OpenVPN^TM 2.0 HOWTO

Numbering private subnets

Setting up a VPN often entails linking together private subnets from different locations.

The Internet Assigned Numbers Authority (IANA) has reserved the following three blocks of the IP address space for private internets (codified in RFC 1918):

While addresses from these netblocks should normally be used in VPN configurations, it's important to select addresses that minimize the probability of IP address or subnet conflicts. The types of conflicts that need to be avoided are:

• conflicts from different sites on the VPN using the same LAN subnet numbering, or
• remote access connections from sites which are using private subnets which conflict with your VPN subnets.

For example, suppose you use the popular 192.168.0.0/24 subnet as your private LAN subnet. Now you are trying to connect to the VPN from an internet cafe which is using the same subnet for its WiFi LAN. You will have a routing conflict because your machine won't know if 192.168.0.1 refers to the local WiFi gateway or to the same address on the VPN.

As another example, suppose you want to link together multiple sites by VPN, but each site is using 192.168.0.0/24 as its LAN subnet. This won't work without adding a complexifying layer of NAT translation, because the VPN won't know how to route packets between multiple sites if those sites don't use a subnet which uniquely identifies them.

The best solution is to avoid using 10.0.0.0/24 or 192.168.0.0/24 as private LAN network addresses. Instead, use something that has a lower probability of being used in a WiFi cafe, airport, or hotel where you might expect to connect from remotely. The best candidates are subnets in the middle of the vast 10.0.0.0/8 netblock (for example 10.66.77.0/24).

And to avoid cross-site IP numbering conflicts, always use unique numbering for your LAN subnets.

Setting up your own Certificate Authority (CA) and generating certificates and keys for an OpenVPN server and multiple clients

Overview

The first step in building an OpenVPN 2.0 configuration is to establish a PKI (public key infrastructure). The PKI consists of:

• a separate certificate (also known as a public key) and private key for the server and each client, and
• a master Certificate Authority (CA) certificate and key which is used to sign each of the server and client certificates.

OpenVPN supports bidirectional authentication based on certificates, meaning that the client must authenticate the server certificate and the server must authenticate the client certificate before mutual trust is established.

Both server and client will authenticate the other by first verifying that the presented certificate was signed by the master certificate authority (CA), and then by testing information in the now-authenticated certificate header, such as the certificate common name or certificate type (client or server).

This security model has a number of desirable features from the VPN perspective:

• The server only needs its own certificate/key -- it doesn't need to know the individual certificates of every client which might possibly connect to it.
• The server will only accept clients whose certificates were signed by the master CA certificate (which we will generate below). And because the server can perform this signature verification without needing access to the CA private key itself, it is possible for the CA key (the most sensitive key in the entire PKI) to reside on a completely different machine, even one without a network connection.
• If a private key is compromised, it can be disabled by adding its certificate to a CRL (certificate revocation list). The CRL allows compromised certificates to be selectively rejected without requiring that the entire PKI be rebuilt.
• The server can enforce client-specific access rights based on embedded certificate fields, such as the Common Name.

Generate the master Certificate Authority (CA) certificate & key

In this section we will generate a master CA certificate/key, a server certificate/key, and certificates/keys for 3 separate clients.

For PKI management, we will use a set of scripts bundled with OpenVPN.

If you are using Linux, BSD, or a unix-like OS, open a shell and cd to the easy-rsa subdirectory of the OpenVPN distribution. If you installed OpenVPN from an RPM file, the easy-rsa directory can usually be found in /usr/share/doc/packages/openvpn or /usr/share/doc/openvpn-2.0 (it's best to copy this directory to another location such as /etc/openvpn, before any edits, so that future OpenVPN package upgrades won't overwrite your modifications). If you installed from a .tar.gz file, the easy-rsa directory will be in the top level directory of the expanded source tree.

If you are using Windows, open up a Command Prompt window and cd to \Program Files\OpenVPN\easy-rsa. Run the following batch file to copy configuration files into place (this will overwrite any preexisting vars.bat and openssl.cnf files):

init-config

Now edit the vars file (called vars.bat on Windows) and set the KEY_COUNTRY, KEY_PROVINCE, KEY_CITY, KEY_ORG, and KEY_EMAIL parameters. Don't leave any of these parameters blank.

Next, initialize the PKI. On Linux/BSD/Unix:

. ./vars
./clean-all
./build-ca

On Windows:

vars
clean-all
build-ca

The final command (build-ca) will build the certificate authority (CA) certificate and key by invoking the interactive openssl command:

Note that in the above sequence, most queried parameters were defaulted to the values set in the vars or vars.bat files. The only parameter which must be explicitly entered is the Common Name. In the example above, I used "OpenVPN-CA".

Generate certificate & key for server

Next, we will generate a certificate and private key for the server. On Linux/BSD/Unix:

./build-key-server server

On Windows:

build-key-server server

As in the previous step, most parameters can be defaulted. When the Common Name is queried, enter "server". Two other queries require positive responses, "Sign the certificate? [y/n]" and "1 out of 1 certificate requests certified, commit? [y/n]".

Generate certificates & keys for 3 clients

Generating client certificates is very similar to the previous step. On Linux/BSD/Unix:

./build-key client1
./build-key client2
./build-key client3

On Windows:

build-key client1
build-key client2
build-key client3

If you would like to password-protect your client keys, substitute the build-key-pass script.

Remember that for each client, make sure to type the appropriate Common Name when prompted, i.e. "client1", "client2", or "client3". Always use a unique common name for each client.

Generate Diffie Hellman parameters

Diffie Hellman parameters must be generated for the OpenVPN server. On Linux/BSD/Unix:

./build-dh

On Windows:

build-dh

Output:

Key Files

Now we will find our newly-generated keys and certificates in the keys subdirectory. Here is an explanation of the relevant files:

The final step in the key generation process is to copy all files to the machines which need them, taking care to copy secret files over a secure channel.

Now wait, you may say. Shouldn't it be possible to set up the PKI without a pre-existing secure channel?

The answer is ostensibly yes. In the example above, for the sake of brevity, we generated all private keys in the same place. With a bit more effort, we could have done this differently. For example, instead of generating the client certificate and keys on the server, we could have had the client generate its own private key locally, and then submit a Certificate Signing Request (CSR) to the key-signing machine. In turn, the key-signing machine could have processed the CSR and returned a signed certificate to the client. This could have been done without ever requiring that a secret .key file leave the hard drive of the machine on which it was generated.