VPN and Firewall Interaction
VPNs enable you to set up secure communications between endpoints and are just one weapon in your security arsenal. Firewalls, an older and more established security technology, are common in almost every environment. A VPN should be integrated into your security policy and that means making sure your VPN and firewall play nicely with each other.
Types of Firewalls
Three main types of firewalls are in common use today: packet filters, application gateways, and stateful inspection firewalls. They are described in the following sections.
A packet filter is the simplest form of firewall. A packet filter firewall will compare any IP packet that attempts to traverse the firewall against its access control list (ACL). If the packet is allowed, it is sent through. If not, the packet filter can either silently drop the packet (DENY in ipchains speak) or send back an ICMP error response (REJECT).
Packet filters only look at five things: the source and destination IP addresses, the source and destination ports, and the protocol (UDP, TCP, and so on). These tests are very fast because each packet contains all the data (in the packet headers) necessary to make its determination. Due to its simplicity and speed, a packet filter can be enabled on your routers, eliminating the need for a dedicated firewall.
One problem with packet filters is that they generally do not look deeply enough into the packet to have any idea what is actually being sent in the packet. Though you might have configured a packet filter to allow inbound access to port 25, the Simple Mail Transfer Protocol (SMTP) port, a packet filter would never know if some other protocol was used on that port. For example, a user on one system might run his Secure Shell (SSH) daemon on that port, knowing that the traffic would be allowed by the packet filter, and be able to SSH through the firewall against policy.
Another problem with packet filters is that they are not effectively able to handle protocols that rely on multiple dynamic connections. The FTP protocol, for example, opens a command channel on which the various commands such as USER, RECV, and LIST are sent. Whenever data is transferred between the hosts, such as files or the LIST output, a separate connection is established. You would need to have an ACL that would allow these data connections through for FTP to work. However, packet filters do not read the FTP command channel to know when such an ACL should be permitted.
An application gateway goes one step beyond a packet filter. Instead of simply checking the IP parameters, it actually looks at the application layer data. Individual application gateways are often called proxies, such as an SMTP proxy that understands the SMTP protocol. These proxies inspect the data that is being sent and verify that the specified protocol is being used correctly.
Let's say we were creating an SMTP application gateway. It would need to keep track of the state of the connection: Has the client sent a HELO/ELHO request? Has it sent a MAIL FROM before attempting to send a DATA request? As long as the protocol is obeyed, the proxy will shuttle the commands from the client to the server.
The application gateway must understand the protocol and process both sides of the conversation. As such, it is a much more CPU-intensive process than a simple packet filter. However, this also lends it a greater element of security. You will not be able to run the previously described SSH-over-port-25 trick when an application gateway is in the way because it will realize that SMTP is not in use.
Additionally, because an application gateway understands the protocols in use, it is able to support tricky protocols such as FTP that create random data channels for each file transfer. As it reads the FTP command channel, it will see (and rewrite, if necessary) the data channel declaration and allow the specified port to traverse the firewall only until the data transfer is complete.
Often there is a protocol that is not directly understood by your application gateway but that must be allowed to traverse the firewall. SSH and HTTPS are two simple examples. Because they are encrypted end to end, an application gateway cannot read the traffic actually being sent.
In these cases, there is usually a way to configure your firewall to allow the appropriate packets to be sent without interference by the firewall. You might hear this called a plug, which comes from the plug-gw, a part of the Firewall Toolkit (FWTK) that was used to connect a client and server directly when the protocol was not supported.
It can be difficult to integrate application gateways into your standard routing hardware due to the processing overhead. Some newer high-end routers are able to function as application gateways, but you'll need a lot of CPU power for acceptable performance.
Even application gateways can be fooled if you are crafty enough. For example, you could tunnel any arbitrary protocol over SMTP: The client could send data as the DATA portion of the transaction, and the server could respond in the resulting error/success code message. The nature and ubiquity of the HTTP protocol makes it even easier; SOAP and .NET are just two "accepted" examples of tunneling other protocols across HTTP. Httptunnel, available at www.httptunnel.com, is a good freeware tool capable of tunneling any protocol across HTTP.
Stateful Inspection Firewalls
Stateful inspection firewalls are a middle ground between application gateways and packet filters. Rather than truly reading the whole dialog between client and server, a stateful inspection firewall will read only the amount necessary to determine how it should behave.
Take the SMTP DATA command, for example. When this command is sent, the client will send the data (the text of the email) ending with a line containing a single ".". The server then responds with a success or error code. An application gateway will need to be reading all the data that is sent and looking for the "." and error code. A stateful inspection firewall, however, will realize that the client is sending data until the server responds. Thus, it will simply forward the client's packets without inspection until the server responds. Simply put, a stateful inspection firewall understands the manner in which stateful protocols must conduct themselves and manages that traffic accordingly within the confines of its rulebase.
By not reading all the data sent, a stateful inspection firewall achieves a significant performance gain over an application gateway while maintaining the higher level of security and protocol support. Our VPN traffic, however, will be encrypted end to end. As such, there will be very little that a stateful inspection firewall will need to look at in our VPN datastreamit can't inspect the actual data anyway. Because of this, there is no functional difference between a stateful inspection firewall and an application gateway for our VPN traffic. There is, however, a solid performance boost from using a stateful inspection firewall. Because our VPN is already introducing latency due to the overhead of encryption, the more performance you can get with your firewall, the better.
Stateful Inspection Concerns
Stateful inspection firewalls have suffered from occasional faults. For example, to support the FTP protocol, stateful inspection firewalls would look at the beginning of a packet to see whether a data channel was requested. An attacker could trick the server into sending a large error message containing a data channel request that aligned perfectly at the beginning of a second packet. The firewall would then open a channel dictated by the attacker. A program called ftpd-ozone (available at http://www.monkey.org/~dugsong/) can exploit this vulnerability. This particular problem has been fixed by firewall vendors, but there might be more lurking around.
Common Linux Firewalls
Several firewalls can run on Linux. We will briefly describe them here, but it is beyond the scope of this book to cover them in depth. See New Riders' Linux Firewalls, Second Edition by Robert Ziegler (2002) for more information on firewalls.
The Firewall Toolkit (FWTK). This was the first publicly available application gateway suite and was the basis of the commercial firewall Gauntlet. It is a set of userland applications that support various protocols such as STMP, HTTP, telnet, and X11. It is still available at http://www.fwtk.org, though it has not been officially supported in years.
IPF. A Linux packet filter for 2.0 kernels. If you are still using a 2.0 kernel, you should really upgrade.
IPChains. The Linux packet filter for 2.2 kernels. Though it is a simple packet filter, you can support some protocols via kernel modules. The ip_masq_ftp module enables you to support the FTP protocol, for example. The problem with IPChains is that the kernel packet filters are handled before the modules can see packets, meaning you must allow inbound access to ports that potentially could be required by the kernel modules.
IPTables. The Linux firewall software for 2.4 kernels, also known as Netfilter. IPTables supports both packet filtering and application gateway support together. Taking the FTP protocol as our example, the data channels that are used are accepted as RELATED packets. This makes your firewall much cleaner because you do not need to leave holes in your firewall for packets that might or might not be used by modules.
IPFilter. Created by Darren Reed, IPFilter is the default kernel packet filter of NetBSD and FreeBSD, it and runs on many other UNIX-like systems. IPFilter only runs on the very old Linux 2.0 kernels, so if you want to use IPFilter, we suggest that you use a BSD machine instead. It is available at http://coombs.anu.edu.au/~avalon/.
Dante. Though normally included as part of a larger commercial firewall system, Dante is available as open source under a BSD-style license. It is a circuit-level firewall/proxy that is largely transparent to users. It is available at http://www.inet.no/dante/.
T.REX. Open sourced in April 2000, T.REX is a sophisticated application gateway firewall that includes intrusion-detection features, strong authentication support, and extensive logging. Version 2 of T.REX was not yet available at the time this was written; though precompiled versions are available on CD. You can obtain T.REX at http://www.opensourcefirewall.com.
Darren Reed modified the IPFilter license (or clarified it, depending on your position) to deny modifications to the source, rendering it not completely open source. Because of this, it was removed from OpenBSD. Reed later allowed modifications when used as part of NetBSD and FreeBSD. OpenBSD is creating a new packet filter, PF, from scratch to replace IPFilter. Yes, now and then there doth be strife in the open-source community ...
Placing Your Firewall
We strongly suggest that you use a firewall as part of your security infrastructure. Using a VPN in conjunction with your firewall, however, requires careful planning and configuration. Several different configurations are available when using both a firewall and a VPN server. Each has its pros and cons, and we'll do our best to help you pick the option that makes sense for you.
VPN Server on a Firewall
The solution that feels most natural is to install your VPN software on your firewall itself. Many commercial firewalls include VPN components as an extra option. If you are working in a mixed commercial and freeware VPN environment, you will likely end up supporting this configuration.
As seen in Figure 2.4, we have a single point of entry into our network, which serves three purposes:
The firewall allows outbound access to the Internet.
The firewall prevents inbound access from the Internet.
The VPN service encrypts traffic to remote clients or networks.
The pros of putting your VPN on your firewall are as follows:
You have one place controlling all your security, meaning fewer machines to manage.
You can create firewall rules that apply to your VPN traffic using the same tools you already use to manage your firewall.
The cons of putting your VPN on your firewall are as follows:
You have one place controlling all your security. You'd better make sure this machine is extremely secure.
You must configure your routes carefully to make sure the traffic goes through the appropriate interfaces (eth0 vs. vpn1, for example.)
Improper configuration in your firewall rules could allow traffic from the Internet to get through to the inside by slipping through using VPN addresses.
Your Internet and VPN traffic will compete for resources on the machine, so a larger machine is likely necessary.
VPN Server Parallel to Firewall
Another topology that seems logical is to use your VPN parallel to your firewall. Your internal machines will still point to the firewall as their default route. The firewall will have a route to the VPN-accessible networks via the VPN server and will inform clients to send packets to the VPN machine when appropriate. (The "Routing" section later in this chapter discusses how this works.) You can see this topology in Figure 2.5.Figure 2.5 The VPN server parallel to a firewall.
If you prefer, you can place a router between the internal network and the VPN and firewall machines. You would configure the router to know the networks accessible through the VPN and would set up your routing rules there instead of on the firewall.
The pros of putting your VPN server parallel to your firewall are as follows:
Your VPN traffic is not going through the firewall at all, so you do not need to modify your firewall configuration to support the VPN packets. Because some VPN protocols are not supported by all firewalls, this might be your only option.
You can scale much easier. If you find that a VPN server is under too much load, you can add machines and distribute the VPNs that can be established between them. If you are connecting multiple remote networks, you can have one VPN machine per network.
If you are supporting roaming users in a host-network configuration, you can simply add VPN servers and use round-robin DNS to distribute the load between the VPN servers.
The cons of putting your VPN server parallel to your firewall are as follows:
Your VPN server is directly attached to the Internet. You had better be very sure that it is well secured; otherwise, an attacker could break in and have direct access to your internal network.
You now have two machines that communicate with the Internet, and you must make sure that both are correctly configured to only pass legitimate traffic, thereby increasing the support workload and associated costs.
Cost, of course, increases incrementally with the addition of new servers and related support staff.
VPN Server Behind Your Firewall
Another location for your VPN server is to put it behind your firewall completely, attached to the internal network. As shown in Figure 2.6, our VPN server is not directly reachable from the Internet at all, and all packets must reach it through our dedicated firewall.
As with the previous topology, you will need to add a route to your firewall that redirects VPN traffic from internal machines to the VPN server. You will also need to configure your firewall to pass the encrypted VPN traffic directly to your VPN server.
Figure 2.6 The VPN server behind a firewall.
The pros of putting your VPN server behind your firewall are as follows:
The VPN is completely protected from the Internet by the firewall.
There is only one machine controlling all access to and from the Internet.
Network restrictions for VPN traffic are located only on the VPN server, which can make writing rulesets easier.
The cons of putting your VPN server behind your firewall are as follows:
All the VPN traffic must travel through the firewall as well, adding a degree of latency.
The firewall will need to shuttle VPN traffic from the Internet to the VPN directly. Because it cannot inspect this traffic (it is encrypted, after all), it will require a simple packet filter ACL or application plug proxy.
Getting your firewall to pass the encrypted VPN traffic to the VPN server is sometimes tricky. Some firewalls do not know what to do with IP protocols other than ICMP, TCP, and UDP. This means it might be harder or impossible to support VPNs that use different IP protocols, such as ESP packets for IPSec VPNs or GRE packets for PPTP VPNs.
All the VPN traffic travels through the same LAN wire twice, once from client to VPN server in the clear and once from the VPN to the firewall encrypted. This can degrade LAN performance.
One option to decrease latency would be to connect a second network card (eth1) on the VPN server directly to a network card on the firewall (eth2) with a crossover cable. You could use a hub or switch and create an actual network segment if you prefer, but because you'd only have the two hosts here, a crossover cable would work as well and would be faster because no extra equipment is involved. This enables you to route the encrypted VPN traffic directly over this wire and through the firewall rather than backtracking over the existing LAN segment. This means the VPN traffic won't be competing for LAN resources.
The direct VPN-to-firewall link will need to either be a point-to-point link or be given its own network segment. If you prefer the latter, we suggest a small network such as a /252, which can hold only two hosts, plenty for our purposes. We could allocate 192.168.254.254 to the firewall and 192.168.254.253 to the "external" VPN interface, and the network itself would be 192.168.254.252/252.
Configuring Your VPN with Its Own Firewall
In any of the configurations previously described, it is possible to restrict what traffic can traverse the VPN connection. This might be desired when the networks or hosts you are connecting are not of the same trust level.
For the case in which your Internet firewall and VPN server are the same machine, this can be achieved simply by adding new rules to the firewall using your existing firewall software.
For cases when you have a separate VPN server, you might either install a separate firewall machine in front of your VPN server or simply rely on Linux kernel packet filters on the VPN server itself. For example, if you want to allow only mail-related traffic to traverse your VPN, you might implement kernel restrictions on your VPN server, such as the following:
# Allow only SMTP, POP, IMAP and POPs/IMAPs through our VPN: for port in 25 109 110 143 993 995 do ipchains -A output destination-port $port -i vpn1 -j ACCEPT ipchains -A input source-port $port -i vpn1 -j ACCEPT done ipchains -A output -i vpn1 -j DENY ipchains -A input -i vpn1 -j DENY
These rules assume that your VPN traffic goes through the virtual interface vpn1. This is a very simple example. You can create ACLs tailored for your VPN using whatever tools (IPchains, IPtables, IPfilter, and so on) you are comfortable with.