Paranoid Penguin - Linux VPNs with OpenVPN, Part III

Secure remote networking, the OpenVPN way.

In my previous two columns, I began a series on building Linux-based Virtual Private Network (VPN) solutions using OpenVPN. When I left off last time, I had gotten as far through the OpenVPN server configuration process as creating a simple Public Key Infrastructure (PKI), using it to generate server and client certificates, and creating a few other “support” files involved in building OpenVPN tunnels. In so doing, I worked my way down just the first third or so of the example OpenVPN server configuration, but those PKI/crypto-related configuration parameters represent the most complicated part of OpenVPN configuration tasks.

This month, I describe the rest of that server configuration file and show a corresponding OpenVPN client configuration file (which I'll dissect next month). I also show how to start both server and client processes, although debugging, firewall considerations and other finer points also will need to wait until my next column.

Have no fear—I think you'll find this installment to be plenty action-packed in its own right. Let's get to it!

OpenVPN Server Configuration, Continued

Normally at this point in a multipart series, I'd review at least some details from the prior month's column, but that won't work this time. Last month's article covered a lot of ground, and this month's needs to cover still more. Suffice it to say that I began dissecting an example OpenVPN server configuration file, /etc/openvpn/server.ovpn (Listing 1).

I got as far as generating the files referenced in the ca, cert, key, dh and tls-auth lines, using a combination of OpenVPN “easy-rsa” helper scripts (located in /usr/share/doc/openvpn/examples/easy-rsa/2.0) and the commands openvpn and openssl. I'm going to continue describing Listing 1's parameters, assuming that the aforementioned certificate, key and other helper files are in place.

So, having set up the basic port/protocol/device settings and cryptography-related settings, let's move on to settings that will determine what happens once a client successfully establishes an authenticated, encrypted tunnel. The first such setting is server.

server actually is a helper directive. It expands to an entire block of other parameters. Rather than slogging through all those additional parameters, let's just say that the server directive takes two parameters: a network address and a netmask. For each tunnel established by clients on the port we specified earlier, the OpenVPN server process will carve a little 30-bit subnet from the specified IP space, assign itself the first host IP address in that subrange as its local tunnel endpoint and assign the other host IP in the 30-bit subnet to the connecting client as its remote tunnel endpoint.

In the example, I've specified the network address with a netmask of, which translates to the range of IP addresses from to When the first tunnel is established, the server will use as its local tunnel endpoint address and assign to the client to use as the remote tunnel endpoint address. ( is that subnet's network address, and is its broadcast address.)

For the next client to connect, the server will use as its tunnel endpoint and will assign as the client's tunnel endpoint (with and as network and broadcast addresses, respectively). Get it?

This isn't the most efficient use of an IP range. The server needs a different local IP address for each tunnel it builds, and for each tunnel, the server essentially wastes two others (for network and broadcast addresses). Preceding the server directive with the line topology subnet will cause the server to use the first IP in its server [network address] [netmask] range for its local tunnel IP for all tunnels and for client tunnel-endpoint IPs to be allocated from the entire remainder of possible IPs in that range, as though all remote tunnel endpoints were IP addresses on the same LAN.

This isn't the default behavior, because it's new to OpenVPN 2.1. The “subnet” topology isn't supported by earlier versions or by Windows clients using version 8.1 or lower of the TAP-Win32 driver. Note that if undeclared (as in Listing 1), the topology parameter has a default value of net30, which results in server's specified IP range being split up into 30-bit subnets as described above.

Continuing on in Listing 1, next comes ifconfig-pool-persist, which specifies a file in which to store associations between tunnel clients' Common Names (usually their hostnames, as specified in their respective client certificates) and the IP addresses the server assigns to their tunnels. Although this doesn't guarantee a given client will receive the same tunnel IP every time it connects, it does allow clients to use the --persist-tun option in their client configurations, which keeps tunnel sessions open across disruptions in service (OpenVPN server dæmon restarts, network problems and so forth).

Next comes the statement push "redirect-gateway def1 bypass-dhcp". The push directive causes the subsequent double-quotation-mark-enclosed string to be run on the client as though it was part of the client's local configuration file. In this case, the server will push the redirect-gateway parameter to all clients, with the effect that each time a client connects, the client's local default gateway, DNS servers and other network parameters normally provided by DHCP will be overridden by the server's settings for those things.

This effectively enforces a VPN policy of “local-subnet-only split tunneling”. For those of you new to VPNs, a split tunnel configuration is one in which clients can use their VPN tunnels to connect to some things and their local (non-tunneled) Internet connection to connect to other things.

As I've said in previous columns though, forcing clients to use the remote network's infrastructure (DNS servers, Internet uplink and so forth) makes it much harder for attackers connected to a client's local network, which might be an untrusted environment like a coffee-shop wireless hotspot, to perform various kinds of eavesdropping, session-hijacking and man-in-the-middle attacks.

Even with this setting, a client still will be able to connect to some things on the local network. It just won't be able to use it as a route for anything but connecting back to your OpenVPN server. Again, it's good policy to configure clients to leverage as much of your trusted network's infrastructure as possible.

After the push "redirect-gateway..." directive comes keepalive 10 120. Like server, keepalive is a helper directive that expands to a list of other parameters. Again for the sake of brevity, let me summarize the effect of the example line: every ten seconds, the server will check to see that each client is still connected, and if no reply is received from a given client over any 120-second period, it will assume that client is unreachable at its last known IP address.

For example, if the server sends a query to a particular tunnel client at 9:00:00 and gets a reply, but another at 9:00:10 to which there's no reply, and also receives no reply to any of 11 more queries sent out at 9:00:20, 9:00:30 and so on until 9:02:00, then at 9:02:00 (after 120 seconds of no replies), the server will conclude the client system is unreachable.

At this point, the server will attempt to re-resolve the remote client's name, on the assumption that its IP address may have changed (due to DHCP lease renewal, for example) and, thus, re-establish the tunnel session.

The aforementioned infrastructure settings, such as DNS servers, by the way, will be read by the server's openvpn process from /etc/resolv.conf, the server's running routing table and so forth—no OpenVPN configuration parameters are necessary for those settings unless you want them to be different from the server's. (For now, let's assume you don't!)

I just spent a fair amount of ink on only a handful of settings. But I think this is warranted given that server and keepalive are helper directives that expand to many more settings and given that we're now done with the network configuration portion of our server configuration.

The next parameter is a simple one: cipher BF-CBC, which specifies that each tunnel's data payload will be encrypted with the Blowfish cipher, using 128-bit keys, in Cipher Block Chaining mode (CBC mode makes it harder for an attacker to brute-force-decrypt isolated parts of a given session). BF-CBC is the default setting for cipher, so technically, I don't need to specify this, but it's an important setting. You can use the command openvpn --show-ciphers to see a list of all supported cipher values and their default key sizes.

comp-lzo is even simpler. It tells OpenVPN to compress all session data using the LZO compression algorithm, unless a given portion of data appears to be compressed already (for example, if a JPEG image or a ZIP file is being transferred), in which case OpenVPN won't compress until it detects a return to noncompressed session content. This adaptive behavior helps minimize the data padding that results from trying to compress already-compressed data. Because LZO is a fast algorithm, this is a good setting. Its cost in CPU overhead is generally more than compensated for by the amount of network bandwidth (and, thus, other CPU cycles) it conserves.

The next setting, max-clients 2, specifies that a maximum of two tunnels may be active at one time. If you have only one or two users, there's no good reason to allow more than one or two concurrent tunnels. In my own testing, however, I've found that setting this all the way down to 1 can cause problems even if you have only one user, probably due to how OpenVPN handles tunnel persistence (see keepalive above).

The next four settings are interrelated. user and group specify the names of an unprivileged user account and group (nobody and nogroup, respectively), for the OpenVPN server dæmon to demote itself to after opening necessary tun/tap devices, reading its configuration file, certificates and keys, and other root-only startup actions.

For this to work properly, you also need to set persist-key and persist-tun. persist-key causes OpenVPN to keep key file contents cached in memory across dæmon interruptions (like those caused by tunnels being broken and re-established). persist-tun causes OpenVPN to keep any tun/tap devices that were opened on startup, open across the same kinds of dæmon restarts.

With user and group set to unprivileged user and group, if you were to skip declaring persist-key or persist-tun, the OpenVPN dæmon would lack the necessary privileges to re-read protected key files or re-open the tun or tap device.

You could, of course, skip the user and group settings. Those settings, however, lessen the impact of some unforeseen buffer-overflow vulnerability. It can make the difference from an attacker gaining an unprivileged shell and gaining a root shell. Unfortunately, you can't assume that just because OpenVPN has had a good track record so far with respect to lacking many significant security vulnerabilities, that it never will have any!

The last three settings in Listing 1 concern logging. status specifies a file to which OpenVPN will write dæmon status updates periodically, regardless of actual activity. Unlike most log files, each time this file is updated, OpenVPN will overwrite the previous message. This is what the file /etc/openvpn/openvpn-status.log on my OpenVPN server says right now:

Updated,Fri Jan  1 21:55:11 2010
Common Name,Real Address,Bytes Received,Bytes Sent,Connected Since
minion2,,125761,103329,Fri Jan  1 17:56:21 2010
Virtual Address,Common Name,Real Address,Last Ref,minion2,,Fri Jan  1 20:54:03 2010
Max bcast/mcast queue length,0

As you can see, there's only one client currently connected (minion2), with one corresponding route table entry.

Moving on back in Listing 1, verb 3 sets the overall logging-verbosity level to 3 out of a possible range of 0 (no logging except major errors) and 11 (the most verbose debugging output possible). The default value is 1, but 3 is much more useful for getting things set up and working properly, without presenting any particular danger of log files growing too huge too quickly.

This is especially true with mute 20 set, which tells OpenVPN never to log the same message (in a given event category) more than 20 times in a row.

On my Ubuntu system, OpenVPN writes all its messages to /var/log/daemon if the openvpn command is executed with the --daemon flag, which causes it to run as a background (dæmon) process. If you run openvpn without --daemon, it runs in the foreground and logs all messages to the console or terminal window you started it in (tying up that console in the process, but this is a very handy way to run OpenVPN during initial setup and testing).