ForensicsWiki will continue to operate as it has before and will not be shutting down. Thank you for your continued support of ForensicsWiki.

Bulk extractor

From ForensicsWiki
Revision as of 14:08, 9 November 2012 by Jessek (Talk | contribs) (Updated current version number)

Jump to: navigation, search


bulk_extractor is a C++ program that scans a disk image, a file, or a directory of files and extracts useful information without parsing the file system or file system structures. The results are stored in feature files that can be easily inspected, parsed, or processed with automated tools. bulk_extractor also created a histograms of features that it finds, as features that are more common tend to be more important.

bulk_extractor is distinguished from other forensic tools by its speed and thoroughness. Because it ignores file system structure, bulk_extractor can process different parts of the disk in parallel. In practice, the program splits the disk up into 16MiByte pages and processes one page on each available core. This means that 24-core machines process a disk roughly 24 times faster than a 1-core machine. bulk_extractor is also thorough. That’s because bulk_extractor automatically detects, decompresses, and recursively re-processes compressed data that is compressed with a variety of algorithms. Our testing has shown that there is a significant amount of compressed data in the unallocated regions of file systems that is missed by most forensic tools that are commonly in use today.

Another advantage of ignoring file systems is that bulk_extractor can be used to process any digital media. We have used the program to process hard drives, SSDs, optical media, camera cards, cell phones, network packet dumps, and other kinds of digital information.

Output Feature Files

bulk_extractor now creates an output directory that has the following layout:

Processing errors.
Credit card numbers
Credit card “track 2″ informaiton, which has previously been found in some bank card fraud cases.
Internet domains found on the drive, including dotted-quad addresses found in text.
Email addresses.
Ethernet MAC addresses found through IP packet carving of swap files and compressed system hibernation files and file fragments.
EXIFs from JPEGs and video segments. This feature file contains all of the EXIF fields, expanded as XML records.
The results of specific regular expression search requests.
IP addresses found through IP packet carving.
Email message headers including Date:, Subject: and Message-ID: fields.
TCP flow information found through IP packet carving.
US and international telephone numbers.
URLs, typically found in browser caches, email messages, and pre-compiled into executables.
A histogram of terms used in Internet searches from services such as Google, Bing, Yahoo, and others.
A histogram of the domain name portion of all the URLs found on the media.
A list of all “words” extracted from the disk, useful for password cracking.
The wordlist with duplicates removed, formatted in a form that can be easily imported into a popular password-cracking program.
A file containing information regarding every ZIP file component found on the media. This is exceptionally useful as ZIP files contain internal structure and ZIP is increasingly the compound file format of choice for a variety of products such as Microsoft Office

For each of the above, two additional files may be created:

  • _stopped.txt
bulk_extractor supports a stop list, or a list of items that do not need to be brought to the user’s attention. However rather than simply suppressing this information, which might cause something critical to be hidden, stopped entries are stored in the stopped files.
  • _histogram.txt
bulk_extractor can also create histograms of features. This is important, as experience has shown that email addresses, domain names, URLs, and other informaiton that appear more frequently on a hard drive or in a cell phone’s memory can be used to rapidly create a pattern of life report.

Bulk extractor also creates a file that captures the provenance of the run:

A Digital Forensics XML report that includes information about the source media, how the bulk_extractor program was compiled and run, the time to process the digital evidence, and a meta report of the information that was found.


We have developed four programs for post-processing the bulk_extractor output:
This program reports the differences between two bulk_extractor runs. The intent is to image a computer, run bulk_extractor on a disk image, let the computer run for a period of time, re-image the computer, run bulk_extractor on the second image, and then report the differences. This can be used to infer the user’s activities within a time period.
This tool, currently under development, reads multiple bulk_extractor reports from multiple runs against multiple drives and performs a multi-drive correlation using Garfinkel’s Cross Drive Analysis technique. This can be used to automatically identify new social networks or to identify new members of existing networks.
In the bulk_extractor feature file, each feature is annotated with the byte offset from the beginning of the image in which it was found. The program takes as input a bulk_extractor feature file and a DFXML file containing the locations of each file on the drive (produced with Garfinkel’s fiwalk program) and produces an annotated feature file that contains the offset, feature, and the file in which the feature was found.
Although forensic analysts frequently make “stop lists”—for example, a lsit of email addresses that appear in the operating system and should therefore be ignored—such lists have a significant problem. Because it is relatively easy to get an email address into the binary of an open source application, ignoring all of these email addresses may make it possible to cloak email addresses from forensic analysis. Our solution is to create context-sensitive stop lists, in which the feature to be stopped is presented with the context in which it occures. The program takes the results of multiple bulk_extractor runs and creates a single context-sensitive stop list that can then be used to suppress features when found in a specific context. One such stop list constructed from Windows and Linux operating systems is available on the bulk extractor website.


The current version of bulk_extractor is 1.3. It can be downloaded from

Sample Output

Running on 2.4Ghz iMac with MacOS 10.5.8 on the nps-2009-realistic.aff disk image, bulk extractor version 0.0.10 took 21816 seconds (6 hours, 3 minutes) and produced an output with 14,160 lines.

Here are the first 200 lines:

Input file: /corp/images/nps/nps-2009-domexusers/nps-2009-realistic.aff
Starting page number: 0
Last processed page number: 2559
Time: Tue Aug 11 04:39:03 2009

Top 10 email addresses:
======================= 572 412 319 268 252 243 232 192 145 138

Top 10 email domains:
===================== 1693 630 543 470 390 376 329 298 268 267

Top 10 URLs:
===================== 6257 3030 2241
http://: 1666 1515 1513 1311 1310 1000 974

All email addresses:
==================== 572 412 319 268 252 243 232 192 145 138 135 133 115 111 110 94 91 80 69 67 67 64 53 51 48 47 43 43 41 37 36 35 34 33 33 33
DOMEXUSER2@GMAIL.COM: 32 32 31 30 29 28 28 28 28 27 27 26 25 24 24 22 21
DOMEXUSER1@GMAIL.COM: 20 20 20 20 20 20 20 20 20
CPh@99841.PA: 19 19 19 19 19 19 18 18 17 17 17 17 16 16 16 16 16 16 15 15 15 15 15 14 14 14 14 14 13 13 13 13 12
DOMEXUSER3@GMAIL.COM: 12 12 12 12 12 12 12 11 11 11
St@atus.eU: 11 11 11 11 11 11 11 11 11 11 11 11 11 11 11 11 11 10 10 10 10 10 10 10 10 10 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 8 8 8