Perspective: Defending the DNS

Paul Mockapetris, the inventor of the domain name system, says that a recent denial-of-service attack should serve as a wake-up call for security.

4 min read
The domain name system--the global directory that maps names to Internet Protocol addresses--was designed to distribute authority, making organizations literally "masters of their own domain." But with this mastery comes the responsibility of contributing to the defense of the DNS.

The distributed denial-of-service (DDoS) attacks against the DNS root servers on Oct. 21, 2002, should serve as a wake-up call. The attack was surprisingly successful--most of the root servers were disrupted by a well-known attack strategy that should have been easily defeated. Future attacks against all levels of the DNS--the root at the top; top-level domains like .com, .org and the country codes; and individual high-profile domains--are inevitable.

The October attack was a DDoS "ping" attack. The attackers broke into machines on the Internet (popularly called "zombies") and programmed them to send streams of forged packets at the 13 DNS root servers via intermediary legitimate machines. The goal was to clog the servers, and communication links on the way to the servers, so that useful traffic was gridlocked. The assault is not DNS-specific--the same attack has been used against several popular Web servers in the last few years.

The legitimate use of ping packets is to check whether a server is responding, so a flood of ping packets is clearly either an error or an attack. The typical defense is to program routers to throw away excessive ping packets, which is called rate limiting. While this protects the server, the attack streams can still create traffic jams up to the point where they are discarded.

Excess capacity in the network can help against such attacks, as long as the additional bandwidth can't be used to carry additional attacks. By intent, root servers are deployed at places in the network where multiple Internet service providers intersect. In the October attacks, some networks filtered out the attack traffic while others did not, so a particular root server would seem to be "up" for a network that was filtering and "down" for one that was not.

Unlike most DDoS attacks, which fade away gradually, the October strike on the root servers stopped abruptly after about an hour, probably to make it harder for law enforcement to trace.

Future attacks against all levels of the DNS are inevitable.
DNS caching kept most people from noticing this assault. In very rough terms, if the root servers are disrupted, only about 1 percent of the Internet should notice for every two hours the attack continues--so it would take about a week for an attack to have a full effect. In this cat-and-mouse game between the attackers and network operators, defenders count on having time to respond to an assault.

Defending the root
The root servers are critical Internet resources, but occupy the "high ground" in terms of defensibility. The root server database is small and changes infrequently, and entries have a lifetime of about a week. Any organization can download an entire copy of the root database, check for updates once a day, and stay current with occasional reloads. A few organizations do this already.

Root server operators are also starting to deploy root servers using "anycast" addresses that allow multiple machines in different network locations to look like a single server.

Unfortunately for those of us that depend on the Internet, the attackers seem likely to strengthen their tactics and distribute new attackware.

In short, defending the DNS root is relatively easy since it is possible to minimize the importance of any root server, by creating more copies of the root database--some private, some public.

Top-level domains, or TLDs, will be much harder to defend. The copying strategy that can defend the root server will not work for most TLDs. It is much harder to protect, say, .com or .fr than to defend the root. This is because the data in TLDs is more voluminous and more volatile, and the owner is less inclined to distribute copies for privacy or commercial reasons.

There is no alternative. TLD operators must defend their DNS servers with rate-limiting routers and anycast because consumers of the TLD data cannot insulate themselves from the attacks.

Defending your organization
If your organization has an intranet, you should provide separate views of DNS to your internal users and your external customers. This will isolate the internal DNS from external attacks. Copy the root zone to insulate your organization from future DDoS attacks on the root. Consider also copying DNS zones from business partners on extranets. When DNS updates go over the Internet, they can also be hijacked in transit--use TSIGs (transaction signatures) to sign them or send updates over VPNs (virtual private networks) or other channels.

But understand that until tools for digital signatures in DNS are finished and deployed, you are going to be at risk from the DNS counterfeiting attacks that lie not too far in the future (and that have apparently already occurred in China). Unfortunately for those of us who depend on the Internet, the attackers seem likely to strengthen their tactics and distribute new attackware, while the Internet community struggles to mount a coordinated approach to DNS defense.