Large Packet Attacks (AKA Ping of Death) --------------------------------- [ Introduction ] Recently, the Internet has seen a large surge in denial of service attacks. A denial of service attack in this case is simply an action of some kind that prevents the normal functionality of the network. It denies service. This trend began a few months back with TCP SYN flooding and continues with the "large packet attack". In comparison with SYN flooding, the large packet attack is a much more simple attack in both concept (explained below) and execution (the attack can be carried out by anyone with access to a Windows 95 machine). TCP SYN flooding is more complex in nature and does not exploit a flaw so much as it exploits an implementation weakness. The large packet attack is also much more devastating then TCP SYN flooding. It can quite simply cause a machine to crash, whereas SYN flooding may just deny access to mail or web services of a machine for the duration of the attack. For more information on TCP SYN flooding see Phrack 49, article 13. (NOTE: The large packet attack is somewhat misleadingly referred to as 'Ping of Death` because it is often delivered as a ping packet. Ping is a program that is used to test a machine for reachablity to see if it alive and accepting network requests. Ping also happens to be a convenient way of sending the large packet over to the target.) The large packet attack has caused no end of problems to countless machines across the Internet. Since its discovery, *dozens* of operating system kernels have been found vulnerable, along with many routers, terminal servers, X-terminals, printers, etc. Anything with a TCP/IP stack is in fact, potentially vulnerable. The effects of the attack range from mild to devastating. Some vulnerable machines will hang for a relatively short period time then recover, some hang indefinitely, others dump core (writing a huge file of current memory contents, often followed by a crash), some lose all network connectivity, many rebooted or simply gave up the ghost. [ Relevant IP Basics ] Contrary to popular belief, the problem has nothing to do with the `ping` program. The problem lies in the IP module. More specifically, the problem lies the in the fragmentation/reassembly portion of the IP module. This is portion of the IP protocol where the packets are broken into smaller pieces for transit, and also where they are reassembled for processing. An IP packet has a maximum size constrained by a 16-bit header field (a header is a portion of a packet that contains information about the packet, including where it came from and where it is going). The maximum size of an IP packet is 65,535 (2^16-1) bytes. The IP header itself is usually 20 bytes so this leaves us with 65,515 bytes to stuff our data into. The underlying link layer (the link layer is the network logically under IP, often ethernet) can seldom handle packets this large (ethernet for example, can only handle packets up to 1500 bytes in size). So, in order for the link layer to be able to digest a large packet, the IP module must fragment (break down into smaller pieces) each packet it sends to down to the link layer for transmission on the network. Each individual fragment is a portion of the original packet, with its own header containing information on exactly how the receiving end should put it back together. This putting the individual packets back together is called reassembly. When the receiving end has all of the fragments, it reassembles them into the original IP packet, and then processes it. [ The attack ] The large packet attack is quite simple in concept. A malicious user constructs a large packet and sends it off. If the destination host is vulnerable, something bad happens (see above). The problem lies in the reassembly of these large packets. Recall that we have 65,515 bytes of space in which to stuff data into. As it happens, a few misbehaved applications (and some specially crafted evil ones) will allow one to place slightly more data into the payload (say 65,520 bytes). This, along with a 20 byte IP header, violates the maximum packet size of 65,535 bytes. The IP module will then simply break this oversized packet into fragments and eschew them to their intended destination (target). The receiving host will queue all of the fragments until the last one arrives, then begin the process of reassembly. The problem will surface when the IP module finds that the packet is in fact larger than the maximum allowable size as an internal buffer is overflowed. This is where something bad happens (see above). [ Vulnerability Testing and Patching ] Testing to see if a network device is vulnerable is quite easy. Windows NT and Windows 95 will allow construction of these oversized packets without complaining. Simply type: `ping -l 65508 targethost`. In this case, we are delivering an oversized IP packet inside of a ping packet, which has a header size of 8 bytes. If you add up the totals, 20 bytes of IP header + 8 bytes of ping header + 65,508 bytes of data, you get a 65,536 byte IP packet. This is enough to cause affected systems to have problems. Defense is preventative. The only way to really be safe from this attack is to either ensure your system is patched, or unplug its network tap. There are patches available for just about every vulnerable system. For a copious list of vulnerable systems and patches, check out a 'Ping of Death' webpage near you. daemon9 Editor, Phrack Magazine (daemon9@netcom.com)