Packet fragmentation vs the Intrusion Detection System

Tuesday, December 08, 2009

Fred Williams



1) Snort -

2) Metasploit -

How well does Snort IDS handle packet fragments when the fragments could contain a potentially malicious software attack? Let's read on.... I found a really great article written in 2007 on how an author setup a lab environment to test this theory. Here is the URL:

Before we get into the article let's explore some background information on packet fragmentation. Let's find out what exactly is packet fragmentation and how packets are fragmented.

What is packet fragmentation?

If IP packets are coming into your network and one or more packets are larger than the network's defined Maximum Transmission Unit (MTU), the packet(s) must be broken up into smaller pieces in order to allow the packets to traverse the network. These smaller packets are called fragments. For more information on the protocols involved with packet fragmentation and reassembly, you can visit the RFC's at

What is the MTU?

I just mentioned the MTU. What is an MTU? The MTU is defined as the largest datagram that can be sent over the network. The network admin has some default sizes to work with. For example, on Ethernet networks the default MTU is 1,500 bytes.

What fields are involved in packet fragmentation?

Answer: look to the IP header. Every IP packet has an IP header that stores information about the packet. Some of the fields on the IP header are the IP version (ipv4, ipv6), the identification field, source and destination IP addresses and total length. Three fields involved in packet fragmentation are (1) identification (2) fragbits or flags and (3) fragment offset.

Let's look at these 3 fields in more depth..... For more information, you can peruse the RFCs at this address:

  1. IP header identification field: The identification field is a 16 bit field provided by the sender that aids in packet reassembly.
  2. Fragbits: The fragbits field is a 3 bit field that contains 3 control flags. Bit 0 is reserved and must always be 0. Bit 1 is the DF fragbit that stands for "Don't fragment". This bit can have 2 values: 0 (may fragment) or 1 (don't fragment). Bit 2 is the MF fragbit that stands for "More fragments". This bit can also have 2 values: 0 (last fragment) or 1 (more fragments to come)
  3. Fragment offset: This is a 13 bit field that indicates where in the datagram this fragment belongs. The first fragment will always have an offset = 0

IP Packet fragmentation example

Given the information from above let's take a look at a simple example of how a packet is fragmented. Say we have a 2366 byte packet coming into our Ethernet network. You may remember that Ethernet networks MTU is 1500 bytes so our packet will need to be divided into 2 fragments.

  1. Fragment 1: The first packet will be 1500 bytes in length. The first packet's DF fragbit will be set to 0 that means "may fragment" and the MF fragbit will be set to 1 which means More fragments to come. Since this is the first fragment, the fragment offset will be 0
  2. Fragment 2: The second packet's DF flag will still be set to 0 to mean "May fragment" but the MF flag will be set to 0 that means this will be the last fragment. The fragment offset of this packet will be somewhere around 910 or so. This is calculated based upon the data portion of previous packets and doesn't include the 20 or 40 bytes for the packet header lengths.
How does packet fragmentation lead to attacks?

Let's take our 2 packet example from earlier and see what an attacker may be able to with it.
What if an attacker wanted to telnet into our remote computer using TCP port 23 for whatever reason and what if that port is blocked by packet filtering firewall. The attacker would probably do a port scan and see which ports are open on our remote computer and what if he sees that the SMTP port 25 is open? Most likely he decides to craft a packet fragmentation attack where the first packet has the following: the fragment offset of 0 (since its the first packet), the DF flag = 0 (may fragment), the MF flag = 1 (more fragments) and the destination port = 25.

The second packet the attacker will force the fragment offset to be 1 - the reason is that offset 1 will be so low that instead of appending it to the first packet it will overwrite everything in the first packet except the first 8 bits. The attacker will also set the second packet's TCP destination port to port 23 - which normally would be blocked but not in this case since the attacker has set a fragmented packet. The packet filter sees that the offset of this second packet is greater than 0 so it will think that this is a fragment of another packet and won't put it through the ruleset.

When the 2 packets arrive at the target host, it will be reassembled and most of the first packet will be overwritten by the second and the combined packet will be allowed to go to port 23.

The article.

Now, we finally get to the actual article. In the article the author states that intrusion detection systems have traditionally had problems with packet reassembly and that they still have issues today. Even though IDS's have gotten a lot smarter in how it reassembles packets, the author wanted to see how well Snort IDS performs when it comes to detecting some simple packet fragmentation attacks. The article's goals are to (1) Show how well Snort can detect simple packet fragmentation attacks and (2) use Metasploit and fragrouter to fragment packets sent to a victim computer running Snort IDS.

The author sets up a lab environment to launch and measure the attacks. The author sets up 3 computers - the attack computer will have Metasploit installed. The middle computer will have fragrouter installed and the victim computer will have the packet sniffer and Snort installed.

What is fragrouter?

We have all seen and used Metasploit and a packet sniffer (Wireshark) so I won't explain those two software tools but I"ll briefly describe fragrouter. According to, fragrouter is one of the top 100 security tools of all time. It is used mainly as a "Network Intrusion detection evasion toolkit". Packets are sent to fragrouter which transforms them into fragments and forwarded to the victim. The author started fragrouter with the F1 option which means to send the fragmented packets in order. Other options, like F2, F3, etc are meant to allow the attacker to send packets in any order they wish.

You can see all of fragrouter's options by listing with the fragrouter -help option. You can see all of the options to run fragrouter with different combinations.

Once the author has his lab environment setup, he is ready to launch the attack. In a nutshell, the author wants to launch a Metasploit MS03-026 attack that is routed through a middle computer running fragrouter. Fragrouter will break up the attack in multiple fragments and send them on to a victim computer. The victim computer is running Snort IDS and the author wishes to see how well Snort detects the attack through fragmented packets. So, Snort has to reassemble the packets, detect the attack, and list any fragmented packets it finds.

The Metasploit MS03-026 attack targets a buffer overflow vulnerability in Microsoft XP. The author then used the win32-reverse payload to actually try to get a remote shell access on the victim computer. Once the author gains shell access on the victim computer, he stops the attack and views Snort's statistics. Since the route was setup to forward all packets from the attack computer to the victim computer through the middle computer running fragrouter, the victim computer should see fragmented packets. I actually tried this exploit on my VMWare setup and attacking the XP VM. Metasploit told me that this VM was not susecpible to this vulnerability so I would imagine that it could be a service pack issue.

Fragrouter does produce output information to the console. It will list the fragments as well as the offset as it is doing the fragmentation.

What did Snort detect? Snort logged 7 items and 2 alerts. The interesting thing was that Snort detected 271 fragmented IP packets during the attack session. Without using fragrouter, the author performed the same attack with the same payload and Snort detected 0 fragmented packets.

Snort foiled our attempts at being stealthy by using packet fragmentation. It detected the exploit use as well as detecting the fragmented packets.

Conclusion: We can conclude from the experiment that indeed Snort is effective at detecting some simple packet fragmentation attacks. We have been shown how to use Metasploit to launch an attack going through a middle computer that fragments the attack into fragments and sent on to the victim computer with Snort running on it. There are certainly more scenarios at using fragrouter and Metasploit and that can be a future point of experimentation and is left up to the reader.
Possibly Related Articles:
IDS/IDP Enterprise Security
IDS Snort fragmentation
Post Rating I Like this!
derek thomas Was Snort installed on the Windows XP box? If so, wouldn't XP reassemble the fragmented packets?
Fred Williams Yes, both Snort and XP would reassemble the packets. Normally, the destination host is the device responsible for packet reassembly. However intermediate devices such as routers can and will reassemble packets.

Snort has built in abilities to reassemble packets. The idea behind the original article is testing Snort's ability to reassemble packets and detect exploits. You can see the results were positive.
The views expressed in this post are the opinions of the Infosec Island member that posted this content. Infosec Island is not responsible for the content or messaging of this post.

Unauthorized reproduction of this article (in part or in whole) is prohibited without the express written permission of Infosec Island and the Infosec Island member that posted this content--this includes using our RSS feed for any purpose other than personal use.