__ __ High Octane Triage Analysis __
|| _||______ __ __________ _____ ||
|| \||___ \__| ____/ ______/___ / ____\ ||
==||=====|| | __/ |/ \ /==| / __ \ __\===]|
'======|| | \ | | \_ _| \ ___/| | ||
||____ /__|___|__/ / | \____]| | ||
=========''====\/=========/ /==|__|=====|__|======'
\ /
\/
The Binary Refinery™ is a collection of Python scripts that implement transformations of binary data such as compression and encryption. We will often refer to it simply by refinery, which is also the name of the corresponding package. The scripts are designed to exclusively read input from stdin and write output to stdout. This way, the individual units can be chained with the piping operator | on the commandline to perform more complex tasks. The project was born to aid with malware triage, and is an attempt to implement something like CyberChef on the commandline.
The main philosophy of the refinery is that every script should be a unit in the sense that it does one job. It is always a case by case decision, but it is generally desirable to reduce the number of possible arguments of each script to a minimum and prefer strong capsulation if some functionality could be provided by a separate unit.
The main documentation is generated automatically from the source code; it serves primarily as a reference but also contains the specification for the three fundamental concepts of the toolkit: framing, multibin arguments, and meta variables.
In recognition of the fact that reference documentation can be somewhat dry, there is an ongoing effort to produce a series of tutorials and there are a few additional resources:
- OALabs was kind enough to let me demo the toolkit in a feature video.
- Johannes Bader wrote an amazing blog post about how to analyze TA551 malspam with binary refinery. It should really be turned into a tutorial at some point.
The Binary Refinery is (c) 2019 Jesko Hüttenhain, and published under a 3-Clause BSD License. This repository also contains a copy of the full license text. If you want to do something with it that's not covered by this license, please feel free to contact the author.
The refinery requires at least Python 3.7. It is recommended to install it into its own virtual environment: The package has a lot of dependencies, and installing it into your global Python is somewhat prone to dependency conflicts. Also, since the toolkit introduces a large number of new commands, there is a good chance that some of these will clash on some systems, and keeping them in their own separate virtual environment is one way to prevent that.
If you want to have all refinery commands available in your shell at all times, you also have the option to choose a prefix for the installation, which will be put in front of every command shim that is installed. For example, if you choose r. as your prefix, then the emit unit will be installed as the command r.emit. An added benefit is that you can type r. and hammer Tab twice to get a list of all available refinery commands. Note however that no prefix is assumed in documentation and it is a development goal of refinery to not clash on most systems. The author does not use a prefix and provides this option as a safety blanket.
If you specify the special prefix ! (a single exclamation mark), then the refinery will be installed in library mode and no command shims will be created at all.
To install or update refinery manually, simply set the environment variable REFINERY_PREFIX to the prefix you want and use pip. For example:
REFINERY_PREFIX=r. pip3 install -U binary-refinery
to install refinery into the current Python environment with prefix r.. As mentioned above, the special prefix ! will have the effect that no shell commands are created and the refinery will be installed only as a library. If you want to install the current refinery HEAD, you can install the master of this repository instead. The following will install the very latest refinery commit:
pip3 install -U git+git://github.com/binref/refinery.git
To update refinery, the most reliable option is to run
pip uninstall -y binary-refinery
and then install it again. This will be a lot faster than installing it, because resolving dependencies is what takes a lot of time during the install of binary refinery.
There are some units that have rather heavy-weight dependencies. For example, pcap is the only unit that requires a packet capture file parsing library. These libraries are not installed by default to keep the installation time for refinery at a reasonable level for first-time users. The corresponding units will tell you what to do when their dependency is missing:
$ emit data.pcap | pcap [| peek ]
(13:37:00) failure in pcap: dependencies missing; install 'pypcapkit[scapy]'
You can then install these missing dependencies manually. If you do not want to be bothered by missing dependencies and don't mind a long refinery installation, you can install the package as follows:
pip install -U binary-refinery[all]
which will install all dependencies on top of the required ones.
Alternatively, you can clone this repository and use the scripts update.sh (on Linux) or update.ps1 (on Windows) to install the refinery package into a local virtual environment. The installation and update process for this method is to simply run the script:
- it pulls the repository,
- activates the virtual environment,
- uninstalls
binary-refinery, - and then installs
binary-refinery[all].
You can also generate all documentation locally. To do so, execute the run-pdoc3.py script. This will fail unless you run it from an environment where binary refinery has been installed as a Python package. To run it, you have to specify the path of a virtual environment as the first command line argument to run-pdoc3.py, which will cause the script to run itself again using the interpreter of that environment. If you are certain that you want to run run-pdoc3.py, there is a command line switch to force the script to run with the current default Python interpreter. The script installs the pdoc3 package and uses it to generate a HTML documentation for the refinery package. The documentation can then be found in the subdirectory html directly next to this readme file.
The tutorials are Jupyter notebooks which you can simply run and execute if your virtual environment has Jupyter installed. It's worth pointing out that Visual Studio Code has very comfortable support for Jupyter.
The units emit and dump play a special role: The former is for outputting data while the latter is for dumping data to the clipboard or to disk. As an example, consider the following pipeline:
emit M7EwMzVzBkI3IwNTczM3cyMg2wQA | b64 | zl | hex
Here, we emit the string M7EwMzVzBkI3IwNTczM3cyMg2wQA, base64-decode it using b64, zlib-decompress the result using zl, and finally hex-decode the decompressed data. Each unit performs the "decoding" operation of a certain transformation by default, but some of them also implement the reverse operation. If they do, this is always done by providing the command line switch -R, or --reverse. You can produce the above base64 string using the following command because hex, zl, and b64 all provide the reverse operation:
emit "Hello World" | hex -R | zl -R | b64 -R
Given a file packed.bin containing a base64 encoded payload buffer, the following pipeline extracts said payload to payload.bin:
emit packed.bin | carve -l -t1 b64 | b64 | dump payload.bin
The carve unit can be used to carve blocks of data out of the input buffer, in this case it looks for base64 encoded data, sorts them by length (-l) and returns the first of these (-t1), which carves the largest base64-looking chunk of data from packed.bin. The data is then base64-decoded and dumped to the file payload.bin.
The unit pack, will pick all numeric expressions from a text buffer and turn them into their binary representation. A simple example is the pipeline
emit "0xBA 0xAD 0xC0 0xFF 0xEE" | pack | hex -R
which will output the string BAADC0FFEE.
Extract the largest piece of base64 encoded data from a BLOB and decode it:
emit file.exe | carve -ds b64
Carve a ZIP file from a buffer, pick a DLL from it, and display information about it:
emit file.bin | carve-zip | xtzip file.dll | pemeta
List PE file sections with their corresponding SHA-256 hash:
emit file.exe | vsect [| sha256 -t | cfmt {} {path} ]]
Recursively list all files in the current directory SHA-256 hash:
ef "**" [| sha256 -t | cfmt {} {path} ]]
Extract indicators from all files recursively enumerated inside the current directory:
ef "**" [| xtp -qn6 ipv4 socket url email | dedup ]]
Convert the hard-coded IP address 0xC0A80C2A in network byte order to a readable format:
emit 0xC0A80C2A | pack -EB4 | pack -R [| sep . ]
Perform a single byte XOR brute force and attempt to extract a PE file payload in every iteration:
emit file.bin | rep 0x100 [|cm| xor var:index | carve-pe -R | peek | dump {name} ]
Extract a RemCos C2 server:
emit c0019718c4d4538452affb97c70d16b7af3e4816d059010c277c4e579075c944 \
| perc SETTINGS [| put keylen cut::1 | rc4 cut::keylen | xtp socket ]
Extract an AgentTesla configuration:
emit fb47a566911905d37bdb464a08ca66b9078f18f10411ce019e9d5ab747571b40 \
| dnfields [| aes x::32 --iv x::16 -Q ]] \
| rex -M "((??email))\n(.*)\n(.*)\n:Zone" addr={1} pass={2} host={3}
Extract the PowerShell payload from a malicious XLS macro dropper:
emit 81a1fca7a1fb97fe021a1f2cf0bf9011dd2e72a5864aad674f8fea4ef009417b [ \
| xlxtr 9.5:11.5 15.15 12.5:14.5 [ \
| scope -n 3 | chop -t 5 [| sorted -a | snip 2: | sep ] \
| pack 10 | alu --dec -sN B-S ]] \
| dump payload.ps1
And get the domains for the next stage:
emit payload.ps1
| carve -sd b64 | zl | deob-ps1
| carve -sd b64 | zl | deob-ps1
| xtp -f domain
Extract the configuration of unpacked HawkEye samples:
emit ee790d6f09c2292d457cbe92729937e06b3e21eb6b212bf2e32386ba7c2ff22c \
| put cfg perc[RCDATA]:c:: [\
| xtp guid | pbkdf2 48 rep[8]:h:00 | cca xvar:cfg | aes -Q x::32 --iv x::16 ] \
| dnds
Warzone RAT:
emit 4537fab9de768a668ab4e72ae2cce3169b7af2dd36a1723ddab09c04d31d61a5 \
| vsect .bss | struct I{key:{}}{} [\
| rc4 xvar:key | struct I{host:{}}{port:H} {host:u16}:{port} ]
Extract payload from a shellcode loader and carve its c2:
emit 58ba30052d249805caae0107a0e2a5a3cb85f3000ba5479fafb7767e2a5a78f3 \
| rex yara:50607080.* [| struct LL{s:L}{} | xor -B2 accu[s]:$msvc | xtp url ]
Extract the malicious downloader payload from a malicious document's text body:
emit ee103f8d64cd8fa884ff6a041db2f7aa403c502f54e26337c606044c2f205394 \
| doctxt | repl drp:c: | carve -s b64 | rev | b64 | rev | ppjscript
Assume that data is a file which was encrypted with 256 bit AES in CBC mode. The key was derived from the secret passphrase swordfish using the PBKDF2 key derivation routine using the salt s4lty. The IV is prefixed to the buffer as the first 16 bytes. It can be decrypted with the following pipeline:
emit data | aes --mode cbc --iv cut::16 pbkdf2[32,s4lty]:swordfish
Here, both cut:0:16 and pbkdf2[32,s4lty]:swordfish are multibin arguments that use a special handler. In this case, cut:0:16 extracts the slice 0:16 (i.e. the first 16 bytes) from the input data - after application of this multibin handler, the input data has the first 16 bytes removed and the argument iv is set to these exact 16 bytes. The final argument specifies the 32 byte encryption key: The handler pbkdf2[32,s4lty] on the other hand instructs refinery to create an instance of the pbkdf2 unit as if it had been given the command line parameters 32 and s4lty in this order and process the byte string swordfish with this unit. As a simple test, the following pipeline will encrypt and decrypt a sample piece of text:
emit "Once upon a time, at the foot of a great mountain ..." ^
| aes pbkdf2[32,s4lty]:swordfish --iv md5:X -R | ccp md5:X ^
| aes pbkdf2[32,s4lty]:swordfish --iv cut:0:16