bulk_extractor is a high-performance digital forensics exploitation tool. It is a "get evidence" button that rapidly scans any kind of input (disk images, files, directories of files, etc) and extracts structured information such as email addresses, credit card numbers, JPEGs and JSON snippets without parsing the file system or file system structures. The results are stored in text files that are easily inspected, searched, or used as inputs for other forensic processing. bulk_extractor also creates histograms of certain kinds of features that it finds, such as Google search terms and email addresses, as previous research has shown that such histograms are especially useful in investigative and law enforcement applications.

Unlike other digital forensics tools, bulk_extractor probes every byte of data to see if it is the start of a sequence that can be decompressed or otherwise decoded. If so, the decoded data are recursively re-examined. As a result, bulk_extractor can find things like BASE64-encoded JPEGs and compressed JSON objects that traditional carving tools miss.

This is the bulk_extractor 2.1 development branch! It is reliable, but if you want to have a well-tested production quality release, download a release from https://github.com/simsong/bulk_extractor/releases.

We recommend building from sources. We provide a number of bash scripts in the etc/ directory that will configure a clean virtual machine:

git clone --recurse-submodules https://github.com/simsong/bulk_extractor.git
./bootstrap.sh
./configure
make
make install

For detailed instructions on installing packages and building bulk_extractor, read the wiki page here: https://github.com/simsong/bulk_extractor/wiki/Installing-bulk_extractor

For more information on bulk_extractor, visit: https://forensics.wiki/bulk_extractor

This release of bulk_extractor requires C++17 and has been tested to compile on the following platforms:

• Amazon Linux as of 2023-05-25
• Fedora 36 (most recently)
• Ubuntu 20.04LTS
• MacOS 13.2.1

You should always start with a fresh VM and prepare the system using the appropriate prep script in the etc/ directory.

• Debian 10 (is not supported for native builds))

If you are writing a scientific paper and using bulk_extractor, please cite it with:

Garfinkel, Simson, Digital media triage with bulk data analysis and bulk_extractor. Computers and Security 32: 56-72 (2013)

• Science Direct
• Bibliometrics
• Author's website

@article{10.5555/2748150.2748581,
author = {Garfinkel, Simson L.},
title = {Digital Media Triage with Bulk Data Analysis and Bulk_extractor},
year = {2013},
issue_date = {February 2013},
publisher = {Elsevier Advanced Technology Publications},
address = {GBR},
volume = {32},
number = {C},
issn = {0167-4048},
journal = {Comput. Secur.},
month = feb,
pages = {56–72},
numpages = {17},
keywords = {Digital forensics, Bulk data analysis, bulk_extractor, Stream-based forensics, Windows hibernation files, Parallelized forensic analysis, Optimistic decompression, Forensic path, Margin, EnCase}
}

The following environment variables can be set to change the operation of bulk_extractor:

Other hints for debugging:

• Run -xall to run without any scanners.
• Run with a random sampling of 0.001% to debug reading image size and a few quick seeks.

Note: Currenlty bulk_extractor 2.1 does not build on windows, but 2.0 does.

If you wish to build for Windows, you should cross-compile from a Fedora system. Start with a clean VM and use these commands:

$ git clone --recurse-submodules https://github.com/simsong/bulk_extractor.git
$ cd bulk_extractor/etc
$ bash CONFIGURE_FEDORA36_win64.bash
$ cd ..
$ make win64

Renamed jpeg_carved feature recorder to jpeg, so that the jpeg carve mode can be set with -S jpeg_carve_mode=2, rather than -S jpeg_carved_carve_mode=2, which was confusing.

bulk_extractor 2.0 (BE2) is now operational. Although it works with the Java-based viewer, we do not currently have an installer that runs under Windows.

BE2 requires C++17 to compile. It requires https://github.com/simsong/be13_api.git as a sub-module, which in turn requires dfxml as a sub-module.

The project took longer than anticipated. In addition to updating to C++17, It was used as an opportunity for massive code refactoring and general increase in code quality, testability and reliability. An article about the experiment will appear in a forthcoming issue of ACM Queue