McAfee Labs

Travnet Botnet Steals Huge Amount of Sensitive Data

0
By on Apr 23, 2013

In a McAfee Labs blog by my colleague Vikas Taneja last month, he discussed high-level functioning in the malware Travnet. Since then we have continued to analyze different samples and now classify Travnet as a botnet rather than a Trojan because of the presence of control code, and the malware’s ability to wait for further commands from the malicious control server.

The Travnet bot not only steals sensitive information from a victim’s machine; it also steals document files. Generally speaking, we store most of our sensitive information in Office files, PDFs, etc. Using data compression and data-encoding methods allows Travnet to steal huge amount of data including large files.

The bot at first gathers sensitive information about victim’s machine. Then searches for document files (doc, docx, xls, xlsx, txt, rtf, pdf). Here is snippet of code:Travnet_source

The preceding code includes computer name, IP address, username, operating system, list of running processes, IP config details, and information about different accounts present on the system. The malware creates the file system_t.dll to store this information in plain text. It also creates the file travelbackinfo-(SystemTime).dll, which will be used in an HTTP GET request.

The data stored in the file can be huge, depending upon running processes and IP config details. The bot will use data compression and encoding methods to send the sensitive data to a remote server. The packet capture looks like this:

Travnet_get_Request

The bot sends the stolen data with the parameter “&filetext,” which starts with “begin::.” But the compressed file can be too big to send over the HTTP, so the bot sends the compressed file in chunks of 1,024 bytes. To track this, it uses the parameter “&filestart.” The bot appends the string “::end” to signal the end of the file.

Data compression and encoding techniques

The bot processes the original data in two passes:

  • In the first pass, it uses a data compression method similar to LZSS (Lempel–Ziv–Storer–Szymanski) to compress the original data
  • In the second pass, it encodes the compressed data using custom Base64

First pass data compression

The bot’s data compression maintains a dictionary (a sliding window) of previously seen data that is similar to data compression with LZSS.

The bot uses a similar method to maintain a large sliding window size (to achieve a high compression ratio) but outputs variable-length “Length- Offset” pairs (the number of bits required to represent the number). We have not seen yet any references or implementation that outputs variable lengths and variable offsets, so for now we will call this method a variant of the LZSS data compression algorithm.

The bot starts compression by reading original data in chunks of 65,536 bytes (so it has to maintain sliding windows of this size). The final output of compression will be in chunks following this format:

Original Length (2 bytes) + Compressed Length (2 bytes) + Compressed Data

This method achieves a high compression ratio and reduces the size of the original data, allowing the bot to upload large files on the remote server. The decompression process is very easy to write because it does not need to search for the longest match but needs only to take care of variable-length values.

Second pass custom Base64 encoding

The Travnet bot uses custom Base64 encoding to encode the compressed binary data. The key and character set used in standard Base64 is “ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/” with “=” used for padding; the key used by the bot is “ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789-/” with “*” used for padding.

We wrote a small tool to decompress the data stolen by Travnet.

travnet_tool

As we look at the output, we see the size of the decompressed file (the original data) is much higher than that of the compressed file. Let’s now look the decompressed data:

Travnet_decompressed_data

The preceding is the original data stolen from the victim’s machine. Interestingly, the unreadable characters in the decompressed file are in Chinese. While writing the sensitive information in a DLL file, the bot writes some hardcoded strings that are in Chinese. If we convert those strings to English, here is how the file looks:

Travnet_chinese_converted_data

Stealing files

The bot doesn’t stop; it steals more data. Next we see the functions called by the bot:

Travnet_send_files

The bot will send the following:

  • A file containing lists of all filenames on the system drives
  • All files that have doc, docx, xls, xlsx, txt, rtf, and pdf extensions
  • All files from victim’s desktop

Once it sends all the files to the remote server, the bot will go into sleep mode and wait for further commands.

Server commands

  • UNINSTALL
  • UPDATE
  • RESET
  • UPLOAD

Next we see a command from the server telling the bot to upload more data:Travnet_cnc_command

Although the botnet uses a simple mechanism to infect and steal information, a few elements make a Travnet botnet unique:

  • Using lossless data compression to steal large data files
  • Stealing documents files with extensions doc, docx, xls, xlsx, txt, rtf, and pdf
  • Stealing all files on the system drives

These unique features and the presence of Chinese strings lead us to conclude that the Travnet botnet may be a targeted attack for stealing sensitive data. We suspect the attackers are using the initial data–computer information, IP’s–to steal sensitive data from a particular group or identity. We also believe that the data uploaded to malicious severs is actively monitored by the attackers. We have found new domains registered to carry out the attack. We believe that huge amounts of data have been stolen from victims whose machines were infected with Travnet.

I would like to thank my colleagues Vikas Taneja, Anil Aphale, Arunpreet Singh, and Subrat Sarkar for their research and assistance.

Leave a Reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>