Detect TCP content injection attacks with findject

NSA's QUANTUM INSERT attack is probably the most well-known TCP packet injection attack due to the Snowden revelations regarding how GCHQ used this method to hack into Belgacom. However, the “Five Eyes” are not the only ones who perform this type of attack on the Internet. We now release a tool to help incident responders to find these types of packet injection attacks.

I had the opportunity to attend and present at SharkFest Europe last week. My presentation, titled “Dissecting Man-on-the-Side Attacks”, showed how TCP packet injection attacks can be analyzed if they have been recorded in a packet capture. In my talk I used a python script called “finject.py”, which reads PCAP files to find TCP packets with duplicate sequence numbers but different content. This script has previously only been shared with vetted parties, but as of my SharkFest presentation findject is now freely available for everyone.

Findject is not the first tool made available to detect TCP content injection attacks. Other detection methods include Suricata's reassembly_overlap_different_data alert, Fox-IT's Bro policy to check for inconsistencies in the first packet with payload, David Stainton's HoneyBadger and Martin Bruse's qisniff. Even though these are all great solutions we found that some of them didn't detect all TCP content injection attacks while others gave too many false positives. We also wanted to have a tool that was fast, portable and simple to use. This led us to create our own TCP injection detection tool.

In the example execution above we can see that findject.py detected an injected TCP packet in the file ppp0.150923_081337.pcap, while the other analyzed files contained no injections. The application layer data of the two conflicting TCP segments are printed to standard output with a header indicating whether the segment was the FIRST or LAST one. To find out which segment is the real one and which was the injected one we need to open the PCAP file in either Wireshark, tshark or CapLoader.

The tshark execution above reveals that three packets sent from the web server's IP (42.96.141.35) are carrying data and have the same sequence number (402877220). Packet 25 and 27 are actually identical, while packet 28 is smaller (170 bytes) and has a different payload. The first displayed frame in the tshark output above is the SYN+ACK packet from the TCP 3-way handshake.

So how can we determine which of packets 25, 27 and 28 are real verses injected? Look at the IP-ID and IP-TTL values! Frame 28 has IP-ID and TTL values in line with what we see in the TCP 3-way handshake (TTL=94, IP-ID=0x00002e3e), which implies that this packet is probably authentic. Frames 25 and 27 on the other hand deviate from what we would expect from the server, which tells us that these packets were likely injected (spoofed) into the TCP session through a “man-on-the-side” attack.

To learn more about findject.py and download the tool, please visit: https://www.netresec.com/?page=findject

Example captures containing TCP content injection attacks can be found on our Publicly Available PCAP Files page under the “Packet Injection Attacks / Man-on-the-Side Attacks” section.

You can also read our blog posts Covert Man-on-the-Side Attacks and Packet Injection Attacks in the Wild to learn more about TCP packet injection attacks.

Posted by Erik Hjelmvik on Tuesday, 25 October 2016 08:45:00 (UTC/GMT)

Tags: #Netresec​ #PCAP​ #packet injection​ #find​ #python​ #SharkFest​

Packet Injection Attacks in the Wild

I have previously blogged about packet injection attacks, such as the Chinese DDoS of GitHub and Covert Man-on-the-Side Attacks. However, this time I've decided to share some intelligence on real-world packet injection attacks that have been running for several months and that are still active today.

Packet Injection by Network Operators

Gabi Nakibly, Jaime Schcolnik and Yossi Rubin recently released a very interesting research paper titled “Website-Targeted False Content Injection by Network Operators”, where they analyzed packet injection attacks in the wild. Here's a snippet from the paper's abstract:

It is known that some network operators inject false content into users’ network traffic. Yet all previous works that investigate this practice focus on edge ISPs (Internet Service Providers), namely, those that provide Internet access to end users. Edge ISPs that inject false content affect their customers only. However, in this work we show that not only edge ISPs may inject false content, but also core network operators. These operators can potentially alter the traffic of all Internet users who visit predetermined websites.

The researchers analyzed 1.4 petabits of HTTP traffic, captured at four different locations; three universities and one corporation. Some of their findings have been made available as anonymized PCAP files here:
http://www.cs.technion.ac.il/~gnakibly/TCPInjections/samples.zip

We have attempted to recreate these packet injections by visiting the same URLs again. Unfortunately most of our attempts didn't generate any injected responses, but we did manage to trigger injections for two of the groups listed by Nakibly et al. (“hao” and “GPWA”).

Redirect Race between hao.360.cn and hao123.com

We managed to get very reliable packet injections when visiting the website www.02995.com. We have decided to share one such PCAP file containing a packet injection attack here:
https://media.netresec.com/pcap/hao123-com_packet-injection.pcap

This is what it looks like when loading that PCAP file into CapLoader and doing a “Flow Transcript” on the first TCP session:

We can see in the screenshot above that the client requests http://www.02995.com/ and receives two different responses with the same sequence number (3820080905):

• The first response is a “302 Found”, forwarding the client to:
  http://www.hao123.com/?tn=93803173_s_hao_pg
• The second response is a “302 Moved Temporarily”, that attempts a redirect to:
  http://hao.360.cn/?src=lm&ls=n4a2f6f3a91

Judging from the IP Time-To-Live (TTL) values we assume that the first response (hao123.com) was an injected packet, while the second response (hao.360.cn) was coming from the real webserver for www.02995.com.

If you have an eye for details, then you might notice that the injected packet doesn't use the standard CR-LF (0x0d 0x0a) line breaks in the HTTP response. The injected packet only uses LF (0x0a) as line feed in the HTTP header.

Since the injected response arrived before the real response the client followed the injected redirect to www.hao123.com. This is what the browser showed after trying to load www.02995.com:

SSL encryption is an effective protection against packet injection attacks. So if the user instead enters https://www.02995.com then the browser follows the real redirect to hao.360.cn

id1.cn redirected to batit.aliyun.com

Prior to the release of Gabi's packet injection paper, the only publicly available PCAP file showing a real-world packet injection was this one:
https://github.com/fox-it/quantuminsert/blob/master/presentations/brocon2015/pcaps/id1.cn-inject.pcap

That PCAP file was released after Yun Zheng Hu (of Fox-IT) gave a presentation titled “Detecting Quantum Insert” at BroCon 2015. A video recording of Yun Zheng's talk is available online, including a live demo of the packet injection.

We have managed to re-trigger this packet injection attack as well, simply by visiting http://id1.cn. Doing so triggers two injected HTTP responses that attempts to do a redirect to http://batit.aliyun.com/alww.html. The target page of the injected responses has a message from the Alibaba Group (aliyun.com) saying that the page has been blocked.

We have decided to also share a PCAP file containing a packet injection attack for id1.cn here:
https://media.netresec.com/pcap/id1-cn_packet-injection.pcap

This is what it looks like when that PCAP file is loaded into NetworkMiner Professional, and the Browsers tab is opened in order to analyze the various HTTP redirections:

Image: Browsers tab in NetworkMiner Professional 2.0

Here's a short recap of what is happening in our shared PCAP file for id1.cn:

• Frame 13 : http://id1.cn is opened
• Frame 18 : Real server responds with an HTML refresh leading to http://id1.cn/rd.s/Btc5n4unOP4UrIfE?url=http://id1.cn/
• Frame 20 : The client also receives two injected packets trying to do a “403 Forbidden” that redirects to http://batit.aliyun.com/alww.html. However, these injected packets arrived too late.
• Frame 24 : The client proceeds by loading http://id1.cn/rd.s/Btc5n4unOP4UrIfE?url=http://id1.cn/
• Frame 25 : Two new injected responses are sent, this time successfully redirecting the client to the Alibaba page.
• Frame 28 : The real response arrives too late.
• Frame 43 : The client opens the Alibaba page with message about the site being blocked

Protecting against Packet Injection Attacks

The best way to protect against TCP packet injection attacks is to use SSL encryption. Relying on HTTP websites to do a redirect to an HTTPS url isn't enough, since that redirect could be targeted by packet injection. So make sure to actually type “https://” (or use a browser plug-in) in order to avoid being affected by injected TCP packets.

Referenced Capture Files

The following PCAP files have been referenced in this blog post:

• http://www.cs.technion.ac.il/~gnakibly/TCPInjections/samples.zip
• https://media.netresec.com/pcap/hao123-com_packet-injection.pcap
• https://github.com/fox-it/quantuminsert/blob/master/presentations/brocon2015/pcaps/id1.cn-inject.pcap
• https://media.netresec.com/pcap/id1-cn_packet-injection.pcap

For more PCAP files, please visit our list of publicly available PCAP files here: https://www.netresec.com/?page=PcapFiles

Posted by Erik Hjelmvik on Tuesday, 01 March 2016 13:37:00 (UTC/GMT)

Tags: #NetworkMiner​ #HTTP​ #browser​ #CapLoader​ #MOTS​ #HTTPS​ #TCP​ #PCAP​