So do not reboot for now. ### Creating the VPN interface A __wg0__ interface for the VPN is going to be setup. For this create the `/etc/hostname.wg0` file, containing (using the fake information form the provider shown above): up wgkey <your wg private key> wgpeer Qay+wLckmRMr08t/9psjhrltJ0T4xb0m8++gQj0FpEA= wgrtable 1 wgendpoint 198.51.100.116 51820 wgaip 0.0.0.0/0 wgaip ::/0 inet 203.0.113.3 255.255.255.255 inet6 2001:db8:8087:300::1 56 ! route add default -link -iface wg0 ! route add -inet6 default -link -iface wg0 The last two lines set up the default route for rdomain 0 via the `wg0` interface. Once `/etc/hostname.wg0` is created, the `/etc/wg0.key` file created previously can be removed. It was only used to keep the private key while obtaining the remaining information needed to create `hostname.wg0` ### Setting up the DNS There are different options to tell the VPN client which DNS resolvers to use. The issue is that `/et/resolv.conf` still contains the adresses of the DNS servers on the LAN, provided by dhclient(8). Generally when using a VPN you don't want to use those, because they will leak the domains that the users of the VPN are connecting to. #### Simple forwarding The simplest solution is to use the _supersede_ option of [`dhclient.conf(5)`](https://man.openbsd.org/dhclient.conf.5) to replace the DHCP provided servers by the ones provided by the VPN provider. This will be the contents of `/etc/dhclient.conf`: supersede domain-name-servers 198.51.100.10, 192.51.100.11; #### Using unwind(8) Another option is to set up [unwind(8)](https://man.openbsd.org/unwind.8), with an [`unwind.conf(5)`](https://man.openbsd.org/unwind.conf.5) configuration file specifying the VPN name servers as forwarders. Add the DNS servers provided by the VPN provider in `/etc/unwind.conf` (see [`unwind.conf(5)`](https://man.openbsd.org/unwind.conf.5) for details): forwarder { 198.51.100.10 192.51.200.11 } Activate unwind with rcctl enable unwind And edit `/etc/dhclient.conf` to contain: prepend domain-name-servers 127.0.0.1 ; This solution also allows processes in routing domain 1 to use the DNS servers provided by the DHCP server. ### SSH access from the LAN If you want to be able to continue accessing the machine from the lan it is connected to, `/etc/ssh/sshd_config` needs to be told to also listen to routing domain 1. ListenAddress 0.0.0.0 rdomain 0 ListenAddress :: rdomain 0 ListenAddress 0.0.0.0 rdomain 1 Otherwise, after reboot, the machine will only be accessible from the console and with SSH though the VPN. ### Rebooting And that's it. The machine can now be rebooted. All processes in the default routing domain (0) will only see the `lo0` and `wg0` interfaces, while the dhclient process needed to bring up `re0` will be in routing domain 1. The `wgrtable` keyword in the wg0 interface configuration tells it ue use the routing table of routing domain 1 to reach the remote VPN endpoint. ## Checking that it works After rebooting, let's review the setup by cheching the `wg0` interface: # ifconfig wg0 wg0: flags=80c3<UP,BROADCAST,RUNNING,NOARP,MULTICAST> mtu 1420 index 6 priority 0 llprio 3 wgport 47326 wgrtable 1 wgpubkey <public key> wgpeer Qay+wLckmRMr08t/9psjhrltJ0T4xb0m8++gQj0FpEA= wgendpoint 198.51.100.116 51820 tx: 45564, rx: 71200 last handshake: 107 seconds ago wgaip ::/0 wgaip 0.0.0.0/0 groups: wg egress inet 203.0.113.3 netmask 0xffffffff inet6 2001:db8:8087:300::1 prefixlen 56 And the default routes: # route -n show | grep default default link#6 ULS 5 229 - 8 wg0 default link#6 ULS 0 4 - 8 wg0 You can test that all traffic will go through the VPN using [traceroute(8)](https://man.openbsd.org/traceroute.8). The machine is reachable from the internet using both its public IPv4 and IPv6 addresses. Finally, the `ps` command with option `-o rtable` will show routing table information in the last column: # ps xau -o rtable USER PID %CPU %MEM VSZ RSS TT STAT STARTED TIME COMMAND RTABLE root 1 0.0 0.0 468 408 ?? I 4:43PM 0:01.02 /sbin/init 0 root 95458 0.0 0.0 788 620 ?? Ip 4:43PM 0:00.24 /sbin/slaacd 0 _slaacd 30350 0.0 0.0 792 676 ?? Ip 4:43PM 0:00.08 slaacd: engine ( 0 _slaacd 69167 0.0 0.0 800 720 ?? Ip 4:43PM 0:00.08 slaacd: frontend 0 root 59774 0.0 0.0 704 528 ?? IU 4:43PM 0:00.02 dhclient: re0 [p 1 _dhcp 30649 0.0 0.0 828 656 ?? Ip 4:43PM 0:02.77 dhclient: re0 (d 1 root 35197 0.0 0.1 548 2252 ?? IpU 4:44PM 0:00.02 syslogd: [priv] 0 _syslogd 1991 0.0 0.1 1120 1384 ?? Ip 4:44PM 0:00.14 /usr/sbin/syslog 0 root 76166 0.0 0.0 836 584 ?? IU 4:44PM 0:00.01 pflogd: [priv] ( 0 _pflogd 9499 0.0 0.0 884 564 ?? Sp 4:44PM 0:01.26 pflogd: [running 0 _ntp 45580 0.0 0.1 1044 2872 ?? I<p 4:44PM 0:00.72 ntpd: ntp engine 0 _ntp 59996 0.0 0.1 888 2560 ?? Ip 4:44PM 0:00.09 ntpd: dns engine 0 root 30954 0.0 0.1 848 1616 ?? I<pU 4:44PM 0:00.06 /usr/sbin/ntpd 0 root 75315 0.0 0.1 1328 1496 ?? I 4:44PM 0:00.16 sshd: /usr/sbin/ 0 root 60900 0.0 0.1 1716 2028 ?? Ip 4:44PM 0:00.07 /usr/sbin/smtpd 0 _smtpd 65700 0.0 0.2 1448 3832 ?? Ip 4:44PM 0:00.08 smtpd: klondike 0 _smtpd 90699 0.0 0.2 1656 4044 ?? Ip 4:44PM 0:00.07 smtpd: control ( 0 _smtpd 31369 0.0 0.2 1536 3936 ?? Ip 4:44PM 0:00.07 smtpd: lookup (s 0 _smtpd 45982 0.0 0.2 1596 4064 ?? Ip 4:44PM 0:00.06 smtpd: pony expr 0 _smtpq 17106 0.0 0.2 1664 4048 ?? Ip 4:44PM 0:00.08 smtpd: queue (sm 0 _smtpd 94920 0.0 0.2 1444 3872 ?? Ip 4:44PM 0:00.05 smtpd: scheduler 0 _sndiop 35803 0.0 0.0 524 936 ?? IpU 4:44PM 0:00.00 sndiod: helper ( 0 _sndio 61376 0.0 0.0 536 808 ?? I<p 4:44PM 0:00.00 /usr/bin/sndiod 0 root 97224 0.0 0.0 204 836 ?? I 4:44PM 0:00.01 /usr/sbin/apmd - 0 root 98386 0.0 0.1 668 1312 ?? Sp 4:44PM 0:00.08 /usr/sbin/cron 0 root 27323 0.0 0.0 940 932 p0 Ip 5:13PM 0:00.36 -ksh (ksh) 0 root 49147 0.0 0.0 484 432 p0 R+pU/3 6:30PM 0:00.00 ps -xau -o rtabl 0 root 42404 0.0 0.1 316 1312 C1 I+pU 4:44PM 0:00.03 /usr/libexec/get 0 root 52228 0.0 0.1 320 1308 C2 I+pU 4:44PM 0:00.02 /usr/libexec/get 0 root 71695 0.0 0.1 316 1308 C3 I+pU 4:44PM 0:00.03 /usr/libexec/get 0 root 90445 0.0 0.1 320 1320 C5 I+pU 4:44PM 0:00.03 /usr/libexec/get 0 Only the dhclient(8) processes are in the non-default routing domain 1. ## What's next ? As already said above, the machine is now reachable from the internet. So perhaps [pf(4)](https://man.openbsd.org/pf.4) needs to be configured to be more restrictive. [sshd(8)](https://man.openbsd.org/sshd.8) can be configured to refuse password based authentication, in order to avoid brute force attacks. You can then point a domain (leased from a registrar) to the IP addresses of the machine and [httpd(8)](https://man.openbsd.org/httpd.8) can be set-up to provide web based services on this domain. Or another interface can be added to that host to convert it into a router to connect to another network. This can be either another physical interface or a vlan interface over the existing re0 interface.