Distributed-denial-of-service (DDoS) attacks have become the scourge of the internet. DDoS attacks use compromised internet devices to generate enormous volumes of data and direct that data at a particular target such as a web server or router. That target either keels over due to some critical resource becoming exhausted, or it finds its connection to the internet saturated by garbage traffic.
DDoS attacks are simultaneously cheap to carry out and expensive to defend against. Almost anyone can order a DDoS attack against any target with no technical knowledge required. All that’s necessary is a website from which to order the attack (yes, such things exist) and some bitcoins with which to pay for it. The attacks generally use botnets with devices that have been compromised and infected with malware. Building internet infrastructure capable of withstanding the volume of data generated by a botnet requires costly over-engineering, commercial DDoS mitigation services or both.
Unfortunately, DDoS attacks have a special relationship to the Domain Name System: DDoS attacks both target and exploit DNS servers. By “target,” I mean that attackers frequently direct DDoS attacks at an organization’s authoritative DNS servers. These are the DNS servers responsible for advertising your DNS data to the rest of the internet; a successful DDoS attack against them will render your customers unable to visit your website or send you email. Every organization with a presence on the internet must have a set of authoritative DNS servers, and given even the most basic information — for example, one of your email addresses or the domain name of your website — a would-be attacker can find the names and addresses of those DNS servers, giving them a list of targets.
However, DNS servers are not just opportune targets of DDoS attacks. Clever attackers will use DNS servers to make their attacks more effective and to conceal their origins. This is possible for two main reasons: 1) Relatively small DNS queries can elicit large responses, and 2) DNS works over a “connectionless” protocol that’s easily spoofed.
Amplification wouldn’t be a problem if DNS responses were always sent back to the source of the query. However, DNS’s use of the User Datagram Protocol (UDP) makes it easy to spoof queries — that is, to send queries that look as though they came from another address. UDP is connectionless: Each UDP “datagram” is independent, like a postcard sent through the postal service rather than a text message in a stream of such messages. All an attacker needs to do is to use the address of his target as the source address in the packet that contains a DNS query — like writing a bogus return address on a postcard — and the DNS server will send the reply to the target rather than the real source of the query.
This makes it easy to enlist DNS servers as unwitting accomplices in a DDoS attack. An attacker can use a botnet to generate a high volume of queries to well-connected DNS servers on the internet, spoofing the source address of their target, and the DNS servers amplify the query traffic into a larger volume of response traffic. Moreover, the traffic that arrives at the target comes from the DNS servers rather than the attacker, making it difficult to trace the attack back to its origin.
Thankfully, there are several mechanisms that can help DNS servers defend against DDoS attacks. One is “anycast,” a configuration technique that lets a distributed group of DNS servers share a single address. The internet’s routing infrastructure directs queries sent to that address to the closest DNS server in the anycast group. This is efficient, of course, but it also implies that an attack launched from one part of the internet can only reach a single DNS server in an anycast group at any time. For example, a DDoS attack using a botnet based in China and targeting the anycast address used by a group of DNS servers would find all of its traffic directed to the closest DNS server in the anycast group. As a result, many organizations, including most DNS hosting companies, use anycast to make their DNS infrastructures resistant to DDoS attacks.
Newer DNS servers also incorporate a mechanism called Response Rate Limiting (RRL) to prevent their use as amplifiers in DDoS attacks. RRL limits the rate at which a particular response is sent to the source of a query. For example, if a DNS server receives too many queries for any records about Infoblox.com from the same address, it will throttle responses to that address. If the source of the query is legitimate, this won’t cause a problem: It will cache the response, making duplicate responses unnecessary. But if the queries are spoofed and the DNS server is being used as an amplifier, this will limit the amplification and therefore the damage it can do.
Companies need to anticipate the possibility that their DNS services could be the target of these attacks. Without DNS, all internet applications and services are unreachable, bringing business to a grinding halt. In fact, recent research from Infoblox found that 24% of companies lost $100,000 or more due to downtime from their last DNS attack. Today, far too many businesses put all their eggs in one basket, relying on a single cloud-based DNS provider, leaving them vulnerable to an attack like we saw on Dyn.