Spoofing a UDP Packet

When spoofing a UDP packet, you need to know the IP addresses and ports in use, which is trivial with a DNS query because the IP addresses are known (the server making the query and the server answering it) and, because it is a DNS request, the destination port is always 53. This leaves only the source port to determine, and because many operating systems simply use a static port for outgoing connections or ports incremented by one for each outgoing request, it's relatively easy for an attacker to guess.

Transaction ID

In an attempt to address the packet-spoofing issue within the DNS protocol, a transaction ID was added. A simple 16-bit number – with 65,536 possibilities – that is sent in the request and that must be copied into the answering packet theoretically prevents an attacker from blindly spoofing the replies because it must now guess the transaction ID as well.

Unfortunately, creating really good random values is surprisingly tricky, and some implementations of Bind simply use transaction IDs that increment by 1 for each request, making them completely predictable. Now you're back to the place where an attacker can easily spoof a packet and insert hostile data into a DNS server.

How the Attack Works

So how do attackers exploit this issue? The first thing they do is find a vulnerable server and a domain that they want to control (e.g., http://www.your-bank.com). Then they find a machine that is allowed to use the vulnerable server for DNS lookups.

Large ISPs – such as mine, which has two DNS servers for the city – are likely targets because compromising them gives the attackers access to thousands of clients, so compromising a single machine to launch the attack does not present a significant hurdle.