DIY VPNsNovember 13, 2019
I recently started working on a project to set up and run my own VPN using OpenVPN and easy-rsa. Someone at work suggested I check out WireGuard, a more modern and performant VPN that's open-source. Still under development at the time of writing, WireGuard is already supported by several major VPN providers. I decided to bootstrap my own WireGuard VPN and compare it with the OpenVPN solution.
If you'd like to learn more about WireGuard, check out the website documentation and, if you really want to dig deep, read the whitepaper. There are many resources that detail the technical differences between OpenVPN and WireGuard. The purpose of this post is to detail the procedural differences between the two if you want to set up your own VPN. Of course, the procedural differences stem from the technical differences so there's no escaping a little technical discussion ;)
A good place to start is authentication. A VPN server needs to be sure clients are who they say they are. Even if you're the only person using your VPN, there still might be multiple clients (e.g. your laptop and phone). Likewise, clients need to be sure the server is who it claims to be and not a malicious party masquerading as the VPN server. OpenVPN uses Transport Layer Security, or TLS, which provides authentication using digital certificates.
Digital certificates are critical for the web today. When you visit an HTTPS site, your browser verifies the site's certificate, which was signed and issued by an entity called a Certificate Authority (CA). The (public) certificate asserts the site's ownership of a public key so we can trust signatures made by the corresponding (secret) private key (read up on public-key/asymmetric cryptography if you're unfamiliar). We assume the site's owners/administrators are the only ones in possession of the private key and, consequently, the only ones who can generate a valid signature using the private key. Therefore, if we receive a valid signature, we can be fairly certain the party we're talking to is the owner of the private key and, according to the trusted Certificate Authority, the rightful owner of the public key and domain.
So the web and our VPN use digital certificates, but not all certificates are the same. Servers have certificates but clients can have them too. Servers sometimes authenticate clients with their certificates. You don't usually have to provide a certificate when you visit an HTTPS site, however. Access to most sites is open, unless some entity is blocking traffic or the site has blacklisted your IP. The site administrator and development team can implement other forms of authentication (e.g. username/password, HTTP Basic auth, cookies, JSON Web Tokens (JWTs)) to verify users, paying subscribers, or premium members on the site. These categories may carry certain permissions or allow access to particular site functionality. The point being, these forms of gatekeeping are implemented above TLS.
We'll be authenticating server and client certificates in our OpenVPN infrastructure. This way, the server's sure a client is who it claims to be and vice-versa. So during set-up, we need to create a Public Key Infrastructure (PKI). This includes a private key and self-signed certificate for our own trusted CA and a private key and CA-signed certificate for the VPN server. For each client, we'll generate a private key and CA-signed certificate; these may be generated later however. In addition, we'll create a Certificate Revocation List (CRL) for keeping track of clients who have been removed/shouldn't be able to authenticate with the server any longer.
While authentication with digital certificates makes sense for an open infrastructure like the web where ownership of domain names matters, it's a pretty heavy-handed approach for our personal VPN. The server host has to generate keys and certificates, which then have to be distributed to clients along with the CA certificate and most likely a config file. We can zip and encrypt a directory containing these files to ease distribution, but it's still kind of a pain e.g. to
sftp the files to my laptop, unzip them, and iTunes sync them to my iPhone.
WireGuard does away with TLS, so no certificates required! Instead it uses Cryptokey Routing to authenticate peers (i.e. clients and servers) and route packets. Each peer has a public/private key pair and a list of allowed IP addresses. When peer1 wants to send a packet to peer2, it encrypts the packet using a symmetric key established during the handshake with peer2, so that peer2 is the only party who can feasibly decrypt it. Peer1 then sends the encrypted packet over the wire to peer2. Peer2 receives the packet, decrypts it with the symmetric key, and checks the source IP. If the address falls in the range of allowed IPs for peer1, peer2 accepts the packet. Otherwise, the packet's dropped.
Suppose peer2 accepts a packet from peer1 and the destination IP address matches an allowed IP address for peer3. Peer2 then encrypts the packet with the secure session key negotiated with peer3 and sends the result to peer3. The following quote from the website documentation nicely summarizes this functionality:
when sending packets, the list of allowed IPs behaves as a sort of routing table, and when receiving packets, the list of allowed IPs behaves as a sort of access control list.
The server is certain client packets are from the client who claims to have sent them thanks to source IP checking and public-key authentication (i.e. the "access control list" component). The destination IP address doesn't have to be that of another peer. It could be an IP address for "reddit.com". If we're just using the VPN for web browsing, packets sent from the client to the server will be forwarded to "reddit.com" or wherever else we're browsing; not to another peer. The server then receives packets from the site, encrypts them with the session key, and sends them to the peer.
How is the installation process and user flow different for WireGuard? For one, we don't have to architect a whole PKI; we just generate server credentials during set up. In addition, it's easier to add clients. We no longer have to generate credentials server-side, zip directories, and distribute them to clients. We can generate the credentials client-side (even on your iPhone!) and execute a single command on the server to add the client. This command requires two inputs, the client's tunnel IP address and its public key. The former is easy to type and the latter, if you can't copy-and-paste, is much less painful to type than a certificate. WireGuard uses 32-byte, base64-encoded public keys; this comes out to 44 characters. Not ideal, but certainly better than this:
-----BEGIN CERTIFICATE----- MIIDQjCCAioCCQCxJTU+JC8brDANBgkqhkiG9w0BAQsFADBjMQswCQYDVQQGEwJV UzELMAkGA1UECAwCTlkxDDAKBgNVBAcMA05ZQzEPMA0GA1UECgwGZm9vYmFyMRQw EgYDVQQLDAtFbmdpbmVlcmluZzESMBAGA1UEAwwJMTI3LjAuMC4xMB4XDTE5MDgx MzE3NTEyNFoXDTIwMDgxMjE3NTEyNFowYzELMAkGA1UEBhMCVVMxCzAJBgNVBAgM Ak5ZMQwwCgYDVQQHDANOWUMxDzANBgNVBAoMBmZvb2JhcjEUMBIGA1UECwwLRW5n aW5lZXJpbmcxEjAQBgNVBAMMCTEyNy4wLjAuMTCCASIwDQYJKoZIhvcNAQEBBQAD ggEPADCCAQoCggEBAMuF/5O5QCz8/oCLw6WC5RIQtO+8w4yQvJhW4ul8C9ZFfKN9 5qV17VM7V/6Nq2mlUQbuv8nL2P+KNS594CgOgn7zHe916V7E0pkwNbGfvoRVpYk1 tyLmo7JYRa/ZmZd2Gzct5nJsHSQ1CoZMnihzD5aMhlORXqhhj70Io5PUvQK74sO8 8xUHgb3dhCoPki4MEaNvOs6ur1Ldvbq9Fk9q/HxNxXzAKYDG7qsIn4JXIVeleSeE Pa3fuqAhIYEeGI/MLhhYvZPf49uMxn8dat0AIlPGosKmRCEvfp5REuSr10PKpEn7 Hf4+fSCByAoSZgJ3Ge1qvg6YXnxZNS3hrWFNskkCAwEAATANBgkqhkiG9w0BAQsF AAOCAQEAwYtWSQo6bToy1Bewyn+id7/iGd54wt6fKuFzGIFTSaaxRR6ACu7mKjKF xYoz5WC8j7cI53DbX7TLIZ594WrGBTTw1f9QDoin1lxT6QAcVhsCAPpToebXzljf q9YToAdeeB7VS6NFWN39+ygOg2SBpIAZ+WQuPp0voVCn40yNKUd/VL+IEKWAQ1N9 f7gAuvP5TFIEhRCoJZDhZ1DOvyT+VI1mXD4JxhV9hXPppcM2LWllDHrU+atWmX22 9IgAtQwg6C2cZgwt3C7NnU6iy2y/uuI8GhwfjL3KNhAiK3lGScn96NVjoIyzpp3f ovDJKOCXlo9RAvyDehNOBYqzAMIkRQ== -----END CERTIFICATE-----
Removing clients is also simpler with WireGuard. It's a single command on the server where you specify the public key of the client you want to remove. Earlier I mentioned that our OpenVPN infrastructure has a certificate revocation list (CRL). This is required if we want to remove clients. Essentially, the CA maintains state for which certificates it's signed/issued to clients and which certificates it's signed but it no longer deems valid for authentication purposes (this is the CRL). With WireGuard's Cryptokey Routing, the server only maintains a whitelist of public keys and tunnel addresses for clients that can authenticate with the server. Removing a client means removing that client's public key and associated address from the whitelist such that it can't authenticate. From a user perspective, adding/removing clients is relatively fluid for both OpenVPN and WireGuard, especially with some bespoke shell scripting. However, architecting a PKI introduces a lot of complexity we don't necessarily want. Less unneeded complexity hopefully means fewer problems when we try to extend or modify the infrastructure and easier troubleshooting when things go wrong.
I've talked about the server side of things. Configuring things client-side is still a bit of work. With WireGuard, we have to specify the server public key, public IP, and the client's tunnel address during the set-up process. And if we're on mobile/not using a pre-made config file, we need to specify allowed addresses for the server and potentially a tunnel address for the DNS server. With OpenVPN, we circumvent some of these steps by distributing zipped directories with config files containing information such as the server IP. Having to fill in a few extra fields on the WireGuard iPhone app bugs me less than having to zip, encrypt, and move directories around though.
Have a look at the repo if you want more detailed steps for setting up/running your own VPN! I've been using both flavors, OpenVPN and WireGuard, and haven't experienced major problems. But if you have ideas to make
diy-vpn better, feel free to share!