Naturally, you need to do more than create and populate a DMZ to build a strong perimeter network. What ultimately distinguishes the DMZ from your internal network is your firewall.
Your firewall (or firewalls) provides the first and last word as to which traffic may enter and leave each of your networks. Although it's a mistake to mentally elevate firewalls to a panacea, which can lead to complacency and thus to bad security, it's imperative that your firewalls are carefully configured, diligently maintained, and closely watched.
As I mentioned earlier, in-depth coverage of firewall architecture and specific configuration procedures are beyond the scope of this chapter. What we will discuss are some essential firewall concepts and some general principles of good firewall construction.
In increasing order of strength, the three primary types of firewall are the simple packet-filter, the so-called "stateful" packet-filter, and the application-layer proxy. Most packaged firewall products use some combination of these three technologies.
Simple packet-filters evaluate packets based solely on IP headers (Figure 2-5). Accordingly, this is a relatively fast way to regulate traffic, but it is also easy to subvert. Source-IP spoofing attacks generally aren't blocked by packet-filters, and since allowed packets are literally passed through the firewall, packets with "legitimate" IP headers but dangerous data payloads (as in buffer-overflow attacks) can often be sent intact to "protected" targets.
An example of an open source packet-filtering software package is Linux 2.2's ipchains kernel modules (superceded by Linux 2.4's netfilter/iptables, which is a stateful packet-filter). In the commercial world, simple packet-filters are increasingly rare: all major firewall products have some degree of state-tracking ability.
Stateful packet-filtering comes in two flavors: generic and Check Point. Let's discuss the generic type first.
At its simplest, the term refers to the tracking of TCP connections, beginning with the "three-way handshake" (SYN, SYN/ACK, ACK), which occurs at the start of each TCP transaction and ends with the session's last packet (a FIN or RST). Most packet-filtering firewalls now support some degree of low-level connection tracking.
Typically, after a stateful packet-filtering firewall verifies that a given transaction is allowable (based on source/destination IP addresses and ports), it monitors this initial TCP handshake. If the handshake completes within a reasonable period of time, the TCP headers of all subsequent packets for that transaction are checked against the firewall's "state table" and passed until the TCP session is closed ? i.e., until one side or the other closes it with a FIN or RST. (See Figure 2-6.) Specifically, each packet's source IP address, source port, destination IP address, destination port, and TCP sequence numbers are tracked.
This has several important advantages over simple (stateless) packet-filtering. The first is bidirectionality: without some sort of connection-state tracking, a packet-filter isn't really smart enough to know whether an incoming packet is part of an existing connection (e.g., one initiated by an internal host) or the first packet in a new (inbound) connection. Simple packet filters can be told to assume that any TCP packet with the ACK flag set is part of an established session, but this leaves the door open for various " spoofing" attacks.
Another advantage of state tracking is protection against certain kinds of port scanning and even some attacks. For example, the powerful port scanner nmap supports advanced " stealth scans" (FIN, Xmas-Tree, and NULL scans) that, rather than simply attempting to initiate legitimate TCP handshakes with target hosts, involve sending out-of-sequence or otherwise nonstandard packets. When you filter packets based not only on IP-header information but also on their relationship to other packets (i.e., whether they're part of established connections), you increase the odds of detecting such a scan and blocking it.
The second type of stateful packet-filtering is that used by Check Point technologies in its Firewall-1 and VPN-1 products: StatefulInspection . Check Point's Stateful Inspection technology combines generic TCP state tracking with a certain amount of application-level intelligence.
For example, when a Check Point firewall examines packets from an HTTP transaction, it looks not only at IP headers and TCP handshaking; it also examines the data payloads to verify that the transaction's initiator is in fact attempting a legitimate HTTP session instead of, say, some sort of denial-of-service attack on TCP port 80.
Check Point's application-layer intelligence is dependant on the "INSPECT code" (Check Point's proprietary packet-inspection language) built into its various service filters. TCP services, particularly common ones like FTP, Telnet, and HTTP, have fairly sophisticated INSPECT code behind them. UDP services such as NTP and RTTP, on the other hand, tend to have much less. Furthermore, Check Point users who add custom services to their firewalls usually do so without adding any INSPECT code at all and instead define the new services strictly by port number.
Check Point technology is sort of a hybrid between packet-filtering and application-layer proxying. Due to the marked variance in sophistication with which it handles different services, however, its true strength is probably much closer to simple packet-filters than it is to that of the better proxying firewalls (i.e., Application Gateway firewalls).
Although Stateful Inspection is a Check Point trademark, other stateful firewalls such as Cisco PIX and even Linux iptables have similar Application-Layer intelligence in tracking certain types of applications' sessions.
The third category of common firewall technologies is application-layer proxying. Unlike simple and stateful packet-filters, which inspect but do not alter packets (except, in some cases, readdressing or redirecting them), a proxying firewall acts as an intermediary in all transactions that traverse it (see Figure 2-7).
Proxying firewalls are often called "application-layer" proxies because, unlike other types of proxies that enhance performance but not necessarily security, proxying firewalls usually have a large amount of application-specific intelligence about the services they broker.
For example, a proxying firewall's FTP proxy might be configured to allow external clients of an internal FTP server to issue USER, PASS, DIR, PORT, and GET commands, but not PUT commands. Its SMTP proxy might be configured to allow external hosts to issue HELO, FROM, MAILTO, and DATA commands to your SMTP gateway, but not VRFY or EXPN. In short, an application-layer proxy not only distinguishes between allowed and forbidden source- and destination-IP addresses and ports; it also distinguishes between allowable and forbidden application behavior.
As if that in itself weren't good enough, by definition, proxying firewalls also afford a great deal of protection against stack-based attacks on protected hosts. For example, suppose your DMZed web server is, unbeknownst to you, vulnerable to denial-of-service attacks in which deliberately malformed TCP "SYN" packets can cause its TCP/IP stack to crash, hanging the system. An application-layer proxy won't forward those malformed packets; instead, it will initiate a new SYN packet from itself (the firewall) to the protected host and reply to the attacker itself.
The primary disadvantages of proxying firewalls are performance and flexibility. Since a proxying firewall actively participates in, rather than merely monitoring, the connections it brokers, it must expend much more of its own resources for each transaction than a packet-filter does ? even a stateful one. Furthermore, whereas a packet-filter can very easily accommodate new services, since it deals with them only at low levels (e.g., via low-level protocols common to many applications), an application-layer proxy firewall can usually provide full protection only to a relatively small variety of known services.
However, both limitations can be mitigated to some degree. A proxying firewall run on clustered server-class machines can easily manage large (T3-sized) Internet connections. Most proxy suites now include some sort of Generic Service Proxy (GSP), a proxy that lacks application-specific intelligence but can ? by rewriting IP and TCP/UDP headers, but passing data payloads as is ? still provide protection against attacks on TCP/IP anomalies. A GSP can be configured to listen on any port (or multiple ports) for which the firewall has no application-specific proxy.
As a last resort, most proxying firewalls also support packet-filtering. However, this is very seldom preferable to using GSPs.
Commercial application-layer proxy firewalls include Secure Computing Corp.'s Sidewinder, Symantec Enterprise Firewall (formerly called Raptor), and Watchguard Technologies' Firebox. (Actually, Firebox is a hybrid, with application proxies only for HTTP, SMTP, DNS, and FTP, and stateful packet-filtering for everything else.)
Free/open source application-layer proxy packages include Dante, the TIS Firewall Toolkit (now largely obsolete, but the ancestor of Gauntlet), and Balazs Scheidler's new firewall suite, Zorp.
Choosing which type of firewall to use, which hardware platform to run it on, and which commercial or free firewall package to build it with depends on your particular needs, financial and technical resources, and, to some extent, subjective considerations. For example, a business or government entity who must protect their data integrity to the highest possible degree (because customer data, state secrets, etc. are at stake) is probably best served by an application-gateway (proxy) firewall. If 24/7 support is important, a commercial product may be a good choice.
A public school system, on the other hand, may lack the technical resources (i.e., full-time professional network engineers) to support a proxying firewall, and very likely lacks the financial resources to purchase and maintain an enterprise-class commercial product. Such an organization may find an inexpensive stateful packet-filtering firewall "appliance" or even a Linux or FreeBSD firewall (if they have some engineering talent) to be more than adequate.
Application-gateway firewalls are generally the strongest, but they are the most complex to administer and have the highest hardware speed and capacity requirements. Stateful packet-filtering firewalls move packets faster and are simpler to administer, but tend to provide much better protection for some services than for others. Simple packet-filters are fastest of all and generally the cheapest as well, but are also the easiest to subvert. (Simple packet filters are increasingly rare, thanks to the rapid adoption of stateful packet-filtering in even entry-level firewall products.)
Free/open source firewall packages are obviously much cheaper than commercial products, but since technical support is somewhat harder to obtain for them, they require more in-house expertise than commercial packages. This is mitigated somewhat by the ease with which one can find and exchange information with other users over the Internet: most major open source initiatives have enthusiastic and helpful communities of users and developers.
In addition, free firewall products may or may not benefit from the public scrutiny of their source code for security vulnerabilities. Such scrutiny is often assumed but seldom assured (except for systems like OpenBSD, in which security audits of source code is an explicit and essential part of the development process).
On the other hand, most open source security projects' development teams have excellent track records in responding to and fixing reported security bugs. When open source systems or applications are vulnerable to bugs that also affect commercial operating systems, patches and fixes to the open source products are often released much more quickly than for the affected commercial systems.
Another consideration is the firewall's feature set. Most but not all commercial firewalls support Virtual Private Networking (VPN), which allows you to connect remote networks and even remote users to your firewall through an encrypted "tunnel." (Linux firewalls support VPNs via the separately maintained FreeS/Wan package.) Centralized administration is less common, but desirable: pushing firewall policies to multiple firewalls from a single management platform makes it easier to manage complex networks with numerous entry points or "compartmentalized" (firewalled) internal networks.
Ultimately, the firewall you select should reflect the needs of your perimeter network design. These needs are almost always predicated on the assets, threats, and risks you've previously identified, but are also subject to the political, financial, and technical limitations of your environment.
Precisely how you configure your firewall will naturally depend on what type you've chosen and on your specific environment. However, some general principles should be observed.
First, before installing firewall software, you should harden the firewall's underlying operating environment to at least as high a degree as you would harden, for example, a web server. Unnecessary software should be removed; unnecessary startup scripts should be disabled; important daemons should be run without root privileges and chrooted if possible; and all OS and application software should be kept patched and current. As soon as possible after OS installation (and before the system is connected to the Internet), an integrity checker such as tripwire or AIDE should be installed and initialized.
In addition, you'll need to decide who receives administrative access to the firewall, with particular attention to who will edit or create firewall policies. No administrator should be given a higher level of access privileges than they actually need.
For example, the Operations Technician who backs up the system periodically should have an account and group membership that give him read-access to all filesystems that he needs to back up, but not write-access. Furthermore, his account should not belong to the groups wheel or root (i.e., he shouldn't be able to su to root).
If your firewall runs on Linux, see Chapter 3 for detailed system-hardening instructions.
If your firewall supports anti-IP-spoofing features, configure and use them. Many network attacks involved spoofed packets, i.e., packets with forged source-IP-addresses. This technique is used most commonly in Denial of Service (DoS) attacks to mask the atttack's origin, as well as in attempts to make packets appear to originate from trusted (internal) networks. The ability to detect spoofed packets is so important that if your firewall doesn't support it, I strongly recommend you consider upgrading to a firewall that does.
For example, suppose your firewall has three ethernet interfaces: eth0, with the IP 220.127.116.11, faces the outside; eth1, with the IP address 192.168.111.2, faces your DMZ network; and eth2, with the IP address 10.23.23.2, faces your internal network. No packets arriving at eth0 should have source IPs beginning "192.168." or "10.": only packets originating in your DMZ or internal network are expected to have such source addresses. Furthermore, eth0 faces an Internet-routable address space, and 10.0.0.0/8 and 192.168.0.0/16 are both non-Internet-routable networks.
 The range of addresses from 172.16.0.0 to 172.31.255.255 (or, in shorthand, "172.16.0.0/12") is also non-Internet-routable and therefore should also be included in your antispoofing rules, though for brevity's sake, I left it out of Example 2-1. These ranges of IPs are specified by RFC 1918.
Therefore, in this example, your firewall would contain rules along the lines of these:
"Drop packets arriving at eth0 whose source IP is within 192.168.0.0/16 or 10.0.0.0/8"
"Drop packets arriving on eth1 whose source IP isn't within 192.168.111/24"
"Drop packets arriving on eth2 whose source IP isn't within 10.0.0.0/8"
(The last rule is unnecessary if you're not worried about IP spoofing attacks originating from your internal network.) Anti-IP-spoofing rules should be at or near the top of the applicable firewall policy.
Example 2-1 shows the iptables commands equivalent to the three previous rules.
iptables -I INPUT 1 -i eth0 -s 192.168.0.0/16 -j DROP iptables -I INPUT 2 -i eth0 -s 10.0.0.0/8 -j DROP iptables -I INPUT 3 -i eth1 -s ! 192.168.111.0/24 -j DROP iptables -I INPUT 4 -i eth2 -s ! 10.0.0.0/8 -j DROP iptables -I FORWARD 1 -i eth0 -s 192.168.0.0/16 -j DROP iptables -I FORWARD 2 -i eth0 -s 10.0.0.0/8 -j DROP iptables -I FORWARD 3 -i eth1 -s ! 192.168.111.0/24 -j DROP iptables -I FORWARD 4 -i eth2 -s ! 10.0.0.0/8 -j DROP
For complete iptables documentation, see http://netfilter.samba.org and the iptables(8) manpage.
In the words of Marcus Ranum, "That which is not explicitly permitted is prohibited." A firewall should be configured to drop any connection it doesn't know what to do with. Therefore, set all default policies to deny requests that aren't explicitly allowed elsewhere. Although this is the default behavior of netfilter, Example 2-2 lists the iptables commands to set the default policy of all three built-in chains to DROP.
iptables -P INPUT DROP iptables -P FORWARD DROP iptables -P OUTPUT DROP
Note that most firewalls, including Linux 2.4's iptables, can be configured to reject packets two different ways. The first method, usually called Dropping, is to discard denied packets "silently" ? i.e., with no notification ? to the packet's sender. The second method, usually called Rejecting, involves returning a TCP RST (reset) packet if the denied request was via the TCP protocol, or an ICMP "Port Unreachable" message if the request was via UDP.
In most cases, you'll probably prefer to use the Drop method, since this adds significant delay to port scans. Note, however, that it runs contrary to relevant RFCs, which instead specify the TCP-RST and ICMP-Port-Unreachable behavior used in the Reject method.. The Drop method is therefore used only by firewalls, which means that while a port-scanning attacker will experience delay, he'll know precisely why.
Most firewalls that support the Drop method can be configured to log the dropped packet if desired.
The most obvious job of a firewall is to block incoming attacks from external hosts. Therefore, allow incoming connections only to specific (hopefully DMZed) servers. Furthermore, limit those connections to the absolute minimum services/ports necessary ? e.g., to TCP 80 on your public web server, TCP 25 on your SMTP gateway, etc.
A central assumption with DMZs is that its hosts are at significant risk of being compromised. So to contain this risk, you should restrict traffic out of the DMZ to known-necessary services/ports. A DMZed web server, for example, needs to receive HTTP sessions on TCP 80, but does not need to initiate sessions on TCP 80, so it should not be allowed to. If that web server is somehow infected with, say, the Code Red virus, Code Red's attempts to identify and infect other systems from your server will be blocked.
Give particular consideration to traffic from the DMZ to your internal network, and design your environments to minimize the need for such traffic. For example, if a DMZed host needs to make DNS queries, configure it to use the DNS server in the DMZ (if you have one) rather than your internal DNS server. A compromised DMZ server with poorly controlled access to the Internet is a legal liability due to the threat it poses to other networks; one with poorly controlled access into your internal network is an egregious threat to your own network's security.
It's common practice to configure firewalls with the philosophy that "inbound transactions are mostly forbidden, but all outbound transactions are permitted." This is usually the result not only of politics ("surely we trust our own users!"), but also of expedience, since a large set of outbound services may legitimately be required, resulting in a long list of firewall rules.
However, many "necessary" outbound services are, on closer examination, actually "desirable" services (e.g., stock-ticker applets, Internet radio, etc.). Furthermore, once the large list of allowed services is in place, it's in place: requests for additional services can be reviewed as needed.
There are two reasons to restrict outbound access from the internal network. First, it helps conserve bandwidth on your Internet connection. Certainly, it's often possible for users to pull audio streams in over TCP 80 to get around firewall restrictions, but the ramifications of doing so will be different than if outbound access is uncontrolled.
Second, as with the DMZ, restricting outbound access from the inside helps mitigate the risk of compromised internal systems being used to attack hosts on other networks, especially where viruses and other hostile code is the culprit.
By now, there should be no mistaking my stance on proxying firewalls: if you have the technical wherewithal and can devote sufficient hardware resources, Application-Gateway firewalls provide superior protection over even stateful packet-filtering firewalls. If you must, use application proxies for some services and packet-filtering only part of the time. (Proxying firewalls nearly always let you use some amount of filtering, if you so choose.)
Linux 2.4's netfilter code, while a marked improvement over 2.2's ipchains, will be even better if/when Balazs Scheidler adds Linux 2.4 support to his open source Zorp proxy suite. (It's at least partly supported now.)
My final piece of firewall advice is that you must avoid the trap of ever considering your firewall to be a provider of absolute security. The only absolute protection from network attacks is a cut network cable. Do configure your firewall as carefully and granularly as you possibly can; don't skip hardening your DMZ servers, for example, on the assumption that the firewall provides all the protection they need.
In particular, you should harden publicly accessible servers such as those you might place in a DMZ, as though you have no firewall at all. "Security in depth" is extremely important: the more layers of protection you can construct around your important data and systems, the more time-consuming and therefore unattractive a target they'll represent to prospective attackers.